The impact of online consumer reviews
and trust at price comparison sites
The case of a price comparison site for health insurances
2
The impact of online consumer reviews
and trust at price comparison sites
The case of a price comparison site for health insurances
University of Groningen
Faculty of Economics and Business
MSc. Marketing Intelligence and Marketing Management
Author: Albert Koller Date: January 16, 2017 Address: Leeuwarderstraat 23
9718HV Groningen Phone number: +31636491082
E-mail address: a.koller.1@student.rug.nl Student number: s2206226
First Supervisor: dr. H. (Hans) Risselada
3
MANAGERIAL SUMMARY
The internet causes consumers to be able to obtain tons of information before doing a purchase. A widely used online tool for consumers which enables consumers to obtain information on price and non-price factors are price comparison sites (PCS). In the setting of a PCS, a consumer can obtain information on the price rank and the review characteristics of a certain product or service. This research investigates the impact these factors have on the choice consumers make and the purchase intention consumers show on a PCS. A possible moderating effect comes from the trust people perceive when visiting a PCS. This is
investigated both with a main effect on purchase intention and as a moderating effect with the other attributes. To gather a first insight in the heterogeneity that is possibly present in the population a latent class analysis is conducted. Data was collected through an online survey with choice based conjoint tasks where respondents had to choose their preferred health insurance. To investigate the transition respondents make when the setting of the PCS changes from fully independent to insurer-owned, a within-subject design is used with 5 choice sets in the first situation and 5 choice sets in the second situation. The results show that the attributes of price ranking, review valence and review volume are all important for
4
ABSTRACT
Every year millions of Dutch residents have the possibility to switch to another health
insurance company. To assist this choice behavior, various price comparison sites are present to show customers their desired information. This research investigates what drives customers to choose a certain health insurance via a PCS. Using a choice based conjoint design, it is shown that review valence is the most important attribute for customers to choose their preferred health insurance contract, before price ranking and review volume. Trust is
important when it comes to actual purchase intention and as moderator on valence: customers who perceive more trust appreciate review valence more than people who perceive less trust. Independence is furthermore important for customers visiting a PCS when it comes to purchase intention. The price sensitivity of certain customers disappear when the context of the situation of the PCS changes as an explorative latent class segmentation shows, indicating a relationship between trust and price sensitivity. Further research should investigate the possible importance of other PCS attributes, as well as the generalization of these results towards other products and services.
Key words
5
PREFACE
With this thesis I will finalize my Master Marketing Management and Marketing Intelligence as well as my time as a student. The past five and half years here in Groningen were great and I am looking back at a very inspiring time here at the University of Groningen.
Writing this thesis was a very challenging but rewarding process. I would like to thank dr. Hans Risselada for his guidance and very valuable feedback throughout the process. I would like to thank dr. ir. Maarten Gijsenberg for his role as second supervisor and my fellow group students who provided valuable insights during the group discussions. A special thanks to my family and girlfriend who supported me during my whole career as a student.
I hope you enjoy reading my thesis. All the best,
6 TABLE OF CONTENT MANAGERIAL SUMMARY ... 3 ABSTRACT ... 4 Key words ... 4 PREFACE ... 5 1. INTRODUCTION ... 8 2. THEORETICAL FRAMEWORK ... 13 2.1 Conceptual model ... 13
2.2 Choice and Purchase intention. ... 13
2.3 Price comparison websites. ... 14
2.4 (Electronic) Word Of Mouth. ... 15
2.5 Perceived trust. ... 18 3. RESEARCH DESIGN ... 22 3.1 Methodology ... 22 3.2 Data collection ... 23 3.2.1 Within-subject design ... 24 3.3 Plan of analysis ... 26 4. RESULTS ... 27 4.1 Descriptive analysis ... 27 4.2 Construct validity ... 28
4.3 Choice-based conjoint analyses ... 29
4.3.1 Model validity ... 29
4.3.2 Main effects of used attributes. ... 29
4.3.2 Interaction effects of attributes ... 31
4.4 Trust analyses ... 33
4.4.1 Main effects of trust ... 33
4.4.2 The moderating effect of trust on attributes ... 33
4.4.3 The effect of independence of the PCS ... 34
4.5 Latent Class analysis ... 35
4.5.1 Model selection ... 35
4.5.2. Two segment solution ... 36
4.5.3. Transition in utility ... 37
4.6 Overview of hypotheses ... 39
7
5.1 Theoretical implications ... 40
5.1.1 Review characteristics and price ranking ... 40
5.1.2 The influence of trust ... 41
5.1.3 Segment differences ... 43
5.2 Managerial implications ... 43
5.3 Limitations and further research ... 44
6. REFERENCES ... 46
7. APPENDICES ... 52
Appendix A – Pre-test interview ... 52
Appendix B – Survey Design ... 53
8
1. INTRODUCTION
With seven billion devices connected to the internet, spread over more than three billion potential customers, e-commerce is a major topic for (online) businesses (International Telecommunication Union, 2016). Until this moment one billion online shoppers worldwide made at least once an online purchase. Projections are that online retail sales will rise from $334 billion in 2015 to $480 billion in 2019 in the United States alone (Forrester Research, 2015). With similar growth rates for the European Union and an ever increasing wallet share for online sales over offline sales, online sellers want to make sure that their products and services are seen and bought. A possible downside for consumers of the massive shift to online retailing and the enormous amounts of online shops is that they are overloaded by (online) information, and are consequently not able to evaluate to what extent information is valuable or worthless (Lee and Lee, 2004). Consequently, they are constantly searching for online tools and shortcuts that help them making the best online decisions.
This is most certain the case for health insurances. Dutch residents are able to change their health insurance at the end of the year and are then overwhelmed with marketing communication from health insurance companies claiming to have the best quality, least expensive or most popular health insurance. To obtain an overview, consumers often visit a price comparison site (PCS) as Independer (Google Trends, 2016) to assist them in their decision. These PCSs address different aspects of the health insurances. What aspects matter for consumers at those PCSs is the subject of this research.
9
using a PCS can reduce these costs 30-fold compared to offline searching (Smith and Brynjofsson, 2001). Nowadays, PCSs are the primary source for consumers to gain knowledge about the market he or she is searching in, during all stages of the buying process of the customer. (Bodur, Klein and Arora, 2015).
10
One of the two roles of online consumer reviews for consumers is the informant role. When too much online review information is provided, this could lead to information overload and a worse perceived choice quality (Park and Lee, 2007). This could lead to less confident, less satisfied and more confused consumers (Lee and Lee, 2004). Whether the additional information given besides price on a PCS such as online consumer reviews is perceived as overload or as valuable to consumers has not yet received many attention in literature.
