SEARCH BEHAVIOUR AT AUCTION WEBSITES

SEARCH BEHAVIOUR AT

AUCTION WEBSITES

by

Anton Buning

University of Groningen

SEARCH BEHAVIOUR AT AUCTION WEBSITES

Abstract

The purpose of this study is to provide insights in search behaviour of consumers at auction websites and distinguish segments based on this search behaviour. Four product markets at auction websites are investigated with the help of an experiment and several questionnaires. Generalized Linear Models are estimated to investigate the relation between search behaviour at the dependent side and risk, trust, experience and gender at the independent side. Furthermore, Latent Class analyses are performed to distinguish segments in the market. The main finding is that there are five different segments based on search behaviour in the auction market and that there are main difference between males and females.

Keywords: Online auctions; search behaviour; market segmentation

Master Thesis

Msc Business Administration: Marketing Management & Marketing Research

Author: Anton Buning Adress: Anemoonstraat 1

8922 GR Leeuwarden Phone number: (06) 30201069

E-mail: a.buning.1@student.rug.nl Student number: 1384074

PREFACE

This master thesis is part of the E-rep project at the University of Groningen. The objective for this project is to investigate reputation systems for e-communities. In line of this project, this thesis is about search behaviour of consumers at auction websites. Many auction websites make use of a reputation system.

In this preface, I would like to thank a couple of persons. I want to thank Wander Jager, my supervisor, for his support and useful feedback during the research and writing process. Furthermore, I want to thank Debra Trampe and Thijs Broekhuizen, who also participate in the E-rep project and were with Wander Jager responsible for the data collection.

Anton Buning

TABLE OF CONTENT 1 INTRODUCTION... 5 1.1 Problem analysis ... 5 1.2 Problem statement ... 6 1.3 Conceptual model... 6 1.4 Sub-research questions ... 9

1.5 Theoretical and social relevance ... 9

1.6 Structure of the thesis ... 9

2 THEORETICAL FRAMEWORK ... 10 2.1 Risk... 10 2.2 Trust ... 11 2.3 Experience ... 12 2.4 Gender ... 13 2.5 Conceptual model... 14 3 RESEARCH DESIGN ... 16 3.1 Research method ... 16 3.2 Data collection... 17 3.2.1 Experiment... 17 3.2.2 Questionnaire... 19 3.3 Sample ... 19 3.4 Scale measurement ... 20 3.5 Plan of analysis... 21 4 RESULTS... 25

4.1 Stage 1: Generalized Linear Models ... 25

4.1.1 Auction 1: New television... 25

4.1.2 Auction 2: Second hand television... 27

4.1.3 Auction 3: Travel guide... 28

4.1.4 Auction 4: Second hand chair... 30

4.1.5 Conclusion... 31

4.2 Stage 2: Exploring differences ... 33

4.3 Stage 3: Latent Class analysis ... 36

4.3.1 Indicators and covariates... 37

4.3.2 Models selection... 37

4.3.3 Auction 1: New television... 39

4.3.4 Auction 2: Second hand television... 41

4.3.5 Auction 3: Travel Guide... 43

4.3.6 Auction 4: Second hand chair... 45

4.3.7 Conclusion... 47

5 CONCLUSIONS ... 50

5.1 Main findings ... 50

5.2 Recommendations ... 52

5.3 Limitations and future research... 53

REFERENCES... 54

APPENDIX I: QUESTIONNAIRE ... 57

1 INTRODUCTION

In this first chapter, the topic of the thesis will be introduced with a problem analysis. Subsequently, the problem statement is presented with a conceptual model to support the statement. The conceptual model is translated in four sub-research questions. Finally, the relevance and the structure of the thesis will be made clear.

1.1 Problem analysis

Online marketplaces are communities of buyers and sellers who exchange product information, coordinate, and transact using Internet technologies. Online marketplaces can be separated into two categories: business-to-business marketplaces that facilitate exchange relationships among organizations and consumer-to-business or consumer-to-consumer marketplaces that facilitate transactions involving consumers (Pavlou & Gefen, 2004). A certain form of consumer-to-consumer marketplaces are online auctions. Online auctions have become one of the greatest successes of the Internet. Online auctions have several distinguishing characteristics compared with traditional auction, which explain their growing popularity. First, online auctions eliminate geographical limitations of many traditional auctions. Second, Internet auctions can last for several days (usually a week) and allow asynchronous bidding, which gives both sellers and bidders more flexibility. Third, these websites can run auctions at substantially lower operational costs than traditional auction houses and can thus charge lower commission fees and attract more sellers and buyers (Ariely & Simonson, 2003). Good examples of websites with these characteristics and which are very popular is Ebay.com on a global scale and Marktplaats.nl on a national scale. On these websites is an enormous supply of consumer products, which are being auctioned.

knowledge from data and, for instance, can be performed to optimize the performance of a website, to discover which products are being purchased together, or to identify whether the site is being used as expected (Spiliopoulou, 2000). A web usage mining technique can capture the consumer’s navigational behaviour from web log data (Huang et al., 2006). Web log data can also be referred as click stream data.

The behaviour of consumers on auction websites will be very different. Traditionally, marketers deal with heterogeneity by segmenting the market (Bhatnagar & Ghose, 2004). Market segmentation has become a dominant concept in marketing literature and practice (Wind, 1978). Market segmentation consist of viewing a heterogeneous market as a number of smaller homogeneous markets in response to differing preferences, attributable to the desires of consumers for more precise satisfaction of their varying wants (Smith, 1956). According to Dias and Vermunt (2007), the most effective segmentation strategy is that which best captures differences in the behaviour of target subpopulations. For online auction websites the search behaviour towards a product can be relevant. With the help of segmentation it is possible to organize information from the consumers search behaviour into groups/ collections (of similar objects), in order to facilitate data availability and accessing, and at the same time to meet consumer preferences (Pallis, Angelis & Vakali, 2007). Therefore, this thesis is focused on search behaviour of consumers on auction websites.

1.2 Problem statement

The problem statement is a result of the problem analysis. The problem statement consists of a research objective and a research question.

Research objective: To provide insight in search patterns/ behaviour of consumers on auction websites and distinguish groups/ segments based on this search behaviour.

Research question: Provide search patterns/ behaviour of consumers on auction websites a useful basis for a segmentation of the market?

1.3 Conceptual model

In this model is on the right hand side the main focus of the thesis, namely the search behaviour of consumers. The sequences of visited pages on a website can be measured with the use of click stream data. Subsequently, it is possible to make up the number of pages that are visited by a consumer. Besides, the time spent on each page can be measured with these kinds of data. In this part of the study the model concentrates first at the length of the search for a product during an auction website visit. For this purpose two features will be measured, namely the number of visited pages and the average time on a page during an auction. It is also possible to include total time spend during an auction. However, the number of visited pages and the total time spend have probably a high correlation, because how more pages a consumer visit, how more time is spend. The variable number of visited pages is selected, because this variable gives a better indication of the amount of information a consumer utilize. Two different characteristics about search length will be measured with this variable and average time.

