The Effect of Continuously Rewarding Customers in a Frequency Reward
Program: An Assessment Based on the World Championship Soccer Reward
Programs
1.
August 22, 2011 Alec Minnema
Supervisors: Tammo H.A. Bijmolt and Mariëlle C. Non
Master thesis, Faculty Economics and Business, University of Groningen. A.Minnema.1@student.rug.nl
Abstract
This research presents the first empirical study on the effectiveness of Frequency Reward Programs (FRPs) with continuous reinforcement schedules, reward benefit incongruence and collaboration with manufacturers (i.e. offering extra premiums at some products). We examine FRPs at five different supermarket chains using a market wide household panel. The results show that the effectiveness of the examined FRPs on spending level is higher than in previous studies on FRP effectiveness. However, we demonstrate that the effect on spending level vanishes if the FRP is ceased which contradicts literature. Furthermore, we find that the FRP has no significant effect on non-participating households. This study corroborates the negative moderation of Share of Wallet (SOW) on FRPs effectiveness, FRPs are most effective in increasing spending levels for households with a small SOW. A second significant positive moderator is the appreciation of the FRP. Moreover, the results indicate that participating in competing FRPs does not have a significant effect on FRPs effectiveness at the focal retailer. We find strong evidence for self-selection since participating households visiting frequency and spending level at the focal supermarket are higher before the start of the FRP. Households who participate in competing FRPs visit the supermarket less frequent and spend a smaller amount at the supermarket in the period before the start of the FRP. Finally, we show that offering an extra premium with a product (i.e. supply funded premiums) positively affects the purchased amount of the sponsored products and this effect does not differ significantly between the examined product categories.
Keywords: Frequency Reward Programs; Grocery Retailing; Tobit-II model; Free Gifts; Retailer Manufacturer Collaboration.
1
1. Introduction
In the continuous competition to draw consumers’ attention, retailers renewed Frequency Reward Programs (FRPs). During the summer of 2010, Dutch supermarkets presented their customers FRPs related to the World Championship Soccer with three new characteristics: rewards only have hedonic value; customers are continuously rewarded; and collaboration between supermarket and manufacturers in the FRP. To illustrate, the FRP at the Plus supermarket chain is outlined: Customer received a free wristband in the colors of a participating country in the World Championship soccer for every 15 euro spend. Furthermore, manufacturers were offered the opportunity to participate in the costs of the FRP and get in return their product in promotion with an extra reward for the customer (e.g. An extra wristband with two packs of dry soup).
To the best of our knowledge, it is not possible to infer the effectiveness of FRPs similar to the FRP at Plus described above from literature. Current research in FRPs focuses on programs with utilitarian benefits. These programs offer participating customers rewards like a free stay at a hotel (Kopalle et al. 2011), a turkey (Taylor and Neslin 2005) or monetary rewards (Liu 2007). In this study, we will explore the effectiveness of a FRP that offers consumers hedonic rewards at the supermarket. This might affect FRPs effectiveness because of benefit incongruence (Chandon, Wansink and Laurent 2000). Benefit congruence denotes that the benefits of the reward must correspond with those sought for the purchase occasion (i.e. hedonic reward with hedonic purchase occasion). This potentially decreases FRP effectiveness because hedonic rewards are obtained at purchase occasions where mainly utilitarian benefits are sought.
In previous studies on FRP effectiveness customers are rewarded after reaching a specific threshold, where the threshold is normally not met during one purchase occasion. However, a new trend exists wherein a reward is obtained after every purchase. This difference in rewarding will affect FRPs effectiveness. A continuous reinforcement schedule (i.e. after every purchase a reward is given) is very effective in learning the desired behavior but it will not result in long-term changes in behavior if the reinforcement is ceased. In contrast, in a fixed ratio reinforcement schedule (i.e. reward after every x times desired behavior occurs) the behavior is learned gradually but retains more persistent after the FRP is ceased (Huczynksi and Buchanan 2007, p.113). This indicates that FRPs with continuous reward schedules compared to FRPs previously studied, are more effective throughout the FRP but are less effective in creating a persistent effect. Therefore, FRPs with continuous reward schedules long term impact (i.e. reward behavior) is undetermined regardless the demonstrated positive long-term impact of FRPs with fixed ratio reward schedules in the study of Taylor and Neslin (2005).
As pointed out by Ailawadi et al. (2010) collaboration trough FRPs can provide benefits to both retailer and manufacturer. In the studied FRPs, manufacturers collaborate with supermarkets to get their product in promotion with an extra reward in return (i.e. supply funded premiums). Zhang and Breugelmans (2011) show that it is more effective to give extra LP points at some product than to give a monetary reward with identical value. However, LP points have utilitarian value while the premiums have hedonic value. Furthermore, offering products with free gifts is not without risk. Simonson, Carmon and Curry (1994) show that offering a premium that is unattractive negatively affects sales and the reputation of the product. In addition, Raghubir and Celly (2011) prove that a large picture of the free gift in the advertisement deteriorates the perceived product quality. However, Lee-Wingate and Corfman (2010) show that when the premium is intended to be enjoyed by someone else (e.g. children) consumers will buy more. Therefore, supply funded premiums effectiveness is undetermined from previous research.
in the following manner: Section 2 elaborates on the conceptual framework of the relation between FRPs and purchase behavior. In Section 3 the data used for our empirical study is discussed while in section 4 we propose the econometric models. Section 5 presents the results and in Section 6 we validate the results of the rewarded behavior effect. Finally, the concluding remarks will be made in Section 7.
2. Conceptual framework
Decomposition of purchase behaviorGrocery retailing is characterized by very frequent buying and most consumers buy at different stores. The latter is often called polygamous loyalty (Dowling and Uncles 1997). FRPs are intended to make consumers more loyal to one retailer by means of creating switching barriers for the customer (Leenheer et al. 2007). Switching barriers are created because customers lose value when they terminate the relation with the retailer (i.e. have an incomplete collection of rewards). By means of creating switching barriers, pressure on retailers to decrease prices diminishes (Kim, Shi and Srinivasan 2001). For retailers it is top priority to make their costumers more loyal as it very difficult to differentiate from competition. The aim of FRPs is to increase loyalty of the households and therefore the ultimate outcome variable is loyalty towards the supermarket. Henderson et al. (2011) define customer loyalty as “Repeat seller specific consumption that may result from desires to self-enhance, learned memory advantages or forward looking desires to maintain a strong relationship based on trust and reciprocity with an organization”. Therefore, two important behavioral variables that characterize the behavioral loyalty towards supermarkets will be used; the frequency of visiting the supermarket (traffic) and the amount spent at the supermarkets. Consequentially, the hypotheses and conceptual framework will focus on the antecedents of both traffic and the spending level at the supermarkets.
Effectiveness of Frequency Reward Programs on purchase behavior
The existence of a FRP at the supermarket might attract customers to visit the supermarket even if they do not participate in the FRP. FRPs are largely advertised which might positively influence the place of the supermarket in the consumer’s consideration set (Shapiro, MacInnis and Heckler 1997). Contrary, studies on exclusive promotions indicate that some excluded groups respond negatively to exclusive promotions (Barone and Roy 2010). In addition, Henderson et al. (2011) argue that the effect on excluded costumers might be negative; exclusion might appeal feelings of unfairness. Furthermore, Samaha, Palmatier and Dant (2011) show that perceived unfairness works as poison for the long-term relationship. The same may hold for FRPs and therefore the net effect of these opposing effects is undetermined and no a prior expectations will be stated both for traffic and spending level.
