Improving promotion image & Gaining more share-of-wallet

(1)

Faculty of Economics and Business University of Groningen

9700 AB Groningen, The Netherlands

June 29, 2020 Supervisor: Prof. Dr. L.M. Sloot

2nd Supervisor: Dr. P.S. van Eck

Improving promotion image

&

Gaining more share-of-wallet

Steven van Middendorp s.van.middendorp@student.rug.nl

S3850749

Master Thesis

MSc. Marketing Intelligence

(2)

ACKNOWLEDGEMENT

(3)

ABSTRACT

This study not only examines the main effects of promotions on promotion image and share-of-wallet, but also if the moderating effects of deal-proneness. More specific, the effects of promotion intensity, promotion depth, type of brand promotions and absolute promotion value on both promotion image and share-of-wallet are examined and in addition if these effects are different for deal-prone consumers. Monthly data of 5595 consumers who (partially) did their groceries at Albert Heijn, Jumbo, Plus, Hoogvliet or Dirk from June 2017 to December 2019 is collected. This data is combined with promotional data of these retailers for the corresponding period. While formally no conclusions can be drawn from this study, the pooled models still indicate that promotions are effective in increasing share-of-wallet and that promoting national brands is effective in both increasing promotion image and share-of-wallet. Furthermore, the pooled models also revealed that deeper promotions of perishable food-products and non-food products is effective in increasing promotion image. Moreover, the pooled models also indicate that for deal-prone people, deeper promoting non-perishable product is effect in increasing promotion image, however, deeper promoting perishable food-products is ineffective in increasing promotion image. Additionally, for deal-prone people the absolute promotion value also matters, as this leads to both an increase in promotion image and share-of-wallet.

(4)

4

TABLE OF CONTENT

1 INTRODUCTION ...6

2 LITERATURE REVIEW...9

2.1 Promotion image & share-of-wallet ...9

2.2 Promotion intensity ... 10

2.3 Promotion depth ... 11

2.4 Type of brand ... 12

2.5 Absolute promotion value ... 13

2.6 Moderating effect of deal-proneness ... 14

2.7 Conceptual model & control variables ... 15

3 METHODOLOGY ... 16

3.1 Measurement ... 17

3.2 Data preparation ... 18

3.3 Data cleaning... 19

3.4 Factor analysis for deal-proneness ... 19

3.5 Variable overview ... 21

4 RESULTS ... 23

4.1 Descriptive statistics ... 23

4.1.1 Promotion image & share-of-wallet ... 23

4.1.2 Promotion intensity - National brands & Private labels ... 24

4.1.3 Promotion depth ... 26

4.1.4 Type of brand promotions - National brands & Private labels ... 28

4.1.5 Absolute promotion value... 30

4.2 Empirical results ... 33

4.2.1 Multicollinearity ... 33

4.2.2 Pooled regression model ... 34

(5)

5

4.2.4 Unpooled regression models (unit-by-unit) ... 38

4.2.5 Model validation ... 45

4.3 Additional analysis ... 46

5 CONCLUSION & DISCUSSION ... 48

6 LIMITATIONS & FUTURE RESEARCH ... 51

REFERENCES ... 53

(6)

6

1 INTRODUCTION

Multiple-store shopping behavior is one of the most important trends characterizing todays grocery retailing business (Gijsbrechts, Campo, & Nisol, 2008; Kahn & McAlister, 1997). Rather than staying loyal to one store and revisiting the same store, consumers actively exploit opportunities offered by retailers and visit two or more grocery stores on regular basis. In fact, multiple-store shopping has become the standard in today’s shopping environment as multiple studies confirmed that roughly 75% of all grocery shoppers regularly visits more than one store each week (Fox & Hoch, 2005; Gijsbrechts et al., 2008).

(7)

7 to these thresholds, promotions are also bounded by saturation effects which are caused by the limited amount consumers can stockpile (Martínez-Ruiz et al., 2006; Van Heerde, Leeflang, & Wittink, 2001). Ultimately, retailers make decisions about which promotions from which categories they offer in order to capture the largest share of the customers’ total expenditures. Besides the distinction of promotions between perishable and non-perishable products, retailers also differentiate promotions between different types of brands. For instance, retailers are inclined to promote national brands prior to private label brands as promoting private label brands is generally less effective (Cotterill & Putsis, 2000). In line with these findings, Eales (2016) reported that promotional pressures were higher for national brands (24% in 2016) than for private label brands (13% in 2016). Private label brands, or store brands, are brands developed by retailers themselves. In contrast, national brands are brands developed by manufacturers. Therefore, retailers must hand over revenues captured from national brands to the manufacturers themselves, and retailers only get a fee in return. Hence, retailers earn more from selling private label brands than from national brands. Private label brands compete with national brands by using lower prices and less promotional expenses. Conversely, national brands are more expensive, which they justify by offering higher quality products (Bao, Bao, & Sheng, 2011). A.C. Nielsen (2005) reported that prices for national brands exceeded prices for private label brands by roughly 31%. From retailer’s perspective, it is therefore interesting to look into the relationships between different kinds of promotions and share-of-wallet. For instance, do promotion effects differ between national brands and private label brands? Do promotion effects differ between perishability and non-perishability? All these questions will be addressed in this study.

(8)

8 satisfaction and an important driver in store choice. From retailer’s perspective it is therefore interesting how sales promotion helps them in increasing their promotion image. How are these promotions related to promotion image? And which types of brand helps retailers predominantly in building promotion image? Answering these types of questions helps retailers understanding how consumers form a promotion image and how consumers decide on allocation of budgets across competing stores. Therefore, this study is going to shed new light on these questions.

From an academic perspective, this study contributes to the literature by shedding new light on the impact of different types of promotion measures on a retailers’ promotion image and share-of-wallet. While much literature is available regarding the effects on store image, there is not much literature to go on regarding the effects on promotion image. As such that in current literature promotion image is mostly studied as a part of store image, but less as a concept on its own. Therefore, this study will fill in this gap. Furthermore, current literature points toward effects promotions on share-of-wallet due to its effects on consumer behavior. Additionally, this study is also going to take into account that the effects of promotions might differ between people as one person can be more prone to deals than the other. Therefore, this study will analyze the moderating effect of deal-proneness on the impact of different types of promotion measures on promotion image and share-of-wallet.

(9)

9

2 LITERATURE REVIEW

Grocery retailing in the Netherlands can be considered as a highly competitive environment. Although the market grows bit by bit, this is mainly due to inflation. Retailers such as Albert Heijn and Jumbo, offer many sales promotions which are primarily intended to attract customers to their store in order to enlarge their share-of-wallet. Consumers benefit from these sales promotions as they buy certain products in promotion at one store and other products in promotion at another store, which is also known as ‘cherry-picking’. Hence, multiple-store shopping behavior is a predominantly popular trend in today’s grocery retail business (Kahn & McAlister, 1997). This trend has led to a share-of-wallet (SOW) competition among retailers due to the fact that every retailer wants to capture the largest possible share of the consumers’ grocery expenditures (Gijsbrechts et al., 2008).

2.1 Promotion image & share-of-wallet

(10)

10 expenditure (i.e., size-of-wallet) captured by the firm” (Du, Kamakura, & Mela, 2007; Jang, Prasad, & Ratchford, 2016; Mägi, 2003).

