The effect of short-term loyalty programs on (the relationship between) attitudinal and behavioral loyalty change in a grocery retail setting

(1)

The effect of short-term loyalty programs on (the relationship

between) attitudinal and behavioral loyalty change in a

grocery retail setting

(2)

2

The effect of short-term loyalty programs on (the relationship

between) attitudinal and behavioral loyalty change in a

grocery retail setting

January, 2016

Master Thesis

MSc Marketing Intelligence University of Groningen

Faculty of Economics and Business Department of Marketing

PO Box 800, 9700 AV Groningen

Lars Jellinek S1770322

Van Vollenhovenlaan 107

3527 JB Utrecht, The Netherlands 06 24824588

l.jellinek@student.rug.nl

(3)

3

Management summary

With growing competition and increasing numbers of substitutes, achieving customer loyalty is of huge importance within the grocery retail market. One of the marketing tools widely used all over the world are short-term loyalty programs. Retailers invest lots of money into the implementation of these programs, in order to increase loyalty among their customers.

Customer loyalty refers to the strength of the relationship between customers and a retailer. It is composed of customer’s behavior on the one hand and customer’s attitudes on the other hand (Day, 1969; Dick & Basu, 1994; Uncles et al., 2003). When both dimensions are high, customers are ‘truly loyal’ towards a firm. This study examines the effect of a short-term loyalty program on both behavioral and attitudinal loyalty, as well as the relationship between these two.

Making use of a pre-program and post-program survey regarding an short-term loyalty program implement in Japan, the change in behavioral and attitudinal loyalty is captured and the effect of participation on this change was tested. Furthermore, the relationship between attitudinal and behavioral loyalty change in the context of a short-term loyalty program is addressed making use of (multiple moderated) mediation models.

Findings of this study suggest that participation in the program under study resulted in an increase in basket size, but share-of-wallet and visit frequency did not change as a result of program participation. Furthermore, no main effect was found on change in attitudinal loyalty, suggesting that short-term loyalty programs are not efficient in terms of increasing customer’s attitudes towards the retailer in general.

The effect of program participation on basket size is found to be dependent on both the number of competitors a customer visited before the program started and the perceived reward attractiveness of the customer. Under the condition that customers visit a low number of competitors, basket size significantly increased. The same yields for a high level of reward attractiveness. Under the condition that reward attractiveness was perceived as high, the effect of participation became stronger for higher levels of perceived program achievability. From these results, retailers can induce that the success of short-term loyalty programs is dependent on the level of perceived reward attractiveness. Furthermore, it seems to be the case that a short-term loyalty program generates effect on behavior under customers that are typically in contact with low degrees of competition. Lastly, short-term loyalty programs only enhance an increase in attitudes towards the retailer for households with high levels of income.

(4)

4

Preface

Exploring the world of my real interest. With that in mind I decided to start a new adventure after finishing my BSc Communication and Information Science and step into the world of marketing, with this study as the end result. During these last two years of pre-MSc Marketing and MSc Marketing Intelligence, I found my passion in data analytics and can truly say that I will never regret the switch towards the inspiring field of marketing intelligence. I am grateful towards BrandLoyalty, and in particular Ms. A. ten Brink, for giving me the opportunity to spend the final period of my studies at its dynamic and instructive working environment.

Furthermore, I really appreciate the help of my academic supervisor Prof. Dr. T.H.A. Bijmolt for providing valuable feedback during the process of writing this study and I thank second supervisor A. Minnema MSc for his interest in and feedback on this study. Besides, I would like to thank the whole customer intelligence team of BrandLoyalty, and in particular supervisor Ms. K. Heeren, for their feedback and willingness to help all along the road. I am grateful for the support of my parents during my entire period as a student. Giving me the opportunity to finish this MSc, while always believing in my capabilities, really means a lot to me. Lastly, I thank my girlfriend Esthelle and my sister Manon, for their interest in this project throughout the whole process and their unlimited support and positivity.

(5)

5

1. Introduction

Within markets nowadays, where purchase options for consumers grow every single day, achieving customer loyalty is very important. Possibilities for customers to purchase online become partial or even complete substitutes for the offline providers. This development of growing competition and substitutes can also be found within the grocery retail market. With the introduction of Instacart for instance, consumers order groceries via a mobile application and get it delivered at home within an hour. Thus, new formulas - online as well as offline – appear in high speed. In order to survive in such a competitive environment, it is very important to be able to rely on a firm customer base with loyal customers. In line with the central thought that preserving existing customers is less expensive - and thus more beneficial - than bringing in new customers, Gupta et al. (2004) state that focusing on customer retention is more beneficial to the value of a firm, than increasing profit margins or lower acquisition costs for example.

Main benefits of customer loyalty are its positive impact on repeat purchases, an increase in purchase amount and higher share-of-wallet of customers (Meyer-Waarden 2007; Ashley et al., 2011). Customer loyalty can be defined as the combination of behavioral and attitudinal loyalty (Day, 1969; Dick & Basu, 1994; Bijmolt et al., 2011). Bijmolt et al. (2011) define behavioral loyalty as certain purchase actions, like the frequency and volume of buying. Attitudinal loyalty on the other hand, refers to perceptions that customers have regarding a certain product, service or firm (Furinto et al. 2009). Bijmolt et al. (2011) suggest that behavioral loyalty relates more to short-term purchasing patterns, whereas attitudinal loyalty creates favorable attitudes and a true long-term affect. Therefore, the importance of attitudinal loyalty cannot be underestimated.

(8)

8 months. Within this time, customers can earn (a) tangible reward(s) if they repeatedly purchase the company’s products (Lal & Bell, 2003; Taylor & Neslin, 2005).

This research will focus on the effect of short-term loyalty programs. Earlier studies found a positive effect of SLPs on purchase behavior, mainly due to increasing purchases by moderate and light users and new customers (Dréze & Hoch, 1998: Lal & Bell, 2003; Taylor & Neslin; 2005). While these studies found positive effects on behavioral loyalty, little is known about the potential effect of SLPs on attitudinal loyalty and how this is relating towards purchase behavior. Analyzing the interplay between behavioral and attitudinal loyalty in an SLP-context is important, in order to be able to understand and influence the effectiveness of SLPs. Looking at the underlying construct of attitudinal loyalty and its link towards actual behavior, should provide valuable insights in terms of influencing consumers towards a higher loyalty level with the use of SLPs.

Based on a program at a Japanese retailer, which ran between June 2nd 2015 and September 18th 2015, survey data is gathered from a representative sample of 591 Japanese households. As the setup consists of two parts, including a pre- and post-program survey, the total dataset consists of pre- and post-program measurements for all respondents. This data is used to analyze the scope of this study: the relationship between attitudinal and behavioral loyalty in an SLP-setting.

