An Investigation of the Impact of the Local Labour Markets on Staff Shortages and Staff Mix of Hospitals in England and France

Jean-Baptiste Combes

National Diploma in Economics (University of Marseille, France)

MSc in Statistics for Health Economics (National School of Statistics and Information Analysis, Rennes, France)

A thesis presented for the Degree of Doctor of Philosophy (PhD) at the University of Aberdeen

June 2012

Declaration of originality I, Jean-Baptiste Combes, declare that this thesis is my own work and has not been previously submitted for a degree or award in any university. It does not contain any material or verbatim extracts previously previously published or written by another person except where due reference is made in the text. Jean-Baptiste Combes

i

Table of Contents Declaration of originality............................................................................................................................i Acknowledgements....................................................................................................................................vi Executive Summary....................................................................................................................................xi Chapter 1 Introduction.................................................................................................................................1 Chapter 2 Literature......................................................................................................................................5 2.1. Local Labour Markets.....................................................................................................................5 2.1.1. Compensating wage differentials, a theoretical perspective.......................................................................6 2.1.2. The theory of compensating geographical wage differentials, empirical results and discussion.......6 2.1.3. The role of pay setting........................................................................................................................................ 14

2.2. Impact of the different pay setting institutions on public services...................................19 2.3. Literature review of research on shortage of staff................................................................24 2.4. Literature review of research on staff-mix..............................................................................31 2.4.1. Staff-mix and the delivery of care.................................................................................................................... 32 2.4.2. Drivers of staff-mix.............................................................................................................................................. 35 2.4.3. Complementarity/substitutability of health professionals........................................................................39

2.5. Paving the way for empirical analyses ....................................................................................42 2.5.1. Cross effects of labour markets........................................................................................................................ 42 2.5.2. Staff-mix and local labour markets.................................................................................................................. 45

2.6. Conclusion........................................................................................................................................47 Chapter 3 Institutional Settings and Data.............................................................................................49 3.1. UK healthcare setting...................................................................................................................49 3.1.1. Hospitals and secondary care in the UK: Organisation..............................................................................50 3.1.2. Hospitals and secondary care in England: Funding of hospital trusts and primary care trusts.......51 3.1.3. Pay setting ............................................................................................................................................................ 51 3.1.4. Assistant nurses and health care assistants, a clarification.......................................................................53 3.1.5. Agenda for Change.............................................................................................................................................. 54

3.2. French health-care setting...........................................................................................................55 3.2.1. Organisation of care............................................................................................................................................ 55 3.2.2. Hospitals................................................................................................................................................................. 56 3.2.3. Status...................................................................................................................................................................... 56 3.2.4. Participation to the public service................................................................................................................... 57 3.2.5. Funding................................................................................................................................................................... 57 3.2.6. Pay and employment........................................................................................................................................... 59 3.2.7. Description of collective agreements and public sector pay grid.............................................................61 3.2.8. Geography............................................................................................................................................................. 65 3.2.9. Social Occupational Coding.............................................................................................................................. 65

3.3. Comparing the two institutional settings, why studying France is of interest.............67 3.4. English hospitals data...................................................................................................................68 3.4.1. Dependent variables for English analyses...................................................................................................... 69 3.4.2. Independent variables for English analyses, presentations and expected effects................................72

3.5. French hospital data......................................................................................................................76 3.5.1. Administrative and juridical characteristics of hospitals...........................................................................78 3.5.2. Descriptives of the workforce variables across status and year ..............................................................79 3.5.3. Detailed descriptives for public hospitals...................................................................................................... 83 3.5.4. Activity variables, construction........................................................................................................................ 84 3.5.5. Activity variables, descriptive statistics......................................................................................................... 90

3.6. Conclusion........................................................................................................................................95

Chapter 4 Creation of Standardised Spatial Wage Differentials.....................................................98 4.1. Creation of Standardised Spatial Wage Differentials for England...................................98 4.1.1. Empirical model to estimate SSWDs............................................................................................................... 99 4.1.2. Cost of living supplements ............................................................................................................................. 102 4.1.3. Using SSWDs...................................................................................................................................................... 102

4.2. Strategies to deal with “missing” SSWDs parameters in England..................................105 4.2.1. Using information for surrounding LADs.................................................................................................... 105 4.2.2. Using more years................................................................................................................................................ 106 4.2.3. Using information of surrounding LADs and more years........................................................................107 4.2.4. Testing the method used to recode missing SSWDs values....................................................................108

4.3. Definition of pay gaps and their descriptive statistics for England...............................109 4.4. Creation of Standardised Spatial Wage Differentials for France....................................112 4.4.1. Labour market data........................................................................................................................................... 112 4.4.2. Labour Market models...................................................................................................................................... 114 4.4.3. Previous method, fully stratified method..................................................................................................... 115 4.4.4. New method, fully interacted method ......................................................................................................... 116 4.4.5. Results of the estimations................................................................................................................................ 117 4.4.6. SSWDs descriptive statistics........................................................................................................................... 118 4.4.7. Gaps descriptive statistics............................................................................................................................... 119

4.5. Conclusion.....................................................................................................................................122 4.5.1. Comparing data for the two countries, limitations of the data..............................................................122 4.5.2. Summary.............................................................................................................................................................. 124

Chapter 5 Empirical Strategies...............................................................................................................126 5.1. Ordinary Least Squares..............................................................................................................126 5.2. Endogeneity: a small review of the causes............................................................................128 5.3. Endogeneity a review of the independent variables...........................................................129 5.3.1. Endogeneity of the competitiveness of pay ................................................................................................ 129 5.3.2. Endogeneity of other independent variables............................................................................................... 130

5.4. Why not panel estimations?......................................................................................................131 5.5. Shortage of staff...........................................................................................................................132 5.5.1. Measures of shortage of staff......................................................................................................................... 132 5.5.2. Econometric specification................................................................................................................................ 135 5.5.3. Vacancy rates models, only for England....................................................................................................... 137 5.5.4. Staff levels models for England and France................................................................................................. 140 5.5.5. Staff levels models with more than one slope for the effect of gaps.....................................................142 5.5.6. Complexity index estimations......................................................................................................................... 145

5.6. Staff mix models...........................................................................................................................145 5.6.1. Econometric specification................................................................................................................................ 146 5.6.2. Staff-mix models................................................................................................................................................. 146 5.6.3. Specific staff mix models for each country.................................................................................................. 148

5.7. Conclusion.....................................................................................................................................153 Chapter 6 Shortage of Staff in English Hospitals..............................................................................154 6.1. Results.............................................................................................................................................155 6.1.1. Assistant nurses vacancy rates, pay gaps, equations 5.7 and 5.8...........................................................155 6.1.2. Assistant nurses vacancy rates, introducing control variables, Equation 5.9......................................155 6.1.3. Assistant nurses vacancy rates, two slopes for the assistant nurses gap, Equation 5.10.................157 6.1.4. Registered nurses vacancy rates, pay gaps, equations 5.7 and 5.8.........................................................158 6.1.5. Registered nurse vacancy rates, introducing control variables, Equation 5.9.....................................158 6.1.6. Registered nurses vacancy rates, two slopes for registered nurses gap, Equation 5.10....................159 6.1.7. Vacancy rates models with the complexity index......................................................................................160 6.1.8. Assistant nurses staff levels models, pay gaps, equations 5.11 and 5.12..............................................162 6.1.9. Assistant nurses staff levels models, two slopes for the own gap, Equation 5.14..............................163 6.1.10. Registered nurses staff levels models, pay gaps, equations 5.11 and 5.12..........................................164 6.1.11. Registered nurses staff levels models, two slopes for the own gap, Equation 5.14..........................165 6.1.12. Staffing levels models with the complexity index....................................................................................166

6.2. Summary of the results for the pay gaps...............................................................................167 iv

6.3. Conclusions....................................................................................................................................168 6.3.1. Registered nurses pay gap plays the bigger role........................................................................................169 6.3.2. Distinguishing two slopes for gaps................................................................................................................ 170 6.3.3. Complexity index............................................................................................................................................... 170 6.3.4. Critics of the measures of shortage of staff................................................................................................. 170 6.3.5. From shortage of staff to skill mix................................................................................................................. 171

Chapter 7 Nursing Staff Mix in English Hospitals............................................................................174 7.1. Results.............................................................................................................................................174 7.1.1. Pay gaps, Equation 5.17.................................................................................................................................... 174 7.1.2. Pay gaps with covariates, equations 5.17 and 5.18....................................................................................175 7.1.3. Two parameters for each gap, equations 5.17, 5.18 and 5.19. .................................................................176 7.1.4. Introducing more than one slope for gaps, equations 5.18, 5.19 and 5.20. ..........................................177 7.1.5. Specific effects of gaps for foundation status, equations 5.18 and 5.21................................................179 7.1.6. Complexity index models................................................................................................................................. 180

7.2. Conclusions....................................................................................................................................182 7.2.1. Pay gaps, shortage and staff-mix................................................................................................................... 182 7.2.2. Complexity index models................................................................................................................................. 183

Chapter 8 Staff Levels in French Hospitals.........................................................................................185 8.1. Results.............................................................................................................................................186 8.1.1. Results for assistant nurses staff levels........................................................................................................ 186 8.1.2. Results for registered nurses staff levels...................................................................................................... 194

8.2. Conclusion.....................................................................................................................................202 Chapter 9 Staff Mix in French Hospitals.............................................................................................206 9.1. Results for the proportion of registered nurses....................................................................207 9.2. Discussion and conclusion.........................................................................................................213 Chapter 10 Conclusion.............................................................................................................................217 References..................................................................................................................................................224 Table of Contents of Annexes...............................................................................................................232 A Merging hospitals in England............................................................................................................233 B Changes in number of staff, England...............................................................................................234 C Flow diagram, England........................................................................................................................241 D Creating a variable for size of hospitals, England........................................................................242 E Cost of living supplements, England................................................................................................243 F Descriptives when restricting the sample to those hospitals trusts with observations for nursing SSWDs, England.........................................................................................................................245 G Descriptives when restricting the sample to those hospitals with observations for nursing SSWDs and complexity index, England...............................................................................................246 H Regressions with Complexity index, England...............................................................................247 I Flow diagram, France.............................................................................................................................255 J Description of the collective agreement for cancer treatment centres, France.......................256 v

Index of Tables..........................................................................................................................................257 Index of Diagrams....................................................................................................................................260 Index of Equations....................................................................................................................................261

vi

Acknowledgements I thank everyone I have met over the last four years. Life is all interdependent, even if I have signed above a declaration of originality saying that I am the author of this document, it is difficult to believe that one could come with a complete original idea without interactions with others. This does not mean that I do not have full responsibility for my actions. I am very grateful to my two supervisors Bob Elliott and Diane Skåtun. One of the key element for doing a PhD, a prospective student should think of, after the topic, is with who. This thesis would not have ended successfully without Bob and Diane. Thank you. My girlfriend and now partner, Laura, I do not know how she has handled this, I have tried to understand but I do not think any economist, unless also being a poet, which I am not, can anyhow understand love. I will try to behave more for the next PhD … in baking or bike mechanics! There is something in my education that tells me never ever give up. I owe this to my parents. It makes me a bit stubborn sometimes but it is a necessary quality in this process. Solange, Jacques, thank you to you both for this and all the rest. I am also very grateful to the ENSAI which hosted some of my stays in France and to every people that helped me in doing this thesis: Eric, Valentin, Mhenni, Matt, Ioannis and al.. I have now plenty of cabbage recipes that I tried on my my former flatmates Hannah and Frauke, and former neighbours Patrycja and Fabian, you are great people supporting my cooking and saying it is good even when it is disgusting, back then that was some of the only petty things that helped me stay afloat during some of the most difficult days. You are most than welcomed wherever life brings me to. For the five o'clock pint after work and the nice lunches in is the not so nice IMS, I would like to thank you all in HERU, to single out some of you would be unfair but for sure those of you who came regularly or quite regularly to lunches are very warmly thanked. HERU would not be HERU without its administrative staff: Shona and her team, Anne, Alison, Lesley, thank you. Knoydart, Lochnagar, Glen Affric, The Cairngorms, Ben Macdui and the people associated with these very nice places especially when it was sunny: Sophie, Laura, Lydiane, Thomas, Graham,

Bob, Frauke, Hannah, Lucie, Paolo, Paola, Sheetale, Yves, Antoine, Tristan, Flora, Laura, Xavier, Blandine … that was cool. By the way, to all Scottish people when Sophie is coming to Scotland, there is not a drop of rain! Yet some whisky is poured into glasses. Thank you to all who came to Scotland for a visit. My cousins (thank you Diane for hosting two of them) and Friends: Mathieu, Aude, Katia, Benoit, Sophie, Yolène, Romain, Flora, Tristan, Yves, Lydiane, Nelly. Jorge Cham and his PhD comic series. Maybe you thought I would forget you guys but no, this time in Aberdeen without you is unthinkable: A very warm thank you to Arthur, Robin, Gordon, Frida, Emily, Anna, Lavania, Megan, Megan, Hanna, Shared Planet, Pavol, Claire, Steven, Stefan, AUCU, Mike, Sue, Sarah, Caspar, Becycle, Claudia, Dominic, Ela, Eva, Mark, Mike, Natalia and all the others I forget with who I have shared some of the greatest experiences, especially when occupying university offices. Free Palestine, Free Education, Free housing and Free food.

viii

Pour mon grand-père,

Abbreviations Acronyms

Definitions

AfC

Agenda for Change

AHA

American Hospital Association

AMI

Acute Myocardial Infraction (heart attacks)

AN

Assistant Nurses

ARH

Agence Régionales d'Hospitalisation

ASHE

Annual Survey of Hours and Earnings

BMA

British Medical Association

CDD

Contrats à Durée Déterminée

CDI

Contrats à Durée Indéterminée

CI

Complexity Index

COLS

Cost-Of-Living Supplements

CV

Coefficient of Variation

DADS

Déclaration Annuelles de Données Sociales

DHOS

Direction de l'Hospitalisation et de l'Organisation des Soins

DRASS

Direction Régionale des Affaires Sanitaires et Sociales

DRG

Diagnosed Related Groups

FE

Finished Episodes

FEHAP

Fédération des Etablissements d'Hospitalisation et d'Aide à la Personne

FHP

Fédération des cliniques et Hôpitaux Privés

FT

Foundation Trusts

GA

Global Allowance

GHM

Groupes Homogène de Maladies

GLM

General Labour Market

GMM

General Method of Moments

HA

Health Authority

HES

Hospital Episodes Statistics

HRG

Health Related Groups

INSEE

Institut National de la Statistique et des Etudes Economiques

KSF

Knowledge and Skills Framework

LAD

Local Authority Districts

LST

Long Stay

MFF

Market Forces Factor

MRI

Magnetic Resonance Imaging

MSO

Medecine, Surgery and Obstetrics

NHS

National Health System

NQG

National Quantified Goals

OLS

Ordinary Least Squares

PCS

Professions et Catégories Socioprofessionnelles

PCG

Primary Care Groups

PCT

Primary Care Trusts

PNH

Private Non Hospital

PNP

Private Not for Profit hospitals

PP

Private for Profit hospitals

x

Acronyms

Definitions

PPS

Prospective Payment Systems

PSY

Psychiatric

ONDAM

Objectif National des Dépenses d'Assurances Maladie

RN

Registered Nurses

RRP

Recruitment and Retention Premia

SAE

Statistique Annuelle des Etablissments

SD

Standard Deviation

SOC

Standard Occupational Classification

SSWD

Standardised Spatial Wage Differentials

TTWA

Travel To Work Areas

UK

United Kingdom

USA

United States of America

VML

Virtual Microdata Library

xi

Executive Summary i.

Shortage of nurses in England and France is widely recognised. Earlier research provided some evidence that vacancy rates of nurses in the UK were higher in areas for which the relative standardised pay of nurses was lower. The spatial pattern of pay in the public and in the private sectors of the economy are expected to differ. The pay in the private sector is expected to be more flexible and to adapt more easily to local conditions. The pay in the public sector is expected to be less likely to adapt to local conditions due to national wage agreements.

ii. Earlier research that has dealt with similar issues gave evidence of the impact of the pay competitiveness of one group of staff on shortages of this particular group of staff. The literature review revealed that none of the empirical research on the topic has investigated the impact of the competitiveness of pay of an alternative nursing group (assistant nurses). Moreover, this thesis brings some evidence of the cross correlation of the labour markets of two nursing groups. It also shows that hospitals alter their staff-mix in order to dampen the consequences of the shortage of staff and the thesis shows that this differ according to the care performed by the hospital. Finally, it extends the research to another country, France, which was yet to be explored regarding this issue. In France there are more widespread use of collective bargaining throughout the economy, even in the private sector. France also has different types of hospitals with different wage regulations. iii. Standardised Spatial Wage Differentials (SSWDs) measure the average wage in each area of the country. They reveal the spatial pattern of pay for a specific sector, a group of occupation or the whole of the economy. This research investigates the impact of Standardised Spatial Wage Differentials (SSWDs) on local variations of shortage of nurses. It is hypothesised that ◦ Areas for which the gap between the SSWDs for nurses working in the public sector and the SSWDs for a comparator group will be correlated with difficult recruitment of staff.

◦ Further, hospitals are expected to have an adaptive behaviour when faced with uncompetitive pay and thus alter their relative demand for staff as a consequence. iv. This research uses Standardised Spatial Wage Differentials (SSWDs) to compute gaps between nurses in the public sector and their appropriate comparative group. The Annual Survey of Hours and Earnings (ASHE, 2003-05) was used to compute SSWDs in England and the Déclaration Annuelle de Données Sociales (DADS, 2006-08) was used for France. v. Gaps are computed as the difference of the set of SSWDs for nurses working in the public sector with their comparator group. Two nursing groups are investigated, assistant nurses and registered nurses. vi. For England, the comparator groups are the corresponding employees in the private sector in the same Standard Occupational Classification as the two nursing groups. For assistant nurses and registered nurses, the comparator group is made of the employees working in the private sector classified in the SOC group 6 and 3 respectively. vii. For France, there are three comparator groups for each nursing group. The first type of comparator group is made of employees working in the private for profit sector in occupations similar to the nursing ones. The pay for assistant nurses and registered nurses are compared to the pay for employees in the non hospital private for profit sector in the Professions et Catégories Socio Professionnelles (PCS) code number 4 and 2 respectively. The pay of nursing groups in France are also compared to the same nursing groups in the private not for profit hospital sector and in the private for profit hospital sector. viii.Hospital trusts data for the years 2003-2005 and public hospitals data for the years 20062008 were obtained from the Department of Health (England) and from the Ministry of Health (France). ix. For hospital trusts in England and for public hospitals in France, regressions are computed using the gaps as defined above. Vacancy rates (only for England) and a sizestandardised measure of staff numbers are used to analyse the impact of the gaps on the supply of nursing staff. Then the proportion of registered nurses (to assistant nurses) is used as the relative demand of hospitals for nursing staff. x. For England, the vacancy rate of assistant nurses and registered nurses is lower and the number of assistant nurses is higher in areas with more competitive pay for registered

xiv

nurses. The impact of the pay for registered nurses on the vacancy rates of registered nurses is stronger and only holds where the pay for the assistant nurses is also competitive. The relative demand for nursing staff differs according the competitiveness of the pay for the two nursing groups and also differs according to the care performed by hospital trusts. xi. For France, the supply of nurses is not associated with the competitiveness of the pay for nurses when compared with the pay for nurses in the private not for profit or private for profit hospitals. However, the supply of assistant nurses is higher when the pay for assistant nurses is more competitive compared to the pay for employees in the private non hospital sector. A more competitive pay for assistant nurses (compared to the pay for employees in the non hospitals private sector) is associated with a higher relative demand for registered nurses. xii. This thesis contributes to the literature in its area by first investigating a group of staff which had not been explored before: assistant nurses. This thesis gives some evidence of the importance of the cross correlations of labour markets of different nursing groups of staff. Then it shows that hospitals performing different types of care have different abilities in changing their staff-mix. Finally, this thesis investigate a country which had never been investigated regarding that topic before and the thesis gives some evidence that the pay gaps matter also in countries where pay settings are very regulated.

xv

Chapter 1 Introduction This thesis investigates the impact of geographical pay differences on shortage of nursing staff in public hospitals. It also investigates whether public hospitals change staff-mix to dampen the impact of the staff shortage. The analyses will focus on two countries, England and France, and it will investigate the shortage of two nursing groups, assistant nurses and registered nurses. Wage regulation has been a debate in economics for nearly one century and the seminal work of Arthur Cecil Pigou. On one hand wage regulation is supposed to impede markets to clear and thus create distortion in labour allocation. On the other hand wage regulation is a consequence of a process for a fairer allocation of revenues between labour and capital. Regulation can also be the consequence of a second best choice to unregulated markets. Unregulated markets may not be perfectly competitive and agents as informed as required for unregulated markets to clear. Therefore, regulations may improve allocations and meet societies objectives for distributions that are perceived to be fair. This thesis explores a situation in which one wage regulated market operates within/alongside an unregulated environment and a situation in which two types of regulations operate. As a result within a same region different wage rates are applied to similar jobs. The specific question this thesis will try to give some evidence on is what is the impact of such differences on regional nursing workforce in hospitals? Local labour markets have been studied in England and France and show variations in pay structure when standardised on occupation and industry. Prior to the recent recession and deterioration in public funding which has caused many countries to reduce funding for public service, shortage of staff in hospitals was a key problem for most OECD countries. This research is driven by prior evidence of an impact of the competitiveness of pay on shortage of registered nurses in the UK. This result assumed that the competitiveness of pay for one group of staff would only impact on the shortage of that group of staff. This thesis expands this previous work to consider the interactions of occupational groups. This thesis contributes to research by

1

Chapter 1

Introduction

introducing the interdependence of two nursing groups: assistant nurses and registered nurses in two health care settings: England and France. Chapter 2 presents some results from the literature which show that there is much more variation of pay for workers in the unregulated sector than for workers in the sector for which employees are covered by collective agreements. The sector for which the market sets the pay rate is expected to have a pay rate that compensate for cost-of-living and amenity levels. Pay in public hospitals for both England and France is set at national levels. Competitiveness of pay for public hospitals is computed by differencing the pay for the regulated sector by the pay for the unregulated sector. Under the assumption of compensating wage differentials, where hospitals have uncompetitive pay, hospitals should have difficulties in attracting and retaining staff. This assumption is made for England. With France, this thesis is testing whether wage gaps also matter in settings where there is no unregulated sector. This thesis introduces the risk of reverse causality of shortage of staff on the competitiveness of pay. Workers looking for jobs in the public hospital sector may change their mind where the shortage of staff in the hospital public sector is low and start looking for jobs in the comparator sector. By doing this, workers would depress the wage rate in the comparator sector and thus increase the competitiveness of pay for the hospital public sector. This is not expected to be the case for France as the wage rate is supposed to be fixed in all sectors of the economy whatever is the supply of workers due to the widespread use of collective agreements. This thesis contributes to the research by providing analyses of a country (France) in which most of the working population is covered by a collective bargaining. Therefore the assumption spelt out above cannot be used for such a country. On the other hand, for France there is no risk of endogeneity of competitiveness of pay on the shortage of staff. This thesis will show that even in settings where there is no unregulated sector, pay gaps matter for public hospitals. The institutional setting and the data for both countries are presented and compared in Chapter 3. Chapter 4 explains the creation of the competitiveness of pay for nursing staff and their comparators in England and France and then assess the differences between the two countries. Before the analyses, Chapter 5 presents the empirical strategy for all empirical chapters. Chapter 6 is the first empirical chapter and will explain the impact of the pay competitiveness on the shortage of staff in England. In previous research shortage of staff has been measured only with vacancy rates. It is likely, as any measure, that it does not perfectly measure the underlying concept of shortage of staff. For England, this thesis will use vacancy rates and use another measure of shortage: staff levels. 2

Chapter 1

Introduction

This thesis expands the research to a different health care setting for two reasons. First, as already mentioned, France has a wide coverage of collective agreements, in consequence there are no unregulated sectors. Even if the comparator group for France can not be assumed to be paid at a market rate due to the high coverage of collective bargaining for France the same strategy is followed. Therefore, this thesis tests whether a regulated pay for an alternative sector to public hospitals matters to nursing staff working in public hospitals. Investigating France allows to test whether there is still an impact of pay gaps on shortages of staff. Second, in Section 3.2 is explained that France has a health care setting in which there are more than one sector of hospitals. One of the main assumption in this thesis is that the pay for nursing staff can be mapped to the pay of another group of staff in the private sector. This is argued because the other group of staff is expected to have similar preferences and therefore it provides the exact rate of pay that these workers accept to work in different areas. For France, there is also a direct comparator group: nurses working in the hospital private sector. Because of being direct potential alternative employments in this sector, the assumption that workers in the comparator group have similar preferences is not necessary. Section 4.4 explains the mapping of the nursing pay to the different comparator groups. Chapter 8 will investigate the impact of the pay gaps on the shortage of public hospitals for France. Hospitals may have to find alternative solutions to shortage of staff in order to provide health care services. One of this solution may be to use the substitutional possibilities among their nursing staff. Where hospitals cannot recruit registered nurses they might hire assistant nurses instead; where this happen, the skill mix of their workforce will be altered. Altering the skill mix is only possible if there is a certain number of skills that are shared by the two nursing staff and if the care performed by hospitals use this shared set of skills. It is assumed that this set of skills exist as the two nursing group of staff work together and the lower group of staff can, after undertaking some training, achieve the higher grade of nursing, and this is possible in both England and France. However, not all hospitals may use the shared set of skills and the impact of the pay competitiveness may differ accordingly. This will be tested for England. The skill mix will be measured by creating a proportion of registered nurses in the total number of nursing staff for each hospital. The impact of the competitiveness of pay for two nursing groups on the skill mix of hospitals will be investigated in Chapter 7 for England and in Chapter 9 for France. This thesis is concluded in Chapter 10. It will conclude that this thesis contributes to the research on the topic by including the cross effects of the labour market for one nursing group of staff on an other nursing group of staff. The thesis also brings evidence of the role played by hospitals, as 3

Chapter 1

Introduction

they alter their staff mix where the pay of nurses is not competitive. Hospitals alter their staff mix depending on the care they perform, some alter their staff mix more easily than others because they may not all use the same set of skills that nursing groups provide. Finally, this thesis also contributes to research by investigating a country for which the assumption that the private sector labour market clears is not possible. It shows that in such a setting, the competitiveness of pay constructed as the difference between the public sector pay and the the pay in another sector still matters. The conclusion will try to give a policy perspective to the work done in this thesis, then a political perspective on the potential impact will also be discussed before thoughts about future works are given.

4

Chapter 2 Literature Nurses comprise the largest single staff group in the health service workforce in all developed countries. The period studied here is a time of shortages for nurses (Simoens et al., 2005). In March 2001 the English vacancy rate was 3.4% (Review Body for Nursing Staff, Midwives, Health Visitors and Professions Allied to Medicine, 2002). The Department of Health 1 planned to increase the whole time equivalent number of nurses by 26,300 from 1999 to 2005 which represents an increase of 8.7% of the total whole time equivalent number of qualified nurses and an annual increase of 1.4%. The increase in whole time equivalent number of qualified nurses between 1999 and 2000 was 2.1%. Local variations in shortages of nurses are linked to the attractiveness of regions. In a first approach it is possible to separate the attractiveness of regions between what is in the hospital scope of change and what is not. What is not part of hospitals actions may be the climate, environment, housing costs, cost-of-living. The attractiveness that depends on hospitals themselves are their ability to provide an exciting working environment and good working conditions. Wages are decided by different outside work elements such as collective agreements, law on minimum wages, and are observed only if job seekers take on the job. Moreover, it might not be the nominal wages that staff are concerned about. It might be the wage relative to the cost-of-living and amenities. Economists have developed a theory: compensating wage differentials, how much wages can compensate for cost-of-living or a low level of amenities in an area?

2.1. Local Labour Markets Local variation of staff shortages depends on the relative attractiveness of areas. Assume, in the first instance, that the demand for staff does not vary due to changes in local attractiveness of regions. One region may be more attractive if it has more amenities and a lower cost-of-living. Following the neoclassical theory and the assumption of a clearing market, local differences 1

Responsible for providing strategic leadership for public health, the NHS and social care in England → http://www.dh.gov.uk/health/about-us/

5

Chapter 2

Literature

should be compensated by wages which will remove local shortages. Subsection 2.1.1 introduces the theory of compensating wage differentials and Subsection 2.1.2 presents the corresponding empirical results published in research articles. Shortages of staff may occur if pay is set below the market rate, this is expected in sectors of the economy where pay is covered by collective agreements. Subsection 2.1.3 gives empirical evidence that wages differ in sectors of the economy where pay is set by collective agreements.

2.1.1. Compensating wage differentials, a theoretical perspective This subsection does not intend to give the full literature coverage of this theory. Its aims are just to highlight the theory as it is key to the construction of the empirical work performed in this thesis. The theory of compensating wage differentials provides a theoretical framework to explain why the ‘underlying’ structure of pay differs between geographical areas (A Smith, 1776; Rosen, 1986). Competition in labour markets ensures that the net advantages of different jobs will tend to equality. Thus, higher pay in some areas of the country is expected where the cost-of-living is higher while higher pay is also necessary to compensate for a less pleasant working environment. The rate of pay in the private sector represents (according to the hypothesis) the exact rate necessary to attract and retain staff. Thus all else equal a higher rate of pay in one area means that this area is less attractive (either has low amenity levels or higher cost-of-living). The pay offered in this area is set to counter the relative unattractiveness of the region. Some empirical studies have tried to test this assumption. Most of this research is interested in inter geographical wage disparities. The research ask the question: how can geographical wage differentials be explained?

2.1.2. The theory of compensating geographical wage differentials, empirical results and discussion This subsection2 aims to give a broad overview of the empirical research that test the existence of geographical wage differentials and the explanations for these. This subsection will therefore give empirical grounds to the underlying theory this thesis is built on: wages vary across areas.

2

This research was performed by using different search engines and the key words: geographical wage differentials, labour economics, compensating wage differentials and combinations of such keywords. Some articles were drawn to the attention of the candidate in conferences and by reading other articles.

6

Chapter 2

Literature

Reilly (1992) decomposes (with the well known Oaxaca (1973) technique) wage differentials between 6 local labour markets3 in the UK into two parts, one which is the consequence of individual characteristics in those 6 labour markets and the other one which is due to unexplained differences. The author finds that the differences in wages between labour markets is at around 20%, and that between Aberdeen and Rochdale, 50% of this difference is explained by workers' characteristics, the other part is not explained. The author concludes that those remaining differences are due to distinct wage determining processes thus suggesting that wages are not only a consequence of workers characteristics. García and Molina (2002) decompose observed mean wage differentials in Spain between different areas. They use a cross sectional data for 1994 (European Community Household panel) combined with regional price and population data from the National Statistics Institute (INE). They have 4450 individual observations that were divided into 5 main regions (North, South, East, Centre and Madrid). They introduce controls into their wage equations for occupation, education, industry, gender, age, tenure and whether they have a second language. They run regressions for each of the 5 regions and then use the parameters of the regressions and mean characteristics to decompose the difference in average wage between each region and Madrid. They follow the Oaxaca (1973) decomposition technique. The Oaxaca technique allows the researcher to decompose the difference in average wages between two regions into a characteristics component and in a remuneration of those characteristics component. Half of the differences in the mean wage are not explained, therefore they are not the consequences of differences in characteristics included in the regressions. Some of those unexplained differences in regional wage differentials should be the consequence of amenity and cost-of-living as no variables for these were introduced. Pereira and Galego (2011) analyse wage differentials in Portugal using dynamics. They decompose differences in average wage using the Oaxaca (1973) technique and another technique that is more suited to dynamics (JMP technique Juhn, Murphy and Pierce (1991)). They use data from the Quadros de Pessoal for 1995 and 2002. This data is compulsory for all employers with at least one employee. It is completed and sent to the ministry of employment of Portugal. It does not concern employees of public administration, the armed forces and self employed professionals. They considered only workers between 16 and 65 and excluded fisheries,

3

Aberdeen, Coventry, Kircaldy, Northampton, Rochdale and Swindon are the six local labour markets used.

7

Chapter 2

Literature

agriculture and unpaid family workers and apprentices. They also excluded workers in Madeira and the Açores. They estimated wage equations for each region (North, Centre, Alentejo, Algarve and Lisbon). The wages were deflated by the regional price index (available from the Instituto Nacional de Estatica) to take into account cost-of-living (the cost of housing is not included in the regional price index). They controlled their regressions with occupational dummies, industrial dummies, dummies for education, tenure, workers experience, and firm size 4. They separate regressions for men and women. They find that a large part of the cross sectional differences for 1995 in the mean wage found between Lisbon and other regions in Portugal are the consequences of observed characteristics of workers (higher educated workers), larger firms (larger firms pay better wages), and higher paid occupations. The dynamic of the wage differentials shows that the changes in the wage differentials over time were very small and that Lisbon increased the number of large firms and the proportion of higher educated people, while the regional gap due to unobserved characteristics decreased over time. Though they acknowledge they cannot distinguish whether the observed changes in wage dispersion across areas which are caused by the regional gap in unobserved characteristics result from an increase in the disequilibrium between regions or a change of unobserved productivity due to individual characteristics. Shah and Walker (1983) estimate a wage equation for male white workers in the UK using the general household survey of 1973. They adjust their regressions on individual characteristics, industry mix and cost-of-living. They introduced regional dummies for 15 regions 5. The cost-of-living proxy is taken from Reward Regional Surveys Ltd (1981) which publish reports on cost-of-living and regional comparisons from 1974 up to at least 1996 (Johnston et al., 1996). The description of the construction of those cost-of-living is made in Johnston et al. (1996) 6. Wage differentials change and sometimes are reversed when they introduce the cost-of-living, Scotland and the South East of England are worse off when introducing cost-of-living compared

4

5 6

They do not control for selectivity bias as they do not have enough information for this. They argue that sample selection will not be too much a problem as they do not compare men with women but men (women) in some regions with men (women) in other regions. For the authors, excluding agriculture and fisheries from the analysis removes the influence of different participation decisions in the labour market. They also argue that the selection bias will not be too big even when considering dynamics as regional mobility is very low in Portugal. Greater London, South East, West Midlands, North West, Scotland (West Coast, East Coast, North and South), Wales (South East and not South East), South West, East Midlands, East Anglia, York/Humberside, North East. The descriptions in Shad and Walker (1983) and in Johnston et al. (1996) are not very clear. Though that is how it can be summarised: the Reward Group assumes that households with two adults and two children of age 10 and 13 are representatives of the “communities”. From those households they gather 200 retail prices in 100 locations to which are added 18 regional prices for fuel and car insurances and 31 national prices covering newspapers, stamps, licences etc. The prices are weighted according to eight patterns of expenditure of the households.

8

Chapter 2

Literature

to the Midlands regions. They conclude that the differences in monetary reward are not the same as the differences in real-term awards. Differences still remain. Though there is some evidence to suggest that the differences in wages between regions compensate at least partly for differences in cost-of-living. Blackaby and Murphy (1991) estimate a wage equation with a complete set of industry and regional dummies and interactions between the two on the General Household Survey of 1982 7,8. Interactions were not significant and thus not used. They report that wages after controlling for this industry mix still have some variations between regions. Area specific fixed effects are then regressed on another set of variables. Geographical differences in wages are supposed, according to the compensating wages theory, to compensate for labour markets and area characteristics. Some of those area characteristics relate to the weather or environment. The authors include weather 9 and environmental10 variables. Though none of those were significant. They matched their data with the New Earnings Survey data of the same year in order to get some extra control variables. They introduced additional control for size of the plant and effort based on workers paid by results. They also included data on union wages11. The authors include other variables based on two other theories: efficiency wage and search theory. The following two paragraphs are brief overviews of these two theories. They are not very important for the understanding of the thesis and the informed reader surely will know more on these theories than what is presented. An overview of these theories is given as they are important for the paper of Blackaby and Murphy (1991). a) A brief overview of the efficiency wage theory The efficiency wage theory has two variants. At least two costs are associated with an employee leaving the firm, first the cost of recruiting a new employee and then the cost of training her/him. The first variant of the efficiency theory then says that the employer will set the rate of pay above the market rate in order to minimise the risk of the employees leaving. Secondly, employees will be less likely to shirk if their wage is large enough that the marginal utility taken 7

They restricted their sample to individuals who had worked more than 27 hours and had reported weekly earnings which resulted in a sample of 6999 individuals. 8 They also controlled for individual characteristics (age, age square, gender, experience and education). 9 They used year averages of rainfall, sunshine, gale days … Taken from the monthly weather report of the meteorological offices. 10 They used atmospheric pollution, road congestion and the overall provision of recreational facilities. 11 In the UK in 1982, wages, if set by collective bargaining, have national collective bargain as well as local ones. Thus they used proportions of people covered by either national bargaining, national and local ones for each regions of their analysis.

9

Chapter 2

Literature

by shirking does not compensate the loss they would incur if they get fired. By awarding higher wages the employers minimise the risk of the employee shirking. Therefore wages would be set at a higher rate than what would the competitive theory predicts. b) A brief overview of the search theory The search theory states that the information available both to workers and employers is scarce. Each marginal piece of information will come at a cost in time and resources, employees (and employers) will have non revealed reservation wage for which a marginal piece of information will not be compensated by the expected increased wage (decreased cost) that they may get. Thus workers will accept a job only if the the wage rate meets or is higher than their reservation wage. Employers will hire only if the wage rate at which they can hire meets or is lower than their reservation wage. Therefore they regressed on the area specific fixed effects the following area variables: price 12, unemployment rate, population density, average age, average experience, proportion of workers paid by payment by results, proportion of the different types of collective agreements and average plant size. The parameter for price equals 1 and is interpreted as workers not suffering from money illusion and asking for compensation for any higher cost-of-living. For the authors, population density is associated with higher dis amenities (more pollution) and is found as expected to be positively associated with regional wage fixed effects. The unemployment rate is negatively associated with regional fixed effects and the authors argue that it might be the consequence of an efficiency or search wage theory type of effect. Though the authors also argue that the unemployment rate coefficient can be interpreted as a competitive theory of wage determination if the unemployment rate measures differences in excess demand across the areas. They also find that union variables (proportion of workers with a national coverage and proportion of workers with national and supplementary coverage) are fully significant. The results of this second specification provides according to the authors weak support for a noncompetitive model of wage setting. Though identifying what is non competitive and competitive wage setting is not very straightforward. In a further article, Blackaby and Murphy (1995), using a tobit model for the North and South of England13 show that wages when controlled for individual characteristics, occupations characteristics, cost-of-living and industry mix are better in the North than in the South for 12 The authors use the same regional price as in Shah and Walker (1983) published by the Reward Regional Surveys Ltd. 13 The South is defined as South-East, South-West, East Anglia, East Midlands.

10

Chapter 2

Literature

manual workers. Thus they conclude that it makes sense for unemployees in the North to wait and look for a job in the North as that is where they will get relatively better pay. This give some extra evidence that wage differentials compensate, at least partly, for geographical differences in cost-of-living. Combes et al. (2008) use French data on all employees in France for different years, it is the Déclaration Annuelle de Données Sociales, also used in this thesis. They use the panel version of this data for the years 1976, 1980, 1984, 1988, 1992 and 1996. They are thus able to take into account observed and unobserved workers characteristics. They show large raw wage differences across geographical areas. These include the ratio of the highest geographical average to the lowest one (1.62 to 1.88), the ratio of the ninth to the first decile (1.19 to 1.23), and the coefficient of variation (0.08 to 0.09). They estimate changes in wages with panel data, which allows for individual unobserved characteristics. They include industry, areas, log share of employment, log numbers of plants and share of workers in professional occupations (proxy for level of education). They first estimate a wage equation, as they have a panel they can estimate wages as a function of observed and unobserved characteristics of the workers (workers fixed effect), the area year in which workers are employed (area fixed effect), the industry, and local characteristics of the industry. After computing the regression the authors multiply the parameter of each variable with the corresponding employee value for this variable: this new vector of values is then labelled as the effect of the variable on wages. They calculate the standard deviation of the effect of the variable which is argued to be the explanatory power of the variable and compute correlations with the log of wages. The standard deviation of the effect of workers (observed and unobserved) characteristics on wages is 0.29 and it has a correlation of 0.8 with log of wages. The workers characteristics are the most predominant effects on wages. The second most important effect is the area fixed effects even after de trending it (standard deviation of 0.14 and correlation of 0.34). The authors use the area fixed effect slightly modified so that it can be interpreted as a wage 14 and compare this standardised wage with the raw mean wage for each year-area and conclude that differences between the two are the results of the individual characteristics. There is between 40 to 50% differences between the two indexes, thus individual characteristics explain up to 50% in local wage differentials. They also correlate the average worker fixed effects within each area with de-trended area fixed effects. They conclude that the correlation is high enough (0.29) to conclude that workers sort themselves out, productive workers work in the same areas where average productivity is high. 14 For the details, see the article footnote 14 page 731.

