This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier’s archiving and manuscript policies are encouraged to visit: http://www.elsevier.com/copyright

Author's personal copy Journal of Retailing and Consumer Services 18 (2011) 81–91

Contents lists available at ScienceDirect

Journal of Retailing and Consumer Services journal homepage: www.elsevier.com/locate/jretconser

Consumer learning as a determinant of a multi-partner loyalty program’s effectiveness: A behaviorist and long-term perspective$ Jean Frisou a,n, He´le ne Yildiz b a b

INSEEC Business School Research Laboratory, 27, avenue Claude Vellefaux, 75010 Paris, France CEREFIGE Laboratory, 13, rue Michel Ney, 54037 NANCY, Nancy University IUT UHP, France

a r t i c l e in f o

Keywords: Loyalty program’s effectiveness Theory of commitment Cognitive evaluation theory Operant learning Points accumulation Points redemption Latent growth curve model

abstract In this article, we argue that a loyalty program’s effectiveness depends on how the consumer learns to use its rules. The proposed theoretical framework was tested on the behavioral trajectories of 1380 individuals observed over a four-year period. The tendency of customers to spend more over four years became increasingly pronounced as they learned how to accumulate loyalty points and asked for these to be redeemed. This finding suggests that a loyalty program’s effectiveness does not depend on the program alone. To obtain the loyalty behaviors, firms should take specific measures to help their customers familiarize themselves with the program’s rules. & 2010 Elsevier Ltd. All rights reserved.

1. Loyalty programs: learning versus automatic reflexes A key question facing firms is knowing whether their loyalty programs do in fact produce the intended results. Despite the efforts of researchers to provide an answer to this question, the results obtained are still very mixed. Thus the pioneering study by Sharp and Sharp (1997) reveals a low impact of loyalty programs on buying behavior. Moreover Mauri (2003) and Allaway et al. (2006) show that a high proportion of a loyalty program’s members do not remain loyal to it. Other studies, however, draw more positive conclusions. Some loyalty programs enhance customer retention (Verhoef, 2003), increase their share of wallet (Verhoef, 2003; Meyer-Waarden, 2007) and prolong the duration of the relationship (Meyer-Waarden, 2007). Such divergent findings have reinforced researchers’ skepticism and given rise to a succession of questions: ‘‘Do rewards really create loyalty?’’ (O’Brien and Jones, 1995), ‘‘Do customer loyalty programs really work?’’ (Dowling and Uncles, 1997), ‘‘Do reward programs build loyalty for services?’’ (Keh and Lee, 2006), ‘‘Do loyalty programs really enhance behavioral loyalty?’’ (Leenheer et al., 2007), ‘‘Brand loyalty programs: Are they shams?’’ (Shugan, 2005). Such questions seem to attribute the effectiveness of programs to the programs themselves and ignore the nevertheless determining roles of the customers using them.

$ This research has been financially supported by INSEEC Research Laboratory. The authors would like to thank the company DOING GROUPE NEYRIAL, and in particular Marie-Violette Gellet and Laurence Cote for making available the sample group data, which enabled the empirical study to be implemented. n Corresponding author. E-mail addresses: [email protected] (J. Frisou), [email protected] (H. Yildiz).

0969-6989/$ - see front matter & 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.jretconser.2010.10.002

In this paper we shall look at the specific case of multi-partner loyalty programs. The question of the effectiveness of loyalty programs has often been raised in relation to proprietary loyalty programs that involve only one retailer. Consumer loyalty to the program then implies loyalty to the retailer. The effectiveness of multi-partner programs, in which the same program is applied by several partner retailers, has been less often considered (Lara and Madariaga, 2007). In this second case, the consumer’s loyalty to the program does not necessarily entail loyalty toward each of the partner retailers. Indeed the consumer distributes his expenditure and points redemption among the different partners. Although these programs have become much more widespread (e.g., http://www.interrapro ject.org/about-us/communities/), they are only advantageous to a retailer if the accumulation of rewards in terms of the whole program has a positive effect on the purchases made and rewards obtained by the consumer at that particular retailer. It is important to understand the processes linking these two learning levels. In this study we make three contributions that distinguish it from those of our predecessors: (i) The first contribution is to change how we think about a program’s effectiveness. We propose approaching the effectiveness of loyalty programs not only on the basis of their ‘‘a priori’’ effects and their ‘‘a posteriori’’ results, but according to the way in which each participant learns how to use them. (ii) The second contribution is a behaviorist conceptual and theoretical framework that takes account of the interactions between the user and the program. Operant learning, which makes behavior dependent on its consequences, is currently used to explain how reward programs work (Foxall, 1997; Taylor and Neslin, 2005; Liu, 2007). However, the authors continue to view

Author's personal copy 82

J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

perceptual and cognitive processes as the main factors influencing the programs’ effectiveness (Taylor and Neslin, 2005; Keh and Lee, 2006; Wirtz et al., 2007). We propose re-adopting the behaviorist theory of operant learning, while linking a program’s effectiveness to the manifest behaviors of its users (Skinner, 1965). By ‘‘behaviors’’, we understand all behaviors that the program gives rise to in users: not only buying behavior, but also loyalty point accumulation and redemption behaviors, which structure their learning about the program. (iii) The third contribution concerns the methodology used for modeling the learning process. The need to take account of the long term for measuring the effects of loyalty programs is frequently mentioned in the literature (Kopalle and Neslin, 2003; Taylor and Neslin, 2005; Liu, 2007). The latent growth curve models that we use place the emphasis on people’s behavioral trajectories over a number of years, rather than their immediate responses on a specific occasion. The paper is organized as follows. In the second part, we show that the evolving literature leads us to rethink the effectiveness of loyalty programs both as a result and as all the psychological processes that give rise to it. In the third part, we look at the diversity of reward schemes and the psychological theories able to explain them. In the fourth part, we offer a behaviorist and dynamic theoretical framework capable of understanding the learning processes involved in a loyalty program bringing together several partner retailers. In the fifth part we empirically test the hypotheses underpinning this model. Finally, we conclude with a discussion of possible future research and the managerial lessons to be drawn from this study.

