A SHORT TERM LOYALTY PROGRAM
IN THE GROCERY RETAILING
AN EMPIRICAL RESEARCH BASED ON LONGITUDINAL DATA
JULY 7 , 2014
thA SHORT TERM LOYALTY PROGRAM
IN THE GROCERY RETAILING
AN EMPIRICAL RESEARCH BASED ON LONGITUDINAL DATA
Name:
Sjanne van Velsen
Studentnumber:
1704656
Address:
Damsport 461
9728 PS Groningen
Email:
s.van.velsen@student.rug.nl
Telephone:
+31 (0)625132256
First supervisor:
A. Minnema MSc
Second supervisor:
Prof. Dr. T. H.A. Bijmolt
Supervisor BrandLoyalty: K. Heeren MSc
MASTER THESIS
MSc MARKETING INTELLIGENCE & MANAGEMENT
UNIVERSITY OF GRONINGEN
1
MANAGEMENT SUMMARY
Frequency reward programs are playing an important role in the retail grocery sector and became a key tool in marketing strategies of many retailers. The financial impact of these programs sometimes fail to meet retailers’ expectations, however firms and retailers spend over billions on those programs (Dowling and Uncles 1997; Wagner, Hennig-Thurau, and Rudolph 2009). This underlines the importance of determining the impact of a frequency reward program.
This empirical research analyses the effects of an implemented frequency reward program in the retail grocery sector. The impact of the program is assessed by determining the stretch in customer spending. In the grocery retailing, customers often patronize multiple stores, so it is a challenge for retailers to gain a greater share of wallet by inducing switching behaviour (Meyer-Waarden 2007). The effects on customer spending are determined by using longitudinal transactional data on an implemented frequency reward program by a major retailer. The data includes weeks prior to the program. Data was provided by BrandLoyalty International, which is a leading company in the area of designing tailor-made loyalty programs. The effects of the program are estimated by using a linear regression model with customer specific random effects.
The findings suggest frequency reward programs can successfully enhance customer spending among low spenders, compared to moderate spenders. Moderate spenders are found to increase spending as well, compared to heavy spenders. This reflect the points pressure effect (Taylor and Neslin 2005). Furthermore, customers are found to increase spending as they approach their goal of collecting a reward. This implies customers excel more motivation when being close towards reaching the program requirements to redeem, which is known as the goal-gradient hypothesis (Kivetz, Urminsky, and Zheng 2006; Kivetz 2000) This stretch in spending is found to endure in the subsequent week after a redemption event. This is known in literature as the rewarded behaviour effect (Taylor and Neslin 2005; Lal and Bell 2003). Moreover, the findings of this empirical research suggest that promotions which give customers the opportunity to earn additional points, positively influence customer spending. These promotions are referred to in this research as suppler funded products.
Lastly, the effects on customers’ likelihood of redemption are analysed by estimating a binary logistic regression with customer specific random effects. Results indicate that the weekly spending level of customers influence their likelihood to redeem. The findings also suggest that heavy spenders are more likely to redeem during the frequency reward program, compared to low spenders. Moreover, the decision to redeem increases across the weeks of the program period. This indicates that the length of the program is important in customers’ decision to redeem in a frequency reward program. The findings of this research provide various managerial implications. From the results of this research, managers can learn that frequency reward programs can enhance customer spending, and thereby positively influence retail turnover. The findings of this research also suggest that the performance of a frequency reward programs need to be evaluated in the short- and long-term to determine the impact of the points pressure and rewarded behaviour mechanisms. This study concludes by suggesting directions for further research.
Key words: Frequency reward program, loyalty, points pressure effect, rewarded behaviour, goal-gradient hypothesis,
2
PREFACE
‘The only way around it, is through’
Stuck-in-the-middle is one of the concepts which is discussed and tested for in this empirical research. Although, this phenomenon is not visible during the execution of the studied frequency reward program, the stuck-in-the-middle effect does apply to the process of writing my thesis. Fortunately, with all the help of my supervisors, friends and family around me, I managed to finalize and complete my thesis. It was a challenging, but rewarding process of five months and the result lies in front of you. This research also represents the end of my six years of studying. I am grateful for the opportunity to write my thesis in cooperation with BrandLoyalty and with the obtained data, which resulted in fruitful insights and results. This research would not have been realized without the help of my supervisors, friends and family.
First of all, I would like to offer my gratitude to my supervisors Mr. A. Minnema and Prof. Dr. T.H.A Bijmolt for supporting me throughout this process. Especially Mr. A. Minnema for providing insightful feedback, which improved my thesis. Also a special thanks goes out to Ms. K. Heeren and Mr. R. Jansen from BrandLoyalty for their feedback and cooperation. I would also like to thank my parents, who have always supported me in everything I do. Without them, I would never have been where I am today. My boyfriend, Arjan, was a very great support as well, who always kept believing that I could pull it off. Last, but certainly not the least, my (study) friend Amanda. We followed each other’s process of writing our theses every step of the way, so whenever I needed a positive word she was there.
3
TABLE OF CONTENTS MANAGEMENT SUMMARY ... 1 PREFACE ... 2 LIST OF FIGURES ... 5 LIST OF TABLES ... 5 1. INTRODUCTION ... 6 1.1 Research Context... 7 1.2 Objective ... 8 1.3 Contribution ... 9 1.4 Structure ... 10 2. THEORETICAL FRAMEWORK ... 11 2.1 Turnover ... 112.2 Points Pressure and the Rewarded Behaviour Mechanisms ... 12
2.2.1 Points Pressure Mechanism ... 12
2.2.2 Goal-Gradient Hypothesis ... 13
2.2.3 Stuck-in-the-Middle Effect ... 14
2.2.4 Supplier Funded Products. ... 15
2.2.5 Rewarded Behaviour Mechanism ... 15
2.2.6 Moderating Effect of Pre-Program Spending Level ... 16
2.3 Redemption ... 17
2.3.1 Weekly Spending Level ... 18
2.3.2 Saved Rewards ... 18
2.3.3 Type of Spender ... 19
2.3.4 Supplier Funded Products ... 19
4
4.4.1 Goodness of Fit ... 31 4.4.2 Parameters Estimates ... 31 4.5 Model Validation ... 33 4.5.1 Face Validity ... 33 4.5.2 Predictive Validity ... 34 5. REDEMPTION ... 35 5.1 Descriptives ... 35 5.2 Redemption Behaviour ... 35 5.2 Model Specification ... 36 5.2.1 Model Notation ... 36 5.3 Results ... 37 5.3.1 Goodness of Fit ... 37 5.3.2 Parameter Estimates ... 376. CONCLUSION AND RECOMMENDATIONS ... 39
6.1 Conclusion ... 39
6.1.1 Customer Spending ... 39
6.1.2 Redemption ... 41
6.2 Managerial Implications... 41
6.3 Limitations and Suggestions for Further Research ... 42
6.3.1 Program Evaluation ... 42
6.3.2 Type of Reward Program and Context... 43
6.3.3 Demographics ... 43
REFERENCES ... 44
APPENDIX A CHAPTER 4:CUSTOMER SPENDING ... 51
