Micevska_GREDEG Workshop

Employment ratio (number of employed as % of population aged 15-59) ..... tandem in all communities, although with different intensities: the correlation ... dummy variable indicating the initial status of employment of the individual and the.
292KB taille 4 téléchargements 427 vues
The Overlooked Russian Marriage Drama

by

Maja Micevska University of Klagenfurt; ESCE Economic and Social Research Center, Cologne and Eisenstadt

and Oded Stark Universities of Bonn, Klagenfurt and Vienna; Warsaw University; Warsaw School of Economics; ESCE Economic and Social Research Center, Cologne and Eisenstadt

Mailing Address: Maja Micevska Department of Economics University of Klagenfurt Universitaetsstrasse 65-67 A-9020 Klagenfurt Austria E-mail Address: [email protected] Fax number: + 43 463 2700 4148

The Overlooked Russian Marriage Drama

Abstract

During the decade of 1989-1998, Russia experienced remarkable political and economic shake-ups. These transformations have been researched extensively. The concomitant social and demographic changes received lesser attention, even though, in more than one way, they were not less far reaching. In this paper we study what we refer to as a “drama:” the powerful transformation of the Russian marriage scene. We document the significant decline in the incidence of marriages, and we seek to explain the decline. We consider four appealing explanations: increased cost of forming marriages, a shift toward western norms, a bleak perception of what the future holds, and a potential correlation in economic outcomes across would-be spouses. Our analysis suggests that the higher cost of forming marriages and the positively correlated incomes of individuals are the reasons for the dramatic decline in, what we term, the “couple formation.” We reflect on the long-term repercussions of the changed marriage landscape: while at the outset we explore the repercussions of a crossover from the economic sphere to the marriage scene, in our conclusions we refer to a crossover from the marriage arena back onto the economic domain.

JEL classification: J12; J64; P20 Keywords: Russia; Decline of the rate of marriage; Unemployment; Cost of marriage; Correlation of incomes

1. Introduction In spite of high hopes that Russia will be transformed rapidly into a market economy, the transition process in the 1989-1998 period encountered formidable difficulties. GNP per capita declined by more than 40 percent. Traditional public transfers and support systems were disrupted, and economic and labor-market uncertainties increased considerably. One of the most pressing problems that “the new Russia” faced was rising unemployment: from 1989 to 1998 the unemployment rate rose from 0 to 13 percent. The young have been particularly affected: in 1998 the unemployment rate among the 15-24 year-olds reached 27 percent. During the same (1989-1998) period, Russia also experienced a substantial decline in the marriage rate: the crude marriage rate (marriages per 1,000 of the mid-year population) dropped by 38 percent, almost in unison with the drop in the GNP per capita. The decline in the Russian marriage rate was nearly three times as large as the average decline in marriage rates recorded in the OECD countries during the same period. Sharp declines in marriage rates occurred also in other states of the former Soviet Union as well as in many countries in Central and Eastern Europe that underwent the transition process.1 Since 1999, the social, economic, and political conditions in Russia have improved substantially. Russia has achieved a fair measure of macroeconomic stability and the growth rate of the economy has exceeded 4 percent per annum. The labor market has witnessed a slight decline in the unemployment rate and a decrease in labor-market uncertainties. The decline in the marriage rate has been brought to a halt, and by 2003 the marriage rate has even posted a recovery of about 33 percent in comparison with the 1998 minimum. These developments have led observers of the Russian socioeconomic scene to infer that the drastic decline in the marriage rate in the early and mid 1990s and the subsequent 1

Even the steepest reductions in marriage rates in EU countries in the 1970s and 1990s pale in comparison with the reductions experienced by many transition countries.

1

partial recovery since 1999 were intrinsically connected to, indeed trekked, the concomitant political and economic processes. Considerable difference of opinion exists, however, with regard to the mechanisms that link the social, economic, and political transformations to the variation in the incidence of the formation of couples.2 Some observers advanced a “cost argument:” falling income levels, increasing unemployment, and rising housing costs have prompted the young to postpone or avoid getting married. This argument is akin to the declining “marriageability” argument of Wilson (1987) who uses the status of employment as a proxy for a man’s ability to support a family, hence a man’s “marriageability.”3 Other observers viewed the transformation as a process of convergence toward a “western” social and economic preference structure pertaining to marital formation, a process that is characterized by a gradual rise of the age at first marriage, and by an increased non-marital cohabitation.4 A natural addition to this catalogue of reasons is a perception of what the future holds. While marriage is a “final good,” it is also an “intermediate good” in the production of children. Bearing children is inherently a statement about the future. When the perception of the future is bleak - as it was in Russia during most of the 1990s - the desire to have children is attenuated and the derived demand for marriages is affected adversely. Once the economic conditions begin to improve, optimism ensues, which in turn translates into an increased inclination to form marriages. Indeed, as already mentioned, the marriage rate in Russia has risen since 1999.

2 In this paper we use the term couple formation as an all-encompassing term of marriages and non-marital cohabitation. 3 Wilson’s index considers a man who is currently employed as “marriageable,” and a man who is currently unemployed as “unmarriageable.” Several authors have elaborated on Wilson’s hypothesis that a decline in the incidence of marriage is attributable to a decline in the supply of marriageable males (Mare and Winship, 1991; Wood, 1995; Brien, 1997; Neal, 2001). 4 An overview of the “cost argument” and of the shift toward “western” norms is provided by Kohler and Kohler (2002).

2

To the three possible explanations for the decline and the subsequent recovery in the Russian marriage rate, viz., a “cost story,” a shift toward “western” norms, and a “perception of the future” line of argument, we add a fourth explanation that appears to render the decline in the incidence of marriages even more perplexing. Marriage has long been recognized to constitute a risk-pooling device (Kotlikoff and Spivak, 1981; Rosenzweig and Stark, 1989; Weiss, 1997; Ogaki and Zhang, 2001; Hess, 2004).5 Presumably, the weaker the market in mitigating risks, the more valuable the role that one’s spouse can play as an insurance provider of a last resort. While lowered incomes reduce the financial ability to form and sustain marriages, a reduction in incomes in a setting in which the state relinquishes many of its transfer and support roles could render the institution of marriage an increasingly appealing substitute. In the context of Russia in the 1990s, this consideration prompts us to pose the following question: in the face of virtual absence of market institutions, rising economic and labor-market uncertainties, and vanishing state paternalism, why did not the incidence of marriages rise if marriages could have assumed the role of a risk-mitigating institution? Below we attempt to respond to this question by looking at the correlation between male incomes and female incomes: for a marriage to function as an effective riskmitigating device, the market outcomes of the spouses need to be correlated negatively or if positively, only by a little. If the incomes of males and females are strongly positively correlated, the expected risk-sharing gain will be quite miniscule. The joint working of a “cost argument” and an “income correlation” explanation can be heuristically illustrated by an example. Consider a risk-averse individual with uncertain income. Let us proxy the “cost argument” by the status of employment of the individual, that

5

Marriage can be conceived as an alliance in which two individuals pool and share incomes, thereby providing consumption insurance to each other. Such an arrangement serves to mitigate the effects of idiosyncratic shocks to individual incomes.

3

is, if the individual is employed, marriage is more affordable.6 The “income correlation” argument implies that the individual (regardless of his or her status of employment) can expect insurance gains upon marriage if the individual marries a partner whose income or, for that matter, labor-market prospects are not perfectly correlated with his or her own income, or with his or her market prospects. In the following Table, let the potential benefits from marriage, X, Y, Z, W maintain X > Y, X > Z, Y > W, Z > W. In times of economic crisis, when the prospect of becoming unemployed rises, both male and female incomes decrease and become increasingly positively correlated. Consequently, the benefits from marriage decline (a move along the main diagonal). Thus, in addition to becoming increasingly costly, marriages become less desirable as a consumption-smoothing device.

Table 1. Potential Benefits from Marriage Employed No

Low

X

Y

High

Z

W

decrease

Income Correlation with a Potential Partner

Yes

decrease

We seek to explore the aforementioned four possible explanations for “the Russian marriage drama” (“cost story,” shift toward “western” norms, “perception of the future” argument, and “income correlation”) by drawing on macro-level aggregate data as well as on

6

The intuition is that employed individuals are in a better position to afford the costs associated with marriage (the wedding ceremony, housing expenditures, purchase of household goods), and that they face less restrictive borrowing constraints.

4

micro-level data. The aggregate data were collected from the Demographic Yearbooks of Russia, the World Development Indicators, and the TransMONEE database. The micro-level data are derived from the second wave of the Russian Longitudinal Monitoring Survey (RLMS). In comparison with studies that resort to marriage stocks, our data sets enable us to derive measures of marriage flows; examining flows into marriage is essential for an analysis that seeks to establish how current marital decisions are affected by current labor market and marriage market conditions. In section 2 we present raw macro data pertaining to the deteriorating employment landscape and portray the changing marriage picture. In section 3, drawing on micro data, we explore and test possible explanations for the incidence and the patterns associated with couple formation. We obtain considerable support for the hypothesis that it is the “cost argument,” in conjunction with a positive correlation of male incomes and female incomes, which depressed the inclination for couple formation. In section 4 we provide concluding remarks and delineate several policy repercussions.