A higher price level shows to have a significant negative influence on the eventual rating given by consumers; two identical products with their only difference in price, lead to lower ratings of the more expensive products. Whether this is the case for different products and via different sources including PCSs is suggested as further research (Li and Hitt, 2010). The exact role of online consumer reviews and the impact it has on consumer decision making is not yet a fully understood topic, where further research is required to better understand the new (online) information requirement, where PCSs are a serious subject of (Simonson, 2015; De Langhe, Fernbach and Lichtenstein, 2016).
Combining the topics on online reviews and price comparison sites also has to deal with the topic of the customer journey, one of the most important topics nowadays in literature. This customer journey is made up of all touch points that a customer has with a firm over time, and all these touch points can influence the eventual purchase decision of customers. Many of these touchpoints are not fully understood yet, and the effect of e.g. the interrelationship of various marketing instruments as eWOM and pricing need further research to understand their impact. (Simonson, 2015; Lemon and Verhoef, 2016).
Besides the different marketing instruments online retailers can use to sell their products, there is a very important other reason why consumers use or do not use a certain e-commerce shop; trust. The amount of trust consumers perceive shows to be of key importance if and where to shop online (Lee, Park and Han, 2011). McKnight, Choudhury and Kacmar (2002) developed and validated multiple trust measures that should help the e-commerce in
monitoring the trust visitors perceive when visiting e-commerce retailers. The impact of PCSs and the amount of trust was expressed by an extensive report of the Financial Conduct
11
companies displayed on that PCS (Leeflang et al., 2014). Does it decrease the perceived trust towards the PCS in general, towards the company that owns the PCS or does is have no effect at all?
With these relevant topics in mind the following main research question is derived: The corresponding hypotheses are discussed in the next chapter.
What is the effect of online consumer reviews on the choice and purchase intention of displayed products on price comparison sites and what is the role of perceived trust?
In short, this study investigates the impact of online reviews in the setting of a price comparison site for health insurances. The possible moderating effects of the review characteristics and perceived trust when visiting a PCS are as well taken into account to provide an insight of how price and non-price factors are related in determining the choice and purchase intention of customers visiting a PCS. This research contributes to the current literature of online purchase decisions of consumers by analyzing the choice behavior of consumers at a price comparison site for a complex product, a health insurance. Whether the independence of a PCS is relevant for customers is addressed for the first time with this research, as well as the trust that customers perceive at a PCS. Online vendors and PCSs can benefit from the outcomes by focusing on the elements customers value most.
12
utility on price: the first segment can be interpreted as the very price sensitive group, where segment 2 is more appealed by the review characteristics (mainly review volume).
PCSs and online retailers can benefit from these findings in practice in multiple ways. Online retailers should at first cause satisfied consumers to also review their purchased product or service on a PCS. The PCSs should on the other hand clearly communicate the review characteristics of the products they show to provide customers with their preferred information, without causing information overload. Owning a PCS as an online retailer should at last be clearly communicated to the customers to prevent customers from becoming suspicious, which eventually could lead to less usage of the PCS and less buying intention towards the company that owns the PCS. Finally, retailers should take into account what consumer group they target in their decision on displaying/focusing on price or consumer review elements in their offering.
13
2. THEORETICAL FRAMEWORK
In the following chapter a deeper understanding of the research setting will be provided via a review of the existing literature. The conceptual model including the expected direction of hypotheses is shown to give a clear first insight of what this research is about. The outcome variable will briefly be discussed here after. As the setting of this research is a product comparison site (PCS), the relevant literature on PCSs will be discussed. The main subject of this research is to find out what the impact of online consumer reviews on a PCS is, therefore a deeper understanding of eWOM and, more specific, online consumer reviews is provided. Furthermore, the concept of perceived trust in general and at a PCS is discussed, including the possible moderating effects this topic has.
2.1 Conceptual model
FIGURE 1: Conceptual model
2.2 Choice and Purchase intention.
As the method used in this study is Choice-based conjoint (CBC) analysis, the first outcome variable of this study is the choice respondents make in their evaluation of the difference choice sets. Kostyra et al. (2016) show that online review characteristics are well able to use as attributes is a CBC to predict consumer choices. There are various ways to improve the data gained from CBC analysis. To gain more insight in the choice respondents make a follow-up question can be asked that represent the actual likelihood of choosing their
preferred choice (Qualtrics, 2012). Therefore, as often used in literature on online topics (e.g. Chang and Chen, 2008; Brown, Pope and Voges, 2003) the second, more informative,
14 2.3 Price comparison websites.
As stated in the introduction, product comparison sites (PCS) take a prominent role in the customer journey towards a purchase. Price comparison sites are defined as “electronic intermediaries that assist buyers when they search for product and price information in the internet.” With the provided price (and non-price) information consumers can easily compare offers and choose their preferred deal (Moraga-Gonzales and Wildenbeest, 2012). A PCS is a profit-motivated enterprise which has to balance the desires of shoppers, the desires of sellers and the costs of providing their services (Allen and Wu, 2010). Jung, Cho and Lee (2014) found that a PCS reduces the variability in the acceptable price range for consumers. This means that less knowledgeable customers are likely to perceive lower prices than knowledgeable customers: the knowledge gained from a PCS causes more certainty about the acceptable price range. Another finding is that the influence of a PCS differs over product types: transaction and acquisition value are more increased when using a PCS for buying non-look-and-feel products (e.g. consumer electronics) than for buying a non-look-and-feel product (e.g. clothes). A study of Brown and Goolsbee (2002) showed furthermore that when more people started using the internet to compare prices of insurances, prices decreased significantly, indicating the importance of price transparency for customers. A 10% increase of people comparing their prices before purchasing, led to a 5% decrease of average prices. Pricing shows over time to be a key influencer of purchase intention. Findings show that the average price elasticity (the percentage change in demand after a percentage change in price) is -2.6%. (Bijmolt, Van Heerde and Pieters, 2005). More specifically in the context on online sales, it is shown that, pricing strategy is at least as influential to sales as in offline situations, with product-price elasticities being negative in general (Ghose and Sundararajan, 2006). In the field of health insurance some studies on price elasticity of demand for health insurance are conducted. In the Netherlands this price elasticity is shown to be relatively inelastic compared to other countries with a price elasticity of about -0.5. However, results on price elasticity after the health insurance reform in 2006 in the Netherlands are scarce and are expected to nowadays be more elastic. (NZA, 2006; Pendzialek, Simic and Stock, 2016). Based on above analysis we state:
15 2.4 (Electronic) Word Of Mouth.