11

Chapter 2

Literature

They then use the area parameters estimated in the previous stage as the dependent variable in a second stage. They explain the parameters for area fixed effects with the density of local employment as suggested by Ciccone and Hall (1996), they use the log of land area to distinguish density from pure scale effects. They capture the diversity of the local composition of economic activity with a Herfindhal index 15 (Glaeser et al., 1992). They also use 4 variables to describe the area (sea shore, mountains, lakes and water and outstanding cultural or architectural heritage). They use the residual of the regression and compare them with the raw mean wages of each area. Two thirds of the differences in wages between regions can be accounted for by the variables and fixed effects that are introduced in the two stages of this analysis. Combes et al. (2008) findings suggest that the differences in wages across regions may not be only the consequence of compensations for amenities and cost-of-living but might be endogenous, wages in some areas are higher because workers who have better labour market characteristics agglomerate themselves in denser and more skilled regions. Explanations for such effects were already paved by Alfred Marshall in 1890 “the most enterprising, the most highly gifted, those with the highest physique and strongest character go [to the large towns] to find scope for their abilities” Marshall (1997) cited in Combes et al. (2008, pages 737-738 footnote 27). Another reason (Glaeser and Maré, 2001; Wheeler, 2006) is that workers may learn more in larger cities, thus more skilled workers are attracted by this. However, once one has controlled for workers and industry (observed and unobserved) characteristics the then standardised wage differentials should be the consequences of compensations for disparities in amenities and cost-of-living. Vermeulen & Van Ommeren (2009) looked at the potential compensating effect of an increase in unemployment and a decrease in wage on housing costs. They used a Dutch data set on housing cost (self reported housing value) for the year 2002. They seek to explain the self-reported value of houses by unemployment, wages and the observed quality of the house through a set of indicators (free standing, semi-detached, number of rooms, garden, period of construction and central heating). They regress hourly wage of full time working men on age dummies, educational attainment and 40 region dummies16. They then use the estimated parameters for the regions dummies as standardised wages of the region. They use this wage in an equation along with unemployment and housing characteristics to explain self-reported housing prices. They find that unemployment is negatively correlated with housing prices, they also find that the relationship is always negative even when introducing a wage variable which means that the 15 In that case a Herfindahl index is the ratio of the square of total employment for an activity k divided by the sum of the square of employment of each activity. 16 Each region, say the author, includes at least one city and some adjoining municipalities.

12

Chapter 2

Literature

effect on housing prices of unemployment is partially captured by wage differentials but remains after controlling for wages. They did a sensitivity analysis to control for the sorting of workers (the fact that productive workers tend to migrate where average productivity is already high, as argued in Combes et al. (2008)) with different regressions on sub samples clustered by the level of education. They do not find any evidence of sorting of workers. They find that unemployment is more important for lower skilled workers than in higher skilled ones. They conclude that it is consistent with the idea that wage bargaining and minimum wage policy are more relevant to this group. The strong conclusion of their study is that housing cost is compensated by wages. Thus they say, unemployment may be sustained in equilibrium. The literature is not clear cut about how to interpret area wage differences. Some authors (Combes et al., 2008) find that wage differentials between regions are linked to a self-selection (sorting effect) of workers with similar (observed and unobserved) characteristics. However, Vermeulen & Van Ommeren (2009) find, even when controlling for sorting effects, that housing costs are compensated by wages, thus supporting compensating differentials theory. Blackaby and Murphy (1991) tested different labour markets theories and they concluded that their study provides weak support for non-competitive models of wage determination. Though they state that their study does not give either strong or even a weak support for competitive models (of which is the compensating theory) of wage specification. Hendricks (1977, 1975) gave some evidence that wages in regulated industries would pay their workers lower than in non regulated industries. If industries have different market product regulations (such as a price regulation, a maximum profit or an entry regulation) and are set in different areas, then it would be of interest to study a market product effect on geographical wage differentials. No article has been found testing this. Compensating wage differentials is an area of continuing research. This work employs the assumption that wages, once standardised for workers characteristics and industry mix, partially compensates for cost-of-living and amenities. It follows that shortages of staff may occur if pay is set below the market rate, this may be an outcome in sectors of the economy where pay is covered by collective agreements. Subsection 2.1.3 provides empirical evidence to reveal that wages do differ between the public and the private sectors of the economy where pay is set by collective agreements.

13

Chapter 2

Literature

2.1.3. The role of pay setting This subsection17 highlights the articles that relate the role of unions/collective bargaining to pay variations. It gives to the thesis the underlying idea that workers in the same area paid by different wage settings (either a market or a collective bargaining) will be paid at different rates of pay. In consequence, the pay rates of two groups of employees will differ within a same area. As the previous subsection showed that pay rates vary across areas the combination of the two gives that differences between two average wages set by different processes will vary across areas. For the net advantages of jobs in different areas to be equalised according to the compensating wage theory, labour must be mobile, labour markets must be integrated and pay structures flexible. However, pay structures may reflect the preferences of those participating in the institutions that set pay. Where trade unions have an important role in pay setting, pay is likely to deviate from the rates that would otherwise be paid in the market. Trade unions are likely to be concerned about equity and fair pay, and often seek to negotiate a national rate for the job. Metcalf et al., (2001) gives statistics on the collective agreements in the UK. Thirty five percent of the employees across the UK in 1998 were covered by collective agreements 18,19, 25% of the private sector employees and of these only one out of ten were covered by a national collective agreement, the others are covered by firm or workplace agreements. This is a switch from the previous era where wage councils used to set pay in many sectors for up to 2.5 million workers in 1992. This was before the government abolished them (Machin and Manning, 1994). In 1980, the proportion of workers for which the pay was set by collective bargaining was 71% (Traxler and Brandl, 2011). Because pay is set by firm or workplace agreements in the private sector Metcalf et al. (2001) conclude that private sector pay dispersion is likely to be higher in the collective agreement sector now than it was before, however, it is still likely that pay dispersion is still lower in the organised sector than in the non organised one. The authors note that employers may have an interest in having collective agreements as it helps employers to avoid some competition on the labour side.

17 This literature was searched through various search engines with the key words labour markets, unions, wages, regulation and any combination of these keywords. Meurs & Edon (2007) was highlighted to the candidate by his supervisors. 18 The authors are making a distinction between collective bargaining and the way the wage is set up in the NHS. They call the wage set up process in the NHS quasi arbitration. Though, collective bargaining refers to any of those in succeeding discussions. 19 Traxler and Brandl (2011) say that in 2000 there were 35% of the workforce that was covered by a collective bargaining down from 71% in 1980 in the UK.

14

Standardised Spatial Wage Differentials

15

Areas

Public Sector

Private Sector

Diagram 2.1: Geographical wage dispersion across sectors

Chapter 2 Literature

Chapter 2

Literature

Metcalf et al. (2001) argue that union wage policies will narrow the wage gap between the non discriminated employees (white, educated, males) and the discriminated employees (black, women, not educated). They hypothesise that the return to human capital will be lower for unionised workers than for non unionised ones, thus a higher level of education will offer a lower return in the unionised workplace. They test this assumption using the Labour Force Survey in 1998 and the Workplace Employee Relations Survey (WERS) for the same year. This survey collects data on detailed information from 2191 workplaces with more than ten employees. This represent three quarters of all the employees in the UK. Their study reveals that the dispersion in pay measured by the coefficient of variation (standard deviation divided by the mean) is lower among union members than among non union ones. For example, the average female union members pay is £8.37 which is £2 higher than for non union members, the standard deviation is 0.487 compared to 0.672 in the non unionised group of workers. When they control wage dispersion for differences in age, qualifications, occupation and workplace size they find that the wage dispersion remains. They also find that the dispersion of pay for both union members and non union members has increased other the last 20 years (data used is of 1998) and is taken as confirmation, by the authors, that the move from national pay bargaining to firm and workplace one is one of the causes. This shows that in the presence of unions the wage distribution is truncated at the lower tail. Only 2% of the workplaces with recognised unions have workers earning below £3.5 an hour (the minimum wage was set up at £3.6, 8 months later) while 16% of the workplace which do not have any unions have workers earning less than £3.5 an hour. They run a regression on the proportion of workers earning below £3.5 an hour. This regression is ran for all workplaces, private or public sectors. They control by collective bargaining, recognition of unions and union membership. The regressions are run at the workplace levels. They find that where there is union recognition and union membership above 60% the incidence of low pay is reduced. They run another regression on pay structures and they show that females, blacks and not educated achieve higher pay when working in unionised place. They find that collective bargaining and unionisation narrow the wage structure for each pair of groups (female/male; white/blacks; healthy/non healthy; Non manual/manual). Hayter and Weinberg (2011) conduct a literature review of the link between pay dispersion and

16

Chapter 2

Literature

collective bargaining/union share. They provide some evidence that the dispersion of pay within the union sector is lower than would have occurred without the unions. The authors review the theory first. They say that there are two theories explaining the effect of unions, the distortionist view and the institutionalist one. The distortionists, such as Milton Friedman, say that unions increase the wage of union members over the market rate and they tend to increase pay dispersion among the workforce as union members are usually those who would have been better paid anyway. The insider/outsider theory also tends to be used to justify the fact that unions could exacerbate pay dispersion. By being insiders, employees have a better rate of pay than they would have if the employer unilaterally set pay. Moreover they might make it costly to fire people and thus protect their situation at others expense. The unified theory (which is an extension of the neoclassical and distortionist theory) says that unions keep the wages of low paid workers high at the expense of unemployment. In Europe pay bargaining should make the low paid workers better paid than in the US and this should be associated with higher unemployment. However, this is not found to be the case as the OECD report published in 2006 (Bassanini and Duval, 2006) reveals. The institutionalist view recognises that unions tend to have different goals at the same time. One of these is to increase pay for its members but they are also concerned about employment and inequality. Therefore, a higher share of unionised workers or of collective bargaining should reduce pay dispersion and unemployment as well as enhance pay for unionised employees. In their literature review Hayter and Weinberg (2011) say that there is an extended body of literature that confirms that unions tend to compress wage structure. The wage structure is compressed within the unionised sector but also wage dispersion is less between the union and non union sectors when collective bargaining is more widespread. Overall, unions decrease wage dispersion. The author also point out that those results stand the test of time and that more sophisticated econometric analysis confirm that overall unions reduce wage inequality (Card et al., 2003). The decrease in union coverage and also the rise in decentralised bargaining has been found to be associated with a rise in wage inequality. In Sweden for example, wage dispersion decreased during the centralised bargaining period (1960 to 1983) and then increased when employers withdraw from collective bargaining in 1983 (Hayter and Weinberg, 2011). If the power of trade unions differs between areas (see Blackaby and Murphy, 1991) this will also affect geographical patterns of pay. 17

Chapter 2

Literature

In France, no research on wage differentials between union/non union or collective bargaining/market settings is known. The union membership is very low in France (13% in the public service and 5% in the private sector in 2003, (Amossé, 2004) and paradoxically, there is a very high coverage of collective bargaining for up to 95% of all employees (Howell, 2009; Traxler and Brandl, 2011). Even with such low figures of union membership the proportion of employees declaring that there is a union member at their workplace is around 40% (Amossé 2004). One study gives some evidence that pay dispersion differ between the public and the private sectors (Meurs and Edon, 2007). Meurs & Edon (2007) have studied French labour markets using the Labour Force Survey for 2002. They analysed the premium paid by the public sector (excluding public owned companies) compared to the private sector. Public employees can receive a 3% premium on the grade they occupy for cost-of-living reasons, 3% is only available for those working in Paris. They distinguish 22 French regions and use quantile regressions to estimate wage equations, controlling for individual and occupational characteristics. The main finding is that Île de France (the region that includes Paris) and Alsace pay a similar standardised wage in both the public sector and in the private sector and are the regions in which pay is the highest. In the private sector, Île de France has a premium of 13.5% compared to the reference region PACA 20. In the public sector the premium compared to the reference region is 3.2% for Île de France. They also show that Brittany, Limousin, Auvergne, Aquitaine, Pays de la Loire and Provence Alpes Côtes d'Azur pay a premium of around 10% compared to the private sector in each of these regions. Low skilled public sector workers are relatively better paid than their private sector counterparts. The reverse is true for high skilled workers. The differences between the two sectors are still relatively small but the authors conclude that the wage variation in the private sector is larger and more influenced by economic factors. It appears that the large coverage of collective agreements in France does not result in a totally flat pay structure across the country. As reviewed above, previous research has revealed quite clear effects of unions/collective agreements on wage dispersion. It seems likely that the geographical pattern of pay in the unionised sector exhibits a much flatter distribution than in the non unionised sector. Following a neoclassical assumption, the sector with a market clearing mechanism to set pay would have a pay rate that compensate for geographical variations in amenities and cost-of-living; in the sector in which pay is covered by a collective agreement, employers in regions where pay is set at a lower rate than the market one will have more difficulties to attract staff, therefore employers 20 Provence Alpes Côte d'Azur region, South East of France.

18

Chapter 2

Literature

may face more acute shortage of staff. Under the compensating wage differential assumption, pay gaps should explain a large part of geographical variations in shortages. Section 2.2 will review the literature investigating the impact of pay gaps between the public and the private sectors on the delivery of public services. The compensating wage differential assumption, as used in this thesis, is a supply side theory. In the union/collective bargaining sector wages are set externally to employers behaviour. Demand for labour is supposed to be set following other features outside of the local labour market. However, in this thesis, employers are supposed to alter their demand for labour based on the pay competitiveness they can offer. This thesis investigates the impact of local labour markets conditions on the relative demand for staff: staff-mix. Section 2.4 will then review the literature on the impact of the change in staff-mix on service delivery and on the drivers of staff-mix independently of the local labour market conditions.

2.2. Impact of the different pay setting institutions on public services The literature on union/collective bargaining provides evidence that the unionised sector exhibits less wage variation than the non unionised sector, at least for the UK. The compensating wage theory gives an underlying background for why there should be wage dispersion across regions. The public sector is expected to be more subject to a greater incidence of the union/collective bargaining than the private sector. Therefore the public sector should exhibit less geographical pay dispersion than the private sector. Even though union membership is low in France and collective coverage very high in both sectors of the economy, union is stronger in the public sector than in the private sector: 5% in the private sector and 13% in the public service in 2003 (Amossé, 2004). The pay gap between the two sectors should have consequences for the more regulated sector (public sector) as the pay offered in this sector cannot reflect local amenity and cost-of-living levels. This may make it difficult for the public sector to attract and retain staff. As a result, additional, non monetary costs may occur. Shortage of staff/difficulties to recruit can be one of those non monetary costs. The gap between the public and the private sectors pay should give a monetary value to this extra cost. At the other end of the pay gap distribution, where it is more competitive for the public sector, a null indirect cost would be expected. Hospitals in these areas would not have to overcome an extra burden, it is not expected that they would get an extra 19

Chapter 2

Literature

advantage either. The effect of the pay gap is therefore expected to be non linear. Hospitals are expected to suffer an extra cost where pay is non competitive and nothing where pay is competitive21. Visually, this gap in pay should be expected to be like in diagram 2.1 (page 15). In continuous line the geographical wage distribution for the private sector shows large variations between the regions. In dashed the geographical wage distribution for the public sector shows a flattened rate of pay across a fictional country. This section reviews22 in details the small number of articles found that bring together geographical differences in pay and their impact on the public service. This section is dealing with articles that are similar to this thesis. This is expected to be the whole of the literature dealing with the impact of wage gaps on public services. This review brings some evidence of the impact of the competitiveness of wages on public services. This thesis contributes to this research by investigating a nursing group of staff that had not been investigated before. The thesis will also highlight the cross correlation of labour markets of different nursing staff. Public services may adapt to uncompetitive pay by modifying their skill mix and this may differ according to the type of services they offer: specifically hospitals that perform different care have different abilities to alter their skill mix. Finally, this thesis also perform analyses for a country which was yet to be investigated: France. With the investigation of this country, this thesis shows that pay gaps also matter in settings where nearly the totality of the workforce is covered by a collective bargaining. Elliott et al. (2006, 2007, 2010) in three different pieces of research presented the impact of pay gaps on the vacancy rates of qualified nurses in the NHS in England 23. In the NHS funding for hospitals is allocated according to a formula which assesses population need for health services. This formula is then adjusted for local variations in unavoidable costs using the Market Forces Factor (MFF). A formula to determine the MFF has been used for more than 30 years by the NHS. The underpinning idea originates in the the Resource Allocation Working Party report in 1976. At that time the MFF was introduced to help the NHS compete in hiring ancillary, clerical and administrative staff in local labour markets. In 1996, the General Labour Market (GLM) method which uses wage regressions to estimate MFF was introduced. The GLM method consists of running wage regressions on private sector employees pay where the regression includes

21 At at the other end of the pay gap distribution, the private sector pay may not be competitive compared to the public sector and the impact of pay gaps on the private sector can also be investigated. 22 The following articles were highlighted through discussions with the candidate's supervisors. 23 Elliott et al. (2006, 2010) estimates wage gaps for the whole of Great Britain (England, Scotland and Wales) but present results only for England as Market Forces Factors are only used in England. Elliott et al. (2007) analyse data and report results for Great Britain.

20

Chapter 2

Literature

geographical dummies. The estimation controls for age, age-square, gender, industry dummies, occupational dummies and the coefficient on the dummies give Standardised Spatial Wage Differentials (SSWDs). These are argued to measure the unavoidable local costs and are therefore used to adjust funding to local NHS providers. Elliott et al. (2006, 2010) compute NHS SSWDs and test whether the gaps between nursing and medical SSWDs and the private sector are connected to local shortages of nurses and medical doctors. Data for the GLM and nurses comes from the Annual Survey of Hours and Earnings (ASHE), for doctors the authors used pay bill data provided by the Department of Health. ASHE data is provided by employers and is therefore very reliable. They use data for the financial years 2002-03 to 2004-05. They use turnover and vacancy rates as measures of shortages for all hospital trusts. The data is from the NHS vacancy survey which interrogates all English hospital trusts. The authors dismiss turnover as it is a measure that counts the leaving and joining rate, therefore a hospital trust in a region where hiring and retaining staff would be difficult will have a high leaving rate but a low joining rate. Therefore it may ineffectively proxy shortages of NHS staff. Vacancy rates are their preferred measure of shortages. They find that the gaps are associated with hospital vacancy rates for nurses and doctors. The direction of the effects are different. The greater the competitiveness of nurses (doctors) pay, the lower (higher) the hospital trusts average vacancy rates for nurses (doctors). Therefore, the gaps between the NHS SSWDs and the MFF are connected with hospital vacancy rates. Elliott et al. (2007) focused on qualified nurses and midwives using data for 1999-2002. They use the Quarterly Labour Force Survey (QLFS) 24. They compute a switching regression model to control for the choices to work in the NHS or in the private sector when computing SSWDs. They use data on family characteristics (household income, property owner, illness or disability for members of the family, number of children, provision of parental care, age, ethnicity, education of family members) to compute the switching equation (the choice to work for the NHS) as those variables are sometimes argued to be uncorrelated to pay but related to the decision to work. The authors in this paper focus on qualified nurses and midwives and on a comparator group defined as female employees working in an occupation in the private sector which contained at least one employee with a nursing education for England, Scotland and Wales. Vacancy data, for this article, is available at the Health Authority level. They find that selection significantly affects the

24 In the UK, two datasets provides data on wages. The trade off between the two is that the QLFS is rich in terms of workers characteristics but wages are self declared while the ASHE data provides very reliable wage data and scarce individual characteristics. In Elliott et al. (2007), the authors need more individual characteristics than what ASHE would be able to provide.

21

Chapter 2

Literature

estimation of SSWDs. They also find, as in Elliott et al. (2006, 2009) that when the competitiveness of the nurses pay is higher, the vacancy rates of health authorities are lower. The empirical research conducted by Elliott et al. (2006, 2007, 2010) confirms an impact of local pay gaps on local variations in nursing shortages. However, it is not clear how the result for doctors found in Elliott et al. (2006, 2010) is to be interpreted. The authors conclude that doctors may be attracted by other (unobserved) characteristics. Propper and Van Reenen (2010) examine hospital quality using Acute Myocardial Infraction (AMI) rates (heart attacks) for English hospitals for patients aged 55 and over. They used the death rate within the 30 days of an admission following an AMI. The authors argue that hospital quality is affected by the external wage that nurses could have get had they not worked for the NHS. The authors uses three “outside” wages, one is reviewed here. The outside wage is calculated using the Labour Force Survey between 1996 and 2007. They run a regression on the ln wage of non manual female workers (excluding nurses and teachers). They control for regions and year and all individual characteristics available from the LFS (qualifications, ethnicity, country of origin, marital status, number of children...). By using the characteristics of nurses and the parameters estimated with the wage equation, they calculate an outside wage for each nurse. This is then averaged at the area-year level. They call it the regression corrected outside wage. The decomposition of this regression corrected wage show that the characteristics of nurses do not differ much from the characteristics of their female counterparts. Therefore they will use the results of the regression directly but show that their results are robust to another specification. The authors regress the private sector wage on the hospital performance using panel data. Panel data allows for hospital fixed effects along with measures of case mix. The case mix includes the degree of ill health of the population from which the hospital draws its cases 25 and the age gender distribution of the admissions for AMI 26. They also control for hospital size and the types of hospitals27. Case mix and AMI rates were obtained from the Hospital Episodes Statistics (HES). They run pooled OLS, between, within and GMM estimations. The authors use the panel estimation to treat the wage as endogenous. They use the within and between information as akin instrumental variables. They find a positive and significant relationship between the wage and

25 This is captured by all-cause time-varying mortality of the area around hospitals. 26 The authors also use the severity of AMI in robustness checks. 27 Acute or teaching hospital trusts.

22

Chapter 2

Literature

AMI giving evidence that hospitals with higher outside wages incur unavoidable costs that is reflected in higher AMI rates. The authors argue that the effect of the outside wage should not be linear and there might be a larger effect in higher outside wage regions than in lower outside wage regions. Thus they ran three different regressions on three regions. Each region is defined as having a different level of outside wage regions (low, medium and high) 28. The effect of the wage remains significant only in regions with a high outside wage. The main result is that an increase of 10% of the private sector wage imply an increase of 15% of AMI mortality rates in high wage regions, 8% in middle wage regions and 1% in low wage regions. In further research into the impact of local wage differences on public sector performance Ma et al. (2009) provide an analysis of the impact of the pay gaps between the public and the private sectors on school's teachers vacancies in England and Wales. This report intended to inform the deliberations of the School Teachers Review Body. This body is responsible to the government for the pay structure and pay awards for teachers working in Local Authorities maintained schools in England and Wales. The pay structure has pay bands with pay rates set at different levels for different areas. Pay differs according to whether the teacher works in London, Outer London, the Fringe area and the rest of England and Wales. These different pay bands compensate teachers for local differences in amenities and cost-of-living. The report was commissioned to review the appropriateness of the different bands. The report used a similar method to the one in Elliott et al. (2006). The authors calculated Standardised Spatial Wage Differentials (SSWDs) to compute pay gaps between the public (teachers) and the private sector. They used data for 2004 to 2007. Pay data was taken, as in Elliott (2006), from the Annual Survey of Hours and Earnings. They used data on school teachers vacancy rates provided by the two countries, England and Wales. The data differ in the two countries. For England, data was extracted from the School Workforce in England which is a publicly available dataset that report results from an annual survey carried out in all Local Authority maintained schools. Vacancy rates are reported for qualified and unqualified teachers together, though the authors note that “when schools advertise vacancies they advertise for qualified teachers” (page 34). The authors find a positive relationship between the SSWD gap and the teachers vacancy rate. The result is robust for secondary schools teachers. Though the marginal effect is very small, for an increase of half a standard deviation of the gap, the reduction of the vacancy rate is estimated at 0.09%, the effect is significant. To test for the robustness of this result they introduce the share of unqualified teachers. Unqualified teachers pay is not set by the 28 The high wage region is London and the South East, the low wage is the South West and North East.

23

Chapter 2

Literature

pay review body, it is set locally and therefore more likely to be set at a rate which compensate for local amenities and cost-of-living. They find that the share of unqualified teachers is negatively correlated with vacancies. Therefore, unqualified teachers might be used to substitute for qualified teachers and schools may therefore not report the post as vacant any more. For Wales the results are similar but the number of observations is much lower and therefore the results are not significant. This section highlighted the evidence from the literature on the impact of pay gaps on public services. It shows that pay gaps have an impact on shortage of staff (Elliott et al., 2006, 2009, 2007; Ma et al., 2009) and on public service delivery (Propper and Van Reenen, 2010). This study will extend the work done by Elliott et al. for English hospitals and will also investigate French public hospitals. It will be extended by investigating shortage of another group of staff. Hospitals faced with shortages of staff may try to find alternative solutions to dampen the effect of pay gaps. Altering the staff-mix is one response to shortage. In economic term this means that the behaviour of the demand side will also be investigated. In the research reported above pay gaps were thought as affecting only the supply of staff and the demand for staff was taken for granted. Existing research on skill mix in health care investigate the feasibility of changing staff-mix, substituting nurses by less qualified staff or doctors by nursing. This involves research on costs and outcomes. Section 2.4 will review this literature, are better outcomes and lower costs observed when changing the staff-mix in hospitals?

2.3. Literature review of research on shortage of staff This section gives a review29 of the research on shortage of staff. Most research has been focusing on individual reasons for quitting nursing jobs. This body of research has mainly found that job satisfaction is the main driver of nurses quitting jobs. This literature is partly reviewed here but, as it focuses on drivers of shortage at the individual level, it does not help much for building the empirical strategy on what type of factors to include in the analyses. A very small number of articles has been found investigating the characteristics of hospitals which are associated with a greater or a smaller shortage of staff. There is a wide literature dealing with the reasons for the turnover among nursing staff. Most of 29 This literature has been found using Scopus using keywords such as “magnet hospitals”, staff, shortage, retention, characteristics, hospitals, nurs* .

24

Chapter 2

Literature

this research is based on nursing surveys. The aim of this section is not to provide a systematic review of the topic but to provide this thesis with arguments for the empirical strategy. Carter and Tourangeau (2012) found, in a study of 16,000 British nurses surveyed in 2009 that nurses are more likely to stay in their current workplace where they felt psychologically engaged with their job, if they perceived the availability of development of their career paths, if they could have a good work-life balance and whether they encountered work pressures. The authors conclude by arguing that at the local level, the focus of employers should be to promote employee well being. The article by Tai et al. (1998) reviews the literature on nursing turnover published between 1976 and 1996. They only considered articles published in English. It aims to review both hospital and individual characteristics of turnover. The authors note that the definition of turnover varies among the different articles reviewed, some defining turnover as stayers vs leavers; vacancy posts vs filled ones or intention to quit vs intention to stay. Evidence for the role of size is mixed with some authors finding that there is a positive relation between the size of the health care facility and turnover and others finding a negative relation. Others did not find any effects of size on turnover. Tai et al. (1998) argue that this mixed evidence is likely to be linked to the inconsistency of the definition of both size and turnover in the articles reviewed. The article of Shields and Ward (2001) is about the NHS in the UK. They investigate the retention and recruitment policies that have been put in place in the NHS at the end of the 90s and beginning of the 2000s. Those policies focus on improving pay and working conditions. Shields and Ward (2001) note from their introductory literature review that job satisfaction depend on the level of income relative to expected levels of comparison groups. These expected levels vary according to education, age and occupation. The authors focus on discovering whether low pay or perceived low pay increase the importance of pay comparisons. Shields and Ward (2001) note that overall satisfaction at work depends on the establishment size, hours of work, union membership and occupations. Freeman (1978) finds that job satisfaction is quantitatively more important in quitting jobs than pay. Gordon and Denisi (1995) and Laband and Lentz (1998) find that job satisfaction is related to job quitting. The data used in the article by Shields and Ward (2001) is from a national survey of NHS staff

25

Chapter 2

Literature

conducted by the Policy Studies Institute for the NHS. The data was gathered from postal questionnaires in 1994 from 91 NHS employers. The sample was stratified. The final sample is of 14,400 NHS staff of which 9625 are nurses. This paper focuses on nurses observations. The survey asked nurses to rank overall satisfaction as well as 13 other job related questions with four items, very satisfied, satisfied, dissatisfied and very dissatisfied. They defined a variable to proxy the prospects of respondents. This variable is based on a question with 14 possible answers of what the nurse think her/his job will be in three years, three of these items represent an intention to stay in the NHS while the others represent an intention to quit. The authors define the intention to stay as a binary variable (14 individuals responded both and were put in the quitting category as they showed some uncertainties). They estimate satisfaction with a multinomial logit and include a wide range of controls based on a pseudo economic utility function. They used an alternative wage called relative wage which is supposed to control for what nurses would expect in terms of wages. They argue that only public sector wages should be included in the relative wage measure as that is what nurses are comparing themselves to. They built the relative wage by using the Quartely Labour Force Survey of Spring 1994 (so that it is matched with the survey for nurses used in this article). They used employees of the public sector between 21 and 60 years old and they obtained a sample of 1876 employees. They estimated a log weekly wage regression, controlling for age and age square, gender, ethnicity, marital status and highest qualification and part time status. They used the predicted parameters of the equation and used the characteristics of the nurses to estimate their relative wage. Their logit estimation of job satisfaction include as independent variables: the absolute wage, the number of hours worked, relative wage (as defined above), individual specific characteristics, job characteristics, employer characteristics, a vector of characteristics which are important in the NHS nursing profession and a vector of individual specific work values. The last two are included in an extended regression only. They report that the relative wage is negatively associated with job satisfaction. Where the wage in the public non NHS sector is higher the lower the job satisfaction. An interesting result is that the number of hours is not significantly and negatively associated to job satisfaction when a variable coding an inadequate shift pattern is introduced. This inadequate shift pattern is the number of hours worked compared to the preferred number of hours the person would work.

26

Chapter 2

Literature

The authors report that nurses reporting that they put more values on helping others are more likely to have higher job satisfaction, therefore those personal characteristics may offer positive compensating differentials to lower pay. Then the authors run a regression on the intention of quitting. The job satisfaction is used as an independent variable. The intention of quitting was found to be the strongest predictor of turn over in 1979 by Mercer for the UK (1979). Seventy nine percent of 1400 nurses working in the NHS in a longitudinal study reporting an intention of quitting did so. Even though the study is old, the authors say, it is the only one. Steel and Ovalle (1984) in a psychology literature review of workers intentions confirm a correlation of 0.5 between the intention of quitting and job satisfaction. Shields and Wards (2001) used a PCA on job satisfaction variables and included the first five components of the PCA in a regression model on the intention to quit. Along with job satisfaction they included controls for nursing seniority, nursing speciality, hours worked and job tenure. They also control for type and size of NHS employer, and 8 geographical dummies. As expected, they find that job satisfaction is negatively correlated with the probability of quitting the NHS. They calculate a saving cost, how much would be saved in turn over cost by increasing job satisfaction of NHS employees. Individuals reporting to be highly dissatisfied are 65% more likely to hold an intention to quit their jobs. The article by Stordeur et al. (2007) investigates the organisational configuration of hospitals succeeding in attracting and retaining staff. They report an article in their literature review that large urban hospitals are located in areas where there are more job opportunities, therefore, these hospitals may face higher turnover rates (Bloom et al., 1992). Stordeur et al. recruited 12 Belgian representative hospitals and surveyed more than 2000 Registered Nurses out of the nearly 3800 Registered Nurses employed by these hospitals. The authors analysed turnover and included organisational characteristics based on a questionnaire sent to the nursing director and human resources director. They included legal status of the institution, general characteristics such as level of care, professional model of care, activity and financial balance. The questionnaire also asked for the number of nurses, specialised nurses and assistant nurses, number of permanent contracts, age distribution, turnover, absenteeism, vacancies. For turnover, from September 2002 to October 2003, nurse executives at the 12 hospitals recorded the number of nurses that voluntarily left the institution. They only focused on the hospitals that were in the lower quartile of turnover (below 3.6%) and the hospitals in the

27

Chapter 2

Literature

higher quartile of turnover (above 11.8%) reducing the number of nurses surveyed to 1100 approximatively. This reduce the hospital sample size to 6 and therefore it is not surprising that hospital characteristics do not come as significantly associated to low or high turnover. The authors also explain that the fact that hospitals in Belgium are highly regulated may explain the fact that the characteristics of hospitals are equally distributed among the high and low turnover hospitals. The authors compare qualitatively the characteristics of hospitals whether they have a high or low turnover. Hospitals with low turnover were associated with nurses of a higher age, higher number of children, higher tenure and lower opportunities for jobs in the area. They find that hospitals with higher turnover were also associated with higher proportion of permanent contracts in their workforce. The article by Gifford et al. (2002) use a survey of nurses in obstetrics units in 7 different large urban hospitals (average number of beds is more than 400) in Colorado (USA). They sent a survey to nurses with response rates from 25% to 77%. They asked nurses questions about the values of the organisation and the quality of work life. The values of the organisation are measured with five questions. Each question presents respondents with four scenarios representing values each hospital may have. Respondents are then asked to divide 100 points among these four scenarios depending on how similar the nurses think they were in their organisation. The quality of work life is measured by four broad elements thought of describing the quality of work life: organisation commitment, empowerment, job involvement and intent to turnover. Each were measured by five questions. Each questions contained five points Likert scale that are then summed. Local differences of shortage of staff may depend on job satisfaction, as argued by Coward et al. organisational theory suggests that job satisfaction should vary by size of the facilities. In a study of more than 700 nurses providing direct patient care in the USA, Coward et al. (1992) showed that overall job satisfaction among nurses was higher in hospitals of smaller size (1 to 49 beds) and that to the exception of pay, nurses are more satisfied with the other four sub dimensions (professional status, task requirements, organisational policies and autonomy) of the concept of job satisfaction in small rural hospitals. In a study involving two hospitals, Benedict et al. (1989) showed that the teaching hospital was more likely to retain staff than the corporation of 5 community hospitals. Retention of staff is difficult to measure as often it is measured with turnover but turnover does not tell the employment pattern of the workforce: whether the workforce employed for a long time or a short time. The authors of this article focus on two hospitals as they argue that retention cannot be 28

Chapter 2

Literature

well assessed with common available statistics. They use a technique usually used in survival analysis called life table. The authors were able to measure the employment history of more than 3000 employees over ten years. The two hospitals are stable over time with a relatively stable number of beds for more than two decades. The two hospitals provide similar type of care: short stay general hospitals. The corporation hospital has more centralised decision making than the teaching hospital. The data tells whether an employee had terminated her/his job for every quarter of a year. At the end of the 15th to 18th month of employment, only 50% of the employees were still in the workforce. After 12.6 months, 50% of the employees in the corporation of community hospitals were still employed and the same proportion is attained after 18.6 months in the teaching hospital. Overall they find that there is a sharp turn over in the early months of employment followed by a tapering. The authors then try to explain these differences between hospitals by demographic, professional and employment characteristics. They find that low satisfaction is consistently the major reason for leaving the workforce of the hospital. The article by Holmås (2002) propose a duration analysis of nurses in the Norwegian hospital public sector. They have data on nurses for 5 years and they analyse the exit of the public sector of nurses. In Norway, except for 1% of the total number of beds, all beds are in the public hospital sector. The author combines data from administrative sources and survey data. The first administrative source is a register that gather information from hospitals themselves and they give information about wages, working time and occupation. The author had to exclude two large national hospitals and five small and highly specialised hospitals as they do not report administrative data to the same body. Municipalities and counties hospitals were included in the data but local municipalities were not. Moreover, about 40% of the hospitals in the final data did not provide detailed data about shift work which is, according to the author, an important reason for quitting and therefore the nurses working in these hospitals were not used in the analysis. The author compared the descriptive statistics of these remaining hospitals with the total sample of hospital and observed that they were quite similar, the author concludes that the sample is representative. The author also excluded male nurses as male and female workers are thought to behave differently on the labour market. The author treated nurses quitting their job for a job in a hospital for which the author does not have data as censored. The exit or transition is recorded as nurses leaving the labour market for more than a year (a maternity leave can be of a maximum of 52 weeks, therefore, the cutting off point of one year avoids nurses leaving for a maternity leave) and as nurses being recorded to start a new job outside the public health sector. The author then merged this data with data from Statistics Norway to obtain information about children or spouse

29

Chapter 2

Literature

characteristics. The number of shift work is proxied by the share of wages that is the consequence of shift work in the total wage. The sample goes from the first of January 1993 to the 31st of December 1997 and the year is divided into 4 quarters which leads to 20 time periods. They estimate the model with a gamma distribution and with a non parametric model which approximate the distribution by discrete mass points. They used two mass points as introducing a third one did not increase the likelihood. The author choose the gamma distribution based on AIC and BIC statistics. The result show that wage is negatively associated with the hazard of quitting the public health care sector. An increase of 1 Norwegian Krown leads to a decrease hazard rate of quitting of 3.4%. This effect is smaller for nurses holding a specialist qualification and for those who have managing responsibilities. Shift work is associated with higher probabilities of quitting. The author tested a specification without the covariable “share bonuses” (the proxy for shift work) and shows that the wage effect is smaller by 2 points for each Norwegian Krown, the author argues that this is some evidence of the bias of not taking into account shift work when estimating the decision to quit. Nurses with a larger number of years of experience have a lower hazard of quitting. Age works the other way around up to a turnaround of 41 years, the older they are the more likely nurses are to quit up to 41 years old and then the older they are the less likely they are to quit. Part timers are less likely to quit the public health care sector. The author reckons that some studies have given evidence that poor working conditions are likely to explain turnover rates because they affect job satisfaction of nurses. As the author do not have any measure of job satisfaction or working conditions, occupancy rate, beds per nurse and length of stay are explained to proxy working conditions. The results show that nurses working in hospitals with high occupancy rates or a relatively larger number of beds per nurse are more inclined to leave nursing than others. Long stay is associated with the hazard to quit in the opposite direction: the more patients stay, the less nurses quit. Nurses, even when controlling for type of hospitals, are more likely to quit when hospitals are larger. Nurses with younger kids are less likely to quit nursing than others, and the explanation given to this is that in Norway, women have more abilities to reduce their working time, therefore the supply of nursing might be lower but the nurses stay in the job. Married nurses have lower hazard rates of quitting than non married women. In order to summarise the results presented above in terms of future empirical strategies, it 30

Chapter 2

Literature

should be noted that two articles provides some evidence that smaller hospitals are more likely to retain staff (Holmås, 2002) or that nursing staff working in smaller hospitals have more job satisfaction (Coward et al., 1992) which in turn is said to be the best driver of nursing quitting their job (Shields and Ward, 2001; Coward et al., 1992; Benedict et al., 1989; Mercer, 1979; Steel and Ovalle, 1984; Freeman, 1978; Gordon and Denisi, 1995; Laband and Lentz, 1998; Carter and Tourangeau, 2012); though in a literature review, Tai et al. (1998) say that there is mixed evidence of the effect of size on staff turnover. Holmås (2002) provides some evidence that occupancy rates and relatively larger number of beds per nurses are associated with higher risk of quitting. Benedict et al. (1989) is the only article providing some evidence of the role of the type of care on retaining staff. This article gives some evidence that teaching hospitals are more likely to retain nursing staff.

2.4. Literature review of research on staff-mix This thesis investigates the impact of nursing staff pay competitiveness on shortage of nursing staff and skill mix. The assumption is that hospitals alter the staff-mix to mitigate the effects of shortages. The following literature30 review whether altering nursing staff-mix is possible and also the impact on outcomes of such changes. This literature sets in this thesis as a necessary pre condition for the empirical work on staff-mix to be undertaken. This literature brings evidence that changing the nursing staff mix is feasible. This literature also highlight the consequences of changing the staff-mix. As this thesis will show that hospital may alter their staff-mix in order to cope with shortages of staff, it is relevant to highlight whether this behaviour will have consequences on patients care. However, the literature reviewed in this section does not bring much evidence on whether changing the staff-mix alters the outcomes of hospitals (Subsection 2.4.1). Changing the skill-mix is going to be analysed as a consequence of hospitals behaviour when faced with shortage of staff, in order to take into account potential drivers of staff-mix within the analyses presented in this thesis, this section will also review 31 previous research into the drivers of staff-mix (Subsection 2.4.2). Finally, Subsection 2.4.3 will review articles32 that look specifically at the degree of substitutability between nursing groups of staff. This final subsection provides no 30 This literature has been found with Scopus as a search engine with the key words labour market, skill mix, workforce, nurses and any combination of these key words. This search was done three years ago but an alert was set to keep the candidate up to date with newly published articles. 31 This literature was searched with different search engine with the key words: drivers of skill-mix, nursing skill mix, secondary care supply of staff and combinations of these words. 32 This literature was searched using hospital productivity, skill mix, hospital, production functions, productivity with the search engines Scopus and ScienceDirect.

31

Chapter 2

Literature

evidence on whether the two nursing staff are substitutable or complementary. If one clear result had emerged from this literature, the results concerning staff-mix analyses would have been expected to be similar.

