The influence of loyalty
vouchers on customer value
Does it pay of to reward your best customers for their loyalty?
BY
LEONE VAN DER VEEN
The influence of loyalty
vouchers on customer value
Does it pay of to reward your best customers for their loyalty?
BY
LEONE VAN DER VEEN
January 13, 2017
S2799421
Tuinstraat 34, 8011 HC Zwolle leonevdveen@hotmail.com
+31 6 23 80 32 24
Master Thesis Marketing Intelligence
University of Groningen Faculty of Economics & Business
Department of Marketing PO Box 800, 9700 AV Groningen
Abstract
Preface
People always say that writing a thesis is synonymous to going through a lot of ups and downs. The last four months I have experienced these ups and downs, but it was worth it. This end result makes me proud but most of all it makes me graceful to all the people who enabled me to come so far.
First of all, I want to thank my first supervisor Prof. Tammo Bijmolt. During the first meeting I told Tammo that I lost my self-confidence, about writing a thesis, during my previous study. Tammo learned me how to get this confidence back by the way he motivated and supported me. Thank you Tammo!
Furthermore I want to thank my following students, Jeroen Aarts and Esther Menting, best friend, Sandra van de Beld, and my boyfriend, Thomas, for their input, support and feedback.
Good academic skills are important for writing a thesis, but data is even more important. So last but not least, I want to thank the anonymous construction and furnishing store for sharing their data with me.
Table of content
Chapter 1 Introduction ... 1
1.1 Loyalty programs ... 1
1.2 Problem statement ... 2
1.3 Academic and managerial contribution ... 2
1.4 Structure of the study ... 3
Chapter 2 Theoretical framework ... 3
2.1 Customer lifetime value ... 4
2.2 Effect of voucher possession on purchase incidence ... 5
2.3 Effect of voucher redemption on purchase quantity ... 6
2.4 Moderators; Face value, geographic distance, relationship length ... 7
2.6 Theoretical framework ... 10
Chapter 3 Data description ... 10
3.1 Data collection ... 10
3.2 Data overview ... 11
3.3 Control variables ... 13
3.4 Data cleaning ... 13
3.5 Descriptive statistics ... 13
Chapter 4 Empirical model ... 15
4.1 Tobit model ... 15
4.2 Tobit model type II / Heckman correction ... 18
4.3 Panel data ... 19
4.4 Tobit model type II with panel data ... 20
Chapter 5 Results ... 20
5.1 Purchase incidence ... 20
5.2 Purchase quantity ... 23
5.3 Multicollinearity ... 26
5.4 Vouchers and customer value ... 26
5.5 Summary of results ... 29
Chapter 6 Conclusions and recommendations ... 29
6.1 Conclusion ... 29
6.2 Limitations ... 31
6.3 Implications ... 32
References ... 34
Appendix ... 37
A. Formulas & commands used for calculating the variable purchase incidence ... 37
1
Chapter 1
Introduction
1.1 Loyalty programs
Customers are nowadays likely to be confronted with loyalty programs. There are a lot of companies in different branches (supermarkets, gasoline stations and clothing stores) that make use of a loyalty program (Leenheer et al., 2007; Lewis, 1997). The definition of a loyalty program can be defined as follow: “A loyalty program is an integrated system of marketing
actions that aims to make members customers more profitable by enhancing their loyalty.”
(Leenheer, 2004, p. 8). Research done by GfK and BrandLoyalty shows that loyalty programs are integrated in the society. In 2002 the total population of the Dutch society owned over 10 million loyalty cards in total (GfK 2002). Today, 14 years later, the total number of loyalty cards is increased, and maybe even be doubled (BrandLoyalty, 2015). This also means that households are not only member of one or two loyalty programs, but they are sometimes member of more than 10 loyalty programs (BrandLoyalty, 2015). The question that rises is; how effective are those loyalty programs if people are member of so many different loyalty programs (Meyer-Waarden, 2007)?
2 demonstrated in their study that customer loyalty positively influence CLV. So the more loyal a customer is, the higher the CLV will be. But does the use of a loyalty voucher, having the purpose to increase the loyalty of the customer, also positively influences the CLV?
This study is based on the loyalty program of a construction and furnishing store. This company makes use of a loyalty voucher, as described above. A more specific description of the loyalty program that is used by the company is given in section 3.1.
1.2 Problem statement
Based on the literature described in section 1.1, the problem statement of this study is formulated as follow: What is the effect of a delayed reward system, in the form of a loyalty
voucher, on the customer lifetime value? To answer this problem statement the concepts that
are part of the problem statement need to be fully understood. That is why the first research question goes about one of the concepts, CLV. The first question is formulated as follows: What is customer lifetime value? This CLV is an equation that consists of several components. Purchase incidence and purchase quantity are two of them. They are used to measure the effect of the drivers of voucher, voucher possession and voucher redemption, on CLV. That is why the second and third subquestions are formulated as follows: How does voucher possession
affect purchase incidence? and how does voucher redemption affect purchase quantity? The
effect of voucher possession is only measured on purchase incidence, and not on purchase quantity, to higher the quality of the model that is used. A more specific explanation is given in section 4.2. The last subquestion goes about the variables; face value, geographic distance and relationship length, that possibly influence the effect of vouchers on CLV, also called a moderation effect. What is the moderation effect of face value, geographic distance and
relationship length?
1.3 Academic and managerial contribution
3 Because of this differences, research on the effectiveness of a loyalty voucher will give new insights in the effectiveness of loyalty programs. Beside this, the study provides empirical insight in a branch in which almost no research is done about loyalty programs.
The managerial contribution of this study is, giving insight in effectiveness of a delayed reward loyalty program in the current retail environment. In the first section of this chapter the retail environment, and the changes in the use of loyalty programs, is already discussed. But another big change in the retail environment, that also influenced the use of loyalty programs, need to be mentioned. In 2008 an economic crisis began in the Netherlands, this has made customers more price sensitive than ten years ago. This price sensitivity had as a consequence that customer became less loyal to retailers (Capgemini, 2013). This can be seen in the growing number of loyalty cards that customers hold (BrandLoyalty, 2015). Customers are in particular focused on the rewards they receive for being part of a loyalty program. They do not hold loyalty cards because they feel a close relationship with the retailer. Customers want to pay the lowest price for a product, it does not matter which retailer the product is offering (Capgemini, 2013). With this study, marketing managers receive insights in the effectiveness of a delayed reward system, in the form of a voucher, on purchase behaviour of customers in the 2010s.
1.4 Structure of the study
The structure of the study is as follow; in chapter 2 the theoretical framework for the analysis of this study is set out. Chapter 3 consists of a data description. In chapter 4 the empirical model and the research methods are described. In chapter 5 the results of the research are shown and discussed. And the final chapter, chapter 6, consists of the conclusion, implications and limitations of the study.