FIGURE 1: Conceptual model

On the left hand side are four additional variables in random sequence. These are the variables risk, trust, experience and gender. These variables are admitted in the conceptual model, because in the literature are indications that there is a relation between the four variables and search behaviour of consumers. Below is discussed for every variable the motive to include the variable into the model and the indication, which is found in the literature. The theoretical framework in chapter two provide more detailed information about the relations. If there are indeed relations between the variables and the length of the search pattern, then there are explanatory differences between consumers. Subsequently, these differences can provide the basis for market segmentation.

Search pattern:

Number of visited pages Average time on a visited page

Risk

Trust

Experience

The first additional variable is the perceived risk of a purchase. The risk of a purchase is an important factor with Internet transactions. This is true for Internet retailers, but as one moves from retailers to unfamiliar sellers in online auctions, the perceived risks for buyers increase even more (Finch, 2007). Besides, buyers are most of the times required to pay in advance for a product to be delivered sometime into the future, often with little or no ability to examine the product in advance (Standifird, 2001). Consequently, it is possible that a product will be delivered very late or with defects. In this line of reasoning, a purchase on an auction website involves a certain kind of risk. It is interesting to investigate if perceived risk has influence on the search pattern of a consumer.

The second additional variable is trust. Online auctions involve millions of buyer and seller pairs who are unfamiliar with each other (Finch, 2007). Despite of this unfamiliarity, buyers and sellers deal with, and rely on each other. Therefore, trust between parties is a key issue in online auction transactions. Thus, the variable trust has probably a relation with the search pattern of consumers.

The third additional variable in the conceptual model is the experience of consumers. Wilcox (2000) states that as consumers gain experience, they learn bidding strategies which are more likely to be successful. Hence, it is reasonable that consumers with more experience have different online behaviour in comparison with consumers with less experience. Therefore, the relation between experience and search patterns will be studied in this thesis.

The fourth and last additional variable in the model is gender. Black (2005 & 2007) found that females are more likely to purchase at an auction website. Moreover, males are more willing to pay higher prices. This finding indicates that males and females have different online behaviour at auction websites. It is interesting to investigate if this is also true for the search behaviour of consumers. Therefore, gender is the fourth variable which is taken in consideration.

1.4 Sub-research questions

The relations in the conceptual model can be translated in the following sub-research questions:

To what extent is perceived risk of a purchase of influence on the search behaviour of consumers at auction websites?

To what extent is trust in sellers of influence on the search behaviour of consumers at auction websites?

To what extent is experience of consumers of influence on the search behaviour of consumers at auction websites?

To what extent is the gender of consumers of influence on the search behaviour of consumers at auction websites?

1.5 Theoretical and social relevance

Most of the marketing literature about search behaviour of consumers is focused on traditional offline markets. The last decade studies more and more focused on search behaviour in online environments. This thesis is also about search behaviour in an online setting. The relevance of this thesis lies in the fact that it concerns online auctions. Other studies with this topic concentrated on retail websites and news websites (e.g. Zhang et al., 2006; Dias & Vermunt, 2007; Pallis et al., 2007). Besides, this study explores a relation between the behaviour of consumers and four specific variables, which are relevant for online auctions.

Moreover, this study can make general recommendations towards most of the auction websites, because the used data is from an experiment and not from a specific auction company.

1.6 Structure of the thesis

2 THEORETICAL FRAMEWORK

In this chapter, the conceptual model (figure 1) is examined from a theoretical point of view. The aim is to create hypothesises regarding the sub-research questions in subsection 1.4. Subsequently, the hypothesises will be used for the analyses in chapter 4. This chapter is divided in paragraphs about the four additional variables and ends with an overview of the conceptual model with the assumed relations.

2.1 Risk

The study of Mitra, Reiss & Capella (1999) is about perceived risk and information search for services. Although, they did not investigate search behaviour on the internet, their findings are corresponding with the above line of reasoning. Namely, the authors conclude in their research, that the search time is greater for the service with the highest risk. This relation applies possibly also for search behaviour at online auction. Based on above theories and way of thinking are two hypothesises created for the relation between risk and search length:

H1a: Perceived risk of a purchase has a positive influence on the number of visited pages during an auction.

H1b: Perceived risk of a purchase has a positive influence on the average time spend on a page during an auction.

2.2 Trust

Trust in a seller is the second additional variable to investigate in relation with the search length. This variable has a link with the previous additional variable. Namely, the trust of consumers in sellers facilitates online transactions by reducing perceived risk (Pavlou & Gefen, 2004). Trust, in a broad sense, is the confidence a person has in his or her favourable expectations of what other people will do, based, in many cases, on previous interactions (Gefen, 2000). Besides, Gefen (2000) states that trust is an important condition for e-commerce. Consumers with a low level of trust are less likely to shop online due to their heightened concerns with the security (Das et al., 2003).

about this kind of mechanisms and found that these systems engender trust and contributes to effective online marketplaces. In addition, Ba & Pavlou (2002) found that trust is important at online transactions and that a seller’s reputation plays a very important role in buyers’ willingness to pay premium prices. In the above line of reasoning it is assumable that there is a positive relation between the search length and trust in sellers. Therefore, the following two hypothesises are created for the relation between search length and trust:

H2a: The number of visited pages during an auction has a positive influence on the trust in a seller.

H2b: The average time spend on a page during an auction has a positive influence on the trust in a seller.

2.3 Experience

Experience of consumers with auction websites is the third additional variable. With the help of literature two hypothesises will be created to investigate the relation with the length of a search pattern during an auction. In chapter 1 is previously mentioned that as consumers gain experience, they learn bidding strategies which are more likely to be successful (Wilcox, 2000). Thus, these learned bidding strategies are more effective, then the bidding strategies (or a lack of strategies) of inexperienced consumers. Besides, it is reasonable that the strategy of experienced buyers is more efficient qua the number of biddings. In this line of reasoning it assumable that experienced consumers also have a more effective and efficient search strategy to choose a product at an online auction.

There is mentioned above that at some auction websites it is possible to visit a reputation system. For example, these systems engender trust. If consumers visit by every seller the page with the reputation system, then is the length of search probably longer. Einwiller (2003) found in her study that reputation for inexperienced buyers in a business-to-consumer setting is a more important determinant for trust. Therefore, there is a motive to assume that inexperienced consumers make more use of these reputation pages and as a result of this fact they visit more pages. Besides, Einwiller (2003) state that if experience of consumers provides not enough information to overcome uncertainty, the consumer needs to search for additional information.

university website. Some important findings were that users with less experience tended to visit more pages and used the “home” button more frequently. Besides, inexperienced users navigated in a more linear mode in comparison with more experienced users.