An important factor to consider when researching promotion effects is the difference between promotion effects for national brands and private label brands. Grewal et al. (1998) state that the effect of price promotions on perceived value is different for higher quality product due to the positive effect of brand name and brand perceived quality, compared to lower quality products. This suggest that national brand promotions result in a higher perceived value and subsequently have a higher perceived promotion attractiveness. Therefore, a division is made between national brands and private label brands. Furthermore, another important factor to consider is deal-proneness. According to Inman, McAlister, and Hoyer (1990), promotion effects are always moderated by individual difference variables. Deal proneness is defined as: “the extent that the proportion of purchases made on deal” (Bawa & Shoemaker, 1987; Blattberg, Buesing, Peacock, & Sen, 1978). Deal-prone customers have a high percentage of on-deal purchases compared to deal-proneness customers (Hackleman & Duker, 1980). Additionally, a previous study done by Cacioppo, Petty, Chuan, and Rodriguez (1986) showed that deal-prone people rely on promotional signals to evaluate. Therefore, deal-prone consumers are more likely to be influenced by the effect evoked from promotions.

2.2 Promotion intensity

(11)

11 promotion attractiveness, compared to promotions for private label brands. Based upon this reasoning, the following hypotheses are formed:

- Hypothesis 1A: promotion intensity for national brands has a positive relationship with promotion image.

- Hypothesis 1B: promotion intensity for private labels has a positive relationship with promotion image.

In addition to the findings above, several studies also suggest positive effects of price promotions on SOW due to its positive effect on consumer behavior. For instance, Gilbert & Jackaria (2002) found a positive effect of price promotions on consumer purchase behavior. Furthermore, Kumar & Leone (1988) found a positive relationship between price promotions and store substitution. While these results could point towards a positive relationship between price promotions and SOW, Cotterill and Putsis (2000) state that price promoting with private label brands is generally a less effective strategy for stimulating sales than it is for national brands. This is also in line with the findings of Grewal et al. (1988), who showed that the effect of price promotions on perceived value is different for higher quality product due to the positive effect of brand name and brand perceived quality, compared to lower quality products. Based upon this reasoning, the following hypotheses are formed:

- Hypothesis 2A: promotion intensity for national brands has a positive relationship with SOW.

- Hypothesis 2B: promotion intensity for private labels has a positive relationship with SOW.

2.3 Promotion depth

(12)

12 Cooper, 1992). Additionally, the saturation effect can be explained by the limited amount consumers can stockpile (Blattberg, Briesch, & Fox, 1995) and consume in response to the price promotions.

A study done by Van Heerde et al. (2001) provides evidence for the fact that the deal shape of packaged food products fits an S-shaped curve. More specific, Van Heerde et al. (2001) reports thresholds effects between 5% and 10%, and saturation levels between 15% and 25%. Furthermore, Martínez-Ruiz et al. (2006) found that promotions in perishable categories are most effective at levels between 5% and 25%, and promotions in non-perishable categories are most effective at levels between 2% and 15% percent. Additionally, Narasimhan, Neslin and Sen (1996) confirmed this as they found that stock items generally had a higher promotion elasticity. More specific, according to Gijsbrechts, Campo and Goossens (2003) it is clear that produce and fish/meat categories share a number of deal-prone related characteristics as these categories predominantly comprise frequently bought, high penetration items which make them prime candidates for price promotions (Fader & Lodisch, 1990). Build upon this reasoning, the following hypotheses are stated:

- Hypothesis 3A: promotion depth of non-perishable food-products has a positive effect on promotion image.

- Hypothesis 3B: promotion depth of non-food products has a positive effect on promotion image.

- Hypothesis 3C: promotion depth of perishable food-products has a negative effect on promotion image.

- Hypothesis 4A: promotion depth of non-perishable food-products has a positive effect on SOW.

- Hypothesis 4B: promotion depth of non-food products has a positive effect on SOW. - Hypothesis 4C: promotion depth of perishable food-products has a negative effect on

SOW.

2.4 Type of brand

(13)

13 than it is for national brands (Cotterill & Putsis, 2000). Therefore, it is expected that promotions for national brands are more attractive to customers and also result in a higher SOW for the retailer. Based upon this reasoning, the following hypotheses are stated:

- Hypothesis 5: promotions of national brands are positively related to promotion image. - Hypothesis 6: promotions of national brands are positively related to SOW.

2.5 Absolute promotion value

It is generally accepted that consumers compare a products price to an internal reference price when judging the attractiveness of a products’ price (Janiszewski & Lichtenstein, 1999). For instance, Thaler (1985) states “The measure of transaction utility depends on the price an individual pays compared to some reference price.” Likewise, Kalyanaram & Winer (1995) observe that “there is a significant body of literature to support the notion that individuals make judgments and choices based on the comparison of observed phenomena to an internal reference price”. For this reason, retailers typically want customers to perceive their products to have high reference prices, so perceived savings are greater when a price promotion is offered (Grewal et al., 1998). This suggests that consumers are more attracted to price promotions from expensive products, yet less attracted to price promotions from cheap products.

Expensive products are products which price exceeds the average price of the category. Although there is not much literature about consumer behavior regarding the effects of expensive products, Grewal et al. (1998) found that customers respond different to high- and low-priced products. In addition to these findings, Choi & Coulter (2012) found that consumers are likely to assess the magnitude of a price promotion in absolute value when the reference price is the marketer’s own price. Furthermore, their analysis also confirms that price promotions are more likely to be assessed in absolute terms when the higher prices refer to the retailer’s own regular prices. Since consumers show different buying behavior to high- and low-priced products, and consumers are likely to assess price promotions in absolute terms to the reference price, it is expected that the total absolute promotion value offered is positively related to both promotion image and SOW. Therefore, the hypotheses state:

- Hypothesis 7: absolute promotion value has a positive relationship with promotion image.

(14)

14

2.6 Moderating effect of deal-proneness

When examining promotion effects, it is important to take promotional proneness or deal-proneness into account. According to Inman et al. (1990), promotion effects are always moderated by individual difference variables. Deal proneness is defined as: “the extent that the proportion of purchases made on deal” (Bawa & Shoemaker, 1987; Blattberg et al., 1978). Deal-prone customers have a high percentage of on-deal purchases compared to deal-Deal-proneness customers (Hackleman & Duker, 1980). Cacioppo et al. (1986) showed that deal-prone consumers rely on promotional signals to evaluate products. Therefore, deal-prone customers are more likely to be influenced by the effect evoked from promotions. Based upon these findings it is expected that the promotion effects are higher for deal-prone consumers, than for deal-proneness consumers. The hypotheses states:

Dependent variable promotion image:

- Hypothesis 9a: deal-proneness positively influences the relationship between promotion intensity and promotional image.

- Hypothesis 9b: deal-proneness positively influences the relationship between promotion depth and promotional image.

- Hypothesis 9c: deal-proneness positively influences the relationship between type of brand promotions and promotional image.

- Hypothesis 9d: deal-proneness positively influences the relationship between absolute promotion value and promotional image.

Dependent variable share-of-wallet

- Hypothesis 10a: deal-proneness positively influences the relationship between promotion intensity and SOW.