2. The effectiveness of loyalty programs as result and process Measuring the performance of loyalty programs involves making two choices. The first concerns the concept to be adopted for assessing performance. The second is a question of how to envisage this concept. Assessing loyalty programs is simply one specific instance in the measurement of marketing activities. By ‘‘activities’’, researchers mean the tactics of operational marketing rather than the underlying strategies (Rust et al., 2004; O’Sullivan and Abela, 2007). Because loyalty programs are levers of operational marketing, measuring their performance becomes a key issue. But this is not a straightforward task, since three dimensions of the performance of a marketing action are currently offered in the literature. A first dimension of performance is adaptability. This expresses the degree of fit of the action to the marketing environment. The second dimension is efficiency. This measures the relation between the results of an action and the means used to implement it (Clark, 2000; Morgan et al., 2002). Lastly, the third dimension is effectiveness, which expresses the match between the results obtained and the results expected from the marketing action (Clark, 2000; Morgan et al., 2002). Effectiveness, however, seems to be the generally preferred aspect of performance. It represents the most significant dimension for managers when they assess the performance of their marketing programs (Clark, 2000). But researchers also use it to measure the performance of loyalty programs (Sharp and Sharp, 1997; Wirtz et al., 2007; Liu, 2007; Leenheer et al., 2007). Although effectiveness has become a generally agreed criterion for measuring the performance of loyalty programs, there are at least two other ways of viewing this. In the one, effectiveness is measured at the end of the marketing operation (Clark, 2000). A loyalty program is then effective if it produces the expected results. In the other, effectiveness is measured as the marketing operation takes place. It is assimilated to a process that aims to optimize the results of the operation (Kahn and Myers, 2005). From this

The two perspectives of the loyalty program’s effectiveness Loyalty Program Membership

Effectiveness as “processes…

Effectiveness as “result”

Control of psychological processes induced by rewards

Loyalty Behaviors

…and results”

Fig. 1. The two perspectives of the loyalty program’s effectiveness.

standpoint a loyalty program is effective if the monitoring and control of its functioning enables the best results to be obtained (Fig. 1). Within the perspective of effectiveness seen as a ‘‘result’’, researchers leave aside the way in which the loyalty program operates and simply make sure that the targeted objectives are attained. Their methods are based on three types of comparison (Liu, 2007): comparison of the results of firms or brands using a loyalty program and the results of those that do not use them ¨ (Sharp and Sharp, 1997; Magi, 2003; Leenheer et al., 2007); comparison of the buying behavior of customers belonging to loyalty programs and the behavior of those who do not (Bolton et al., 2000; Verhoef, 2003); and comparisons between the behavior of people belonging to a loyalty program, from one period to the next (Lal and Bell, 2003; Taylor and Neslin, 2005). These comparisons are intended to show whether or not a loyalty program has an effect on users’ buying behavior. They provide no explanation as to the causes of any differences that may be noticed. In the conception of effectiveness viewed as a process and a result, researchers focus their attention on the functioning of the loyalty program. They are interested in the psychological processes liable to influence users’ behavior. Thus Taylor and Neslin (2005) point to the psychological pressure of loyalty points on the user. Keh and Lee (2006) examine the link between rewards and the user’s buying behavior. Wirtz et al. (2007) look at the effect on the user’s buying behavior of the program’s attractiveness and switching costs to another program. Nevertheless, while these studies situate effectiveness within the program’s functioning, it is still the program that, through its intrinsic properties, is the main source of effectiveness. It seems to us to be important to change this view of things, by showing that a program’s effectiveness also has extrinsic causes. It depends on what the user does with it.

3. Which psychological theory best accounts for reward schemes? Researchers locate the main sources of the effectiveness of loyalty programs in the reward schemes accompanying them. But two aspects of the question are of particular interest: on the one hand, how customer loyalty is rewarded, and on the other, what kind of responses elicit such rewards. These two aspects are closely linked, since customer responses vary according to the type of reward (Dowling and Uncles, 1997). 3.1. The varied effects of reward schemes A loyalty program is defined as an integrated system of marketing operations with the aim of making customers subscribing to it more loyal (Leenheer et al., 2007). Such operations involve offering rewards to customers according to the frequency and volume of

Author's personal copy J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

their purchases (Yi and Jeon, 2003). Loyalty programs and reward programs are thus often synonymous in the literature (Taylor and Neslin, 2005; Keh and Lee, 2006; Wirtz et al., 2007). Nevertheless, the idea of winning customers’ loyalty by promising them rewards is not limited to loyalty programs. In recent years, companies have attempted to develop customer loyalty by incorporating into their offering the promise of a higher quality of products or services (Zeithaml et al., 1998). Loyalty programs have the same goal, but their way of achieving it differs. The promised reward is not part and parcel of the product, but is external to it and is obtained followed a succession of purchases (e.g. a gift or a discount). The literature offers three classifications of reward schemes. In the first, there is a posited opposition between direct and indirect rewards (Dowling and Uncles, 1997; Yi and Jeon, 2003; Keh and Lee, 2006). A direct reward is a benefit that supports the value proposition of the product or service. Free flights on US Airways obtained with US Airways Dividend Miles are direct rewards. Conversely, a reward is indirect if it has no direct link with the product or service provided. In the American Express Membership Rewards program, the gifts (e.g. high-tech products, gourmet products, holidays, etc.) obtained by paying with their credit card are indirect rewards. The second classification makes a distinction between immediate and delayed rewards. Immediate rewards are obtained after making the first purchase, whereas delayed rewards apply only after a number of purchases have been made (Dowling and Uncles, 1997; Yi and Jeon, 2003; Keh and Lee, 2006). The rewards offered by Etihad Airways, which come into effect from the first air mile flown, are immediate rewards, while the free flights with American Airlines obtained by exchanging air miles accumulated over a period of time are delayed rewards. The third classification is based on the nature of the rewards, with a distinction being made between tangible and intangible rewards. Tangible rewards consist of discounts, purchase vouchers, and various gifts of a material nature. Conversely, intangible rewards provide the consumer with non-material benefits, such as personalized information or the feeling of being a favored customer (Roehm et al., 2002; Leenheer et al., 2007). The marketing literature distinguishes various effects produced by rewards, including effects on buying behavior as well as how the loyalty program is perceived. The first studies concluded that loyalty programs had a limited effect on buying behavior (Sharp and Sharp, 1997, 1999), but more recent work confirms that this effect is substantial (Taylor and Neslin, 2005; Leenheer et al., 2007; Meyer-Waarden, 2007). The program’s value as perceived by the consumer is also influenced by rewards (O’Brien and Jones, 1995; Yi and Jeon, 2003). In situations of high consumer involvement, Yi and Jeon (2003) show that direct rewards have a greater effect on the program’s perceived value than indirect rewards. They also show that in a low involvement situation, the effect of immediate rewards on the program’s perceived value is stronger than that of delayed rewards. Nevertheless, the impact of rewards does not necessarily benefit the brand. The perceived value of a loyalty program depends more on the attraction of the rewards than the brand itself (Yi and Jeon, 2003). Thus such programs are more liable to create loyalty to the program than loyalty to the brand. The brand does not therefore gain greater protection from its competitors through a loyalty program. In the customer’s mind, the loyalty program is dissociated from the brand and itself is subject to competition from other, more attractive programs (Wirtz et al., 2007). These studies clearly show the many effects of rewards on users’ attitudes and buying behavior. But they do not explain in sufficient detail the underlying psychological processes. Theories derived from social psychology offer perspectives that can deepen our understanding of these phenomena.

