Marketing mix, sales, and the moderating role of brand type, brand size and product category type

(1)

1

Marketing mix, sales, and the

moderating role of brand type, brand size

and product category type

(2)

(3)

3

Marketing mix, sales, and the

moderating role of brand type, brand size

and product category type

A two-stage analysis

June 20, 2016

MSc Marketing, Intelligence Master Thesis

University of Groningen Faculty of Economics and Business

Department of Marketing PO Box 800, 9700 AV Groningen

Supervisors:

First Supervisor: Dr. ir. M.J. Gijsenberg Second Supervisor: Prof. dr. T.H.A. Bijmolt

Author: Name: Wilco de Boer Student number: S2813041

Address: JC Kapteynlaan 21C, 9714CM Groningen Phone: +31611922928

(4)

(5)

5

MANAGEMENT SUMMARY

This study examines the influence of brand and product category characteristics on the relation of price, advertising and distribution on volume of sales. This is the first study that combines these moderating variables in the relation of marketing mix effectiveness. The construct brand characteristic is split into two different variables. The first one is defined as brand type, and operationalized as national brands and private labels. The other variable is brand size, which is operationalized as the market share of a brand. The second moderating variable (i.e., product category characteristics) is defined by the hedonic or utilitarian characteristics of a product. Furthermore, the direct effects of price, advertising, and distribution on volume of sales are measured.

The dataset that is used in this study is a dataset of the Fast Moving Consumer Goods (FMCG) market. The dataset comprises 15 product categories, including 60 brands in which several variables are measured for 208 consecutive weeks, from 1994 till 1998. The method that was used to examine the data embodies a two-stage analysis. In the first stage, 60 ordinary least squares (OLS) regressions were carried out, in which the direct effects of price, advertising and distribution on sales are examined. This means that direct effects were estimated for all 60 brands separately. The second stage of the analysis consisted of three weighted least squares (WLS) regressions, in which the inverse of the standard deviations of the OLS-regressions functions as the WLS-weight. The dependent variables of the three WLS-regressions were the OLS-estimates of the first stage analysis.

(6)

6

These results also have managerial implications. Managers of national brands can use temporary price reductions to generate sales. Consumers that find national brands usually too expensive are then willing to buy those brands, because of the lowered price. In addition, earlier scientific research showed that price reductions of national brands do harm the sales of private labels. But these temporary price reductions for national brands need to be used with caution, because deploying too many price reductions in a short period of time can cause a lowered willingness to pay and a lowered reference price.

This study has various limitations. First, the data that was used was collected between 1994 and 1998. The private label market at that point in time was not yet very developed. This means that for some private labels there is a lack of data regarding price, advertising and distribution. Second, the traditional marketing mix consists of four instruments. In this study there was no data available about the product line length, so only three traditional marketing mix instruments are included in this study. It could be that the results change when the product line length variable is included. Lastly, there is only information about the advertising expenditures. It could be interesting to study another dimension of advertising in this context, namely the content of the advertisements. Based on these limitations, further research on marketing mix effectiveness is warranted.

(7)

7

PREFACE

Two years ago I started with the pre-MSc Marketing program. This thesis is the result of a process that had many ups and downs, but I never regretted the fact that I signed up for this master program after I finished my Bachelor Communication at the Hanzehogeschool. This thesis is not only the end of this master program, but also the end has come for me being a student. I look back on two great years on the Faculty of Economics and Business. I would not have been able to write this thesis, without the help of several people.

First, I would like to thank dr. ir. Maarten Gijsenberg for his feedback and great help during the process of writing this thesis. Second, I want to say thanks to my friends and family, and in particular my parents. Without them it would have been impossible to study at the University of Groningen. Finally, a special thank to my girlfriend Suzanne for her endless support and motivation during the entire academic curriculum.

(8)

8

1. INTRODUCTION

A recent press release of Market Research Company AC Nielsen said that the volume of sales of the European Fast Moving Consumer Goods (FMCG) market increased for sixth straight quarters (Beston, 2015). The FMCG-market is a large and dynamic market all over the world. It is also a well-known subject of interest in marketing science. Scientific research into the marketing mix has a long history, and many academics and scholars have investigated marketing mix effectiveness in the FMCG market (e.g., Ataman, Van Heerde, & Mela, 2010; Dekimpe & Hanssens, 1995; Leeflang & Olivier, 1985; Van Heerde, Leeflang, & Wittink, 2000). There is an extensive use of moderating variables in the existing literature on marketing mix effectiveness. This thesis is an extension on a number of those moderating variables. In this thesis I investigate three moderating constructs in the relation of three traditional marketing mix instruments on sales.

(10)

10

further research of previous papers (e.g., Arnold & Reynolds, 2012; Chandon et al., 2000). To the authors knowledge this is the first paper that empirically test the moderating role of hedonic and utilitarian product category characteristics in the relation between marketing actions and sales.

Apart from the fact that there are hedonic and utilitarian product categories, another factor that affects the buying process of the consumer is brand preference. A widely studied topic in brand types is private labels. According to the private label report of Nielsen (2014) the European market is the most developed market in private labels in the world. Switzerland and the United Kingdom have the highest private-label dollar share of the world with respectively 45% and 41% (Nielsen, 2014). Brand type is also a popular topic of research for academics in marketing science. Many studies have focused on private labels in combination with attitude or loyalty (e.g., Ailawadi, Pauwels, & Steenkamp, 2008; Koschate-Fischer, Cramer, & Hoyer, 2014). However, there is no research that examines the moderating effect of private labels and national brands in the relation between marketing actions and sales in an FMCG-market. Koschate-Fischer et al. (2014) and Ailawadi et al. (2008) have made some suggestions for further research, particularly to combine the hedonic and utilitarian aspects and private labels. This study adheres to those suggestions by including both constructs as moderator in this study. By including these constructs, marketing managers and the field of marketing science get an extensive overview of possible moderating variables in the relation of price, advertising, and distribution on sales.

(11)

11

brands with a high market share have an advantage in terms of marketing mix effectiveness, because they are market leader. While also taking product category characteristics and brand type into account, this thesis will test this assumption.

Derived from the earlier mentioned suggestions for further research from other authors, it is evident that the combination of product category type, brand type, and brand size could be an interesting research area. In a paper of Ailawadi and Keller (2004) three directions of research priorities are highlighted. It is noticeable that the three moderators in this thesis are mentioned a lot through this article as research priorities. Derived from this paper and the above-mentioned papers, it can be stated that it is justified and relevant to conduct a study on the impact of these moderators in the relationship of the marketing mix on sales. This research contributes to the academic literature in one major way. It provides an overview of the effect of three matching moderating variables in the relation between marketing mix and sales. These three variables (brand size, national brands/private labels, and hedonic/utilitarian product categories) are moderators that complement each other and therefore contribute to an extension of the already existing literature regarding marketing effectiveness dynamics. To the author’s knowledge this is the first time that this specific combination of moderators is investigated in the relation of price, promotion, and place with sales.

