The impact of door-to-door store flyer content
decisions on customer behavior: a quantitative
research
Assessing the effect of a retailers’ door-to-door store flyer content decisions on
multiple store performance metrics
by
Nick Roode
The impact of door-to-door store flyer content
decisions on customer behavior: a quantitative
research
Assessing the effect of a retailers’ door-to-door store flyer content decisions on
multiple store performance metrics
by
Nick Roode
University of Groningen Faculty of Economics and Business
MSc Marketing
Master thesis
February 2018
Louise Henriëttestraat 25a 9717LJ Groningen
+316 51583039 N.Roode.1@student.rug.nl
Student number: 3027481
First supervisor: prof. dr. T.H.A. Bijmolt Second supervisor: prof. dr. J.E. Wieringa
PREFACE
This is it: my Master thesis. The final work in my pursuit of a double-track Master’s degree in Marketing Intelligence & Management at the University of Groningen. An article in which I assess the effects of a retailers’ store flyer content decisions on multiple store performance metrics using the knowledge and skills acquired during my academic journey.
This Master thesis was written under the supervision of prof. dr. Tammo Bijmolt. I benefited greatly from the valuable comments on my research and the insightful discussions. It was an honor and pleasure to work with such an established academic. Second, I would like to thank PhD researcher Saeid Vafainia from KU Leuven for his concise feedback and attention for detail. Next to this academic guidance, I also want to thank my parents, sister, close friends and girlfriend for their unconditional support during my academic career. They have always been there for me and I can honestly say that I could not have done it without them. Lastly, I gratefully thank the retailer for providing me with the data for the study.
I hope that you have an equal amount of fun reading this article as I put in the effort to make it readable.
Nick Roode
TABLE OF CONTENTS
PREFACE ... 3
TABLE OF CONTENTS ... 4
ABSTRACT ... 6
1. INTRODUCTION ... 8
1.1 Problem statement and research questions ... 8
1.2 Academic and managerial relevance ... 9
1.3 Structure of the study ... 10
2. THEORETICAL FRAMEWORK ... 11
2.1 Store flyer ... 11
2.2 Store performance decomposition framework ... 11
2.3 Factors influencing store performance ... 12
3.3.2 Heteroskedasticity and nonnormality ... 36
3.3.3 Autocorrelation ... 36
4. RESULTS ... 37
4.1 Main results ... 37
4.1.1 Discount Size ... 39
4.1.2 Promotion Scope Width ... 41
4.2 Interaction effects of loyalty membership ... 42
4.3 Control variables ... 43
5. DISCUSSION ... 45
5.1 Conclusions and recommendations ... 46
5.2 Academic implications ... 48
5.3 Managerial implications ... 48
5.4 Limitations and future research ... 49
REFERENCES ... 51
APPENDIX ... 58
Appendix I: Distribution of store performance metrics ... 58
ABSTRACT
The store flyer has become of increasing importance to attract customers and subsequently influence their behavior. The creation of a store flyer involves significant costs and many complex decisions. In the face of this complexity and importance, retail managers should understand the effects of the store flyer content decisions on various store performance metrics.
The objectives of a retailers’ marketing expenditures can be classified into three categories: 1.) attraction, 2.) conversion, and 3.) spending effects. The thesis uses a decomposition framework to estimate the effects of store flyer content decisions on all the retailers’ objectives simultaneously and examine their effect on store sales. Specifically, store sales are decomposed into four components: store traffic, conversion ratio, number of products sold per basket and average price of products in basket. The latter two represent the spending effects.
A set of six hypotheses based on existing literature and theoretical argumentation are proposed. Data from a Dutch relatively large single-store retailer is used. The retailer sells a wide range of product categories in the home improvement sector. The data regarding the performance metrics, also split into loyalty program members and nonmembers, was gathered at the retailer for a period of approximately 2 years and 3 months. In that period 41 door-to-door store flyers have been sent out by the retailer. Data about the discount size and promotion scope (depth and width), along with other flyer variables (e.g. number of pages and volume) were extracted from the flyers and connected to retailers’ performance metrics dataset.
increase in store visitors and a decrease in conversion ratio. Both were found to positively affect the number of transactions. Also, partial support is found for our expectation that nonmembers compared to loyalty program members are more responsive to store flyer content decisions.
The findings have implications for both academics and practitioners. Academics can use the framework and modeling approach to more thoroughly assess the effects of promotions on multiple store performance metrics simultaneously. Practitioners are invited to use the insights from this thesis on the effect of flyer content decisions in their store flyer content strategy.
1. INTRODUCTION
The world may be going digital, but the brightly colored advertising flyers that spill out our mailboxes on a weekly basis somehow missed the memo about the decline of print. Going more than a century back in time, the promotional flyer has taken its name from the times when airplanes would fly over enemy cities during wartime and drop the ‘flyers’ for propaganda purposes. Nowadays, the store flyer is a frequently used promotional tool in retail business. In 2010, retailers in the US spent more than $20 billion on flyers (PRIMIR, 2013), which on average accounts for 65% of their marketing budget (Nielsen, 2012). European retailers are similar in their use of store flyers. Danish marketers spent €373 million on store flyers in 2010, whereas Spanish retailers have spent €595 million on store flyers in 2013 (Infoadex, 2014). In Italy and France, store flyers represent 50% and 60% of the average marketing budget (Nielsen, 2012).
Feature advertising can be defined as printed promotion materials run by retailers to inform consumers about the availability, price and promotions of products in their assortments (Pieters, Wedel, & Zhang, 2007:1815). The present research will focus specifically on the store flyer as a promotional tool. The store flyer is a crucial tool for practitioners, which is shown by its massive use by retailers. The increasing use of the store flyers is, in part, because they influence shoppers both at home and in store (Ziliani and Ieva, 2015) and, in part, since manufacturers often rely heavily on them to reach customers directly (Arnold, Kozinets, & Handelmanm, 2001; Srinivasan, Leone, & Mulhern, 1995). The promotions featured in the flyer represents therefore two sources, the manufacturer and the retailer, who both pursue different objectives with the flyer. The manufacturer main objective is to increase the sales of their branded products and retailers wants to increase the store traffic, overall sales and ultimately the store performance. This article has a focus on the objectives of the retailer and data from a relatively large single-store retailer is used.
1.1 Problem statement and research questions
get an understanding of the effect on overall store performance. Previous studies have made frameworks for analyzing multiple metrics of store performance (Lam, Vandenbosch, Hulland, & Pearce, 2001; van Heerde and Bijmolt, 2005), but to the best of our knowledge no previous study has utilized such a framework to empirically test the effects of the store flyer content decisions on multiple metrics of store performance. Given these current research gaps, this paper examines the following main research question:
What is the impact of door-to-door store flyer content decisions (discount depth and promotion scope) on a.) store traffic, b.) conversion ratio, c.) number of products bought per basket, d.) average price of products per basket and ultimately on e.) store sales?
