The empirical investigation of how "multiple store patronage" influences consumer reaction towards promotions

(1)

The Empirical Investigation of How “Multiple Store Patronage”

Influences Consumer Reaction Towards Promotions

Master’s Thesis

Research Masters (MSc) in Business in Society

Supervisor: Dr. Jonne Guyt

Faculty of Economics and Business

University of Amsterdam

Student Name: Muhammad Adnan Ahmad

Student ID: (10602755 UvA) / (2574306 VU)

(2)

DEDICATION

This thesis is dedicated to my father (late),

Muhammad Ibrahim.

I want to thank him for all the sacrifices he

made to fulfill my dreams.

(3)

Acknowledgements

I begin with praising almighty ALLAH and His prophet MUHAMMAD (PBUH) for giving me this wonderful life with beautiful people around me in the form of family, friends and teachers.

Next, I want to thank my parents and sisters for their unconditional love and

boundless support. They never forced me to do any part-time jobs, while I am studying.

They never told me the hardships they are bearing, so I can focus only on my studies.

Finally, I would not be doing justice to this thesis without mentioning the name of

my supervisor, Jonne Guyt. I want to express my gratitude to him for providing me the

guidelines and directions at every single stage of my thesis. I would also like to thank him for

showing a great amount of flexibility in scheduling appointments, which helped me a lot.

(4)

Abstract

This study investigates the phenomenon of multiple store patronage and how it influences consumer reaction to promotions. The main objective is to empirically examine how consumers with widespread store patronage differ from consumers with narrow store patronage in terms of their response to promotions. We exclusively focus on large-scale promotional events known as “Savings Weeks” across several Dutch retailers. In doing so, we conduct a segment-level analysis of Savings Week events based on the underlying patronage behavior of consumers. The results reveal that the promotional effects of such events differ across segments. Segments in which consumers display an increased tendency to patronize multiple stores are more responsive towards promotional events as compared to segments in which consumers usually prefer their primary stores and patronize few stores. Additionally, consumers may stay loyal to their primary stores while switch to competing stores during promotions.

Keywords: multiple store patronage, Savings Week, promotions, store performance, price search behavior, customer segmentation, retail management

(7)

Introduction

Given the widespread use and magnitude of dollars spent on promotions, managers and academics have shown large interest in understanding how consumers respond to promotions and how promotions influence retailer performance (Ailawadi et al. 2006; McAlister, George, and Chien 2009). A considerable amount of research has been devoted to examine the impact of promotions on brand and category level (Bucklin, Gupta, and Siddarth 1998; Bell, Chiang, and Padmanabhan 1999; Gupta 1988; van Heerde, Leeflang, and Wittink 2004). Though, relatively little research has been done on their impact on store performance such as store traffic, sales, and spending (see Gauri et al. 2017). While the amount of literature on sales promotion is impressive, a recent and largely neglected phenomenon in grocery retailing is the use of “Saving Weeks”.

“Savings Weeks” refers to “large scale promotional events in which supermarket chains advertise promotions across multiple categories simultaneously, under a common theme, and across several weeks” (Guyt 2015). Examples are Albert Heijn’s “Hamsterweken Event” during which the leading Dutch retailer offers a promotion on more than 1500 SKUs or American retailer Kroger’s “Cart Buster Savings Event” which claims to offer over $100 in savings on a set of products. Recently, a study by Guyt (2015) in which the author examines Savings Week events across several Dutch retailers found an incremental impact of such events on store traffic and spending than regular promotions. Moreover, the author argues that the promotional impact of these events varies across customers based on their loyalty towards the stores.

A stylized fact about supermarket patronage is that consumers may patronize multiple stores, a phenomenon known as “multiple store patronage” (Baltas, Argouslidis, and Skarmeas 2010). Perfectly loyal consumers shop exclusively at their favorite store whereas many consumers patronize a set of secondary stores in addition to their primary store (Ailawadi and Keller 2004; Rhee and Bell 2002). The extent to which secondary stores are used also differ across consumers (Popkowski Leszczyc and Timmermans 1997). While multiple store patronage has always been noticed, it is not until recently that scholars have started giving attention to this phenomenon (e.g. Baltas et al. 2010; Luceri and Latusi 2012;

(8)

Maruyama and Wu 2014). Still, how the tendency to use multiple stores influences consumers’ response towards promotions is unexplored.

The present study aims to accommodate the presence of multiple store patronage among consumers while examining the impact of Savings Week events. Guyt (2015)

indicated that the promotional impact of such events differs across customers depending on their relative visit share at the retailer. Although it is worth mentioning that primary store choice is a relatively stable decision (Rhee and Bell 2002), it has been argued that the degree of loyalty towards the primary store depends on shoppers’ overall patronage pattern,

especially the number of stores patronized (Mägi 2003). Moreover, the studies that disregard the use of multiple stores may discard a significant aspect of store patronage (Popkowski Leszczyc and Timmermans 1997). Therefore, we intend to evaluate the impact the Savings Week events by taking into account the number of stores patronized by consumers as well as their overall shopping activity.

This study is the first attempt to empirically examine how multiple store patronage influence consumer reaction towards promotions. In order to do that, we propose a segment-level analysis of Savings Week events. Specifically, we examine the impact of promo events across segments based on the underlying patronage behavior of consumers. From an academic view point, this study will add to the literature on sales promotion by examining the differential effects of promo events across customers given their

heterogenous patronage behavior. We can examine the differences across consumers with widespread and narrow store patronage. Additionally, the modelling approach we adopt will render segment-level conclusions, which are not readily discernible in the typical household level models.

The extent to which consumers use competing stores is also very relevant from a managerial perspective, as this variable is closely associated with consumers’ loyalty potential. Through promo events, retailers hope to draw new customers and enhance the spending of their established customers. By investigating the differential effects of such events across segments, retailers can examine the ability of their promotional events to haul in customers with the greatest loyalty potential. Furthermore, they can examine the

(9)

Utilizing these insights, retailers can further improve their promotional events for targeting specific shopper groups and enhance their store performance.

To sum up, this study extends Guyt (2015) research by explicitly incorporating the use of multiple stores by consumers. In doing so, this study addresses several interrelated questions: How do various shopper segments based on the number of stores they patronize react to promotions during “Savings Week” events at supermarkets? Do shoppers who patronize multiple supermarkets react differently than shoppers who patronize few supermarkets? Do they differ in their visit propensity during the Savings Week promotions versus regular promotions? Do they differ in their spending during the Savings Week promotions versus regular promotions? And, finally, how does such large-scale events influence store performance?

The paper is organized as follows. After reviewing the exiting literature, we describe Savings Week events and their impact on store visit and spending. Next, we discuss multiple store patronage, the focus of this study, and how consumers with widespread and narrow store patronage may differ in their response towards promo events. Next, we present our modelling framework and estimation results. We conclude with summary insights,

(10)

Relevant Literature

In this section, we review the existing literature on the impact of promotions and how that influences store performance. First, we discuss the theoretical literature on the impact of promotions. Next, we summarize the findings of empirical studies related to the impact of promotions on store performance.