- Hypothesis 10b: deal-proneness positively influences the relationship between promotion depth and SOW.

- Hypothesis 10c: deal-proneness positively influences the relationship between type of brand promotions and SOW.

(15)

15

2.7 Conceptual model & control variables

In conclusion, this study is going to investigate the main-effects and moderated effects of: promotion depth, promotion intensity, type of brand and absolute promotion value on promotion image and SOW. The conceptual model is shown below:

Figure 1 - Conceptual model

(16)

16

3 METHODOLOGY

This study uses two different datasets to test the hypotheses. The datasets are the EFMI Shopping Monitor and Superscanner.nl. The datasets contain information about the Dutch grocery retailing market for a period of 31 months ranging from June 2017 to December 2019. The EFMI shopping monitor is used to obtain data regarding the variables SOW, promotion image, deal-proneness, income, age, household size, education supermarket choice, supermarket close by and enough shops in the area. This dataset contains self-reported answers to a survey from the primary grocery shopper within a household. In general, the EFMI shopping monitor collects survey data from roughly 200 different respondents every month and the data are collected in during the second week of the month. The Superscanner.nl dataset is used to obtain information regarding the promotion related variables such as promotion intensity and depth. This dataset contains crawled data of daily prices and promotions of all products offered by Albert Heijn, Jumbo, Plus, Hoogvliet and Dirk. Table 1 gives an overview from which variables are obtained from the different datasets.

Dataset Obtained information

EFMI Shopper Monitor SOW, promotion image and deal-proneness of respondents for Albert Heijn, Jumbo, Plus, Hoogvliet and Dirk.

Control variables: income, age, household size, education, supermarket choice, supermarket close by and enough shops in the area.

Superscanner.nl Promotion intensity, promotion depth, type of brand promotions and absolute promotion value for Albert Heijn, Jumbo, Plus, Hoogvliet and Dirk.

Table 1 - Used datasets with corresponding obtained information

(17)

17 EFMI Shopper Monitor. Furthermore, in order to account for lag effects, which probably exist, the promotion variables will be calculated four weeks prior to the week of data collection.

3.1 Measurement

The previous section covered how the data for this study is collected. Table 2 gives an overview of how the data is measured and computed in order to make it usable for analysis.

Variable Measure

SOW “Can you estimate what percentage of your total grocery purchases last month you purchased at which supermarket?”

Promotion image “How would you rate the following supermarkets on the

attractiveness of their promotions?” (1 = very poor, 10 = excellent) Deal-proneness I always look at the leaflets of supermarkets to see which discounts

they currently offer (totally disagree, totally agree).

I am well-informed about the promotion deal of the different grocery retailers (totally disagree, totally agree).

I will trade in my usually grocery retailer, because of a very interesting promotion deal (totally disagree, totally agree). I pay a lot of attention to promotion deals when grocery shopping (totally disagree, totally agree).

I like comparing different grocery retailers on promotion deals (totally disagree, totally agree).

Promotion intensity national brands

Number of national brand Stock Keeping Units (SKUs) in promotion per month divided by total number of national brand SKUs in assortment (in %).

Promotion intensity private label brands

Number of private label brand Stock Keeping Units (SKUs) in promotion per month divided by total number of private label brand SKUs in assortment (in %).

Promotion depth perishable food products

Average discount (in %) of all perishable food SKUs in promotion per month.

Promotion depth non-perishable food products

Average discount (in %) of all non-perishable food SKUs in promotion per month.

Promotion depth non-food products

(18)

18 Share of national brands

promotions

Number of national brand Stock Keeping Units (SKUs) in promotion per month divided by total number of SKUs in promotion that month (in %).

Share of private label promotions

Number of private label brand Stock Keeping Units (SKUs) in promotion per month divided by total number of SKUs in promotion that month (in %).

Absolute promotion value Total absolute discount of all SKUs in promotion per month (in €).

Table 2 - Overview variables with their corresponding measurements

3.2 Data preparation

In order to compute the variables for promotion depth, the different categories of each retailer needed to be divided into universal categories across retailers. Table 3 gives an overview of the division of categories for promotion depth. It is important to note that the supermarket categories in table 3 are already aggregated, this is done for each retailer specific so they are universal across retailers.

Promotion depth Supermarket categories

Non-perishable Coffee, tea, juices & soft drinks Wine, beer & liquor

Frozen food

Breakfast cereals, spreads & snacks Sweets, biscuits, chips & baking products Pasta, rice & world cuisine

Soups sauces, herbs & oil Non-food Baby & drugstore

Pet- & household products

Cooking equipment & service desk

Perishable AGF

Meat, fish & chicken Bread & pastries Tapas & meats

Salads, pizzas & meals Dairy produce & cheese

(19)

19

3.3 Data cleaning

Before calculating the variables, the data needs to be checked for outliers. An outlier is considered to be a datapoint which differs significantly from the other observations. A possible explanation for an outlier would in this case that around Christmas retailers give more promotions. For this reason, it follows that the number of promotions around Christmas are significantly higher than the number of promotions for the rest of the year. However, there are also cases in which there is no logical explanation for an outlier, in this case the outlier is probably caused by a crawling mistake in Superscanner.nl and is removed from the data. After checking the data for Albert Heijn, some outliers are found in the number of national brands and private label brands in the assortment for week 44, 45 46 and 47 of 2017. Since the data is very stable the outliers are replaced by an average value of week 42, 43, 48 and 49 of 2017. After checking the data for Jumbo and Plus, no outliers are found. However, when checking the data of Hoogvliet some strange values appeared for week 39 and 40 of 2018 and week 28 and 29 from 2019 for all variables. Since the data is very dynamic, they cannot be replaced with an average value. Therefore, those weeks are removed from the data of Hoogvliet. After checking the data for Dirk, lots of outliers are found. Some outliers are found in the number of national brands in the assortment for week 24 and 25 of 2018, as the data is very stable the outliers are replaced by an average value of week 22, 23, 26 and 27 of 2018. However, there are more outliers for several weeks and for several variables. Those outliers cannot be replaced due to dynamics in the data. Therefore, they are removed from the data of Dirk. Finally, due to a general crawling mistake week 15 and 28 of 2019 are removed from the data for all retailers.

3.4 Factor analysis for deal-proneness

(20)

20 be concluded that factor analysis is appropriate. In order to examine which number of factors should be chosen the eigenvalues are calculated. The eigenvalues are calculated using a Principle Component Analysis. Only factors with eigenvalues above 1 should be selected, factors that explain more than 5% of the variance each and the total explained variance of those factors should be more than 60%. The results are showed in table 4 below.

Initial Eigenvalues

Components Total % of variance explained Cumulative %

1 1.841 0.678 0.678

2 0.758 0.115 0.793

3 0.688 0.095 0.888

4 0.577 0.067 0.954

5 0.478 0.046 1.000

Table 4 - Eigen values from Principle Component Analysis

(21)

21

3.5 Variable overview

In order to give an overview of the dependent and independent variables used in this study, table 5 and 6 present the mean values of all variables, together with their standard deviation and minimum and maximum value.

