Consumption Motives and Hypothetical Biases

(1)

Consumption Motives and

Hypothetical Biases

The differing effect of consumption motivation on the magnitude

of hypothetical bias in choice-based conjoint analyses.

Author: Samradha Sanjeev

Student number: S2884577

Email: s.sanjeev@student.rug.nl

Supervisor: Dr Felix Eggers

Co-assessor: Dr Keyvan Dehmamy

Faculty of Economics and Business

University of Groningen

Duisenberg Building, Nettelbosje 2, 9747 AE Groningen, The

Netherlands

P.O. Box 800, 9700 AV Groningen, The Netherlands

http://www.rug.nl/feb

(2)

List of Figures

1 Conceptual Model . . . 14 2 Experimental Design . . . 15

3 Example Choice Set . . . 18

4 Market Simulation . . . 30

5 Market Simulation Hedonic Condition . . . 34

6 Market Simulation Utilitarian Condition . . . 35

List of Tables

1 Consumption motivation and Hypothetical Bias based on ELM and ATT 13 2 Attributes and Attribute Levels . . . 17

3 Sample Comparison of Group Composition . . . 23

4 Mean HED/UT Scores . . . 25

5 Model Goodness of Fit Results . . . 27

6 M5 Output Key . . . 27

7 M5 Output . . . 28

8 Price and None Option Coefficients . . . 29

9 Attribute Importance . . . 31

10 Price and None Option Per Condition . . . 33

11 M1, M2 and M3 Output . . . 46

(4)

Abstract

(5)

1 Introduction

The choice between options occurs on a daily basis, be it which meal option seems the most attractive for dinner, which type of camcorder is the best buy or even which holiday cruise would be the most enjoyable. This choosing process that occurs can be analysed using Choice Based Conjoint (CBC) analyses, a method of choice experimentation. It is a useful method of studying customers’ decision processes and determining trade-offs (Rao, 2014). The ideology behind CBC analyses is based on the assumption that prod-ucts are essentially a bundle of attributes (Eggers, Sattler, Teichert, & V¨olckner, 2018). This method of analysis also brings light to attribute importance, price sensitivity and valuation, such as willingness-to-pay (WTP). CBC analyses have been used in various sectors including the public sector, health care and technological industries. It is impor-tant to have valid valuation as CBC analyses are one of the most popular market research methods for product designing and pricing (Natter & Feurstein, 2002) and is commonly used in the market (Eggers et al., 2018).

CBC analyses tend to be conducted on a hypothetical basis. This indicates that in prac-tice, consumers are asked to make their choices from hypothetical bundles of goods. This is due to a couple of reasons; firstly, this analysis is useful when a product has reached a prototype stage (Eggers et al., 2018). Therefore, making a physical final product is not always executable, and thus it is made easier to rely on a supposed bundle of goods. Secondly, there are high costs associated with making the collection of products with different attribute variations. Due to its hypothetical nature, CBC analyses can result in significant differences between revealed preferences (results from experimental testing) and revealed behaviour (actual market evidence) (Brownstone & Small, 2005). Therefore, it faces external validity issues known as hypothetical bias.

Hypothetical bias refers to the difference between stated (hypothetical) and actual valu-ations, in which actual valuations refer to statements of value collected from experiments with real economic commitments (Hensher, 2010; List & Gallet, 2001). The hypothetical bias affects the external validity of CBC analyses as it does not allow for accurate valua-tion of attributes. Due to the importance of valid valuavalua-tion, researchers have attempted to mitigate the bias through various methods, one of which, that has shown promise, is incentive-aligned (IA) CBC analyses.

(6)

a random choice set (Ding, Grewal & Liechty, 2005). Ding et al., (2005) explain that its ability to mitigate the hypothetical bias stems from the presence of salience, a condition required in conjoint analyses, according to Smith’s (1976) induced value theory (Ding et al., 2005). By using an IA CBC analysis, salience can be induced into the participant as there are consumption consequences to the choices they make in the experiment. This leads the participant to evaluate their choice more accurately before selecting a preference.

Another stream of research focuses on the origins of hypothetical bias. It has been at-tributed to various factors that stem from human psychology such as uncertainty (Hen-sher, Rose, & Beck, 2012) or social desirability (Norwood & Lusk, 2011). However, no research has been conducted on the effect of consumption motivation on hypothetical bias. Literature suggests that there are two basic reasons for consumption behaviours: hedonic and utilitarian consumption. Hedonic consumption involves affective gratification involv-ing sensory and experiential consumption such as fun, pleasure and excitement (Batra & Ahtola, 1991; Dhar & Wertenbroch, 2000; Hirschman & Holbrook, 1982). While utilitar-ian consumption involves instrumentality and functionality related to the expectations of consequences (Batra & Ahtola, 1991; Dhar & Wertenbroch, 2000). A key difference be-tween hedonic and utilitarian consumption is that while utilitarian consumption focuses on utility maximisation, hedonic consumption focuses on subjective symbols of experience and fantasy (Hirschman & Holbrook, 1982). Therefore, one can expect that product pref-erences and choices in hypothetical CBC analyses would differ based on the consmption motivation. However, a question arises as to whether preferences change when conse-quence to choices is induced. Further resulting in a differing level of hypothetical bias between hypothetical CBC analyses and IA CBC analyses for the different consumption motivations.

One stream of research suggests that the difference can occur due to the level of evalua-tion. In this sense, higher evaluation occurs in IA CBC analyses compared to hypothetical CBC analyses. For example, if a utilitarian product is preferred in a hypothetical CBC analysis and an IA CBC is also conducted on the same choice sets, two theories propose opposing results. Dhar and Wertenbroch (2000) posit that consumers, who increase eval-uation tend to prefer hedonic products (Dhar & Wertenbroch, 2000). This is based on the idea that hedonic products create positive stimuli, enhancing favourableness of judgement (Tybout & Artz, 1994). However, on the other hand, the Elaboration Likelihood Model (ELM) suggest that when evaluation increases, a consumer would prefer the utilitarian product (Petty, Cacioppo, & Schumann, 1983).

(7)

et. al. (1983) indicates the importance for empirical testing to determine which con-sumption motivation would lead to a lower or higher hypothetical bias. Therefore, this research will focus on the question: How does the differentiation between hedo-nic and utilitarian consumption motivation affect hypothetical bias in choice based conjoint analyses?

The implications of this research would benefit both academic and managerial marketing fields. Firstly, results from this research would provide empirical evidence in support of either one of the contradicting arguments mentioned previously. Next, it would also sup-port the supposition of consumption motivation having an effect on hypothetical biases in choice experiments. Finally, companies with the intention of conducting hypothetical CBC for their products can take into consideration the magnitude of hypothetical bias depending on their consumption motive. If it is evident that the specific motive may not face high levels of bias, the need for a more expensive IA testing can be reconsidered.

(8)

2 Literature Review

2.1 Incentive Aligned CBC analyses and Hypothetical Bias

Recent literature has proposed that salience is a fundamental prerequisite in CBC anal-yses (Ding et al., 2005). They explain that in hypothetical CBC analanal-yses, participants can receive a fixed compensation for their participation. Hence, there is no relationship between the compensation the participant receives and their responses in the experiment. The lack of direct connection between these aspects shows that there is a lack of con-sequential thinking. From which, the lack of salience contributes to hypothetical bias. Evidence from other researchers have suggested the importance of salience as participants step away from a socially desirable presentation of themselves to a more realistic one in the context of generosity and risk-seeking behaviour (Camerer & Hogarth, 1999).