When a consumer participates in a FRP, a point pressure effect is expected (Taylor and Neslin 2005). Previous empirical research shows the effectiveness of FRPs on purchase behavior (Lal and Bell 2003; Taylor and Neslin 2005). Point pressure occurs when a consumer wants to complete a set and therefore will take each opportunity to collect rewards. This results in a positive expectation of the effect of the FRP on purchase behavior. However, the reverse holds if a consumer participates in a competing FRP, then point pressure will decrease the likelihood of shopping at the focal supermarket. This leads to the following hypotheses:
H1: Participating in a FRP has a positive effect on (a) traffic and (b) the spending level at the focal supermarket. H2: Participating in a competing FRP has a negative effect on (a) traffic and (b) the spending level at the focal supermarket.
When positive affect towards the retailers causes persistent improvement in spending level, no difference should be observed between FRPs with continuous versus fixed ratio reinforcement schedules. However, when learned behavior is the cause for the rewarded behavior, the effect should be stronger for fixed ratio reinforcement schedules compared to continuous reinforcement schedules. Therefore, it is proposed that the rewarded behavior effect is smaller in our research setting as compared to Taylor and Neslin (2005) study.
H3: Participating in a FRP has a positive effect on (a) traffic and (b) spending level at the focal supermarket after the program is ceased.
The moderating effects on FRP participation effectiveness
The effectiveness of participation in a FRP will be moderated by four variables: The appreciation of the FRP, SOW, the time since the introduction of the FRP and participation in competing FRPs. The appreciation of the FRP affects the effectiveness of the FRP on buying behavior. With a higher appreciation, the point pressure effect of the FRP is stronger and the rewards from the reinforcement schedule are higher appreciated which improves the likelihood of visiting and raises the expected purchase amount. This leads to the following hypotheses:
H4: A higher appreciation of the FRP enhances FRP effectiveness on (a) traffic and (b) the spending level.
The effect of the FRP on purchase behavior for a consumer partly depends on the ability to increase the SOW at the focal supermarket. Lal and Bell (2003) find that FRPs are most effective on consumers with a small SOW in the period prior to the introduction of the FRP. Therefore, we hypothesize that the FRP is less effective for households with a high SOW, because they are unable to increase their purchase level further due to a ceiling effect (Du, Kamakura and Mela 2007). Kim et al. (2009) support this finding by showing that a VIP program is most effective for consumers who spent a small purchase amount before the start of the program.
H5: FRPs effectiveness on (a) traffic and (b) the spending level is negatively affected by the SOW at the focal supermarket.
The time since the start of the FRP will likely influence the effectiveness of the FRP on purchase behavior. Kivetz, Urminsky and Zheng (2006) find that the point pressure effect increases as consumers get closer to the reward. They call this purchase acceleration the goal gradient hypothesis. For the FRPs studied, it is undetermined if the goal gradient hypothesis will hold. The consumer will be rewarded after every X euro spend regardless how close they are to the completion of the set, so part of the reward will be obtained independent of the distance to the end goal. A second reward is obtained when the set is complete because of the need for set completion (Carey 2008). Contrary, Pauwels (2004) shows that price promotions have no lag before reaching the peak in impact and for advertising this is only 1.2 weeks. Because the FRPs of interest have both characteristics of a promotion and of a traditional reward program the effect is undetermined. In sum, we hypothesize a positive moderation of the time since introduction on the effectiveness of the FRP on purchase behavior.
H6: The effectiveness of participating in a FRP on (a) traffic and (b) the spending level becomes stronger during the time-span of the FRPs.
H7: The effectiveness of participating in a FRP on (a) traffic and (b) the spending level decreases if the household participates in a competing FRP.
As described in the previous section manufacturers sponsored the FRP to have their product in promotion with an extra premium. The type of premium used is a near pack direct premium (D’Astous and Jacob 2002), which means that the premium is obtained directly and that it is not fixed on the product. D’Astous and Jacob (2002) conclude that the effect of this type of premium is likely to result in a positive appreciation and that the appreciation further improves if the consumer is interested is in the premium. However, previous research shows that premiums can backfire if they are not appreciated by the consumer (Simonson et al. 1994), the value of the premium is too high (Chung-Hui 2010) or if there is too much focus on the premium (Raghubir and Celly 2011). Lee-Wingate and Corfman (2010) show that the effectiveness of premiums increases if they are intended to be used by someone else (e.g. children). Nevertheless, because of the relative high appreciation of this type of premiums, its low value and the focus on children we expect a positive effect of supply funded premiums on sales.
H8: Having a supply funded premium will improve the sales of the product.
We expect that the effectiveness of the supply funded premium depends on other variables. The first moderator will be the supermarket chain. This because there is heterogeneity between the FRPs offered (e.g. the difficulty to complete the set) and heterogeneity between the supermarket’s customers (i.e. the proneness to participate in a FRP). Therefore, we hypothesize that the effectiveness differs significantly between the supermarkets and is the strongest at Dirk (i.e. highest participation degree in a FRP, most difficult to complete a set).
H9a: The effectiveness of the supply funded premium differs significantly between supermarkets.
A second moderator is the difference in effectiveness between the product categories. Similar to sales promotions, the effectiveness of the supply funded premiums differs between categories. Pauwels, Srinivasan and Franses (2007) show that price elasticity is lower in categories with longer purchase cycles and is higher for products in more expensive categories. Secondly, Lee-Wingate and Corfman (2010) argue that premiums intended to be used by someone else are more effective in hedonic categories. Therefore, we expect structural differences in effectiveness between the product categories.
H9b: The effectiveness of supply funded premiums differs significantly between product categories.
A third moderator of the effectiveness of supply funded premiums is the week in which the product is promoted. We expect an effect similar to FRP effectiveness of the FRP, thus the effectiveness increases over time.
H9c: The effectiveness of supply funded promotions becomes stronger during the time-span of the FRP The effects of store and household characteristics on purchase behavior
Figure 1: Conceptual framework of choice to buy at a supermarket
Note: In the figure a dashed line connotes a negative effect and a solid line a positive effect.
3. Data
The data used in this study is obtained from a household panel that registers all purchases at Dutch supermarkets. The Dutch supermarket industry is interesting to study FRPs due to the intense competition and the large number of FRPs. At the dawn of the summer of 2010 before the World Championship soccer, many Dutch supermarkets offered a FRP related to the World Championship soccer. Supermarket purchase data is available from a large Dutch panel of 2308 households. The panel members scan all their purchases, which give insights in the SOW of supermarkets and their spending level. The data covers 20 weeks in 2010 surrounding the period of the World Championship soccer FRPs. The purchases at six main supermarkets will be assessed, where only Jumbo does not offer a World Championship related FRP. Next to purchase behavior, information is available about household characteristics and their preferences for supermarket characteristics. This data is collected by the annual panel survey completed by all households. In addition, after the World Championship Soccer respondents were surveyed about their participation and appreciation of the FRP.
The Frequency Reward Programs
The FRPs differ in the starting date but all start before the World Championship soccer (Table 1). The minimum amount of purchase indicates the smallest purchase amount necessary to complete the set, which differs substantial between the FRPs(60-900 euro). The FRP of AH, Plus and Dirk are better appreciated compared to the other FRPs. Further, the FRPs differ in the number of supply funded premiums offered. C1000 does not offer supply funded promotions, while supply funded premiums were frequently used at Deka and Dirk.