APPENDIX B CHAPTER 4:PREDICTIVE VALIDITY ... 55
APPENDIX C:MODEL VALIDATION CUSTOMER SPENDING ... 56
APPENDIX D:MODEL VALIDATION REDEMPTION ... 58
APPENDIX ECHAPTER 5: REDEMPTION ... 59
5
LIST OF FIGURES
Figure 1 Points pressure and the rewarded behaviour mechanisms (Taylor and Neslin 2005) ... 12
Figure 2 Conceptual Model ... 21
Figure 3 Checking for outliers ... 23
Figure 4 Development Weekly Spending ... 24
Figure 5 Development Frequency of Visits ... 24
Figure 6 Checking for the assumption of normality after log transformation ... 29
Figure 7 Residual Plot ... 30
Figure 8 Overview Observed Spending Per Segment ... 33
Figure 9 Scatterplot Actual And Predicted Spending ... 34
Figure 10 Development Redemption during FRP ... 35
Figure 11 Development Weekly Spendng During FRP ... 35
Figure 12 Development Sales Supplier Funded Products ... 53
Figure 13 Histogram residuals ... 53
Figure 14 Normality probability plot of the residuals ... 53
Figure 15 Moderation effects ... 54
Figure 16 Weekly spending validation sample ... 56
Figure 17 Redeemed Rewards ... 59
Figure 18 Spending (non)- participants ... 59
Figure 19 Development Customer Participation Estimation Sample ... 60
Figure 20 Development Customer Participation Validation Sample ... 60
LIST OF TABLES
Table 1 Descriptive Statistics ... 24Table 2 Multicollinearity diagnostics: Variance Inflation Factors ... 31
Table 3 Linear Random Effects Model Results ... 32
Table 4 Comparison Redemption vs. Non-redemption ... 36
Table 5 Logistic Random Effects Regression Results ... 38
Table 6 Overview Of Findings ... 39
Table 7 Breusch – Pagan Multiplier Test for RE ... 51
Table 8 Hausman Test linear regression ... 51
Table 9 Hausman test after logtransformation ... 52
Table 10 Overview coefficients alternative models ... 52
Table 11 Linear random effects model validation... 57
Table 12 Comparison estimates estimation and validation sample ... 57
Table 13 Binary random effects model validation ... 58
6
1. INTRODUCTION
Loyalty programs are playing an increasingly important role in the customer relationship management efforts of companies in the current marketplace. The use of loyalty programs have spread into a diverse range of industries like gaming, travelling, financial services and department stores (Zhang and Breugelmans 2012). Loyalty programs are also becoming more important in the retail grocery sector, as they play a crucial role in the marketing strategy of a lot of retailers. The involvement of retail companies in loyalty programs grew from 45% in 2009 to 52% in 2010 (García-Gómez, Gutiérrez-Arranz, and Gutiérrez-Cillán 2012a). According to Evanschtizky et al. (2011), the use of retail loyalty programs is growing with 11% per year. Moreover, the popularity of these programs have increased around the world (Sharp and Sharp 1997). A recent survey, conducted by The Logic Group, shows that almost two- thirds of people, which is 62%, say that they belong to at least one loyalty program (Kopalle et al. 2012). Based on these insights, it can be concluded that loyalty programs are becoming more important, not only for customers but also for firms. This is confirmed by McKinsey research, which found that around half of the ten largest retailers in the U.S. and the U.K. implemented loyalty programs (Cigliano et al. 2000). According to Leenheer and Bijmolt (2007), around 37% of all Dutch retailers have a loyalty card system.
Loyalty programs are defined in literature in many ways. For this research, the definition by Allaway, Berkowitz and D’Souza (2003) is most fitting. The authors of this study define loyalty programs as a concentrated effort by retailers with the aim to enhance basket size, visit frequency and to attract more customers. Loyalty programs in turn strengthen relationships with customers. In prior literature, many classifications are made of loyalty programs. The study of García-Gómez et al. (2012a) , make a distinction in loyalty programs that are often implemented in the retail grocery sector. The authors classify reward programs, loyalty cards and VIP programs as the most common loyalty programs used by grocery retailers. Each type of program operates differently and offers the participants different kinds of rewards (García-Gómez et al. 2012a).
Reward programs are comparable to the concept of frequency reward programs (Blattberg, Kim and Neslin 2008; Kopalle et al. 2012). Frequency reward programs represent the original trading stamp programs, where customers have to buy a number of times to meet the program requirement to obtain a reward (Bijmolt, Dorotic and, Verhoef 2010; Kopalle et al. 2012). Customers participating and reaching the program threshold in these type of programs are rewarded with discounts or gifts. Frequency reward programs can be defined as a tool to increase customer value, by means of offering tangible benefits to customers that make repeated purchases at a specific company. The customers are rewarded proportionally to their level of purchases or profitability (Blattberg et al. 2008; Taylor and Neslin 2005). Especially the retail grocery sector moved towards adopting the frequent shopper program, as the competition for customers intensified. Moreover, there are not many ways for grocery retailers to distinguish themselves in the market with respect to the grocery store appearance (Lal and Bell 2003).
7
towards relationship building while short-term loyalty programs are more effective in achieving a direct effect on sales (Dorotic, Bijmolt, and Verhoef 2012).
The type of loyalty program that will be analysed in this empirical research can be classified as a FRP since the program had a short time span of 16 weeks. The FRP was implemented by a major retailer, aimed at enhancing retail turnover. Prior research found evidence for the increased effect of FRPs on sales, share of wallet, customer lifetime duration and found to lower switching behaviour (Drèze and Hoch 1998; Lal & Bell 2003; Meyer-Waarden 2007; Nunes and Drèze 2006a; Taylor and Neslin 2005). Turnover can be increased by a change in customer purchasing behaviour, meaning that customers stretch their spending level as a result of participating in the FRP. FRPs are found to increase sales through two mechanisms: points pressure and rewarded behaviour, which are directly related to reward redemption. Though, several studies also point out that FRPs could be sometimes ineffective, without beneficial outcomes for the offering retailer (Dowling and Uncles 1997; Liu and Yang 2009; Sharp and Sharp 1997). Therefore, the financial impact of these program sometimes fail to meet expectations, while firms spend over billions on FRPs (Wagner et al. 2009). This research aims at analysing the effects of an implemented FRP on retail turnover.
Besides the effect of a FRP on retail turnover, analysing redemption in retail FRPs is important, since customers that receive value from a retailer are found to generate a feeling of obligation towards the retailer (Smith and Sparks 2009b; Wulf, Odekerken-Schröder, and Iacobucci 2001). In the context of a FRP, customers receive value in the form of a reward. Therefore, the obtained rewards might lead to reciprocity between the retailer and the customers’ participating in the FRP (Kumar and Shah 2004). Based on the importance of redemption, the influences on customers’ likelihood to redeem are examined in this research.