2. Marriage Trends in Russia during the Transition: A Macro-Level Perspective In this section we present and study raw macro data. Doing so enables us to shed some light on the first two possible explanations of the observed trends in the Russian marriage rate: the “cost story” and the shift toward “western” norms. In addition, this section serves to motivate the micro-level analysis presented in section 3. During the period 1960-1989, the Russian crude marriage rate (CMR) declined gradually from 12.5 to 9.4 marriages per 1,000 inhabitants. Starting in 1990, the decline of the CMR accelerated considerably: the CMR declined from 9.4 to 5.8 during 1989-1998. It then increased to 7.1 in 2002. The path of, and the steep decline in the Russian CMR after the

5

onset of the economic and political transformation in 1990, are depicted in Figure 1. The emerging picture is of a considerable association between the trends in the CMR, employment, and real GDP.7 Taken at face value, this co-movement speaks in favor of a “cost story,” that is, it is consistent with the notion that falling incomes and a deteriorating employment landscape were the culprit behind an increasing reluctance to form marriages. Since 1999, the economic conditions have improved and marriages may have become more “affordable” (less costly).

Figure 1. Trends in Marriage, GDP, and Employment

100 80 (1989=100)

CMR, AMR, GDP, and Employment

120

60 40 20 0 1988

1990

1992

1994

1996

1998

2000

2002

Year GDP (constant 1995 US$) CMR (marriages per 1,000 population) AMR (marriages per 1,000 15-44 year-olds) Employment ratio (number of employed as % of population aged 15-59)

Sources: World Development Indicators 2004, the TransMONEE database (2004), and the Demographic Yearbook of Russia (various editions). 7 Notwithstanding the strong positive correlation between the CMR and the GDP reported in the text below, we note, in passing, that between 1992 and 1995 the CMR remained relatively constant, while the GDP kept declining. While we do not have a compelling explanation for this interruption in the co-movement of the two indices, we suggest a tentative reasoning. In 1992, the CMR was already at a very low level of 7.1 marriages per 1,000 inhabitants. It could well stand to reason that at very low levels, the CMR is quite downwards inelastic. Thus, incomes have to fall by a substantial amount before the CMR is dragged along, which finally it was (to 5.8) in 1996. It is as if, akin to the case of a hard core unemployment, there is a “hard core” of adherents to marriage such that at very low levels of the CMR, that “hard core” is paramount.

6

The depiction in Figure 1 does not appear to support the argument of a shift toward “western” norms. While a declining CMR is often linked to an increasing average age at first marriage, a somewhat unanticipated pattern for Russia is a constant and relatively low average age at first marriage: about 23 years for women, and 25 years for men. The nonpostponement of marriages is somewhat of a challenge for explanations that highlight a convergence of marital (and fertility) patterns of behavior to “western” styles. Consequently, there has been practically no difference between the trends in the CMR and the adjusted marriage rate (AMR) - which reflects marriages per 1,000 inhabitants of the 15-44 year-olds (Figure 1). Hence, the decline in Russia’s marriage rate in the 1990s amounts to a quantum effect rather than to a postponement effect.8 Table 2 provides insights into the relationships between variables of interest. The CMR and the AMR are perfectly positively correlated. The strong positive correlation of the CMR with GDP and with employment, and the strong negative correlation of the CMR with unemployment and with youth unemployment are consistent with our hypothesis: difficulties in gaining employment, especially for young adults, are deleterious to marriage formation. The average age at first marriage for both men and women is not significantly correlated with any other variable of interest. It is worth noting that the crude divorce rate (CDR) is not significantly correlated with other variables, except for a negative correlation with youth unemployment. Apparently, the relationship between divorce and economic conditions is less pronounced than the relationship between marriage and economic conditions. We provide an explanation for the lack of correlation between the CDR and the deteriorating labor market conditions in the concluding section of the paper.

8

Philipov and Kohler (2001) also find that, in comparison to the fertility patterns in other Central and Eastern European countries, Russia’s fertility decline in the early 1990s is characterized by a substantially larger relevance of quantum effects and a lesser relevance of postponement effects.

7

Table 2. Correlation coefficients, 1989-2002

CMR AMR Age_w Age_m

CMR

AMR

Age_w

Age_m

CDR

GDP

Employ.

Unempl.

Youth unempl.

1.000

0.997**

0.304

-0.143

0.075

0.926**

0.916**

-0.796**

-0.819**

1.000

0.346

-0.104

0.017

0.927**

0.921**

-0.811**

-0.821**

1.000

0.790*

-0.608

0.368

0.255

-

-

1.000

-0.536

-0.110

-0.248

-

-

1.000

0.006

0.006

-0.586

-0.660*

1.000

0.960**

-0.857**

0.842**

1.000

-0.924**

-0.916**

1.000

0.984**

CDR GDP Employment Unemployment Youth unemployment

1.000

Notes: CMR is marriages per 1,000 population; AMR is marriages per 1,000 15-44 year-olds; Age_w is the average age of women at first marriage; Age_m is the average age of men at first marriage; CDR is divorces per 1,000 population; GDP is in constant 1995 US$; Employment is the number of employed as % of population aged 15-59; Unemployment is the annual unemployment rate based on the Labor Force Survey (LFS) concept; Youth unemployment is the unemployment rate among 15-24 year-olds based on the LFS concept. Data on Age_w and Age_m are available only until 1996. Data on Unemployment and Youth unemployment are available only from 1992. ** and * indicate significance at the 1% and 5% level, respectively. Sources: World Development Indicators 2004, the TransMONEE database (2004), and the Demographic Yearbook of Russia (various editions).

8

The raw data appear then to point to economic hardship or to a “cost argument” as a possible explanation for the decline in the marriage rate in Russia during the early and mid 1990s. The marriage trend followed the GDP and the employment trends, and marriage rates declined almost proportionally across all age groups with no indication of a distinct inclination to postpone marriages. Looking at a more disaggregate level, the CMR declined in all Russian regions, albeit with different intensities.9 The available statistical reporting is for 11 standard “macro” regions (North, Northwest, Central, Volga-Vyatsky, Central Chernozem, Povolzhsky, North Caucuses, Urals, West Siberia, East Siberia, and the Far East) and for 89 smaller regions. Among the “macro” regions, between 1990 and 1997 the decline of the CMR in the Central region was 21 percent, while in the Far East it amounted to 41 percent. At a further disaggregate level, the variation was even more pronounced: the CMR decline in the KomiPermyatzky autonomous area and in the Republic of Tuva, for instance, was about six times as large as the decline recorded in the Moscow region. Relating the regional variation in the decline of the CMR to the regional unemployment rates produces a statistically significant correlation coefficient of 0.30. Disaggregated level data can be used to further assess the argument of a possible discernable shift toward “western” norms. If a convergence of marital patterns of behavior to “western” styles were at work, we would have expected to observe the largest decline in CMR in regions that are most exposed to ”western” norms. However, Moscow experienced what turns out to be among the smallest declines in marriage rates. Yet, at the same time Moscow was among the regions that were least affected by increasing unemployment. This tends to

9

The marriage rates are reported in the Demographic Yearbook of Russia (various editions) published by Goskomstat - Russia’s official statistical agency.

9

suggest that in affecting the propensity to marry, economic factors played a more important role than cultural factors. An obvious weakness of the foregoing impressions and interpretation is reliance on macro-level data: a correlation of the movements in aggregate employment rates and real GDP with marriage trends falls short of comprising a compelling argument that economic hardships are the cause of a significant discouragement for marriage formation at the individual level. Aggregate correlations disregard the variation across social groups and distant locations, and they do not answer the question of who is not getting married: those who are unemployed or those who have benefited from the transition? To further investigate these issues, we turn our attention to micro-level data.

3. Couple Formation: A Micro-Level Perspective We draw on micro-level data in order to investigate the four possible explanations of the trends in the Russian marriage rate. We separate the pre-1999 period (the period of economic crisis) from the post-1999 period (the period of economic recovery). We study these relatively short periods for two reasons. First, the profound changes in the socioeconomic conditions and the labor market outcomes that occurred during these periods provide a fertile ground and a stark background for seeking to understand how these changes affected couple formation. Second, data requirements impose more stringent sample selection requirements the longer the period of analysis. While we attend to the post-1999 period, the pre-1999 period is of paramount interest to us since it is then that the propensity to marry changed most dramatically.

