The Effect of Vertical Shelf Positions on Sales
of Private Labels and National Brands:
An Experimental Study at Albert Heijn
Rebekka Morgenstern
The Effect of Vertical Shelf Position on Sales
of Private Labels and National Brands:
An Experimental Study at Albert Heijn
Master Thesis
University of Groningen Faculty Economics and Business Department: Marketing Management
29. September 2009 Rebekka Morgenstern Strausberger Platz 9 10243 Berlin Germany +31 623920248 Supervisors:
rebekkamorgenstern@gmail.com Dr. Laurens Sloot
Management Summary
Shelf space is one of the scarcest resources in the retail environment. Thus, researchers and practitioners seek ways to allocate it more optimally. While there are several studies concerning the effect of shelf space on sales, few focus on shelf layout.
The objective of this research is to investigate in how far the vertical shelf position of products has an effect on their sales. More specifically, the goal is to find out how changes between eye- and hip-level affect store and national brands and how changes between the top- and floor-level of the shelf affect sales of discount private labels and specialty items.
To this end, a field experiment is performed at the Dutch retailer Albert Heijn and 18 shelves are switched within nine different categories. Three different types of changes are implemented: in three categories, private labels are moved down from eye- to hip-level and national brands are moved up. In three more product groups, the opposite changes are performed. Finally, in three categories, discount private labels and specialty items are switched between the bottom and the top of the shelf.
The results show that an eye-level position has a positive effect on the sales of private labels, while is has a negative effect on national brands. A hip-level position, on the other hand, increases sales of both store and national brands. Moving discount private labels from the floor-level to the top of the shelf causes an increase in their sales. The opposite change – moving specialty items from the top-level to the floor – has no effect on their turnover.
Preface
First of all, I would like to thank Dr. Laurens Sloot for his help in developing and executing this research project. His vast knowledge and insights into the retail environment have helped me tremendously throughout the different phases of this research. Furthermore, I would like to thank Dr. Erjen van Nierop, who offered his valuable time and knowledge, especially concerning the methodology of this research. His detailed feedback was also very helpful for improving this report.
My gratitude also goes to Frits van der Heide and Paul Tammeling from Albert Heijn, who made the research possible. Their cooperation, input and support gave me the opportunity to conduct this research as a field experiment. In addition, I would like to thank the friendly staff at the respective Albert Heijn store for their assistance.
Finally, I am very grateful to my parents and my boyfriend Benjamin for their support and encouragement during these last few months.
Table of Contents
1 Introduction ... 1
2 Literature Review... 4
2.1 Shelf Management... 4
2.1.1 Shelf Space Allocation ... 5
2.1.2 Shelf Layout ... 6
2.1.3 Private Labels versus National Brands... 11
2.1.4 Hedonic versus Utilitarian Products... 13
2.2 In-Store Experiments ... 15
3 Research Objectives and Hypotheses ... 17
3.1 Hypotheses Regarding Private Labels and National Brands ... 17
3.2 Hypotheses Regarding Discount Private Labels and Specialty Items... 18
3.3 Hypothesis Regarding Hedonic and Utilitarian Product Groups... 19
4 Research Methodology ... 20
4.1 Store Characteristics... 20
4.2 Experimental Design... 21
4.2.1 Study One: Private Labels and National Brands... 22
4.2.2 Study Two: Discount Private Labels and Specialty Items ... 23
4.2.3 Study Three: Hedonic and Utilitarian Product Groups ... 24
4.2.4 Timing of Experiment ... 25 4.2.5 Confounding Variables ... 26 4.3 Data Collection ... 26 4.4 Plan of Analysis ... 27 5 Results ... 31 5.1 Descriptives ... 31
5.2 Findings Per Product Group... 33
5.3 Findings Per Type of Change ... 37
5.4 Hypotheses Testing ... 39
5.4.1 Private Labels and National Brands ... 39
5.4.2 Discount Private Labels and Specialty Items... 40
6 Discussion ... 43
7 Managerial Implications ... 45
8 Limitations and Further Research... 47
9 Conclusions... 49
INTRODUCTION
1 1 Introduction
A large portion of marketing effort in food retailing takes place outside the retail store, e.g. marketing research and advertising. Hubbard (1969-1970) calls these the “external promotional techniques”. If customers were a hundred percent brand loyal, these external measures would be sufficient. However, research shows that many consumers are willing to compromise their initial choice and switch to another product. This could either be because their brand is not available or that their decision is influenced by the shelf display. Therefore, it is crucial for retailers to focus on “internal promotional techniques”, measures that they can take inside the store. These tools may even be more important than the external ones, because there is evidence that many consumers make their choice at the point of purchase – in front of the retail shelf (Hubbard 1969-1970, Borin and Farris 1995).
While manufacturers primarily seek to increase the sales and profits of their respective brands and products, retailers aim at maximizing sales and profits of the category as a whole (Drèze et al. 1994). For that endeavor, they need to allocate a fixed amount of shelf space, one of the scarcest goods in a retail environment, in the most optimal way. Important issues to address are the amount of space that is allocated to each stock keeping unit (SKU) as well as the question where items are placed on the shelf. To make these complex decisions and tremendous amounts of data more manageable, a number of computer-based shelf management systems (e.g. Apollo from IRI and Spaceman from Nielsen) have been developed to help retailers allocate shelf space more profitably.
INTRODUCTION
2
While there have already been several studies on the allocation of shelf space, few of them deal with the shelf space management of store brands (Gómez and Okazaki 2009, Suárez 2005). Store brands often provide a higher percent profit margin than national brands. Yet, their rate of turnover and their absolute margin might be lower compared to national brands (Quelch and Harding 1996). Van Nierop et al. (2008) state that retailers often favor high margin items in the allocation of shelf space, rather than products with a high turnover. According to their findings, this practice is not always optimal. When looking at how retailers also favor store brands qualitatively and place them in the most favorable positions on the shelf, one may also wonder if this is in fact best practice. However, the effect of shelf positions on the performance of private labels compared to national brands has not yet been investigated.
The objective of the present study is to contribute to closing this research gap and to offer information regarding shelf layout, with a special emphasis on the distinction between national and store brands. Since the realm of this present study is restricted, the focus will be limited to vertical shelf positions. Even though literature supports an eye-level position on the shelf as the most favorable one, it has not yet been shown if this applies to national brands and private labels alike. Thus, following research question will be investigated:
How do vertical shelf positions affect sales of private labels and national brands?
To make this rather generic question more specific, following sub questions are stated: Will an eye-level position improve sales, when items are moved up?
Does the opposite change (moving items from eye- to hip-level) have a negative effect on sales?
Is there a difference in those effects for private labels and national brands?