Albert Heijn Jumbo Plus

Mean (SD)

Min Max Mean (SD)

Min Max Mean (SD) Min Max SOW 48.55 (33.75) 0 100 41.55 (32.79 ) 0.00 100 34.28 (30.56 ) 0 100 Promotion Image 7.73 (1.41) 0 10 7.22 (1.63) 0 10 7.48 (1.40) 0 10 Promotion Intensity NB 12.85 (3.40) 6.09 20.53 6.74 (1.20) 3.43 9.50 4.07 (0.63) 2.24 5.89 Promotion Intensity PL 6.30 (2.07) 2.97 16.29 7.72 (1.99) 3.87 10.93 4.84 (1.90) 1.73 11.81 Promotion Depth: Perishable 28.48 (3.85) 22.9 6 41.20 22.79 (1.87) 19.77 31.51 29.54 (2.76) 23.83 42.46 Promotion Depth: Non-Perishable 28.03 (4.28) 20.0 3 41.88 21.92 (0.89) 20.17 25.22 32.63 (2.94) 25.54 43.53 Promotion Depth: Non-food 39.10 (3.57) 30.4 2 42.03 31.86 (4.67) 23.94 44.27 45.05 (3.67) 35.67 55.62 Share of NB 86.78 (3.67) 76.0 1 94.39 71.98 (5.86) 59.82 84.80 73.26 (6.73) 53.74 88.28 Share of PL 13.22 (3.67) 5.61 23.99 28.02 (5.86) 15.20 40.18 26.74 (6.73) 11.72 46.26 Absolute promotion value 7889 (2850) 2978 16866 6 1234 (249) 747.8 2077 901 (260) 405.4 1529 Deal-proneness 3.62 (0.83) 1.00 5.00 3.59 (0.85) 1.00 5.00 3.77 (0.82) 1.00 5.00

(22)

22

Hoogvliet Dirk

Mean (SD)

Min Max Mean (SD) Min Max SOW 44.89 (32.65) 0 100 41.68 (30.91) 0 100 Promotion Image 7.76 (1.35) 0 10 8.06 (1.35) 0 10 Promotion Intensity NB 7.81 (2.65) 1.88 14.74 5.34 (0.81) 2.46 7.49 Promotion Intensity PL 6.79 (1.71) 3.41 11.18 1.54 (0.89) 0.02 3.80 Promotion Depth: Perishable 28.19

(3.40)

20.46 39.76 28.23 (2.46)

23.37 36.89 Promotion Depth: Non-Perishable 30.53

(2.84)

24.59 41.52 29.75 (2.65)

24.70 38.65 Promotion Depth: Non-food 45.07

(3.20) 36.38 51.81 42.70 (3.92) 34.25 52.04 Share of NB 83.94 (4.57) 71.46 92.17 97.04 (1.76) 92.70 99.97 Share of PL 16.06 (4.57) 7.83 28.55 2.96 (1.76) 0.03 7.30 Absolute promotion value 954

(384) 177.8 2132 369 (91) 92.27 571 Deal-proneness 3.76 (0.78) 1.00 5.00 3.89 (0.72) 1.00 5.00

(23)

23

4 RESULTS

This chapter presents the descriptive statistics along with the empirical results of this study. The first sub-section shows descriptive statistics by presenting graphs in order to gain more knowledge about the data. The second session present the empirical results which will answer the hypotheses presented in the literature review (chapter 2).

4.1 Descriptive statistics

In order to get some feeling with the data, plots are made to see the development a variable over time. Chapter 3 already presented the mean, standard deviation, minimum and maximum for the independent and dependent variables, now a more elaborate description about the variables is given which compares the retailer over time.

4.1.1 Promotion image & share-of-wallet

Figure 2 shows the development of the average promotion image over time. The graph shows that overall, promotion image is the highest for Dirk and the lowest for Jumbo. This means that when a retailer is available to the respondent, they consider Dirk to have the most attractive promotions and Jumbo to have the least attractive promotions.

Figure 2 - Promotion Image

Promotion Image

(24)

24 Figure 3 shows the development of the average share-of-wallet over time. The graph shows that overall, Albert Heijn and Hoogvliet have the highest share-of-wallet and Plus has the lowest share-of-wallet. This means that when a retailer is available for a respondent, they shop a lot at Albert Heijn and Hoogvliet, and less at Plus.

4.1.2 Promotion intensity - National brands & Private labels

Promotion intensity is a variable which might have an effect on share-of-wallet and promotion image. Figure 4 shows the promotion intensity for national brands per retailer over a time period 137 weeks. It is notable from the graph that promotion intensity of national brands is the highest for Albert Heijn, compared to the other retailers whose promotion intensity of national brands is much lower. Additionally, these retailers are also more grouped together at the bottom of the graph which means that these retailers are more or less on the same level regarding promotion intensity for national brands. Furthermore, the graph also shows that the behavior of Albert Heijn and Hoogvliet is different from the other retailers as Albert Heijn is the only retailer which slowly increased his promotion intensity over time and the promotion intensity for Hoogvliet shows varying movements over time. The other retailers did not increase their promotion intensity but stayed more or less the same.

Figure 3 - Share-of-wallet

Share-of-wallet

(25)

25 Figure 5 shows the promotion intensity for private label brands per retailer. First thing notable is that the promotion intensity of private label brands is the lowest for Dirk. The other retailers are mixed together which means that these retailers are more or less on the same level regarding promotion intensity for private label brands. Furthermore, the graph also shows that both Albert Heijn and Plus increased their promotion intensity for private label brands. Especially for Plus, as the promotion intensity for Plus is increasing more substantially than it is for Albert Heijn. However, while the promotion intensity for Albert Heijn and Plus is increasing, the promotion intensity for Jumbo, Hoogvliet and Dirk shows more varying movements over time. In addition, both Jumbo and Hoogvliet show also similar movements after week 50.

Promotion Intensity - National Brands

Retailer:

Figure 4 - Promotion Intensity for National Brands

Promotion Intensity - Private Labels

Retailer:

(26)

26 Overall, it can be concluded that the promotion intensity of national brands has the upper hand. Furthermore, the behavior of Albert Heijn, Jumbo, Plus and Dirk regarding the promotion intensity of national brands is different from the promotion intensity of private label brands. Jumbo for example, while the promotion intensity of national brands stays more or less the same for this retailer, the promotion intensity of private label brands shows varying movements. This is also true for Plus as the promotion intensity of national brands for Plus stays more or less the same, the promotion intensity of private label brands in contrast shows a continuous increase over time. However, Hoogvliet is the only retailer for which the behavior regarding the promotion intensity of national brands is not different from the behavior of the promotion intensity for private label brands. Both figure 3 and 4 show that Hoogvliet decreased his promotion intensity for both national brands and private label brands up to week 95, after which this retailer increased it again.

4.1.3 Promotion depth

Promotion depth is another variable which might have an effect on share-of-wallet and promotion image. Figure 6 shows the promotion depth of non-perishable food-products over a time period of 137 weeks.

First thing notable is that the promotion depth of non-perishable food-products is different for Jumbo than it is for the other retailers. Where the lines for Albert Heijn, Plus, Hoogvliet and

Promotion Depth - Non-perishable food-products

Retailer:

(27)

27 Dirk are grouped together, the line for Jumbo is at a substantially different level at the bottom of the graph. This means that Jumbo does not promote non-perishable food-products as deep as the other retailers. Furthermore, where the lines from retailers Albert Heijn, Plus, Hoogvliet and Dirk vary over time, the line from Jumbo does not vary that much and is more stable over time. This means that Jumbo also not changes his average discount when it comes to non-perishable food products. In addition, the graph also shows that Plus offers on average the highest discount for non-perishable food-products compared to the other retailers. Furthermore, the graph shows similar movements for Plus and Dirk up to week 50 and a substantial decrease after week 100 in the promotion depth for Albert Heijn.