Theoretical insights on impact of promotions

Past research suggests that consumers respond to price promotions by engaging in price search and by comparing the benefits of price search with the opportunity costs associated with it (Putrevu and Ratchford 1997; Urbany, Dickson, and Kalapurakal 1996). Based on this cost-benefit framework, consumers can be seen as attempting to maximize the total benefits of their price search. Bell, Ho, and Tang (1998) in their study on consumer store choice decisions argue that consumers evaluate the total shopping cost of each shopping trip for each store. The underlying implication is that consumers compare the expected savings on purchasing the shopping basket from different stores and visit the store which offers the maximum savings. Similarly, Hosken and Reiffen (2007) show that retailers use a mixture of promotional strategies to attract various shopper segments and the retailer with the most attractive offers gets the majority of shopper traffic. Therefore, promotions serve as a communication device for retailers through which they convey the magnitude of savings consumers can get by visiting their outlets.

Consumers may also utilize different mechanisms to search for prices. The

theoretical literature provides two main rationales on consumer price search behavior. The first explanation suggests that consumers may engage in spatial price search across stores (Varian 1980). Such consumers are typically price sensitive who visit multiple stores in search for best prices. The second explanation suggests that consumers may engage in temporal price search across time within a store (Conlisk, Gerstner, and Sobel 1984). Such consumers have low willingness to pay for regular prices, and shift their purchases to promotional periods at their preferred stores rather than switching stores (Mace and Neslin 2004; Neslin, Henderson, and Quelch 1985). Therefore, promotions act as a price

discrimination mechanism for retailers to boost store traffic and sales. In relation to Savings Week events, such promotional events may appear more attractive to consumers who

(11)

search across stores given their high price sensitivity and increased willingness to switch stores. However, the recurring nature of Savings Week events might also appeal to consumers who search across time.

From a psychological perspective, it has been suggested that items with feature promotions and deep discounts appear more salient and get consumer attention (Gotlieb and Swan 1990; Grewal, Marmorstein, and Sharma 1996). Moreover, recent advancements in the neuroscience literature have examined the link between expectation and attention (Friston 2005). This literature indicates that unexpected stimuli result in stronger neural responses than expected stimuli (Garrido et al. 2009; Melloni et al. 2011). Since deeper discounts are incongruent to consumers’ expectation, they may trigger such stimuli among consumers’ brain and gain their increased attention (Gauri et al. 2017). Consequently, stores with more feature promotions and deeper discounts have a higher likelihood to be noticed by consumers. However, out-of-store communication of promotions is essential for such an effect to take place. In this regard, Savings Week events have greater potential to generate additional store traffic because these events utilize nation-wide advertising campaigns with increased usage of mass media.

Empirical studies on impact of promotions on store performance

Table 1 provides an overview of the relevant empirical studies on various store performance metrics. The empirical evidence regarding the impact of promotions on store performance has remained inconclusive, especially on store traffic and sales. Walters and Mackenzie (1988) found that the effect of promotions on store traffic was significant in only one of the eight categories examined. Srinivasan et al. (2004) investigated the traffic effects of price promotions on 63 brands in 21 categories and found significant positive effects of promotions for only 15% of the brands. In contrast, Lams et al. (2001) found a significant positive effect of promotions on store traffic whereas the impact on consumer spending was dependent on the type of promotion.

Similar to the store traffic effects, the evidence about the impact of promotions on store sales and profits is also mixed. While Mulhern and Leone (1990) did not find a positive impact of promotions on store traffic, the authors did find a positive impact of promotions on store sales when retailers switch from promoting many items at a shallow discount to a

(12)

few items at deep discount. On the contrary, Srinivasan et al. (2004) found an insignificant impact of brand promotions on store sales. One limitation of the studies discussed above is that these studies typically focus on the brand and category level promotions which may not represent store level effects in a comprehensive manner (Gauri et al. 2017).

Another stream of research related to promotional effects on store performance is more focused on the factors that influence the effectiveness of promotions and in turn store performance (e.g. Ailawadi et al. 2006; Gijsbrechts, Campo, and Goossens 2003; Gauri et al. 2017; Lams et al. 2001). Researchers in this vein have examined the impact of various promotional characteristics of discounts such as the discount depth (the size of discount) and discount breadth (no. of items on discount). These studies conclude that both the depth and breadth of discount are positively associated with store level impact of promotions (e.g. Ailawadi et al. 2006). In addition to that, the nature of product categories being promoted also influences the promotional impact on store performance (Gauri et al. 2017). Thus, studies examining the drivers that influence the effectiveness of promotions have found more positive results for store performance.

A recent entry in retailers’ promotional toolkit is the use of large scale promotional events known as “Savings Weeks”. The unique nature of such events by offering a plethora of discounts under a common ‘Savings Theme’ offers retailers the opportunity to steal additional customer traffic and boost their sales. As of now, Guyt (2015) is the only study which investigates the impact of such events on store performance. By examining 25 large-scale events across four leading Dutch retailers, the author concluded that the store visit frequency and the amount spent at the focal retailer increases during the Savings week events. Another interesting finding is that the effects on store visit frequency and store spending vary across customers based on their degree of loyalty towards supermarkets. This study is an extension of Guyt (2015) work in which we investigate the differential effects of Savings Week events across customer segments based on their underlying patronage behavior.

(13)

Table 1: Relevant empirical studies on impact of promotions on store performance

Reference Dependent

Variable(s)

Sample

Size Promotion Type Key Findings

Walters and Rinne (1986)

Store sales, traffic, and profit

Three grocery

stores

Price promotions on loss leader product portfolios with featured

ads in newspaper

Certain product portfolios affected store sales and traffic. No significant effect

on profits Walters and

Mackenzie (1988)

Two grocery

stores

Price promotions on loss leader product

categories with featured ads in

newspaper

Certain product categories affected store sales and traffic. No significant effect

on profits

Mulhern and Leone (1990)

Store sales and traffic

One grocery

chain

Discount depth and breadth of featured

items

Featuring few items at deep discounts as compared to

featuring many items at small discounts positively affects store sales but not

store traffic Lam et al.

(2001)

Front traffic, store-entry ratio, closing ratio, and average

spending Two chains (apparel and sporting goods)

Price promotions and newspaper advertisements

Price promotions increases store entry, spending, and

purchase likelihood

Gujsbrechts et al. (2003)

Store sales and traffic

55 grocery stores

Price promotions with feature ads on store

flyers

The compositional characteristics of store flyer

ads positively affect store traffic and sales Srinivasan et al. (2004) Various measures for manufacturer and retailer performance (incl. store traffic) 96 grocery

stores Promotion depth

Price promotions positively affects manufacturer revenues but for retailer the

results were inconclusive Ailawadi et

al. (2006

Store sales and

profit 3803 stores

Promotions with deep discounts and featured

ads

Promotions on high “consumer-pull” brands

increase total sales but lower net profits

Guyt (2015) Store traffic and spending

Seven grocery

chains

Savings Week events (large scale promotional events

with massive advertising campaigns)

Savings Week events increase store traffic whereas the mixed effects were found for spending at

the chain Gauri et al.

(2017)

24 grocery stores

Discount depth and breadth and feature

depth

The impact of promotions on store traffic, spending and profit is moderated by

category characteristics

This study Store traffic and

spending

Seven grocery

chains

Savings Week events

Segment level differences of promotional effects exists

between consumers with widespread and narrow

(14)

Savings Week Promotion Events

Guyt (2015) claims that “Savings Week” promotion events differ from regular promotions on content, communication, and timing. In terms of content, the author argues that such events promote store wide savings by offering deals across multiple categories and unusual discounts, which are presented under a common retailer-specific “Savings” theme. With respect to communication, the author suggests that promo events are supported by nation-wide advertising with increased out-of-store advertising stressing (store wide) savings as opposed to regular promotions which are mostly communicated through in-store advertising. As for timing, the author shows that promo events last for longer period (three weeks on average) as compared to regular promotions which typically lasts for a week and that these events are recurring, often once or twice a year. Together these distinguishing characteristics of “Savings Week” promotions may, indeed, generate additional store traffic and incremental store spending as compared to regular promotions. Below we explain why.