83

behavior: the theory of operant learning, the theory of personal commitment and the theory of cognitive evaluation. Skinner’s (1965) theory of operant learning was first put forward to explain the effect of sales promotions (Rothschild and Gaidis, 1981), then to explain the mechanisms of loyalty programs (Foxall, 1996, 1997). This psychological theory is based on a straightforward idea. When an individual manifests a behavior in the presence of a stimulus and receives a reward or punishment, this stimulus becomes discriminative. In similar situations, the presence of the stimulus increases the probability that the individual will repeat the behavior if it has previously been rewarded, and will decrease the probability if it has been penalized (contingencies of reinforcement). Reward programs are based on the same principle. Points gained are indirect rewards functioning as secondary reinforcers (Foxall, 1996). The reductions or discounts obtained in exchange for the points gained play the role of primary reinforcers (Foxall, 1996). The loss of points is a penalty, and encourages the customer to ask for his points to be redeemed within the given time frame (negative reinforcement). Loyalty is therefore not automatically produced by the program, but depends on the ability of the consumer to learn the often complex and changing rules of the system. In the theory of commitment, the rewards offered to induce a particular action tend, on the contrary, to reduce the probability that it will be repeated. This theory, which has been corroborated by the facts on many occasions, asserts that carrying out a relatively easy, freely consented to and publicly exhibited action commits its perpetrator to repeat it subsequently (Kiesler, 1971). Belonging to a loyalty program corresponds to this type of action. It calls for very little effort on the part of the customer – the provision of personal information, for example – is visible to all and agreed to without pressure. The commitment conditions identified by Kiesler (1971) are all present, and these encourage the customer to continue in a relationship the principle of which he has freely accepted. The individual feels committed because he can attribute his participation in the program to himself. Kiesler (1971) nevertheless showed that when an action is rewarded, the individual no longer attributed this action to himself but to the rewards he obtains in exchange. The commitment effect then tends to disappear. The rewards received can therefore reduce users’ commitment to a program over time and can account for their progressive withdrawal. Finally, the theory of cognitive evaluation maintains that people’s intrinsic motivations lie in their need for autonomy and competence (Deci et al., 1999). According to whether or not the rewards obtained satisfy these needs, they respectively increase or reduce people’s motivation to repeat the rewarded behavior. If, when an individual carries out a task, he gets a reward independent of the task, he does not feel controlled by this reward. His autonomy and competence needs are preserved and his motivation to repeat the behavior is unchanged. On the other hand, if the reward depends on the task, the individual can experience this as a restraint on his behavior. His need for autonomy and competence will be thwarted and his motivation to continue acting in the same way will be reduced (Deci et al., 1999; Jordan, 1986). For example, a participant in a loyalty program whose rules are too complicated can feel controlled by the rewards and abandon the program because he feels his autonomy is compromised (Shugan, 2005). Conversely, if the rules allocating points call on the user’s astuteness, he can then experience a feeling of competence that will encourage him to continue with the program.

4. Learning in a multi-partner program: a behaviorist and dynamic theoretical framework

3.2. Contradictory behavior/reward psychological theories We shall limit our presentation to the three main contributions of social psychology dealing with the effect of rewards on human

The many contributions from the literature on the performance of marketing actions invite us to rethink the theoretical framework around the effectiveness of loyalty programs, by viewing this as an

Author's personal copy 84

J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

outcome and as all the processes resulting in this outcome (Kahn and Myers, 2005). It is then a matter of identifying which of the various processes described in the literature lead to the success or failure of the program. For this reason, we will not use behavioral commitment theory, which predicts that rewards reduce people’s commitment and are therefore a potential source of failure for programs primarily based on rewards (Kiesler, 1971). Nor will we use, and for the same reason, cognitive evaluation theory. According to its authors, rewards associated with purchases can frustrate the purchaser’s need for independence and competence, and reduce his motivation to adhere to the program (Deci et al., 1999). Instead we shall draw on operant learning theory, which is most often referred to in the marketing literature to explain the behaviors induced by a loyalty program (Foxall, 1996, 1997 ; Liu, 2007). This theory maintains that people tend to reproduce behaviors that have been rewarded—precisely the expected outcome of loyalty programs. The impact of loyalty programs has been mainly studied in relation to changes in users’ buying behavior (Sharp and Sharp, 1997, 1999; Taylor and Neslin, 2005; Meyer-Waarden, 2007). However, the use of a loyalty program gives rise to behaviors other than purchasing. A loyalty program is comparable to a board game, where the consumer has to learn the rules in order to maximize his rewards. As well as buying behavior, learning to play the loyalty program game involves two other forms of behavior – accumulating points and redeeming points – that we must define and conceptualize. Foxall (1996 p.286) considers accumulation to be a ‘‘planned acquisition of a series of reinforcers which have some hedonic content but which are principally informational’’. Accumulating points thus presupposes learning the rules of the program in order to apply them better and win maximum points. The customer will, for example, make his purchases on Tuesday rather than Saturday, because on Tuesday the points gained are doubled or trebled. The bonus points inform the customer positively as to his ability to manage the program successfully (Foxall, 1996). Points redemption behavior can be seen in a similar way as the planned acquisition of hedonic or utilitarian reinforcers. It consists of converting the points gained (intangible rewards) into reductions or gifts (tangible rewards).