Chapter 2
Theoretical framework
4
2.1 Customer lifetime value
The goal of this study is to assess what the influence of loyalty vouchers is on the CLV. CLV is a metric that measures the firm profitability, literally it is ‘the present value of all future
profits obtained from a customer over his or her life of relationship with a firm’ (Gutpa, et al,.
2006). Because a loyalty reward system has as goal to make customers more profitable by enhancing their loyalty (Leenheer, 2004), CLV is used in this study as a measurement of this profitability. CLV consist of three components (Kumar and Pansari, 2016);
1) purchase frequency & quantity 2) gross contribution margin 3) marketing costs
When those components are multiplied and subtracted of each other the net cash flow of a customer is calculated. The elements are represented in equation 1, a general equation that often is used for calculating CLV. To obtain a net cash flow this equation multiplies the retention probability (r) with the earned margin on individual i (m). The net cash flow is divided by one plus the discount rate (d) which gives the CLV (Verhoef, 2015).
(1)
r = Retention probability;
mi = Earned margin on household i;
d = Discount rate.
In this study the retention probability cannot be measured because there is no data available about retention. This makes it impossible to measure the lifetime value, that is why, from now on, we do not speak about CLV but about customer value CV. The equation for CV that is used in this study, divides the calculation of the net cash flow in two parts; 1) a calculation of the total gross profit of each household and 2) a calculation of the total marketing costs made on each household. This gives the following equation (Leenheer, 2004)
5 PIit = Purchase incidence; 1 if household i makes a transaction in week t,
otherwise 0
PQit = Purchase quantity; total expenditures of household i makes in week t,
in Euros;
VOit = Voucher received by household i makes in week t;
vm = Marginal costs of mailing a voucher to a household; DCnvit = Direct marketing costs not related to the voucher
For the calculation of the gross profit the probability of purchase incidence (PIit) is multiplied
by expected purchase quantity (PQit) which is multiplied by the average margin (m). The
expected purchase quantity is the total expenditures household i does in week t. The face value of the voucher, if it is redeemed, is included in the variable purchase quantity.
The calculation of the costs that are made for household i in week t consist of costs that are made for mailing the voucher to the customer i (VOit) and direct marketing costs that are not
related to the voucher (DCnvit).
In order to assess the influence of loyalty vouchers on CV, the effect of these vouchers on purchase incidence and quantity, two components of CV, need to be estimated first.
2.2 Effect of voucher possession on purchase incidence
Vouchers are, next to price promotions and coupons, a type of sale promotion. That revenues increase during sale promotion is generally known. Gupta (1988) researched the effect of sale promotions. His main research question was; do sales increase due to consumers switching from other brands or is the brand borrowing sales from the future as consumers advance their purchases in time or stockpile the product? The conclusion of this research is that 84% of sales increase due to promotion comes from brand switching. The other 16% comes from purchase acceleration and stockpiling (Gupta, 1988). Brand switching lead to additional purchases; purchase acceleration means that customers do their purchase earlier in time. Both brand switching and acceleration lead to an increase in purchase incidences. The research of Gupta did not specify the type of sales promotion, he looked at the overall effect of sales promotions. This study focuses on one specific type, the voucher. Does receiving a voucher also leads to more purchase incidences?
6 customer to return to the store and make another purchase. That is why, in this study, a voucher is called a type of advertisement (Leone and Mulhern, 1991). There is a lot of existing literature about advertisements and the effectiveness of it. One of the findings is that when the personal relevance is high, customers are more intended to pay attention to the advertisement. When the customer pays a lot of attention to an advertisement this brand will become easier accessible in memory. Purchase decisions are strongly influenced by the extent to which brand names are accessible in memory (Fennis, 2016). The easier the brand name comes to mind, the more likely it is that the customer purchases a product with the brand name. This is also applicable to the voucher that is used for this study. The member has a high personal relevance to the voucher because the member has explicitly earned the voucher. So when a member needs to make a purchase decision, the member is more likely to go the store of which the member has received a voucher recently, because the name of that store is easily accessible in mind. This leads to more purchase incidences.
Bawa and Schoemaker (1987) researched the effect of voucher possession on purchase incidence and they found a positive effect. Receiving a voucher increases the purchase incidence of customers. According to them, the reason for this effect is that members tend to make a purchase earlier in time when they receive a voucher because they afraid to forget or to lose the voucher, purchase acceleration (Bawa and Schoemaker, 1987). Based on the literature a positive effect is expected in this study for the effect of voucher possession on purchase incidence.
H1a: Voucher possession has a positive effect on purchase incidence.
2.3 Effect of voucher redemption on purchase quantity
There exist arguments for both a positive and negative effect of voucher redemption on the purchase quantity. It depends, among other things, on the form of reduction of the voucher. Possible reduction forms are proportional reduction (%) and fixed reduction (€). In this study the vouchers have a fixed reduction. A voucher with a fixed reduction does not give an incentive to purchase more. In comparison a voucher that has a proportional reduction the total reduction depends on the quantity you purchase. The higher the quantity the higher the total price reduction, so it is an incentive to purchase more.
7 1992). But it is important to take into account that there is a difference between vouchers and coupons. A coupon is only valid for a specific product and a voucher is valid for the total assortment in the store. Lee and Steckel (1999) found that customers prefer a large assortment. Based on this, the assumption is that it is more likely to see an effect of redemption on purchase quantity for voucher than for coupons.
Another difference between a coupon and a voucher is the reason why people get the sales promotion. Coupons are usually given to every customer, it does not matter if they are member of a loyalty program or not. They do not have to do anything to receive the coupon. For a loyalty voucher it holds that the customer need to be a member of the loyalty program and have to do something if they want to receive the voucher. They need to collect points by making purchases at the store. Because the customers have earned the voucher, this will stimulate them to do larger transactions or purchase more luxury items (Kivetz and Simons, 2002; Mick and Faure, 1998). Smith and Sparks (2009a,b) also researched the effect of redemption on the type of products that customers buy. They found that if customers can do a redemption, and they plan this redemption, they have a focus on hedonic goods. If the redemption is impulsive customers tend to focus on utilitarian goods (Smith and Sparks, 2009a,b ). In this study all redemptions are planned because the customers receive the voucher at home and need to bring it to the store if they want to redeem it. The focus on hedonic goods, because of planned redemption, will positively influence the purchase quantity. Why is this? Hedonic goods are goods that provide more experiential consumption, fun, pleasure, and excitement (designer clothes, sports cars, luxury watches, etc.) (Dahr and Wertenbroch, 2000). These goods are in general more expensive than utilitarian goods, this will increase the purchase quantity.
Despite there are arguments for both a positive and negative effect of voucher redemption on purchase quantity, most of the arguments are positive. Because of that, this study expects that when a customer redeems the voucher the purchase quantity will increase. The hypothesis for this study is formulated as follow:
H2: Voucher redemption has a positive effect on purchase quantity.