The above mentioned theories have almost the same line of reasoning. It seems that experienced internet users search more effective and more efficient. Therefore, this thought can be translated to hypothesis with regard to online auctions:

H3a: Experience of consumers has a negative influence on the number of visited pages during an auction.

H3b: Experience of consumers has a negative influence on the average time spend on a page during an auction.

2.4 Gender

There are not many studies about gender in combination with auction websites. One of the few researchers is Black (2005 & 2007) with his studies about consumer demographics at auction website eBay. His main finding about gender is that females are more likely to purchase at eBay, but males are more willing to pay higher prices. That females are more likely to purchase at eBay is contradictory with researches concerning common online purchases. These researches suggest that males are more likely to purchase online (e.g. Van Slyke, Comunale & Belanger, 2002; Kwak, Fox & Zinkhan, 2002). A possible explanation is that bidding on products online may provide a different experience than merely purchasing products on the Internet. For instance, females, like to shop, and the excitement of the auction may replace the satisfaction of a normal shopping experience (Black, 2005). This difference between males and females is probably an indication that there is also a difference in search behaviour.

that males visit more pages during an auction and that they spend on average less time on a page. However, children are most of the times not the consumers of online auction and a study of Lorigo et al. (2006) with adults is taken in consideration before the hypothesises can be formed.

Lorigo et al. (2006) had an eye-tracking experiment with a search engine task, and their participants were college students. Their result about gender is that males look at a greater number of abstracts and spend more time examining the result page. Besides, males follow a linear path when they observing to result, while females make more regressions (jump back). Although, this study is about eye contacts and not about visited pages, the results suggest that the length of searching by males is longer in comparison with females. With the studies of Large et al. (2002) and Lorigo et al. (2006) in mind the following hypothesises can be formulated:

H4a: Males visit more pages during an auction in comparison with females.

H4b: Males spend on average more time on one page during an auction in comparison with females.

2.5 Conceptual model

The hypothesises are schematic described in the conceptual model (figure 2). This model will be explored and is the starting point for further investigation.

FIGURE 2: Conceptual model with hypothesises

The objective is to get insights in the four relations separately and not to investigate complete causality for the whole model. In other words, if all relations are significant, then this would

Search pattern:

a) Number of visited pages b) Average time on a visited page

3 RESEARCH DESIGN

In this chapter, the research design, to investigate the conceptual model and to distinguish the market in groups, is explained. First, the research design is explained by means of a classification. Next, the manner of data collection is presented and is showed how the experiment and the questionnaire are conducted. Subsequently, the used sample and the scale measurement for the analysis are introduced. This chapter ends with a plan of analysis and this plan will be used in chapter 4.

3.1 Research method

Roughly, this study can be divided in two parts. The first part is to investigate the conceptual model, which is formulated in chapter two. This part serves as input for the second part. Namely, if there are indeed significant relations in the model, then there is probably a basis to distinguish groups of consumers in the online auction market. To distinguish groups based on search patterns is the second part of this research and the main objective of this study. According to Malhorta (2004) above parts in this research can be classified in conclusive research design. The author states that conclusive research is more formal and structured, than exploratory research. Besides, findings from this research are considered to be conclusive in nature and that they are used as input into managerial decision making. Furthermore, conclusive research design can be descriptive or causal. The first part in this research is more causal research, because the objective is to investigate relationships. On the other hand the second part is more descriptive research, because the main objective of this study is to describe segments in the market of online auctions.

drawn from the target population, and the information is obtained from the sample only once (Malhorta, 2004).

3.2 Data collection

There is mentioned above that the data collection consist of an experiment and a questionnaire. In this paragraph both instruments are explained.

3.2.1 Experiment

An experiment is used to obtain data about the search behaviour of consumers at a controlled consumer-to-consumer auction environment. Respondents are asked to fulfil four shopping tasks for different markets at an experimentally designed auction site. Respondent are able to choose between six sellers and are asked to make a bid at one seller. In figure 3 is presented which pages are available for respondents to obtain information about the product and the different sellers. Basically, there are three levels of pages to visit. At each level there are differences between sellers. The experiment will produce click stream data. This data will reveal the sequence of visited pages for every respondent. Besides, the time on each page is also available. Below will be shown which information is available for respondents at each level and what possible differences between sellers at each level are.

FIGURE 3: Lay-out experimentally designed auction website

1

2

3

Product descriptions page (Level 1): This is the main page of the experimentally designed auction website. At this page respondents can read short product descriptions of the sellers. Furthermore, there is presented a picture of the product and the possible starting price of the

sellers. These three characteristics vary between sellers. Respondents can visit the seller description pages from this page. Besides, respondent can make a bid on this page.

Seller description pages (Level 2): On these pages is information available with more detail about the product in comparison with the short description at the main page. However, the six seller pages are different in the amount of detail. Another difference between sellers is the scores at the reputation system (Resnick et al., 2000). A score is calculated as follows: # positive reactions - # negative reactions. At this page respondents can see the reputation score of a seller and the percentage of positive reaction, but not how this score is build up. Therefore, respondents have to visit level 3.

Feedback pages (Level 3): At this level respondents can check in detail the reputation scores. Thus, here is also presented at how many reaction the score is based and how many months ago the reactions were placed. In the experiment is made a distinction between three kinds of reputation scores. The first feedback score is a general score based on reaction of other online auction consumers at the same website. The second feedback score is the friends score page. This feedback score is based on reaction of fictive familiar consumers of the respondent. The third feedback score is the rumour score page. This feedback score is based on reactions about the seller on other websites.

Now, there has been explained how the experiment is conducted. Finally, the four different markets will be introduced. It is interesting to investigate if there are differences between markets. Besides, these markets are chosen based to certain extent on price, because Finch (2007) state that risk is for a certain part a function of the amount of stake. This are the four product categories: a new television, a second hand television, a travel guide and a second hand chair. The respondents fulfil the tasks in different order. Beneath, the four product categories and the corresponding shopping task are explained in more detail.

New television: The respondents are asked to choose a seller and make a bid on a new Philips Ambilight television. The retail price of this product is € 2.599,-. The starting price of the six sellers in the auction is around € 1.000,-.

Second hand television: The shopping task is to buy a two year old Philips television. A year ago was the retail price of the television € 799,-. The starting price of the six sellers at the website is around € 250,-.

Second hand chair: The shopping task is to make a bid on a second hand chair of the brand Gispen from 1931. Most of the times these chairs are sold on auction websites between € 25,- and € 75,-. The six sellers do not give a starting price.

3.2.2 Questionnaire

The second data collection instrument is a questionnaire. The questionnaire(s) can be found in Appendix I and is used to obtain data about the four additional variables. After every shopping task, the respondent have to complete a questionnaire about the specific auction and the product. These questions are about the risk of the purchase, trust in the seller en information about the search behaviour during the auction. This study make use of the questions about risk (part A) and trust (part B). These questions are asked on a seven-point Likert scale. The advantage of this ordinal scale is that the data can be treated continuous, because the scales have more than five categories (Torra et al., 2006).