In assessing the usefulness of our decomposition framework, we formulate the following sub-research question: Which modeling approach for predicting store sales performs best? With store flyer content decisions in the main research question, we are specifically interested in the effect of price discount depth and promotion scope (depth and width). So, What is the effect of deeper price discounts in store flyers on multiple store performance metrics? and what is the effect of a larger promotion scope (width and depth)? Given the limited flyer space we are also interested in the relative effects of promotion scope width and promotion scope depth. Hence, the following sub-research question is formulated: What is the relative effect of communicating more products per category versus adding more categories in the flyer on store sales? To complicate matters further, there are indications in the marketing literature that flyer effects might be moderated by the loyalty of customers (van Heerde and Bijmolt, 2005). This study will therefore also aim to specify the moderating role of customer loyalty. That is why the following sub-research questions are formulated: How does customer loyalty affect the responsiveness to store flyer content decisions? and What flyer content decisions have the most effect on loyal customers?
1.2 Academic and managerial relevance
Several studies have examined the effect of feature advertising on consumer behavior (Gedenk and Neslin, 1999; Niraj, Padmanabhan, & Seetharaman, 2008; Pancras and Sudhir, 2007; Zhang, Wedel, & Pieters, 2009), however they provide few generalizable insights for store flyer design, because store flyers characteristics strongly differ from those of advertisements published in newspapers or feature advertising using other media (Swoboda, Elsner, Foscht, & Schramm-Klein, 2010). Relatively few studies have specifically focused on the effect of promotions advertised through store flyers on customer behavior (Gijsbrechts, Campo, & Goossens, 2003; Gupta, 1988; Miranda and Kónya, 2007; Luceri, Latusi, Vergura, & Lugli, 2014; Schneider and Currim, 1991; van Heerde, Leeflang, & Wittink, 2004). This study has a significant academic contribution because it answers the call for research on how the store flyers influences consumer behavior. The offered decomposition framework in this study enables academics to more thoroughly and precisely estimate the effects of flyer content decisions on various performance metrics simultaneously. Besides that, this study gives empirical insights into the effects of, in the store flyer communicated, discount size and promotions scope on multiple store performance metrics. With these separate estimates, the effect on store sales can be calculated. The managerial contribution of this study is giving insights and guidelines for managers in retail environments to develop more effective store flyer content strategies. Managers will have insight into what the specific effects of their decisions are on store traffic, conversion ratio and customer spending.
1.3 Structure of the study
2. THEORETICAL FRAMEWORK
2.1 Store flyerVarious terms are used to refer to a store flyer, such as brochure, circular, shopping guide, handbill, pamphlet, catalogue and leaflet. In the present study we adopt the definition by Schmidt and Bjerre (2003: 379), in which they define a ‘flyer’ as “a frequently distributed free printed matter, part of the mass communication marketing from the sender(s), with a minimum of four pages, immediately readable, targeted at private households or firms”. A small adaptation in the current study is made to the definition offered by Schmidt and Bjerre (2003) by also including flyers with two pages. The retailer under study often uses flyers of two pages in the same way as flyers with more pages. A clear distinction is still made with feature promotions in newspapers; the store flyer comprises of multiple pages for typically a larger number of products. Also, a clear distinction is being made with the word: ‘printed’ in the definition, which indicates that this study will only involve the offline store flyer.
2.2 Store performance decomposition framework
attraction effect. Our decomposition of store performance allows us to examine these separate effects. The metrics combined can be used for an overall indicator of store performance.
Figure 1: Framework for analyzing Store Performance
2.3 Factors influencing store performance
Existing literature and theoretical arguments are used to form hypotheses about the effects of store flyer content decisions on the highlighted store performance metrics in figure 1. The moderating role of loyalty membership on these effects is also hypothesized.
2.3.1 Discount Size
Price promotions are a key marketing instrument used by retailers to generate more sales and increase their market share (Grewal, Ailawadi, Gauri, Hall, & Robertson, 2011). The size or depth of a discount is measured as the relative difference between the price during the price promotion period compared to the regular price. The “deep discount” strategy, in which some items are promoted at deep discounts to attract customers who will buy more profitable items once inside the store is a widely used strategy by retailers (Bliss, 1988; Lal and Matutes, 1994). Most retailers in practice use their store flyer to communicate discounts to their target audience (Gauri, Ratchford, Pancras, & Talukdard, 2017).
and their probability to buy. Literature suggests that the presence of a discount can enhance elaboration and make the item more salient (Gotlieb and Swan, 1990; Grewal, Marmorstein, & Sharma, 1996). This occurs because the presence of a discount increases involvement, which in turn increases elaboration. The level of elaboration in turn is argued to vary with the discount depth (Grewal et al., 1996). In this reasoning deeper discounts could provide such unexpected stimuli leading to increased attention towards them by shoppers. Therefore, deeper discounts may be more salient in the minds of consumers, they attract more attention by being incongruent with the expectations of the consumer, and therefore cross the threshold of attention for the consumer.
Deeper price discounts may be attractive for reasons beyond saving of money (Garretson and Burton, 2003). Not only do the reduced prices increase the customer’s variable shopping utility (Bell, Ho, & Tang, 1998), they also lead to greater psychological pleasure resulting from having to pay less than the normal price (Grewal, Krishnan, Baker, & Borin, 1998). When buying deep discounted products a consumer might obtain different hedonic benefits such as entertainment and self-expression. Deeper discounts provide customers with a saver feeling and reduce the pain of paying (Gázquez-Abad, Martines-López, & Barrales-Molina, 2014). Also, prior research implies that shoppers will do more unplanned buying when the prices in a store are lower, due to normative justification. (Rook and Fisher, 1995). The reduced feeling of pain with paying and the normative justification gives us the foundation to expect that deeper discounts might increase the number of products sold per basket.
products are expected to compensate for the decrease in price of products sold, hence we hypothesize the following:
H1: Deeper discounts communicated in the store flyer will have a positive effect on a.) store traffic, b.) conversion ratio and c.) number of products sold per basket, a negative effect on d.) average price of products sold in basket and ultimately a positive effect on e.) store sales.