Impact of Savings Week events on store traffic

A commonly held marketing belief is that price promotions build store traffic and that this effect increases with the depth and breadth of promotions (Levy, Weitz, and Grewal 2014). Ailawadi et al. (2006) found that the increased number of product categories or items on promotion is generates additional store traffic. The underlying implication is that increased assortment of promoted items draws a broader group of shoppers with diverse needs. In a recent study, Gauri et al. (2017) argued that the ability of featured discounts to build store traffic depends on the magnitude of potential savings they offer to consumers. This leads to the expectation that both the depth and breadth of discounts are positively associated with store traffic. Since the range of promoted items and the depth of discounts during ‘Savings’ weeks is massively huge than non-Savings weeks (Guyt 2015), promo events have the potential to trigger more significant changes in store patronage.

Also, this holds for the fact that Savings Week events are supported by increased out-of-store promotions on TV and newspapers, thereby, create awareness among the masses (Guyt 2015). Researchers have found an incremental effect of out-of-store promotions on store performance in the past (Walters and MacKenzie 1988; Bawa and

(15)

Shoemaker 1989). It has been argued that the impact of advertising depends on its reach i.e. the number of consumers in the target market exposed to the ads (Levy and Weitz 1998). Finally, the large-scale advertising campaigns minimize consumers’ search cost by reducing their effort to find best deals which strongly influences cross-store shopping (Gauri et al. 2010).

Impact of Savings Week events on store spending

Savings Week events may also have a stronger impact on store spending than regular promotions given the huge savings opportunities available to consumers. For new customers, any amount spent is an increase. However, the impact of promo events on existing customers’ spending depends on their ability to generate incremental monetary spending on non-promoted items that offsets the price cuts on promoted items (Guyt 2015). On one hand, the opportunity to buy promoted items from multiple categories may

encourage consumers to concentrate their shopping towards the promoting store which will increase their average spending (Lams et al. 2001). Once consumers are in the store they are likely to buy items other than those on promotion and engage in more impulse purchases (Belle et al. 2011; Stilley et al. 2010). Similarly, the unusual depth of discounts during promo events maximize total savings on the entire shopping basket and entice consumers to spend more (Mulhern and Leone 1990). Finally, the messages carried by promo events advertisements signal larger savings opportunities (Guyt 2015), which usually exceed consumer expectations and may trigger stronger neural responses (Gauri et al. 2017).

(16)

Multiple Store Patronage

A stylized fact about supermarket patronage is that shoppers may patronize multiple supermarkets to fulfill their shopping requirements, a phenomenon known as “multiple store patronage” (Baltas et al. 2010). Although consumers tend to visit their preferred stores, they may patronize a set of secondary stores while spending the majority of their purchases at their primary store (Ailawadi and Keller 2004; Desmet and Volle 1996; Knox and Denison 200). For instance, price-sensitive or deal prone consumers are likely to shop at multiple stores in search for best deals (McGoldrick and Andre 1997). Researchers have recommended cost-benefit analysis as an appropriate framework to study multiple store patronage (Baltas et al. 2010; Luceri and Latusi 2012). Particularly, Baltas et al. (2010)

suggests that consumers will patronize multiple stores if their search benefits are higher and search costs are lowers and vice versa based on individual specific characteristics (Baltas et al. 2010). However, discussing all of those characteristics is beyond the scope of this study. Our focus is on the number of stores consumers patronize.

The number of stores patronized by consumers is a key quantitative measure which provides several insights on consumers’ shopping behavior, especially the diversity in their store preferences (Baltas et al. 2010). It is also closely associated with consumer price search behavior discussed earlier, as patronizing multiple stores is positively related to search effort (Stigler 1961). Consumers who adopt spatial price search strategy (across stores) are likely to patronize multiple stores because search benefits for their search efforts are higher. On the other hand, consumers who adopt temporal price search strategy (across time) are likely to patronize few stores because the opportunity costs for their search efforts are higher. Therefore, spatial price search leads to widespread patronage whereas temporal price search leads to narrow store patronage and possibly stockpiling, purchase acceleration and purchase delays (e.g. see Gauri et al. 2008).

Despite this interrelationship between consumer price search and store patronage behavior, the link between multiple store patronage and how it influences consumer reaction to promotions is missing in the sales promotion literature. Since Savings Week events maximize search benefits with massive savings opportunities and reduce search costs with large-scale advertising (Guyt 2015), we speculate that Savings Week events will have

(17)

differential effects across shoppers depending on their patronage set size. Below we explain why.

Differential effects of Savings Week events across shoppers

Consumers who are likely to visit multiple stores may display more significant changes in their visit propensity during Savings Week events. One explanation for this behavior emerges from their underlying price search strategy (spatial price search), as indicated above. Researchers focusing on the spatial price search dimension have argued that such consumers search a lot across stores to find cheap prices (Carlson and Gieseke 1983; Fox and Hoch 2005; Putrevu and Ratchford 1997). These consumers may buy the majority of their purchases at their preferred store but switch to competing stores to buy deal-items (Gauri et al. 2008). Given the high price sensitivity and increased willingness to switch to competing stores, store switching will appear more attractive to consumers with widespread store patronage. Furthermore, it has been argued that price-sensitive

consumers who compare prices across stores to find the best deals distribute their purchases evenly across stores (Kim et al. 1999; Laaksonen 1993; Mägi 1999; McGoldrick and Andre 1997).

On the other hand, consumers who patronize few stores may not display that much change in their visit propensity because such consumers are typically more loyal to their primary store and search for cheap prices across time within a store (Gauri et al. 2008). Past research argues that such consumers will shift their purchases, rather than switching stores, to promotional periods and engage in stockpiling, purchase acceleration, and purchase delays to avail the full benefits of promotions (Mela, Jedidi, and Bowman 1998; Neslin, Henderson, and Quelch 1985).

Similar to the impact on visit propensity, we expect the impact of Savings Week events on consumers with widespread store patronage to be larger. Baltas et al. (2010) argues that consumers with larger patronage set size tend to have greater expenditures because it increases the benefits of patronizing multiple stores. Since the discounts during promo events are unusually deep which create larger savings opportunities on the shopping basket, we expect its impact on this group of customers to be larger. However, consumers with narrow store patronage may also exhibit a positive change in their spending, but that

(18)

depends on if their incremental spending exceeds the price cuts they obtain on promoted items (Guyt 2015).

We begin our analysis of multiple store patronage and its impact on consumer reaction to promotions as an empirical issue that is yet to be examined. Our analysis will shed light on how consumers based on their patronage behavior react to Savings Week events and provide several useful insights.

(19)

Modelling Framework

Similar to Zhang and Breugelmans (2012) and Guyt (2015), we build a joint model to examine the effects of Savings weeks promotions on store visit incidence and store

spending. To estimate the model, we use a two-step approach. In the first step, we model store visit incidence based on whether household (h) visits retailer (r) in a given week (w) or not. In the second step, we model household’s spending level conditional on a visit to retailer in that week. We use the same households for both models.