This study is done on behalf of BrandLoyalty, a company that implements and manages frequency reward programs within the grocery retail industry all over the world. BrandLoyalty is in the FRP-business for over twenty years and currently serving retailers in more than thirty countries (BrandLoyalty, 2015a). The firm uses a full service approach and aims for generating immediate changes in consumer behavior on the one hand and on the other hand increase retailer brand equity (BrandLoyalty, 2015a). Their programs all run for a certain time span (short-term) in which the customer can redeem rewards. Getting a better understanding on the relationship between attitudinal and behavioral loyalty in this SLP-context is highly valuable for BrandLoyalty in order to increase the success of their programs and especially in terms of increasing retailer brand equity.

1.1 Research questions

(9)

9 SLPs on attitudinal and behavioral loyalty. In order to address this issue, the main purpose of this research is to examine the interrelationship between behavioral loyalty and attitudinal loyalty in an SLP-context. The main research question addressed in this research then becomes:

What is the effect of consumer’s participation in SLPs on both behavioral and attitudinal loyalty and how do these two constructs relate?

In order to provide a clear answer to this question, it is necessary to formulate and answer a number of sub questions. First, the constructs of behavioral and attitudinal loyalty have to be defined and explained. Besides, it is important to identify how these constructs can be measured. Previous literature is extensively available on both types of loyalty, regarding both definitions and measurements. Therefore, the following questions will be answered in the theoretical framework:

1) How can behavioral loyalty be defined and measured? 2) How can attitudinal loyalty be defined and measured?

As the purpose of this study is to find the effect of SLP-participation on customer loyalty and the relationship between attitudinal and behavioral loyalty, a participation variable has to be defined as well. There are different ways of defining participants. Every person that is collecting can be included as a participant, irrespective of redemption. However, participants can also be defined as the ones that redeemed at least one product during the SLP. In order to decide which definition is most usable, it is important to understand the difference between both definitions, as well as the possible difference in effect it may have on customer loyalty. This leads to the following question:

3) How can SLP-participation be defined most accurate in this study?

(10)

10 4) What is the effect of SLP-participation on both attitudinal loyalty and behavioral

loyalty?

5) Is behavioral loyalty change affected by attitudinal loyalty change in an SLP-environment?

Also, multiple factors could strengthen or weaken the effect of SLP-participation on behavioral loyalty and attitudinal loyalty. These factors can for instance be customer-related or program-related. To make the final model as complete as possible, it is important to test for supposable moderation effects as well. Therefore, different factors that could influence the effect of SLPs on customer loyalty will be identified in the theoretical framework:

6) What customer- and program-related factors moderate the effect of SLP-participation on customer loyalty?

In the end, when variables that could affect the SLP-effectiveness in terms of customer loyalty are identified and the total model is tested, it should be possible to describe the effect of SLPs on both attitudinal and behavioral loyalty. Furthermore, the possible interplay between consumer’s attitudinal and behavioral loyalty in an SLP-context could be demonstrated. Because of the inclusion of customer- and program-related moderators, the outcomes possibly provide information on how change in customer loyalty can actually be created using SLPs, as well as what customer segment(s) can actually be triggered into customer loyalty change. Throughout the whole study, attitudinal loyalty and behavioral loyalty are operationalized as the change in these scores between before and after the program.

1.2 Relevance

(11)

11 existing academic literature by focusing on the unexplored interplay of attitudinal and behavioral loyalty for short-term loyalty programs specifically.

Besides the academic contribution, this study aims to provide useful insights for the loyalty program business in general and BrandLoyalty in specific. As BrandLoyalty is trying to add value to grocery retailers by providing tailor-made programs that are - besides boosting turnover - will increase retailer brand equity, it is essential to understand the relationship between customer’s attitudes and behavior and how this is affected by SLPs. As firms are really interested in creating repeat and more frequent purchases by tapping into consumers’ underlying psychological processes (Kumar & Shah, 2004; Nunes & Dréze, 2006), studying the effect of SLPs on the combination of behavioral and attitudinal loyalty capitalizes this. Gaining more knowledge on the underlying psychological dimensions consumers potentially develop towards retailers by means of SLPs should lead to higher ability to provide retailers with programs that actually increase customer’s attitudinal loyalty and therewith brand equity.

1.3 Outline

(12)

12

2. Theoretical framework

This chapter starts with a visual representation of the conceptual model. Next, the construct of customer loyalty will be discussed. Then, definitions for both behavioral and attitudinal loyalty (the two main dimensions of customer loyalty) are provided. Further, a description of short-term loyalty programs and the definition of participation are given, whereupon its assumed effect on customer loyalty will be outlined. This is followed by earlier findings on the relationship between behavioral and attitudinal loyalty. Lastly, possible moderators of the effect of SLP-participation on customer loyalty are identified, control variables are introduced and an overview of hypotheses is given.

2.1 Conceptual model

Figure 1: Conceptual model.

(13)

13 2.2 Customer loyalty

Customer loyalty is a concept that refers to the strength of the relationship between customers and a firm. Oliver (1999) defines customer loyalty as a situation in where a customer has preferable beliefs and attitudes, as well as a higher buying intention for a certain brand/at a certain company compared to competitive options. Specifically, it consists of two dimensions: behavioral loyalty and attitudinal loyalty (Day, 1969; Dick & Basu, 1994; Uncles et al., 2003). Where behavioral loyalty refers to purchase patterns of the customer (such as frequency, basket size and share of wallet), attitudinal loyalty captures the level of commitment, positivity of the attitude, and so on (Bijmolt et al., 2001). The presence of behavioral loyalty does not directly mean that a customer also is attitudinally loyal. In a situation where alternatives lack, customers could show behavioral loyalty through repurchases, without creating positive attitudes (Dick & Basu, 1994; Whyte, 2004). Therefore, Bijmolt et al. (2011) suggest that ‘true loyalty’, inducing sustainable effects on customer loyalty, will only be reached via a strong positive attitudinal attachment. In order to measure customer loyalty, both behavioral and attitudinal loyalty have to be (and will be) considered in this study.

2.3 Short-term loyalty program participation

The implementation of a loyalty program (LP) is a widely used tool in numerous markets nowadays. Its main purpose is to foster and reward customer loyalty (Bijmolt et al., 2011). Setups and designs of loyalty programs can vary among different aspects. A long-term loyalty program versus short-term loyalty programs is one of these distinctions in terms of LP-setup. Where a long-term loyalty program in essence is a continuous and ongoing investment, a short-term loyalty program (SLP) contains only a pre specified time span (Bijmolt et al., 2011). The main intuition of an SLP is rewarding consumers that spend a certain amount of money within a specified time frame with an economic incentive. These SLPs are a common used marketing tool in the retail industry. As consumers in this industry are characterized by having high purchase frequencies, variations in basket size and often shop regularly at different firms (Kahn & Schmittlein, 1992; Kahn & McAllister, 1997), this market is very suitable for influencing consumers making use of SLPs.

(14)

14 (Hirschman & Holbrook, 1982; Tauber, 1972). It is argued by Johnson (1999) that the customers are attracted by loyalty programs because they get pleasure out of collecting and redeeming. In line with these studies, it is assumed that collecting on itself is influential in changing customer loyalty. Therefore, the main predictor variable SLP-participation will be conceptualized as customers that collected during the SLP.