Table 1: Overview Frequency Reward Program characteristics
Supermarket Starting week End week Minimum spending 1 premium for x euro spend Appreciation (grade 1-9) Supply funded premiums per week
To measure the effectiveness of participation in a FRP compared to non-participants two variables will be used. The first variable is if the FRP runs (FRPt), this variable has only variance over time and estimates the exposure
effect of a FRP. Secondly, the variable PartFRPit estimates the supplementary effect of participation in a FRP. To
measure the effect of competing FRPs, the exposure and participation variable will be included for competing supermarkets. The variable WkSinceStartFRPt indicates the number of weeks since the FRP runs, in the first
week this variable is zero and it increments with one per week. Because the appreciation of the FRP is only asked to households who participate in the FRP, appreciation is mean centered and for non-participating households a zero is assigned as their appreciation. Persistency of the effect of participating in the FRP is examined by PostPartFRP. This variable measures the effect of participating in the FRP on the decision to visit the supermarket and spending level in the four weeks after the FRP is terminated (i.e. Rewarded Behavior). The variables WCSoccert en Tempt are included to control for the effect of temperature en the World Championship
soccer on purchase behavior.
FRPt If the FRP of the focal supermarket is running
PartFRPit If household i is participating in the FRP of the focal supermarket
FRPCompt The number of competing FRPs running
PartFRPCompit The number of competing FRPs in which household i participates
WkSinceStartFRPt The number of weeks since the start of the FRP
ApprFRPit The appreciation of the focal FRP by household i, mean centered
PostPartFRPit If household i participated in the FRP in the four weeks after the FRP is ceased
WCSoccert If the World Championship soccer is running
Tempt The average temperature in week t
FRP Participation and household characteristics
The average spending level per week and the participation degree in the FRP of the households differs between supermarkets. The spending level varies from 28.66 euro at Deka to 40.06 euro at Jumbo and the percentage of non-participating households ranges from 65.7% at Jumbo to 46.2% at Dirk. In the third column the percentage of consumers of the supermarket who participates in one FRP is shown, this can be at the focal FRP or at a competing FRP. The household participation in only one FRP does not differ substantially between the different supermarkets, but is the highest at AH (24.9%) and the smallest at Deka (21.8%). At Dirk the highest percentage of households participates in more than on FRP (30.2%), while at Jumbo this percentage is substantially smaller (13.1%). The participation in the focal FRP is the highest at Dirk (35.1%) and the smallest at Deka (23.8%).
Table 2: Household participation in FRPs and buying behavior
Supermarket Spending per week in eurocents No participation in FRP Participates in 1 FRP Participates > 1 FRP Participates in focal FRP AH 3914 56.0% 24.9% 19.1% 33.8% C1000 3428 55.7% 24.7% 19.6% 27.7% Deka 2866 56.4% 19.1% 24.5% 23.8% Dirk 3120 46.2% 23.6% 30.2% 35.1% Plus 3377 52.5% 22.4% 25.1% 26.4% Jumbo 4006 65.7% 21.2% 13.1%
Childreni If household i has children living at home
Incomei The net income of household i
SClosesti If the focal supermarket is closest to household i
S2Closesti If the focal supermarket is second closest to household i
S3Closesti If the focal supermarket is third closest to household i
S4Closesti If the focal supermarket is fourth closest to household i
QualitySupermarketi The importance of quality at the supermarket for household i
Conveniencei The importance of convenience at the supermarket for household i
Pricei The importance of the price level at the supermarket for household i
Household buying behavior
The behavior of household i is characterized by the choice to buy at a supermarket in week t, which will be a dependent variable. Secondly, if the household decides to buy at a supermarket the sales per week at the supermarket are measured, which will be the second dependent variable. To control for heterogeneity in previous behavior, two variables will be included. The average share of wallet of the focal supermarket and grocery expenses will be included for every household. Both averages are estimated over the first 11 weeks of 2010.
BuyAtSupermarketit The decision of household i to buy in week t at the focal supermarket
SpendAtSupermarketit The amount of money spend in eurocents by household i in week t at focal supermarket
SOWi The average SOW of household i in the first 11 weeks of 2010 at the focal supermarket
AvSpendingi Total grocery expenses of household i in the first 11 weeks of 2010
Disaggregate Data for Supply Funded Premiums
To assess the effectiveness of the supply funded premiums, the purchases of the panel will be examined on a lower aggregation level. The interest will be on the number of products of brand X bought in week t by household i, which will be a dependent variable. Because C1000 does not offer supply funded premiums (Table 1), the supermarkets AH, Deka, Dirk and Plus are studied. Furthermore, we examine brands in the margarine/ butter category, savory food snacks, canned soup, ice cream and carbonated soft drinks. At minimum, 15 times the brand must be bought. In total 473 brands meet this requirement and after checking for multicollinearity and if a supply funded premium is offered 49 brands are selected. Because of the examination on product level, the amount of rebate for the brand in week t and if the product was promoted is controlled for. Additionally, a dummy variable indicates if the product is offered with a supply funded premium in week t.
NrBrandBoughtit The total bought of brand X in week t by household i
Rebatet The rebate in eurocents in week t for brand X
Promotiont If brand X was promoted in week t in leaflet of focal supermarket
SupplyFundedPrt If at brand X a free premium could be obtained in week t
4. Model specification
Because participants of a FRP might have higher visiting and spending levels compared to non-participants before the introduction of the FRP, a simple comparison of participants versus non-participants cannot establish a conclusive causal relationship (Liu 2007). Therefore, we estimate a model that has the spending level at the supermarket of household i in week t as the dependent variable and that accounts for both variation over time and for variance between households. Secondly, the model must be able to estimate the effect of non time-varying variables, because of the interest in household characteristics. The estimator that fits these requirements is the linear panel regression model with random effects estimation. With a panel model we control for the dependence between observations for household i over time. The resulting model is:
(1) = +
Where is spending level for household i in week t and are the explanatory variables.
Equation 1 is a random effects panel model and the error term will consists of two parts = + ; where
is a random effects parameter and an error term that is assumed to be i.i.d. Because of the inclusion of in
the error term, the assumption must be made that the random effects parameter is independent of .
However, the random effect accounts for household heterogeneity and it is plausible that household
heterogeneity is related to the observed household characteristics in Xit. Dependence between and is
called the endogeneity problem and will lead to inconsistent parameter estimates. Mundlak (1978) proposes a method to correct for this endogeneity problem between the time-varying parameters and the random effects parameter. Mundlak argues that = + , where is an error term which is unrelated to . Therefore,
we can write:
(2) = + + +
In equation 2 both uit and wi are i.i.d., so we can write = + which simplifies the equation to:
(3) = + +
In equation (3) the correlation between the random effect parameter and the time-varying independent variables is controlled for. However, a second problem arises due to the large number of observations where household i in week t does not visit the focal supermarket. In these weeks there is no observation for the spending level of household i in which gives a problem called incidental truncation (Wooldridge 2010: p. 802). Unobserved heterogeneity can influence both the spending level and the choice to buy at the supermarket and thereby biases the estimates of the coefficients in the spending level regression. Wooldridge recommends a Heckman selection model to resolve this problem. In the Heckman selection model a probit panel model is included to estimate the probability that household i will buy in week t. The model we use is:
(4) ∗ = + , where = 1 if ∗ > 0.
Where is decision to visit supermarket for household i in week t and are the explanatory variables.
The structure of the error term of the probit model is identical to the structure of the error term of the linear panel model described in (1). Thus, a similar endogeneity problem arises for equation 4 and Mundlak’s approach will be applied:
(5) ∗ = + + , where = 1 if ∗ > 0.