4.1. Learning: concepts and definitions Learning is defined in the literature in two different ways. Within a factual perspective, it is defined by the observable events characterizing it. Within a theoretical perspective, learning is, on the contrary, defined by the non-observable psychological processes that explain it (Kimble, 1961). We favor this second approach, and learning in relation to a loyalty program will therefore be understood through latent variables. These represent the tendencies acquired through learning and give rise to the manifest purchasing behavior, points accumulation behavior and points redemption behavior induced by the program. The conceptualization of the behaviors involved in learning a program has to resolve two problems. On the one hand, such behaviors are oriented toward the long term and their definition should take account of this temporal aspect (Liu, 2007; MeyerWaarden, 2007). On the other hand, product purchases and the acquisition or redemption of points are moves by the consumer that often come up against obstacles (Bagozzi and Warshaw, 1990). A stockout in a store or forgetting one’s loyalty card prevent these behaviors achieving their aim. Thus behavior achieved and observed must be distinguished from the tendency that guides it. This latent tendency is defined as a ‘‘probability of response’’ which produces effective behavior if there is no obstacle in the customer’s environment that conflicts with it (Coutu, 1949;

Skinner, 1965). Taylor and Neslin (2005) examine two stages in learning a loyalty program. During the first, the consumer is subject to the pressure of points until such time as the desired gift or discount is obtained. The program is deemed to be effective if the buying behavior is maintained or increases after this reward has been obtained. Our longitudinal approach covers several years, but also makes a distinction between two periods. The first year, or initial period, is an adaptation stage during which the customer familiarizes himself with the loyalty program. He receives his first rewards and establishes his initial use levels. In subsequent years, or the development period, the consumer adjusts his initial use levels, either maintaining them or increasing them if his learning has been successful, or reducing them if it has failed. The tendency to increase purchases over time represents an effort to control on the part of the consumer, in order to overcome the obstacles arising in purchasing situations (Oliver, 1999). We therefore define the tendency to loyalty behavior as the customer’s propensity to maintain or increase his initial level of expenditure from one period to the next despite the obstacles he encounters (Oliver, 1999). Similarly, we define the tendency to accumulate points as the customer’s propensity to maintain or increase the expected initial level of points from one period to the next. Finally, we define the tendency to redeem points for rewards as the customer’s propensity to maintain or increase over time the initial level of his redemption demands. 4.2. Research hypotheses Our research hypotheses apply to a multi-partner loyalty program. Each store participating in this type of program benefits from the increased purchasing by its customers obtained from the points given out by all the participating stores. By positioning ourselves from the standpoint of one of the partner retailers – which we call the witness store – three long-term processes characterize learning about the program by that store’s customers, which are to its advantage: (1) an accumulation of points process, involving all the participant retailers, the trajectory of which reveals the customer’s loyalty or lack of loyalty to the multipartner program, (2) points redemption process that only concerns the witness store and whose trajectory expresses the consumer’s loyalty or lack of loyalty to that store, and (3) a purchasing process in the witness store, the trajectory of which expresses the consumer’s loyalty or lack of loyalty to the witness store. Our research hypotheses concern the relations between the accumulation of points, the redemption of points, and purchasing, both during the initial adaptation period and in the consolidation period. In the initial period, the points (secondary, conditioned reinforcers) control buying behavior because the customer associates them with tangible delayed rewards (backup reinforcers). During this adaptation period, we therefore expect a positive effect from the customer’s anticipated points level on his intended level of expenditure in the witness store he frequents. Hence the hypothesis: H1. The expected initial points level has a positive effect on the expected initial level of expenditure in the witness store. But the rules of a loyalty program also set the time limits beyond which the points gained by the customer are lost. These rules exert pressure through the points pushing the customer to act in order to preserve his points capital (Taylor and Neslin, 2005). Redeeming them avoids the risk of losing them and brings the customer a sense of relief that is equivalent to a reward (Skinner, 1965). This negative reinforcement of redemption behavior is all the stronger the higher the initial expected level of points. Hence the second hypothesis: H2. The expected initial points level has a positive effect on the expected initial redemption level in the witness store.

Author's personal copy J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

Redeeming points also brings the customer tangible rewards which reinforce his buying behavior in the store which gives him these rewards. Many researchers emphasize the determining role of these primary reinforcers which shape buying behavior (Nord and Peter, 1980; Rothschild and Gaidis, 1981; Foxall, 1996, 1997; Sharp and Sharp, 1997). Hence our third hypothesis: H3. In the initial period the level of redemption the customer claims in the witness store has a positive effect on the expected initial level of expenditure in this store. In subsequent years, the success of the program depends on how in the long term the customer’s initial use levels change. He has learned which behaviors are best remunerated in terms of points. Accumulating points, redeeming points, and the amount spent on purchases should therefore progressively increase over the initial levels. We therefore have good grounds for extending to the long term the hypotheses advanced for the initial use levels. Hence the three following additional hypotheses: H4. The tendency to change the initial level of points has a positive effect on the tendency to change the initial level of expenditure in the witness store. H5. The tendency to change the initial level of points has a positive effect on the tendency to change the redemption level in the witness store. H6. The tendency to change the redemption level has a positive effect on the tendency to change the initial level of expenditure in the witness store. 5. Longitudinal research methodology

85

percentages should be treated with caution, since each loyalty card can be used by any member of the family. The principle of the program is simple. Every h3 spent in any of the 33 participant stores gives the bearer of the loyalty card one loyalty point. When 100 points are accumulated, the bearer has the right to a reduction or a purchase voucher worth h10. Any points not converted into rewards within a year are lost. We have three series of individual data for the years 2002 to 2005: (1) the total number of points won in all the stores, (2) the total number of points redeemed in the stationery store (direct tangible rewards), and (3) total expenditure in the stationery store. The main descriptive statistics pertaining to these data are summarized in Table 1.