2.4 Moderators; Face value, geographic distance, relationship length
8
Face value
Existing research shows that face value is an important determinant for coupon effectiveness. Face value is the value of the coupon of voucher that the customer receives. The higher the value of the coupon, the more likely the voucher is creating a feeling of saving (Barat and Paswan, 2005). Dutch customers are very sensitive for savings and price promotions. 85% of the Dutch customers takes discounts into account during their purchase decisions, both online and offline (Nationale Kortingsmonitor, 2016). This is a reason to assume that the face value of vouchers influence purchase decisions.
Krishna and Shoemaker (1992) researched the influence of face value of coupons on purchase quantity. They found an positive effect for face value on redemption rates. A higher face value of a coupon increases the probability to redeem the coupon and increase the incremental purchases of prior non-buyers (Krishna and Shoemaker, 1992). For prior buyers it also holds that the higher the face value the higher the probability that the customer will redeem the coupon (Krishna and Shoemaker, 1992; Leone and Srinavasan, 1996). A corresponding increase in incremental purchases for the brand is not found (Krishna and Shoemaker, 1992).
The moderations effect of face value on the relationship between voucher possession and purchase incidence is not found yet. Because of the fact that face value has found to be an important determinant for coupon effectiveness, it is plausible to expect that a higher face value will lead to more purchase incidences.
H3a: Face value positively moderates the relationship between voucher possession and purchase incidence.
H3b: Face value positively moderates the relationship between voucher redemption and purchase quantity.
Geographic distance
9 they purchase. But for purchase quantity the opposite is expected. People who live close to the store will buy more often but are also visiting the store for only one cheaper product, for example a screw. People who are living further from the store do not have the intention to buy a screw, they probably only buy products that are more expensive. The hypotheses are formulated as follow:
H4a: Geographic distance negatively moderates the relationship between voucher possession and purchase incidence.
H4b: Geographic distance positively moderates the relationship between voucher redemption and purchase quantity.
Relationship length
The relationship between the customer and the store is the number of months the customer is part of the loyalty program. The longer the customer member is of the loyalty program the more experienced the customer is with the loyalty program. Loyalty program experience has found to be a positive driver of purchase behaviour (Bolton and Kannan, 2000).
In this study the moderation effect of relationship length (loyalty program experience) is researched. Liu and Brock (2007) also researched the effect of customer’s relationship length with a loyalty program on redemption activity. They found a positive effect between those two variables. The longer the relationship is the higher the probability that the customer redeems the voucher. In this study a positive effect of redeeming a voucher on purchase quantity is expected (see section 2.3). Based on this expected effect and the research of Liu and Brock (2007) the assumption is that relationship length positively moderates the expected effect between redemption activity and purchase quantity.
This study also examines the effect of voucher possession on purchase incidence. If relationship length also moderates this effect is not researched yet. That is why it is included in this study. Because of the absence of existing research the sign, whether it is positive or a negative moderator effect, cannot be defined yet.
H5a: Relationship length moderates the relationship between voucher possession and purchase incidence.
10
2.6 Theoretical framework
All variables and relationship between the variables are described. Figure 2.1 visualises the relationships and moderation effects that are described in previous sections.
Figure 2.1 Theoretical framework
Chapter 3
Data description
Chapter 3 gives a description of the dataset. How is collected, which variables are included, which control variables are used and how is the dataset cleaned. In the last section the descriptive statistics of the dataset are given.
3.1 Data collection
In the chapter 1 there is already mentioned that this empirical study uses the real life customer data of the loyalty program of a construction and furnishing store. Because of confidentiality reasons the name of the store cannot be identified.
The company is a family company that is owned for almost 60 years by the family. They started as an demolition company who sold the demolition material they collected. Nowadays they are the biggest construction and furnishing store of the northern and eastern part of Holland, with a surface of 22.000 m2. The company is running a loyalty program since 1978.
11 The company has around 22.000 loyalty members, from which 2700 are active members (i.e. members who collected and/or redeemed points last year). The data is collected on household level. Each household that is a member of the loyalty program is linked to a unique member ID and can hold 1 or more loyalty card(s). This makes it possible for a family to hold more loyalty cards, one for each member of the family, but collect points together.
A key component of the loyalty program is the saving feature that is just explained. Next to this component, the loyalty program consists of newsletters that are send via email. Not all members receive this newsletter because some of them did not accept the conditions to receive the newsletter or did not give their email address to the company. Exclusive price promotions for loyalty program members are incidentally given. During the observation period of this study the company did not give exclusive price promotions to their loyalty members.
Real life data about customer purchases and marketing instruments are available over four years (2012 - 2016). In this study an observation period of one year (2015 till 1-8-2016) is used, to keep the data manageable. The data consist of more than 2600 members, each members received a voucher during the observation period. Five vouchers are send out during the observation period, in week 32, 46, 52, 8 and 16. For calculating the CV, only the observation period of voucher 1 is used, week 32 t/m 46. Figure 3.1 gives an overview of the observation period.
Figure 3.1 Overview of observation period
3.2 Data overview
12
Table 3.1 Explanatory variables
Variable Description
Voucher related variables;
Voucher possessed Valid voucher in possessed in this week (0/1) Voucher redemption Voucher is used in this week (0/1)
Moderators;
Face value The face value of the voucher in euros. Minimum value of €10,-
Relationship length The length of the relationship between the member and the company in months. The length is calculated from the date that the member took part of the loyalty program until 1-8-2016
Geographic distance The distance between the house of the member and the store in km
Control variables;
Marketing-mix variables
Door-to-door flyer Whether the customer has a valid door-to-door flyer in possession in this week (0/1)
Newsletter Whether the customer receives a newsletter (0/1)
Household characteristics
Number of loyalty cards # of loyalty cards hold by the household
13
3.3 Control variables
To asses and clarify the relationship between the dependent and independent variable control variables are used. The control variable itself is not of primary interest in the study. The following control variables are used in this study:
- Marketing mix variables; different marketing variables are used to check whether the effect on purchase behaviour does not come from other marketing mix variables than the loyalty program reward. The marketing mix variables are: door-to-door flyers and newsletter send via email.
- Customer characteristics; in this study a lot of customer characteristics that normally are used as control variables, like age and gender, are not applicable because the data is on a household level. In section 3.1 there is already explained that each member can hold more than 1 loyalty card. As a customer/household characteristic the variable number of loyalty cards hold by the household is included as a control variable. - Time variable; purchase behaviour does strongly depend on time variables in retail
stores. In the home store branch during the summer holidays the sales decrease because most of the customers are on holiday. To control for this effect the time variable month is included.
3.4 Data cleaning
Before the data can be used for analysis it need to be cleaned based on missing values and outliers. The dataset consist of 2626 unique members. Four members where fake accounts, they were created by the company itself for different purposes. Those Member are deleted from the dataset. Five other members had very extreme values for the purchase quantity. Those values were four time bigger than the standard deviation (s.d. 482). The members with those extreme values are also deleted from the dataset. For the other variables there were no extreme outliers detected. In the end there are 2617 unique members left. In those 2617 members there are no missing values found.
