• No results found

Measuring the Effect of Brands and Customer Reviews on Uncertainty by Eliciting Choice Probabilities

N/A
N/A
Protected

Academic year: 2021

Share "Measuring the Effect of Brands and Customer Reviews on Uncertainty by Eliciting Choice Probabilities"

Copied!
78
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Measuring the Effect of Brands and Customer Reviews on

Uncertainty by Eliciting Choice Probabilities

(2)

Measuring the Effect of Brands and Customer Reviews on

Uncertainty by Eliciting Choice Probabilities

Master Thesis

Faculty of Economics and Business

MSc Marketing Management and Marketing Intelligence

20th June 2016

First Supervisor: dr. Keyvan Dehmamy Second Supervisor: dr. Felix Eggers

(3)

Abstract

Uncertainty is omnipresent in many consumer decisions causing consumers to delay purchase decisions. By eliciting choice probabilities in a conjoint analysis it is tested whether brands and customer reviews reduce the uncertainty for innovative products. Furthermore, it is investigated whether customer reviews change the brand perceptions of consumers. While the effect of brands is captured by the level of brand equity, customer reviews are represented by the average rating (valence) and the number of reviews available (volume) for the product.

Results reveal that brands do not decrease uncertainty, while the valence and volume of customer reviews even increase uncertainty for the product and its characteristics. However, a negative interaction between brands and customer reviews has been identified, implying that both acting together are capable of reducing uncertainty. Lastly, despite their increasing effect on uncertainty, customer reviews were found to increase the choice-share towards the no-choice option by almost 50%.

(4)

Preface

The finalization of this master thesis also represents the end of my double-track master in Marketing Management and Marketing Intelligence at the University of Groningen. My journey in Groningen now stops after two wonderful years and I will keep them in good memory.

Writing this master thesis and in particular conducting its research has been the greatest academic challenge for me so far. In the last months, I experienced both ups and downs, including many moments of pride and enthusiasm but also feelings of doubts and uncertainty. Therefore I would like to thank all the people without whom I would not have mastered this academic challenge.

Above all my parents earn my deepest gratitude for motivating me with their endless faith and cheering me up in times of doubt. Moreover, they made it possible for me to enjoy five years of academic education.

Furthermore, I would like to thank my first supervisor dr. Keyvan Dehmamy for his guidance throughout the thesis and his helpful feedback. At this point I would also like to thank the members of my master group for the good cooperation, valuable discussions and exchanges.

Another big thank you goes to my friends who always had a sympathetic ear for my concerns. A special thanks goes to my housemate Yvonne, who was the first to listen to me, laugh with me and suffer with me. Sandra and Tobi, you were also two of my biggest supporters. Sandra, you had to go through the same and I think we managed quite well to motivate each other through troublesome times. And Tobi, you managed to support me, provide me with input and always make me laugh despite the great distance.

Finally, I owe a big thank you to everyone who took the time and filled in my survey and especially those who again reached out to their own network and acquired even more participants.

Groningen, 20th June 2016

(5)

Table of Contents

List of Figures and Tables ... V List of Abbreviations ... VI

1. Introduction ... 1

2. Theoretical foundation ... 4

2.1 Uncertainty at innovations ... 4

2.1.1 Categorizing the innovation under study... 8

2.2 The effect of brand equity on uncertainty ... 9

2.3 The effects of customer reviews on uncertainty ... 13

2.4 The interaction between brand equity and customer reviews ... 18

2.5 Hypotheses and conceptual model ... 19

3. Methodology ... 23

3.1 Research method ... 23

3.2 Study design ... 23

3.2.1 Attributes and levels ... 24

3.2.2 Experimental Design ... 26

3.3 Data collection ... 28

3.4 Model specification ... 28

4. Results ... 31

4.1 Descriptive statistics ... 31

4.2 Relevance of no-choice option ... 32

4.3 Conjoint analysis using HB ... 32

4.3.1 Conjoint analysis without reviews ... 32

4.3.2 Conjoint analysis with reviews ... 35

4.4 Testing hypotheses ... 38

4.4.1 Effect of brands on uncertainty ... 38

(6)

4.4.3 Interaction effect between brands and customer reviews ... 40

4.4.4 Overall effect of customer reviews ... 41

5. Discussion ... 43

5.1 Conclusion of findings ... 43

5.2 Managerial implications ... 47

(7)

List of Figures and Tables

Figures

Figure 1 iSkin can be worn on the finger, forearm, hand and ear (Weigel et al., 2015) ... 8

Figure 2 Conceptual Model ... 22

Figure 3 Drawing of part-worths for valence and volume of reviews ... 35

Tables Table 1 Typology for types of innovation according to Chandy and Tellis (1998) ... 5

Table 2 Six attributes to form an intention to buy an innovative product according to Rogers (1983) and Holak & Lehmann (1990) ... 6

Table 3 Extract of the studies on product ratings based on Moe and Trusov (2011) and supplemented by studies since 2011 ... 17

Table 4 Attributes and their levels ... 26

Table 5 Average attribute importance for conjoint analysis 1 and 2 ... 34

Table 6 Comparison of model indicators for part-worth, valence linear, volume linear and valence and volume linear ... 35

Table 7 Paired t-tests to test Hypothesis 1 ... 39

Table 8 Number of respondents who decreased their variance of preferences from conjoint 1 to conjoint 2 (n=62) ... 40

Table 9 Paired t-tests to test Hypothesis 3 (valence as linear parameter) ... 41

Table 10 Paired t-tests to test Hypothesis 3 (valence as part-worth) ... 41

Table 11 Paired t-tests for the attribute brand after the introduction of customer reviews ... 42

(8)

List of Abbreviations

CBC choice-based conjoint

FMCG fast-moving consumer goods HB Hierarchical Bayes

LL log likelihood

M mean

min minutes

NR no reviews

(9)

1. Introduction

“There is nothing certain, but the uncertain.” (Proverb)

Uncertainty is a phenomenon widely researched in many areas such as Finance (Hirshleifer, 1965), Social Sciences (Pfeffer, 1976) and Consumer Behavior (Taylor, 1974). The reason is that in those areas consumers are faced with difficult decisions involving a certain level of uncertainty in their everyday life, such as investments, gambling and purchase decisions. Uncertainty describes situations in which the current state of knowledge does not allow to determine the nature of things, the outcomes and the probabilities for these possible outcomes (Rogers, 1973). This research focuses on uncertainty in consumer behavior, more specifically on the adoption of innovative products. Here, the radicalness of an innovation determines the level of knowledge needed in order to resolve uncertainty and reluctance towards a new product. The more radical the innovation is, the more knowledge and information are needed (Dewar & Dutton, 1986).

As a consequence, companies should focus on providing the necessary knowledge and information to legitimate their product on a cognitive level (Aldrich & Fiol, 1994). If they fail to do so, at worst the product will not be viable on the market, at best, consumers will likely defer the decision to avoid this uncertain situation and will rather choose a product they are familiar with (Dhar, 1997). For that reason, it is not surprising that uncertainty is negatively related to purchase intention as it might dampen consumer motivation to take action in the market place (Shiu, Walsh, Hassan & Shaw, 2011).

