How pictorial messages could influence the helpfulness and trustworthiness in online reviews

Faculty of Economics and Business

Master Thesis MSc Business Administration - Marketing Track

How pictorial messages could influence the helpfulness and

trustworthiness in online reviews.

Student name：Yating Wu

Student number: 11099062

Supervisor: Situmeang, Frederik

Statement of Originality

This document is written by Student Yating Wu who declares to take full

responsibility for the contents of this document.

I declare that the text and the work presented in this document is original

and that no sources other than those mentioned in the text and its

references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the

supervision of completion of the work, not for the contents.

Abstract

Online reviews play the important role in customers’ online shopping process because they believe online reviews are more credible than information compared with other sources such as product details posted by sellers. Then the problems of how to increase helpfulness or trustworthiness of online reviews draws attentions of researchers. Even though many previous researches have demonstrated many antecedents to influence helpfulness and trustworthiness of online reviews, there are still some antecedents missing. This research test one of the missing factors, the presence of pictures, about how and when pictures could influence helpfulness and trustworthiness of online reviews. Based on the prediction of various function of pictorial messages, we developed and tested a model of describing the relationship between pictures and helpfulness/trustworthiness. Collecting data from 157 students among under-graduated and graduated students. Results showed that in clothes shopping, the presence of pictures in online reviews would increase both helpfulness and trustworthiness, and high relevance pictures is more effective than medium or no relevance pictures. And we also found the positive relationship between helpfulness and trustworthiness. In contrast, pictures are less influential in electronic devices settings and only medium relevance pictures would affect helpfulness and trustworthiness. The moderating role of involvement, propensity to trust and proficiency of shopping online were all rejected with the hypothesis we stated. These results call the theory used in literature review into question. And there is a need to revise these theories about their effectiveness of working in online settings. In last few parts, we discuss the implication in both theory and practice and limitation of our findings.

TABLE OF CONTENTS

Abstract ... 3

1. Introduction ... 5

2. literature reviews ... 8

2.1 Background ... 8

2.2 Various functions of pictorial messages ... 11

2.3 various relevance with products ... 13

2.4 Helpfulness ... 15 2.5 trustworthiness ... 20 2.6 involvement ... 23 2.6 conceptual model ... 29

3. method ... 30

The pilot tests ... 35

3.3 Results ... 35

Test of H1: Pictorial messages role to influence perceived helpfulness ... 36

Test of H2: Pictorial messages role to influence perceived trustworthiness ... 38

Test of H4: The moderator role of involvement ... 39

H3a test: ‘Proficiency of using online service’ as a moderator: ... 42

H3b test: ‘Propensity to trust’ as a moderator ... 43

Further analysis ... 43

4. Discussion ... 45

5. Limitations and further research ... 51

6. Reference ... 54

7. Appendix ... 61

Experiment treatment ... 61

1. Introduction

Online reviews, electronic word-of-mouth (e-WOM), nowadays plays important roles in helping consumers to make purchasing decision when shopping online. This is because, for customer, information provided by other consumers are more credible compared with product information provided by producers (Bickart & Schindler, 2001). However, not every online reviews are useful, reviews will be useful only when consumers trust these reviews and then use these reviews to make decision (Ardion Beldad, Menno de Jong, Michaël Steehouder,2010). Thus, it is important to know which kind of online reviews are helpful and when customers trust information provided by reviewers. Antecedents of helpfulness and trustworthiness are defined as factors which will influence customers’ perceived helpfulness and trustworthiness on online reviews. Many previous researchers have already made huge progress about investigating various antecedents. For example, Mudambi and Schuff (2010) found that product type moderates the helpfulness of review for experience goods, while review depth has a greater positive effect on the helpfulness of the review for search goods. Raffaele Filieri (2014) used duel process to divide consumer evaluation process into two ways, normative evaluations and informational evaluations, and he found that informational influence is larger than the normative influence on helpfulness and information quality is tested to be the most effective antecedent of helpfulness of online reviews. Salam, Iyer, Palvia, & Singh (2005) proved that people’s propensity to trust anyone and anything influence the trust in online reviews. However, there are still some antecedents missing, even though co many researches have involved in researches about online reviews. In this paper, we will

investigate one of the missing antecedents, pictorial messages, to fill the gap of researches in online reviews’ helpfulness and trustworthiness.

