Measuring Willingness to Pay: A Comparison of Direct Hypothetical Willingness to Pay Methods

Measuring Willingness to Pay: A Comparison of Direct

Hypothetical Willingness to Pay Methods

Richard Hendrik Burema - Author

3 | P a g e

Measuring Willingness to Pay: A Comparison of

Direct Hypothetical Willingness to Pay Methods

by R.H. Burema

University of Groningen

Faculty of Economics and Business

Msc Marketing Intelligence

June 27th 2013

Zwarteweg 54, 9717 HS Groningen

(06) 30072296 Supervisor: dr. J.E.M. van Nierop

r.h.burema@student.rug.nl Second supervisor: Prof. dr. L.M. Sloot Student number: 2220628

4 | P a g e

PREFACE

By means of this Master Thesis, an end has come to a beautiful and valuable period in my life. Six years ago my life as a student started by studying Facility Management at the Hanze University of Applied Sciences in Groningen. After my graduation in 2011, I was looking for a bigger challenge to make full use of my potential. In September 2011 I started the Pre-Master Marketing at the University of Groningen, a decision I have never regretted. After finishing the Pre-Master within the year, I started my Msc Marketing Intelligence in September 2012. A big driver for my enthusiasm for Marketing Intelligence is that I like to provide the evidence that marketing contributes to the bottom line results.

I would like to thank everyone who contributed to this study. Special thanks are there for dr. J.E.M. van Nierop for the guidance and valuable feedback during the process. Furthermore, I would like to thank my girlfriend Wisanne for her endless support and patience. Lastly, I would like to thank all of my close friends, fellow students and family for their great support throughout my studies.

Richard Hendrik Burema

5 | P a g e

MANAGEMENT SUMMARY

Methods determining willingness to pay (WTP) by consumers have received a lot of interest in the research literature (Miller et al., 2011, Voelckner, 2006, Wertenbroch and Skiera, 2002, Backhaus et al., 2005). But still there is little evidence about which method to measure WTP is the most accurate and whether there are significant differences between certain methods. This study tried to get insights in a small selection of willingness to pay (WTP) methods. The methods in this study all are direct hypothetical WTP methods. The aim of the research was to find out to what extent the methods differ in their outcomes. Three methods have been selected on the basis of their ease of applicability and the ability to directly compare the methods with each other. The three direct WTP methods used in this study are the Gabor Granger method, the van Westendorp method (also known as the price sensitivity meter), and the Self-Explicated approach. As far as known, this study is the first in comparing these methods directly with each other. The reason for only including hypothetical WTP methods is that these methods, compared to real WTP methods, are much cheaper, can be performed within a short time span with less effort, and the execution is more straightforward (Gabor and Granger, 1970).

6 | P a g e

The results from the median test and Kruskal-Wallis test show that there are differences between the used methods in their values and distributions. The median test shows that the vW method most frequently elicits higher values compared to the other methods. The vW method shows significantly higher distributions for the beer category over the SE method and on aggregate level over both the SE and GG method. Based on this finding, it can be said that the vW method has a significantly higher median and mean rank compared to the other methods.

The model estimation by means of linear regression resulted in very low R² values and a large amount of insignificant effects of the variables on the WTP. In these models only income, age, and favorite brand seem to influence the amount a respondent is willing to pay. Because the linear regression models did not account for heterogeneity among respondents, latent class regression has been chosen for subsequent analysis. With latent class regression the respondents are grouped on the basis of similar parameter effects on their WTP, hence this is a more sophisticated method for defining segments than an ordinary cluster method. The results of the latent class regression show that in almost all classes the WTP value elicited by the SE method is right in between the other two methods. Furthermore, the effects of category experience and price consciousness on WTP are inconclusive. Next, for the majority of the classes the moderation effect of the category experience with the vW method has a significant positive effect on the WTP value. Hence, when people are confronted with the vW method to elicit their WTP and are experienced within the category, they are willing to pay more compared with the SE method and GG method. As expected, in most of the cases the WTP value is higher when the favorite brand of the respondents is an A-brand. Noticeable is that the WTP for a discount brand is higher in some classes. Finally, the direction of the effect of the respondents’ socio-demographics (age, gender, and income) on the WTP is not consistent.

7 | P a g e

CONTENTS

1 INTRODUCTION ... 10

2 LITERATURE REVIEW ... 13

2.1 Conceptual definition WTP ... 13

2.2 Gabor-Granger Price Study ... 14

2.3 Van Westendorp ... 15

2.4 Self Explicated approach ... 17

2.5 Price Consciousness ... 18

2.6 Category Experience ... 20

2.7 Covariate: Income ... 21

2.8 Conceptual Model ... 21

3 METHODOLOGY ... 22

3.1 Participants and Stimulus ... 22

3.2 Determining Price Ranges ... 23

3.3 Experimental Design ... 25

3.4 Data description ... 26

3.4.1 Methods and Products ... 26

3.4.2 Price Consciousness ... 27

3.4.3 Category Experience ... 27

3.4.4 Favorite brand ... 28

3.4.5 Willingness to pay value ... 28

3.5 The model ... 29

4 RESULTS... 30

4.1 Face validity ... 30

8 | P a g e 4.3 Gabor Granger ... 32 4.4 van Westendorp ... 34 4.5 Self-Explicated ... 36 4.6 Comparing methods ... 38 4.7 Model estimation ... 41 4.7.1 Beer model ... 41 4.7.2 Coffee model ... 42 4.7.3 Cola Model ... 42 4.7.4 Aggregate Model ... 43 4.8 Normality ... 45 4.9 Multicollinearity ... 45 4.10 Heteroscedasticity ... 45 4.11 Model Fit ... 46 4.12 Classes ... 46 4.12.1 Model fit ... 47