(10)

Does brand equity reduce uncertainty for an innovative product?

If consumers are uncertain and information is difficult to obtain, they often seek the opinion from others (Huang & Chen, 2006). One marketing tool inspired by social influence is word-of-mouth (WOM). Due to the emergence of online retailing, WOM became electronically available which facilitated the access to opinions from other consumers worldwide. Customer reviews play a major role in this, as recent studies show: According to an online survey conducted by Weber Shandwick and KRC Research (2012) 65% of potential consumers selected a brand which was not in their consideration set due to the availability of customer reviews. Moreover, in an international study 5,000 shoppers were asked for their three most important sources of information they use for making a buying decision. Online ratings and reviews (52%) were named ahead of advice from family and friends (40%) as well as advice from store employees (12%) (Cisco Internet Business Solutions Group, 2013). In the context of innovation, social influence is the requirement for a product to diffuse through the market. Customer reviews can be seen as a means of diffusion as consumers share their experiences with others about the new product and thereby spread information about the product. Consequently, the second research question of this thesis is:

Do customer reviews reduce uncertainty for an innovative product?

In order to capture the effectiveness of customer reviews, two metrics are applied: valence and volume of customer reviews. Whereas a high rating might persuade a consumer about the quality of the product, the number of reviews might signify the stage in the diffusion process and social acceptance of the product. For this reason, the second research question can be divided in two sub questions:

Does the valence (average rating) of customer reviews reduce uncertainty for an innovative product?

Does the volume (number) of customer reviews reduce uncertainty for an innovative product?

(11)

Does the valence of customer reviews affect the impact of brands on uncertainty for an innovative product?

For this research, an innovative product was chosen which can be categorized as a technological breakthrough due to its outstanding technology, but missing customer benefits (Chandy & Tellis, 1998). The product was chosen intentionally to guarantee a high level of uncertainty and verify whether it can be reduced by the two marketing tools.

This thesis contributes to existing research in the following way. First, innovative products and the two constructs, brands and customer reviews, have been studied separately but have hardly been studied in relations with one another. If they had been related to each other, the degree of innovativeness was not further determined, but generally it was referred to new products. Second, in these studies, uncertainty may have been used as an independent variable, including moderators and mediators, but to the best of the author’s knowledge never as a dependent variable. Third, as mentioned above, literature has scarcely addressed the effect of interaction between brands and customer reviews. Fourth, this study applies a conjoint analysis which elicits choice probabilities. Compared to other popular conjoint approaches, it offers the advantage that uncertainty in answers can be expressed as respondents do not have to decide for one alternative, but can distribute probabilities to all alternatives (Blass, Lach & Manski, 2010). This approach, although useful for analysis and interpretation, has not been widely employed so far. Lastly, the model is estimated as a Hierarchical Bayes model which has become widespread in marketing due to its superior results towards other methods. It benefits from the requirement of limited data to determine individual utilities and therefore provides more precise answers compared to an aggregate or latent class model (Rossi, Allenby & McCulloch, 2012).

(12)

2. Theoretical foundation

2.1 Uncertainty at innovations Defining uncertainty

In the following, basic concepts such as uncertainty, innovation types, information search and social influence are discussed to better classify the innovation at hand and to set a framework for the two marketing tools, brands and customer reviews.

The definition of uncertainty is as uncertain as the term already indicates. As a result, several definitions exist. Rogers describes uncertainty as “(…) the degree to which a number of alternatives are perceived with respect to the occurrence of an event and the relative probabilities of these alternatives” (1983, p. 6) implying a lack of predictability of the possible outcomes and consequences. Knight (1921) draws a strict line between measurable uncertainty which can be represented by probabilities and unmeasurable uncertainty which cannot. Literature refers to measurable uncertainty as risk, whereas unmeasurable uncertainty is either simply termed

uncertainty or ambiguity (Camerer & Weber, 1992; Fox & Tversky, 1995; Knight, 1921). While

risk is closely connected to utility theories and deals with perceived exposure to harm or loss on a psychological level, ambiguity arises out of missing information which is relevant and could be known (Venkatraman, Aloysius & Davis, 2006). The focus of this thesis lies on uncertainty created by missing information. However, marketing research often does not differentiate between risk and uncertainty/ambiguity (Mitchell, 1999; Taylor, 1974), which is why the following literature review also covers findings about perceived risk.

Defining innovation

(13)

technology is different from prior technologies, the second dimension determines the extent to which the product fulfills customer needs better than existing products. Table 1 shows the typology in accordance with Chandy and Tellis (1998).

Table 1 Typology for types of innovation according to Chandy and Tellis (1998)

However, literature mainly focuses on examining both extreme points of the innovation continuum, i.e. incremental and radical innovation (Garcia & Calantone, 2001). Whereas “radical innovations are fundamental changes that represent revolutionary changes in technology”, “(…) incremental innovations are minor improvements or simple adjustments in current technology” (Dewar & Dutton, 1986, p.1422-1423).

Risk and success factors of innovations

The radicalness of an innovation can also be defined in terms of risk: Since radical innovations often require new knowledge compared to incremental, consumers will vary in their familiarity and experience with the innovation and consequently perceived level of risk (Dewar & Dutton, 1986; Holak & Lehmann, 1990). The perceived level of risk for an innovation can comprise the degree of physical risk, economic risk, functional risk and lastly social risk, indicating purchaser’s concern of other people’s opinion using the item (Kleijnen, Lee & Wetzels, 2009). Based on Rogers (1983), Holak and Lehmann (1990) identified further attributes determining purchase intention for innovative products, especially with respect to technologically-intensive durable products: relative advantage, compatibility, complexity, divisibility and communicability. However, in their typology perceived level of risk only consists of product performance risk (equivalent to functional risk) and psychosocial risk (equivalent to social risk). Table 2 depicts the definitions for all six attributes.

Attributes Definitions

Relative advantage Degree to which an innovative product is perceived to be superior to those that preceded it

Compatibility Degree to which an innovation is consistent with adopters’ behavior

patterns, life-styles and values

Complexity Degree to which an innovation is perceived to be relatively difficult to understand and use

Newness of technology Customer need fulfillment

Low High

Low Incremental Innovation Market Breakthrough

(14)

Divisibility Degree to which an innovation may be tried on a limited basis or without vast initial commitment

Communicability Degree to which product results or benefits are perceived easily and

expressed readily

Perceived level of risk Degree to which product performance and/or psychosocial risks are

attributed to a product

Table 2 Six attributes to form an intention to buy an innovative product according to Rogers (1983) and Holak & Lehmann (1990)

Sources and consequences of uncertainty

Several factors can be found, when addressing the sources of consumers’ uncertainty. First, consumers are not aware of their own needs, purchase goals and acceptance levels. Second, they struggle with knowledge uncertainty which includes defining the range of alternatives and the relative importance of their attributes. Third, consumers find it difficult to make an overall evaluation of the alternatives and decide for an alternative which is referred to as choice uncertainty (Mitchell, 1999; Urbany, Dickson & Wilkie, 1989). The logical consequence of a high level of uncertainty is that a purchase becomes less likely.