Moreover, there is hardly any literature to be found on the effect that a vertical shelf positions have on discount private labels. The question arises whether their constant floor position is due to practical reasons, such as bulk sized packages or if it actually sells better on the bottom of the shelf. Hence, the research should contribute to answering following questions:
INTRODUCTION
3
Does the opposite change (moving specialty items from the top of the shelf to the bottom) have the opposite effect?
When comparing both outcomes, which one is stronger?
Lastly, it shall be investigated, in how far the nature of each product group plays a role in the effect that the vertical shelf position has on sales. Following question will be investigated:
Is the positive effect from improving the vertical shelf position of an item larger in hedonic or utilitarian product groups?
Besides adding to academic literature, findings regarding shelf layout are especially relevant for practitioners. Even though shelf space is such a scarce good in the retail environment, Curhan (1973) states that many managers rely on subjective experience for their shelf management. While this may have changed to some extent over the past decades, this research can offer further empirical evidence concerning vertical shelf layout, which managers can apply. The attained knowledge should prove to be valuable for retailers as well as manufacturers.
LITERATURE REVIEW
4 2 Literature Review
In this section, relevant literature concerning shelf management and store experiments is provided. Shelf management lays the foundation and creates the context of this research. In this first section, the focus is on the results of previous studies. Since this present empirical research endeavor is conducted as a field experiment, it is also important to gather literature on in-store experiments that have been performed in the past. This second section is not concerned with the results of these experiments but with the field experiment as a research method, its prerequisites and the challenges that can arise when performing an in-store experiment.
2.1 Shelf Management
Already in 1962, Cairns addressed the fact that retailers should think about shelf space in terms of opportunity cost. He regards retailers as “sellers of space”. If a retailer sells shelf space to a supplier, he should consider the opportunity costs of that space. Opportunity costs, in this case, are the forgone profits that the retailer could have earned by giving the retail space to the most profitable item not yet in the assortment or to the most profitable combination of items that are already stocked (Cairns 1962).
For quite some time now, the allocation of shelf space and the layout of shelves have been of interest to practitioners. They have also been investigated by many researchers and academics. However, the problem of shelf space as opposed to shelf layout has gotten considerably more attention in literature. Most articles that do consider layout effects include them into an optimization model as a variable beside space. In the following literature review, the two topics are considered separately, as far as possible. This is done to be able to focus on layout effects and to summarize what knowledge presently exists about them.
LITERATURE REVIEW
5 2.1.1 Shelf Space Allocation
Shelf space allocation is not the main research interest of this paper. However, shelf layout and shelf space are usually interlinked – in practice and in literature. Therefore, a brief overview of the research regarding shelf space allocation will be provided. This is also important, as it creates a historical context for the research that also takes location into account.
In the 60s and 70s of the last century, various experiments were conducted to measure the relationship between shelf space and sales. This relationship is often called “space elasticity” and is defined as the ratio of relative change in unit sales to relative change in shelf space (Curhan 1972). Cox (1964) was one of the first authors to test the effect that shelf space had on food product sales. In his research, with a rather limited set of product groups, he differentiated products into staples and impulse goods. His findings did not support the hypothesis that impulse items responded more to variations in shelf space than staples. In a subsequent research (Cox 1970) the author also took the effect of the brand into account. He found no relationship between the amount of shelf space given to a staple product brand and unit sales; whereas the relationship between the amount of shelf space given to an impulse product brand and unit sales was significant. Kotzan and Evanson (1969) conducted an experiment at a drug store and found a significant relationship between shelf facings and sales. Curhan (1972, 1973) also concludes that there is a small positive relationship between shelf space and sales. He remarks, however, the inconsistency of this relationship between products, across stores and different locations within a store. Furthermore, he found that changes in shelf space have significantly more impact on private labels than on national brands. Unlike the research of Cox (1964), his findings also show a larger impact on impulse goods as opposed to staples. The results of a more recent empirical study by Desmet and Renaudin (1998) show a stronger effect of space on sales for impulse goods as well.
LITERATURE REVIEW
6
In a comprehensive study, Drèze et al. (1994) conducted a series of different field experiments. They examined the effects of changes in shelf space allocation and location on two levels: for the category as a whole (which is important for retailers) as well as each brand (which is the central interest of manufacturers). In regards to shelf space, the authors found that a minimum threshold of shelf space needed to be maintained to avoid out of stocks, but additional facings did not lead to additional returns. In fact, most of the products were over-allocated. They concluded that the number of facings was one of the least important success factors. The position of a product on the shelf, on the other hand, had a far larger impact on sales.
Several studies consider these findings and include location variables into their shelf optimization models (see, for example, Yang and Chen 1999, Yang 2001, Lim et al. 2004). Van Nierop et al. (2008) state, however, that the previously listed authors assume that the effect of shelf layout on sales is known. In real life this is not the case though and the effect needs to be estimated for each situation. Van Nierop and his colleagues thus investigate in their research the relationship between shelf layout and sales, as well as between shelf layout and marketing instrument effectiveness. Their findings regarding shelf layout, among others, will be considered hereafter.
2.1.2 Shelf Layout
Location can play a role on the store and category level. Campo et al. (2000) show how location characteristics on the store level have an impact on different categories. Yet, also on the product level, location plays a large role. There have only been a few studies that examine the effect that a product's shelf position has on sales. This creates a noteworthy research gap because findings by Drèze et al. (1994) acknowledge the significance of location, which is even larger than that of shelf space. Shelf positions are defined by a horizontal and vertical axis. They are thus discussed in this order as well.
Horizontal Shelf Positions
LITERATURE REVIEW
7
It is also important to consider a product's horizontal position in relations to the beginning and end of the whole aisle and consumers' shopping paths. Larson et al. (2005) investigated shoppers' in-store travel behaviors. They found that many customers only travel selected aisles. It appears that many shoppers do not follow the systematic pattern up and down every aisle, nor do they always travel the entire length of the aisle. Larson and his colleagues therefore concluded that products that are placed in the middle of the aisle will receive much less “face time” than the ones that are placed close to the end of the aisle.
Even though Van Nierop et al. (2008) understand the horizontal position of an SKU within the realm of the category; they still take the findings of Larson et al. (2005) regarding shopping paths into account. One of the zones that Larson et al. (2005) identify is the outer ring of the store – the “racetrack” – that circles around the outer ends of the aisles and has a higher amount and speed of travel than other zones within the supermarket. Van Nierop et al. (2008) found that the distance to the shelf end, and thus the racetrack, has a negative effect on sales. Therefore, a horizontal location close to the racetrack is optimal. The horizontal distance to the middle of the shelf, on the other hand, appears to have a positive effect on sales. This goes back to the findings of Larson et al. (2008) that customers often only take short excursions in and out of the aisle. Van Nierop et al. (2008) also found that promotional sensitivity is also higher for the outer shelf positions. Yet, price sensitivity is highest for these items as well. In contrast, positions in the middle of the shelf are approached with a lower rate of price sensitivity.