In addition, increased salience can increase price sensitivity of the participant which is more realistic for WTP predictions (Wl¨omert & Eggers, 2016). Since these analyses tend to have a price attribute in order to deem valuation, an IA experiment can provide a participant with a budget. By making the participant consider the price attribute as well, they are made to consider consumption consequences and its effect on their budget when making choices (Ding et al., 2005). Based on the alternative they have been re-warded, the remainder of the budget can be provided to them as a monetary reward. It is important to note that in CBC analyses choice sets, a none option can be selected if the participant would choose not to acquire any of the product alternatives. In the case they do not select any product alternative (none option is selected), they would be given the full budget value instead.

(9)

Another distinction is in the increased preference for the none option in IA CBC analyses compared to hypothetical CBC analyses. In hypothetical analyses, participants show in-terest in acquiring one of the options presented in the choice sets. However, when asked to make an actual purchasing decision (IA analyses), they behave differently (Ding et al., 2005; Miller, Hofstetter, Krohmer, & Zhang, 2011). From Ding et al. (2005), results show that in the hypothetical analyses, the none option is selected 25% of the time while in an IA analysis, it is selected close to 50% of the time. These results are supported in the study conducted by Miller et al. (2011). The combination of this literature demonstrates that IA CBC analyses is a more accurate method of valuation for product alternatives based on their attributes, price and the none option. Hence, the following hypotheses are proposed:

H1a: A hypothetical experimental condition leads to an overvaluation of WTP, hence

moderating the relationship between product attributes and price and product utility.

H1b: An incentive aligned experimental condition leads to a higher preference for the none

option, hence moderating the relationship between the none option and product utility.

2.2 Hedonic and Utilitarian Consumption and Utility

Consumption behaviour can be separated into two basic reasonings: hedonic and utilitar-ian. A hedonic product can be defined as a good that is consumed based on an affective experience due to its multi-sensory, fantasy and emotive elements (Batra & Ahtola, 1991; Dhar & Wertenbroch, 2000; Hirschman & Holbrook, 1982). For example, the result of a consumer smelling a perfume is the generation of internal imagery that include sounds and tactile sensations (Hirschman & Holbrook, 1982). Due to this sense of imagery, he-donic products are not seen as objective entities of utility maximisation, but rather as subjective symbols of experience and fantasy (Hirschman & Holbrook, 1982).

(10)

However, consumption behaviour being based on this bidimensional reasoning does not mean that they are mutually exclusive (Batra & Ahtola, 1991). Instead, consumption of a specific product will lead to evaluation of certain attributes more than others. For example, a person buying toothpaste for a rational purpose would put more evaluative weight on utilitarian attributes such as cleaning and whitening. Meanwhile, a person buying toothpaste for experiential reasons, would put more evaluative weight on hedonic attributes, such as flavour (Batra & Ahtola, 1991). This suggests that attribute impor-tance can differ based on the consumption motivation.

Batra and Athola (1991) empirically tested the Semantic Differential (SD) scale sug-gested by Osgood, Suci and Tannenbaum (1957). It measures consumers attitudes to-wards brands and consumption behaviours. Using a common factor analysis, two factors emerged, hedonic and utilitarian. Hedonic loaded heavily on attitudes related to pleasant-ness, agreeableness and niceness. While utilitarian heavily loaded on usefulpleasant-ness, extent of benefit and importance. This analysis being applied on four brands, Pepsi, Cadillac, Comet cleanser and Listerine, showed that cleansers such as Comet are superior in utili-tarian aspects such as usefulness and beneficial. Soft drinks such as Pepsi are superior in hedonic aspects such as pleasant and nice. Their conclusion from this research was that there is nomological validity for the hedonic and utilitarian distinction (Batra & Ahtola, 1991).

Crowley, Spagenberg and Hughes (1992) develop this research further. They conduct a factor analysis on 24 product categories with different consumption motivations, in-cluding categories used by Batra and Athola (1991). Even though, they are unable to replicate the results of Batra and Athola (1991), they provide evidence that the hedonic and utilitarian dimensions of consumer attitudes are separate and measurable (Crowley, Spangenberg, & Hughes, 1992).

(11)

Implications from these studies provide evidence that a prominent distinction can be made between hedonic and utilitarian products (Batra & Ahtola, 1991; Crowley et al., 1992; Voss et al., 2003). Consumers attitudes towards these products differ based on this distinction. Hence, specific attributes can be dominant during evaluation of the product (Batra & Ahtola, 1991; Hirschman & Holbrook, 1982). This would suggest that in CBC analyses, attribute importance can differ based on the product when determining utility. Specifically, if a product is being purchased for hedonic reasons, then attributes that focus of hedonic aspects will be more important. Therefore, the effect of product attributes on utility can be segregated into hedonic and utilitarian product attributes.

H2: When the product being analysed is being purchased due to utilitarian reasons,

utili-tarian attributes increase in importance.

H3: When the product being analysed is being purchased due to hedonic reasons, hedonic

attributes increase in importance.

2.3 Incentive Alignment and consumption motivation

When an IA CBC analysis is being used rather than a hypothetical CBC analysis, the participant is required to consider their choice decisions more accurately due to the con-sumption consequences present. Therefore, higher evaluation in an IA CBC analysis compared to hypothetical CBC analyses can be expected. When taking consumption motivation into consideration, the previous section has suggested that there exists differ-ing consumption behaviours between hedonic and utilitarian products. A question arises as to what extent consumption behaviour changes when the level of evaluation changes between hypothetical and IA CBC analyses. Two main contradicting perspectives provide varying angles of explanation for this question.

2.3.1 ELM

(12)

This form of reasoning is much like the reasoning used in utilitarian consumption due to the focus on functionality.

The second route is known as the peripheral route. Unlike the central route where the pros and cons are considered, this route is dependent on positive or negative cues (Petty et al., 1983). This route bases argumentation on affective reasoning, such as the shape and colour of a disposable razor (Petty et al., 1983). This form of reasoning is similar to that of hedonic consumption due to the focus on the experience and attractiveness of the product. Opposing to the central route, attitudinal changes occurring through the peripheral route tend to be temporary and less predictive of behaviour (Petty et al., 1983). Therefore, the central route is associated with quality argumentation while the peripheral route is associated with weaker argumentation.

ELM extends its theory into the concept of involvement level. Petty et al. (1983) indicate that the level of involvement plays an important role in information processing. They suggest that higher involvement is related to greater personal relevance and consequence (Petty et al., 1983). Furthermore, several papers have established that there is positive association between elaboration and involvement (Brown & Stayman, 1992; Lord, Lee, & Sauer, 1995; Miniard, Bhatla, & Rose, 1990; Mittal, 1990; Petty et al., 1983). Therefore, high involvement can be a proxy for cognitive elaboration (Voss et al., 2003).

Situations of higher involvement can be induced by consequence, which is present in IA CBC analyses. The Elaboration Likelihood Model (ELM) suggests that in situations of higher involvement, quality arguments (utilitarian arguments) have a higher impact on attitude than weak arguments (hedonic arguments) as a person would use the cen-tral route (high elaboration) rather than peripheral route (low elaboration) (Petty et al., 1983). This would suggest that in hypothetical CBC analyses, if a product is be-ing selected for utilitarian reasons, a product with more utilitarian attributes would be preferred. Likewise, in the case of hedonic consumption reasons, a product with more hedonic reasons would be preferred. However, if the same analyses are conducted with IA, the product with more utilitarian attributes would be preferred.

2.3.2 Acquisition Task Theory

(13)

connected to want preferences while products high on utilitarian values can be connected to should preferences (Dhar & Wertenbroch, 2000). Continuing in their research, they perpetuate that in acquisition tasks, lower levels of elaboration lead to a relative salience and preference for utilitarian features (Dhar & Wertenbroch, 2000).