4.12.2 Classes for Beer ... 48

4.12.3 Classes for Coffee ... 51

4.12.4 Classes for Cola ... 54

5 CONCLUSION AND DISCUSSION ... 57

5.1 WTP values ... 57

5.2 WTP distributions ... 58

5.3 Price Consciousness ... 58

5.4 Category Experience ... 59

5.5 Income ... 59

5.6 Difference WTP for Products ... 60

9 | P a g e

REFERENCES ... 62

APPENDIX A: Graphs ... 67

APPENDIX B: Price Ranges Supermarkets ... 73

APPENDIX C: Median per Product Category ... 74

APPENDIX D: Homogeneity of Variances ... 75

10 | P a g e

1

INTRODUCTION

Methods determining willingness to pay (WTP) by consumers have received a lot of interest in the research literature (Miller et al., 2011, Voelckner, 2006, Wertenbroch and Skiera, 2002, Backhaus et al., 2005). But still there is little evidence about which method to measure WTP is the most accurate and whether there are significant differences between certain methods. This study tries to get insights into a small selection of WTP methods. When designing optimal price policies, companies need to know which procedure for measuring consumers’ WTP is most valid in what situation (Voelckner, 2006). It can be said that WTP denotes the maximum value a buyer is willing to pay for a given quantity of a good (Kalish & Nelson, 1991; Wertenbroch & Skiera, 2002). There are several methods to determine the value that consumers are willing to pay for a given good. WTP methods can be divided into two main groups, namely hypothetical and real approaches (Miller et al., 2011; Voelckner, 2006). Both hypothetical and real approaches can be divided into direct and indirect approaches. Hypothetical approaches try to elicit a price which respondents would pay for a given good as if they were in a real shopping situation, but without the financial consequences. Real methods to elicit willingness to pay are approaches in which the respondent is expected to pay the price that is elicited in the study for a given good. Research has shown that the hypothetical methods can lead to a hypothetical bias (Backhaus et al., 2005, Miller et al., 2011, Wang et al., 2007) which might lead to inaccurate estimates of the WTP value. Prior studies provide initial evidence that hypothetical WTP methods elicit a substantially higher WTP value than real methods (e.g., Neill et al., 1994; Botelho and Pinto, 2002; Johannesson, Liljas and O’Conor, 1997;Wertenbroch and Skiera, 2002) In a hypothetical context the WTP value that is elicited might differ from the WTP values in a real context. This is defined as the hypothetical response bias (Nape et al., 2003).

11 | P a g e

analysis, elicit a WTP value by means of showing a profile of certain attributes of the product in which respondents choose or rate the profiles. This way, researchers can decompose the answers of the respondents and calculate the WTP value. Respondents might not directly see what the research purpose is and therefore it is likely that the strategic response bias is somewhat less apparent with an indirect method compared to a direct method.In the research of Miller et al. (2011) it has been found that open-ended (OE) (direct method) can outperform choice based conjoint (CBC) (indirect method) when looking at the WTP means and WTP distributions, compared to the actual demand curves, as well as in making pricing decisions for an inexpensive, frequently purchased, nondurable product category. This has been found for a cleaning product which is relatively low priced, more frequently purchased, in a nondurable product category with no direct competition. Backhaus et al. (2005) and Voelckner (2006) found that indirect approaches are more suitable for a product category in which a more extensive decision process is involved, while direct methods are more suitable for frequently purchased product categories (Gabor and Granger, 1966) and more suitable for product categories without explicit competitive offering (Huber, 1997).

The main advantages of the hypothetical methods over real methods are that the former is very much cheaper than the latter, that it can be performed within a short space of time and effort and the simplicity of execution. In this respect it is somewhat removed from reality, hence perfect correspondence of the results with behavior under actual shop conditions could not be expected (Gabor and Granger, 1970).

This study aims to get insights into three different hypothetical direct WTP methods. The methods have been selected on the basis of their ease of applicability and the ability to directly compare the methods with each other. The three methods used in this study are Gabor Granger, Self-Explicated approach, and van Westendorp.

The main aim of this study is the find out to what extent these methods differ in outcomes (WTP value) from each other. To find out whether there exists a difference between the methods, a comparison between different product categories has been done. Other variables of interest are the effect of the price consciousness on the WTP value for the different methods, and the effect of category experience on the WTP value. The main research question of interest in this study is:

12 | P a g e

And sub-research questions are:

• If there is a difference; to what extent and how (mean, distribution etc.) do the willingness to pay values differ between and within product categories?

• What is the effect of the price consciousness for a product category on the WTP value? • What is the effect of category experience of consumers on their WTP value?

• What is the effect of income on consumers WTP?

13 | P a g e

2

LITERATURE REVIEW

2.1

Conceptual definition WTP

In the literature many different definitions of WTP can be distinguished. In Table 1, a number of these definitions are shown. As already mentioned the definition of Kalish & Nelson (1991) and Wertenbroch & Skiera (2002) is used as the definition of WTP in this study.

Table 1 - Definitions WTP

Authors Definition willingness to pay

Kalish & Nelson (1991); Wertenbroch & Skiera (2002) “WTP denotes the maximum value a buyer is willing to pay for a given quantity of a good”

Hauser & Urban (1986) “the minimum price at which a consumer will no longer purchase the product”

Varian (1992) “the price at or below which a consumer will purchase one unit of the good.”

Ariely, Loewenstein & Prelec (2003)

“(1) a threshold price up to which a consumer definitely buys the product, (2) another threshold above which the consumer simply walks away, (3) and a range of intermediate prices between these two thresholds in which consumer response is ambiguous.

Wang, Venkatesh & Chatterjee (2007)

“(1) floor reservation price, the maximum price at or below which a consumer will definitely buy one unit of the product; (2) indifference reservation price, the maximum price at which a consumer is indifferent between buying and not buying; and, (3) ceiling reservation price, the minimum price at or above which a consumer will definitely not buy the product.”

Jedidi & Zhang (2002) ‘the price at which a consumer is indifferent between buying and not buying the product’

14 | P a g e

This is contradictory to most of the other definition in which the consumers’ WTP is a maximum price which a consumer is willing to pay for a given good.

2.2

Gabor-Granger Price Study

The Gabor-Granger price study originates from 1961, developed by the researchers A. Gabor and C.W.J. Granger (1961,1966). The purpose of the Gabor-Granger studies is to establish a response curve which shows the percentages of respondents who choose to buy the product at various prices (Wedel and Leeflang, 1998). The respondents are offered products at various prices and must decide whether to buy or not to buy the product at the specific price.

The advantage of Gabor-Granger studies over observational data is that price levels can be determined by the researchers. This way, more price variation can be obtained through the study. Generally, this method results in a downward sloping demand curve (Gendall et al., 1997). Therefore, price sensitivity can be measured at each price point tested (Gabor and Granger, 1964). As Gabor and Granger indicate, consumers most probably decide to purchase a product when the price range falls within an acceptable price range which is based on the current market prices and the prices of the product normally purchased. Gabor and Granger (1965, 1966) not only confirmed the acceptable price range hypothesis, but also found that the range shifted downward as income fell. Moreover, as income fell, the upper price threshold dropped less than the lower one, implying that low price was a more potent deterrent to the higher-income groups than was high price to lower income groups (Monroe, 1973).

Comparable to this study, Gabor and Granger (1970) have been studying product categories as a whole to determine the price sensitivity of the category specifically. The product categories have been distinguished based on the price categories but not in any other way between different brands. This is also what this study is about.

15 | P a g e

A limitation of the Gabor Granger price study that has been discussed by Wedel and Leeflang (1998) is that the effect of prices of competitive brands are not considered explicitly. Leeflang (1994) also found that the method does not take the reactions of competitors on price changes into account. Furthermore they state that usage frequency and the probability that a respondent will purchase the product is not covered by the method. Another limitation of the method is that the sequence of shown prices and questioning affects the responses of the respondents. Respondents will use the first shown price as their reference price (Leeflang, 1994), to which the respondents relate the successive prices to. Therefore a fully randomized design needs to be used to prevent such structural biases.