1.1 Research Context
T
he present research is conducted in cooperation with BrandLoyalty, an international company founded in 1995. BrandLoyalty creates innovative, tailor-made loyalty programs for leading food retailers with the aim to create consumers engagement and to create changes in customer behaviour. This in turn leads to an additional increase in sales and profit for the retailers (WikiLab 2014). The main mechanisms that make up a loyalty program designed by BrandLoyalty are the amount spend, the type of reward, the timeframe and support. The first three mentioned mechanisms are most crucial for a FRP, according to the company. The loyalty programs BrandLoyalty designs and executes are FRPs that run between 12 and 20 weeks, which can be categorized as short-term programs (WikiLab 2014). BrandLoyalty runs these short-term loyalty programs in more than 200.000 retail stores across the world. These kind of programs are designed for grocery retailing companies, which is one of the least profitable sectors since the net margins are between 1% and 2% (Lal and Bell 2003).8
from July 30th – October 31st 2013. The FRP lasted however until November 18th, unfortunately the transactional
data of this period was not available for this research.
1.2 Objective
The main focus of this research is to look into the impact of an implemented FRP on retail turnover. Besides, the impacts on the decision of a customer to redeem during the program will be analysed. More specifically, the main research question can be defined as follows:
What are the effects of the FRP on retail turnover and what influences customers’ likelihood of redemption?
The first part of the theoretical framework will explore what the effects are of the points pressure mechanisms and the rewarded behaviour on retail turnover, by assessing the stretch in individual customer spending. The research questions that will be answered in this part are:
- What is the impact of the points pressure effect on retail turnover during the FRP? - What is the impact of the rewarded behaviour effect on retail turnover after redemption?
The consecutive part will analyse the consequences of the goal-gradient hypothesis and the stuck-in-the-middle phenomenon on retail turnover. The goal-gradient hypothesis implies that customers expend more effort as they approach the goal of attaining a reward. The stuck-in-the-middle effect implies a decrease in customer motivation halfway the FRP. The research questions that will be answered in this part are:
- What is the impact of the goal-gradient hypothesis on retail turnover?
- Is there a stuck-in-the-middle effect and if so, what is the impact on retail turnover?
The subsequent chapter of the theoretical framework will explore the impact of supplier funded products on retail turnover. By purchasing these supplier funded products, customers could receive additional points. This empirical research will also investigate the moderating effect of the FRP on the relation between supplier funded products and retail turnover. The research questions that will be answered in this part are:
- What is the impact of supplier funded products on retail turnover?
- What is the effect of supplier funded products on retail turnover during the FRP?
The following chapter will examine the moderating effect of pre-program spending level on retail turnover. Pre-program spending level is important to include, as customers can maximize rewards during a FRP by stretching their spending level during a single visit. This increased spending level is an essential part of their value contribution to the retailer (Liu, 2007). The research questions that will be answered in this part is:
9
After the effects of the described elements on retail turnover are determined, the redemption behaviour and the effects on customers’ likelihood of redemption will be explored. The research questions that will be answered in this part are:
- Does customer spending influence the likelihood of redemption?
- Do the number of customers’ saved rewards influence the likelihood of redemption?
- Do the number of weeks since the program has started influence the likelihood of redemption? - What is the influence of buying supplier funded products on the likelihood of redemption? - What is the influence of customers’ spending before the start of the program on the likelihood of
redemption?
1.3 Contribution
This empirical research will provide academics with additional insights in the effects of FRPs, by assessing the impact on retail turnover in the short run as well as in the long run. This research also investigates the effects of the program over a longer time span, which contribute to current empirical studies. According to Liu (2007), relatively few empirical studies have examined the longitudinal effects of FRPs, which leaves a gap in understanding the impact of a FRP. Nevertheless, according to the same author, it is important to measure this longitudinal effect since the effects of a FRP develop over multiple periods or can be time dependent. FRPs are often evaluated at a single point in time, without analysing temporary effects. Therefore, the rewarded behaviour effect after a redemption event will be analysed. There are studies that have assessed the effects of points pressure and rewarded behaviour, but more insights are needed on the long-term impacts of these mechanisms. Besides the points pressure effect and rewarded behaviour, the effect of a customer approaching their redemption goal on customer motivation will be explored. Increased achievement motivation might result in an increase in spending as a customer gets closer towards reaching the redemption requirement, which is known as the goal-gradient hypothesis. The goal-gradient hypothesis has important implications for goal pursuit, but is understudied in terms of human behaviour (Kivetz et al. 2006). This research extends the effects accommodated by the gradient hypothesis on customer behaviour, by determining the impact of the goal-gradient effect on customer spending during a FRP. Moreover, this empirical research will examine the effect of supplier funded products on retail turnover. Relatively few studies have explored the impact of supplier funded products during a FRP.
10
1.4 Structure
11
2. THEORETICAL FRAMEWORK
The goal of the theoretical framework is to identify the elements that can possibly influence the dependent variables. The first chapter will look into the first dependent variable, retail turnover. This chapter will be followed by chapter 2,2, which shortly introduce the points pressure and the rewarded behaviour mechanisms. The subchapters will explain the possible predictors and concludes with the proposed hypotheses. Chapter 2.3 will define and explain the second dependent variable, which is redemption. The subsequent subchapters will explain the possible predictors on the likelihood of redemption and concludes with the proposed hypotheses. The theoretical framework will conclude with the conceptual model in chapter 2.4.
2.1 Turnover
Retailers that have implemented a successful FRP in the past claim that redemption activity directly generates additional turnover through better knowledge of customers and causes a stretch in spending by satisfied consumers (Drèze and Hoch 1998; Taylor and Neslin 2005). Turnover will be operationalized by measuring the effect of the program on customer spending. Customer spending is defined as the average amount of money a customer spends at a particular retailer for the time period of one week (Fernandez-Villaverde and Krueger 2007). According to Meyer-Waarden and Benavent (2009), spending is a measure of the retailer’s store loyalty. Furthermore, spending is an important determinant of the retailers profit margin however, transaction size not often included in prior studies (Liu, 2007).
Customers visit several grocery stores and make frequent purchases in more than one store. Besides, customers often purchase small amounts. Usually, customers visit several stores depending on circumstances like promotions or FRPs. Customers also maintain a range of competitive stores that they regularly visit, but they have one main store where most of their share of wallet is spend (Meyer-Waarden and Benavent 2009). A FRP can encourage the consolidation of purchases and can enhance customer commitment by giving the customer a reason to concentrate more purchases at one retailer (Liu 2007; Nunes and Drèze 2006a). However, the increase in spending might also be due stockpiling, which occurs because customers are induced to buy more or sooner than they would have done otherwise. The result is that customers end up with more products than they would normally buy if there was no promotion (Ailawadi et al. 2007). The proposed effect on spending could be due to brand switching, which can be defined as the percentage of the sales promotion elasticity that can be assigned to secondary demand effects (Van Heerde, Gupta, and Wittink 2003). Besides the effect of the FRP on spending, the program can also have an effect on store attraction. More customers might be attracted by the FRP, which leads to more traffic while customers spending level might remain unchanged. Leenheer et al. (2007), found in their study that program membership has a positive influence on store attraction. The same study also underline that FRPs can create commitment and attachment towards the concerning retailer.