Word of mouth has been a subject of research for a long time. Definitions vary but nowadays the widely used definition is “the act of consumers providing information about goods, services, brands or companies to other consumers” (Babic Rosario et al., 2016.) It is often called the most effective, yet least understood marketing strategy (Godes and Mayzlin, 2004). When this information is communicated over the internet in any form (e.g. tweets, reviews or blog posts) this is called electronic word of mouth (eWOM). The influence of WOM was first studied back in 1955 by Katz and Lazarsfeld. They found that WOM was seven times more effective compared to print advertising when influencing customers to switch brands. With the growing importance of eWOM nowadays, the growing interest in this subject was investigated through a meta-analysis of Babic Rosario et al. (2016), researching 96 studies that involved eWOM research.
Between 1960 and 2000 over 70 marketing studies have been conducted to investigate the effect of word of mouth (Trusov, Bucklin and Pauwels, 2008). The introduction of eWOM in 1995 can overcome the limitation of traditional WOM that direct observation is not possible. All available eWOM can namely at some point in time be observed and some types are even controllable for marketers. Online marketers can for example decide in what format certain comments should be placed and whether or not to show online reviews and star ratings on their e-commerce website (Park and Kim, 2008). However, the majority of eWOM is uncontrollable for companies and marketers. Especially with the rise of social networking sites the last years a huge amount of (unstructured) data about brands and companies is publicly placed on the world wide web every day (Chu and Kim, 2011).
16
In this study, one specific type of eWOM is investigated: the online consumer reviews. Chen and Xie (2004) did extensive research on what functions online consumer reviews should serve. They found e.g. that complicated, high-tech products benefit more from consumer reviews than simple, low-tech products. Furthermore, mass-products gain more from consumer reviews than niche products.
In the field of health insurances, little scientific research has been done on the effect of online consumer reviews on the choice of health insurances. However, influential reports from e.g. consumerreports.org show the relevance of rating health insurers, and the value customers generate from (expert) reviews. (Consumerreports, 2015).
Based on the above findings the following hypotheses are stated and tested:
H2a: The volume intensity of online consumer reviews is positively associated with the choice probability and purchase intention of the displayed product.
H2b: The valence intensity of online consumer reviews is positively associated with the choice probability and purchase intention of the displayed product.
A reasonable or attractive assumed price showed on a PCS could boost the purchase intention of online consumers, more positive and a higher number of consumer reviews could as well, but an interesting next step which makes a PCS relatively unique is the interaction that could be present between price ranking and online reviews, and the relative importance of these factors.
17
rising importance for consumers of online consumer ratings. Thinking one step further this finding could also be tested for the case of a PCS. The positivity of review characteristics could weaken the influence of price ranking.
When it comes specific to valence of online reviews, Ho-Dac, Carson and Moore (2013) suggest a mechanism that explains the effect of positivity of online reviews on the influence on price. When a product is rated more positively this indicates a higher quality product and a higher willingness to pay for the product. This fits in in the line with the finding of De Langhe, Fernbach and Lichtenstein (2015) that the rating of products becomes more important and that this rating in combination with a higher price reflects the best quality. Kostyra et al. (2016) analyzed furthermore the different aspects of review characteristic in a choice based conjoint setting. Willingness to pay (WTP) was included in this study and it is found that an increase in valence increased the WTP for customers. This suggests that higher rated products are less influenced by a higher price. It seems that when people choose a high rated product, the effect of price diminishes.
It is shown that a dynamic pricing model based on online consumer reviews has a positive impact on profit maximization. It is based on the finding that the quantity of consumer information available has a positive impact on the sales of a product or service. At the early stage of information sharing (low volume intensity) the optimal price should be slightly lower than when an average amount of information is available. When more (consumer) information about a product or service is available, the optimal price becomes slightly higher. This suggests that the price importance varies over review volume, holding the valence of online reviews constant, with a decreasing importance of price when the volume of reviews of a certain product or service increases. (Wang, Zhang and Zheng, 2011).
Based on these previous researches it is therefore suggested that more online reviews and a better consumer rating for a product causes the influence of price to decrease. The tested hypotheses will be dealing with the explained findings that more consumer reviews and a higher rated product decrease the importance of price factors for consumers.
H3a: A higher volume intensity of online consumer reviews diminishes the effect of price rating on choice probability
18
As the primary reason for visiting a PCS is price information (Moraga-Gonzales and Wildenbeest, 2012) the question is what the absolute importance is of online consumer reviews. Although experience goods as health insurances show to be relatively more influenced by consumer reviews than search goods (Mudami and Schuff, 2010), customers are usually further in their pre-purchase phase of the customer journey at the moment they visit a PCS. They are, furthermore, very price sensitive when visiting a PCS (Smith and Brynjolfsson, 2001).
Comparing price and review impact in the current literature shows some interesting results. In the context of hotel room sales, both price and review characteristics (volume and valence) had a highly significant effect on sales, with not a significant difference in importance towards a purchase decisions found in this study. (Ye, Law and Gu, 2009). De Langhe, Fernbach and Lichtenstein (2015) found that review valence is often of more importance for customers than price over different product categories at amazon.com. This was even the case when the online consumer reviews show little similarity with the quality defined as in external (expert) reports. A contradicting finding is found by Fagerstorm and Ghinea (2011), who researched the relative impact of online recommendations as online consumer reviews using a conjoint analysis. Aggregating results from this study shows an importance distribution of about 3-to-2 in favor of price. When the price is either low or high (compared to market average) this displayed price had even more impact. However, when the price is about average compared to the market price the impact of online recommendations (customer reviews) and price are equal. The products in this case were, however, MP3-players. Whether the results are generalizable to other products and services is subject to further research. Based on the above analysis the following hypothesis is stated:
H4: Price ranking has a bigger influence on the choice decision of consumers in choosing a health insurance than online consumer review characteristics do.
2.5 Perceived trust.
19
well-known PCSs are in general perceived as well trustable but often not informative enough, this is the result of the not well working specific interface of the PCS. This could eventually have serious impact on the purchase intention towards the displayed product.
Kimery and McCard (2002) define (online) trust as “customers’ willingness to accept
weakness in an online transaction based on their positive expectations regarding future online store behavior.” McKnight, Choudhury and Kacmar (2002) built an extensive framework which empirically validated the measures for trust in an e-commerce setting. They show e.g. that benevolence, integrity and competence are main dimensions of online trust but they are very distinct: consumers gauge e-commerce sites in terms of specific attributes instead of in broad. Consumers demand these dimensions often all together, a site can therefore be
perceived as honest (integer) and benevolent but not competent, which in the end will result in a consumer not doing business on that website.