(15)

Reducing uncertainty for a new product

If additional product information is not easily available to relieve uncertainty, consumers have several other options to reduce it, such as by using quality signals (Erdem & Swait, 1998; Kirmani & Rao, 2000) and heuristics (Gigerenzer & Gaissmaier, 2011) or seeking evaluations from peers who already adopted the product (Rogers, 1983). Here, brands could serve either as a quality signal or a short cut. Whereas a quality signal could be inferred by the reputation and credibility of the brand (Erdem & Swait, 2004), a short cut can include the familiarity with the brand which evokes feelings of safety and trust in the consumer and as a result facilitates decision-making (Benedicktus, Brady, Darke & Voorhees, 2010; Keller, 1993).

Likewise, seeking evaluations from others reduces uncertainty and is found to be a big influencer in decision-making (Huang & Chen, 2006). Bass (1969) acknowledges this influence by adding an imitation parameter to his well-known diffusion model for adopting new products and by distinguishing adopters in innovators and imitators. Furthermore, several experiments have been conducted to demonstrate group mimicking behavior, meaning that behavior is matched to group norms (Asch, 1956; Sherif, 1936). Reasons for this conforming behavior include being accepted, feeling safe as well as the belief that people may have better information about the product than oneself (Bonabeau, 2004; Huang & Chen, 2006). Two types of influence were identified by Deutsch and Gerhard (1955): informative and normative. Whereas informative influences occurs when information is accepted by others and is seen as “(…) evidence about reality”, normative influence occurs when one “(…) conforms with the positive expectations of another” (p. 629).

(16)

who already adopted a product (Mudambi & Schuff, 2010). Thus, they are perceived more trustworthy than marketer-generated content (Godes & Mayzlin, 2004).

2.1.1 Categorizing the innovation under study

The investigated innovation is called iSkin which is a “(…) thin, flexible, stretchable and visually customizable touch sensor that can be worn directly on the skin”1

(Weigel et al., 2015, p. 2991). It is made of silicone and can be produced in different shapes to fit different parts of the body such as the finger, hand, forearm or the ear (see figure 1).

The exceptional feature of this touch sensor is that it can be connected to other mobile devices such as cell phones, music players or smart watches and is used to operate these devices. iSkin fulfills the general definition of an innovation as it is the first gadget to use electronic skin “(…) for on-body interactions to control mobile computing devices” and is therefore new to the individual (Weigel et al., 2015, p. 2992-2993). More precisely it can be classified as a technological-breakthrough as it uses a different technology not used by existing products before, however it does not offer additional customer benefit, since mobile devices can be also operated directly.

Applying the six indicators for potential adoption of new products leads to the following. The relative advantage seems to be low as mobile devices can be also operated directly and the innovation rather appears as a gimmick. At the same time this might affect the communicability as benefits and results cannot be communicated easily due to non-availability or difficulty of describing them. In terms of compatibility, iSkin is succeeding in fulfilling the requirements, as it can be connected to already existing devices such as smartphones or smartwatches. Nevertheless, it might be questionable, whether iSkin can succeed internationally as showing skin is not compatible with cultural values of Eastern countries. The requirements for complexity are also fulfilled, since the operation of the sensor is straightforward and is equivalent to other devices. In

1https://www.youtube.com/watch?v=9cvZnhvzrBI

(17)

terms of divisibility, there is no information given, but its functionality could probably be demonstrated to the potential customer at the store. Lastly, regarding perceived level of risk, risk can be split in four factors: Physical risk could cover the issue whether iSkin does any harm to the skin functionality, while no information is given to an economic risk. Functional risk can be connected to the concept of search and experience goods. While the search attributes price, brand, operating function and placement of the gadget can be easily verified, functional risk is represented by experience attributes such as responsiveness of the skin, unintended touching, feeling on the skin or stability of the silicone. Some functional risk factors could be relieved by testing the product before purchase. Finally, iSkin might bear a social risk as it is uncertain what the social environment thinks of the individual when wearing a wired tattoo- shaped silicone sensor on the body and whether it is socially acceptable.

Overall, potential customers will face many uncertainties when considering to adopt iSkin, which is likely to be a consequence of the missing additional customer benefits and of direct product-related issues such as physical, functional and social risk. Hereafter two tools will be introduced how uncertainty can be reduced.

2.2 The effect of brand equity on uncertainty

(18)

Brands as a quality signal

Keller’s view of brand equity is based on cognitive processes in the consumer’s mind and does not include other dynamics such as the interaction between firms and consumers (Erdem & Swait, 1998). As a result, research has also followed another approach in describing brand equity, which is rooted in information economics and signaling theories (Erdem & Swait, 1998; Kirmani & Rao, 2000). The information economics approach implies that different information levels are available to different parties resulting in an information asymmetry. When partner A needs information which partner B owns, partner A makes inferences from the information he is provided by partner B and thus, inference formation should be taken into consideration by partner B (Kirmani & Rao, 2000). Applying this scenario to consumer decision-making, buyer’s uncertainty about the quality of the product provided by the seller is omnipresent. Here, quality signals can be used to reduce uncertainty. These signals come in many forms such as brand name, price, advertising and warranty (Kirmani & Rao, 2000). The main determinant to convey a quality signal with a brand is its credibility (Erdem & Swait, 1998).

Brand credibility describes the believability of the product information provided by a brand and can be subdivided in the two dimensions - trustworthiness and expertise (Erdem & Swait, 2004). Whereas trustworthiness captures the willingness to deliver what was promised, expertise captures the actual ability of succeeding in it. Trustworthiness is found to be more influential in consumer’s brand consideration and choice than expertise; however this relative importance varies from one product category to another. For categories with a higher degree of uncertainty about attributes and associated information costs as well as perceived risk of consumption, such as personal computers, the relative importance of trustworthiness is higherthan for more certain categories such as juice. Moreover, the overall impact of credibility is higher for product categories facing more uncertainty (Erdem & Swait, 2004). Further evidence suggests that besides reducing uncertainty, credibility likewise reduces information costs and increases perceived quality leading to an increase in the level of confidence associated to the brand (Aaker, 1991; Erdem & Swait, 1998; 2004).

(19)

credible. Building up reputation is effortful and is lagged in time since the former behaviors of the firm lead to consumers’ current beliefs. The power of reputation is highest for products that are alike such as commodities or cannot be seen such as services. If a firm owns a good reputation it can be referred to having goodwill, an asset comprised of brand and corporate logos, as well as customer loyalty. It is of high value since it cannot be easily manipulated by firms as a false quality signal in comparison to other cues such as price and warranty (Akdeniz, Calantone & Voorhees, 2013). It must be said that reputation can be lost very easily and restoring it requires tremendous effort (Herbig & Milewicz, 1993).

In line with the construct of brand reputation is the introduction of brand extensions. A firm uses its reputation to introduce further products under the same brand. Introducing further products serves as a quality signal due to the fact that a firm would not introduce a new product under the same brand, if it believed that its quality was not good (Wernerfelt, 1988). Advantages are that the new product will be recognized immediately and that it will benefit “(...) from the ‘halo effect’ of the brand’s established reputation” (Herbig & Milewicz, 1993, p. 22).

A reason why many approaches exist to convey quality could be related to its many facets.