LITERATURE REVIEW
8
Drèze et al. (1994) also considered the effect of horizontal positions within a category but came to no consensus on whether an outer or inner position was more favorable. For half the categories, the edges seemed favorable, for the other half the center positions. The authors found, however, that moving a brand from the worst to best horizontal position can increase sales by an average 15 percent.
Another way to think about horizontal shelf positions is left versus right, as opposed to inner versus outer positions. To this end, one needs to take walking directions into account. If a person walked along a shelf from left to right, the question is whether he or she would be more prone to pick a product from the first part of the shelf (primacy effect) or from the second half of the shelf (recency effect). To test this, the EFMI Business School performed a study in 1999. Their findings were in favor of the primacy over the recency effect. The results showed that consumers chose 15 percent more products from the first part of the shelf than from the second. This, however, was only the case for consumers who were not brand loyal (Broere et al. 1999).
LITERATURE REVIEW
9
Paper Definition of Shelf Division of Shelf Optimal location
Drèze et al. (1994) merchandising box inner vs. outer positions no consensus Christenfeld (1995) merchandising box center vs. outer positions Center Broere et al. (1999) merchandising box first half vs. second half first half Shaw et al. (2000) merchandising box center vs. outer positions Center Larson et al. (2005) length of aisle middle vs. end positions End
Van Nierop et al. (2008) merchandising box middle vs. end positions end, close to racetrack
Table 1: Summary of Findings Concerning Horizontal Shelf Positions
As far as the optimal location goes, there are various different results. However, since Christenfeld (1995) and Shaw et al. (2000) perform their research amongst identical choices, their findings cannot easily be generalized. Larson et al. (2005) and Van Nierop et al (2008) apply a more comprehensive approach and consider the shelf within the supermarket as a whole. Their findings seem thus more useful. The results presented by Broere et al. (1999) are not necessarily contradicting those of Larson et al. and Van Nierop et al., but can be seen as a valuable addition. In conclusion it can thus be stated that positions toward the end of the shelf seem to be more favorable. Which end of the shelf is more preferable, might depend on the direction the consumer comes from.
Vertical Shelf Positions
Another dimension to consider is the vertical shelf position of an SKU. Vertical positions, in comparison to horizontal ones, tend to have a more pronounced effect on sales (Breugelmans et al. 2006).
LITERATURE REVIEW
10
eye-level. The average height of Dutch males and females, for example, exceed that of US American's by about five centimeters (McDowell et al. 2008, Centraal Bureau voor de Statistiek 2009).Drèze et al. (1994) assume that most consumers stand about 1,20 meters away from the shelf, so that central shelf positions become most desirable. Yet, how far away from the shelf consumers stand might also differ by country and by the layout of the store. Wal-Mart supercenters in the United States, for instance, can be expected to provide more aisle space than a small grocery store within the center of Amsterdam. If people stand further away from the shelf, the range of the products they initially see becomes larger. If they stand closer to the shelf, they would predominantly notice the shelves that are exactly at eye-level. The range of shelves that can be considered what is eye-level thus differs by geographic location and the store layout. Drèze et al. (1994) found that though most manufacturers and retailers agree on the level as being the best position, when asked to specify what they meant by eye-level, most experts referred to a range of shelves above the knees but below two meters.
Hubbard (1969-1970) states that manufacturers and marketers both agree that an eye-level shelf position provides the highest sales in food retailing. Frank and Massy (1970) found only a modest effect on sales when varying a product's vertical position. Curhan (1973) notes, however, that changes on shelves closer to eye-level would probably have a larger effect on sales than changes of shelves above or below eye-level. One of the findings of Drèze et al. (1994) was that a couple of facings at eye-level generated more turnover than five facings on the bottom shelf. More specifically, the authors state that a central vertical position, slightly below eye-level (on average at around 132 cm above the floor), is most desirable. Corstjens and Corstjens (1995) state also that, shelf space at eye-level or hand-level is worth more than shelf space at the bottom of the shelf. The authors furthermore present the rule of thumb that sales of a brand will be reduced by two thirds if it is moved from eye- to floor-level. Van Nierop et al. (2008) found that a higher shelf location is positively related to sales. They additionally found higher price sensitivity and promotion effectiveness for products on higher shelves. The positive effect on sales flattens toward the top shelf.
LITERATURE REVIEW
11
or top-level shelves are more preferable is arguable. Drèze et al. (1994) found different effects in different product categories.
As far as the vertical organization of the shelf goes, generally bulk items (often discount private labels) are found at the bottom of the shelf. Specialty items are usually found on the top shelf. Retailers often devote more shelf space to products with high margins, rather than the best selling products. Therefore, high margin items are frequently displayed at eye-level positions (Curhan 1972). Devoting more shelf space and valuable shelf positions to products with high margins as opposed to products with high turnover is not always the optimal solution (Van Nierop et al. 2008).
2.1.3 Private Labels versus National Brands
Private labels1 (PLs), also called store brands, are “products developed and marketed by a retailer and only available for sale by that retailer” (Levy and Weitz 2009, p.645). They are the only brands for which retailers must take on all responsibility (Dhar and Hoch 1997). There are many reasons for retailers to offer PLs within their assortment. Ailawadi et al. (2008) summarize the main ones as follows: higher margins on PLs, negotiation leverage with national brands (NBs) manufacturers and a higher customer store loyalty.
ACNielsen research showed that PLs currently account for 17 percent of total supermarket sales around the world. In Europe this share is even larger (23 percent) with a growth rate of 4 percent, which makes Europe the most developed private label (PL) region. The Netherlands hold a PL share of 22 percent with an 8 percent growth rate, which is the 7th highest share globally. In terms of PL customers, it can be stated that almost all households purchase store brands. Not surprisingly, the PL share is especially high for large families and households with lower income (ACNielsen 2005). Ailawadi and Keller (2004) found that PL customers are price sensitive but not image sensitive, have middle-income and are educated. This alludes to the fact that PL strength varies with economic conditions (Quelch and Harding 1996).