This reasoning is based on the counter reasoning for preference of hedonic features due to more elaboration. Previous research suggests elaboration on positive stimuli can enhance favourableness of judgement (Tybout & Artz, 1994). In addition, the elaboration leads to a greater influence of easily imaginable attributes, making them more salient (Shiv & Huber, 2000). Basing on the definition of hedonic products being more emotionally invoking (Hirschman & Holbrook, 1982), products with stronger hedonic features would become more attractive in the situation of higher elaboration and products with stronger utilitarian features would be preferred in the situation of lower elaboration (Dhar & Wertenbroch, 2000).

Considering theory on consumption motivation, in a hypothetical CBC of a utilitarian product, it can be expected that functional attributes and instrumentality of the product are being mainly considered. However, when the same product is used in an IA CBC analysis, elaboration is induced due to consequences. Based on Dhar and Wertenbroch’s (2000) research, the higher elaboration in IA CBC analyses would lead to the preference of products with stronger hedonic attributes. Meanwhile, in the case of a hypothetical CBC analysis on a hedonic product, a product with more hedonic attributes would be preferred. Unlike the previous case, this preference would remain the same even in a IA CBC analysis. Therefore, the hypothetical bias for utilitarian products would be higher compared to hedonic products.

Table 1: Consumption motivation and Hypothetical Bias based on ELM and ATT

Consumption Reason Hypothetical CBC IA CBC

ELM Hedonic Hedonic Product Utilitarian Product

Utilitarian Utilitarian Product Utilitarian Product

ATT Hedonic Hedonic Product Hedonic Product

Utilitarian Utilitarian Product Hedonic Product

ATT = Acquisition Task Theory

(14)

However, as previously explained, the ELM suggests the opposite that, with reasoned arguments, products with utilitarian features would be preferred in a situation of higher elaboration. Table 1 illustrates the consumption reason that would lead to a higher hy-pothetical bias according to each theory. This demonstrates that there are contradictions in reasoning for the preference of a specific consumption motivation in CBC analyses. A possible origin for this contradiction could root from the basis of research of Dhar and Wertenbroch (2000) being decision tasks. Nevertheless, these opposing views leave a research gap to test the relationship between a hypothetical experimental condition and consumption motivation. Therefore, the two opposing hypotheses can be put forward in the attempt to support one and reject the other.

H4: When the product being analysed is being purchased due to utilitarian reasons, there

is a lower hypothetical bias compared to when it is a hedonic product.

H5: When the product being analysed is being purchased due to hedonic reasons, there is

a lower hypothetical bias compared to when it is a utilitarian product.

Based on the hypotheses proposed in the literature review, a conceptual model illustration is presented in Figure 1.

(15)

3 Methodology

3.1 Experimental Design

In order to test the aforementioned hypotheses, multiple CBC analyses will be conducted in varying experimental conditions. The sample is made up of 247 randomly selected participants based in the USA, using the Prolific Academic (ProA) platform. This crowd sourcing platform has shown to have participants that have a higher level of attention and a lower propensity of dishonest behaviour and cheating, compared to platforms such as, MTurk (Peer, Brandimarte, Samat, & Acquisti, 2017). The sample is randomly dis-tributed into a 3 x 2 design. The six experimental conditions are based on utilitarian, hedonic and neutral consumption motivations. Within which, a segregation between the hypothetical and IA experimental conditions are made (Figure 2). For ease of explana-tion, hedonic, utilitarian and neutral conditions will herewith referred to as conditions while hypothetical and IA conditions will be referred to as situations.

A between-subject design is used based on the conclusions of Johansson-Stenman & Sveds¨ater (2008). Their study used a within-subject design for hypothetical and incen-tivised testing. They concluded that due to a person’s need for consistency, experiment results can be affected (Johansson-Stenman & Sveds¨ater, 2008). Furthermore, additional conclusions made suggests that the hypothetical bias is underestimated in within-subject designs. In order to control for a person’s need for consistency, a between-subject design is used.

(16)

3.2 Utilitarian versus Hedonic Condition

In order to make the segregation between hedonic and utilitarian consumption motives, a single product is used under different pretences. The same product needs to be used in all conditions in order to reduce external factors that may affect choice preferences. Therefore, a product with both hedonic and utilitarian attributes is employed. According to Batra and Athola (1991), toothpaste is a product with both hedonic and utilitarian benefits (Batra & Ahtola, 1991). From their research on toothpaste, instrumental at-tributes identified are plaque/tartar, decay and whitening, while sensory atat-tributes are mouth freshness and taste/flavour. On this basis, toothpaste will be used as the product.

To induce the hedonic and utilitarian condition, the consumption motive of the partici-pant is based on the respective conditions. Botti and McGill (2011) explain that a hedonic consumption motivation is induced by focusing on purchasing the product due to sheer pleasure reasons. While a utilitarian consumption motivation is induced by focusing on purchasing the product due to goal serving reasons. Using the similar priming script of Botti and McGill (2011), the participants in the hedonic condition are told they are purchasing a toothpaste to use before socialising with friends. Meanwhile, the partici-pants in the utilitarian condition are told they are purchasing the toothpaste based on their dentist’s advice. As a control group, the neutral condition is not given any priming except explaining some facts about toothpaste. Detailed scripts for each condition are presented in Appendix I.

3.3 Hypothetical versus Incentive Aligned Situation

The segregation between hypothetical and IA situations are based on the the type of compensation the participant receives. In the hypothetical situation, the participant is given their compensation regardless of their performance in the CBC analysis. The com-pensation is worth $5 and is given to one in every 10 participants based on a random draw.

(17)

3.4 Attributes & Levels

Based on existing options of toothpastes in the market, the attributes that will be used are presented in Table 2. The first three attributes as explained earlier represent hedonic attributes and the following three represent utilitarian attributes. The price attribute is determined by reducing market prices by 20%. By doing this, it reduces situations in which the participant chooses the none option in every set in order to simply attain the compensation and make the purchase on their own.

Table 2: Attributes and Attribute Levels

Attribute Levels

Flavours Fennel, Peppermint, Watermelon

Freshness Max Fresh, Cooling Blast, Fresh breath (regular freshness)

Colour Triple colour paste, white paste, black paste

Whitening Advanced whitening, regular whitening, no whitening

Cleaning Deep clean (Advanced cavity, tartar and enamel protection),

Everyday cleaning (Regular cavity, tartar and enamel

protection), Sensitive clean (Gentle cavity, tartar and enamel protection)

Ingredients Contains fluoride (regular), Fluoride-free (natural), Paraben-free (ecofriendly)

Price $2.00, $2.80, $3.60, $4.40

3.5 Choice Design

(18)

Figure 3: Example Choice Set

3.6 Measures

3.6.1 Manipulation Measures

Three main manipulations are measured and calculated. First, the manipulation be-tween the hedonic and utilitarian condition. Next, the hedonic and utilitarian scores of the product attributes. Finally, the manipulation between the hypothetical and IA situation. Each of these are elaborated on in this section.

First, in order to determine if the hedonic and utilitarian consumption motive is success-fully manipulated on the sample groups, participants are asked, “when you were choosing among different types of toothpaste, what attributes were you looking for?” To answer this question, participants are made to employ the adjective pairs HED/UT scale re-searched by Voss, Spangenberg and Grohmann (2003)(Appendix III). This seven point Likert scale consists of 10 adjective pairs. The first five pairs represent utilitarian adjec-tives, in which selecting low scores implies the presence of utilitarian consumption motive. The last five adjective pairs represent hedonic adjectives, in which selecting higher scores implies the presence of hedonic consumption motives.