2.4.1. Staff-mix and the delivery of care Most of the research into staff-mix has focused on the impact of staff-mix on the delivery of care and patient outcomes (McGillis Hall, 1997; Cavanagh and Bamford, 1997; Spilsbury, 2001; J Buchan and MR Buchan, 2002; Aiken et al., 2003; T-Y Lee et al., 2005). Some research has also focused on the economic impact of changing staff-mix. It appears to have established that in a few settings substituting nurses for less qualified ones may improve patient satisfaction with no adverse impact on patients outcomes. The review by McGillis Hall (1997) mainly reports articles from the United States of America and those articles mainly reveal that there is a financial advantage of introducing more health care assistant and decreasing the number of qualified nurses. The author the article critics the fact that there are not many articles examining the effect over time of changes in staff-mix. The author also comments that the articles reviewed have a task based approach which divides the work nurses undertake into tasks whereas the work nurses do is more of a cognitive and a conceptual practice. Cavanagh & Bamford (1997) review the substitution of nurses by unlicensed staff and the substitution of medical staff by advanced practice nurses. Advanced practice roles have developed lately and allow nurses to undertake some medical activities. Their review of the impact of changing skill mix on the quality of care reports that there is no evidence that changing the skill mix of hospitals results in a deterioration in the quality of care in hospitals using more unlicensed personnel instead of nurses. They also report an increase in patient satisfaction when care is performed by advanced practice nurse compared to doctors. However the authors note that the evidence is based on “shaky ground[s]” (page 336). The McGillis Hall (1997) and Cavanagh & Bamford (1997) literature reviews are not systematic reviews. Spilsbury (2001) and Buchan & Dal Poz (2002) undertake systematic reviews and find more mixed results. Spilsbury (2001) in a systematic review of the literature for the UK aims at understanding the impact of skilled nurses on patients’ outcomes and consequently reviews the literature on skillmix between unqualified and qualified nurses (unqualified and qualified are the terms used for licensed nurses and registered nurses in the USA). They focus specifically on literature for the UK 32

Chapter 2

Literature

from 1992 onwards. They used the Cochrane Literature Database, the National Research Register, LION (Department of Health library online), CINAHL, Medline, and British Nursing Index. The key words used in the search were: Nursing care, Nursing contributions, Skill-mix, Advanced practice, Specialist practice, Nurse-led care/initiatives, New nursing roles, nursing development units, physicians hours, doctors hours/workload, Health care assistant, Magnet Hospitals, those terms were searched and linked with patient outcomes and cost effectiveness. Some of the articles reviewed reveal that higher quality of care is associated with qualified staff. Some articles reveal that less qualified than nurses staff groups can do the same job as nurses as efficiently. Moreover, some articles reports that some tasks done by nurses could be fulfilled easily by other staff because they are not related to care directly. In consequence, Spilsbury concludes that the results from the literature are not clear cut. Buchan & Dal Poz (2002) undertake a systematic literature review of the impact of skill mix on outcomes and costs. The article is based on two literature searches, the first one was made for the World Health Organization; they reviewed English-language material published between 1986 and 1996 on the databases CINAHL, Medline, RCN Nurse Rom, ASSIA plus and FirstSearch. The second systematic review reviewed articles between 1996 and 2000 on Medline, CINAHL, ASSIA and Nurse Online. The key words used were: skill mix, skill substitution, personal mix, reprofiling, staffing levels and staffing mix; changing role was also included for the second search. The authors argue that the articles reviewed are weak and do not provide a lot of evidence. Most of the articles are descriptive or when they are not, methods are poorly employed. However, some results can be derived from those articles; there is no evidence that the substitution of nurses by unqualified professions is effective in every situation or leads to a reduction of costs. However it seems that there is an increase in quality when substituting doctors by nurses. Most of the studies reviewed are from the U.S.A. Aiken et al. (2003) report the results of a cross section study of 168 non federal adult general Pennsylvanian hospitals (1999). They analysed the impact of changing skill mix on risk adjusted patient mortality and failure to rescue within 30 days of admittance. The authors adjust for patient characteristics and hospital structures and reveal that a 10% increase in the share of nurses holding a bachelor degree was linked to a 5% reduction in the likelihood of patients dying within 30 days of patients admissions and a 5% reduction in the odds of failure to rescue. The authors conclude that hospitals with a higher proportion of nurses with higher degrees of education have lower mortality rates.

33

34

n211 n212

n222

n221

I22

N2

I21

n122

n121 n112

n111

C2

P1

C1

I12

I11

P2

N1

Diagram 2.2: Effect of a change in funding depends on the types of production functions

Chapter 2 Literature

Chapter 2

Literature

Lee et al. (2005) analysed a quasi experiment in a Taiwanese hospital. The quasi experiment involved a change in skill mix within the hospital. The study reveals that in the short term there was no impact on patient outcomes of a reduction in the proportion of registered nurses compared to nursing aides. They also report an economic evaluation showing that the change resulted in a decrease in cost. However the results given do not assess the long term effect of the change in skill-mix. Reviewing the different definitions of nursing is out of the scope of this thesis. However, it should be noted that changing the staff-mix might result in deterioration in the quality of nursing jobs and this may have an impact on health care delivery and patient outcomes. Some studies have focused on the effect of a change in staff-mix on the health care workforce: on workload, satisfaction, hierarchy, and organisation. Rafferty et al. (2007) analysed the impact of different nursing staff levels on the outcomes of general, orthopaedic and vascular surgery patients in 30 English acute hospitals in 1999. The study reveals that hospitals with fewer nurses per patient have lower levels of satisfaction among nursing staff. Moreover, hospitals with low nurse to patient ratios had higher mortality rates. Walker et al. (2007) observed nurses activity before (in the year 2000) and after (2002) a change in the organisation of care in an Australian hospital. The study reveals that a change in staff-mix (with a higher proportion of unqualified staff) in acute care wards and operating services changed the way the work was done. Such changes in the organisation of work have been argued to be counter productive by Hancock (1992), Ivor (1994) and McGillis Hall (1997). They have argued that changes in the organisation of work can result in the jobs of nurses changing from holistic practice towards more task based work. A study which elicited the opinions of nurses and health care assistants on changes in staff-mix in a hospital ward (Daykin and B Clarke, 2000) reported that nurses viewed this as deskilling, resulting in a change from a holistic approach to a task oriented practice. In contrast health care assistants saw it as an opportunity to obtain new skills. Further the authors showed that according to nurses their job was already task oriented. In line with some authors arguing that the success in a staff-mix depends on good strategic planning and human resource management (Sibbald et al. 2004). Sibbald et al. (2004) and Druss et al. (2003) argue that staff-mix in health is seen as a competitive policy. When the staff-mix will not be seen any more as a substitution between staff but as complimentary, the argument follows, it would be better both for the staff and patients. Thus research provides no strong evidence of any harmful effects of changing the staff-mix by employing a greater proportion of less qualified nursing staff though this may be because such 35

Chapter 2

Literature

effects are difficult to measure. There is some evidence that quality is related to the presence of qualified nursing staff. Clearly nurses have concerns that changing staff-mix changes practice and this again may lead to deterioration in quality and in the patient experience. This literature reveals that changing staff-mix is possible, though its impact is not well evidenced. Therefore, if this thesis reveals that the competitiveness of pay of nursing staff has an impact on skill mix, then it may also have some adverse consequences on outcomes. Skill mix may be driven by other sources than pay competitiveness, the following subsection will review the evidence from the literature on drivers of staff-mix.

2.4.2. Drivers of staff-mix Changes in factors other than relative pay can result in changes in staff-mix. In the UK there was a change in the balance of care provided by the primary and secondary sectors in the nineties as a result of the drive to reduce the time patients spent in hospital. Richards et al., (2000) and Hunter (1996) have proposed that this development may well have changed skill-mix in hospitals. The change in the balance of care was accompanied by a switch in resources from the secondary to the primary sector. Such a development will have shifted the hospital budget constraint toward the origin; depending on the shape of the production function a change in staff-mix could occur 33. More generally Buchan & Calman (2005) argue that the institutional setting will drive the staffmix of the the health workforce. They interviewed 13 informants in both England and the USA (informants were such as government representatives, professional associations, educators, employers and policy analyst). Respondents identified drivers for the introduction or extension of advanced practice nurse. Skills shortages, substitution, establishing a new type of services, government policy, and the medical profession are the drivers identified by English respondents. The USA respondents also identified value for money and nurse led initiatives but the medical profession was not identified by the USA respondents. In the USA the main facilitators were related to the nurse profession itself while the medical profession is seen as a constraint 34. In England the approach is more state led not linked to any profession in particular. The main constraint is the lack of visibility of career patterns that a nurse can have as an advance nurse; moreover there is no visibility on the pay that an advance nurses may receive. This two case studies highlight that drivers of skill mix would change according to the institutional setting, key 33 The usual assumption in neoclassical economics is to have a production function homogeneous of degree 0 which means that the production function is a linear function of the factors of production: in that case there would not be any change in staffmix. No a priori reasons would tell the particular shape of the production function. 34 In England the British Medical Association (BMA) was supportive of nurses taking more advanced roles as they were concerned by the shortage of medical doctors and the pressure on consultants.

36

Chapter 2

Literature

informants in England and the USA identify different drivers. If taken at the lower level, assuming that there are different regulations for different types of hospitals, drivers may differ according to the regulation. Hospitals in acute care are expected to be regulated differently than those in mental care, therefore, drivers of skill mix in the two types of hospitals are expected to differ. Acemoglu & Finkelstein (2008) focus on the impact of regulatory change for American hospitals on input mix (labour and capital) and technology choices. They find that the introduction of the Medicare Prospective Payment System (PPS) in 1983 tended to increase the ratio capital to labour due to a decline in labour input. The PPS was a new regulatory regime that reimbursed only capital expenditures while labours expenses were supposed to be covered by the fixed price paid per unit of output. Therefore, it increased the relative price of labour inputs. The reimbursement of labour was prospective with a fixed price, while the reimbursement of capital was retrospective and based on the share of Medicare patients a hospital had. Data was from the American Hospital Association census; the authors took 6 years of data, 3 years before the PPS implementation and 3 years after (1980-1986). Some states were excluded as they had some specific features. The data was for 6200 hospitals per year and contained information on local labour input expenditures and its components. The most interesting finding of the article was that there was an increase in the proportion of skilled nurses; the increase in this proportion was argued to be an effect of the adoption of more advanced technologies as skilled labour is often said to be complementary to technology. However the tendency to increase the proportion of skilled nurses existed in the years before the introduction of PPS; the observed effect before the introduction of PPS was half the magnitude after the introduction. This paper provides evidence that technology is complementary to skilled labour. For the UK Buchan & Calman (2005) have argued that technological innovation also increases the demand for skilled nurses and that such developments reduce the scope for substituting assistant nurses for nurses and may change the production function which consequently changes the slope of the production function isoquants.

37

n21 n22b n22a n22c

N2

n12

n11

Pc

38

Pb

Slope= -w11/w21

Pa

N1

Diagram 2.3: Change in staff-mix with an exogenous production function

Chapter 2 Literature

Chapter 2

Literature

An improvement in the general level of real terms funding for hospitals, as occurred in the UK over the period studied shifts the budget constraint out. Czuber-Dochan et al. (2006) report from a survey of ophthalmic units in the UK that specialised nurses are more likely to be present in bigger institutions. The impact of a change in the general level of funding, workload or size depends on the potential returns to scale which in turn depends on the form of the production function. The diagram 2.2 (page 34) presents the effect of a change in the budget constraint 35 on the staff-mix in two hospitals with different production functions (P1 and P2). The isoquants are represented by dashed lines with the first subscript identifying the production function which this isoquant belongs to and the second subscript identify the budget constraint. The budget constraints slope represent the relative price of one group of staff to the other. The intersection of the budget constraint with the vertical (horizontal) axis gives the maximum number of staff N2 (N1) that the hospital can hire out of a given budget. An inward shift of the budget constraint represent a reduction in budget. The number of staff in each case are given on the horizontal and vertical axes. The first subscript represents the staff group (N1 or N2), the second subscript represents the production function (P1 or P2) and the third subscript is the number of the budget constraint (C1 or C2). A reduction in the size of the budget as represented by an inward shift from C1 to C2 causes a reduction in the number of both types of staff employed. This reduction is bigger for the staff group of N1 on production function P2 ( n111 −n 112n211 −n212 ) and is bigger for the group of staff N2 on production function P1 ( n221 −n 222n121 −n122 ). As a consequence a reduction in budget as represented by an inward shift in the budget constraint will result in a change in staff-mix and the direction of this change depends on the form of the production function. If the production function is not linear, represented by a straight line through the origin, then any change in budget will change the skill mix. Drivers of staff-mix that were highlighted in this section are the technology, institutional settings and workload/funding/size. If possible the following empirical analysis will try to take that into account. The

final

subsection

will

review

articles

focusing

on

the

degree

of

complementarity/substitutability of health professionals.

2.4.3. Complementarity/substitutability of health professionals Subsection 2.4.1 has reviewed evidence of changing the skill mix on delivery of care, this 35 A change in the budget constraint and a change in size are similar in microeconomics as the budget constraint represent the maximum inputs possible.

39

Chapter 2

Literature

subsection reviews articles that specifically focus on the degree of substitutability between health professionals. The research revealed here involve empirical testing of production functions. No article has been found with the specific goal of estimating production functions in order to quantify the degree of substitutability of two nursing groups. Articles that estimate substitutability of nursing groups, or at least mention it, do so as a by product of their analyses. Four articles (Van Montfort, 1981; Jensen and Morrisey, 1986; McGuire, 1987; Eastaugh, 2010) use production functions to give some insights to the substitution/complementarity of nursing staff. Among these articles, only Eastaugh (2010) estimates substitutability between two nursing groups. Jensen & Morrisey (1986) estimate the degree of substitutability between doctor specialities. Van Montfort (1981) estimates a Cobb Douglas, a Constant Elasticity of Substitution and a translog production function on 100 general hospitals in the Netherlands for 1971. As outputs the author uses weighted admissions (the weighting technique was not made clear). Inputs were measured as the number of registered nurses, student nurses, other nursing staff, paramedical staff and doctors in hospitals. The author also measured inputs of drugs and paramedical devices. The author reminds the reader that elasticities of substitutions “do not indicate how, over a period of time, a hospital might substitute one input for another; rather how, in practice, hospitals with differing input ratios realise a certain average output level.” (Van Montfort, 1981; 93). The author then argues that there is some degree of substitution between certain types of labour. However, the author does not calculate any degree of substitution, arguing that it is not possible to assess fully (in terms of significance) the sign of the elasticities. Therefore, for this study it is only possible to conclude that there might be some substitutability among nursing staff. Jensen & Morrisey (1986) focus on different medical specialities and the degree of substitutability between them and other inputs. They used data from the American Hospitals Association (USA associations of hospitals) survey of 1983 and a case mix index from the United States of America Department of Health and Human Services. They excluded federal and teaching hospitals and focused on short-term acute hospitals.

40

41

n22a

n21

n22b

N2

n12 n11

Pa

Slope= -w11/w21

Pb

N1

Diagram 2.4: Change in staff-mix with an endogenous production function

Chapter 2 Literature

Chapter 2

Literature

They show that general practitioners in inpatient care (meaning that they work at the hospital) are complements to other hospital medical specialities. Obstetricians/Gynecologists are complements to pediatrics, the former, as the authors argue, creating a demand for the latter. Nurses are complements to

primary care physicians

and pediatrics

but not to

obstetricians/gynecologists and surgeons. This article does not provide any data on the degree of substitutability between different groups of nursing staff. McGuire (1987) analysed hospital allocative efficiency. The author uses Scottish data for the financial year 1983/84 for non teaching mainly acute hospitals with more than 50 beds, the total sample is therefore 28 hospitals. The analysis uses as inputs the number of hours of medical labour, number of hours of nursing labour and number of beds and service department labour (paramedical and housekeeping). The authors obtained the data from the Scottish Health Service Costs. The total number of nursing hours worked was collected from the New Earning Survey, the total number of medical hours was collected from the Office of Manpower Economics supporting the Review Body of Doctors and Dentists Remuneration. Labour costs were imputed from the Annual Accounts of the Individual Scottish health boards. The author was unable to distinguish between the different nursing staff groups. The author did not provide a direct estimate of the substitutability of inputs, but stated that “[t]he results also revealed that the own (Allen) elasticities of substitution were negative” (McGuire, 1987; 721)36, the interpretation of this statement is that nursing staff and medical staff are found to be complementary. Eastaugh (2010) use a sample of 58 hospitals in the USA with measurements of outputs and inputs between 2002 and 2005. Hospitals had between 100 and 900 beds. They estimated a production function on those hospitals with nursing output measured by a score sold by the largest proprietary vendor of nurse workload and scheduling systems in Ohio (Cardinal Health Information). No detail is given about the score or the hospitals in the data. Qualified nurses are found to be complements with less qualified types of nurses with a partial elasticity of -0.438. It is a pity that the only estimate of the elasticity of substitution that is given is in an article that provides little information about the data. Therefore evidence from the literature on the degree of substitutability between groups of staff is small and cannot help to draw many useful conclusions on which groups can be regarded as substitutes. This section aimed at reviewing the literature on skill mix in order to pave the way for the empirical analysis of this thesis. The thesis aims at analysing the impact of the competitiveness of 36 Allen elasticities are input elasticities see (Behar and Stevens, 2009).

42

Chapter 2

Literature

pay on shortages in hospitals and of a potential adaptive behaviour of hospitals. Hospitals may wish to dampen the effects of shortage of staff by altering staff-mix. Therefore, reviewing the impact of a change of staff-mix on outcomes and costs seemed necessary. Subsection 2.4.1 reviewed and concluded that there is no clear evidence of any adverse effects of changing staffmix on patient outcomes. Non labour market drivers of staff-mix were reviewed in Subsection 2.4.2; technology, institutions and workloads/funding were highlighted as potential drivers. The empirical analysis which follows will use these results in setting the estimation procedures. If this is not possible the discussion will acknowledge it. Finally, in Section 2.4.3 the review of estimations of whether nursing staff are substitutable or complementary was not found to be very conclusive. Therefore, in the subsequent empirical analysis of changes in staff-mix, no a priori expectations can be drawn based on previous empirical research. An attempt to draw some theoretical conclusions of the impact of pay competitiveness in one local labour market on hospital behaviours is made in the following section.

2.5. Paving the way for empirical analyses This section explains the rationale of the empirical strategy that will be used in subsequent chapters. For both England and France, the shortage and the skill mix of nursing staff is analysed. Subsection 2.5.1 explains the potential impact of the competitiveness of pay for one nursing group on the shortage of the other nursing group. Subsection 2.5.2 explains the expectations of the impact of local labour markets on skill mix.

2.5.1. Cross effects of labour markets In paragraph “i” it is argued that the competitiveness of pay for one nursing group will have consequences for the recruitment of another nursing group. In paragraph “ii” it is argued that the effect of the competitiveness of pay for one nursing group will differ whether the pay for the other group is competitive or not. Assume that there are two groups of staff N1 and N2. i Competitiveness of pay for one nursing group has consequences for recruitment/retention of the other nursing group On one hand, if N2 is the most skilled nursing group, the inability to recruit N1 may mean that tasks usually done by N1 might be given to N2. Because N2 is more skilled it is easier to transfer tasks done by less skilled staff to the more skilled workers. Tasks undertaken by N2 are not transferable to N1 because they simply do not have the skills to do them. Therefore the pay 43

Chapter 2

Literature

competitiveness for N1 may mark an increased workload for the other nursing group, higher tensions and worse working conditions: less attractive workplace. Staff N2 may not be attracted as a result. On the other hand, if the group N1 is hierarchically lower (less skilled) than the group N2 and if bridges between the two nursing groups exist it might be expected that N1 would be attracted to areas where they would be able to get a competitive pay as they become N2 staff. Those bridges exist between assistant nurses and registered nurses (Grimshaw 2009) in the UK as the form of secondment in training paid at 80% by hospitals. In France, workers in the public service are eligible for a training vacation that can last for up to three years (nurse training is 3 years) and which is funded for up to two years37. Both cases give the same conclusion, the pay competitiveness of one group of staff is the marker of unobserved characteristics of the workplace. Those unobserved characteristics have an impact on the recruitment of staff. Therefore, the empirical strategy will test the impact of the competitiveness of pay for one group of staff on the shortage of the other group of staff. ii Effect of the competitiveness of pay for N1 differs according to competitiveness of pay for N2 If as suggested above the pay competitiveness of one group of staff explains partly the shortage of the other group of staff, it might also be the case that the own gap may have a different effect when the pay for the alternative group of staff is more or less competitive. Say that there are two nursing groups N1 and N2. It is expected that the more competitive the pay for N1 the lower the shortage of N1 all else equal. In a linear model, the effect (slope) of the competitiveness of pay for N1 will be expected to be the same whatever are the values of the other variables. It is possible to lessen a bit this assumption by introducing an interaction effect. The empirical strategy will test for the impact of the competitiveness of pay for N1 when the competitiveness of pay for N2 is above or below the national average (for France, the models will be slightly more subtle). Choosing the national average is arbitrary but easy to understand compared to an interaction of two continuous variables. The results will show two effects one for areas where the pay competitiveness for the alternative group of staff is above the competitiveness of pay nationwide and one when it is below.

37 Décret N° 2008-824 Du 21 Août 2008 Relatif à La Formation Professionnelle Tout Au Long De La Vie Des Agents De La Fonction Publique Hospitalière, 2008. http://www.legifrance.gouv.fr/affichTexte.do?cidTexte=JORFTEXT000019354799. Accessed the 23rd of March 2012.

44

Chapter 2

Literature

2.5.2. Staff-mix and local labour markets Section 2.2 discussed the impact of pay gaps on hospital staffing and public service delivery, however no literature has been found which explores the impact of uncompetitive pay on the demand for labour. If hospitals are unable to recruit because of uncompetitive pay (and result in shortages of staff) they may employ adaptive behaviours to mitigate the impact on hospital output. Changing staff-mix can be one of these behaviours. A behaviour that implies modifying the skill mix means that hospitals have some scope in altering the organisation of care, said in other words, they can change the production function. If a hospital cannot recruit the nursing group of staff N1 because of an uncompetitive pay, then to reorganise the way care is performed, the hospital need to find tasks that were done by N1 which can be undertaken by an alternative nursing group of staff: N2. This is possible if some of the skills held by the nursing group N1 overlap with the skills of the nursing group N2 and if those skills are actually used in the care performed by the specific hospital suffering from an uncompetitive pay for N1. a) Impact of non competitive pay on hospitals which cannot change their production functions If none of the skills held by the group of staff N1 overlap with skills held by the group of staff N2 or if hospitals do not use the overlapping set of skills, then hospitals facing a shortage will not be able to reorganise the way care is performed. What is the impact of such a situation on the skill mix? In a simple micro economic framework, this situation is described by Diagram 2.3 (page 38). Initial situation is represented by the intersection point between the production functions Pa, Pb or Pc and the budget constraint. This situation is one that would happen if pay for the two groups of staff were competitive which means that hospitals can hire as many staff as they wish. Assume that hospitals cannot hire N1 at the level they wish, and that the production function cannot be altered by hospitals. Hospitals can only hire n12 number of N1. The optimum level of output is not achieved because there is an external constraint: not enough supply of N1 (n12 < n11). Depending on the production function of the hospital a different skill mix may occur. The observed skill mix if the production function is Pa will not change because the production function is linear. Therefore, the empirical scientist would not observe any difference in skill mix between one hospital which do not face uncompetitive pay for any staff and one hospital which does. In this case, the uncompetitive pay will not have any impact on the proportion of staff as for 45

Chapter 2

Literature

each level of staff a decrease in N1 needs a decrease of N2 of x times N1 (where x is a fixed number above 0). If the production function is Pb, then the skill mix will have a larger proportion of N2 at n22b level. However, if the production function is Pc, then the skill mix will have a larger proportion of N1, the group of staff which is difficult to hire. This is observed because with the production function Pc, the needed level of N2 is lower than with Pa. b) Impact of non competitive pay on hospitals which can change their production functions If hospitals can reorganise the health care performed, then the skill mix should be altered because this is the most sensible behaviour, this situation is described by Diagram 2.4 (page 41). Initial situation is described by the intersection of the production function Pa and the budget constraint. The number of staff N1 that hospitals can hire is n12 which is lower than the sought one n11. At that level, hospitals only employ N2 staff at level n22a and hospitals do not use the whole of the budget. As they can alter the way care is performed they will seek to change the production function up to use the maximum of the budget constraint. They move the production function to Pb which uses more of staff N2, hospitals have transferred tasks from N1 to N2 and employ n22b numbers of N2. Hospitals in this case are responsible for the change in skill mix. In the introduction of this subsection, it was argued that mitigating the effect of an uncompetitive pay for one nursing group of staff depends on whether the skills of the two nursing groups overlap and if the nursing staff use these skills to perform the care. Whether the overlapping set of skills are used by staff to perform the care may well depend on the type of care performed. Mental and acute hospitals perform different care and one may use the set of skills which the nursing staff groups N1 and N2 share and the other one may not. If the type of activity performed by hospitals allows a change in the production function, it would be expected that hospitals would employ the nursing staff which can be recruited in that area. Therefore hospitals would employ more of the nursing group for which the pay is competitive and less of the nursing group for which the pay is uncompetitive. In that case a competitive pay for the nursing group N1 should lead to a higher proportion of N1. Therefore the impact local labour market conditions on shortage may differ according to the type of care performed. One way of measuring the fact that hospitals perform different types of care is to take into account the types of hospitals (for example mental or acute) and allow the impact of local labour market conditions to vary according to the different types of hospitals.

46

Chapter 2

Literature

2.6. Conclusion Chapter 2 reviewed the literature on three topics. Subsections 2.1.1 and 2.1.2 reviewed the empirical support for compensating wage differential theory. The compensating wage differentials theory describes why employers in local areas with different levels of amenities and cost-of-living should pay different rates to their employees. The empirical tests of this theory did not reveal strong empirical evidence to support the theory. Subsection 2.1.3 provides strong empirical arguments for different pays across different sectors of the economy. Sectors with a higher degree of collective bargaining are likely to exhibit less variations in pay, both geographically and across occupations than sectors with few or no collective agreements. Section 2.2 described the small number of studies that have explored the impact of pay gaps between the public and the private sectors on the delivery of public services. It gives some evidence that pay competitiveness of public sector staff has an impact on the public service staffing shortages and service delivery. Where the public service pay is less competitive than the private sector pay, the public service delivery is, to some extent, affected. Section 2.3 reviewed some of the literature on the drivers of shortage of staff in hospitals. This reviews mainly conclude that the drivers of shortage and turnover is job satisfaction. This conclusion does not help to draw empirical strategies to measure the drivers of shortage at the hospital level. There was mixed evidence of the sign of the impact of size on shortage of staff. One article reviewed showed that the type of hospitals matter in the turnover of nursing staff. Employees of hospitals in England and France are covered by collective agreements. Where pay is non competitive hospitals may try to find an alternative solution. Altering staff-mix is one response that is investigated here. Section 2.4 reviewed the literature on staff-mix in hospitals. The literature (2.4.1) was not clear cut on the impact of changing nursing staff-mix on patient outcomes. Drivers of staff-mix (2.4.2) highlighted in the literature are technology levels, workloads, funding and institutional settings. Economic literature (2.4.3) on testing the substitutability/complementarity of the nursing workforce is weak and did not provide any reliable measure of potential substitutability. Section 2.5 describe what could be expected from uncompetitive pay in different circumstances and therefore it provided a straightforward empirical strategy. This thesis will extend the work on staff shortages made by Elliott et al. (2006, 2007, 2009). Registered nurses (also called qualified nurses in the UK, or simply nurses in France) are one group of staff. Assistant nurses (also called, licensed practice nurse in the USA, unqualified nurses, nurse auxiliaries or health care assistants in the UK and health care assistants in France) is a second, less qualified, group of nursing staff. Chapter 6 (for England) and 8 (for France) will 47

Chapter 2

Literature

analyse the impact of the competitiveness of the pay of assistant nurses and registered nurses on shortages of assistant nurses and registered nurses. However, changing the staff-mix may provide a way of adjusting to uncompetitive public sector pay provided hospitals can change the production function. By changing the production function, hospital will substitute one staff for the other. At the aggregated level substitution possibilities between different nursing group of staff are supposed to exist. Chapter 7 (England) and 9 (France) will investigate the impact of the local labour market on the result of an adaptive behaviour of hospitals.

48

Chapter 3 Institutional Settings and Data This thesis tests for the impact of pay gaps between public hospitals and the private sector on shortages of staff in public hospitals and on the way hospitals might adapt by altering their behaviour in consequence. Therefore, the thesis tests for the impact of pay gaps on the supply side of the labour market (shortage) assuming in a first instance that the consequences of different pay competitiveness have no impact on the demand for staff. Demand for staff, in a first approach, is assumed not to be affected by local labour markets. The thesis then attempts to test the impact of the pay gaps on the demand side of the labour market (staff-mix) assuming that an observed different staff-mix for hospitals with different pay gaps is the result of hospitals behaviour and not a constrained outcome as paragraph “a” in Subsection 2.5.2 (page 45) would suggest. In the afore mentioned paragraph, hospitals cannot change their staff mix and the result of an uncompetitive pay may well be an observed change in staff-mix, though, this staff-mix is constrained as it is not sought by the hospital. This chapter describes the institutional setting and provides some preliminary analysis for England and France. The English and French organisation of health care is presented in Section 3.1 and 3.2 respectively, they are compared in Section 3.3. The data used for hospitals is described in Section 3.4 for England and in Section 3.5 for France and compared in Section 3.6.

3.1. UK healthcare setting In this section the organisation of healthcare in the UK will be presented in order to inform the analyses that will be performed later on. This section will focus on what matters for the thesis: hospitals, secondary care organisation and pay setting. When necessary the text will refer to some reforms implemented over the last 20 years. The general organisation of hospitals will be presented along with the funding system, then pay setting at the beginning of the 21 st century. A slight detour to staff categories in the data will be made in order to clarify some issues regarding the wording of staff in official documents. Finally recent reforms 1 of pay setting will be presented. 1

Agenda for Change.

49

Chapter 3

Institutional Settings and Data

3.1.1. Hospitals and secondary care in the UK: Organisation In the United Kingdom, health care is provided by a national system free at the point of delivery and financed by general taxation. Secondary care is mainly provided by public hospitals although a small sector of private hospitals does exist. The structure of the NHS in the UK has been subject to many reforms2. In the beginning of the ninety nineties, the government introduced a separation between purchasers and providers of health care services. This allowed hospitals to compete within an internal market to win contracts with local health authorities who held the budgets for local health care expenditure. Within this new framework, both providers and purchasers have themselves undergone various reforms. These changes are both substantive in the sense they are fundamental changes to their remit or regulations that govern them and administrative for instance mergers between adjoining units of organisations. Finally, in 1999, devolution to the 4 nations constituting the UK (England, Scotland, Wales and Northern Ireland) was introduced. The result has been that the organisation of care differs between the nations. However, pay to NHS staff is set and agreed UK-wide. The organisation described here reflect the one in England only, as no analyses will be conducted in the other nations. The text may refer to the UK when the description concerns the whole of the UK. In 1997, Primary Care Groups which consisted of representatives of general practitioners, community nurses and other individuals involved in primary care were created. They commissioned hospitals services for their patients. First they were an advisory group to the Health Authority (HA), then they had their own budget to purchase services but still remained part of the HA, thirdly they became free standing bodies but accountable to HA (Fisher, 1999). Finally, they became Primary Care Trusts (PCT), independent and responsible for primary health care and also responsible for the population access to secondary health care. Their number narrowed down from 303 in 2005 to 152 in 2006 (BBC 2006). Hospitals in 1991 became hospital trusts. This status gave hospitals more freedom; they would be run by a board of appointed directors, they would employ their own medical and non medical staff but they would not set their pay. They would own the buildings and they could sell them off (Fisher, 1999). Each hospital trust comprise a number of sites. The number of sites for each hospital trust increased over time as some trusts would merge together. Between 2003 and 2005 12 hospital trusts merged into 6 (see annex A, page 240). Hospital trusts are grouped in different categories according to the main care they perform (slightly more than half of them are acute 2

Current management of the National Health Service (NHS) is undertaken separately by the devolved administrations within the U.K. The discussion in this section focuses on the NHS in England as the analyses are done for English hospitals.

50

Chapter 3

Institutional Settings and Data

hospitals). Those hospital trusts are the focus of the study in the following chapters. By the end of the period studied here, some hospital trusts achieved Foundation status (then called Foundation Trusts, FT). FT status allows more freedom in the way the hospital is managed (20 hospital trusts achieved status in 2004 and twelve in 2005). This may mean more flexibility over staff-mix. Hospitals had to apply for foundation status and were judged according to a set of criteria that emphasised both financial and quality aspects of performance. FTs became no longer accountable to the health department and would manage themselves. FTs have more freedom to spend their income, they can form companies, obtain ownership of shares of existing companies and make independent investments. FTs can keep surpluses and do not have to break even year on year. FTs can borrow money, however the limit is set by Monitor 3.“There is no good evidence currently available about how FTs are operating compared to other NHS trusts (other than their financial outcomes, which are reported by Monitor).” (Allen, 2009; 8). Moreover, FTs were granted more freedom on setting pay, following a reform that started to be put in place in 2004 called Agenda for Change. The extended power of FTs raised concerns that these trusts may become more competitive than neighbouring NHS trusts (Booth et al., 2004).

3.1.2. Hospitals and secondary care in England: Funding of hospital trusts and primary care trusts Funding in the NHS aims at providing equal access to health care for equal needs. Primary Care Trusts are financed through a formula based on the health needs of the population corrected for unavoidable variations in input prices. Different formulae have been used over the years, the York formula (Carr-Hill et al., 1994), the Allocation of Resources to English Areas (Sutton et al., 2002), and more recently the Combining Age-Related and Additional Needs formula (Morris et al., 2007). The correction for unavoidable variations in input prices is based on the Market Forces Factor formula, which is calculated by using Standardised Spatial Wage Differentials (SSWDs). PCT receive 75%-80% of the budget of the Department of Health in England and then commission hospital trusts to provide health care for the local population in the remit of the PCT (BBC, 2006).

3.1.3. Pay setting Seifert (1992) argued that the pay setting in the NHS created a two tier system as groups of staff had their pay increased through a dozen of centralised institutions; nurses pay was set by one pay Review Body, doctors pay was set by another Review Body, while cleaners and porters pay was set

3

The independent body in charge of monitoring FTs.

51

Chapter 3

Institutional Settings and Data

through collective bargaining. Grimshaw (1997) provided evidence that the difference in pay increases between staff narrowed during the 1990s. Since 1983 nurses pay has been set by a pay Review Body which makes recommendations to the government based on interviews, analysis and research of both representing bodies of the employees and the employers as well as the Department of Health 4 itself. The Review Body makes recommendations on pay increases for all staff and also specific increases for groups of staff which are judged difficult to recruit and retain. When introduced in 1992, along with changes to registered nurses training, health care assistants were hired on local contracts in a move by the government of the time to decentralise the pay setting. Health care assistants are another type of assistant nurses and as the paragraph 3.1.4 argues both will be treated as part of the same staff group in the analysis. In 1997, the government changed and those efforts were stalled. From 2001 onwards health care assistants increasingly came within the scope of the nursing Review Body. In 2001 the Review Body for Nursing Staff, Midwifes, Health Visitors and Professions Allied to Medicine argued that the Review Body pay scales for nurses should “provide a wider framework for Trust negotiations on the pay and conditions of nursing staff on distinct Trusts contracts.”5 (Review Body for Nursing Staff, Midwives, Health Visitors and Professions Allied to Medicine, 2001; para. 7.29) and that “[w]e will continue to monitor the position of these staff”(para.7.29). The last report before the introduction of Agenda for Change (see page 54 Subsection 3.1.5) in 20026 said that the staff side were asking the Review Body that the implementation of extra pay for staff on lower grades who achieved the National Vocational Qualifications at level two and three should also be recommended for health care assistants on local contracts (Review Body for Nursing Staff, Midwives, Health Visitors and Professions Allied to Medicine, 2002; para. 7.64). Assistant nurses are on grades A, B or C on the pay scale. To be on grade B or C, they need to hold National Vocational Qualification level II or III (respectively). On the first of April 2004, the lowest annual salary an assistant nurse could get was £10,375 per annum, with the highest being £17,060 p.a. Registered nurses are on grades D to I, and on the first of April 2004, the salary range goes from £17,060 to £33,920 (£34,920 on the last discretionary point). The pay scale is thus linear and there are no jumps from one grade to the other. Except for grade C to D where the highest spinal 4 5 6

The Department of Health is in charge of health care in England, the review body would also take evidence from the respective health directorate in Scotland, Wales and Northern Ireland. Trust is the generic terms when referring to hospital trusts or primary care trusts. Agenda for Change was only implemented in 2005 but the agreement occurred in 2003 and thus the parties asked the Review Body to publish a small report saying that the parties have agreed on the Agenda for Change and that during the time of the implementation there should not be any report submitted by the Review Body.

52

Chapter 3

Institutional Settings and Data

point on grade C is exactly the amount on the lowest spinal point on grade D, each grade overlaps with the grade below and the grade above (Review Body for Nursing Staff, Midwives, Health Visitors and Professions Allied to Medicine, 2003; 8). In April 2001, the government decided to implement Cost-Of-Living Supplements (COLS) on top of existing London Allowances for all registered nurses and midwives (grade D to I) and the highest grade of assistant nurses (grade C) working in London and parts of the South of England 7. These supplements were not paid to grade A and B (the two lowest grades for assistant nurses). COLS were implemented as recognition that those working in some areas face higher costs. As a result hospitals face indirect costs in these areas as staff are not attracted because the pay is not competitive. The COLS was 4 per cent of basic salary in London subject to a minimum of £400 and a maximum of £1000. Outside London it was 2.5 percent of basic salary with a minimum of £400 and a maximum of £600 (Incomes Data Services, 2002; 185). In 2002, the COLS were extended to more regions8 outside London (Incomes Data Services, 2003; 160).

3.1.4. Assistant nurses and health care assistants, a clarification In the UK, registered nurses are assisted by two kinds of support workers: assistant nurses and health care assistants9. Assistant nurses comprise nursery nurses and nursing auxiliaries and the former in particular have long been a feature of the NHS. Health care assistants were introduced in the early 90s to allow hospitals to cope with the absence of nurse students who had until then provided support for registered nurses (nurses training used to be based at the hospital level and was replaced by a degree obtained at the university level), their pay was not set by a review body. Grimshaw (1999) shows that the roles of health care assistants and assistant nurses are not distinct. Further evidence from NHS careers 10 and the British Medical Association, (BMA 11) shows that there is no substantive difference in the roles performed by the two groups; that any difference between the two groups is a mere coding artefact and that when support jobs are advertised, the two names, health care assistant and assistant nurses or nurse auxiliaries are used interchangeably. In Subsection 3.1.3 (page 51) it was explained that during the 1990s, the pay of health care assistants drifted from local agreements towards the Nurses pay review body. Subsection 3.1.5 (page 54) explains that the subsequent reform put the two groups under the 7

They applied to the areas of London, East and West Surrey, East, West and North Hertfordshire, Berkshire, West Sussex, Oxfordshire, Buckinghamshire, Bedfordshire, Cambridgeshire, North, mid South East and South West Hampshire, Wiltshire and Avon. 8 The added regions in 2002 are East Sussex, Essex, Kent, Northamptonshire. 9 http://www.ic.nhs.uk/webfiles/data%20collections/NHS_Occupation_Code_Manual_Version_8_1.pdf accessed May the 24th, 2011 10 http://www.nhscareers.nhs.uk/details/Default.aspx?Id=485 accessed May the 24th, 2011 11 http://www.bma.org.uk/patients_public/whos_who_healthcare/glossnurses.jsp accessed May the 24th, 2011

53

Chapter 3

Institutional Settings and Data

same pay agreement. In annex B (page 241) it is shown that the correlations between the whole time equivalent of assistant nurses and health care assistants for the data employed in the subsequent empirical analysis is highly negative (-0.25). This evidence supports the argument that assistant nurses and health care assistants are to be seen as substitutes. In the analysis which follows assistant nurses and health care assistants numbers are aggregated and throughout this analysis referred to as ‘assistant nurses’.