10

The Data We draw primarily on the second wave (1994-2002) of the Russian Longitudinal Monitoring Survey (RLMS).10 The RLMS is the first nationally-representative random sample for Russia.11 Although, the main unit of observation is the household, data were collected both for households and for individual household members. Information about the residential community was recorded as well. The households were selected by employing a multi-stage probability sample and were clustered into primary sampling units. In the sample, the distribution of households by size within the urban and the rural areas matches well the corresponding 1989 census distribution. The data available to us consist of seven rounds (data were collected in 1994, 1995, 1996, 1998, 2000, 2001, and 2002). Although the target sample size for the second wave was set at 4,000 households, in anticipation of a nonresponse rate of about 15 percent, the number of households drawn into the sample was 4,728. The net effect of non-response and attrition, however, was fairly small.12 Although the RLMS is a longitudinal study of Russian households, it does not present a real panel design. For example (few exceptions notwithstanding), individuals and households were not interviewed if they moved away from the initial dwelling. This is a primary shortcoming of the RLMS data set, at least from our perspective. In particular, nonrandom attrition could bias the estimation results. It might be expected that single individuals are more likely to exit from the panel, either because they are more likely to migrate or

10

The first phase of the survey was conducted in 1992 and 1993. The data facilitate longitudinal analyses within each phase but not for the entire period of the survey: the two phases are based on entirely different population samples. 11 The RLMS was conducted with the support of the World Bank, the Agency for International Development (USAID), the National Science Foundation, the National Institute of Health, and the North Carolina Population Center. Detailed information about the RLMS dataset can be downloaded from: http://www.cpc.unc.edu/projects/rlms/rlms_home.html. 12 Attrition was highest for respondents from the metropolitan areas of Moscow and St. Petersburg. Because lower rates of participation in these regions were anticipated, the initial allocations to them were increased.

11

because they migrate upon marriage. In the last part of this section we employ two alternative procedures to investigate whether sample selection is likely to pose a real problem. To begin with, we supplement the descriptive account that was based on the macro data with summary statistics based on the RLMS data. Table 3 presents data on gender, urban residence, education, employment indicators, and the attitudes of married and single individuals aged between 16 and 45 in the survey years.13 Panel A of Table 3 reveals that the proportion of single men is slightly higher than the proportion of single women. There does not seem to be a significant difference in marital formation between urban and rural areas. Married individuals are, on average, better educated.14 In line with the analysis in section 2, Panel B of Table 3 reveals that the employment landscape deteriorated until 1998, but posted some improvement thereafter. A similar variation appears to pertain to measures of labor-market uncertainties. In particular, the percentages of individuals to whom employers owed wage payments declined drastically after 1998. Overall, single individuals were more affected by unemployment than married individuals. On the other hand, the proportions of married persons sent on unpaid leave and with unpaid or delayed wages were mostly higher than the corresponding proportions of singles.

13

The legal minimum age at marriage in the Russian Federation is 18 years for both men and women. In some cases, upon the request of those to be married, local authorities are granted authority to permit marriage for individuals who have attained the age of 16. 14 The married are, however, also older than the singles.

12

Table 3. Summary Statistics by Marital Status (percentages of complete samples of persons aged 16-45, unless indicated otherwise) Year

1994

1995

A. Gender, urban residence, education Married and cohabitating Men 47.4 47.1 Living in urban areas 70.9 70.3 With high-school diploma 88.9 1) With university degree 21.2 18.7 Single 54.6 Men 52.3 Living in urban areas 73.7 71.6 With high-school diploma 71.3 With university degree1) 15.1 12.5 B. Employment status Married and cohabitating Unemployed2) Employed3) On unpaid leave4) With unpaid or delayed wages4) Single Unemployed2) Employed3) On unpaid leave4) With unpaid or delayed wages4)

1996

1998

2000

2001

2002

47.6 68.8 87.4 18.2

47.3 67.5 85.1 18.6

46.9 65.6 84.7 18.7

46.5 68.4 87.1 19.7

46.8 68.3 86.6 20.4

52.7 69.2 68.5 11.6

51.9 65.3 69.3 11.8

50.8 65.4 68.9 14.1

50.6 68.4 71.1 16.9

51.7 68.8 73.0 18.6

9.0 80.5 1.0 43.5

9.2 79.4 1.0 43.9

11.7 77.0 0.6 60.6

13.4 73.7 0.8 63.3

11.5 75.2 0.4 30.4

9.8 76.1 0.5 24.3

10.3 76.7 0.2 21.4

14.9 44.4 1.1 36.2

15.2 42.8 0.6 34.0

16.4 40.4 0.2 53.3

18.6 33.9 0.3 59.4

13.6 36.4 0.0 27.4

13.6 38.1 0.4 25.4

13.3 38.8 0.5 20.7

59.8 80.1 36.1 28.9 14.9

61.2 81.5 35.1 25.7 12.3

68.8 86.3 46.3 21.8 13.2

57.4 75.4 17.3 33.6 19.7

50.5 71.7 10.7 43.0 26.5

51.6 74.4 8.4 42.2 36.0

53.9 70.6 26.0 30.2 18.1

57.6 71.6 22.3 34.0 18.1

59.8 76.8 34.4 23.4 16.8

48.2 63.8 11.2 44.5 23.3

46.5 61.1 7.2 46.6 28.9

44.4 63.2 8.1 49.3 39.1

C. Attitudinal characteristics Married and cohabitating Concerned about chance of job loss 58.5 Concerned about getting necessities 79.7 Pessimistic about the future 37.8 Certain to find another job4) 28.1 Satisfied with life at present 14.6 Single Concerned about chance of job loss 49.2 Concerned about getting necessities 67.8 Pessimistic about the future 25.8 Certain to find another job4) 33.3 Satisfied with life at present 18.1 1)

Samples used in the analyses are persons aged 20-45. The question about university education was slightly changed in 1995 relative to 1994 and then it was modified again in 2001. 2) Samples used in the analyses are persons who are in the labor force. 3) Includes persons who are currently working or are on paid leave. 4) Samples used in the analyses are persons who have a primary employment. Source: The Russian Longitudinal Monitoring Survey.

13

Table 3 also displays attitudinal characteristics. Panel C shows that, on average, married individuals were more concerned and more pessimistic than singles: the married were more concerned than singles about job losses, more concerned about obtaining daily necessities, less certain to find another job, and a higher proportion of married believed that they would “live worse” in the coming year. On average, the married were also less satisfied with their lives. The attitudinal characteristics for both married and single individuals registered a discernible improvement after 1998.

Baseline Analysis The descriptive analysis presented so far serves to highlight several patterns: the macro-level data reveal that to a certain extent, the marriage trend followed the employment trend throughout the study period. The RLMS micro-level data portray a more differentiated picture: while the proportion of the unemployed among single individuals was higher than the corresponding proportion among married individuals, married individuals were more likely than single individuals to be sent on unpaid leave or to receive no payment on their primary job. However, the percentages presented in Table 3 are unconditional statistics, calculated for the entire sample. For sure, an individual’s decision to get married or to enter a non-marital cohabitation depends on personal and contextual characteristics that are specific to the individual. To account for the impact of these characteristics, we perform probit regressions for both men and women. Prior to conducting the regression analysis, however, we utilize the clustered sample structure of the RLMS to calculate for each sampling unit the correlation of male incomes and female incomes.

14

The households in the RLMS were clustered into 38 sampling units, henceforth “communities.”15 For each community we calculate the correlation between male incomes and female incomes. Communities differ widely in their industrial structure, wealth, amenities and, of course, geographical location. In particular, occupational segregation in Russia seems to be quite strong and thus the correlation between male incomes and female incomes in a given community could be strongly related to the industrial structure of the community. For example, in communities with mainly professional jobs, incomes are likely to be highly (positively) correlated because men and women are working in similar white-collar jobs, while in manufacturing communities, where men and women are likely to have very different jobs (women are employed in white-collar jobs, men - in blue-collar jobs), the income correlations are weaker. We apply two methods to calculate a measure of the expected spousal income correlation.16 While each measure has advantages and disadvantages, both measures yield quite similar estimates. Below we report results that were obtained using a measure of income correlation that draws on the occupational structure of a community.17 During the 1994-1998 period, the real incomes of males and females were declining in tandem in all communities, although with different intensities: the correlation coefficients of male incomes and female incomes across communities ranged from about 0.25 to 0.82. Men faired poorly relative to women, that is, the female-male income differential increased in

15

Communities were identified using confidential RLMS community data. The communities include large cities and consolidated “raions” (a Russian term for regions) that were defined on the basis of geographical factors and the level of urbanization, as well as on the basis of ethnicity. Hence, a community seems to be an appropriate geographic sphere of a local marriage market. 16 The Appendix describes the criteria for the sample selection and elucidates the methods used to calculate the income correlations. 17 This measure is based on what we consider to constitute a more appealing source of exogenous variation. Results based on an alternative measure of income correlation derived from a simultaneous estimation of the probability of work and the expected hourly wages of potential spouses were qualitatively similar to the results reported in the text.