2.3.2 Promotion Scope
The promotion scope is defined as “the number of product categories or items on discount” (Lam et al., 2001: 199). To drive store traffic and increase sales, retailers need to ensure that consumers’ perceptions of their assortment are positive (Oppewal and Timmermans, 1997). The store flyers help retailers communicate about the variety present in their store assortment (Arnold et al., 2001). Therefor Lam et al. (2001) found that the promotion scope was positively related to store traffic and in one of their two samples positive related to spending effects. This finding would suggest that adding more and more categories and products in the store flyer will have a positive effect on store performance. However, logical reasoning provides that this finding cannot hold indefinitely. Consumers have limited attention and cannot process all information in a flyer, especially when the flyer gets to large. So, there must be a cut of point after which adding more products or categories has a negative effect on store performance. In the current study we are not interested in finding that specific cut off point, we are interested in finding what practitioners should decide until that cut of point. Basically, fill the flyer with more products per category or more different categories. We therefore make a further distinction between promotion scope depth and width. We define promotion scope depth as the number of products promoted in the flyer per category and promotion scope width as the number of product categories promoted in the flyer. This distinction will help practitioners to make better informed decisions about the trade-off between adding more categories or adding more products per category in the store flyer.
their resources into going to a store. The perception of the assortment variety could therefore be a trigger. According to the heuristics framework presented by Tversky and Kahneman (1974), when people face uncertainty or complexity in a decision making, they rely on heuristics (i.e., rules of thumb) to simplify their decision-making process. To manage the store flyers’ complexity and cope with their limited cognitive ability, people may use the perceived attractiveness of the assortment in the store flyer as a salient cue of the variety of the retailer’s assortment. In that sense customers use the availability heuristic (i.e., thing that comes up first in your mind). They see a lot of variety in the store flyer and therefore assume that the store assortment is varied. Then they base their decision whether the store assortment is attractive enough to visit based on this information. This theoretical reasoning also works the other way around, such that decreasing the promotion scope communicates less variety in the assortment therefore decreasing the persuasiveness for people to visit the store. Literature on assortment perceptions show that assortment perception depends on the space that is allocated to a given product category (Kahn and Wansink, 2004), as well as category attributes (Broniarczyk, Hoyer & McAlister, 1998). The more the different attributes in a product category are distinct, adding more products per category in the store flyer, the more the consumers perceive the assortment as varied (Van Harpen and Pieters, 2002). In sum, adding more products per category and adding more product categories help to convey an image of variety. It then becomes more likely that a consumer finds the promoted set of items sufficiently attractive to visit the store (Gijsbrechts et al., 2003). In the absence of other information about the stores assortment, the consumer uses this information to infer an image of assortment variety for the retailer and base the decision whether to visit the store on this assumption. Hence, we expect that a larger promotion scope (depth and width) increases the store traffic.
A deeper promotion scope, in that, more products per category are promoted in the flyer could attract more task-oriented customers. For example, people who are looking only to buy a new lamp might see in the store flyer that the retailer has a large assortment of lamps and therefor decide to go to the retailer with the task of buying a lamp. Therefore, it might be expected that these more task-oriented customers buy less in terms of quantity (e.g. only the lamp). At the same time, we can expect that task-oriented customers might be less price sensitive. They went to the store for the variety in assortment of lamps and not necessarily to get the bargain of a discounted lamp in the flyer. We therefore expect that a deeper promotion scope might decrease the number of products sold per basket, while having a positive effect on average price of products sold. The reasoning might be the other way around for promotion scope width, such that, advertising more categories in the store flyer might attract less task-oriented customers and more hedonic-oriented customers. These more hedonic-oriented customers are expected to spend more time in the store going through all the department. This in turn increases the probability of finding products that meet their utility requirements and therefore increase the number of products they will buy. However, the hedonic-oriented customer, who enjoy the process of shopping more, might enjoy the process of ‘winning’ a bargain (Bawa and Shoemaker, 1987) more than task-oriented customers. Therefore, they might be more price sensitive leading to a decrease in average price of products sold. To conclude, we have argued that both a deeper promotional scope, that is, more promoted products per category and a wider promotional scope, that is, promoting more different product categories are expected to have a positive effect on store traffic and conversion ratio.We expect that the task-oriented customer, attracted through a deeper promotional scope, will create less sales in quantity, but sales of higher priced items compared to more hedonic-oriented customers who are expected to buy a higher quantity of lower priced items. Both types of customers are expected to have a positive effect on the store sales and therefore we hypothesize the following:
H2: A deeper promotion scope in the store flyer will have a positive effect on a.) store traffic and b.) conversion ratio, a negative effect on c.) number of products sold per basket, a positive effect on d.) average price of products sold and ultimately e.) store sales.
Continuing the reasoning about the task- versus the hedonic-oriented customer we expect that attracting the task-oriented customer generates overall more sales. Those customers are expected to be less price sensitive and thus buy higher priced items. Selling one lamp for 150 euros to a task-oriented customer generates mores sales than selling 10 times a +/- 5 euro’s item (150 > 10*5). We do not expect these numbers, but we expect that this reasoning holds. Therefore, we hypothesize the following:
H4: The positive effect of a deeper promotion scope on store sales is larger than the positive effect of a wider promotion scope on store sales.
2.3.3 Loyalty Membership
Loyalty programs are prevalent across a wide range of industries and have in recent years enjoyed an increase in membership participation (Berry, 2013). In the current study we use loyalty program membership as a measure for loyalty. We denote a loyalty program member as a ‘loyal customer’ and all other customers as ‘nonloyal customers’. Rational reasoning would suggest that nonloyal customers are less aware of the assortment of the retailer and are therefore expected to be more responsive to the store flyer content. The loyal customer needs less persuasion by the retailer to visit the store and are therefore expected to be less responsive to deeper price discounts and larger promotion scopes compared to nonloyal customers. Also, empirical research found nonloyal customers to be more responsive to promotions than loyal customers (Bell and Lal, 2003; Zeithaml, Berry and Parasuraman, 1996). Anderson and Simester (2004) for example studied the effect of discount size on new versus established customer and indicate that the percentage of new customers (nonloyal) that react to a promotional offer is greater than the percentage of established (more loyal) customers. In addition, Van Heerde and Bijmolt (2005) also found nonmembers to be more responsive to price discounts than members.
expected to be less persuaded by deeper discounts in the store flyer compared to nonloyal customers, however we expect that they can still be influenced by deeper discounts since these loyal customers might be on the lookout for a bargain in the store flyer. Thus, we expect that loyal customers are less responsive to the flyer content decisions than nonloyal customers and this negative moderation effect is stronger for promotion scope compared to discount size, such that a loyal customer is more influenced by deeper discounts than by a larger promotion scope. Therefore, we hypothesize the following:
H5: The effects of flyer content decisions on store performance is lower for loyalty program members than for nonmembers.
H6: The effects of discount size on store performance for loyalty program members is larger than the effect of promotion scope on store performance for loyalty program members.
2.4 Conceptual framework
Figure 2 visualizes the described relations in the current study. It serves as a guide to better understand the research issues at hand with their hypothesized relations. Control variables are also added which will be described in section 3.1.3.