Lee, Park and Han (2011) found that trust is the most important factor in online business transactions, and that a high level of trust leads to a higher purchase intention and more perceived trust in online consumer reviews. Online sellers could therefore increase their sales by creating a trustworthy shopping environment. This could also be the case for price
comparison sites; a trustworthy environment (PCS) which displays the important perceived factors for consumers could have a positive impact on the purchase intention towards the product a consumer is looking at. Furthermore, Ganguly et al. (2010) investigated the effect of trust and website design on purchase intention and found among other things that perceived trust in an online store is significant predictor of purchase intention. In this research there will be investigated if this results holds as well in the case of a PCS. Therefore the following hypothesis is stated:
H5: High perceived trust when visiting a PCS increases the purchase intention of a consumer visiting a PCS.
20
more trust in an online environment, online consumer reviews have eventually more impact on the purchase intention of consumers.
Perceived trust and perceived price both show to have a serious impact on consumers’ purchase intention. Research of the interacting effects of trust and price levels show that customers are more influenced by trust when buying a high-priced product or service than when buying a less expensive product or service (Kim and Benbasat, 2009). Comparing these two elements show that perceived trust exerted a stronger effect on purchase intention than perceived price (Kim, Xu and Gupta, 2012). Analyzing the possible interaction that is present between perceived trust and price shows that high perceived trust leads to consumers that appear to be willing to pay a higher price premium. A cut-off value for trust of 7.2 on a 9-point scale was found from where customers are willing to pay a price premium. This is especially the case for relative expensive products (Ba and Pavlou, 2002). Smith and Brynjolfsson (2001) showed in their research on PCSs that price was not the sole attribute when it comes to consumer decision making on these sites. Brand was used as a proxy for the perceived credibility, where a credible (i.e. well trusted) offer leaded to a higher willingness to pay. These findings can be interpreted as the diminishing effect of price on consumer choice when a high level of trust is perceived. In contrast to the negative effect trust could have on price ranking, the effect of review characteristics on consumer choice could become more important due to higher perceived trust. Xu (2014) shows that the perceived credibility of eWOM is of major importance before it will serve as a decision-making aid, many factors contribute to this credibility. Smith, Menon and Sivakumar (2005) show that consumers who perceive recommendation information on a website as more credible (trustworthy), utilize these recommendations more in their decision making. This finding implies the strengthened effect trust can have on review characteristics in consumer choice decisions.
The following moderating effects of trust are, based on the above analysis, hypothesized: H6a: High perceived trust positively moderates the effect of review characteristics. H6b: High perceived trust negatively moderates the effect of price ranking.
21
mobile commerce, but is likely to hold as well for a PCS (Siau and Shen, 2003). The independent nature of a third-party certification helps customers feel more secure in doing business. This could be a seal of approval which is often perceived as valuable by consumers. This could consequently lead to a higher purchase intention (Pennington, Wilcox and Grover, 2004).
Based on the above information the following hypotheses will be tested:
H7: Consumers perceive higher trust at a fully independent PCS than at an insurance-owned PCS.
A next step in the analyses is to account for the possible presence of heterogeneity. Whether the found effects are different for different segments in the population is an additional research possibility. It is found that price sensitivity is different for different segments of people. Younger and healthier customers are e.g. more price sensitive when it comes to choosing a health insurance (Strombom, Buchmueller and Feldstein, 2002). Whether a brand or online vendor is perceived as credible is also an important condition for the price sensitivity of customers. A credible brand or vendor could decrease the price sensitivity of customers and is therefore less dependent of the corresponding price rank (Erden, Swait, Louviere, 2002). Latent class analysis should indicate whether there is a difference in price sensitivity between respondents.
22
3. RESEARCH DESIGN
3.1 Methodology
To evaluate the different attributes that play their part at a price comparison site a Choice Based Conjoint (CBC) design is used. For a long time, conjoint analysis is marketers’ favorite methodology for finding out how buyers make trade-offs among competing products, services and suppliers (Green, Krieger and Wind, 2001). As this study aims to research consumer choices in the setting of a price comparison site with a limited number of attributes, choice based conjoint analysis is the most appropriate method.
Combining this method with traditional survey questions in the form of Likert-scale questions and basic control questions leads to the desired data set to conduct the necessary analysis to test the stated hypothesis. The choice for CBC is strengthened by the limited number of levels that have to be taken into account, the absence of the number-of-levels effect and the large population of possible respondents. The only requirement for a respondent to attend in the research is that (s)he is a Dutch resident of 18 years or older. This indicates a large population and the possibility to obtain a sample size that is big enough. To investigate the relevance and importance of the review characteristics, price rank and insurance company on a PCS, a choice based conjoint design is an appropriate method to research the included attributes and levels (e.g. Decker and Trusov, 2010). As the attributes and levels (see table 1) are shown to be an accurate representation of a real situation, CBC will generate valuable insights. There are four important criteria when designing a CBC design, this research account for these in the following manner. As every choice set consist of four alternatives with four attributes, all levels appear an equal number of time (balanced design) and every level pair appears an equal number of times as well (orthogonal design). With four levels per attribute and four choice alternatives, every level is only seen once in a choice set, thereby guaranteeing minimal overlap. At last, as there is an expected order in level preferences, there can be accounted for fully dominating alternatives by indicating the expected orders in the survey program, thereby creating utility balance (Eggers and Satler, 2011).
23
visible as could have been the case when balanced level overlap was created in the choice design (Rao, 2014; Chrzan and Orme, 2000). To take into account the moderating effect of trust a multi-item scale is adapted from McKnight, Choudhury and Kacmar (2002) and
(partially) included in the survey. By analyzing this construct using the Cronbach’s Alpha one can analyze if this construct can be included. The moderating effect of trust can thereafter be analyzed by creating additional variables consisting of the trust measure (split in two
dummies, low and high) and one of the attributes. 3.2 Data collection
The data in this research is collected using the preference measurement software
Mypreferencelab, which is able to develop a choice based conjoint analysis in a structured way. Qualitative research with six different people of the research population was conducted on forehand to come up with a research design, attributes and levels that are easy to follow but realistic as well. Furthermore, incentive alignment is used to minimize the dropout rate by giving respondents the opportunity to leave their e-mail address to have a chance to win a coupon (Malhotra, 2010). Appendix A gives an overview of the interview questions.
