The Impact of Successive Short-Term Reward Programs on Turnover in Retailing:
an Empirical Research Study using Longitudinal Data
Klaas van der Veen
2
The Impact of Successive Short-Term Reward Programs on Turnover in Retailing:
an Empirical Research Study using Longitudinal Data
Master Thesis
University of Groningen
Faculty of Economics and Business Msc Business Administration Department of Marketing
Marketing Management and Marketing Research January 31st, 2012
Name: Klaas van der Veen
Student number: 1620096
Address: Schaepmanlaan 59
9722 NR Groningen
E-mail: klaas_vanderveen@hotmail.com
Telephone: +31 (0)6 110 532 84
First Supervisor: Prof. Dr. Tammo H.A. Bijmolt
Second Supervisor: Jacob H. Wiebenga MSc
Supervisors BrandLoyalty International B.V.: Ms. Renée Wessels
3
Management Summary
Reward programs have become a popular instrument to increase customer spending and retention. Despite the popularity of these programs, studies providing evidence for the turnover impact of reward programs are limited and were based on single programs. This empirical research investigates the impact of multiple successive short-term reward programs by analyzing weekly turnover data of a supermarket chain in Luxembourg over a four-year period. In specific, this research studies whether customers increase their spending in order to obtain the reward (points pressure effect) and after they received the reward (rewarded behavior effect) (Taylor and Neslin 2005). Furthermore, it studies the moderating effect of program design characteristics on reward program effectiveness.
A multiplicative model was used to test the hypotheses. Points pressure and rewarded behaviour periods were estimated to be 8 and 4 weeks, respectively. Pooling tests showed significant differences between store formats. Therefore, the largest store format (Hypers) was chosen as base case to which the smaller ones (Supers and Marches) were compared. Several model validation tests support the robustness of the empirical findings.
The results show that there is a significant points pressure effect on turnover. On average, a reward program increases customer spending with 8.8% before redemption. However, this effect does not persist after the reward program has ended, indicating that customers want to attain the reward but do not remain loyal. Furthermore, the effectiveness of reward programs in increasing turnover is most evident among the largest stores. There are several moderating effects on the effectiveness of reward programs. Customer spending is higher when saved stamps/points are only valid for a specific reward program, rather than having an unlimited validity. Moreover, the amount of money customers need to spend to be able to redeem a reward influences the success of reward program. Reward programs are more effective in increasing turnover when the investment (e.g. advertising) in the program is higher and the number of weeks between successive programs increases.
4
Preface
This empirical research concludes the final part of my MSc degree in Business Administration with the specializations Marketing Management and Marketing Research. This research assesses the impact of successive short-term reward programs on turnover in retailing, which corresponds perfectly with my own interests and multiple marketing courses I followed through the last couple of years.
First of all, I would like to thank my supervisors Mr. Tammo H.A. Bijmolt and Mr. Jacob H. Wiebenga for their guidance, critical view and positive feedback during the whole process. Moreover, I would like to thank BrandLoyalty International and especially Ms. Renée Wessels and Mr. Hans Hoppenbrouwers for their cooperation and hospitality at the office in `s Hertogenbosch. I’m very grateful for their feedback and provision of the data, which was of critical importance for conducting this research. Furthermore, I would like to thank Aliza van der Veen for the time she invested in reading this thesis. Her experience in writing a thesis, as well as her positive feedback, proved to be very helpful. Finally I would like to thank my family and friends for their unconditional support and interest in my thesis.
Klaas van der Veen
5 Table of Contents MANAGEMENT SUMMARY ... 3 PREFACE ... 4 1. INTRODUCTION ... 6 1.1OBJECTIVE ... 6 1.2CONTRIBUTION ... 7 1.3STRUCTURE ... 8
2. CONCEPTUAL FRAMEWORK AND HYPOTHESES ... 9
2.1THE LOYALTY CONCEPT ... 9
2.2POINTS PRESSURE AND REWARDED BEHAVIOR EFFECTS ... 9
2.2.1 Points Pressure Effect ... 10
2.2.2 Rewarded Behavior Effect ... 11
2.3MODERATING EFFECT OF PROGRAM TYPE ... 11
2.4MODERATING EFFECT OF PROGRAM DURATION ... 12
2.5MODERATING EFFECT OF SPEND REQUIREMENT ... 13
2.6MODERATING EFFECT OF PROGRAM INVESTMENT ... 13
2.7MODERATING EFFECT OF TIME BETWEEN PROGRAMS ... 14
3. RESEARCH DESIGN ... 16 3.1STUDY SETTING... 16 3.2SAMPLE DESCRIPTION ... 17 3.3RESEARCH METHOD ... 18 3.4POOLING ... 18 3.5MODEL SPECIFICATION ... 19
3.6POINTS PRESSURE AND REWARDED BEHAVIOR PERIOD ... 21
3.7ERROR TERM ASSUMPTIONS ... 21
4. RESULTS ... 23
4.1ANALYSIS OF MAIN AND MODERATING EFFECTS ... 23
4.2INTERACTION EFFECTS OF STORE FORMATS ... 26
4.3PREDICTIVE VALIDITY ... 27
5. CONCLUSIONS ... 30
5.1LIMITATIONS AND SUGGESTIONS FOR FURTHER RESEARCH ... 32
5.2MANAGERIAL IMPLICATIONS ... 32
6. REFERENCES ... 34
APPENDICES ... 37
APPENDIX 1:REGRESSION ANALYSES OF MAIN AND MODERATING EFFECTS ... 37
APPENDIX 2:VISUAL REPRESENTATION OF MODERATING EFFECTS ... 43
APPENDIX 3:VISUAL REPRESENTATION OF STORE FORMAT INTERACTION EFFECTS ... 45
APPENDIX 4:PREDICTIVE VALIDITY MEASURES ... 48
Appendix 4.1: Mean Absolute Percentage Error (MAPE) ... 48
6
1. Introduction
In the last couple of years, loyalty programs have become omnipresent in the marketplace. In 2008, membership rates in the US reached 1.8 billion, of which 153.2 million in the grocery sector (Colloquy 2009). Moreover, McKinsey research found that about half of the ten largest US retailers have launched loyalty programs, and similar rates have been found among UK and Dutch retailers (Cigliano et al. 2000; Leenheer and Bijmolt 2003). Loyalty programs, often called reward programs (Yi and Jeon 2003) or frequency reward programs (Kopalle and Neslin 2003), usually provide rewards based on cumulative purchases and attempt to increase customer retention and product usage. Furthermore, they serve to attract new customers by promising additional benefits (Liu and Brock 2008). Customers’ motivation to participate in these programs mainly stems from their appreciation to get something for nothing (Uncles 1994). Loyal customers are considered an asset to the business, since relative costs of customer retention are considerably lower than those of acquisition (Fornell and Wernerfelt 1987). Moreover, customer loyalty is considered an important key to organizational success and profit (Oliver 1997).
