How far will you go to pay the price? - The influence of online- and offline social networks and online customer reviews on consume

(1)

How far will you go to pay the price? - The influence of online- and offline social networks and online customer reviews on consumers’ willingness-to-pay for high-involvement, durable products and the implications for retailers and manufacturers.

E.A.M. Folbert

Marketing Department University of Groningen Master thesis June 2018 Akerkhof 10-4, 9711JB Groningen +31 6 81 17 32 64 e.a.m.folbert@student.rug.nl s1891979

Dr. J.E.M. Van Nierop Dr. F. Eggers

(2)

2

Abstract

The Internet, and more specifically social networks, are rich sources of information. Social networks, such as Facebook, but also the offline social network of friends and family provide reviews about products and services alike. This work investigates how the online- and offline social network and their reviews influence what consumers are willing to pay for a high-involvement product, such as a laptop. The average rating and the combination of reviews from strong- and weak-ties prove most important. Implications for retailers and manufacturers are discussed.

Keywords: reviews, conjoint, online- and offline social network, consumer

(3)

3

1. Introduction

1.1 Identifying a research gap.

Nowadays, it is challenging to compete on price or product assortment, for both retailers and manufacturers alike, according to Ahmetoglu, Furnham and Fagan (2014). For example, the challenge to compete on price could provide increased deal value for consumers through discounts and promotions. However, Darke and Chung (2005) find that discounts and promotions can cause negative quality inferences if there is no guarantee of product quality.

Therefore, retailers and manufacturers shift their focus increasingly to ways that influence price perceptions, since these ways can influence buyers’ perceptions and purchase decisions while it does not have to involve any change to prices (Ahmetoglu et al., 2014). Shifting buyers’ perceptions influence how much consumers are willing to pay for a product or service.

As such, retailers’ and manufacturers’ pricing strategies constitute one factor that influences consumers’ willingness-to-pay. Consumers themselves are another: through online consumer reviews (OCRs). Consumer opinions, user experiences and product reviews impact consumer purchase decisions (Gu, Park & Konana, 2012) and consumer purchase intentions (Park, Lee & Han, 2007). What helps to facilitate this are platforms such as social media (Chen, Fay & Wang, 2011), as further highlighted in the work of Erkan and Evans (2016).

Online consumer reviews can be powerful, since they influence consumer purchase intention through their quality and quantity (Park et al., 2007). Yet, their influence differs for low-involvement versus high-involvement consumers.

(6)

high-6

involvement ones, since the findings for the latter are more sensitive to the quality of OCRs.

This may warrant corroboration, and herein lays an opportunity: to further the research of OCRs’ influence on high-involvement consumers, for a high-involvement product category, and their influence on consumer willingness-to-pay.

In addition, consumers’ peers in their offline social network may influence consumer willingness-to-pay too. Duhan, Johnson, Wilcox and Harrell (1997) observe that a consumer may turn to persons with whom they have strong ties (i.e. family and friends) for advice and assurance on product alternatives. Naturally, weak-ties (e.g. acquaintances) may also serve as a source of recommendation and advice to a consumer.

The influence of both the online- and offline (social) networks play a role in determining consumers’ willingness-to-pay, while the level of involvement of the consumer with the product category has a moderating role (Park et al., 2007).

Currently, however, there are not many articles that consider the influence of online- and offline networks jointly. Furthermore, the work of Gu et al. (2012) indicates that there is a need for further research on the manner in which (online) word-of-mouth influences consumers’ willingness-to-pay for high-involvement products.

This research may help to fill that gap by addressing it. The problem statement becomes: “To what extent do online- and offline social networks influence consumers’ willingness-to-pay for high-involvement products, and how can retailers and manufacturers optimise their strategy given this influence?”

(7)

7

The first research question leads to second question, and that is: how can retailers and manufacturers cope with influences of online- and offline social networks, so that they can benefit from those influences?

A third question, that builds on the second one, is: are there any particular coping strategies for those retailers and manufacturers that employ a multi- or omnichannel approach towards consumers, so that they can maximise the benefits of online- and offline social networks influence on consumers’ willingness-to-pay?

A fourth and final matter is the consumer level of involvement with a product. As mentioned by Park et al. (2007), the level of product involvement moderates the influence of online- and offline social networks’ influence on consumers’ willingness-to-pay.

This moderation effect is in part determined by the quantity and quality of online consumer reviews (Gu et al., 2012), so research questions related to this are: Is there a lower or upper limit of the quantity of OCR to have an influence on consumers and their willingness-to-pay? Also, what constitutes a qualitative OCR? What does this mean for product categories, and are there implications for retailers and manufacturers?

1.2 Benefits of this research

In seeking answers to the research questions lie potential benefits that can be reaped by practitioners and academics.

(8)

8

If consumers’ willingness-to-pay is positively affected, consumers may be induced to spend more compared to their initial willingness-to-pay. For retailers and manufacturers this means that their revenues may increase, if the price level of their products and services increases, but stays within the limits of consumers’ willingness-to-pay. If sales volume stays the same, revenues for manufacturers and retailers increase (Ahmetoglu et al., 2014). It provides another tool in the pricing strategy toolbox of retailers and manufacturers.

Another benefit of this research is that it considers both the influence from online- and offline (social) networks on consumer willingness-to-pay, which may help retailers and manufacturers in dealing with consumers that shop across channels. This may especially hold for multi- and omnichannel retailers for whom consumers move through their channels (Verhoef, Kannan, & Inman, 2015). They have a more difficult challenge in finding fitting strategies that positively influence consumer willingness-to-pay.

Several research issues have been raised, and for a better understanding a theoretical framework is presented in the next chapter. It gives a foundation on which the remainder of this thesis is built.

It focuses on consumers’ willingness-to-pay, and how it is influenced by online consumer reviews (OCR) and by opinions stemming from the consumer’s social network (i.e. family, friends and acquaintances), as documented by academics so far. Other topics that play a mediating or moderating role, such as level of involvement (Gu et al., 2012), are addressed too.

Chapter three addresses the research methodology that is suitable for the research in this thesis. It sheds light on what kind of data is required and how it is collected. Last, a plan of analysis for this research is presented.

(9)

9

2. Literature review

This chapter provides more in-depth knowledge from literature regarding the research questions that have been raised in chapter 1. First, a discussion of factors that in general influence consumer willingness-to-pay is presented. This is to raise awareness of the breadth of factors that influence consumer willingness-to-pay. From that general discussion, a more detailed one follows on the influence of the online social networks, as well as the offline social networks on consumer willingness-to-pay. Afterwards, any moderators that significantly influence the relationship between consumer willingness-to-pay and online social networks for instance, are discussed.