15

almost all communities.18 On the other hand, the 2000-2002 period witnessed increasing real incomes for both males and females, albeit with a weaker correlation. The female-male income ratio increased in some communities, but declined significantly in others. Below we exploit the variation in the community-level correlation of male incomes and female incomes with the purpose in mind of investigating whether a potential correlation in economic outcomes across would-be spouses can help us explain the propensity for couple formation. Akin to Duflo (2001) and to Pischke (2005), our estimation strategy is as follows: a dummy variable indicating the initial status of employment of the individual and the correlation of male incomes and female incomes in a community are plausibly exogenous variables.19 The idea that gives rise to this strategy can be illustrated by using simple two-bytwo tables. Table 4 shows percentages of men and women who participated in couple formation, by the initial status of employment and the community-level correlation of male incomes and female incomes. Communities are divided into “low-correlation” communities and “high-correlation” communities. High-correlation communities are defined as communities with correlations of male incomes and female incomes that are higher than the median.20 In Panel A of Table 4, the analysis of couple formation is based on all single men and single women aged 16-45 in 1994. In Panel B of Table 4, the sample consists of all single men and single women aged 16-45 in 2000.21 Because of the small number of new couples 18

In contrast to this, Brainerd (1998) finds a decreasing female-male wage differential during the period 19911994. 19 Unlike the current status of employment, the initial status of employment is unlikely to be endogenous with the couple formation decision. This characterization should particularly hold during times of economic crisis and increasing unemployment. Below we test for the exogeneity of the status of employment and find support of the exogeneity assumption. 20 The median correlation is 0.63. The difference between the average correlations in high-correlation and lowcorrelation communities is about 0.55. 21 The reason for including individuals aged 16-45 years is the compatibility of the sample with the definition of the AMR (which includes individuals old enough to marry legally and still young enough to bear children). It could be argued that the sample is choice-based since single individuals older than, say, 30 form a peculiar group. For this reason, we have experimented with different sample definitions. For example, restricting the sample to individuals aged 16-30 years yielded similar results to those that we have presented, with only slightly

16

that are not in a formal marriage, we do not refer to cohabitation as a separate marital status category and merge the individuals who started cohabitating with the individuals who entered a formal marriage.22

Table 4. Percentages of Individuals Who Participated in Couple Formation by Employment Status and Community-Level Correlation of Male and Female Incomes Men Correlation of male and female incomes Low High Difference

Women Correlation of male and female incomes Low High Difference

(1)

(2)

(3)

(4)

(5)

(6)

55.94 (4.09)

38.67 (2.17)

17.27 (5.23)

46.61 (3.78)

39.73 (5.06)

6.91 (5.05)

Not employed in 1994

33.86 (1.95)

27.07 (2.02)

6.79 (2.19)

40.29 (3.05)

37.06 (2.69)

3.65 (3.11)

Difference

22.08 (5.06)

11.60 (4.38)

10.48 (7.29)

6.32 (4.11)

2.67 (3.96)

3.26 (5.15)

32.02 (4.53)

16.45 (4.42)

15.57 (7.47)

21.87 (3.05)

15.02 (2.69)

6.85 (3.21)

Not employed in 2000

15.37 (4.59)

8.03 (1.54)

8.23 (4.73)

17.93 (2.97)

13.15 (2.76)

4.78 (3.32)

Difference

16.65 (4.80)

8.42 (4.61)

7.34 (7.09)

3.94 (3.13)

1.87 (3.01)

2.07 (5.16)

Panel A: Couples formed during 1995-1998 Employed in 1994

Panel B: Couples formed during 2001-2002 Employed in 2000

Notes: In Panel A the sample consists of all individuals who were single and aged 16-45 in 1994. In Panel B the sample consists of all individuals who were single and aged 16-45 in 2000. Standard errors are in parentheses. Low-correlation (high-correlation) communities are defined as communities with correlations of male incomes and female incomes that are lower (higher) than the median.

In Panel A of Table 4, we compare couple formation during the period 1995-1998 by men and women who were employed in 1994 to couple formation by men and women who higher overall quality of the statistical fit. Consequently, we have elected to refer to the sample of individuals aged 16-45 years since this sample is larger and since it yields results that are asymptotically more valid. 22 The moderate increase in cohabitation does not appear to support the argument of a formidable shift toward “western” norms.

17

were not employed in 1994, in both types of communities.23 For both men and women, the percentage of individuals who participated in couple formation in low-correlation communities was higher than the corresponding percentage in high-correlation communities. In addition, both men and women who were employed in 1994 were more likely to form a couple during 1995-1998. The differences are particularly pronounced for men. On average, participation in couple formation by employed men living in low-correlation communities was 10.5 percentage points higher (for women it was 3.3 percentage points higher). Although the differences in differences are not statistically different from zero, the results in Table 4 suggest that, especially for men, being employed and living in a low-correlation community could impinge positively on the decision to form a couple.24 In Panel B of Table 4 we present the corresponding estimates for the period 20012002, differentiating between men and women who were employed in 2000, and men and women who were not employed in 2000. Recall that this period witnessed economic recovery and improved economic conditions.25 Although statistically significant to a lesser extent, the results are qualitatively the same as the results presented in Panel A. Note that the pattern of couple formation could vary systematically across communities. That is, in interpreting the preceding results we need to be aware of the possibility that the association between couple formation and the status of employment may emanate from other features of the labor and marriage markets in the different communities, and that the association with income correlation across communities may be purely incidental. Yet, the simple percentages reported in Table 4 appear to suggest that both the status of 23

The difference-in-difference estimator controls for systematic variation of the propensity for couple formation both across the status of employment and across communities. 24 It can be argued that for many individuals who are younger than 22 and who have yet to complete their formal education, the labor market does not feature importantly in their decision to form a couple. For this reason, we have repeated the analysis eliminating individuals aged 21 or younger. The results were qualitatively similar. 25 During this period male and females incomes were less correlated compared to the late 1990s.

18

employment of an individual and the correlation of male incomes and female incomes in a community were at work in the couple formation calculus of young Russians. In the remainder of this section we further investigate this association.

Regression Results To discern more clearly whether the “cost argument” is indeed supported by the micro-level data, we begin with the following relatively parsimonious specification:

(1) Mij = c + Piα + Eiβ + εij

where Mij is a dummy variable equal to one if individual i who is living in community j participated in a couple formation (and equal to zero otherwise) within the periods 1995-1998 and 2001-2002; c is a constant; Pi is a vector of individual characteristics (age, age squared, completed highest grade level in school, a dummy variable for high education, a dummy variable for urban residence) and attitudinal characteristics (dummy variables for pessimism, concern about obtaining daily necessities, concern about job loss, and satisfaction with life at present); Ei is a dummy variable indicating whether the individual is employed or not; εij is a random term; and α and β are coefficients to be estimated. All the explanatory variables are measured at the beginning of the observation periods, that is, at 1994 and at 2000. The main reason for this choice is that we are interested in finding how subsequent couple formation propensities are related to the initial individual, attitudinal, and employment characteristics. This choice also helps us to deal with the potential problem of reverse causality. One possible objection to our single-equation approach is the simultaneity between employment status and couple formation. Because of the intertemporal sequencing (the status

19

of employment is measured at the beginning of the observation period) we treat employment status as predetermined, that is, we assume that it is independent of subsequent disturbances in the probit. Nevertheless, it could be argued that the status of employment is likely to be intertemporally correlated and thus, the initial status of employment could be determined by the individual’s expected marriage decision. For example, individuals who seek to attract a good mate know that they will need to hold a job for several years prior to that. To account for the possible simultaneity between the status of employment and couple formation, we conducted a test of weak exogeneity, drawing on a procedure proposed by Smith and Blundell (1986). Following Wood (1995) and Blau et al. (2000), we have addressed the issue of identification by including the occupational structure of the community in the equation for the status of employment. We also used a dummy variable indicating whether any government enterprise was closed down in the community prior to the survey (this variable is most plausibly exogenous to the propensities to form marriages). We found evidence strongly supportive of the exogeneity assumption.26 A plausible explanation for the exogeneity of the status of employment is that holding a job is less likely to be choice-related in times of labor market crisis with high unemployment. It could also be argued that due to the very nature of the transition to a market economy, the status of employment was determined primarily by exogenous labor demand shocks induced by the transition.27 To test the causal framework of Table 1, we employ the following regression model:

(2) Mij = c + Piα + (Cj Ei)γ + (Rj Ei)δ + εij

26

It was not possible to reject the hypothesis of weak exogeneity of the status of employment variable for the parameters of the equation for couple formation even at the 0.25 level. 27 The Russian transition process was quite erratic. The country experienced price liberalization, nearhyperinflation, and a near-collapse of aggregate demand.

20

The propensity for couple formation of individual i who is living in community j depends on the interaction of the individual’s status of employment with the community-level correlation of male incomes and female incomes, Cj, and with a vector of region-specific variables, Rj; γ and δ are coefficients to be estimated; Pi and Ei are as defined in equation (1). In this specification, Cj and Rj can also be taken to serve as proxies for exogenous marriage market conditions in the communities. This is in line with the evidence that individuals tend to search for marriage partners within particular marriage markets. Given the clustered sample structure of the RLMS, standard errors in all the regressions are clustered at the community level to correct the inference for the inter-dependence of individual unobservables within communities. A test of the specification in equation (1), reported in Panel A of Table 5, reveals that for both women and men, employment is significantly associated with the probability of couple formation during the period 1995-1998. That is, with the onslaught of rising unemployment, being unemployed decreased the probability of couple formation for both sexes. The effect of the status of employment was greater for men: as shown in the first line in the Table, unemployment reduced the couple formation propensity of men by 9.1 percent.28

28

The significantly negative effect of unemployment on the probability of couple formation is particularly troublesome in light of the fact that the unemployment in Russia is not short-term, with many transitions in and out of the labor market, but rather long-term: for instance, about 56 percent of the single men aged 16-45 who were not employed in 1994 were still not employed in 1996 and 43 percent were still not employed in 1998.