Brucks, Zeithaml and Naylar (2000) showed that quality of consumer durables can be divided into six dimensions: ease of use, versatility, durability, serviceability, performance, and prestige. As a result different cues are used for evaluating these dimensions. Their results point out that price and brands are rather used to evaluate prestige, which is referred to the superiority of the product in terms of visible characteristics but also intangible characteristics such as a social component in the product. This social component represents a symbolic need which includes a need for self-enhancement and group membership (Park, Jaworski & MacInnis, 1986). Baek, Kim and Yu (2010) confirmed the importance of prestige and identified brand prestige as another brand signal. They found that brand prestige has a positive influence on perceived quality, information costs saved, and perceived risk. Additionally it was found that brand prestige has a higher impact on purchase intention than brand credibility for high-expressive products (Baek et al., 2010).

Brands as shortcuts

(20)

information, whereas signaling assumes a rational consumer who needs to be rewarded for his/her implicit commitment (Kirmani & Rao, 2000). However, one could argue that it is difficult to distinguish from the outside whether a consumer has less motivation or is rationally processing signaling cues. Certainly, it can be assumed that consumers have less motivation in a low-involvement than in a high-low-involvement decision. But even in a high-low-involvement decision brands can be used as a shortcut, if ability to process systematically is low (Chaiken & Eagly, 1989; Keller, 1993). Using brands as shortcuts includes brand recognition and brand familiarity, which are both related to Keller’s described brand awareness and brand image mentioned above.

Consumers usually buy products they recognize. The same goes for brands, with the difference that brand recognition can dominate functional attributes to a degree that perceptions of the product change. This became obvious in several blind tests with jars of peanut butter, that high quality peanut butter was preferred in the blind setting. However, if a mediocre product was combined with a familiar brand the preference in taste switched to the mediocre product (Hoyer & Brown, 1990.) There can be two interpretations for this reaction. First, the brand name dominates the taste cues or second, the familiar brand serves as a “halo effect” and changes the taste itself, so that the consumers taste the brand (Gigerenzer & Goldstein, 2011). However, this blind test has been executed for a common product which is likely to be repurchased and cannot be easily transferred to a high-involvement purchase.

Related to brand recognition is brand familiarity. Whereas brand recognition only confirms the prior exposure to a brand, brand familiarity goes a step further and comprises brand knowledge structures, more specifically brand associations within consumer’s memory (Campbell & Keller, 2003). Thus, brand familiarity reflects the degree of direct and indirect experience with the brand (Alba & Hutchinson, 1987). In the study of Laroche et al. (1996) it was shown that brand familiarity influences confidence which is the subjective certainty that the judgement of the brand is correct. Confidence in turn affects the intention to buy that brand.

(21)

studies, such as Huang, Schrank and Dubinsky (2004) who found that perceived risk was not reduced by brand familiarity for online shopping, but was rather increased.

2.3 The effects of customer reviews on uncertainty

With the emergence of the internet, a huge variety of choices and information sources have appeared for consumers (Huang & Chen, 2006; Ward & Lee, 2000). However, unlike in the offline shopping environments, products cannot be seen and touched and therefore cannot be experienced before purchase. As a result, decisions have become very complex and consumers tend to delay purchase decisions, not only because of the complexity, but also because of uncertainty about the set of alternatives and the fact that quality might not meet their expectations (Huang & Chen, 2006). In order to reduce this uncertainty and risk perceived, consumers start searching on the internet (Peterson & Merino, 2003). Here, consumers tend to trust other consumers’ opinions more than company-generated content (Chiou & Cheng, 2003; Godes & Mayzlin, 2004). As a consequence, customer reviews have become one of the most popular information source before purchase (Ludwig et al., 2013). Mudambi and Schuff refer to customer reviews as “(…) peer generated product evaluations posted on company or third party websites” (2010, p. 186).

The functional mechanism of customer reviews is related to a herding behavior and explained by social influence. Consumers are influenced by other consumers and imitate the latter (Huang & Chen, 2006). Huang and Cheng (2006) state that on the internet informational rather than normative influence plays a key role, as consumers do not need to conform to expectation of others and have more informational motives. Nevertheless, consumers can neglect their own information and follow the decisions of other people leading to information cascades.

(22)

Due to their facilitating role in the purchase process, customer reviews have been set in context with an increase in sales. Positive customer reviews were found to impact sales (Chevalier & Mayzlin, 2006; Duan, Gu & Whinston, 2008), but also simply the availability versus the absence of customer reviews with only seller-created content may result in a difference in sales (Chen & Xie, 2008). However, it has to be considered that the causality between reviews and sales goes in both directions: Sold goods lead to further customer reviews (Duan et al., 2008). Moe and Trusov (2011) disagree with the causality, though claim that sales and reviews are correlated since a “good” product is likely to receive higher sales and more customer reviews, than a “bad” product.

Besides all the positive effects of customer reviews, they have also been criticized. First, as customer reviews are publicly available, they are not free of manipulation and interested parties can easily add some favorable reviews about their products leading to reviews lacking credibility (Chevalier & Mayzlin, 2006). Second, for some products such as books or movies a self-selection bias can be found, since consumers watch only a movie or read a book, if they believe that they could like it. As a result, the reviewers are likely to have a positive bias in their evaluation compared to other consumers (Chevalier & Mayzlin, 2006; Duan et al., 2008).

Classification of customer reviews

(23)

The following will focus on two of the most investigated determinants, valence and volume (Floyd et al., 2014; Liu, 2006). Besides their popularity, another reason for choosing these two determinants is that they have led to mixed findings so far (Kostyra et al., 2016; Ludwig et al., 2013). Consequently, there is also no leading opinion whether valence or volume is more important (Floyd et al., 2014).

Valence

The valence of reviews is mostly indicated as an average rating with a rating on a five-star scale with 1 as the worst and 5 as the best score (Moe & Trusov, 2011). The rating of reviews can be seen to have a persuasive effect on consumers, since it shapes their evaluations and attitudes towards a product (Duan et al., 2008). Though, this persuasive effect could not be confirmed for all studies about (average) ratings and rather led to mixed results (see table 3).

Possible reasons for these mixed findings are the following. First, the dependent variable differs between studies (such as sales, helpfulness of the review and click-out) and therefore the influence of reviews can vary (Dellarocas et al., 2007). Second, as mentioned above, some studies consider WOM as an exogenous factor and do not include the correlation between WOM and sales in their study (Moe & Trusov, 2011). Third, the importance of valence can differ across product categories. There might be a tendency for movies and book to be less affected by valence, but rather influenced by awareness factors (Floyd et al., 2014). Finally, social and time dynamics can have an impact on the findings. Product ratings can reflect both consumers’ experience with the product and the influence of other’s ratings (Moe & Trusov, 2011). Likewise, Schlosser (2005) claimed that consumer tend to negatively adjust their opinions if they have read negative reviews confirming the influence of valence of previous reviews. Regarding time dynamics, it was shown that reviews and their rating tend to become more negative over time. This could be the result of the product life cycle: The product evolves together with different buyers and their different tastes (Li & Hitt, 2008).

(24)

(2012), stating that negative chatter has a significant effect on stock performance, whereas positive chatter has not.