Combining these insights with previous research leaded to a situation that formed a realistic representation of a PCS, using four attributes, all with four different levels. Table 1 shows the four attributes and their corresponding levels. For an example of a choice set, appendix B gives an overview of the different elements of the survey, including the choice sets.
Attribute Levels
Price Ranking €, €€, €€€, €€€€
Health insurance company R, S, W, X Volume of online consumer reviews 5, 25, 100, 500
Valence of online consumer reviews 2 stars, 3 stars, 4 stars, 5 stars TABLE 1 : Overview of attributes and levels for the choice-based conjoint structure.
24
this included question the no-choice option is not included. No-choice options are especially useful during e.g. the test phase of new products (Eggers and Eggers, 2011). In this case with complex and difficult decisions with health insurances, no-choice options could lead to a loss of valuable choice data of respondents when they perceive the question as too difficult or hard to answer at this time.
After the first five choice sets the situation changes and additional information on one of the health insurance companies (and the PCS) is given. The given message provides a situation where additional information about the PCS is obtained via a friend: one of the health insurance companies is the owner of the PCS, even though the PCS guarantees their
independence. Via this manipulation it is possible to measure the difference in utility, trust and purchase intention for the first five and latter five choice sets, including the transition different respondents make (Bachman and Zaheer, 2006)
For the sake of reality the exact prices of the health insurances are not shown in the choice sets. Respondents only see the relative differences in price, displayed by 1 to 4 euro signs. To improve realism of a PCS, the lowest price alternative is in every choice set shown at first. To prevent any confounding factors for the health insurance company, only letters are used to display the insurance company. Four letters were chosen that do not cause any associations with existing (real) health insurance companies. For the review characteristics four levels were chosen that show almost the minimum and maximum of the possible levels.
The online questionnaire will solely be fulfilled by Dutch respondents and to improve the understanding of the choice sets and questions will consequently be written in Dutch. The main reason why no non-Dutch respondent can be allowed is because of the choice sets involving health insurances. In many other countries than the Netherlands the market and mechanisms of health insurances are different so including non-Dutch respondents could lead to biased results (Van de Ven and Schut, 2008)
3.2.1 Within-subject design
This choice-based conjoint study uses a within-subject research design. This design has
25
less respondents. Although the advantage with a between-subject design is that more choice sets can be asked per respondent, the need for more respondents to identify possible
heterogeneity (different segments) is present (Desarbo, Ramaswamy and Cohen, 1995). When it comes to determining what the minimum amount of respondents is to draw conclusions from a choice-based conjoint analysis the following rule of thumb is often used:
𝑛 ∗ 𝑡 ∗ 𝑎
𝑐 ≥ 500
Here is n the number of respondents, t the number of tasks, a the number of alternatives per task and c the largest number of levels of an attribute.
With the limited amount of attributes (4) and levels (4), the number of choice sets that have to be evaluated to obtain valid results and utilities is limited as long as the number of
respondents meets this criteria. For this research the corresponding minimal sample size would consequently be 100. This is however seen as a bare minimum and a bigger sample is advised (Sawtooth Software, 2010). Literature on conjoint analysis furthermore shows that the gain in validity after 5 choice sets per respondent is limited when a design is used with a limited number of attributed and levels (Haaijer and Wedel, 2000), as is the case with this research. At last the available time to collect the data is rather limited with only about three weeks to collect the majority of the data. This means that maximizing the information
obtained from every respondent is necessary to draw clear and well-founded conclusions from the collected data.
With respect to possible analyses, the within-subject design offers some additional
possibilities over other designs. Besides the aggregated analyses between the two conditions (which are also possible with a between-subject design) there is the possibility to investigate the transition certain segments make in the new described situation. After the first five choice questions the situation changes and to accurately show the possible differences in change of preferences, the within-subject design is capable of doing so. In the context of this research the within-subject design can provide valuable insights in the context of exact transition in trust, attribute importance and eventually choice and purchase intention of certain segments can be determined. Other researches show the strength of this design as well (Ashraf, Bohnet and Piankov, 2006; Desarbo, Ramaswamy and Cohen, 1995)
26
unknown and a context-change happens often when online consumers find out more about a PCS.
3.3 Plan of analysis
After collecting the data using Mypreferencelab the data will be transported to SPSS and checked on missing and false data. This includes checking the results on reliability by looking at the completion time and the given answers. A very fast (less than 3 minutes) response, little variation in the answers and a wrong answer on the control question (the question which company was the owner of the corresponding PCS) could indicate that the respondent paid little to no attention to the asked questions, and will therefore be excluded from further analysis (Malhotra, 2010).
Various analyses will be used to test the stated hypotheses. Choice based conjoint analyses uses a multinomial logit model to obtain utilities from respondents. The utility of consumer n for health insurance i is consequently the sum of the part-worth utilities which can be modeled using the following utility function (Haaijer and Wedel, 2000):
𝑈𝑖 = ∑𝛽𝑘𝑋𝐾𝑗 𝐾 𝑘=1 Where: U = utility sum
i = health insurance offer
∑𝛽𝑘𝑋𝐾𝑗 = 𝑠𝑢𝑚 𝑜𝑓 𝑎𝑙𝑙 𝑢𝑡𝑖𝑙𝑖𝑡𝑖𝑒𝑠 𝑜𝑓 𝑡ℎ𝑒 𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒𝑠 (𝑋: 𝑘 = 1, . . , 𝐾) 𝐾
𝑘=1
Using the attribute betas, one can derive the average utility from a specific choice set over the respondent sample. When aggregating these utilities, however, it is assumed that the sample is homogenous in their preferences (Haaijer and Wedel, 2000) This is not often the case and therefore an additional latent class analyses will be conducted to investigate eventual
27
4. RESULTS
In this chapter the results of the different analyses are elaborated. After showing descriptive statistics of the used data sample some constructs are validated. Thereafter various validation and robustness checks are performed to make sure the data can be used to analyze the different hypotheses in the right manner. The choice based conjoint analyses are conducted, consisting of the main effects of the attributes and the interaction effect between price ranking and review characteristics. Hereafter the effect of trust is taken into account by analyzing trust as predictor for the shown average purchase intention and as a moderator on the attributes of price ranking, review volume and review valence. Furthermore the independence of the PCS will be analyzed with additional tests. From here, there is a switch back to the choice based conjoint analysis, where a latent class analysis is conducted from the first five choice sets to create segments and take into account the heterogeneity of the sample. Thereafter the transition in utilities are analyzed by fixing these segments and analyzing the utility in the choice sets 6 to 10 where the PCS is owned by one of the companies. The chapter is finalized with an overview of the tested hypotheses.