This research may also be of value to managers for multiple reasons. The author will, inter alia, be able to (i) draw conclusions whether private labels or national brands depend stronger on the relation between price, promotion and place and sales. Besides that, (ii) this thesis will provide greater clarity on which product category type is most responsive in the relation of price, promotion and place on sales. As stated earlier in this introduction, product category type distinguishes between hedonic and utilitarian characteristics. Finally, (iii) there will be a clarification for academics and marketing managers if it is true that brand size positively moderates the influence of price, advertising and distribution on sales. These results could help managers to better allocate their marketing budget and other resources.

In conclusion, the three moderators are in line with each other. This study is academically relevant and useful for marketers in practice. The problem statement of this thesis can be summarized in the following research question:

(12)

12

(13)

13

2. THEORETICAL FRAMEWORK AND HYPOTHESES

2.1 Marketing mix effectiveness

In this study the focus for the direct effects is on the effectiveness of the traditional marketing mix for different product categories and brands. In the analyses I investigate the differences in effectiveness for three well-known marketing mix instruments: price, promotion, and place (McCarthy, 1960).

2.1.1 Price

Price is an important instrument for retailers to generate sales. Sprott, Manning and Miyazaki (2003) found that consumers use prices of frequently purchased products to form an opinion about retail prices and store-price image. But forming an opinion, creating an image, and purchasing products is not only a rational process. Prices (in particular, price setting) have a psychological effect on the human brain. A good example of this can be found in an interesting study of Schindler and Kibarian (1996). These authors empirically examined the effect of so called 99-ending prices (e.g. €14,99 instead of €15,00). They found that a 99-ending actually led to an increase of customers purchasing the product. In addition to this, it is plausible to argue that higher prices lead to lower sales and lower prices lead to higher sales. A good measure for this phenomenon is price elasticity. According to the meta-analysis of Bijmolt, van Heerde and Pieters (2005) the average price elasticity is -2.62. In their article, Bijmolt, van Heerde and Pieters (2005) accounted for the different characteristics of brands, categories, and economic circumstances. These conditions make the price elasticity they found generalizable for the dataset of FMCG in this thesis. Also Hanssens et al. (2014) found empirical evidence for the direct effect of price on sales. They found support for the claim that lowering the price will cause an increase in volume of sales. A price increase will, in turn, cause a decrease in volume of sales. Thus lowering prices can generate sales and increase revenue performance (Hanssens et al., 2014). This result does not only mean that permanent price cuts have a positive effect on sales, but also that price promotions / discounts are part of the marketing mix and could have a positive effect on sales. Ataman, van Heerde and Mela (2010) conducted research regarding the effect between marketing actions and sales and found that lowering prices generates more sales in both short-term and the long-term.

(14)

14

H1a: A decrease of price leads to an increase of sales. 2.1.2 Promotion

Advertising is an important instrument for manufacturers, because advertisements can build positive customer-based brand equity and brand knowledge (Keller, 1993). According to Keller (1993), “customer-based brand equity occurs when the consumer is familiar with the brand and hold some favorable, strong and unique brand associations in memory” (p.1). A measure that is often used to examine the effectiveness in advertising is advertising elasticity, which can be described as “the percentage increase in sales or market share for a 1% increase in advertising” (Sethuraman, Tellis, & Briesch, 2011, p. 457). In the meta-analysis of Sethuraman, Telis and Briesch (2011), short-term elasticity was .12, whereas long-term advertising elasticity was .24. Up until now, many other studies have shown that advertising expenditures are positively related with an increase in sales (e.g., Ataman et al., 2010; Dekimpe & Hanssens, 1999; Hanssens et al., 2014). Ataman et al. (2010) showed that advertising indeed has a positive effect on sales, although it is not as effective as distribution. Hanssens et al. (2014) found that advertising significantly predicted sales in three out of four categories in the dataset that was used in that particular study.

Considering these studies, it seems plausible to assume that advertising has a positive effect on sales. Therefore, the following hypothesis is formulated:

H1b: Advertising positively affects sales. 2.1.3 Place

According to McCarthy (1960) the marketing instrument place, refers to the distribution coverage of a particular product. Ataman, van Heerde and Mela (2010) found that distribution is positively related to sales. This means that when the distribution coverage increases, the volume of sales will increase as well. In fact, they found that distribution is the second most influencing marketing mix instrument in terms of short- and long-term sales elasticity. In their study concerning marketing mix effectiveness, Yoo, Donthu and Lee (2000) found that high distribution intensity is a significant predictor of brand equity. Brand equity, in turn, positively correlates with brand awareness and sales (Huang & Sarigöllü, 2012).

(15)

15

consumers can easily find the particular brand or product. The brand or product is simply more visible to customers, because it is located in more stores. This increased visibility could lead to an increase of sales (Ataman et al., 2010; Bronnenberg, Mahajan, & Vanhonacker, 2000).

Overall, there seems to be evidence that indicates a positive relationship between distribution and sales. Hence, the following hypothesis is formulated:

H1c: Large distribution coverage positively affects sales.

As mentioned earlier, Ataman et al., (2010) demonstrated that distribution is the strongest predictor of the marketing actions that are used in this thesis. Therefore, the following hypothesis is formulated:

H1d: Distribution is the strongest predictor of sales in comparison with advertising and price.

2.2 Product category characteristics: hedonic and utilitarian

(16)

16

feeling of satisfaction is simply more important than the money they spent on that product. By this reasoning, it is plausible to argue that price has a stronger influence on sales in product categories with utilitarian characteristics than product categories with hedonic characteristics.

This expected difference in reaction also relates to psychological processes that take place in the consumer’s minds. It can be said that buyers of hedonic products mainly act on an emotional basis, whereas buyers of utilitarian products are more rational during the buying process (Drolet, Williams, & Lau-Gesk, 2007). Therefore, the content of the advertisement plays a key role in determining to which type of audience it will appeal (Iyer, Soberman, & Villas-Boas, 2005).

The content of advertising hence is an important factor to take into account when persuading consumers to buy a certain product. According to the Elaboration Likelihood Model of Petty and Cacioppo (1984), there are two ways to persuade a consumer, namely through the central route (i.a. arguments, active thinking) and the peripheral route (i.a. simple cues). In the present FMCG-dataset there is only data of advertising expenditures. This means that the content of advertising cannot be taken into account in the analysis. Still, it is important to state that there are different approaches possible in persuading customers. In this study the focus is on the hedonic and utilitarian characteristics of product categories.