The most important driver of PL share is its perceived quality (Ailawadi and Keller 2004). This is why it is profitable to position a PL close to the leading national brand (NB), as far as variables such as labeling, package design and shelf placement are concerned (Sayman et al. 2002). Choi and Coughlan (2006) also investigated how PLs
1
LITERATURE REVIEW
12
should be positioned against NBs in terms of product quality and features. They came to the conclusion that a PL's best positioning depends on the nature of the NB it is competing with. If the NB is differentiated, a high quality PL should be positioned closer to a stronger NB and a low quality PL closer to a weaker NB. When NBs are undifferentiated, however, the PL should be differentiated from both NBs. As a consequence, times have past when PLs only represented lower quality product alternatives at lower prices and where they were mainly present in staple categories (Burt 2000). PLs have become much more diverse. The authors Ailawadi and Keller identify four tiers of PLs, “ranging from low quality, no-name generics to cheap, medium quality own labels to somewhat less expensive, comparable quality private labels, to premium quality, high value added private labels that are not priced lower than national brands” (Ailawadi and Keller 2004, p. 338).The quality gap between store brands and NBs has narrowed and PLs have expanded into new and diverse categories. Retailers that offer premium PLs often deliver quality superior to NBs (Quelch and Harding 1996). By use of premium PLs, retailers can respond to NBs' ability to cater to heterogeneous preferences. This appears more likely in categories where PLs already offer high quality comparable to NBs (Dhar and Hoch 1997).
LITERATURE REVIEW
13
As far as a PL's position on the shelf is concerned, it depends on the kind of PL and the tier it is in. Research showed, for example, that low price and quality brands were more likely to be chosen when placed next to competing brands (Nowlis and Simonson 1997). This applies to PLs that are cheaper than the corresponding NBs. These PLs should therefore be placed just next to the respective NBs. High price and quality brands, on the other hand, were more successful when displayed separately (Nowlis and Simonson 1997).
2.1.4 Hedonic versus Utilitarian Products
Another distinction that can be made is between hedonic and utilitarian products. Though some researchers divide products categories into staple and impulse goods, the characterization of hedonic and utilitarian items appears more useful for food retailing. One way of making this differentiation is to consider the attitude with which consumers approach the product and brand. Batra and Ahtola (1990) state that consumer attitudes are two-dimensional; customers either consume for affective, hedonic gratification or for instrumental, utilitarian reasons. Voss et al. (2003) give an even more applicable distinction, when they state that the hedonic consumption dimension results from the sensation derived from the experience of using the product., The utilitarian dimension, one the other hand, is derived from functions performed by product. They put forth a ten-item scale to measure product groups against, in order to find out whether they are more utilitarian or hedonic in nature. Utilitarian items include: effective/ineffective, helpful/unhelpful, functional/not functional, necessary/unnecessary and practical/impractical. Items to identify a product's hedonic nature are: not fun/fun, dull/exciting, not delightful/delightful, not thrilling/thrilling and enjoyable/unenjoyable (Voss et al. 2003). Hedonic and utilitarian motivations need not be mutually exclusive. Batra and Ahtola (1990) give the example of toothpaste that provides a pleasurable taste as well as preventing cavities. Yet, in some cases they are exclusive. For instance, some products or services provide pleasure, yet have no instrumental value, e.g. smoking. Others might be instrumental but give no pleasure, e.g. going to the dentist (Batra and Ahtola 1990).
LITERATURE REVIEW
14
shopping trip itself is the goal, not the end result – the purchase. The shopping experience is concerned with hedonic fulfillment, such as fun, amusement, fantasy and sensory stimulation; even if it does not lead to a final purchase (Babin et al. 1994). Arnold and Reynolds (2003) identify six types of hedonic shopping motives. These consist of adventure, gratification, role, value, social and idea shopping motivations. Sloot et al. (2005) summarize that there is more opportunity to differentiate the brand of a hedonic product in the mind of the customer than a brand in a utilitarian product group. In hedonic product groups, brands can be positioned through means of emotional and symbolic aspects. Brands of utilitarian products, however, mainly are differentiated by quality. Furthermore, the authors state that high-equity brands in hedonic categories usually provide more items on the shelf than high equity brands in utilitarian categories. In addition, they state that in utilitarian categories (e.g. milk), the leading brand only provides a few items, whereas there is a large selection of different sizes and flavors of leading brands in hedonic categories (e.g. soft drinks). This may also be due to the fact that customers seek more variety in hedonic categories, as repeated consumption of these products may lead to boredom (Van Trijp et al. 1996). Research also showed that consumers of high-equity items were more likely to switch to another item in a hedonic product group than in a utilitarian one (Sloot et al. 2005).
Moreover, Ailawadi and Keller (2004) state that PLs perform better in utilitarian than in hedonic categories. The reason the authors give is that consumers choose compromise alternatives in situations where differences between choices (marginal values) are decreasing. Customers do not settle for an alternative, however, when products are strongly differentiated and marginal values are increasing. As stated before, hedonic products give brand manufacturers more opportunity to differentiate themselves from competition and thus to create marginal values. In utilitarian product groups, many consumers might not even be able to detect quality differences, which makes marginal values rather small. This whole premise is based on the US market however. Since the PL market there is not quite as developed as it is in Europe, it can be assumed that the first statement of this paragraph pertains mainly to discount and regular PL. Specialty and premium PL are highly differentiated and created marginal value, similar to or even higher than NBs.
LITERATURE REVIEW
15
assumed that an eye-level position is more crucial in hedonic than in utilitarian ones categories.
2.2 In-Store Experiments
Many studies that test the relationship between shelf space and sales contain the issue of causality. It is often not clear if more shelf space leads to more sales; or if a product had higher sales and therefore got an increased share of shelf. Frank and Massy (1970) were the first to point at this problem. The same issue of causality also applies to shelf layout, which leads to the question: are products with a high turnover placed in more favorable positions on the shelf or high quality position increase sales?
When studying the effects of shelf space and/or position on sales, researchers have three data sources: time series, cross sectional data and experimental data. There are some advantages of the first two sources. First of all, the data already exists, which leads to lower costs, no interference with the operations of the store and faster results. Secondly, they usually provide a much larger and thus more representative sample size. However, the issue of causality is not solved by time series and cross sectional studies, which limits their usefulness. Experimental studies, on the other hand, are capable of identifying causal relationships. They can thus be considered the most conclusive of scientific testing procedure (Doyle and Gidengil 1977).
An experiment is conducted when one or more independent variables are manipulated or controlled and their effect on one or more dependent variables is measured (Malhotra 2007). Experiments can be conducted within retail stores to test the consumer response to various marketing instruments. In the case of in-store experiments, the dependent variables are studied in an actual environment, which makes it a field experiment.
LITERATURE REVIEW
16
An example of the useful academic and managerial evidence that field experiments can provide, are the findings by Drèze et al. (1994).