Previous studies have either focused on long-term (continuous) reward programs (e.g. Kopalle et al. 2007), such as frequent-flyer programs, or on short-term reward programs (e.g. Taylor and Neslin 2005), where customers need to save points or stamps before an expiration date in order to obtain the reward. Though empirical research has found evidence for the turnover impact of short-term reward programs (Lal and Bell 2003; Taylor and Neslin 2005), these effects were based on a single program. This research is unique in that it investigates multiple successive short-term reward programs over a four-year period. Furthermore, this is the first research studying the effect of transition from a stamps-based program (where stamps are only valid for the specific program) to a card-based program (where loyalty points have unlimited validity).
1.1 Objective
7
programs, which represent the increase in customers’ expenditures in order to obtain the reward and the increase in customers’ expenditures after they received the reward, respectively (Taylor and Neslin 2005). In addition, this study assesses the role of moderating variables influencing the point pressure and rewarded behavior effects, including program type, spend requirement, and program investment. Figure 1 shows the main and moderating effects in a simplified conceptual model. The relations between the variables will be further discussed in Chapter 2. In order to address the objectives of this research, a dataset is used of a supermarket chain located in Luxembourg consisting of turnover and specific reward program data.
1.2 Contribution
This research contributes to the understanding of reward programs in several ways. First, it provides additional insights into the effectiveness of reward programs by examining their impact on turnover in the short-term as well as in the long-term. Second, this research studies the effectiveness of successive short-term reward programs over a longer period in time instead of a single short-term (or long-term) program over a short (or longer) period. These results can be used to assess whether reward programs are able to build loyalty in the long-run. Third, this research examines how the design of reward programs affects its effectiveness.
8
1.3 Structure
9
2. Conceptual Framework and Hypotheses
2.1 The Loyalty Concept
Customer loyalty can be divided into attitudinal and behavioral loyalty. Both a favourable attitude towards the store and repeated store patronage are required for loyalty (Dick and Basu 1994). Although both constructs are important, this empirical research focuses on behavioral loyalty and uses turnover as a proxy. Reward programs rewarding customers based on their purchasing behavior attempt to enhance retention by providing incentives for customers to purchase more frequently and/or in larger volumes. Previous studies have either focused on long-term (continuous) reward programs (e.g. Kopalle et al. 2007) or on short-term reward programs (e.g. Taylor and Neslin 2005). However, existing research has always based their findings on a single program rather than multiple, successive programs. Moreover, research on the effectiveness of these programs is limited and contradictory. Lal and Bell (2003) found that, as a whole, reward programs in grocery retailing are profitable, despite an unprofitable segment of high spending customers. Similar results were found by Taylor and Neslin (2005). However, Sharp and Sharp (1997) and Meyer-Waarden and Benavent (2006) found that reward programs had no impact on repeat purchases. These results indicate that there is still no consensus regarding whether a reward program is effective in enhancing profitability or not.
2.2 Points Pressure and Rewarded Behavior Effects
This empirical research aims to improve the understanding of the impact of reward programs on turnover. An effective reward program increases turnover in both the points pressure and rewarded behavior period. These direct effects may be moderated by various program design characteristics, as well as the speed with which reward programs succeed each other.
10
FIGURE 2–POINTS PRESSURE AND REWARDED BEHAVIOR EFFECTS (TAYLOR AND NESLIN 2005)
2.2.1 Points Pressure Effect
11
H1: Each consecutive reward program has a positive impact on turnover for the last weeks of the program
2.2.2 Rewarded Behavior Effect
The rewarded behavior effect represents the long-term impact, by which customers increase their purchase rate after they have received the reward. This effect can be generated either by behavioral or cognitive processes (Taylor and Neslin 2005). Research has found rewarded behavior effects on purchase behavior in retailing (Taylor and Neslin 2005, Lal and Bell 2003) and in aviation (Kopalle et al. 2007). On the other hand, Kivetz et al. (2006) find that the positive change in customer behavior vanishes after they have received the reward.
Learning theories posit that when purchase behavior is rewarded, and this reward is pleasing (e.g. meets customer’s needs), then the rewarded behavior is more likely to persist (Rothschild and Gaidis 1981). The customer learns that his or her purchase behavior is compensated and therefore will repeat this behavior. According to the cognitive perspective, a reward is able to increase subsequent purchase behavior if the rewarded customer develops positive feelings toward the store which translate into higher patronage (Taylor and Neslin 2005). Therefore, the hypothesis is as follows:
H2: Each consecutive reward program has a positive impact on turnover for the first weeks after the program
2.3 Moderating Effect of Program Type
12
stamps compared to points on a card. Since stamps are tangible, customers are reminded to their progress in the program every time they receive stamps at the store. Consequently, a higher awareness of their progress, due to increased visibility, is likely to increase customer spending.
The first reward programs in the dataset are stamp-based and are followed up by a program which is card-based. Hence, it seems plausible that customers, who are used to spending all their stamps in one program, will save up loyalty points in a card-based program (where points are valid over consecutive programs) because they are more selective in reward redemption. Therefore, the hypotheses are as follows:
H3: Card-based programs, in contrast to stamp-based programs, will decrease the positive points pressure effect on turnover.
H4: Card-based programs, in contrast to stamp-based programs, will decrease the positive rewarded behavior effect on turnover.
2.4 Moderating Effect of Program Duration
Although the reward programs in this study only last for 14 to 19 weeks, the duration of these programs may have a moderating impact on the points pressure and rewarded behavior effects. When the duration of the program increases, customers are likely to have collected sufficient stamps/points before the final weeks of the program, and thus reducing the points pressure effect. However, the longer a reward program lasts, the more time (low spending) customers have to collect the threshold level of stamps or loyalty points. Therefore, the likelihood of attaining the reward increases which should positively affect the participation of customers in the program and hence the effectiveness of the program in increasing turnover. The importance of engaging low spending customers in the reward program was shown by Lal and Bell (2003), who found the most significant behavior changes among low and moderate buyers (cherry pickers), in contrast to minor changes among heavy spenders. Since the positive effect of program duration seems more plausible, the hypotheses are as follows:
13
2.5 Moderating Effect of Spend Requirement
Since customers participating in a reward program experience switching costs, the retailer is able to retain these customers over a longer period of time, also called the customer lock-in effect. This effect is likely to be influenced by the divisibility of the reward, which is the number of discrete reward-redemption opportunities within a program (Nunes and Drèze 2006). The authors argue that there is a trade-off, since customers prefer a highly divisible program (providing various redemption opportunities and thus reducing loyalty points waste) and retailers prefer lower divisibility of rewards (which is more effective in creating a customer lock-in effect). In case the points threshold for obtaining a free reward is too high, customers will consider it as unobtainable and thus will consider it as irrelevant (O’Brien and Jones 1995). However, a program with a low spend requirement may not be challenging enough for customers and therefore will not participate in the program (Kivetz and Simonson 2003; Nunes and Drèze 2006). This suggests an inverted U-shape effect where a spend requirement which is either too low or too high will have no (or a negative) impact on the effectiveness of the reward program. Section 3.2 will show that the spend requirement of the programs studied in this research are relatively low. Therefore, the hypotheses are as follows:
H7: Spend requirement will increase the positive points pressure effect on turnover. H8: Spend requirement will increase the positive rewarded behavior effect on turnover.