Furthermore, it was dealt with the question whether a moderate rating of three-star or an extreme five-star rating is more helpful (Mudambi & Schuff, 2010). Here, literature identified two possibilities on how a three-star rating evolves. Either it can be the result of a true moderate rating (indifference) or the result of positive and negative reviews which cancel each other out (ambivalence) (Kaplan, 1972). Similar to the one-star rating, a three-star rating could have higher credibility which can be related to the theory of message sidedness. Two-sided messages are characterized by admitting weaknesses next to strengths and therefore make the whole message more credible resulting in more positive attitudes (Crowley & Hoyer, 1994). Results are mostly ambiguous whether three- or five-star ratings are more helpful. However, Mudambi and Schuff (2010) identified that helpfulness of these two ratings varies according to the type of good. Moderate reviews were found to be more helpful for experience goods than for search goods, addressing the questionable credibility of extreme positive reviews for experience goods.

Authors Product Category Dependent Variable Significant results?

Chevalier & Mayzlin (2006) Books Sales rank Valence: yes Volume: yes

Chintagunta et al. (2010) Movies Sales Valence: yes Volume: no

Clemons, Gao & Hitt (2006) Beer Sales growth rate Valence: yes Volume: no

Dellarocas et al (2007) Movies Sales diffusion

parameters

Valence: yes Volume: yes

Duan, et al.(2008) Movies Sales Valence: no Volume: yes

Godes & Mayzlin (2004) Television shows Television- viewership rating

Volume: no Huang et al. (2015) Cell phone, printer,

camera, music player, CD, video game

Helpfulness of a review Valence:yes

Kostyra et. al (2016) E-Book reader Product choice Valence: yes Volume: no

(moderates valence)

Liu (2006) Movies Box office revenue Valence: no Volume: yes

Moe & Trusov (2011) Bath, fragrance and beauty products

Cross-product temporal variation in ratings and sales

Valence: yes (static & dynamic)

Ögüt & Tas (2012) Hotels Sales Valence: yes

Olbrich & Holsing (2011) Fashion, living & lifestyle Click-out Valence: yes Pan & Zhang (2011) DVD, CD, video games,

consumer electronics, software, healthcare

(25)

Sun (2012) Books Sales rank Valence: partially Volume: yes

Tirunillai & Tellis (2012) Personal Computer, cell phone, PDA/smartphone, data storage, toys, footwear

Stock market performance

Valence: yes (for negative chatter)

Volume: yes

Table 3 Extract of the studies on product ratings based on Moe and Trusov (2011) and supplemented by studies since 2011

Volume

The volume of reviews is mostly reflected by the number of reviews available for a product (Floyd et al, 2014). The number of reviews can represent an awareness effect, the more reviews are available, the more aware consumers become about a product (Dellarocas et al., 2007). Duan et al. (2008) claim that the awareness effect is highest when reviews are posted on communities, which are relatively unaware of a product. However, if consumers search for a product and then click to read the views, they have already become aware of the product. Thus, the number of reviews is rather an indicator of the intensity of WOM (Duan et al., 2008). Volume can be seen as an extrinsic high-scope cue which increases the persuasiveness of WOM, underlined by the logic that opinions expressed by many people seem to be true (Khare, Labrecque, and Asare, 2011). Furthermore, a higher number of customer reviews can be processed heuristically resulting in a higher perceived quality (Duan et al., 2008).

(26)

2.4 The interaction between brand equity and customer reviews

Research regarding the interaction between brand equity and customer reviews is limited (Ho-Dac, Carson & Moore, 2013; Kostyra et al., 2016). Only few researchers have attended to the interaction between these two constructs so far and those who did, focused mostly on the effect of negative reviews on brand equity. An explanation for exploring only negative reviews is that consumers are more interested in sharing their experience if it was negative and negative reviews tend to be more descriptive and informative (Bambauer-Sachse & Mangold, 2011; Beneke, de Sousa, Mbuyu & Wickham, 2015 ).

Bambauer-Sachse and Mangold (2011) identified a detrimental effect of negative customer reviews on brand equity which leads to a significant “brand equity dilution”. This brand equity dilution consequently leads to a lower purchase intention. They explain the effect with a search and alignment theory: Consumers with initially positive attributes in their minds are challenged by negative attributes included in the reviews and therefore tend to align or change their impression in the direction of the challenging information. Beneke et al. (2015) argue here, that negative reviews have a higher negative effect on brand equity and purchase intention, than no reviews at all. Furthermore, they state that the effect on brand equity is more detrimental for high involvement products than for low involvement products. They claim that negative reviews signify that potential product-related risks exist, which directly enhances perceived uncertainty (Beneke et al., 2015). Lastly, Ullrich and Brunner (2015) found out that negative reviews have a higher negative impact on strong brands than on weaker brands.

(27)

also increase their brand equity. Overall, these results oppose the view of Ullrich and Brunner (2015) who claim that stronger brands are more affected by negative views. The disagreement towards the research about negative reviews is further enhanced by the fact, that cumulative numbers of negative reviews were not found to be related to the brand’s transition from a weak to a strong brand. Ho-Dac et al. (2013) do admit to the fact that in their study there were more positive than negative reviews which could have decreased credibility and impact of the latter.

Another stream of literature also disagrees with Ho-Dac et al. (2013) regarding the importance of brands in the presence of customer reviews in general. Kostyra et al. (2016) claim that customer reviews decrease the importance for customer purchase decisions and point to the essential difference that Ho-Dac et al. (2013) did not consider a choice situation but investigated sales. Their rationale is that customer reviews provide additional information about product quality and replace other product attributes, such as the brand. As a consequence, it seems that reviews reduce uncertainty and will replace brands as a quality indicator due to their higher credibility.

2.5 Hypotheses and conceptual model

The discussed literature provides the basis for the present research and the conceptual model with its hypotheses (figure 2). Prior research has focused on the categorization of innovations (Chandy & Tellis, 1998), success factors of innovations (Holak & Lehmann, 1990) and types of risks related to innovations (Kleijnen, Lee & Wetzels, 2009), though it has not discovered how innovative products are affected by brands and customer reviews. On the other hand, brands and customer reviews have been investigated as to the extent to which they affect the choice of a product (Huang & Cheng; Keller, 1993). However they have hardly been studied in the context of new products and especially not with uncertainty faced when considering innovative products. Exceptions include the introduction of product extensions under the same brand (Wernerfelt, 1988) and the studies in which movies were investigated from their release date on in context with customer reviews (Dellarocas et al., 2007; Duan et al., 2008). This research aims to investigate the effect of brands, customer reviews and their interaction on the uncertainty towards an innovative product.

(28)

brand reputation can be used to refer a quality signal (Erdem & Swait, 1998; 2004; Herbig & Milewicz, 1993), brand prestige can signal the social acceptance of a brand (Baek et al., 2010) thereby possibly reducing uncertainty. Brand recognition and brand familiarity as heuristics could reduce uncertainty by simply recognizing a familiar brand and having former experience with it. iSkin is an innovative product carrying experience characteristics creating uncertainty in consumers’ minds. Therefore, this innovative product could benefit from a well-known, socially-accepted and credible brand with a good reputation. A strong brand which owns these characteristics is referred to as having strong brand equity. Based on this, the first hypothesis can be formed:

H1: Higher brand equity leads to a reduction in uncertainty for an innovative product.