5.2. Measurement models of constructs The operationalization of the concept of tendency, applied to buying behavior, points accumulation behavior and points redemption behavior, is implemented by means of latent growth curve models (Duncan et al., 1999; Kaplan, 2000; Bollen and Curran, 2006). Although we considered the tendencies pertaining to three distinct behaviors (accumulation, buying, redemption), the way of operationalizing them is the same and will be presented by taking buying behavior as an example. In a given purchasing situation, the buying behavior observed can be split up into two components. One is an underlying, nonobservable or latent buying behavior, the other an adaptation behavior due to the situation’s positive or negative influence on buying behavior. We therefore have the following identity: Observed buying behavior ¼ non-observable underlying buying behavior þ adaptation to the situation behavior:

5.1. Data collection The data used in this study concern members of a multi-partner loyalty program. This reward scheme covers 33 stores in a mediumsized French town near a metropolitan region with a million inhabitants. The participating stores belong to various sectors in retail distribution (e.g. food, clothing, personal services, IT, stationery, etc.). The cohort studied is made up of the customers of a stationery store which joined the program in 2001. This store was chosen because of its great involvement as a driver of the loyalty program. The sample represents the entire cohort, and comprises 1380 people of whom 54% are women and 46% men. These

If, for reasons of simplification, we accept a linear development of buying behavior, buying behavior observed in the long term can be represented by two trajectories: a manifest trajectory of the consumer’s repeated expenditure, represented by the dotted line (Fig. 2A); and a non-observable latent trajectory of the underlying or probable buying behavior, represented by the solid straight line (Fig. 2A). The gap observed in each period between these two lines (e.g. ei1) represents the customer’s adaptation to the purchasing situation (e.g. stockout, anticipated purchase, etc.). For example, in the event of a stockout, the consumer will switch to another brand.

Table 1 Descriptive statistics (correlations, means, standard deviations). Years

N ¼1380

Expenditures e1

2002 2003 2004 2005 2002 2003 2004 2005 2002 2003 2004 2005 Means Std-dev

e1 e2 e3 e4 p1 p2 p3 p4 r1 r2 r3 r4

1.00 0.60 0.48 0.51 0.26 0.14 0.15 0.14 0.34 0.28 0.10 0.14 54.6 91.1

e2

Points e3

1.00 0.67 0.63 0.21 0.28 0.28 0.21 0.12 0.35 0.16 0.26 33.9 83.2

1.00 0.68 0.18 0.25 0.34 0.24 0.11 0.30 0.43 0.32 31.7 80.1

All the correlations greater than 0.05 are significant.

e4

1.00 0.14 0.20 0.24 0.28 0.08 0.39 0.24 0.37 30.0 84.8

p1

1.00 0.66 0.58 0.51 0.09 0.03 0.02 0.09 267.5 370.1

Rewards p2

1.00 0.74 0.66 0.02 0.09 0.05 0.12 195.7 294.3

p3

1.00 0.74 0.08 0.11 0.11 0.15 155.8 246.1

p4

1.00 0.01 0.07 0.06 0.15 127.6 240.4

r1

1.00 0.07 0.12 0.16 12.7 50.3

r2

1.00 0.29 0.23 11.3 48.4

r3

1.00 0.31 9.3 45.0

r4

1.00 9.7 41.6

Author's personal copy J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

86

e it (expenditures)

e i3

1

Latent trajectory

μβ

μα

ζα

ei1

ζβ

ε i1 ei2 αi

1

ei0

Latent trajectory

βi

αi

β i1

1 1

1

1

2

3

eit = αi + βi * t + εit ei0

0

1

2

3

εi0

time

ei1 εi1

ei2 εi2

ei3

Observed trajectory

εi3

Fig. 2. Latent and observed individual trajectories and associated structural model: (a) Latent and observed trajectories and (b) associated path diagram.

The measurement model of repeated buying behavior comprises two latent factors (Fig. 2B). These are interpreted as chronometric common factors. The latent factor a represents the initial level of customers’ expenditure during the first period ‘‘t¼0’’. The latent factor b expresses customers’ long-term latent tendency to change the initial level of expenditure a from one period to the next. In mathematical terms, a is an ‘‘intercept’’ and b a ‘‘slope’’. The amount of observed expenditure ‘‘eit’’ for customer ‘‘i’’, at each period ‘‘t’’, is written eit ¼ ai þ bi* t þ eit

ð1Þ

In the first period t ¼0, Eq. (1) takes the following form (2): ei0 ¼ ai þ ei0

ð2Þ

ai would therefore be equal to the initial level of expenditure ei0 if the behavior due to the purchasing situation ei0 had no effect on the consumer’s choice (i.e. ei0 ¼0). For the following periods (i.e. t ¼ 1, 2, 3), the levels of expenditure are given by Eq. (1). In this equation ai represents the initial level of expenditure aimed for at t ¼0 by consumer ‘‘i’’, bi*t the linear variation of the initial level of his expenditure during period t, and eit the positive or negative influence of the purchasing situation on his expenditure in period ‘‘t’’. If bi Z0, the buying tendency is increasing or unchanged and corresponds to a tendency to loyal behavior. If bi o0, the buying tendency is decreasing and corresponds to disloyal behavior. ai and bi characterize the latent buying trajectory of consumer ‘‘i’’ in the program. The following Eqs. (3) and (4) complete the specification of the measurement model: ai ¼ ma þ zai