From these discussions and what literature finds with regard to the research questions, hypotheses are formulated. An overview of all hypotheses is presented at the end of this chapter. Additionally, a conceptual model is presented which visualises the relationships between the variables in this research.

2.1 Consumer willingness-to-pay

What consumers are willing-to-pay (willingness-to-pay) for products and services is influenced by a variety of factors. A couple of factors that in general influence what consumers are willing-to-pay are discussed here.

For starters, stress can reduce consumers’ willingness-to-pay, as demonstrated by Maier & Wilken (2014) using the construal-level theory. Yet, this influence is moderated by the product category, meaning that for product categories with less (abstract) information that has to be processed by consumers, the decline in willingness-to-pay is smaller. In other words, retailers and manufacturers would do well if they can reduce consumers’ stress by conveying product or service information as comprehensive as possible, thereby reducing levels of abstraction of the information consumers have to process.

(10)

10

positive impact of customer satisfaction on consumers’ willingness-to-pay, albeit that this relationship is more cumulative in form.

This means that a single satisfied consumption of a product or service may not necessarily increase the willingness-to-pay of consumers, but the accumulation of consumptions does (Homburg, Koschate & Hoyer, 2005). For retailers and manufacturers this entails that satisfying consumer demands at every occasion, as well as making sure that consumer become returning customers, helps to increase consumers’ willingness-to-pay.

In a different context, textual cues can also help influence consumers’ willingness-to-pay in the case of radically new products (Kuijken, Gemser and Wijnberg, 2017). For products that are incrementally new, the willingness-to-pay is not (positively) influenced by cues referring to a product category with a higher than average monetary value, according to Kuijken, Gemser and Wijnberg (2017). This may be valuable for retailers and manufacturers who may on occasion cooperate on (the introduction of) a radically new product on the market (e.g. Philips and Douwe Egberts with the Senseo coffee machine in 2001). If consumers are willing to spend more on a radically new product, it can provide a larger than anticipated revenue stream for retailers and manufacturers.

Retailers also need to consider the research of Chatterjee and Kumar (2017) if they would like to capitalise on consumers’ willingness-to-pay. Their study indicates that consumers are willing to pay higher prices for household products at omnichannel retailers compared to pure-play (i.e. only) online retailers. Yet, this differs for functional and expressive goods, for which consumers may not be willing to pay a higher price at omnichannel retailers (Chatterjee & Kumar, 2017).

(11)

11 2.2 Online social networks.

Online social networks can be considered as online sources to which consumers can turn for information, advice, reassurance, and (perceived) risk-reduction regarding a prospective purchase (Gu et al, 2012; Gunawan & Huarng, 2015; Wang, Yu & Wei, 2012). Most commonly this is done through consumer usage of social network media (SNM) sites and online consumer reviews (OCRs) (Chen et al., 2011; Park et al., 2007; ).

In a more specific case of online stores, consumers exhibit a higher willingness-to-pay if rich media was displayed at a site such as product videos and virtual product experience (Li & Meshkova, 2013). This is due to the greater feeling of informedness that consumers experience, as well as increased excitement regarding the shopping experience in the online store.

An additional benefit for retailers and manufacturers is that interaction is made possible between consumer and online store; this entails that consumers’ willingness-to-pay for experience products becomes higher than for search products (Li & Meshkova, 2013). Hence, when feasible for a company to do so, interaction should made possible for experience products to benefit from the higher willingness-to-pay of consumers.

For sake of clarity, “experience product” is defined here as “a product whose information is not readily available before purchase or cannot be observed or assed properly before experiencing the product, like a perfume for instance (Bae and Lee, 2011, p. 256)”. Conversely, a “search product” is a product whose information is readily available before purchase, like a watch (Pantilla, 2015).

(12)

12

Next to pervasiveness of online consumer reviews, their evaluative consistency needs to be considered as well. Schlosser (2011) finds that presenting pros and cons of a product in a review is not always more helpful and persuasive. This depends on the perceived consistency between a reviewer’s arguments and rating, where consumers find a review helpful and are influenced by it if a pros-and-cons review corresponds with a moderate rating (Schlosser, 2011). Likewise, a product review containing only favourable arguments that corresponds with an extreme rating (e.g., 9/10), is considered helpful by consumers, according to Schlosser (2011), since the arguments of the review and the rating are evaluative consistent. Only then does an online consumer review (positively) influence consumers’ willingness-to-pay and is that review considered to be helpful by consumers.

How helpful an (online) reviews are, is a matter where the works of Singh, Irani, Rana, Dwivedi, Saumya and Roy (2017), and Saumya, Singh, Baabdullah, Rana and Dwivedi (2018) focus on. As these studies state that there are hundreds, if not thousands of reviews on products, they establish a “helpfulness” score of reviews. The helpfulness score they propose is predicted using features extracted from review text, product description, and customer question-answer data of a product, for which Saumya et al. (2018) use random-forest classifier and gradient boosting regressor.

The helpfulness score may help to distinguish low-quality reviews from high-quality ones, which helps consumers to signal out the high-quality reviews they are looking for. In turn, consumers could find faster the information, advice, reassurance, and reduction of (perceived) risk they are looking for, with regard to a prospective purchase (Gu et al, 2012; Gunawan & Huarng, 2015; Wang, Yu & Wei, 2012).

As far as online consumer reviews and their quality are concerned, indications of what constitutes a qualitative online consumer review have been presented above. The quantity of online consumer reviews is found to matter for low-involvement consumers, while it also represents a measure of the popularity of a product or service (Park et al, 2007).

(13)

13

product has, the more likely it is consumers may experience a reduction or elimination of risk. This may in turn increase consumer purchase intention or willingness-to-pay. Online consumer reviews are not the only online source of influence; social networking media (SNM) sites like Facebook, Instragram, Youtube also influence consumer willingness-to-pay (Gunawan & Huarng, 2015). Gunawan and Huarng (2015) find that social influence together with perceived risk in social network media sites, affect consumers’ purchase intention. The social influence relates to the influence from friends and relatives, who exercise their pressure on a consumer’s subjective norms. In turn, subjective norms, together with consumer attitude, are key determinants of consumer’s purchase intention for virally marketed products/services through SNM (Gunawan & Huarng, 2015). Thus, in a different part of the online social network, social network media sites also influence a consumer’s willingness-to-pay.