3. RESEARCH DESIGN
3.1 Data descriptionDue to confidentiality issues the name of the retailer cannot be identified. However, some descriptions about the retailer will be given to understand the context of our current research. The retailer started as a demolition company 60 years ago selling the materials they collected during the demolition of buildings out of the back of their truck and garage. Nowadays they are one of the biggest construction and furnishing stores in the Netherlands with a physical store of 22.000 m2, which also has their own in-store bistro. The retailer sells a large variety of product categories: kitchens, bathroom materials, home decoration, home improvement, garden materials, do it yourself and professional construction materials. Customers typically visit infrequent compared to other for example grocery retailers. This is mainly due to the infrequent need for products sold at the retailer. Another reason is the higher costs associated with visiting the store, since the retailer is located in a less dense populated area. The retailer offers a medium- to high-end assortment and is situated in a highly religious area. Due to the owners’ religious principles the retailer is never opened on Sundays and Christian holidays and the retailer does not use radio and tv advertising for the same reason. Most used advertising besides the flyer is feature advertising in newspapers and online advertising with the aim to strengthen retailers brand. The retailer has a customer loyalty program since 1978 with currently around 22.000 members. The study period examined here ran initially from the 1st of January 2015 till the 1st of January 2018 (3 years).
3.1.1 Dependent variables
Table 1: Computation of implied dependent variables
Variable Computation
Conversion ratio Transactions / Store Traffic
Basket size Store Sales / Transactions
Products sold per basket Number of products sold / Transactions
Price Basket size / Products sold per basket
3.1.2 Independent variables
A door-to-door store flyer is sent 53 times during the duration of this study. The retailer renewed the store flyer strategy at the beginning of 2016. Previously the store flyer typically advertised multiple if not all categories, with only a few products per category and relatively deep price discounts. After 2016, the retailer decided to focus more on being a specialist in a certain category by reducing the number of categories in a flyer and displaying more products per category. The chosen timeframe gives us the data to assess the flyer effects with varying discount depth and promotion scope.
Table 2: Computation independent variables
Variable Computation
Discount size |Averaged over all advertised products|:
(Discounted price – Regular price) / Regular price Promotion scope width Count the number of categories advertised in the flyer
Promotion scope depth Count the total number of products advertised in the flyer / Promotion scope width
Figure 3: Flyer data connection to retailers’ performance metrics dataset
3.1.3 Control variables
In our study we control for flyer variables that are not part of the content decisions. Hence, we control for campaign volume, since its logical to assume that when there are more flyers sent out we see an effect on performance metrics, which cannot be attributed to the flyer content decisions. We use number of pages of the flyer to captures the effect of flyer size. Since the effect of a store flyer is expected to decline over time, we also control for number of weeks since the start of the flyer.
shopping street where potential customers have to go to by foot, which is critical difference with the research by Lam et al., (2001). In the current study it might be expected that when it is raining compared to sunny people might be more tempted to visit the retailer since it would be a nice activity to avoid the bad weather. When weather conditions are so extreme that they influence customers’ ability to reach the retailer by car we expect a negative effect on the performance metrics. So, there might be an effect which has to be controlled for. Next, the retailers under study sells many items which might be subject to seasonal fluctuations. We should therefore control for seasonality using dummies for the season. With the same reasoning we also control for the year. Lastly, the retailer has besides the flyer also other marketing activities such as feature advertising in newspapers, online advertising and events such as VAT-free days. This effect should also be controlled for. Table 3 provides an overview of the dependent and explanatory variables with descriptive statistics after corrections based on 3.1.4.
Table 3: Overview of variables
Variable Description Median Mean SD
Dependent variables
Store traffic Amount of people who visited the store 907.00 991.32 400.56
Transactions Amount of people who made a purchase 556.00 578.07 175.62
Conversion ratio Percentage of people made a purchase 0.61 0.61 0.08
Products sold per basket Number of products sold per basket 3.85 4.01 0.68
Average price Average prices of products sold 16.44 16.93 4.07
Average basket size Average basket size per transaction 63.84 67.10 18.12
Store sales Sales of the retailer 34999 40505 23505
Independent variables
Discount size Depth of the discounts in the flyer in percentages 19% 20% 9
Promotion scope width Number of categories advertised in the flyer 2 4.2 2.8
Promotion scope depth Number of products per category advertised in the flyer 6.29 11.34 10.48
Other flyer variables
Number of pages Number of pages in the flyer 4.00 4.43 4.07
Flyers active Number of flyers active at the same time 1.00 1.11 0.64
Weeks since active Weeks since the flyer is active 3.00 2.94 1.34
Control variables
Day of week Dummy for day of the week
Seasonal effect Dummy for season of the year
Year Dummy for year
Offline advertising Euro’s spend on offline advertising per month 4250 3909.47 2146.82 Online advertising Euro’s spend on online advertising per month 0 1370.79 1494.55 Other promotion Dummy for occasion when there is a special event
Activity day Dummy for activity day
Temperature Temperature in degrees 9 9.58 6.22
Overcast Measure of cloudiness (9= sky not visible at all) 6 5.74 2.25
Rain Measure of hours rain per day 0.30 1.98 3.04
3.1.4 Data cleaning
Figure 4: Sales plotted over time
On the 10th of October in 2015 the sales are €2,013,797 (the store sales in the figure 4 are mean aggregated per week for visual fluency purposes). This extreme outlier does not make any logical sense and there are no other explanations like for example the VAT-free days that could explain such high sales. We therefore decide to treat it as a measurement error and the delete the value. We find several other outliers in store sales using the IQR calculation method (Courtney, 2017), but we decide not to treat them, since they seem possible and might therefore contain valuable information.
There are also outliers in the average number of items per basket (figure 5). On the 11th of
December 2016 the average number of items per basket is negative resulting in a total of 34,148 items that are returned on this day. Here again there is no logical explanation and after consulting with the retailer, we decided to also treat these values as measurement errors. In total four values are deleted.
To preserve statistical power of our analysis the now missing values are imputed. Simplistic approaches such as mean-substitution usually introduces bias into the data, for instance, applying mean-substitution strengthens the mean while decreasing the variance (Donders, van der Heijden, Stijnen, & Moons, 2006). Therefore, a more sophisticated method will be used in the current study. The four missing values are imputed using the multivariate imputation by chained equations package in R, or ‘MICE’ (Buuren and Groothuis-Oudshoorn, 2011). Since the imputed variable is numeric the used method in MICE is predictive mean matching. The method regresses the other selected variables in the dataset on the missing values to predict what the values should be. This is done five times and one of the five values is selected and used to impute the missing value (Azur, Stuart, Frangakis, & Leaf, 2011).
Lastly, the days on which the retailer is closed are deleted from the dataset. Thus, all Sundays and Christian holidays. On these days the store traffic and other store performance metrics are equal to zero. This is not desired in our data since it 1.) might bias the results and 2.) because we might need to log transform the dependent variables and we cannot take the log of zero. This leaves us with 665 observations for our analysis.