ð3Þ

bi ¼ mb þ zbi

ð4Þ

In Eq. (3) the initial level of expenditure ai of consumer ‘‘i’’ is a function of the mean of all the consumers’ initial expenditure levels (ma) and of a random component specific to consumer ‘‘i’’, zai. Similarly the latent tendency bi of consumer ‘‘i’’ to change his initial level of expenditure is a function of the mean of all consumers’ tendencies (mb) and of a random component specific to consumer ‘‘i’’, zbi. Eqs. (1) and (2) define level 1 of the model (‘‘within person model’’) and Eqs. (3) and (4) define level 2 of the model (‘‘between person model’’) (Curran, 2000). In Fig. 2A we show the latent and observed trajectories of individual ‘‘i’’. In Fig. 2B, we reproduce the measurement model’s path diagram of the tendency associated with these curves. This measurement model also applies to points accumulation and redemption behaviors. A growing tendency to accumulate points means that the loyalty program member is increasingly well able to use the program’s rules. Similarly a

growing tendency to redeem points means that the customer is increasingly well able to manage his points account. 5.3. Models and results Testing the research hypotheses was carried out using two distinct models. The first is a multivariate model combining three latent growth curves (M1), and allows the six hypotheses to be tested. The second is a multi-group model applied to the single latent curve of buying behavior (M2). This exploratory multi-group model is a methodological procedure that enables the results of the previous model to be deepened, by distinguishing in advance four distinct forms of learning. To make this paper easier to read, we present the mathematical aspects of models M1 and M2 in Appendix A and B. A preliminary analysis of the data collected did not enable their multinormality to be established. The Mardia-based kappa index was estimated at 6.403, while it should be around zero if the data were multinormal. The violation of the multivariate normality hypothesis affects the estimation of the chi-square and standard errors (Bollen, 1989). To overcome this problem we estimated the two models with Muthe´n and Muthe´n’s (2004) appropriate MLMV estimator, which produces a scaled chi-square and robust standard errors (Satorra and Bentler, 1988). We re-estimated the standard errors of the parameters with a 1000-sample bootstrap. 5.3.1. First model: testing the hypotheses The first structural model M1 describes the operant learning process defined by the first six hypotheses. The model is estimated with the data available from the 1380 individuals making up the cohort. Its path-diagram is shown in Fig. 3. The model’s goodness-of-fit is acceptable, on the basis of the cut-off criteria for fit indices accepted in the literature (Hu and Bentler, 1999). The estimated value of w2 is 13.985, with 5 degrees of freedom and a p-value estimated at 0.016. The estimations of the CFI and TLI indices come out at 0.94 and 0.90, respectively. They are slightly lower than the minimum threshold recommended for these indices ( Z0.95). The estimation of the SRMR (0.056) is higher but close to the maximum recommended threshold ( o0.05) and that of the RMSEA (0.036) is lower than the generally accepted maximum threshold ( o0.05). The estimations of the structural model M1 parameters are shown in Table 1. Hypothesis H1 predicts that the expected initial points level (ap) has a positive effect on the expected initial level of expenditure (ae). The results show a positive and statistically significant standardized coefficient, which confirms this hypothesis (gap1 ¼0.258, po0.001).

Author's personal copy J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

Tendency to change the initial level of points in the 33 stores βp

87

Expected initial level of expenditure in the stationery αe

γαp1

γβp1 Tendency to change the initial level of expenditure in the stationery βe

Expected initial points level in the 33 stores αp

γαr

γαp2

Expected initial redemption level in the stationery αr

γβp2

γβr

Tendency to change the redemption level in the stationery store βr

Fig. 3. Path diagram of model M1.

The more a customer accumulates points in this initial period, the more he buys in the stationery store during the same period. The loyalty program therefore has a positive impact on the customer’s buying behavior. Hypothesis H2 predicts that the expected initial points level (ap) has a positive effect on the expected initial redemption level (ar). The positive and statistically significant standardized coefficient corroborates this hypothesis (gap2 ¼0.291, po0.001). To retain the points gained in all the stores, the customer asks for them to be redeemed in the stationery store within the allotted time period. These direct rewards reinforce the symbolic value of the points and loyalty to the program (Foxall, 1996, 1997). Hypothesis H3 predicts that in the initial period the level of redemption the customer claims in the stationery store (ar) has a positive effect on the expected initial level of expenditure in this store (ae). This hypothesis is also confirmed by the results. We find a positive and statistically significant standardized coefficient (gar ¼0.615, p o0.001). The reinforcement of buying behavior by tangible rewards is therefore also established. Hypotheses H4, H5, and H6 predict similar results among the evolution tendencies in the long term. They too are confirmed. The tendency to change the initial level of points (bp) has a positive effect on the tendency to change the initial level of expenditure in the stationery store (be). The standardized coefficient is positive and significant (H4: gbp1 ¼0.379, po0.001). The tendency to change the initial level of points (bp) also has a positive significant effect on the tendency to change the redemption level in the stationery store (br H5: gbp2 ¼0.246, po0.001). Similarly the tendency to change the redemption level has a positive and significant effect on the tendency to change the initial level of expenditure in the stationery store (be H6: gbr ¼0.834, po0.001). Two indirect effects of points on expenditure, mediated by redemption, should be noted. The indirect effects between the two variables are mediated by at least one other variables. They are calculated as the product of the direct effects concerned (Bollen, 1989). In regard to the initial levels this effect is positive and significant (gap2  gar ¼0.179 po0.001); in regard to tendencies this effect is positive but less clearly significant (gbp2  gbr ¼0.205 po0.100).

The r2 of the initial level of expenditure (ae) and of the tendency to change expenditure (be) are estimated to be 0.54 and 0.99, respectively. These are good indicators of nomological validity. The results of model M1 concern all individuals in the cohort whatever their way of learning. Model M2 enables us to understand these users’ behavior in more depth according to their way of learning.