Other interesting findings with regard to social network media sites are that users attribute more value to source credibility of a review than argument quality, as credible sources form positive attitudes towards purchase intention. Additionally, the more the information transparency of a review, the more positive effect it has on consumer willingness-to-pay. (Gunawan & Huarng, 2015). Therefore, the higher the number of credible sources and their information transparency, the higher the purchase intention of consumers who are users of social network media sites (Gunawan & Huarng, 2015). Thus, the online social networks, in particular online consumer reviews and social network media sites, have an influence on consumers and their willingness to pay for products and services.

It is therefore hypothesised that:

- H1: The less variance (less pervasive) online consumer reviews are, the more positive

online consumer reviews influence consumer willingness-to-pay.

- H2: The more consistent the arguments and rating of an online consumer review are with each other, the more positive influence on consumer willingness-to-pay.

- H3: Helpfulness scores of online consumer reviews reinforces the positive effect of

(14)

14

- H4: More online consumer reviews on a product or service will increase the consumer

willingness-to-pay for that product or service.

- H5: The more credible the source of an online consumer review or review on a social

network media site is, the more positive influence the review will have on consumers’ willingness-to-pay for the reviewed product or service.

2.3 Offline social network influence.

By offline social network influence is meant the influence that family, friends and acquaintances can have on a consumer; especially on the consumer’s willingness-to-pay and purchase intention with regard to a product or service. Family, friends and acquaintances can generally be referred to as ‘ties’ to a consumer, like Duhan et al. (1997) do.

In their work, they indicate that the influence of the word-of-mouth of the recommendation sources, like the ties, depends in part on their closeness in relationship to a consumer. Friends who accompany a consumer on a shopping trip are considered as strong-ties, whereas the shop assistant from whom a consumer may seek advice is considered as weak-tie. Prior knowledge and task difficulty influence the likelihood of consumers’ consulting with a strong-tie source like a friend, whereas importance of instrumental cues and subjective prior knowledge influence a consumer’s likelihood to seek advice from a weak-tie, like a shop assistant (Duhan et al, 1997).

Hence, for a consumer tasked with buying a novel item, making an infrequent purchase, or facing a difficult choice set, strong-ties are most likely chosen for advice. Kapferer and Laurent (1993) also notice this reliance on another person’s advice in the behaviour of consumers who are highly involved with a product category. Thus, the level of involvement may also influence the relationship between offline social networks and consumer willingness-to-pay.

(15)

15

blend with that of the online social network; that is, an interaction effect between the two types of social networks may occur.

Kim, Kim, Choi & Trivedi (article in press) also study the interaction between the online- and offline social network, although they take a more general approach, since their research finds that overall offline social interactions have a positive informational influence on online shopping demand. Moreover, it studies only the direction of the influence of offline social interactions on online shopping demand, while not addressing any influence of online shopping on offline social interactions (Kim et al., article in press). Still, the article does point out that consumers of high-involvement products have lower preferences for online shopping and rely on offline social interactions to quell any uncertainties such consumers may have, before they make their purchase decision (Kim et al., article in press). Hence, an offline social interaction with a friend, or a shop assistant with expert product knowledge, could give a consumer of a high-involvement product the assurance through which the consumer’s willingness-to-pay is positively influenced. This may cause that consumer to shop online and purchase the product, so that offline social interaction(s) have a positive influence on online shopping, just as Kim et al. (article in press) find.

Overall, there are indications that the offline social networks of consumers have an influence on their willingness-to-pay.

It is therefore hypothesised that:

- H6: The more positive strong-ties of a consumer are about a product or service, the more

positive it influences the consumer’s willingness-to-pay.

- H7: The more positive weak-ties of a consumer are about a product or service, the more

positive it influences the consumer’s willingness-to-pay.

(16)

16

- H8: Strong-ties in the offline social network have a positive influence on consumer willingness-to-pay in the online shopping environment through their assurances to consumers that are highly-involved with a product.

-H9: A negative review from strong-ties in the offline social network towards the consumer, decreases consumer willingness-to-pay in the online shopping environment for consumers that are highly-involved with a product.

2.4 Level of involvement with a product

The matter of consumer involvement with a product plays a moderating role on the relationship between online consumer reviews (OCRs) and consumer willingness-to-pay, as indicated by the studies of Park et al. (2007) and Gu et al. (2012).

For the sake of clarity, the definition from Gu et al. (2012) will be used here to define the concept of involvement with a product, or product involvement as Gu et al. (2012) call it.

Product involvement is defined as a consumer’s perceived importance of, and interest in,

a product. If a consumer does not show interest in, or perceive the product as important, then the concept of product involvement does not have a moderating effect on the relationship between the online- and offline social networks and consumer willingness-to-pay.

There is a further distinction between low- and high levels of product involvement of consumers, and this is based on involvement theory (Gu et al., 2012). This theory states that for low involvement levels, consumers engage in a limited pre-purchase information search, whereas extensive searches for information are done by consumers in case of high involvement levels.

(17)

17

Naturally, the perceived risks vary from consumer to consumer. Hence, not every consumer experiences the same level of involvement with a specific product, or product category versus another, as Laurent and Kapferer find (1985).

Laurent and Kapferer (1985) also find other antecedent conditions of involvement in literature, next to perceived risks by consumers. In their study, Laurent and Kapferer (1985) subdivide the antecedent condition of perceived risk into perceived importance of

negative consequences in case of a poor choice, and the perceived probability of making such a mistake. The other antecedents they find are perceived importance of the product or situation, perceived sign value, and perceived pleasure value.

Laurent and Kapferer (1985) argue that these antecedent conditions could be used to make a profile of consumer involvement regarding a product category. Making a consumer profile for a specific product may lead to interesting insights, but these can be the opposite for another product in the same category. However, for an entire category of products, there may be insights or trends of consumer involvement that apply to the product category in general. Hence, it is better make profiles of consumer involvement at the product category level (Laurent & Kapferer, 1985).

A profile of consumer involvement can be constructed for product categories for which consumers exhibit high involvement. This entails behaviour that is associated with high involvement, such as information gathering, time spend on decision making, and reliance on other person’s advice (Kapferer & Laurent, 1993). Similar behavioural associations are found by the studies of Moorthy, Ratchford and Talukdar (1997), Murray (1991), and Viswanathan, Gosaïn, Kuruzovich and Agarwal (2007),

(18)

18

based on facts (Park et al., 2007). It is also expected that the consumer seeks advice in the offline social networks to reduce perceived risks, among others.