(19)

Next, to ensure the hedonic and utilitarian attributes are perceived as such by the par-ticipants, the HED/UT scale is used again. Participants are asked to scale the different attributes of toothpaste against the different adjective pairs. For example, the phrase, ”I think the product attribute ‘Flavour’ in a toothpaste is...”, is shown to the participant. From which, they are made to answer the seven point HED/UT Likert scale. By calculat-ing the average score for each attribute, the hedonic (score > 3) and utilitarian attributes (score < 3) are determined. In the case that attributes are considered both hedonic and utilitarian, a Principal Component Analysis (PCA) will be conducted to give a weighting to the hedonic and utilitarian aspects of the attributes instead.

Finally, in order to determine if the manipulation between the hypothetical and IA situa-tion successfully alters the level of involvement, Mittal’s (1989) Involvement Scale (MIS) is used (Mittal, 1989). This scale entails four seven-point Likert scale statements (Ap-pendix IV). MIS is a scale to measure purchase-decision involvement by collecting the average scores for the scale and forming an overall purchase-decision involvement (PDI) score. It is used to test if evaluation in fact increases in the IA CBC situation compares to the hypothetical situation. A score > 3 suggests high involvement and a score < 3 suggest low involvement. The PDI scores of the participants in the different situations are compared to check if there is a significant difference. This is important to determine as it would show that the behavioural change between the hypothetical and IA CBC situation is present.

3.6.2 Hypotheses Measures

Based on the hypotheses presented in the literature review, two measures are calculated to develop results. Specifically, WTP and attribute importance. WTP is calculated to determine results for H1, H4, H5. Comparisons of the results from WTP calculations for

each experimental condition are examined. Specifically, a comparison is made between hypothetical and IA situation for both consumption motivations. Significant differences between hypothetical and IA situations demonstrate the moderating effect of the hy-pothetical condition as postulated in H1, hence suggesting the presence of hypothetical

bias. The differing magnitude of hypothetical bias between the consumption motivations will demonstrate the moderating effect of consumption motivation on hypothetical bias postulated in either H4 or H5.

(20)

on, the first function shows the preference for acquiring the toothpaste at any price level, ceteris paribus. The second shows the preference of the none option. The market share is calculated using (Eggers et al., 2018):

P (ji|J) = exp(Vi) P exp(VJ) (1) in which: Vi = βixi (2) Where: Vi = utility of product i

x = attribute level of product i

β = partworth utility for attribute level of product i

Attribute importance is used to determine results for H2and H3. Attribute importance is

calculated for the different consumption motivations. Under each motivation condition, the respective attributes are expected to increase in importance, as postulated in H2 and

H3. Attribute groups (hedonic and utilitarian attributes) are compared between the

con-sumption motivations. If there is significant difference between attribute importance for both attribute groups between the motivation conditions, H2 and H3 are supported.

At-tribute importance ωn is calculated using the relative range of partworth utilities (Eggers

et al., 2018):

ωn=

max(βn) − min(βn)

PN

i=1(max(βn) − min(βn))

(21)

3.7 Model Definitions

In order to test the hypotheses, five models are specified. Each model includes addi-tional independent variables that allow for analysis for the different hypotheses. Since the dependent variable is nominal with four levels (four choice options per choice set), multinomial logistic regression models are used (Equation 1).

Model 1 (M1) is a partworth model consisting of all the attribute level variables

(effect-coded) and the none option. Model 2 (M2) holds the same independent variables as M1

except uses price as a linear function. In order to test for H1, Model 3 (M3) expands on

M2 by adding interactions between the IA situation and price and the none option.

Model 4 (M4) focuses on testing H2 and H3, therefore, builds on H3 by adding

interac-tions between HED/UT scores, experimental condiinterac-tions and the most preferred attribute levels. In order to implicitly measure the effect on attribute importance, dummy coded attribute levels are used instead of effect coded levels. The least preferred attribute lev-els determined from the previous modlev-els, are used as the base levlev-els for each attribute dummy coding. By doing this, the estimate for the most preferred attribute level is the range value that is used to calculate attribute importance. Hence, the interactions re-quired in the model to test H2 and H3 are only for the most preferred attribute levels.

This model accounts for the possibility of requiring the factor variables for the hedonic and utilitarian aspects of the attributes. M5 is developed to test H4 and H5, therefore,

interactions between the experimental condition and incentive alignment are done on price and the none option.

M1 =⇒ Vi1=β1Fennel + β2Peppermint + β3Max Fresh + β4Cool Blast+

β5Triple Colour + β6White + β7Whiten Adv + β8Whiten Reg+

β9Deep Clean + β10Everyday Clean + β11Fluoride + β12Fluoride Free+

β13Price2.00 + β14Price2.80 + β15Price3.60 + β16None + ε

M2 =⇒ Vi2=Vi1(β1−12) + β13Price + β14None + ε

(22)

M4 =⇒ Vi4=β1Flavour1 + β2Flavour2 + β3Fresh1 + β4Fresh2 + β5Colour1+

β6Colour2 + β7Whiten1 + β8Whiten2 + β9Clean1 + β10Clean2 + β11Ingredient1+

β12Ingredient2 + β13Price + β14None + β15(Price*IA) + β16(None*IA)+

β17(Flavour1*FlavourU T I) + β18(Flavour1*FlavourHED)+

β19(Fresh1*FreshU T I) + β20(Fresh1*FreshHED)+

β21(Colour1*ColourU T I) + β22(Colour1*ColourHED)+

β23(Whiten1*WhitenU T I) + β24(Whiten1*WhitenHED)

β25(Clean1*CleanU T I) + β26(Clean1*CleanHED)+

β27(Ingredient1*IngredU T I) + β28(Ingredient1*IngredHED)

β29(Flavour1*FlavourU T I*Utilitarian) + β30(Flavour1*FlavourU T I*Hedonic)+

β31(Flavour1*FlavourHED*Utilitarian) + β32(Flavour1*FlavourHED*Hedonic)

β33(Fresh1*FreshU T I*Utilitarian) + β34(Fresh1*FreshU T I*Hedonic)+

β35(Fresh1*FreshHED*Utilitarian) + β36(Fresh1*FreshHED*Hedonic)

β37(Colour1*ColourU T I*Utilitarian) + β38(Colour1*ColourU T I*Hedonic)+

β39(Colour1*ColourHED*Utilitarian) + β40(Colour1*ColourHED*Hedonic)

β41(Whiten1*WhitenU T I*Utilitarian) + β42(Whiten1*WhitenU T I*Hedonic)+

β43(Whiten1*WhitenHED*Utilitarian) + β44(Whiten1*WhitenHED*Hedonic)

β45(Clean1*CleanU T I*Utilitarian) + β46(Clean1*CleanU T I*Hedonic)+

β47(Clean1*CleanHED*Utilitarian) + β48(Clean1*CleanHED*Hedonic)

β49(Ingredient1*IngredU T I*Utilitarian) + β50(Ingredient1*IngredU T I*Hedonic)+

β51(Ingredient1*IngredHED*Utilitarian) + β52(Ingredient1*IngredHED*Hedonic) + ε

M5 =⇒ Vi5=Vi4+ β53(Price*Hedonic) + β54(Price*Utilitarian) + β55(None*Hedonic)+

β56(None*Utilitarian) + β57(Price*IA*Hedonic) + β58(Price*IA*Utilitarian)+

(23)

4 Data Analysis

4.1 Pre Analysis Examination

Before conducting analyses to determine results for the hypotheses, an initial examination is conducted on the data for participant composition across different experimental groups (Table 3). After which, the manipulation results (HED/UT scale and PDI score) are calculated. Finally, the models specified in Section 3.7 are compared against each other to determine the most appropriate model for interpretation and analysis in the results section.