5.3.2. Second model: extending the results Another significant finding reported in the literature concerns the heterogeneity of behavior among a program’s users. Loyalty programs seem to work on some customers and not on others. Different user profiles have even been distinguished (Mauri, 2003; Allaway et al., 2006 ; Lewis, 2004; Liu, 2007). The underlying idea is that people do not all go through the same learning process in regard to the loyalty program, although the conditions for acquiring or converting points are the same for everyone. Indeed we can distinguish different learning profiles. We have thus segmented the sample of 1380 card bearers into four distinct groups (Fig. 4). Assigning individuals into each group was done according to the estimated factor scores in model 1. Group 1 contains 1022 individuals who underwent no learning. Their tendencies to change the initial level of points (bip) and the initial level of redemption (bir) are negative. Group 2 contains 233 individuals whose learning was motivated only by points. Their tendency to change the initial level of points (bip) is positive or zero and their tendency to change the initial level of redemption (bir) is negative. Group 3 is made up of the 75 individuals whose learning was motivated only by the rewards. Their tendency to change the initial level of points (bip) is negative and their tendency to change the initial level of redemption (bir) is positive. Finally group 4 contains the 50 individuals whose learning was motivated both by the points and the rewards. Their tendencies to change the initial level of points (bip) and to change the initial level of redemption (bir) are positive or zero (Table 2). The second model, M2, is a multi-group model applied solely to the consumer’s expenditure latent growth curve. It specification and mathematical features are shown in Appendix B. Customers

Author's personal copy J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

88

Segmentation of individuals into four groups Individuals not increasing their redeeming at the stationery βir < 0

Individuals increasing their redeeming at the stationery βir ≥ 0

Individuals not increasing their points βip < 0

Group 1: no learning either through points or redeeming 1022 individuals

Group 3 learning through redeeming 75 individuals

Individuals increasing their points βip ≥ 0

Group 2 learning through points 233 individuals

Group 4 Learning through points and redeeming 50 individuals

Fig. 4. Segmentation of individuals into four groups.

Table 2 Standardized estimations of structural relations in model M1. Model Hypothesis H1 H2 H3 H4 H5 H6 H2 and H3 H5 and H6 a b c

Robust estimation (MLMV) Coefficient

gap1 gap2 gar gbp1 gbp2 gbr gap2  gar gbp2  gbr

Process

ap-ae ap-ar ar-ae bp-be bp-br br-be

ap-ar-ae bp-br-be

Est. 0.258 0.291 0.615 0.379 0.246 0.834 0.179 0.205

S.E 0.069 0.068 0.105 0.142 0.100 0.083 0.051 0.086

Bootstrap (1000 samples) Est./S.E a

3.749 4.268a 5.868a 2.671a 2.467b 10.100a 3.504a 2.385b

S.E

Est./S.E

0.085 0.061 0.126 0.148 0.135 0.098 0.065 0.118

3.022a 4.073a 4.878a 2.559b 1.819c 8.498a 2.734a 1.744c

p r 0.001. p r0.05. pr 0.10.

having a strong tendency to accumulate and redeem points should on average tend to change their expenditure to a greater extent than other customers. The model’s goodness-of-fit is acceptable in terms of the cut-offs recommended in the literature. The estimated chi-square value is 8.291 with 4 degrees of freedom and an estimated p-value of 0.0815. The estimated values of the CFI and TLI indices are 0.96 and 0.93 are higher than or close to the minimum required cut-off ( Z0.95). The estimated RMSEA index of 0.05 is lower than the maximum cut-off of 0.08 suggested by Brown and Cudeck (1993). Only the estimated SRMR index of 0.09 is higher than the recommended maximum cut-off (r0,05) (Hu and Bentler, 1999). The results of the model are shown in Table 3. The mean latent trajectories of each group are described by their mean latent intercept and by their mean latent slope. The estimated means slopes for each of the groups (mgb) vary from one group to the next. In group 1, made up of the 1022 individuals who undergo no learning, the mean tendency to change the initial level of expenditure is negative. It is estimated to be h  9.299, a statistically significant value (p o0.001). This indicates that on average the people in this group in period ‘‘t’’ reduce their planned expenditure by h (t  9.299). In 2005, for example, (i.e. t ¼3) on average they reduced their planned expenditure in the stationery store by h  27.897 (3  h 9.299). The total planned expenditure during this period thus rises to h 16,562 (h 44.459–h 27.897). In group 2, made up of the 233 individuals whose learning was exclusively based on accumulating points, the mean tendency to change their initial level of expenditure has an estimated value of h 1.081. This value is not significantly different from zero. In group 3,

composed of the 75 individuals whose learning was based solely on redeeming points, the estimated mean tendency to change the initial level of expenditure is h 5.037. This value is significant only at the threshold p o0.10. In group 4, where the individuals underwent learning based both on points and their redemption, the estimated mean tendency to change the initial level of expenditure is significant and has a value of h 20.352 (p o0.001). The differences between these mean tendencies are significant, as shown by a specific chi-square difference test of the robust estimator MLMV (Muthe´n and Muthe´n, 2001). The model M2, freely estimating the means, was compared to the nested M2 model constraining the means to equality across the four groups. The estimated chi-square difference has a value of 53.306 with 3 degrees of freedom and a p-value lower than 0.0000. It is therefore unlikely that the model with equal means is correct. The Fig. 5, shows distribution of tendencies in each group. In group 1, where the individuals have not undergone any learning, only 5% (50 out of 1022) have a mean tendency to change the initial positive level of expenditure. In group 2, comprising individuals who learn in order to gain points, 100% of them have a mean tendency to change the initial positive level of expenditure. In group 3, made up of individuals who learn in order to redeem points, 51% (38 out of 75) have a mean tendency to change the initial positive level of expenditure. Finally 82% (41 out of 50) of the individuals in group 4, who learn while accumulating points and transforming them into rewards, have a mean tendency to change the initial positive level of expenditure. A difference test of these proportions confirms that they are statistically significant at the threshold po0.01.

Author's personal copy J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

Table 3 Average latent trajectories. Estimations of latent intercepts and slopes of each group. Group 1 n¼ 1022 (h)

Group 2 n¼ 233 (h)

Group 3 n¼ 75 (h)

Group 4 n¼50 (h)

Mean intercept (mga) Estimation (est.) Robust standard error (r.s.e) Robust est./r.s.e Bootstrap standard error (b.s.e) Bootstrap est./b.s.e

44.459 2.546 17.464a 3.088 14.398a

33.705 4.647 7.254a 5.249 6.421a

65.579 7.306 8.976a 7.499 8.745a

65.647 16.443 3.992a 15.432 4.254a

Mean slope (mgb) Estimation (est.) Robust standard error (r.s.e) Robust est./r.s.e Bootstrap standard error (b.s.e) Bootstrap est./b.s.e

 9.299 0.718  12.959a 0.846  10.990a

1.081 1.593 0.679c 1.721 0.628c

5.037 3.011 1.673b 2.956 1.704b

20.352 5.764 3.531a 5.807 3.505a

Group 1: no learning, either by points or by rewards. Group 2: learning only by points. Group 3: learning only by rewards. Group 4: full learning, by points and by direct rewards. a b c

p r 0.001. p r0.10. Estimation not significantly different from zero.