Due to the emergence of the internet, a wide range of alternatives has appeared, leading to complex decisions and uncertainty about sets of alternatives and their quality (Huang & Chen, 2006). In order to overcome uncertainty and perceived risk, consumers search on the internet for information (Peterson & Merino, 2003). As consumers are influenced by other consumers and imitate them (Huang & Chen, 2006), customer reviews tend to be a popular method to inform oneself about the opinion of others (Ludwig et al., 2013).

(29)

H2a: An increase in valence of customer reviews (average rating of customer reviews) leads to a reduction in uncertainty for an innovative product.

The volume of reviews represents an awareness effect as a high number of reviews implies that many consumers become aware about a product (Dellarocas et al., 2007). Additionally, volume can also be considered to have a persuasive effect as opinions expressed by many people seem to be true (Khare et al., 2011). In the context of an innovation, the number of ratings available for a product can serve as an indicator that already many consumers adopted the product and that it is accepted by society. This might decrease the reluctance towards the product as well as increase the social pressure to consider the product and form an opinion about it. As a result, the following hypothesis is formed:

H2b: An increase in volume of customer reviews (number of customer reviews) leads to a reduction in uncertainty for an innovative product.

In general, consumers tend to trust other consumers’ opinions more than company-generated content (Chiou & Cheng, 2003; Godes & Mayzlin, 2004). This finding is in line with the claim by Kostyra et al. (2016) that customer reviews might counteract the effect of brands. Further evidence can be found in the research on the effect of negative reviews on brands. It was shown that negative customer reviews have a detrimental effect on brand equity (Bambauer-Sachse & Mangold, 2011) which is even more enhanced for stronger brands than for weaker brands (Ullrich & Brunner, 2015). In contrast, Ho-Dac et al. (2013) claim that stronger brands are less affected by negative reviews than weaker brands. On the other hand, weaker brands also benefit more from positive reviews than stronger brands. Following Ho-Dac’s approach, it is hypothesized:

(30)
(31)

3. Methodology

In the previous chapter, relevant literature on the topic has been reviewed and hypotheses as well as the conceptual model have been introduced. So, in order to test the hypotheses, empirical research has been conducted. The present chapter outlines the research methodology including research method, study design, data collection and model specification.

3.1 Research method

In order to test the hypotheses, it was decided to perform a conjoint analysis. The basic idea of conjoint analysis is to reveal the preferences of consumers by asking them about products, which can be represented by attribute bundles. Out of the preferences for these attribute bundles, preferences for each individual attribute can be derived (Eggers & Sattler, 2011). This is why marketers regularly use this method to find out how consumers make trade-offs between available products and their attributes (Green, Krieger & Wind, 2001). Several approaches of conjoint analyses are available, among the popular ones are traditional conjoint analysis which asks respondents to rate or rank the products and choice-based conjoint (CBC) analysis which asks respondents to choose one product within a choice set of several alternatives (Eggers & Sattler, 2011; Green & Srinivasan, 1990). Drawbacks of these popular approaches include first that they do not allow respondents to express uncertainty about their behavior (Blass et al., 2010) and second that no information is given how the unselected alternatives are compared to each other. Therefore, it was decided to use a conjoint analysis which elicits choice probabilities for all alternatives in the given sets. In contrast, for a CBC analysis data is captured only for the actual choice and possible information about the other alternatives is lost. A CBC will not cover whether there is a clear preference over the other alternatives, such as 80% to 20% or whether it was a close decision of 55% to 45%.

3.2 Study design

(32)

the independent variables of the model and were nominal (function and placement), ordinal (brand) and metric (number and average rating of reviews). In the following the choice of attributes and their levels is explained.

3.2.1 Attributes and levels Brand

To measure the effect of brand equity, it was decided to select three brands varying in their brand equity. Following the argumentation that a brand extension should fit the existing product category of a brand, only brands selling electronics were considered (Aaker & Keller, 1990; Wernerfelt, 1988). In order to determine the brand equity or value of a brand, reference was made to the Interbrand ranking. Apple was selected as the brand with high equity, since it was elected as the best global brand of 2015 (Interbrand Ranking, 2016). LG was selected as a brand with medium brand equity, as it achieved rank 97 of the Interbrand ranking in 2007 (Syncforce, 2016). However, in 2015 it was not listed anymore in the top 100 compared to other electronic brands such as Samsung (rank 7), Sony (rank 58) and Panasonic (rank 65) (Interbrand Ranking, 2016). Lastly, to create a control situation in which the product carried low brand equity, it was decided to introduce a fictional brand named Ink’d which was inspired by the tattoo-like shape of the gadget.

Function and Placement

The function and placement of iSkin represent the two main characteristics with which the product can be described and at the same time they can be varied in their levels. When choosing the attribute levels, it was considered that all functions could be combined with all placements on the body. Therefore, it was decided not to choose “ear” as an attribute level, since it could not be combined with message typing. All other variations seemed compatible, although some variations appeared more reasonable than others, such as a music player on the finger versus message typing on the finger.

Average rating of reviews

(33)

(2016). They classified low from 1.1 to 2.33, medium from 3 to 3.3 and high from 3.67 to 4.9. The scores chosen according to the scale were: 2.2, 3.1 and 4 to also allow a linear modelling of the valence attribute.

Number of reviews available

In order to identify appropriate levels for the volume, it was again oriented towards the conjoint attributes of Kostyra et al. (2016). The volume levels they used were 6, 10, 30, 100 and 200 reviews. For this study the following levels were chosen: 10 as low, 100 as medium and 1000 as high.

No-choice option

In decision-making, consumers usually also have the choice to not choose an alternative. In addition, free choice resolves psychological discomfort related to forced choice under preference uncertainty (Dhar & Simonson, 2003). Consequently, it was decided to include a no-choice option in the study stating that “I would not choose any of these”. However, if a respondent distributed 100% to the no-choice option, information about the preference for the other alternatives would be lost. So, in a worst case scenario no preferences at all would be retrieved, if a respondent distributed 100% to all choice sets (Eggers & Sattler, 2011). Reasons for choosing a no-choice include avoidance of difficult trade-offs, idleness to decide and lacking of attractiveness of the alternatives (Dhar, 1997). Therefore, Brazell et al. (2006) advise to use a dual-response no-choice option which asks the respondent first to decide between alternatives, but in a second question asks whether the preferred alternative would be actually bought. Advantage of this method is that no information is lost. However, the software used for analysis does not offer a dual response option in the combination with choice probabilities, as a distribution of probabilities also allows for equal preference of multiple alternatives. Hence, it was decided to stick with the regular no-choice option.

Capturing probabilities

(34)

how likely would you choose any of these alternatives? Please distribute 100 % to mark your choice.” The constant sum has an ordinal scale due to its comparative nature, however the resulting values of the dependent variable were treated as metric (Malhotra, 2010).