For notational simplicity we will stack with to and with to . To overcome the incidental
truncation bias in (3) the inverse Mill’s ratio is included as is explained in detail by Wooldridge (2010:805). The inverse Mill’s ratio is ()
(), where connotes the normal probability density function and ϕ connotes the
normal cumulative density function. However we do not have δ, but from the first step in the Heckman selection model we have a consistent estimator . Including the inverse Mill’s ratio in equation 3 will lead to the following equation:
(6) = + + ! (")
(")+
In equation 6 heteroskedastiscity arises, caused by the inclusion of the estimated inverse Mill’s ratio and in
The model specification of equation (7) and (8) is grounded on section 2 and specified in equation (5) and (6). In equation (7) the variables that indicate which supermarkets are nearby the households residence and the importance of supermarket characteristics are included. These variables are not included in equation (8) because they will not influence the spending level at the supermarket. Adjacently, for C1000 a dummy variable is included for the weeks in which all meats are on promotion. We need to control for these promotions because they surrounded the weeks in which the FRP was running and affected the coefficients of FRP effectiveness. As explained above, to account for endogeneity the mean of the variables that are both time- and household-varying are included.
(7) BuyAtSupermarketit = β20+ β21SClosesti +β22S2Closesti +β23S3Closesti +β24S4Closesti + β25Incomei
+β26Childreni +β27 QualitySupermarketi+ β28Conveniencei + β29Pricei + β210SOWi + β211AvSpendingi +
β212FRPt +β213WkSinceStartFRPi + β214PartFRPit + β215SOWi*PartFRPit + β216WkSinceStartFRPi*PartFRPit +
β217PartFRPCompit * PartFRPit + β218 PostPartFRPit +β219ApprFRPit + β220FRPCompt + β221PartFRPCompit +
β222WCSoccert + β223Tempt + γ21M(PartFRPi )+ γ22M(PartFRPCompi)+ γ23M(ApprFRPi) +
The second step in the regression is to estimate the amount purchased by respondent i in week t. This will be done with random effects linear panel regression estimated with the FGLS method (Verbeek 2008). The dependent variable is the natural logarithm of the sales for household i in week t. This makes the comparison between the results meaningful and the interpretation of the coefficients becomes 100*coefficient is percentage change in spending level.
(8) Ln(SpendAtSupermarketit)= β0+ β1Incomei +β2Childreni +β3QualitySupermarketi + β4Conveniencei +
β5Pricei + β6SOWi + β7 AvSpendingi + β8FRPt +β9WkSinceStartFRPt + β10 PartFRPit + β11SOWi *PartFRPit +
β12WkSinceStartFRPt*PartFRPit + β13PartFRPCompit * PartFRPit + β14 PostPartFRPit +β15ApprFRPit +
β16FRPCompt + β17PartFRPCompit + β18WCSoccert + β19Tempt + γ11M(PartFRPi )+ γ12M(PartFRPCompi) +
γ13M(ApprFRPi) +λ
(")
(") +
The model specified in equation (7) and (8) will be estimated for all six supermarkets separately. However, the interest of this study is to infer general conclusions about FRP effectiveness. Therefore, one test is preferred to see the overall effect of the independent variable on the traffic and/or purchase amount. In this test, we want to account for the difference in confidence between the coefficients. Furthermore, the difference between coefficients can be random or structural different. Consequently, we test if the dispersion from the mean can be expected from sampling error (i.e. homogeneity test). This homogeneity test examines if the effects have a common mean. If the test indicates homogeneity, a fixed effect weighted mean effect size test will be used. When the coefficients are heterogeneous, the random effects weighted mean effects size test will be applied (Lipsey and Wilson 2001:p.114). These tests give every observation a weight based on its standard error. The Jumbo supermarket chain does not have a FRP and therefore the variables regarding the own FRP will be excluded for Jumbo.
Supply funded premiums
In this model, we are only interested in the effect of the supply funded premiums on purchase behavior. However, brands who offer a supply funded premium are frequently simultaneously promoted in leaflets and/or discounted. Furthermore, for some of the product categories, temperature and the World Championship soccer influence purchase behavior (e.g. carbonated soft drinks and chips). Adjacently, we expect in hypothesis 1 that participation in the FRP will positively affect purchase behavior and therefore PartFRP is included. Without controlling for these covariates, the effect for supply funded premiums can be overstated. The resulting Poisson model is described in equation (9) and (10).
(9) λit = exp(β1Rebatet+ β2Promotiont+ β3SupplyFundedPrt+ β4FRPt+ β5PartFRPit+ β6Temperaturet+ β7
WCSoccert +uit )=exp(xit’β)
(10) P{yit=y|xit}
=
#$% {'()*}()* ,
-!
,
y=0,1,2, . . . ,The interpretation the coefficient for supply funded premiums, 100* exp{β3} is the percentage change in
purchase amount due to the supply funded premiums (Wooldridge 2010:726). The aim of this study is to give an overall conclusion of the effectiveness of supply funded premiums and therefore we prefer a single test. Because of the 49 brands we test if other variables influence the effect of the supply funded premiums. First, a homogeneity test will check if the variation in β3 is larger than can be expected from the randomly imprecision
of results in each study. If the test rejects homogeneity, we explain the variance in β3 with covariates in a meta
regression (Thompson and Sharp 1999). To account for the difference in precision of the estimation of β3, the
coefficients are weighted by their standard errors. Furthermore, the model should account for heterogeneous variance in β3 because otherwise the estimated standard errors will be too small, which is done by adding a
between study variance parameter to the total variance. The model will be estimated with restricted maximum likelihood as proposed by Thompson and Sharp (1999). The covariates in the equation (11) are based on hypotheses 9a to 9c from section 3.
(11) β3p=γ0 + γ1 AHp+ γ2 Dekap+ γ3 Dirkp+ γ4 CarbonatedSoftDrinksp+ γ5 Margarine+ γ6 Snacksp+ γ7 Soupp+ γ8
IceCreamp+ γ9 WeekSupplyFundedPremiump+υp
5. Results
The estimated coefficients for both the decision to visit the supermarket and the spending level are of interest and will be discussed. The first step is the decision to buy at the supermarket; this is done for each supermarket separately. The hit rates indicate that the regressions are good in predicting the choice of household i to visit the supermarket in week t (from 80% at AH to 98% at Deka). For all supermarkets, the probability of visiting increases if the relative distance to the supermarket is small. Net income and having children has no effect on the decision to choose a supermarket (Income=0.00, p=0.32; Children=0.05, p=0.70). When quality of the supermarket is important to the household, supermarkets will be visited more often (0.02, p=0.09). Households who give high importance to convenience, will visit supermarkets less frequently (-0.05, p=0.04) while when price is important, supermarkets are visited more frequent (0.10, p<0.01). The coefficients for previous share of wallet at the supermarket chain (6.07, p<0.01) and the average spending level (-0.02, p=0.05) are significant predictors of the choice to visit the supermarket.
by the Meta test (-.05, p=0.01) but only the supermarket specific coefficient of Jumbo is significant (-0.08, p=0.09). We accept hypothesis 3a that there is a persistent effect of participation in the FRP, given the significant effect at AH (0.10, p=0.02), Plus (0.15, p=0.06) and the Meta test (0.06, p=0.04).