(17)

17

Reynolds, 2012), one could argue that advertising has a stronger influence on sales for hedonic product categories than for utilitarian product categories.

When a product is advertised, it also needs to be in store, otherwise there are obviously no sales. A reason why distribution could have a positive effect on sales is that the product categories are easier to find in stores because of the distribution breadth (Ataman et al., 2010; Bronnenberg et al., 2000). But why would there be differences in effects with regards to the differences between hedonic and utilitarian product category characteristics? Research found that consumers are willing to browse different stores and participate in a permanent search for products in which they are highly involved, to increase levels of fun and pleasure (e.g., Bloch, Sherrell, & Ridgway, 1986; Jones, Reynolds, Weun, & Beatty, 2003). In another study, Okada (2005) reported that consumers are wiling to spend more time on searching for hedonic goods than for utilitarian goods. This could be because of the fact that they are more motivated to find products that give them pleasure and happiness. Okada (2005) stated in her article: “Most people are more excited about the prospect of fun than the prospect of practicality” (p. 52). Thus, as consumers are willing to search longer and more intensely for hedonic products, and at the same time the distribution coverage increases; the consumers that search for hedonic products will benefit more from this increase. With this possible theory in mind, a logical next step is to assume that distribution would have a stronger influence on hedonic products than utilitarian products. Another line of reasoning starts with another proposition: hedonic products are more often purchased impulsively than utilitarian products (Fennis & Stroebe, 2015; Jones et al., 2003; Rook, 1987; Yim, Yoo, Sauer, & Seo, 2014). It may well be that for utilitarian products the threshold for saturation of needs is lower than it is for hedonic products. When the distribution coverage increases for both hedonic and utilitarian products, consumers are confronted with more products in both categories. But according to the aforementioned authors, hedonic products are more likely to be bought on impulse. Thus, the expectation is that distribution has a stronger influence on sales for hedonic products than for utilitarian products.

Considering all of these studies, it seems that there is reason to assume that price, advertising, and distribution differ in effectiveness regarding to hedonic and utilitarian product categories. Therefore, the following three hypotheses are formulated:

(18)

18

H2b: Advertising has a stronger influence on sales for hedonic product categories than for utilitarian product categories.

H2c: Distribution has a stronger influence on sales for hedonic product categories than for utilitarian product categories.

2.3 Brand characteristics: brand size

The general assumption in the field of marketing is that brands with a high market share have beneficial results in terms of marketing mix effectiveness, as compared to brands with a smaller market share. So the question is whether the three marketing instruments included in this study (price, advertising, and distribution) have more effect on sales for large brands or for smaller brands.

(19)

19

Major brands have a benefit regarding visibility (Bloom & Kotler, 1975). Through their high distribution coverage these brands are more visible for customers than smaller brands. It could be the case that because of the higher visibility and the products on eye-level aspect, consumers have the tendency to trust the major brand more than smaller brands, and are therefore willing to pay a higher price for the major brand.

In conclusion, there seems to be reasonable convincing evidence to claim that large brands have more influence on the relation between marketing and sales than smaller brands. There is no scientific evidence for the argument that marketing actions have more effect when a brand has a low market share. Therefore, the following three hypotheses have been derived:

H3a: Price has a stronger influence on sales for large brands than for small brands.

H3b: Advertising has a stronger influence on sales for large brands than for small brands.

H3c: Distribution has a stronger influence on sales for large brands than for small brands.

2.4 Brand characteristics: private labels and national brands

(20)

20

products than national brands. Thus, it is clear that nowadays there is not that much of a difference in quality between private labels and national brands, because there is an extension of the product line length of private labels. However, the dataset that will be used in this study comprises data that is collected from 1994 till 1998. In that period of time, the private label products that were offered were primarily distinctive due to their low prices in comparison with national brands. Therefore, in this thesis, private labels will be approached in accordance with the paper of Steenkamp, van Heerde & Geyskens (2010), which states that there could be a perceived gap in quality between private labels and national brands. The low price of private labels could be the cause of this gap.

It may be important for private label owners to reduce the perceived quality gap in order to generate sales. This can be achieved by the use of marketing tools. However, because of the differences in brand type, it is possible that brand type also differs in impact on the relation between marketing and sales. In other words, private labels and national brands may differ in strength of influence on the relations of price, place, and promotion with sales. As mentioned earlier, in general the average price of private label products are lower than the substitutes of national brands.

(21)

21

Another argument that strengthens this expectation is obtained from the meta-analysis of Sethuraman (1995). He examined the effect of discount of national brands on the sales of private labels and vice versa. The outcomes were interesting, indicating that large national brands can lower private label sales through price cuts, but the same national brands are not affected in sales when private labels are in discount. Overall, there seems to be enough scientific evidence to assume that price has a stronger impact on sales for national brands than for private labels.

It is well known that advertising can increase brand awareness and knowledge about products. Thus advertising could be of influence in this case. This presumption is in line with the study of Steenkamp et al. (2010) who found evidence that advertising investments can decrease the perceived quality gap between private labels and national brands. An interesting notion of this author is that manufacturers of national brands invest more in advertising than the retailers that own private labels, and therefore the perceived quality gap is not becoming smaller. This may be because of the fact that retailers aim at low costs and high margins for their private label products.

Another reason why national brands would want to invest more in advertising is because it can be used as an approach to control the impact of private labels (Corstjens & Lal, 2000). Hence, it may well be that because of the fact that national brands are investing more in advertising, those products are also better known than private label products, with the consequence that national brands are selling more than private labels.

(22)

22

Thus, national brands could have more loyal customers and higher brand awareness because they advertise more than private labels. This also causes an increase in the perceived advertising expenditures that customers have of a national brand. Sloot, Verhoef and Franses (2005) found that consumers are more willing to buy high-equity brands than low-equity brands. Moreover, they claim that low-equity brands often consist of private labels.

In conclusion, national brands have an advantage in multiple facets of retailing. Their market share is higher and customers are more aware of their products. Together, these studies are the underlying evidence to assume that advertising has a stronger influence on sales for national brands than for private labels.