3.5 Descriptive statistics
14
Table 3.2 Voucher redemption
The redemption behaviour of the members is graphically shown in figure 3.2. Most of the vouchers are redeemed in the first two weeks after receiving the voucher. In the following weeks the number of vouchers that are redeemed declined. But in week 13 and 14 a peak of redemptions has been observed. This can be explained by the validity of the voucher. A voucher has a validity of three months, which is 13/14 weeks. There are also some redemptions after three months (>14 weeks). This can happen because of the procedure that the company handles related to late redemptions of vouchers.
Figure 3.2 Redemption behaviour
A voucher can have different face values, depending on the number of points that are collected by the member. The average face value is €15.52 (s.d. 8.6). The range of the face value shows that 25% of the vouchers have a face value of €10.67 or lower and 25% of the vouchers have a face value of €16.50 or higher. The minimum face value is €10 and the highest face value during this observation period is €93.50.
Looking at the purchase behaviour of the members they on average visit the store 6,3 times per year. During those visits they spend on average €486 per year in the store, this is an
# Vouchers # cases received # cases redeemed
15 average of €74 per purchase incidence. When a member does not redeem a voucher the average purchase quantity is €73.70 (s.d. 122). When a member redeems a voucher during a purchase incidence the average increases to €79.40 per incidence (s.d.102). There should be mentioned that the face value of the voucher is included in the purchase quantity.
The average geographic distance is 12.2 km (s.d. 13). This means that most of the members live pretty close to the store. The member that lives closest to the store has a geographic distance of 0.2 km to the store, the maximum is 190 km.
Looking at the marketing variables the following conclusion can be drawn. Almost all members receive the door-to-door flyers at home. Only 3% of the members does not receive any flyer at all. During the observation period 22 flyers are send out. On average the customers received 11.6 (s.d. 7) flyers. The newsletters are received by only 17% of the members.
Chapter 4
Empirical model
In this chapter the empirical model that is used to test the hypotheses will be explained. The dataset of this study consist of a few thousand unobserved observations, selection bias, which makes that the data is censored. Because of this censored data, a normal OLS cannot be performed. A model that deals with this problem is called a Tobit model. The second type of this Tobit model is used because it gives the independent variables the opportunity to have different effects on purchase incidence and purchase quantity. For the panel data structure a correction is used, this correction is called XT. So purchase incidence and purchase quantity are both modelled with the use of a XT Tobit type II model.
4.1 Tobit model
16
Figure 4.1 Histogram purchase quantity
This problem with the selection bias is a well-known problem within the economic data, and is called censored or truncated data. Censored data leads to biased estimates when a regular OLS is used. This problems arise due to the measurement of the variables (for example, there are a lot of zeros within the dataset) or because the dataset does not capture all the information (for example, only a specific part of the population is included). To control for censored data, James Tobin created the Tobit model in 1958. The basic formula for a Tobit model is:
(3)
𝑦
𝑖= {
𝑦
𝑖∗
𝑖𝑓𝑦
𝑖∗
> 0
0 𝑖𝑓 𝑦
𝑖∗≤ 0
𝑦
𝑖∗= 𝛽𝑥
𝑖+ 𝜀
𝑖𝑦𝑖 = observed outcome variable of interest 𝑦𝑖∗ = latent variable
𝑥𝑖 = explanatory variables
𝛽 = parameters specifying relationshipbetweeen 𝑥 and 𝑦 𝜀𝑖 = error term
17 The estimation of the parameter of the model is done by the use of maximum likelihood method. This likelihood function consist of two parts:
1) Probit model; censored observations
For the censored observations, y = 0, it means that 𝑦𝑖∗ ≤ 0. Maximum likelihood estimation is done with the following formula:
(4) f (0│xi ) = Pr ( 𝑦𝑖∗ ≤ 0 │𝑥𝑖 ) = Pr (εi ≤ -x’iβ) = Pr
(
𝜀𝑖 𝜎≤ −
𝑥′𝑖𝛽 𝜎)
= Փ( −
𝑥′𝑖𝛽 𝜎)
= 1 - Փ(
𝑥′𝑖𝛽 𝜎)
f = Density Pr = Probability σ = Standard deviationՓ = Cumulative distribution function (cdf), probability that the standard normal is smaller than some number
2) Truncated regression model; uncensored observations
For the uncensored observations, y > 0, it means that 𝑦𝑖∗ > 0. This means that 𝑦𝑖∗ is equal to yi and because of that they have the same pdfs. This means that the ordinary
least square (OLS) method can be used to estimate the parameters. This OLS equals a normal linear regression.
(5) f (yi│xi ) = f ( 𝑦𝑖∗│𝑥𝑖 ) for all yi > 0 = Pr (εi ≤ yi - xiβ │𝑥𝑖) = Pr
(
𝜀𝑖 𝜎≤
𝑦𝑖− 𝑥𝑖𝛽 𝜎│𝑥
𝑖)
= Փ(
𝑦𝑖− 𝑥𝑖𝛽 𝜎)
F (εi) = 1 𝜎ϕ(
𝑦𝑖− 𝑥𝑖𝛽 𝜎)
Փ = Cumulative distribution function (cdf), probability that the standard normal is smaller than some number.
18 3) Likelihood function
Those two parts together generate the following likelihood function.
(6)
This likelihood function gives us the parameters for both the censored and uncensored data. In this version of the Tobit model, also called type I, the latent variable include both the process of participation, whether y = 0 or y > 0, and the outcome of interest. But for this study we have two different dependent variable. The first dependent variable, purchase incidence, is the selection criteria whether y is zero or bigger than zero. The other dependent variable, purchase quantity, is used to get the outcome of interest. The model that fits this interpretation is called Tobit model type II or Heckman correction. This model allows to let the participation process, and the outcome of interest, be independent.
4.2 Tobit model type II / Heckman correction
Heckman created this correction in 1976 to control for the selection bias that is explained in section 4.1. To control for this he used the inverse mills ratio; this is the ratio of the pdf to the cdf of a distribution. His two-step model consist of the following two steps;
1a. Probit model: this model is used to observe the selection process. So it determines whether y = 0 or y > 0. A negative outcome of y is not possible for a tobit model.
1.b Generate the inverse mills ratio: the estimations of the probit model are used to generate the inverse mills ratio. This is done by generating the lambda which is calculated by dividing the standard normal pdf by the standard normal cdf.
2. Linear regression: a linear regression is used to calculate the outcome of interest. This only done for the observations that are selected in step 1.
19 incidence would be the same. If a type II model is used the effects of the independent variables on purchase incidence and purchase quantity can be different. This gives us the opportunity to also research what influence is of purchase incidence, which is necessary to calculate the CV.