12
2.2 Points Pressure and the Rewarded Behaviour Mechanisms
A successful FRP can lead to an increase in turnover by the before mentioned points pressure effect and/or by rewarded behaviour. In figure 1, the impacts of the points pressure and the rewarded behaviour effects are visualized. Before the FRP is implemented, the customer’s baseline purchase rate is reflected in period A. During the program (period B), the customer might increase their spending level in order to meet the requirement to redeem a reward. At the end of period B, the customer might receive a reward depending on the purchase level. The potential long-term impact of the reward is displayed in period C, where the purchase level could be equal or higher than the baseline. The latter indicates the existence of the rewarded behaviour effect. However, the simple effects showed in figure 1 are not guaranteed to operate linearly (Taylor and Neslin 2005). In fact, the strength of both mechanisms depend on the program requirements; if the requirements are too low, there will be minimal points pressure and a high redemption rate, but ample rewarded behaviour. If the requirements are high, there will be more points pressure but lower redemption rates so the rewarded behaviour will less (Kopalle et al. 2007).
Figure 1 Points pressure and the rewarded behaviour mechanisms (Taylor and Neslin 2005)
2.2.1 Points Pressure Mechanism
The points pressure mechanism represents a temporary shock in the level of spending, since customers increase this level in order to qualify for the reward (Liu 2007). FRPs primarily create a points pressure effect caused by the program’s clear time period and goal (Liu 2007). For the points pressure effect to occur, a customer must value the reward and the reward must be attractive (Bijmolt et al. 2010). One psychological reason for the points pressure is the goal distance model, which implies that humans make judgements relative to a benchmark. During a FRP, the benchmark is the amount of effort needed to reach the goal, which is to earn a reward (Blattberg et al. 2008). Customers judge their progress by assessing the percentage of distance that still needs to be covered toward receiving the reward.
13
obtain the reward (Bijmolt et al. 2010). To care about the opportunity to collect a reward by saving points, the customer must be future oriented. This means that the customer has to care about a future reward in order to start saving points (Taylor and Neslin 2005). These switching costs and future orientation causes customers to increase their spending level to qualify for a reward. However, according to Lal and Bell (2003), the best customer may already visit the retailers’ stores to a certain degree. This might make it difficult to increase their basket size. Therefore, a different customer segment should be targeted (Lal and Bell (2003). Though, some studies argue it is best to focus on existing customers, as FRPs do not lead to increased penetration (Meyer-Waarden and Benavent 2006). Indicating that these programs have little effect on recruiting new customers, but mainly influence purchase behaviour of existing customers. These different studies indicate the importance of focussing on the right target group that are most likely to change their behaviour. In this way, the greatest stretch in customers’ spending could be received by the FRP.
A study by Lewis (2004) found forward-looking behaviour to be present during a FRP offered by an internet grocery retailer, which motivated customers to increase spending. However, customers are only stimulated to accelerate their purchases and increase spending, if the FRP offers a valued reward (Nunes and Drèze 2006a). Current studies found evidence for the points pressure effect and the increased effect on sales during the FRP (Drèze and Hoch 1998; Kivetz et al. 2006; Kopelle et al. 2007; Lal and Bell 2003; Lewis 2004; Taylor and Neslin 2005). Based on the aforementioned arguments, the hypothesis will be defined as follows:
H1: During the FRP, the points pressure effect will lead to a positive increase in customer spending.
2.2.2 Goal-Gradient Hypothesis
Kivetz et al. (2006) found that customers increase spending as they get closer towards receiving a reward. Customers reaching their redemption goal is found to results in purchase acceleration, which in turn increases retail turnover (Taylor and Neslin 2005; WikiLab 2014). This effect is known as the goal-gradient hypothesis, which indicates that the tendency to reach a goal increases when being close towards the goal. The closer a customer is toward the end goal, the more motivated he or she is to complete the task (Kivetz et al. 2006). The goal-gradient hypothesis is an important element of the points pressure effect. A study by Garland and Conlon (1998), focused on the work-completion hypothesis and found that completing a certain goal gets more priority over other goals, with increasing progress towards that specific goal. The goal-gradient effect especially holds in a single goal pursuit context where there is one goal, since motivation is shown to increase as people make progress. This causes a greater overall success (Fitzsimons and Fishbach 2010; Kivetz et al. 2006). Motivation can be defined as the desire to engage in a goal-oriented behaviour, for example loyalty (Meyer-Waarden, Benavent, and Castéran 2013).
14
The effect of the goal-gradient hypothesis has important issues for customer behaviour in FRPs (Lal and Bell 2003). Kivetz (2000), propose that a smaller goal distance leads to an increase in achievement motivation. However, these results do not always apply, as a study byBonezzi, Brendln and De Angelis (2011) found that motivation does not always increases with progress toward the desired end state. The classic gradient effect of increasing motivation as people approach their goal is found to occur when individuals focus on the desired end state throughout the process. A decreasing motivational gradient effect occurs when individuals focus on the initial state (Bonezzi et al. 2011). Though, a study by Kivetz et al. (2006) found that customers continue and accelerate in their efforts as they near the program threshold, which is the reward requirement in a FRP. The same study implies that the operationalization of increased effort depends on the structure of that specific reward program. Acceleration might be identified through shorter interpurchase times as well as increased spending levels if the program requires a more intense activity. In this study, increased customer effort to reach the end goal is measured by analysing the effect on spending if a customer approaches their redemption goal.
The goal-gradient effect might influence spending in the analysed FRP, since a study by Cheema and Bagchi (2011) found that people who approach their goal exert more effort and commitment if they can easily visualize their goal. Since the participants in the studied FRP had a graphic representation in the form a saving card, it is expected that customers exert more commitment towards reaching their redemption goal when being close toward their goal. Based on the aforementioned arguments, the hypothesis will be defined as follows:
H2: Accommodated by the goal-gradient effect, customers motivation increases as they approach their redemption goal.
2.2.3 Stuck-in-the-Middle Effect
Studies indicate that even if customers are willing to participate in a FRP, about 75-80% drops out halfway during the process. This indicates that customers who are halfway in the process of collecting rewards, are becoming less motivated since they are likely to loosen goal pursuit (Wiebenga and Fennis 2014). The study by Bonezzi et al. (2011) show that individuals exert less effort when being halfway toward reaching their goal, compared to the end or the beginning of the process. Hence, motivation decreased in the middle of the process, which indicates there is a vulnerability halfway the process of reaching a goal. This phenomenon is often mentioned as the stuck-in-the-middle effect, which is also found to be present in the mentioned study by Wiebenga and Fennis (2014). This stuck-in-the-middle effect might negatively impact customer spending and thereby turnover. If customers exert less effort halfway the program, it can be expected that customers will visit the retailer less often or decrease spending. Based on the aforementioned arguments, the hypothesis will be defined as follows:
15
2.2.4 Supplier Funded Products.
Many retailers implement promotions during a FRP. By purchasing these promotional products, customers are given the opportunity to earn additional rewards or points besides those earned on their total purchases (Minnema, Bijmolt, and Non 2014). In this research, these promotional products will be referred to as supplier funded products. A customer can earn extra points by purchasing these specially promoted items, which are often sponsored by the supplier but sometimes also by the retailer (Fredregill and Schrum 2000). One of the reasons why supplier funded products are implemented, is due to the relatively small effect of a FRP on total sales (Leenheer et al. 2007). Supplier funded products can lead to an increase in sales, since customers are likely overreact to promotions that include a free component. Mainly because adding a free component is found to increase the perceived value by the customer (Shampanier, Mazar, and Ariely 2007). Besides, the effectiveness of the FRP for the manufacturers’ own product sales can enhance if they arrange to get a special promotion for their product, in exchange for their participation in the program (Minnema et al. 2014).