Attributes Level 1 Level 2 Level 3

Brand Ink’d LG Apple

Function Music player Message typing Telephone

Placement Finger Back of the hand Forearm

Average rating of reviews 2.2 stars 3.1 stars 4 stars

Number of reviews available 10 reviews 100 reviews 1000 reviews

Table 4 Attributes and their levels

Control variables

Lastly, several demographic variables were included as control variables in the study to rule out that the dependent variable, the choice probabilities, would be affected by these variables and to increase the statistical power (Becker, 2005). Furthermore, it was aimed at measuring the effect of these control variables on the attributes. The control variables included were gender, age, country of origin as well as level of education and were captured in the following way: “What is your gender?” was captured by a nominal variable with two answer possibilities (1) male and (2) female. “What is your age?” represented an open question leading to interval data. The question “What is your country of origin?” provided three answer possibilities (1) Germany, (2) Netherlands and (3) “Other, please specify”, leading to nominal data. Finally, “What is your highest level of education?” is an ordinal variable with a 6-point scale. Answer options were: main school/secondary school, high school, bachelor, master, PhD and other.

3.2.2 Experimental Design

(35)

product. While the first conjoint analysis was designed to measure the effect of brand on uncertainty (H1), the second conjoint was designed to investigate the effect of customer reviews on uncertainty (H2a and H2b) as well as the moderating effect of valence of customer reviews on brands (H3).

The experimental design chosen for both analyses was fractional factorial due to the high number of possible attribute level combinations (3*3*3 =27 for the first conjoint analysis and 3*3*3*3*3= 243 for the second). Eggers and Sattler (2011) recommend 12-15 choice sets per respondent for a CBC. As it is more tedious to fill out sets with a constant sum, it was decided to randomly create five choice sets per respondent for each conjoint analysis. Additionally, one fixed set was created for the second conjoint analysis in order to measure the interaction effect between brands and valence of customer reviews. This fixed set contained alternatives combining high brand equity with low valence of customer reviews versus low brand equity with high valence of customer reviews (compare with figure A6 in Appendix A). During the survey each respondent was faced with four alternatives per choice set, three product combinations and the no-choice option. Furthermore, all attributes were shown within the conjoint analyses and none was left out entailing a traditional full-profile design.

For each respondent a unique questionnaire version was created to avoid order effects of the attributes. The randomized design used for generating tasks was balanced overlap. This method is recommended by Sawtooth, as it allows for a higher level of overlap which was found to be valuable in recent research. Furthermore, this method is considered to be a good compromise as it tries to achieve orthogonality, yet at the same time allows to measure the effect of interactions (SSI Web Help, 2016).

(36)

3.3 Data collection

The online survey was accessible for a week in April and its link was distributed via the social media platform Facebook as well as via mail. This fast and wide distribution, along with cost efficiency and efficient processing of the results were the main reasons why an online survey was selected (Malhotra, 2010). Screenshots of the whole survey can be found in Appendix A.

The setup of the online questionnaire was as follows. First, the respondents were welcomed and received some information about the author of the study as well as about the confidentiality of their answers. The survey continued with a short video about iSkin introducing the product and its functions2. Subsequently, the participants were presented with the first conjoint analysis consisting of five sets. Afterwards, they were given the information that one year had passed and customer reviews were available now, which served as a means to introduce customer reviews as attributes. Consequently, the respondents were presented with the second conjoint analysis encompassing six sets which differed from the first conjoint analysis in the way that the two attributes, average rating and number of customer reviews, were added. At the end of the survey respondents were asked to answer demographic questions about gender, age, nationality and level of education. Finally, the respondents were thanked for their participation in the survey.

3.4 Model specification

In order to analyze the two conducted conjoint analyses a random utility model was used (Manski, 1977). A random utility model is characterized by the fact that choices are based on overall utilities of alternatives and assumes that products are combinations of attributes. Here, consumers attach part-worth utilities to each attribute. Therefore, the systematic utility for iSkin can be represented by the sum of part-worth utilities:

First conjoint analysis (without customer reviews): 𝑉𝑛,𝑖,𝑗= 𝛽𝐹𝑛,𝑖,𝑗 𝐹𝑛,𝑖,𝑗+ 𝛽𝑃𝑙𝑛,𝑖,𝑗 𝑃𝑙𝑛,𝑖,𝑗+ 𝛽 𝐵𝑟𝑛,𝑖,𝑗 𝐵𝑟𝑛,𝑖,𝑗 𝑉𝑛𝑜−𝑐ℎ𝑜𝑖𝑐𝑒 = 𝛽𝑛𝑜−𝑐ℎ𝑜𝑖𝑐𝑒

n=1,…,N i=1,…,I j=1,…,J

2https://www.youtube.com/watch?v=9cvZnhvzrBI

(37)

with N=250, I=5, J=3

Second conjoint analysis (including customer reviews and the effect of interaction between brand and customer reviews):

𝑉𝑛,𝑖,𝑗 = 𝛽𝐹𝑛,𝑖,𝑗 𝐹𝑛,𝑖,𝑗+ 𝛽𝑃𝑙𝑛,𝑖,𝑗 𝑃𝑙𝑛,𝑖,𝑗+ 𝛽𝐵𝑟𝑛,𝑖,𝑗 𝐵𝑟𝑛,𝑖,𝑗+ 𝛽𝑉𝑎𝑙𝑛,𝑖,𝑗 𝑉𝑎𝑙𝑛,𝑖,𝑗+ 𝛽𝑉𝑜𝑙𝑛,𝑖,𝑗 𝑉𝑜𝑙𝑛,𝑖,𝑗 + 𝛽 𝐵𝑟𝑛,𝑖,𝑗∗𝑉𝑎𝑙𝑛,𝑖,𝑗 𝐵𝑟𝑛,𝑖,𝑗∗ 𝑉𝑎𝑙𝑛,𝑖,𝑗 n=1,…,N i=1,…,I j=1,…,J with N=250, I=6, J=3 where, V = utility n = individual i = choice set j = choice F = function Pl = placement Br = brand

Val = valence of reviews Vol = volume of reviews

To derive probabilities from the given utilities a constant sum model is applied. 𝑝𝑟𝑜𝑏 ({𝑦𝑛,𝑖,𝑗}𝑛|𝑿, 𝛽 ) =

∏𝐽𝑗=1[exp (𝑉𝑛,𝑖,𝑗)]𝑦𝑛,𝑖,𝑗

∑𝐽𝑗=1 exp 𝑉𝑛,𝑖,𝑗 where,

X= product attributes

(38)

from other respondents to stabilize the estimates (Orme, 2000). Therefore, Hierarchical Bayes is suited for analysis with limited data as given in the study at hand. Above all, recent research shows that HB models lead to superior results compared to other procedures, which is why this model was used for the following analysis (Rossi et al., 2012).

(39)

4. Results

The present chapter reports the data gathered through the quantitative study and analyzes them. First, the descriptive statistics of the data set are presented in order to provide insights about the demographics of the participants and distribution of the responses. Then, insights about the frequency of the no-choice option are provided and subsequently, results of the two conjoint analyses using a Hierarchical Bayes model are presented. Afterwards, the hypotheses are tested using the standard deviations of the utilities gathered by the conjoint analyses and lastly, further insights about customer reviews are revealed.