21

Table 5. Marginal Effects on the Incidence of Couple Formation Dependent Variable: Couple Formation (1=Yes, 0=No) Men (1)

Panel A: Couples formed during 1995-1998 (1) Not employed in 1994 Log-likelihood (2) Not employed in 1994 * income correlation Log-likelihood

Log-likelihood (2) Not employed in 2000 * income correlation Log-likelihood

-0.114 (0.063) -417

(4)

-0.119 (0.062) -416

-0.210 (0.066) -412

-0.032 (0.016) -636

-0.035 (0.017) -635

-0.038 (0.017) -633

-0.021 (0.018) -711 -0.074 (0.040) -303

-0.074 (0.039) -302

-0.025 (0.018) -729

552 68 484

No No

(6)

502 201 301

-0.063 (0.034) -315 -0.072 (0.037) -306

(5)

-0.030 (0.017) -642

373 141 232

Number of observations Formed a couple Remained single

Control variables: Not employed*regional income Not employed*regional sex ratio

Women (3)

-0.091 (0.043) -427

Number of observations Formed a couple Remained single

Panel B: Couples formed during 2001-2002 (1) Not employed in 2000

(2)

Yes No

-0.029 (0.020) -725

-0.030 (0.021) -724

747 133 614

Yes Yes

No No

Yes No

Yes Yes

Notes: In Panel A the sample consists of all individuals who were single and aged 16-45 in 1994. In Panel B the sample consists of all individuals who were single and aged 16-45 in 2000. All specifications include individual characteristics (age, age squared, completed grade level in school, a dummy for high education, a dummy for urban residence) and attitudinal characteristics (dummies for pessimism, concern about obtaining daily necessities, concern about job loss, and satisfaction with life at present). Each entry corresponds to a separate regression according to the specifications in equations (1) and (2) in the text. Robust standard errors are in parentheses. All log-likelihoods are significant at the 5% level based on the Chisquare test.

For women, employment seems less important: unemployment reduced the couple formation propensity of women by 3 percent. Nevertheless, the negative and significant coefficient on female unemployment is inconsistent with the traditional model of women’s

22

labor and marriage market behavior as well as with much of the evidence (for example, Schultz, 1994; Wood, 1995; Blau et al., 2000; Loughran, 2002). Our finding that unemployment is harmful for women’s chances of couple formation is in accordance with the argument that in the presence of increasing labor market uncertainties (as was the case in Russia during the late 1990s), employed women should be able to compensate for male unemployment. In such a setting, a woman’s employment would be valued in the marriage market as a means of insurance.29 Overall, we can conclude that the “cost argument,” which seemed a plausible explanation according to our reading of the macro data, is also led credence to by the micro-level data. An empirical test of equation (2), reported in Panel A of Table 5, suggests that the positive community-level correlation of male incomes and female incomes (that is, incomes co-declining for both males and females) also contributed to a reduction in the marriage rate during the era of the economic crisis. Unemployed individuals living in communities where the incomes of males and females were more positively correlated were less likely to participate in couple formation. For unemployed men living in high-correlation communities this effect amounted to 11.4-21 percent, while for women the effect was less pronounced - it was 3.2-3.8 percent.30 These patterns, consistent with the framework outlined in Table 1, indicate that the Russian “marriage drama” would have been an attenuated drama in the absence of the positive correlation of the incomes of the would-be spouses.

29

We do not know, however, whether an unemployed woman was less likely to form a couple due to a voluntary decision of the woman to wait until she becomes employed, or due to lack of offers from men who prefer a spouse with a job. 30 As pointed out by Ai and Norton (2003), the probit model is non-linear and the interaction effect might not be evaluated correctly by simply looking at the coefficient on the interaction term. Since the interaction term might differ across individuals, we also plotted the estimated z-statistics against the predicted probability of forming a couple. For the great majority of observations, the interaction term was indeed negative and significant.

23

The estimated coefficients on the various attitudinal characteristics were mostly insignificant and are not reported in Table 5.31 In summary, the results support both the “cost story” and the “income correlation” explanation, but not the “perception of the future” argument.32 Panel B of Table 5 reveals that the coefficient of female unemployment no longer significantly explains the probability of couple formation during the period 2001-2002. Male unemployment still has a significantly negative effect on couple formation propensities.33 In other words, with improving economic and labor market conditions, unemployment only “penalizes” men: presumably, unemployed women have more of an option of withdrawing from the labor market into family life than is the case with regard to unemployed men. In line with the framework of Table 1, the negative effect of unemployment on couple formation is larger in communities with relatively large positive correlation of male and female incomes, although these results are not as robust as in Panel A. The preceding inference draws on the assumption that there are no omitted regionspecific effects that are correlated with the community-level income correlation. It could though be argued that the community-level income correlation is proxying for other regionspecific variables that influence both male and female propensity for couple formation. In particular, the income correlations in themselves might be highly correlated with the overall economic and social conditions in different regions. To account for this possibility, we

31

We also do not report the estimated coefficients for individual characteristics, which yield relatively standard or statistically insignificant results. 32 Regressions using a restricted sample of individuals aged 22-45 produced qualitatively similar results, with the magnitude of the marginal effects being slightly higher. This is not surprising since for these individuals, the labor market is likely to play a more important role in the decision to form a couple. 33 The negative effect of unemployment on the propensity for couple formation during the period 2001-2002 is weaker than during the period 1995-1998. It is possible that unemployment was less of a problem in the early 2000s as, with the improving economic conditions, it was less likely to last for a long time. Indeed, according to the Russian Labor Force Surveys the share long-term unemployed increased from 18 to 47 percent during 19931999, and then declined to 37 percent in 2001.

24

present in Table 5 specifications that control for the interaction between the employment status dummy variable and the regional income level, as well as for the interaction between the status of employment dummy variable and the ratio of single men to single women in a region. Apparently, controlling for both the regional income level and for the sex ratio does not diminish the explanatory power of the interaction term of the status of employment dummy variable with the income correlation (columns 2, 3, 5, and 6), suggesting that the coefficients are not overestimating the effect of income correlation on couple formation. We have repeated the analysis, including other region-specific variables: the change of the regional income during the study periods, the regional male mortality rate, and a measure of regional housing availability. Again, the explanatory power of the interaction term of the status of employment dummy variable with the income correlation has not decreased.34 In an alternative specification we have included community dummy variables to sweep out unobservable factors that vary across communities.35 We find it encouraging that the inclusion of community dummies has no significant impact on the estimates of the effect of the status of employment. It could be argued that the “cost argument” may not be well proxied by the status of employment of the individual. By the mid-1990s, delays in wage payment and the extensive use of unpaid administrative leave became a chronic problem even in highly profitable Russian enterprises (Table 3). Therefore, a simple distinction between being employed and being unemployed may not reflect meaningfully the degree of economic uncertainty and

34 In view of recent research on the link between income inequality and marriage rates (Loughran, 2002; Gould and Paserman, 2003) it could be argued that we should also control for an interaction of the status of employment with a regional measure of income inequality. However, data on inequality at the community level are not readily available, and the samples from the RLMS are not large enough to allow us to construct accurate community measures of inequality. 35 The inclusion of community dummies could be considered an overly conservative identification strategy because we are throwing away all the strong cross-sectional variation in marriage rates as well as the correlation between male incomes and female incomes.

25

unemployment risk in Russia during the transition period. In response to this blurring we reran the regressions, relegating the individuals with unpaid or delayed wages to a separate category. The reconstructed variable is also less likely to be endogenous with the propensity to form marriages than the employment status, since it is quite difficult to ascribe a systematic lack of interest in marriage to individuals who experience earnings shocks due to unpaid or delayed wages. Results presented in Table 6 reveal that wage arrears had indeed an additional effect on couple formation. Once again, the effect was greater for men: wage arrears reduced the couple formation propensity of men by 2 percent. The negative effect of unemployment on couple formation increased, gaining statistical significance. Consistent with out earlier findings, unemployed individuals and individuals with unpaid or delayed wages who lived in communities where the incomes of males and females were more positively correlated were less likely to participate in couple formation.

26

Table 6. Marginal Effects on the Incidence of Couple Formation: An Alternative Specification Dependent Variable: Couple Formation (1=Yes, 0=No)

Panel A: Couples formed during 1995-1998 (1) Not employed in 1994 With unpaid or delayed wages in 1994 (2) Not employed in 1994 * income correlation With unpaid or delayed wages in 1994 * income correlation

Panel B: Couples formed during 2001-2002 (1) Not employed in 2000 With unpaid or delayed wages 2000 (2) Not employed in 2000 * income correlation With unpaid or delayed wages in 2000 * income correlation

Men

Women

-0.097 (0.041)

-0.031 (0.015)

-0.020 (0.009)

-0.012 (0.006)

-0.115 (0.045)

-0.034 (0.013)

-0.028 (0.010)

-0.013 (0.007)

-0.067 (0.028)

-0.023 (0.019)

-0.009 (0.007)

-0.014 (0.026)

-0.072 (0.035)

-0.024 (0.020)

-0.015 (0.017)

-0.016 (0.034)

Notes: In Panel A the sample consists of all individuals who were single and aged 16-45 in 1994. In Panel B the sample consists of all individuals who were single and aged 16-45 in 2000. All specifications include individual characteristics (age, age squared, completed grade level in school, a dummy for high education, a dummy for urban residence) and attitudinal characteristics (dummies for pessimism, concern about obtaining daily necessities, concern about job loss, and satisfaction with life at present). Each entry corresponds to a separate regression according to the specifications in equations (1) and (2) in the text, including individuals with unpaid or delayed wages as a separate category. Robust standard errors are in parentheses. All log-likelihoods are significant at the 5% level based on the Chi-square test.