Thus, for a consumer that is highly involved with a product category, their involvement level has a moderating impact on the influence the online- and offline social network have on consumer willingness-to-pay.

Therefore, it is hypothesised that:

- H10: a high involvement level with a product category positively moderates the influence

of the online consumer reviews and reviews from strong- and weak-ties in social network media sites and offline social network on consumer willingness-to-pay.

The implications for retailers and manufacturers are that online consumer reviews, and to a lesser degree their own product attribute information (Chen & Xie, 2008), can help to reduce the perceived risks consumer have towards high involvement product categories. This gives consumers reassurance and a guarantee that the product or service they are considering to buy, is money well spent. Therefore, their willingness-to-buy increases, which in turn has the potential to increase sales volumes at retailers and manufacturers.

As a graphical representation of what is discussed up till now, a conceptual model is presented on the next page. On the left are the independent variables (IVs) who are hypothesised to have an influence on consumers’ willingness-to-pay, which is the dependent variable (DV).

(19)

19

Figure 1 – Conceptual model.

(20)

20

3. Research design.

This chapter presents the details on the product category that is chosen for this research, how the hypotheses are tested, as well as which research method would be most suitable for this research. Furthermore, a plan on how to obtain sufficient data and respondents is presented.

3.1 Product category choice.

The opportunity to further research on the effect of word-of-mouth on high-involvement product categories (Gu et al., 2012) requires a selection of a product from that category. Examples of product categories that are typically considered as high involvement are durables like consumer electronics for instance (Gu et al., 2012). Consumer electronics concern products like digital cameras, (smart)phones, and laptops.

As digital cameras have been researched before (Gu et al., 2012; Wang et al., 2015), and smartphones often come in combination with a subscription at a mobile phone company or provider. This combination of a (smart)phone with a subscription means that the type and details of that subscription can influence the choice and willingness-to-pay for this product. This is beyond the scope of the current research. The most suitable choice for this research would therefore be to focus on laptops.

3.2 Suitable research method.

One suitable research method is the Choice-Based Conjoint (CBC) Models, which allows for the testing of different attributes that influence the utility a consumer derives from the choice set or choice sets.

In total, there are five factors for the choice-based conjoint analysis, namely:

Factors / levels Number of online reviews Average rating in stars of online reviews Review from store assistant Review from friend or family member Price of the laptop

Level 1 2 2* Negative Negative €599,-

Level 2 44 3* Neutral Neutral €899,-

Level 3 93 4* Positive Positive €1199,-

(21)

21

Most factors and their levels have been explained, but the levels of the reviews from the store assistant and friend or family member need some further clarification. What these reviews state, is presented here:

Factor Level 1 Level 2 Level 3

Store assistant

1= negative: “To be honest, this one is not the best laptop.”

2= neutral: “Not the best deal, but decent quality for an affordable price.”

3= positive review: “One of the best laptops on sale: if you want a good laptop, this is it.”

Friend or family member

1= negative: “This laptop is not so great, there a better ones for this price. I would recommend to look for another one.”

2= neutral: “Laptop is not too bad.”

Friend/family: 3= positive: “I bought this laptop last year, works like a charm: buy it!”

Table 2 – Reviews store assistant and friend or family member.

In sum, the research design would entail a factorial design. An example of a choice set is given in section 3.3, and a full overview of the creation of the choice sets can be found in appendix A.

3.3 Testing of hypotheses

(22)

22

Figure 2 – example choice set conjoint.

It is expected that the respondent’s preference or choice goes to a choice set with a high number of reviews that corresponds with a high(er) average rating compared to the alternatives.

(23)

23

preferred choice of respondents, compared to alternatives that exhibit less congruence. This is measured through the choice-based conjoint analysis.

Next, for hypothesis 3, the inclusion of an attribute level indicating helpfulness score of an online consumer review into the choice sets of reviews can help to find out whether consumers prefer to have an indication of how helpful other consumers found the (set of) reviews. Through presenting a set of reviews, where some do have a helpfulness score and others do not, it is expected that consumers (respondents) prefer the set(s) of reviews that include a helpfulness score, which would indicate that these consumers have a higher willingness-to-pay. Yet, this factor was not included in the choice-based conjoint analysis, since there were already five factors included in the conjoint analysis which can become too big if a sixth factor was included. It was also considered a lesser relevant factor compared to the ones who are included.

Hypothesis 4 can be tested by including an attribute in choice sets that indicates the amount of online consumer reviews for the laptop. By varying this amount through the choice sets, it is expected that for higher total amounts of online consumer reviews respondents would be more likely to pay for their preferred choice set.

(24)

24

Figure 3 – example question source credibility.

It gives an indication which source plays a more influential role on the willingness-to-pay of consumers relative to the other sources. The sources included in the five questions are: 1) a consumer review from a previous buyer; 2) a friend you are shopping with; 3) a study friend; 4) your uncle ; 5) a product tester of a laptop. These a presented through a direct question to the respondent, as indicated by the example above.

Regarding hypotheses 6 (strong-tie, like a friend or family member) and 7 (weak-tie, such as the shop assistant), these can be tested by alternatively presenting a positively, a more-balanced, and a negatively worded review in choice sets presented to respondents. In case a friend or family member, or the shop assistant give a positive review, the willingness-to-pay of consumers is expected to increase. This is measured by presenting these reviews in the choice-based conjoint analysis.

(25)

25

that product involvement is determined by the perceived importance of the product by the consumer, as well as how much interest the consumer takes into the product. Through directly asked questions, an attempt is made to establish a measure of the respondents’ level of product involvement. These questions are as follows:

Question/ level or option

How important do you find a laptop as a product?

How much interest do you have in a product like a laptop?

Level 1 “Not at all important” “None at all”

Level 2 “Slightly important” “A little”

Level 3 “Moderately important” “A moderate amount”

Level 4 “Very important” “A lot”

Level 5 “Extremely important” “A great deal”

Table 3 – Product involvement questions and response levels.

3.4 Data collection plan.

(26)

26

4. Results.

4.1. General population.

4.1.1. Descriptive statistics.

The survey that was distributed, has resulted in 210 recorded responses. 3 of them were not fully answered and have been deleted, leaving 207 fully recorded responses. The following descriptive statistics apply to the general population (n=207).