2.3

Van Westendorp

The van Westendorp (vW) method also known as the price sensitivity meter (PSM) is a WTP method that has been developed by the Dutch researcher Peter van Westendorp in 1976 (van Westendorp, 1976). The model is an extension of the Gabor Granger method that has been described previously. The van Westendorp consists of 4 basic questions which are answered by the respondents:

1. At which price on this scale are you beginning to experience (test-product) as cheap? 2. At which price on this scale are you beginning to experience (test-product) as expensive? 3. At which price on this scale you are beginning to experience (test-product) as too expensive – so

that you would never consider buying it yourself?

4. At which price on this scale you are beginning to experience (test-product) as too cheap – so that you say “at this price the quality cannot be good?

This way the researcher can obtain four price points to base its conclusion on. As can be seen in figure 1, the first question (cheap) will show a declining cumulative frequency distribution. Combined with the increasing cumulative frequency distribution resulting from question 2 (expensive), the indifference price can be seen at the intersection of the two distributions. Hence, this is the price at which an equal number of respondents see the product as “cheap” and “expensive”.

16 | P a g e

within the product category. Furthermore, the IDP can vary for different sub-markets such as people who buy cheap or expensive brands.

Figure 1 - PSM Distribution Example

Source: Roll, Achterberg, and Herbert (2010) Conference Proceeding

The ideal price point is said to be obtained at the intersection of the distributions resulting from question 3 (too expensive) and question 4 (too cheap). At this price point there are an equal number of respondents that find the product “too cheap” and “too expensive”. It is called the ideal price point because there is a minimum of resistance at this price. As can be seen in Figure 1, at the ideal price point there is around 10% of resistance to the price (10% of the people find the price “too cheap” and “too expensive”). At the indifference price the resistance is already around 35%. Research of van Westendorp (1976) shows that it might happen that the two lines do not cross. In that case van Westendorp refers to an optimal price range.

17 | P a g e

According to Lyon (2002) a common limitation that has been found in many research studies is that surprisingly high numbers of respondents give four answers that do not show the expected relationship. For example, their “expensive” price is higher than their “too expensive” price. It has been found that as many as 20% of the respondents may give incorrect answers which leads to inconsistent figures.

2.4

Self Explicated approach

The self-explicated (SE) WTP method, also known as contingent valuation, direct elicitation method, direct questioning and price matching, is the third direct hypothetical WTP method. With SE the respondent explicitly state their reservation prices for the product categories.

According to Jeddi and Jagpal (2003) the use of self-stated reservation prices makes two implicit assumptions: reservation prices are deterministic (i.e., within-subject variability is zero) and provide unbiased estimates of the true, unobservable reservation prices. As already mentioned before, direct methods to elicit WTP suffer from a bias that is due to strategic responses but also from social responsible answers (Green and Srinivasan, 1990). This is in line with the findings of Gabor and Granger (1965) which shows that SE WTP methods can lead to significant measurement error, especially for products that are not purchased frequently or are new to the market. This is because consumers do not yet have a internalized WTP for those products. This however, does not apply to this study because the product categories of interest will be, generally speaking, frequently purchased FMCGs.

SE WTP is likely to be biased downward because consumers think that it is in their best interest to keep prices down (Monroe 1990). Miller et al. (2011) found that although a method such as SE suffers from a hypothetical bias it may still lead to a correct demand curve and good pricing decisions can be based on it. A possible explanation given by Miller et al. (2011) for this finding is that the optimal price(p*) in their research is less sensitive to changes in intercept (a) than to changes in slope. They found that, when comparing intercepts and slopes, hypothetical methods are largely biased in terms of intercept but no t much in terms of slope. This implies that methods that yield demand curves, which are highly biased in terms of intercept but not so much in terms of slope, may still yield good estimates of optimal price (p*).

18 | P a g e

be found with the Gabor Granger price study and the van Westendorp. Also, consumers might not have the incentive to state their true WTP; they might understate or overstate their WTP. This might be due to the strategic response bias as mentioned earlier in which respondents try to keep the prices down by presenting themselves as price conscious consumers (Nessim and Dodge, 1995, p.72). The opposite can also occur, Nagle and Holden (2002, p.344) observed that respondents overstated their WTP because they did not want to seem stingy towards the researcher.

The primary advantage of the SE method is its simplicity and thus the low demand on the cognitive capabilities of the respondents, which leads to a relatively low abandonment rate (Backhaus et al. 2005, Green and Srinivasan 1990). The empirical results (e.g. Leigh, MacKay, and Summers 1984; Srinivasan 1988; Wright and Kriewall 1980) prior to the research of Green and Srinivasan (1990) shows that the SE approach is likely to yield predictive validities that are roughly comparable to those of traditional conjoint analysis which shows that a direct WTP method is likely to result in comparable results as an indirect WTP method.

The ‘floor reservation price’ from the research of Wang et al. (2007) is used in this study to determine the maximum WTP value for the consumers. The floor reservation price is the maximum price at or below which a consumer will definitely buy one unit of the product (i.e., 100% purchase probability) (Wang et al. 2007).

It is argued that the single-point representation of reservation price requires the central assumption that a consumer knows with certainty how much he or she would be willing to pay for one unit of the product (Hanemann 1984). This is a strong assumption because consumers may lack a well-defined preference structure or the cognitive ability to make such an assessment (Gregory, Lichtenstein, and Slovic 1993), except perhaps in the most familiar contexts (Fischhoff 1991). Therefore, it is chosen to include the variable product familiarity in the research.

H1: (a) The WTP methods will yield different WTP values and (b) different distributions.

2.5

Price Consciousness

19 | P a g e

category (Sinha and Batra, 1999). Price consciousness might depend on the consumers’ perceptions of the category characteristics and their preference towards the product category. As Tai and Tam (1997) stated, price consciousness is an attitude-like enduring predisposition that varies in intensity across individuals. Therefore it can be said that price consciousness varies across consumers and across products and situations for the same individual (Lichtenstein et al. 1988). This is in line with earlier research in which Monroe and Krishnan (1985) showed that consumers can be more or less price consciousness when shopping for certain products in contrast to others. The contrast in price consciousness between certain products can be ascribed to a difference in perceived risk and other reasons. To be able to show the difference in price consciousness between different products it is necessary to obtain data from each individual that covers multiple product categories (Sinha and Batra, 1999).