Pictorial messages are normally seen in online reviews and many online sellers encourage reviewers pose pictures on their online reviews. And some online shopping interface highlight pictures posed by reviewers. Take Yelp.com as an example, it is a famous online review website gathering reviewers for all kinds of products or services. This website shows its interest on the pictures provided by reviewers and makes fully use of the pictorial messages. in Yelp.com, interface for service store or product store, pictures posed by reviewers are always placed on the top of website in order to attract other customers’ attentions. However, little researches investigated whether posing pictures in reviews will always make online reviews more helpful and trustworthy to other customers and if yes, whether there are other antecedents to influence the relationship between pictorial messages and perceived helpfulness? whether there are also some moderating factors to influence that relationship? So in my research, we will fill in this gap and focus on two research questions: when and how pictorial messages are helpful in online reviews; whether pictorial messages will increase trustworthiness towards online reviews and if yes, when it will increase trustworthiness.

Understanding pictorial messages as an antecedent of helpfulness and trustworthiness of reviews is important to both sellers and reviewers. Vendors should know whether invest money on improving the content by encouraging reviewers to add pictorial messages or just designing the interface of online reviews to image only web page. For reviewers, they will

know when they are supposed to add picture on their reviews to make their reviews more helpful and trustworthy for others.

In the first section, we introduced background knowledge of online reviews including some achievement of existing researches about online reviews and model that existing researchers always use to state their hypothesis when involved in the issue about helpfulness and trustworthiness in online reviews. In second section, we discussed about the different roles of pictures in online reviews. The third section deals with one possible antecedent, relevance, in influencing the relationship between picture and helpfulness/ trustworthiness. And we categorized pictures into three: high relevance, medium relevance and no relevance. The fourth and fifth sections presented the concept of helpfulness and trustworthiness in online reviews and postulate some hypothesis based on three levels of relevance. In sixth section, we stated one possible moderator, involvement. Then in the experiment design section, we described our experiment which is to test the hypothesis and presented the results of our experiment. The paper ended with discussion of the results and contribution, limitation of our research.

2. literature reviews

2.1 Background

Online reviews (e-WOM) refers to ‘‘statement wrote by former, actual consumers about presented product on interface, which is available to other website viewers’’ (Hennig-Thurau, Gwinner, Walsh, & Gremler, 2004, p. 39). In this research, online reviews refer to one piece of reviews which contain all of the messages including verbal messages, pictorial messages, consumer profiles and so forth. So many researches were on online reviews, especially the antecedents of online reviews, these days as the prevalence of online shopping. There are mainly two parts for the research, one is about the final performance of antecedents. For example, some previous researches have investigated the effect of online reviews on sales (Dellarocas et al., 2007; Godes & Mayzlin, 2004; Vermeulen & Seegers, 2009). Other researches focus on how antecedents in online reviews influence consumer behaviors, which including information adoption (Xiao-Liang Shen, kem Z. K. Zhang, Sesia J. Zhao, 2014), Purchase intentions (Park, Lee, & Han, 2010), and attitudes changes towards products (Lee, Park, & Han, 2008). For example, researches convinced that some antecedents such as ratings, word count and characteristics of reviewers have positive influence on customers’ purchase decision making process. To be specific, Mudambi and Schuff (2010) convinced that review extremity, review depth as well as product categories influence helpfulness of reviews. In this research, we will investigate one of these antecedents, pictorial messages attached with one piece of reviews, and discuss about how this antecedent influence consumer behavior.