The formula that is used for a type II model is:
(7)
𝑦
1𝑖∗= 𝛽
1𝑥
1𝑖+ 𝜀
1𝑖,
𝑦
2𝑖∗= 𝛽
2𝑥
2𝑖+ 𝜀
2𝑖𝑦
2𝑖= 𝑦
2𝑖∗if
𝑦
1𝑖∗> 0
𝑦
2𝑖= 0
if𝑦
1𝑖∗≤ 0
This second type is a more flexible model than the type one model. But the model works best when exclusionary variables are added to both parts. This means that there need to be a difference in the independent variables that are added in both models. In this study the difference will be in the variables about voucher possession, voucher redemption and face value.
In Stata, the statistical package that is used for this study, the Heckman two-step model is included as a standard command. It models all the three steps by giving one command.
4.3 Panel data
20
Random or fixed effect
There are two different techniques that can be used for panel data, random and fixed effects. Fixed effects can be used if the interest of the study is only on the impact of the variables that vary over time. In a model in which random effects are used, also time invariant variables like gender can be included in the model (Torres-Reyna, 2007). In this study some of the variables are time invariant like geographic distance. That is why models with a random effect are used in this study.
4.4 Tobit model type II with panel data
The combination of a Tobit model type II and panel data is a combination that cannot automatically be made in any statistical package. This means that this need to be done by hand by the use of the steps that are explained in section 4.2. Instead of a normal probit and linear regression a panel data (xt) probit and linear regression are used.
Chapter 5
Results
This section presents the results of the model that is used for purchase incidence and purchase quantity. The effects of the moderators are also discussed in this section. Finally, the effect of a voucher on the CV is analysed.
5.1 Purchase incidence
Purchase incidence is analysed with a XT-probit analysis. Because the odds of a probit cannot be interpreted directly, the marginal effects of the probit model are also shown in table 5.1. The overall model is significant (χ2 = 336.71; p < .001) and has a log-likelihood of – 46189.4. A likelihood ratio test is applied to check whether the estimated model performs better than a null model (a model with no explanatory variables). The following formula is used:
( 8 ) Likelihood Ratio : -2 (Log Likelihood null – Log Likelihood estimated) -2 (-46746.055+46189.387) = 1113.3
21 Main effects
To measure the main effect the moderation variables are centred. For the variables relationship length and geographic distance this means that they are set equal to their average. The moderator face value is set equal to the minimum value, which is ten euro. The main effect, voucher possession, has as expected a positive and significant effect on purchase incidence (β1.0 = 0.0261 p < 0.01). This means that the purchase probability of customer, with an average
relationship length and geographic distance that has a valid voucher in possession with a face value of ten euro, increases with 2.61% in comparison of not possessing this voucher. The main effect of the moderator face value has a positive and significant effect on the purchase incidence probability (β2 = 0.0009; p < 0.01). This means that if the member has a valid voucher in
possession in a specific week, and the face value of that voucher would increase with one euro, the probability that the member will make a purchase increases with 0.09%. The moderator relationship length also shows a positive but not significant effect (β3 = 0.0005; p = 0.660)
which makes that it is impossible to make an interpretation for this effect. For the moderator geographic distance a negative but significant main effect is found (β4 = -0.0028; p < 0.01).
This means that if a member lives one kilometre further from the store, assuming that the member has a valid voucher in possession that specific week, the probability that the member purchase in that week decreases with 0.28%.
Moderation effects
The moderation effect of the moderator are all as expected, but are much smaller than the main effects. The moderation effect of face value (β1.1 = 0.0011; p < 0.001) and relationship length
(β1.2 = 0.0001; p < 0.001) are positive and significant. This means that the effect of voucher
possession on purchase incidence is positively influenced by those two variables. If the face value of the voucher or the relationship between the member and the company increases with respectively one euro or one month, the voucher will have a bigger effect on purchase incidence.
The moderation effect of geographic distance is negative and significant (β1.3 = -0.0003;
p < 0.001). So if the member is living further away from the store, the effect of voucher possession on purchase incidence decreases.
22
Table 5.1 Random effects xt probit model for purchase incidence
Coefficients Marginal effects P-value
Constant β0 -1.337 <0.001 Voucher possessed β1.0 0.145 0.0261 <0.001 Moderators: Face value β1.1 0.006 0.0011 <0.001 Relationship length β1.2 0.001 0.0001 <0.001 Geographic distance β1.3 -0.001 -0.0003 <0.001 Face value β2 0.005 0.0009 <0.001 Relationship length β3 0.003 0.0005 0.660 Geographic distance β4 -0.016 -0.0028 <0.001 Door-to-door flyer β5 -0.029 -0.0053 0.121 Newsletter β6 -0.002 -0.0004 0.940
Number of loyalty cards β7 0.118 0.0213 <0.001
Seasonal effects September β8.1 -0.095 -0.0171 <0.001 October β8.2 -0.139 -0.0251 <0.001 November β8.3 -0.089 -0.0159 <0.001 December β8.4 -0.068 -0.0122 0.002 January β8.5 -0.003 -0.0006 0.890 February β8.6 -0.020 -0.0036 0.407 March β8.7 -0.176 -0.0318 <0.001 April β8.8 -0.192 -0.0348 <0.001 May β8.9 -0.112 -0.0202 <0.001 June β8.10 -0.232 -0.0418 <0.001 July β8.11 -0.144 -0.0260 <0.001 Control variables
Both marketing mix variables do not show a significant effect. This can be partially explained by the fact that for the variable door-to-door flyer there is too little variation within the variable. Almost everybody has each week a valid flyer in possession. This makes it impossible to see what the effect is of this variable. For the variable newsletter the opposite applies. From the total dataset, with 2617 unique members, only 449 members receive the newsletter each week. This can be the reason that the effect, if it exist, is not captured in this dataset.
The variable number of loyalty cards has a positive and significant effect (β7 = 0.0213; p <0.01).
23 household has. So the higher the chance that somebody of a household buys something at the store. The month dummies show clear seasonal fluctuations.
5.2 Purchase quantity
Purchase quantity is analysed with a XT-regression analysis, with an additional variable inverse mills ratio to control for the selection bias. Purchase incidence is used as a selection variable. This means that only for the observations that had a value that was bigger than zero, for the variable purchase incidence, a regression analysis is performed. The estimates of the xt-regression are shown in table 5.2. To check whether this overall model performs better than the null model the likelihood ratio test, that also is performed in section 5.1, is used.
( 9 ) Likelihood Ratio : -2 (Log Likelihood null – Log Likelihood estimated) -2 (-102106.74+101935.23) = 343.02
This likelihood ratio needs be higher than the critical value of a chi-square test. For this test a degrees of freedom of 19 (number of parameter in the estimated model) is used. The critical value is 36.191 for a probability level of 0.01. The likelihood ratio is higher than the critical value which means that the estimated model is significantly better than the null model.