Gaur and Fisher (2005) state that there are also some challenges when using in-store experiments. In order for results to be generalizable, the store where the experiment is performed needs to be representative of the entire retail chain. Furthermore, the experiment needs to be conducted in a controlled environment, where non-experimental factors are kept at a minimum. This is usually quite a difficult task.
There are also drawbacks to using experimental designs. One is linked to the high costs of implementing a controlled experiment in the field (Drèze et al. 1994). Another is the bias in the selection of test stores, which are mostly chosen on the basis of convenience or because of their willingness to cooperate. Therefore findings usually have a limited external validity. Another danger is the misinterpretation of findings. Since in-store experiments usually only include a limited amount of product groups, they cannot be generalized indefinitely (Doyle and Gidengil 1977).
RESEARCH OBJECTIVES AND HYPOTHESES
17 3 Research Objectives and Hypotheses
Drèze et al. (1994) showed that an item’s shelf position has a larger impact on its sales than the number of facings it occupies. Breugelmans et al. (2007) furthermore summarize that in terms of absolute shelf placement, the vertical shelf position exceeds the horizontal one in regards to its effect on sales. Yet, few studies have focused on shelf layout. To the author’s knowledge, none has concentrated on the effect of vertical shelf positions alone. There is thus only general knowledge concerning the vertical aspects of shelf layout, e.g. that a position on eye-level is the most favorable one. However, whether or not higher quality positions have a stronger effect on the sales of PLs or on those of NBs has not yet been investigated. Nor has a distinction between the effect of an eye-level position on sales in hedonic and utilitarian product groups been made. This study seeks to deepen and differentiate the existing knowledge on the effect that vertical shelf positions have on sales of various products. Therefore, six hypotheses are formulated in the following subparagraphs, which are to be tested in the research that is to follow. They are based on the reviewed literature. As current knowledge concerning more specific aspects of vertical shelf positions is rather slim, assumptions are developed as well.
3.1 Hypotheses Regarding Private Labels and National Brands
If a product is moved from a less favorable position (hip-level) to a higher quality (eye-level) position, this should result in a higher visibility and thus higher sales (e.g. according to Drèze et al. 1994 and Van Nierop et al. 2008). The opposite change, on the other hand, should cause a decrease in sales (Drèze et al. 1994, Corstjens and Corstjens 1995 and Van Nierop et al. 2008). These expectations can also be expressed in the following hypotheses:
H1: Moving items from hip- to eye-level has a positive effect on their sales.
H2: Moving items from eye- to hip-level has a negative effect on their sales.
While most researchers and managers would probably agree with these rather general hypotheses, even without performing any empirical research, there is little knowledge on whether these premises apply more to PLs or to NBs. One could actually argue each case in a plausible way.
RESEARCH OBJECTIVES AND HYPOTHESES
18
ones on the shelf. Thus, customers might believe that retailers also put their most valuable, high quality products there, while lower quality products are placed at the vertical ends of the shelf. A shelf position on eye-level could thus serve as a cue for quality and add to the brand equity of PLs and to their sales. The reason that an eye-level position could have a stronger effect on sales of PLs than of NBs is that retailers do not usually advertise their PLs as much outside of the store as NB manufacturers do. So, even if NBs are not placed on eye-level, they can still have a strong brand equity that is based on external marketing measures, such as brand communication and advertising. There might also be higher brand loyalty for NBs so that people are willing to look for these products even if they are not immediately seen. This is less likely to apply to PLs, which makes an eye-level position more crucial for them.
On the other hand, one could also imagine that an eye-level position has a stronger effect on NBs. The rationale behind this assumption is that customers choosing products on eye-level positions seem to be strongly influenced by the shelf display. They stand in front of the shelf and chose the product that is most visible. It can furthermore be assumed that customers buying products from eye-level are less price-sensitive, because else they would bend down and look for cheaper alternatives. Based on these assumptions, one could make the case that an eye-level position is in fact more vital for NBs than it is for PLs.
3.2 Hypotheses Regarding Discount Private Labels and Specialty Items
One may wonder if discount PLs are always put on floor-level due to their heavy, bulk size packages or if their static floor position is part of the shelf optimization. It seems obvious that retailers would not allocate their most valuable shelf positions (eye-level) to discount PLs. Even if their percent profit margins may be high, their absolute margins are quite low. As discount PLs usually have a sales price far below average, price conscious consumers are more likely to look for them and go through the effort of bending down. It is not quite clear, however if the floor-level position is of practical nature, based on some retail heuristic or actually the optimal position for discount PLs. In addition, it has not been investigated, if sales of specialty items, which are usually located on top-level, would decrease if these items were moved to the floor-level. As they are very particular items, customers might be used to search for them and even buy them if they were located at the bottom of the shelf instead of the top.
RESEARCH OBJECTIVES AND HYPOTHESES
19
shelf borders on eye-level rows, whereas the floor-level is farthest away from eye-level. The top-level can thus be considered slightly higher in quality than the floor-level. Moving discount PLs up to the top-level would thus mean improving their position which should lead to an increase in sales. Moving specialty items down should however decrease their sales, as they are put in a slightly worse shelf position. In accordance with the assumption that consumers would be more willing to search for those products, this negative effect should not be as strong as the positive effect from moving discount PLs to the top of the shelf. These assumptions are expressed in following hypotheses:
H3: Moving discount PLs from floor- to top-level results in an increase of their sales.
H4: Movingspecialty items from top- to floor-level results in a decrease of their sales.
H5: The absolute positive effect of moving discount PLs from floor-level up to top-level is bigger
that than the absolute negative effect of moving specialty items from top-level down to floor-level.
3.3 Hypothesis Regarding Hedonic and Utilitarian Product Groups
In categories where impulse buying is important and variety seeking is the norm, eye-level positions are the critical factor for customers to choose a brand (Corstjens and Corstjens 1995). While the differentiation applied in this report is between hedonic and utilitarian product groups, impulse buying and differentiation is stronger related to hedonic categories, where affective components play a role (Hausman 2000, Sloot et al. 2005). Therefore, the following hypothesis can be formulated:
H6: Moving items closer to eye-level will have a stronger positive effect on sales in hedonic than
RESEARCH METHODOLOGY
20 4 Research Methodology
The proposed hypotheses could be studied by using time series, cross sectional data or an experiment. Using the first research methods, it is difficult to conclude on the causal relationship between a product's position on the shelf and its sales. Through the use of an experiment, this causality can be found more easily. Due to the advantages mentioned under 2.2, it was decided to conduct this research as a field experiment. The hypotheses of both studies, introduced in the preceding subparagraph, can be tested in the same research. For both studies, the independent variables are the change in the vertical shelf position. This change can take the form be an upward change (which is expected to be an improvement) or a downward change (which is hypothesized to be a deterioration). Sales are the dependent variable. In order to make sales comparable throughout all product groups, with different price levels, the relevant sales figures will be the number of units sold for each product.