3.1.5 Descriptive Statistics
The data consists of 665 observations over a period of approximately two years and two months. On 578 of those days a flyer was active, a total of 41 flyers have been sent. Those flyers had an average discount of 20% (Discount size) and promoted on average 4.2 categories (PSW) with 11.34 products per category (PSD). Since the store performance metrics are right-skewed (Appendix I), hence the mean is higher than the performance on most days we communicate the median of the performance metrics. The median of store visitors is 991. 61 percent of the visitors who enter the store also make a purchase, which leads to a median of 556 transactions. The median of the number sold products per basket is 3.85 and the median of the average price of the products in the basket is 16.44, which lead to a median of 63.84 in basket size or average expenditures per customer. The median of the daily revenue is €34,999.
VAT, which causes people to buy higher priced items. However, the promotion bump seems to be much higher for nonmember compared to loyalty member. This is in line with the findings by van Heerde and Bijmolt (2005), who found that nonmembers tend to be more responsive to discounts than members. Next, it is noticeable that loyalty members compared to nonmembers buy more products per visit, however they buy products of a lower average value. Combining the two metrics into expenditures per visit we see that nonmember on average spend more per visit compared to members. We should note however that this might also be caused by how we identify a loyalty member in the current study: ‘a customer that shows their loyalty card at the cashier’. The terms and conditions of the loyalty card state that typically higher priced products such as kitchens and bathrooms are excluded from the loyalty program point rewarding system, such that even if a loyalty member would buy an expensive priced kitchen we identify the customer as a nonmember. Another more theoretical explanation for our finding that loyalty members buy on average more products per store visit could be based on our earlier reasoning about task- versus hedonic-orientation. Identifying loyalty members as more hedonic-oriented customers, who visit more often, spend more time in the store going through all departments and thus buying more in terms of quantity, yet lower priced products. The reasoning is reversed for the nonmember who are identified as task-oriented customers who visit less often and go to retailer with a goal to buy a certain product. Hence, they buy less in terms of quality, yet on average higher priced products.
Figure 6: Continued
To illustrate the nature of variation in the metrics from our store performance framework (figure 1), in figure 7, we graphically display the time series for the seven store performance metrics. We add the promotion scope and discount size over time for a visual observation of a potential relation between our dependent variables and independent variables. All metrics are shown with their average level per week for a period of approximately 2 years and 3 months. We observe a declining trend in promotion scope width and an increasing trend in promotion scope depth, which is in line with the retailers’ strategy as described in section 3.1.2. However, it is difficult to visually detect any correlation between the store performance metrics and the flyer content decisions (independent variables). Figure 7, therefore, underscores the need for a statistical model to find the effect of the flyer content decisions on the multiple store performance metrics.
3.2 Plan of analysis
To formalize the performance metrics framework in the current study we specify store revenue at day t as the product of two components, number of transactions and the average basket size on day t. In turn, transactions on day t is expressed as a product of store traffic and conversion ratio and average basket size as a product of number of products sold per basket and the average price of products sold on day t:
(1) 𝑇𝑅𝑡= 𝑆𝑇𝑡∗ 𝐶𝑅𝑡 (2) 𝐵𝑆𝑡 = 𝑁𝑆𝑡∗ 𝐴𝑃𝑡 (3) 𝑆𝐴𝑡= 𝑇𝑅𝑡∗ 𝐵𝑆𝑡
Where,
STt = Store traffic on day t;
CRt = Conversion ratio on day t;
NSt = Number of products sold per basket on day t;
APt = Average price of products sold on day t;
TRt = Transactions on day t;
BSt = Average basket size on day t;
SAt = Store sales on day t;
Equation 1-3 enables us to decompose the effect of store flyer content decisions on store sales as the sum of four sources:
(4) ∆ SA = ∆ ST (a: change in number of store visitors)
To test the first four hypotheses, we specify regression models for the natural logarithm of the dependent variables. Except for conversion ratio, which is normally distributed, hence we do not log-transform conversion ratio (Appendix I). The store performance metrics are explained by variation in the flyer content decisions (discount size and promotion scope) and the control variables as specified in section 3.1.3.
The nature of regression model is such that it describes the behavior of a dependent variable based on a set of explanatory variables. However, in the current study we aim to explain a system of dependent variables. We use multiple simultaneous regressions, each explaining a different metric of the store performance. It seems reasonable to assume that if we overestimate the performance of one metric on day t we might also overestimate the performance of other metrics on that day. In other words, the regressions may be linked statistically through the distribution of the error term. Thus, to draw reliable statistical inferences about the model parameters we need to consider the separate equations collectively. Therefore, in the current study we use seemingly unrelated regression (SUR) models (Zellner, 1962). To estimate the four dependent variables in equation 4 we specify the SUR model as follows:
(5) 𝑦𝑡𝑖 = ∑ 𝑥𝑡𝑖𝑗 𝑘𝑖 𝑗=1 𝛽𝑖𝑗+ 𝜀𝑡𝑖 Where,
yti = log-transformed dependent variable on day t, for regression equation i (i = ST, CR, NS or
AP);
xtij = a vector of explanatory variables on day t, for regression equation i, explained by
explanatory variable j (j = 1, 2, …, ki);
βij = a vector of parameters associated with xtij;
εti = disturbance term on day t, for regression equation i.
with the correlation of the residuals in table 4 and the covariance matrix of the residuals that is used for estimation in table 5.
Figure 8: Fitting Seemingly Unrelated Regression model
Table 4: The correlations of the residuals
Traffic Conversion Ratio Count Products Price
Traffic 1 -0.654 0.093 0.079
Conversion Ratio -0.654 1 -0.013 -0.078
Count Products 0.093 -0.013 1 -0.473
Price 0.079 -0.078 -0.473 1
Table 5: The covariance matrix of the residuals used for estimation
Traffic Conversion Ratio Count Products Price
Traffic 0.0324 -0.006 0.002 0.003
Conversion Ratio -0.006 0.002 -0.000 -0.001
Count Products 0.002 -0.000 0.021 -0.013
Price 0.003 -0.001 -0.013 0.038
(6) 𝑦𝑡= ∑ 𝑥𝑡 𝑘 𝑗=1 𝛽 + 𝜀𝑡 Where,
yt = log-transformed store sales on day t;
xt = a vector of explanatory variables on day t;
β = a vector of parameters; εt = disturbance term on day t.
We compare the models by taking the anti-log transformed fitted values of the SUR model for the different performance metrics. We then use equation 1-3 to calculate the predicted value for sales and compare the sum of squares residuals (SSR) with the SSR of the more simplistic model. The SSR is calculated as follows:
(7) 𝑆𝑆𝑅 = ∑(𝑦𝑡− 𝑇 𝑡=1 ŷ𝑡)2 Where,
yt = observed store sales on day t;
ŷt = estimated store sales on day t.