Zhang and Breugelmans (2012) found that the impact of supplier funded products on customer purchase behaviour was mainly through its influence on customers’ decision to visit the store. The direct effect on spending was found to be small. The same study found that offering supplier funded products reduced attrition among existing customers and attracted more unique customers, which accounted for the majority of the increase in total turnover. The study by Minnema et al. (2014) found that these promotional products during a FRP influence product level sales by a positive effect on the number of buyers and by a small, positive effect on mean purchase volume.
Besides, a positive effect of supplier funded products on spending during the FRP can be expected. Customers’ could purchase these promotional products during the program period in order to reach their redemption goal and to enhance progress towards obtaining rewards. Based on the aforementioned arguments, the hypotheses will be defined as follows:
H4a: Supplier funded products will positively influence customer spending.
H4b: The FRP will positively influence the effect of supplier funded products on customer spending.
2.2.5 Rewarded Behaviour Mechanism
16
Put differently, receiving a reward produces affect, that can increase purchase intentions and translates into behaviour. If a FRP adequately rewards loyalty, meaning that the utilities are higher than the costs, customers’ repeat buying behaviour should continue. Therefore, the rewarded behaviour mechanism can cause the increase in spending level to endure after the customer receives a reward. The result is a movement of their spending levels above baseline. Mean spending level, store purchase frequencies and share of wallet should all increase as well (Meyer-Waarden and Benavent 2009). Research found evidence for the positive rewarded behaviour effect on spending level in the retailing sector (Kopelle et al. 2007; Lal and Bell 2003; Taylor and Neslin 2005).
However, the effect of this rewarded behaviour can be temporary if the change in customer behaviour is not rewarded enough or if customers views the FRP as a promotional mechanism (Meyer-Waarden and Benavent 2009). The principal motivation for customer’s loyalty is the reward and if customers do not receive sufficient rewards, there is no rewarded behaviour effect. Moreover, if this occurs, the small changes in repeat purchase can drop back to baseline level 6 months after the customer redeemed during the FRP (Taylor and Neslin 2005). Based on the aforementioned arguments, the hypothesis will be defined as follows:
H5: In post-redemption weeks, the rewarded behaviour effect will positively influence customer spending.
2.2.6 Moderating Effect of Pre-Program Spending Level
One of the major drivers of the effectiveness of a short term FRP is the before mentioned points pressure effect. This effect has to be high enough in order for customers to increase their spending level. Therefore, this research will analyse how the effects of FRP differ between various customer groups, depending on their pre-program spending level. Mainly since the extent to which customers increase their spending level depends on their initial level (Liu 2007).
When customer evaluate the attractiveness of a FRP, there are two main components that customers are likely to consider, which are the required effort and the rewards that can be earned (Kivetz and Simonson 2003). The same authors imply that rewards that can be collected during a FRP are conditional on reaching a certain requirement level, which is the required amount of points. Perceived program effort can be defined as any inconvenience attached in meeting the requirements of the program, for example purchasing more than the customer normally buys. According to the same study by Kivetz and Simonson (2003), heavy and moderate buyers have an effort advantage over light buyers. Customers belonging to these segments do not need to work as hard or wait so long for the rewards as low spenders. Therefore, the latter segment might not consider the FRP as relevant but the effort advantage can increase the perceived fit and attractiveness of the program for heavy and moderate buyers.
17
reward literature, which suggest that when rewards do not offer sufficient challenge to the performance of a task, they lack the effect on customers’ motivation to lead to a change in behaviour (Eisenberger and Rhoades 2001). Low to moderate spenders are found to gradually increase their spending level and also are found to become more loyal to the retailer (Lal and Bell 2003; Liu 2007). Several studies also found the rewarded behaviour effect to be strongest among customer with a lower sales baseline (Lal and Bell 2003; Taylor and Neslin 2005). This can be due to the fact that a low speed in progressing implies extra effort is needed to attain goals, which concerns individuals with a lower initial progress level and with uncertainty regarding goal achievement (Huang and Zhang 2011). The motivation of further goal pursuit of these individuals was found to increase, when one made sufficient progress and when goal achievement was sufficiently secured. In conclusion, it can be argued that the stretch in spending level, accommodated by the points pressure effect, of low and moderate spenders will be larger compared to heavy spenders. Based on the aforementioned arguments, the hypothesis will be defined as follows:
H6: Spending level of low and moderate spenders will increase faster during the FRP, compared to heavy spenders.
2.3 Redemption
Participation in a FRP is influenced by various design elements, including issues of understanding, perceived fairness, speed of accumulation and the possibility of obtaining rewards (Nunes and Drèze 2006a). Low understanding of points accumulation and redemption might result in customer frustration. Frustration can also be a result of barriers to redemption for example a low value of the reward, which in turn impact motivation and behaviour (Stauss, Schmidt, and Schoeler 2005).
Assessing reward redemption behaviour in FRPs is important for a number of reasons. Firstly, retailers spend considerable money and effort on developing FRPs (Smith and Sparks 2009b). Furthermore, from a management perspective, redemption rates measure both the success and failure of customers’ engagement with the retailer and the resulting loyalty activities. Lastly, for customers, redemption is the most tangible aspect of their participation in the FRP, since the benefits of their participation becomes most salient to them during redemption. Redeeming accumulated points might result in gratification and satisfaction from the perspective of the customer. The collected reward might positively reinforces customers’ purchase behaviour, which stimulates them to revisit the retailer (Sheth and Parvatiyar 1995). This in turn might create loyalty tothe retailer, caused through the development of involvement and attachment, causing repeat purchase behaviour.
18
2.3.1 Weekly Spending Level
Customers’ cumulative purchase volume must be sufficient to reach the threshold for redemption. Weekly spending could therefore be an important influencer on customers’ likelihood to redeem during the FRP. Mainly since the weekly spending level of a customer determines how many progress a customer has made in saving points. A customer decide to redeem during the FRP, and thereby to start saving points, if the expected benefits are larger than the expected costs. The program is evaluated by weighing the value of a reward against the perceived effort to obtain that reward (Leenheer et al. 2007; Soman 1998). For low spenders, redeeming during a FRP will be less attractive, as they need to work harder and exert more effort to qualify for a reward. These customers can however increase their spending level, in accordance with the points pressure effect (Taylor and Neslin 2005). The fact that saving points for low spenders during a FRP to redeem is less attractive has to do with the so-called idiosyncratic fit. This is a theory that states when customers perceive to have an effort advantage over others, they will feel more attracted towards the FRP (Kivetz and Simonson 2003). Effort advantage can be described as the easiness for a particular customer to meet the program requirements to collect rewards, compared to others. The idiosyncratic fit is considered to be an driving factor of redemption, as it indicates an attractive opportunity for a customer who perceives to have an effort advantage over others. Therefore, spending level is expected to influence redemption, as customers who purchase frequently and thus have higher spending levels, could benefit more from the FRP. Besides, the FRP might change the shopping behaviour of some customers, even if a customer is already a loyal buyer and rewarded for his or her purchases (Meyer-Waarden 2007). Based on the aforementioned arguments, the hypothesis will be defined as follows:
H7: Customers weekly spending level positively influences the likelihood of redemption.