Hypothesis 4a states that a higher appreciation of the FRP improves the effectiveness of participating in the FRP on visiting frequency. However, the hypothesis is rejected because the coefficient for the appreciation of the FRP is insignificant at all supermarkets and the Meta test (0.01, p=0.49). In hypothesis 5a, we hypothesize a negative interaction between SOW and participation in the FRP. This is confirmed because both the Meta test (-0.79, p<0.01) and for AH (-0.88, p<0.01), C1000 (-0.75, p<0.01) and Plus (-0.96, p<0.01) a significant negative effect exists. Hypothesis 6a states that the effect of participation in the FRP is strengthened during the time-span. This hypothesis is rejected because of the insignificant Meta test (-.08, p=0.13) and incongruent results at the supermarkets. From hypothesis 7a, we infer a negative moderation of participating in a competing FRP on the effectiveness of participating in the focal FRP. However, both the Meta test (0.04, p=0.35) and the supermarket specific coefficients show no effect and consequently hypothesis 7a is rejected.
The World Championship soccer (-0.03, p=0.42) has no effect on visiting behavior while with higher temperatures households visit supermarkets less often (-0.01, p=0.03). From the mean of participation in a FRP, we infer that participating households visit the supermarket substantially more frequently in the period surrounding the FRP (1.40, p<0.01). Furthermore, households participating in a competing FRP visit the focal supermarket significantly less often (-1.05, p<0.01) in the period surrounding the FRP. Finally, the coefficients of quality of the supermarket, the interaction between SOW and participation in a FRP, the reward behavior effect and participation in a competing FRP are similar for all supermarkets (i.e. homogeneity test).
Table 3: Estimated coefficients of probit regression for supermarket choice
(PartFRP) (7.38)*** (3.55)*** (2.15)** (3.40)*** (4.82)*** (6.17)*** SOW*Participation FRP -0.88 -0.75 -0.63 -0.19 -0.96 -0.79 (SOW*PartFRP) (5.00)*** (3.45)*** (1.17) (0.47) (2.81)*** (6.61)*** Week since start FRP
*Participation FRP -0.26 -0.01 0.02 -0.02 -0.10 -0.08 (WkSinceFRP*PartFRP) (4.43)*** (0.32) (0.24) (0.41) (1.50) (1.53) PartFRP_PartComp 0.06 0.03 0.02 0.16 -0.06 0.04 (0.74) (0.31) (0.13) (1.30) (0.48) (0.94) PostPartFRP 0.10 -0.02 0.09 -0.04 0.15 0.06 (2.34)** (0.39) (0.64) (0.48) (1.85)* (2.04)** Apprecation FRP 0.02 0.01 0.04 0.02 -0.05 0.01 (ApprFRP) (0.78) (0.37) (0.71) (0.57) (1.19) (0.69) Competing FRP -0.00 0.03 0.01 -0.02 0.03 -0.01 0.00 (CompFRP) (0.24) (1.05) (0.15) (0.86) (1.25) (0.95) (0.45) Participation Competing FRP -0.06 -0.06 -0.08 -0.08 0.03 -0.08 -0.05 (PartFRPComp) (1.57) (1.13) (0.87) (1.19) (0.52) (1.70)* (2.50)** WC Soccer 0.02 -0.01 -0.20 0.12 -0.07 -0.06 -0.03 (WCSoccer) (0.64) (0.12) (2.35)** (1.68)* (1.47) (1.55) (0.82) Temperature -0.01 -0.01 0.01 -0.01 -0.01 0.00 -0.01 (Temp) (5.12)*** (2.95)*** (1.08) (2.18)** (1.50) (0.41) (2.12)** Mean Participation FRP 0.61 0.73 1.34 3.67 0.70 1.40 (MPartFRP) (9.87)*** (9.01)*** (6.01)*** (21.43)*** (6.04)*** (3.44)*** Mean Participation competing FRP -1.35 -1.12 -0.37 -2.26 -0.39 -0.98 -1.05 (MPartCompFRP) (7.18)*** (5.22)*** (1.11) (5.78)*** (1.63) (4.33)*** (4.68)*** Mean Appreciation FRP -0.01 -0.16 -0.27 -0.02 0.51 -0.06 (MApprFRP) (0.06) (1.04) (1.16) (0.08) (1.66)* (0.50) MeatWeek 0.04 (0.89) Constant -1.48 -2.09 -3.66 -2.98 -2.75 -2.82 (19.24)*** (24.39)*** (18.11)*** (21.48)*** (23.89)*** (24.54)*** Number of Observations 47600 47600 47600 47600 47600 47600 Number of Households 2380 2380 2380 2380 2380 2380 Hit rate 80% 88% 98% 92% 94% 93%
* significant at 10%; ** significant at 5%; *** significant at 1%, t-value inside brackets2
If a household decides to buy at the supermarket, the spending level at the supermarket is estimated. As described above, a linear panel regression is used to estimate the model with the natural logarithm of the spending level per week as dependent variable. The R2 for the regression equations varies between .22 at Dirk and .40 at Plus. In the regressions, Heckman’s lambda is included to account for the selection bias and this coefficient is significant and positive for AH and significantly negative for Deka. A positive coefficient for Heckman’s lambda indicates that the error terms of both regressions are positively correlated (Verbeek 2008: 245). This implies unobserved heterogeneity that affects both the decision to visit the supermarket and the spending level. While a negative coefficient implies the presence of unobserved heterogeneity that affects the decision to visit positively but has a negative effect on the spending level.
The results show that households with a higher income spend more per week (0.02, p<0.01) and that having children does not significantly affect the spending level (0.11, p=0.12). The significant effect of SOW (1.55, p<0.01) and average spending level (0.02, p<0.01) control for habitual behavior.
The effect of the presence of the FRP on sales is congruent, all supermarket coefficients and the Meta test (0.02, p=0.28) are insignificant. Hypothesis 1b states that participating in a FRP has a positive effect on spending level, the coefficient of AH (0.44, p<0.01), C1000 (0.15, p<0.01) and Plus (0.26, p<0.01) confirm this. Additionally, the
2
Meta test supports hypothesis 1b (0.23, p<0.01). When households participate in competing FRPs the effect on supermarket sales is significantly negative at AH (-0.06, p=0.02), Dirk (-0.10, p=0.03) and Jumbo (-0.08, p=0.04). The Meta analysis supports this hypothesis (-0.06, p<0.01). Hypothesis 3b is rejected because there is no effect of participation in a FRP after the program is ceased at all supermarkets except at Dirk (0.10, p=0.02), and also the Meta test indicates no rewarded behavior (0.01, p=0.63). In hypothesis 4b a positive effect of appreciation of the FRP on FRP effectiveness is expected. This is true for AH (0.02, p=0.07) and Deka (0.04, p=0.05) and confirmed by the Meta test (0.01, p<0.01). Hypothesis 5b predicts that a larger SOW results in a smaller effect of participating in the FRP on spending level. This holds for AH 0.41, p<0.01), C1000 0.26, p<0.01) and Plus (-0.23, p=0.03) and these were the only supermarkets with a significant effect of participating in a FRP. Furthermore, the Meta test confirms hypothesis 5b (-0.22, p<0.01). We find no time trend in FRP effectiveness (Meta test: -0.04, p=0.26) and therefore hypothesis 6b will be rejected. For the supermarket coefficients only AH has a significant but negative effect (-0.13, p<0.01). From hypothesis 7b, we infer a negative relation on the effect of participating in a FRP if households simultaneously participate in competing FRPs. This hypothesis cannot be confirmed because both the supermarket specific coefficients and the Meta test are insignificant (0.01, p=0.67).