Table 3: Sample Comparison of Group Composition

Full Sample

Hedonic Utilitarian Neutral

IA Hyp. IA Hyp. IA Hyp.

N 247 53 43 48 28 41 34

Gender(Male)a 38% 26% 35% 45% 43% 20% 28%

Av. Household Size

3 3 3 3 3 3 3

Av. Age Range 25-29 30-34 25-29 25-29 30-34 25-29 25-29

(Highest frequency age range) (≤ 17) (≤ 17) (≤ 17) (≤ 17) (≤ 17) (25-29) (≤ 17) Most common US Region (Proportion) SE (24%) SW (34%) NE (28%) NE (27%) SW (36%) SW (37%) SW (29%)

a_{Not inclusive of 2 participants who selected “Other/Prefer not to answer”.}

SE = Southeast, SW = Southwest, NE = Northeast

The sample is distributed in the six different experimental groups. The demographics of the groups are tested to determine if there are any relationships between the proportions of the groups. This is done through Chi-squared tests of independence and Fisher Exact tests where appropriate. The tests are run based on the samples used for each hypothesis to ensure the proportions do not interfere with the analysis.

(24)

situa-tions, which are related to H1. A Fisher exact test is conducted for gender and the results

show that the null hypothesis, suggesting independence of proportions between variables, cannot be rejected (p = 0.332). Chi-squared tests are conducted for household size, age range and US regions. Results from these tests show that the null hypotheses, suggesting no relationship between variables, cannot be rejected (X2 _{= 2.219, p = 0.818), (X}2 ₌

6.632, p = 0.760), (X2 _{= 2.942, p = 0.816) respectively. Hence, the conclusion can be}

made that there is no relationship between the demographic variables and being in the different situations.

Next, the demographics are compared based on being in the utilitarian, hedonic and neu-tral conditions, which are related to H2 and H3. Chi-squared tests are conducted for all

the demographics: gender, household size, age range and US regions. Results show that the tests for gender (X2 _{= 4.196, p = 0.380), household size (X}2 _{= 13.559, p = 0.194),}

age range (X2 _{= 22.191, p = 0.330) and US regions (X}2 _{= 5.397, p = 0.943) cannot reject}

the null hypotheses. This suggests that there is no relationship between the demographic variables and being in the different conditions.

Finally, the demographics are compared based on their grouping as shown in Table 3, which are related to H4and H5. Chi-squared tests are conducted for all the demographics:

gender, household size, age range and US regions. Results show that the tests for gender (X2 _{= 8.297, p = 0.405), household size (X}2_{= 19.998, p = 0.458), age range (X}2_{= 42.877,}

p = 0.349) and US regions (X2 _{= 15.537, p = 0.904) cannot reject the null hypotheses.}

This suggests that there is no relationship between the demographic variables and being in the different groups. From this, an overall conclusion can be made that there is no relationship between the demographic variables and the various groupings for situations and conditions, hence are independent of each other. Therefore, no extra measures to account for such an issue need to be considered.

4.2 Calculating Manipulations

4.2.1 HED/UT Scale

(25)

(F = 0.982, p = 0.376). An independent sample t-test is conducted on the hedonic and utilitarian condition. However, once again, the results show that there is no significant difference between the conditions (t = -1.3762, p = 0.171). This implies that the ma-nipulation is not sufficient to prime the samples into the different consumption reasons. These results can affect the results of H2 to H5.

In order to determine which attributes are considered hedonic and which are considered utilitarian, the average scores of the HED/UT scales are calculated (Table 4). Once again, ”enjoyment” is removed as it is used as an attention check. At initial sight, it is evident that none of the attributes are heavily hedonic nor are they heavily utilitarian. To further investigate this, an ANOVA test is conducted to determine if the HED/UT scores for each attribute are significantly different from each other. Results from the ANOVA test show that the scores are significantly different (p < 0.001). Upon closer inspection, using a Tukey HSD test, it is evident that only cleaning and ingredients are not significantly different from each other (p = 0.999). However, this can be expected as both attributes are expected to be utilitarian.

Table 4: Mean HED/UT Scores

Flavour Freshness Colour Whitening Cleaning Ingredients

2.924 2.648 3.617 2.315 1.866 1.891

Since the values are surrounding the central score of three, it can be concluded that the attributes are considered both hedonic and utilitarian. This result does not meet the expectations of the attribute manipulation. Therefore, a PCA is conducted for the HED/UT scales of each attribute. Before conducting the PCA for each attribute, KMO, Bartlett and communality tests are conducted for each attribute’s HED/UT scale. Re-sults from these tests showed that a PCA would be appropriate (KMO > 0.5, significant Bartlett tests and communality > 0.4). To determine the number of factors to be formed, the eigenvalues (> 1), total explained variance (> 5%), cumulative variance (> 60%) and scree plots are inspected. In addition, VARIMAX rotations are employed to ensure vari-able loading on a single factor.

(26)

the factors for each attribute are formed. These results support the research conducted by Voss et al., (2003). By doing this, the attributes differ in weighting between hedonic and utilitarian rather than being defined as one or the other. These weightings will be used in the modelling to test for H2 and H3.

4.2.2 PDI Scale

In order to calculate PDI score of each participant, their average score across the four MIS scales are calculated. The average PDI score for the IA situation = 4.129 while for the hypothetical situation = 4.188. This does not match the expectations as higher involvement is expected to be in the IA situation. An independent samples t-test is con-ducted to test if the PDI scores are significantly different between the situations. The test result is not significant (t = 0.36833, p = 0.713), therefore, the null hypothesis of equal means is not rejected. Since the PDI scores are not significantly different between the situations, the manipulation check does not support the theory suggested by the ELM and ATT, hence, the results for H1, H4 and H5 can be affected.

4.3 Model Comparison

The five models specified in Section 3.7 are compared to determine the best model to further estimate for results. The best model is determined based on the model fit and quality through its log-likelihood, McFadden Pseudo-R2

adj and information criterion. A

likelihood ratio test comparing M1 and M2 is not significant (p = 0.1274). Hence, M1

(larger model) is not significantly better than M2. Even though the partworth model has

a better log likelihood score (Table 5), the linear model is used to continue building the following models in order to reduce the size of the models. By doing this, more parsimo-nious models are being formed by pushing for simplicity but also completeness (Leeflang, Wieringa, Bijmolt, & Pauwels, 2015).

The results (Table 5) show that M3 improves from M2. The likelihood ratio tests shows a

significant result (p= 0.003725), therefore, showing that M3 (larger model) is significantly

better than M2. M4 is better than M3 even taking into consideration the larger size of

the model. A likelihood ratio test allows for the conclusion that M4 (larger model) is

significantly better than M3 (p < 0.001). Finally, the results shows that M5 is the best

model as it has the highest Pseudo-R2_adj and lowest AIC scores. The likelihood ratio test results show that M5 (larger model) is significantly better than M4 (p < 0.001).

(27)

Table 5: Model Goodness of Fit Results

Model # Parameters Log-Likelihood Pseudo-R2

adj AIC M1 16 −2938.9 0.1370 5909.828 M2 14 −2941 0.1370 5909.949 M3 16 −2935.4 0.1381 5902.764 M4 52 −2848.6 0.1529 5801.186 M5 60 −2816.2 0.1600 5752.455

5 Results

Based on the data analysis, M5 is chosen as the best model. Hence it is further used for

estimation to test the hypotheses of this research (Table 7).