3.1.5. Agenda for Change By the 21st report (2006) of the Review Body on nursing staff, midwives and health visitors pay was within the framework of Agenda for Change (AfC), the new conditions of employment contract introduced in October 2004. In 1999 when the government published its proposals for a new pay structure for NHS staff in the UK, the government aimed to have more pay local flexibility with a national jobs evaluation framework and minimum threshold for jobs with grades and actual pay rates determined at the local level. This was said in 2002 to be a key point of tension between the Department of Health and the unions (Incomes Data Services, 2002; 190). The original plan was to have a new pay structure for all staff and the pay structure would be in three spines (the first one for doctors, the second one for nurses and the last one for non Review Body staff). Doctors unions and representatives have not been part of the negotiations and in the end were not part of the new pay structure, however a new pay pay scale was agreed independently: the new consultant contract. AfC has less local flexibility than originally planned. The structure of pay is set nationally and pay flexibility is achieved under only certain conditions. Pay in high costs areas such as the London allowance and the Cost of Living Supplements (COLS) in London and the fringe would be awarded by a system of high cost area allowances. Moreover, AfC introduced the Recruitment and Retention Premia (RRP). Those RRP would be allowed either locally or nationally. If set locally it would be done in accordance with neighbouring trusts and staff side organisations after all non monetary recruitment and retention initiatives have been exhausted. Nationally RRP should be put in place under recommendations of the pay Review Body or the negotiating councils for non review bodies. It should be set up “where market pressures would otherwise prevent the employer from being able to recruit and retain staff in sufficient numbers for the post concerned at the normal salary level and for a job of that weight” (Incomes Data Services, 2004; 153, 2005; 161). The total of locally and nationally RRPs should not exceed 30% of basic salary pay except for Foundation trusts. Foundation trusts were given more scope to set local pay premium above the 30 percent of basic

54

Chapter 3

Institutional Settings and Data

pay but they need to make sure that rewards are set according to NHS principles and that it will not undermine the ability of other providers to meet their NHS obligations. Jobs are evaluated through the NHS Knowledge and Skills Framework (KSF) which is a nationally agreed set of competencies and dimensions. Six dimensions are core to every job within the NHS and the AfC agreement. Some further ones are used for specific jobs. Every employee has their job assessed within the KSF, for each dimension their jobs will be evaluated and the pay on AfC will be matched with the assessment. Employees and line manager should meet annually in order to assess if the skills needed to do the job are part of the employee skills, what kind of training is needed for the employee to fulfil his/her job and what training is required to help the employee in the long term. Progression is not based on quotas. It is intended that every individual with the skills to progress will do so. Gateways of progression are intended to be accessible and not closed, individuals should be able to meet with their line managers in order to look at the training that will help them to progress (Incomes Data Services, 2004; 157–158). In 2004 there were 12 early implementers12 with subsequent implementation rolled out in 2005.

3.2. French health-care setting In France the coverage of the population by collective agreements is more widespread than in the UK. This and the existence of direct alternative employers are the main interesting features of this country13. Results for France may give some insights on what may happen for the UK, as the hospital private sector has been growing over the last 10 years. In 2006, there are 214 737 whole time equivalent nurses in public hospitals, 29 737 in private not for profit hospitals and 39 237 in private for profit hospitals (Afrita et al., 2008; 24).

3.2.1. Organisation of care The organisation of care is the responsibility of the Hospital Directorate of Health Care Supply (Direction d'Hospitalisation et d'Organisation des Soins (DHOS)) and of the Regional Hospital Agency (Agences Régionales d'Hospitalisation) 14. The former is in charge of organising care 12 Aintree Hospitals NHS Trust; Avon and Wiltshire Mental Health Partnership NHS Trust; Central Cheshire Primary Care Trust; City Hospitals Sunderland NHS Trust; East Anglian Ambulance NHS Trust; Guy’s and St Thomas’ Hospital NHS Trust; Herefordshire NHS Primary Care Trust; James Paget Healthcare NHS Trust; North East Ambulance Service NHS Trust; Papworth Hospital NHS Trust; South West London and St George’s Mental Health NHS Trust West; Kent NHS and Social Care Trust. In bold the trusts in the data used here. The data will be described in 3.4 (page 68). 13 Part of this work has been submitted to the Journal of Health Economics. The French chapters have received contributions from Eric Delattre, one of the co author on this paper. 14 With the last reform of 2010, this has changed and is only accurate for the analyses done here.

55

Chapter 3

Institutional Settings and Data

nationwide, supervising the creation of health care networks, making sure that patients are getting the appropriate care. Thus the DHOS is in charge of all types of hospitals (public and private) whether they participate in the public service or not. The latter are responsible for the organisation within their regions, organise care through Regional Schemes and negotiate funding of all types of hospitals according to their goals (Direction de l’information légale et administrative, 2005). Those public bodies depend on the ministry of health.

3.2.2. Hospitals Hospitals in France are classified as “établissements de santé”. Etablissements de santé are all those institutions providing health care. It includes dialysis premises which would receive patients just for a few hours, premises that are specialised in home care 15 and obviously, premises providing acute (medicine, surgery and obstetric), mental and long stay care.

3.2.3. Status Hospitals are divided into three different legal types: public hospitals, private not for profit hospitals and private for profit hospitals. Public hospitals 16 can be large regional hospitals providing education, research, acute care, and advanced treatments, local hospitals which run a smaller acute ward or a maternity hospital. Local hospitals depend on large regional hospitals for most of the technical care. They also rely on ambulatory care medical doctors to provide some of their services. All public hospitals run an emergency service. Not for profit hospitals are associations, mutual organisations, foundations, in which any surplus over costs is reinvested into the services provided for patients (it could be invested in new technologies for example). For profit hospitals have a commercial status and aim to make profit. Usually of smaller size, none of the private hospitals are as large as regional public hospitals. Private for profit hospitals tend to specialise in routine procedures. Some private not for profit hospitals participate in the public service and therefore have to fulfil similar functions as public hospitals (see the following section). Both public and private hospitals are regulated by the same body. Regulation covers both financial and non financial matters. Both public and private hospitals (whether or not they participate in the public service, see 15 Home care is a policy that will allow people who would have needed to be at the hospitals to be at home (patients with a cancer in a terminal phase for example). 16 Within public hospitals are included military hospitals, they depend on another entity (the ministry of defence) but represent only 9 (out of 942 public hospitals) hospitals and 0.56% and 0.68% of the total number of assistant nurses and registered nurses.

56

Chapter 3

Institutional Settings and Data

Subsection 3.2.4) are regulated by the same body. Regulation covers both financial and non financial matters (for regulating bodies see Subection 3.2.1).

3.2.4. Participation to the public service The hospital public service is provided by public hospitals and some private hospitals. The former are mandated to participate in the public service (“service public hospitalier”) while the latter opt in (Code de la Santé Publique, 1993). The private hospitals which participate in the public service have to accept any patient and provide equal access to all. They have to be able to make their premises available at any time of the day and night and run an emergency ward or be able to redirect patients to a hospital that runs one (equal access to all; continuous access even for potentials emergencies; continuity of care: patients are reoriented if necessary and given everything to continue the care needed once they left the hospital) (Afrita et al., 2008; 8). They also participate in education and research, for example small entities will welcome junior doctors (Adaius et al., 2007; 8). Private hospitals, whether for profit or not for profit, can participate in the public hospital service in either of the two following ways. The first one is an agreement between the private hospital and the government that covers all the activity done at the hospital. The second is an agreement for just some procedures17,18, only a small number of private hospitals participate in this way (less than 0.01%, Statistique Annuelle des Etablissements, 2006-2008). The advantages of taking part in the public service for the private hospitals is that they receive subsidies for equipment, on the other hand they are required to run an emergency ward or have to redirect patients to hospitals that will provide emergency. The private hospitals which do not participate in the public service get funded by the government (except a very small number of hospitals) only for the care they perform 19 (see Subsection 3.2.5). They do not get extra subsidies for equipment.

3.2.5. Funding The global budget to be allocated to hospitals is a sub target of the Objectif National de Dépenses de l'Assurance Maladie (ONDAM) which has been set by the parliament since 1996 every 17 In that case hospitals are being conceded procedures through an agreement with the ministry of health, and within a region no other hospital can propose such procedures, this is called the non concurrence agreement. 18 There is in fact an other type of hospitals participating in the hospital public service. Those hospitals (less than 0.1% of the hospitals) have an agreement with another hospital participating in the hospital public service. 19 Procedures performed by private hospitals not participating in the public service are, nevertheless, subject to authorisations by the regulatory body (which depends on the ministry of health).

57

Chapter 3

Institutional Settings and Data

September for the coming year. The budget for the coming year is based on the expenditure of the current year plus a growth rate decided by the parliament. The total expenditure of the current year is not perfectly known at the time of the decision (September) and is based on an estimation. Six sub targets (of which one is for the care performed in public hospitals and one for the care performed in private hospitals) are then derived with a different growth rate for each. Each region is then allocated a proportion of the budget with the aim of reducing inequalities between regions, consequently the budget for each region is adjusted following three elements: –

“a theoretical volume of hospitals stays, derived by applying national occupancy rates to the region's demographic structure;

–

weighting by a comparative mortality index (that is, the differential mortality of the region when controlling for age and gender)

–

the productivity of the hospital in the region.”

(Chevreul et al., 2010; 89). Since 2004, French hospitals have received payments for activity for short stays in Medicine, Surgery and Obstetrics (MSO, MCO in French). In 2005, 100% of MSO wards in private for profit hospitals have received payments for activity. These payments were gradually implemented for public hospitals: from 10% of MSO wards in 2004 through 50% in 2005 to 100% in 2008. Even though MSO is funded by payment for activity for all types of hospitals, the rate of pay differs between the public and the private hospitals. Those rates should converge 20 by 201821. For other activities not concerned by the payment for activity, long term stay and mental care, public hospitals and the private hospitals participating in the hospital public service are all theoretically funded by the Global Allowance (Dotation Globale). Other private hospitals whether for profit or not are funded by the National Quantified Goal (NQG or ONQ in French). From the ONDAM the national tariffs for payment for activity are revised, the functioning

21 Except for local hospital which are yet to use GHMs (Ministry of Health, 2010; 3.2).

58

Chapter 3

Institutional Settings and Data

allowance (given to hospitals on top of the national tariffs) and global allowances are also revised22,23. To summarise (Table 3.1) hospitals can be divided into different categories: public, private not for profit and private for profit; and the latter two categories are further distinguished by whether or not they participate in the public hospital service and this determines their non MSO funding. Table 3.1: Summary of the types of hospitals, participation in the public service and the types of funding Status of hospitals

Hospital public service

Funding except payment for activity24

Public hospitals

Yes

Global Allowance

Private not for profit

May participate

Mainly Global Allowance

Private for profit

May participate

Mainly National Quantified Goal

The descriptive statistics (see Tables 6.2 and 6.3, page 157) will give evidence of the non complete matching. Though in statistical terms it is a near complete separation of data which means that the three variables, status, participation to the public service and funding could not be included together in a regression.

3.2.6. Pay and employment This section describes the pay of employees within the public and the private hospital sectors. The definition of net wages differ between France and the UK. In both countries the gross wage is the wage before any deductions and is therefore not the wage that employees receive every month or week in their bank accounts. The net wage, in economics and in the UK, is the wage after tax and other statutory deductions. In the UK the net wage is the wage after payment of income tax and other statutory deductions such as national insurance. The net wage is the wage available for daily consumption and savings. In France, payments for social benefits (pensions, health insurance, unemployment benefits) are deducted from the gross wage. However, income tax is not deducted directly from the gross wages 22 From the patient perspective those different funding above do not make any difference to what they have to pay. As long as they are accepted as patients they pay what the social security do not reimburse. For out-patients visits, they may have to pay at the point of delivery the GP but they will get reimbursed two thirds by the social security and the totality by the complementary cover if they have one. 23 Another small funding system concern hospitals which are out-network, those hospitals ask patients to pay at the point of delivery, a very small part is then reimbursed by the social security. Network/out network are US terms referring to GPs who are in network (they have signed and agreement with a health insurance or HMO) and out network. The out network hospitals in France are these with no agreement with the social security to have their patients reimburses. Twenty six (over eight thousands observations for all hospitals in every year) hospitals like these ones are in the final dataset. 24 All status of hospitals have some procedures funded by the payment for activity, procedures in Medicine, Surgery and Obstetric.

59

Chapter 3

Institutional Settings and Data

by the employer. Each year every employee is required to fill in a form, around March-April, on which everyone declares how much income they received during the previous year. The tax office after checking the truthfulness of the statements then request payment of tax. So, it is not possible to know the net pay of employees until after year n+1. In France employees working in the public service pay lower social contributions than the private sector employees. They do not pay for unemployment benefits as they have long life jobs. Their contribution to social security is also smaller (Bartoli and Bras, 2007). The government is supposed25 to transfer directly certain sum to the social security. The state prefers to transfer the amount straight from its budget to the budget of the publicly managed body in charge of the social security. Therefore, net pay in France is not comparable across sectors. Two individuals working in the two different sectors would have to transform their gross pay in order to compare it. Gross pay is useful for a study that is interested in costs as it would be closer to what employers have to pay. This study is interested in the attractiveness of pay to employees, therefore, it seems more sensible to use the net pay. Moreover, Meurs & Edon (2007) used the “net” 26 pay, as they investigate the average difference in wages between the private and the public sectors. Aude & Raynaud (2010), on the other hand, used the gross wages as they were interested in hospital costs. Their study was within the larger debate around funding of hospitals. In public hospitals medical doctors are largely salaried, they may however have different status, those who are also employed as professors or lecturers 27 receive part of their salary from the education ministry. In local hospitals (public sector) and in private hospitals that do not participate in the hospital public service scheme, medical doctors are paid like general practitioners, they receive a fee for each consultation. In the private sector that participate in the public service they are salaried and employed under a private sector contract (Adaius et al., 2007; 9). Non medical staff working in public hospitals mainly have the status of civil servants, but hospitals can also hire staff under non statutory contracts. Private hospitals hire staff under the appropriate contracts for the private sector. They may be either, open ended contracts (Contrats à durée indéterminée) or fixed term contract (Contrats à durée déterminée).

25 The government sometimes do not transfer the sums which does not help to resolve the deficit of the social security. 26 As defined in France, gross pay minus social contributions but without deducting income tax. 27 Medics students are taught within large regional hospitals called Centre Hospitalier Universitaire (University Hospital Centre).

60

Chapter 3

Institutional Settings and Data

3.2.7. Description of collective agreements and public sector pay grid As already mentioned, one of the feature of the French health care system that makes it interesting to study is the existence of a non public hospital sector which provides direct alternative employments to nursing staff. The pay for two nursing groups working in the public hospital sector will be mapped with the pay of the respective nursing groups in the private hospital sector. This subsection describes the different pay agreements for employees in the three hospital sectors, public, private not for profit and private for profit. The following paragraphs describe the pay as it ought to be following the collective agreements. In public hospitals assistant nurses had three levels of pay in 2007. The minimum hourly wage was 9.56 € and climbed up to 14.06 €/hour in the last level which was attainable after around 20 years. Over a 35 years career for a hypothetical person who climbed all the ladders, this person would have had a 1.11% increase every year in his/her wage. The maximum increase over the period as a whole would be 47%. Registered nurses in public hospitals received 10.409 28 €/hour at the start of their career. They could get up to 16.256 €/hour after 21 years. There was a second level for registered nurses which could be attained after 10 years at the previous level minimum and achieving a certain grade which was attained on average after 12 years. A condition is attached to this, namely that at this higher level there should not be more than 34% of the registered nurses of the hospital (Direction de l’Hospitalisation et de l’Organisation des Soins, 2009). The minimum wage on this scale was 13.89 €/hour and the maximum was 18.047 €/hour. Over a 35 years career for a hypothetical person who climbed all the ladders, they would have had a 1.58% wage increase every year on average. The total maximum increase over the 35 years period was 73%. Wage grids for registered nurses in a speciality were different and above the regular registered nurse grid. Registered nurses specialised in surgery had a different grid going from 11.52 €/hour in the first point of the first level to 19.26 €/hour in the last point of the second level. To be a specialised registered nurse in surgery, nurses had to add to their initial training a 18 months speciality training. This training was available only after at least 2 years working as a regular nurse.

28 308 number of points and the point being 55.4113 in 2007 308*55.4113/1610=10.409€/hour. 1610 is the number of hours per year for a 35 hours week.

61

Chapter 3

Institutional Settings and Data

Night shifts in the public sector were between 9pm and 6am and were paid an extra 0.90 €/hour29,30. Sundays and bank holidays were paid the regular pay plus 5.61 31 €/hour of work32,33. Hospitals in the private sector have different agreements. These collective agreements are called conventions collectives34, they set the grid for the employees and they are negotiated between the employers grouped under a federation and the unions. Five main collective agreements existed for hospitals. Four of these concern not for profit hospitals (though some hospitals which are for profit have a not for profit collective agreement). Three of these collective agreements were easily accessible and covered the large majority of hospitals. Two which represented the majority of hospitals will be described (a summary for the third one is available in annex J, page 263). The main one for private not for profit hospitals was called FEHAP 35 and covered not for profit hospitals who have joined the so called Federation of not for profit hospital and private assistance premises (Federation des établissements hospitaliers et d'assistance privés à but non lucratif). The federation represented 30% of private hospitals and 70% of the not for profit private hospitals (Table 6.1, page 156). Staff covered by this collective agreement had a pay based on a basic rate of 10.92 36 €/hour for assistant nurses and 14.7537 €/hour for nurses. Tenure was rewarded at 1% for each year in the limit of 30% maximum38. Hospitals could give allowances on the basis of the tasks that the employee was doing. Since 2002, the collective agreement stated that the local employer could offer an award in order to adapt the pay to local situations 39. However the decentralised allowance was caped at 5% of the total sum of gross wages. The pay was set by a notional point, this point was then multiplied by the value agreed by the parties. For night shifts the hourly rate was multiplied by 1.03 points which results in 4.3€/hour at the time of the study. The night shift was between 9 pm and 6 am (Article A3.2.1 of the collective

29 Décret n°88-1084 du 30 novembre 1988 relatif à l’indemnité horaire pour travail normal de nuit et à la majoration pour travail intensif, 30 November 1988 (revised in 1992) 30 Arrêté du 20 avril 2001 fixant le taux de la majoration pour travail intensif, 20 April 2001 31 The décret says 44.89 for each 8 hours of work, anything below is a proportional share of the indemnity. 32 Décret n°92-7 du 2 janvier 1992 instituant une indemnité forfaitaire pour travail des dimanches et jours fériés, 2 January 1992 33 Arrêté du 16 novembre 2004 fixant le montant de l’indemnité forfaitaire pour travail des dimanches et jours fériés , 16 November 2004 34 http://fr.wikipedia.org/wiki/Convention_collective accessed the 7th of November 2011. 35 Convention collective nationale des établissements privés d’hospitalisation, de soins, de cure et de garde à but non lucratif du 31 octobre 1951, 31 October 1951 (Revised by Avenant n° 97-09 du 25 novembre 1997 BO conventions collectives 98-35) 36 Assistant nurses have 351 points times 4.1757 which was the value of the point at the end of 2005 351*4.1757 gives the monthly salary 351*4.1757*12/1610=10.924. 1610 is the number of hours worked in a year. 37 Registered nurses have 474 points times 4.1757 which was the value of the point at the end of 2005. 474*4.1757 gives the monthly salary, 474*4.1757*12/1610=14.752. 1610 is the number of hours worked in a year. 38 Managers can also get up 1% more each year up to 20% maximum. 39 Avenant n° 2002-01 2002-03-25 BO conventions collectives 2003-23, 25 March 2002

62

Chapter 3

Institutional Settings and Data

agreement, see footnote 35). Bank holidays and Sundays were paid on top of regular pay 6.43€/hour (Article A3.3 of the collective agreement, see footnote 35). Employees in the private for profit hospitals were under the private hospitals collective agreement40 (FHP). The basic wage was calculated according to a point which values increase over time. This point was multiplied by the grade the person was in. The value of the point in the middle of the period studied was 6.54 euros (1st April 2007), at the beginning of the period it was 6.28 (1st January 2006). This collective agreement distinguished two separate levels for both assistant nurses and registered nurses. At the start of their career assistant nurses were paid 8.58 € an hour for a 35 hours week. After more than 30 years they would receive 10.29€/hour. The second level correspond to a job where some low level qualification is necessary, assistant nurses were paid 9.26 €/hour at the beginning of the their career and 12.48 €/hour at the end. The first level for registered nurses was regular nurses, the second corresponded to nurses who had a qualification in a specialised theme (such as ophthalmology). The first (second) level base wage was 11.99 (13.02) €/hour, the end of career monthly wage was 16.18 (17.55) €/hour. For a hypothetical person with a career of 30 years who would have climbed all the ladders, the maximum increase that they would get is 45% for an assistant nurse (46% for a registered nurse) which results in a 1.22% (1.24%) average annual increase. The night shift was between 7pm and 8am. Employees received an allowance for night shifts of 10% of salary per hour worked 41. Sunday and bank holiday allowance equalled 0.4 points of each hour done (article 82 of the reference in footnote 40). Tables 3.2 and 3.3 summarise the pay for the different collective agreements along with the public wage grid. For assistant nurses (Table 3.2), the best rate of pay when starting to work was with hospitals covered by the FEHAP collective agreement (8.41€, net salary). The best average annual increase (1.22) was obtained with the private for profit collective agreement (FHP). The best allowances for night shifts and Sunday and bank holidays pay were obtained for workers covered by the FEHAP collective agreement (4.3%/hour for night shifts 6.43€/hour for Sunday/bank holiday shifts). The public sector had the best maximum increase.

40 Convention collective nationale de l’hospitalisation privée du 18 avril 2002, 18 April 2002 41 Those 10% allowance are available for employees having a night post, or for someone with a day post who has a night shift (working hours between 7pm and 8am) of at least 4 hours in a row.

63

Chapter 3

Institutional Settings and Data

Table 3.2: Pay of assistant nurses, summary of collective agreements Basic wage €/hour Gross

Basic wage €/hour Net

Annual increase (average in %)42

Maximum increase end career in %

Night shifts

Sunday, bank holiday shifts

Public

9.56

7.699

1.10

47

0.99 €/hour

5.61 €/hour

FEHAP

10.924

8.41

1.00

30

4.3 €/hour

6.43€/hour

FHP 8.58 6.6 1.22 45 10% /hour 2.62€/hour Source: Convention collective nationale des établissements privés d’hospitalisation, de soins, de cure et de garde à but non lucratif du 31 octobre 1951, 31 October 1951 (Revised by Avenant n° 97-09 du 25 novembre 1997 BO conventions collectives 98-35), Convention collective nationale de l’hospitalisation privée du 18 avril 2002, 18 April 2002 and (Direction de l’Hospitalisation et de l’Organisation des Soins, 2009).

For registered nurses (table 3.3), the starting wage was better for employees in hospitals covered by the FEHAP collective agreement, the maximum and average annual increase was better for the public sector employees (73% and 1.58%), the allowances for night shifts and Sunday/holiday shifts were better in hospitals covered by the FEHAP collective agreement. Table 3.3: Pay of registered nurses, summary of collective agreements Basic wage €/hour Gross

Basic wage €/hour Net

Annual increase (average in %)42

Maximum increase end career in %

Night shifts

Sunday, bank holiday shifts

Public

10.41

8.38

1.58

73

1 €/hour

5.61 €/hour

FEHAP

14.752

11.358

1

30

4.3 €/hour

6.43€/hour

43

FHP 11.99 9.23 1.24 46 10% /hour 2.62€/hour Source: Convention collective nationale des établissements privés d’hospitalisation, de soins, de cure et de garde à but non lucratif du 31 octobre 1951, 31 October 1951 (Revised by Avenant n° 97-09 du 25 novembre 1997 BO conventions collectives 98-35), Convention collective nationale de l’hospitalisation privée du 18 avril 2002, 18 April 2002 and (Direction de l’Hospitalisation et de l’Organisation des Soins 2009).

Public sector hospitals have some advantages, in terms of careers, that can attract staff. Statutory contracts are life times ones. A household with one member with such a contract will have less difficulties in getting a mortgage compared to a household with the same income without a contract in the public service. Public hospitals and the private hospitals participating in the public service also have research activities in which nurses might become involved. Recall that hospitals participating in the public service have to receive all patients at any time, this might be regarded as attractive or work as a deterrent with more varied activities weighted against less sociable hours. Not all collective agreements are covered above, the main ones are however covered (FHP and FEHAP). Descriptive statistics will show the share of hospitals by the different types of collective agreements. It will show that there is a near complete separation of hospitals by collective

42 Taking a fictive career of 30 years in the private sector, 35 years in the public for a person who climbs all the ladders. 43 Availability for local discretionary increases.

64

Chapter 3

Institutional Settings and Data

agreements and status (public, private not for profit and private for profit) and that FEHAP represented 70% of the private not for profit hospitals and FHP represented 90% of the private for profit hospitals. In conclusion it is possible to say that differences in pay agreements are observed. The empirical results will also show that there are differences in pay between different types of hospitals in the same area.

3.2.8. Geography For administrative purposes France is divided into communes, départments and régions. Communes are the smallest administrative areas. In 2007 there are 36,000 of these covering an average population of 1700 with a median of 400 inhabitants. There are 96 départements (including Corsica but excluding the overseas départements) each covering an average population of 647 000 inhabitants (median 530 000). There are twenty two régions (including Corsica) each covering 2,800,000 inhabitants on average (2,120,000 being the median). The départements are very different from each other. Paris is the only département which is also a commune. Paris has the highest density with 20775 inhabitants per square kilometre, then followed by all the other département in the région Île de France (which includes Paris). The first non Île de France département is the Rhône which the main commune is Lyon (density 514 inhabitants per square kilometre). The Lozère département has only 15 inhabitants per square kilometre in 2007.

3.2.9. Social Occupational Coding As for the analysis made for England, the pay for assistant and registered nurses will be mapped with a comparator group of workers in the non hospital private sector. On what grounds nursing groups will be mapped to other workers in the private sector? This subsection answers this question and compares the social occupational coding in France and the one in the UK. Assistant nurses and registered nurses are clearly identified in the Social Occupational Code and in the Professions et Catégories Socioprofessionnelles respectively for the UK and for France. In the UK the SOC code is constructed on the the kind of work performed and the content of the tasks (Office for National Statistics, 2008). For France the PCS code distinguishes individuals in the coding regarding the kind of work, the status (employee or self-employed), the number of people employed in the firm, the professional position the employee is assigned to in the collective

65

Chapter 3

Institutional Settings and Data

agreement, the economic activity of the firm and the post (Institut National de la Statistique et de l’Analyse Economique, 2012). In France it is possible to identify midwives from nurses, and within nurses those with a special qualification (such as surgical). Both codes have a set of digits to identify each occupation. The first digit represent the larger group to which the occupation belongs, each subsequent digit is a more refined group of occupations. At the one digit level, nurses in France and in the UK are in a very diverse groups of workers. The differences between the two countries are highlighted in this subsection. In France the SOC code is called Professions et Catégories Socioprofessionnelles (PCS), it was first created in 1982 by the National Institute of Statistics and Economic Studies (INSEE, Institut National de la Statistique et des Études Économiques) and revised in 2003 (which is used here). It categorises people within 8 groups: −

Farmers

−

Craftsmen, retailers, head of firms

−

Managers and intellectual professions (such as medics, lawyers, lecturers, professors)

−

Semi-Professionals (technicians, teachers, nurses)

−

White collars workers (clerks, secretaries, salesman, social workers, assistant nurses)

−

Blue collars workers

−

Pensioners44

−

Other people with no activity44

The groups for assistant nurses in the two countries both include Personal Service occupations. Sales assistants are in the same group in France but coded in the Group 7 in the UK SOC code. The French code also includes administrative and secretarial occupations which are in the Group 4: Administrative and secretarial occupations in the UK. Policeman and soldiers are with the group assistant nurses in France and are with registered nurses in the UK.

44 In the labour market data available for this study, the last two groups are not available. The data focuses on employees so by definition will not include people with not activity.

66

Chapter 3

Institutional Settings and Data

Table 3.4: Occupations which are not classified with the same groups as assistant nurses in France and UK Other groups not with assistant nurses in France and with assistant nurses in UK

Other groups not with assistant nurses in UK and with assistant nurses in France

Animal care service45

Sales Assistants, Administrative and secretarial occupations, policeman and soldiers

Registered nurses, in France, are included with secondary and primary school teachers, social workers, technicians-semi professionals. In the same professional group in England, social workers, teachers are not part of the same occupational code, they are part of the Group 2 which includes doctors. In the British SOC code which includes registered nurses we also find artistic and literary occupations, media and associate professionals all of which are coded in the group managers and intellectual professions in France. Table 3.5: Occupations which are not classified with the same groups as registered nurses in France and the UK Other groups not with registered nurses in France and with Other groups not with registered nurses in the UK and with registered nurses in the UK registered nurses in France Artistic and literary professions, media professions, policeman and soldiers.

Teachers46, social workers,

The groups are different in the two countries. Yet they may also reflect two different societies and thus both are showing a different reality in a different way. As for the analysis made for England, the nursing group working in the public hospital sector in France will be mapped with the workers employed in the private non hospital for profit sector 47 in the same PCS group.

3.3. Comparing the two institutional settings, why studying France is of interest France has some very interesting features that motivate the study of the impact of pay gaps on hospital staffing shortage and skill mix. First the number of employees covered by collective agreements is more widespread in France than in England (95% of the total employees in France compared to 35% in England), in consequence it is then expected that pay gaps of public sector pay minus private sector pay show less variations in France than in England. This thesis will then

45 Nothing explicit has been found regarding this occupation. 46 There are two categories of school teachers in France, certifié and agrégé, the last one is classified in the group intellectual professions. Agrégé teachers teach in secondary and high school and can teach undergraduates. Certified teachers will teach in secondary and high school and will have more hours and be less paid than agrégé for the same job. This is not possible to translate it properly as there is nothing similar in the USA or the UK. In the French education system, agrégation is the highest degree a prospective teacher can obtain. 47 For the rest of the thesis, this comparator group will be referred to workers in the private non hospital sector.

67

Chapter 3

Institutional Settings and Data

be able to conclude whether the competitiveness of pay in countries with less widespread pay variations will have an impact on the shortage of nursing staff and on the skill mix of hospitals. In France, nursing staff have two direct alternative employers: private not for profit and for profit hospitals. This is a distinct feature from England. In England, there are private hospitals but the number of private hospitals, though rising, is small at the time of study. In England it is expected that labour is more mobile than in France, nursing are expected to move more easily from one sector to the other. The comparator group of employees working in non hospitals private sector in both countries is expected to have similar preferences to nurses. Though, in England, it is also expected that this group is more of an alternative employer than for France. In Elliott et al. (2007), the authors seek an alternative employer for NHS staff. They use occupational groups of private employees for which some employees have a nursing qualification. They only choose women as nurses are 90% women. It is expected that employees within these occupational codes have similar preferences to nurses. This suggests that in England, nurses work outside of the sector employing nursing skills. Though there is no evidence for this not existing in a large scale for France, it is clearly not expected as employers in France rely more on the diplomas. Moreover, nurses can be employed as nurses outside hospitals: nursing homes, schools and self employed for France and in agencies for England. This thesis does not take that information into account. The results may not be robust to the introduction of this information. English hospitals are compensated for extra cost due to higher market forces factor, this is not the case for France. In England, hospitals in areas where the local labour market conditions are uncompetitive, receive a higher budget than hospitals in areas where there are no extra costs caused by the labour market conditions. With this extra funding, hospital trusts cannot increase the pay of their workers as the pay grid is rigid. It is expected that hospitals use more of agency staff and this is the topic of some future research.

3.4. English hospitals data This thesis tests for the impact of pay gaps between public hospitals and the private sector on shortages of staff in public hospitals and looks at the way hospitals may respond by changing staff-mix. Empirical analysis will be performed using data for English and French hospitals. Up to Chapter 7 the analyses will focus on English hospitals using English data. The English data was obtained from publicly available data from the Department of Health for the years 2003 to 2005.

68

Chapter 3

Institutional Settings and Data

Head counts and whole time equivalent data available for each of the years included in the study were for October of the preceding year. Vacancy rates are from March of 2003, 2004, 2005. The number of assistant nurses in each Hospital Trust was computed by aggregating the numbers of unqualified nurse and health care assistants employed in each Hospital Trust (see 3.1.4 page 53). The pattern of pay is mapped by the computation of Standardised Spatial Wage Differentials. How this is done will be explained in Section 4.1 (page 99). Assistant nurses and registered nurses are the two groups of staff analysed here. During the period studied some hospitals merged, where this happened observations were aggregated to the merged structure (see annex A, page 240, for the details of merging hospitals). Observations for “ambulance trusts” were dropped from the data set because these Trusts have a very different workforce structure and employ very few nurses. The data set comprised 230 hospital trusts in each of the three years resulting in 690 observations (see annex C page 248 for the flow chart). This section (3.4) describes the variables used in the analyses for England (chapters 6 and 7). Subsection 3.4.1 describes the variables used as dependent variables, Subsection 3.4.2 describes control variables. Pay data is described in the following chapter (4). Section 3.5 presents the French data and Section 3.6 compares hospital data for the two countries respectively.

3.4.1. Dependent variables for English analyses One focus of this thesis is about shortages of staff. Shortages can theoretically be measured by the number of staff that is desired minus the number of staff currently employed. However, one does not know the desired level of staff, administrative statistics such as the vacancy numbers can proxy this. Grumbach et al. (2001) propose different measures of shortage in hospitals, staff level, vacancy rates, turn over and reported levels of staff shortages. Grumbach et al. (2001) argue that there is no perfect definition of what should be the level of staff. However, they correlate vacancy, turnover and staff levels with reported shortages by managers for USA hospitals48 and find that there are weak correlations which, according to the authors, are interpreted as each measure revealing different aspects of the shortage of nurses. Thus different measures should be used to capture the different aspects of shortage. Vacancy rates measure the proportion of posts that are unfilled. As noted in Bach (2010), vacancy rates do

48 They used data from the Annual Survey of Hospitals carried out by the American Hospital Associations, and Nursing Personnel Survey from 1990 and 1992.

69

Chapter 3

Institutional Settings and Data

not say anything about the freezing of posts. Buchan (2002) argues that vacancy rates for nurses understate the number of actual vacancies because of at least two reasons: managers may fail to advertise the post because they have no expectations of a successful recruitment; or they might hire other less qualified people to do the job. Staff levels, imperfectly reflect shortage because they fail to capture posts which are frozen; hospital trusts with a larger share of frozen posts should, all else equal, have a lower staff size. Staff levels measure the relative size of the workforce, a higher level of staff should, all else equal, correspond to a lower shortage. For English data, vacancy rates and staff levels are used. Vacancy rates are measured by vacancy counts of hard to fill positions which have been advertised for more than 3 months. The number of vacancy counts is then divided by the total number of filled and unfilled posts. The distribution of vacancy rates across the different groups of staff are presented in Table 3.6. On average there was a vacancy rate of 1.2% for assistant nurses over the period 2003-2005 while for registered nurses it was 2.5%. More than half of the hospital trusts do not have any vacancies for assistant nurses while for registered nurses it is the case for only 17% of hospital trusts. Table 3.6: Vacancy rates distribution in English hospitals, pooled 2003-2005 (690 obs.) Vacancy rates Assistant nurses

Mean 0.012

SD 0.028

NB Null 53.00%

P30 0

P50 0

P70 0.007

P90 0.037

Registered Nurses

0.025

0.033

17.00%

0.004

0.013

0.025

0.068

In order to use correctly a measure such as staff levels it is necessary to control for hospital trusts activities. Some activities might be more labour intensive than others and therefore require higher staff levels. By relying on controls for the different types of activity it would be possible to argue that any other variations in staff levels reflect staff shortages which are impeding hospitals for providing the required level of services. This is, however, a very strong assumption stating that controls that are available for types of activity will control for all the activity performed in the hospital. There might be non measured activity and regression models usually control by adding controls one to the other. Equation 3.1 presents the definition of level of staff which is a ratio of the number of staff divided by the first component of a Principal Component Analysis (PCA) where this PCA is performed on four variables reflecting the size of hospital trusts: the number of beds and the total of whole time equivalent of assistant nurses, registered nurses and doctors (see Section 3.4.2 below and annex D page 249). It is expected that the number of staff carries information about the size of hospitals.

70

Chapter 3

Institutional Settings and Data

This thesis does not seek to explain size of hospitals. The information that is part of staff numbers and that this thesis tries to explain is the variation in staff numbers, between hospitals which are doing similar activities, that underlie a shortage of staff. The size dimension contained in the number of staff is removed by standardising it by another measure of size that would not contain the information about shortage. A PCA gathers the maximum of the variance between the variables to construct the first component. Therefore, if variables included in the PCA are underlying a similar dimension the first component should contain this information that relates only to this dimension. The higher the value of the first component, the higher the size of a hospital trust. Using the PCA allows to have a more reliable variable, the first component of the PCA captures what all four variables have in common, what differs go to the subsequent components. By dividing staff numbers by this component, only the variability in staff numbers that does not refer to size is kept. Equation 3.1: Definition of staff levels

Staff StaffLevel = Size j N j

N j

(3.1)

Where N is the whole time equivalent number of either assistant nurses or registered nurses in hospital j. Staff level is a measure of staff divided by a measure of size which has no unit, therefore the values for staff levels as defined above are not giving much information at the descriptive level. Therefore, descriptive statistics (Table 3.7) presents the number of staff per bed. On average there are 1.4 registered nurses per bed and 0.9 assistant nurses. Table 3.7: Staff numbers per bed in English hospitals, pooled 2003-2005 (690 obs.) Staff Levels

Mean

SD

Median

P10

P30

P70

P90

Assistant nurses

0.982

0.511

0.909

0.590

0.778

1.054

1.324

Registered Nurses

1.461

0.546

1.341

1.065

1.228

1.499

1.906

Hospitals facing shortage of staff may try to reduce the impact of such shortages. Variations of pay gaps are supposed to reflect variations in attractiveness of regions to assistant nurses or registered nurses. Staff-mix is supposed to reflect hospital trusts behaviours. Hospitals in regions where the pay of one of the nursing group of staff is not competitive may try to alter their staffmix in order to cope with the inability to recruit this group of staff (Section 2.5). Staff-mix is

71

Chapter 3

Institutional Settings and Data

defined as a proportion. The proportions of staff were computed using the whole time equivalent numbers for each group of staff in each hospital, see Equation 3.2. Equation 3.2: Definition of staff-mix

Staffmix RN j =

Staff RN j

(3.2)

RN Staff AN j Staff j

Where RN is the whole time equivalent number of registered nurses in hospital j and AN is the whole time equivalent of assistant nurses in hospital j. The distribution of registered nurses staff-mix in hospitals in England is presented in Table 3.8. On average, hospital trusts have 60% of registered nurses and 40% of assistant nurses. Table 3.8: Proportion of registered nurses in English hospitals, pooled 2003-2005 (690 obs.) Proportions

Mean

SD

Median

Min

Max

RN

0.601

0.091

0.6

0.294

0.887

3.4.2. Independent variables for English analyses, presentations and expected effects This thesis investigates the impact of pay gaps on staff shortages in hospitals and hospitals staffmix. Yet, geographical differences of shortages and staff-mix may be caused by other factors than just pay gaps. Therefore a range of variables were also used to control for other differences between hospitals. This subsection specifically present and explain those variables. a) Size of hospital trusts Size should have an impact on shortages. Though its effect are not clear cut from the literature (Tai et al., 1998; Coward et al., 1992; Holmås, 2002). Larger hospitals may offer a wider range of different tasks and challenges and so offer a more exciting working environment, making it easier to attract staff all things equal. However, more tasks might work as a deterrent as it can also increase stress levels. Size should have an impact on staff-mix as explained in 2.4.2 (page 36). Czuber-Dochan et al. (2006) argue that more specialised nurses should be present in bigger institutions. The aggregated data used here will not allow to distinguish between specialised and non specialised nurses. However, one effect of size could be a higher proportion of registered nurses compared to assistant nurses if bigger hospitals require more specialised skills. The size of hospital trust also 72

Chapter 3

Institutional Settings and Data

proxies the general funding and workload as described in paragraph 2.4.2 (page 36). Its effect depends on the form of the hospital trust production function. Table 3.9 gives the descriptive statistics for the number of beds. The mean is around 700 beds with a standard deviation of more than 400 beds. Table 3.9: Number of beds in hospital trusts (690 obs.) Number of beds

Mean

sd

Min

Max

P10

P30

P50

P70

P90

709.0

406.9

44.12

2669

246.9

485.2

619.0

847.4

1232

There are several measures of size available. There is no strong prior knowledge for selecting one over the others. The number of beds, the total number of whole time equivalent of assistant nurses, the total number of whole time equivalent of registered nurses and the total number of whole time equivalent of doctors, each contains some unique information but they also have in common information about size. A Principal Component Analysis (PCA) maximise the variance among a set of variables and provides weights associated to each variable. The linear combination of the product of these weights and the original variables gives the first component of the PCA (see annex D page 249). This first component should then gather all the information about the dimension that those variables have in common: size. PCA analysis is not an end in this thesis, its goal is to reduce the number of variables to put in the regressions. It is particularly recommended in this case as there are different variables which express the same idea (Everitt and Dunn, 2001). It is also used to define the staff level variable (see 3.4.1, above). The higher the value of the component, the higher the size of a hospital trust. Hospital trusts in England are large institutions, there are only 230 hospital trusts in the analyses. By comparison, in the French analyses, there are 900 public 49 hospitals in the dataset. On average, hospital trusts have four hospitals (table 3.11). In order to take this into account, the number of hospitals per hospital trusts is used. The effect of this variable on shortages and staff-mix is difficult to predict. The number of hospitals might matter as they may reflect the different types of organisations, therefore it would be expected that different way of organising care should lead to different staff-mix. The data on number of hospital per trust has been taken from the Binley's Management Directory book in the 2004-2005 edition 50 (Binley, 2004). Table 3.11 (page 75) gives the descriptive for the

49 In France, public hospitals are only a third of the total number of hospitals. 50 Therefore, this statistic is only available for one year, only 230 different observations are available.