Bolton et al. (2000) find that LP-members in general perceive larger gains than losses on price and quality aspects. One could argue that if redemption takes place the feeling of gaining would be stronger. In that sense, redemption occurrence should not be disregarded, as it might lead to higher levels of change in both behavioral and attitudinal loyalty. Therefore, this study will define SLP-participation as all customers that collected during the program, while including redemption in the model estimation in order to test whether redemption is leading to bigger changes in behavioral in attitudinal loyalty.

2.4 Behavioral loyalty

One of the building blocks of customer loyalty is behavioral loyalty. This construct is focusing on the actual outcome of customer loyalty, as it reflects customer’s purchase behavior. More concrete, share-of-wallet, visit frequency and purchase volume are certain concepts that reflect behavioral loyalty (Bijmolt et al., 2011). Bijmolt et al. (2011) also notice that earlier studies used a number of different forms of operationalization for customer metrics, which makes comparisons between studies hard, if not impossible. In order to test effects of SLP-participation on behavioral loyalty, this study will address three constructs that measure behavioral loyalty: share-of wallet, visit frequency and total spend at retailer.

Because of the high purchase frequency and differentiation in regular firm visits in the retail industry, a highly suitable measure for behavioral loyalty is suggested to be share-of-wallet (Berger et al., 2002; Mägi, 2003). This study also follows that intuition and therefore the behavioral loyalty construct will be operationalized via share-of-wallet. Share-of-wallet reflects the percentage of expenditures within a certain category at a specific retailer (Leenheer et al., 2007).

(15)

15 programs. Bijmolt et al. (2011) found that SLPs (Dréze and Hoch, 1998; Taylor and Neslin, 2005) tend to be more effective in generating additional sales and thus in affecting behavioral loyalty of customers than continuous loyalty programs (Leenheer et al., 2007; Magi, 2003; Sharp and Sharp, 1997). Following this reasoning, it is plausible to assume that there will be a positive effect of SLP-participation on share-of-wallet.

H1a: SLP-participation has a positive effect on share-of-wallet change.

Most retailers use some combination of building traffic and increase purchase value in order to increase profits (Dréze & Hoch, 1998). Building traffic can be done by attracting new customers or increasing the visit frequency of existing customers. Purchase value can increase by customers purchasing more expensive alternatives or buy more products. In the end, the total effect on turnover for the retailer is visit frequency multiplied by average basket size. In line with Van Heerde & Bijmolt (2005), to specify the change in turnover attribution for participants and non-participants the effect on visit frequency and basket size will be tested.

By including both changes in visit frequency and basket size in the analysis, it can not only be identified if SLP-participation leads to a positive effect on turnover, but also whether this effect is coming from visit frequency, basket size, or a combination of those two. Dréze & Hoch (1998) found in their study that loyalty program implementation increased store traffic by 5%. The same study reported an even bigger positive effect (of 25%) regarding purchase value. This leads to the assumption that there will be positive effects of SLP-participation on both visit frequency and average basket size.

H1b: SLP-participation has a positive effect on visit frequency change. H1c: SLP-participation has a positive effect on average basket size change. 2.5 Attitudinal loyalty

(16)

16 Attitudinal loyalty is composed of satisfaction, trust and commitment (Bijmolt et al., 2011). Satisfaction can be defined by the favorability of attitudes and the positivity of affect customers have towards a certain brand or firm (Bijmolt et al. 2011), which is affecting the strength of preference over alternatives (Jones & Taylor, 2007). According to Morgan & Hunt (1994), in terms of customer loyalty, trust exists when a customer has confidence in the reliability and integrity of a firm/retailer. Trust is created when customers believe a firm has high integrity and reliability. These values are associated with consistency, honesty, competence, fairness, responsibility and helpfulness (Altman & Taylor, 1973). Lastly, Morgan & Hunt (1994) define commitment as “a believe that a certain relationship is worth working on to ensure that it endures.” In terms of customer loyalty, this can be translated into the situation where a customer wants to make an effort to maintain the customer-firm relationship (s)he has.

In essence, the total construct of satisfaction, trust and commitment is focusing on the consumer’s state of mind. Throughout this study, the level of attitudinal loyalty will be conceptualized via the combination of satisfaction, trust and commitment. It is preferred to analyze the total construct of attitudinal loyalty over the more detailed dimensions, as Crosby, Evans, & Cowles (1990) found that consumers find it difficult to make distinctions between them and tend to lump them together.

Based on the concept of reciprocity, if a retailer shows that it invests in its relationship with customers, this would lead to an increase in customer’s attitudinal loyalty (De Wulf et al., 2011). The principle of reciprocity states that, in proportion to what people receive, they return good for good (Bagozzi, 1995). The reciprocal action theory states that “actions taken by one party in an exchange relationship will be reciprocated in kind by the other party, because each party anticipates the feelings of guilt it would have if it violated the norm of reciprocity” (Li & Dant, 1997). Translated towards a consumer-retailer relationship, consumers show loyalty to certain retailers in reciprocation of the relationship investment of this retailer in combination with an obligated feeling to pay back the “kindness” of the retailer (Kang & Ridgway, 1996).

(17)

17 Two out of three countries showed positive effects of tangible rewards on perceived relationship investment, leading towards higher attitudinal loyalty (De Wulf et al., 2001). Therefore, it is assumed that SLP-participation affects attitudinal loyalty in a positive way.

H2: SLP-participation has a positive effect on attitudinal loyalty change.

2.6 Mediating effect of attitudinal loyalty on behavioral loyalty

De Wulf et al. (2001) found a significant positive relationship between attitudinal and behavioral loyalty across different countries and industries, suggesting that behavioral loyalty can be explained by attitudinal loyalty. Furthermore, empirical evidence has been found regarding the relationship between the more detailed levels of attitudinal loyalty (being satisfaction, trust and commitment) and behavioral loyalty measurements. First, positive paths are demonstrated between relationship satisfaction (attitudinal) and relationship duration/purchase intentions (behavioral) in earlier studies of Bolton (1998) and MacIntosh & Lockshin (1997). Second, Smith & Barclay (1997) found a positive relation between trust (attitudinal) and forbearance from opportunism (behavioral). Third, Morgan & Hunt (2004) provide empirical support for the relation between commitment (attitudinal) and acquiescence/propensity to leave/cooperation (behavioral). Despite the fact that different behavioral loyalty measures are used throughout these studies, the conclusions at least indicate the existence of a relationship between attitudinal and behavioral loyalty.

In line with these studies, Jensen & Hansen (2006) also find that increasing attitudinal loyalty (as combination of satisfaction, trust and commitment) leads to repeat purchases and customer loyalty (share-of-wallet increase). They conclude that “customers with a high relative attitude are much more likely to stick with your brand, even when it is not available.” Furthermore, several earlier studies showed that attitudinal loyalty is an important driver of behavioral loyalty for participants of loyalty programs (Daams et al., 2008; Furinto et al., 2009; Hansen et al. 2010; Lacey, 2009; Lacey et al., 2007). That leads to the suggestion that attitudinal loyalty can influence customer behavior, mediating the effects of marketing instruments (as SLPs) on behavioral loyalty (Chaudhuri & Holbrook, 2001). Taken together, this leads to the assumption that the effect of SLP-participation on behavioral loyalty is mediated by attitudinal loyalty.