(27)

27

In terms of age, there is a wide range among the respondents, with the youngest being 19 years old, and the oldest 83 years old. The majority of respondents is quite young though, with 169 respondents aged 30 years or younger. To check how this influences the logit models for the conjoint analyses, a subset is made under section 4.2.3. and used in two logit models: one with all the variables and no moderators (ml15), and a second one including all variables and moderating effects (ml16). The results can be found in section 4.2.3.

On the subject of product involvement which is hypothesised to have a moderating effect on the relationship between reviews and consumer willingness-to-pay, a substantial portion of respondents indicate that they find a laptop a very important (=106) or extremely important (=68). Based on this measurement, 174 out of 207 respondents (approximately 84%) indicate to have a high involvement level with laptops. These numbers are slightly lower for level of interest in a laptop, with 44 respondents indicating to have a lot of interest and 73 a great deal, totalling 117 out of 207 respondents (approximately 57%). It indicates that the majority of respondents have a high level of product involvement, based on these two measures. A full overview of all responses to the product involvement questions is presented below.

Variable/ level Level 1 Level 2 Level 3 Level 4 Level 5 Total

Involv_Import n=2

(0,97%) n=4 (1,93%) n=27 (13,04%) n=106 (51,21%) n=68 (32,85%) n=207 (100%)

Involv_Interest n=3

(1,45%) n=16 (7,73%) n=71 (34,30%) n=44 (21,26%) n=73 (35,27%) n=207 (100%)

Table 4 – Descriptive statistics product involvement measures.

The other moderator, source credibility, has five reviews which differ in terms of their source. These are as follows:

Review1 = A consumer review from a previous buyer Review2 = A friend you are shopping with

Review3 = A study friend Review4 = Your uncle

Review 5 = A product tester of a laptop

(28)

28

A further description of each review can be found in appendix B. Each review has five choice options (levels), as can be observed from the following table with the distribution of responses:

Table 6 – Response distribution source credibility of reviews.

From table 6 it becomes clear that most respondents find reviews from a previous buyer (1), a study friend (3), and a product tester (5) the most credible. Hence, for these three review it is expected that they have a positive sign and a greater estimate size in the logit models compared to reviews 2 and 4. These reviews for the moderator source credibility are introduced in logit model ml9 in section 4.1.3. This also counts for the moderator product involvement.

The next descriptive statistic highlights which alternative was mostly chosen per question in the conjoint part of the Qualtrics survey. For details on the choice sets/alternative, see appendix A. The absolute and relative importance of the alternatives are presented in the following table:

(29)

29 Selection_dummy=TRUE (=1) Question/Alternative 1 2 3 4 Total 1 116 33 43 15 207 2 22 39 132 14 207 3 83 36 53 35 207 4 35 48 98 26 207 5 32 114 34 27 207 6 21 101 35 50 207 7 54 78 26 49 207 8 90 39 38 40 207 Relative importance/Percentage Question/Alternative 1 2 3 4 Total 1 56,04% 15,94% 20,77% 7,25% 100,00% 2 10,63% 18,84% 63,77% 6,76% 100,00% 3 40,10% 17,39% 25,60% 16,91% 100,00% 4 16,91% 23,19% 47,34% 12,56% 100,00% 5 15,46% 55,07% 16,43% 13,04% 100,00% 6 10,14% 48,79% 16,91% 24,15% 100,00% 7 26,09% 37,68% 12,56% 23,67% 100,00% 8 43,48% 18,84% 18,36% 19,32% 100,00%

Table 7 – (Relative) importance alternatives per question in CBC part survey.

For all choice sets (questions), there is one alternative that is clearly preferred over the others, and this also holds for choice set 7, though the choices for alternatives are more balanced here compared to the other choice sets. The most popular choice of each choice set is discussed to establish whether there is a pattern in the choice behaviour of respondents.

(30)

30

rating is 3stars, neutral review from store-assistant, positive one from a friend/family member, and price is €899,-.

The second choice set is dominated by alternative 3, which is characterised by a number of reviews of 93, average rating of 4stars, positive reviews from both the store-assistant and a friend/family member, and a price of €1199,-

The third choice set has alternative 1 as the most popular choice, which is characterised by 93 reviews, average rating of 4stars, a neutral review from the store-assistant and a positive one from a friend/family member, and a price of €1199,- for the laptop.

In the fourth set, alternative 3 is most popular. Its characteristics are 2 reviews, average rating of 4stars, positive review from store assistant and a neutral one from a friend/family member, with a price tag of €899,-

In the fifth set, alternative 2 is mostly chosen, and this alternative is characterised by 2 reviews, average rating of 4stars, neutral review from store-assistant and a positive one from a friend/family member, with a price tag of €599,-

The sixth set has alternative 2 as the most popular choice, and this choice is characterised by 93 reviews, average rating of 4stars, a neutral review from the store-assistant and a positive one from a friend/family member. The price in this alternative is €899,-

In the seventh set, alternative 2 is the popular choice, and it is characterised by 93 reviews, average rating of 3stars, a positive review from the store-assistant and a negative one from a friend/family member, and with a price of €599,-

The eighth and final set has alternative 1 as the most popular choice, which consists of 44 reviews, an average rating of 4stars, a negative review from the store-assistant and a neutral one from a friend/family member, with a price of €599,-

(31)

31

neutral and positive, or both positive, the willingness-to-pay for the laptop increases: again, judging from the prices that are associated with the alternatives which are €899,- and €1199,-. The true effect of variables such as number of reviews is further addressed in the logit models in section 4.2.1.

4.1.2. Effect coding.

Prior to building and executing the logit models, certain variables need to be adjusted in order for the logit models to run. These are the variables that are included in the choice-based conjoint, which are: number of reviews, average rating(stars), review from

store-assistant, review from friend/family member, and price. These variables are effect coded,

whereby the following levels are taken as the reference level:

Variable Number of reviews Average rating(stars) Store-assistant Friend or family member Price Reference level Level 3= 93reviews Level 3= 4stars Level 3 = positive review Level 3 = positive review Level 3 = €1199,-

Table 9 – Reference levels variables for effect coding.