When the respondents had completed all four shopping tasks, they had to fill in a fifth and final questionnaire. In this final survey are asked more general questions about trust and buying behaviour. Besides, there are background questions about gender and experience (part 2). These last questions will be used in this research. In paragraph 3.4 will be explained how the used questions are scaled to measure the variables from the conceptual model.

3.3 Sample

The sample consist of 173 students of the University of Groningen. 127 males and 46 females joined the experiment and they answered the corresponding questionnaires. The average age of the respondents is 20,96 year old and the range is between 17 and 29 year old. The experiment is held on 7 and 10 January 2008 in a computer room at the University of Groningen.

An advantage is the size of the sample, because the sample is large enough to execute statistical analysis. For example, a preferred ratio of observations to variables is 15:1 for a multiple regression (Hair et al., 2006). This means here, that it is possible to make a multiple regression with 11 variables.

3.4 Scale measurement

Scales must be created for the variables to test the hypothesises. Risk and trust are the first two variables to create a scale. These two variables are measured with multiple questions in the four questionnaires about the specific auctions (part A and part B in Appendix I). The coefficient Cronbach’s alpha will be used to measure the internal consistency reliability. A value of 0.6 or less generally indicates unsatisfactory internal consistency reliability (Malhorta, 2004; Hair, 2006). For the first two questions about perceived risk of the purchase is the scale reversed. In table 1 are displayed the Cronbach’s alphas for the variables at the four auctions. Only the Cronbach’s alpha for risk at the fourth auction is below 0,6. This coefficient is not higher when one of the five questions will be deleted from the scale. Besides, the coefficient is almost 0,6. Therefore, this low coefficient is ignored and the variable risk for the four auctions will be calculated by summing up the five questions and divide through five. All Cronbach’s alphas for the variable trust are widely above 0,6. Consequentially, the scale for trust at the four auctions will be calculated by summing up the three question and divide through three.

TABLE 1: Cronbach’s Alphas

Risk Trust Auction 1: New television .686 .838 Auction 2: Second hand television .624 .801 Auction 3: Travel guide .643 .721 Auction 4: Second hand chair .570 .764

Finally, the variables regarding the search length can be calculated easily from the click stream data. The average time on a page will be measured in seconds. It is indeed true that there is a high correlation between number of pages visited during an auction and total time spend during an auction. The correlations at the four auctions are all higher than 0,7. This confirms the decision to include the proposed two variables into the analysis.

3.5 Plan of analysis

The plan of analysis is explained in this paragraph. The analysis will be divided in three stages. In the first stage, analysis will be executed regarding the conceptual model. The second stage consist of analysis, which will serve as a link between the conceptual model and the main objective. In the third stage, analysis will be performed with the main objective to divide the population in groups.

between the variate and the probability distribution. Concluding, the below two equations will be estimated in stage one of the analysis. Possibly non-linear effects will be added to optimize these models in terms of their relevance, significance and a lower deviance for the total model.

[ ]

e

y

E

=

y = Number of visited pages T = Trust i = Respondent: 1, …, 173 E = Experience a = Auction: 1,…, 4 G = Gender

R = Risk u = Disturbance term

[ ]

y

R

T

E

G

u

E

=

α

+

β

+

β

+

β

+

β

+

y = Average time on a visited page T = Trust i = Respondent: 1, …, 173 E = Experience a = Auction: 1,…, 4 G = Gender

R = Risk u = Disturbance term

The second stage consist of analysis as t-tests and ANOVA to explore differences between the four auctions and search behaviour of consumers. These analyses can give some more information about differences or similarities between the four markets. Furthermore, detailed information about the founded relations during stage one can be provided with this analysis. The execution of this stage is thus partially dependent of the results from the first stage. Besides, the results of this stage will be used as input for the third stage, because the detail information can suggest differences in search behaviour between consumers.

reason, why there is chosen for a LC analysis instead of a more standard cluster analysis (e.g. hierarchical clustering). Besides, it is possible to include variables of mixed scale types (e.g. nominal, ordinal) in the same analysis (Magidson & Vermunt 2002).

Several models with cluster solutions will be estimated in the analysis. Next, the best solution will be chosen with the help of information criteria, in contrast with GLM, where the deviance will be used. In the best case, the solutions for all four auctions are the same in terms of the number of cluster. Popular information criteria for LC analyses are AIC, AIC3, BIC and CAIC (Vermunt & Magidson, 2002). These criteria are a function of the log likelihood plus a penalty for the number of parameters. The penalty is the lowest at the AIC criterion, while the penalty for adding a parameter is the highest at the CAIC criterion. The penalty at the AIC3 criterion is slightly larger, then the penalty at the AIC criterion. The penalty of BIC criterion is slightly smaller, then the penalty of the CAIC criterion. Besides, the penalties of the BIC and CAIC criteria are dependent of the sample size. The model with the lowest value at a specific information criterion will be chosen. Thus, the information criteria can help to decide which model has the optimal solution with an appropriate number of parameters. Researchers have different opinions about the effectiveness of these criteria. According to Andrews and Currim (2003), AIC3 is the best criterion to use for a large variety of data configurations. Furthermore, the authors state that BIC and CAIC are appropriate for larger samples. Other authors found that CAIC is superior to AIC for relatively simple models (Lin & Dayton, 1997) and that AIC and BIC have the tendency to select a model with to many clusters (Naik, Shi & Tsai, 2007). This is a result of the small penalty of these criteria in comparison with the CAIC criterion. Too many clusters are hard for interpretation of the models. An other criterion is the approximate weight of evidence (AWE). This criterion combines information about the model fit with information about classification error (Vermunt & Magidson, 2002). Hence, the AWE criterion has the tendency to select a model with less clusters in comparison with the other criteria. In sum can be stated, that there is not one superior selection criteria. Therefore, for the LC analyses of the four auctions will be made use of the criterion which is most appropriate for the specific situation. For this purpose is interpretation of the different models important.

4 RESULTS

In this fourth chapter the results of the analysis are presented. The analysis are performed with the help of the analysis plan in paragraph 3.5. The analysis are divided in three stages and in this chapter are discussed the relevant results for each stage.

4.1 Stage 1: Generalized Linear Models

In this paragraph the results of the Generalized Linear Models are discussed. For every auction are estimated equations 1 and 2. The results of these equations give insights in the assumed relations of the conceptual model. The models for auction four are based on 173 respondents, while the models for other auctions are based on 172 respondents as a result of missing values.

4.1.1 Auction 1: New television

highest perceived risk level visit more pages then the consumers with the lowest perceived risk level. The relations of trust and experience(²) are positive and relative small with the number of visited pages based on the exponent of the value (respectively, 1,05, 1,02 and 0,99). Consequently, the consumers who visit more pages during the search phase have a higher trust level in the seller. The fourth relation between gender and visited pages is larger and positive (exp(Value) is 1,42). This contains that females visited significant more pages in comparison with males during auction 1.