Store Visit Incidence model

Like Guyt (2015), we only focus on whether households visit a retailer or not in a given week, instead of which retailer they visit. To model store visit incidence, we assume that household (h) visits retailer (r) in week (w) if and only if the utility of visiting a retailer exceeds a threshold. Let = 1 if household (h) visits retailer (r) in week (w), and let = household’s (h) utility of visiting retailer (r) in week (w). A household may visit multiple retailers: = 1 for all retailers that household visits. Therefore,

(1) = 1, > 0

= 0, ℎ

The utility is specified as:

(2) = +

= + +

where and are the parameters to be estimated, and is a matrix of explanatory variables (explained in the following section) which contains household,

retailer-week specific drivers for the utility function. Assuming that the random component follows an extreme value distribution, the probability of household (h) visiting retailer (r) in week (w) can be obtained using the familiar logit formulation given below. Therefore, the store visit incidence model becomes a simple binary logit model.

(3) { = 1} = ( )

(20)

Store Spending model

Let ∗ be a latent variable that derives household (h) spending at retailer (r) in week (w), and let be household (h) observed spending at retailer (r) in week (w).

Conditional on a household’s (h) visit to retailer (r) in week (w) i.e. ( = 1), store spending can be modelled such that

(4) = ,

∗

= 1 = 0, = 0

Therefore, conditional store spending ( | = 1) is modelled as follows:

(5) ∗ = + + log( ) + (1 − ) ∗ (1 − ) +

where and are the parameters to be estimated, and is a matrix of

explanatory variables which contains household, retailer-week specific variables related to spending (explained in the following section). is a random component which follows a normal distribution: ~ (0, ).

Similar to Guyt (2015), we link the visit and spending equations by using the

approach suggested by Dubin and McFadden (1984): we include the term log( ) +

∗ ( ) in order to correct for the non-random occurrence of store visits1_.

Otherwise the parameter estimates will be biased because store visit and spending are highly interrelated decisions. The correction factor ( )captures that correlation between store visit and store spending.

Customer Segmentation

We use a concomitant variable finite-mixture model to examine the differential effects of Savings week events across shopper segments. In finite-mixture model, all the parameters are segment specific and the latent segment membership of individuals is formulated as a function of household characteristics (Gupta and Chintagunta 1994). Since

1_{The Dubin and McFadden term requires individual probabilities for each retailer-week combination. As equation 3}

(21)

we are interested in how shoppers with a wide store patronage and narrow store patronage may differ in their response towards promotional events, we include characteristics that capture households’ underlying store patronage behavior. Specifically, we include the average weekly number of stores visited by households, the average weekly number of trips made by households, and the average weekly shopping expenditures of households as concomitant variables to derive households’ segment membership probability. Together, these variables describe the overall patronage behavior of households towards the stores and may assist in distinguishing between shoppers with wide and narrow store patronage (see Baltas et al. 2010). Similar to Zhang and Breugelmans (2012), we adopt the following formulation for segment membership probability of each household ( ):

(6) = exp +

1 + ∑ exp( + )

Where and ( = 1 ) are the parameters to be estimated indicating the intercept and the effects of concomitant variables on prior segment probability. The parameters and are fixed to zero for identification purpose. Also, notice that is a time-invariant matrix of household specific variables. We use maximum likelihood

estimation to estimate equation 3 and 5 using concomitant variable finite mixture model. Note that we use concomitant variable approach separately in visit incidence and store spending equation. Due to technical issues driven by the large number of observations per household, it was not feasible to estimate the joint segment membership probability based on visit incidence and store spending. Therefore, we separately include household

characteristics as concomitant variables to determine the segment membership probability in visit incidence model ( ) and store spending model ( ).

Thus, the store visit model becomes a binary logit with finite mixture distribution, also known as concomitant variable mixture model (Wedel 2002) and the log likelihood is:

(22)

And, the store spending model becomes the latent class regression with concomitant variable or mixture regression (DeSarbo and Cron 1988) and the log likelihood is:

(8) = ∑ ∑ ∏ ∏ 2 ∗ − − |

∗

(23)

Data and Operationalization

Data Description

The analysis is performed using GfK scanner panel data which contain information on household purchases and promotional activities of all Dutch retailers for the period of three years (2009-2011). For analysis purpose, we focus on the top seven Dutch retailers based on their market share. To estimate the models, we take a random sample of 500 households who remained present during the entire observation period. All the households on average make 2.36 trips in a week each year. Table 2 provides descriptive statistics of the chosen retailers. As you can see, Albert Heijn has the highest market share in the sample followed by C1000, Aldi, Lidl, Jumbo, Plus, and SDB. Together these retailers capture 60% of the Dutch grocery market. Albert Heijn also has the highest visit rate averaged across

households followed by C1000 and Aldi. The average weekly spending among households at Albert Heijn and Jumbo is approximately similar. Furthermore, it appears that the average visit rate per week at chains depends on their relative market share whereas the average spending is more related to retail format as the HILO retailers seem to steal more cash on average than the Hard Discounters.

Table 2: Descriptive Statistics of Retailers

Retailer Format Market Share

(in sample)

Average Visit Ratea

Average Spending (Euros)b

Albert Heijn Hi-Lo .35 .46 37.37

C1000 Hi-Lo .18 .28 31.55

Aldi Hard discounter .14 .26 26

Lidl Hard discounter .10 .23 22.29

Jumbo EDLP .10 .13 37.39

Plus Hi-Lo .08 .12 32.07

SDB Hi-Lo .05 .10 27.73

a_{Fraction of weeks with a household visit, averaged across households}

b_{Average weekly spending of households at each retailer, averaged across households}

To identify the promotional weeks (Savings Weeks) we use the similar methodology as Guyt (2015). We utilize the dataset containing descriptive information on the

promotional events across retailers. The dataset contains names of the events as

communicated by the retailer and their timing. We use the same events as examined by Guyt (2015). In total, five events are chosen and all of them occur at HILO retailers.

(24)

Moreover, all the selected events run at least twice in 3 years and have above-normal promotional activity considering the number of SKUs on promotions during the event weeks and regular weeks.

Table 3 provides descriptive statistics of the selected “Savings Weeks” events. A promotional event lasts three weeks on average with a maximum of 4 weeks. Albert Heijn’s “Hamsterweken” and SDB’s “Super Toeter Weken” have the longest average duration in terms of weeks. Furthermore, the total number of SKUs on themed promotion at each retailer is substantially larger than the regular weeks which indicates the distinct nature of the “Savings Weeks” promotions as compared to regular promotions.

Table 3: Descriptive Statistics of Savings Week Events

Retailer Event Mean duration in weeks Total Weeks Frequency SKUs on themed promotiona

Albert Heijn Hamsterweken 3.17 19 6 1543 (653)

C1000 Euroweken 2.57 18 7 1168 (526) C1000 I Love Gratis Weken 3 6 2 1203 (526) Plus Hollandse Prijzweken 3 12 4 652 (335) SDB Super Toeter Weken 3.33 20 6 695 (337)

a _{Number of SKUs on promotion during Savings weeks and regular weeks (in bracket)}

Variables and Operationalization

Here we explain the operationalization of explanatory variables used in the store visit and spending model, as shown in Table 6. We include several control variables to capture the true effects of promotional events on store visit incidence and store spending. First, we add a dummy variable ( _ ) for each retailer representing the baseline visit propensity and spending at each retailer. To capture intrinsic preferences of households to visit a particular retailer, we include a state dependence variable

( _ ) in the visit incidence model which equals to1 if households visited the focal retailer in the previous week. Similarly, we add last spending variable ( _ ) in the spending model representing households’ expenditures (in Euros) at the focal retailer in

(25)

the previous week. We also include average weekly expenditures of households at each retailer ( _ ) in the spending model. Next to that, we add a retailer share variable ( _ ℎ ) measured as the share of household visits to the focal retailer during the observation period. We also take into account the seasonality affects by including dummy variables for Easter ( _ ), Christmas ( _ ℎ ), and the first week of the year ( _ 1 ).