4.1 Descriptive statistics

In total, 454 participants have started the survey, of which 250 completed the questionnaire until the end. This amounts to a drop-out rate of 44.9%. All incomplete responses were removed from the data set, as most of these participants dropped out of the survey after seeing the introduction or the video and therefore no insights could be gained. There was no need for data cleaning, as the demographic answers seemed plausible and participants could not continue the survey if they had not distributed 100% to the alternatives.

In the dataset with complete answers, 66% were female and 34% were male respondents. This high inequality could be due to the distribution in the author’s network which entails a majority of female contacts. The average age of the respondents was 30.56 years, ranging from 16 to 69 years old. The distribution of the country of origin was as follows: Germany (72.4%), Netherlands (6%) and others (21.6%), half of them coming from the USA. Lastly, the level of the participants’ highest education reached ranged from “main school/secondary school” (3.2%), “high school” (16.4%), “bachelor” (44.8%), “master” (27.2%), “PhD” (2.8%) and “other” (5.6%). On average, it took the respondents 8.48 minutes (min) to complete the survey, with values ranging from 1.25 min to 139.72 min.

(40)

4.2 Relevance of no-choice option

The no-choice option had a high relevance in this quantitative study and will be therefore emphasized. Overall, 18.4 % of the respondents distributed 100% to the no-choice option in the whole survey implying a low interest and/or resistance towards this innovative product. The average response time of these participants was also significantly lower with a value of 5.15 min, ranging from 1.25 min to 9.52 min. This signifies that many of these respondents rushed through the survey, as it already takes 1.27 min to watch the video until the end.

Excluding the participants who always distributed 100% to the no-choice option, 66.8% of all respondents distributed 100% to the no-choice option at least once, leaving 14.8% to respondents who never distributed to the no-choice option. These results already indicate that participants were uncertain about the product in general and also doubted the attractiveness of the given alternatives. Interestingly, the distribution of 100% to the no-choice option decreased in the second conjoint analysis with the introduction of the customer reviews: On average, 105 respondents distributed 100% to each set belonging to the first conjoint analysis, whereas for the sets of the second conjoint analysis on average 91 respondents did so. This had an impact on the preference share for the combination with highest utility, as will be shown in the following evaluation of the conjoint analyses.

4.3 Conjoint analysis using HB

4.3.1 Conjoint analysis without reviews

The model of conjoint analysis 1 consisted of the three attributes brand, function and placement. Attributes were represented as part-worths because an application of a vector specification was not useful here. The final model comprised seven parameters and 10 covariates. The number of parameters can be determined by the number of levels minus one due to effect coding, times the number of attributes and lastly, adding the none-option. The 10 covariates are composed of one parameter for gender, one for age, three minus one for the country-of-origin, six minus one for education and lastly of one intercept which indicates a base case, if all covariates are zero.

(41)

model. As values vary between 0 and 1, a resulting value of 0.617 for the given model implied that the LL is 61.7 % of the way that would be expected by chance and the value for a perfect fit. For RLH the nth root of the likelihood is taken where n refers to the total number of choices made in all tasks which was in this case 1250 (= 250 respondents * 5 choice sets). If no information about the part-worths was given, the probability for each alternative in a choice set would be one divided by the number of alternatives available per set, which is in that case four. Consequently, RLH would be ¼ = 0.25 for a random model. A value of 0.588 signified that the given model was 2.3 times better than a random model (Sawtooth Software, 2009).

Findings

A summary of the output for conjoint analysis 1 is given in Appendix B under table B1. The product with the highest utility was represented by a combination of Apple as the brand with the function of a music player and placement on the forearm. The levels with the lowest utility within an attribute included LG as the brand, message typing as the function and placement on the finger. It is noticeable that the average utility for LG was lower than for the fictional brand Ink’d, although both still had a negative utility compared to Apple.

The importance for each attribute in descending order was function as the most important attribute with 39.11 %, followed by brand with 32.05% and lastly placement with 28.84%, denoting a smooth distribution (table 5). When grouping respondents by age, it could be seen that the importance of the brand slightly increased with age (see Appendix table B2).

The no-choice option had a high utility of 504.065 compared to the combination with the highest utility having a value of 82.907, which led to an average choice share of 0.0146 (= (exp 82.907) / [(exp 82.907) + (exp 504.065)]). This means that only 1.46% would decide for the combination with the highest utility, instead of the no-choice option. This confirms the high importance of the no-choice option, as already mentioned above.

Analysis of utilities

(42)

For the brand Apple, 186 respondents had positive utilities, while 64 had negative utilities. Analyzing the last or fourth quartile in order to filter Apple fans, subsequent insights were gained. They tended to be younger (M = 28.39 years vs. M = 30.56 years), a high proportion of them were males (41.9% vs. 34%) and their level of education was slightly lower than the distribution of the overall sample. The level of education could be read by an increase in percentage for main school/secondary school (9.7% vs. 3.2%) and high school degree (21% vs. 16.4%), while a decrease in the master’s degree was observed (19.4% vs. 27.2%).

Considering the utilities for LG and Ink’d the following insights were gained. While for the fictional brand Ink’d half of the respondents (121 people) had positive utilities, for LG only a quarter (62) of the respondents had positive utilities. When verifying the last quartiles for these two brands, it could be seen that the respondents were older on average than the overall sample, 32.63 years for Ink’d and 34.68 years for LG. While Ink’d is characterized by a higher proportion of women (72.6% vs. 66% for the overall sample), the fourth quartile of LG contains more master (40.3% vs. 27.2%) and PhD graduates (6.5% vs. 2.8%). However, due to the low sample size of PhD graduates in the whole sample, this group is not very representative.

Lastly, the parameters for individual respondents were tested for their significance on a 95% confidence level. For this, the individual standard deviations were converted into standard errors and t-tests were conducted. The analysis revealed that the range of significant parameters on a 95% confidence level varied between 57.2% and 85.2 % within the attribute levels (Appendix table B4).

Attributes Conjoint 1 (without reviews) Conjoint 2 (with customer reviews)3

Brand 32.05% 20.68% (43.82%)4

Function 39.11% 14.51% (30.75%)

Placement 28.84% 12.00% (25.43%)

Valence of review n.a. 30.94%

Volume of review n.a. 21.86%

Sum: 100% 100%

Table 5 Average attribute importance for conjoint analysis 1 and 2

3 Does not add up to 100% due to rounding

Referenties

GERELATEERDE DOCUMENTEN

By conducting the main analysis 1 (i.e. zero measurement) by means of a quantitative questionnaire, the score of three luxury brands (e.g. Chanel, Armani and

[r]

[r]

The moderating effect of culture on purchase intention has therefore been examined by conducting both an experiment and questionnaire simultaneously

How does the valence of online customer reviews written by unknown consumers and the valence of peer opinions impact the purchase intention of sportswear products, and how is

The main objective of this research was to understand the relationship between CSR and sales and the moderating effects of brand equity and GDP on the relation between CSR and

MEASURING THE EFFECT OF BRANDS AND CUSTOMER REVIEWS ON UNCERTAINTY BY ELICITING

From this research it can be concluded that there are no significant differences between humorous and non-humorous reviews, and no significant differences between