2.6 Moderating Effect of Program Investment
Promotions are an important element for the success of reward programs. Mastroberte (2005) argued that even some very good reward programs may fail due to lack of proper promotion. When promotional themes are developed which focus on information about the value-added features of a particular credit card, it may induce customers to select and use it (Goyal 2004). Furthermore, Liu and Brock (2008) found that reward point redemption rates in the Chinese credit card industry can be increased by communications that raise program awareness.Therefore, the hypotheses are as follows:
14
2.7 Moderating Effect of Time Between Programs
Two goals which can be served by reward programs are increasing share of wallet and increasing demand (Nunes and Drèze 2006). From a turnover promotions point of view, it is known that customers often accelerate their purchases in time and/or quantity due to these promotions, which should lead to a dip in purchases in the weeks following a promotion (Van Heerde et al. 2000). The authors find that the post-promotion dip lies between 4 and 25 percent of the current turnover effect. Although these effects are resulting from price promotions of specific products, it can be expected that turnover will decrease after a reward program has ended, for example due to stockpiling. In fact, when a brand increases the use of price promotions, baseline turnover of this brand will decrease (Foekens et al. 1999; Kopalle et al. 1999) and the height of the deal spike will be reduced (Blattberg et al. 1995). Hence, more time between two programs will increase the effectiveness of the succeeding program in both the points pressure and rewarded behavior period. Therefore, the hypotheses are as follows:
H11: Time between two successive programs will increase the positive points pressure effect on turnover.
H12: Time between two successive programs will increase the positive rewarded behavior effect on turnover.
15
16
3. Research Design
3.1 Study Setting
Data is used from a supermarket chain in Luxembourg, consisting of 18 stores. The stores are divided into three categories: Hypers (2), Supers (10), and Marches (6). The distinction between these formats is mainly based on location, as well as assortment size and floor surface. All stores offer an average of two short-term reward programs a year to its customers, which are similar for all stores. At the beginning, customers collected stamps for every ten euro’s spent, which could only be redeemed for the specific program and after a certain threshold had been reached (the spend requirement). Typically, these stamp-based programs included one or two clean-up weeks, in which no stamps were issued but could still be redeemed. Every reward program had a different theme, varying from kitchen utensils to gardening tools. Moreover, within a program customers could choose from multiple rewards, where each reward had a different spend requirement. After several stamp-based programs, the chain shifted for all store types to a card-based program. In this new format, customers collected credits for every euro spent. In addition, customers were not restricted to redeem their points in a specific reward program. This gave them the ability to be more selective among different reward programs. Since loyalty points were not program specific, the clean-up weeks were dissolved.
17
TABLE 1-OVERVIEW OF THE RELEVANT VARIABLES
Variables Measurement Description Hypothesis
Dependent: Turnover Independent:
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Moderating:
Program Type (PT)
Program Duration (PD) Spend Requirement (SR)
Program Investment (PI)
Time Between Programs (TBP) Control: Store formats € Weeks Weeks a. Stamps b. Loyalty card Weeks € € Weeks a. Hypers b. Supers c. Marches
Average weekly turnover per store format
Turnover increasing mechanisms
within a reward program
Mechanism through which points are collected
Duration of the reward program Amount of money spent needed to obtain a reward
Total investment in communication per reward program
Period between two successive reward programs
Store formats differ in their size, location and assortment
H1 H2 H3 & H4 H5 & H6 H7 & H8 H9 & H10 H11 & H12 3.2 Sample Description
18
consideration, customers need to spend between €21.39 and €46.75 per week to be able to redeem the reward of the particular program. Furthermore, to enhance the interpretability of parameter estimates in the regressions, spend requirement and program investment were re-scaled by dividing them by 1,000 and 1,000,000 respectively, although Table 3 still contains the original values.
TABLE 2–DESCRIPTIVE STATISTICS (AVERAGE WEEKLY TURNOVER)
Store Format Min Max Mean SD
Hypers (N=211) Supers (N=211) Marches (N=211) €2,082,342 €531,277 €92,940 €5,279,560 €1,058,683 €178,574 €2,809,503 €687,235 €136,628 €453,313 €71,087 €13,112
TABLE 3–OVERVIEW OF THE MODERATING VARIABLES
Reward Program Program Type Program Duration (weeks) Spend Requirement (€) Program Investment (€) Time Between Programs (weeks) 1 2 3 4 5 6 7 Stamp Stamp Stamp Stamp Stamp Stamp Card 14 18 15 19 15 18 15 500 539.98 338.77 888.30 503.59 384.95 740.40 2,381,665 3,860,451 2,227,014 4,263,692 NA NA NA 38 38 6 13 5 14 8 3.3 Research Method
In order to test the hypotheses and to explain the effect of reward programs on turnover, a multiplicative model was used. Using such a model is necessary because it can be assumed that (1) the independent variables have no constant returns to scale and that (2) interactions exist between these independent variables. Another advantage of using a multiplicative model is that it has a simple economic interpretation, since the parameters are constant elasticities (Leeflang et al. 2000). By taking the logarithm of the multiplicative model it becomes linear and can be analyzed by using multiple linear regression in SPSS.
3.4 Pooling
19
pooling test was conducted to test whether the store formats could be combined for parameter estimation. Table 4 provides the F-values and corresponding critical values for the estimated models. Results show that for all the models pooling is not allowed indicating that separate models have to be estimated for the store formats, which would harm the comprehensibility of this research. To overcome this problem and simultaneously accounting for differences between store formats, they were included in the models as dummies. More specifically, the Hyper store format was chosen as base case to which the Supers and Marches were compared.
TABLE 4–POOLING RESULTS*
Model F value Critical value
Main Effects Program Type Program Duration Spend Requirement Program Investment Time Between Programs
7.530 7.192 7.237 6.852 6.478 7.192 1.517 1.499 1.482 1.482 1.485 1.482
* The OLSDV (Ordinary Least Squares with Dummy Variables) pooling method was chosen. This method assumes that the slope parameters are fixed and common for all store formats, but the intercepts are unique for each store format (Leeflang et al. 2000).