TABLE 2: Number of visited pages during auction 1

Variable Value exp(Value) Std. Error t-value p-value Intercept 2,1080 8,2319 0,2380 8,8579 0,0000* Risk 0,5018 1,6516 0,1307 3,8385 0,0000* Risk² -0,0769 0,9259 0,0209 -3,6889 0,0003* Trust 0,0576 1,0593 0,0210 2,7492 0,0066* Experience 0,0203 1,0205 0,0038 5,2870 0,0000* Experience² -0,0004 0,9996 0,0001 -4,9239 0,0000* Gender 0,3522 1,4221 0,0321 10,9786 0,0000* Total Model Null Deviance 2122,788 Df 171 Residual Dev. 1968,565 Df 165 Difference 154,223 6 0,0000* * Significant at p < 0,05

TABLE 3: Average time on a visited page during auction 1 Variable Value Std. Error t-value p-value Intercept 13,0445 5,7662 2,2622 0,0250* Risk -0,2050 0,8618 -0,2378 0,8123 Trust -0,2557 0,7609 -0,3360 0,7373 Experience -0,0120 0,0655 -0,1827 0,8553 Gender -2,5649 1,2325 -2,0811 0,0390* Total Model Null Deviance 8283,479 Df 171 Residual Dev. 8064,266 Df 167 Difference 219,213 4 0,0000* * Significant at p < 0,05

be taken, because the identity-link is used for this model. Therefore, the value indicates that males visit a page 2,5 seconds longer in comparison with females. The other three variables are negative but not significant. Nevertheless, the variables contribute to the model, because the difference between the null deviance and the residual deviance is significant.

4.1.2 Auction 2: Second hand television

For the following auctions comparable analysis are executed as for the first auction. Therefore, not every result will be explained in detail and the emphasis is on the main results. For the dependent variable number of visited pages during auction 2 the results are displayed in table 4. Only for the variable risk is added a quadratic effect. The residual deviance of a model without this effect is 1736,684. Besides, this effect is significant. Other significant effects are the linear risk effect and gender. These two effects are in line with the previous auction and quite large, namely the exponent of the value is 2,41 and 1,40. These values indicate that when the perceived risk level of a consumer is higher, then consumers visit more pages and that females visit more pages during an auction in comparison with males. Nevertheless, risk has also a quadratic effect, which point out that the consumers who visit the most pages have a high risk perception, but not the highest risk perception. The variables trust and experience have a positive, but not a significant relation with the number of visited pages during auction 2. However, the variable experience is significant at a 10 % significance level. Finally, the variables contribute to the model.

TABLE 4: Number of visited pages during auction 2

Variable Value exp(Value) Std. Error t-value p-value Intercept 2,0333 7,6392 0,3125 6,5059 0,0000* Risk 0,8783 2,4069 0,1698 5,1730 0,0000* Risk² -0,1713 0,8426 0,0270 -6,3379 0,0000* Trust 0,0239 1,0242 0,0252 0,9493 0,3439 Experience 0,0037 1,0037 0,0020 1,8797 0,0619** Gender 0,3377 1,4017 0,0327 10,3271 0,0000* Total Model Null Deviance 1901,062 Df 171 Residual Dev. 1693,838 Df 166 Difference 207,224 5 0,0000* * Significant at p < 0,05 ** Significant at p < 0,1

significant (-4,78). This means that males are on average longer at one page in comparison with females during auction 2. Furthermore, the value of the variable trust (-22,74) and the quadratic effect of trust are significant. This is an enormous effect which is contradictory with the hypothesis en the finding about trust at auction 1. There is no logical explanation for the effect that trust decrease enormous, when consumers are longer on a page during the second hand television auction. Moreover, the quadratic effect contains that trust is decreasing when a consumer is on average longer on a visited page until a certain level of time, after which trust is increasing. The residual deviance of the model without the non-linear effect is 11443,85. Risk is positive and significant at a 10 % level and experience is not significant, nevertheless the four additional variables contributes to the model.

TABLE 5: Average time on a visited page during auction 2 Variable Value Std. Error t-value p-value Intercept 65,4996 20,7801 3,1520 0,0019* Risk 1,8743 1,0358 1,8096 0,0722** Trust -22,7424 8,0621 -2,8209 0,0054* Trust² 2,1710 0,8090 2,6836 0,0080* Experience -0,0602 0,0768 -0,7847 0,4337 Gender -4,7800 1,4430 -3,3125 0,0011* Total Model Null Deviance 12765,890 Df 171 Residual Dev. 10968,020 Df 166 Difference 1797,870 5 0,0000* * Significant at p < 0,05 ** Significant at p <0,1

4.1.3 Auction 3: Travel guide

explore this finding in stage 2 of the analysis. Trust has a small, positive relation with the dependent variable and is significant at a 10% significance level. It is remarkable that the value of risk is negative and not significant, while it had positive and relatively large influence at auction 1 and 2. A possible explanation for this fact could be that the travel guide is a different market, where risk do not play an important role. Finally, there can be stated that the independent variables contribute to the model.

TABLE 6: Number of visited pages during auction 3

Variable Value exp(Value) Std. Error t-value p-value Intercept 2,9542 19,1864 0,1878 15,7311 0,0000* Risk -0,0329 0,9677 0,0281 -1,1679 0,2445 Trust 0,0468 1,0479 0,0250 1,8679 0,0635** Experience -0,0056 0,9944 0,0021 -2,5943 0,0103* Gender 0,3990 1,4903 0,0327 12,2190 0,0000* Total Model Null Deviance 1711,242 Df 171 Residual Dev. 1543,278 Df 167 Difference 167,964 4 0,0000* * Significant at p < 0,05 ** Significant at p < 0,1

The estimation of the model with the dependent variable average time on a visited page during auction three are presented in table 7. Non-linear effects did not give reason to include in the model. In line with the previous two models gender has again a negative influence and is the only significant additional variable. Consequently, the average time on a page is at the travel guide auction longer for males. Finally, the four variables contributes to the model based on the significant difference between the null and the residual deviance.

4.1.4 Auction 4: Second hand chair

The last and fourth model with the dependent variable number of visited pages consists of the four additional variables and two quadratic variables (table 8). The residual deviance of the model without linear effects is 1376,544. All variables except the variable experience are significant. The effect of the variable gender is constant over the four auctions and is again relatively large and positive (exp(Value) is 1,33). The effect of risk is in this model in comparison with the previous auction again positive and relatively large (exp(Value) is1,75). Besides, risk has a relatively small quadratic effect. This contains that consumers with a higher perceived risk perception visit more pages. However, the consumers with the highest risk perception visit not the most pages, because of the small quadratic effect. The exponent of the value of the variable trust is 2,69. This value is very large and not in line with the other three auctions. For trust is also found a relatively small quadratic effect. Experience is the only variable which is not significant. In comparison with the first two auctions experience has a negative effect. This is in line with the previous auction and suggest that females have less experience. The additional variables contribute again to the total model.