Figure 7 shows the development of promotion depth for perishable food-products over time. Just like the promotion depth for non-perishable food-product, the promotion depth for perishable food-products is different for Jumbo than it is for the other retailers. While the promotion depths for Albert Heijn, Plus, Hoogvliet and Dirk are at an equivalent level in top-middle, the promotion depth for Jumbo is substantially lower at the bottom of the graph. This means that Jumbo does not promote perishable food-products as deep as the other retailers. Furthermore, the graph also shows that promotion depth for Jumbo varies, which means that Jumbo changed their average discounts for perishable food-products over time. Conversely, this is in contrast with the average discounts of non-perishable food-products offered by Jumbo as these were more stable over time and did not change as much as the perishable food-products.

Promotion Depth - Perishable food-products

Retailer:

(28)

28 Figure 8 shows the development of promotion depth for non-food products over time. Just like the promotion depth for non-perishable and perishable food-product, the promotion depth for non-food products is different for Jumbo than it is for the other retailers. Additionally, where retailers Albert Heijn, Plus, Hoogvliet and Dirk are grouped together at an equivalent level, Jumbo is at a lower level at the bottom of the graph. This means that Jumbo does not promote non-food products as deep as the other retailers. Furthermore, the promotion depth of non-food products for Jumbo is only stable up to week 50 after which it substantially increases up to week 110. Another thing notable is that both Plus and Dirk slowly increase their promotion depth of non-food products as both lines steadily increase over time.

Overall, it can be concluded that promotion depth is the highest for non-food products, and promotion depth for non-perishable and perishable food-products are at an approximate same level. Furthermore, it can also be concluded that Jumbo does not promote their products as deep as the other retailers do because Jumbo scores the lowest in promotion depth for non-perishable, perishable and non-food products. Furthermore, promotion depth for the other retailers is at an equivalent level.

4.1.4 Type of brand promotions - National brands & Private labels

Type of brand promotions is another variable which might affect share-of-wallet and promotion image. Figure 9 shows the development over time for the share of national brand products in promotion. It is important to note that all shares are relative to all the products in promotion.

Promotion Depth - Non-food products

Retailer:

(29)

29 From the graph is clear that Dirk has the highest share of national brands in promotion. The other retailers start at an equivalent level, however at the end, in week 137, Albert Heijn and Hoogvliet are still at an approximate same level, but Jumbo and Plus are at a lower level. This means that Jumbo and Plus decreased their share of national brands in promotion substantially over time, but the share of Albert Heijn only shows minor increases and decreases. The latter is also true for retailer Dirk which also shows minor increases and decreases. Conversely, the share of national brands in promotion for Hoogvliet shows a substantial decrease up to week 90 after which it stays at an approximate same level.

Figure 10 shows the development over time for the shares of private label products in promotion. Just like the shares of national brands, it is clear that the share of Dirk is different from the other retailers. The shares of private label products for Dirk are low which is in contrast to the shares of national brands for Dirk in which they were at top. This means that Dirk mainly promotes national brands and only promotes a few private label brands. Furthermore, while the shares of Albert Heijn only show minor increases and decreases over time, the shares for Jumbo and Plus show a substantial increase over time. Moreover, the shares for Hoogvliet only show a substantial increase up to week 90 after which it stays at an approximate same level. It is important to note that the share of private label brand promotions cannot be used in a regression model as this is the opposite of the share of national brand promotions due to the fact that both variables sum up to 1. Therefore, the effects of national brands (mentioned in section 4.2) are relative to the share of national brands.

Figure 9 - Type of brand for National Brands

Type of brand - National Brands

(30)

30 Overall, it can be concluded that all retailers promote more national brands than private label products. Especially Dirk, as this retailer mainly promotes national brands and only a few private label brands. Additionally, the shares of Dirk show only minor increases and decreases over time. The latter is also true for Albert Heijn. However, it is not true for Hoogvliet as the shares of Hoogvliet vary over time. This retailer decreased their share of national brands substantially up to week 100, but as they decreased it, they also increased their share of private label brands. In addition to Hoogvliet, retailers Jumbo and Plus also vary over time. Plus decreased their share of national brands substantially over time, however, as they decreased it, they also increased their share of private label brands with the same rate. Jumbo decreased their share of national brands substantially up to week 50, but meanwhile they also increased their share of private label brands. After week 50 Jumbo shows minor increased and decreases. These findings are in line with literatures about national brands versus private label brands as it is suggested that price promotions for national brands might have a stronger effect because they are perceived as higher quality products (Deleersnyder et al., 2007; Grewal et al., 1998). Therefore, retailers are more inclined to promote national brands as promoting private label brands is generally a less effective strategy in stimulating sales (Cotterill & Putsis, 2000).

4.1.5 Absolute promotion value

The absolute promotion value offered by retailers is another variable which might have an impact on share-of-wallet and promotion image. Figure 11 shows the absolute promotion value

Type of brand - Private Labels

Retailer:

(31)

31 per retailer over time. From the graph is clear that Albert Heijn is superior regarding absolute promotion value. In comparison, the absolute promotion value of Albert Heijn around week 100 is about 8.000 euros, however, the absolute promotion value of the other retailers is around week 100 on average about 1.000 euros. This means that in week 100 the absolute promotion value of Albert Heijn is 8 times higher than the other retailers. Moreover, this difference becomes also bigger after week 100 as the absolute promotion value of Albert Heijn shows a substantial increase up to almost 15.000 euros in week 137. The difference between Albert Heijn and the other retailers can be explained by the size of the assortment of Albert Heijn as Albert Heijn also has a larger assortment than the other retailers have. Furthermore, Albert Heijn also uses a special bonus-promotion system, these are promotions which they offer on top of the normal promotions. Moreover, when Albert Heijn promotes a brand, all products from that brand are in promotion. On top of that, Albert Heijn uses also a promotion policy in which they not only promote one variant of one product for a specific brand at a time, but all variants of that product for that specific brand at a time.

In order to get some insights about the absolute promotion value over time for the other retailers, another graph was made. Figure 12 shows the development of absolute promotion value over time for the Jumbo, Plus, Hoogvliet and Dirk. The graph shows that the absolute promotion value is more or less the highest for Jumbo. Additionally, the absolute promotion value of Jumbo is also substantially increasing after week 100. Furthermore, from the graph is also notable that Dirk is stable over time and only shows minor increases and decrease. Additionally, the absolute promotion value of Plus increases slowly and the absolute promotion value of

Absolute promotion value

Retailer:

(32)

32 Hoogvliet varies a lot as the value for this retailer decreases substantially up to week 90 after which it increased.

Absolute promotion value without Albert Heijn

Retailer:

(33)

33

4.2 Empirical results

This section explains the different models used for testing the hypotheses. However, before building a model it is important to check if the assumptions are met, therefore these will be discussed. After that the different types of models will be presented along with testing for the appropriate pooling level. Finally, this section ends with a discussion about the model assumptions which are violated in the models.