Main effect
The main effect, voucher redemption, has an negative and significant effect on purchase quantity (purchase quantity is inclusive the face value of the voucher that is redeemed) (β1 =
24 during purchase incidence that are not planned ahead. This means that the average purchase quantity for a purchase incidence in which a voucher is redeemed will be lower than the normal average purchase quantity. The main effect of the moderator face value is positive and significant (β2 = 1.27; p < 0.01). This means that if the face value of the voucher increases with
€1 the purchase quantity increase with €1.27. Because the face value of the voucher is included in the purchase quantity we need to subtract the additional €1 of the face value from the purchase quantity. But this still means that a higher face value stimulates the member to spend more than only the additional face value of the voucher. For the moderator relationship length the main effect is negative and significant (β3 = -0.49; p < 0.01). This means that the longer the
member has a relationship with the company the lower the purchase quantity in a week will be. For geographic distance it holds that the effect is positive and significant (β4 = 1.91; p < 0.01).
This means that each additional km will let the purchase quantity rise with €1.91. So members who live further from the store are spending more during each purchase incidence.
Moderation effects
The moderation effect of the moderator are not all as expected. The moderation effect of face value (β1.1 = 1.37; p < 0.001) and relationship length (β1.2 = 0.16; p < 0.042) are, as expected,
positive and significant. This means that the effect of voucher possession on purchase incidence is positively influenced by those two variables. If the face value of the voucher or the relationship between the member and the company increases with respectively one euro or one month, voucher redemption will have a bigger effect on purchase quantity. The moderation effect of geographic distance is negative and significant (β1.3 = -0.76; p < 0.001) while a positive
effect was expected. If the member is living further away from the store, the effect of voucher redemption on purchase quantity decreases.
Control variables
The marketing mix variables show both a not significant effect, the same like it was in section 5.1. It can be assumed that the reason for this not significant effect can be explained with the same reason that is used in section 5.1. The number of loyalty cards has a negative and significant effect (β7 = -4.12; p = 0.02). This means that the number loyalty cards that a
25 observations is only 16.278 (because of the selection that is made, purchase incidence > 0), which is a very limited number to draw conclusion on.
Table 5.2 Random effects xt regression model for purchase quantity
Coefficients P-value Constant β0 146.42 <0.001 Voucher redeemed β1.0 -18.69 0.001 Moderators: Face value β1.1 1.37 <0.001 Relationship length β1.2 0.16 0.042 Geographic distance β1.3 -0.76 0.001 Face value β2 1.27 <0.001 Relationship length β3 -0.39 <0.001 Geographic distance β4 1.91 <0.001 Door-to-door flyer β5 -2.47 0.294 Newsletter β6 5.52 0.138
Number of loyalty cards β7 -4.12 0.139
Seasonal effects September β8.1 -5.17 0.235 October β8.2 -6.62 0.182 November β8.3 -1.56 0.725 December β8.4 -1.41 0.735 January β8.5 -0.58 0.895 February β8.6 -6.39 0.156 March β8.7 -5.63 0.243 April β8.8 -1.83 0.727 May β8.9 -12.60 0.008 June β8.10 -8.94 0.103 July β8.11 -15.20 0.003 Mills Β9 -20.03 0.001
Inverse mills ratio
The inverse mills ratio is added to the analysis to control for the selection bias, explained in section 4.1. The mills ratio has a negative and significant effect (β9 = -20.03; p = 0.001). The
26 negative values of the inverse mills ratio means that when OLS would be used the estimates would be downwardly biased.
5.3 Multicollinearity
Multicollinearity exist when two or more explanatory variables in a regression model are highly correlated. This can bias the outcomes of the regression which results in a limitation of the conclusions that can be drawn form a regression (Malholtra, 2012). To check for multicollinearity a correlation matrix can be made or the variance inflation factor (VIF) scores can be calculated. When the score of the VIF is above 10 there is high multicollinearity within the dataset. (Kutner, 2004).
If Stata is used as statistical package, the check for multicollinearity is already done by the software. If the software finds high multicollinearity within a model, the variable that ensures the multicollinearity will be deleted from the model. For the model that is used in this study no variables where deleted. But to be sure, the VIF scores are checked by hand. Indeed only the VIF score of the mills ratio (VIF = 24.7) is above 10. This variable is not deleted from the model because it is logical that this variable has a high VIF score. It is calculated based upon the variables that are used in the probit model, and 80% of that variables are equal to the variables that are used in the regression model.
5.4 Vouchers and customer value
The empirical model estimates make it possible to calculate the voucher effect on CV. The CV formula is described and explained in section 2.1, but to remember it again it is also written down here; ( 10 ) CV𝑖𝑡 = ∑ PI𝑖𝑡∗ PQ𝑖𝑡∗ 𝑚 − VO𝑖𝑡 ∗ 𝑣𝑚 − DCnv𝑖𝑡 (1 + 𝑑)𝑡 13 𝑡=0
27
Preference rate, margin and mailing costs
Previous studies on CV used a discount rate between 12% and 20% (Berger and Nasr, 1998). For this study a time discount rate of 15% on annual level is taken. This means an weekly rate of 0.00269 (r = 1 - 52√1.15 = 0.00269). The margin that is used in this study is equal to 40% which is the average margin for mixed furnishing stores as reported by the Centraal Bureau Statistiek (detailhandelinfo, 2016). The mailing costs are set at €1.
Purchase incidence
By calculating purchase incidence all variables are set equal to the mean, except voucher possession. This variable has two different values 0/1. For both values the purchase incidence is calculated in the following way. In the first place a probit model with panel data is performed, based on this model the coefficients are accessed. Those coefficients are required to calculate the predicted probabilities, for which the formula is:
( 11 ) pi = F (xi’* beta)
F = cumulative normal distribution
xi = the data vector for the i-th observation
beta = vector of coefficient estimates.
Those predicted probabilities give the opportunity to calculate the purchase incidence for different values of x. In this study only the values of the variable voucher possession are changing (0/1), the rest was set equal to the mean. The formulas and commands that are used to calculate the purchase incidence can be found in the appendix.
Purchase quantity
To calculate the purchase quantity all variable are set equal to the mean except voucher redemption and face value. For a linear regression it holds that the formula can be filled in with the estimates that are derived in the regression analysis and the values of the variables that want to be studied. The total formula for this model is also attached in the appendix.
With the use of all these variables, the voucher effect is calculated for seven different situations. 1. No voucher is received
28 4. A voucher is received but not redeemed and has a face value of €20,-
5. A voucher is received, is redeemed and has a face value of €10,- 6. A voucher is received, is redeemed and has a face value of €15,- 7. A voucher is received, is redeemed and has a face value of €20,-
The outcomes are shown in table 5.3. It seems that giving out vouchers has a positive effect on the profitability of the firm. A member who receives a voucher is in 13 weeks almost 24 euros more worth than a member who did not receive a voucher. If the face value of the voucher, that the member has in possession, increases, the member will also be worth more. However the conclusion that a higher face value also lead to higher firm profitability cannot be made. If the minimum face value of the voucher increases from 10 euro to 15 euro the members who receive a voucher will be worth more. But it also means that it takes more time for the members to collect enough points. So there are more members who do not receive a voucher. Which means that there are members who now have a lower CV than they probably would have before. So if a decision needs to be taken about the minimum face value of the voucher, those two effects need to be weighted out against each other.