TABLE 8: Number of visited pages during auction 4

Variable Value exp(Value) Std. Error t-value p-value Intercept -0,3542 0,7017 0,6002 -0,5901 0,5559 Risk 0,5614 1,7531 0,2151 2,6099 0,0099* Risk² -0,0960 0,9085 0,0366 -2,6193 0,0096* Trust 0,9913 2,6947 0,2423 4,0907 0,0001* Trust² -0,0894 0,9145 0,0234 -3,8208 0,0002* Experience -0,0018 0,9982 0,0020 -0,8848 0,3775 Gender 0,2829 1,3269 0,0339 8,3564 0,0000* Total Model Null Deviance 1460,070 Df 172 Residual Dev. 1338,835 Df 166 Difference 121,235 6 0,0000* * Significant at p < 0,05

dependent variable. For this model there is again a significant difference between the null deviance and the residual deviance.

TABLE 9: Average time on a visited page during auction 4 Variable Value Std. Error t-value p-value Intercept 17,9753 4,8079 3,7387 0,0003* Risk -1,0907 0,7200 -1,5149 0,1317 Trust -1,0348 0,6439 -1,6071 0,1099 Experience 0,0074 0,0476 0,1552 0,8769 Gender -1,9341 0,8817 -2,1937 0,0296* Total Model Null Deviance 4487,684 Df 172 Residual Dev. 4268,981 Df 168 Difference 218,703 4 0,0000* * Significant at p < 0,05 4.1.5 Conclusion

In this sub-paragraph is decided if hypothesises 1 till 4 will be accepted or rejected. For this purpose, the results of the Generalized Linear Models of the four auctions will be taken in consideration. Table 10 and 11 summarize the results of the models and will help to make the decisions concerning the conceptual model. In this summary, only the linear effects are included. The non-linear effects are not included, because these effects are most of the times relatively small and do not display the main direction of the relation. If the exponent of the value is above 1, then there is a positive relation. If the exponent of the value is below 1, then there is a negative relation with the number of visited pages during an auction.

TABLE 11: Summary results of average time on a visited pages Auction 1: Value Auction 2: Value Auction 3: Value Auction 4: Value Risk -0,2050 1,8743** 1,0548 -1,0907 Trust -0,2557 -22,7424* 0,3326 -1,0348 Experience -0,0120 -0,0602 -0,0718 0,0074 Gender -2,5649* -4,7800* -3,2776* -1,9341* * significant at p < 0,05 ** significant at p < 0,1

In the theoretical framework is assumed a positive relation between risk and search length. For the number of visited pages is found that there is in three of the four auctions a positive significant relation (table 10). Only at the travel guide auction there is not a significant relation. This is a relative cheap product and as a result risk is possibly not important for this category. Another finding is that the two second hand product have a larger effect on risk in comparison with the most expensive product. A reason for this could be that second hand products are used and consumers do not know if the products have some defects. For the average time on a page is only traced one significant relation at a 10% significance level (table 11). Furthermore, there are two positive and two negative relations. Hence, it seems that there is no relation between risk and the average time on a visited page. In conclusion, hypothesis 1a can be partially accepted, because at three of the four auctions is found a positive relation. Besides, hypothesis 1b can be rejected, because it seems that there is not a constant significant relation over four auctions.

findings of the other two auctions suggest small, negative relations of which one is insignificant. A possible reason for this occurrence can be that experienced consumer consider relative cheap purchases as a routine task and do not need much information. While an experienced consumer is more cautions by making a relative expensive product. For the dependent variable, average time, are not found significant relation with experience. However, it correspondents with the hypothesis that three of the four relations are negative. In conclusion, hypothesises 3a and 3b could be rejected, because the findings suggest not a stable and/ or significant relation.

The last additional variable to investigate is gender. In the theoretical framework is assumed that males visited more pages and are spending more time on a page. The findings of the number of pages in table 10 suggest a significant, positive relation with stable values over the four auctions. A positive relation suggest that females visited more pages and this is in contrast with the hypothesis. A reason for this fact could be that females like the excitement of an online auction (Black, 2005). For average time on a page is discovered a significant, negative relation at the four auctions. Moreover, the values are in a limited range. Thus, this finding corresponds with the hypothesis, but is in contrast with the relation between gender and the number of visited pages during an auction. Hence, it is interesting that females visited more pages, while males are on average significant longer at one page. In sum, hypothesis 4a is rejected and there will be accepted that female visited more pages during an online auction. Besides, hypothesis 4b is accepted, because there is discovered enough evidence for this relation.

4.2 Stage 2: Exploring differences

have less auction experience in comparison with males. This difference will be explored in this paragraph. This finding can be a possible explanation for the difference in search behaviour between males and females.

In table 12 are the results presented of the comparison of the auctions. An ANOVA test is used to investigate if the search length is different between markets. For the number of visited pages is the p-value > 0,05. This means that the averages of this variable do not differ significantly between markets. For the average time spend on a page is the p-value < 0,05. This indicates that the averages between the four auctions are not equal and that at least two averages are different. This is remarkable, because this range between the averages is quite small. Besides, it is interesting that the two products with the highest price have by both variables a higher average in comparison with the two cheapest products. Nevertheless, the differences in search length are not enormous between the four markets. Hence, in stage 3 it is perhaps possible to distinguish comparable groups of consumers based on search behaviour in the four markets.

TABLE 12: Difference in search length Average number

of visited pages

Average spend time at a page Auction 1: New television 27,58 10,44 Auction 2: Second hand television 26,57 11,30 Auction 3: Travel guide 24,84 9,712 Auction 4: Second hand chair 24,46 8,996 P-value ANOVA 0,281 0,020

comparison with males after receiving a recommendation. In this line of reasoning, females attach more value to the feedback pages and as a result females visit more of these pages during an auction. Apparently, there is variance in the number of visited pages between levels and consumers. This information can be used for stage 3 of the analysis. In the segmentation must be taken in account that there are differences between page levels. The consumers visit on average less feedback pages. Besides, this page levels can give insights in the amount and type of information, which is used, before consumers place a bid at an online auction.