Table 6: M5 Output Key

Variable Description

PepperD Peppermint (Dummy coded)

WaterD Watermelon (Dummy coded)

MaxFreshD Max Fresh (Dummy coded)

FreshBD Fresh Breath (Dummy coded)

WhiteD White paste (Dummy coded)

TCPD Triple Colour Paste (Dummy coded)

AdvD Advanced Whitening (Dummy coded)

RegD Regular Whitening (Dummy coded)

DeepD Deep Clean (Dummy coded)

SenseD Sensitive Clean (Dummy coded)

FluorideD Contains Fluoride (Dummy coded)

(28)

(29)

5.1 Hypothesis 1

H1a: A hypothetical experimental condition leads to an overvaluation of WTP, hence

moderating the relationship between product attributes and price and product utility. H1b: An incentive aligned experimental condition leads to a higher preference for the none

option, hence moderating the relationship between the none option and product utility.

In order to test this hypothesis, the coefficients from Table 7 are used to calculate the overall effect (main effect + interaction) of price and the none option in the hypothetical and IA condition (Table 8). Using the utilities for each price level and none option, the predicted market shares can be determined (Figure 4).

Table 8: Price and None Option Coefficients

Attributes Coefficients

Price (Hypothetical Situation) -0.1908075

None Option (Hypothetical Situation) 0.32913488

Price (IA Situation) -0.1526313

None Option (IA Situation) 0.90120426

(30)

Figure 4: Market Simulation

At the same time, the option to not acquire the product in both conditions has a higher probability than acquiring the product at any price level. This is evident as the none op-tions do not intersect with the price funcop-tions. Therefore, the consideration WTP cannot be determined without extrapolation. When focusing on the none option between condi-tions, the probability of choosing not to acquire the product is higher for participants in the IA condition compared to the hypothetical condition. This implies that participants overestimate their WTP in the hypothetical condition compared to the participants in the IA condition. This is evident as having a lower probability of selecting the none option suggests a higher probability of willingness to purchase the product. Since the interaction of the none option and IA is significant and positive (p < 0.05), the effect of the none option is significantly higher in the IA condition.

Based on these results, the conclusion can be made that without having an incentive to accurately select preferences between the different choice options, participants overesti-mate their WTP for the product. However, the overestimation through the price utility is not significantly higher for the hypothetical group, yet at the same time, the effect of the none option is significantly lower. Therefore, H1a is not significantly supported while

(31)

5.2 Hypotheses 2 & 3

H2: When the product being analysed is being purchased due to utilitarian reasons,

utili-tarian attributes increase in importance.

H3: When the product being analysed is being purchased due to hedonic reasons, hedonic

attributes increase in importance.

To test H2 and H3, the individual attribute ranges are calculated in the different

con-ditions to obtain attribute importance. This is calculated using: Direct effect + (ef-fect*attribute factor*condition). Within which, specific attribute importance for when attributes are considered more hedonic (Hed) or utilitarian (Util) are presented (Table 10). From Table 7, a majority of the interactions are not significant. This suggests that there is no significant difference in the effect of attributes in the different conditions based on their hedonic and utilitarian scores. However, the analysis of attribute importance is continued to test the hypotheses.

Table 9: Attribute Importance

Attributes

Flavour Freshness Colour Whitening Cleaning Ingredients

(32)

Table 9 shows that in the neutral condition when scoring is higher on the utilitarian scale, flavour, colour and whitening are the most important attributes. This is consistent in the neutral condition when hedonic scoring is higher. This implies that the aforementioned attributes are considered both hedonic and utilitarian, consistent with the results from the HED/UT scale (Table 4). Though the direct effects of flavour, colour and whitening are significant (p < 0.001), the difference in utility in the utilitarian and hedonic scoring are not significant (p = 0.833 and p = 0.329) for colour and (p = 0.876 and p = 0.200) for whitening. This suggests that when the utilitarian and hedonic scales are considered, there is no significant change in the utility of the aforementioned attributes.

When the utilitarian score is higher, ingredients increases in importance compared to when it is scored on a hedonic scale. However, the utility does not increase significantly (p = 0.922). The vice versa occurs for cleaning, in which, it increases in importance when compared on the hedonic scale than utilitarian scale and the utility increases significantly (p < 0.05). This suggests, that ingredients is considered more utilitarian and cleaning is considered more hedonic. Though this does not replicate the results from Table 4, it can be expected as the results in Table 4 suggest that the HED/UT scores of cleaning and ingredients are not significantly different from each other.

Testing the proposition of H2, the attributes that are of higher importance in the neutral

condition (when utilitarian scores are higher) are compared to the attributes of higher importance in the utilitarian condition (when utilitarian scores are higher). Results show that the same attributes; flavour, colour and whitening are the most important attributes. That being said, only flavour increases, though its utility does not increase significantly (p = 0.120), while colour remains similar and whitening decreases, though its utility does not decrease significantly (p = 0.641).

At the same time, ingredients does not increase in importance, however, it decreases, though its utility does not decrease significantly (p = 0.101). This suggests that the more utilitarian attributes do not increase in importance when the product is being pur-chased for utilitarian reasons. At the same time, attributes that are considered the most important in the neutral condition that have higher mean HED/UT scores, suggesting higher hedonic aspects, do not change significantly when in the utilitarian condition, hence, rejecting H2.

Testing H3, the attributes that are of higher importance in the neutral condition (when

(33)

hedonic condition (when hedonic scores are higher). The table shows that the same attributes flavour, colour and whitening are the most important attributes both in the neutral and hedonic condition. The importance of flavour increases by 32.21% and its increase in utility is significant (p < 0.001). In this comparison even though the same attributes are the most important, the individual importance of colour and whitening decrease in the hedonic condition. However, their utilities do not significantly decrease (p = 0.581 and p = 0.823 respectively).

At the same time, cleaning and ingredients decrease in importance by 17.22% and 0.96% respectively. The utility of cleaning decreases significantly (p < 0.05) while the utility of ingredients does not (p = 0.846). This can be expected as they have relatively lower HED/UT scores, suggesting they are more utilitarian. However, at the same time, the increase in the more hedonic attributes, colour and whitening, does not occur. This suggests that when the product is being purchased for hedonic reasons, hedonic attributes do not increase in importance , rejecting H3.

5.3 Hypotheses 4 & 5

H4: When the product being analysed is being purchased due to utilitarian reasons, there

is a lower hypothetical bias compared when it is a hedonic product.

H5: When the product being analysed is being purchased due to hedonic reasons, there is

a lower hypothetical bias compared when it is a utilitarian product.

In order to test H4 and H5, the coefficients from Table 7 are used to calculate the

overall effect of price and the none option for the IA and hypothetical group within the neutral, hedonic and utilitarian condition (Table 10). Since the neutral situation has the same coefficients as presented in Table 8, only the hedonic and utilitarian condition coefficients are calculated. This is calculated using: Direct effect + (effect*condition) + (effect*condition*IA).

Table 10: Price and None Option Per Condition

Attributes Hedonic Utilitarian

Price (Hypothetical Situation) -0.311 -0.473

None Option (Hypothetical Situation) -0.366 -1.131

Price (IA Situation) -0.174 -0.247

(34)

Table 7 shows that there is no significant difference (p = 0.144) in the effect of price being in the hedonic condition compared to being in another condition. Within the hedonic condition, there is no significant difference (p = 0.215) in the effect of price between the IA and hypothetical group. Meanwhile, the effect of the none option is significant and negative (p < 0.05) in the hedonic condition. This shows that the none option has a significantly lower utility for participants in the hedonic condition compared to other conditions. Within the hedonic condition, there is no significant difference (p = 0.140) in the effect of the none option between the IA and hypothetical situations.