4.2.1 Multicollinearity

In order to obtain the coefficients for the variables in the model, a linear regression model needs to be created. However, when creating a linear regression model, some assumptions should be satisfied. One of these assumptions is ‘multicollinearity’. Multicollinearity arises when there is a strong resemblance in the evolution of two or more independent variables. This forms a problem because due to these correlations among the independent variables one cannot distinguish which variable is causing the effects in promotion image or share-of-wallet. This leads to misinterpretation of the regression coefficients as the regression model might give unreliable coefficients. In order to detect if multicollinearity is present, the Variance Inflation Factors (VIF) should be calculated. Additionally, also a correlation matrix of the independent variables will help to expose specific correlations between variables.

When estimating a linear regression model, a multicollinearity issue (VIF > 5) is found in promotion intensity of national brands (VIF: 21.80). This variable is positive and significantly correlated with absolute promotion value (COR: 0.90, p < 0.01). Hence, when promotion intensity for national brands increases, absolute promotion value also increases. In order to solve this issue, promotion intensity for national brands and private label brands were summed up which resulted in a lower VIF-statistic. It follows that it is not possible anymore to test H1 and H2, however it is possible to test the overall effect of promotion intensity on promotion image and share-of-wallet.

(34)

34 (Curto & Pinto, 2011). Table 7 gives an overview of the VIF-statistics as they are used in the analysis.

Variable VIF Variable VIF

Avg. Promotion intensity 7.12 Age 1.13

Avg. PD: non-perishable food-products 4.98 Income 1.09 Avg. PD: non-food products 2.94 Household size 1.10 Avg. PD: perishable food-products 4.94 Education level 1.11 Avg. Share of national brands 1.63 Supermarket close by 1.12 Avg. Absolute promotion value 8.37 Supermarket choice 1.05

Enough Shops 1.12

Table 7 - VIF statistics

4.2.2 Pooled regression model

Two pooled regression models were created to analyze the effects of the explanatory variables on promotion image and share-of-wallet. A pooled regression model assumes that both the intercept and coefficients are the same for all retailers. The advantage for this approach is that it allows for the use of more observations in the model (Leeflang, Wieringa, Bijmolt, & Pauwels, 2015).

The results of the pooled models are shown in table 8. Both models are statistically significant (p < 0.01) which means that the included variables improve the fit of both models. The R² for the model predicting promotion image is 0.213 which means that 21.3 percent of the variance in promotion image is explained by the explanatory variables in the model. The adjusted R² is 0.212, which means that when accounting for the number of parameters estimated, 21.2 percent of the variance in promotion image is explained by the exploratory variables in the model. As the difference between the R² and adjusted R² is small (0.001), it can be concluded that no superfluous parameters are present in the model for promotion image.

(35)

35 only the significant coefficients can be interpreted. It is decided to consider p-values lower than 0.05 as significant (Masicampo & Lalande, 2012).

Both the estimate for the main effect of promotion intensity and the estimate for the moderated effect of promotion intensity reveal to have a significant and negative effect on promotion image (Promotion intensity main-effect: β = -2.32, p < 0.05 | Promotion intensity moderated-effect: β = -6.47, p < 0.01). This is remarkable as it is contradicting the expectations. Although the VIF-statistics for promotion intensity and absolute promotion value are still high (VIF > 5), they are not the cause for this negative effect of promotion intensity. The negative effect of promotion intensity can be explained by including the share of national brands in the model. While these variables are not highly correlated with each other (COR: 0.18), removing the share of national brands from the model turns the effect of promotion intensity from negative to positive. However, regarding the hypotheses, H1, H9A and H10A are not supported by this model. Furthermore, while the estimate for the main effect of promotion intensity in the model for predicting promotion image is negative, it is positive in the model predicting SOW (β = 57.18, p < 0.05). Since is in line with the expectations, H2 is supported by this model.

Furthermore, the estimates of promotion depth for both perishable food-products and non-food products reveal to have a positive effect on promotion image (PD non-perishable non- food-products: β = 2.15, p < 0.01 | PD non-food food-products: β = 1.45, p < 0.01). As both effects are in line with the expectations, H3A and H3B are supported by this model. Furthermore, both estimates of share of national brands reveal to have a positive effect on promotion image as well as a positive effect on SOW (Promotion image: β = 2.05, p < 0.01 | SOW: β = 24.00, p < 0.01). Since both effects are in line with the expectations, H6 and H7 are supported by their corresponding model.

(36)

36

Main effects Promotion

Image

(SE) SOW (SE)

Intercept 1.98*** 0.23 38.67*** 5.55

Avg. Promotion intensity -2.32** 1.15 57.18** 27.66

Avg. PD: non-perishable food-products 2.15*** 0.73 6.86 17.64

Avg. PD: non-food products 1.45*** 0.41 5.35 10.03

Avg. PD: perishable food-products -0.96 0.87 -37.36 21.29 Avg. Share of national brands 2.05*** 0.19 24.00*** 4.75 Avg. Absolute promotion value (* 1000) 0.01 0.01 0.11 0.25

Moderating effects of Deal-proneness

Avg. Promotion intensity -6.47*** 1.10 -33.70 26.51

Avg. PD: non-food products 0.31 0.48 10.27 11.48

Avg. PD: perishable food-products -2.29** 1.00 -32.25 24.22

Avg. Share of national brands 0.27 0.20 -8.80 4.93

Avg. Absolute promotion value (* 1000) 0.06*** 0.01 0.44** 0.21

Control variables Age (* 10) 0.04*** 0.01 -1.53*** 0.26 Income 0.01 0.01 0.33 0.18 Household size 0.04*** 0.01 -0.43 0.28 Education level -0.06*** 0.01 0.13 0.24 Supermarket close by 0.25*** 0.01 3.28*** 0.24 Supermarket choice -0.03*** 0.01 -6.59*** 0.16

Enough Shops in the area 0.19*** 0.01 -0.57*** 0.19

Observations 9495 7737

R² 0.213 0.255

Adjusted R² 0.212 0.254

Residual St. Error (df) 1.33 (9475) 28.69 (7717)

F-statistic 135*** 139***

Table 8 - Pooled models Note: ** p<0.05, *** p<0.01

(37)

37 pooling is not allowed for both models and no conclusions can be drawn from the pooled models predicting promotion image and SOW.

4.2.3 Partially pooled regression model

In order to introduce more flexibility into the model a partially pooled regression model is created. A partially pooled model (also known as an Ordinary Least Squares with Dummy Variables-model) assumes the coefficients are the same across retailers, however in contrast to a pooled model, a partially pooled model allows for different intercepts across retailers. This results in a type of model which strikes a suitable balance between flexibility and efficiency (Leeflang et al., 2015).