2.3.2 Saved Rewards
Customers could be attracted towards redeeming in a FRP based on the economic reason of obtaining value in the form of a reward. Customers value rewards, in an absolute sense and relative to others (Feinberg, Krishna, and Zhan 2002). When being provided with better value than others, this creates a feeling of being a special customer (Leenheer et al. 2007). Furthermore, customers tend to maximize the offered value by the FRP, instead of the final outcome. This reflect that customers generally aim to maximize saving points, resulting in an overvaluation of the collected rewards (Van Osselaer, Alba, and Manchanda 2004)
.
A FRP creates a certain extent of calculative commitment in customers’ relation with the retailer, as a result of the before mentioned switching costs if a customer stop purchasing at the retailer. Calculative commitment is defined here as the perceived need of a customer to maintain a relationship, given the switching costs that are associated with leaving (Leenheer et al. 2007). From these studies can be concluded that customers who have saved a certain number of rewards, generally continue saving points as a result of switching costs in the sense of forgoing points (Patterson and Smith 2001). Or in the sense of a missed feeling of being preferred when receiving a reward, reflecting switching costs as a loss of sense of belonging (Dowling and Uncles 1997).
19
successful goal attainment (Drèze and Nunes 2011). Though, successfully reaching a goal only influences subsequent goal success, if the attained goal was challenging enough. In the context of a FRP, success in terms of obtained rewards might influence subsequent redemption successes. Based on the findings of Drèze and Nunes (2011), the success of reaching a reward goal is expected to influence effort in subsequent goal attempts. According to the same study, attaining a goal depends on the number of rewards and the spend requirement for collecting a reward. After goal attainment, people reconsider their likelihood of success, which in turn influences their effort in reaching the same goal again. Moreover the received rewards influences the level of customers’ satisfaction with the FRP (Lara and de Madariaga 2007). Hence, the rewards are considered to greatly determine the success of an implemented FRP. Based on the aforementioned arguments, the hypothesis will be defined as follows:
H8: Customers cumulative saved rewards positively influences the likelihood of redemption.
2.3.3 Type of Spender
FRP are designed to target the most valuable customers of a retailer, therefore the program serves as a tool to reward existing purchase patterns (Mägi 2003). However, being a loyal frequent shopper does not necessarily imply participation in a FRP (García-Gómez et al. 2012b). Research has found that FRPs with cumulative rewards attract heavy spenders first, since these customers purchase enough to gain fast and substantial benefits, compared to the costs of redeeming during the FRP (Demoulin and Zidda 2009; Meyer-Waarden and Benavent 2009). Later adopters of the program, which are customers that redeem later during the program period, display lower spending levels since they expect lower program utilities and higher costs.
The research by Taylor (2001) found that customers’ prior purchase behaviour, and therefore spending level, is a good indicator for their likelihood of redemption. Early adopters of a FRP are found to visit the store more often and spend more at each visit, than do late adopters (Demoulin and Zidda 2009). These customers adopt earlier, as they are more likely to meet the opportunity to redeem, since frequent shoppers with higher spend levels will reach the threshold to collect rewards earlier, in contrast to the lower spenders. In conclusion, it will be expected that being a heavy or moderate spender, in contrast to a low spender, will influence the probability of redemption. The type of spender will be classified based on customers’ spending level before the start of the program, in order to assess customers’ prior spending behaviour. Based on the aforementioned arguments, the hypotheses will be defined as follows:
H9a: Moderate spenders are more likely to redeem during the FRP, compared to low spenders. H9b: Heavy spenders are more likely to redeem during the FRP, compared to low spenders.
2.3.4 Supplier Funded Products
20
This phenomenon is called the endowed progress effect. In the setting of a FRP, the same study found that this effect increases the likelihood of a customer redeeming during the program. Besides, the overall attractiveness of the program increased. Customers are more likely to meet the program requirement due to the advancement, as they were found to increase effort as they approach their goal (Nunes and Drèze 2006b). This is in line with the goal-gradient hypothesis. By purchasing supplier funded products, customers’ motivation to obtain rewards can increase, since customers can reach the redemption threshold sooner. This in turn reduces the time a customer have to wait for the reward. Based on the aforementioned arguments, the hypothesis will be defined as follows:
H10: Supplier funded products positively influences the likelihood of redemption.
2.3.5 Week
The decision of a customer to redeem might depend on the number of weeks since the start of the program. Firstly, redemption events are likely to occur after a couple of weeks since the program has started, as customers will have saved up enough points to redeem. Secondly, due to forward-looking, a customer who is collecting points might feel pressured to reach the threshold to redeem. Also, the benefits of collecting rewards may grow with the time horizon of the program. On the other hand, if the time to reward redemption becomes too long, the possibility exist that customers will lose interest in the FRP or even refuse to join (Krafft and Mantrala 2006). Since this research analyses the effects of a program with a short-time span, it is expected that the number of weeks of the FRP will not negatively influence the probability of a customer to redeem. Hence, the program week since the start of the FRP might influence a customers’ decision to redeem. If a program exist for a longer accumulation period, customers will be locked in the program by their build up assets, which consist of saved points in the setting of a FRP. Furthermore, the amount of saved points function as switching costs (Taylor and Neslin 2005). Therefore, the probability of redemption during a FRP can depend on the week since the start of the program. Based on the aforementioned arguments, the hypothesis will be defined as follows:
21
Figure 2 Conceptual Model
2.4 Conceptual Model
22
3. RESEARCH DESIGN
3.1 Study Setting
This empirical research analyses the impacts of a FRP using data provided by a grocery retailer in Brazil. The retailer is an example of a Cash & Carry, which targets professional customers rather than end-customers, and the concept itself is based around self-service and bulk buying (Wikilab Brandloyalty 2014). The retailer implemented a FRP, with the aim to achieve turnover growth. The clientele of the grocery stores are mainly businesses owners, but the general public are also allowed at the supermarkets. However, all customers are registered members as customers cannot enter the store if they do not have a personal identification card. The retail grocery sector is an interesting market to study the effects of a FRP, mainly because customers make frequent transactions in supermarkets and often visit several supermarkets within a short- time period.
Longitudinal data was gathered via personal identification cards in about 80 different stores in Brazil, indicating that data is measured over time for the same cross section units. The retailer offered a FRP, where the customers had the opportunity to participate in the program by collecting points for every 8,25 euro spent. According to BrandLoyalty, the most attractive group for Cash & Carry stores are the customers belonging to the middle group in terms of spending. This group generates the most turnover and was therefore the focus group of this FRP, as the largest stretch in terms of spending could be gained here. Customers also had the opportunity to participate in the FRP by redeeming rewards with an additional payment. These customers could redeem with a lower amount of points in exchange for a certain payment. The accumulated points could be redeemed after reaching the spend requirement of 8,25 euros. As was mentioned, customers could save for a complete knife set and at various thresholds, a different item of the set could be redeemed. The total set existed of ten rewards. The reward program included a two weeks clean-up period, were the stores did not hand out points anymore, but customers where given the opportunity to redeem their saved points.