Table 2 - Definitions Price Consciousness

Authors Definition Price Consciousness

Lichtenstein, Bloch, and Black (1988) “the degree to which the consumer uses price in its negative role as a decision-making criterion”

Lichtenstein, Ridgway, and Netemeyer (1993)

“the degree to which the consumer focuses exclusively on paying low prices”

“the degree to which a buyer is ‘‘unwillingness’’ to pay a higher price for a product”

Sinha and Batra (1999)

“a consumer’s reluctance to pay for the distinguishing features of a product if the price difference for these features is too large’’

In this study the definition for price consciousness of Lichtenstein et al. (1993) is used; it is the degree to which the consumer focuses exclusively on paying low prices. In this way quality perception is not taken into account by the consumer and therefore it will not be part of this research. From this the following hypothesis is formulated:

20 | P a g e

2.6

Category Experience

In the literature it has been found that the extent to which a customer has knowledge about and experience in a product category has an influence on their willingness to pay. Consumers’ product knowledge can be seen as familiarity with the product (Alba and Hutchinson, 1987). Familiarity is based on purchase frequency, price knowledge, and other experiences. Therefore, consumers who are more familiar with a product or product category are more aware of the current market prices and the total assortment of products and brands in the products in the category. Kalyanaram and Little (1994) found that consumers with a higher frequency of purchase have narrower latitude of price acceptance, because they are more aware of the range of price distributions in a store environment. As Monroe (1990) suggests, previous and present prices affect internal reference prices, or consumers’ expectation of a reasonable price level. As already mentioned, WTP can be captured in a price range, the acceptable price range is conceived to have upper and lower price limits or endpoints (Gabor and Granger 1966; Monroe and Venkatesan 1969). The upper price limit shows the price at which consumers consider the product to be too expensive, while the lower price limit shows the price below which consumers would be suspicious of the quality of the product. This is in line with the findings of Roa and Sieben (1991), which shows that acceptable price range end points (price limits) were found to be lowest for low-knowledge subjects. Also they found that an increase of prior product low-knowledge is accompanied by an increase in both limits of the acceptable price range. As found by Roa and Sieben (1991) this is also suggested in the behavioral pricing literature which shows that prior product knowledge moderates the effect of price on perceived quality and influences the prices consumers are willing to pay the acceptable price range for a given level of quality (Lichtenstein, Bloch, and Black 1988; Rao and Monroe 1988). In a subsequent study by Janiszewski and Lichtenstein (1999) this finding is also confirmed. It is expected that consumers’ with an above average category knowledge yield higher WTP values than low category knowledge consumers. This would be in line with the results or Roa and Sieben (1991) who found that low category knowledge consumers systematically yield lower price limits than moderately or high knowledgeable consumers. This is because a higher degree of uncertainty leads to greater discounting of the inferred products’ value (Ford and Smith 1987, p. 363) and uncertainty is triggered due to missing product information (Roa and Sieben, 1991).

21 | P a g e

2.7

Covariate: Income

As already mentioned in chapter 2.3, Gabor and Granger (1965, 1966) stated that consumers most probably decide to purchase a product when the price range falls within an acceptable price range which is based on the current market prices and the prices of the product normally purchased. Gabor and Granger (1965, 1966) not only confirmed the acceptable price range hypothesis, but also found that the range shifted downward as income fell. Moreover, as income fell, the upper price threshold dropped less than the lower one, implying that low price was a more potent deterrent to the higher-income groups than was high price to lower income groups (Monroe, 1973). Income is expected to be positively related to the willingness to pay.

H4: Consumers with a higher income elicit a significantly higher willingness to pay value.

2.8

Conceptual Model

The conceptual model graphically represents all the variables and relationships addressed in this study. As already mentioned the main focus of this paper is to find out to what extent the direct hypothetical willingness to pay methods differ from each other in their outcomes. This is done on the basis of three different product categories. As presented, the relationship is moderated by the price consciousness of the respondents and their category experience. Price consciousness is expected to negatively influence the WTP values elicited by the WTP methods. Furthermore, category experience is expected to have a positive influence on the WTP value that is elicited by the WTP methods. Income of the respondents is expected to have a negative direct influence on the WTP value. Finally, for estimating the aggregate model, the product categories are also included as moderating variables.

22 | P a g e

3

METHODOLOGY

3.1

Participants and Stimulus

For this study, consumers are asked to participate in the study. To recruit participants, consumers were contacted through social media sites (Facebook, LinkedIn, Twitter etc.), email invitations, face-to-face contacts, forums (Wetenschapsforum, Trosradar) and by spreading 240 flyers around the neighborhood in Groningen.

The stimuli used were three different product categories from the FMCG segment. The categories used were: beer, coffee and soft drinks. In the study the respondents give their WTP for a crate of beer (24 bottles, 30cl/33cl), coffee pads (36 pads), and cola (1,5 liter bottles).

In total, 380 consumers have participated in the study. After analyzing the data, a total of 249 respondents had a completion rate over 80%, which were found to be useful for the study. The sample consists of 54% males (132) and 46% females (114) with an average age of 32 years. Between the three versions of the survey the average completion rate (64%) differed, resulting in a total of 90 (70%) useful respondents in version 1, 70 (54%) respondents in version 2, and 89 (68%) respondents in version 3. In Table 3, the income levels and employment status of the sample are shown.

Table 3 - Income and Employment status

Income Employment status

0 - 9.999 31% Unemployed 2%

10.000 - 19.999 20% Student, unemployed 12%

20.000 - 29.999 11% Student, employed 28%

30.000 - 49.999 20% Entrepreneur 3%

50.000 - 59.999 3% Employed, Part-time (0 -19 hours) 8%

60.000 - 69.999 6% Employed, Part-time (20 - 29 hours) 11%

70.000 + 9% Employed (30 - 40 hours) 32%

Retired 2%

23 | P a g e

3.2

Determining Price Ranges

To determine the price ranges for the products in this study a pre-test has been conducted. The method is partly adopted from the research of Dodds, Monroe, and Grewal (1991). The results are used as input for the final price ranges in the study. The pre-test has been conducted as follows, a total of X consumers are asked to state the price ranges they expect to be present in the product categories. They are also asked to state the price ranges across different stores (discount stores; such as Aldi and Lidl and general supermarkets; such as AH, Jumbo, PLUS etc.). Dodds et al. (1991) determined on the basis of their pre-tests three different price points, high price, medium price, and low price. Additionally, one price treatment (too high) was above subject’s acceptable price ranges. Different from the research of Dodds et al. (1991) the respondents in this study provide the low and high price for each product category, not the medium price. Furthermore, as Dodds et al. (1991) indicates, the price range they used in their study was too narrow. The addition of more price points at both ends of the price continuum would provide a stronger test (Dodds et al. 1991). The general shopping behavior of the respondents is also questioned. This way the reliability of the provided price ranges can be determined. Following the pre-test with the test panel, a field research is conducted to verify the price ranges provided and to determine the final price ranges used in the study. The pre-test has been conducted among a total of 11 people consisting of 6 (55%) males and 5 (45%) females. They were directly questioned what prices they thought that were present in the Dutch supermarkets for the product categories of interest (beer, coffee, cola). In Table 5, the results of the pre-test are shown. As previously mentioned, the final price ranges are determined by validating the price ranges from the pre-test with the actual price ranges in the Dutch supermarkets.