(18)

18 H3b: The positive effect of SLP-participation on visit frequency change is mediated by attitudinal loyalty change.

H3c: The positive effect of SLP-participation on average basket size change mediated by attitudinal loyalty change.

2.7 Factors moderating the effect of SLP-participation

Both customer-related and program-related factors could influence the effect of SLP-participation on customer loyalty change. Therefore, conceptualizations of these factors are described below and their moderating effects will be tested throughout this study.

2.7.1 Customer-related factors: initial spending and number of competitors

When determining the effectiveness of loyalty programs, customers’ initial (pre-program) loyalty have to be taken into account. Lal & Bell (2003) found significantly less behavior change amongst heavy spenders compared to moderate and low spenders. Studies from Dréze & Hoch (1998) and Taylor & Neslin (2005) found the same effect of purchase volume increase amongst moderate, light and new customers. This can be explained by the fact that consumers will only raise their purchase behavior if it is below their consumption limit. Heavy spenders are the ones that already score very high on purchase behavior, so these are the ones that will most likely be affected by this ceiling effect (Lal & Bell, 2003). Therefore, it is expected that the behavioral loyalty change for heavy spenders is lower than for moderate and low spenders. This leads to the assumption that the indirect effect of SLP-participation on behavioral loyalty change is dependent on the initial spending, in such a way that initial spending is negatively affecting behavioral loyalty change. In line with Liu (2007), this study includes consumer’s initial spending at the beginning of the program, in order to increase the predictive usefulness of possible findings.

H4a: The positive (in)direct effect of SLP-participation on behavioral loyalty change is negatively moderated by the customer’s initial spending at the retailer.

(19)

19 industry, Liu (2007) found that, in order to create a successful loyalty program, “the lower spending among light and moderate users should be mainly due to polygamous loyalty … rather than insufficient need for the product/service category.” When customers spend little money because of the fact their absolute demand is low, it is hard to affect behavior using loyalty programs. In contrast, if customers are light or moderate users because of flying with multiple airlines (or, translating it to the grocery industry, visiting multiple super markets), loyalty programs are more successful (Liu, 2007).

Following this reasoning, it can be hypothesized that among people shopping at a lot of different stores, the effect of SLP-participation on customer loyalty will be higher. If people do shop at one or a few stores on the other hand, it is plausible to assume that these people already are quite loyal towards a certain retailer and there is not much room for growth in terms of purchase behavior. This room for growth will be bigger among consumers whom shop at multiple stores, as their absolute demand is not low, but it is spent at multiple retailers. This leads to the assumption that the indirect effect of SLP-participation on behavioral loyalty change is positively affected by the number of competitors consumers were regularly visiting before the program.

H4b: The positive (in)direct effect of SLP-participation on behavioral loyalty change is positively moderated by the consumer’s number of competitors visited.

2.7.2 Program-related factors: reward attractiveness and program achievability

Wirtz et al. (2007) demonstrate that the perceived attractiveness of a reward program positively affects customer’s behavioral loyalty, regardless of the attitudinal loyalty level of customers. In that study, it is concluded that when the perception of a reward program increases, the greater the perception of rewards gained from participation is (Gruen, 1994; Wright & Sparks, 1999), leading to an increase in share-of-wallet. This study will examine the (moderated) effect of SLP-attractiveness on behavioral loyalty change via two constructs: the perceived attractiveness of the reward and the perceived program achievability.

(20)

20 H5a: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is positively moderated by the consumer’s perceived reward attractiveness.

Next to the perceived reward attractiveness, another factor that could possibly influence the effectiveness of an SLP is the perceived achievability of the program. Liu (2007) concludes that the structure of a loyalty program should be set in such a way that light and moderate users will strive for the rewards. In that sense, rewards that are easier to achieve are likely to be more effective than large rewards that require a significant amount of effort, because less consumers will simply give up and see the program as irrelevant (Liu, 2007). Therefore, it is assumed that consumers with high perceived program achievability will show higher effects on behavioral and attitudinal loyalty change.

H5b: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is positively moderated by the consumer’s perceived program achievability.

2.7.3 Control variables: household size and household income

In line with Leenheer et al. (2007), this study includes household size and household income as control variables. Larger household sizes tend to have higher grocery spending then smaller households. Therefore, it is assumed that having a larger households leads to a stronger effect of SLP-participation on both change in behavioral and attitudinal loyalty. Furthermore, households with higher incomes also tend to have higher grocery spending, as they buy more luxurious products (Leenheer et al., 2007). Based on this, stronger effects on behavioral and attitudinal loyalty change would be expected for households with larger income. However, it is possible that households with a high income will value the rewards less than households with lower income (Leenheer et al., 2007). This might then lead to a weaker effect of SLP-participation on customer loyalty. Therefore, the moderating effect of household income on both behavioral and attitudinal loyalty is not defined a priori.

H6a: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is

positively moderated by household size.

H6b: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is moderated by the household income.

(21)

21 principle, people return good for good in order to what they receive (Bagozzi, 1995). When customers redeem, they receive (a) tangible reward(s). Therefore, it is expected that redemption during an SLP would lead to positive effects on both attitudinal and behavioral loyalty in addition to the positive effect of participation.

H6c: There is a positive effect of redemption on behavioral and attitudinal loyalty change.

Hypotheses

H1a: SLP-participation has a positive effect on share-of-wallet change. H1b: SLP-participation has a positive effect on visit frequency change. H1c: SLP-participation has a positive effect on average basket size change. H2: SLP-participation has a positive effect on attitudinal loyalty change.

H3a: The positive effect of SLP-participation on share-of-wallet change is mediated by attitudinal loyalty change.

H3b: The positive effect of SLP-participation on visit frequency change is mediated by attitudinal loyalty change.

H3c: The positive effect of SLP-participation on average basket size change mediated by attitudinal loyalty change.

H4a: The positive effect of SLP-participation on behavioral loyalty change is negatively moderated by the initial spending at the retailer.

H4b: The positive effect of SLP-participation on behavioral loyalty change is positively moderated by the consumer’s number of competitors visited.

H5a: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is positively moderated by the consumer’s perceived reward attractiveness.

H5b: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is positively moderated by the consumer’s perceived program achievability.

H6a: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is positively moderated by household size.

H6b: The positive effect of SLP-participation on behavioral and attitudinal loyalty change is moderated by the household income.

(22)

22

3. Methodology

This chapter will start with an elaboration on the questionnaire design and computation of variables. Thereafter, the model specifications and research method are explained. Following, the data collection is described and a sample description is given. Lastly, statistical power issues are addressed.

3.1 Questionnaire design and variable computation

As this study analyzes the effect of SLP-participation on attitudinal and behavioral loyalty, it is crucial that these constructs are operationalized in a correct way in the survey. Naturally, the same yields for the customer- and program-related factors included in the model. Furthermore, it is important that the ‘right’ people fill in the survey. Therefore, criteria are set up that the respondent must meet.