TABLE 13: Gender differences by visited pages at page level

Auction Variable Males Females P-value t-test Number of visited product descriptions pages 10,16 12,83 0,022 Number of visited seller description pages 9,16 11,83 0,022 1

Number of visited feedback pages 5,83 9,63 0,023 Number of visited product descriptions pages 9,72 12,24 0,012 Number of visited seller description pages 8,72 11,24 0,012 2

Number of visited feedback pages 6,06 8,80 0,065 Number of visited product descriptions pages 9,04 12,37 0,000 Number of visited seller description pages 8,04 11,37 0,000 3

Number of visited feedback pages 4,85 9,13 0,013 Number of visited product descriptions pages 9,39 11,04 0,068 Number of visited seller descriptions pages 8,39 10,04 0,068 4

Number of visited feedback pages 4,72 8,76 0,008

TABLE 14: Gender differences by average time at page level

Auction Variable Males Females P-value t-test Average time at a product descriptions page 11,7562 7,9235 0,009 Average time at a seller description page 10,2784 10,5831 0,745 1

Average time at a feedback page 10,3142 7,6938 0,139 Average time at a product descriptions page 12,7515 7,4498 0,000 Average time at a seller description page 12,3901 10,9899 0,226 2

Average time at a feedback page 8,9924 6,6728 0,007 Average time at a product descriptions page 10,9865 6,3529 0,000 Average time at a seller description page 10,6727 9,4402 0,119 3

Average time at a feedback page 8,2443 7,0038 0,276 Average time at a product descriptions page 9,8930 6,4949 0,001 Average time at a seller description page 9,4554 9,2045 0,680 4

Average time at a feedback page 9,1645 6,4432 0,005 Finally, the relation between gender and experience is explored. For this purpose is used a t-test. From the t-test can be made clear that the amount of experience is significant different between males and females (p-value = 0,003). The males in the sample have made on average 5,17 transactions on auction websites in the past twelve months, while this average for females is 2,15. Probably males visit pages more effective and efficient in comparison with females as a result of more experience. In summary, the difference in the number of visited pages between males and females can now be explained with a combination of issues. On the whole auction the difference can possibly be explained by the fact the males are searching more effective and efficient, while the larger difference at the feedback pages is reasonable, because females attach more value to these pages.

4.3 Stage 3: Latent Class analysis

4.3.1 Indicators and covariates

In the plan of analysis is mentioned that variables can be distinguished as indicators and covariates. The indicator for the LC analysis is the search behaviour of the respondents. In stage one and two of the analysis became clear that the number of visited pages is a more interesting variable in comparison with the average time on a page. Besides, it became clear that there are differences for this variable between page levels on the experimentally designed website. Therefore, variables are included which measure the number of times a respondent visits a certain pages level. Furthermore, the feedback level is split of in the three different feedback scores to get insights in which score is used more. By including these variables into the models can be made clear, which type and the amount of information are used by consumers to make a decision regarding the sellers. The variable average time on a page, which is used in stage one and two of the analysis, will also be included in the LC model. Based on the results of the previous stages, it is not interesting to divide this variable in page levels.

In table 15 are displayed the indicators and covariates, which will be included in the Latent Class analysis. The covariates are not different in comparison with the proposed covariates in the plan of analysis.

TABLE 15: Indicators and covariates Indicators Covariates Number of visited product descriptions pages Risk Number of visited seller description pages Trust Number of visited feedback score pages Experience Number of visited friends score pages Gender Number of visited rumour score pages Bid price Average time on a visited page

4.3.2 Models selection

clusters. The seven models consist of cluster solutions with to 2 till 8 clusters, because this number of cluster is manageable for interpretation. All 173 respondents are used in the segmentation models.

TABLE 16: Information criteria

Auction 1: New television Auction 2: Second hand television

# Clusters AIC AIC3 BIC CAIC AWE AIC AIC3 BIC CAIC AWE

2 5446,816 5461,816 5494,116 5509,116 5636,787 5192,301 5207,301 5239,6 5254,6 5371,542 3 5050,114 5073,114 5122,64 5145,64 5330,682 4821,996 4844,996 4894,522 4917,522 5108,989 4 4827,831 4858,831 4925,583 4956,583 5209,319 4672,84 4703,84 4770,592 4801,592 5058,688 5 4718,86 4757,86 4841,839 4880,839 5211,323 4542,424 4581,424 4665,402 4704,402 5023,063 6 4628,808 4675,808 4777,013 4824,013 5211,442 4495,73 4542,73 4643,934 4690,934 5082,096 7 4602,147 4657,147 4775,578 4830,578 5305,88 4475,115 4530,115 4648,546 4703,546 5170,156 D ep en d en t 8 4587,355 4650,355 4786,012 4849,012 5394,364 4457,327 4520,327 4655,984 4718,984 5242,065 2 5521,363 5535,363 5565,509 5579,509 5692,783 5281,589 5295,589 5325,735 5339,735 5449,228 3 5144,553 5165,553 5210,772 5231,772 5402,318 4986,46 5007,46 5052,679 5073,679 5239,327 4 4917,346 4945,346 5005,638 5033,638 5265,648 4788,413 4816,413 4876,705 4904,705 5133,203 5 4819,107 4854,107 4929,472 4964,472 5272,027 4658,941 4693,941 4769,307 4804,307 5096,967 6 4710,816 4752,816 4843,254 4885,254 5225,011 4583,636 4625,636 4716,075 4758,075 5108,34 7 4622,812 4671,812 4777,323 4826,323 5229,355 4532,394 4581,394 4686,905 4735,905 5142,547 In d ep en d en t 8 4611,038 4667,038 4787,622 4843,622 5325,459 4519,492 4575,492 4696,076 4752,076 5215,211

Auction 3: Travel guide Auction 4: Second hand chair

# Clusters AIC AIC3 BIC CAIC AWE AIC AIC3 BIC CAIC AWE

2 5155,119 5170,119 5202,419 5217,419 5340,437 4964,1 4979,1 5011,399 5026,399 5149,834 3 4724,939 4747,939 4797,465 4820,465 5015,792 4601,677 4624,677 4674,203 4697,203 4887,796 4 4482,632 4513,632 4580,384 4611,384 4865,82 4480,102 4511,102 4577,854 4608,854 4860,103 5 4378,532 4417,532 4501,51 4540,51 4878,41 4404,086 4443,086 4527,064 4566,064 4900,165 6 4343,817 4390,817 4492,022 4539,022 4947,081 4363,037 4410,037 4511,242 4558,242 4968,957 7 4322,646 4377,646 4496,077 4551,077 5027,693 4337,32 4392,32 4510,751 4565,751 5036,866 D ep en d en t 8 4305,666 4368,666 4504,323 4567,323 5101,587 4311,975 4374,975 4510,632 4573,632 5106,594 2 5236,587 5250,587 5280,733 5294,733 5409,88 5029,323 5043,323 5073,469 5087,469 5205,094 3 4911,897 4932,897 4978,116 4999,116 5179,976 4700,513 4721,513 4766,732 4787,732 4966,678 4 4678,768 4706,768 4767,06 4795,06 5027,024 4585,679 4613,679 4673,971 4701,971 4932,328 5 4577,647 4612,647 4688,013 4723,013 5009,422 4514,901 4549,901 4625,267 4660,267 4968,306 6 4475,521 4517,521 4607,959 4649,959 5007,82 4432,82 4474,82 4565,258 4607,258 4965,294 7 4398,28 4447,28 4552,791 4601,791 5012,93 4386,559 4435,559 4541,07 4590,07 5007,213 In d ep en d en t 8 4365,709 4421,709 4542,293 4598,293 5069,414 4365,921 4421,921 4542,505 4598,504 5073,872

The bold values are the lowest value of a specific information criterion.

respectively 7, 6, 6 and 8 clusters. The AIC(3) and BIC criteria selects relatively many clusters and this is harder for interpretation. The CAIC criterion select at all the auctions models with 6 clusters, while AWE select three times models with 4 clusters and one time a model with 5 clusters. These models are easier in use for interpretation and consequently the CAIC and AWE criteria are used to select the models. Based on these two criteria the dependent models with 5 clusters are selected for every auction, because this solution is a compromise at three of the four auctions, while at auction two this is the optimize model according the AWE criteria. In the next four sub-paragraphs, the models with five clusters at the four auctions are discussed.