3.5 Model Specification
In this section the multiplicative model and the additive model predicting the turnover of the different store formats are presented, which are specified as follows:
Taking the logarithm of the multiplicative model above results in the following additive model, which is analyzed by using multiple linear regression:
20
Skt denotes the turnover of store format k in week t and may depend on the type of store format Fk, a
vector indicating the period within or after the reward program (Plkt), program design characteristics
(Mlkt) and time-varying indicators (Qlt). Specifically, the vectors contain the following explanatory
variables:
2 #222 , 342(
+ #21 , 25, 3, 26, 142(
7 8 7$,&/&2CA,'DE;F , 6=GH,/>?=2006, 7$,&/&3, 7$,&/&4K, 6=GH,/>?=2007, ;,'/& M, <'-='>?=, 6=GH,/>?=2008, 2=/-?'/ , 6=GH,/>?=2009@, $AA&B ,P
where:
222 = an indicator variable for the points pressure period: 1 for the final 8 weeks of the program for store format k, week t, 0 otherwise;
342 = an indicator variable for the rewarded behavior period: 1 for the first 4 weeks after the program for store format k, week t, 0 otherwise;
21 = an indicator variable for program type: 1 if the program uses a loyalty card for store format k,
week t, 0 if the program uses stamps;
25 = an indicator variable for the duration of the program: the length of the program in weeks if
PPP1kt and RBP2kt are 1, store format k, week t, 0 otherwise;
3 = an indicator variable for the spend requirement of the program: the amount of euro’s if PPP1kt
and RBP2kt are 1, store format k, week t, 0 otherwise;
26 = an indicator variable for the program investment of the program: the amount of euro’s if
PPP1kt and RBP2kt are 1, store format k, week t, 0 otherwise;
142 = an indicator variable for the time between two successive programs: the number of weeks between the current and previous program if PPP1kt and RBP2kt are 1, store format k, week t, 0
otherwise;
7 = an indicator variable for time-varying effects: 1 if the observation is in week t, 0 otherwise;
= the store format intercept multiplier if k ≠ 0, base case if k = 0;
21
= the multiplier for the program design characteristics for store format k, week t; = the seasonal multiplier for week t;
= disturbance term for store k, week t.
3.6 Points Pressure and Rewarded Behavior Period
The points pressure effect is likely to increase when customers approach the end of the reward program (Kivetz et al. 2006). Therefore, it can be assumed that the points pressure effect is only present in the final weeks of the program instead of the during the whole program period. To empirically assess the length of this period, as well as the rewarded behavior period, weekly indicators were examined (by using a dummy for every single week) for up to 14 weeks before and 6 weeks after the end of the reward program. Results demonstrated a points pressure and rewarded behavior period of respectively 8 and 4 weeks.
3.7 Error Term Assumptions
After the separate regressions were estimated, the estimates of the error term were investigated. The assumption of homoscedasticity (similar variance for all possible values of a predictor variable) was not validated because heteroscedasticity especially occurs in case cross-sectional data is used for estimation (Leeflang et al. 2000). Therefore it is assumed that the error term is homoscedastic.
To test whether the error terms are not independent but follow a first-order autoregressive process, a Durbin-Watson (D.W.) test was conducted based on the variance of the difference between two successive disturbances (Leeflang et al. 2000). The D.W. statistic varies between 0 and 4 and autocorrelation does not exist at a value of 2. The results in Table 5 show that all models experience significant positive autocorrelation (D.W. statistics are below the lower bounds). However, after accommodating for autocorrelation by using GLS (Generalized Least Squares), significance levels of parameter estimates worsened. This was mainly due to the dummy coding of the variables. Assuming that the model has the correct functional form, it was decided to allow for positive autocorrelation in order to preserve the relevance of the models.
22
TABLE 5–AUTOCORRELATION RESULTS
Model Durbin Watson Lower Bound* Upper Bound*
Main Effects Program Type Program Duration Spend Requirement Program Investment Time Between Programs
1.545 1.525 1.565 1.536 1.707 1.678 1.789 1.793 1.785 1.785 1.767 1.785 1.932 1.928 1.936 1.936 1.937 1.936 * At 5% significance level
TABLE 6–NONNORMALITY RESULTS
Model Kolmogorov-Smirnov Z P-value
Main Effects Program Type Program Duration Spend Requirement Program Investment Time Between Programs
23
4. Results
4.1 Analysis of Main and Moderating Effects
The results described in this section are based on the largest store format, the Hypers. Section 4.2 will elaborate on the differential effects between stores. Table 7 shows the estimated main effects of points pressure and rewarded behavior on turnover. It also shows the moderating effects influencing these main effects. The main and moderating effects have been estimated in separate regressions in order to minimize multicollinearity. The elaborated models can be found in Appendix 1, Table 1-6. The points pressure effect, the effect of the last 8 weeks of the reward program on turnover, has a significant positive impact (β=.085, p=.000), supporting H1. On average, a reward program increases turnover with 8.8% during the final weeks of the program. However, no significant effect was found for the rewarded behavior period (β=.013, p=.642), which rejects H2. These results indicate that customers want to attain the reward but do not remain loyal. However, it is important to notice that the increase in turnover during the program is not at the expense of a decrease in turnover after the program.
TABLE 7–MAIN EFFECTS AND MODERATING EFFECTS
βa SE Exp (β)b P-value
Main effects
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Moderating effects Program Type (PT) * PPP Program Duration (PD) * PPP Program Duration (PD) * RBP Spend Requirement (SR) * PPP Spend Requirement (SR) * RBP Program Investment (PI) * PPP Program Investment (PI) * RBP Time Between Programs (TBP) * PPP Time Between Programs (TBP) * RBP
.085*** .013 -.115*** .009** -.014* .087* -.169* .042*** -.044** .004*** .004** .016 .028 .038 .005 .007 .049 .090 .011 .021 .001 .002 1.088 1.013 0.891 1.009 0.986 1.089 0.841 1.043 0.957 1.004 1.004 .000 .642 .003 .047 .060 .077 .060 .000 .041 .000 .012 a *** p<0.01; ** p<0.05; * p <0.1
bIncludes a bias correction: Exp(β)*Exp(-0.5*Var(β))
24
the data included only one card-based program. It is likely that other factors play a role in the highly negative effect of card-based programs. The moderating effect of program type during the rewarded behavior period could not be estimated due to the lack of turnover data. Therefore H4 remains inconclusive.
Customers’ spending increases in the final weeks of the reward program when the duration of the program as a whole increases (β=.009, p=.047), supporting H5. This effect is reverse for the rewarded behavior period, where spending decreases when program duration increases (β=-.014, p=.060), which contradicts H6.However, the Variance Inflation Factors (VIF) for this model are well above the cutoff threshold value of 10 (Hair et al. 2006), which indicates multicollinearity (Appendix 1, Table 3). Therefore, the estimates must be considered with caution.
As the spend requirement of a reward program increases, customers’ spending during the points pressure period increases (β=.087, p=.077), supporting H7. This result confirms that the spend requirement levels are rather low for most of the programs. In contrast, a higher spend requirement has a negative effect on spending during the rewarded behavior period (β=-.169, p=.060), which contradicts H8. These outcomes indicate, for example, that increasing the spend requirement from €250 to €500 will increase the points pressure effect with 2.15%, whereas it reduces the rewarded behavior effect with 4.24%. These effects are shown in Figure 4 for different spend requirement levels. The negative effect of spend requirement on turnover after redemption is probably due to increased customer spending before redemption.