A large and obvious difference between private labels and national brands is distribution coverage. Whereas retailers can only allocate shelf space in their own store(s), large national brands with a high market share are available in a lot of different FMCG retailing stores (Narasimhan & Wilcox, 1998). Added to this, it is certain that there are simply more national brands present in stores than private labels. The distribution of these national brands is widely spread out through multiple retailers across the entire country, whereas private labels are less distributed. Off course there are private labels fore sale in almost every retailer store, but there is not one specific private label brand that is for sale in every store in the entire country. In other words, a consumer can buy his favorite national brand in every grocery store, but he cannot buy his favorite private brand in every store. This statement is supported by results from a previously mentioned study showing that private labels are often low-equity brands that have a lower distribution level than high-equity national brands (Sloot et al., 2005). It is not unthinkable that due to this distribution coverage, national brands have an advantage in generating sales. In addition to this, as stated earlier, retailers need to offer those national brands in order to make sure that customers are willing to go to their store (Webster, 2000). From that perspective, manufacturers have a strong position in terms of distribution, because retailers are dependent of their brands. Because of the difference in distribution coverage and the dependence of retailers on manufacturers there is sufficient reason to assume that national brands have a stronger influence on the relation of distribution on sales than private labels.

(23)

23

H4a: Price has a stronger influence on sales for national brands than for private labels.

H4b: Advertising has a stronger influence on sales for national brands than for private labels.

H4c: Distribution has a stronger influence on sales for national brands than for private labels.

2.5 Conceptual model

In figure 1 the conceptual model of the current study is shown. The marketing mix in this thesis consists of price, promotion, and place. H1 represents the direct effects of the three marketing variables on volume of sales. As indicated in the literature review, H2, H3 and H4 all contain three hypotheses regarding moderating effects between price, advertising, and distribution on sales.

H2 is the second hypothesized effect, which is labeled as product category characteristics. This variable is split up in hedonic and utilitarian product categories. As shown in section 2.2 this moderator will be tested in three hypotheses. The objective is to examine whether price, advertising, and distribution have a stronger impact on sales for hedonic or for utilitarian product categories.

H3 and H4 are together labeled as brand characteristics. This construct is split up into two different moderating variables, namely brand size and private labels / national brands. H3 represents the third hypothesized effect; brand size. Brand size is a continuous variable that is measured on brand level. The result of the analysis will provide insights whether price, advertising, and distribution have a stronger impact on sales for large or for small brands.

H4 represents the fourth and last hypothesized effect. This construct is measured at brand level and distinguishes between private labels and national brands. The objective is to examine if private labels or national brands have a stronger influence between price, advertising, and distribution on sales.

(24)

24

Note: This conceptual model will be tested for the FMCG-market by means of a dataset that contains 15 product categories and 60 brands. For an overview of these product categories and brands, see Appendix A.

H1

H3 H4

Figure 1 Conceptual model

H 1

Brand characteristics:

-‐ Private labels / National brands

-‐ Brand size (Brand level) Marketing Mix: -‐ Price -‐ Place (Distribution) -‐ Promotion (Advertising) Sales Product category characteristics:

-‐ Hedonic / utilitarian (Category level)

(25)

25

3. DATA AND METHODOLOGY

3.1 Data description

In order to answer the research question and to test the hypotheses, I use a dataset of the Dutch FMCG-market. This dataset contains weekly data in the period of 1994 to 1998. This means that there are data from 208 consecutive weeks. There are 57 different product categories; each category contains data about a number of brands, including private labels. The categories are classified according to the IRI classification. Only categories with sufficient activity of private labels will be included in the analysis. For the analysis, the dataset will be narrowed down to 15 product categories. Each of those 15 categories contains the top 3 brands and 1 private label. This top 3 of brands is composed of the three brands that have the highest market share in the first week that the data was collected. The hedonic or utilitarian natures of the product categories are derived from a different data set. This dataset is from the FMCG-market of the UK and is collected from 2002 to 2008. The 15 chosen product categories from the Dutch dataset match with the product categories from the UK-dataset. Thus, only a small subset from the UK-dataset will be used to measure the moderating role of hedonic / utilitarian product categories. Following the template of Gijsenberg (2014), I divided the categories in product classes. In table 1 an overview of the product classes and categories is presented. For the full list of categories and brands, see Appendix A.

3.1.1 Brand market share

To get a better understanding of the brands that are included in the analysis, table 2 presents a number of statistics regarding the product classes and brands. The percentage represents the average market share of the top 3 brands and private label

Product class Examples product categories Example brands

Dairy products -‐ Milk

-‐ Yoghurt drinks & juices -‐ Campina -‐ Coberco

Snacks -‐ Crisps _{-‐ Instant hot snacks} -‐ Bahlsen _{-‐ Aviko}

Sauces & spices -‐ Mustard

-‐ Cooking sauces and condiments -‐ Knorr -‐ Lowensenf

Supper -‐ Rice and savoury pasta

-‐ Frozen Fish -‐ Lassie Toverrijst -‐ Ocean Catch

Non-Food -‐ Razor blades

-‐ Machine wash products -‐ Sun -‐ Gilette

(26)

26

of that particular product class. Also the average market share of the first week of the top 3 brands is calculated. Most private labels had no or little market share in the first week of the dataset. Therefore the market share for the top 3 brands and private labels is presented separately. The number in between the parentheses represents the standard deviation. It is noticeable that non-food scores relatively low. The variability of the brands in terms of market share is quite divergent. Some brands have a market share of +50% whereas other brands have only 8% market share. Indications for this are the high standard deviations. These high standard errors mean that the data, with regards to market share, is dispersed (Babbie, 2016). In other words, the market shares of the brands are far from similar.

3.1.2 Advertising and distribution

The marketing mix as presented in this study also includes advertising and distribution. To gain some additional insights in how the product classes deploy these marketing mix instruments, a graph is presented to visualize the relation between

Product Class Market share (top 3 brands) Market share (Private label)

Dairy products 21,9% (23.1) 0.002% (0,003) Snacks 19,6% (17.4) X

Sauces & Spices 24,9%

(16.3) 0.04% (0.07) Supper 24,4% (18) X Non-food 12,9% (9.9) X

Table 2 Product classes and market share (X=no private label market share in first week)

Figure 2 Advertising and distribution

7524,06 11177,26 22715,91 1812,04 6110,62 77,6 86,6 63,6 67,7 77,9 0 10 20 30 40 50 60 70 80 90 100 0 5000 10000 15000 20000 25000

Dairy Products Snacks Non-Food Sauces & Spices Supper

(27)

27

advertising and distribution (figure 2). The black bars indicate the average spending on advertising for the four brands in the product categories. The grey line indicates the average distribution coverage for the top 3 brands in the product categories. The private label brands are not included in these percentages because the distribution coverage of the private labels were relatively low. It is clear that the product class non-food has the highest average advertising spending of all product classes. However, this product class also has the lowest average distribution coverage, namely 63,6%. Snacks have the highest distribution coverage with 86,6%. Sauces & Spices is not a very active product class in terms of advertising. They have the lowest average spending on advertising, namely €1812,04. The analysis of the moderating variables and other statistics will be discussed in the results chapter of this thesis.