4.1.3. Correlation statistics.

After effect coding, an investigation on correlations between variables is executed on the variables that are used in the logit models. Using the Pearson method, only correlations with a value of 0.3 or higher are reported. This results in the following findings:

Correlated variables Correlations value

Involv_import & Involv_interest 0.623

Price.1 & Price.2 0.500

Price.2 & Number_of_Reviews.2 0.313

Price.1 & Average_rating_stars2 0.313

Number_of_Reviews.1 & Number_of_Reviews.2 0.500

Price.2 & Average_rating_stars2 0.438

(32)

32

Average_rating_stars2 & Average_rating_stars3 0.500 Average_rating_stars2 & Friend_or_family.1 0.688 Store_assistant.1 & Store_assistant.2 0.500 Store_assistant.2 & Friend_or_family.1 -0.500 Store_assistant.2 & Friend_or_family.2 -0.438 Friend_or_family.1 & Friend_or_family.2 0.500

Table 10 - Correlations.

Of these correlations, those between Price.1&Price.2, Number_of_Reviews.1&Number_of_Reviews.2, Store_assistant.1&Store_assistant.2, and Friend_or_family.1&Friend_or_family.2 can be explained through their mutual

relationship with the reference level, since those variable levels are relative to the reference level. Hence, a correlation value of 0.500 is expected.

The correlations between Price.2&Number_of_Reviews.2 and

Price.1&Average_rating_stars2 indicate some relation between the levels of the variables,

but this is not expected to be an issue when running the logit models.

The levels of variables that have a correlation with each other that is higher than 0.6 however, indicates that if the value of (for example) Average_rating_stars2 increases,

Friend_or_family.1 will also increase. This particular relationship can be explained, since

a lower rating of stars (e.g. 2stars) usually occurs in combination with a negative review, in this instance from a friend or family member.

(33)

33

4.2 Logit models.

4.2.1. Logit models general population.

For the conjoint analyses, a data-frame called CBC4 is created, see appendix D. As the dependent variable, Selection_dummy is used where a 0 indicates no choice, and a 1 indicates a choice.

The following independent variables are used in the logit models:

Price.1,Price.2,Number_of_reviews.1,Number_of_reviews.2,Average_rating_stars2,Average_ rating_stars3,Store_assistant.1,Store_assistant.2,Friend_or_family.1,Friend_or_family.2, None_option.

As for the moderators, these are included in model ml9 through the following way:

I(Review1*Price.1). The is consequently applied for all (partworth) variables and

moderators that are included in the logit models.

The models that are discussed, are the intercept only model ml14, the initial logit model ml3 (includes variables Price.1,Price.2, and None_option), the model with all the variables ml7, the model with all the variables and moderating effects ml9. For a full overview of logit models and their output, see appendix C.

Coefficients and significance general population model.

(34)

34

From model ml3, more models were built, each adding another variables at a time. For the sake of brevity however, not all of these models are discussed here, although the code for them is added in appendix D.

Model ml7 results from the iterations as the model with all the partworth variables included for estimation. From its output, all variables were found significant, except for

Price.2 (p=0.1502103) and Average_rating_stars3 (p=0.5136439). These are both the

second and middle levels of their original variables. Respondents find the value of these levels insignificant compared to the reference level of the variables (which is level 3), and the small estimate values (0.072168 and -0.029425) underline this. All other variables have a significant influence on the choice behaviour of respondents.

Model ml9 includes the moderating effects, next to all the variables. This is quite an extensive model as can be observed from the output and code in appendices C and D, respectively. After executing the model, only the moderating relationships I(Review1 *

None_option), I(Review3 * Price.2), I(Review3 * Number_of_reviews.1), and I(Review5 * None_option) were found significant. Their estimates, p-values and odds ratios are as

follows:

Moderator Estimate p-value Odds ratio

I(Review1*None_option) -0.1627161 0.006513 <0.01 0.8498325

I(Review3 * Price.2) -0.1136882 0.037344<0.05 0.8925362

I(Review3*Number_of_reviews.1) -0.1266159 0.018677<0.05 0.8810720

I(Review5 *None_option) 0.1808011 0.021843 <0.05 1.1981769

Table 11 - Significant moderating relationship ml9

Review1, Review3, and Review5 are considered (very) credible by most respondents. If Review1 is encountered with None_option in the conjoint analysis, then this has a

(35)

35

Marginal effects.

For model ml3, as with the coefficients and their significance, the marginal effects of

Price.1 and None_option are significant at p-levels of 0.05 and 0.001, respectively. The

sizes of these marginal effects are 0.0182991 for Price.1, and -0.1268871 for the

None_option. It means that for the Price.1 the probability of observing 1 in the dependent

variable Selection_dummy increases with 0.0182991 if Price.1 is the choice set.

As for model ml7, all marginal effects are significant, except for Average_rating_stars3 as can be observed from the output in appendix C. Thus, for all variables, except

Average_rating_stars3, a 1 unit increase in the independent variables, will lead to an

positive or negative effect on the probability of observing 1 in Selection_dummy, depending on the sign of the estimates, as can be seen from the marginal effects table of model ml7.

The marginal effects for model ml9 present a slightly different picture, compared to the findings related to the coefficients for the model. The marginal effects that are significant are I(Review1 * None_option), I(Review2 * Average_rating_stars3), I(Review3 * Price.2),

I(Review3 * Number_of_reviews.1), I(Review3 * Average_rating_stars2), I(Review4 * Price.2), I(Review5 * Number_of_reviews.2), I(Review5 * None_option), I(Involv_Import * Store_assistant.2), and I(Involv_Interest * Store_assistant.2).

Apparently, the influence of the moderators is much more present when considering the marginal effects of model ml9, compared the actual variables that are included in this model as well. Again, none of the variables were found significant, this time with regard to the marginal effects. The constructs for source credibility and product involvement leave their mark such that the original variables are rendered insignificant in comparison to model ml7. In model ml7, almost all variables are significant.

(36)

36

Performance/evaluation of logit models general population.

Model Loglikelihood-value

AIC BIC Hitrate

ml3(start model) -2248.5 4502.912 NA 42,27%

ml7(model with all variables, without moderators)

-2060.4 4142.864 NA 53,59%

ml9 (model with all variables and moderators)

-2017.5 4210.945 NA 53,59%

ml14(intercept only) -2250.4 4506.73 NA 42,27%

Table 12 - Validation measures general population.

Model ml7 performs best on the AIC criterion on which it holds the lowest value. For the Loglikelihood however, model ml9 performs best, holding the lowest value on this dimension. The BIC was not available for all models, and both model ml7 and ml9 hold the same hitrate in terms of their predictive capability. Therefore, the likelihood ratio tests determine which model is preferred over the other.