25
A higher investment level makes the reward program more successful with respect to the points pressure period (β=.042, p=.000) with an impact of 4.27% for every additional €1,000,000 invested, which supports H9. The opposite effect is true for the rewarded behavior period (β=-.044, p=.041), indicating a decrease in turnover of 4.3% with a similar expansion of investment level, rejecting H10
As predicted, reward programs are more successful when the time between successive programs increases. This applies for both the final weeks of the program (β=.004, p=.000) and the first weeks after the program (β=.004, p=.012), increasing turnover with 4.07% and 4.09% respectively in case the time between programs doubles from 10 to 20 weeks. These results support both H11 and H12. Figure 5 shows the positive effect of increasing the number of weeks between reward programs on turnover. A visual representation of the moderating effects of program duration and program investment can be found in Appendix 2.
26
4.2 Interaction Effects of Store Formats
The main and moderating effects described in the previous section were based on the largest store format. However, it seems plausible that these effects differ between store formats. Therefore, this research also accounts for differential effects between these formats.
Although customers increase their spending considerably during the final weeks of a reward program at Hyper stores (β=.085, p=.000), this effect is significantly lower at Supers (β=-.044, p=.035) and Marches (β=-.071, p=.001), where the points pressure effect is 4.12% and 1.29% on turnover, respectively. No rewarded behavior effect was found at Hypers (β=.013, p=.642) and this effect did not significantly differ at Supers (β=.003, p=.940) or Marches (β=.021, p=.583). These effects are shown in Figure 6. To enhance the interpretability of the interaction effects, the graphs for Supers and Marches also include the points pressure and rewarded behavior effects of the Hypers.
27
Customers shopping at Supers (β=.077, p=.082) and Marches (β=.123, p=.006) did not respond that negatively to a card program in the points pressure period compared to Hyper customers (β=-.115, p=.003). Although spend requirement had a significant positive impact on customers’ spending at Hypers (β=.087, p=.077) during the final weeks of the reward program, this effect diminishes at Supers (β=-.075, p=.025) and Marches (β=-.124, p=.000). In addition, the negative impact of spend requirement in the rewarded behavior period for Hypers (β=-.169, p=.060) did not significantly differ for Supers (β=.029, p=.610) and Marches (β=.069, p=.228). Higher investment levels to, for example, promote the reward program, have significantly less pre-rewarding impact at customer spending at Supers (β=-.033, p=.000) and Marches (β=-.053, p=.000) than at Hypers (β=.042, p=.000). The negative effect of program investment in the rewarded behavior period at Hypers (β=-.044, p=.041) was not significantly different for Supers (β=.006, p=.668) or Marches (β=-.009, p=.498). Finally, increasing the time between sequential programs is not effective in boosting spending levels during the program at both Supers (β=-.004, p=.000) and Marches (β=-.006, p=.000) when compared to the Hyper format (β=(β=-.004, p=.000). For Marches this is also true during the rewarded behavior period (β=-.004, p=.015) when compared to Hypers (β=.004, p=.012).
The effects of program duration are not discussed due to high multicollinearity, causing unreliable parameter estimates. Figures 1-6 in Appendix 3 give a visual representation of the differential effects described above.
4.3 Predictive Validity
In order to assess the predictive validity of the models, the predictive performance is compared with a holdout sample. This sample is the 6th reward program in the data and comprises 30 weeks (22 program period and rewarded behavior period weeks and 4 weeks before and after this period). This sample was chosen because leaving out the final program in the data would exclude the only card-based reward program, and thus excludes the whole variable.
28
Furthermore, Theil’s U-statistic is computed to test the predictive validity of the models (Appendix 4.2). If this statistic is less than one, the model outperforms the naïve model, which is the difference between the turnover of the next and current period. The last column in Table 8 shows that the models make good predictions apart from program investment, due to the absence of program investment data in the holdout sample. Consequently, only predictions could be made before the points pressure period and after the rewarded behavior period (i.e. when there is no program investment). These predictions (with a simplified model) were not able to outperform the naïve model, resulting in a Theil’s U-statistic greater than one. Figure 7 shows the actual and predicted turnover of all three store formats for the main effects model. The model predicts the extreme turnover values of Hypers and Supers quite accurately, although actual turnover during Christmas and New Year’s Eve is somewhat higher than anticipated. However, the model tends to overpredict the turnover of Marches, especially in the first year and after the third quarter in 2008. This can be explained by the increase of average store turnover during 2005, which levels out during 2006. Moreover, turnover patterns during Christmas (2007 and 2008) and Easter (2007, 2008 and 2009) are not in accordance with previous years, at both Marches and the other store formats. Intuitively, customers tend to switch to the larger store formats (which have for example broader assortments) when their anticipated spending increases.
TABLE 8–PREDICTIVE VALIDITY RESULTS
Model T* T-T* MAPE Theil’s U-statistic
Main Effects Program Type Program Duration Spend Requirement Program Investment Time Between Programs
29
30
5. Conclusions
This empirical research aimed to assess the impact of successive short-term reward programs by analyzing weekly turnover data of a supermarket chain in Luxembourg over a four-year period. Table 9 summarizes the findings of this study. The main findings of this research are that reward programs significantly increase store turnover in the final weeks of the program (the points pressure effect), but are not able to increase customer spending for some time after they have received the reward (the rewarded behavior effect). Although not hypothesized, the points pressure effect is significantly lower at smaller store formats (Supers and Marches) compared to the largest one (Hypers). In specific, turnover at Hypers increases with 8.8% during the final 8 weeks of the reward program, whereas this is 4.5% and 1.9% for Supers and Marches, respectively. Apparently, the number of customers shopping at Hypers and/or their corresponding basket size increases in the final weeks of a reward program. This may be caused by customers switching to Hypers, allowing them to increase their basket size with products which are not offered at the smaller formats. In this way they are better able to reach their threshold level of stamps/points in order to redeem the reward.
This research confirms that customers increase their spending levels during the final weeks of the program (Lal and Bell 2003; Lewis 2004; Taylor and Neslin 2005; Kopalle et al. 2007). Apparently, through a combination of switching costs and future orientation, customers are motivated to increase spending during the reward program and will accelerate their purchases when they near the end of the program (Nunes and Drèze 2006; Kivetz et al. 2006). In contrast to prior studies (Lal and Bell 2003; Taylor and Neslin 2005; Kopalle et al. 2007), which found evidence for post-rewarding effects (although these were weaker than pre-rewarding effects), this research finds no significant rewarded behavior effect. The absence of post-rewarding effects is in line with Kopalle and Neslin (2003), who found that reward programs are fundamentally powerful, multi-period promotions. Specifically, a firm is able to lock in the customer for multiple purchase occasions because he or she experiences switching costs. However, as in most promotions, this effect is non-existent in the post-promotion period.