The results for both OLSDV-models can be found in Appendix A. Both models are statistically significant (p < 0.01) which means that the included variables improve the fit of both models. The R² for the model predicting promotion image is 0.970 which means that 97.0 percent of the variance in promotion image is explained by the explanatory variables in the model. The adjusted R² is 0.970, which means that when accounting for the numbers of parameters estimated, 97.0 percent of the variance in share-of-wallet is explained by the exploratory variables in the model. As the difference between the R² and adjusted R² is very small, it can be concluded that no superfluous parameters are present in the model for promotion image. The R² for the model predicting share-of-wallet is 0.729 which means that 72.9 percent of the variance in share-of-wallet is explained by the explanatory variables in the model. The adjusted R² is 0.729 which means that when accounting for the numbers of parameters estimated, 72.9 percent of the variance in share-of-wallet is explained by the exploratory variables in the model. As the difference between the R² and adjusted R² is small (0.002), it can be concluded that no superfluous parameters are present in the model for share-of-wallet.

(38)

38 4.2.4 Unpooled regression models (unit-by-unit)

In order to test the hypotheses for promotion image and share-of-wallet, ten different unit-by-unit models are created. Five models for predicting promotion image and five models for predicting share-of-wallet. The advantage of using a unit-by-unit model is that both the intercepts as the coefficients are specific to each retailer which provides the model with maximum flexibility. However, a disadvantage of using unit-by-unit models is that it lacks efficiency as the results are valid for a specific retailer, however they are difficult to generalize across retailers. Another disadvantage of unit-by-unit models is that they are using less data than a fully pooled model uses which could lead to less significant estimations (Leeflang et al., 2015).

Albert Heijn

Table 9 shows the results of the unpooled models for predicting promotion image and share-of-wallet for Albert Heijn. Both models are statistically significant (P < 0.01) which means that the includes variables improve the fit of both models. The R² for the model predicting promotion image is 0.217 which means that 21.7 percent of the variance in promotion image is explained by the model. Furthermore, the R² for the model predicting share-of-wallet is 0.240 which means that 24 percent of the variance in share-of-wallet is explained by the model. As mentioned before, when testing the hypotheses only the significant coefficients can be interpreted. It is decided to consider a p-value lower than 0.05 as significant (Masicampo & Lalande, 2012).

(39)

39 the other estimates are not significant, they cannot be interpreted for testing the remaining hypotheses.

Image

(SE) SOW (SE)

Intercept 3.50*** 0.95 27.16 24.45

Avg. Promotion intensity -4.78*** 1.60 42.66 40.77

Avg. PD: non-perishable food-products 0.43 1.63 30.12 42.00

Avg. PD: perishable food-products 0.82 1.70 -61.86 43.56 Avg. Share of national brands 0.01 0.93 13.59 23.90 Avg. Absolute promotion value (* 1000) 0.03 0.02 -0.09 0.52

Avg. Promotion intensity -1.71 1.94 66.81 49.60

Avg. PD: non-perishable food-products -0.65 1.90 -13.14 48.61

Avg. PD: perishable food-products 0.54 1.91 -4.86 48.78 Avg. Share of national brands 0.24 0.44 -27.99** 11.22 Avg. Absolute promotion value (* 1000) 0.02 0.02 0.46 0.62

Control variables Age (* 10) -0.01 0.02 -1.99*** 0.39 Income -0.01 0.01 0.83** 0.27 Household size 0.03 0.02 -0.74 0.42 Education level -0.03** 0.01 1.28*** 0.36 Supermarket close by 0.29*** 0.01 3.60*** 0.38 Supermarket choice -0.03*** 0.01 -6.64*** 0.24

Enough Shops in the area 0.19*** 0.01 -0.28 0.31

R² 0.217 0.240

Adjusted R² 0.213 0.236

F-statistic 58.37*** 58.40***

Table 9 - Unit-by-unit model for Albert Heijn Note: ** p<0.05, *** p<0.01

Jumbo

(40)

40 image is 0.165 which means that 16.5 percent of the variance in promotion image is explained by the model. Furthermore, the R² for the model predicting share-of-wallet is 0.260 which means that 26 percent of the variance in share-of-wallet is explained by the model. Since all estimates of both models for Jumbo are not significant, they cannot be used for hypotheses testing.

Image

(SE) SOW (SE)

Intercept 2.72 1.52 51.24 33.53

Avg. Promotion intensity 1.33 5.15 -59.32 113.40

Avg. PD: non-perishable food-products 1.42 5.38 109.51 118.88

Avg. PD: perishable food-products 0.27 1.83 -37.37 39.76 Avg. Share of national brands 1.02 0.73 7.94 16.03 Avg. Absolute promotion value (* 1000) -0.09 0.25 0.91 5.45

Avg. PD: non-perishable food-products -0.82 3.13 102.69 66.58

Avg. PD: non-food products 0.73 1.11 -1.34 23.79

Avg. PD: perishable food-products 0.35 2.20 -46.74 46.55 Avg. Share of national brands -0.41 0.50 -15.09 10.46 Avg. Absolute promotion value (* 1000) -0.49 0.28 -4.06 5.98

Control variables Age (* 10) 0.04** 0.02 -2.24*** 0.45 Income 0.04*** 0.01 -0.08 0.31 Household size 0.04 0.02 -0.61 0.49 Education level -0.07*** 0.02 -0.83** 0.42 Supermarket close by 0.20*** 0.02 2.40*** 0.42 Supermarket choice -0.05*** 0.01 -7.13*** 0.29

R² 0.165 0.260

Adjusted R² 0.160 0.255

F-statistic 33.11*** 46.24***

(41)

41 Plus

Table 11 shows the results of the unpooled models for predicting promotion image and share-of-wallet for Plus. Both models are statistically significant (P < 0.01) which means that the includes variables improve the fit of both models. The R² for the model predicting promotion image is 0.229 which means that 22.9 percent of the variance in promotion image is explained by the model. Furthermore, the R² for the model predicting share-of-wallet is 0.289 which means that 28.9 percent of the variance in share-of-wallet is explained by the model.

Image

(SE) SOW (SE)

Intercept 1.67 1.75 -30.97 43.92

Avg. Promotion intensity 20.14 13.86 506.85 343.46

Avg. PD: non-food products -0.84 1.92 -55.83 47.48

Avg. PD: perishable food-products 0.63 2.85 32.56 72.27 Avg. Share of national brands 1.05 1.20 57.27** 29.79 Avg. Absolute promotion value (* 1000) -0.94** 0.37 1.45 9.22

Avg. Promotion intensity -3.66 13.20 -266.83 324.79

Avg. PD: non-perishable food-products 1.44 2.74 -3.40 68.65

Avg. PD: non-food products -0.14 1.91 59.59 47.12

Avg. PD: perishable food-products -2.32 3.22 17.13 79.77 Avg. Share of national brands 1.07 1.16 -39.67 28.74 Avg. Absolute promotion value (* 1000) -0.07 0.40 3.51 9.90

Control variables Age (* 10) 0.10*** 0.03 1.21 0.75 Income (* 10) 0.01 0.02 -0.58 0.47 Household size 0.05 0.03 1.19 0.81 Education level -0.08*** 0.03 -0.41 0.63 Supermarket close by 0.22*** 0.02 3.74*** 0.59 Supermarket choice -0.03** 0.02 -5.73*** 0.40

Enough Shops in the area 0.17*** 0.02 -1.48*** 0.45

R² 0.229 0.289

Adjusted R² 0.217 0.274

Residual St. Error (df) 1.24 (1253) 2604 (889)

F-statistic 19.55*** 19.06***

(42)

42 Firstly, the estimate of promotion depth for non-perishable food-products reveals to have a positive effect on promotion image (β = -7.36, p < 0.01). Hence, 1% increase in promotion depth for non-perishable food-products, increases the expected promotion image with 7.36 units. Since this is in line with the expectations, evidence is found in support of H3A. Secondly, absolute promotion value reveals to have a negative effect on promotion image (β = -0.94, p < 0.05). As this is the opposite of what was expected, no evidence is found in support of H7. Lastly, the share of national brands reveals to have a positive effect on SOW (β = 57.27, p < 0.05) which means that 1% increase in the share of national brands, increases the expected SOW with 57.27%. As this is in line with the expectations it can be concluded that evidence is found in support of H6. Since the other estimates are not significant, they cannot be interpreted for testing the remaining hypotheses.