For Savings week events, we add a separate dummy variable ( ) for each of the five promotional events which equals to 1 if there was an event going on at the focal retailer and zero otherwise. These event dummies capture the immediate effect of promotional events on store visit incidence and spending. Moreover, we control for

competitive events by adding a ( _ ) dummy variable which equals to 1 if there is an event going on at any competing retailer during the focal week and zero otherwise. Such large scale promotional events are more likely to induce consumers to engage in stock piling and purchase acceleration. Therefore, we also include lag ( _ ) and lead event ( _ ) dummy variables to capture the dynamic effects of Savings week events.

As indicated earlier, the aim of this study is to examine the segment specific effects of promotional events on store visit incidence and spending based on consumers’ patronage behavior towards the stores. Therefore, we add some household specific variables for segmentation purpose. First, we include the average weekly number of stores patronized by the household ( _ ) during the entire observation period. This variable captures the extent to which households use multiple stores for their shopping needs based on the average of total number of stores households visit each week. Next to that, we add average weekly number of trips ( _ ) made by the households to capture their shopping frequency. Finally, we include average weekly expenditures in Euros ( _ ) of households which captures another significant aspect of consumer shopping behavior and assist in differentiating between large versus small basket shoppers.

Our choice of these household characteristics for segmentation is based on the fact these variables determine consumers’ tendency to use multiple stores (see Baltas et al. 2010). The author also indicates other individual specific factors such as demographic

(26)

characteristics which may influence multiple store patronage, however, such individual level data is not easily available. Also, note that we use the same households for store visit and spending model, and therefore, the same characteristics to segment households in both models. We acknowledge the fact that the factors which influence households’ visit

decisions might not be same as for households’ spending decisions. However, those factors are beyond the scope of this study. Our focal of interest is to distinguish households based on their multiple store patronage behavior which may in turn impact their store visit and spending decisions.

Below we present descriptive statistics of household characteristics used for segmentation and their correlation (Table 5 and 6). Generally speaking, household

characteristics indeed display a fair amount of dispersion in households’ shopping behavior, which will be useful for segmentation. Furthermore, the number of stores households patronize is in accordance with the previous research (Popkowski Leszczyc and Timmermans 1997) which found that most of the households patronize 2 to 5 different stores. The

number of stores households patronize and their frequency of visits seems to be highly correlated, as expected. However, with respect to their spending the correlation is reasonable and positive.

Table 4: Descriptive Statistics of Household Characteristics

Characteristics Min. Mean Max.

_ 1 1.71 4.93 _ 1 2.39 7.85 _ 3.5 49.65 162.1 Table 5: Correlation _ _ _ _ 1 _ .82 1 _ .33 .44 1

(27)

Table 6: Variables and Operationalization

Variable Name Description Visit Spending

C O N TR O L V A R IA B LE S

_ Dummy equal to 1 for retailer r  

_ Dummy equal to 1 if household h visited

retailer r in the preceding week 

_ Weekly spending (Euros) of household h at

retailer r in the preceding week 

_ Average weekly expenditures (Euros) of

household h at retailer r 

_ ℎ Share of household h’s visits at retailer r

during the entire observation period  

_ Dummy equal to 1 during the Easter week  

_ ℎ Dummy equal to 1 during the Christmas

week  

_ 1 Dummy equal to 1 for the first week of the

year   P R O M O TI O N V A R IA B LE S

1 Dummy equal to 1 during “Euroweken”

event at C1000  

2 Dummy equal to 1 during

“Hamsterweken” event at Albert Heijn   3 Dummy equal to 1 during “Hollandse

Prijsweken” event at Plus  

4 Dummy equal to 1 during “I Love Gratis

Weken” event at C1000  

5 Dummy equal to 1 during “Super Toeter

Weken” event at SDB  

_ Dummy equal to 1 if there is an event at

any competing retailer in week w  

_ Dummy equal to 1 indicating an event in

the preceding week  

_ Dummy equal to 1 indicating an event in

following week   SE G M EN TA TI O N V A R IA B LE S _

Average weekly no. of stores visited by household h during the entire observation period

 

_ Average weekly no. of trips of household h

during the entire observation period  

_

Average weekly expenditures (Euros) of household h during the entire observation period

(28)

Estimation Results

Store Visit Incidence

Table 7 presents model comparison for the visit incidence models using different number of latent segments. A six-segment model appears to fit the data best among the estimated models according to Bayesian information criterion (BIC)2_{. We present the}

estimation results in Table 8. Overall, the model shows distinct patterns among households’ purchase behavior and their response towards promotional events. The magnitude of the impact of promotional events on store visit propensity differ largely across segments. Also, the underlying characteristics of households across segments differ.

Table 7: Comparison of Store Visit Incidence Models

No. of segments No. of parameters N LL AIC3 BIC

1 20 450177 -112844.7 225749.4 225949.7 2 44 450177 -100484.8 201101.6 201542.4 3 68 450177 -99245.1 198694.2 199375.4 4 92 450177 -98373.5 197023 197944.6 5 116 450177 -97814.618 195977.2 197139.3 6 140 450177 -97522.397 195464.8 196867.2

Table 8 reports parameter estimates for the six-segment visit incidence model. For most of the control variables, we find significant effects with the expected signs. Last visit variable which indicates shoppers’ tendency to revisit the retailer they shopped at before has a positive significant effect for all segments. However, the magnitude of effects differs across segments with the largest coefficient (4.19) for segment 5 and the smallest

coefficient (.176) for segment 2. Similarly, the coefficient of retailer share is positive and significant in line with the previous findings. Retailer share variable indicates the impact on the likelihood to visit retailer based on their respective share i.e. the higher the retailer share the higher the visit likelihood. Loosely speaking, segment 3 and segment 6 appear to be the most loyal segments in terms of the size of the retailer share coefficient (14.593 and 12.578). Additionally, most of the seasonality effects are insignificant across segments. We only find a positive significant effect for Easter on segment 2 and segment 3, and a positive significant effect for Christmas on segment 3.