4.1 Descriptive analysis
The first check of the data is an indicative analysis of the completion time and the reliability of the answers. Respondents that showed to be an outlier in terms of completion time (less than 3 minutes) and showed little variance in the answers (same answers on (almost) all Likert-scale questions) were marked as potential unreliable. After deleting the most unreliable results (3 respondents) 166 respondents completed the online survey and analyses will be done over this sample. The dropout rate after reading the title page was 10%, which is rather low compared to the average web survey dropout of 30% (Galesic, 2006). With an average completion rate of 8 minutes respondent fatigue is as much as possible prevented while still collecting the preferred data.
28
25% between 25% and 50% of their online bought services and products, 24% between 50% and 75% of their online bought services and products and 13% use a PCS more than 75% of the time they buy something online.
Characteristic Frequency Percentage
Gender Male Female 71 95 43% 57% Age <23 23-30 31-46 46-53 >53 58 82 4 8 10 35% 49% 4% 5% 7% Highest level of education
Secondary School MBO HBO University 47 10 22 87 28% 6% 14% 52% Employment Working Student Other 41 116 9 25% 70% 5% PCS Usage <25% 25%-50% 50%-75% >75% 63 41 40 22 38% 25% 24% 13% TABLE 2: Sample characteristics
4.2 Construct validity
29
The second construct to be validated is that of familiarity. The questions concerning familiarity with e-commerce, the use of PCS and the use of online consumer reviews were derived from Gefen (2000). A reliability analysis showed that this construct could be taken together and be used as one construct of familiarity with online buying. The corresponding Cronbach’s Alpha is 0.811 and therefore the e-commerce familiarity variable can be generated. Appendix C shows more information on the reliability analyses.
As respondents state their actual purchase intention towards the chosen health insurance after every choice set, the 5 purchase intentions of the choice sets with the fully independent PCS are averaged and saved as a new variable purchase intention. The mean of this variable over all respondents has a value of 5.47 on a 7 point scale. The same has been done with the choice sets 6-10 (with one of the insurance companies owning the PCS), which results in an average purchase intention of 5.19
4.3 Choice-based conjoint analyses
To test the first three hypotheses the data from the choice sets is analyzed using the software package Latent Gold. After validating the model, the main effects from the four attributes are firstly interpreted for the first five choice sets, thereafter the change in attribute parameters and importance will be briefly analyzed. Hereafter the hypothesized interaction effects will be analyzed.
4.3.1 Model validity
To assess the model validity different metrics can be used to analyze. At first a likelihood ratio test is conducted. This compares the tested model with attributes to a null model without any parameters (Chrzan, 1994). The null model shows a log-likelihood value of -1150.62, where the model with the four main attributes results in a log-likelihood -569.50. The corresponding chi-square test statistic is -2*(-1150.62-(-569.50))=1162.24. With 9 degrees of freedom the corresponding p-value is <0.001; the model with attributes therefore performs significantly better than the null model without parameters.
4.3.2 Main effects of used attributes.
30
perceived PCS. Besides the importance of the attributes the aim is as well to discover the differences in utilities between specific levels, therefore the part-worth utility function is used. The first attribute (price ranking) deals with H1. From the utilities of the levels can be derived that a price rank displaying a more expensive health insurance is less preferred by respondents. The range in utility is 3.07 which reflects an importance of 33%. The largest shift in utility comes from price rank €€€ to price rank €€€€, with a decrease in utility of 1.4. Based on these utilities hypothesis 1 can be confirmed.
Analyzing the impact of insurance company, no interesting results are found. In the first regular PCS situation (choice set 1 to 5) no additional information about the companies is given, the non-significant and very limited differences in preferences are therefore not surprising.
The two tested review characteristics both show an significant effect in the hypothesized way. Valence is with a range of 4.13 and a relative importance of 44% the most important attribute in this model. With a negative utility of -2.63 a two star rated health insurance has the strongest negative effect of all levels in this study. Interestingly, the utility gain from a 4 to 5 star rated insurance is only limited (from 1.27 to 1.50). Volume of reviews show a similar pattern of utility development, but due to the lower importance (21%) and therefore range (1,95) the impact on the utility is smaller. The largest gain in utility is from 5 consumer reviews (-1.20) to 25 (-0.10) consumer reviews, while moving from 100 to 500 consumer reviews only has a limited effect on utility (0.2 gain).
31
Attribute Parameters P-Value Range Relative
importance Price ranking € €€ €€€ €€€€ 1.30 0.77 -0.31 -1.77 <0.001** 3.07 0.33 Average rating 2 stars 3 stars 4 stars 5 stars -2.63 -0.13 1.27 1.50 <0.001** 4.13 0.44 Number of consumer reviews 5 25 100 500 -1.20 -0.10 0.55 0.75 <0.001** 1.95 0.21 Health insurance company R S W X 0.05 0.10 0.00 -0.15 0.25 0.20 0.03
TABLE 3: Attribute and levels parameters for the first five choice set **: Significant on the 0.05 level.
When looking at the results of the choice sets 6 to 10 (the new situation with one of the companies being the owner of the used PCS), some transition in utility arise. The most important change comes from the fact that insurance company now becomes a significant predictor of choice, with insurer R (the company owning the PCS) showing a negative utility. This effect is further analyzed in paragraph 4.4.3.
4.3.2 Interaction effects of attributes
This study uses the minimal overlap approach to ensure a design that can precisely measure the main effects of the different attributes. The drawback here is, however, the fact that interactions are not well visible as could have been the case when balanced level overlap was created in the choice design (Rao, 2014; Chrzan and Orme, 2000). The main goal of this research was however to provide consumers with an as realistic as possible choice set. The interaction effects as described below are therefore more explorative, and further research should be conducted to further validate these findings.
32
VolumeHigh*Price) are created that capture the interaction effect: Volume and valence are split into a low and high variant and then taken together with the price effect as attribute to investigate the interactions. This method was shown to be an effective one to discover interactions by e.g. Chrzan (1994).
The analysis conducted for the five choice sets with the regular PCS situation. When the two interaction effects are included one by one in the choice prediction model the interaction between the variables can be interpreted. Table 4 shows statistics of the tested interaction effects. Two models were estimated, both with two additional interaction variables next to the four main attributed. One with the two variables of valence*price included, the other with the two variables of volume*price included. With these included variables, a new choice based conjoint model is estimated. There is a significant interaction effect between price and the dummies for low and high valence (p-values of both added variables <0.001, relative importance is 23%). Comparing the ranges of the low and high valence interactions shows that high valence has a 0,40 bigger range, thereby contradicting the hypothesized effect that when valence is higher, price would be of less importance. Looking at the specific level parameters, the difference in range is mainly caused by the bigger negative utility from the highest price rank (-1.84 versus -1.28). Hypothesis 3b can consequently not be supported; review valence does not diminish the effect of price on choice decision. When analyzing the interaction between volume intensity and price there seems to be no significant interaction effect; review volume and price are not interacted with each other. Respondents do not generate less (or more) utility from price over different values of review valence and H3a is consequently not supported.