The table also shows a negative effect of redeeming the voucher. The reason for this negative effect is already explained in section 5.3. The average purchase quantity if a member redeems a voucher is lower than if they do not redeem a voucher because of the type of purchase that is done in both purchase incidences.
Table 5.3 The effect of voucher on customer value
1 For one week 2 In euros
3 This is the customer value per customer calculated over 13 weeks
Voucher possession Voucher redemption Face value1 Purchase incidence Purchase quantity1 CV 2, 3
Voucher not obtained 8.8% 3.51 44.32
29
5.5 Summary of results
In figure 5.2 the results are summarized in the conceptual model. Hypothesis H1, H3a, H3b, H4a, H5a and H5b are all supported, they are shown in the model with bold figures. The other two hypothesis, H2 and H4b, are not supported.
Figure 5.2 Conceptual model with results
* p < 0.001 ** p < 0.05
Chapter 6
Conclusions and recommendations
A considerable amount of literature is published in the past about the effectiveness of reward systems from loyalty programs (Yi and Jeon, 2003; Dréze and Nunes, 2004; Kivetz and Simonson, 2002). In this study a delayed reward system is research. Yi and Jeon (2003) also researched this system and they found that a delayed reward system is more effective than an immediate reward system. This study extends on previous research by examining the effect of delayed reward system, in the form of a loyalty voucher, on the customer value. CV is a metric that is commonly used by retailers to measure company profitability. In this chapter the findings of the study are discussed, the limitation are considered an implications and directions for further research are presented.
6.1 Conclusion
To make it possible to write the overall conclusion of the study, the effect of delayed reward system om CV, the conclusions of the subquestions need to made up first.
Customer lifetime value
30 purchase quantity, the margin and subtracting the costs. To measure what the effect of the voucher is on the CLV, the effect of this voucher is modelled on purchase incidence and quantity. Because this study only measures the effect of the voucher on the period in which one voucher is valid, three months, and not include the retention probability the term CV instead of CLV is used.
Purchase incidence
The results of this model are in line with the expected results that are based on previous researches. Having a voucher in possession has a positive and significant effect on purchase incidence (H1 is supported). This finding is in line with previous research of Bawa and Schoemaker (1987). They claimed that receiving a voucher will let members purchase more often. The assumption that face value has a positive and significant moderation effect is supported by the results (H3a is supported). A higher face value of the voucher will let member purchase more often. For the moderation effect of geographic distance it holds that the effect is negative and significant (H4a is supported). This means that if a member is living further away from the store and holds a voucher the purchase incidence probability decreases. The moderation effect of relationship length is positive and significant (H5a is supported). The longer the member takes part of the loyalty program the more effective the voucher is.
Purchase quantity
31 The assumption that face value has a positive and significant moderation effect is supported by the results (H3b is supported). A higher face value of the voucher will let member purchase more, even when the additional face value is subtracted from the purchase quantity. For the moderation effect of geographic distance it holds that the effect is negative and significant (H4b is not supported). This means that if a member is living further away from the store the voucher effectiveness decreases. The moderation effect of relationship length is positive and significant (H5b is supported). The longer the members is member of the loyalty program the more effective the voucher gets.
Voucher effect on CV
The empirical model estimates makes it possible to calculate the voucher effect on CV. The CV is calculated for seven different situations; Whether there is a voucher received or not, if this voucher is redeemed or not and what the face value of the voucher was. The conclusion is that receiving a voucher let the CV, per customer over 13 weeks, increase extremely. So giving out vouchers as a company will have positive effect on the profitability. If a voucher is received the face value also have influence on the CV. The higher the face value, the higher the CV. If a voucher is redeemed or not also influence the CV but negatively. This negative effect is caused by the negative effect of voucher redemption to purchase quantity. But it must be said that this negative effect is very small, one euro over 13 weeks.
6.2 Limitations
This study has some limitations that need to be taken into account. First of all, the model that is used, a type II Tobit model with panel data, is a model that, until now, not often is used for econometric data. There are no previous researches that used this type of model before, and therefore there is also no standard command available. The only possible way to estimate the model is by doing it by hand. This could have an influence on the results of the model, because there is no way to control if everything is done in the right way.
32 Finally, the effect of voucher possession on purchase quantity is not captured in the model. This is done because the tobit type II model performs best if exclusionary variables are added in both parts of the model. In this study is decided that voucher possession, because of this reason, could not be added in the regression model.
6.3 Implications
This study is done in collaboration with a construction and furnishing store to see if their loyalty program is still effective. With the results of this study the conclusion can be made that indeed giving out vouchers is still profitable for the company. On this moment the minimum face value of the voucher is €10. A higher face value gives a higher CV, but if the company increases the minimum face value to for example €15, there need to be taken into account that is takes more time for members to receive the number of points before they receive a voucher at home. This means that on average less vouchers will be given out each time. This also means that there are less members that have a valid voucher in possession, which will lower the profitability of those customers. These two things need to be weighed against each other. Figure 6.1 is a visualization of the situation that is described above.
Figure 6.1 Visualization of increasing minimum face value
33 If the minimum face value becomes €15 instead of €10, the number of vouchers that need to be given out, to get the same amount of CV (break-even point), should be at least 85% of what would be given out if the minimum face value was €10. This means that if 500 vouchers are given out before, the number of vouchers that need to be given out in the new situation, to get the same CV, is 425 vouchers. If there are given out more than 425 vouchers this makes the new situation more profitable than the old one. If the face value is €20 instead of €10, the factor is 73%. This means that if 500 vouchers are given before, the number of vouchers that need to be given out in the new situation to get the same CV is 365 vouchers. The dataset that is used for this study does not give the opportunity to check what would happen with the number of vouchers that would be given out if the minimum face value increases. To get an answer to this question other data, about the total number of points collected by the members, need to be researched. This research need to check how much members would receive a voucher if the minimum value is €15 or €20 instead of €10. In addition, the company can try a new minimum value for one year and do the same research again to check if the total CV of all members increases or not. If the company is going to start this test, the advice is to use a minimum face value of €15 instead of €20 because the average face value on this moment is around €15 and 75% of the vouchers have a face value that is lower than €16.50. The probability that the 73% will be reached if the minimum face value is €20, is very small.
6.4 Directions for further research
34
References
Allison, P. D. (2010). Definition of the Event. The Reviewer’s Guide to Quantitative Methods
in the Social Sciences, 413.
Arnold, S., Oum, T., & Tigert, D. (1983). Determinant attributes in retail patronage: seasonal, temporal, regional and international comparisons. Journal of Marketing Research, 20 (2), 149 – 157.
Baltagi, B. (2012). Econometric analysis of panel data. Hoboken, USA: John Wiley & Sons. Barat, S., & Paswan, A. K. (2005). Do higher face-value coupons cost more than they are worth in increased sales?. Journal of Product & Brand Management, 14(6), 379-386.