In the following, the characteristics of the retailer where the experiment is conducted are elaborated on. Furthermore, the design of the experiment is explained.
4.1 Store Characteristics
The tests are carried out at an Albert Heijn (AH) franchise store in Groningen, NL. Albert Heijn is the flagship of Royal Ahold, one of the world's largest grocery retailers, with a market share of approximately 27 percent in the Netherlands (Ailawadi et al. 2008). It is widely perceived to be the most innovative Dutch grocery retailer (Steenkamp and Dekimpe 1997). The chain is positioned on service. PL makes up about 23 percent of all SKUs and 42 percent of purchases (Ailawadi et al. 2008). The high PL-share makes AH stores very suitable environment to study effects between PLs and NBs. The retail chain offers four types of PLs: a discount PL (Euroshopper), a regular PL (AH), a specialty PL (AH Biologisch2) and a premium PL (AH Excellent). Each week products are offered as a “Bonus” promotion. This entails a reduction in price when buying a single or several products. This promotion is valid for all holders of a “Bonus Card”, which customers can obtain for free.
The relevant franchise store used to be a C1000 store and was remodeled in September of 2008. The available data is therefore limited to a year. The research is furthermore restrained by the data that is available on store level. Profit margins for each product, for example, are not disclosed to the store management. Other
2
RESEARCH METHODOLOGY
21
information, such as implementations of promotions and changes on the shelf are only stored as long as they are relevant for store operations. Beyond that, the information usually has little value for the store management and in most cases is not stored. Additionally, upcoming promotions can only be viewed 1-2 weeks in advance. Finally, one can only access data through a closed intranet at AH. For the purpose of this research, all data thus has to be copied manually. The research is therefore performed within the realm of these information restrictions.
4.2 Experimental Design
As explained in the research objectives, there are two parts to this research. The first deals with the regular PL and NBs, whose shelf position is altered between eye- and hip-level. The second seeks answers concerning the effect of floor and top-level shelf positions on discount PLs and specialty items. Both objectives are sought to be met through an in-store experiment. Though both studies differ to some extent, their general experimental design is quite similar. Hence, the general design is explained first, before details about each study are elaborated on.
In a nutshell, the experiment consists of vertical changes of two rows within one-meter shelves. Only one-meter shelves were considered, in order to be able to manipulate the independent variable (vertical shelf position) without changing the horizontal shelf position. If this was not the case, there would be too much variation within the data, so that this study would most likely not produce any clear results.
In some categories, one meter shelves are occupied by a single product group (e.g. milk). In others, it is a mix of several smaller product groups that occupy a couple of rows each (e.g. ice cream, frozen fruit, frozen cakes, etc.). In order to be able to switch two rows within the shelf (and to compare apples with apples), it was a prerequisite that the majority of each shelf selected was occupied by a single product group. The goal was to switch two rows within each shelf while keeping all other variables (e.g. facings, horizontal position, tags, price, etc.) constant. Therefore, the rows had to additional conditions:
The whole row should be occupied by the same type of product.
The whole row should be occupied by either NBs, the regular PL or the discount PL, not a mix of them.
RESEARCH METHODOLOGY
22
For each of the studies, shelves in different categories were selected. For study one, changes were performed between eye- and hip-level. For study two, rows on floor- and eye-level were changed. Since most shelves encompass a different number of rows (in the experiment between six and eight), the following definition of top-, eye- hip- and floor-level rows were applied: The top- and floor-level consist of one row each, on the top and on the bottom of the shelf. The eye-level shelves are all the rows from the middle of the shelf up minus the top-level row. The hip-level shelves are all the rows from the middle of the shelf down minus the floor-level row. This categorization is following that of the EFMI study performed in 2008 (Sloot et al. 2008).
Top-level
Eye-level
Hip-level
Floor-level
Figure 1: Categorization of Shelf
4.2.1 Study One: Private Labels and National Brands
To test the first two hypotheses, two types of changes were performed in six categories. In three categories, where the PL was currently on eye-level, the PL was moved down to hip-level and the row with NBs was moved up. The relevant product groups for this first type of change were meal kits (packaged meal mixes, to which one has to only add fresh ingredients), pasta and cereal (Appendix A gives an overview of the shelves). This change in the shelf layout created an improvement of vertical shelf positions for the NBs and deterioration of position for PLs. A graphical example of this type of change is given below.
RESEARCH METHODOLOGY
23
Initial Shelf Layout Shelf Layout During Experiment
NB NB Specialty PL Specialty PL PL NB NB NB NB PL Discount PL Discount PL NB NB
Figure 2: Example of First Type of Shelf Change
In three other categories, where the NB currently occupied shelf positions on eye-level, the second type of change was performed. There, the NB was moved down from eye- to hip-level and the PL up. The relevant categories were “ontbijtkoek” (a Dutch breakfast cake), cooled juice and canned vegetables. In those three categories, the changes created an improvement for PLs and deterioration for NBs compared to the initial vertical shelf position (see Appendix B). The graphic below illustrates an example of the second type of change.
Initial Shelf Layout Shelf Layout During Experiment
NB NB NB PL Specialty PL, NB Specialty PL, NB PL PL PL NB Discount PL Discount PL
Figure 3: Example of Second Type of Shelf Change
4.2.2 Study Two: Discount Private Labels and Specialty Items
RESEARCH METHODOLOGY
24
Initial Shelf Layout Shelf Layout During Experiment
NB Discount PL NB NB NB NB NB NB PL PL PL PL PL PL Discount PL NB
Figure 4: Example of Third Type of Shelf Change
4.2.3 Study Three: Hedonic and Utilitarian Product Groups
RESEARCH METHODOLOGY
25
Study Performed Change Product
Groups
Examples of Switched
Products Classification
Meal kit Utilitarian /
Hedonic Pasta Utilitarian Change Type 1: PL: Eye- to hip-level (deterioration) NB: Hip- to eye-level (improvement) Cereal Utilitarian / Hedonic Ontbijtkoek Utilitarian / Hedonic
Cooled juice Hedonic
Study One (PLs vs. NBs) Change Type 2: PL: Hip- to eye-level (improvement) NB: Eye- to hip-level (deterioration) Canned vegetables Utilitarian Licorice Hedonic Hagelslag Utilitarian / Hedonic Study Two (Discount PLs vs. Specialty Items) Change Type 3: Discount PL: Floor- to top-level (improvement) Specialty NB: Top- to floor-level (deterioration)
Asian chips Hedonic
Table 2: Overview of Performed Changes (Image Source: www.ah.nl)
4.2.4 Timing of Experiment
RESEARCH METHODOLOGY
26
store operations proceeded as normal. This included promotional activities, which are directed by the central AH office, and the regular restocking practices, so that no out-of-stock situations occurred. The shelves were changed back to the original planogram in calendar week 32, with August 4th as the last day of the experiment. The author performed all the shelf changes herself, with support of the AH staff. This was to ensure that all changes were carried out as planned and that no other variables (such as facings, vertical position of items or shelf height) were altered.