To be certain that gender does not have an influence on the price ranges in this pre-test, an ANOVA has been performed. The results of the ANOVA indicate that there are no significant differences between males and females in the pre-test. In Appendix B the actual price ranges in several Dutch supermarkets are shown. The overall price range across the supermarkets can be seen in Table 4.

Table 4 - Overall price ranges across supermarkets

24 | P a g e

Table 5 - Results Pre-test Price Ranges

Perceived Actual Perceived Actual

Supermarket Product Minimum Minimum % Difference Maximum Maximum % Difference

Normal Beer € 4,78 € 4,69 2% € 16,39 € 15,07 9% Discount Beer € 4,11 € 5,99 -31% € 11,02 € 6,99 58% Normal Coffee € 0,93 € 1,99 -53% € 7,35 € 3,94 87% Discount Coffee € 0,38 € 1,99 -81% € 6,20 € 1,99 212% Normal Cola € 0,35 € 0,39 -10% € 2,52 € 1,74 45% Discount Cola € 0,16 € 0,55 -71% € 1,84 € 1,65 12%

25 | P a g e

=Max. P − Min. P 14

The price interval is then 14 times added to the minimum price to get to the maximum price. For the beer, coffee, and cola category the price interval are determined, respectively € 0,89, € 0,22, and € 0,16.

3.3

Experimental Design

Respondents are randomly assigned to one of the three surveys. Thus, three groups have been formed with the experimental design. In Table 6, it can be seen that the groups all use different methods for the various product categories. This way results from every WTP method is collected for every product category. The respondents have not been informed about these groups, thus have no prior awareness of the sequence of asking. Before they start with the methods to determine their WTP, the respondents are told that they have to imagine that they are in a shopping situation in which they went out to get some ‘Product X’ for themselves, and saw the product they were looking for (beer, coffee, and soft drinks). The respondents are told that they have to look critically and realistically to the prices in the study. Each respondent will answer some questions before stating their WTP for the products. Prior to each product category the respondents need to answer questions regarding their experience and their favorite brand in the specific product category. Furthermore, the respondents need to state in which stores they normally go shopping for these products. In the Gabor Granger study the respondents need to complete a survey in which they are asked to state if they would buy their favorite brand within the product category at a particular price Y (‘Would you buy product X at price Y?’). Following the research of Wedel and Leeflang (1998), the order of fifteen different price levels, varying from X to X, were randomly called out to each of the respondents. At each price point respondents were required to respond with a statement whether they intend to buy the brand (1) at that price or not (0). This will result in a so-called buy-response curve, which depicts the percentages of consumers buying a certain brand at various prices (Wedel and Leeflang, 1998). Frequently, in Gabor Granger price studies, a bell-shaped price-response function is observed, in which the percentage of consumers buying the brand rises at lower prices, and declines again at higher prices. Gabor and Granger (1966) attributed this effect to consumers' use of price as an indicator of quality (Wedel and Leeflang, 1998).

26 | P a g e

self-explicated method the maximum value a buyer is willing to pay for a given good is directly asked from the respondents.

Table 6 – Experimental Groups

Method Product Category

Group 1 Gabor Granger study Beer Van Westendorp (PSM) Coffee Self-Explicated Soft drinks Group 2 Van Westendorp (PSM) Beer

Self-Explicated Coffee Gabor Granger study Soft drinks Group 3 Self-Explicated Beer

Gabor Granger study Coffee Van Westendorp (PSM) Soft drinks

First, the respondents need to answer questions about their age and gender followed by the first WTP method in the sequence. Thereafter, the respondents answer a question about their price consciousness for the preceding product category. Then the second method in the sequence is shown and answered by the respondents, here the respondents again indicate their price consciousness of the preceding product category, followed by questions regarding (household)income and employment status. The remainder of the survey will include the third WTP method in the sequence, with another question about the price consciousness in the preceding product category. Price consciousness was measured using a scale by Lichtenstein et al. (1993), with the responses anchored at 1 = strongly disagree, and 9 = strongly agree. These questions are asked at the end of the survey to minimize the strategic response bias in the WTP statements.

3.4

Data description

Before the models can be estimated correctly, the data needed to be transformed and recoded. In this chapter the steps that were taken are explained.

3.4.1 Methods and Products

27 | P a g e

method of dummy coding has been applied with beer as the base category. These dummy variables will be used to estimate the aggregate model.

3.4.2 Price Consciousness

The extent to which consumers are price conscious is measured with the scale developed by Lichtenstein et al. (1993). The five questions that are asked to measure this construct are factored into one variable. First, a cronbach’s alpha (α) has been performed to measure the internal consistency. According to Malhotra (2010), the cronbach’s α is the average of all possible split-half coefficients resulting from different ways of splitting the scale items which varies from 0 to 1, and a value of 0,6 or less generally indicates unsatisfactory internal consistency reliability. With the five questions considering the price consciousness of consumers for a crate of beer, a cronbach’s α test has been performed which resulted in a cronbach’s α of 0,845. For the coffee pads a cronbach’s α of 0,906 and for a bottle of cola a cronbach’s α of 0,911 has been found. Therefore, it can be said that for all the products the five price consciousness questions have sufficient internal consistency for factoring. For factoring the scores, a factor analysis has been performed with the principal components extraction method. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy is used to examine the appropriateness of factor analysis. High values (between 0,5 and 1,0) indicate factor analysis is appropriate (Malhotra, 2010). Furthermore, the Bartlett’s test of sphericity has been performed to examine the hypothesis that the variables are uncorrelated in the population, where H₀ = variables are uncorrelated. The KMO measure for the price consciousness regarding the crate of beer is found to be 0,830 and the Barlett’s test shows that the variables are correlated (p = 0,000). Furthermore, the total explained variance must exceed 60% to be able to factor, in this case the total explained variance is 62% which is sufficient. For the coffee pads the KMO is found to be 0,851 and the Barlett’s test shows that the variables are correlated (p = 0,000). The total explained variance in this case is 73%. For the bottle of cola the KMO measure is found to be 0,860 and the Barlett’s test shows that the variables are correlated (p = 0,000). The total explained variance for this factor is 74% which is also sufficient. Therefore it can be concluded that factoring these five questions is appropriate and allows using the provided factor scores in the estimation.

3.4.3 Category Experience

28 | P a g e

For a crate of beer the cronbach’s α of the three variables is found to be 0,631, for coffee the cronbach’s α is 0,593, and for cola the cronbach’s α is 0,645. Only the internal consistency of the variables for the coffee pads is slightly under 0,6. This might be due to the low price knowledge of the respondents for coffee compared to the other products. By means of a post hoc test, it has been found that the price knowledge of the respondents is significantly lower for both beer and cola, with a mean difference of respectively -0,588 (p = 0,000) and -0,406 (p = 0,012). Because the crobach’s α for beer and cola is sufficient and the α of coffee is just slightly under the cutoff value it determined to still factor try to factor these scores. For a crate of beer, the KMO measure is found to be 0,602, for coffee the KMO measure is 0,621, and for cola the KMO measure is 0,593. Hence, all measures are exceed 0,5 therefore it indicates that factor analysis is appropriate. Also the Barlett’s test of sphericity shows that for all products the Hᵒ (variables are uncorrelated) can be rejected (p = 0,000). The total explained variance also exceeds 60% when the three variables are belonging to one factor (beer = 63%, coffee = 63%, and cola = 66%). The factor analysis shows that it is appropriate to grasp the variables into one construct, namely category experience.