2_{Please note that the estimation time of these models is rather prohibitive, hence it was not possible to test these models}

(29)

Table 8: Estimation Results of Store Visit Incidence Model

Variables Segment 1 Segment 2 Segment 3 Segment 4 Segment 5 Segment 6

Store Visit Incidencea: -5.672*** -3.481*** -3.747*** -3.774*** -5.344*** -6.146*** _ 1000 2.06*** 0.025 -0.008 0.403*** 0.306** 2.198*** _ 2.353*** 0.025 0.261*** 0.293*** 0.224 2.531*** _ 1.094*** -0.474*** -0.139* -0.011 -0.578*** 2.128*** _ 2.358*** 0.216*** 0.242*** 0.334*** 0.148 2.307*** _ 1.402*** -0.638*** 0.134* 0.206*** 0.045 2.021*** _ 2.375*** 0.352*** 0.342*** 0.37*** 0.145 2.551*** _ 0.467*** 0.176*** 0.791*** 1.803*** 4.19*** 0.254*** _ ℎ 7.288*** 9.547*** 14.593*** 7.299*** 6.891*** 12.578*** _ 0.178 0.221* 0.319*** 0.182 0.287 0.057 _ ℎ -0.008 -0.024 0.277** 0.115 -0.196 -0.103 _ 1 0.118 -0.146 0.034 0.064 -0.194 0.053 1 0.677*** 0.367** 0.593*** 0.253** 0.351 0.58*** 2 0.289*** 0.494*** 0.281** 0.293*** 0.318 0.396*** 3 -0.259 0.098 0.066 0.12 0.613 0.211 4 0.247 -0.017 0.019 0.217 0.519 -0.433** 5 -0.389 0.24 0.258** 0.326** 0.58** 0.451 _ -0.001 0.038 0.041 0.001 -0.028 0.014 _ -0.073 -0.029 0.323*** 0.016 -0.052 0.18 _ -0.314** -0.14 0.156 0.029 0.205 -0.237* Segment Membership: - -11.436*** -25.934*** -10.426*** 6.067*** -12.743*** _ - 5*** 9.314*** 3.024** -9.967*** 2.976** _ - 2.007** 3.616*** 2.774*** 2.924*** 3.686*** _ - 0.008 0.033** 0.024** 0.029*** 0.032*** 0.205 0.13 0.197 0.135 0.153 0.179 No. of Parameters 140 Loglikelihood -97522.397

a_{Retailer SDB is set as reference} * _{p-value < .10}

** _{p-value < .05} *** _{p-value < .01}

(30)

Turning to the Savings Week effects, we find significant dummy coefficients for event 1 (“Euroweken” at C1000), event 2 (“Hamsterweken” at Albert Heijn), event 4 (“I Love Gratis Weken” at C1000), and event 5 (“Super Toeter Weken” at SDB). “Euroweken” and “Hamsterweken” has a positive significant effect on the store visit propensity of all segments except segment 5. However, the magnitude of event effects varies across segments. Segment 1, 3, and 6 react more strongly during “Euroweken” as compared to other segments whereas Segment 2 and 4 react more strongly during “Hamsterweken” as compared to other segments. We do not find a significant effect for event 3 (“Hollandse Prjseweken” at Plus) on any of the segments. “I Love Gratis Weken” is found to reduce the visit likelihood of segment 6 during promotion weeks. “Super Toeter Weken” has a positive significant effect on the visit propensity of segment 3, 4, and 5 with the largest impact on segment 5. The coefficient for the competing events dummy is insignificant for all segments suggesting that competing events do not affect the likelihood to visit competing retailers.

Furthermore, we find evidence that segment 1 and segment 6 postpone store visits in anticipation of events, as shown by the negative significant coefficient for lead effect. However, we do not find any significant effects for the lagged event coefficient, except for segment 3 for which the sign is positive indicating an increase in the revisit tendency of segment 3 during the post-event weeks. Therefore, there is no substantial evidence to conclude post-event effects.

In the concomitant variable model, most of the effects of concomitant variables on the prior segment membership probabilities are significant. As compared to segment 1, the average weekly number of stores has a positive significant impact on the prior latent class membership for segment 2, 3, 4, and 6. This suggests that households who on average visit multiple stores in a week are more likely to fall in segment 2, 3, 4, and 6. Similarly, weekly average of the number of trips made by the households and spending have a positive significant impact on the prior class membership for segment 2, 3, 4, 5, and 6 as compared to segment 1. Furthermore, the size of segment 1 is the largest (20.5%) followed by segment 3 (19.7%), 6 (17.9%), 5 (15.3%), 4 (13.5%), and 2 (13.0%).

(31)

Segment description and Savings Week impact

Combining the effects of concomitant variables (household characteristics) and the estimated segment membership probabilities derived from them indicates that households with tendency to visit multiple stores belong to segment 2, 3, 4, and 6. Table 9 provides an average of number of stores patronized by households, average of number of trips made by households, and average spending of households within each segment over the entire observation period. It appears that on average households in segment 3 visit a larger number of multiple stores followed by segment 6, 2, 4, 1, and 5. Based on the usage of multiple stores, shopping frequency, and spending, segment 3 perfectly depicts the characteristics of households with a multiple store patronage (e.g. Baltas et al. 2010). Segment 2, 4 and 6 also share similar characteristics with segment 3. Households in segment 1 and 5 on average are more likely to stick to their preferred stores with fewer amount of shopping trips.

Table 9: Segment Description in Store Visit Incidence Model

Segments Mean Stores Mean Trips Mean Spending

1 1.14 1.30 32.91 2 1.50 1.90 40.99 3 2.36 3.59 59.97 4 1.48 2.11 50.93 5 1.07 1.68 50.42 6 1.56 2.58 54.83

To ensure that the segment membership does not adhere to the user base of any specific retailer, we cross examine the households in each segment with their share of each retailer. Below we present the percentage of retailers’ loyal customer base within each segment (see Table 10). Retailer’s loyal customer base represents households with the retailer share of 50% or higher. There is no clear evidence that loyal customer base of retailers is confined to any single segment, though some segments contains a much larger proportion of loyal customers. If we look at segment 3 in which multiple store patronage is most common, we find that this segment contains the least amount of loyal customers for almost all of the retailers. Whereas the rest of the loyal customers are distributed across other segments.

(32)

Table 10: Percentage of Retailers’ Loyal Customer Base across Segments

Segment 1 Segment 2 Segment 3 Segment 4 Segment 5 Segment 6

SDB 0% 0% 0% 44.4% 44.4% 11.1% C1000 23.7% 17.0% 0% 10.2% 18.6% 30.51% Albert Heijn 24.6% 9.5% 6.3% 15.1% 26.2% 18.25% Plus 27.3% 13.6% 4.5% 9.1% 18.2% 27.3% Aldi 32.6% 13.0% 4.3% 10.9% 17.4% 21.8% Jumbo 25.0% 8.3% 8.3% 4.2% 25.0% 29.2% Lidl 48.6% 20.0% 2.9% 8.6% 11.4% 8.6%

The segment specific effects we found above are mostly according to our

expectations. For instance, the segment specific effects we found for the last visit coefficient pointed that the tendency to revisit the retailer visited before is substantially large for segment 5 as compared to other segments. It makes sense given the fact households in segment 5 tend to stick to a single store. We also found that segment 3 and segment 6 have the strongest impact for the retailer share variable as compared to other segments which appears counter intuitive considering the large number of multiple stores visited by these segments. However, a logical explanation is that segment 3 and segment 6 are more likely to revisit their preferred retailers but also visit other retailers to find best deals which is in accordance with consumer price search behavior (e.g. Gauri et al. 2008). Moreover, we found that 3 out 5 Savings Week events have a positive significant impact on the store visit propensity of segment 3. Similarly, we found a positive significant effect of 3 out 5 Savings Week events on the visit propensity of segment 4. Segment 2 and segment 4 have a positive significant impact of 2 Savings Week events. On the other hand, segment 5 which includes households with the least usage of multiple stores has a positive significant effect for only one Savings week event. This suggest that, indeed, households who are likely to visit multiple stores are more responsive towards Savings Week events.