3.1.3 Volume of Sales

Another important variable is off course the variable sales. Figure 3 presents the progress of sales of the five product classes. For an overview of these product classes, see table 1. There are five data points that are used for this graph, week 1, 53, 105, 157, 208. This means that the volume of sales is measured on a yearly basis. The variable volume of sales is expressed in liters, kilo’s etc., thus in order to compare the trend of sales through time, the average sales is expressed in an index, with week 1 as the base (100). This means that the volume of sales of every year is compared to the

85 90 95 100 105 110 1 53 105 157 208 V o lu m e o f S al es ( In d ex ) Weeknumbers

Dairy Products Snacks Non-Food Sauces & Spices Supper

(28)

28

sales of year one. This average sales-score is based on the top 3 brands and private labels of the product categories in that product class. As one can see in the figure 3, the product classes Sauces & Spices, Supper and Dairy products are relatively stable over time. The most notable aspect of this figure is the decrease in sales, compared with year one, of Snacks and Non-Food. In the first and second year the sales were increasing, but the index shows that year 3 was a bad year for these two product classes. Almost all the brands in these product classes show a decrease in sales in this period. Despite this strong decrease in sales, it is not yet that extreme that it can be attributed to a certain cause, such as an economic regression. This is useful information for the upcoming analysis, because this means that for this variable no corrections are required.

3.2 Operationalizing variables 3.2.1 Marketing mix and sales

In this thesis there are three independent variables, three moderating variables, and one independent variable. The independent variables are the marketing mix instruments: price, promotion, and place, also known as price, advertising, and distribution. These three variables are measured on ratio scale, which is a numeric scale.

Price is defined as the average price per volume of sales. This is the value of sales divided by the volume of sales. Price is measured in euros.

Advertising is defined as the advertising expenditures per week, per brand. These expenditures are measured in euros.

Distribution is defined as the ratio of the product category sales where the product is available and the sales of the product category in all stores.

Sales is measured on a ratio scale and is defined as ‘Volume of Sales’. That means that the sales is expressed in liters, kilo’s, et cetera.

3.2.2 Moderating variables

(29)

29

National brands / private labels is a variable with two levels. This variable is measured on brand level. Hedonic / utilitarian product characteristics is also a dichotomous variable. Brand size will be operationalized as the market share of a brand in a product category. Market share is measured as the percentage share of a brand of the total sales in that particular product category. Table 3 shows an overview of how the variables are operationalized.

Variables How operationalized?

Independent variables: -‐ Price

-‐ Advertising -‐ Distribution

Price per volume of sales (in €)

Advertising expenditures per week, per brand (in €) Ratio: product category sales where available / sales of all categories (in %)

Dependent variable:

-‐ Sales Defined as volume in sales (e.g. kilo, liter)

Moderating variables: -‐ Brand size

-‐ Private labels / national brands -‐ Utilitarian / hedonic

Defined as market share (in %) (brand level)

Private labels (0) vs. national brands (1) (brand level) Hedonic (0) vs. Utilitarian (1) (Category level)

3.3 Plan of analysis

The method of analysis will be a time series regression. To overcome the issue of the multi-level moderators, I will use a two-stage regression method. This method of analysis is partly derived from previous research (Nijs, Dekimpe, Steenkamp, & Hanssens, 2001; Steenkamp, Nijs, Hanssens, & Dekimpe, 2005) .

The first stage consists of the estimation of a time series model to examine the direct effects of price, advertising and distribution coverage on sales. For this first stage analysis, Ordinary Least Squares (OLS) will be used. Effects are estimated for all brands separately. This means that the first stage analysis consists of 60 OLS-regressions. To reject or accept the hypotheses regarding the direct effects, the added Z-method will be applied (Gijsenberg, 2014; Rosenthal, 1991). In this method the estimate is divided by the standard error of that estimate, which results in the t-statistic. With this t-statistic the p-value can be determined. Via the inverse standard normal distribution, the Z-score of this p-value is calculated. When dividing the sum of the z-scores by the square root of the number of brands, the overall Z-score is calculated. This Z-score is then used to determine the overall p-value of the independent variables. This p-value is decisive in rejecting or accepting hypothesis 1a, 1b and 1c. The long-term effects of advertising will be determined by applying the

(30)

30

delta rule and the added Z-method. I will elaborate on this analysis in the results section.

The 60 OLS-regressions of the direct effects give estimates for the βs and the standard deviations for the effects of price, advertising, and distribution on sales. These estimates will be the basis for the second stage of the analysis.

The first-stage effects (β) will function as the dependent variables in the second stage of the analysis. The three moderating variables will function as independent variables. In other words, the difference with a moderation in a linear regression is that there are no interaction effects in the model specifications. The difference between the first- and second stage analysis is that in the first stage OLS is applied and in the second stage Weighted Least Squares (WLS) is applied. In WLS the independent variables and dependent variable are weighted by the inverse of the standard error (Nijs et al., 2001). WLS is well applicable to this specific model, because it is a remedy for heteroscedasticity and ensures that the estimates of the second stage analysis are unbiased (Leeflang, Wieringa, Bijmolt, & Pauwels, 2015; Steenkamp et al., 2005) .

3.3.1 Error term assumptions

According to Leeflang et al. (2015) there are five important error term assumptions that need to be checked. First, the mean needs to be zero. By using the Ramsey-RESET test this violation can be detected (Ramsey, 1969). A consequence of a nonzero mean is that the parameters can be biased. A possible cause of a nonzero mean is that the model does not have the correct functional form (Leeflang et al., 2015). When confronted with this violation, an appropriate remedy will be applied.

Second is the assumption of heteroscedasticity, in this case a set of residuals having a different variance than another set of residuals (Leeflang et al., 2015). The Levene’s Homogeneity of Variance Test can detect this violation. If this assumption is violated the parameters could be inefficient and cannot be trusted anymore.

Third, the error term needs to be normally distributed. The Kolmogorov-Smirnov and the Shapiro Wilk test can detect violation of this assumption (Leeflang et al., 2015). If the assumption of normal distribution is not met, a bootstrap procedure will be incorporated in the analyses.

(31)

31

turns out that there are violations of the error term assumptions, actions will be taken in order to obtain the most efficient estimates as possible.