(8)
𝑡 = 𝛽1 − 𝛽2 √𝑠𝛽21+ 𝑠𝛽22
3.3 Model assumptions 3.3.1 Multicollinearity
Multicollinearity occurs when there is a high correlation of at least one explanatory variable with a combination of the other explanatory variables (Leeflang, Wieringa, Bijmolt, & Pauwels, 2015). It is therefore difficult to disentangle the effect of an individual explanatory variable on the dependent variable. Hence, with multicollinearity we may get biased regression coefficients, which might even be in the other direction than expected. In the current study we use the variance inflation factor (VIF) as an indicator of multicollinearity. When we first estimated the model, we found some multicollinearity issues.
1.) To control for seasonality, we had dummies for the months of the year (VIF = 295), which highly correlated with the variable that controlled for long duration flyers. The variable equaled one when there is a long duration flyer and zero when there is no long duration flyer. Typically, the long duration flyer is active for one or two seasons, which causes high correlation with seasonality.
2.) Multicollinearity was found for the variable that controls for the time between the sent-out date of the flyer and the current date (VIF = 13.4), measured in weeks since sent sent-out date. This variable was found the be highly correlated with number of flyers active. Hence, when flyers active is zero the dummy for weeks since flyer active always equals ‘no flyer active’ and then the two control variables follow a similar pattern such that when number of flyers active is two it is often the first week that the flyer is active. This makes logical sense since figure 3 shows, the flyers often overlap for a period, in which it is the first week for a flyer and there are two flyers active.
A way to solve multicollinearity is to reformulate the model through recoding or the deletion of variables (Leeflang et al., 2015). Hence, to solve the first issue, we deleted the control variable for long duration flyers and recoded seasonality into dummies for the seasons. The effect of the seasonal flyer is now captured by the dummies for season. In our attempt to solve the second issue we recoded the number of flyers active into zero or one, in which zero indicates that there is no flyer active and one indicates that there is at least one flyer active. However, this gives errors with singularities, indicating that independent variables in our model are perfectly colinear. We tested this with alias command in R and found that the number of flyers active when recoded is perfectly collinear with weeks since active, hence we use the previous version of flyers active. To also solve the third issue, we found that when season, year and week since start flyer are deleted, we get a model without any multicollinearity (all VIF < 10). However, the adjusted R2 for this model is lower compared to when these variables are in the
model (0.7558 < 0.7609). Hence, these variables contain valuable information about the variation in the dependent variable. It is now a tradeoff between a model that also contains variables with high VIF scores or allowing for omitted variable bias into our model. The variables with high VIF’s are control variables and are not collinear with the variables of interest. Hence, the coefficients of the variables of interest are not affected by the high VIF’s, and the performance of the control variables as controls is not impaired. We thus to decide to keep these variables in the model, but not interpret their estimates. The VIF’s for explanatory variables can be found in table 6. In the current study we follow the view of Hair, Anderson, Tatham and Black (1995) that VIF values smaller than 10 are acceptable. Thus, we do not interpret the estimates of the control variables for weeks since active and year, the dummies for seasonality are to be interpreted with caution.
Table 6: VIF scores
Variables VIF Variables VIF
Discount Size 3.35 Year 26.64
Promotion Scope Width 6.68 Offline advertising 3.65 Promotion Scope Depth 2.95 Online advertising 6.89
Campaign volume 3.54 Other promotions 5.18
Number of pages 2.74 Activity days 1.26
Flyers active 2.77 Temperature 2.98
Weeks since active 13.40 Overcast 1.33
Day of week 1.22 Rain 1.35
3.3.2 Heteroskedasticity and nonnormality
Checked the linear model for sales with Breusch–Pagan test (Breuscha and Pagan, 1979). Found significant at < .05 level, not at <.01 level. Checked for non-normality with Shapiro-Wilk test (Shapiro and Wilk, 1965), which was also found significant (p < .05). The solution for nonnormality in the residuals would normally be to log-transform the dependent variables or delete outliers. However, both solutions are already performed. Section 3.1.4 describes what we did with the outliers and since the distribution of the dependent variables are right skewed we log-transformed them (Appendix II). Only for conversion ratio, which seems to be normally distributed we decide to not log-transform the dependent variable. Appendix II shows for store revenue how the log-transformation makes the distribution of the performance metric more resembling a normal distribution. Hence, we decide to not further treat the issues since the regression models are robust against violation of heteroskedasticity and normality (Schützenmeister, Jensen, and Pieho, 2012).
3.3.3 Autocorrelation
4. RESULTS
First, the results show that our proposed framework is better at predicting the store revenue compared to a model with only store sales as a dependent variable. Figure 9 graphically shows that the predicted values with both modeling methods are almost identical. Hence, the blue line and red line are almost always overlapping. We calculated the SSR as in equation 7 and found that the SSR of our decomposition is lower than the SSR of the more simplistic model (136202542565 < 136203033539). SSR is a statistical metric that measures the amount of variance in the data that is not explained by the model (Leeflang et al., 2015). Since the SSR is smaller with our decomposition framework, we show that our framework is better at predicting the store revenue then a more simplistic model. These findings are similar to the study by Lam et al. (2001). It should be noted however that our performance improvement is very small.
The model performs quite well at predicting store revenue peaks (figure 9). However, at the end of May in both 2015 and 2016 the model does not anticipate the peak. We conclude that in that period there must be a sales promotion that we did not control for.
Figure 9: Predicted store revenue decomposition framework versus model for only sales
4.1 Main results
The estimates of our models can be found in table 7. The intercepts are significant, indicating that if all explanatory variables in the model are set to zero the dependent variable will take the value of β0. For example, when all explanatory variables are set to zero the number of expected
This indicates that we are much better able to predict the number of store visitors with our explanatory variables compared to the average number of products customers buy per visit.
To test hypothesis 1-4, the estimates of the independent variables in table 7 are evaluated. Four significant main effects are found, which will be described in detail. The other theorized hypotheses are not supported in the current study (table 7). Hypothesis 4, promotion scope depth has a stronger positive effect on sales compared to promotion scope width, is automatically not supported since both promotion scope depth and width are not found to significantly affect store sales.