73

Chapter 3

Institutional Settings and Data

number of hospitals in hospital trusts. The mean is 4.2 and the standard deviation is four hospitals. Table 3.10: Number of hospitals per hospital trusts (230 obs.) Number of hospitals

Mean

sd

Min

Max

P10

P30

P50

P70

P90

4.226

4.011

1

21

1

2

3

5

9.1

b) Institutional setting Benedict et al. (1989) give some evidence of the effect of the institution on the retention of staff. This article shows that teaching hospitals are more likely to retain staff. Buchan & Calman (2005, see Subsection 2.4.2, p. 36), have argued that differences in the types of organisations should drive differences in staff-mix. They argue that institutional settings may imply that different regulations apply. Three variables are used to proxy institutions: number of hospitals per hospital trust, dummies for the types of activity performed and having achieved the foundation status. Activities undertaken in different hospitals may imply different set of skills, for example, acute hospitals may look for staff with different skills than mental health hospitals. It is expected that shortages of staff differ by the different types of hospitals. The effect is unknown. For medical staff Elliott et al. (2006, p. 71) argued that the types of hospitals are included for two reasons “first to capture variations in speciality mix and second to distinguish teaching hospitals”. Teaching hospitals offer this opportunity to undertake research and therefore might be able to attract staff more easily (Benedict et al., 1989). Hospital trusts which undertake different activities may have different production functions which, in turn, may result in different possibilities for substitution/complementarity. Some hospital trusts may have more scope substituting registered nurses with assistant nurses than others. The variable was identified from Crilly et al. (2007) data51, where this data distinguished between Specialist, Acute, Teaching, Mental Health and Other Trusts. Other hospital trusts include children, social care and orthopaedic hospitals. Slightly more than half of all hospitals are acute hospitals (Table 3.11) and only around 5% are other or specialist hospitals.

51 Data from the Department of Health did not identify as many groups. In the DoH data Specialist hospital trusts are with “Others”. See page 75, below, where Crilly et al. provided another variable.

74

Chapter 3

Institutional Settings and Data

Table 3.11: Dummies for activity of hospital trusts (690 obs.) N

Proportion

165

0.239

Teaching

75

0.108

Acute

381

0.552

Other

33

0.047

Specialist

36

0.052

Mental

During the period studied foundation trusts (FT) were created. FTs were granted more freedom on setting pay, following a reform that started to be put in place in 2004 called Agenda for Change. FTs should have more scope to reduce shortages. Those hospital trusts which achieved the foundation status did so by meeting a set of criteria required to change status. The criteria included both financial and quality of care aspects of performance. The status was awarded by the government. This process of selection may hide some characteristics of hospital trusts that would not be captured otherwise. It is only at the end of the period studied here that the foundation status was awarded to some hospitals52. Hospital trusts which were awarded foundation status will be marked as foundation status hospital trusts in this study for all years preceding the award. Around 13% of hospitals are marked with foundation status 53 (table 3.12). The rationale for doing that, is that the foundation status was achieved by meeting some requirements, in consequence, hospitals which achieved that status early on may have had some specific features beforehand that allowed them to be among the first to be awarded the status. Table 3.12: Frequency and proportion of hospital trusts with foundation status (690 obs.) Foundation

N

Proportion

93

0.134

c) Complexity index A further variable, available for non-Mental hospital trusts only 54, measures the complexity of activities undertaken in each hospital trust. This variable was created by Crilly et. al (2007) who constructed a complexity index to measure the complexity of the activities undertaken by each hospital. They did this by summing the national average cost of the different Health Related 52 The Foundation status were awarded in 2004 and 2005; 20 hospitals achieved the status in 2004 and 12 in 2005. Only 11 out of 12 hospital trusts that were awarded the status in 2005 are in the data used in this thesis. 53 The data used here has information from 2003, a hospital which would get the foundation status in 2005 would also be coded as Foundation in 2004 and 2003. 54 Nine “Other” Hospital Trusts ( 27 observations) do not have a complexity index measure on top of the mental trusts.

75

Chapter 3

Institutional Settings and Data

Groups (HRGs) performed in each Hospital Trust and dividing this by the total number of Finished Episodes (FEs) for the Trust. The data was made available by the authors. Only 172 hospital trusts have an index, this corresponds to 516 out of the total of 690 observations in the dataset. Complexity would be expected to be determined by the technology employed and the technology as explained in paragraph 2.4.2 (page 36) is likely to alter staff-mix. It is expected that the more technically advanced the procedures, the more highly qualified the nursing workforce (Pope, Menke, 1990; Acemoglu & Finkelstein, 2008). Table 3.13 shows the distribution of the index. The effect of technology on shortage of staff is yet to be analysed, as there is not many articles investigating it. As Section 2.3 shows, the literature is quite clear cut on the drivers of shortage: at the individual level it is satisfaction at work that is the most important feature. Technology can be thought of having an impact on satisfaction as it may provide more exciting working environment but also more stressful. The impact of the complexity index on shortages could go both ways. On one hand, the more technical the procedures performed by hospitals, the more likely hospitals will look for more skilled workers, though it is likely that the shortage of more skilled workers would be more acute. Hospitals with a larger value for the complexity index will have a higher level of vacancy rates. On the other hand, hospitals performing more complex procedure may find it easier to attract staff. None of those effects were found in the literature. Table 3.13: Distribution of Complexity index (516 obs.) Complexity index

N

Mean

sd

Min

Max

P10

P30

P50

P70

P90

516

1.275

0.290

0.83

3.04

1.11

1.16

1.21

1.27

1.45

This section presented the dependent and independent variables that will be used in the analyses for English data. Those analyses will also use Standardised Spatial Wage Differentials, those are presented in the following chapter (4).

3.5. French hospital data This thesis is testing the impact of pay gaps on nursing staff shortages and further analysing how hospitals might adapt their behaviour as a result. France has three types of hospitals: public, private not for profit and private for profit. The goal is to test the impact of pay gaps between nurses working in public hospitals, and –

employees working in the private non hospital sector, 76

Chapter 3

Institutional Settings and Data

–

nurses working in private not for profit hospitals and

–

nurses working in private for profit hospitals

on public hospitals. This section describes the hospital data that will be used in chapters 8 and 9. The hospital data comes from the “Statistiques Annuelles des Etablissements de Santé” (SAE) which is a dataset available from the French ministry of health 55. This data gives information on health care premises (“Etablissement de santé”) in France. All health care premises have to complete the survey, it is a legal requirement, thus this data is exhaustive 56. The Ministry of Health survey all hospitals. Questionnaires are either sent to juridical legal entities or to institutions. The latter depend on juridical entities. A juridical entity in French law is a legal person. Juridical entities are responsible for what is happening inside each institution. Juridical entities are surveyed for public hospitals, except for a very marginal number of health institutions which do not depend on a health juridical entity (those health care premises would depend on the ministry of justice, city councils, conseil général57 or ministry of defence). Institutions are surveyed for private hospitals, except for some singular cases for which the juridical entity is survey: cancer treatment centres, dialysis institutions, and mental health hospitals. It is assumed that economic decisions are taken at the level surveyed. Only public hospitals are investigated in this document, information on private hospitals is presented in descriptive statistics and then summarised at the département level when used in regressions. Hospitals employing no assistant nurses or no registered nurses have been excluded from the data. Those which are not in all the years (2006-2008) for which the analysis is conducted have also been excluded. Overall there are 942 public hospitals in the final data in each of the three years of interest (2006 to 2008), 668 private not for profit hospitals and 910 private for profit hospitals. This gives a total of 2826 public hospitals, 2004 private not for profit hospitals and 2730 private for profit hospitals over the three years. The changes in the number of nurses over years have been calculated by using one more year of data (2005). Only hospitals which do not change status over time have been included (see annex I, page 262). Three types of variables will be used. The first type is the group of the workforce variables, nursing in hospitals can be measured with different variables. The second group comprises

55 http://www.sae-diffusion.sante.gouv.fr/ accessed the 14th of April 2012. 56 http://www.sante.gouv.fr/statistique-annuelle-des-etablissements-sae.html accessed the 14th of April 2012. 57 A conseil général is the elected public body that runs a département.

77

Chapter 3

Institutional Settings and Data

activity variables. These measure hospital activity. The final group of variables concerns labour market data and their construction will be explained in Section 4.4 (page 113). Before describing these variables, the descriptive statistics for the administrative characteristics of hospitals will be presented in 3.5.1, then descriptives of the workforce variables will be presented in 3.5.2 (page 79) and 3.5.3 (page 83). Activity variables are explained in 3.5.4 (page 84) and their descriptives are in 3.5.5 (page 90).

3.5.1. Administrative and juridical characteristics of hospitals Pay will be set differently in hospitals covered by different collective agreements. Table 3.14 shows that there is a near complete separation between the status of hospitals and the types of collective agreements. The distinction between private not for profit and private for profit hospitals can almost distinguish by which collective agreement hospitals are covered. Table 3.14 shows that more than seventy percent of private not for profit hospitals are covered by the FEHAP collective agreement, while 91% of private for profit are covered by the FHP agreement. Table 3.14: Collective agreements and status of hospitals Collective Agreements Other58

FEHAP

FNLCC59

Red Cross60

UCANSS61

FHP

Total

Private not for profit

180 (9%)

1423 (71%)

57 (3%)

54 (3%)

248 (12%)

42 (2%)

2004

Private for profit

240 (9%)

8 (0%)

3 (0%)

1 (0%)

0 (0%)

2478 (91%)

2730

The funding can differ between hospitals, table 3.15 presents the types of funding for long stay and psychiatry wards and the status of hospitals. Seventy five percent of private not for profit hospitals have the Global Allowance funding, while, except for nine hospitals, all private for profit hospitals have the National Quantified Goal. Medicine, Surgery and Obstetric funding is paid by activity62 (see para. 3.2.5 page 57 for details about funding). Funding differ by sector, therefore it can be expected that the strategy for hiring nurses would differ by the status of hospitals.

58 There is no information on those hospitals except that they have other types of collective agreements, this might mean that they have local agreements and are not attached to any large collective agreements. 59 Collective agreement for anti cancer treatment centres. 60 Collective agreement for Red Cross hospitals. 61 Collective agreement for the Social Security Staff. 62 A cost is identified for each procedure and this is paid to a hospital when it undertakes that procedure. The system is based on the “Groupes Homogènes de Maladie (GHM)” which is similar to the British Diagnosed Related Groups (DRG).

78

Chapter 3

Institutional Settings and Data

Table 3.15: Status of hospitals and the types of funding for long stay and mental care Public

Global Allowance

National Quantified Goal

2826

0

0

Private not for profit

0

1698

306

Private for profit

0

9

2721

Public

Table 3.16 presents the status of hospitals crossed with participation in the public service. It is possible to conclude that nearly all private hospitals participating in the public service are not for profit. However, hospitals that do not participate in the public service may not all be for profit hospitals. Table 3.16: Status of hospitals and participation in the public service Participate in the public service

Do not participate in the public service

Public

2826

0

Private not for profit

1446

558

60

2670

Private for profit

The descriptive statistics show a high degree of overlap between the different ways of classifying the three hospital status. For example the vast majority of private for profit hospitals are covered by the FHP agreement, are funded under the National Quantified Goal (for long stay and psychiatric wards) and do not participate in public service. Accordingly in the following analyses when categorising hospitals only one of the variables, collective agreement, participation in the public service, the type of funding or status will be used. Using all of them would give rise to issues in multicollinearity. Specifically the status is going to be used as the wage data is based on the status of hospitals and not on the collective agreements.

3.5.2. Descriptives of the workforce variables across status and year To introduce the reader to some basic statistics, Table 3.17 provides aggregated data at the national level. This data comes from Eco-Santé data 63 which provides readily available aggregated data on the French health care system. The overall proportion of registered nurses working in hospitals remained the same at 56.3% over the years 2006-2008 (Table 3.17). Over the period the national sum of both assistant nurses and registered nurses increased slightly by 1.25% for assistant nurses and by 1.29% for registered nurses. These results confirms those of Com-Ruelle et al. (2000) who found that in 2000, the proportion of assistant nurses and registered nurses in

63 www.ecosante.fr accessed on the 14th of April 2012.

79

Chapter 3

Institutional Settings and Data

France among the total of the health care professionals (including doctors) is the same, as shown here (Com-Ruelle et al., 2000). Table 3.17: Proportion and numbers of assistant nurses and registered nurses in French hospitals for 2006-2008 Year

AN

RN

Proportion of RN

2006

243669

313485

56.3

2007

245709

316027

56.3

2008

246737

317534

56.3

Variations 1.25% 1.29% Source: www.ecosante.fr based on the SAE données administratives - ministère chargé de la santé, DREES, 2006-2008.

The following tables will then use the data described in the introduction of this section. On average there are 110 registered nurses and 88 assistant nurses per hospital (Table 3.18). Public hospitals have on average more assistant nurses and registered nurses than private hospitals (230 registered nurses against 39 for hospital in both private sectors and 190 assistant nurses against 28 for private not for profit and 25 for private for profit). The dispersion is also greater for public hospitals. Private for profit and private not for profit have a similar size of workforce on average but not for profit hospitals have a higher dispersion. Table 3.18: Size of the workforce by year and status of hospitals year

Assistant nurses

Registered nurses

Average

SD

Average

SD

(all)

(all)

88.07

448.78

110.82

473.35

Public hospitals

(all)

190.58

721.53

230.81

756.07

2006

189.57

721.22

229.78

750.28

2007

190.64

721.77

231.11

759.93

2008

191.53

722.35

231.55

758.77

(all)

28.43

38.59

39.68

68.24

2006

27.77

38.82

38.80

67.54

2007

28.38

38.33

39.44

67.75

2008

29.16

38.67

40.79

69.49

(all)

25.79

26.46

38.90

43.51

2006

25.51

26.06

38.09

42.13

2007

25.85

26.58

38.85

43.32

2008

26.00

26.74

39.75

45.05

Private not for profit

Private for profit

Chapter 3 (Section 3.4.1 page 69) discussed the use of the Principal Component Analysis as a way to standardise the staff numbers. The measure of staff numbers will carry a size dimension which is removed by dividing it by a variable that captures the size dimension in a set of 6 variables.

80

Chapter 3

Institutional Settings and Data

Table 3.19 presents the nursing levels per unit of size as defined in Equation 3.3: Equation 3.3: Definition of Staff levels, French data

Staff Nj StaffLevel = Size j N j

(3.3)

Where N is either assistant nurses or registered nurses. “j” is the indexing letter for hospitals. Size is the first component of the Principal Component Analysis explained in 3.5.4 (page 84, tables 3.22 and 3.23), the higher the value on this variable the higher the number of beds for complete 64 and weekly stay at hospitals 65, number of places for ambulatory surgery 66, day care67, night care68 or at home care69. Staff is defined as the number of whole time equivalent employed by hospitals. A hospital with a staff level of “a” registered nurses does not have twice as less registered nurses than another hospital with a staff level of “2a” registered nurses. It is only possible to say that the second hospital has a larger standardised number of registered nurses than the first one. However, it can be noted from Table 3.19 that the number of nurses per unit of size (both assistant and registered) is larger in the public sector than in the two private sectors (65 to 23 in the private not for profit and 22 in the private for profit for registered nurses). The standard deviation is also larger for the public sector. The next largest is the standard deviation in the private not for profit sector.

64 65 66 67

Complete care is for heavy treatments that require the patient to stay at the hospital for a long period. Week care is for patients who need to come for less than 5 days, usually from Monday to Friday. Ambulatory surgery is for small surgeries that do not need patients to stay overnight. Day care is for patients who come in the morning and leave during the same day. They might come every day. They do not stay overnight. 68 Night care is for patients who need to stay at the hospital overnight but have a daily activity outside the hospital. 69 The at home care is a scheme for people in terminal phases of illness that would require to be at the hospital. Hospitals make the care that they would need available from home.

81

Chapter 3

Institutional Settings and Data

Table 3.19: Nursing levels per unit of size year

Registered nurses

Assistant nurses

Average

SD

Average

SD

(all)

(all)

38.57

45.93

35.19

37.14

Public hospitals

(all)

65.48

60.01

66.32

42.30

2006

66.61

62.19

66.81

44.05

2007

65.52

60.03

66.29

42.22

2008

64.30

57.76

65.87

40.58

(all)

23.00

28.84

18.61

18.08

2006

22.75

29.05

18.36

18.64

2007

22.88

28.48

18.60

17.91

2008

23.36

29.04

18.86

17.69

(all)

22.16

17.56

15.17

11.06

2006

22.10

17.46

15.23

11.19

2007

22.25

17.74

15.26

11.19

2008

22.14

17.50

15.02

10.80

Private not for profit

Private for profit

Table 3.20 presents average and standard deviation statistics for proportion of registered nurses in hospitals. The proportion of registered nurses is defined simply as follow: Equation 3.4: Definition of the proportion of registered nurses

Staffmix RN j =

Staff RN j RN Staff AN j Staff j

(3.4)

Where AN and RN are the whole time equivalent number of Assistant Nurses and Registered Nurses in hospital j . The grand mean of the proportion of registered nurses is at just above 50% which is lower than the proportion obtained through the Eco-Santé data reported in table 3.17 (page 80). The data from Eco-Santé also originate from the SAE data which has been used to compute Table 3.20. The difference in the two may come from the restriction that were applied to the data used in this thesis. Those restrictions meant that hospitals must be present in all years, do not change status, or of type of funding. Therefore, this sample might be biased compared to reality as the selection is definitively not at random. Public hospitals have the lowest proportion at 43% (against 50% in the private not for profit sector and 59% in the private for profit). The standard deviation is quite stable over the three status, it is around 0.17-0.18 which means that the average difference from the mean is around 17 to 18 percentage points.

82

Chapter 3

Institutional Settings and Data

Table 3.20: Proportion of registered nurses year

Proportion of registered nurses Average

SD

(all)

(all)

0.51

0.17

Public hospitals

(all)

0.43

0.17

2006

0.43

0.17

2007

0.43

0.17

2008

0.43

0.17

(all)

0.50

0.18

2006

0.50

0.18

2007

0.50

0.18

2008

0.50

0.18

(all)

0.59

0.13

2006

0.59

0.13

2007

0.59

0.13

2008

0.59

0.13

Private not for profit

Private for profit

The above reveals there are quite substantial differences between public and private hospitals on the variables presented here. There are also differences between private not for profit and private for profit hospitals. Private not for profit hospitals seem to have a greater variety of hospitals; the standard deviations for the number of staff and nursing levels are higher than for private for profit hospitals.

3.5.3. Detailed descriptives for public hospitals This section gives more detailed descriptives for just public hospitals on the variables that were presented above; they are given in Table 3.21. The average number of assistant nurses is lower than the average number of registered nurses, though the median is higher at 79 assistant nurses. The mean number standardised for size of registered nurses is approximatively the same as the mean number standardised for size of assistant nurses. The median is, however, different, with 50% of hospitals with less than 54 assistant nurses and 50% of hospitals have less than 41 registered nurses. The number of registered nurses varies greatly between the different public hospitals, 10% of the hospitals have less than 10 registered nurses, the median is 59, and the mean is 230. This shows that more than half of all public hospitals have fewer than twice the average number of registered nurses. Ten percent of hospitals have more than 379 assistant nurses and 523 registered nurses, so there are also some large ones.

83

Chapter 3

Institutional Settings and Data

A similar pattern (higher mean for registered nurses and lower median) is observed for the increase in the number of staff. On average, hospitals have increased the number of registered nurses by 1.6 per year, but half of them increased by less than 0.43. The corresponding figures for assistant nurses are 1.13 and less than 0.8. The growth in staff numbers, at both the mean and the median is higher for assistant nurses than it is for registered nurses. Finally, the mean proportion of registered nurses in the nursing workforce in public hospitals is 0.43. While the median is very close at 0.42. It is interesting to note that ten percent of hospitals have 23% or fewer registered nurses among the nursing workforce. Table 3.21: Detailed descriptive statistics for public hospitals70 Mean

SD

Median

P10

P30

P70

P90

Nb. of AN

190.58

721.53

79.40

27.40

47.35

149.82

379.05

Nb. of RN

230.81

756.07

59.70

10.25

19.32

202.95

523.16

AN per Size

65.69

41.50

54.87

22.51

37.75

80.61

122.86

RN per Size

65.00

59.73

41.05

9.72

17.59

92.19

149.82

Increase in Nb of AN

1.13

14.70

0.80

-6.87

-1.20

3.56

10.82

Increase in Nb of RN

1.60

19.13

0.43

-7.40

-1.00

2.40

11.77

Growth in Nb of AN

3.15%

39.28

1.03%

-5.70%

-1.44%

4.42%

11.62%

Growth in Nb of RN

2.58%

14.37

0.86%

-7.35%

-1.79%

4.43%

13.34%

0.43

0.17

0.42

0.23

0.30

0.53

0.65

Proportion

3.5.4. Activity variables, construction Tai et al. (1998) provide no evidence of an effect of size on shortage of staff. They say that the articles they review provide either evidence on a positive relationship (larger institutions have larger turnover) or negative or no relationship. Holmås (2002) on the other hand with good econometric work provides some evidence of smaller health care facilities in Norway being associated with less quitting from their nursing staff. It has been noted by Czuber-Dochan et al. (2006) that in ophthalmic units in the UK, specialised nurses are more likely to be present in bigger institutions. Though this result was obtained by the authors only with bivariate analysis it has some implications for the analyses reported here. More skilled workers may prefer bigger institutions so that they are confronted to more cases. It could also be that larger institutions undertake the more specialist procedures. Controlling for size is therefore critical when analysing staff recruitment/retention. A principal component analysis (PCA) is performed on a set of different variables thought as describing the

70 RN stands for Registered Nurses, AN stands for Assistant Nurses.

84

Chapter 3

Institutional Settings and Data

common factor size. As for English chapters PCA analysis is not an end in this chapter. PCA are performed throughout the thesis for two reasons, either to divide staff numbers by a component of size or to reduce the number of variables being used in the regression models: introducing the variables in regressions might lead to collinearity (Everitt and Dunn, 2001; 49). In SAE dataset under the name “Capacité d'accueil” there are several variables related to size. It gives the number of beds for complete71 and weekly stay at hospitals 72 (both with the number of beds on a daily basis for the total year, and the number of beds on the 31 st of December) and the number of places for ambulatory surgery73, day care74, night care75 or at home care76. The number of beds on the 31 st of December was not used in the analysis as it has a correlation coefficient of 0.99 with the total number of beds per day throughout the year for both complete stays and weekly ones. Using the six other variables a PCA was performed for public hospitals 77 in France. The resulting standard deviation and proportion of variance explained by each factor is described in table 3.2278. Each variable has been centred and scaled to 0 so that all variables have the same weight. A standard deviation value greater than one shows that the factor contains more information than one of the original variables. A way of selecting factors is to take only the factors which explain more variance than one of the original variables. In which case components with standard deviations above one would be selected (Everitt and Dunn, 2001; 53)79. On these criteria the first two factors are selected. Those two variables are then described in terms of correlations with the original six variables (Table 3.23). Table 3.22: Standard deviation and variance of the principal component analysis for size PC1

PC2

PC3

PC4

PC5

PC6

Standard deviation

1.86

1.09

0.84

0.62

0.45

0.27

Proportion of Variance

0.58

0.20

0.12

0.06

0.03

0.01

Cumulative Proportion

0.58

0.77

0.89

0.95

0.99

1.00

Table 3.23 shows the correlation between the original variables and the factors that are retained. A high correlation between a factor and one of the original variable shows that hospitals with 71 72 73 74 75 76 77 78 79

Complete care is for heavy treatments that require the patient to stay at the hospital for a long period. Week care is for patients who need to come for less than 5 days, usually from Monday to Friday. Ambulatory surgery is for small surgeries that do not need patients to stay overnight. Day care is for patients who come in the morning and leave during the same day. They might come every day. They do not stay overnight. Night care is for patients who need to stay at the hospital overnight but have a daily activity outside the hospital. The at home care is a scheme for people in terminal phases of illness that would require to be at the hospital. Hospitals make the care that they would need available from home. All PCA performed here were re performed for the whole sample of hospitals in order to get average of PCA components at the département level for non public hospitals. The principal component analysis performed in this document are done with the singular value decomposition technique with R software and the prcomp program. Other techniques to select components exist, this is one is preferred as it is simple and straightforward (Everitt & Dunn 2001).

85

Chapter 3

Institutional Settings and Data

large values on the original variable also have a large value on the component. The first component (or factor) is positively correlated with all the variables. This indicates that factor one gives a general indication of size, the higher the value of the component the higher the number of beds (complete care and week care), the number of ambulatory surgery, day care, night care and at home care places. The second factor, however, shows a new dimension of size, hospitals with a higher number of day and night care places have a large value on this factor and hospitals with a large number of places for ambulatory surgery have a small value on this factor. These two factors are orthogonal to each other and share a correlation close to 0. As a consequence both factors can be put in a regression analysis without the risk of introducing multicollinearity. Table 3.23: Correlation of retained factors with original variables, Principal Component Analysis for size Size1

Size2

Day beds complete care

0.95

-0.13

Day beds week care

0.90

-0.18

Ambulatory Surgery

0.55

-0.40

Day care

0.83

0.40

Night care

0.32

0.89

At home care

0.79

-0.14

Following this PCA on size, another was performed on occupancy rates. Holmås (2002) provides some evidence that nurses working in more demanding health care facilities are more likely to quit. Occupancy is one of way the author measured the more demanding working conditions. As seen above, the number of beds and places per hospital are available; the number of days and patients that have attended are also available in the data. Thus proxies of occupancy rates are calculated by dividing the number of day care by the number of day beds for complete and weekly care. The number of patients that have been treated is divided by the number of places available for ambulatory surgery, day care, night care and at home care. Then a PCA is performed on these six variables. The results are given in Table 3.24 and show that the first three components bring more information than one of the original variables and thus are retained. Table 3.24: Standard deviation and variance of the principal component analysis for occupancy rates PC1

PC2

PC3

PC4

PC5

PC6

Standard deviation

1.19

1.06

1.01

0.95

0.89

0.86

Proportion of Variance

0.23

0.19

0.17

0.15

0.13

0.12

Cumulative Proportion

0.23

0.42

0.59

0.74

0.88

1

86

Chapter 3

Institutional Settings and Data

Table 3.25 shows the correlations of the retained factors with the original variables. The first component describes hospitals with low occupancy rates on all variables except occupancy rate of day beds for complete care. In order to make it more straightforward the first component when used in regressions will be reversed. As changed, the higher the value a hospital has on this component the higher the occupancy on beds week care, ambulatory surgery, day care, night care and at home care. The second component contrasts hospitals with a high occupancy rate for beds for complete care and night care with those with a high occupancy rate in ambulatory surgery. The last component contrasts hospitals with high occupancy rates on beds for complete care, ambulatory surgery and day care as opposed to those with a high occupancy rate on at home care. Table 3.25: Correlation of retained factors with original variables, PCA for occupancy rates Occupancy rates 1

Occupancy rates 2

Occupancy rates 3

Beds complete care

0.12

-0.61

0.57

Beds week care

-0.67

-0.29

-0.02

Ambulatory Surgery

-0.36

0.69

0.32

Day care

-0.66

0.16

0.36

Night care

-0.47

-0.40

0.02

At home care

-0.41

-0.11

-0.69

Correlations of these factors with Size components were computed (Table 3.26), the first component of the occupancy rates PCA and the first component of the size PCA has a value of -0.40, which is high but not high enough to argue that the first component of the occupancy rates PCA is just another proxy for size. The correlation of the second component of the size PCA with the second component of occupancy PCA is -0.22 which is large enough to argue that there is some correlations but not large enough to dismiss the second component of the occupancy rate PCA. The third component is not correlated with any size components. The three Occupancy rates components will be introduced in regressions along with size components. Table 3.26: Correlation between PCAs of Size and Occupancy rates Size1

Size2

OccRate1

-0.39

-0.12

OccRate2

-0.04

-0.22

OccRate3

-0.08

-0.05

The type of care has been found by Benedict et al. (1989) to have an impact on the shortage of staff, teaching hospitals are associated with a better retention of staff. A certain mix or number of

87

Chapter 3

Institutional Settings and Data

staff might be needed for different types of activity undertaken at hospital. Mental care is supposed to require relatively less skilled staff while acute care is supposed to require more specialised staff. In France registered nurses in surgical theatres are supposed to have undertaken an extended training on top of the regular one to work in a surgical theatre. The following document, published by a regional (Midi-Pyrénées) Social and Sanitary board (Direction Régionale des Affaires Sanitaires et Sociales, DRASS), argues that there should be a minimum number of staff required per surgical theatre, one registered nurse (preferably, says the document, with the special training for surgical theatres) is necessary, however, the exact number should be set according to the acuteness of the operation (Commission de Coordination Régionale des Vigilances, 2007). It is possible to conclude that there is no strong obligations on the qualifications of staff per ward or surgical theatres. Though it is still important to control for different types of activity. Three dummies were constructed from the SAE dataset: hospitals with acute wards, long term stay wards or mental care wards. Some hospitals may have all three. From the data available for this study it is possible to know the number of specialised registered nurses for each hospital and the share of these among the total number of registered nurses. It is expected that the higher the share of specialised registered nurses the more technical the care will be. If it is more technical then more registered nurses (relatively to assistant nurses) are requested. The technology employed in hospitals in some studies has been found to be skilled intensive (Pope and Menke, 1990; Acemoglu and Finkelstein, 2008) and the more advanced the technology the greater the need for skilled nurses. The equipments with French hospitals is recorded in the SAE. Though record is exhaustive, putting all the variables in a regression model will lead to multicollinearity. The data records the number of scanners, MRI (Magnetic Resonance Imaging), gamma

cameras,

Positron

emission

tomography,

lithotriptor,

diagnostic

sonography

(ultrasonography), number of non digital radiography rooms, number of digital radiography rooms, number of vascular radiography rooms, number of electrophysiology rooms, number of coronary catheterisation rooms and number of rooms for functional explorations. A PCA is used as described in the previous paragraphs in order to reduce the scale of the data to just a few dimensions. Variables have been centred and scaled so that none has more weight than another one. Table 3.27 shows the standard deviation of all the factors that describe technology and the proportion of variance explained by each factor. A standard deviation that is above one shows

88

Chapter 3

Institutional Settings and Data

that the factor has more information than one of the single variable, thus only factors with a standard deviation above one are retained. Table 3.27: Principal component analysis on equipments, proxy for technology PC1

PC2

PC3

PC4

PC5

PC6

PC7

PC8

PC9

PC10

PC11

PC12

Standard deviation

2.17

1.15

1.07

0.92

0.91

0.82

0.75

0.74

0.69

0.58

0.54

0.44

Proportion of Variance

0.39

0.11

0.10

0.07

0.07

0.06

0.05

0.05

0.04

0.03

0.02

0.02

Cumulative Proportion

0.39

0.50

0.60

0.67

0.74

0.79

0.84

0.89

0.93

0.95

0.98

1.00

The first three components were used. Those three components incorporate 60% of the total variance. Table 3.28 shows the correlation between the three factors retained and the original variables. A high, positive correlation between a factor and one of the original variables shows that hospitals with high values on the original variable have a high value on the component. The first component ranks hospitals according to the number of equipments they have, the more equipments a hospital has the higher the value of the component for this hospital. The second factor contrasts hospitals with Gamma cameras (0.68) and positron emission tomography (0.75) with those with diagnostic sonography (-0.22), non digital radiography rooms (-0.32), electrophysiology rooms (-0.20) and functional explorations rooms (-0.28). The last component contrasts hospitals with diagnostic sonography rooms (0.32), non digital radiography rooms (0.44) and digital radiography rooms (0.22) with hospitals with lithotriptor rooms (-0.42), vascular radiography rooms (-0.24), electrophysiology rooms (-0.43) and coronary catheterization rooms (0.56). With the three components it is possible to argue that technology is not linear and that hospitals have different amount of technology (first component). Different types of technology are captured by the second and third components as they contrast hospitals. Some hospitals have more of one type of equipment while others will have more of another type.

89

Chapter 3

Institutional Settings and Data

Table 3.28: Principal components analysis for equipments, correlations with original variables EQ1

EQ2

EQ3

Scanners

0.81

-0.04

0.17

MRI

0.78

-0.02

0.03

Gamma cameras

0.58

0.68

0.10

Positron emission tomography

0.46

0.75

0.13

Lithotriptor

0.38

0.08

-0.42

Diagnostic sonography

0.70

-0.22

0.32

Non digital radiography rooms

0.45

-0.32

0.44

Digital radiography rooms

0.75

-0.08

0.22

Vascular radiography

0.68

-0.07

-0.24

Electrophysiology

0.61

-0.20

-0.43

Coronary catheterization

0.55

-0.04

-0.56

Functional explorations

0.61

-0.28

0.06

Larger hospitals are likely to employ a wider range and a larger number of equipments. In order to check that the first principal component of equipment is not just another proxy of size, Table 3.29 presents correlations of the PCAs for components of size and equipments. There is a correlation between the first component of the size PCA and the equipment PCA but it is weak (0.15). The first component of the equipment PCA has some different features than just size. The second and third components of the equipment PCA are hardly correlated with size components. Thus these three components of the equipment PCA are retained. Table 3.29: Correlation between PCAs of Size and Equipments Size1

Size2

EQ1

0.15

-0.08

EQ2

-0.01

-0.04

EQ3

-0.01

0.08

3.5.5. Activity variables, descriptive statistics Private and public hospitals do not perform the same activities, public hospitals specialise in more difficult cases. For example, while 64% of obstetrics is performed by public hospitals 74% of all complicated births are performed by public hospitals (Audric and Buisson, 2005). This implies that even though public and private hospitals might appear to employ the same groups of staff, the working environment is different and the skills required are different. Public hospitals in large regional premises perform research and education, operate 24 hours, 7 days a week, while private for profit hospitals are open during the day mainly for non complicated elective care. As a consequence, the activities that hospitals undertake are likely to affect recruitment/retention in 90

Chapter 3

Institutional Settings and Data

public hospitals. The working environment in the private hospitals which operate in the same area as public hospitals might be expected to have an impact on recruitment of staff by public hospitals. Table 3.30 presents the differences between hospitals in the activities they undertake. A larger proportion of public hospitals have medicine, surgery and obstetric and long stay activities (72% and 85% respectively; than either private not for profit hospitals at 29% and 67% and private for profit hospitals at 54% and 33% respectively). Twenty two percent of public hospitals have a psychiatric activity, while only 7% of not for profit hospitals and 14% of for profit hospitals undertake this activity. Table 3.30: Proportion of hospitals with Medicine, Surgery and Obstetric activity, psychiatric activity, and long stay activity by status of hospitals year

Medicine, Surgery and Obstetric

Psychiatric

Long stay

Average

SD

Average

SD

Average

SD

(all)

(all)

0.54

0.50

0.15

0.36

0.61

0.49

Public hospitals

(all)

0.72

0.45

0.22

0.41

0.85

0.36

2006

0.72

0.45

0.22

0.41

0.85

0.36

2007

0.72

0.45

0.22

0.41

0.85

0.36

2008

0.72

0.45

0.22

0.41

0.85

0.36

(all)

0.29

0.45

0.07

0.26

0.67

0.47

2006

0.29

0.45

0.07

0.26

0.67

0.47

2007

0.29

0.45

0.07

0.26

0.67

0.47

2008

0.29

0.45

0.07

0.26

0.67

0.47

(all)

0.54

0.50

0.14

0.34

0.33

0.47

2006

0.54

0.50

0.14

0.34

0.33

0.47

2007

0.54

0.50

0.14

0.34

0.33

0.47

2008

0.54

0.50

0.14

0.34

0.33

0.47

Private not for profit

Private for profit

The proportion of specialised nurses (Table 3.31) is small around 5-6% and on average very similar across hospitals of different status. However the standard deviations are large, and larger for private not for profit than either private for profit or public hospitals.

91

Chapter 3

Institutional Settings and Data

Table 3.31: Proportion of specialised nurses by the status of hospitals year

Proportion of specialised nurses Average

SD

(all)

(all)

0.06

0.15

Public hospitals

(all)

0.05

0.10

2006

0.05

0.10

2007

0.05

0.10

2008

0.05

0.10

(all)

0.06

0.19

2006

0.06

0.19

2007

0.06

0.19

2008

0.06

0.18

(all)

0.06

0.16

2006

0.06

0.15

2007

0.06

0.15

2008

0.07

0.17

Private not for profit

Private for profit

The following tables constitute a series of descriptive tables for each of the PCA components. All components have a 0 mean, due to the scaling. This means that hospitals of a certain type which on average have a negative mean will have scored low on the original variables. Moreover, components are constructed to be orthogonal to the previous component, this means that what the first component has captured is not present in subsequent components. Therefore, hospitals can have high values on all the original variables of the PCA and, relatively to other hospitals, have small values on some of the original variables. Table 3.32 presents the average and standard deviations for the size components of the principal component analysis (explained in 3.5.4, page 84, Tables 3.22 and 3.23). Size 1 is indexing hospitals according to how many beds they have for complete80 and weekly stay at hospitals 81, number of places for ambulatory surgery82, day care83, night care84 or at home care85. Size 2 is indexing hospitals with higher values for those which have a larger number of places for either night care or day care and small number of places for ambulatory care. Public hospitals are larger on both indexes than any of the two other sectors. Private for profit hospitals have, on average, a negative value on the second component which shows that private for profit hospitals are relatively large

80 81 82 83 84 85

Complete care is for heavy treatments that require the patient to stay at the hospital for a long period. Week care is for patients who need to come for less than 5 days, usually from Monday to Friday. Ambulatory surgery is for small surgeries that do not need patients to stay overnight. Day care is for patients who come in the morning and leave during the same day. They do not stay overnight. The night care is for patients who need to stay at the hospital overnight but have a daily activity outside the hospital. The at home care is a scheme for people in terminal phases of illness that would require to be at the hospital. Hospitals make the care that they would need available from home.

92

Chapter 3

Institutional Settings and Data

on ambulatory care but that they are smaller than private not for profit hospitals for night or day care. Table 3.32: Average size of hospitals, based on PCA components year

Size 1

Size 2

Average

SD

Average

SD

(all)

(all)

0.00

1.86

0.00

1.09

Public hospitals

(all)

0.37

2.94

0.31

1.53

2006

0.34

2.87

0.31

1.49

2007

0.38

2.93

0.33

1.54

2008

0.41

3.01

0.31

1.57

(all)

-0.27

0.54

0.00

0.74

2006

-0.28

0.54

0.00

0.74

2007

-0.27

0.54

-0.01

0.74

2008

-0.26

0.56

0.00

0.75

(all)

-0.19

0.42

-0.32

0.47

2006

-0.21

0.39

-0.31

0.44

2007

-0.19

0.41

-0.32

0.46

2008

-0.17

0.46

-0.34

0.51

Private not for profit

Private for profit

Table 3.33 presents the means and standard deviations of the occupancy rates components. Public hospitals have large values on the first component. As this component indexes hospitals by the occupancy on beds for week care, ambulatory surgery, day care, night care and at home care, this means that public hospitals tend to have a high rate of occupancy on those variables. On the other hand private for profit and not for profit hospitals have on average negative values on this component, which means that on average, private hospitals have low occupancy rates on all but day care beds. The second component shows that private for profit hospitals have large values meaning that they tend to have high occupancy rates for ambulatory surgery and a low one on complete and night care. Public hospitals have high occupancy rates on complete and night care and low occupancy for beds for ambulatory care. Private not for profit hospitals have negative values on average in the last component, showing that they have a high occupancy rate on at home care and low occupancy rates on beds for complete care, ambulatory surgery and day care. Private for profit hospitals have opposite values, thus they have on average high occupancy rates on complete care, ambulatory surgery and day care and low occupancy rates on at home care.

93

Chapter 3

Institutional Settings and Data

Table 3.33: Average occupancy rates of hospitals, based on PCA components year

Occupancy Rate 1 Average

SD

Occupancy Rate 2 Average

SD

Occupancy Rate 3 Average

SD

(all)

(all)

0.00

1.19

0.00

1.06

0.00

1.01

Public hospitals

(all)

0.23

1.43

-0.40

0.76

-0.08

0.81

2006

0.20

1.35

-0.41

0.82

-0.05

0.76

2007

0.23

1.46

-0.39

0.72

-0.07

0.80

2008

0.26

1.48

-0.39

0.73

-0.11

0.88

(all)

-0.18

1.13

-0.14

1.20

-0.22

1.48

2006

-0.19

0.85

-0.17

1.86

-0.12

1.96

2007

-0.16

1.38

-0.14

0.71

-0.26

1.22

2008

-0.17

1.12

-0.12

0.62

-0.27

1.10

(all)

-0.11

0.86

0.51

1.00

0.24

0.68

2006

-0.13

0.62

0.55

1.01

0.25

0.55

2007

-0.09

0.94

0.51

1.01

0.24

0.72

2008

-0.09

0.98

0.48

1.00

0.22

0.75

Private not for profit

Private for profit

Table 3.34 presents the means and standard deviations of the PCA for the different types of equipments. Hospitals with the higher level of equipments are found on average to be public hospitals, while private (not for profit and for profit) do not have many equipments. This being taken into account, private not for profit hospitals have higher values on the second component than private for profit hospitals which in turn, have higher values than public hospitals. On average, private for profit hospitals have a positive value on this component while public sector hospitals have a negative one, showing that on this component public hospitals tend to have relatively lower numbers of the equipments which this component is positively associated with and larger numbers of the equipments this component is negatively associated with 86. The third component shows the public sector hospitals have, on average, a large value on this index. Private not for profit hospitals have an average above 0, meaning that these hospitals have on average a larger value than the grand mean. Private for profit hospitals are at the lower end on this component87.