Likelihood Ratio (LR)-test.

Lrtest(ml7,ml9)

#Df LogLik Df Chisq Pr(>Chisq) 1 11 -2060.4

2 88 -2017.5 77 85.919 0.2278

Table 13 – LR-test logit models ml7 and ml9

(37)

37

4.2.2 Subset: population highly educated respondents.

A subset of highly educated respondents was created to investigate if this population exhibits different response behaviour than the general population. This population holds all respondents with an education level of 4 (WO, scientific education) with n=119. In a similar structure as for the general population, the coefficients and significance are discussed first.

Coefficients and significance.

For this subset, models ml12 and ml13 are created. These are similar to ml7 and ml9 for the general population, with ml12 being the model which includes all variables, and ml13 the model which includes all variables and all moderating effects.

For model ml12, Price.1, Number_of_reviews.1, Number_of_reviews.2, Average_rating_stars2, Store_assistant.1, Store_assistant.2, Friend_or_family.1, and the None_option are all found significant. The sizes of their estimates and odds ratios are as

follows:

Variable Estimate p-value Exponent of estimates (odds ratio) Price.1 0.440716 7.759e-08<0.001 1.5538186 Number_of_reviews.1 -0.253064 0.0001516<0.001 0.7764181 Number_of_reviews.2 0.136896 0.0628960<0.1 1.1467085 Average_rating_stars2 -0.420140 3.482e-05<0.001 0.6569547 Store_assistant.1 -0.202909 0.0138866<0.05 0.8163527 Store_assistant.2 -0.164510 0.0592325<0.1 0.8483093 Friend_or_family.1 -0.468682 0.0001965<0.001 0.6258265 None_option -0.311309 0.0004235<0.001 0.7324878

Table 14 – Coefficients and odds ratios.

(38)

38

opinion has therefore quite some impact on highly educated respondents. The other estimates can be interpreted in a similar way.

Compared to model ml7 of the general populations, there are some difference in sizes of the coefficients for the two distinct populations. For example, the estimate for

Store_assistant.1 is -0.245097 for ml7, but -0.202909 for ml12. This means that a

negative review from a store-assistant has less negative impact on the probability that a highly educated respondent chooses to buy a laptop.

For another variable, Price.2, the size coefficient is more positive for highly educated respondents (0.099315 vs. 0.072168). This indicates that highly educated respondents are willing to pay more for a laptop, since the size coefficient of Price.2 compared to the reference level is higher than that of the general population. In other words, highly educated respondents are more often willing to pay the price of €899,- for a laptop. For model ml13, Price.1 is found significant, just as None_option, I(Review1 * None_option), I(Review2 * None_option), I(Review3 * None_option), I(Review4 * None_option), I(Review5 * Price.1), I(Review5 * Price.2), I(Review5 * Average_rating_stars2), I(Involv_Interest * None_option). For ml13 compared to ml9 there are also some size differences for the coefficients, as well as for significance. Variable Price.1 is not significant in model ml9, but is in model ml13. The sign and size of the coefficient changes from -0.2147422 to 1.3798e+00.

Another measure to address these models are its marginal effects, and these are presented next.

All marginal effects for model ml12 are significant, except for Average_rating_stars3. The sizes of the marginal effects are comparable to those of the general population. There are some minor differences: the size coefficient for Average_rating_stars2 is -0.1006459 for highly educated respondents, and -0.0754665 for the general population, and for

None_option the sizes of the marginal effects are -0.0783270 and -0.1055748 for the

(39)

39

is part of a choice set, then the probability of Selection_dummy becoming 1 is less negatively influenced for highly educated respondents, compared to the general population.

For model ml13, the marginal effects of variables Price.1, None_option,

I(Review1*None_option), I(Review2*None_option), I(Review3 *Number_of_reviews.1), I(Review4*None_option), I(Review5 * Price.2), I(Review5 * Price.1), I(Review5*Average_rating_stars2), I(Involv_Interest * None_option) are significant.

Performance/evaluation of ml12 and ml13.

Model Loglikelihood-value

AIC BIC Hitrate

ml12 (subset highly educated, all variables, no moderators (3808 observations/32=

169respondents))

-1179.2 2380.326 NA 64,39%

ml13 (subset highly educated, all variables, all moderators)

-1126.8 2429.504 NA 64,39%

Table 15 – Performance and predictive capability logit models ml12 and ml13.

Both ml12 and ml13 have the same hitrate and predictive capability in that term, but the AIC is lower for model ml12. The Loglikelihood however, is lower for model ml13.

LR-test.

lrtest(ml12,ml13) Likelihood ratio test

#Df LogLik Df Chisq Pr(>Chisq) 1 11 -1179.2

2 88 -1126.8 77 104.82 0.01923 * ---

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(40)

40

The LR-test indicates that model ml13 explains significantly more compared to model ml12. Therefore, on the basis of the lowest Loglikelihood, a similar hitrate, and a significant LR-test, model ml13 is the preferred model for the population of highly educated respondents.

4.2.3. Subset: young respondents (age<=30 years old).

The subset involving respondents aged 30 years or younger, was created to compare their response behaviour relative to the general population. This population holds 169 respondents.

Coefficients and significance.

Again, the models that are used for this subset are comparable to ml7 and ml9 (and therefore also to ml12 and ml13). Model ml15 entails all variables, and model ml16 consists of all variables and moderating effects.

For model ml15, all variables are significant, except Average_rating_stars3, see the output in appendix C. Price.1 has the greatest significant size estimate (0.454811, p=<0.001), meaning that for respondents younger than 30 years, the a price level of €599,- in a choice set increases the probability of Selection_dummy becoming 1 by 0.454811.

As for model ml16, Number_of_reviews.2, I(Review1 * None_option), I(Review3 * Price.2),

I(Review3 * Number_of_reviews.1), I(Review3 * Average_rating_stars2), I(Review4 * Price.2), I(Review4 * None_option), I(Review5 * Number_of_reviews.2), and I(Review5 * None_option) are significant.

For model ml15, the marginal effects of all variables are significant, except for

Average_rating_stars3 (again).

With regard to model ml16, this counts for Price.2, Number_of_reviews.2, I(Review1 *

(41)

41

Performance/evaluation of models.

Model Loglikelihood-value AIC BIC Hitrate

ml15 (Respondents aged <=30y)(5408 observations/32 = 169 respondents) -1668.9 3359.772 NA 64,98% ml16 (Respondents aged <=30y) -1621.6 3419.188 NA 64,98%

Table 17 - Validation measures logit models ml15 and ml16.