3.4.4 Favorite brand

The respondents were asked to state what their favorite brand was before answering all other questions regarding the price consciousness, category experience, and willingness to pay. To be able to use this information the brands have been divided into three different categories (A-brand, private label, and discount brand), a fourth category for this variable was ‘no preferred brand’. After this dummy variables have been made with the ‘A-brand’ as base category.

3.4.5 Willingness to pay value

29 | P a g e

3.5

The model

With the following general model the effects of the used variables on the WTP value from customer i for product j can be estimated:

= ₀ + ₁ + + ₃ ! + ₄ ! + ₅ ∗ ! + ₆ ∗ ! + ₇

∗ ! + ₈ ∗ ! + ₉() ,+ + ₁₀() ,, -./0 + ₁₁() , /+ 1+ ₁₂3

+ ₁₃4 + ₁₄5,6

The second model is used for estimating the aggregate model over the product categories where the effect of the different price levels of the products is eliminated for comparison purposes. It estimates the indexed WTP value for customer i :

, 7 = ₀ + ₁ + + ₃ ! + ₄ ! + ₅ ∗ ! + ₆ ∗ ! + ₇

∗ ! + ₈ ∗ ! + _{9 ./11} + _{10 ./} + ₁₁ ∗ _./11 + ₁₂ ∗ _./

+ ₁₃ ∗ ./11 + ₁₄ ∗ ./ + ₁₅(),+ + ₁₆(),, -./0

+ ₁₇(), /+ 1+ ₁₈3 + ₁₉4 + ₂₀5,6

M Method used for eliciting willingness to pay, dummy variable with Gabor Granger as base method.

PC Price consciousness level for the specific product of the respondent, factor scores.

CE Category experience the respondent has with the product, factor scores.

P Product category, dummy variable with beer as base product.

FB The price level of the favorite brand of the respondent given a specific product, dummy variable with A-brand as base brand.

I (Household) income level of the respondent, ordinal scale variable.

A Age of the respondent.

30 | P a g e

4

RESULTS

4.1

Face validity

To find out whether the methods used measure the same construct, which is the willingness to pay value for the specific product, the cumulative demand (%) of the product at a specific price (€) has been plotted. This can be seen in the Figures 3, 4, and 5. It can be said that, although the demand functions are not exactly the same, they all follow the same trend. Further analysis shows to what extent the methods are different from or similar to each other. At first sight, it can be said that every method measures the same construct, willingness to pay, and that the methods seem to measure somewhat the same demand at the price points for every product. Second, the biggest difference can be seen between the van Westendorp method and the self-explicated willingness to pay method within the beer category. Further analysis will show whether these methods are statistically different from each other. Another measure of face validity is the correlation between the WTP value and the likelihood of purchasing the product the upcoming month (Voelckner, 2006). As expected, there exists a significant correlation of 0,142 (p = 0,000) has been found between the respondent’s WTP values for the products and their likelihood of purchasing. The correlation is based on the aggregate data to be sure that it holds over the different products.

Figure 3 - Cumulative demand beer per method

31 | P a g e

Table 7 - WTP means and percentage difference between methods

Gabor Granger Van Westendorp Self-Explicated GG vs vW GG vs SE vW vs SE Beer € 11,84 € 12,16 € 11,49 -2,65% 2,95% 5,45%

Coffee € 3,46 € 3,56 € 3,35 2,94% 3,04% 5,81%

Cola € 1,64 € 1,71 € 1,63 -4,02% 1,10% 4,92%

Figure 4 - Cumulative demand coffee per method

Figure 5 - Cumulative demand cola per method

32 | P a g e

4.2

Normality

To make sure the right methods are used in this study, the normality assumption is tested. This has been done by running a Kolmogorov-Smirnov (KS) test for the dependent variable, WTP values per product category. All KS statistics are found to be significant (p < 0,05), which infers that it can be assumed that the WTP values do not follow a normal distribution, as can be seen in Table 8.

Table 8 – Test for normality Kolmogorov-Smirnov Statistic Beer 0,072*** Coffee 0,092*** Cola 0,060** *** p < 0,01;** p < 0,05.

4.3

Gabor Granger

33 | P a g e

prices, B(P) may remain unaffected and his own brand shares decreases. Next, the median of the maximum willingness to pay elicited by the Gabor Granger method is compared to the price points discussed earlier. The median is chosen because it is less affected by outliers and due to the non-normality of the WTP values. For a crate of beer, the median WTP value of the consumers is €12,02 (μ = €11,84) which is already higher than the point at which a fast decline of demand is visible. It can be said that for a crate of beer, a supermarket is likely to have the most sales with a price ranging between €10,24 - €12,02, with demand ranging between 58% - 86%. As can be seen in Figure 6 (for more detailed figures see Appendix A), the revenue in this research is the highest at a price of €10,24 (€788,-). For a pack of coffee pads, the median WTP value is €3,49 (μ = €3,46) which is reasonably higher than the point at which a fast decline is visible in the B(P) curve. Therefore, prices ranging between €2,82 - €3,49 are likely to result in high sales figures, with a demand between 52% - 82% . The highest revenue is obtained in this research at a price of €2,82 (€206,-) as can be seen in Figure 6. Finally, for a bottle of cola prices ranging between €1,37 - €1,53 are most likely for a high demand, with a median value of €1,53 (μ = €1,64) with a demand between 56% - 81%. The highest revenue for a retailer is reached at the price of €1,37 with a revenue of €78,- (see Figure 6). Unfortunately, no figures about cost prices were available to calculate the profit/loss at the specific prices.

34 | P a g e

4.4

van Westendorp

35 | P a g e

experience the price to be too high. At OPP, least resistance against the price can be expected from the consumers.

Table 9 - Price Thresholds and Revenue from van Westendorp

Beer Coffee Cola

Indifference Price(IDP) € 10,35 € 3,03 € 1,46 Revenue IDP € 580,00 € 188,00 € 99,00 Optimal Price Point (OPP) € 9,73 € 2,45 € 2,45 Revenue OPP € 566,00 € 186,00 € 186,00 Highest Revenue Price (HRP) €11,44 € 2,85 € 1,49 Revenue HRP € 583,00 € 205,00 € 100,00

Lower Bound € 7,51 € 1,98 € 0,80

Higher Bound € 13,10 € 3,95 € 1,85

% of Total Price Range 45% 65% 47%

When the OPP is situated to the left of the OPP – that is, at a lower price – the space in between is called the “stress” in price consciousness. This stress is due to the fact that consumers’ perception of a normal price is a price that is too high. Hence, the greater the separation between the OPP and IDP, the greater the feeling that the normal prices are too high (van Westendorp, 1976 and Travis, 1982). The largest difference between the IDP and OPP is found with the coffee pads (19%), followed by cola (18%), and finally beer (6%). From theory, this infers that the retail prices are generally perceived to be too high for coffee pads and bottles of cola.