Pictorial messages, compared with verbal messages, is ‘‘a visual representation of a person, object, or scene, as a painting, drawing, photograph and so forth’’. However, little researchers investigated about how the pictorial messages exert their impacts under the settings of online reviews. Pranjal Gupta (2010) ever did a research about how antecedents, emotional expressions which is one kind of pictorial messages, influence online reviews’ helpfulness. And he found that negative emotional expressions would decrease customers’ informative value and product evaluations in online reviews. Instead, there are many researches investigating pictorial messages’ performance on consumer behavior. For example, some researchers studied the importance of pictorial message during information adoption process. Majooni, Azam, Masood, Mona and Akhavan, Amir (2015) conduct a research with Felder-Silverman learning style model test in order to identify preferred learning style of the participants on electronic device. And they found that ‘75% of the participants were visual learners and 25% verbal learners’. Furthermore, many researches have proved that the combination of text and picture provides better learning compared to text only and picture only situations (Fletcher & Tobias, 2005). All the researches above stated the importance of pictorial message in information adoption. However, in online settings, consumer have various purposes when reading the online reviews and they have various involvement for different goods (Hyunmi Baek, JoongHo Ahn, and Youngseok Choi, 2012). Information presented in online reviews might serve in different cues in various situations. That means it is possible that pictures have various roles in such different situation and then not all of the pictorial messages in their roles could exert positive influence on online reviews’ performance. However, function of pictorial messages on the process of influencing review

adoption is still unclear. Thus, this research will fill this gap by investigating the various roles of pictorial message during the process of customer decision making when reading online reviews and their influence on customer behavior.

The Elaboration Likelihood Model (ELM) has been widely used in online review research. This model states that consumer in different involvement with product will process information differently (Gupta & Harris, 2010; Park & Lee, 2008). In online settings, when consumers look through online reviews, the information contained in one specific review will be a cue to influence consumers’ evaluation to this product (central route), or be a cue to motivate consumers to make purchase decision directly (peripheral route). To be specific, In other words, when the elaboration likelihood is high, people will make decisions in central route and that means the central cues to persuasion should be effective, but when elaboration likelihood is low, people will make decisions in peripheral route and thus the peripheral cues should be better (Petty,R. E. & Cacioppo,J.T. 1983). Therefore, this model provides us with basics for understanding role of function messages in online reviews. There is evidence that one factor of object depicted by that factor can serve as different roles in different contexts (Petty and Cacioppo, 1983). To be specific, one variable which is perceived as a peripheral cue can also serve as an issue-relevant argument in other experimental study. For example, in advertisement, humor function as a message argument when promote products including humorous television shows, funny movies and others that have the quality of funniness. The funnier the advertisement is, in other words, higher quality of the arguments, the more effective the humor advertisement will be. However, humor can also be and always be a peripheral cue in other products’ advertisement such as chocolate advertisement (Yong Zhang

and George M. Zinkhan, 2006). Thus, when pictures are used in persuasive communication, we predict that functions of pictures would vary based on the way that pictorial messages are presented by reviewers.

2.2 Various functions of pictorial messages

Pictorial message as a message argument

Compared with verbal messages, pictorial messages are always seen as lack of argumentation in persuasive communication. However, as stated before, one variable which is always perceived as a secondary information also can serve as an issue-relevant argument. According to example of humor’s various function, humor will be issue-relevant argument when humor is central to the merits of object (Petty and Cacioppo, 1986). Thus, in online reviews, picture might serve as an issue-relevant argument if pictorial messages could display central merits of related products. That means customers can find the information which they want to know about the product through watching the picture. In this case, picture serves as a cue to the quality of the product. For example, when customer book hotel room online, they usually care more about environment, facility of the room, as well as other factors such as transportation, services. Then pictures contain the information above will function as issue-relevant arguments. And since environment around and facility are attributes which can be described better by picture than words, we can predict that high quality pictures are more influential to consumers than verbal words. And the more information the picture contains or relevant with the product, the higher probability that pictorial messages would positively influence consumer behavior when reading online reviews. Examples of this kind of products include hotels, restaurants, clothing and others that possess visual attribution as their main

merits. Then we assume that pictorial messages would function as issue-relevant arguments when it shown in reviews for these kind of products.

Pictorial messages as peripheral cue

While pictorial messages can act as issue relevant argument that relates to the main attributes of the product, there are still some other products that are difficult depicted by pictures such as laptops, movies. Take laptops as an example, customers prefer reading some important dimensions such as storages of hard disc which these information is impossible to get when just see a picture. Thus, in such cases, pictures in online reviews are more likely to serve as a peripheral factor.