Attribute Utility low valence Utility high valence Utility difference (low – high valence) P-Values Relative importance Valence x Price € €€ €€€ €€€€ Range 1.37 0.27 -0.20 -1.28 2.65 1.21 0.94 -0.34 -1.84 3.05 0.16 -0.74 0.14 0.56 <0.001 0.23 Volume x Price NS 0.99 0.07
33 4.4 Trust analyses
To investigate the influence trust has on choice and purchase intention, three separate analyses are conducted. Firstly, the construct of trust is analyzed as a predictor for purchase intention (the main effect of trust), the construct of trust is taken into account as a moderator in the choice based conjoint design, where a low and high level of trust are taken into account and interacted with the different attributes. At last the effect of independence is taken into account by analyzing the choice sets involving the regular choice sets and the choice sets where one of the health insurances is the owner of the PCS.
4.4.1 Main effects of trust
As trust is not incorporated as an attribute in the choice design, it is at first treated as a separate construct in the study design. To discover the main effect of trust as described in hypothesis 4 the five survey questions concerning perceived trust towards the PCS, adapted from McKnight, Choudhury and Kacmar (2002) are analyzed. This set of five questions is asked twice in the two settings. The means of these five questions are used as the measures for trust in this research.
To investigate if higher perceived trust leads to a higher purchase intention the answers on the follow-up question of the choice set are analyzed. The mean purchase intentions of these first five and second five choice sets are taken. A first insight is gathered from the correlation between trust and the average purchase intention which shows a significant correlation of 0.328. When conducting a linear regression with trust as predictor and the average purchase intention as outcome variable this significant effect is confirmed (p-value <0.001): a higher level of perceived trust leads to a higher purchase intention, hereby supporting H4.
4.4.2 The moderating effect of trust on attributes
To investigate the effect trust has as a moderator on the different attributes in the choice based conjoint model, the construct of trust is included as an interacting variable in the data, and subsequently a new model is estimated. The perceived trust constructs is split in two variables using a median split (trust value under and above 4.2) and renamed as low perceived trust or high perceived trust. When including the created variables in the same manner as with the previously described interaction variables the importance of e.g. price for the two groups of respondents (high and low level of trust) become visible.
34
When comparing the ranges of the low and high trust variable, the high trust variant has a 0.88 greater importance for review valence, suggesting that respondents generate more value from a highly rated health insurance when they perceive more trust at the PCS. Looking at the specific part-worths it can be seen that this effect is about similar for the 2 star rated health insurances (a 0.46 lower utility when perceived trust is high) and the 5 star rated health insurance (a 0.42 higher utility when perceived trust is high). This finding is line with H5a, which suggested that high perceived trust positively moderates the effect of review characteristics on consumer choice.
The other tested review characteristic, review volume, has no significant interaction effect with trust. The same holds for the effect of trust on price: Respondents who showed a high level of perceived trust towards the PCS do not generate a different utility from a low (or high) price than respondents that showed a low level of perceived trust, hereby rejecting H5a. Attribute Utility when low trust Utility when high trust Utility difference (low – high trust) P-values Relative importance Trust x Valence 2 stars 3 stars 4 stars 5 stars Range -2.45 -0.05 0.99 1.25 3.70 -2.91 -0.08 1.45 1.67 4.58 0.46 0.03 -0.46 -0.42 <0.01** 0.07 Trust x Volume NS 0.99 0.02 Trust x Price NS 0.74 0.03
TABLE 5: Overview of the tested interaction variables **: significant on the 0.05 level.
4.4.3 The effect of independence of the PCS
To see what the effect of information shift involving independence of the PCS is, the attribute of health insurance companies in choice sets 6 to 10 is analyzed. Where the attribute was previously highly insignificant, it now has a significant influence on the choice respondents make (p < 0.001). The utility of company R over these choice sets is -0.64. This indicates that owning a PCS as a health insurance company (and consumers notice it) results in being less likely chosen.
35
point scale. This drop in level of trust is highly significant (p < 0.001). It can therefore be concluded that when people are told that the PCS is not fully independent, their perceived trust drops significantly, confirming H6.
4.5 Latent Class analysis
One of the main advantages of using a choice-based conjoint design is the possibility to account for heterogeneity (Ramaswamy and Cohen, 2000). In this paragraph the described attributes and covariates are used to find out different segments in the respondent base. Because of the use of the within-subject design, the different segments can thereafter be analyzed in their transition of preferences with the introduction of the new situation.
The segmentation of the respondents is based on the four main attributes and the covariates gender, age and trust.
4.5.1 Model selection
With the main attributes and the covariates several models are built to see how many classes create the best segmentation. The AIC3 and BIC values are therefore analyzed, together with the classification error and R² of the model.
Number of classes
BIC AIC3 R² Classification
error 1 12.003.479 11.750.040 0.4192 0.0000 2 11.376.267 10.784.910 0.5878 0.0965 3 11.632.821 10.703.546 0.6350 0.1080 4 11.830.870 10.563.678 0.7044 0.1318 5 12.265.312 10.660.201 0.7506 0.1636 6 12.716.544 10.773.516 0.7729 0.1384
TABLE 6: Overview latent class analysis
36 4.5.2. Two segment solution
Table 7 gives an overview of the differences in utilities between those segments. Appendix C shows the relative importance of the attributes and an example of the classification table. The segments consisted of respectively 89 and 77 respondents. The two segments differ significantly in their utility of price ranking (Wald(=) 87.27, p-value <0.001) and number of reviews (Wald(=) 9.03, p-value 0.029). Segment 1 can be considered as the very price sensitive segment, these respondents generate a lot of utility from a low price ranking: the lowest price ranking can increase utility with 3.68 while the highest price ranking decreases utility with 6.87. This range 10.73) varies enormous from the range of segment 2 (1.60), This segment even values the second (0.50 utility) and third (0.48 utility) price rank more than the first price rank (0.12). The second segment can be seen as the segment that is less price sensitive, but more influenced by reviews (especially review volume). The review valence shows no significant differences in utility, although segment two loses some more utility with a two star rated health insurance (-3.50 versus -2.97) with the lowest number of reviews segment 2 loses more utility: -1.99 versus -1.13, segment 2 also generates clearly more utility from 500 reviews (1.24 versus 0.71). As expected, no differences in insurance company preferences arise.