36 | P a g e

Figure 7 - Demand and Revenue Curves van Westendorp

4.5

Self-Explicated

The Self-Explicated approach results in one demand curve per product category. Also for this approach the demand curve and the revenue curve is discussed, the graphs are presented in Figure 8. For a crate of beer the price threshold where the demand is rapidly decreasing and where simultaneously the highest revenue is attained is at the price of €9,80 with a demand of 86% and a revenue of €739,-. For coffee no real price threshold is visible due to a reasonable linear decrease of the demand. However, there is a range of acceptable price ranges visible, ranging from €2,38 until €2,95 where the revenue is rather stable (€143-€142). For a pack of coffee pads the highest revenue is attained at a price of €2,36 until €2,50 (€143,-) but there is only a €1,- difference in revenue at the price of €2,95. Thus, it can be said that this price would be more attractive from retailer perspective. The demand curve for the cola category shows a price threshold at €1,14, where also the highest revenue is attained of €88,-. From retailer perspective a price of €1,47 would be more attractive because revenue only decreases with €1,-.

37 | P a g e

Figure 8 - Demand and Revenue Curves Self-Explicated

38 | P a g e

4.6

Comparing methods

To answer the main research question, namely whether the three methods to elicit WTP yield different WTP values; the data has been analyzed at product level. Because the dependent variable, WTP value, does not follow a normal distribution, non-parametric tests need to be conducted to find out whether the values and distributions differ between the methods. Because an ANOVA could not be applied due to non-normality a non-parametric median test has been conducted to find out whether the methods differ in the values that are elicited. As can be seen in Table 10, the median (M) test shows that there are statistical significant difference between the medians for the methods within the coffee (M = €3,49; X² = 7,082; p = 0,029) and cola category (M = €1,69; X² = 13,810; p = 0,001). Within the beer category (M = €12,01) there is only a marginal significant difference between the methods when comparing their median (X² = 5,764; p = 0,056). Finally, there are differences between the methods when excluding the effects of the product categories (X² = 6,422; p = 0,04).

Table 10 - Median comparison methods

Category X² Method > median ≤ median Beer 5,764* Gabor Granger 53 37 Van Westendorp 34 35 Self-Explicated 36 52 Coffee 7,082** Gabor Granger 31 52 Van Westendorp 48 35 Self-Explicated 33 33 Cola 13,810*** Gabor Granger 20 47 Van Westendorp 52 35 Self-Explicated 43 43 Aggregate 6,422** Gabor Granger 134 106 Van Westendorp 140 99 Self-Explicated 114 126 * p < 0,10; **p < 0,05; *** p < 0,01

39 | P a g e

or above the median (=1). This way the differences between the methods can be shown. The medians per method and product category can be found in Appendix C in Table 2. For comparison purposes the methods are compared on a 1-to-1 base, this way it is possible to find out which methods differ from each other. As can be seen in Table 11, the differences between the methods do not hold for every product category. First, given the median within the beer category, the GG method significantly has more observations above the median compared to the SE method (X² = 5,754; p = 0,016). Within the beer category no other differences between methods are observed with the median test. Within the coffee category there is found to be a significant difference between the GG method and the vW method.In this case the vW method significantly has more observations above the median compared to the GG method (X² = 6,980; p = 0,008). Finally, within the cola category there is a significant difference between the GG method and both the vW method as the SE method. The GG method has significantly more observations below the median compared to the vW method (X² = 13,610; p = 0,000) and to the SE method (X² = 6,312; p = 0,012). No differences have been found in the median test when comparing the vW method with the SE method for specific products.

Table 11 - Median comparison methods

Category Methods X² Phi (φ) Sig.

40 | P a g e

Finally, the differences between the methods at aggregate level are discussed. When the differences in the price levels between the products are eliminated, by means of index values, other differences between methods are visible. At aggregate level, significant difference has been found between the vW method and the SE method (X² = 5,899; p = 0,015). The vW method significantly has more observations above the median compared to the SE method. Furthermore, a marginal significant difference has been found between the GG method and the SE method (X² = 3,337; p = 0,068). The GG method has more observations above the median compared to the SE method. In the table, φ denotes the estimated of effect size and is used as a measure of strength of association in the special case of a table with two rows and two columns. The phi coefficient is proportional to the square root of the chi-square (X²) statistic (Malhotra, 2010).

Table 12 - Distribution comparison methods

Category X² Method Mean rank

Beer 5,569* Gabor Granger 123,80

Van Westendorp 139,30 Self-Explicated 112,21

Coffee 2,675 Gabor Granger 114,04

Van Westendorp 125,62 Self-Explicated 108,13

Cola 3,613 Gabor Granger 112,93

Van Westendorp 131,74 Self-Explicated 115,03 Aggregate 11,005** Gabor Granger 346,50 Van Westendorp 396,05

Self-Explicated 337,60

* p < 0,10; ** p < 0,05

41 | P a g e

This difference is due to the statistical significant difference between the vW method and the GG method W U = 24698; p = 0,009) and the difference between the vW and the SE method (Mann-W U = 24047; p = 0,002). It can be said that the mean rank of the v(Mann-W method is significantly higher compared to the other methods.

4.7

Model estimation

This part of the paper includes the model estimation per product category. For every product used in this study a model has been estimated. This resulted in different coefficients and effects of the variables on the willingness to pay. First, the model for a crate of beer is presented followed by the coffee pads and the cola, respectively. For estimation purposes dummy variables have been included. In Table 13, it is shown which category is used as base category.

Table 13 - Base Categories

Base

Gender Female

Method Gabor Granger Favorite brand A-Brand

Product Beer

4.7.1 Beer model

In this model the variables, as shown in the conceptual framework, are used to explain the variation in the price the respondents are willing to pay for a crate of beer (24 bottles). As can be seen in Table 14, most of the variables that were taken into account in this research seem not to have a significant influence.

The total amount of variation in the WTP value that is explained by the independent variable is expressed by the R² and the adjusted R². The R² of the model is 0,195, which means that the used independent variables explain 19,5% of the variation in the WTP value. The adjusted R², which is the R² with subtracted explained variation because of the number of variables used, shows that 13,9% of the variation in the WTP value is explained by the used variables.

42 | P a g e

experience has a marginal significant positive effect on the WTP value (p = 0,082) this result is in line with the expectations. However, no moderation effect has been found. Finally, there is a significant negative effect on the WTP value when the favorite brand of the respondents is among the discount brands (p = 0,000). When a customer’s favorite brand is a discount brand the WTP value for a crate of beer will decrease with €5,64. Finally, it was expected that the more price conscious people are the lower the WTP value would be.