When focusing on the utilitarian condition, Table 7 shows that there is a significant neg-ative difference (p < 0.01) in the effect of price, being in the utilitarian condition. This implies that the utility of price is significantly lower for participants in the utilitarian condition compared to other conditions. Likewise, the effect of the none option is sig-nificantly lower (p < 0.001) for the utilitarian condition compared to being in another condition. Within the utilitarian condition, there is no significant difference (p = 0.057) in the effect of price between the IA and hypothetical situations. At the same time, there is no significant difference (p=0.155) in the effect of the none option between the IA and hypothetical situations.

(35)

The utilities for all price levels and the none option are used to calculate the predicted market shares for the hedonic (Figure 5) and utilitarian (Figure 6) conditions. The hedonic condition shows that in the hypothetical situation, the none option always has a higher probability of selection over acquiring the product and as price increases, this probability increases as well. However, in the IA situation, the probability of acquiring the product is always higher than the none option. Based on the trajectory of the lines, the consideration WTP for the IA situation will occur at a price greater than $4.40 while the same would occur at less than $2.00 for the hypothetical situation. This demonstrates that there is a hypothetical bias in the hedonic condition.

Figure 6: Market Simulation Utilitarian Condition

The utilitarian condition shows that in the IA situation the probability of acquiring the product starts higher than the none option, however, is less preferred than the none op-tion after $2.08. Similarly, in the hypothetical situaop-tion, the probability of acquiring the product starts higher than the none option but decreases to a lower probability at $2.39. The intersection of the price functions with their respective none options represent the consideration WTP for each situation. The difference in the situational consideration WTP , in which the hypothetical WTP is higher than the IA WTP, suggests that there is a hypothetical bias in the utilitarian condition.

(36)

for utilitarian reasons than hedonic reasons. Hence, this supports H4 but rejects H5,

suggesting that the theory posited by ELM is supported.

6 Discussion

Past research has delved into the origins of the hypothetical bias investigating factors such as, uncertainty (Hensher et al., 2012) and social desirability (Norwood & Lusk, 2011). These factors focus on the psychological characteristics of the participants. However, this research takes a different angle by investigating the effect of differing the consumption behaviour instead. Namely, this report focuses on researching the differential effect of consumption motivation, specifically, hedonic and utilitarian, on hypothetical biases in choice based conjoint analyses. The research investigates this question is through three main lines of testing.

Through the first hypothesis, the presence of a hypothetical bias is partially supported regardless of consumption motivation through the significant difference in the preference of the none option. The none option in this research is selected 12.82% of the time in the IA situation and 10.66% in the hypothetical situation. The magnitude of difference does not reach the levels compared to the results presented by Ding et al. (2005) and Miller et al. (2011), 50% and 25% respectively. At the same time, there is lack of significant difference between the price functions between the IA and hypothetical situation. Such a results could be attributed to an experimental limitation, specifically, the product chosen for analysis.

(37)

these products could be more suitable for analysis.

The next pair of hypotheses reject the proposition of attribute importance being affected based on consumption motivation. Though these hypotheses are based on theory (Batra & Ahtola, 1991; Crowley et al., 1992; Voss et al., 2003) the result occurrence can be attributed to two main factors. First, the attributes do not mimic results suggested by Batra & Ahtola (1991) in which flavour and freshness are expected to be hedonic at-tributes, while cleaning and whitening are expected to be utilitarian attributes. Hence, the attributes in each given condition do not increase in importance as expected.

The second reasoning can be due to the HED/UT manipulation check only partially being achieved, in which the HED/UT score is, as expected, higher in the hedonic compared to the utilitarian condition but not significantly (p = 0.171). Hence, suggesting that partic-ipants may not have been primed sufficiently to differ in consumption motivation. This leads to the lack of differing attribute importance. However, the HED/UT manipulation check continues to be examined later in this section.

Finally, the last hypotheses support the proposition of a difference in hypothetical bias based on consumption motivation. Since H4 and H5 are opposing hypotheses suggested

by different theoretical background only H4 is successfully supported, namely, the theory

suggested by the ELM. This postulates that in a situation of higher involvement (IA situation) utilitarian reasoning is emphasised (Petty et al., 1983). This is demonstrated in the results by the lower magnitude of hypothetical bias in the utilitarian condition compared to the hedonic condition as depicted in Figure 5 and 6. Though this result occurs, it is interesting given that there are no significant differences in price or the none option between situations in either condition. As explained earlier, due to toothpaste not being an elastic good, consequential thinking which would denote to higher involvement might not be induced in the purchase of the product. This is also evident in the PDI scores not having a significant difference in involvement between the IA and hypothetical situation.

(38)

factors such as, social desirability (Norwood & Lusk, 2011) cannot be ruled out.

In addition, the manipulation for the conditions has shown to not be sufficient for a be-havioural change to occur, therefore, it would not have been expected to have a differing magnitude of hypothetical bias. To investigate this, we consider the HED/UT scale de-veloped by Voss et al. (2003). Their suggested set of 10 adjective pairs to use as scales is from a larger list of scales developed. The larger list of scales have adjective pairs such as not sensuous/sensuous and unpleasant/pleasant. Given that Batra and Athola (1991) explain the hedonic aspects of toothpaste to be ”sensory”, perhaps the use of these scales instead to determine the consumption motivation of toothpaste would have been more appropriate. Therefore, though the manipulation check suggests that the condition priming is not sufficient, perhaps the priming does occur but is not captured by the scale.

Overall, the results from this research have been able to partially support previous the-ory in the presence of hypothetical biases in CBC analyses. However, given the lack of visible success in the manipulations of the experiment, conclusions for some hypothe-ses are not possible. That being said, investigation into these hypothehypothe-ses have drawn interesting results when comparing hypothetical biases based on differing consumption motivations. From which, it calls for further implications and future research possibilities.

7 Implications and Limitations

The main implications of this research are multi-faceted and relate to the construction of choice based conjoint analyses. To make the results from the analysis more representa-tive of the real market, an IA method should be adopted. However, the challenges faced with applying said method, such as cost and lack of a physical product, would still exist. Therefore, considering the price sensitivity of the product can help determine the neces-sity of an IA method. Furthermore, considering that the condition manipulation might be successful, the IA method being used for the products being purchased for utilitarian reasons may not be as necessary compared to products being purchased for hedonic rea-sons.

(39)

larger sample size to accommodate the different experimental condition or repeating the experiment to increase reliability should be considered.

Next, from Table 5 the Pseudo R2

adj scores are low (< 0.2) and a possible reason for

this can be attributed to the need for segmentation within the sample. According to Johansson-Stenman & Sveds¨ater (2008), women have a higher mean WTP compared to men. Therefore, latent classing can be used to determine groups within the sample to increase the model fit. Running an initial test on M2, while considering class

member-ship based on gender show the significant results. With a better log likelihood (-2646.8) compared to M2 and a classification error of 0.022, an initial conclusion can be made that

segmentation based on gender can improve the model performance and fit.

Another limitation is the lack of a pre-tests. Pre-tests were not conducted for the HED/UT scales for the manipulation and attributes. Hence, the HED/UT scale as a manipulation check and the attributes selected to represent hedonic and utilitarian im-portance may not be the most appropriate. In addition, current toothpaste preferences and purchases could have an effect on the extent of hedonic or utilitarian view of tooth-paste. For example, some brands, such as Hello and Tom’s of Maine, advertise more hedonic attributes and variety, such as, flavour. Consumers who purchase from this brand more often could seek out such attributes when selecting preferences in the exper-iment, even if they are assigned to the utilitarian condition.