3.1.2 Data Purification
23
SFP Weekly Spending Saved Rewards 0 20 40 0 20 40 0 2000 4000 0 2000 4000 0 20 40 60 0 20 40 60Figure 3 Checking for outliers
Before estimating the model, the sample is checked on outliers. Since the cumulative saved number of rewards, the number of purchased supplier funded products and customer spending are continuous variables, a scatterplot matrix of these variables is made to graphically check for outliers. The presence of outliers is pertinent in panel data, since large panels of customers are likely to contain irregular observations (Bramati and Croux 2007). The presence of outliers can strongly bias panel data estimators, however, according to the same study, visual inspection of panel data is less obvious than for cross-sectional data. In figure 3, can be seen that there are a few outliers. By removing these outliers from the sample, valuable information will be lost. Therefore, the robust random effects model will be used for estimation to cope with the outlying observations.
3.2 Sample Description
Daily transactional data of five months was obtained, with a total of 114.377.796 data points. A random sample was made of 1000 customers, based on their unique customer IDs. Through merging transactional data of each month with Customer ID as key variable, it was ensured that all transactional behaviour of those 1000 customers was gathered. The sample is made of the transactional data of the month July, which is the month before the start of the FRP. As was discussed in the theoretical framework, a FRP attracts more customers to the store, including customers which are not considered to be loyal customers of the retailer. Since one aim of this empirical research is to track the effect of the FRP on existing customer behaviour, the sample of customer IDs is made of the month July.The daily transactional data was aggregated to weekly transactional data. The data covered customer purchases during eight 8 weeks before the FRP and 12 weeks during the FRP.
24
0 100 200 300 400 500 600 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 FR EQ UE N CY WEEKFREQUENCY OF VISITS
0 5.000 10.000 15.000 20.000 25.000 30.000 35.000 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 EURO WEEKWEEKLY SPENDING
Start FRP Start FRPFigure 4 Development Weekly Spending Figure 5 Development Frequency of Visits
two months before the start of the program was €22.462. It can be seen that mean customer spending increased during the FRP from €85.34 to €106.20. As can be seen, the mean sales of supplier funded products did not increased during period of the FRP. In figure 12 in Appendix A, the development of the sales of supplier funded products can be seen.
Variables Minimum Maximum Standard deviation Mean
Pre FRP Weekly Spending €0 €2365.33 €119.76 €70.5
Pre FRP Sales SFP €1557.44 €3643.04 €714.60 €2531.43
Pre FRP Weekly Turnover Pre FRP Frequency of Visits
€14864.31 1 €27923.65 6 €4852.9 0.74 €22.462 1.31
During FRP Weekly Spending €0 €3277 €154.3 €88
During FRP Sales SFP €1707.93 €3207 €495.86 €2464.51
During FRP Weekly Turnover €17211.77 €30313.17 €3494.09 €22.792
During FRP Frequency of Visits 1 7 0.81 1.39
Table 1 Descriptive Statistics
25
3.3 Research Method
Leeflang et al. (2000) define a model as a representation of the observed elements of a real world system that are most important. According to these authors models are also often defined as simplified pictures and the use of models enhance decision makers’ understanding of the systematic parts that underlie reality. The use of models became important as the degree of the production and reports of marketing decision modelling knowledge in the academic literature has tripled in the past 25 years (Lilien 2011). The same study by Lilien (2011) also argue that models improve the consistency in the decision making by managers, and consistency is desirable. Models can be classified according to their purpose or intended use and reflects the reason why a manager wants to use a model. A distinction can be made between descriptive, predictive and normative models (Leeflang et al. 2014). Descriptive models are used to describe decisions and other processes, whereas the purpose of a predictive model is to forecast future events. Lastly, a normative model leads to a recommended course of action. Hence, for this research a descriptive model will be specified in order to describe the effects of an implemented FRP for the retailer as well for the designer of the program.
In order to assess the effects of the FRP on customer spending and redemption, several models were specified by using longitudinal panel data. The data was measured over multiple time periods for the same customers, therefore in this research, panel data models were used. Panel data is useful for several reasons. It allows to look at dynamic relationships, which is not possible with a single cross section. Furthermore, it allows to control for unobserved individual heterogeneity (Woolridge 2010). Panel data studies not controlling for such heterogeneity run the risk of getting biased results (Baltagi 2013). According to the same author, longitudinal data is also more informative and more efficient. Besides, this kind of data has more variability and less multicollinearity among the included variables (Baltagi 2013). One model that can be specified for panel data is the fixed effects model, which is a linear panel data model. The fixed effects model is a linear regression model in which the intercept term vary over the individual units, meaning that an individual-specific intercept terms is included, which is fixed through time (Verbeek 2012). This fixed intercept captures all (un)observable time-invariant differences across different customers and is allowed to be correlated with one or more explanatory variables. When t represents different time periods for the same individuals, the unobserved effect can be interpreted as capturing features of an individual, for example motivation (Woolridge 2010).
Besides the fixed effects model, a model can be specified which contain time-invariant, unobserved effects in the error term. These type of panel data model are called random effects models, where the unobserved heterogeneity, 𝜇𝑖 is uncorrelated with the explanatory variables 𝑋𝑖𝑡.Therefore time-invariant
26
4. CUSTOMER SPENDING
The first part of the framework contained research questions about the impact of the FRP on customer spending, and subsequently on retail turnover. Therefore, the first part of the model specification will analyse the effects of the FRP on customer spending
.
4.1.Pooled Model
Breusch and Pagan devised a Lagrange multiplier test to test for the random effects model, based on the OLS residuals (Greene, 2012). In the ordinary regression model, the intercept is the same for all individuals. Under the null hypothesis, the variances across individuals
𝑢
𝑖 is zero, indicating there is no significant effect across unitsand therefore no panel effect: H0 : 𝜎𝑢2= 0
H1 : 𝜎𝑢2≠ 0
In table 7 in Appendix A can be seen that the p-value was significant (.000), meaning that there is significant variance in
𝑢
𝑖j. Therefore, the pooled cross-sectional model was not used since the test indicated that therandom effects model was strongly preferred.
4.2 Model Specification
The Hausman test was be used to analyse whether the unique errors are correlated with the explanatory variables. The fixed effects model is consistent when the unobserved individual specific effects
𝑢
𝑖are correlatedwith Xit, whereas random effects model is inconsistent. Under the null hypothesis, they are not correlated which
indicates to use a random effects model. Therefore, a statistically significant difference is evidence against the use of the random effects model. The test statistic is:
𝐻 = (𝑏
𝐹𝐸− 𝛽̂
𝑅𝐸[𝑉
𝐹𝐸− 𝑉
𝑅𝐸] − 1(𝑏
𝐹𝐸− 𝛽̂
𝑅𝐸)
Table 8 in the Appendix A show the results of the Hausman test. The value of the test statistic was -6799.89 which confirms the shortcoming of this test, as nothing prevents the statistic from being negative. Mainly because there is no guarantee that the difference of the two covariance matrices will be positive definite. According to Greene (2012), it can be concluded that the random effects model is not rejected because the similarity of the covariance matrices causes the negative statistic. Therefore, the random effects model was preferred. The model specification is as follows:
𝐶𝑆
𝑖𝑡= 𝛽
0+ 𝛽
1𝐹𝑅𝑃
𝑡+ 𝛽
2𝑆𝐼𝑀
𝑡+ 𝛽
3𝐺𝐺
𝑖𝑡+ 𝛽
4𝑆𝐹𝑃
𝑖𝑡+ 𝛽
5𝑆𝑅
𝑖𝑡−1+ 𝛽
6𝐿𝑆
𝑖+ 𝛽
7𝐻𝑆
𝑖+
27
i = 1,.., 1000 customers.t = 1,..,20 weeks.