6. Conclusion To conclude our study, we successively discuss its main contributions, emphasize the major managerial lessons to be drawn from it, and point out its limitations. 6.1. General discussion The aim of this paper was to show that the effectiveness of loyalty programs does not depend solely on their design, but on how they are learned. This first contribution of our study has been empirically confirmed over a lengthy time period. Our choice of methodology conforms to the long-term orientation of loyalty programs, which encourage their customers to maximize their utility over a long time period rather than in the short term (Kopalle and Neslin, 2003). The impact of rewards on buying behavior is demonstrated both during the initial ‘‘running in’’ period of the first year and over the four years during which we observed the cohort. A second contribution of our study is that we take account of real, actually observed behaviors, which always differ from stated behaviors. Our behaviorist approach thus extends and enriches cognitivist approaches to the effectiveness of loyalty programs, which measure the effects of such programs on loyalty intention or on attitudinal loyalty (Yi and Jeon, 2003; Keh and Lee, 2006). A third contribution lies in revealing ‘‘learning profiles’’ among the users. Points accumulation and redemption behaviors differ from one individual to another, and allow the users to be segmented. In the segmentation proposed by Allaway et al. (2006), the users are distinguished according to their buying behavior (e.g. number of purchase, average interval between purchases, total expenditure). The segment with the loyalest customers is also the smallest segment in terms of size. In making our segmentation according to learning profiles we came up with the same observation. However, the advantage of our segmentation is being able to explain why the customers in the smallest group are loyal. They are so because they know how to accumulate points and use them wisely. Since they know how to use the program more effectively than other customers, they benefit from it more and became the stationery store’s best customer. 6.2. Implications for managers A number of lessons for managers may be drawn from this study. The loyalty program on which we tested our hypotheses is a

89

multi-partner program. The distinctive feature of this type of program is that it brings customers to each participating business from other participating businesses. In this study the points accumulated are those gained by the user in one of the 33 participating stores. The tangible rewards (e.g. reductions) and the purchases made concern only the stationery store. The findings show that the points gained by the customer through his expenditure on, say, food or clothing, favor the growth of his expenditure in the stationery store. They suggest that managers benefit from using multi-partner programs for their customers that bring in complementary stores. The greater the synergy among the stores participating in the program, the greater the number of opportunities for using the program and the more each business will directly or indirectly profit from it. Moreover this study shows that the effectiveness of a program is linked to the type of learning about the program by the user. Learning sustained solely by points accumulated or solely by reductions can only stabilize medium- to long-term expenditure. On the other hand, learning sustained simultaneously by points and by reductions leads to a marked growth in average expenditure. This important finding suggests that businesses should encourage users to fully profit from the points gained. Check-out receipts, for example, should inform customers of their current points level and of the reductions they can obtain. The management of points would be facilitated by making available an account accessible on the Internet, which would enable customers to keep track of their points and to manage their points account efficiently. In addition we see that only 26% (362 out of 1380) of the program’s users increase or maintain their level of expenditure over four years. These findings have been previously noted by other authors (Mauri, 2003; Allaway et al., 2006). The fact that loyalty cards and their wide distribution are free of charge can partly explain this low rate. Since the card is obtained without any effort, users are little involved in the program and their commitment is weak or even non-existent. A free card and a few points gained are insufficient to offset switching costs to another program (Wirtz et al., 2007). On the other hand, a paid-for loyalty card with an immediate points credit without making a purchase could create a minimum level of commitment that would induce customers to persevere in their learning. One final managerial lesson should be drawn. The effectiveness of a loyalty program should not be viewed as something that is only surveyed periodically. It is a permanent process, involving an effort to monitor and check that users are making progress in learning about the program. The sophisticated CRM programs that currently support loyalty programs provide a business with all the information needed to carry out this monitoring and checking and to point customers toward the most effective forms of learning.

6.3. Limitations and further research Our study naturally has a certain number of limitations that call for future research. These concern the methodological as well as the theoretical aspects of the research. By avoiding comparative methods, we reduced the self-selection bias. Nevertheless the longitudinal methods based on the observation of a cohort of users is not without weaknesses. The lack of purchases on the part of a customers is not always a sign of disloyalty. Indeed a customer may remain loyal to the program and to the store, but may have lost or forgotten his loyalty card. Purchases made without the card by program members are neither rewarded nor recorded in the data base. The data used is therefore tainted with errors and the latent slopes probably have underestimated means. Another methodological limitation stems from the relatively limited size of our sample (n ¼1381). Certain segments therefore

Author's personal copy J. Frisou, He´. Yildiz / Journal of Retailing and Consumer Services 18 (2011) 81–91

90

Groupe 1: no learning, neither by the points nor by direct rewards

890

Percentage of observations

88% 78% 68% 59% 49%

Groupe 3 : learning only by direct rewards

1000

53%

900

47%

800

β 0 tendency towards loyalty behavior 5 % of individuals

700 600 500

39%

400

29%

300

20%

200

10% 0%

8

14

55

50

4

1

Percentage of observations

98%

40 34

40% 33% 27% 20%

30

7%

0

0%

233; 100%

β 0 tendency towards loyalty behavior 100 % of individuals

β = or > 0 tendency towards loyalty behavior 38 individuals 10

13%

100

3

25 20 15 10

222

2

5 0

-60 -40 -20 0 20 40 60 Histogram of β's individual slopes ( n = 75) 260 240 220 200 180 160 140 120 100 80 60 40 20 0

Groupe 4 : complete learning, by the points and by direct rewards

Percentage of observations

Percentage of observations

Groupe 2 : learning only by the points

24

β