86 The second factor contrasts hospitals with Gamma cameras (0.68) and positron emission tomography (0.75) with those with diagnostic sonography (-0.22), non digital radiography rooms (-0.32), electrophysiology (-0.20) and functional explorations (0.28). 87 The last component contrasts diagnostic sonography (0.32), non digital radiography rooms (0.44) and digital radiography rooms (0.22) with Lithotriptor (-0.42), vascular radiography (-0.24), electrophysiology (-0.43) and coronary catheterization (0.56).

94

Chapter 3

Institutional Settings and Data

Table 3.34: Average equipments of hospitals, based on PCA components year

EQ1

EQ2

Average

SD

EQ3

Average

SD

Average

SD

(all)

(all)

0.00

2.17

0.00

1.15

0.00

1.07

Public hospitals

(all)

0.44

2.67

-0.19

1.11

0.24

0.91

2006

0.41

2.65

-0.22

1.12

0.26

0.89

2007

0.45

2.67

-0.20

1.12

0.24

0.92

2008

0.47

2.69

-0.16

1.09

0.23

0.93

(all)

-0.26

1.88

0.15

1.49

0.09

1.05

2006

-0.28

1.76

0.14

1.41

0.08

0.96

2007

-0.25

1.97

0.15

1.55

0.10

1.12

2008

-0.24

1.90

0.17

1.52

0.09

1.05

(all)

-0.27

1.68

0.09

0.82

-0.32

1.16

2006

-0.34

1.56

0.08

0.80

-0.27

1.04

2007

-0.28

1.67

0.08

0.76

-0.33

1.27

2008

-0.18

1.79

0.10

0.90

-0.35

1.16

Private not for profit

Private for profit

This section has presented data on hospitals activity, size, occupancy rates and equipments. It was shown that there are differences between the different types of hospitals in all these respects. Furthermore, it has shown that variations in staffing levels and nursing staff-mix exist among public hospitals. The goals of this thesis is to test for impact of the external labour market, as captured by pay gaps, on public hospitals staffing levels and skill-mix. The following chapter will explain how are created the Standardised Spatial Wage Differentials for the English and French data in order to compute these pay gaps. The next section will compare the data sets used in both countries.

3.6. Conclusion This chapter presented the institutional setting and the data of the two countries investigated in this thesis. It showed that there are differences between the two countries. In England, the pay setting for the private sector is a market rate. In France, the pay setting for all sectors of the economy is made of different collective agreements. This major difference between the two countries implies that the assumption of clearing labour market for France cannot be made. The implication in terms of results is that it is expected that the pay vary less for France than for England and that in consequence, the role of the competitiveness of pay in shortage and skill mix for France is expected to be of a lesser magnitude. Unfortunately, due to the construction of the

95

Chapter 3

Institutional Settings and Data

variables for staff levels88 and the fact that the competitiveness of pay is used differently in staff mix models in England and France, the magnitude of the impact cannot be tested. France differ from England by the size of the number of private hospitals. In France, nurses have direct alternative employers in the same sector as public hospitals. In England, it is likely that nurses can get hired more easily than in France by the non hospitals private sector 89. Though, in this thesis there is no information about the direct alternative employers nurses may have in England. By using the data for France, this thesis will be able to provide a test of whether the pay competitiveness of nurses in the public sector compared to nurses in the hospital private sector has an impact on the attraction of staff and the staff-mix in public hospitals. Compared to previous research, this thesis will investigate a group of staff which has never been investigated before: assistant nurses. In previous research, such as in Propper and Van Reenen (2010) the pay for assistant nurses in England was described as not being covered by a pay review body, this chapter showed that assistant nurses are covered by the Nurses Pay Review Body. In consequence, for England, the same assumption as for registered nurses can be made, that the gap between the pay for assistant nurses and the pay for the private sector counterpart should exhibit an indirect cost that hospitals have to face in order to recruit staff. Data for this thesis combines different datasets from different sources. The construction of the data is then peculiar. A summary is provided here and at the end of Chapter 4. For both countries, data for hospitals is an administrative data provided by the Department and Ministry of Health respectively for England and France. The data for England is smaller and considers only 230 hospital trusts while there around 900 public hospitals in France. Any drop in the number of observations for the English hospital results in speculation on the power of results. There is one variable which is only available for some hospitals (non mental hospital trust) in England and this will reduce the power of the estimations. Results with the sub sample of hospitals may not show any significant results (or less of significant results). When compared with the results obtained with the full sample, the discussion of this outcome will imply considering the fact that there is a lack of power to be able to find results significant. This will not be the case for France. For both countries data at the hospital level provides information on shortage and number of whole time equivalent staff for each year in the data (2003-2005 for England and 2006-2008 for

88 Using PCA in the denominator of the definition of staff levels forbid any comparison of values across countries. 89 To date, the candidate has no evidence for this, though it is common to think about France and the UK in this way: in France, employers gives more importance to the type of diploma and what school (Grande Ecole) the potential employee did, in the UK, employers are likely to give more importance to the career path of the potential employee.

96

Chapter 3

Institutional Settings and Data

France). In the empirical analysis, shortage and skill mix vary over time. Though, for England and France, institutional settings do not vary over time, hospitals which provide acute care in 2003 provide acute care in 2005. The number of hospitals per hospital trusts in England does not vary over time as this was taken from a one year book directory from the NHS (Binley, 2004). Most of the hospital observations vary over time except those just mentioned. Chapter 4 will explain that the geographical parameters for wages do not vary over time. The wage data for France is of a similar quality to the one for England. Both are filled by employers. Though, the data for France concerns the total number of employees, for England, the data concerns “only” one percent of the total number of employees and there might be some areas in England for which there are approximatively less than 100 nurses, in the data these areas may not have enough, if any, observations. This will reduce the number of areas for which a local labour market competitiveness can be computed. Finally, this thesis will analyse the interdependence of the labour markets of assistant nurses and registered nurses. This is a contribution of this thesis to analyse the effect of the labour market of one nursing staff on the shortage of the other group of staff. As presented in Section 2.5.1, paragraph ii, the underlying idea is that the labour market of one nursing staff is an indicator of some unobserved characteristics of the workplace. For registered nurses, a non competitive pay for assistant nurses might indicate more strenuous working conditions due to lack of staffing. On the other hand, for assistant nurses, a non competitive pay for registered nurses might indicate a poorer prospect in terms of career paths as they may want to become registered nurses themselves.

97

Chapter 4 Creation of Standardised Spatial Wage Differentials In the UK the nurses union covers the whole of the UK (including England) and they are paid on a national salary scale that exhibits little spatial variation (see Section 3.1 for more details on UK pay setting). In contrast the private sector pay is likely to exhibit greater spatial variation. Differences between the public sector pay and the private sector pay are proxied by differences in Standardised Spatial Wage Differentials (SSWDs). Variations in SSWDs proxy geographical variations in pay competitiveness. SSWDs will be used throughout the thesis, however the technique is taken from the literature (Elliott et al., 2006) with no additional specific research done on it. Sections 4.1 and 4.2 explain the creation of SSWDs. Section 4.3 gives the definition of the gap between the public sector SSWD and the private sector one. This latter section will also gives some descriptive statistics. One of the main attraction for using French data is that unlike for England the pay for the private non hospital sector is also regulated and covered by collective agreements. Still some variations are expected but France may offer a setting in which it is possible to test whether the same test lead to very different results. It is likely that the dispersion of pay will be less in France than in England but these remaining variations may still have tremendous impact on the shortage of staff. Section 4.4 will presents the wage data for France and how SSWDs are computed. In Subsection 4.5.1 the two countries, regarding the wage data and SSWDs, are compared.

4.1. Creation of Standardised Spatial Wage Differentials for England The central hypothesis that the empirical chapters investigate is the impact of pay gaps on staff shortages and staff-mix. The geographical pattern of pay in the public sector should exhibit a much flatter distribution than in the private sector as explained in Section 2.1.3 (page 14). Even though average pay may be the same in the two sectors, the private sector pay structure, for the reasons explained in Chapter 2, seems likely to exhibit greater dispersion than the public sector.

99

Chapter 4

Creation of Standardised Spatial Wage Differentials

Differences in wages between the public and the private sectors are likely to result in difficulties to attract and retain staff and thus shortages of assistant nurses and registered nurses staff in those areas where public pay is not competitive. Comparing raw wage averages between the two sectors would be too crude. Wages differ between different geographical areas due to differences in workers characteristics, industries and occupations. Therefore pay is standardised to take account of these differences. Standardised measures of pay can be calculated by computing Standardised Spatial Wage Differentials (SSWDs). SSWDs for the private sector are assumed to represent the equilibrium on the local labour market: it is the standardised pay that should be offered to an employee to attract her/him to work in this area. SSWDs are standardised for differences in the industrial and workforce composition of the areas. Thus the difference in SSWDs between the public and the private sectors represent a distortion of the market rate. In areas where the public sector under pay, it is likely that the public sector will incur indirect costs such as shortages of staff. In areas where the public sector pay is above the private sector rate, it is expected that the public sector would not incur such costs 1. The following subsections explain the estimation procedures (Subsection 4.1.1), the relation between SSWDs and Cost-Of-Living Supplements (Subsection 4.1.2, page 103) and descriptive statistics (Subsection 4.1.3, page 103). The last section will then explain how some missing SSWDs were recoded (Section 4.2, page 106).

4.1.1. Empirical model to estimate SSWDs The data used to compute the SSWDs is the the Annual Survey of Hours and Earnings (ASHE). This data is at the individual level 2. The survey is run every year, and the years used are from 2003 to 2005. This data is 1% of the total labour workforce of the UK and gives accurate information on wages as employers fill the survey. As stated above (see introduction of Chapter 3, page 49) the technique used to calculate SSWDs has been used before and is taken from the literature (Elliott et al, 2006). Local labour markets need to be identified. Elliott et al. (2006) have already estimated SSWDs for the whole of the UK and they compared different geographical boundaries (Elliott, 2006, Chapter 4 page 34). They compared Local Authority Districts (LADs), Primary Care Trusts (PCTs) and Travel To Work Areas (TTWAs) to distinguish which of these geographical descriptions best explained the variations in the Market Forces Factors (MFF). MFF are SSWDs calculated for the 1 2

In those areas where the private sector pay is below the public sector pay, the private sector might have some difficulties in recruiting staff. Though this is not studied in this thesis. Accessed from the Virtual Micro Laboratory (VML) at the Office of National Statistics in London

100

Chapter 4

Creation of Standardised Spatial Wage Differentials

private sector for the whole of the UK 3. Based on goodness of fit statistics TTWAs were excluded. LADs were then preferred to PCTs based on the stability over time of LADs boundaries. Hospital trusts are associated to the LAD which hosts the headquarters of the trust. There were 408 LADs in England in 2003-2005, and theses were used to define SSWDs. The previous chapter described in details the theory underpinning the use of SSWDs. The private sector local labour market is supposed to clear. Therefore the rate of pay that is offered to workers in the private sector is assumed to reveal what is requested to compensate for differences in cost-of-living and amenities. SSWDs calculated for the private sector should reveal those differences. This pattern of pay is then compared to the public sector pay of two nursing groups: assistant nurses and registered nurses. When mapped to public sector SSWDs the gap between the two should reveal how much more competitive is the public sector pay in one area compared to another. The SSWDs for registered nurses are available from Elliott et al. (2006) and are thus in the public domain. The SSWDs for assistant nurses was calculated for this thesis using the same methods (Elliott et al. 2006) with data available from the Virtual Microdata Laboratory (VML) 4. Because the Office for National Statistics (the body in charge of the data) updates the data after the initial release to incorporate late returns, SSWDs for registered nurses were also re computed using the VML data. It also provides a benchmark to ensure the calculation of new SSWDs for assistant nurses was based on correct coding. Correlations between the SSWD for registered nurses in the public domain (Elliott et al. 2006) and the one calculated here were above 97% (see paragraph 4.2.2, page 107). Only those LADs with observations above ten were used. With this restriction and the fact that the data is a 1% sample of the employees in the UK, all LADs for private sector SSWDs could be used but, for SSWDs computed for assistant nurses and registered nurses, not all LADs had enough or even any observations. Subsection 4.2 (page 106) will recode some of those missing LADs. The two nursing SSWDs can be mapped with a General Labour Market SSWD that use observations for workers in any occupation within the private sector. This is the straightforward gap that follows the assumption that the local labour private sector markets are clearing. An alternative can be to map the public sector pay with workers sharing similar occupations. It is then 3 4

Compared to the SSWDs for the private sector that are calculated in this thesis, the MFF are calculated for the whole of the UK, they are smoothed across areas in order to mitigate the cliff edges that exists between areas and they introduce an adjustment for higher responsibility using the Labour Force Survey. The VML is a facility available from the Office for National Statistics (ONS) that allow academics to use restrictive sensitive data held by the ONS.

101

Chapter 4

Creation of Standardised Spatial Wage Differentials

hypothesised that nurses and workers in similar occupations in the private sector share the same utility function. Both groups, nurses and their private sector counterparts, are sensitive to similar levels of cost-of-living and amenities. The gap would more accurately reveal what is requested to compensate for differences between areas. It is obviously difficult to decide which occupations in the private sector are likely to reflect nursing utility. The Standard Occupational Classification rank workers based on the kind of work performed and the content performance of the tasks and duties. The SOC 2000 groups workers in 9 major groups, 25 sub major groups, 81 minor groups and 353 unit groups (Office for National Statistics, 2008). Registered nurses are part of the SOC group 3 and assistant nurses are part of the SOC group 6. Two SSWDs on private sector workers in group 3 and 6 are computed and two for nursing staff: assistant nurses and registered nurses. Using data from the Annual Survey of Hours and Earnings (ASHE) for the years 2003-2005 the following models (Equation 4.1) were estimated: Equation 4.1: Equations for SSWDs estimation

lh ik =x i∗ k  ik ,i=Assistants Nurses AN  lh ik =x i∗ k  ik ,i=Registered Nurses RN  (4.1)

lh ik =x i∗ k  ik ,i=SOC6 private comparator to Assistant Nurses lh ik =x i∗ k  ik ,i=SOC3 private comparator to Registered Nurses Where lh ik is the log of hourly earnings of individual i who works in area k . The vector x contains all the control variables (age, age-square, gender, year dummies, industry dummies 5 and occupational dummies5),  ik are the individual-specific error terms and  k are the area-specific effects. SOC6 stands for individuals classified in the Occupational main group 6 working for the private sector which is the comparator group for assistant nurses. SOC3 stands for individuals classified in the Occupational main group 3 working for the private sector which is the comparator group for registered nurses. The area-specific effects represent the SSWDs and are estimated using a dummy variable for each area. As in Meurs & Edon (2007) and in Pereira & Galego (2011) it is argued here that there is no need to control for sample selection. Meurs & Edon (2007) assume that the mechanisms that underpins the choices of sector have little regional variation. Moreover, Meurs & Edon (2007) argue that the correction for selection is sensitive to the chosen specification. 5

Only for the private sector SSWDs.

102

Chapter 4

Creation of Standardised Spatial Wage Differentials

Four sets of SSWDs are estimated, all SSWDs values are LAD specific. Each hospital trust is matched to one LAD, in some cases more than one hospital trust will be matched to one LAD. Some LADs contain no hospitals and therefore do not have a SSWD estimate. Table 4.1, (below), presents the number of those hospitals without a SSWD estimate. Section 4.2, (page 106) investigates this issue.

4.1.2. Cost of living supplements In the Subsection 3.1.3 (page 51) it was explained that the Cost-Of-Living Supplements was set on top of the London Allowance and (implicitly) recognised that the differences between the pay in the public sector and the pay in the private sector is too large for some regions. COLS were introduced in order to compensate for this. There is no need to use this information in the empirical analyses as the SSWDs will capture all elements of pay. Annex E (page 250) shows the relative impact of the COLS on assistant and registered nurses vacancy rates.

4.1.3. Using SSWDs Two sets of SSWDs for nursing are supposed to be associated with each hospital, one for registered nurses and one for assistant nurses. However, the ASHE datasets cover one percent of the employed workforce, as a consequence there are two few observations to compute with confidence SSWDs for some smaller hospitals. Because there are fewer assistant nurses than registered nurses this has a particular impact on observations for assistant nurses. Hospitals would have the two sets of SSWDs for assistant nurses and registered nurses, neither of the two or either the set for assistant nurses or the set for registered nurses. Table 4.1 shows the number of “missing values”. Table 4.1: “Missing values” for assistant nurses and registered nurses SSWDs Set of SSWDs 2003-2005

Assistant Nurses

Registered Nurses

All

Nb of hospitals without a SSWD estimate

87/230 (37.8%)

18/230 (7.8%)

90/230 (39%)

Both sets of SSWDs for assistant nurses and registered nurses will be used at the same time in the analyses, therefore a comparison of hospital trusts with both SSWD sets and those with at least one missing is made below. Ninety out of 230 hospital trusts per year have a missing for either the assistant nurses SSWDs or the registered nurses SSWDs. Table 4.2 gives the descriptive statistics for the number of staff and the staff levels distinguishing by whether or not hospitals have at least one missing SSWD estimate. Hospital trusts without one 103

Chapter 4

Creation of Standardised Spatial Wage Differentials

or the two or without both the assistant nurses and registered nurses SSWD estimate have a lower number of beds (649 against 747), have a higher vacancy rates for the two groups of staff (0.017 against 0.009 for assistant nurses). Table 4.2: Differences in dependent variables by whether or not hospitals have at least one missing SSWD estimate (690 obs.) N

Mean

sd

Min

Max

P10

P50

P90

Number of Beds

All obs.

690

709.0

406.9

44.12

2669

246.9

619.0

1232

Number of Beds

One SSWD missing

270

649.2

303.8

44.12

1556

317.8

588.7

1054

Number of Beds

No SSWD missing

420

747.5

457.4

53.78

2669

238.9

652

1294

AN6 vac. rate

All obs.

690

0.012

0.028

0

0.282

0

0

0.037

AN vac. rate

One SSWD missing

270

0.017

0.035

0

0.282

0

0.000

0.049

AN vac. rate

No SSWD missing

420

0.009

0.022

0

0.212

0

0

0.030

RN vac. rate

All obs.

690

0.025

0.033

0

0.204

0

0.013

0.068

RN vac. rate

One SSWD missing

270

0.032

0.038

0

0.192

0

0.018

0.091

RN vac. rate

No SSWD missing

420

0.020

0.028

0

0.204

0

0.010

0.052

AN staff levels

All obs.

690

0.982

0.511

0.276

9.076

0.590

0.909

1.324

AN staff levels

One SSWD missing

270

0.897

0.395

0.276

3.361

0.520

0.831

1.294

AN staff levels

No SSWD missing

420

1.037

0.568

0.375

9.076

0.646

0.956

1.360

RN staff levels

All obs.

690

1.461

0.546

0.729

8.385

1.065

1.341

1.906

RN staff levels

One SSWD missing

270

1.441

0.502

0.731

6.317

1.053

1.355

1.860

RN staff levels

No SSWD missing

420

1.474

0.573

0.729

8.385

1.083

1.327

1.954

Table 4.3 gives the proportion of registered nurses whether hospital trusts have two SSWDs estimate or not. Hospital trusts with one missing SSWD estimate have higher proportions of registered nurses (62% against 59%). Table 4.3: Differences in proportion of staff by whether or not hospitals have at least one missing SSWD estimate (690 obs.) N

Mean

sd

Min

Max

P10

P50

P90

RN proportion

All obs.

690

0.601

0.091

0.294

0.887

0.495

0.595

0.714

RN proportion

One SSWD missing

270

0.619

0.099

0.335

0.887

0.506

0.611

0.750

RN proportion

No SSWD missing

420

0.590

0.084

0.294

0.851

0.489

0.585

0.695

Table 4.4 gives the descriptive statistics of the different types of hospitals distinguishing for hospitals with an estimate for the two sets of SSWDs (assistant nurses and registered nurses) from hospitals with at least one missing estimate on one of the two sets. Hospitals with a missing value for one set of SSWDs are more likely to be Acute hospitals (59% of all hospitals with a missing value for one set of SSWDs are acute compared to 53% for the ones with no missing values). 6

AN stands for assistant nurses and RN for registered nurses.

104

Chapter 4

Creation of Standardised Spatial Wage Differentials

Table 4.4: Types of hospital trusts by whether or not hospital trusts have at least one missing SSWD estimate (690 obs.) N

Mean

SD

Specialist

All obs.

690

0.052

0.22

Specialist

One SSWD missing

270

0.044

0.206

Specialist

No SSWD missing

420

0.057

0.232

Other

All obs.

690

0.047

0.213

Other

One SSWD missing

270

0.033

0.179

Other

No SSWD missing

420

0.057

0.232

Acute

All obs.

690

0.552

0.497

Acute

One SSWD missing

270

0.588

0.492

Acute

No SSWD missing

420

0.528

0.499

Teaching

All obs.

690

0.108

0.311

Teaching

One SSWD missing

270

0.088

0.285

Teaching

No SSWD missing

420

0.121

0.327

Mental

All obs.

690

0.239

0.426

Mental

One SSWD missing

270

0.244

0.430

Mental

No SSWD missing

420

0.235

0.424

Table 4.5 presents the proportions of hospitals with the foundation status whether hospitals have two SSWD estimates. A bit more than eleven percent of the hospital trusts which have one SSWD estimate missing have the foundation status. Table 4.5: Types of hospital trusts by whether or not hospital trusts have at least one missing SSWD estimate (690 obs.) N

Mean

SD

Foundation Status

All obs.

690

0.134

0.341

Foundation Status

One SSWD missing

270

0.111

0.314

Foundation Status

No SSWD missing

420

0.15

0.357

Table 4.6 presents the descriptive statistics for the complexity index by whether or not hospitals have at least one missing SSWD estimate. Only slight differences are observed in the technicality of the procedures of hospital trusts between hospital trusts with and those without at least one set of parameters for the SSWDs (Table 4.6). The number of observations for this table is lower because mental hospitals do not have a Complexity Index. Table 4.6: Differences between hospital trusts Complexity Index by whether or not hospitals have at least one missing SSWD estimate (516 obs.) Miss

N

Mean

sd

Min

Max

P10

P50

P90

Complexity index

All obs.

516

1.275

0.290

0.83

3.04

1.11

1.16

1.21

Complexity index

One SSWD missing

204

1.292

0.354

0.95

2.56

1.086

1.16

1.19

Complexity index

No SSWD missing

312

1.264

0.238

0.83

3.04

1.11

1.16

1.22

105

Chapter 4

Creation of Standardised Spatial Wage Differentials

Tables 4.2, 4.3, 4.4, 4.5 and 4.6 suggest that hospital trusts with at least one missing SSWD estimate on one of the two SSWD sets (assistant nurses or registered nurses) differ from hospitals with an estimate for both sets of SSWDs. Therefore the “missing” SSWDs estimates may not be at random. The following section shall describe how this issue was dealt with.

4.2. Strategies to deal with “missing” SSWDs parameters in England As seen in Section 4.1 (above) hospitals without at least one SSWD value for one of the SSWDs set appear to be different to those with SSWD values for the two nursing groups. Moreover, a large number (39%, Table 4.1, page 103) of hospitals do not have SSWD estimates for at least one group of nurses. Some strategies to recode the “missing values” have been explored and are presented below.

4.2.1. Using information for surrounding LADs The first two solutions are similar, both imply using informations of the surrounding LADs: –

Before estimation, LADs with small numbers of observations can be merged with the surrounding LADs, those small LADs shall then disappear.

–

After the SSWDs have been estimated LADs with no SSWD can have one imputed by averaging the surrounding SSWDs.

These two solutions imply that surrounding LADs are similar to the missing ones. However, LADs with no SSWDs are not at random as shown in Subsection 4.1.3 (page 103). Hospitals in those LADs are smaller and are likely to have some features that are different from those surrounding areas. These would not be captured when merged. Those two solutions also raise another issue. SSWDs are calculated for LADs in which there is at least one hospital headquarters, in some rural areas none of the surrounding areas have headquarters of a hospital. Consequently, using neighbouring areas is not helpful. The nearest trust with a SSWD may be far away. For these reasons the two solutions above were not exposed further. Instead an alternative solution has been adopted as discussed in Subsection 4.2.3 (page 108).

106

Chapter 4

Creation of Standardised Spatial Wage Differentials

4.2.2. Using more years Even though the labour market in the private sector is supposed to clear in the short term, it is unlikely to do so very quickly. Pooling years together recognises this feature and mitigates the effects of short term shocks. However, with more years the estimation will reflect less current local labour market conditions. SSWDs are estimated to capture the underlying pattern of spatial wage differentials and those are unlikely to be revealed in data from a single year. It is for this reason that data is pooled. That said, the right number of years is difficult to assess. Elliott et al. (2006) used 3 years pooled estimates. Increasing the number of years reduces the number of LADs with few observations and thus reduces the number of LADs with no SSWD. Five sets of SSWDs for NHS group of staff were estimated (estimations were also made for private sector group of staff7) with a different number of years pooled each time: 1997-2005; 1998-2005; 1999-2005; 2000-2005; 2001-2005; 2002-2005; 2003-2005. Table 4.7 shows that the number of hospital trusts with missing values for the SSWDs is decreasing as the number of years pooled is increased. It falls from 39% to 9%. Table 4.7: Frequencies for the number of hospital trusts without an estimate for the assistant nurses or registered nurses SSWD Set 03-05

Set 02-05

Set 01-05

Set 00-05

Set 99-05

Set 98-05

Set 97-05

230

230

230

230

230

230

230

Non missing

140 (61%)

169 (73%)

181 (79%)

196 (85%)

200 (87%)

204 (89%)

209 (91%)

Missing

90 (39%)

61 (27%)

49 (21%)

34 (15%)

30 (13%)

26 (11%)

21 (9%)

All

Tables 4.8 and 4.9 compare the different sets of SSWDs calculated pooling different number of years for each of the two nursing staff groups 8. The registered nurses table has one more column and row than that for assistant nurses as the SSWDs already in the public domain have been set alongside those calculated for this thesis. In both tables a similar pattern can be observed, the correlations with the 2003-05 set are decreasing when more years are added, assistant nurses SSWDs are slightly less correlated with a correlation of “only” 0.78 between the pooled data for the earliest period 2003-2005 and that for 1997-2005. One possibility would seem to use the pooled SSWDs with the least number of “missing values”. Subsection 4.2.3 (page 108) explains the solution chosen.

7 8

Seven sets of SSWDs for 5 group of staff, that is 35 estimations. The observations used are only the ones for which there are two values for the SSWDs, for example the 2002-05 set has its correlation with the 2003-05 set calculated only for hospitals with a value on both sets.

107

Chapter 4

Creation of Standardised Spatial Wage Differentials

Table 4.8: Correlations of SSWDs estimated using the assistant nurses population 2003-05

2002-05

2001-05

2000-05

1999-2005

1998-2005

1997-2005

2003-05

1.000

2002-05

0.964

1.000

2001-05

0.922

0.969

1.000

2000-05

0.892

0.949

0.988

1.000

1999-2005

0.859

0.922

0.971

0.985

1.000

1998-2005

0.828

0.908

0.950

0.965

0.982

1.000

0.985

1997-2005

0.782

0.876

0.930

0.946

0.966

0.985

1.000

Table 4.9: Correlations of SSWDs estimated using the nurses population Pub Dom

2003-05

2002-05

2001-05

2000-05

1999-2005

1998-2005

1997-2005

Pub Dom

1.000

2003-05

0.975

1.000

2002-05

0.955

0.973

1.000

2001-05

0.906

0.932

0.973

1.000

2000-05

0.879

0.893

0.947

0.984

1.000

1999-2005

0.860

0.865

0.925

0.958

0.985

1.000

1998-2005

0.835

0.837

0.902

0.940

0.971

0.992

1.000

0.992

1997-2005

0.824

0.816

0.883

0.922

0.955

0.980

0.992

1.000

4.2.3. Using information of surrounding LADs and more years An alternative is to use the set of SSWDs for 1997-2005 to impute “missing values” for the 20032005 set. Several LADs have values for the set 1997-2005 but not in the 2003-2005 one (for whichever group of staff: assistant nurses or registered nurses). The goal is to impute values for these observations. As stated earlier choosing the surrounding LADs is in many ways inappropriate as there are some LADs which are not surrounded by any other LAD which has a SSWD estimate. A regression is run on the set of 2003-2005 with the only explanatory variable being the set for 1997-2005. This regression is run on observations for which SSWDs values for the 1997-2005 set and for the 2003-2005 set are available. Then it is possible to impute the missing values in the 2003-2005 set with the parameters of this regression and the SSWDs values in the 1997-2005 set. This is done for both NHS groups of nursing staff.

108

Chapter 4

Creation of Standardised Spatial Wage Differentials

Equation 4.2: Recoding SSWDs /2005  ANSSWDk2003/ 2005=∗ANSSWD1997  k k  2003/ 2005 1997 / 2005 RNSSWD k =∗RNSSWD k  k

(4.2)

Where k index LADs, AN stands for assistant nurses and RN stands for registered nurses. Once   and  have been estimated, it is then easy to impute values for areas with missing observations for the SSWDk2003/2005 which have a value in the SSWDk1997/2005 . Using the SSWD estimated for 1997-05 to impute values for the 2003-05 rely on the assumption that the differences between the regions with “missing values” and the regions without have remained the same over the years: it is assumed that, had observations for the “missing values” for 2003-2005 been available, they would be a fixed transformation of the ones that are available for the 1997-2005.

4.2.4. Testing the method used to recode missing SSWDs values The assumption made in Subsection 4.2.3 can be evaluated by using the areas for which there are values in both sets 1997-05 and 2003-05. First the same regression as in Subsection 4.2.3 is computed on a sub-sample of observations which have SSWDs for both sets (2003-2005 and 19972005). Note the sub-sample represent p% of the total of the sample. Then values for the 2003-2005 set of observations that are not in the sub-sample but are in the whole sample are imputed as in 4.2.3. Then the values imputed are compared with the true values. This strategy has been repeated 1000 times with p equal to the true proportion of missing values in the set 2003-2005. For the assistant nurses regression, the regression was computed on 62% of the sample, for registered nurses 92% of the sample was used. Table 4.10 presents average correlations between the computed and the true values. Those correlations are high but obviously different from one. Table 4.10: Average correlations between the true values and the imputed ones for each group of NHS staff Average correlations [CI@95%]

9

9

Assistant nurses

Registered nurses

0.925 [0.9247;0.9258]

0.987 [0.9859;0.9864]

The confidence interval is computed as usually done in bootstrapping methods. The lowest 2.5% values and the highest 97.5% values are removed. Then the lowest and highest of the remaining distribution are the lowest and highest bounds of the confidence interval.

109

Chapter 4

Creation of Standardised Spatial Wage Differentials

This test is sound as it shows that the methods used to compute the SSWDs gives for both staff groups close values to the true ones. This method is used to impute those missing SSWDs. As table 4.11 shows, there is now only 9% of hospitals without a value for at least one of the nursing groups SSWD. This means that the sample of hospitals used is then reduced from 690 to 627 resulting in 63 “missing values”. Table 4.11: Number of observations that will be used in subsequent chapters Nb of Observations (%)

Non Missing

Missing

627 (91%)

63 (9%)

In Annex F (page 252), the descriptives statistics presented in Tables 3.9, 3.11, 3.11, 3.12 and 3.13 (pages 73 to 76) are given for the restricted sample of observations with SSWD estimates for the two nursing groups. Annex G (page 253) gives the same descriptives when restricting the sample to observations with both SSWDs and the complexity index. To conclude the recoding sections, the number of hospitals without a SSWD estimated has been reduced to just below 10%.

4.3. Definition of pay gaps and their descriptive statistics for England It was explained earlier that the gap between two SSWDs for a nursing staff group and that of the comparator group provides a measure of the competitiveness of nursing staff groups pay. Here is explained how the gap is calculated. First, the interpretation of SSWDs is reviewed. SSWDs are estimated based on a log wage regression. Two papers in the early eighties have discussed how to interpret the parameter  associated to dummy variables in semi logarithmic regressions (Halvorsen and Palmquist, 1980; Kennedy, 1981). Kennedy (1981) refined the method proposed by Harlvorsen & Palmquist (1980) to interpret parameters associated to dummy variables. Harlvorsen & Palmquist (1980) argued that the formula in 4.3 transform the dummy parameter and can then be interpreted in how much percentage change in the dependent variable is due to the dummy taking the value 1. Equation 4.3: Effect of a dummy variable following Harlvorsen & Palmquist (1980)

=100exp −1

(4.3)

Where exp  is the exponential function,  is the parameter of one of the area dummy,  is the wage percentage difference for workers in this area compared to the reference. If  equals 0.1, 110

Chapter 4

Creation of Standardised Spatial Wage Differentials

then wages for workers in this area are 10.5% greater than wages for workers in the reference area. Kennedy (1981) proposed a corrected , for  is not known but estimated by  . Therefore  should be corrected into  * by the variance of the estimation of . For small values of  , between -0.5 and 0.5,  is close to  and biased downward. The difference between  and  * depends on the variance of  . Because the focus of the thesis is to provide evidence of a significant impact of pay competitiveness on staff shortages, and skill mix and because  * is, according to Kennedy (1981), still biased, the gaps are computed using  . Differences between two sets of SSWDs will be used in empirical analyses in subsequent chapters. Gaps are defined as in Equation 4.4. Equation 4.4: Definition of gaps

gapAN =ANSSWD −SOC6SSWD gapRN =RNSSWD −SOC3SSWD

(4.4)

Where AN stands for Assistant Nurses and RN for Registered Nurses. SOC6 and SOC3 represent the groups of workers of the private sector in the main SOC group number 6 and 3 respectively. The gaps used in subsequent chapters are then centred, a region with a positive gap has a relative nurses SSWD larger than the private sector one compared to the national mean. Gaps are differences of two percentages, if the value is 0.1, the standardised nurses wage in this particular region is more competitive by approximatively 10 percent compared to the national mean. Table 4.12 shows descriptive statistics of the gaps. The range is larger for assistant nurses (min=0.33 and max=0.40) than for registered nurses and so is the standard deviation. For both gaps the median is slightly greater than the mean. Table 4.12: Gaps descriptive values N

Mean

sd

Min

Max

P10

P50

P90

Assistant Nurses gap

627

0

0.123

-0.33

0.400

-0.16

0.003

0.127

Registered Nurses gap

627

0

0.111

-0.28

0.245

-0.16

0.007

0.129

As explained in Subsection 2.5.1 (page 43) the gap of, say, registered nurses is supposed to have an impact on the shortage of assistant nurses. This may occur in two ways. First the gap for registered nurses may drive the shortage of assistant nurses directly. If the registered nurses pay is competitive, then assistant nurses may have better working conditions due to better staffing

111

Chapter 4

Creation of Standardised Spatial Wage Differentials

and may look forward to the competitive registered nurses pay as they may want to become registered nurses themselves. In non competitive areas for the registered nurses pay a local shortage of assistant nurses may be observed. Secondly, the effect of the assistant nurses pay gap may be stronger when the pay of the registered nurses is also competitive. A slightly more competitive pay for assistant nurses is expected to have a larger effect for areas with an already competitive pay for registered nurses. To test this, the effect of pay gaps will be interacted with a dummy variable indicating the competitiveness of pay of the other group of staff. Those two dummy variables are simply built by taking the value one when the gap is above its national mean, therefore the interaction will test the impact of the gap for registered nurses (assistant nurses) when the assistant nurses (registered nurses) pay gap is more competitive than the competitiveness of the gap at the national level. The cut off point for constructing the dummy is quite arbitrary, pay for hospitals for which the dummy value is one may not be competitive, it just says that it is more competitive than the average national pay. Table 4.13 gives descriptive statistics for the gap of registered nurses by whether the gap for assistant nurses is above or below the national mean. The registered nurses gap is on average negative when the assistant nurses gap is below its national mean. Table 4.13: Registered nurses pay gap distinguished for whether the assistant nurses gap is above or below its national mean RN gap 10

AN10 gap is … the national mean

N

Mean

sd

Min

Max

P10

P50

P90

(all)

627

0

0.111

-0.28

0.245

-0.16

0.007

0.129

below

306

-0.02

0.103

-0.26

0.226

-0.17

-0.00

0.089

above

321

0.020

0.114

-0.28

0.245

-0.14

0.033

0.170

Table 4.14 gives the descriptive statistics for the gap of assistant nurses when the gap for registered nurses is either below or above the national mean. The assistant nurses gap is positive on average when the registered nurses gap is above the national mean.

10 AN stands for Assistant Nurses and RN stands for Registered Nurses.

112

Chapter 4

Creation of Standardised Spatial Wage Differentials

Table 4.14: Assistant nurses pay gap distinguished for whether the registered nurses gap is above or below its national mean AN gap 11

RN11 gap is … the national mean

N

Mean

sd

Min

Max

P10

P50

P90

(all)

627

0

0.123

-0.33

0.400

-0.16

0.003

0.127

below

288

-0.02

0.119

-0.31

0.247

-0.22

-0.02

0.111

above

339

0.022

0.122

-0.33

0.400

-0.10

0.027

0.151

4.4. Creation of Standardised Spatial Wage Differentials for France Pay gaps are computed by differencing two Standardised Spatial Wage Differentials. The latter are explained in 4.4.2 below. Beforehand the pay data that is available for France and is used in the computations is presented in Subsection 4.4.1. Subsections 4.4.3, 4.4.4 and 4.4.5 present the methods used to estimate SSWDs and the results, while the last subsection (4.4.6) will present the descriptive statistics of the SSWDs and the gaps.

4.4.1. Labour market data The “Déclarations Annuelles des Données Sociales” (DADS) is an administrative data set which gives details of the pay and employment of all employees in all firms in France 12. Before the 2008 version employees of the state were excluded from this data though administrative employees, teachers and nurses were all present before 2008 13. Each year all companies have to provide to the fiscal and social administrations the names of all their employees during the year and information on these employees such as their sex, age, address, hours worked, position, qualification and the wage they received. The National Institute of Statistics and Economic Studies (INSEE, Institut National de la Statistique et des Études Économiques) then merge this data with other data on the firms. Some controls and tests are then conducted by INSEE to validate the data. Different sets of data are available: jobs, main jobs, premises and companies 14. The premises and companies dataset gives information at the premise and company level which is not of interest here. The two other sets of data gives the gross and the net 15 wage of the individual, the number of hours she/he has worked during the year, the location of the workplace, the types of

11 AN stands for Assistant Nurses and RN stands for Registered Nurses. 12 This data has been accessed thanks to Eric Delattre, co author on the paper submitted to the Journal of Health Economics based on this work, and helped improve the work presented here. 13 Employees of private individuals or families are not surveyed either. 14 There is a fifth one which is a panel version of the data, it is used by Combes et al. (2008). 15 As defined in France, gross pay minus social contributions but without deducting income tax.

113

Chapter 4

Creation of Standardised Spatial Wage Differentials

companies the individual is working for. Consequently it is possible to compute the average hourly wage of the individual each year for each job. The data does not distinguish basic from overtime and other types of pay, there is just one value for the total pay the person received for the job. The jobs data gives all the positions an individual occupied during the year (in the data an individual may be represented by more than one row). The main jobs data gives, for each individual, the main position held during the year. There is also information on the premises and companies16 at which employees work. The jobs data was used in this thesis; more than one row was available for each individual. Two formats of this data exist, one which is exhaustive and which was used when analysing nurses only and the other which is a sample at 1/12th and which was used when analysing the largest groups of employees such as those working in the private non hospital sector. Three years of data are used, 2006-2007-2008; the choice of these three years is data driven, since only from 2006 assistant nurses and registered nurses can be distinguished from other group of carers such as social assistants. For each employee the hourly wage is calculated by dividing the total annual salary by the number of hours worked during the year. The net wage is used. The aim is to compare the attraction of different jobs, therefore using the “net” wages is more appropriate than gross as deductions for unemployment insurance, health insurance are different in the public and the private sectors. Identifying local labour markets is a challenging task. By knowing the time and/or distance people travel to work then local labour markets based on the pattern of travel to work of staff could be created. Combes et al. (2008) use such data called “employment areas”, there are 341 of them in metropolitan France. This boundary was not used for three reasons, first its existence was not known at the time of analysing the data and it is not straightforward to match those areas with hospital data, the “employment areas” may match correctly the travel to work of the general population but may not do so for nurses. Moreover, for the English data Elliott et al. (2006) find that the administrative boundaries compared to travel to work areas performed better following a goodness of fit model on the variation of wages. Départments are the second level of local administrative boundaries (see 3.2.8, page 65). Communes 16 Those two datasets are aggregated information at the company or premise level to make easier the analysis of data at that level. It is not really of use for the analyses done here.