4.2.5 Confounding Variables
The start of the experiment coincided with the first week of the Dutch summer vacation. This is expected to add variation to the data that is unrelated to the experiment. However, as this factor is known, it will be considered in the analysis. The advantage to performing the experiment during the summer is that there are fewer changes in the planograms, as was disclosed by the store management. As mentioned before, on the store level, Bonus promotions can maximally be predicted 2 weeks in advance. Even with a careful selection of product groups, it was thus not possible to avoid this influence completely.
In order to control for the confounding variables, it was considered to use a similar and nearby store as a control store. Yet, the demographics of its customers were quite different. The vacation period, for instance, showed a bigger effect at the potential control store than at the test store. In addition, the floor plan of the store was quite different, as well as the meters allocated to each category. This also made the vertical shelf positions of the included SKUs unequal. Hence, it was decided, to add control variables from the same store instead of from a control store.
4.3 Data Collection
The Albert Heijn shelf management system comprises real-time scanner data. While this creates a good basis for the data collection, there are certain practical limitations to the attainable data. While it is possible to access daily sales data for each SKU, the output can only be generated per category. Thus, each SKU has to be looked up in a rather extensive list.
RESEARCH METHODOLOGY
27
NBs. Furthermore, sales data for the relevant categories was collected. This was also done to get a better idea of the turnover and size of the categories, also in comparison to one another.
The rows that were manipulated include 82 SKUs. Two SKUs had to be excluded, because they were delisted during the run of the experiment. For the analysis, eight weeks of data were selected, four weeks before the intervention and four weeks after. Due to practical reasons, the first week of the experiment only included Wednesday through Saturday and the last week Monday and Tuesday. Calendar weeks 28 and 32 were hence combined and treated as one week. While this was not optimal, the combined week still included all weekdays with their specific characteristics (e.g. Saturday is usually a busier day than Tuesday).
To be able to estimate the effect of the vacation period, control items were selected for each of the manipulated 80 SKUs. The position of these control items was not manipulated during the experiment. In the selection of control items, it was important to decide on items that were as similar to the manipulated SKUs as possible. Thus, items were chosen that held a similar vertical shelf position and were of the same brand. If there were no such items in the relevant category, an SKU of the same tier was chosen (e.g. another national A-brand, national B-brand, Albert Heijn PL, etc.). As the control items were kept constant during the run of the experiment, the effect of the vacation period can be estimated. The same eight weeks of data were included in the data set, which amounted to a total of 1280 observations over the nine product groups.
Due to the limited amount of weeks, promotional sales could not simply be excluded, as was done by e.g. Drèze et al. (1994). All promotions were recorded on a weekly basis during the run of the experiment. This included promotions of the SKUs that were manipulated, as well as promotions of other articles within the same product group on the shelf. These competitive promotions were documented as they were expected to have a negative effect on the sales of the manipulated items, which is unrelated to the experiment.
4.4 Plan of Analysis
RESEARCH METHODOLOGY
28
Furthermore, for each type of change the three relevant categories are combined into one analysis. This means that, e.g. the meal kit, pasta and cereal product groups (where the PLs were moved down and the NBs up) are aggregated into one data base. As there are category specific differences, variables are added to account for them. They are entered into the regression as three dummy variables, one for each product group. The advantage to combining the product groups is that it makes the data base larger, which also makes results more generalizable. All nine categories cannot be combined, however, because the three performed types of changes are very different in nature (PL down / NB up; PL up / NB down; discount PL up / specialty NB down). The analysis is divided into descriptive statistics, regression analysis and correlations.
Descriptives
As a first step of an analysis, descriptive statistics provide a useful overview of the data. In addition, they are a tool to examine the distribution of the sample and to check for outliers. This is important because subsequent statistical techniques, such as the regression analysis, are very sensitive to outliers. In some cases, several SKUs are part of a brand or product wide Bonus promotion. The promotion increases and often multiplies their sales. Yet, since this effect is due to the special Bonus promotion and several items are affected by it, these values can be explained through the variables of the analysis. In other cases, however, where there is a single outlier that cannot be explained by the independent variables, it is double checked to ensure there was no error in entry. Some authors suggest removing these values from the data set. Others take a less radical view and convert them into less extreme values (Pallant 2005). It could already be told from the data entry, however, that the extreme values only make up a minor fraction of the overall data. Thus removing them should not diminish the quality of analysis.
Regression Analysis
To test in how far a change in the vertical position of an SKU changes its sales, a linear regression is used. A regression analysis, in comparison to e.g. correlations, has the advantage that causal relationships can be identified, which is crucial for this research and the proposed hypotheses. Another advantage (over techniques such as e.g. t-tests) is that other effects can be controlled for. A multiple regression thus provides more opportunity to account for real life situations.
RESEARCH METHODOLOGY
29
independent variables is binary, with values of either 0 or 1. Since there are several independent variables, a multiple linear regression is used. This statistical technique helps to explore the predictive ability of a set of independent variables on one continuous variable (Pallant 2005). As mentioned before, there are other variables affecting sales (Bonus promotions, competitive Bonus promotions and the start of the vacation period). Within the multiple regression, these factors are entered as dummy variables. All independent variables are hence binary. In this way, their effect, which is unrelated to the experiment, is accounted for. The retail sales price is not taken into account as a variable, because it is not assumed to drastically change over an eight week period. Extreme discounts are taken into account through the dummy variable for Bonus promotions. For the multiple regression, the enter method is used. This pursues the goal to look at all independent variables simultaneously.
There are several factors that need to be considered before applying a multiple regression. The first is the sample size. As a rule of thumb, the required sample size for a multiple regression is n > 50 + 8 m, with m being the number of independent variables (Pallant 2005). There are five independent variables (upward change, downward change, and three dummy variables that also enter the regression as an independent variable). Therefore, the minimum sample size should be 90. If the dependent variable is skewed, even more cases are needed (Pallant 2005). Furthermore, multicollinearity can be an issue. This refers to the level of correlation between the independent variables. Since this research is performed as an experiment, however, the independent variables are controlled. Each observation either has received a treatment (either upward or downward movement) or it has not. So these independent variables do not overlap and multicollinearity between them should not be a problem. There can however, be correlations with the dummy variables, which would make the model weaker. This needs to be examined before the analysis. Variance inflation factor (VIF) values above 10 (or tolerance values of less than ,10) are an indication that multicollinearity is present (Pallant 2005). As these values still allow for quite high correlations between the independent variables, correlations between them are also examined.