The fifth violation is serial correlation. In the first stage analysis a lagged dependent variable will be included in order to account for serial correlation (Leeflang et al., 2015). The second stage of this analysis will be a WLS-regression. This regression will be conducted with cross-sectional data. Hence, autocorrelation is not a problem in cross-sectional data. So this assumption will not going to be tested.

3.4 Model specification

The first model is a unit-by-unit model since all the brands are analyzed separately. Model 1 is a partial adjustment model because of the lagged sales variable (λ) that is included (Leeflang et al., 2015). This variable takes into account how the market was in the week before the present week. As mentioned in the previous section, including this variable is a remedy for autocorrelation (Leeflang et al., 2015).

Models 2a, 2b, 2c are partially pooled models, because the hedonic / utilitarian product characteristics are measured on product category level. So, each brand within a category has the same characteristics in terms of hedonic or utilitarian nature. As indicated earlier in this thesis, the dependent variables of models 2a, 2b and 2c are the estimates of the betas (β) of model 1.

To be able to interpret the parameters of model 1, the dependent and the independent variables are all transformed to the natural logarithm.

(1)

S

_!"

= α

_!

+ β

_!"

PR

_!"

+ β

_!"

ADS

_!"

+ β

_!"

DIST

_!"

+ λS

_!"!!

+ ε

_!"

(2a)

β

_!"

= φ + ψ

_!

HEUT

_!

+ ψ

_!

NBPL

_!

+ ψ

_!

BS

_!

+ ε

_!

(2b)

β

_!"

= φ + ψ

_!

HEUT

_!

+ ψ

_!

NBPL

_!

+ ψ

_!

BS

_!

+ ε

_!

(2c)

β

_!"

= φ + ψ

_!

HEUT

_!

+ ψ

_!

NBPL

_!

+ ψBS

_!

+ ε

_!

Where:

Sit = Volume of Sales brand i (i=1….60) in time t (t=1….208)

α! = Intercept brand i (i=1….60)

(32)

32 β2iADSit

= Advertising expenditures for brand i (i=1….60) in time t (t=1….208)

β3iDISTRit = Distribution for brand i (i=1….60) in time t (t=1….208)

λS!"!! = Lagged sales (Retention rate) for brand i (i=1….60)

ψ1i = Effect hedonic / utilitarian product category for brand i (i=1….60)

ψ2i = Effect National Brands / Private Labels for brand i (i=1….60)

ψ3i = Effect Brand size for brand i (i=1….60)

ϕ Intercept

β!" = Parameter estimate of β1 for brand i (i=1….60)

β_!" = Parameter estimate of β2 for brand i (i=1….60)

β_!" = Parameter estimate of β3 for brand i (i=1….60)

(33)

33

4. RESULTS

The first model contained 60 linear regressions (OLS): one for each brand. For the independent variable advertising 27 of the 60 brands had no data regarding advertising expenditures. On average, in 30.45 weeks (SD = 40.66) of the 208 weeks, there was money spent on advertising. This means that the results regarding advertising expenditures need to be interpreted with care. Model 2 contained three linear regressions (WLS), where the inverse of the standard error of the independent variable of model 1 was used as WLS weight. For a complete overview of the results of model 1 see appendix B.

4.1 Direct effects

Model fit was assessed by examining the percentage of explained variance (R2_{) of}

the dependent variable. Figure 4 visualizes the distribution of the R2_{s of all}

OLS-regressions of the first model. R2 _{ranged from 0.071 to 0.993, with an average of}

0.667. This means that, on average, 66,7% of the variance of the dependent variable sales is explained by the independent variables price, advertising, distribution, and lagged sales. According to Malhotra (2010) a R2_{of 0.6 is acceptable, so the model fit}

of these regression is also acceptable. In addition, all the 60 models were statistically significant (p<0.01). The conclusion therefore is that the model fit of the first model is sufficient.

As said earlier, model 1 consisted of 60 regressions. This resulted in 240 parameters. To give an indication of the variability of the estimates and the extent to which the estimates were significant, a shortened overview is provided in table 4. For all the estimates (β) of all brands, see appendix B.

(34)

34 4.1.1 Results estimates direct effects

Price. In figure 5 the histograms of the estimates of the independent variables are shown. The average price elasticity of this dataset is -1.74. This is less strong than the

1_{Median β’s:} _Price _{= -1.69}

Advertising = 0.003 Distribution = 0.9 Lagged sales = 0.31

Table 4 Overview direct results (OLS)

Independent

variables Average β1 negative sig.* β Positive / Positive / negative sig. * β (in %)

Price -1.74 1 / 35 1,7 % / 58,3 %

Advertising 0.04 6 / 0 10 % / 0 %

Distribution 1.25 31 / 4 51,7 % / 6,7 %

Lagged Sales 0.38 42 / 0 70 % / 0 %

* p<0.05 & p<0.1

Figure 5 Distribution of parameter estimates per variable

Panel A: Price Estimates Panel B: Advertising Estimates

(35)

35

price elasticity of -2.62 that was found in the meta-analysis of Bijmolt, van Heerde and Pieters (2005). In figure 5 the distributions of the estimates are displayed. All of the estimates are included in the histograms, as all of these parameters are used as dependent variables in the WLS. Most of the price elasticities are between the 0 and the -4. This result was to be expected, because in the FMCG-market there are no expensive, luxury inelastic products for sale. More than half of the parameters are significant and pointing in the expected direction. The hypothesis regarding price as an instrument in the marketing mix referred to the fact that a decrease of price leads to an increase of sales. The added Z-method, which is explained in section 3.3, showed that the overall effect of price on sales is statistically significant (p<0.01) on a one-sided test. This means that there is empirical evidence to support hypothesis 1a. This result is in line with other research on marketing mix effectiveness (Ataman et al., 2010; Bijmolt et al., 2005; Hanssens et al., 2014).

Advertising.

Short-term. The significances of the outcomes of the independent variable advertising are less convincing than they were for the independent variable price. This elasticity is the short-term effect of advertising. For six of the 60 brands, advertising was found to be a significant predictor of volume of sales. The average advertising elasticity is .04. This means that when the advertising expenditures increase with 1%, the volume of sales will, ceteris paribus, increase with 0,04%. This is relatively low, but in line with other research (e.g., Ataman et al., 2010). Panel B of figure 5 shows the distribution of the significant parameters of advertising. Although the evidence is not that convincing, there are indications that advertising is positively related to sales. Namely, as is shown in table 4, there are some significant parameters. In addition, the significant parameters are all suggesting positive advertising elasticity, which is logical. The added Z-method points out that the one sided test is statistically significant (p<0.01). According to this p-value I conclude that there is enough evidence to assume that advertising positively affects sales. In other words, hypothesis 1b is supported.