Table 7: Estimation results
log (Traffic) Conversion Ratio log (Count
Products) log (Price) log (Sales) Constant/Intercept β0 6.349*** 0.686*** 1.350*** 2.698*** 10.019*** Independent variables
Discount Size β1 0.005*** -0.001*** 0.001 <.001 -0.003
Promotion Scope Width β2 0.021*** -0.004* 0.003 -0.009 0.009
Promotion Scope Depth β3 0.001 <.001 0.001 -0.001 0.002
Control variables
Campaign volume β4 <.001 <.001 <.001 >-.001 <.001
Number of pages β5 -0.006* 0.001 -0.001 -0.001 -0.007
Flyers active β6 0.038* 0.001 <.001 0.018 0.057*
Weeks since active
First week β7.1 -0.168** 0.005 -0.079 0.106 -0.127
Second week β7.2 -0.174** 0.003 -0.067 0.087 -0.140
Third week β7.3 -0.224*** 0.016 -0.090 0.140* -0.140
Fourth week β7.4 -0.167* -0.007 -0.091 0.178* -0.082
Fifth week or later β7.5 -0.140* 0.002 -0.094 0.142 -0.081
Day of week
Tuesday β8.1 0.092*** -0.028*** -0.010 -0.024 0.014
Wednesday β8.2 0.192*** -0.059*** -0.062** 0.047 0.081*
Friday β8.4 0.533*** -0.100*** -0.001 0.160*** 0.530*** Saturday β8.5 0.846*** -0.148*** -0.041* 0.269*** 0.824*** Seasonal effect Spring β9.1 -0.009 0.048*** 0.057** -0.050 0.077* Summer β9.2 0.036 0.024** 0.054* -0.025 0.104* Winter β9.3 -0.050 0.035*** -0.002 0.022 0.029 Year 2016 β10.1 -0.020 0.005 -0.055 -0.017 -0.083 2017 β10.2 -0.065 -0.055** -0.052 -0.153* -0.367*** Offline advertising β11 0.009 -0.004* 0.019*** -0.006 0.015 Online advertising β12 0.022 -0.009** 0.005 0.014 0.028 Other promotions Vat-free β13.1 0.432*** -0.016 0.181*** 0.332*** 0.920*** Knipkorting β13.2 0.143*** -0.012 0.027 0.010 0.156*** Waardecheque β13.3 0.366*** -0.063** 0.143* 0.206* 0.592*** Nieuwjaarskorting β13.4 0.353*** -0.003 0.101** 0.235*** 0.686*** Kortingstapelen β13.5 0.001 0.025 -0.005 0.194 0.236 Activity day β14 0.079 >-.001 0.050 0.001 0.129* Temperature β15 -0.008*** 0.002*** -0.002 -0.002 -0.008** Overcast β16 0.010** -0.001 -0.001 -0.001 0.006 Rain β17 0.008** -0.002*** 0.001 0.002 0.007* R2 0.772 0.611 0.125 0.332 0.718 Adjusted R2 0.761 0.591 0.081 0.298 0.703 Residual Std. Error (df = 632) 0.180 0.048 0.145 0.196 0.245 Note: *p<0.1; **p<0.05; ***p<0.01 4.1.1 Discount Size
antilog-transform the coefficient for the interpretation. When all other variables in the model are held constant, with a one unit (1%) increase in discount size, we expect the store traffic to increase by 0.501 (100(e^ β – 1)) percent. Thus, when the average discount in the flyer would be 20% instead of the median discount size of 19%, we expect in that period to have 0.501 percent more visitors in the store. Note that this effect is location specific, such that on a day when the store traffic would have been 907 (median), the store traffic is now 912, which is a difference of 5 store visitors. On a day when the traffic would be 1600, the increased store traffic with a higher discount (1%) in the flyer will be 0.501 percent of 1600, which is 8 extra store visitors.
Table 7 further reveals that, contrary to expectations, the coefficient for the effect of discount size on conversion ratio is significant and negative (-0.001). We hypothesized the sign to be positive, thus hypothesis 1b is rejected. Deeper discounts communicated in the store flyer do not have a positive effect on the conversion ratio. The contrary is true, deeper discounts have a negative effect on conversion ratio. When all other variables are held constant, a one unit increase in discount size leads to a decrease in conversion ratio of 0.001. Hence, we expect a 0.001 lower conversion ratio in a period when the average discount in the flyer would be 20% instead of the median discount size of 19%. This effect is not location specific.
A deeper discount has a positive effect on the number of store visitors, but a negative effect on the percentage of visitors who buy. Following equation 1 we can calculate the effects of the discount size on the number of transactions (buyers), which is found to be positive. See figure 10 for a visualization of the effects of various discount sizes communicated in the store flyer in a scenario with deviation from the median for both the communicated discount size and the median of the dependent variable.
Figure 10: Continued
4.1.2 Promotion Scope Width
Promotion scope width is significant for store traffic with a positive coefficient of 0.021. Thus, hypothesis 3a is supported. A wider promotion scope in the store flyer has a positive effect on store traffic. Interpretation is similar to the effect of discount size on store traffic. When all other variables in the model are held constant, with a one unit increase in promotion scope width, we expect the store traffic to increase by 2.122 percent. In managerial terms: when the flyer would contain products of 3 different product categories, compared to 2, we can expect to have 2.122 percent more visitors in the store in that period. Similar to the effect of discount size on store traffic, this effect is location specific.
Figure 11: Visualization of effects of Promotion Scope Width
4.2 Interaction effects of loyalty membership
Separate models were estimated for only loyalty members and nonmembers (Appendix II: B & C). To test hypothesis 5, we use equation 8 to compare the slopes of the estimates and we find a significant difference in slopes between the two groups for 3 of the 15 estimates. The effect of discount size on number of transactions (T = 4.347, P < .01) and store sales (T = 2.949, P < .01) and the effect of promotion scope width on store sales (T = 1.964, P < .05) are all significantly smaller for loyalty members compared to nonmembers. Thus, we find partial support for our hypothesis that the effects of flyer content decisions on store performance is lower for loyalty member compared to nonmembers. Additionally, a deeper discount communicated in the store flyer is found to significantly increase the sales for only nonmembers. Keeping all the other variables in the model equal, with a one unit increase in discount size communicated in the store flyer, the sales for only nonmember is expected to increase by 0.070 percent.
between slopes of variables with different scales (Schroeder, Sjoquist, & Stephen, 1986). The coefficients in table 8 refer to how many standard deviations the performance metric will change, per standard deviation increase in the independent variable. The slopes of the estimates of discount size on the store performance metrics are compared with the slopes of promotion scope depth and width. We find no significant differences between the slopes of discount size and promotions scope depth. We do find that the effect of discount size on the number of transactions (T = 3.161, P < .01) and store sales (T = 2.278, P < .05) is significantly smaller compared to the effect of promotion scope width. This is in contrast with hypothesis 6 and we thus reject it.
Table 8: Standardized coefficients loyalty members
log (Count
Products) log (Price)
log (Transactions)
log (Basket
Size) log (Sales)
Discount Size 0.005 -0.017 -0.016 0.012 -0.028
(0.013) (0.015) (0.010) (0.012) (0.017)
Promotion Scope Width 0.015 -0.016 0.041** -0.001 0.039
(0.018) (0.020) (0.015) (0.017) (0.024)
Promotion Scope Depth 0.016 -0.027* -0.003 -0.011 -0.008
(0.012) (0.013) (0.010) (0.011) (0.016)
Note: *p<0.1; **p<0.05; ***p<0.01
4.3 Control variables
When we look at the control variables (table 7) we surprisingly find that having a flyer active in any week after the sent-out date compared to having no flyer active, would result in less store traffic. However due to the high VIF (13.40), as reported in table 6, we do not interpret these results. Similarly, the dummy for year 2017 would indicate that the sales have dropped compared 2015. We do not interpret this due to multicollinearity (VIF = 26.64).