The hitrate of both models is the same, Loglikelihood value is lowest for ml16, while the AIC is lowest for ml15. BIC could not be produced. The following LR-test determines if model ml16 explains significantly more than ml15.

LR-test.

lrtest(ml15,ml16) Likelihood ratio test

#Df LogLik Df Chisq Pr(>Chisq) 1 11 -1668.9

2 88 -1621.6 77 94.584 0.08467 . ---

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Table 18 – LR-test models ml15 and ml16.

(42)

42

Chapter 5 – Theory vs. research findings.

In this chapter, it is time to reflect on the results from chapter four, relative to the theory from chapter two. First and foremost, the research questions and the problem statement are addressed.

Online- and offline social networks have the ability to influence consumers’ behaviour. This can be generally applied, if consumers search for products and services and make inquires to gain information about product- or service offerings in the market. For a high-involvement, electronics product such as a laptop, this is no different. Through inclusion of constructs like the number of reviews and the average rating of those reviews, the influence of the online social network was represented.

In most instances, the number of reviews in the chosen alternatives was medium (level 2 = 44 reviews) or high (level 3 = 93 reviews). Furthermore, for the average rating in numbers of stars given to a laptop, most respondent chose a choice set where the number of stars was at least three, but often the majority went for four stars (highest level, level 3). The combination of a medium or high number of reviews with an average rating of three or four stars is often observed. For the most popular alternatives among the eight choice sets, none of them had an average rating of two stars. It did occur that the most popular alternative involved two reviews, but never two stars. This indicates that the average rating in number of stars is stronger valued than the number of reviews by the respondents. In other words, in the online social network, the average rating of a laptop is more important than the number reviews that have been given.

(43)

43

the offline social network, both weak- and strong-ties influence consumer willingness-to-pay with their reviews.

For retailers and manufacturers this means that they should strive for the best possible score on the combination of number of reviews and the average thereof, though the average rating should be the more important element. Rather to have 2 reviews with an average rating of four stars, than 93 reviews with an average rating of two stars.

As for the offline social network, instructing store-assistant or other clerics employed in a company’s store to give fair reviews, but avoid negative ones if possible. This helps to increase the consumers’ willingness-to-pay. Training of employees in soft skills can also help to give fair and positive reviews to (potential) customers.

The reviews and recommendations from friends and family are not under direct control of retailers and manufacturers, but ensuring customer satisfaction with products and/or services, presenting an appealing and pleasurable image of the company, and offering certain products and services for free to customers (whether they subsequently bring new customers in or not) can help to create a favourable word-of-mouth in the offline social network. Yet, this remains a challenging area of the social network, not just offline, but online as well.

This counts particularly so for those companies that employ multi- or omnichannel approach towards consumers. There are coping strategies such companies can make use of, and many do so already by integrating the systems and points-of-contact with (potential) customers, whether those points-of-contact be online or offline. For example, a system with a free flow of information between stores, sales agents, and customer service can help to get information about all the encounters a person has had with company, what happened during those interactions. This can help a company’s employee to have the most recent and accurate information about a (potential) customer, so that the employee can give the best advice and review. Making efforts to give the best advice and review can in turn increase the willingness-to-pay of the customer.

(44)

44

reviews are read by a customer may differ, but this research did not find a lower or upper limit to the amount of online consumer reviews to have an influence on consumer willingness-to-pay. Both a high number of reviews (93), as well as a low number (2 review) can influence consumer willingness-to-pay.

As far as what constitutes are qualitative online consumer review (OCR), Schlosser (2011) and Gunawan and Huarng (2015) find that evaluative consistency and source

credibility indicate the quality of an OCR, more so than argument quality. This research

has evidence for this.

Earlier it was established that if the reviews of the weak- and strong-ties (store-assistant vs. friend/family member) are evaluative consistent with each other, the consumer willingness-to-pay increases. However, if one was a negative review and the other one neutral or positive, the willingness-to-pay for a laptop decreased.

Furthermore, the combination of the average rating in stars, with the reviews from the store-assistant and friend/family member also indicated that respondents favour a medium or high average rating (3 or 4 stars) with evaluative consistent reviews (neutral and/or positive) from weak- and strong-ties. This hints at the importance of evaluative consistency across the review-measures of the online- and offline social networks. For product categories of high-involvement product like a laptop, this means that it is important to have credible, evaluative consistent review measures (ratings, number of reviews, but also other frequently used measures, like net promoter score and endorsement from an authoritative body (e.g. a good review from Tweakers on a laptop). Then, the consumer willingness-to-pay is expected to be higher compared to the situation where review measures are not credible and evaluative consistent.

This may not apply though for low-involvement product categories, where the need for information gain and assurance of quality is presumably less important compared to high-involvement product categories.

(45)

45

review measures, retailers and manufacturers can turn to those consumers who gave a review. From those consumers, retailers and manufacturers can find out which part(s) of the product or service they are providing was/were dissatisfactory. These parts are probably the cause of consumers being satisfied with certain elements of the provided service or product, but not with other elements. In turn, this can create evaluative inconsistency in the review measures.

As for providing an answer to the problem statement: the online- and offline social networks can influence consumers’ willingness-to-pay to a (great) extent. Not so much through the influence of their own, separate influence on willingness-to-pay, but more so through their combined power. In particular the factors of source credibility and evaluative consistency are influential, before argument quality plays a role. In response to this, retailers and manufacturers can optimise their strategy to maximise consumer willingness-to-pay through safeguarding the balance in the review measures.

Furthermore, source credibility plays an influential role as a moderator, since most significant moderating effects in the models in chapter 4 are those from the source credibility constructs. Moreover, the analyses found that reviews from previous buyers (Review1), an authoritative and trustworthy review body like Tweakers (Review3), and the opinion of a product tester (Review5) are found most credible by respondents. Thus, positive and balanced review from such sources can also optimise consumer willingness-to-pay for high-involvement products, such as a laptop. After all, this optimum is generally speaking the strategy retailers and manufacturers would like to employ.

As for hypothesis 1 of this research: is was not fully evident if less variance of online consumer reviews leads to a positive influence on consumer willingness-to-pay. However, there are indications that less variance in terms of number of reviews and

average rating in stars in an alternative is associated with a higher price if that

alternative is chosen. A consumer is willing to pay more in that case.