During the World Championship soccer households do spend significantly more at the supermarket (0.03, p<0.01) and temperature has a negative effect on spending level (-0.01, p<0.01). The positive and significant value for the mean of participation in the FRP indicates that participating households spend significantly more in the period surrounding the FRP (0.33, p<0.01). Furthermore, we find that households who participate in a competing FRP spend less (-.33, p=0.05) in the period surrounding the FRP. As mentioned before, a dummy for the weeks at which C1000 offers special meat promotions is included because it affected the coefficients for the FRP. During the weeks of the meat promotions, households spend significantly more (0.16, p<0.01). Finally, the coefficients of Appreciation FRP, Participation in competing FRPs, World Championship Soccer and Temperature are similar for all supermarkets (i.e. homogeneity test).
Table 4: Linear regression for spending level at supermarket.
AH C1000 Deka Dirk Plus Jumbo Meta Analysis Net household income 0.03 0.01 -0.01 0.02 0.02 0.03 0.02 (Income) (7.97)*** (2.85)*** (0.69) (1.62) (2.61)*** (3.22)*** (3.71)*** Children -0.04 0.24 -0.29 0.25 0.22 0.21 0.11
(1.39) (6.21)*** (2.11)** (3.26)*** (4.09)*** (3.00)*** (1.61) Share of Wallet 2.20 1.58 0.69 0.95 1.84 1.68 1.54 (SOW) (23.77)*** (11.77)*** (2.10)** (4.57)*** (11.31)*** (7.53)*** (7.17)*** Average Spending level 0.01 0.01 0.06 0.02 0.01 0.02 0.02 (AVSpending) (15.44)*** (5.14)*** (6.93)*** (3.88)*** (2.68)*** (3.72)*** (4.38)***
FRP -0.00 0.07 0.15 0.08 -0.03 0.02
(0.02) (1.20) (0.75) (1.62) (0.41) (1.09) Week since start FRP 0.04 -0.02 0.07 -0.02 0.00 0.01 (WkSinceFRP) (1.82)* (1.50) (1.60) (0.78) (0.07) (0.32) Participation FRP 0.44 0.15 0.17 0.08 0.26 0.23 (PartFRP) (8.80)*** (2.58)*** (1.02) (1.32) (2.87)*** (2.76)*** SOW*ParticipationFRP -0.41 -0.26 -0.04 -0.04 -0.23 -0.22 (SOW*PartFRP) (6.74)*** (3.60)*** (0.20) (0.48) (2.17)** (2.72)*** Week Since start
Apprecation FRP 0.02 0.01 0.04 -0.02 0.00 0.01 (ApprFRP) (1.80)* (0.90) (1.95)* (0.96) (0.13) (1.81)* Competing FRP -0.01 0.01 -0.12 0.01 0.01 0.01 0.00 (CompFRP) (2.18)** (0.36) (1.86)* (0.89) (1.05) (1.93)* (0.51) Participation Competing FRP -0.06 -0.02 -0.06 -0.10 -0.01 -0.08 -0.06 (PartFRPComp) (2.37)** (0.46) (0.58) (2.19)** (0.32) (2.04)** (3.51)*** WC Soccer 0.01 0.05 -0.07 -0.01 0.06 0.08 0.03 (WCSoccer) (0.68) (2.00)** (1.02) (0.24) (1.80)* (3.11)*** (3.05)*** Temperature -0.01 -0.01 -0.01 -0.00 -0.00 -0.01 -0.01 (Temp) (5.12)*** (5.91)*** (2.33)** (1.96)* (1.58) (4.79)*** (9.40)*** Lambda 0.40 -0.02 -0.36 -0.03 0.01 -0.03 0.00 (7.35)*** (0.58) (6.12)*** (0.62) (0.15) (0.52) (0.05) Mean Participation FRP 0.39 0.27 0.11 0.38 0.27 0.33 (MPartFRP) (10.50)*** (6.08)*** (0.71) (3.08)*** (3.92)*** (12.87)*** Mean Participation competing FRP -0.69 -0.39 0.08 -0.91 0.04 0.06 -0.33 (MPartCompFRP) (5.36)*** (2.45)** (0.17) (3.03)*** (0.22) (0.25) (2.00)** Mean Appreciation FRP 0.03 0.05 0.06 0.00 0.24 0.07 (MApprFRP) (0.33) (0.61) (0.42) (0.03) (1.81)* (1.33) MeatWeek 0.16 (5.98)*** Constant 5.91 6.82 7.42 6.92 6.59 6.89 (68.72)*** (82.51)*** (33.51)*** (44.62)*** (61.46)*** (46.16)*** Observations 22832 12143 1859 5219 5220 6058 Number of Households 1938 1253 257 570 640 750 Overall R2 .34 .35 .35 .22 .40 .28
* significant at 10%; ** significant at 5%; *** significant at 1%, t-value inside brackets3
To provide insight in program effectiveness, the average net effect of all FRPs is calculated. The effect is calculated for every household controlling for their appreciation of the FRP, participation in competing FRPs and their SOW at the focal supermarket. The expected net increase in sales per week by participating in the FRP is the smallest at C1000 with 6.88 percent and the highest at AH with 26.27 percent. Dirk (7.44%), Deka (15.73%) and Plus (17.19%) are in between these extremes.
Supply Funded Premiums
To estimate the effectiveness of the supply funded premiums, first equation (9) is estimated for all 49 brands. The estimated coefficients for supply funded premiums are saved and a homogeneity test indicates that the variance in coefficients is larger than can be explained from sampling error (Q=374.70, p<0.01). A random effects weighted mean effects size test confirms hypothesis 8 that the mean effect is significantly positive (1.27, p<0.01). These results verify the estimation of equation (11) which is presented in Table 5.
The equation estimates presented in Table 5 includes no dummy variables for product categories, this because the inclusion of the product categories decreased the adjusted R2 and there was no significant variation in effectiveness between the categories. This result rejects hypothesis 9b. Furthermore, the variable promotion is included to control for the possible overestimation of β3 because for some of the brands the variable promotion
was dropped from the equation due to multicollinearity with supply funded premiums. The adjusted R2 of the final regression is 27.68%. The results indicate substantial differences between effectiveness of supply funded premiums at the different supermarkets. The effectiveness at AH is significantly lower (-2.15, p<0.01) compared to the base category Dirk which confirms hypothesis 9a. A possible explanation for this difference is that more households at Dirk participate in FRPs compared to the other supermarkets (Table 2) and the difficulty to
3
complete the set (Table 1). This difference in participation might indicate diversity in proneness to participating in a FRP. Secondly, including a dummy for the regressions without a control variable for promotions shows that the exclusion significantly overestimates the effectiveness of supply funded premiums (-1.02, p=0.04). The coefficient for WkSinceFRP shows that the effectiveness of the supply funded premiums does not significantly change during the time-span of the FRP (-.29, p=0.15), which rejects hypothesis 9c. The constant confirms that the effect of the supply funded premium is significant and positive (2.60, p<0.01).
Table 5: Moderators of Supply Funded Premiums
β3 (Effectiveness Supply Funded Premium)
AH -2.15 (3.22)*** Plus -0.74 (0.89) Deka 0.06 (0.09) NoPromotion -1.02 (2.18)** WkSinceFRP -0.29 (1.48) Constant 2.60 (4.90)*** Observations 49 Adjusted R2 27.68%
Absolute value of t statistics in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
6. Validating Rewarded Behavior
Because of the unexpected result of no rewarded behavior in spending level, another test will be conducted to verify this result. For a validating test of rewarded behavior, a simple comparison between participating households and non-participating household does not test the effectiveness of the FRP because there might be a structural difference between the two groups. Alternatively, a comparison of the spending level before and after the FRP for participating households cannot control for time trends simultaneous with the FRP.