Consumer decisions, in this case, are more likely influenced by meaningful arguments, which could be verbal messages from online reviews. Therefore, if picture and text coexist, the text is seen as an anchor and guides the identification and the interpretation of pictorial message (Barthes, 1978). In addition to that, there are another research found that in this occasion, people will spend more time on text, however, they might have better memory about information from picture (Keith Rayner and Caren M. Rotello ,2001). In other words, customers would pay more attention on verbal messages and make decisions based more on verbal messages. But how pictorial message as a peripheral cue to influence decision making? It seems that compared with verbal messages without picture attached, pictures could attract customers to read the verbal reviews with that picture attached. That means pictures increase possibility of adoption of that reviews and increase the interest of reading the verbal messages. However, customers would not take pictorial messages into consideration in decision making.

Example of that such products include movies, electronic devices, drugs and others that do not possess the quality which can be displayed by pictures. And we assume that pictorial messages would serve as peripheral cues in this case.

2.3 various relevance with products

Relevance has been defined as ’the concept of one topic being connected to another topic in a way that makes it useful to consider the first topic when considering the second’. (Hjorland & Sejer Christensen,2002, p961). Petty and Cacioppo (1986) argued that people perceive two arguments to be relevant when one meets the person’s subjective determination of the true merits of the other one argument. Here is always the case that when pictorial messages could serve as issue-relevant argument, however, reviewers pose weak and less relevant picture such as posing a cat picture under laptop’ online customer reviews. In this situation, picture might be seen as a distraction and will engender unfavorable attitudes to this reviews (Yong Zhang and George M. Zinkhan, 2006). Thus, this research will take different level of relevance into consideration to investigate the perceived influence of relevance on relationship between picture and helpfulness/trustworthiness.

High-relevance

Pictorial messages can serve as issue-relevant argument or peripheral cue only when pictures are highly related with products. Then it is possible for this picture to depict the true merits of product (Petty and Cacioppo, 1986). No matter what function picture served and whether it can reveal main attributes of product, these various functions could work when customers could found true information from seeing this pictures. Otherwise, we predict that influence might be reduced because of this irrelevance.

No-relevance

While pictures can function as various roles in different products’ reviews, there are still some picture shown in reviews with low relevant information. For example, reviewers pose a picture of a tree outside the hotel when writing hotel online reviews. This irrelevant picture will be a distraction cue which decrease review recipient’s motivation and ability to process the message (Petty, Cacioppo, and Schumann, 1983). Thus, the picture might generate recipient’s dislike attitudes towards this review and will be detrimental to customer decision making process. In this case, pictures in online reviews will act as neither issue-relevant argument nor peripheral cues. Instead, it will act as a distraction cue which might hurt elaboration likelihood in ELM or act as cue which can generate unfavorable attitudes during customer decision making in online shopping.

Medium-relevance:

In addition to that, there is another scenario, reviewers do not think too much about which kind of pictures are highly relevant with product and they just put some pictures they think is related to product. For example, when consumers write reviews, they might pose a logo as a picture on their reviews. On the one hand, the logo reveals some attributes towards product, on the other hand, it is hard to say that this logo depicts the entirely true merits which other customers will care about. Based on definition of relevance, the degree of relevance might operate on a continuum, which means that every pictures are relevant with product in a degree between high and low relevance. Thus we assume that relevance in this case is medium. Less researches investigate about whether medium relevance cues would influence consumer decision making process when shopping online. However, this level of relevance

must be taken into consideration since there are more online shop sellers encourage reviewers pose pictures with some rewards such as credit and money refund. This might cause reviewers pose pictures without deliberate thinking when writing online reviews. Then it comes to question that whether this kind of picture exert same effect on consumer behavior as highly related picture.