Long-term. Because OLS provides estimates for β1 and λ (λ = lagged sales), the

long-term effect of advertising can be calculated. This effect can be calculated by using the next formula (Leeflang et al., 2015):

(36)

36

Thus, by dividing the beta of advertising by 1-λ, the elasticity of the long-term effect of advertising is calculated. It is possible that the advertising estimate is not significant on the short-term but is significant on the long-term. Therefore, for all the brands in the dataset the long-term effect of advertising is calculated. The added Z-method is used to assess whether the estimates are significant. The added Z-Z-method is also used to examine if the overall long-term effect of advertising is significant. In order to obtain the standard error of the estimate of the long-term effect of advertising, the delta rule is applied (𝑉𝐴𝑅 _!! = !!!

!_!! 𝜎!!+ _!!

!

! 𝜎!!). The estimates of the long-term effects of advertising are shown in Appendix C.

The added Z-method shows that the overall long-term effect of advertising is significant (p<0.01). The average long-term advertising elasticity is .07, which is stronger than the average short-term elasticity of .04. This is a reproduction of the study of Ataman et al. (2010). They also found that advertising is more effective in the long-term than in the short-term. These authors found a long-term elasticity of .12, which is higher than the .07 in this analysis. This long-term elasticity is in line with the study of Sethuraman et al. (2011), because they found that long-term advertising elasticity is double the short-term advertising elasticity. Looking at the results of the brands individually, it is clear that there are more significant parameters for long-term effect of advertising than for the short-term effect. For 33 brands there is data regarding advertising expenditures. In the short-term there were six significant estimates, in the long-term this number increases to 23 significant parameters. These results indicate that advertising on the long-term is much more effective than on the short-term.

(37)

37

Z-method points out that the variable distribution is overall significant (p<0.01). I conclude that there is enough evidence to accept hypothesis 1c, that large distribution coverage positively affects sales. Hypothesis 1d, regarding that distribution is the strongest predictor of sales in comparison with advertising and price, is rejected. Namely, price has a stronger average elasticity (-1.74) than distribution (1.25). Still, both marketing instruments, including advertising, can be deployed to generate sales.

4.2 Second stage analysis

The second stage of the analysis consisted of three regressions (WLS). The three estimates of model 1 (price, advertising, and distribution) functioned as the dependent variables in model 2a, 2b, and 2c. The inverse of the standard error of the estimates functioned as the WLS-weight. Three independent variables were included. Hedonic (0) and utilitarian (1) product category types, and private label (0) and national brands (1) were both dummy coded. Market share was measured as a continuous variable.

4.2.1 Model 2a: Price as dependent variable

In figure 6 the average price elasticity for the three independent variables of model 2a are shown. There is a small difference in price elasticity for national brands and private labels. This difference is larger for brand size. For each of the 15 categories, the two brands with the largest market shares represent the elasticity for the large brands. The other two brands represent the elasticity for the small brands.

In total there are 30 brands labeled as large brands, and 30 brands labeled as small brands. According to this bar chart hypotheses 3a and 4a are supported, since

Figure 6 Average price elasticities of brand and product characteristics

Hedonic

(38)

38

national brands and large brands are both stronger in terms of price elasticity. But the hypotheses can only be supported when there is statistical evidence that there is a significant difference between those elasticities. To test if the hypotheses hold, a WLS is applied.

In the WLS-regressions, the estimated parameters of β1 (price) functioned as the

dependent variable in model 2a. The WLS-weights were the corresponding inverse of the standard error of the estimates. Before interpreting the results, some model assumptions were checked (Leeflang et al., 2015). The first assumption is to check if the model is correctly specified. The RESET-test checks this assumption. The corresponding F-test is not significant (F=0.545, p>0.05), so the conclusion is that model 2a has the correct functional form. The next assumption is that the error term is homoscedastic. According to the scatterplot of the unstandardized residuals and the unstandardized predicted values (figure 8), it is not clear whether there is presence of heteroscedasticity. To statistically test this assumption the Levene’s Homogeneity of Variance Test is applied. The p-value of the Levene statistic is .67. This means that there is no reason to assume heteroscedasticity. The next assumption is the assumption of normality. In figure 7 the histogram of the standardized residuals is shown. It seems like the histogram is relatively normally distributed. To statistically test this assumption, the Kolmogorov-Smirnov test and the Shapiro Wilk tests are applied. Both p-values are statistically significant (p<0.05). This means that the null hypothesis is rejected and that the standardized residuals are not normally distributed.

This result has no direct consequences for the betas, but the p-values are no longer reliable. Also, there are no clear outliers that can be deleted. A remedy for this violation is the bootstrapping method. To accommodate for the non-normality issue

Figure 7 Scatterplot of residuals and predicted values

(39)

39

the bootstrapping procedure is repeated 10.000 times. The p-values of the parameters differ in significance level after the bootstrap procedure. Hence, the regression that included the bootstrap will be interpreted. Another assumption is that there is no multicollinearity. The VIF-values (Variance Inflation Factor) are a good indication if multicollinearity is an issue. These VIF-values are all smaller than five, which means that the multicollinearity assumption is not violated. The assumption of autocorrelation is not relevant in this case, because the data of the second stage analysis is not time series data.

The result of the regression is shown in table 5. The R2_{of the model is 0.223. This}

means that 22,3% of the variance in the dependent variable is explained by the three independent variables. The model as a whole is significant (F=5.349, p<0.01). The independent variable Dummy National Brands is significant (p<0.1, β=-0.932), which means that price has a stronger influence on national brands than for private labels. This means that there is empirical evidence to support hypothesis 4a. The other two independent variables are not statistically significant (p>0.1), which means that there is no evidence to support hypothesis 2a and 3a.

4.2.2 Model 2b: Advertising as dependent variable

In figure 9, the descriptives of the advertising elasticity are shown. There seems to be a large difference in advertising effectiveness for national brands and private labels. One important note is that there is data about advertising for only four of the 15 private labels. So this average advertising elasticity of 0.28 for private labels needs to be interpreted with caution, because the representativeness and generalizability of the results are at stake. But it could be that there is a significant difference in advertising elasticity between private labels and national brands. This difference is also tested with a WLS-regression.