Bawa, K. (1996). Influences on consumer response to direct mail coupons: an integrative review. Psychology & Marketing, 13(2), 129 – 156.
Bawa, K., & Shoemaker, R. W. (1987). The effects of a direct mail coupon on brand choice behavior. Journal of Marketing Research, 370 – 376.
Bell, D.E., & Lal, R. (2003). The impact of frequent shopper programs in grocery retailing. Quantitative marketing and economics, 1(2), 179 – 202.
Bian, H. (2015, May 19). Survival analysis using SPSS. Retrieved from http://core.ecu.edu/ofe/StatisticsResearch/Survival%20Analysis%20Using%20SPSS.pdf. Bijmolt, T. H., Dorotic, M., & Verhoef, P. C. (2010). Loyalty programs: generalizations on their adoption, effectiveness and design. Foundations and Trends in Marketing, 5(4), 197-258. Bond Brand Loyalty. (2015). The 2015 Loyalty Report. Retrieved from http://info.bondbrandloyalty.com/the-loyalty-report-2015.
Capgemini. (2013). Trends in retail 2013-2014; Band met de klant. Retrieved from https://www.nl.capgemini.com/resource-file-access/resource/pdf/trends_in_retail_report.pdf. Centraal Plan Bureau. (2016, March 21). Centraal economisch plan 2016. Retrieved from http://www.cpb.nl/sites/default/files/omnidownload/Centraal-Economisch-Plan-2016.pdf. Detailhandel.info. (n.d.). Gemengde woonzaken. Retrieved December 15, 2016, from http://detailhandel.info/index.cfm/branches/woninginrichting/gemengde-woonzaken/.
DeWulf, K., Odekerken-Schröder, M. H., & van Oppen, C. (2003). What drives Consumers Participation to Loyalty Shcemes? A Conjoint Analytical Approach. Journal of Relationship
Marketing, 2(1-2), 69 – 83.
35 Dréze, X., & Nunes, J.C. (2004). Using combined-currency prices to lower consumers’ perceived cost. Journal of Marketing Research, 41(1); 59 – 72.
GfK. (2002). GfK Jaargids 2002: kerncijfers voor marketing- en beleidsplannen.
Gupta, S. (1988). Impact of sales promotions on when, what, and how much to buy. Journal of
Marketing Research, 25(4), 342 – 355.
Gupta, S., Hanssens, D., Hardie, B., Kahn, W., Kumar, V., & Sriram, S. (2006). Modeling customer lifetime value. Journal of service research, 9(2), 139 – 155.
Haisley, E., & G. Loewenstein. (2011). ‘It’s not what you get but when you get it: The effect of gift sequence on deposit balances and customer sentiment in a commercial bank’. Journal of
Marketing Research, 48(1), 103–115.
Jain, D. C., & Singh, S. S. (2002). Customer lifetime value research in marketing: a review and future directions. Journal of Interactive Marketing, 16(2), 34 – 46.
Kahn, B. E., & Schmittlein, D. C. (1992). The relationship between purchases made on promotion and shopping trip behaviour. Journal of Retailing, 68(3), 294 – 315.
Kivetz, R., & Simonson, I. (2002). Earning the right to indulge: effort as a determinant of customer preferences towards frequency program rewards. Journal of Marketing Research,
39(2), 155 – 170.
Krishna, A., & Shoemaker, R. W. (1992). Estimating the effects of higher coupon face values on the timing of redemptions, the mix of coupon redeemers, and purchase quantity. Psychology
and Marketing. 9(6), 453 – 467. .
Kumar, V., & Pansari, A. (2016). National Culture, Economy, and Customer Lifetime Value: Assessing the Relative Impact of the Drivers of Customer Lifetime Value for a Global Retailer. Journal of International Marketing, 24(1), 1 – 21.
Kutner, M. H., Nachtsheim, C. J., & Neter, J. (2004). Applied Linear Regression Models. New York, USA: McGraw-Hill.
Liu, M. T., & Brock, J. L. (2009). Antecedents of redemption of reward points: Credit card market in China and international comparison. Journal of International Consumer
Marketing, 22(1); 33-45.
Lee, J. K. H., & Steckel, J. H. (1999). Consumer strategies for purchasing assortments within a single product class. Journal of Retailing, 75(3), 387 – 403.
Leenheer, J., Van Heerde, H.J., Bijmolt, T.H.A., & Smidts, A. (2007). Do loyalty programs enhance behavioral loyalty: an empirical analysis accounting for self-selecting members.
International Journal of Research in Marketing, 24(1), 31 – 47.
Leenheer, J. (2004). The adoption and effectiveness of loyalty programs in retailing. Thesis,
36 Leone, R.P., & Srinavasan, S. S. (1996). Coupon face value: its impact on coupon redemptions, brand sales, and brand profitability. Journal of Retailing, 72(3), 273 – 289.
Leone, R.P. & Mulhern, F.J. (1991). Implicit Price Bundling of Retail Products: A multiproduct approach to maximizing store profitability. Journal of Marketing, 55(October); 63-76.
Lewis, H. G. (1997). Does your loyalty program inspire any loyalty? Direct Marketing, 59(6), 46 – 48.
Mägi, A. W. (2003). Share of Wallet in Retailing: the Effects of Customer satisfaction, Loyalty Cards and Shopper Characteristics. Journal of Retailing, 109(2), 1-11.
Meyer – Waarden, L. (2007). The effects of loyalty programs on customer lifetime duration and share of wallet. Journal of Retailing, 83(2), 223 – 236.
Mick, D. G., & Faure, C. (1998). Consumer self-gits in achievement contexts: The role of outcomes, attributions, emotions and deservingness. International Journal of Research in
Marketing, 15(4), 293 – 307.
Parvizian, M., Kianersi, M. E., & Albadvi, A. (1995). How perceived service quality and customer loyalty can affect customer lifetime value? International Business Management,
10(10), 1973 – 1988.
Soderbom, M. (2009, January 16). Estimation of Binary Choice Models with Panel Data. Retrieved from https://www.nuffield.ox.ac.uk/users/bond/binarychoice_notes_mans.pdf Scribney, W. (n.d.). Predict probabilities after probit. Retrieved from http://www.stata.com/support/faqs/statistics/predicted-probabilities-after-probit/
Smith, A., & Sparks, L. (2009a). It’s nice to get a wee treat if you’ve had a bad week: Consumer motivations in retail loyalty scheme points redemptions’. Journal of Business Research, 62; 542 – 547.
Smith, A., & Sparks, L. (2009b). Reward redemption behaviour in retail loyalty schemes’.
British Journal of Management, 20, 204 – 218.
Taylor, G. A., & Neslin, S. A. 2005. The current and future sales impact of a retail frequency reward program. Journal of Retailing, 81(4), 293-305.
Torres-Reyna, O. (2007). Panel Data Analysis Fixed and Random Effects using Stata. Retrieved from https://www.princeton.edu/~otorres/Panel101.pdf