However, we also found positive significant effects of two Savings Week events on segment 1 which also includes households with less usage of multiple stores. One

reasonable explanation is that these households may not be inclined towards visiting other stores for promotional benefits, but rather wait for promotions within their preferred stores. Consistent with this, we found a negative significant lead effects, which indicates that households in segment 1 delay their purchases in anticipation of promotional events.

(33)

To get a better grip on the magnitude of Savings Week effects across segments, we also calculate the percentage change in the average visit propensity of households at the focal retailer across segments (see Table 11). In segment 3, the average visit propensity of households increases by 5.6%, 3.7%, and 2.4% during “Euroweken”, “Hamsterweken”, and “Super Toeter Weken”. In segment 4, the average propensity of households increases by 3.1%, 4.3%, and 3% in response to these events. “Euroweken” has lower impact (3.5%) on segment 2 as compared to segment 3 whereas “Hamsterweken” has a larger impact (5.9%). On the other hand, “Super Toeter Weken” increases the average visit propensity of

households in segment 5 by 1.1% which is minimal as compared to the promotional impact on segments with multiple store patronage. Finally, “Euroweken” and “Hamsterweken” bring an increase of 6% and 3.6% in the average visit propensity of segment 1.

Table 11: Percentage Change in Average Visit Propensity of Households during Savings Week Events3

Events Segment 1 Segment 2 Segment 3 Segment 4 Segment 5 Segment 6

Euroweken 6%* 3.5%* 5.6%* 3.1%* 1.9% 4%* Hamsterweken 3.6%* 5.9%* 3.7%* 4.3%* 1.1% 3.7%* Hollands Prijsweken -.3% 0% .1% .5% .8% .7% I Love Gratis Weken .2% -1.3% .1% 3.9% 1.7% -4%* Super Toeter Weken -.1% 1.7% 2.4%* 3%* 1.1%* .8% * significant events

Store Spending Model (conditional on visit)

Table 12 presents model comparison for the store spending models using different number of latent segments. A six-segment model appears to fit the data best among the estimated models according to Bayesian information criterion (BIC)4_{. The estimation results} are given in Table 10. Overall, the model shows distinct patterns among households’

spending behavior and their response towards promotional events. In contrast with visit incidence model, strong seasonality effects are present in the model.

3_{To compute the percentage change in the average visit propensity of households at the focal retailer, we compute the}

predicted probabilities of households with and without the event taking place within each segment using the segment specific estimates in Table 6, then took the average across households.

(34)

Table 12: Model Comparison of Store Spending Models

No. of segments No. of parameters N LL AIC3 BIC

1 23 101001 -435627.8 871324.6 871520.6 2 50 101001 -423566.6 847283.2 847709.3 3 77 101001 -420015.3 840261.6 840917.9 4 104 101001 -418679.1 837670.2 838556.6 5 131 101001 -418100.3 836593.6 837710.1 6 158 101001 -417585.9 835645.8 836992.4

Table 13 reports the parameter estimates of the six-segment store spending model (conditional on visit). Similar to the visit incidence model, we find strong influence of household’s previous spending at retailer (r) on their current spending at the retailer. We find positive significant effect of last spending coefficient across all segments. We

additionally control for average weekly spending of households at retailer (r) during the entire observation period, which also have positive significant for all segments. However, we do not find a significant effect for retailer share coefficient in the store spending model. Contrary to the visit incidence model, there are strong seasonality effects on households weekly spending at the retailer. We find positive significant effects for Easter, Christmas, and Year’s first week dummy coefficients in the all segments. Based on the size of the coefficients, segment 2 and 4 seems to have the largest increase in their weekly spending level at the focal retailer during these holiday periods.

Coming to the Savings Week effects, we find significant dummy coefficients for event 1 (“Euroweken” at C1000), event 2 (“Hamsterweken” at Albert Heijn), event 3 (“Hollandse Prjseweken” at Plus), and event 4 (“I Love Gratis Weken” at C1000). However, not all events affect equally across segments. The events at C1000 (“Euroweken” and “I Love Gratis Weken”) have a positive significant effect on segment 1 whereas “Euroweken” and “Hamsterweken” have a positive significant effect on segment 3 and segment 5. On the other hand, “Hollandse Prijsweken” has a negative effect on segment 2 and

“Hamsterweken” has a negative effect on segment 4. We do not find any significant effect for promotional events on segment 6. Like visit incidence model, the coefficient for the competing events dummy is insignificant for all segments except segment 5, suggesting that competing events do not affect spending at competing retailers for most of the segments.

(35)

Table 13: Estimation Results of Store Spending Model (Conditional on visit)

Variables Segment 1 Segment 2 Segment 3 Segment 4 Segment 5 Segment 6 Store Spendinga_: -0.085 -0.971 -0.029 0.740 -0.280 0.270 _ 1000 -0.766 0.161 -0.207 -0.326 -0.361 -0.264 _ -0.200 -0.199 -0.236 -0.242 -0.331 -0.255 _ -0.154 0.657 -0.150 -0.180 -0.195 0.024 _ -0.060 0.298 0.010 -0.870 -0.051 -0.297 _ -0.058 -0.432 0.042 -0.787 -0.153 -0.361 _ -0.047 0.010 -0.024 -0.720 -0.042 -0.269 _ 0.073*** 0.098*** 0.079*** 0.089*** 0.089*** 0.064*** _ 0.932*** 0.906*** 0.924*** 0.913*** 0.913*** 0.939*** _ ℎ -0.017 0.214 -0.191 -0.515 -0.011 -0.290 _ 1.093** 11.793*** 2.009*** 10.886*** 5.070*** 3.798*** _ ℎ 2.484*** 17.805*** 4.902*** 8.940*** 7.632*** 5.305*** _ 1 2.212*** 9.043** 0.712 7.049*** 1.300 3.912*** 1 3.424*** -1.385 1.253** 0.382 1.730** 0.780 2 0.748 0.340 1.634*** -2.171** 2.390*** 0.340 3 0.780 -12.229** 1.348 -3.465 1.012 -1.700 4 2.604** -1.035 -0.343 -2.082 2.066 -0.417 5 -0.935 0.534 0.291 -1.440 0.669 -0.969 _ 0.073 1.277 0.115 -0.727 0.533* 0.064 _ -0.575 -0.445 -0.142 -1.726 -0.693 -1.628*** _ 1.553*** -4.436* 0.507 -4.095*** -0.585 -1.251** 0.013 0.004 0.079 0.164 0.013 0.118 6.409 34.930 9.398 23.748 17.441 12.978 Segment Membership: - -4.743*** -0.667 -2.816*** -1.091 -1.190 _ - -12.746*** -2.653** -10.685*** -7.377*** -5.409*** _ - 1.211 0.512 1.357* 0.801 0.915 _ - 0.571*** 0.199*** 0.510*** 0.422*** 0.340*** .084 .084 .156 .162 .286 .228 No. of Parameters 158 Loglikelihood -417585.9

a_{Retailer SDB is set as reference}

*_{p-value < .10}

** _{p-value < .05}

(36)

Furthermore, we find pre-event decrease in the spending across segment 2, 4, and 6, as shown by the negative lead effect coefficient for these segments. Surprisingly, we find an increase in the spending level of households in segments 1 before the events. Segment 6 also shows post-event effects by reducing their spending in the weeks after event.

Therefore, we find evidence for pre-event effects for 3 out 6 segments. For segment 6, both the pre- and post-event effects are present. Other than that, there is no substantial

evidence for post-event effects.