Hoogvliet

Table 12 shows the results of the unpooled models for predicting promotion image and share-of-wallet for Hoogvliet. Both models are statistically significant (P < 0.01) which means that the includes variables improve the fit of both models. The R² for the model predicting promotion image is 0.305 which means that 30.5 percent of the variance in promotion image is explained by the model. Furthermore, the R² for the model predicting share-of-wallet is 0.282 which means that 28.2 percent of the variance in share-of-wallet is explained by the model.

(43)

43

Image

(SE) SOW (SE)

Intercept 0.28 4.20 -109.85 132.91

Avg. PD: non-perishable food-products 1.45 5.26 361.45** 165.53

Avg. PD: perishable food-products -3.12 5.31 -325.31** 166.18 Avg. Share of national brands 4.49 3.90 126.73 123.68 Avg. Absolute promotion value (* 1000) 0.27 0.70 -13.48 22.06

Avg. PD: non-perishable food-products -0.89 6.20 36.49 188.53

Avg. PD: perishable food-products 1.96 5.11 151.29 152.17 Avg. Share of national brands 3.17 2.37 -102.32 79.80 Avg. Absolute promotion value (* 1000) -0.36 0.81 4.61 26.82

Control variables Age (* 10) 0.03 0.05 -0.29 1.49 Income -0.02 0.03 -0.04 1.10 Household size 0.08 0.05 -0.38 1.68 Education level -0.18*** 0.04 -0.75 1.37 Supermarket close by 0.35*** 0.04 2.57 1.34 Supermarket choice 0.03 0.03 -6.04*** 0.83

Enough Shops in the area 0.12*** 0.03 0.13 0.95

R² 0.305 0.282

Adjusted R² 0.263 0.230

F-statistic 7.33*** 4.94***

(44)

44 Dirk

Table 13 shows the results of the unpooled models for predicting promotion image and share-of-wallet for Dirk. Both models are statistically significant (P < 0.01) which means that the includes variables improve the fit of both models. The R² for the model predicting promotion image is 0.276 which means that 27.6 percent of the variance in promotion image is explained by the model. Furthermore, the R² for the model predicting share-of-wallet is 0.273 which means that 27.3 percent of the variance in share-of-wallet is explained by the model.

Image

(SE) SOW (SE)

Intercept 12.62*** 4.36 10.28 113.09

Avg. PD: perishable food-products -2.31 4.70 -109.73 122.01 Avg. Share of national brands -8.26 4.43 54.06 115.49 Avg. Absolute promotion value (* 1000) -3.45 2.51 15.97 65.70

Avg. PD: perishable food-products 7.57 6.45 456.76** 178.45 Avg. Share of national brands -0.90 3.14 -155.93 88.23 Avg. Absolute promotion value (* 1000) 0.85 3.31 -73.43 93.20

Control variables Age (* 10) 0.14*** 0.04 0.89 1.02 Income -0.02 0.03 0.15 0.68 Household size 0.06 0.04 1.09 1.08 Education level -0.10*** 0.03 -0.80 0.88 Supermarket close by 0.29*** 0.03 5.27*** 0.89 Supermarket choice -0.01 0.02 -5.49*** 0.58

R² 0.276 0.273

Adjusted R² 0.254 0.245

F-statistic 12.64*** 9.87***

(45)

45 The estimate of promotion depth for perishable food-products moderated by deal-proneness reveals to have a positive effect on SOW (β = 456.76, p < 0.05). While this is in line with the expectations, it can be concluded that only partial evidence is found in support of H10B as the moderating effects of deal-proneness for perishable and food products are non-significant. Interestingly, when comparing the positive effect of perishable food-products with moderation to the effect of perishable food-products without moderation, it seems that the effect reverses when it is moderated by deal-proneness. However, no conclusions can be made about this due to the fact that the estimate for perishable food-products without moderation is not significant. Since the other estimates in the model are not significant (P < 0.05), they cannot be interpreted for testing the remaining hypotheses.

4.2.5 Model validation

As mentioned before, in order to know if the results are reliable, some assumptions should be satisfied. Section 4.2.1 already discussed the first assumption of multicollinearity; in this section the other assumptions will be discussed.

Firstly, in order to know if the data suffers from autocorrelation, a Durbin Watson-test is performed. The test turns out to be significant for the unpooled models of Plus and Hoogvliet predicting promotion image. This means that for those models the residuals are correlating to each other which results in wrong estimates of variances of effects. Therefore, the estimates for the unpooled models of Plus and Hoogvliet predicting promotion image should be taken with caution as these estimates could be different from the real estimates. Regarding all other models, no autocorrelation is present as the test is non-significant (P > 0.05) for the other models. Secondly, in order to test whether the variance of the regression residuals is constant, a Breusch-Pagan test is performed. Only the unpooled models for Hoogvliet is not significant (P > 0.05) which means that the variance in the residuals is constant for that model. However, for all other models the test is significant (P < 0.05) which means that heteroscedasticity is present. Therefore, the estimates of those models should be taken with caution as these could be different from the real estimates.

(46)

46

4.3 Additional analysis

In order to yield some extra findings, an extra pooled model is created for explaining SOW. Although not mandatory for this study, it could provide interesting insights as it is interesting to see if and how the model changes when promotion image is added to the model explaining SOW.

(47)

47

Main effects Share-of-wallet (SE)

Intercept 33.55*** 5.54

Promotion Image 2.52*** 0.26

Avg. Promotion intensity 63.58** 27.50

Avg. PD: non-perishable food-products -0.05 17.55

Avg. PD: non-food products 1.99 9.98

Avg. PD: perishable food-products -33.05 21.16

Avg. Share of national brands 18.89*** 4.75

Avg. Absolute promotion value (* 1000) 0.074 0.25

Promotion Image -0.39 0.25

Avg. Promotion intensity -9.56 26.59

Avg. PD: non-perishable food-products 15.16 20.89

Avg. PD: non-food products 10.92 11.42

Avg. PD: perishable food-products -24.25 24.04

Avg. Share of national brands -7.86 5.10

Avg. Absolute promotion value (* 1000) 0.21 0.22

Control variables Age (* 10) -1.66*** 0.26 Income 0.32 0.18 Household size -0.51 0.28 Education level 0.27 0.24 Supermarket close by 2.64*** 0.25 Supermarket choice -6.54*** 0.16

Enough Shops in the area -1.04*** 0.20

Observations 7737

R² 0.267

Adjusted R² 0.265

Residual St. Error (df) 28.46 (7715)

F-statistic 134.1***