Therefore, the inclusion in the model of the difference between households over time and the difference for participants and non-participants is a prerequisite. A model that fits these requirements is the differences-in-differences estimator (Verbeek 2008: p. 362). This model is especially useful to estimate the impact of a treatment on an outcome variable for a subgroup of the population. The data used for this model has two time periods t, where t=1 is the period before the start of the FRP and t=2 the period after the FRP is ceased. The aggregation to two periods is made because by including more than two periods the model estimates can become biased (Bertrand, Duflo and Mullainathan 2004; Wooldridge 2010 p.321). The proposed regression is shown in equation (12).
(12) /0(12304561 237897:36) = ;<976=>< + ?+ +
Where SpendAtSupermarketit is the spending level for household i in period t. This variable is estimated by using
five weeks before and after the FRP to sum the spending level. The PartFRPit is if household i participated in the
FRP, the variable is zero for all households in period 1 and is one for the subset who participated in the FRP in period 2, Tt is the time dummy to account for trends, ci is the individual effect and uit an i.i.d. error term.
However, instead of estimating the equation above directly we take the first difference which results in equation (13).
(13) ∆/0(12304561 237897:36) = ;<976=><+ ? + ∆
The subscript t disappears because by taking the first difference there is only one observation left per household. A main advantage of this model is that it is not a prerequisite that the individual effect ci is
uncorrelated with the PartFRPi because ci is eliminated in (13). This is relevant as it is unlikely that the choice to
The estimates for (13) are shown in Table 6, the results indicate no rewarded behavior effect for the FRPs studied. The coefficient is insignificantly different from zero for all supermarkets and the Meta test confirms this (0.02, p= 0.71). This finding of no significant rewarded behavior is congruent with the results in Table 4 but contradicts the finding of Taylor and Neslin (2005). The time trend is negative for AH 0.13, p=0.01), C1000 (-0.25, p<0.01), Plus (-(-0.25, p=0.02) and the Meta test (-0.12, p=0.02). However, for Dirk the time trend is insignificant but positive (0.04, p=0.72), which might explain why the coefficient for rewarded behavior in Table 4 is positive while in the coefficient in Table 6 is insignificant. For all regressions the R2 0.00 which confirms that participation in a FRP does explain the difference in spending level after the FRP is ceased.
Table 6: Linear regression for difference in average spending level before and after the Frequency Reward Program
AH C1000 Deka Dirk Plus Jumbo Meta
PartFRP 0.09 -0.05 0.13 -0.24 0.11 0.02 (Participants FRP) (1.09) (0.35) (0.41) (1.49) (0.60) (0.37) Constant -0.13 -0.25 -0.08 0.04 -0.25 0.02 -0.12 (Time Trend) (2.53)** (3.42)*** (0.49) (0.36) (2.43)** (0.26) (2.30)** Observations 1776 1102 201 495 540 621 R2 0.00 0.00 0.00 0.00 0.00 0.00
* significant at 10%; ** significant at 5%; *** significant at 1%, Absolute value of t statistics in parentheses4
7. Discussion and conclusion
This research examines the effect of FRPs on supermarket purchases at five different supermarket chains using a market wide panel dataset. Current research is extended by allowing for a difference in effect of the FRP on the decision to spend at the supermarket and on purchase amount. To the best of our knowledge, this is the first study that examines the effectiveness of FRPs that use continuous reward schedules. We find that participating in the FRP has a significant positive effect both on the decision to visit the supermarket and the spending level. This finding is in line with previous research (Lal and Bell 2003, Taylor and Neslin 2005) and shows that FRPs help in attracting customers and in improving the spending level at the supermarket. Taylor and Neslin (2005) find that FRPs increase sales with 6.1 to 6.4 percent while the effect of the FRP on spending in this study varies between 6.9% and 26.3%. The results indicate that the effectiveness of FRPs with continuous reward schedules are at least as high as those with the fixed ratio reward schedule. So, regardless of the benefit incongruence there is a positive effect of participation in a FRP on spending level. Due to the lack of information on the costs of the FRPs no conclusions can be stated about FRPs profitability.
The results indicate that the effect of the FRP on spending level vanishes after the program is ceased. This finding diverges from literature (Taylor and Neslin 2005) where a positive effect was shown. The difference in effectiveness can be explained by the different reinforcement schedule applied because the fixed ratio reinforcement schedule results in more persistence in continuing the desired behavior after the program is ceased.
In addition, we find no significant effect on consumers who do not participate in the FRP. This implies that self- exclusion has no effect on buying behavior. This result demonstrates the importance of encouraging customers to participate because the positive effect on buying behavior only holds for participating customers. We find that participation in a FRP at a competitor decreases the visiting frequency and spending level at the focal supermarket. Therefore, the extra expenses are at the utmost partially generated from market expansion or stockpiling. Kopalle and Neslin (2003) state that if the positive effect of the FRP comes mainly from competition there is a high probability of a strong competitive response that leads to a prisoner’s dilemma. The results from this research indicate that the extra sales come at competitors’ expense and therefore the results points
4
towards the occurrence of a prisoner’s dilemma. Besides, participating in a competing FRP has no effect on the effectiveness of the focal FRP.
The effectiveness of the FRP decreases when the consumer has a higher SOW at the focal supermarket, which is in line with previous research (Lal and Bell 2003, Liu 2007, Kim et al. 2009). This can be explained by the ceiling effect, from the Meta analyses we infer that the effect of the interaction SOW*Participation FRP is as large as the coefficient for participation in the FRP. This denotes that the effectiveness of participating in the FRP completely vanishes if the household has a SOW near 100%. The appreciation of the FRP positively moderates the effect of participation in the FRP on spending level. Therefore, we argue that retailers can take profit from creating a highly appreciated FRP. The effect might be stronger than the results of this study indicate because here the appreciation is only known for households who participate in the FRP. It is likely that the probability of participating in the FRP will increase when the appreciation of the FRP is higher. We find no pattern in the effectiveness of the FRP during its time-span, not for decision to visit the supermarket and for the spending level. AH differs with a negative pattern over time, which can be explained by a difference in the minimum purchase amount to complete the set. At AH the minimum purchase amount was €60 while the second lowest minimum purchase amount was €480 at Plus.
This study is the first that examines the effectiveness of collaboration of manufacturers in FRPs trough offering extra rewards. These supply funded premiums are effective in improving the purchase quantity of the promoted products. The effectiveness differs substantially between supermarkets, we find that supply funded premiums are most effective at supermarkets with households prone to participation in the FRP and where it is more difficult to complete the set. However, no significant difference is found between product categories or during the time-span.
Finally, we find self-selection in participating in the FRP and in participation in a competing FRP. Households who participate in the FRP substantially visit the supermarket more frequently and spend more at the supermarket in the period before the start of the FRP. While households who participate in a competing FRP visit the supermarket less frequent and spend less at the supermarket in the period before the FRP. This is in line with research on Loyalty Programs (Leenheer et al. 2007; Demoulin and Zidda 2009; Meyer-Waarden and Benavent 2009).
Limitations and further research
This study has several limitations that need to be addressed. One of the major findings from this research is the result of no rewarded behavior while this was expected (Taylor and Neslin 2004). The main difference between the FRPs of the focal study and the FRP assessed in the study by Taylor and Neslin (2004) is the reinforcement schedule. Therefore, this is most likely the source of the diverging finding. However, there are other differences between the FRPs studied (e.g. U.S.A. versus The Netherlands and rewards with utilitarian benefits versus hedonic benefits). To validate the statement that the difference is caused by the reinforcement schedules, two complete identical FRPs should be examined in which only the reinforcement schedule diverges.
effectiveness should take into account the successfulness in letting households participate. A first step is made by Kivetz and Simonson (2003) who show that households who perceive that the effort for them compared to others is low to receive the reward are more likely to participate.