3.1.1 Screening criteria

Before respondents were able to fill in the whole survey(s), some screening questions were asked. Starting, respondents have to be at least 21 years old. It is believed that, in most cases, individuals younger than twenty are not yet responsible for their own groceries. Besides, a respondent has to be the main shopper of his/her household. Lastly, respondents had to indicate which retailer(s) they regularly visit out of a list of the eighteen biggest Japanese retailers. If the central retailer of this study - where the SLP was implemented - is regularly visited by the respondent, all criteria are met. In that case the respondent proceeded, filling in the rest of the questionnaire.

3.1.2 Attitudinal and behavioral loyalty

(23)

23 (see Appendix C for complete output). Therefore, new factors were created by taking the average of the items.

Behavioral loyalty is measured via share-of-wallet (SOW), visit frequency and basket size. As transactional data is not available, the values for behavioral loyalty are self-reported. Customer’s SOW and visit frequency are directly asked in the survey. It is believed that these values can be determined quite accurate by respondents. Due to the fact that basket size could fluctuate quite a lot per visit in the grocery industry, average basket size may be harder to determine for respondents. Therefore, the (average) basket size is not asked in the survey, but can be computed by combining the values of SOW and visit frequency, together with the customer’s total spend on groceries per week (asked in the survey, all questions are available in Appendix A2). To compute the customer’s basket size two steps have to be taken. Firstly, categorical values for visit frequency and total spend are turned into continuous values, by taking the mean of each category. For some categories (the lowest and highest), lower or upper bounds were not specified. In these cases the pattern of the other categories is continued as close as possible. For instance, for total spend on groceries per week steps of 3,000 Yen are taken from one category to the next. In the last case with a missing upper bound (32,001 or more) this pattern of +3,000 is maintained (see Appendix B). Secondly, the average basket size is computed per customer using the formula below:

𝐵𝑎𝑠𝑘𝑒𝑡 𝑠𝑖𝑧𝑒 = (𝑇𝑜𝑡𝑎𝑙 𝑠𝑝𝑒𝑛𝑑∗𝑆𝑂𝑊) 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦

3.1.3 Customer- and program-related factors

The customer-related factors that are included in this study are initial spending at the retailer and the number of competitors visited, as introduced in section 2.6.1. The initial spending at the retailer is captured as spending per week. This variable was created by the customer’s total spend on groceries (the continuous value) multiplied with the SOW before the program. The number of competitors visited by the customer is gathered via an easy calculation. Respondents had to identify all recently visited retailers (see also section 3.2.1), from which the number of competitors is calculated by adding up those numbers.

(24)

24 all original variables are at their mean, the mean centered variables will be zero. Then, the constant represents the value of the dependent variable for a hypothetical ‘average’ situation (Gijsenberg, 2014).

Furthermore, perceived reward attractiveness and program achievability are included as program-related factors throughout this study. Reward attractiveness is measured, making use of four dimensions: appeal, relevance, quality and design. The program achievability is measured via offer perception on three dimensions: additional payment, spend requirement and time to collect. All statements were measured on 11-point Likert scales (Appendix A3). The three items for perceived program achievability (Cronbach’s α = 0.830) and the four items for perceived reward attractiveness (Cronbach’s α = 0.917) are also tested on internal consistency and turn out to be reliable (Appendix C). So, new factors were created by taking the average of the items. In the multiple moderated mediation models these factors are mean centered as well.

3.1.4 Control variables

The variables household size and household income were asked with categorical questions. The range for household size was between ‘1’ and ‘6 or more’. Household income was asked on annual basis, also including six different categories. In order to decrease the difficulty of interpretation, the categorical values for household income were converted into continuous values by taking the mean of the category (see Appendix B). These variables are then mean-centered for the (moderated) mediation analysis.

3.2 Model specifications

(25)

25

Figure 2: Visual representation of a simple mediation model (from Hayes, 2013) applied to this study (covariates not included). The abbreviations in parentheses represent the variable names in the dataset.

The coefficients aᵢ, bᵢ and c’ will be obtained using least-squares regression. In order to estimate the values for these coefficients, the following regression equations can be specified for the simple mediation model (where aᵢ = a1, bᵢ =b1):

(1) 𝐵𝐿_𝐶𝐻_1,2,3= 𝑏₀+ 𝑐′𝑃𝐴𝑅𝑇 + 𝑏₁𝐴𝐿_𝐶𝐻 + 𝑏₂𝑅𝐸𝐷 + 𝑏₃𝑆𝐼𝑍𝐸 + 𝑏₄𝐼𝑁𝐶 + 𝑏5𝑃𝐴𝑅𝑇 𝑆𝐼𝑍𝐸 + 𝑏6𝑃𝐴𝑅𝑇 𝐼𝑁𝐶 + 𝑟, where

(2) 𝐴𝐿_𝐶𝐻 = 𝑎₀+ 𝑎₁𝑃𝐴𝑅𝑇 + 𝑎₂𝑅𝐸𝐷 + 𝑎₃𝑆𝐼𝑍𝐸 + 𝑎₄𝐼𝑁𝐶 + 𝑎₅𝑃𝐴𝑅𝑇 𝑆𝐼𝑍𝐸 + 𝑎6𝑃𝐴𝑅𝑇 𝐼𝑁𝐶 + 𝑟

Both 𝑎₀ and 𝑏₀ are intercept terms, while r is the residual of the regression. Furthermore, BL_CH₁ = SOW_CH = change in share-of-wallet between pre and post program. BL_CH₂ = FRQ_CH = change in visit frequency between pre and post program. BL_CH3 = BAS_CH = change in basket size between pre and post program. AL_CH = change in attitudinal loyalty between pre and post program.

PART = binary variable indicating whether consumer did (1) or did not (0) participate. RED = binary variable indicating whether consumer did (1) or did not (0) redeem. SIZE = household size (mean centered).

INC = annual household income in million’s Japanese Yen (mean centered).

In this simple mediation model, the direct effect of SLP-participation on behavioral loyalty change is captured by c’. The indirect effect through attitudinal loyalty is captured by 𝑎1𝑏1. This model is the starting point of analysis in this study. However, as customer- and program-related factors will be implemented into the model, the simple mediation model of figure 2 is

X = SLP-participation (PART)

(26)

26 expanded with moderator effects. Following Preacher et al. (2007), expanding the model with moderations can be done in several ways. Included in this study are: 1) variables W and Z (see figure 3) affecting path aᵢ and c’ and 2) variables V and Q (see figure 4) affecting path bᵢ and c’. The moderation effect of program-related factors will be tested on path aᵢ (and c’) from figure 2, while the moderation effect of customer-related factors will be tested on path bᵢ (and c’). Customer-related and program-related factors are tested separately, due to a practical constriction of the PROCESS-tool that does not allow including all moderators at the same time. In the end this leads to two moderated mediation models. As behavioral loyalty change will be measured on three different constructs, the estimation of every model will be measured on change in 1) share-of-wallet, 2) visit frequency and 3) basket size. The different models will be explained below.