37
Attribute Class 1 utility Class 2 utility Wald(=) p-value Price ranking € €€ €€€ €€€€ Range 3.86 2.70 0.32 -6.87 10.73 0.12 0.50 0.48 -1.10 1.60 87.27 <0.001** Average rating 2 stars 3 stars 4 stars 5 stars Range -2.97 -0.34 1.44 1.86 4.83 -3.50 -0.03 1.69 1.84 5.34 1.69 0.64 Number of reviews 5 25 100 500 Range -1.13 -0.25 0.68 0.71 1.84 -1.99 0.04 0.71 1.24 3.23 9.03 0.029** Health* insurance company NS 0.39 0.94 Covariates Age -0.01 0.01 1.39 0.24 Gender Male Female -0.07 0.07 0.07 -0.07 0.54 0.46 Trust 0.03 -0.03 0.09 0.76
TABLE 7: Overview of the utilities of the two different classes of the first five choice set. *Company was not a significant predictor of choice in the main model, so the utilities are not shown, average rating was a significant predictor but shows no significant differences per segment. **Significant on the 0.05 level.
4.5.3. Transition in utility
38
ranking and number of reviews showed significant differences in utility levels, these differences diminish in the new situation. Only small differences are now present between the different utility levels, with a huge increase in gained (less negative) utility from a high price rank (from -6,87 to -1,99), where segment two shows a slight decrease (more negative) utility for a high price rank (-1,10 to -1,91). Number of reviews (the other significant different attribute between segments) shows a similar pattern where the two segments show less difference in utility for the second set of choices, resulting in almost similar utilities. The health insurance company utility shows (almost) identical results for the two segments (-0,65 and -0,62). Attribute Segment 1 utility Segment 2 utility Transition (new-old) in utility from first five sets Segment 1
Transition (new-old) in utility from first five sets Segment 2 Price ranking € €€ €€€ €€€€ Range 1.63 0.97 -0.61 -1.99 3.62 1.03 1.10 -0.23 -1.91 3.01 -2.23 -1.73 -0.93 4.88 0.91 0.60 -0.25 -0.81 Average rating 2 stars 3 stars 4 stars 5 stars Range -2.23 -0.04 0.95 1.32 3.55 -2.05 -0.49 1.12 1.42 3.47 0.74 0.30 -0.49 -0.54 1.45 -0.46 -0.57 -0.42 Number of reviews 5 25 100 500 Range -0.95 -0.09 0.53 0.50 1.45 -1.22 0.19 0.55 0.49 1.77 0.18 0.16 -0.15 -0.21 0.77 0.15 -0.16 -0.75 Health insurance company R S W X Range -0.65 0.25 0.04 0.34 0.99 -0.62 0.31 0.24 0.06 0.93
39 4.6 Overview of hypotheses
Table 9 shows all tested hypothesis of this research. The attributes price ranking, review volume, review valence all have a significant positive effect on the choice decision of respondents. The hypothesized effect of trust on purchase intention is found as well. The hypothesized interaction effects were not supported. Of the suggested moderating effects of trust, one effect is supported: review valence is strengthened by a higher level of perceived trust
Hypothesis number
Hypothesis Supported?
H1 A relative low price (high ranked price) leads to a higher purchase intention and choice probability of that product.
Yes H2a The volume intensity of online consumer reviews is positively
associated with the purchase intention and choice probability of the displayed product.
Yes
H2b The valence intensity of online consumer reviews is positively associated with the purchase intention and choice probability of the displayed product.
Yes
H3a A higher volume intensity of online consumer reviews diminishes the effect of price rating on choice and purchase intention
No H3b A higher valence intensity of online reviews diminishes the effect of
price rating on choice and purchase intention
No H4 Price ranking has a significant bigger influence for customers in
choosing a health insurance than review characteristics.
No H5 High perceived trust when visiting a PCS increases the purchase
intention of the displayed product.
Yes H6a High perceived trust positively moderates the effect of review
characteristics.
Partly H6b High perceived trust negatively moderates the effect of price ranking. No H7 Consumers perceive higher trust at a fully independent PCS than an
insurance-owned PCS.
40
5. DISCUSSION
The primary goal of this research was to add to the current literature on customer decision making on the internet, with the main focus on electronic word of mouth and perceived trust on a price comparison site for health insurances. In this chapter the findings of the conducted choice based conjoint analysis are discussed, compared to existing research and new findings are addressed. The implications for practitioners in the field of PCSs, health insurances and e-commerce practitioners in general are discussed and at last study limitations and further research possibilities are discussed.
5.1 Theoretical implications
5.1.1 Review characteristics and price ranking
An important finding of this study is the utility consumers generate from review characteristics on a PCS. Although people are expected to focus mainly on price when searching on a PCS, the findings show that people perceive the review valence of the
41
For both volume and valence of reviews the results suggest that consumers want some kind of minimum value before considering the option. This finding is derived from the relatively little utility that is gained when moving from 4 to 5 stars (valence) and from 100 to 500 stars. This pattern is less visible for price ranking where the gained utility is more linear when moving from relatively expensive to relatively cheap. For the review characteristics, consumers seem to reach a satisfactory level, where with price ranking a cheaper health insurance always leads to more utility (maximizing behavior). Further research should however show if there indeed is a difference in choosing a preferred health insurance between maximizers and satisficers (Schwartz, 2004).
One interesting interaction effect is found, although the hypothesized interaction effects are not found. The significant interaction effect is the valence – price interaction: a higher valence shows a greater range in utility, mainly caused by the greater negative utility that is present when the price rank is highest. This is the opposite effect as expected from literature. A possible explanation is the disappointment that arises when consumers see a very well rated health insurance but they are not willing to pay the premium over comparable products or they perceive they are not able to afford it (Chi Lin, 2003). The suggested effect that consumers are willing to pay a premium for a high rated product is consequently not found with this research design. Another research method could however be better able to conclude about this possible interaction, as the interactions in this research are more exploratory of nature because of the limited number of attributes and minimal overlap in this study. The interaction effect between review volume and price was not significant, which could again be contributed to the minimum number of reviews that consumers desire before
considering an option; with in an important product as a health insurance, consumers want to make sure their decision is right and therefore choose a product that meets their requirement on the number of reviews, independent of what the price rankings of the products are. 5.1.2 The influence of trust