4.3.3 Auction 1: New television

In this sub-paragraph are discussed the five clusters and the corresponding model at the auction of the new television. In table 17 is displayed information about the size of the clusters. The first row displays the probability that a consumer belongs to a cluster. For example, the probability that a consumer belongs to cluster 1 is 31,56%. In the second row are these chances translated to the sample and then can be stated that 54 of the 173 respondents belong to cluster 1. Hence, cluster 1 is the largest cluster, while cluster 5 is the smallest cluster. Between this two clusters are three clusters, which are comparable based on their size.

TABLE 17: Cluster size

Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Probability 0,3156 0,226 0,182 0,1569 0,1196 Absolute 54,60 39,10 31,49 27,14 20,69

The clusters are based on the six indicators in table 18. All variables are significant in the model. Besides, the R² value give insight to what extent a variable explains the differences between clusters. Consequently, the visited pages are more different between clusters in comparison with the variable average time on a page. The search behaviour of the consumers in the cluster can be described with the averages of the indicators in table 18.

The consumers in cluster 2 visit approximately the same amount of seller description pages as cluster 1. However, these consumers have more attention for the feedback sections. Probably, this plays a role in the decision of these consumers. Moreover, these consumers spend on average less time on a page, around 7 seconds.

The consumers in the third cluster can be characterized as consumers, who have a brief search regarding the visited pages. However, these consumers stay on average a long period on one page, around 20 seconds. It seems likely that the consumers select a couple of sellers based on the product descriptions and they make their decision based on some additional information at the seller description pages. These consumers do barely not visit the feedback sections.

The consumers in cluster 4 visit many pages during the auction. The consumers visit on average 16 seller description page. If every seller gets the same attention, then consumers visit every seller 2 till 3 times during the search phase. Hence, these consumers make their decisions probably based on comparisons between sellers. The feedback pages are barely visited. The average time on a page is around 7 seconds.

The consumers in cluster 5 visit the most pages and stay on average the fewest time on a page. Around 20 seller description pages are visit during the search. Moreover, the consumers also visit around 26 feedback pages. Thus, these consumers use a lot of information to make the right choice.

TABLE 18: Cluster averages indicators Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 R² P-value # visited product descriptions pages 9,27 9,60 3,17 17,15 21,01 0,6622 0,00 # visited seller description pages 8,27 8,60 2,17 16,15 20,01 0,6621 0,00 # visited feedback score pages 0,47 4,27 0,55 1,31 9,92 0,7202 0,00 # visited friends score pages 0,39 3,90 0,53 1,60 9,26 0,7071 0,00 # visited rumour score pages 0,33 3,00 0,34 0,81 6,84 0,7166 0,00 Average time on a visited page 10,02 7,51 20,39 6,85 6,64 0,4943 0,00

different, however the consumers who visit the most feedback pages have the highest trust rate. The hypothesis about experience is rejected in stage 1 of the analysis. Here it is noticeable that consumers with the least experience, have the most attention for the feedback pages (cluster 2 en 5). In cluster 2 and 5 are also relatively the most females. This confirms the issue in stage 2 that females have less experience and make more use of the feedback pages. Moreover, in the cluster with the short search phase (cluster 3) is the probability 94,22 % that the respondent is a male.

TABLE 19: Cluster averages and probability covariates

Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Bid price € 1408,36 € 1364,52 € 1287,95 € 1455,88 € 1437,33 Risk 2,93 2,98 2,99 3,23 2,91 Trust 4,98 4,92 4,88 4,69 5,05 Experience 5,50 2,35 4,34 5,52 3,70 Gender (Probability): Male 0,7318 0,6542 0,9422 0,7392 0,5676 Female 0,2682 0,3458 0,0578 0,2608 0,4324

4.3.4 Auction 2: Second hand television

The results of the LC model of the second auction are described at the same manner as the previous auction. However, the main results are presented and not every table will be explained in detail. From table 20 can be concluded that there are two small clusters (cluster 4 & 5) and three larger clusters (cluster 1, 2 & 3). This combination of cluster sizes is different in comparison with the first auction. There are two extreme sizes and in between are three comparable cluster sizes.

TABLE 20: Cluster size

Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Probability 0,316 0,2867 0,2253 0,1013 0,0707 Absolute 54,67 49,60 38,98 17,52 12,23

The consumers in cluster 1 are visiting on average almost six seller description pages and sometimes they have interest for the feedback section. The average time on a page is approximately 11 seconds. The behaviour and the size of this cluster are comparable with the first cluster at the new television auction. Although the consumers at this auction visit slightly fewer seller description pages.

The second cluster at this auction exists of consumers who visit on average 12 seller description pages. Hence, if every seller gets the same attention, then every seller has two visits. Besides, these consumers use detailed feedback information to decide which seller to choose. The average time on a page is around 8 seconds.

The consumers in the third cluster visit the same amount of sellers, but these consumers do not use the feedback sections to make a choice. The average time on a page is also comparable with the previous cluster.

The search behaviour of the consumers in the fourth cluster has similarities with the third cluster at the new television action. However, this cluster consists of fewer respondents. It seems like that the consumers make their decision based on the short descriptions at the product description pages. Besides, the consumers stay on average relatively long on a single page (29 seconds).

The size and the behaviour of the fifth cluster are comparable with the fifth cluster at the previous auction. These consumers have attention for the most seller description pages and many feedback pages. Besides, the average time on a page is the shortest. Hence, these consumers use a lot of information to pick a seller.

TABLE 21: Cluster averages indicators Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 R² P-value # visited product descriptions pages 6,81 12,96 11,95 2,52 22,31 0,6666 0,00 # visited seller description pages 5,81 11,95 10,98 1,52 21,31 0,6655 0,00 # visited feedback score pages 1,48 4,76 0,08 0,15 9,73 0,683 0,00 # visited friends score pages 1,08 4,92 0,12 0,01 8,70 0,7238 0,00 # visited rumour score pages 0,79 3,58 0,10 0,01 7,70 0,7317 0,00 Average time on a visited page 11,55 8,00 8,44 29,64 6,58 0,5483 0,00