3.2.1 Customer-related moderated mediation model

To test the moderated mediation effect of the customer-related factors initial spending and number of competitors, the PROCESS procedure for SPSS for multiple moderated mediation analysis from Hayes (2013) is used. Figure 3 shows a graphical representation of the customer-related moderated mediation model.

Figure 3 includes two variables V and Q testing their moderation effect on path bᵢ and c’. This means that model 17 is the appropriate one to use (Hayes, 2013). To estimate the coefficients in this model, equation (1) has to be expanded. This leads to the following regression equations:

Figure 3: Model for customer-related moderators initial spending (number 17 from Hayes, 2013). X = SLP-participation (PART)

M = Attitudinal loyalty change (AL_CH) Y = Behavioral loyalty change (BL_CH1,2,3)

V = customer’s initial spending at the retailer per week in 10’s Yen (SPE)

(27)

27 (3) 𝐵𝐿_𝐶𝐻_1,2,3= 𝑏₀+ 𝑐₁′_{𝑃𝐴𝑅𝑇 + 𝑐} 2′𝑆𝑃𝐸 + 𝑐3′𝐶𝑀𝑃 + 𝑐4′𝑃𝐴𝑅𝑇 𝑆𝑃𝐸 + 𝑐5′𝑃𝐴𝑅𝑇 𝐶𝑀𝑃 + 𝑏₁𝐴𝐿_𝐶𝐻 + 𝑏₂𝐴𝐿_𝐶𝐻 𝑆𝑃𝐸 + 𝑏₃𝐴𝐿_𝐶𝐻 𝐶𝑀𝑃 + 𝑏₄𝑅𝐸𝐷 + 𝑏₅𝑆𝐼𝑍𝐸 + 𝑏₆𝐼𝑁𝐶 + 𝑏₇𝑃𝐴𝑅𝑇 𝑆𝐼𝑍𝐸 + 𝑏₈𝑃𝐴𝑅𝑇 𝐼𝑁𝐶 + 𝑟, where (2) 𝐴𝐿_𝐶𝐻 = 𝑎₀+ 𝑎₁𝑃𝐴𝑅𝑇 + 𝑎₂𝑅𝐸𝐷 + 𝑎₃𝑆𝐼𝑍𝐸 + 𝑎₄𝐼𝑁𝐶 + 𝑎₅𝑃𝐴𝑅𝑇 𝑆𝐼𝑍𝐸 + 𝑎6𝑃𝐴𝑅𝑇 𝐼𝑁𝐶 + 𝑟

The conditional direct effect of SLP-participation on behavioral loyalty change then is captured by 𝑐₁′+ 𝑐₄′ 𝑆𝑃_𝑃𝑅𝐸 + 𝑐₅′𝑁𝑅_𝐶𝑂𝑀𝑃 , while the conditional indirect effect through attitudinal loyalty is captured by (𝑏1+ 𝑏2𝑆𝑃_𝑃𝑅𝐸 + 𝑏3𝑁𝑅_𝐶𝑂𝑀𝑃 )𝑎1 (Hayes, 2013). As these direct and indirect effects depend on the value of the moderator, the effect of participation will differ per condition. Therefore, these effects are conditional effects.

If both moderators take the value zero, in the conditional direct effect for example, the direct effect is captured by 𝑐1′, as 𝑐4′ 𝑆𝑃_𝑃𝑅𝐸 and 𝑐5′𝑁𝑅_𝐶𝑂𝑀𝑃 become zero. Because the moderators will be mean centered, the value zero represents the mean. So 𝑐₁′_{, which is} estimated in the least-squares regression, captures the effect of participation on change in behavioral loyalty if the moderators are at their mean. As this is more useful than the effect of participation if the moderators are at zero (meaning initial spending and number of competitors are zero), this clearly demonstrates the advantage of mean centering.

3.2.2 Program-related moderated mediation model

The program-related factors, reward attractiveness and program achievability, are tested on path aᵢ and c’. This is graphically represented in figure 4.

Figure 4 includes two variables W and Z, testing their moderation effect on path aᵢ and c’. In order to estimate the coefficients in this model, equations (1) and (2) have to be expanded. As

Figure 4: Model for program-related moderators (number 10 from Hayes, 2013). X = SLP-participation (PART)

M = Attitudinal loyalty change (AL_CH) Y = Behavioral loyalty change (BL_CH_1,2,3)

(28)

28 the effect on path aᵢ is tested instead of the effect on path bᵢ, this leads to the following regression equations: (5) 𝐵𝐿_𝐶𝐻1,2,3= 𝑏0+ 𝑐1′𝑃𝐴𝑅𝑇 + 𝑐2′𝑅𝐸𝑊_𝐴𝑇𝑅 + 𝑐3′𝑃𝑅𝐺_𝐴𝐶𝐻 + 𝑐₄′_{𝑃𝐴𝑅𝑇 𝑅𝐸𝑊_𝐴𝑇𝑅 + 𝑐} 5′𝑃𝐴𝑅𝑇 𝑃𝑅𝐺_𝐴𝐶𝐻 + 𝑏1𝐴𝐿_𝐶𝐻 + 𝑏2𝑅𝐸𝐷 + 𝑏3𝑆𝐼𝑍𝐸 + 𝑏4𝐼𝑁𝐶 + 𝑏5𝑃𝐴𝑅𝑇 𝑆𝐼𝑍𝐸 + 𝑏6𝑃𝐴𝑅𝑇 𝐼𝑁𝐶 + 𝑟, where (6) 𝐴𝐿_𝐶𝐻 = 𝑎0+ 𝑎1𝑃𝐴𝑅𝑇 + 𝑎2𝑅𝐸𝑊_𝐴𝑇𝑅 + 𝑎3𝑃𝑅𝐺_𝐴𝐶𝐻 + 𝑎4𝑃𝐴𝑅𝑇 𝑅𝐸𝑊_𝐴𝑇𝑅 + 𝑎₅𝑃𝐴𝑅𝑇 𝑃𝑅𝐺_𝐴𝐶𝐻 + 𝑎₆𝑅𝐸𝐷 + 𝑎₇𝑆𝐼𝑍𝐸 + 𝑎₈𝐼𝑁𝐶 + 𝑎₉𝑃𝐴𝑅𝑇 𝑆𝐼𝑍𝐸 + 𝑎₁₀𝑃𝐴𝑅𝑇 𝐼𝑁𝐶 + 𝑟

In this model, the conditional direct effect of SLP-participation on behavioral loyalty change is captured by 𝑐₁′_{+ 𝑐}

4′ 𝑅𝐸𝑊_𝐴𝑇𝑅 + 𝑐5′𝑃𝑅𝐺_𝐴𝐶𝐻. The conditional indirect effect through attitudinal loyalty is captured by (𝑎₁+ 𝑎₂𝑅𝐸𝑊_𝐴𝑇𝑅 + 𝑎₃𝑃𝑅𝐺_𝐴𝐶𝐻 )𝑏₁ (Hayes, 2013). 3.2.3 Control variables

In order to determine the effect of household size and household income on the change in behavioral and attitudinal loyalty, these variables are included in the estimation of the models. As we are interested in the effect of these variables on the change in both behavioral loyalty and attitudinal loyalty, both variables are included as covariates and tested on the dependent variable as well as on the mediator. These variables are included in equations 1-6, as well as the effect of redemption. Redemption is a binary variable, taking value zero if the respondent did not redeem any rewards and one if one or more redemptions took place. Its main effect is tested on both the mediator and dependent variable.