To gain additional insight and to test the robustness of our findings, we present in Table 7 results equivalent to the specifications in Table 5 (equations (1) and (2)) for various subsamples of communities. Columns (2) and (3) suggest that the effects of the status of

27

employment and income correlation on couple formation were stronger in communities that belonged to regions with higher unemployment rates, while columns (4) and (5) confirm these results for poorer communities. In other words, individuals who lived in relatively poor areas with bleak labor market prospects and who were jobless had a lower probability of forming a couple. The reason why couple formation was most impaired by a dismal labor market with high incidence of unemployment could be that an individual’s expectations about continuing changes in unemployment or job uncertainty are likely to be shaped by current changes in unemployment and labor-market situation: the latest conditions comprise the most pertinent experience that is extrapolated into the future. This “learning based on current experience” is likely to be especially important in transition countries where individuals face new institutional framework that has very little in common with the pre-1990 past.36 In columns (6) and (7) of Table 7 we divide the sample into communities located in regions where the initial CMR was lower or higher than the median CMR. Results indicate that the status of employment and income correlation had a greater impact in communities with higher initial CMR. Presumably in these communities, the benefits from marriage were higher to begin with. The stronger effect of our explanatory variables for these communities provides suggestive evidence that the Russian “marriage drama” was not driven by some initial, unobservable community-level characteristics. However, most of these estimates are not statistically significant.

36

For formal models about the formation of expectations based on current macroeconomic conditions, see for example, Sargent (1993) and Kohler (2000).

28

Table 7. Effects on Couple Formation during 1995-1998 by Categories of Communities Dependent Variable: Couple Formation (1=Yes, 0=No) Characteristics of communities Whole sample

Panel A: Men Not employed in ‘94 Not employed in ’94 *income correlation Panel B: Women Not employed in ‘94 Not employed in ’94 *income correlation

Unemployment rate Median

Regional income Median

1994 CMR Median

(1)

(2)

(3)

(4)

(5)

(6)

(7)

-0.091 (0.043)

-0.080 (0.049)

-0.099 (0.051)

-0.097 (0.044)

-0.075 (0.038)

-0.086 (0.077)

-0.095 (0.093)

-0.210 (0.066)

-0.203 (0.103)

-0.254 (0.114)

-0.275 (0.062)

-0.119 (0.056)

-0.202 (0.176)

-0.219 (0.119)

-0.030 (0.017)

-0.026 (0.011)

-0.035 (0.013)

-0.040 (0.013)

-0.023 (0.010)

-0.029 (0.023)

-0.045 (0.047)

-0.038 (0.017)

-0.033 (0.015)

-0.046 (0.012)

-0.046 (0.022)

-0.035 (0.026)

-0.034 (0.013)

-0.048 (0.030)

Notes: The sample consists of all individuals who were single and aged 16-45 in 1994. Each entry corresponds to a separate regression according to the specifications in equations (1) and (2) in the text. Robust standard errors are in parentheses.

Overall, it appears then that the results of empirical tests across subsamples of communities parallel those obtained by using the entire sample. As an additional check of robustness, we have employed an alternative geographic definition of the marriage markets. We have repeated our analysis using units larger than communities viz., regions.37 While the CMR declined and the male and female incomes were positively correlated in every region, the analysis suggests that the CMR declined relatively less in regions in which the correlation of male and female income was weaker.38 For instance, in the metropolitan areas (Moscow and St. Petersburg) and in the Western Siberian region, where male and female incomes were correlated less than in other regions, individuals 37

The RLMS data can be divided into eight “macro” regional categories: Metropolitan areas (Moscow and St. Petersburg), Northern/Northwestern, Central/Central Black-Earth, Volga/Viask/Volga Basin, North Caucuses, Urals, Western Siberia, and Eastern Siberia/Far East. 38 The coefficient of correlation between the percentage decline in the CMR and the regional correlation in male and female income is about 0.45.

29

could presumably expect higher co-insurance gains upon marriage. And indeed, in these regions, the CMR declined relatively less than in other regions. Re-running the regressions at a regional level yields coefficients similar in pattern to the coefficients obtained when communities are the units of analysis, although the regional coefficients are generally about 15 percent smaller than the community coefficients. The most likely reason for this difference is that communities capture better than regions the relevant marriage market conditions since individuals tend to search for mates in fairly small geographic areas.39

Unobserved Heterogeneity and Simultaneity Bias As reported in the preceding subsection, we tested formally whether the status of employment could be treated as exogenous to the propensity to form couples. The tests showed that it was not possible to reject the exogeneity assumption. In this subsection we seek to address any remaining sources of simultaneity bias in the coefficients of interest. Herewith we briefly summarize our findings.40 It has been long recognized in the received literature that the decision to marry may depend on some unobserved characteristics of the individuals concerned. For instance, individuals who are hard-working, law-abiding, risk-averse, and abstemious may be more attractive to potential spouses as well as to potential employers; individuals who prefer marriage to remaining single may possess a desirable profile of inclination in the labor market and therefore are more likely to end up being employed. To the extent that the unobserved differences between individuals who participate in couple formation and individuals who do not participate are due to fixed, time-invariant individual characteristics such as preferences, 39

Drewianka (2003) finds that individuals in the US are likely to search for mates over relatively large geographic areas. This apparent difference between the span of the marriage markets in the US and in Russia is consistent with the observation that the inter-regional mobility of the Russian population is quite limited. 40 Full detail will be provided upon request.

30

state of health, or endowments, the effect of the status of employment on marriage can be identified by using fixed-effects regressions. In order to perform the required inquiry, we organized our data into a person-year file, and we estimated fixed-effects probit regressions. The dependent variable was a binary variable indicating whether an individual participated in couple formation in the following year. We used the same set of explanatory variables that we used before (the value of the explanatory variables was held at the current year) and in addition we included year dummies in order to capture national trends in the taste for marriage: social norms, changes in marriage and divorce laws, contraceptive methods, and so on.41 The fixed-effects probit regressions produced results that were qualitatively similar to the results presented in Tables 5 and 6, albeit with a slightly reduced magnitude of the coefficient on the employment status.42 An endogeneity of our measure of correlation of male incomes and female incomes with couple formation could be expected, especially if an individual migrated from a highcorrelation community to a low-correlation community prior to the survey. We re-estimated the models in Tables 5 and 6 using the correlation of male incomes and female incomes in the individual’s community of birth as an explanatory variable instead of the correlation in the individual’s current community. We concluded that our main results were not attributable to individuals moving to communities with a lower correlation between male incomes and female incomes.43 We next attend to potential problems that could arise from sample selection due to non-random sample attrition. 41

We had to drop, however, the explanatory variables that were constant across periods since their inclusion would have precluded estimation of the fixed-effects model. 42 In addition, the simple likelihood ratio test of the fixed-effects model against the base model without individual effects (pooled observations) did not lead to rejection of the base model. 43 It is not surprising that we obtained similar results using communities of residence and communities of birth since the overwhelming majority of the population in Russia does not relocate.

31

Correction for Sample Selection Ignoring the potentially non-random sample attrition could cause a serious selectionbias problem. To check for this, we first estimated probit equations to explain whether respondents leave the panel between rounds t and t+1, using dummy variables for the marital status at time t as explanatory variables. For the various specifications that we have tried (controlling for different sets of background variables) and for both males and females, the coefficients on these dummy variables turned out to be insignificant, suggesting that sample attrition does not relate to marital status.44 We implement two alternative procedures to investigate whether sample selection is likely to be an important problem. First, we estimate a nested probit model which adjusts for the sample-selection bias associated with restriction of the sample to single individuals who did not move during the study period. In the nested probit model, the non-attrition and the incidence of couple formation are estimated jointly. The specification of the non-attrition equation is guided by considerations discussed in Fitzgerald et al. (1998).45 However, the estimated selection effects in the nested probit model were mostly statistically insignificant (or only on the margin of statistical significance). There is also little difference between the coefficients in the adjusted estimations in column (2) of Table 8 and the non-adjusted estimates in column (1) of Table 8.

44

The results are available from the authors upon request. Variables entered in the non-attrition equation are the status of employment (or interaction terms between the status of employment and region-specific variables), individual characteristics (age, age squared, completed grade level in school, a dummy for high education, a dummy for urban residence), attitudinal characteristics (dummies for pessimism, concern about obtaining daily necessities, concern about job loss, and satisfaction with life at present), a dummy for house ownership, and a dummy for inter-regional migration prior to the survey. Estimates of the first-stage probit are available upon request. 45

32

Table 8. Effects on Couple Formation during 1995-1998: Alternative Estimates Dependent Variable: Couple Formation (1=Yes, 0=No)

Panel A: Men (1) Not employed in 1994 (2) Not employed in 1994 * income corr. Panel B: Women (1) Not employed in 1994 (2) Not employed in 1994 * income corr.