From this, there is room for further research. Specifically, research into differences be-tween segments can specify which groups of people demonstrate varying hypothetical biases between consumption motivations. Such research would benefit companies as it would identify when an incentive alignment is necessary not just from consumption mo-tivation but from target markets as well. Another line of research could delve into the effect of price elasticity of the products being analysed and the level of hypothetical bias. This line of research could contribute to current research on the effects of consumption motivation and hypothetical bias in CBC analyses.

(40)

8 References

Batra, R., & Ahtola, O. T. 1991. Measuring the hedonic and utilitarian sources of con-sumer attitudes. Marketing Letters, 2(2): 159–170.

Bazerman, M. H., Tenbrunsel, A. E., & Wade-Benzoni, K. 1998. Negotiating with your-self and losing: Understanding and managing conflicting internal preferences. Academy of Management Review .

Botti, S., & McGill, A. L. 2011. The Locus of Choice: Personal Causality and Satisfac-tion with Hedonic and Utilitarian Decisions. Journal of Consumer Research .

Brown, S. P., & Stayman, D. M. 1992. Antecedents and Consequences of Attitude To-ward the Ad: A Meta-Analysis. Journal of Consumer Research .

Brownstone, D., & Small, K. A. 2005. Valuing time and reliability: Assessing the evi-dence from road pricing demonstrations. Transportation Research Part A: Policy and Practice.

Camerer, C. F., & Hogarth, R. M. 1999. The Effects of Financial Incentives in Exper-iments: A Review and Capital-Labor-Production Framework. Journal of Risk and Uncertainty .

Crowley, A. E., Spangenberg, E. R., & Hughes, K. R. 1992. Measuring the hedonic and utilitarian dimensions of attitudes toward product categories. Marketing Letters, 3(3): 239–249.

Dhar, R., & Wertenbroch, K. 2000. Consumer choice between hedonic and utilitarian goods. Journal of Marketing Research, 37(1): 60–71.

Ding, M. I. N., Grewal, R., & Liechty, J. 2005. Incentive-Aligned Conjoint Analysis. Journal of Marketing Research , 42(2): 67–82.

Dong, S., Ding, M., & Huber, J. 2010. A simple mechanism to incentive-align conjoint experiments. International Journal of Research in Marketing , 27(1): 25–32.

Eggers, F., Sattler, H., Teichert, T., & V¨olckner, F. 2018. Choice-Based Conjoint Anal-ysis. Handbook of Market Research .

(41)

Hensher, D. A. 2010. Hypothetical bias, choice experiments and willingness to pay. Transportation Research Part B: Methodological , 44(6): 735–752.

Hensher, D. A., Rose, J. M., & Beck, M. J. 2012. Are there specific design elements of choice experiments and types of people that influence choice response certainty? Jour-nal of Choice Modelling , 5(1): 77–97.

Hirschman, E. C., & Holbrook, M. B. 1982. Hedonic Consumption: Emerging Concepts, Methods and Propositions. Journal of Marketing .

Johansson-Stenman, O., & Sveds¨ater, H. 2008. Measuring hypothetical bias in choice experiments: The importance of cognitive consistency. B.E. Journal of Economic Analysis and Policy , 8(1).

Leeflang, P. S. H., Wieringa, J. E., Bijmolt, T. H. A., & Pauwels, K. H. 2015. Modeling Markets: Analyzing Marketing Phenomena and Improving Marketing Decision Making. Springer .

List, J. A. 2001. Do explicit warnings eliminate the hypothetical bias in elicitation. The American Economic Review .

List, J. A., & Gallet, C. A. 2001. What experimental protocol influence disparities between actual and hypothetical stated values? Environmental and Resource Eco-nomics, 20(3): 241–254.

Lord, K. R., Lee, M. S., & Sauer, P. L. 1995. The combined influence hypothesis: Central and peripheral antecedents of attitude toward the Ad. Journal of Advertising , 24(1): 73–85.

Miller, K. M., Hofstetter, R., Krohmer, H., & Zhang, Z. J. 2011. How should consumers’ willingness to pay be measured? an empirical comparison of state-of-the-art approaches. Journal of Marketing Research , 48(1): 172–184.

Miniard, P. W., Bhatla, S., & Rose, R. L. 1990. On the Formation and Relationship of Ad and Brand Attitudes: An Experimental and Causal Analysis. Journal of Marketing Research , 27(3): 290.

Mittal, B. 1989. Measuring Purchase-decision involvement. Psychology & Marketing .

(42)

Natter, M., & Feurstein, M. 2002. Real world performance of choice-based conjoint mod-els. European Journal of Operational Research , 137(2): 448–458.

Norwood, F. B., & Lusk, J. L. 2011. Social desirability bias in real, hypothetical, and inferred valuation experiments. American Journal of Agricultural Economics, 93(2): 528–534.

Peer, E., Brandimarte, L., Samat, S., & Acquisti, A. 2017. Beyond the Turk: Alternative platforms for crowdsourcing behavioral research. Journal of Experimental Social Psychology .

Petty, R. E., Cacioppo, J. T., & Schumann, D. 1983. Central and Peripheral Routes to Advertising Effectiveness: The Moderating Role of Involvement. Journal of Con-sumer Research , 10(2): 135.

Rao, V. R. 2014. Applied conjoint analysis. Springer .

Shiv, B., & Huber, J. 2000. The Impact of Anticipating Satisfaction on Consumer Choice. Journal of Consumer Research .

Strahilevitz, M., & Myers, J. G. 1998. Donations to Charity as Purchase Incentives: How Well They Work May Depend on What You Are Trying to Sell. Journal of Consumer Research .

Tybout, A. M., & Artz, N. 1994. Consumer Psychology. Annual Review of Psychol-ogy .

Voss, K. E., Spangenberg, E. R., & Grohmann, B. 2003. Measuring the hedonic and utilitarian dimensions of consumer attitude. Journal of Marketing Research , 40(3): 310–320.

Wakefield, K. L., & Inman, J. J. 2003. Situational price sensitivity: The role of consump-tion occasion, social context and income. Journal of Retailing .

(43)

9 Appendices

9.1 Appendix I: Condition Manipulation Script

9.1.1 Utilitarian Condition

On the next pages we would like to present different types of toothpaste to you in order to learn more about what features you value most. When answering these questions please imagine that recently you have been suffering from gum irritation and tooth aches. Your dentist tells you that these are just symptoms of poor dental hygiene and recommends changing your toothpaste. Therefore, you go to the store to pick one up. You can choose one of the options displayed keeping in mind that your goal is to help fix your troubles.

9.1.2 Hedonic Condition

On the next pages we would like to present different types of toothpaste to you in order to learn more about what features you value most. When answering these questions please imagine that you have been working really hard during the year and achieved important successes in your goals. You think that going to celebrate with friends is a great way to reward yourself. However, you always like to brush your teeth before meeting up with people and you have recently run out of toothpaste. Therefore, you go to the store to pick one up. You can choose one of the options displayed keeping in mind that your goal is to get one that makes you feel good before meeting your friends.

9.1.3 Neutral Condition

On the next pages we would like to present different types of toothpaste to you in order to learn more about what features you value most. Toothpastes came into general use in the 19th century. In 1896, the first toothpaste that was sold in a collapsible tube, after having been sold in glass jars for many years. Most people use toothpaste on a daily basis, and The World Health Organization (WHO) has recommended to use 6 tubes of toothpaste and at least 4 toothbrushes per year. Now you are going to a store to pick one toothpaste up. You can choose one of the options displayed.

9.2 Appendix II: Situation Manipulation Script

9.2.1 Hypothetical Situation

We will now show you 12 more of these questions in which we ask you to choose between types of toothpastes.

Consumption Motives and Hypothetical Biases