31
TABLE 9–OVERVIEW OF FINDINGS
Hypothesis Turnovera Findings
Main effects
H1 Positive effect of points pressure period H2 Positive effect of reward redemption period Moderating effects
H3 Reducing effect of program type (points pressure) H4 Reducing effect of program type (rewarded behavior) H5 Enhancing effect of program duration (points pressure) H6 Enhancing effect of program duration (rewarded behavior) H7 Enhancing effect of spend requirement (points pressure) H8 Enhancing effect of spend requirement (rewarded behavior) H9 Enhancing effect of program investment (points pressure) H10 Enhancing effect of program investment (rewarded behavior) H11 Enhancing effect of time between programs (points pressure) H12 Enhancing effect of time between programs (rewarded behavior)
+ N.S. -N.A. N.A. N.A. + -+ -+ + Supported Not supported Supported N.A. N.A. N.A. Supported Mixed support Supported Mixed support Supported Supported a
“+”= significant positive effect; “-”= significant negative effect; “N.S.”= no significant effect; “N.A.”= not applicable
This research also assesses the role of moderating variables influencing the point pressure and rewarded behavior effects. Although not all hypotheses could be confirmed, there were some interesting observations. Stamp-based programs are more successful than card-based programs. This stresses the fact that the presence of switching costs in a reward program is vital for its effectiveness in increasing customer spending. Moreover, customers tend to spend more when the spend requirement increases, which supports H7. A lower spend requirement makes the program probably less challenging. In addition, the attractiveness of the program may suffer when a lower spend requirement relates to less valuable rewards. Furthermore, higher program investment levels and increased time between programs both positively affect customers spending before reward redemption.
32
pressure period an increase in spend requirement has a positive effect on turnover, although the reverse is true for the rewarded behavior period. Third, this research confirms that reward program success depends on its investment level. More interesting is that, despite positive effects of spend requirement and program investment on turnover during the final weeks of a reward program, these effects are negative in the first weeks after the program. A possible explanation for this is that increased purchase acceleration will simultaneously lead to an increased post-promotion dip (Van Heerde et al. 2000). Finally, this research shows that the success of reward program increases when there is more time between successive programs. This is in line with literature on price promotions, which found that a higher frequency of same brand price promotions will reduce the height of the deal spike (Blattberg et al. 1995).
5.1 Limitations and Suggestions for Further Research
This empirical research based the effectiveness of reward programs solely on turnover and did not include costs. Therefore reward program profitability could not be determined, which is of critical importance for a retailer. Furthermore, this research assumes that the points pressure effect exists only in the final weeks of the program. However, due to different spending levels among customers, stamp or point threshold levels will be reached at different moments within the program. Therefore, the points pressure effect is customer specific rather than program specific.
This study provides evidence for the points pressure effect of reward programs, as well as the impact of program design characteristics. However, these results were based on turnover data of the largest store format, whereas these effects diminished for the smaller formats. More detailed analyses are needed to determine the cause of these differential effects. Moreover, the analyzed data included only one reward program based on a loyalty card. In order to better understand the impact of a transition from a short-term reward program to a (basically) continuous reward program, future research should not only analyze more card-based programs, but also include customer specific data regarding spending and reward redemption behavior.
5.2 Managerial Implications
33
34
6. References
Alsem, Karel Jan, Peter S.H. Leeflang, and J.C. Reuyl (1989), “The Forecasting Accuracy of Market Share Models Using Predicted Values of Competitive Marketing Behaviour,” International Journal of Research in Marketing, 6, 183-198.
Blattberg, Robert C., Richard Briesch, and Edward J. Fox (1995), “How Promotions Work,” Marketing Science, 14 (3, Part II), G122-G132.
Cigliano, James, Margaret Georgiadis, Darren Pleasance, and Susan Whalley (2000), “The Price of Loyalty,” The McKinsey Quarterly, 4, 68–77.
Colloquy (2009), "The 2009 Colloquy Loyalty Marketing Census," Colloquy.
Dick, Alan S. and Kunal Basu (1994), “Customer Loyalty: Toward an Integrated Conceptual Framework,” Journal of the Academy of Marketing Science, 22 (2), 99-113.
Foekens, Eijte W., Peter S.H. Leeflang, and Dick R. Wittink (1999), “Varying Parameter Models to Accommodate Dynamic Promotion Effects,” Journal of Econometrics, 89, 249-268.
Fornell, Claes and Birger Wernerfelt (1987), “Defensive Marketing Strategy by Customer Complaint Management: A Theoretical Analysis,” Journal of Marketing Research, 24 (November), 337-46. Goyal, Anita (2004), “Role of Supplementary Services in the Purchase of Credit Card Services in India,”
Asia Pacific Journal of Marketing and Logistics, 16 (4), 36-51.
Hair, J.F. Jr., W.C. Black, B.J. Babin, and R.E. Anderson (2006), Multivariate Data Analysis: a Global Perspective, Prentice Hall, New Jersey.
Hartmann, Wesley R. and V. Brian Viard (2008), “Do Frequency Reward Programs Create Switching Costs? A Dynamic Structural Analysis of Demand in a Reward Program,” Quantitative Marketing and Economics, 6 (2), 109-137.
Kivetz, Ran and Itamar Simonson (2003), “The Idiosyncratic Fit Heuristic: Effort Advantage as a Determinant of Consumer Response to Loyalty Programs,” Journal of Marketing Research, 40 (4), 454-467.
Kivetz, Ran, Oleg Urminsky, and Yuhuang Zheng (2006), “The Goal- Gradient Hypothesis Resurrected: Purchase Acceleration, Illusionary Goal Progress, and Customer Rentention,” Journal of Marketing Research, 43 (February), 39–58.
35
Kopalle, Praveen K., Scott A. Neslin, Baohong Sun, Yacheng Sun, and Vanitha Swaminathan (2007), “A Dynamic Structural Model of the Impact of Loyalty Programs on Customer Behavior.”
Kopalle, Praveen K., Carl F. Mela, and Lawrence Marsh (1999), “The Dynamic Effect of Discounting on Sales: Empirical Analysis and Normative Pricing Implications,” Marketing Science, 18 (3), 317-332. Lal, Rajiv and David E. Bell (2003), “The Impact of Frequent Shopper Programs in Grocery Retailing,”
Quantitative Marketing and Economics, 1 (2), 179–202.
Leeflang, Peter S.H., Dick R. Wittink, Michel Wedel, and Philippe A. Neart (2000), Building Models for Marketing Decisions, Dordrecht, The Netherlands, Kluwer Academic Publishers.
Leenheer, Jorna and Tammo H.A. Bijmolt (2003), “Adoption and Effectiveness of Loyalty Programs: The Retailer's Perspective,” MSI working paper, 3 (4), 91-109.