4.7.2 Coffee model

In this model the variables are used to explain the variation in price the price the respondents are willing to pay for a pack of coffee pads (36 pads). The results of the regression can be seen in Table 14. The total amount of variation that is explained by the used variables is 15,8% (R²) and the adjusted R² shows that 9,5% of the total variation is explained by the model.

The variables with a significant effect on the WTP are discussed in the following part. As also found with the beer model, income has a significant positive effect on the willingness to pay of the respondents (p = 0,019). Furthermore, the fact that the favorite coffee brand is a private label or a discount brand has a significant negative effect on the WTP value compared to the A-brand, respectively with p = 0,000 and p = 0,021. Hence, as expected having a relative cheaper favorite brand lowers the WTP value. However, the coefficient of the private label brands is larger than the coefficient of the discount brand. This infers that consumers, whose favorite brand is a private label, are even willing to pay less than consumers whose favorite brand is a discount brand. This also might be due to the fact that the price knowledge of the respondents was significantly lower for coffee than for beer or cola. Finally, a marginal significant moderation effect has been found for category experience (p = 0,070). The moderation effect is found to be positive, which infer that the WTP is higher when consumers are more experienced in the product category.

4.7.3 Cola Model

43 | P a g e

Estimating the cola model, income has a significant positive effect (p = 0,012). The influence of favorite brand type also has a significant negative effect compared to the A-Brands, for the private label this is p = 0,000, and for the discount brands this is p = 0,000. The difference between the results of coffee and cola is that the coefficients of the favorite brands make more sense now. For cola, the coefficient of the private label (β = -0,399) is higher than it is for the discount brand (β = -0,599). This infers that consumers are willing to pay the least when their favorite brand is a discount brand. Only in this model, the effect of the van Westendorp method (compared to the Gabor Granger method), has a positive significant effect on the WTP value (p = 0,022). This infers a difference in the elicited WTP value for the van Westendorp method.

4.7.4 Aggregate Model

Here the model will be estimated on aggregate level. By means of WTP value index (WTP/mean) the different WTP values for the products can directly be compared. Further, the model that is estimated is almost the same as the other estimated models. The extra variables included are the dummy variables indicating the product, to see whether the effect on WTP value also depends on the product of interest. The total amount of variation that is explained by the used variables is 15,9% (R²) and the adjusted R² shows that 13,1% of the total variation is explained by the model.

As also found in the previous models, income has a positive significant effect on the WTP value (p = 0,000). Furthermore, it has been found that the older people get the less they are willing to pay for a product (p = 0,008). Also for the aggregate model, the favorite brand influences the WTP. For discount brand and private label a negative significant effect has been found, respectively p = 0,000 and p = 0,000. When people have no preferred brand a marginal negative significant effect has been found (p = 0,056). Next, the moderation effect of the category experience with the SE method has found to be marginally significant (p = 0,093). Finally, a significant main effect of coffee on the WTP value has been found which means that the intercept of coffee is significantly higher compared to other products (p = 0,049).

44 | P a g e

Table 14 - Regression Results of all Products

Beer Coffee Cola Aggregate

β β β β

(Constant) 11,782*** 3,65*** 1,698*** 1,003***

Age -0,027** -0,007 -0,003 -0,002***

Dummy Variable Gender Male 0,075 0,006 -0,042 0,00

Income 0,272*** 0,068** 0,039*** 0,022***

Dummy variable for comparing vW 0,524 0,071 0,157** 0,05

Dummy variable for comparing SE 0,137 -0,106 0,047 0,002

Dummy for Private label -1,227 -0,601*** -0,399*** -0,19***

Dummy for Discount brand -5,644*** -0,373** -0,599*** -0,233***

Dummy for No Preference 0,023 -0,132 -0,081 -0,043*

Category Experience 0,413* -0,135 0,00 0,00

Price Consciousness -0,158 -0,049 -0,067 -0,021

Method * Price Consciousness vW -0,103 0,092 0,061 0,019

Method * Price Consciousness SE -0,349 0,092 0,075 0,01

Method * Category Experience vW 0,092 0,207* 0 0,00

Method * Category Experience SE -0,394 0,048 0 0

Dummy variable for Coffee - - - 0,066**

Dummy variable for Cola - - - 0,012

Method vW * Product Coffee - - - -0,025

Method SE * Product Coffee - - - -0,033

Method vW * Product Cola - - - 0,036

Method SE * Product Cola - - - 0,024

45 | P a g e

4.8

Normality

The saved residuals where tested for normality, which is one of the assumptions that has to be met when estimating a model with linear regression. The Kolmogorov-Smirnov (KS) statistic for the estimated model for beer is 0,043 and has been found to be insignificant (p > 0,05). The KS statistic for the estimated model for coffee is found to be 0,054 which is again insignificant (p > 0,05). Also the KS statistic for the estimated model for cola (KS = 0,046) and on aggregate level (KS = 0,025) is found to be insignificant (p > 0,05). All the results show that normality of the residuals can be assumed. Hence, this causes no problems for the estimation and interpretation of the model results.

4.9

Multicollinearity

Another assumption that has to be met is the absence of multicollinearity. Multicollinearity might inflate the estimated variance of the slope coefficients. The cause of the multicollinearity might be due to too less variation in the method variable (Leeflang et al. 2000) or by the inclusion of the moderating effect of price knowledge on the WTP value. When the variable inflation factor (VIF) is >10, multicollinearity is a problem and has to be dealt with. In this case all VIF levels < 5, so for the estimated models multicollinearity is no issue.

4.10

Heteroscedasticity

46 | P a g e

4.11

Model Fit

To be certain that adding variables to the model increases the fit to the data, all estimated models are compared to the intercept only model. For model comparison the Akaike Information Criterion (AIC) has been used which is expressed as follows:

43! = (−299:+ 2;)/>

where N are the number of observations and k are the number of parameters used. The information criterion seeks to incorporate, in model selection, the divergent considerations of accuracy of estimation and the “best” approximation to reality (Leeflang et al. 2000). The model with the lowest AIC value shows the best fit to the data. For beer, the intercept model the AIC is 1077,270 and for the full model the AIC is 908,623 which is found to be significantly different from each other (X² = 46,715; p = 0,000). For coffee, the intercept model the AIC is 483,576 and for the full model the AIC is 422,479 which is found to be significantly different from each other (X² = 34,854; p = 0,002. For cola, the intercept model the AIC is 272,276 and for the full model the AIC is 201,820 which is found to be significantly different from each other (X² = 55,888; p = 0,000). This means that the addition of the variables does improve the fit of the data. For aggregate level, the intercept model the AIC is -183,989 and for the full model the AIC is -233,250 which is found to be significantly different from each other (X² = 109,973; p = 0,000). Hence, the addition of the variables does improve the fit of the data.

4.12

Classes