β (After bootstrap)

Intercept 0.139

Dummy National Brands -0.932*

Dummy Utilitarian -0.615

Market Share -0.012

* p < 0.1

(40)

40

In Model 2b the estimate of advertising functions as the dependent variable. However, as said earlier in this section, advertising has a relatively low number of observations, because in most weeks there are no advertising expenditures. Model 2b is tested for the assumptions regarding functional form, heteroscedasticity, normality and multicollinearity. According to the RESET-test, model 2b the model specification is not correctly specified (F=16.553, p>0.05). The reason could be that there are omitted variables or that the model has not the right functional form. Because the parameter estimates can be biased, the betas need to be interpreted with care. When applying the Homogeneity of Variance Test, it turns out that the heteroscedasticity assumption (Var(εt)=σ2) is not violated (p<0.05). Thus, the error term is homoscedastic (Leeflang

et al., 2015). The Kolmogorov-Smirnov and the Shapiro Wilk are both statistically significant (p<0.05), which means that the unstandardized residual is not normally distributed. To remedy this violation the bootstrapping procedure is repeated 10.000 times for model 2b in order to analyze the regression results. In table 6 the regression with the bootstrap coefficients is displayed.

The bootstrap procedure has a great impact on the significance level of the parameters. Because of the fact that the bootstrap procedure has solved the non-normality violation, the β’s and corresponding p-value of the bootstrap procedure will be interpreted. The multicollinearity assumption is not violated, because all VIF-values

Intercept 0.161

Dummy National Brands -0.158

Dummy Utilitarian 0.003

Market Share 0.000

* P < 0.1, ** P < 0.05

Table 6 Results model 2b (+ bootstrap)

Hedonic 0.01 Utilitarian 0.06 National Brands 0.04 _Private Labels 0.28 Large brands 0.04 Small brands 0.04 0 0,05 0,1 0,15 0,2 0,25 0,3 A d ve rt is in g e la st ic it y

(41)

41

Figure 10 Average distribution elasticity for brand and product characteristics

are <5. Again, serial correlation is not an issue for this regression, because there is no time series data involved.

The R2_{of model 2b is 0.274. Thus 27,4% of the variance in de dependent variable}

is explained by the independent variables. The model as a whole is significant (F=3.781, p<0.05), so the model adds value to the existing literature. Even though none of the parameters turned out to be significant in the bootstrap procedure, there is still a relative high R2_{and the model is significant. Since there are no significant}

parameters, there is no empirical evidence for hypothesis 2b, 3b and 4b. Because the long-term effect of advertising was calculated, also these parameters plus the inverse of the standard errors were used as dependent variable and WLS-weight. It turned out that also in that WLS none of the independent variables were significant (p>0.1). These results of non-significance are surprising, among others because of the large difference in advertising elasticity between national brands and private labels in figure 9. It could be that because of the lack of data-points and variance in the data, the parameters are not significant. This could be a limitation of this study. I will elaborate on this issue in the limitation section of this thesis.

4.2.3 Model 2c: Distribution as dependent variable

In figure 10 the descriptives of the distribution elasticities of the independent variables of model 2c are shown. It is clear that there are visually no large differences between the elasticities. The hedonic / utilitarian product characteristics variable shows the largest difference (0.48). Again, a WLS-regression will be applied in order to analyze if there are significant differences within the variables.

(42)

42

In model 2c the variable distribution is used as dependent variable in the WLS-regression. The inverse of the standard error of the estimates of the variable distribution in model 1, are used as WLS-weight.

The R2_{of this model is relatively low (0.06), so the independent variables}

explained only 6% of the variance in the dependent variable. One major shortcoming of this model is that the model as a whole is not significant (F=1.184, p>0.05). This means that this model does not add any value. In other words, this analysis showed that there is no statistically significant relationship between the independent variables and the dependent variable. All the assumptions were checked in order to see if possible violations of the assumptions could be the cause of the insignificant F-test. Except for the normality violation, no violations were found. According to the F-statistic and the fact that the betas are not significant, there is no empirical evidence to support hypothesis 2c, 3c and 4c. Table 7 shows an overview of the WLS results.

4.2.4 Hypotheses

Table 8 provides an overview of the rejected and supported hypotheses of this study. In total there were 13 hypotheses. Four of the hypotheses were related to the direct effects of price, advertising, and distribution on sales. The other nine hypotheses were related to the moderating effects of brand and product characteristics in the relation of price, advertising, and distribution with sales.

Three out of the four hypotheses regarding the direct effects were found to be significant, so there is empirical evidence to support these hypotheses. With regards to the hypotheses of the moderating effects, there is empirical evidence to support one hypothesis, namely 4a.

When analyzing the bar charts of figure 6, 9 and 10, it seems that there are some differences within the independent variables. However, the WLS-regressions showed that these differences are not statistically significant. I will elaborate on these results on the next section, the discussion.

Intercept 1.550*

Dummy National Brands 0.003

Dummy Utilitarian -1.000

Market Share 0.228

* p < 0.1

(43)

43

Hypotheses: Rejected/Supported:

H1a: A decrease of price leads to an increase of sales, and an increase

of price leads to a decrease of sales. Supported

H1b: Advertising positively affects sales. Supported

H1c: Large distribution coverage positively affects sales. Supported

H1d: Distribution is the strongest predictor of sales in comparison with

advertising and price. Rejected

H2a: Price has a stronger influence on sales for utilitarian product

categories than for hedonic product categories. Rejected

H2b: Advertising has a stronger influence on sales for hedonic product

categories than for utilitarian product categories. Rejected

H2c: Distribution has a stronger influence on sales for hedonic product

categories than for utilitarian product categories. Rejected

H3a: Price has a stronger influence on sales for large brands than for

small brands. Rejected

H3b: Advertising has a stronger influence on sales for large brands

than for small brands. Rejected

H3c: Distribution has a stronger influence on sales for large brands

than for small brands. Rejected

H4a: Price has a stronger influence on sales for national brands than

for private labels. Supported

H4b: Advertising has a stronger influence on sales for national brands

than for private labels. Rejected

H4c: Distribution has a stronger influence on sales for national brands

than for private labels. Rejected

Marketing mix, sales, and the moderating role of brand type, brand size and product category type

Marketing mix, sales, and the

moderating role of brand type, brand size

and product category type

Marketing mix, sales, and the

moderating role of brand type, brand size

and product category type

A two-stage analysis

MANAGEMENT SUMMARY

PREFACE

TABLE OF CONTENTS

1. INTRODUCTION

2. THEORETICAL FRAMEWORK AND HYPOTHESES

3. DATA AND METHODOLOGY

S

= α

+ β

PR

+ β

ADS

+ β

DIST

+ λS

+ ε

β

= φ + ψ

HEUT

+ ψ

NBPL

+ ψ

BS

+ ε

β

= φ + ψ

HEUT

+ ψ

NBPL

+ ψ

BS

+ ε

β

= φ + ψ

HEUT

+ ψ

NBPL

+ ψBS

+ ε

4. RESULTS