Lewis, Michael (2004), “The Influence of Loyalty Programs and Short-Term Promotions on Customer Retention,” Journal of Marketing Research, 41 (August), 281–92.
Liu, Matthew Tingchi and James L. Brock (2008), “Redemption Behavior for Credit Card Reward Programs in China,” International Journal of Bank Marketing, 27 (2), 150-166.
Mastroberte, T. (2005), “Creating Loyal Customers, Flash Foods and Superstop Take Loyalty Programs to the Next Level,” Convenience Store News, 41 (5), 2-5.
Meyer-Waarden, Lars and Christophe Benavent (2006), “The Impact of Loyalty Programmes on Repeat Purchase Behavior,” Journal of Marketing Management, 22 (February), 61–88.
Morgan, Robert M. and Shelby D. Hunt (1994), “The Commitment–Trust Theory of Relationship Marketing,” Journal of Marketing, 58 (July), 20–38.
Nunes, Joseph C. and Xavier Drèze (2006), “Your Loyalty Program Is Betraying You,” Harvard Business Review, 84 (4), 124-131.
O’Brien, Louise and Charles Jones (1995), “Do Rewards Really Create Loyalty?” Harvard Business Review, 73 (May–June), 75–82.
Oliver, Richard L. (1997), Satisfaction: A Behavioral Perspective on the Consumer, Irwin-McGraw-Hill, New York, NY.
Rothschild, Michael L. and William C. Gaidis (1981), “Behavioral Learning Theory: Its Relevance to Marketing and Promotions,” Journal of Marketing, 45 (Spring), 70-78.
Sharp, Byron and Anne Sharp (1997), “Loyalty Programs and Their Impact on Repeat-Purchase Loyalty Patterns,” International Journal of Research in Marketing, 14 (5), 473–86.
36
Uncles, Mark D. (1994), “Do You or Your Customers Need a Loyalty Programme?” Journal of Targeting Measurement and Analysis for Marketing, 2 (4), 335-50.
Van Heerde, Harald J., Peter S.H. Leeflang, and Dick R. Wittink (2000), “The Estimation of Pre- and Postpromotion Dips with Store-Level Scanner Data,” Journal of Marketing Research, 37 (August), 383-95.
37
Appendices
Appendix 1: Regression Analyses of Main and Moderating Effects TABLE 1–MAIN EFFECTS
βa SE Exp (β) b P-value VIF value
Intercept Main effects
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Interaction effects Supers * PPP Marches * PPP Supers * RBP Marches * RBP Control variables Quarter2 Quarter3 Quarter4 Easter Ascension Pentecost Summer
Christmas/New Year’s Eve (XmasNYE) Inflation2006 Inflation2007 Inflation2008 Inflation2009 Supers Marches F value Durbin Watson R2 Adjusted 14.750*** .085*** .013 -.044** -.071*** .003 .021 .018* .047*** .084*** .190*** -.026 .038 -.125*** .147*** .007 .057*** .065*** .041*** -1.390*** -2.999*** 5,629.585*** 1.545 .995 .011 .016 .028 .021 .021 .038 .038 .011 .014 .012 .023 .024 .024 .017 .025 .010 .012 .012 .016 .011 .011 2,546,114.101 1.088 1.013 0.957 0.931 1.002 1.021 1.018 1.048 1.088 1.209 0.974 1.038 0.883 1.158 1.007 1.058 1.067 1.042 0.249 0.050 .000 .000 .642 .035 .001 .940 .583 .096 .001 .000 .000 .263 .111 .000 .000 .456 .000 .000 .010 .000 .000 3.598 3.328 2.527 2.527 2.133 2.133 1.820 2.263 1.858 1.052 1.082 1.082 1.833 1.183 1.456 1.455 1.695 1.406 1.963 1.963 a *** p<0.01; ** p<0.05; * p <0.1
38
TABLE 2–MODERATING EFFECT OF PROGRAM TYPE
βa SE Exp (β) b P-value VIF value
Intercept Main effects
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Moderating effects Program Type (PT) * PPP Interaction effects Supers * (PT * PPP) Marches * (PT * PPP) Control variables Quarter2 Quarter3 Quarter4 Easter Ascension Pentecost Summer
Christmas/New Year’s Eve (XmasNYE) Inflation2006 Inflation2007 Inflation2008 Inflation2009 Supers Marches F value Durbin Watson R2 Adjusted 14.761*** .053*** .024 -.115*** .077* .123*** .018* .049*** .083*** .187*** -.024 .040* -.125*** .145*** .008 .055*** .063*** .064*** -1.405*** -3.022*** 5,908.920*** 1.525 .995 .011 .011 .018 .038 .044 .044 .011 .014 .012 .023 .024 .024 .017 .025 .010 .012 .013 .021 .009 .009 2,574,924.910 1.054 1.024 0.891 1.079 1.129 1.018 1.050 1.086 1.205 0.976 1.041 0.883 1.155 1.008 1.057 1.065 1.066 0.245 0.049 .000 .000 .173 .003 .082 .006 .093 .001 .000 .000 .317 .089 .000 .000 .438 .000 .000 .002 .000 .000 1.771 1.294 4.447 2.061 2.061 1.820 2.273 1.867 1.058 1.087 1.087 1.833 1.188 1.457 1.462 1.714 2.409 1.394 1.394 a *** p<0.01; ** p<0.05; * p <0.1
39
TABLE 3–MODERATING EFFECT OF PROGRAM DURATION
βa SE Exp (β) b P-value VIF value
Intercept Main effects
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Moderating effects Program Duration (PD) * PPP Program Duration (PD) * RBP Interaction effects Supers * (PD * PPP) Supers * (PD * RBP) Marches * (PD * PPP) Marches * (PD * RBP) Control variables Quarter2 Quarter3 Quarter4 Easter Ascension Pentecost Summer
Christmas/New Year’s Eve (XmasNYE) Inflation2006 Inflation2007 Inflation2008 Inflation2009 Supers Marches F value Durbin Watson R2 Adjusted 14.751*** -.060 .230** .009** -.014* -.003** .001 -.005*** .002 .019* .047*** .085*** .189*** -.024 .040* -.124*** .129*** .004 .053*** .065*** .042*** -1.389*** -2.998*** 5,188.593*** 1.565 .995 .011 .074 .117 .005 .007 .001 .002 .001 .002 .011 .014 .012 .023 .023 .023 .017 .026 .010 .012 .013 .016 .011 .011 2,547,842.729 0.939 1.250 1.009 0.986 0.997 1.001 0.995 1.002 1.019 1.048 1.089 1.208 0.976 1.041 0.883 1.137 1.004 1.055 1.067 1.043 0.249 0.050 .000 .420 .049 .047 .060 .017 .778 .000 .368 .089 .001 .000 .000 .310 .087 .000 .000 .724 .000 .000 .008 .000 .000 80.010 57.092 82.646 59.993 2.517 2.129 2.517 2.129 1.985 2.307 1.882 1.053 1.085 1.085 1.834 1.369 1.542 1.545 1.794 1.410 1.950 1.950 a *** p<0.01; ** p<0.05; * p <0.1