3.3 Research method

(29)

29 However, besides investigating the indirect effect of SLP-participation on behavioral loyalty through attitudinal loyalty, this research is also interested in the direct effects of SLP-participation on both behavioral and attitudinal loyalty. These effects are estimated making use of least-squares regression. The two models with attitudinal loyalty change as dependent variable, equations 2 and 6 (equation 4 is exactly the same as 2), lead to two least-squares regression models. Furthermore, three equations (1, 3 and 5) include behavioral loyalty change as dependent variable. As behavioral loyalty is operationalized in three ways, this leads to a total of 3*3 = 9 least-squares regression models.

While Baron & Kenny (1986) state that a mediation effect exists if all relationships are significant and the effect of path c’ is weaker than the direct effect of the predictor variable on the dependent variable without any mediator, Hayes (2009) recommends a bootstrapping method in order to test for mediation. This method is preferred over the causal steps approach, as the conclusion is based on an estimate of the indirect effect itself. Furthermore, it makes no assumptions about the shape of the sampling distribution of the indirect effect. More specifically, it does not assume a normal distribution (Hayes, 2009). Therefore, the bootstrapping method will be conducted in this study, to test for mediation effects. Bootstrapping is repeated 10,000 times, which is sufficient as Hayes (2009) recommends at least 5,000 times.

In order to be able to grasp under what situations and/or if consumer characteristics affect the possible mediation effect tested in this study, customer- and program-related moderators are included. These moderators are variables that may possibly affect the size or sign of the effect of path aᵢ, bᵢ and/or c’ from the simple mediation model. In order to estimate the moderated mediation effect, the bootstrapping method from Hayes’ PROCESS-tool is used. This tool generates a bootstrap confidence interval for the index of moderated mediation. If the mediated effect of SLP-participation to behavioral loyalty change through attitudinal loyalty change is somehow related to another variable, than the effect is moderated by that variable and the confidence interval will not include zero (Hayes, 2015).

(30)

30 program-related moderated mediation models are first stage and direct effect moderation models.

3.4 Data collection

An online survey was randomly distributed by BrandLoyalty in cooperation with an external marketing research office, among a panel existing of a representative sample of the Japanese population. Respondents were contacted via email and asked to fill in the survey. Two surveys were designed and spread at two different moments in time among the respondents: a pre survey before the SLP started and a post survey after the SLP was finished. In order to test the actual effect of the SLP in terms of change in behavioral and attitudinal loyalty, the post survey was only sent to those respondents that filled in the pre survey. The pre survey remained online until 1,000 respondents were gathered. Out of that group, a total of 591 people also filled in the post survey, leading to a total sample of 591 respondents.

It is believed that the setup of this research, including a pre and post measurement, directly solves the problem of self-selection. In most studies regarding the effectiveness of loyalty programs, self-selection seems to be a significant problem. Leenheer et al. (2007) state that one of the reasons past studies regarding LP-effectiveness found different effects may relate (partly) to self-selection bias. Behavioral loyalty differences between members and non-members (Van Heerde & Bijmolt, 2005) may be partly driven by self-selection of the most loyal customers into the loyalty program (Leenheer et al., 2007). Regarding SLPs, loyal customers are the ones that are rewarded the most. Therefore, it is plausible to assume that these customers will be more likely to participate within the program. Then, if the effectiveness of the SLP is measured at the end of a program by comparing the behavioral loyalty of participants and non-participants, it is hard to determine which part of the (eventual) difference is really coming from the SLP. However, using a setup with a pre and post measurement, takes the fact that participants probably already had higher behavioral (and attitudinal) loyalty into account.

(31)

31 enough room for improvement, also for participants. All behavioral loyalty values (SOW, visit frequency, basket size) before the program did not significantly differ between the two groups and there was ample room for improvement on all constructs (see Appendix D). Concluding, as this study calculates the SLP-effect on the change in both types of loyalty, while values were not extremely high for participants before the program, self-selection problems should not be an issue.

Respondents that did not meet the screening criteria were directly removed at the beginning of the online survey, meaning all of the 591 respondents in the dataset met the criteria of age, being the main shopper and regularly shop at the retailer under study. The awareness, participation and redemption rates of the loyalty program at this retailer are shown in figure 5, together with an average of ten random SLPs (‘benchmark’ in figure 5) that were implemented by BrandLoyalty in the last five years.

Figure 5 shows that there is quite a difference in terms of program awareness, where the SLP in this study scores more than 20% lower compared to the benchmark. The participation rate shows the percentage of respondents that were aware of the program and collected stamps (irrespective of redeeming), while the redemption rate shows the percentage of respondents that were aware of the program and redeemed at least once. These percentages are comparable for the SLP under study and the benchmark. For the SLP under study, 53.2% of the aware people (40.1%) did participate. Therefore, the percentage of participants from the total sample then becomes .401*.532 = .209 = 20.9%. Using the same calculation for the benchmark leads to 32.3% participation of the total sample. Redemption rates are 9.3% and 16.3% for respectively the SLP under study and the benchmark. It can be seen that, despite the fact participation and redemption rates for aware people are quite similar, the overall rates differ quite a lot because of the lower awareness of the SLP under study.

61.9% 52.2% 26.4% 40.1% 53.2% 23.2% 0% 10% 20% 30% 40% 50% 60% 70%

Awareness rate Participation rate Redemption rate

Benchmark SLP

(32)

32 Out of the 591 respondents, 237 people were aware of the SLP. Only the aware respondents were able to answer statements about program achievability and reward attractiveness. As these variables are included in one of the moderated mediation models, only these people are included in the analysis. Furthermore, some cases in the dataset appeared to be unreliable as (almost) all of the statements were answered with the same answer. It is believed that these respondents are streamliners, meaning that they answer without even reading the statements. Therefore, if a respondent gave more than 90% of the time the same answer regarding all Likert-statements in the survey, it is removed. Following this rule, a total of 29 respondents was considered as streamliner and thus taken out of the dataset. While excluding these people from the dataset leads to a decrease in sample size of 12%, it is believed that the data quality increases because of this adjustment, leading to more reliable results. This leads to a final sample of 208 respondents. Figure 6 shows the distribution of participants and redeemers regarding the final dataset.

Figure 6: Number of respondents participating and redeeming in final dataset (percentages of total sample). 3.5 Sample description

Out of the 208 respondents in the final dataset, 61.5% is female and 38.5% is male. Most of the respondents are between 31 and 50 years old (63%), 9% is younger than 30 and 28% is 51 or older. The distribution of respondents on household size/income is displayed in table 2.