CSit = Weekly customer spending in euros for individual i in week t.
FRPt = Indicator variable for the FRP period in week t: 1 if the observation is in the program period, 0 otherwise.
SIMt = Indicator variable for the stuck-in-the-middle effect in week t: 1 if the observation is during week 15 and
16; the middle of the program, 0 otherwise.
GGit = Indicator variable for the goal-gradient effect for customer i in week t: 1 if the observation is in the week
before redemption, 0 otherwise.
SFPit = Number of purchased supplier funded products of customer i in week t.
SRit-1 = Lagged cumulative number of saved rewards of customer i in week t.
LSi
=
Indicator variable for customer i belonging to the low spender segmentHSi
=
Indicator variable for customer i belonging to the heavy spender segmentHSiFRPt = Interaction effect between heavy spenders and the FRP for customer i in week t.
LSiFRPt = Interaction effect between low spenders and the FRP for customer i in week t.
SFPitFRPt = Interaction effect between supplier funded products and the FRP for customer i in week t.
𝑢
𝑖 = Random effect for customer i𝜀
it = Random error for customer i in week t.4.2.1 Explanatory Variables
In order to assess the points pressure effect on customer spending during the FRP, an indicator variable was included to indicate whether the observation was measured during the program period. Therefore, the β1 parameter show the main points pressure effect. Parameter β2 indicate the impact of the stuck-in-the-middle effect halfway the program, which was measured by an indicator variable if the observation was during week 15 and 16. The goal-gradient effect was measured by the impact on spending as a customer approaches their redemption goal in the week before they collect the reward. This impact is measured by parameter β3. SFP measured the number of purchased supplier funded products bought by customer i in week t, which is captured by parameter β4. Customers’ saved number of rewards were lagged to assess the effect of redemption on spending in the subsequent week, this indicated the presence of rewarded behaviour in the week after the customer redeemed. Therefore, parameter β5 reflect the rewarded behaviour effect.
The dummy variable indicating if a customer is a low of heavy spender (β6 and β7), was multiplied with the FRP in order to detect differences in spending between low and heavy spenders during the program compared to moderate spenders. These effects are shown by parameter β8 and β9. In order to determine the effect of supplier during the program period, an interaction term was included by multiplying SFP with the FRP (SFPitFRPt). This impact is captured by parameter β10.
𝑢
𝑖controls for unobserved individual specific differences,28
4.3 Model Assumptions
Several assumptions about the error term needed to be satisfied in order to obtain reliable estimates, as violation of one of the assumptions can lead to wrong parameter estimates and wrong estimation of the variance of the parameters (Leeflang et al. 2014). It is wrong to assume that the disturbance term satisfy the assumptions (Leeflang et al. 2000), therefore the following assumptions were tested for:
(1) Normally distributed error term; (2) Uncorrelated disturbances; (3) Homoscedastic error term; (4) No perfect multicollinearity exist
4.3.1 Normality Test
The assumption of a normally distributed disturbances must be tested for, which might be violated due to misspecification (Leeflang et al. 2014). In many research situations, the distribution deviates from normality (Paul and Zhang 2010). The validity of this assumption can be examined indirectly by the residuals. The normality assumption was first graphically assessed by plotting a histogram, which can be seen in figure 13 in Appendix A. The histogram gives a first indication that the residuals are normally distributed, as can be seen from the bell shaped figure. However, the normality probability plot of the residuals, shown in figure 14 in Appendix A, show the residuals are not nicely distributed and deviate from the straight normality line. Therefore, there was evidence against the assumption that the residuals are normally distributed, which could produce biased standard errors.
In order to statistically test for non-normality, the Shapiro – Wilk test was performed. The residuals seems to be non-normal distributed, since the significance level of the test showed to be lower than .05. Though, it is still reasonable to assume normality as given by the large sample size. Leeflang et al. (2014) point out it becomes easier to reject the null hypothesis of normal distributed errors with increased sample sizes. Minor deviations from a normal distribution can indicate a violation of the assumption. However, as the Skewness/Kurtosis tests for normality also indicated a non-normal distribution, the dependent variable was log transformed in order to normalize the residuals 1. Spending data tends to violate the normality assumption and
estimating a lognormal model is therefore appropriate (Cameron and Trivendi 2010). The interpretation of the log transformed dependent variable is also more intuitive, as the parameter estimates can be interpreted as an percentage increase or decrease of the outcome variable. The normality plots of the residuals after the log transformation can be seen in figure 6, which show approximately normally distributed residuals.
1 Log transforming the dependent variable was an appropriate remedy for non-normality, as the square root
29
0. 00 0. 25 0. 50 0. 75 1. 00 N o rm al F [( res R Er o bus t-m )/ s] 0.00 0.25 0.50 0.75 1.00Empirical P[i] = i/(N+1)
0 20 0 40 0 60 0 80 0 Fr equenc y -6 -4 -2 0 2 4 u[CustomerID] + e[CustomerID,t]
Figure 6 Checking for the assumption of normality after log transformation
Before testing the other model assumptions, the Hausman test is performed again in order to verify if the random effects model is still the most efficient model to use after log transforming the dependent variable. The results can be seen in table 9 in Appendix A. The test revealed a significant test statistic, indicating the fixed effects model is more appropriate. This is an unfortunate outcome as some explanatory variables are time-invariant, which will not be estimated when using the fixed effects model. According to Baltagi (2013), the RE specification is appropriate if N individuals are randomly drawn from a larger population. Based on these arguments, it was decided to continue using the RE model. After log-transforming, the model will be specified as follows:
𝑙𝑛𝐶𝑆
𝑖𝑡= 𝛽
0+ 𝛽
1𝐹𝑅𝑃
𝑡+ 𝛽
2𝑆𝐼𝑀
𝑡+ 𝛽
3𝐺𝐺
𝑖𝑡+ 𝛽
4𝑆𝐹𝑃
𝑖𝑡+ 𝛽
5𝑆𝑅
𝑖𝑡−1+ 𝛽
6𝐿𝑆
𝑖+ 𝛽
7𝐻𝑆
𝑖+
𝛽
8𝐻𝑆
𝑖𝐹𝑅𝑃
𝑡+ 𝛽
9𝐿𝑆
𝑖𝐹𝑅𝑃
𝑡+ 𝛽
10𝑆𝐹𝑃
𝑖𝑡𝐹𝑅𝑃
𝑡+ 𝑢
𝑖+ 𝜀
𝑖𝑡(2)
4.3.2 Autoregression Test
Autoregression implies a dependence in the disturbance term and can result from a model misspecification (Leeflang et al. 2014). This misspecification might be caused by omitted variables and any latent effects that are left out will carry across all the time periods. The other included variables will capture the effect of the omitted variable, which leads to biased estimates. Serial correlation across the observations in the groups of customers in panel data could exist, as there might be a correlation between the disturbances of the same individual. Since in panel data, a group of observations can belong to the same individual (Greene 2012).