114

Chapter 4

Creation of Standardised Spatial Wage Differentials

are too small, many communes will not have a hospital and people would travel to another commune to work, large communes such as Lyon or Marseille would be large enough for local labour markets but this geography does not include commuters. The commune definition is restricted compared to other European countries. As reported above (Subsection 3.2.8, page 65) the median number of inhabitants by commune is 400 in France, it is 11 000 in Belgium, 2 000 in Italy and 5 500 in Spain17. Régions, on the other hand, are too large, as the Parisian région (Île de France) contains 19% of the total population. Départements are a fair boundary, it will not remove the issue of people moving from one département to another large town in another département to work. On top of the net hourly wages variables that are available in the DADS, the département of work of each individual is also known. Excluding overseas départements except Corsica, all départements are used. The data distinguishes between part time and full time jobs. The sex and age of the employee will also be used along with year dummies.

4.4.2. Labour Market models Section 4.1 (page 99) explained the creation of Standardised Spatial Wage Differentials for the UK. This section will explain that the technique used for the UK is repeated for France (Section 4.4.3) and that another technique is also used (Section 4.4.4). The new technique measure the wage gaps directly, it tells by how much a group of staff earn more than another one in the same area and test whether this difference is significantly different from zero. This technique is used when presenting descriptive statistics of the wage gaps and when estimating more than one slope for the wage gaps. Because both techniques produce very similar results in terms of ranking local labour markets18 the effect of a gap in a regression model will be the same when using one or the other technique to compute the gap. However, in these regressions, the parameters for this gap estimated with one or the other method will differ because their values differ. The gaps estimated using the method in 4.4.3 will be used in regressions as not all gaps exist when estimated using the method 4.4.4. Assume that SSWDs are estimated for registered nurses. The technique used in 4.1 and in 4.4.3 gives a value for each area which reveals by how much percent registered nurses in this area are over or under paid compared to a reference (national mean or a particular area). Thus a value of 17 http://fr.wikipedia.org/wiki/Commune_(France) accessed the 2nd of November 2011 18 If the technique used in 4.4.4 had been used for English data it would not have changed the ranking of the geographical areas, an area with a high SSWD with technique in 4.1 would still have a SSWD that is high here. Correlations between the results of the two techniques would be perfect.

115

Chapter 4

Creation of Standardised Spatial Wage Differentials

zero indicates that it is the reference. Thus the gap between two SSWDs, calculated with this technique, for registered nurses and a comparator group, tells by how much greater this gap is compared to the reference value. In the technique used in 4.4.4, gaps are estimated directly, SSWDs are not estimated, assume that a gap for registered nurses working the public hospital sector against a comparator is estimated. A gap with the value zero (or close to) will tell that there are no differences between the underlying SSWD for registered nurses and the underlying SSWD for the comparator group. A positive, significantly different form zero, value (say c) will tell that the registered nurses are more paid than their comparator by 100*c percent. Thus this technique gives extra information; not only does it give the index of regions: a positive gap means that registered nurses in the public sector are relatively better paid than their comparator group compared to a region with a smaller gap. It also tells that registered nurses are better paid than their comparator group in this particular area. Following the previous paragraph, a more technical explanation of the differences is given below. Both methods were used on the French data, thus it is possible to test for the differences between the two gaps.

4.4.3. Previous method, fully stratified method SSWDs are estimated from a wage equation and represent the parameters associated with the area dummies (see Equation 4.5). SSWDs are estimated for different groups of nursing staff in public hospitals, private not for profit hospitals, private for profit hospitals and a group comprising workers in the private non hospital sector belonging to the same PCS as the nursing groups (see Subsection 3.2.9, page 65). The value of a SSWD in one area tells by how much the standardised wage differs from a reference (either the national mean or an area set up as the reference). Equation 4.5: Equations for SSWDs estimation

lh ikj=∗x ij  j∗ k  ikj , j=Nurses Public sector lh ikj=∗x ij  j∗ k  ikj , j=Workers in Private Non Hospital lh ikj=∗x ij  j∗ k  ikj , j=Nurses Private Not for Profit lh ikj=∗x ij  j∗ k  ikj , j=Nurses Private for Profit

(4.5)

Where lh ikj is the log of hourly earnings of individual i who works in sector j of the economy in area k . The vector x contains all the control variables (age, age-square, gender, year dummies,

116

Chapter 4

Creation of Standardised Spatial Wage Differentials

part time dummy, industry dummies 19 and occupational dummies19),  ikj are the individualspecific error terms and  k are the area-specific effects and  j their associated parameters. The area-specific effects represent the SSWDs and are estimated using a dummy variable for each area. Equation 4.6: Definition of gaps

gap PNH : Nurses Public vs. Private non hospital =PUSSWD −PNHSSWD gap PNP : Nurses Public vs. Nurses Private not for profit =PUSSWD−PNPSSWD gap PP : Nurses Public vs. Nurses Private for profit =PUSSWD −PPSSWD

(4.6)

Equation 4.6 defines gaps. PUSSWD stands for the SSWD vector for nurses working in the hospital PUblic sector, PNP stands for nurses working in the hospital Private Not for Profit sector, PP stands for nurses working in the hospital Private for Profit sector, finally, PNH stands for workers in the Private Non Hospital sector. Gaps are differences between two SSWDs estimated on a different sample for the same region (see Equation 4.6). The higher the gap the larger the difference between two SSWDs. Though the value of one gap (say it is c) only tells that the gap in this particular region is 100*c percent greater than the gap for the reference.

4.4.4. New method, fully interacted method It is interesting to know if the gap in an area is larger than in another one, but this does not tell if the gap in this particular region is different from 0, or put another way if the SSWDs for nurses in the public sector is different from SSWDs in the private not for profit (or for profit) sector for that particular area? Answering this question implied re-estimating wage equations with a different specification. This re-estimating strategy omits the estimation of SSWDs and directly estimates the gaps. This strategy is not used to calculate the gap between nurses in the public hospital sector and workers in the private non hospital sector. Equation 4.7: Direct estimation of gaps with the fully interacted method

lh ik =α j +β∗x i +β∗x i∗φ j + γref ∗μ k + γ j∗μ k∗φ j +εik

(4.7)

Where lh ik is the log of hourly earnings of individual i who works in area k . The vector x contains all the control variables (age, age-square, gender, part time dummy and year dummies), 19 Only for private non hospital sector workers.

117

Chapter 4

Creation of Standardised Spatial Wage Differentials

 ik are the individual-specific error terms and  k are the area-specific effects and  j their associated parameters. Where  j is a set of dummies for all but the reference sector of the economy. Thus the associated parameters  j represent the Log-standardised wage gap between the sector j of the economy and the reference sector and thus provides a direct estimation of the gaps. Its associated t-statistics reveals if the gap for region k is statistically different from 0. Not all gaps, as in Equation 4.6 are estimated with one run of Equation 4.7. Only two can be estimated (depending on which sector is taken for reference), the missing one is obtained by re estimating Equation 4.7 and changing the reference sector. Gaps are directly estimated, therefore there are no SSWDs estimated from this method. However, the underlying idea of SSWDs is still accurate. Those gaps explain by how much standardised pay for nurses in the public hospital sector is above or below that of nurses in the alternative hospital sector (private not for profit and private for profit). There are perfect correlations between gaps estimated with Equation 4.7 and the ones computed with Equation 4.6. However, the values differ. The gap between nurses working in public hospitals and the workers in the private non hospital sector was not estimated like this, it was only calculated with the “old method” 20.

4.4.5. Results of the estimations SSWDs were estimated with the fully stratified method (Subsection 4.4.3 page 116) for registered nurses, assistant nurses in all hospitals sector and non nursing groups working in the private non hospital sector (see Table 4.15 for details of the different samples). Non nursing groups are defined as employees with the same Professions et catégories socioprofessionnelles (PCS) code as the nursing groups and working in the private non hospital sector. Registered nurses and assistant nurses are coded within the group 4 and 5 respectively. The comparator group for registered nurses and for assistant nurses are thus defined as employees in the group 4 and 5 21 respectively.

20 It is feasible to estimate it like this, the problem laying with the size of the dataset to be used which may be too large (around 2 millions observations). 21 See Subsection 3.2.9 (page 65) for more details about the PCS.

118

Chapter 4

Creation of Standardised Spatial Wage Differentials

Table 4.15: Summary of the data used for estimating different sets of SSWDs Occupations

Sector

Data

Nb of observations

Nb of départements

Registered nurses, PCS code 431,

Hospitals in the public sector

Jobs Exhaustive

870499

96

Registered nurses, PCS code 431,

Hospitals in the private not for profit

Jobs Exhaustive

143100

85

Registered nurses, PCS code 431,

Hospitals in the private for profit

Jobs Exhaustive

213386

94

Assistant nurses, PCS code 526a

Hospitals in the public sector

Jobs Exhaustive

612631

96

Assistant nurses, PCS code Hospitals in the private not 526a for profit

Jobs Exhaustive

93344

85

Assistant nurses, PCS code Hospitals in the private for 526a profit

Jobs Exhaustive

136680

94

Semi-Professionals (PCS code 4) excluding registered nurses

Private for profit sector excluding hospitals

Jobs National Sample, 1/12th of the data

761 260

96

White collar workers (PCS code 5) excluding assistant nurses

Private for profit sector excluding hospitals

Jobs National Sample, 1/12th of the data

1 624 597

96

There are 96 départements in metropolitan France (including Corsica), some départements do not have private not for profit or private for profit hospitals. Therefore there are no SSWD for those département. The next sub sections will give descriptives of SSWDs and gaps in France.

4.4.6. SSWDs descriptive statistics Table 4.16 presents the SSWDs. There are two for each nursing group in the public hospital sector. There are six for each comparator group: two groups of employees working in the private non hospital sector, assistant and registered nurses working in the private not for profit hospital sector and assistant and registered nurses working in the private for profit hospital sector. There are no missing départements for the non nursing comparator groups SSWDs as there are workers in each départements. No département is missing for the hospital public sector. Eleven départements are missing for hospital private not for profit sector 22. Two départements are missing for hospital private for profit hospitals23. These missing values are the consequences of the absence of hospitals (private not for profit or private for profit) in these départements. The range of values, the standard deviation and the confidence interval are lower for the hospital public sector, suggesting that pay in the public sector is less widespread. The standard deviation

22 Allier, Ariège, Cher, the two Corsica département, Gers, Haute-Marne, Meuse, Somme, Tarn et Garonne and Territoire de Belfor are missing for private not for profit hospitals. 23 Ariège and Lozère are missing for private for profit hospitals.

119

Chapter 4

Creation of Standardised Spatial Wage Differentials

for SSWDs for registered nurses working in the private for profit hospitals is the highest compared to registered nurses in the other two hospital sectors. The standard deviation for SSWDs for assistant nurses working in the private not for profit hospitals is the highest compared to assistant nurses in the other two hospital sectors. The standard deviation for the SSWDs for the comparator group for registered nurses working in the private non hospital sector is lower than the standard deviation for the SSWDs for registered nurses in the two private hospital sectors and higher than the standard deviation for the SSWDs for registered nurses in the public hospital sector. The same ranking is observed for the standard deviation for SSWDs for assistant nurses. Table 4.16: Descriptive statistics for SSWDs Registered nurses

Assistant nurses

Hospital sector

Hospital sector

Private Non Hospital

Public

Private not for profit

Private for profit

Private Non Hospital

Public

Private not for profit

Private for profit

Obs

96

96

85

94

96

96

85

94

Missing

0

0

11

2

0

0

11

2

Min

-0.171

-0.081

-0.116

-0.256

-0.096

-0.060

-0.140

-0.126

Max

0.046

0.031

0.128

0.068

0.031

0.024

0.065

0.033

Range

0.217

0.111

0.244

0.324

0.128

0.084

0.204

0.159

Median

-0.103

-0.025

0.014

-0.090

-0.065

-0.014

0.009

-0.035

Mean

-0.097

-0.023

0.012

-0.087

-0.058

-0.013

-0.002

-0.039

CI 0.95

0.008

0.004

0.009

0.012

0.005

0.004

0.009

0.006

SD

0.042

0.022

0.044

0.056

0.023

0.017

0.039

0.029

CV

-0.428

-0.950

3.520

-0.643

-0.398

-1.349

-26.680

-0.762

4.4.7. Gaps descriptive statistics The analyses explores the impact of spatial variations in the competitiveness of nurses pay on spatial variations in nursing staff levels and skill mix in French hospitals. The competitiveness of nursing pay is distinguished by the gap between the pay of two groups being compared. Three gaps are used: −

the differences between the pay for nurses (assistant or registered nurses) in public hospitals and the pay for a comparator group working in the private non-hospital sector, gap PNH;

−

the differences between the pay for nurses (assistant nurses or registered nurses) in public hospitals and the pay for nurses (assistant nurses or registered nurses) in hospitals in the not for profit private sector, gap PNP;

120

Chapter 4

−

Creation of Standardised Spatial Wage Differentials

the differences between the pay for nurses (assistant nurses or registered nurses) in public hospitals and the pay for nurses (assistant nurses or registered nurses) in hospitals in the for profit private sector, gap PP.

Table 4.17 presents the descriptive statistics of the gaps. For the hospital sector, the gaps presented are the ones estimated following the fully interacted method described in Subsection 4.4.4 (page 117). For the gaps computed with the workers in the private non hospital sector, the “old” method was used. The fully interacted method allows the comparison of underlying SSWDs between two sectors in the same département. Thus a negative value for a gap defined as the difference between the underlying SSWD for registered nurses in the public hospital sector and the underlying SSWD for registered nurses in the private not for profit hospital sector means that the registered nurses in the private not for profit sector are, on average, more paid than registered nurses in the public hospital sector. The first comment concern the differences in the gaps for assistant nurses and in the gaps for registered nurses, assistant nurses have better SSWDs in the public sector compared to both private hospital sectors (in all départements). Assistant nurses working in the public hospital sector are paid 20% to 40% more (average is 26%) than assistant nurses working in one of the hospital private sector. Registered nurses tend to have lower SSWDs (negative mean, median) in the public sector compared to nurses working in the private not for profit hospitals. Registered nurses are on average paid 11% and 2% less in the public hospital sector than in the private not for profit hospital sector and private for profit hospital sector respectively. The standard deviation is smaller for the gaps between SSWDS for nurses in the public hospital sector and SSWDs for employees in the private non hospital sector than for the two other gaps. The range, which shows the maximum variation in pay competitiveness across département, shows that there is between 14% and 28% variations in the competitiveness of pay between wages for nurses in the public sector and wages for one of the comparator group.

121

Chapter 4

Creation of Standardised Spatial Wage Differentials

Table 4.17: Descriptive statistics for gaps Registered nurses Non hospital

Assistant nurses

Hospital sector

Non hospital

Hospital sector

PNH24

PNP24

PP24

PNH24

PNP24

PP24

Obs

96

85

94

96

85

94

Missing

0

11

2

0

11

2

Min

-0.107

-0.194

-0.150

-0.109

0.192

0.195

Max

0.102

0.012

0.134

0.047

0.437

0.336

Range

0.209

0.207

0.284

0.156

0.245

0.142

Median

0.003

-0.112

-0.030

0.004

0.262

0.247

Mean

0.000

-0.113

-0.028

0.000

0.269

0.251

CI 0.95

0.007

0.009

0.010

0.005

0.009

0.006

SD

0.034

0.042

0.050

0.026

0.041

0.029

CV

NA

-0.375

-1.766

NA

0.152

0.116

In the empirical analysis, the gaps used are the ones calculated with the fully stratified method in order to have comparable estimates between the gaps of types PNP and PP and gaps of type PNH24. Subsection 2.5.1 (page 43) explained that the gap of, say, registered nurses will have an impact on the shortage of assistant nurses. This may occur in two ways. First if the pay for assistant nurses is competitive, then registered nurses may have better working conditions due to better staffing. Second if the pay for registered nurses is competitive, assistant nurses may look forward to the competitive registered nurses pay as they may want to become registered nurses themselves. When the pay for registered nurses is not competitive a local shortage of assistant nurses may be observed (respectively for the pay for assistant nurses and the shortage of registered nurses). It was observed for English data, that the effect of the wage gap for registered nurses on the registered nurses vacancy rates was stronger when the wage gap for assistant nurses was above its national mean. The effect of pay gaps will be interacted with a dummy variable indicating the competitiveness of pay of the other group of staff. Two slopes for the pay gaps (PNH, PNP, or PP) for registered nurses and for pay gaps for assistant nurses of type PNH will be introduced, one when the alternative gap is below the national mean and one when it is above. If the gaps computed with the new method are negative, then the underlying SSWD for the

24 Where PNH stands for Private Non Hospital sector, PNP stands for the Private Not for Profit hospital sector and PP stands for Private for Profit hospital sector.

122

Chapter 4

Creation of Standardised Spatial Wage Differentials

registered nurses in the hospital public sector has a lower value than the SSWD in one of the comparator group. Different slopes for the gap of type PNP for assistant nurses in assistant nurses staff level models will be introduced based on whether the gap for registered nurses is negative or not significantly different from zero (the registered nurses gap of type PNP is never positive and significantly different from 0). Different slopes for the gap of type PP for assistant nurses in assistant nurses staff level models will be introduced based on whether the gap for registered nurses is negative and significantly different from 0, not significantly different from zero or positive and significantly different from 0.

4.5. Conclusion After the literature review in Chapter 2, the institutional background, the data and some preliminary analyses in Chapter 3 for both England and France, this chapter presented the analyses made on wage data for both countries. These analyses are necessary for the core of the empirical analyses on hospitals in England and France.

4.5.1. Comparing data for the two countries, limitations of the data For England, SSWDs are computed with a more refined definition of local labour markets. England is smaller in terms of inhabitants than France, though the local labour markets used for England are more numerous, they contain less individuals and are likely to have less heterogeneity. However, this advantage is countered by the dataset for wages. The dataset for wages in England is a survey that contains “only” one percent of employees in England while the dataset for France contains the whole of the employees. In consequence for some local labour markets in England, SSWDs were not estimated because of a lack of observations in the dataset. For France, SSWDs were estimated for all départements. As expected in Section 3.3 (page 67), there is more widespread geographical variations between the public sector pay and the private sector one in England than in France. In England the standard deviation for pay gaps for assistant nurses and registered nurses are 0.12 and 0.11 respectively. The average variation around the mean represent a 11-12% variation. In France, the standard deviation of the pay gaps represents only a 3 to 4% variation for the pay gap for assistant nurses and 3 to 5% variation for the pay gap for registered nurses. In consequence, one could expect that the pay gaps for France have less consequences on public hospitals than the pay gaps for England.

123

Chapter 4

Creation of Standardised Spatial Wage Differentials

With the wage data wage equations were estimated. In these wage equations no control for education was introduced as the data used does not provide such information. Instead, the productivity of workers is controlled by the occupational codes. This document argues that it is a much better control as it introduces 353 number of categories for England and 497 number of categories for France. Had education been provided the number of categories used would have been, at best, a handful. Occupational coding are introduced only for non nursing groups. For nursing groups there is only one occupational code per nursing group (registered nurses and assistant nurses) in England and five occupational codes for registered nurses and only one for assistant nurses in France. One could argue that with occupational coding it is not possible to control for different levels of productivity among the nursing workforce for both nursing groups in England and for assistant nurses in France. Though, in both countries, the number of school years for nurses is more than A levels for registered nurses and education usually does not distinguish education levels above A levels 25. On the contrary, with education, this thesis would have been able to distinguish assistant nurses with or without A levels. The shortage of staff may have an impact on SSWDs, this is a cause of reverse causation which is discussed in more econometric terms in Section 5.3.1. Where workers have skills to move from nursing labour markets to the comparator groups labour markets, then where hospitals do not have any shortage of staff, workers may look for jobs in the comparator groups. This is more likely where skills are transferable. Obviously, this is the case for nurses in France as they would use the same set of skills whether for the public or the private hospital sector. Lower qualified nursing staff are expected to have more transferable skill. Though, it is not possible to disregard the ability more qualified nursing staff may have in moving to another occupation. Empirically this will have consequences as the effect of the competitiveness of pay on shortage of staff will be affected by the effect of shortage of staff on the competitiveness of pay. However, the effect of the shortage of staff on the competitiveness of pay through its impact on the supply of labour and on the pay rate exist only and only if pay is flexible. Where pay is not flexible, there is still an impact of shortage of staff in public hospitals on the supply of labour in the comparator group labour market but this does not impact the pay. In consequence, where pay is not flexible there is no reverse causation possible. A larger supply of workers does not depress the wage rate which does not increase the wage gap. In France workers in all sectors are likely to be covered by collective agreements. The supply of labour has no or less impact on wages. It is expected that there is no reverse causation in France.

25 Registered nurses in both England and France would answer that they have a higher education diploma.

124

Chapter 4

Creation of Standardised Spatial Wage Differentials

By merging the geographical parameters of wage equations with hospital data, it is then possible to estimate the impact of the wage gaps on hospital shortage of staff and skill mix. This thesis uses this data as a cross sectional data. Though observations for hospitals are available for three years for both countries. Then one may ask why not using panel estimation strategies in order to control for individual heterogeneity such as in Propper and Van Reenen (2010)? For both countries it has been argued that to avoid shocks in the labour markets wage equations should be ran for a minimum of three years. This taken into account in order to use panel data with fixed effects estimations (fixed effects are unbiased as they fully control for individual heterogeneity) six years of hospitals data would have been needed. Unfortunately, and contrarily to Propper and Van Reenen (2010) only three years of data are available for England. For France, shocks in the labour markets could be argued to be less frequent and one would have not needed to use three years of data to estimate wage equations. In consequence panel data could be used. This is let for future research as so far wage equations were estimated with three years of pooled data. Moreover, for France, as the hospital data is freely available, if access to wage data for the years 2009 to 2011 is granted, it will be possible to estimate two three years pooled wage equations and then use data for hospitals from 2006 to 2008 and from 2009 to 2011.

4.5.2. Summary a) England The results of the literature showed quite clearly the effects of unions/collective agreements on the wage dispersion. It showed that sectors with more collective agreements tend to have a less widespread wage dispersion than sectors without collective agreements. It seems likely that the geographical pattern of pay in the unionised sector exhibits a much flatter distribution than the pattern in the non unionised sector. The sector in which the market sets pay should have pay rates that compensate for geographical variations in amenities and cost-of-living. In the sector in which pay is set by a collective agreement, employers will have more difficulties to attract staff where pay is set at a lower rate than the market one, therefore employers may face more acute shortages of staff. Under the theory of compensating wage differentials, pay gaps should explain a large part of geographical variations in shortages, this will be examined in Chapter 6. The effect of shortages of staff on hospital activity may be dampened by actions by hospitals. In the analyses undertaken in Chapter 6 it is assumed that there is no impact on employers demand for other groups of staff. Later it will be examined whether hospitals may try to alter staff-mix in order to reduce the effect of the shortage of staff. If successful this would be reflected in different 125

Chapter 4

Creation of Standardised Spatial Wage Differentials

proportions of staff between hospitals facing staff shortages and those not. Hospitals facing staff shortages may try to alter the staff-mix by employing other staff for which recruitment difficulties are less acute. Chapter 7 (page 175) will investigate if pay gaps have an impact on hospitals staff-mix. b) France Analysing France is of interest for three reasons. First, there was a shortage of nurses in the year 2000s. This shortage has never been measured but it has been argued that it exists (Depoire, 2011; de Pourvourville, 2002). Secondly, there is a high level of collective agreements among French employees. The wage gaps presented in this chapter show that there are variations in the pay competitiveness for nurses of up to 28%. Thirdly, the French health care sector has different hospital sectors: public, private not for profit and private for profit. There is a direct alternative to a job in the public hospital sector. Consequently, along with the impact of the usual comparator group of non nursing workers in the private non hospital sector, the impact of nurses working in alternative hospital sectors (private not for profit and private for profit) on the public hospital sector will be investigated. As reviewed in Chapter 2 (Section 2.1.3, page 14), the results of the literature on the effects of unions/collective agreements on the wage dispersion are well established. It seems likely that the geographical pattern of pay in the unionised sector exhibits a much flatter distribution than in the non unionised sector. Though in France 95% of employees are covered by collective agreements. Therefore there is no sector with a supposed market clearing system. However, there are variations in the competitiveness of pay for nurses working in the public hospital sector compared to three comparator groups. Those variations might reveal that for the comparator groups it is easier to adjust pay. As a consequence it is expected that the shortage of staff among public hospitals be higher in those départements for which pay competitiveness is low (Chapter 8). The effect of shortage of staff on hospital activity may be dampened by alternative hospital behaviours. Analysis of staff shortage (Chapter 8) assume that there is no impact of employers on the staff shortage. Though, hospitals may try to alter staff-mix in order to reduce the effect of staff shortage. This alternative behaviour should be reflected by different proportion of staff. Hospitals with pay gaps impeding them to recruit may try to modify the skill-mix of their staff by employing other staff for which recruitment difficulties are less acute. Chapter 9 (page 209) will investigate if pay gaps have an impact on hospitals staff-mix. 126

Chapter 5 Empirical Strategies Data and institutional settings for the two countries have different features. Empirical strategies for the two countries are going to differ accordingly. The main advantage of the data for England compared to the one for France is the availability of vacancy rates. When analysing shortages of staff for England, two types of dependent variables can be used: vacancy rates and staff levels. For France only staff levels are available. Staff mix is defined similarly in the two countries. For England, the data contains one variable which is not available for all hospital trusts in the data set. Thus the empirical strategy will consider estimating the models for the reduced sample of hospital trusts. For France, there are three gaps which are defined as the differences between the pay of nurses (assistant nurses or registered nurses) in public hospitals and: −

the pay of a comparator group working in the private non-hospital sector, gaps PNH;

−

the pay of nurses (assistant nurses or registered nurses) in hospitals in the not for profit private sector, gaps PNP;

−

the pay of nurses (assistant nurses or registered nurses) in hospitals in the for profit private sector, gaps PP.

In England, the gap between the hospital sector and private sector is similar to the PNH gap as defined for France.

5.1. Ordinary Least Squares Ordinary Least Squares is the most common technique to estimate linear equations. It consists in estimating an equation such as the following:

127

Chapter 5

Empirical Strategies

Equation 5.1: Equation estimated with OLS

y i=X β+ u i

(5.1)

Where yi is a vector of observations for a population that the researcher would like to explain with the observations contained in the matrix X by estimating the parameters β , and where ui is the vector of residuals. The first key assumption to use OLS is that the error term in the estimated linear equation is uncorrelated with the explanatory variables (Equation 5.2). Where it is not, the estimated parameters are not consistent and thus are biased. An explanatory variable that is correlated to the error term is called endogenous in econometrics (see the following section). Equation 5.2: First key assumption in OLS estimations

E(X ' u)=0

(5.2)

In order to estimate the vector of parameters, the matrix of explanatory variables need to be of full rank (Equation 5.3), this mean that no explanatory variable is a linear combination of the other explanatory variables. If not, then the model is misspecified and cannot be estimated. In practice, the matrix of the explanatory variables is of full rank. Though, sometimes, explanatory variables might be correlated, this will lead to multicollinearity issues which, if strong, might lead to inconsistencies in estimating β . In this thesis, the use of Principal Component Analysis has been undertaken where suspicion or alledged collinearity among explanatory variables was observed. Equation 5.3: Second key assumption in OLS estimations

rank E(x ' x )=K

(5.3)

Under these two requirements, the estimated parameters are identified and consistent. Even if the dependent variable is binary, the estimated parameters are still identified and consistent. The risk is that the predicted values are out of range (above or below the maximum and minimum values taken by the dependent variable). Wooldridge (2002; 455) even argues that the fact that with binary dependent variables the predicted values may be below 0 or above 1 may not matter too much so far as one is only interested in the partial effects around the mean value of the explanatory variable. The only problem in using OLS where the dependent variable is not infinitely distributed is that the error terms cannot be assumed to be homoskedastic. Though 128

Chapter 5

Empirical Strategies

nowadays, it is quite standard to use heteroskedastic matrices of variance covariance to estimate standard errors and pvalues. Usual standard OLS requires homoskedasticity of the error terms. Though, this is not relevant in the case of the thesis as standard robust variance covariance matrices have been used. Heteroskedastic standard errors have been used in the thesis because observations are repeated for all individuals. Fortuitously, the assymptotic variance covariance used to correct for repeated observations also correct heteroskedasticity. Except for the vacancy data which has lots of zeros, OLS is appropriate for the other dependent variables. Normality tests are required only where the number of observations is small. Bardet and Azais (2006) argue that normality tests are required where the number of observations is below a few dozens. Even for England, the number of observations is large enough. In order to test for the consistency of the estimations, staff levels were log transformed and the specification gave similar results to simple OLS. Staff mix was estimated by probit. Yet, probit estimates gave similar results to OLS. In consequence OLS results are presented.

5.2. Endogeneity: a small review of the causes It is suspected that there will be endogeneity in the empirical specifications of models. Endogeneity results when assumption 5.2 does not hold. This happens when there are correlations between the error terms and the explanatory variables. Said otherwise, when an explanatory variable is correlated with an unobserved factor. This section only reminds the reader the different sources of endogeneity. It does not aim to give a full discussion of endogeneity issues. This thesis does not attempt to control for any of the endogeneity. It will only try to discuss it where appropriate and try to assess, where possible, the sign of the corresponding bias. Following Wooldridge (2002) there are three sources of endogeneity. One cannot always distinguish one from the other. Yet this description is useful as it permits the description of the possible sources of endogeneity with their respective econometric specifications. The first source of endogeneity is the omitted variable. One core assumption to estimate, through Ordinary Least Squares (OLS) estimations, the parameters of an equation is that the error term being uncorrelated with the explanatory variables. This uncorrelated assumption can be translated into plain English as the fact that all the information supposed to influence the 129

Chapter 5

Empirical Strategies

dependent variable has been accounted for. This is because the OLS estimation leads to the estimation of the following equation: Equation 5.4: Estimation of the parameters with OLS

̂ β=(X ' X)(−1) X' Y =β+( X' X)(−1) X ' U

(5.4)

In Equation 5.4 one can easily see that β is estimated without any bias if and only if X ' U is null. This translate into the correlation between the error term U and each explanatory variable (contained in the matrix X ) is null. In consequence, if this assumption is not fulfilled: there is still some information that has an impact on the dependent variable that is not accounted for; then the estimated parameters are biased and the sign of this bias is the sign of the correlation between the error term and the explanatory variables. Self selection of individuals is one of the example of the omitted variable problem. The researcher may not be able to measure the decision of individuals to participate in the labour market. The second source of endogeneity, as described by Wooldridge (2002), is the measurement error. This issue is very similar to the precedent one in terms of statistical structure but different in concept. If the researcher cannot measure fully the explanatory variable and that the missing elements that are not measured are correlated with the observed part of the variable then the measured part of the explanatory variable is endogenous. Finally, the dependent and independent variables may be determined simultaneously. This is very common in cross sectional data. Researchers cannot disentangle which of the dependent or the independent are determined first. An example of this occurs when studying in cross sectional data, alcohol addiction and social deprivation, researchers observe that there is a high correlation between the two, though, one cannot tell whether which one is determined first and cause the other one. Only cohort or panel data can help give some evidence about this. In the following section, potential endogeneity of some explanatory variables is discussed.

5.3. Endogeneity a review of the independent variables 5.3.1. Endogeneity of the competitiveness of pay This section specifically reviews the possible reverse causation of the shortage of staff on the competitiveness of pay. Where the shortage in hospitals is small, job seekers may prefer to look 130

Chapter 5

Empirical Strategies

for a job in the comparator sector. If they do so and if it is assumed that the wage rate in the comparator sector is flexible then the standard neo classical theory argues that an extra supply of labour on the labour market depresses the wage rate. In consequence, as the competitiveness of pay of the public sector employees is defined as the public sector SSWD minus the comparator group SSWD, a small shortage of staff is associated with a large wage gap. If estimation is made by OLS the parameters are estimated according to the following equation: Equation 5.5: Estimation of the parameters with OLS

̂ β=(X ' X)(−1) X' Y =β+( X' X)(−1) X ' U

(5.5)

The parameters β are estimated without bias when X ' U is null. This does not hold in the specific case argued in this section as the supply of workers in the labour market of the comparator group is contained in U which is therefore correlated with X. The correlation being positive, an increase in the supply of workers increase the wage gap, the parameter for the competitiveness of pay is therefore underestimated. This does not hold for France, as in France an increase in the supply of workers on a labour market is unlikely to depress the wage rate of this labour market as 95% of the population is covered by a collective agreement. In consequence, X and U as in equation 5.5 is unlikely to be correlated and the parameters are then estimated without bias.

5.3.2. Endogeneity of other independent variables a) Occupancy rates Occupancy rates are variables only defined for France. They are computed with the number of places or beds as the denominator and the number of occupied places or beds as numerator. A PCA is then performed on six of such variables and the first three components are retained. These variables suffer from potential endogeneity as the denominator of occupancy rates is also part of the definition of the denominator of staff levels. Staff levels are staff number divided by an indicator of size. Size was also computed with a PCA performed on the number of places and beds. In consequence occupancy rates are determined simultaneously with staff levels. This endogeneity issue would be dealt with data on vacancy numbers. This data has not made been available by the Ministry of Health.

131

Chapter 5

Empirical Strategies

b) The share of specialist nurses For France, the independent variables will include the share of specialist nurses for each hospitals. The share of specialist nurses is defined as the number of registered nurses holding a specialist qualification in the total number of registered nurses. The share of specialist nurses is likely to be determined simultaneously with the share of registered nurses among the total nursing workforce. This is a case of simultaneity for staff mix thus creating an endogeneity issue for the variable share of specialists nurses. c) Characteristics of neighbouring hospitals Public hospitals in France are competing for staff with private hospitals. The characteristics of private hospitals are included in empirical analyses in order to control for non monetary characteristics of hospitals that would impact on recruitment and retention of staff in public hospitals. Among these characteristics is the staffing levels in private hospitals. Alike the pay gap for England, the supply of staff is an omitted variable in the analyses performed for France. It is argued that in France the supply of staff on the labour market has no or very little impact on the pay gaps because of collective agreements. However, a larger supply of labour on the private hospitals labour market may be due to a low shortage of staff in the public hospital sector. Where there is a larger supply of labour for the private hospitals labour market this may affect hospitals characteristics. They should be able to recruit staff more easily, a better staffing level in private hospitals will then occur which in turn may have an impact on staffing in public hospitals. A better staffing in private hospitals might attract staff to private hospitals as it may increase the working conditions and it might also work as a deterrent as there might be less shortage of staff in private hospitals and thus less opportunities to find jobs. Thus there is a reverse causation of the shortage of staff in public hospitals. There is an omitted variable issue for France which will cause endogeneity of the characteristics of neighbouring hospitals.

5.4. Why not panel estimations? For both countries, the data would seem to allow the use of panel data as more than one observation is available for each hospital. Panel data is quite useful and could remove some of the issues raised by potential endogeneity. Researchers doing empirical economics are interested in the partial effects of some explanatory variables on an explained one. Sometimes, as discussed in the section above, some of the explanatory variables will be correlated with the error term, thus creating endogeneity. One way to deal with endogeneity due to omitted variables is to use panel 132

Chapter 5

Empirical Strategies

data. Panel data has more than one observation for each individual. As presented in Chapters 3 and 4 the data for England and France has more than one observation for each hospital. Then, this data could be used to consistently estimate the parameters if there is some correlation between the explanatory variables and an unobserved effect. This unobserved effect is called individual heterogeneity. It captures all characteristics of hospital that do not change over time. If explanatory variables are correlated with this unobserved effect then there is an endogeneity problem due to omitted variable. By using panel data, it is then possible to estimate consistently the parameters associated with the explanatory variables. In order to do so, all the parameters of not time varying variables will not be estimated. That is where, it is not possible for England and France to use panel data. For both countries SSWDs have been estimated by pooled regressions and therefore do not show any variability over time. The reason for doing this was that by pooling regressions, short term labour shocks would be averaged out. One drawback is that it is then not possible to use the panel feature of the data as the parameter of wage gaps would be lost. Unfortunately, it is in this parameter that lies most of the interest of the this thesis. Moreover, for England, it is not possible to have more than three years of hospital data. Thus extending the number of years is not possible. For France, it would soon be possible to have two sets of three years of data: 2006-2008 and 2009-2011. The hospital and wage data is not yet available for 2011.

5.5. Shortage of staff The rates of pay of nurses in England and France are set nationally and exhibit very little regional variations, beyond higher rates in London and the South East for England and Paris for France. As a result in some areas there may be differences between what hospital trusts in England and public hospitals in France pay and what the private sector pays. These are likely to result in difficulties in attracting and retaining staff and thus shortages of staff.

5.5.1. Measures of shortage of staff Throughout this thesis there are two measures of shortage of staff. One of these statistical series is only available for England: vacancy rates. The other one is available for both England and France: staffing levels. The definition of staffing levels has been explained in Chapter 3 for both countries. The shortage of nursing staff in the NHS can be measured in English data by two statistical series, 133

Chapter 5

Empirical Strategies

vacancy rates and staff levels. If the two measures were in complete harmony, hospital trusts with a more competitive pay would be associated with both lower vacancy rates and higher staff levels all else equal. However it has been argued that the two measures capture different dimension of shortage (Grumbach, 2001). This chapter investigates the role of pay gaps on vacancy rates and nursing staff levels. For France, only the staffing levels variable is available. Local variations in shortages of one staff group may be the results of both the own pay gap and these of other groups of staff. The pay gap of an alternative group of staff may capture hidden characteristics of the workplace. If hospitals have difficulties hiring a specific group of staff, say assistant nurses, because of non competitive pay then the workload for the other, more trained, group of staff may increase, tension may be higher, working conditions may get worse. As a result the pay gap for one group of staff may be an important signal of the unattractiveness of jobs in the same hospital for other groups of staff. The recruitment of other groups of staff may therefore be more difficult because these potential employees are aware of the poorer working conditions. Pay gaps for more trained nursing groups may also reveal the expected wages for lower trained nursing groups once they have achieved higher qualifications. In this chapter the impact of both pay gaps on nursing staff shortages, as measured by vacancy rates and nursing staff levels, shall be investigated. The empirical strategy for vacancy rates (only available for England) will be set as follow, first the gap corresponding to the group of staff for which the vacancy rate is analysed will be tested. Then the gap of the alternative nursing staff is introduced. Other independent variables (types of hospitals, size, foundation status) are introduced in a third model. The last model will test whether the effect of the own gap is affected by the competitiveness of the pay gap for the alternative group of staff. Staffing levels will vary between hospitals for many reasons. Hospitals will differ in size, number of beds, types of procedures undertaken, all of which may affect staff levels. Some procedures are likely to be more labour intensive than others and therefore require higher staff levels. Including controls for the different types of procedures is therefore important. It might appear possible to argue that once such controls have been included the impact of any other variable on staff levels captures factors that prevent hospitals undertaking their activities properly. However, this is a very strong assumption stating that controls for the types of procedures control for all differences between hospitals in the activities they perform. There might be non measured differences in activity. For staff level models, controls to capture the types of hospital trusts and the number of hospitals in each hospital trusts will be included from the start. Staff levels are 134

Chapter 5

Empirical Strategies

defined as a ratio of staffing number divided by size, therefore size cannot be included in controls (see Equation 5.6). First the impact of the gap corresponding to the group of staff for which the staff level is analysed will be tested in a regression model including the control variables. Then, the last model is testing the idea that the effect of the own gap should have different effects where the competitiveness of the pay gap for the alternative group of staff differ. Staffing levels can be seen as a very peculiar way of measuring shortage of staff as it is close to a measure of labour productivity. Productivity usually includes a measure of input and a measure of output. A labour productivity index would measure input by staffing. Here lies the closeness with staffing levels as measured in this thesis. Labour productivity and staffing levels are both ratios and both include labour inputs. Staffing levels divides labour input by size; labour productivity would divide output by labour. If size is adjusted on quality, type of care, case mix, then size can be a sort of output. Helton et al. (2012) uses adjusted occupied beds 1 as the denominator of their labour productivity measure. Therefore, the measure of shortage using staffing levels and controlled by different characteristics of hospitals can be seen as a labour productivity index. If hospitals are doing the same with lower labour inputs, it is not because of shortage of staff but because they are more efficient. This thesis argues the opposite but it is always possible to change the interpretation to one of labour productivity.

5.5.2. Econometric specification Vacancy data for both nursing staff groups has large number of zeros (Table 3.7 in Section 3.4.1, page 71). Modelling such data cannot be used with standard OLS. Elliott et al. (2006, 2009) used a negative binomial2 estimation to allow for the fact that vacancy rates should be estimated with a zero inflated model to take into account the large number of zeros. Zero-inflated models are part of the family of General Linear Models. According to Zeileis et al. (2008) there are three main distributions to estimate such models. Standard Negative-binomial 3 and Poisson distribution are two distributions that assume a dispersion parameter of 1. The Quasipoisson distribution estimate the dispersion parameter from the data, the estimation is the same as the Poisson but inference is then adjusted for dispersion; an alternative would be to use a Poisson distribution and then adjust the inference with a sandwich-adjusted inference. In this chapter the Quasipoisson distribution is used and the dispersion coefficient, when estimated, appeared to be close to 0.054 (it differs 1 2 3

Adjusted by the proportion of the total patients revenues by inpatients revenues. Replicating the results of Elliott (2006) was successful as the difference in the parameter of the gap was less than 10%. The difference in estimation of SSWDs (in this thesis the number of LAD used to estimate SSWDs has been restricted to those with at least 10 nurses) can explain this slight difference. Negative binomial can be estimated with a dispersion parameter which is estimated from the data.