Probit marginal effects

Nested probit marginal effects

Re-weighted estimates

(1)

(2)

(3)

-0.091 (0.043)

-0.086 (0.044)

-0.088 (0.043)

-0.210 (0.066)

-0.199 (0.070)

-0.197 (0.069)

-0.030 (0.017)

-0.033 (0.018)

-0.025 (0.015)

-0.038 (0.017)

-0.031 (0.015)

-0.030 (0.014)

Notes: The sample consists of all individuals who were single and aged 16-45 in 1994. Each entry corresponds to a separate regression according to the specifications in equations (1) and (2) in the text. Robust standard errors are in parentheses. Probit marginal effects are estimated ignoring sample selection issues. Nested probit marginal effects are adjusted for the sample-selection bias by a joint estimation of the non-attrition and the incidence of couple formation. Re-weighted estimates are obtained by re-weighting the sample to take account of non-random attrition using observed characteristics to predict the probability of selection.

The second procedure to correct for potential sample-selection bias is a simple datadriven approach, which re-weights the sample to take account of non-random attrition using observed characteristics to predict the probability of selection (as, for example, in Fitzgerald et al.). The comparative strength of this approach is that it does not rely on a functional form and distributional assumptions for identification, while its comparative weakness is that it does not account for selection on unobservables. After fitting the selection equation,46 in the second stage, each observation in the narrower sample of single individuals who remained in the sample was re-weighted by its predicted selection probability and the couple formation equations were re-estimated. The results from this approach are summarized in column (3) of 46

The set of covariates used to predict selection is identical to the set used in the first stage of the nested probit model discussed above.

33

Table 8. Again, the estimates yielded by this approach are similar to those in columns (1) and (2). In summary, the estimates appear to be robust to alternative methods of adjusting for the non-representative nature of the sample caused by endogenous migration. This reinforces our confidence in the applicability of the results to a broader population of individuals.

4. Concluding Remarks It seems fairly uncontroversial to argue that the high rates of unemployment experienced in Russia, as in many other transition countries during the past 15 years, damaged young people’s chances of a successful labor market early career path. In this paper we ask whether labor market conditions, along with the associated correlation in economic outcomes across would-be spouses, also impact couple formation. We have investigated several plausible explanations of the observed trends in the Russian marriage rate. We have identified a strong crossover effect from the labor market to the marriage market. In particular, we find that sharply increasing unemployment is associated with a sharp decline in the marriage rate. This finding is yielded by both the macro-level evidence and the micro-level evidence. Hence, policy measures aimed at reducing unemployment could also affect the marriage market (indirect “subsidization” of marriages) rather than, as traditionally thought, merely subsidize job creation. We have shown that it is not a simple “cost story” that explains the dramatic decline of marriage rates during the transition. While the era of increasing unemployment favors a “cost argument,” the finding that a positive correlation of male incomes and female incomes contributed to the decline in the propensity for couple formation suggests that the weakening

34

of the appeal of marriage as a risk-mitigating institution played a role.47 Although we cannot categorically rule out the possibility that other social forces were at work, the weight of the evidence points to the important role of the “cost argument” in conjunction with a positive correlation of male incomes and female incomes in accounting for the weakened inclination of couple formation.48 We find that the decline in the marriage rate is not due to postponement of marriages, but rather due to avoidance of marriages. The decline in the marriage rate has translated into a fertility decline: in the period 1989-1998, the total fertility rate declined by 38 percent.49 The fertility decline was mainly due to a decline in within-marriage fertility; during the same period, the share of non-marital births in total births increased from 13.5 to 28 percent. Absent evidence of a trend towards a compensating increase in within-marriage fertility, the incidence of marriage avoidance in the current generation is bound to result in a smaller workforce in the next generation. This repercussion, in turn, could entail new “structural adjustment” difficulties. In other words, even though unemployment is reversible, its demographic repercussions can have irreversible economic consequences. Of course, a test of this scenario requires a joint study of marriage and fertility decisions and behavior. It is an intriguing possibility, however, that a decade of rising unemployment and declining marriage rates in Russia might contribute to formidable changes in the labor market and in family structures in the future. 47 Our findings are in line with results reported by Hess (2004). Analyzing data on first marriages from the National Longitudinal Survey of Youth for the years 1978-1994 in the US, Hess finds that, controlling for several variables that potentially can impact on the survivability of marriages, more positively correlated incomes between spouses are associated with marriages of decreased duration: “the more positively related the spouses’ incomes are, the more likely that the marriage will end in divorce” (p. 312). 48 The goal of our analysis has not been to test an all-inclusive model of couple formation. Rather, we have tried to find out whether or not the stipulated mechanisms have the potential quantitative power to explain the observations on couple formation after the onset of the Russian economic and political transformation. Hence, several factors that may contribute to the understanding of the phenomena under study have been left out of the analysis, both for the sake of clarity and to facilitate tractability. 49 President Putin has recently termed Russia’s dramatic population decline “a creeping catastrophe.”

35

Finally, the decline in the marriage rate is likely to affect adversely productivity and life satisfaction in more subtle ways. If marriage makes spouses more productive by facilitating increased specialization in non-household production (Antonovics and Town, 2004), and if marriage and having children positively affect happiness (Frijters et al., 2004), then the economy that emerges from the era of marriage decline will be weakened, and the society will be less happy. Two additional considerations merit comment. First, it could be argued that the decision to divorce too might be sensitive to changing labor market conditions. After all, in the population of married couples, of those who experience bad events (such as unemployment), the couples most likely to resort to divorce are the ones for whom the withinmarriage utility falls by a particularly large amount. To address this perspective, we estimated a probit model of a married couple’s propensity to divorce. We found no evidence that the employment status or the correlation of male incomes and female incomes bear significantly on marriage dissolution.50 This is not all that surprising though. When mutual insurance gains are harder to come by economy-wide, the appeal of risk-pooling matching alternative to an existing marriage is weak. And as is well recognized, the decisions of getting into and out of marriage are asymmetrical. Once a marriage is formed, dissolving it is costly in a manner that differs in kind from the manner in which forming a marriage is costly. Legal costs arising from the division of joint property, loss of marriage-specific capital, and the suffering of children create frictions that mitigate the attraction of dissolving marriage when the mutual insurance gains from marriage erode. In particular, the accumulation of marriage-specific capital in the form of children increases the cost of divorce and creates an incentive for the

50

This finding seems to be in accord with the conclusion of Charles and Stephens (2004). Using a probit analysis based on the Panel Survey of Income Dynamics for the years 1985-1993 in the US, they find evidence against purely pecuniary motivation for divorce following earnings shocks.

36

marriage to last. And indeed, the probit estimates show that people with children were significantly less likely to divorce.51 Second, it would be of interest to examine systematically the patterns of assortative mating (that is, who marries whom in contemporary Russia). As illuminating as this may be, such an endeavor is beyond the scope of this paper. For sure, the factors that affect the inclination for couple formation are also likely to affect the pattern of marital matching. So far, there is no evidence in hand on whether the degree of assortative mating in Russia has increased or has decreased since the early 1990s. Extensions of our analysis could attend to this effect. Suppose that individuals differ in their labor market productivities. Assume that married males devote all their time to market work, while married females split their time between market work and household work. When choosing a potential mate, the potential mate’s earnings in the labor market will constitute a consideration. This consideration might matter more during times of economic crisis, that is, in such times the earning potential could be a more important criterion when choosing a mate. The degree of assortative mating would then rise.52 Such an analysis would likely imply that the decline in the marriage rate should be larger for individuals in the lower income groups. Additionally, our considerations could imply a particular assortative mating pattern: the individuals’ “permanent” characteristics or endowments influencing the level and variability in incomes would be similar (positive assortative mating with respect to the persistent attributes), but the correlation between income outcomes would be as low as possible. These considerations could gainfully feature in our future research. 51

It should be noted that it is difficult to be certain about the direction of causality in this context since it may be the confidence in the durability of the marriage that prompts people to bear children to begin with. 52 We have conducted a preliminary assessment of the extent of sorting among the married couples in terms of education. Overall, about 50 percent of husbands and wives have the same level of completed schooling. For the couples formed in the 1995-1998 period, the corresponding number is 65 percent. This result is consistent with the hypothesis of increased assortative mating. However, a much more careful research is needed prior to drawing definite conclusions.

37

Appendix: Sample Selection and Computation of Income Correlation

Table 3 summarizes cross-sectional samples of all individuals aged between 16 and 45 in each survey year. The regression analysis in section 3 is based on longitudinal samples of all single individuals aged 16-45 in years 1994 and 2000, which holds constant the cohort composition of the samples for the pre-1999 period and for the post-1999 period. For the pre1999 period, the dependent variable is the propensity for couple formation within the period 1995-1998 (that is, between rounds 6 and 8 of the RLMS) of all single individuals aged 16-45 in 1994 (that is, in round 5 of the RLMS). For the post-1999 period, the dependent variable is the propensity for couple formation within the period 2001-2002 (that is, between rounds 10 and 11 of the RLMS) of all single individuals aged 16-45 in 2000 (that is, in round 9 of the RLMS). When computing the correlations of male incomes and female incomes, we first calculated hourly wages (measured as monthly earnings divided by monthly hours worked).53 Then we expressed the wages in constant (1994 and 2000) rubles, obtained by deflating nominal values by the Consumer Price Index (CPI). Unfortunately, regional CPI data that correspond to the regions as defined in the RLMS are not available. Thus, it is assumed that the national price level is applicable to all regions or, at least, that the regional price levels are uncorrelated with the regional income levels. We apply two methods to calculate a measure of the expected spousal income correlation. Following Schultz (1994), the first method is based on estimating the probability of work and the expected hourly wages of potential spouses. That is, a probit equation for the probability that a single individual is a wage earner is jointly estimated with an equation for 53

The cross-correlation between male wages and female wages were higher than the cross-correlation between male earnings and female earnings, suggesting that hours worked are not independent of hourly wages.