Variable Frequency Percentage Variable Frequency Percentage

Household income Household size

2,000,000 or less 12 5.8% 1 43 20.7% 2,000,000 – 4,000,000 31 14.9% 2 55 26.4% 4,000,001 – 6,000,000 47 22.6% 3 59 28.4% 6,000,001 – 8,000,000 49 23.6% 4 43 20.7% 8,000,001 –10,000,000 25 12% 5 4 1.9% 10,000,000 or more 25 12% 6 or more 4 1.9%

(33)

33 Furthermore, none of the participants is shopping only at the retailer under study, as all respondents regularly visit at least one competitor. More than 80% of the respondents do their grocery shopping at two or three retailers (Appendix E). This finding is in line with previous research from Kahn & McAlister (1997). Besides, the number of redemptions for the people that did redeem during the reward shows that most consumers redeemed a low amount of rewards. More than 65% of the redeemers made a maximum of two redemptions, while only 9% redeemed 5 or more rewards (Appendix E). As can be seen from figure 7, the average reward attractiveness and program achievability scores are lower for the SLP under study than for the benchmark (same as used for figure 5). Especially the reward attractiveness shows quite a big gap, scoring almost a whole point lower on an eleven-point scale, indicating that the reward was perceived less attractive than rewards in an ‘average’ SLP.

Figure 7: Average program achievability and reward attractiveness of SLP compared to benchmark. 3.6 Statistical power

As the sample size of the final dataset decreased quite a lot, statistical power is also decreasing. Statistical power analysis is neglected in lots of studies, but its importance is not questioned among methodologists (Cohen, 1992). Therefore, it is analyzed whether the final sample size leads to a large enough statistical power. Cohen (1992) states that a statistical power of .80 is conventional for general use. This means that an 80% probability is reached to detect relationships between variables, given the fact that this relationship in reality exists. In order to specify the minimum sample size leading to a statistical power of .80 at a 95%-confidence level, the population effect size needs to be set. Following Cohen (1992), the ES index for multiple correlations (or least-squares regression) is defined as 𝑓² = 𝑅²

1−𝑅² . 5.56 _5.43 5.24 4.51 0 1 2 3 4 5 6

Program achievability Reward attractiveness

(34)

34 Specifying the population’s effect size is the most difficult part of statistical power analysis, as it has to be set by the researcher just based on what is believed to hold for the population (Cohen, 1992). This study makes use of the conventional criteria as set by Cohen (1992). Table 3 shows values for small, medium and large effect sizes. Cohen (1992) refers to a medium effect size as “an effect likely to be visible to the naked eye of a careful observer. It has been noted that a medium effect size approximates the average size of observed effects in various fields.” This study assumes that the effect size reaches a medium value (.15), leading to an R² of .1304 when solving the equation above. Next, in order to determine whether the sample size in this study is large enough to reach statistical power of .80 when the population effect size is at medium value, table 3 shows minimum sample sizes for least-squares regression analysis at different amounts of predictor variables (Cohen, 1992).

Table 3: Effect sizes for small, medium and large effects and minimum sample sizes for number of predictors (from Cohen, 1992).

This table stops at 8 predictor variables, while the largest estimated models within this study move up to thirteen predictors (equation 3). While the minimum sample size value cannot be read from the table, it will definitely be below 208. This is inferred from the pattern of table 3, and the increase in minimum sample size by increasing the number of parameters. Going from 6 to 7 and from 7 to 8 predictors means that sample size should increase with 5. Following this pattern, including thirteen predictors leads to 107 + 25 = 132 as minimum sample size, way below the 208 cases in this study. Therefore, it is concluded that for the least-squares regression models that are estimated to determine the effects on attitudinal and behavioral loyalty change, the sample size is large enough under the condition of medium effect size.

Table 3 also clearly shows the consequence of a decrease in effect size. When the effect size drops from medium to small, the impact on the minimum sample size is huge. This has to be kept in mind when estimating the models. A lower R² will lead to a lower f² and therefore to smaller effect size. Therefore, when estimated models are not significant, it is

Small Medium Large

(35)

35 possibly because of the fact that the effect size is smaller than assumed and sample size has to be increased in order to find significant effects with .80 statistical power (Cohen, 1992).

(36)

36

4. Results

This chapter will elaborate on the results of the analysis. First, some exploratory analyses are described. Then, the results of the least-squares regression models testing effects on attitudinal and behavioral loyalty change are provided. Lastly, multiple moderated mediation models are explained and conditional (in)direct effects identified.

4.1 Exploratory analysis

Table 4 shows the correlation coefficients for the variables included in this study. It can be seen that there are no significant correlations between participation (and redemption) and attitudinal and behavioral loyalty change. Participation and redemption are strongly correlated with perceived program achievability and reward attractiveness. Furthermore, except for a significant (p < .05) correlation between attitudinal loyalty change and SOW change for participants (.210), no correlations are found between change in attitudinal loyalty and behavioral loyalty change measures. Initial spending is negatively correlated with behavioral loyalty change for participants, indicating that an increase in participants’ initial spending leads to decreases in change of SOW, visit frequency and basket size.

AL_CH SOW_CH FRQ_CH BAS_CH SPE CMP REW_ATR PRG_ACH SIZE INC PART RED AL_CH 1 .098 .071 .009 -.083 -.085 .145 .056 -.096 -.192 SOW_CH .210* 1 .325** .472** -.122 .028 -.023 -.052 -.096 -.168 FRQ_CH .075 .397** 1 -.238* -.015 -.082 -.007 -.048 .027 .057 BAS_CH .098 .522** -.089 1 -.247* .16 -.087 -.037 -.163 -.135 SPE -.077 -.419** -.249** -.222* 1 -.032 -.046 .208* .083 .260** CMP -.143 -.001 -.13 .072 .007 1 .141 .223* -.096 .061 REW_ATR .007 -.036 -.022 .088 .094 .192* 1 .253* .058 -.089 PRG_ACH -.093 -.088 .047 .026 .288** .290** .611** 1 .003 .036 SIZE -.009 .02 .174 .024 .108 .184 .098 .179 1 .361** INC .146 -.042 .025 .092 .17 .142 .058 -.011 .276** 1 PART ª .01 -.028 -.06 .074 .064 .303** .586** .502** .117 .096 1 RED ª -.007 -.055 -.037 -.005 .158* .237** .452** .419** .069 .019 .484** 1 Table 4: Pearson’s correlation matrix, with * p < 0.05 and ** p < 0.01.

a: as PART and RED are categorical variables, these values are Spearman’s rho coefficients.

Notes: correlations above the diagonal represent correlations for non-participants, correlations below the diagonal for participants. For PART and RED the correlations are for the total sample.

The effect of short-term loyalty programs on (the relationship between) attitudinal and behavioral loyalty change in a grocery retail setting

The effect of short-term loyalty programs on (the relationship

between) attitudinal and behavioral loyalty change in a

grocery retail setting

The effect of short-term loyalty programs on (the relationship

between) attitudinal and behavioral loyalty change in a

grocery retail setting

January, 2016

Management summary

Preface

Table of contents

1. Introduction

2. Theoretical framework

3. Methodology

4. Results