RESEARCH METHODOLOGY
30
RESULTS
31 5 Results
In this chapter, the results of the analysis are presented. This is done in a rather sober fashion, without interpretation of the findings. The reason for this is that all results need to be considered first, before hypotheses can be tested and generalizations can be derived. The proposed hypotheses are tested at the end of the chapter, under 5.4.
Following the structure of the plan of analysis, the descriptive statistics are given first. As a second step, the results of the individual models for each of the product groups are presented. Thirdly, the results of the combined analyses for each type of change are considered.
5.1 Descriptives
To get an idea of the data bases and the categories to which products belong, a short overview is given. Over the span of the eight weeks of the experiment, the following average weekly sales were realized in each of the categories (see figure 5 below). The categories are labeled the way they are defined by AH. The product groups that were part of the experiment, are stated in parentheses.
€ 2.508 € 1.993 € 1.886 € 1.499 € 1.152 € 845 € 821 € 908 € 927 0 € 1.000 € 2.000 € 3.000 € Suike rwer k (L icor ice) AG F C onse rven (Can ned Vege tabl es) Bro odbe leg (Hag elsl ag) Med iterr aan (Pas ta) Cat egor y B asis Inte rnat iona l (M eal K it) Gek oeld Sap (Col led Juic e) Oos ters (Asi an C hips ) Cer eals (Cer eal) Bro odve rvan gers (Ont bijtk oek)
RESULTS
32
Only four rows out of two shelves of each category (two manipulated rows and two control rows) entered the data sets for the analysis. Any statement concerning the share of PLs and NBs within these data bases would be very unrepresentative for the category as a whole. Yet, due to the mentioned limited accessibility of data, entering all items for nine whole categories was not feasible. All items for the nine manipulated one-meter shelves, were entered however. Their shares of shelf space (facings) are presented in the table below. On average the shelves are made up of 42 percent PLs and 58 percent NBs. The PL share of the respective shelves is thus considerably higher than the overall PL share given by Ailawadi et al. (2008), which was said to be 23 percent (see paragraph 4.1).
Product Group SKUs Facings Private Label National Brand
38 % AH 51% Knorr 11% Honig Meal kit 42 55 38 % PL 62 % NB 12 % AH 14 % AH Biologisch 10% Euroshopper 16 % Grand Italia 44 % Honig 4% Napolina Pasta 37 50 36 % PL 64 % NB 30 % AH 10 % Euroshopper 43 % Kelloggs 13 % Quaker 3% Weetabix Cereal 21 30 40% PL 60 % NB 43 % AH 3 % Euroshopper 53 % Peijnenburg Ontbijtkoek 27 30 47 % PL 53 % NB 46 % AH 7 % AH Biologisch 6 % AH Excellent 36 % Coolbest 6 % Hero Fruit & Co
Cooled juice 32 72 58 % PL 42 % NB 26% AH 10% AH Biologisch 18 % Euroshopper 46 % HAK Canned vegetables 29 61 54 % PL 46 % NB 27 % AH 10 % Euroshopper 10 %Autodrop 3 % Harlekijntjes 16 % Klene 8 % Oldtimers 27 % Venco Licorice 55 63 37 % PL 63 % NB 21% AH 13% Euroshopper 3 % Bolletje 3 % Chogal 33 % De Ruijter 3 % Fred & Ed 5 % Fair Trade 1 % K3 1 % Piet 16 % Venz Hagelslag 36 75 35 % PL 65 % NB 20 % AH 15 % Euroshopper 54 % Conimex 5 % Go-Tan 7 % TL Asian chips 13 41 34 % PL 66 % NB
RESULTS
33 5.2 Findings Per Product Group
For all the product groups, the histograms showed positive skewness of the dependent variable sales. This needs to be considered when accessing whether the individual sample sizes are sufficient. Before going into the results of each individual regression analysis, all of them were inspected for multicollinearity. While multicollinearity was not an expected outcome, due to the way the research was set up, it was still checked for completion’s sake. The Pearson correlations between the independent variables were assessed, to see if they had score lower than ,5 for all categories. All of them did. This is a first indication that multicollinearity does not exist. The correlations that were close to ,5 are due to the fact that the experiment overlapped with the beginning of the vacation period. Furthermore, all tolerance values are considerably higher than ,10 and VIF values lower than 2, which implies that multicollinearity is not an issue in the data sets.
Meal Kit
For this category, n is equal to 144. This creates a suitable sample size. Yet, there were no significant correlations to be found between the independent variables and sales. The multiple regression model also did not produce any significant results that were able to explain the variance in sales (see Appendix D for an overview of the results).
Pasta
RESULTS
34
Cereal
The n of the cereal product group is equal to 96. This is more than the amount required by the rule of thumb, which would be a minimum of 90 observations. While it is not a very extensive data set, it should be sufficient for the multiple regression analysis. When looking at the Pearson correlation, one can examine a positive relationship (,344) between the downward change of items and their change in sales. With an adjusted R square of ,23, the independent variables as a whole explain 23 percent of the variation in sales. The model reaches statistical significance. The R square is equal to ,27. The model has a modest fit. Yet, more that 70 percent of the variation is still due to other factors. To conclude on how much the upward and downward change contributed to the prediction of sales, unstandardized coefficients are examined. The downward change has a positive effect on sales, with a somewhat low significance level of 0,093 (for a summary of the results see Appendix F).
Ontbijtkoek
The entered observations for ontbijtkoek amount to 112. The box plot identified one value to be an outlier. It could be that this SKU was part of a special promotion apart from the regular Bonus offers. Even though the AH system showed that the item was on promotion, it is unclear what kind of promotion it was. It was not across the category or product group, since no other items were affected. Therefore this high value cannot be explained by the dummy variable “Bonus promotion”. The observation was removed from the data set, to eliminate distorting effects unrelated to the experiment. The remaining observations still create a suitable sample size. The model proved to be significant, with an adjusted R square of ,088 and an R square of ,113. These values can be considered low, as approximately 90 percent of the variance in sales are explained by other factors. Unlike the previous categories, the upward and downward change of items in the ontbijtkoek product group showed some significant results. The beta coefficient of the upward change (3,286) and the downward change (3,223) are both positive, with that of the upward change being slightly higher (see Appendix G for summary of results).
Cooled Juice