135

Chapter 5

Empirical Strategies

substantially from 1) for the vacancy model estimating Equation 5.9 (see 5.5.3, page 137 for the full specification of this model) for assistant nurses and 0.038 for the same model for registered nurses. Parameters estimated from a Poisson model or a Quasipoisson model are not directly interpreted as for OLS parameters. Cameron and Trivedi (2005; 669) show that a good candidate for an estimate of an average response can be simplified to the product of the estimated parameter with the mean of the dependent variable. The pay gaps are estimated from log wage regressions and caution should be taken when interpreting a change in pay gaps (as explained in 4.3). An area with a gap of 0.01 means that this area compensate public sector workers approximatively 1% more than the national average. A difference of 0.10 between two gaps in two different areas can be interpreted as one area compensate public workers 10% more than the other area. Interpreting the impact of the wage gaps on vacancy rates will focus on the signs, examples of the size are given for some models. A negative sign will mean that when the pay gap for nurses is more competitive, the vacancy rates of nurses is lower. Then a quantitative impact of a change of 10% in the compensation for nurses will be given. If the parameter is negative and equals -2.3, then an increase of 10% in the competitiveness of one of the pay gap will decrease the vacancy rate by -2.3*0.1*2.5=-0.57; -2.3 being the value of the parameter, 0.1 is an increase in ten percent of the gap, and 2.5 is the mean of vacancy rates. Therefore an increase of 10% in the pay gap competitiveness will decrease the vacancy rates by a bit more than half a percent. Equation 5.6 reminds the definition of the staff level. This is the ratio of the number of staff divided by the first component from a Principal Component Analysis (PCA). The PCA is performed for England on four variables each of which measures the size of hospital trusts: the number of beds, the total number of whole time equivalent of assistant nurses, the total number of whole time equivalent of registered nurses and the total number of whole time equivalent of doctors (see Subsection 3.4.2, page 72 and Annex D, page 249). The Principal Component Analysis (PCA) for France is performed on the 6 variables supposed to proxy size: beds for complete4 and weekly stay at hospitals5, number of places for ambulatory surgery 6, day care7, night care8 or at home care9 (see Section 3.5.4, page 84). The higher the value of the component, the higher the size of the hospital. 4 5 6 7 8 9

Complete care is for heavy treatments that require the patient to stay at the hospital for a long period. Week care is for patients who need to come for less than 5 days, usually from Monday to Friday. Ambulatory surgery is for small surgeries that do not need patients to stay overnight. Day care is for patients who come in the morning and leave during the same day. They do not stay overnight. The night care is for patients who need to stay at the hospital overnight but have a daily activity outside the hospital. The at home care is a scheme for people in terminal phases of illness that would require to be at the hospital. Hospitals make the care that they would need available from home.

136

Chapter 5

Empirical Strategies

Equation 5.6: Definition of staff levels, same as Equation 3.1

StaffLevel Nj =

Staff Nj Size j

(5.6)

Where N is the whole time equivalent number of either assistant nurses or registered nurses in hospital j. Staff levels are estimated using standard Ordinary Least Squares (OLS). Considering the distribution of staff levels, a log transformation would have made the different distributions of the staff levels variable look like normal distributions. Though, the estimation of the log transformed staff levels models did not alter the interpretation of the results, in consequence only the simple OLS estimation is presented. As for vacancy rates, the main concern will be the sign of the pay gaps parameters. A quantitative interpretation will also be given but it is more straightforward as the models for staff levels are estimated with a standard OLS. The parameter just need to be multiplied by 0.1 to give the effect of an increase of 10% in the compensation of public sector wages. The data set has repeated observations on each hospital trust. Hospital trusts vacancy rates and staff levels are pooled for three years as explained in Section 3.4 (page 68). With repeated observations for each hospital, the assumption of homoscedasticity of error terms is no longer sustainable. Consequently, the regressions will have the variance covariance matrix modified to take into account the repeated observations on individuals. This modified variance-covariance matrix will give cluster-robust standard errors (Arai 2011). Along with each model is presented the p-value associated to the Chi-square statistic testing the added value of the model compared to the null model (a model with just the intercept), and the pvalue associated to the Chi-square statistic testing the added value of the model compared to the previous embedded model. Those Chi-square statistics are computed by comparing the deviance 10 of the models and tested against a Chi-square distribution.

5.5.3. Vacancy rates models, only for England The first model (Equation 5.7) will test for the impact of pay competitiveness on the vacancy rate of the corresponding group of staff. The gap between the SSWD for assistant nurses (registered

10 The deviance is twice the opposite of the difference between the log likelihood of the reduced model and the log likelihood of the full model. If the Chi Square statistic is significant at the 95% level then the full model has statistically more information than the null or reduced model.

137

Chapter 5

Empirical Strategies

nurses) SSWD and the SSWD for the comparator group will be regressed on the vacancy rate of assistant nurses (registered nurses) in Equation 5.7. Equation 5.7: Vacancy rates simple model with one gap

g Vac Nj = 1∗gapkN  j 1

(5.7)

1

N1

Where Vac j is the vacancy rate of hospital j for assistant nurses (registered nurses), g  is the function used (Quasipoisson distribution),  is the intercept of the model, 1 is the parameter for N1

the gapk which is the difference between the SSWD for the assistant nurses (registered nurses) and the SSWD for the comparator group in the private sector of the area k . It is expected that the gap will be negatively correlated with the vacancy rate. In areas where pay is more competitive (a higher gap), lower vacancy rates are expected. In Equation 5.8 the gap for the other group of staff (either assistant nurses or registered nurses) is introduced. Equation 5.8: Vacancy rates model with two gaps

g Vac Nj = 1∗gapkN  2∗gap Nk  j 1

1

(5.8)

2

Where  2 is the parameter associated with the alternative gap gap kN . It is argued that the pay 2

gap for the alternative group of staff captures some characteristics of the workplace that would not be captured otherwise. In Subsection 2.5.1 (page 43), it was argued that the gap for the alternative group of staff could work in two ways depending on the hierarchy of the nursing staff group. On one hand it was argued that for registered nurses, if hospitals have difficulties hiring assistant nurses because of a non-competitive pay then the workload for registered nurses may increase, tensions may be higher, working conditions may be worse. On the other hand it was argued that assistant nurses may be attracted by areas where the registered nurses pay is competitive as they might look for achieving the registered nurses diploma. Where pay for the alternative group of staff is uncompetitive, the recruitment of the nursing staff may be more difficult.  2 is expected to be negative, because where the pay for the group of staff N2 is less competitive ( gap kN1 gap Nk , a lower gap) staff N1 should not be attracted to this area and the 2

2

vacancy rate for N1 will be higher (larger vacancy rate).

138

Chapter 5

Empirical Strategies

Covariates will be introduced in Equation 5.9. Equation 5.9: Vacancy rates model with two gaps and covariates N1

N1

N2

g(Vac j )=α +β1∗gapk +β2∗gapk + XCj +ε j

(5.9)

X Cj is the matrix of independent variables that constitute the covariates with the corresponding parameters. The effect of size on shortage is not clear cut in the literature. In a review of the literature Tai et al. (1998) do not provide strong evidence of the impact of size on the shortage of staff. Still, two covariates are available to measure the size of hospital trusts for England: the size variable as described in Section 3.4 (page 68) and the number of hospitals within each hospital trust. The type of care has been found to have an impact on the retention of staff (Benedict et al., 1989) in consequence the types of hospital trusts, mental, teaching, acute, “other” and specialist, with acute as the reference group is introduced. Teaching hospitals, because they offer research opportunities might be able to attract more staff and therefore might have smaller vacancy rates. In the literature review by Tai et al. (1998) ownership is discussed and shows mixed results. The foundation status is not a different type of ownership but may imply a different type of management and therefore is included in the analyses. Foundation status are expected to have greater ability to deal with shortages, therefore, the parameter associated to the foundation status dummy is expected to be negative, (foundation trust status is associated with lower vacancy rates). Equation 5.10 distinguishes two slopes for the own gap. It was argued in Subsection that the effect of the gaps might only matter in areas where the gap for the alternative group of staff is competitive. Registered nurses might be attracted by their own gap only in areas where hospitals are well staffed for assistant nurses. This should be captured by a competitive gap for assistant nurses. Therefore, an increase in the competitiveness of the pay gap for registered nurses may have a stronger effect where the gap for assistant nurses is also competitive. Assistant nurses may be attracted by their own gap only in areas where prospects for future expected wages are good. The prospects for future wages might be captured by competitive pay for registered nurses as assistant nurses may wish to become registered nurses. In consequence, the increase of the competitiveness of pay for assistant nurses may matter more in areas where the competitiveness of the pay for registered nurses is good.

139

Chapter 5

Empirical Strategies

Equation 5.10: Vacancy rates model with two slopes for the own gap

g (Vac Nj )=α+β1∗gap Nk ∗Dgap 1N +β2∗gap Nk ∗Dgap N2 +β3∗gap Nk + X Cj +ε 1

1

2

1

2

2

(5.10)

Where Dgap 1N is a dummy variable indicating that the gap for the group of staff N2 is below the 2

national mean and Dgap N2 is a dummy indicating that the gap for the group of staff N2 is above 2

the national mean. The use of two dummies in this setting comes in order to have direct estimates of the effect of the two slopes. Had one dummy been introduced the parameter associated with the interaction of the gap and the dummy would only give the difference in slopes between the gap where the dummy is one and the slope of the gap where the dummy is zero. With this specification, β1 gives the direct effect of the own gap in areas where the alternative gap is below its national mean and β2 gives the direct effect of the own gap in areas where the alternative gap is above its national mean. The gap of the nursing staff group, N2, was argued above to have an impact on the vacancy rates of staff N1. It was argued that the gap for N2 was a signal of the unattractiveness of jobs in the same hospital for group of staff N1. This was argued for the introduction in Equation 5.9 of the gap for the alternative group of staff. In Equation 5.10 this argument is taken one step further by allowing the impact of the own gap to have a different effect whether the competitiveness of the gap for N2 is “good” or “bad”. This dichotomy is a bit crude as it does not say for sure whether hospitals for which the value of the gap for N2 is more competitive than the national mean offer a pay that is above the private sector pay. If the gap for N2 is low it is a signal for unattractiveness of jobs for other groups of staff. An increase in the gap for N1 should have a stronger effect when the gap for N2 is competitive. β2 is the effect of the gap for N1 when hospitals do not have difficulties to recruit N2. β1 is the effect of the gap for N1 when hospitals have difficulties to recruit N2. The effect of the gap for N1 should be stronger when hospitals can also recruit N2, therefore as  1 is expected to be negative, β2 is also expected to be negative but even smaller.

5.5.4. Staff levels models for England and France The first model (Equation 5.11) will test for the impact of the competitiveness of own pay on staffing levels. In Equation 5.11 the gap between the assistant nurses (registered nurses) SSWD and the pay of the comparator group will be regressed on the number of assistant nurses (registered nurses).

140

Chapter 5

Empirical Strategies

Equation 5.11: Staff level model with just one gap and the control variables N1

N1

StaffLevel j =α+β1∗gapk + XCj +ε j

(5.11)

Where StaffLevel Nj is the nursing staff level in hospital j for assistant nurses (registered nurses). 1

N  is the intercept of the model, 1 is the parameter of the gapk where the gap is the difference 1

between the SSWD of the assistant nurses (registered nurses) and the SSWD for their comparator group in the private sector of the area k . For France the gap kN is either gap PNH, PNP 11 or PP for 1

assistant nurses (registered nurses) of the area k . X Cj is the matrix of independent variables with the corresponding parameters. Covariates for England include the same variables as for vacancy rates model but size which is the denominator of staff level. The interpretation of the parameters are expected to be the same as for vacancy rates models. Covariates for France include the activity performed in hospital j, three dummies are introduced for the different activities, medicine, surgery and obstetric (MSO), long stay (LST), and psychiatry (PSY). The proportion of specialised registered nurses among the total number of registered nurses is also introduced and is controlling for more technical procedures being performed in the hospital (the higher the value on this variable the more technical the procedures are supposed to be). Holmås (2002) gave some evidence that worse working conditions as measured by higher occupancy rates are associated with a higher risk of quitting. Three occupancy rates components of the PCA are included, the first one measures higher occupancy as a whole and is therefore expected to be associated with worse working conditions and larger shortage of staff. Though in this setting, a larger occupancy rate may be associated with a higher staff level as hospitals need more staff to deal with higher throughput. The second and third components of the occupancy PCA may lead to different effects as they index for higher values in some of the occupancy variables and lower values for others12. A larger number of equipment, as the first component of the PCA on equipments measures, may mark more complex type of care, the impact of more complex care on shortage of staff has not yet been reviewed and its effect if not known. The second and third components of equipments are measuring different types of activity. In 11 As some départements do not have private hospitals, some SSWDs for private hospitals could not be estimated (see Subsection 4.4.5, page 118). Therefore, some public hospitals are not associated with gaps of types PNP or PP. The estimation of equations with those gaps use an interaction between a dummy variable and the gaps. 12 The second component contrasts hospitals with a high occupancy rate for beds for complete care and night care with those with a high occupancy rate in ambulatory surgery. The third component contrasts hospitals with high occupancy rate on beds complete care, ambulatory surgery and day care opposed to those with a high occupancy rate on at home care.

141

Chapter 5

Empirical Strategies

consequence, it is unlikely that their effect can be predicted. Finally, the component of size included in this regression is measuring larger sizes for some sort of care and smaller for some other sort. Thus it is not possible to predict the impact of size. In Holmås (2002), larger health care facility face a higher risk of their staff quitting their jobs. Equation 5.12 tests the introduction of the gap for the other group of staff. Equation 5.12: Staff level model with two gaps and the control variables N1

N1

N2

StaffLevel j =1∗gapk  2∗gapk XCj  j

(5.12)

Where 2 is the parameter associated with the alternative gap, gap Nk . It has been argued above 2

that the pay gap of the alternative group of nursing staff may capture either characteristics of the job and the expected pay once lower skilled staff have achieved higher training. The greater the wage gap for the alternative group of staff, the larger the nursing staff levels. In France, there are neighbouring hospitals which may attract staff more effectively and this may impact the recruitment of staff. Equation 5.13 presents models introducing the characteristics of surrounding hospitals. Equation 5.13: Staff level model with surrounding hospitals characteristics, only for France N1

N1

N2

StaffLevel j = 1∗gapk  2∗gapk ZDeptPNP jZDeptPP jXCj

(5.13)13

Where PNP stands for private not for profit and PP for private for profit, and Z is the matrix for the characteristics variables of neighbouring hospitals. ZDeptPNP j and ZDeptPP j are the vector of characteristics of surrounding private not for profit hospitals and private for profit hospitals respectively. Surrounding hospitals may attract staff to their premises therefore making difficult for public hospitals to hire any staff. While pay gaps may capture some of this attractiveness, here is tested any non monetary effects of surrounding hospitals. The Z vectors are average of the following variables at the département level for each type of private hospitals: the assistant nurses and registered nurses staff levels, a participation to public service dummy 14 and the activity variables: the proportion of hospitals with wards in the different activities medicine, surgery and 13 Results do not present the parameters of the Z variables. The effect of the Z variables is not of interest. Their inclusions tries to capture non monetary attractions of hospitals. Therefore the effect as a whole of the Z variables may be assessed by looking at their impacts on the parameters of the other variables in the models and at the Chi-square statistic that compare this model with the previous embedded. 14 Only for private not for profit hospital, the number of private for profit hospitals participating in the public service is too low to include a dummy of this type (2.5% of private for profit hospitals are participating in the public service).

142

Chapter 5

Empirical Strategies

obstetric (MSO), long stay (LST), and psychiatry (PSY), the proportion of specialised registered nurses among the total number of registered nurses is also introduced to control for more technical procedures being performed in the hospital. The component of equipments (EQ1, EQ2 and EQ3), the components of size and the three components of the occupancy rates (OccRate1, OccRate2 and OccRate3) are also included. The coefficients for the Z variables are not presented. Their effect is not sought to be understood, but only to control for characteristics of neighbouring hospitals. Their effects will be assessed with the p-value associated to the Chi-square statistic and with possible changes in the other parameters.

5.5.5. Staff levels models with more than one slope for the effect of gaps As explained in Subsection 2.5.1 (page 43) which presented the general idea, Section 4.3 (page 110) which presented the same idea for England and in Subsection 4.4.7 (page 120) for France, gaps for registered nurses may have a different impact on the level of registered nurses for different values of the gap for assistant nurses and vice versa, gaps for assistant nurses may have a different impact on the level of assistant nurses for different values of the gap for registered nurses. It was already argued that the gap for assistant nurses in registered nurses models is expected to have a positive impact on the registered nurses staff levels. This was tested in Equation 5.12. This Subsection argues that there might be an interaction effect, the effect of the gap for registered nurses will be stronger when the gap for assistant nurses is more competitive. This is due to the fact that the pay for registered nurses might play a bigger role in areas where the pay for assistant nurses is competitive as it might indicate a lesser workload. Symmetrically, the pay for assistant nurses might be play a bigger role in areas where the pay for registered nurses is more competitive as it might mark higher expected earnings when assistant nurses climb the ladders to become registered nurses. This consists in estimating different slopes for the gap for registered nurses depending on the values of the alternative gap. Equations 5.14 and 5.15 present such models. Depending on the types of gaps being tested the models will differ. For English models, for registered nurses models for France with gaps PNH, PNP or PP and for assistant nurses models for France with gaps of type PNH the two slopes effect will consist in testing the effect of the gap whether the alternative gap is above its national mean or not (Equation 5.14). Gaps of type PNH and gaps for England are similar. These gaps are using workers in the private non hospital sector as a comparator group. Estimation of these gaps were only done with the fully stratified method for France and values of these gaps do not tell whether the public hospital 143

Chapter 5

Empirical Strategies

sector offer a better wage or not. The national mean is chosen as an arbitrary cutting point value. It is expected that where the pay for the alternative group of staff is above the national mean the larger the impact of the pay gap for the investigated nursing group of staff. Gaps PNP and PP for assistant nurses are positive and significantly different from 0 therefore the two slopes for the gap for registered nurses will refer to an arbitrary value of the gap for assistant nurses, the national mean was chosen. It is expected that where the pay for assistant nurses is more competitive (above the national mean), the registered nurses gap will have a stronger impact. For assistant nurses models with gaps PNP the two slopes effect will consist in testing the effect of the assistant nurses gap whether the registered nurses gap is negative and significantly different from 0 or not significantly different from 0 (Equation 5.14). Where the pay gap for registered nurses is not significantly different from 0 it is expected that hospitals should be able to recruit more easily registered nurses. Where registered nurses are recruited more easily, the effect of a more competitive assistant nurses gap is supposed to be stronger. For assistant nurses models with gap of type PP, a different model has to be used as the effect of the gap will have three slopes following three dummies: the registered nurses gap is significantly different from 0 and negative, not significantly different from zero; positive and significantly different from zero (Equation 5.15). It is expected that the effect of the wage gap for assistant nurses will increase where the wage gap for registered nurses is more competitive. Equation 5.14: Staff level model with different slopes effect for gaps for England and for the following gaps for France: PNH, PNP and PP for registered nurses and for gaps PNH and PNP for assistant nurses N1

N1

N2

N1

N2

N2

StaffLevel j =α+β1∗gapk ∗Dgap1 +β2∗gapk ∗Dgap 2 +β3∗gapk C ( + ZDeptPNP j + ZDeptPP j )+ X j +ε

(5.14)15

N2

For the assistant nurses model with gaps of type PNP, Dgap1 is a dummy variable indicating that N2

the registered nurses gap is negative and statistically different from 0 and Dgap 2 is a dummy variable indicating that the registered nurses gap is not statistically different from 0. For the N2

other models following this equation Dgap1 is a dummy variable indicating that the gap for the N2

group of staff N2 is below the national mean and Dgap 2 is a dummy variable indicating that the 15 The Z variables are only for French models.

144

Chapter 5

Empirical Strategies N2

gap for the group of staff N2 is above the national mean. Where Dgap 2 takes the value one, for all models, the pay gap for the group of staff N2 is expected to be more competitive. Where the gap for N2 is competitive it is expected that hospitals in those areas have less difficulties to recruit N2. Where hospitals have less difficulties recruiting N2, a more competitive pay for N1 may have a stronger effect than in areas where it is difficult to recruit N1 (Subsection 2.5.1 page 43). β2 is the effect of the gap for N1 when the gap for N2 is competitive.  1 is the effect of the gap for N1 where the gap for N2 is uncompetitive. The effect of the gap for N1 should be stronger where the gap for the alternative group of staff is competitive, therefore as  1 is expected to be positive,  2 is also expected to be positive and larger than  1 . Equation 5.15: Staff level model with different slopes effect for the gaps of type PP for assistant nurses (French model) N1

N1

N2

N1

N2

N1

N2

StaffLevel j =α+β1∗gapk ∗Dgap1 +β2∗gapk ∗Dgap2 +β3∗gapk ∗Dgap 3 +β4∗gapNk + ZDeptPNP j+ ZDeptPP j + XCj +ε 2

(5.15)

N2

Where Dgap1 is a dummy variable indicating that the gap for registered nurses is negative and N2

significantly different from 0, Dgap2 is a dummy variable indicating that the gap for registered N2

nurses is not statistically different from 0 and Dgap3 is a dummy variable indicating that the gap for registered nurses is positive and significantly different from 0. If the gap for registered nurses is positive it is expected that hospital trusts in those regions have less difficulties to recruit registered nurses.  1 is the effect of the gap for assistant nurses where the gap for registered nurses is negative. β2 is the effect of the gap for assistant nurses where the gap for registered nurses is not statistically different from 0. β3 is the effect of the gap for assistant nurses where the gap for registered nurses is positive. The effect of the gap for assistant nurses should be stronger where the gap for registered nurses is positive. Therefore  3 is expected to be positive and larger than  1. The effect of  2 is also expected to be positive, in regions where the pay gap for registered nurses does not impede the recruitment of registered nurses, the gap for nurse assistants should be slightly stronger than in areas where the gap for registered nurses is significantly negative.

145

Chapter 5

Empirical Strategies

5.5.6. Complexity index estimations Equations 5.7, 5.8, 5.9, 5.10, 5.11, 5.12 and 5.14 for England will also be estimated for the subsample of hospitals for which there is a complexity index (as described in 3.4.2 Table 3.13, page 76). This adds one covariate to all English models, the complexity index itself and because such an index is not available for non mental health hospital trusts, it removes the mental hospital trust dummy. The number of observations will be reduced to 477 (observations with a complexity index and SSWDs, see Annex G, page 253). These results are presented in Annex H.1 (page 254) for vacancy rates and H.2 (page 258) for staffing levels. Summary of results are provided in the text.

5.6. Staff mix models The estimation strategy is as follows: first the two gaps will be introduced in models of staff-mix. Then control variables will be introduced. In the models for shortage of staff a specification testing for the impact of the gap for the alternative group of staff on the parameter of the own gap was tested. A similar strategy is also experimented here with testing two parameters for each gap depending on the competitiveness of the alternative gap for England and a PCA is performed on gaps for France and the components of this PCA are used in regressions. Hospitals performing different types of activity may have different strategies and abilities to cope when facing shortage of staff, therefore the models for England will allow for different slopes for each of the different types of hospitals. A similar specification for France has not been experimented for lack of information indicating the types of hospitals.

5.6.1. Econometric specification Staff-mix is defined as a proportion. Proportions of staff were computed with the whole time equivalent of each group of staff in each hospital see Equation 3.2 or Equation 5.16. Equation 5.16: Definition of staff-mix, same as Equation 3.2 RN j

Staffmix =

Staff RN j

(5.16)

RN Staff AN j Staff j

Where AN stands for Assistant Nurses, and RN for Registered Nurses. Staff

RN j

is the whole time

equivalent number of registered nurses working in hospital j. The proportion of assistant nurses is one minus the registered nurses proportion.

146

Chapter 5

Empirical Strategies

Proportion models can be estimated with probit link functions. As the proportion of staff is strictly bounded between 0 and 1, OLS has been preferred. Estimations using the probit link function do not change fundamentally the results though the parameters would have been more difficult to interpret. As for the previous chapter, Chi-square statistics testing for the added value of models with null and previous embedded models are computed 16.

5.6.2. Staff-mix models In Equation 5.17 the impact of the two wage gaps on staff-mix is investigated. Equation 5.17: Staff-mix simple model with just the two gaps RN

AN

RN

Staffmix j = 1∗gapk  2∗gapk  j

(5.17)

RN

Where Staffmix j is the registered nurses staff-mix of hospital j.  is the intercept of the model,

2 ( 1 ) is the parameter for the gapkRN ( gapkAN ) which is the difference between the SSWD of the registered nurses (assistant nurses) and the one for the comparator group in the private sector of the area k . It is expected that the effect of the registered nurses pay should be positive. Where the pay gap for registered nurses is more competitive, hospitals have a larger share of registered nurses in their workforce. Then covariates will be introduced (Equation 5.18). Equation 5.18: Staff-mix model with covariates RN

AN

RN

C

Staffmix j = 1∗gapk  2∗gapk X j j

(5.18)

Covariates, for England, include the size variable constructed as described in Section 3.4 (page 68), the number of hospitals within a trust, whether or not hospital trusts have achieved the foundations status and the types of hospital trusts, acute, mental, teaching, “other” and specialist, with acute being the reference category. Achieving the foundation status meant meeting a set of criteria that included financial targets and quality of care. Because they have met

16 Along with each model is presented the p-values associated to the Chi-square statistic testing the added value of the model compared to the null model (a model with just the intercept), and the p-values associated to the Chi-square statistic testing the added value of the model compared to the previous embedded model. Those Chi-square statistics are computed by comparing the deviance of the models and tested against a Chi-square distribution.

147

Chapter 5

Empirical Strategies

these criteria and succeeded in achieving the foundation status, these hospitals should have underlying features that make them more able to achieve quality of care and financial health. It is then expected that they may have some ability in recruiting staff that other trusts do not have. Covariates, for France, include the activity performed in hospital j, three dummies are introduced for the different activities, medicine, surgery and obstetric (MSO), long stay (LST), and psychiatry (PSY). The proportion of specialised registered nurses among the total number of registered nurses is included as a control for activity requiring higher skilled labour. It is expected that a larger share of the number of specialised registered nurses within the registered nurses staff will be associated with higher proportion of registered nurses within the total registered nurses plus assistant nurses. The component of equipments (EQ1, EQ2, EQ3) are introduced to control for higher technical activities. The first component would be expected to be associated with larger proportion of registered nurses as in Pope & Menke (1990) and Acemoglu & Finkelstein (2008). In those articles it was found that the higher the technical procedures the more skilled intensive the workforce is. The second and third components effects are difficult to assess as larger values on those components do not index more technical hospitals but different technicalities. The first component of size is supposed to control for bigger institutions and as in Czuber-Dochan et al. (2006), bigger institutions are supposed to have more skilled staff, thus higher proportions of registered nurses. Where the occupancy rates are higher, it is expected that hospitals have a more sustained activity, this should have an impact on the proportion of staff, though it can be argued that hospitals with more sustained activity might have more of the better qualified staff so that they can deal with difficult cases at all times. It can also be argued that hospitals with greater occupancy rates need more hands and that lower qualified nursing staff are needed. The three components of the occupancy rates (OccRate1, OccRate2 and OccRate3) are then introduced.

5.6.3. Specific staff mix models for each country a) England In areas where the pay for assistant (registered) nurses is more competitive, it would be expected that the effect of the gap for registered (assistant) nurses would be stronger. Equation 5.19 will introduce two slopes for each gap, one when the alternative gap is competitive and one when the alternative gap is not competitive. The model is specified as follow (Equation 5.19).

148

Chapter 5

Empirical Strategies

Equation 5.19: Two parameters for each gap AN RN AN RN Staffmix RN j =α +β1∗gapk ∗Dgap 1 +β2∗gapk ∗Dgap 2 AN RN AN C +β3∗gapRN k ∗Dgap 1 +β4∗gapk ∗Dgap2 +X j +ε

(5.19)

 1 (  2 ) is the effect of the gap for assistant nurses (AN) when the gap for registered nurses is RN

below (above) its national mean. Dgap1 is the dummy indicating that the registered nurses (RN) RN

gap is below its national mean and Dgap2 is the dummy indicating that the registered nurses (RN) gap is above its national mean.  3 (  4 ) is the effect of the gap for registered nurses when the AN

gap for assistant nurses is below (above) its national mean. Dgap 1 is the dummy indicating that AN

the assistant nurses gap is below its national mean and Dgap 2 is the dummy indicating that the assistant nurses gap is above its national mean. In Subsection 2.5.2 (page 45) it was argued that the ability to change skill mix was linked to the ability to change the production function. Changing the production function implies reorganising care in order to have some staff do some tasks which they usually do not perform. This is possible when the skills that are necessary to perform the tasks being reallocated are shared by the two nursing groups. If so, these hospitals might be able to substitute one group of staff with the other. In order to test whether the impact of the pay competitiveness is different for hospitals with different production functions, Equation 5.20 introduces interactions between the different types of hospitals and the pay gaps. Equation 5.20: Staff-mix model with interactions with the types of hospitals AN RN C Staffmix RN j =α +β 1j∗Type j∗gapk +β 2j∗Type j∗gapk + X j+ ε

(5.20)

The effect of the assistant nurses (registered nurses) gap for a one type of hospital trusts is β1j (

β2j ) where j is the index for this type of hospital trust. Hospital trusts which have achieved the foundation status have met a set of criteria on both quality of care and financial health. Though foundation hospitals cannot alter the pay they offer to staff. However, where the pay is more competitive, foundation hospitals may have more abilities to recruit staff. Introducing a slope for each gap for hospital trusts with the foundation status would allow to test whether they have more ability to alter their staff-mix where the pay

149

Chapter 5

Empirical Strategies

they offer is relatively more competitive. Equation 5.21 will test for a specific effect of the pay gaps for foundation status. Equation 5.21: Specific effect of the foundation status AN AN StaffmixRN j =α +β1∗gapk ∗DFound1 +β2∗gapk ∗DFound2 RN C +β3∗gap RN k ∗DFound1+β4∗gapk ∗DFound2+ X j +ε

(5.21)

DFound 1 is the dummy indicating hospitals with the foundation status. DFound 2 is the dummy indicating hospitals with the foundation status.  2 (  1 ) is the effect of the gap for assistant nurses (AN) for foundation trusts (for non foundation hospital trusts).  4 (  3 ) is the effect of the gap for registered nurses (RN) for foundation trusts (for non foundation hospital trusts). Hospital trusts which have achieved the foundation status are expected to have more ability to hire staff, hospital trusts with the foundation status and in an area with more competitive wage gap should have even more ability to substitute assistant nurses for registered nurses. Equations 5.17, 5.18, 5.19, 5.20 and 5.21 for England will also be estimated for the sub-sample of hospitals which have a complexity index (Table 3.13, page 76). This will add one covariate, the complexity index itself and remove the mental hospital trust dummy, the number of observations will be reduced to 477 (observations with a complexity index and SSWDs, see annex H, page 254). Summary of results are provided in the text. b) France In France there are direct alternative employers for nursing staff. Public hospitals face the competition for staff of private not for profit and private for profit hospitals. These hospitals have different features and may offer different working conditions. Except from pay, it is not possible to control for the working conditions that this hospitals may offer. The following equation will attempt to control for these alternative employers. Equation 5.22 presents models introducing the characteristics of surrounding hospitals. Equation 5.22: Staff-mix model with surrounding hospitals characteristics, for France only N2

N1

N2

Staffmix j = 1∗gap k  2∗gapk ZDeptPNP jZDeptPP jXCj

(5.22)

Where ZDeptPNP j and ZDeptPP j are the vectors of characteristics of surrounding private not for profit and for profit hospitals respectively. Surrounding hospitals may attract staff to their 150

Chapter 5

Empirical Strategies

premises therefore making it difficult for public hospitals to hire any staff which in turn should drive hospital behaviours in changing their staff-mix. Pay gaps will capture some of this attractiveness but characteristics of surrounding hospitals are indeed here to test for non monetary effects. Within the Z vectors are included the assistant nurses and registered nurses staff levels, a participation to public service dummy 17 and the activity variables: the proportion of hospitals with wards in the different activities medicine, surgery and obstetric (MSO), long stay (LST), and psychiatry (PSY), the proportion of specialised registered nurses among the total number of registered nurses is also introduced to control for more technical procedures being performed in the hospital. The component of equipments (EQ1, EQ2 and EQ3), the components of size and the three components of the occupancy rates (OccRate1, OccRate2 and OccRate3) are included. The coefficients for the Z variables are not presented. Nothing can be said a priori on whether hospital characteristics would attract more assistant nurses or registered nurses. Their effect is not sought to be understood, the main reason for including them is that they might control for characteristics of neighbouring hospitals. Their effects will be assessed with the pvalues associated to the Chi-square statistic and with possible changes in the parameters of other variables. Compared to the English corresponding specification (5.19, page 149) and to the specification for shortage of staff models in France (equations 5.14 and 5.15, pages 144 and 145 respectively), the types of interactions being tested here is of a different nature. It implies extracting components with a Principal Component Analysis from the gaps for the two nursing staff. The models with PCA components cannot bring any information compared to the corresponding model with the original gaps as it is the exact same information that is put in the model. The difference is that the interpretation of the PCA will bring some better understanding of the effect of the interaction between the gaps for the two nursing staff. PCA will gather in the first component the information that is common to the gaps for the two nursing groups. The dimensions that differ will be gathered in the second component. Moreover, as for other independent variables for which there is a suspicion of multicollinearity, PCA will remove the correlations between the gaps 18. An alternative to this is to introduce the two variables and their interaction terms. This strategy is quite common in applied economics. However, it does not remove the collinearity between the two variables and once one has understood PCA, interpreting the PCA component is much simpler as it does not involve the 17 Only for private not for profit hospital, the number of private for profit hospitals participating in the public service is too low to include a dummy of this type (2.5% of private for profit hospitals are participating in the public service). 18 Introducing the two variables with the interaction term will not do that, the multicollinearity will remain.

151

Chapter 5

Empirical Strategies

reasoning at fixed value for one variable when the other one varies. When introducing the two variables and the interaction terms, the effect of each variable has to be presented in a different table or with a graph that will tell the effect of the variable for a certain number of values of the other variable. With the PCA there are only two parameters. The correlations between the gaps for the two nursing staff are high for France. Table 5.1 presents those correlations for the different types of gaps. This was not observed for England 19. Table 5.1: Correlations of the gaps for the two nursing groups Registered Nurses PNH PNH Assistant Nurses

PNP

PP

0.658

PNP

0.452

PP

0.567

Table 5.2 presents the correlations between the original gaps and the components for each type of gap. Three PCAs are performed, one for each type of gap: PNH, PNP and PP. The first component of each PCA of the different types of gaps vary positively with the two gaps, the second component, contrasts the two gaps. The results of the PCA are very much as what would be expected, one component control for the common dimension of the two gaps and the second component opposes the remaining information of the gaps for the two nursing groups. Therefore the second component tells that for some areas, one gap can be very competitive while the other is uncompetitive. It is expected that in regressions, the second component should have an impact on the strategic behaviours of hospitals. The second components captures the trade off in terms of prices that hospitals can make between the two nursing staff. Table 5.2: Correlations of the original gaps with the components of the PCAs PCA PNH Original Gaps

PNP

2nd Comp.

Assistant Nurses

0.89

-0.46

Registered Nurses

0.89

0.46

Registered Nurses

2 Comp.

0.91

-0.41

0.91

0.41

nd

1 Comp.

2 Comp.

0.85

-0.52

0.85

0.52

st

PNP Assistant Nurses Registered Nurses PP

PP 1 Comp.

PNH Assistant Nurses

1 Comp. st

nd

st

Equation 5.23 and 5.24 are the same as equations 5.18 and 5.22 respectively, except that instead of 19 The corresponding correlation to the gap PNH for England is around 0.25.

152

Chapter 5

Empirical Strategies

the gaps, the PCA components are introduced. The results of the other independent variables will be exactly the same as the ones estimated with the original gaps as the exact same amount of information will be introduced in the two specifications. Equation 5.23: Staff-mix model with the two components of the PCA for gaps and the control variables N2

Staffmix j = 1∗PCA1k 2∗PCA2k  XCj j

(5.23)

Equation 5.24 introduces the Z vectors. Equation 5.24: Staff-mix model with the two components of the PCA for gaps and surrounding hospitals characteristics

Staffmix Nj =α+β1∗PCA1 k +β2∗PCA2 k + ZDeptPNP j + ZDeptPP j + X Cj +ε 2

(5.24)

It is expected that where the pay for one group of staff is relatively better than the pay for the other group of staff (parameter for the second component of the PCA) hospitals will use more of the group of staff for which the pay is relatively more competitive. The first component of the PCA captures more competitive pay for the two nursing staff, thus there are no expectations for its effect on the proportion of staff.

5.7. Conclusion Different institutional settings and different data for the two countries lead to different models. One aspect that is available for England and not for France is the availability of vacancy rates. This the preferred measure of shortage of staff. Still it can have some measurement errors and may not accurately measure shortages. One author (J Buchan, 2002) has argued for England that it may underestimate shortage of staff as hospital trusts may know that they will not be successful in hiring certain groups of staff, they might, the argument follows, do not even loose their time advertising for it. In consequence, the measure of shortage of staff provided by vacancy rates may be underestimated. Three types of gaps have been computed for France, the difference between the public sector pay with the private non hospital sector, the private not for profit hospital sector and the private for profit hospital sector. In consequence each model is repeated three times. Some extra control variables are added for France in order to attempt to control for the differing working conditions the alternative employers may offer. Finally different strategies are used to model the skill mix. 153

Chapter 5

Empirical Strategies

For England, the empirical strategy will focus on the different abilities hospitals may have to alter their staff mix as a consequence of the different care they perform or because they may have different juridical structures (Foundation status). For France, such measures are not available and an innovative approach to control for the interactions between gaps is tested.

154

Chapter 6 Shortage of Staff in English Hospitals This chapter is the first of the four empirical chapters of this thesis. The rates of pay of nurses in England are set nationally and exhibit very little regional variations, beyond higher rates in London and the South East. As a result in areas of England there may be differences between what hospital trusts pay and what the private sector pays. These are likely to result in difficulties in attracting and retaining staff and thus shortages of staff. The shortage of nursing staff in the NHS can be measured in English data by two statistical series, vacancy rates and staff levels. If the two measures were in complete harmony, hospital trusts with a more competitive pay would be associated with both lower vacancy rates and higher staff levels all else equal. However it has been argued that the two measures capture different dimension of shortage (Grumbach, 2001). This chapter investigates the role of pay gaps on vacancy rates and nursing staff levels. Local variations in shortages of one staff group may be the results of both the own pay gap and these of other groups of staff. The pay gap of an alternative group of staff may capture hidden characteristics of the workplace. If hospitals have difficulties hiring a specific group of staff, say assistant nurses, because of non competitive pay then the workload for the other, more trained, group of staff may increase, tension may be higher, working conditions may get worse. As a result the pay gap for one group of staff may be an important signal of the unattractiveness of jobs in the same hospital for other groups of staff. The recruitment of other groups of staff may therefore be more difficult because these potential employees are aware of the poorer working conditions. Pay gaps for more trained nursing groups may also reveal the expected wages for lower trained nursing groups once they have achieved higher qualifications. In this chapter the impact of both pay gaps on nursing staff shortages, as measured by vacancy rates and nursing staff levels, shall be investigated.

155

Chapter 6

Shortage of Staff in English Hospitals

6.1. Results The results of estimating the models for shortage of staff presented in Chapter 5 (Section 5.5, page 133) are presented in the tables with significant results highlighted with stars, where one star refers to a result significant at 10%, two stars at 5% and three stars at 1%. Annex 1 (page 173) gives the definition of the variables used in this chapter.

6.1.1. Assistant nurses vacancy rates, pay gaps, equations 5.7 and 5.8 The following table presents the results of estimating equations 5.7 and 5.8 for assistant nurses. Table 6.1 shows that the gap of the assistant nurses is significant in Equation 5.7 but not in Equation 5.8. Equation 5.7 represents the most simple specification. In Equation 5.7 the gap is significant at the 5% level, though once the gap for registered nurses (Equation 5.8) is introduced this is no longer the case. The effect of the assistant nurses gap is hidden by the effect of the alternative group of staff, registered nurses. It would appear that the competitiveness of the registered nurses pay plays a more important role in explaining the vacancy rates of assistant nurses than does the competitiveness of the assistant nurses pay. Assistant nurses are attracted to regions where they can expect a better pay once they have achieved the registered nurses grade. The two models are significant, and Equation 5.8 adds information compared to Equation 5.7 (last row of Table 6.1). Table 6.1: Assistant nurses vacancy rates, equations 5.7 and 5.8 (627 obs.) Eq. 5.7

Eq. 5.8

Estimate (Std Err)

Pvalue

Estimate (Std Err)

Pvalue

Intercept

-4.429 (0.13)