5. DISCUSSION
A first research question regarding the effectiveness of the proposed decomposition framework was: “Which modeling approach for predicting store sales performs best?”. We found that, in predicting store sales, the decomposition framework marginally outperformed a modeling approach using only the store sales as a dependent variable. A perhaps more significant advantage of the decomposition framework is that it examines the effects of store flyers content decisions more thoroughly than a more simple approach. Allowing for a decomposition of flyer content decision effects into 1.) attraction, 2.) conversion and 3.) spending effects. Hence, in the current study an approach using only store sales as a dependent variable would not have resulted in significant results, concluding that the flyer content decisions do not matter as long as a flyer is sent out. We demonstrate, that changes in store flyer content decisions can have a non-negligible impact on store traffic and conversion ratio. All hypotheses and findings are summarized in table 9, which will be discussed in the following sections.
Table 9: Summary of Hypotheses and Results
Variable Hypothesis Effect Anticipated
Sign Results
Discount Size H1a Attraction + Supported
H1b Conversion + Not supported. Effect reversed.
H1c Quantity + Not supported
H1d Unit Spending – Not supported
H1e Sales + Not supported
Promotion Scope Depth H2a Attraction + Not supported
H2b Conversion + Not supported
H2c Quantity – Not supported
H2d Unit Spending + Not supported
H2e Sales + Not supported
Promotion Scope Width H3a Attraction + Supported
H3b Conversion + Not supported. Effect reversed.
H3c Quantity + Not supported
H3d Unit Spending – Not supported
H3e Sales + Not supported
Promotion Scope H4 Store Performance + Not supported
Loyalty Membership H5 Moderation – Partially supported.
H6 Moderation + Not supported. Effect
5.1 Conclusions and recommendations
The main research question this thesis aimed to answer was: ”What is the effect of door-to-door store flyer content decisions on multiple metrics of store performance?”. We found that communicating deeper discounts in the store flyer yields an increase in store visitors, while decreasing the conversion ratio. The same effect was found for the number of categories promoted in the flyer, such that communicating more categories in the store flyer leads to an increase in store visitors and a decrease in conversion ratio. Both were found to positively affect the number of transactions.
The positive effect on store traffic is in line with our expectations and earlier findings (Lam et al., 2001; Gijsbrechts et al., 2003). The finding that both deeper discount size and a wider promotion scope decreases the conversion ratio is somewhat surprising and contradicts previous findings. Lam et al. (2001) found positive effects of discount size on conversion ratio and Dhar et al. (1999) found a positive effect of promotion scope on the conversion ratio. In evaluating this result, we first consider the negative effect of discount size on conversion ratio. It should be recognized that the current study is done at a single-store retailer. A possible explanation might be that the retailer offers a medium- to high-end assortment. Attracted ‘bargain hunters’ through deeper discounts communicated in the store flyer might not find what they are looking for at the retailer under study and thus not buy. We add to the findings of Lam et al. (2001) that the effect of discount size on conversion ratio might be moderated by the price strategy of the retailer under study. Another explanation might be grounded in reactance theory (Brehm, 1966). Consumers resent transparent marketing tricks aimed to influence their behavior and try to ‘punish’ the offending marketer (Simonson, Carmon, & O’Curry, 1994). When discounts are deeper customers might feel pushed by the retailer to buy products that they do not need. The consumer feels manipulated and might try to ‘punish’ the offending retailer by not buying. Lastly, deep discounts might lead to inferences about the product's value and quality (Simonson et al., 1994). The deep discount might signal that a product is of low quality, thus decreasing a consumer’s willingness to buy.
discount. This is a critical difference with the current study in which it was possible that the promoted products did not have a discount. A possible explanation for the negative effect on conversion ratio might again be found in the type of customers that are attracted to the store. As argued in section 2.3.2 advertising more categories in the store flyer might attract less task-oriented customers and more hedonic-task-oriented customers. Hedonic-task-oriented customers might not visit the store because of the identification of a functional need (e.g. the need for a lamp), but more for the enjoyment of the shopping experience. Hedonic-oriented customers might exhibit more browsing behavior compared to task-oriented customer, who visit the store with a certain purpose (e.g. buying a lamp). Further research should explore the moderating role of customer orientation on the effect of promotion scope width on conversion ratio.
Another research question was: “What is the relative effect of communicating more products per category versus adding more categories in the flyer on store sales?”. Somewhat surprisingly, adding more products per category has no significant effect on any performance metric in our study. Also, adding more categories was not found to significantly affect the store sales. Our research is therefore inconclusive regarding this research question.
Next, the research explored the moderation role of customer loyalty by answering the following research question: “How does customer loyalty affect the responsiveness to store flyer content decisions?”. We found partial support for our expectation that nonloyal customers are more responsive to store flyer content decisions. A deeper discount size is found to increase the sales for only nonloyal customers. Lastly, we aimed to answer the question: “What flyer content decisions have the most effect on loyal customers?”. In contrast to our expectation we found partial evidence that promotion scope width, compared to discount size, has a stronger effect on loyalty members.
previous argument the advertising might be aimed to build the retailers brand rather than pure transactional. Seeing an ad of the retailer might not be a direct trigger to visit, since the customer might not recognize a need of the products offered by the retailer. Once the customer has a need for a certain product offered by the retailer, such as a kitchen, the advertisement might help the retailer to become salient in the mind of the consumer and in such a way positively influence the store performance. A different modeling approach is needed to correctly assess this effect.
5.2 Academic implications
In the current highly competitive retail environments, the store flyer has become of increasing importance to attract customers and subsequently influence their spending behavior (Gijsbrechts et al., 2003). Managers and academics therefore share an interest in understanding how the store flyer influences store performance (Ailawadi et al., 2009). This research answers that call and extents the current literature stream by utilizing a decomposition framework as offered by Lam et al. (2001) and van Heerde and Bijmolt (2005) to model the effects of store flyer content decisions on multiple metrics of store performance. We showed that the framework combined with the modeling approach offers a more precise and thorough estimation of promotional effectiveness. Academics, also in other promotional contexts, can use the framework and modeling approach to more thoroughly assess the effects of promotions on multiple store performance metrics simultaneously. This finds empirical evidence for an attraction effect of deeper discounts communicated in the store flyer and an attraction effect for communicating more variety in the store flyer by promoting multiple categories in the same flyer. Both are found to negatively affect the conversion ratio; however, the attraction effect is stronger. Hence, communicating deeper discounts or more products categories in the store flyer is found to be effective in attracting more customers, but not effective in positively influencing behavior at the retailer. Lastly, our study gives insights in the moderating role of loyalty membership. Partial evidence is found that loyalty members are less influenced by flyer content decisions compared to nonmembers and partial evidence that loyalty members are more influenced by number of categories promoted in the flyer than discount size.
5.3 Managerial implications