8. References
Ailawadi, K., Bradlow, E., Draganska, M., Nijs, V., Rooderkerk, R., Sudhir, K. K., Wilbur, K. & Zhang, J. (2010). Empirical models of manufacturer-retailer interaction: A review and agenda for future research. Marketing Letters, 21(3), 273-285.
Barone, M. J., & Roy, T. (2010). Does exclusivity always pay off? Exclusive price promotions and consumer response. Journal of Marketing, 74(2),121−132.
Bertrand, M., Duflo, E., & Mullainathan, S. (2004) "How Much Should We Trust Differences-in-Differences Estimates?" The Quarterly Journal of Economics, 119(1), 249-275.
Cameron, A. C., & Trivedi, P. K. (2009). Microeconometrics Using Stata. Texas: Stata Press.
Chandon, P., Wansink, B., & Laurent, G. (2000). A Benefit Congruency Framework of Sales Promotion Effectiveness. Journal of Marketing, 64(4), 65-81.
Carey, C. (2008). Modeling Collecting Behavior: The Role of Set Completion. Journal of Economic Psychology, 29(3), 336-347.
Chung-Hui, T. (2010). A Research of the Spillover Effect of Gift Promotion-Its Forming and Fluctuation. Advances in Consumer Research - North American Conference Proceedings, 37916-917.
D'Astous, A., & Jacob, I. (2002). Understanding consumer reactions to premium-based promotional offers. European Journal of Marketing, 36(11/12), 1270-1286.
Demoulin, N. M., & Zidda, P. (2009). Drivers of Customers’ Adoption and Adoption Timing of a New Loyalty Card in the Grocery Retail Market. Journal of Retailing, 85(3), 391-405.
Dowling, G. R., & Uncles, M. (1997). Do customer loyalty programs really work? Sloan Management Review, 38(4), 71−82.
Du, R. Y., Kamakura, W. A., & Mela, C. F. (2007). Size and share of customer wallet. Journal of Marketing, 71(2), 94−113.
Fennis, B. M., and Stroebe, W. (2010). The psychology of advertising, 1st ed. Hove: Psychology Press.
Henderson, C.M., Beck, J.T and R.W. Palmatier (2011), “Review of the theoretical underpinnings of loyalty programs”, Journal of Consumer Psychology, forthcoming.
Huczynski, A. A. and Buchanan, D.A. (2007). Organizational Behavior, sixth edition. Pearson education Limited, Harlow England.
Kivetz, R., Urminsky, O., & Zheng, Y. (2006). The goal-gradient hypothesis resurrected: Purchase acceleration, illusionary goal progress, and customer retention. Journal of Marketing Research, 43(1), 39−58.
Kivetz, R., & Simonson, I. (2003). The Idiosyncratic Fit Heuristic: Effort Advantage as a Determinant of Consumer Response to Loyalty Programs. Journal of Marketing Research, 40(4), 454-467.
Kopalle, P. K., and Neslin S. A. (2003). "The Economic Viability of Frequency Reward Programs in a Strategic Competitive Environment." Review of Marketing Science 1(1), 1-39.
Kopalle, P. K., Neslin, S. A. Sun, B. Sun, Y. and Swaminathan Y. (2011). “The Joint Sales Impact of Frequency Reward and Customer Tier Components of Loyalty Programs,” working paper.
Lal, R., & Bell, D. E. (2003). The Impact of Frequent Shopper Programs in Grocery Retailing. Quantitative Marketing and Economics, 1(2), 179-202.
Leenheer, J., & Bijmolt, T. H. A. (2008). Which retailers adopt a loyalty program? An empirical study. Journal of Retailing & Consumer Services, 15(6), 429-442.
Leenheer, J., van Heerde, H. J., Bijmolt, T. H. A., & Smidts, A. (2007). Do loyalty programs really enhance behavioral loyalty? An empirical analysis accounting for self-selecting members. International Journal of Research in Marketing, 24(1), 31−47.
Lee-Wingate, S., & Corfman, K. (2010). A Little Something for Me and Maybe for You, Too: Promotions that Relieve Guilt. Marketing Letters, 21(4), 385-395.
Lewis, M. (2004). The influence of loyalty programs and short-term promotions on customer retention. Journal of Marketing Research, 41(3), 281−292.
Lipsey M.W. and Wilson D.B. (2001). Practical Meta-Analysis, Applied Social Research Methods Series, Thousand Oaks, California, Sage Publications.
Liu, Y. (2007). The long-term impact of loyalty programs on consumer purchase behavior and loyalty. Journal of Marketing, 71(4), 19−35.
Liu, Y., & Yang, R. (2009). Competing Loyalty Programs: Impact of Market Saturation, Market Share, and Category Expandability. Journal of Marketing, 73(1), 93-108.
Meyer-Waarden, L., & Benavent, C. (2009). Grocery retail loyalty program effects: self-selection or purchase behavior change?. Journal of the Academy of Marketing Science, 37(3), 345-358.
Mundlak, Y. (1978). On the Pooling of Time Series and Cross Section Data. Econometrica, 46(1), 69-85.
Pauwels, K. (2004). How Dynamic Consumer Response, Competitor Response, Company Support, and Company Inertia Shape Long-Term Marketing Effectiveness. Marketing Science, 23(4), 596-610.
Pauwels, K., Srinivasan, S., & Franses, P. (2007). When Do Price Thresholds Matter in Retail Categories?. Marketing Science, 26(1), 83-100.
Simonson, I., Carmon, Z., & O'Curry, S. (1994). Experimental Evidence on the Negative effect of Product Features and Sales Promotions on Brand Choice. Marketing Science, 13(1), 23.
Samaha, S., Palmatier, R., & Dant, R. (2011). Poisoning Relationships: Perceived Unfairness in Channels of Distribution. Journal of Marketing, 75(3), 99-117.
Shapiro,S., MacInnis D.J., & Heckler S.E.(1997). The effect of incidental ad exposure on the formation of consideration sets. Journal of Consumer Research, 24, 94-104.
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.
Thompson, S., & Sharp, S. (1999). Explaining heterogeneity in meta-analysis: a comparison of methods. Statistics In Medicine, 18(20), 2693-2708.
Verbeek, M. (2009), A guide to modern econometrics. Third Edition, Chichester: John Wiley and Sons, LTD. Wooldridge, J. M. (2010), Econometric analysis of cross section and panel data, Second Edition. Cambridge, Massachusetts: The MIT Press.
Zhang J. & Breugelmans E. (2011). The Impact of an Item-Based Loyalty Program on Consumer Purchase Behavior. Journal of Marketing Research, forthcoming.
Appendix A: Importance of Characteristics Supermarket
Extraction Rotated Factor Loadings (Varimax rotation)
Quality of Retailer Convenience Price
Fast Service 0.42 0.55
Attractive Offers 0.80 0.85
Nice Atmosphere 0.61 0.76 High quality of Fresh
products 0.66 0.74 0.32 Large assortment 0.60 0.67 Friendly staff 0.69 0.80 Low Prices 0.78 0.84 Neat Shop 0.69 0.80 Capable staff 0.66 0.77 Opening times 0.70 0.82
Enough parking capacity 0.60 0.74
Good quality 0.63 0.66 0.35