38

the individual’s expected log wage, conditional on the individual being a wage earner. The probit equations for the probability of being employed include all the variables in the wage equations, together with a variable indicating the occupational mix of the community.54 One advantage of the joint estimation of the equations for the probability of being a wage earner and the corresponding log wage equations is that the resulting estimated wage equations are corrected for selection into the samples of individuals with observed wages. This allows us to impute a wage for every single person even if the person does not work.55 As pointed out by Dooley et al. (2000), it is particularly important that when it comes to women, that we measure earnings capacity and not actual earnings since a change (or an expected change) in marital status might be accompanied by a change in work hours and weeks of market work. The equations are estimated separately for single men and for single women. We also imputed the wages of each single individual’s potential (that is, predicted) partner by sample-selection corrected wage equations specified identically as the equations for the individuals themselves. This required data for both spouses and we have used data on already-formed couples. As in Schultz, we regressed each married individual’s wage on a vector of characteristics of the spouse so as to avoid including match-specific variables that constitute jointly determined aspects of the marital unions. Predicted log hourly wages of individuals and their potential spouses are then used to calculate the correlations of the income of the individual with the income of the potential partner separately for the period 1994-1998, and for the period 20002002. The assumption underlying this method is that single individuals use the current hourly wages to form expectations of future earnings prospects. As already noted in the text of the

54

Schultz uses different identifying variables: individual property income and two measures of the generosity of welfare programs. 55 Put differently, the estimated wage equations are used to calculate predicted values of wage offers that would be received by each individual if he worked.

39

paper, this assumption is not out of place for a transition country. This measure of the expected spousal income correlation yields individual-specific estimates. The second method - the results reported in the paper draw on this method constructs, in turn, a community-level measure of income correlation. The method is based on constructing a proxy of the economic opportunities that potential spouses face. Following Wood (1995), we compute a wage index using the occupational structure in community j.56 The index is computed separately for men and for women, and is defined as follows: g W jtg = ∑o wotg eotg s ojt

where g stands for gender, o for occupational category, j for community, t for year, and o for occupational category; wotg is the national mean hourly wage of men (women) in occupation o in year t; eotg is the percentage of those employed in occupation o who are men (women) g divided by the percentage of all workers who are men (women) in year t; and s ojt is the

percentage of men (women) in community j who are employed in occupation o in year t. Since the same weights (wotg eotg ) are used for each community in a given year t, the wage index is driven by community differences in the overall occupation composition of the employed.57 The computed wage indices are then used to calculate the correlations of male incomes and female incomes. This method is less likely to be contaminated by reverse causality biases than the first method. In the empirical analysis, we match each individual to the computed overall measure of correlation between male incomes and female incomes in the individual’s community of residence, therefore disregarding assortative mating on the basis of education. Two reasons motivate this choice. First, changes in the correlation of incomes may 56

Wood actually uses the regional industrial structure. It would be appealing to construct indices based on the regional industry-occupation structure. Unfortunately, measures of the industrial structure of communities are not included in the data sets that are available to us. 57 A caveat of the index is that it may not capture within-occupation shifts in wages.

40

affect individual’s schooling decisions, thus casting doubt on the exogeneity of our measure of income correlation in a given education group. Second, data constraints render it difficult to construct accurate community-level measures of income correlations within finely defined cells.

41

References Ai, Chunrong, and Norton, Edward C. (2003) “Interaction Terms in Logit and Probit Models,” Economics Letters 80(1): 123-29. Antonovics, Kate, and Town, Robert (2004) “Are All the Good Men Married? Uncovering the Sources of the Marital Wage Premium,” American Economic Review 94(2): 317-21. Blau, Francine, Kahn, Lawrence, and Waldfogel, Jane (2000) “Understanding Young Women’s Marriage Decisions: The Role of Labor and Marriage Market Conditions,” Industrial and Labor Relations Review 53(4): 624-47. Brainerd, Elizabeth (1998) “Winners and Losers in Russia’s Economic Transition,” American Economic Review 88(5): 1094-116. Brien, Michael (1997) “Racial Differences in Marriage and the Role of Marriage Markets,” Journal of Human Resources 32(4): 741-78. Charles, Kerwin Kofi, and Stephens, Melvin, Jr. (2004) “Job Displacement, Disability, and Divorce,” Journal of Labor Economics 22(2): 489-522. Dooley, Martin D., Gascon, Stéphane, Lefebvre, Pierre, and Merrigan, Philip (2000) “Lone Female Headship and Welfare Policy in Canada,” Journal of Human Resources 35(3): 587602. Drewianka, Scott (2003) “Estimating Social Effects in Matching Markets: Externalities in Spousal Search,” Review of Economics and Statistics 85(2): 409-23. Duflo, Esther (2001) “Schooling and Labor Market Consequences of School Construction in Indonesia: Evidence from an Unusual Policy Experiment,” American Economic Review 91(4): 795-813. Fitzgerald, John, Gottschalk, Peter, and Moffitt, Robert (1998) “An Analysis of Sample Attrition in Panel Data: The Michigan Panel Study of Income Dynamics,” Journal of Human Resources 33(2): 251-99. Frijters, Paul, Haisken-DeNew, John P., and Shields, Michael A. (2004) “Investigating the Patterns of Life Satisfaction in Germany Following Reunification,” Journal of Human Resources 39 (3): 649-74. Gould, Eric D., and Paserman, Daniele M. (2003) “Waiting for Mr. Right: Rising Inequality and Declining Marriage Rates,” Journal of Urban Economics 53(2): 257-81. Hess, Gregory D. (2004) “Marriage and Consumption Insurance: What’s Love Got to Do with It?,” Journal of Political Economy 112(2): 290-318.

42

Kohler, Hans-Peter (2000) “Fertility Decline as a Coordination Problem,” Journal of Development Economics 63(2): 231-63. Kohler, Hans-Peter, and Kohler, Iliana (2002) “Fertility Decline in Russia in the Early and Mid 1990s: The Role of Economic Uncertainty and Labour Market Crises,” European Journal of Population 18(3): 233-62. Kotlikoff, Laurence J., and Spivak, Avia (1981) “The Family as an Incomplete Annuities Market,” Journal of Political Economy 89(2): 372-91. Loughran, David S. (2002) “The Effect of Male Wage Inequality on Female Age at First Marriage,” Review of Economics and Statistics 84(2): 237-50. Mare, Robert, and Winship, Christopher (1991) “Socioeconomic Change and the Decline of Marriage for Blacks and Whites,” in Jencks, Christopher and Peterson, Paul, eds., The Urban Underclass, pp. 175-202, Washington: The Brookings Institution. Neal, Derek (2001) “The Economics of Family Structure,” National Bureau of Economic Research Working Paper No. 8519. Ogaki, Masao, and Zhang, Qiang (2001) “Decreasing Relative Risk Aversion and Taste for Risk Sharing,” Econometrica 69(2): 515-26. Philipov, Dimiter, and Kohler, Hans-Peter (2001) “Tempo Effects in the Fertility Decline in Eastern Europe: Evidence from Bulgaria, the Czech Republic, Hungary, Poland, and Russia,” European Journal of Population 17(1): 37-60. Pischke, Jorn-Steffen (2005) “Labor Market Institution, Wages, and Investment: Review and Implications,” CESifo Economic Studies 51(1): 47-75. Rosenzweig, Mark R., and Stark, Oded (1989) “Consumption Smoothing, Migration, and Marriage: Evidence from Rural India,” Journal of Political Economy 97(4): 905-26. Sargent, Thomas J. (1993) Bounded Rationality in Macroeconomics. Oxford: Oxford University Press. Schultz, T. Paul (1994) “Marital Status and Fertility in the United States: Welfare and Labor Market Effects,” Journal of Human Resources 29(2): 637-69. Smith, Richard J., and Blundell, Richard W. (1986) “An Exogeneity Test for Simultaneous Equation Tobit Model with an Application to Labor Supply,” Econometrica 54(3): 679-85. Weiss, Yoram (1997) “The Formation and Dissolution of Families: Why Marry? Who Marries Whom? and What Happens upon Divorce,” in Rosenzweig, Mark R. and Stark, Oded, eds., Handbook of Population and Family Economics, Volume 1A, pp. 81-123. Amsterdam, New York, and Oxford: North-Holland.

43

Wilson, William J. (1987) The Truly Disadvantaged. Chicago: University of Chicago Press. Wood, Robert (1995) “Marriage Rates and Marriageable Men: A Test of the Wilson Hypothesis,” Journal of Human Resources 30(1): 163-93.

44