40
TABLE 4–MODERATING EFFECT OF SPEND REQUIREMENT
βa SE Exp (β) b P-value VIF value
Intercept Main effects
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Moderating effects Spend Requirement (SR) * PPP Spend Requirement (SR) * RBP Interaction effects Supers * (SR * PPP) Supers * (SR * RBP) Marches * (SR * PPP) Marches * (SR * RBP) Control variables Quarter2 Quarter3 Quarter4 Easter Ascension Pentecost Summer
Christmas/New Year’s Eve (XmasNYE) Inflation2006 Inflation2007 Inflation2008 Inflation2009 Supers Marches F value Durbin Watson R2 Adjusted 14.752*** .035 .108* .087* -.169* -.075** .029 -.124*** .069 .015 .045*** .083*** .190*** -.025 .039* -.124*** .141*** .007 .057*** .067*** .040*** -1.391*** -3.000*** 5,163.230*** 1.563 .995 .011 .027 .056 .049 .090 .033 .057 .033 .057 .011 .014 .012 .023 .023 .023 .017 .025 .010 .012 .013 .017 .011 .011 2,551,348.551 1.035 1.113 1.089 0.841 0.928 1.028 0.883 1.069 1.015 1.046 1.086 1.209 0.975 1.039 0.884 1.152 1.007 1.059 1.069 1.040 0.249 0.050 .000 .208 .052 .077 .060 .025 .610 .000 .228 .161 .001 .000 .000 .283 .099 .000 .000 .477 .000 .000 .020 .000 .000 10.836 12.952 13.288 15.209 2.463 2.117 2.463 2.117 1.993 2.303 1.870 1.053 1.083 1.083 1.836 1.244 1.457 1.456 1.761 1.626 1.888 1.888 a *** p<0.01; ** p<0.05; * p <0.1
41
TABLE 5–MODERATING EFFECT OF PROGRAM INVESTMENT
βa SE Exp (β) b P-value VIF value
Intercept Main effects
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Moderating effects
Program Investment (PI) * PPP Program Investment (PI) * RBP Interaction effects Supers * (PI * PPP) Supers * (PI * RBP) Marches * (PI * PPP) Marches * (PI * RBP) Control variables Quarter2 Quarter3 Quarter4 Easter Ascension Pentecost Summer
Christmas/New Year’s Eve (XmasNYE) Inflation2006 Inflation2007 Inflation2008 Inflation2009 Supers Marches F value Durbin Watson R2 Adjusted 14.748*** .018 .155** .042*** -.044** -.033*** .006 -.053*** .009 .024** .048*** .083*** .184*** -.024 .036 -.122*** .125*** .005 .055*** .068*** .061*** -1.391*** -3.001*** 4,772.033*** 1.707 .995 .012 .029 .069 .011 .021 .008 .013 .008 .013 .012 .014 .012 .023 .030 .030 .017 .026 .010 .011 .015 .021 .011 .011 2,541,790.743 1.018 1.164 1.043 0.957 0.967 1.006 0.948 1.009 1.024 1.049 1.086 1.202 0.976 1.036 0.885 1.133 1.005 1.057 1.070 1.063 0.249 0.050 .000 .535 .026 .000 .041 .000 .668 .000 .498 .041 .001 .000 .000 .413 .227 .000 .000 .580 .000 .000 .003 .000 .000 8.949 13.425 11.490 15.764 2.298 2.072 2.298 2.072 1.823 2.347 1.859 1.060 1.080 1.080 1.864 1.364 1.413 1.445 1.627 1.189 1.691 1.691 a *** p<0.01; ** p<0.05; * p <0.1
42
TABLE 6–MODERATING EFFECT OF TIME BETWEEN PROGRAMS
βa SE Exp (β) b P-value VIF value
Intercept Main effects
Points Pressure Period (PPP) Rewarded Behavior Period (RBP) Moderating effects
Time Between Programs (TBP) * PPP Time Between Programs (TBP) * RBP Interaction effects Supers * (TBP * PPP) Supers * (TBP * RBP) Marches * (TBP * PPP) Marches * (TBP * RBP) Control variables Quarter2 Quarter3 Quarter4 Easter Ascension Pentecost Summer
Christmas/New Year’s Eve (XmasNYE) Inflation2006 Inflation2007 Inflation2008 Inflation2009 Supers Marches F value Durbin Watson R2 Adjusted 14.738*** .031* -.013 .004*** .004** -.004*** -.002 -.006*** -.004** .023** .050*** .083*** .187*** -.025 .039* -.126*** .142*** .007 .063*** .074*** .046*** -1.381*** -2.984*** 5,588.813*** 1.678 .996 .011 .017 .028 .001 .002 .001 .002 .001 .002 .011 .013 .012 .022 .023 .023 .016 .024 .009 .012 .013 .016 .010 .010 2,514,847.973 1.031 0.987 1.004 1.004 0.996 0.998 0.994 0.996 1.023 1.051 1.086 1.205 0.975 1.039 0.882 1.152 1.007 1.065 1.077 1.047 0.251 0.051 .000 .066 .646 .000 .012 .000 .179 .000 .015 .030 .000 .000 .000 .263 .088 .000 .000 .434 .000 .000 .004 .000 .000 4.293 3.458 6.040 5.102 2.269 2.066 2.269 2.066 2.050 2.303 2.037 1.065 1.090 1.090 1.839 1.226 1.463 1.720 2.049 1.484 1.659 1.659 a *** p<0.01; ** p<0.05; * p <0.1
43
Appendix 2: Visual Representation of Moderating Effects
FIGURE 1-MODERATING EFFECT OF PROGRAM DURATION
44
FIGURE 3-MODERATING EFFECT OF PROGRAM INVESTMENT
45
Appendix 3: Visual Representation of Store Format Interaction Effects
FIGURE 1-MAIN EFFECTS AND THE INTERACTION EFFECT OF STORE FORMATS
46
FIGURE 3-PROGRAM DURATION AND THE INTERACTION EFFECT OF STORE FORMATS
47
FIGURE 5-PROGRAM INVESTMENT AND THE INTERACTION EFFECT OF STORE FORMATS
48
Appendix 4: Predictive Validity Measures
Appendix 4.1: Mean Absolute Percentage Error (MAPE)
+<2; 1 Q 11 ∗ |TTQ TU|
⋅ 100
W W∗X
where:
1 = total number of observations
1∗ = number of observations used for estimation
T#TU( = the actual (predicted) value in period t
Appendix 4.2: Theil’s U-statistic
Y∑∑W W∗X #T#T Q TU(
Q T[ ( W
W∗X
where:
1 = total number of observations
1∗ = number of observations used for estimation