For the concomitant variable model, the effects of average weekly number of stores patronized by the households and their spending has a significant effect on the prior latent class membership of households. However, average weekly number of trips do not have substantial impact on the prior membership probability. Given the negative signs of mean stores on the segment membership probability, it appears households with the increased usage of multiple stores are more likely to fall in segment 1. Contrarily, the positive signs for average spending variable indicates that the average spending of households in segments (2 to 6) is higher as compared to segment 1. Furthermore, the size of segment 5 is the largest (28.6%) followed by segment 6 (22.8%), 4 (16.2%), 3 (15.6%), 2 (8.4%), and 1 (8.4%).

Segment description and Savings Week impact

Using the estimated effects of concomitant variables (household characteristics) segment specific parameters, we calculate the latent segment membership of each household. Table 14 provides average of the household characteristics (averaged across households) within each segment for the entire observation period. It appears that households in segment 1, 3, and 6 on average visit numerous stores for their shopping needs, followed by segment 4 and 5. Whereas segment 2 contains households with the least usage of multiple stores. With respect to their shopping frequency, all segments display similar patterns. However, segments with more usage of multiple stores have on average lower spending level as compared to segments with fewer usage multiple stores. The previous evidence on the relationship between spending level and multiple store patronage is mixed. Some researchers have argued that greater spending increases the search for best deals which in turn enhances the benefits of patronizing multiple stores (Baltas et al. 2010; Magi 2003). However, it has also been suggested that heavy spenders tend to be more loyal and opt for fewer stores (East et al. 1997; McGoldrick and Andre 1997. Since our prime

(37)

focus is on the number of stores households patronize (Mean Stores), clearly segment 1 has the least usage of multiple stores as compared to other segments.

Table 14: Segment Description in Store Spending Model

Segments Mean Stores Mean Trips Mean Spending

1 2.31 3.20 24.48 2 1.58 2.67 93.43 3 2.27 3.09 35.65 4 1.80 2.94 78.03 5 1.90 2.67 56.79 6 2.15 3.06 49.99

Like visit incidence model, we also examine whether the segment membership is driven by the retailer share of households (see Table 15). Again, we do not find any influence of the retailer share on the segment membership of households. The loyal customer base of retailers is fairly dispersed across segments. The distribution of households across segments in the store spending model is quite similar to what we observed in the visit incidence model. Segment 1 which represents households with increased preference for multiple store patronage contains the least amount of loyal customers for most of the retailers. And other segments capture the rest of the loyal customer base.

Table 25: Percentage of Retailers’ Loyal Customer Base across Segments

Segment 1 Segment 2 Segment 3 Segment 4 Segment 5 Segment 6

SDB 11.1% 11.1% 11.1% 22.2% 22.2% 22.2% C1000 3.4% 11.9% 15.3% 18.6% 30.5% 20.3% Albert Heijn 5.6% 15.% 13.5% 23.85 23.0% 19.05 Plus 4.2% 12.5% 25.0% 25.0% 29.2% 4.2% Aldi 10.9% 0% 21.8% 6.5% 26.1% 34.8% Jumbo 8.3% 20.8% 0% 16.7% 37.5% 16.7% Lidl 20.0% 0% 14.3% 11.4% 31.4% 22.9%

Regarding the impact of Savings Week events, we do find a stronger response among segments with the increased usage of multiple stores. However, the results are vague for some segments. For segment 1 which contains households with most usage of multiple stores, we found a positive increase in the spending level during the Savings Week events at C1000. Similarly, segment 3 and 5 are found to respond positively towards promotions

(38)

during “Euroweken” and “Hamsterweken”. On the other hand, the segment with the least usage of multiple stores (segment 2) is found to respond negatively towards “Hollandse Prijs Weken”. We did not find a positive effect of any promotional event on this segment. For segment 4 and 6, the results are unclear, both segments entail households with the higher usage of multiple stores on average, but do not find a significant change in their spending levels during promotional events. In fact, segment 4 respond negatively towards

“Hamsterweken” promotions. Surprisingly, these two segments showed significant pre-event effects with a reduction in their spending before the pre-events.

To examine the magnitude of Savings Week effects, we report the percentage change in the average spending of households at the focal retailer across segments in Table 16. In segment 1, the average spending of households increases by 13.12% and 10% during “Euroweken” and “I Love Gratis Weken”. Similarly, the average spending of households in segment 3 increases by 2% and 9.8% during “Euroweken” and “Hamsterweken”. A small percentage increase of 1.2% and 4.4% is also found for segment 5 during “Euroweken” and “Hamsterweken”. All of these segments include households with a higher tendency to patronize multiple stores. On the other hand, households in segment 2 with the least usage of multiple stores shows a decrease of 21.3% in their average spending during “Hollandse Prijseweken”.

Table 16: Percentage Change in the Average Store Spending of Households during Savings Week Events5

Events Segment 1 Segment 2 Segment 3 Segment 4 Segment 5 Segment 6

Euroweken 13.2%* -4.7% 2%* -2% 1.2%* 1.6% Hamsterweken 8.5% 1.4% 9.8%* -4.8%* 4.4%* .35% Hollands Prijsweken 6.7% -21.3%* 1.1% -8.3% -.16% -12.3% I Love Gratis Weken 10%* -3.4% -7.9% -5.8% 7.8% -4% Super Toeter Weken -8.5% -6.6% 1.5% 2.2% 3.7% -1.4% * significant events

5_{To compute the percentage change in the spending we first compute the predicted weekly spending based}

on the estimates in Table 10 for each segment. Then for the focal retailer we calculate the percentage change in the spending (averaged across households) when the Savings week dummy is on.

(39)

Discussion

Conclusion

Retail patronage has always been a center of attention in grocery retailing. Recent advancements on consumer patronage behavior suggest that consumers have a tendency to patronize multiple stores i.e. “multiple store patronage” (Baltas et al. 2010). Surprisingly, this phenomenon hasn’t received much attention in the sales promotion literature. In this study, we attempt to consolidate multiple store patronage and its impact on consumer reaction towards promotions. By focusing on large scale promotional events known as “Savings Weeks” across several Dutch retailers, we perform a segment-level analysis of such events based on consumer underlying patronage behavior. Specially, we examine how consumers who patronize multiple stores for their shopping needs differ from consumers who patronize few stores in their response towards promo events.

With respect to store visit incidence, we find evidence that segments with multiple store patronage are more responsive towards Savings Week events. All the four segments (segment 2, 3, 4, and 6) which represent households with increased usage of multiple stores show a positive increase in store visit propensity during Savings Week events. Particularly, segment 3 which entails households with highest usage of multiple stores respond positively to 3 out of 5 promo events. Conversely, segment 5 which contains households with the least usage of multiple stores only respond to 1 of the 5 events investigated. For segment 1, we find a positive impact of two events which was unexpected because that segment share similarities with segment 5. Further analysis of this segment reveals that it is composed of a large proportion of loyal customer base of retailers to whom promotional events they respond to.

Similarly, we also find evidence of an increased spending, given that the store is visited, among households who are likely to visit multiple stores for most of the segments. Segment 1, 3, and 5 which includes households with increased preference for visiting multiple stores are found to be more responsive towards the Savings Week events.

Contrarily, segment 2 with the least tendency to visit multiple stores is negatively affected by the promotional events. However, our results remain inconclusive for segment 4 and 6. While these segments represent characteristics of multiple store patronage, we do not

The empirical investigation of how "multiple store patronage" influences consumer reaction towards promotions