2.4 Helpfulness

Online reviews can give customers confidence on purchasing online. However, there are sometimes so many reviews shown on website to make customer confused (Frias, Rodriguez, & Castaneda, 2008). Therefore, it is a vital task for online shop vendors provide easily access to consumers and helpful information in order to increase review diagnosticity (Zhiwei Liu, Sangwon Park, 2014). When customers reading reviews, Two kinds of costs exist. One cost is happened during customers searching information on online shopping interface which is named search cost. The other one is determining which information to adopt and it is named cognitive costs (Mudambi & Schuff, 2010). This means online reviews will be perceived helpful if there are cues to help customers decrease searching cost and at the same time, gain sufficient information of products to decreased cognitive costs. Thus helpful online reviews mean that online platforms provide customers with satisfied experience for helping them filtering useful information in reviews’ pool and providing more value on purchasing because of confidence building (Gupta & Harris, 2010).

In order to decrease costs during customer decision making process, it is vital for online vendors to know how to increase helpfulness of online reviews, in other words, the antecedents of helpfulness. Previous researches have contributed much on discovering factors

to increase helpfulness. By using dual process model, Raffaele Filieri (2014) divided general reviews’ antecedents into two types. Normative influences are antecedents about product rankings and consumer ratings, and informational influences are antecedents about reviews’ quality, reviews’ credibility and the quantity of reviews. He found that consumers are primarily influenced by the quality of information and subsequently influenced by customer ratings and overall rankings. Both antecedents can make reviews helpful. There is another research found that helpfulness itself can also be a cue to influence helpfulness of reviews overall. Many website always encourage consumers to vote on whether they feel this reviews are ‘helpful’ for them (Zhiwei Liu a, Sangwon Park). For example, amazon.com always put first two helpfulness reviews voted by consumers on the top of reviews list. The reason why amazon.com do this is to help consumers filtering reviews and make consumer more confident when reading reviews.

However, the most prevalent classification of antecedents is classifying them into quantitative and qualitative factors. Quantitative antecedents include words cound, review ratings, and qualitative antecedents includes strength of reviews and reliability of reviews. Quantitative antecedents have been fully investigated by researchers so far but researches about qualitative factors are limited. Schindler and Bickart (2012) stated that the message contents and styles are vital factors which make the reviews appealing to consumers. This research will focus on one qualitative antecedent, image, to investigate its relationship with helpfulness. Junyong Kim, Pranjal Gupta (2010) conducted a research about helpfulness of one kind of small images, emotional expressions, which are some emotional icons always contained in reviews. But the images involved in this research are pictures which we can see

some contents or story in them. For example, reviewers share pictures of the food which they eat in the reviews for a specific restaurant.

Some previous researchers used diagnosticity-accessibility model to investigate persuasiveness of e-WOM. Diagnosticity is whether the website convey relevant product information which could help customers evaluate the quality of online products. Herr, Kardes, and Kim (1991) came up with the diagnosticiy-accessibility model firstly and Raffaele Filieri (2014) applied it under e-WOM settings. He stated that e-WOM generates greater influence on product evaluation than WOM for its vividness and this influence comes from greater information diagnosticity of e-WOM. This means the greater the diagnosticity is, the more signals customers could discover in order to evaluate the quality of products (Jiang & Benbasat, 2004). Thus, information diagnosticity also represent the increased knowledge about a product and it also cause information adoption from online shopping interface. Diagnosticity is often defined as the degree of helpfulness of information (Qiu, Pang, & Lim, 2012; Skowronski & Carlston, 1987). Therefore, investigating the level of diagnosticity of online reviews is effective for understanding the level of helpfulness. As diagnosticity is determined by how much information and how ease customer process information during decision making process, it is important to discover how antecedents of helpfulness influence the amount and accessibility of information when reading online reviews.

Decision making process in online shopping includes two stages, information search stage and evaluation of alternative stage (Gerald Häubl and Valerie Trifts, 2000). Baek, Ahn, and Choi (2013) found that customers focus on different cues in different stages. To be specific, customers pay more attention on peripheral cues in information searching stage and

focus on central cues in evaluation of alternative stage. In addition to that, Gerald Häubl and Valerie Trifts (2000) pointed out that a useful cue can influence two variables in decision making process. First one is the size of the first pick-up alternative sets. On first stage of information searching, customer will be end with several satisfied products in a set for further decision making. And the second one is the quality of customers’ purchase decisions. This variable includes the efficiency of purchasing and certainty of customers’ final decision. High related picture and its influence on helpfulness:

Picture as a strong argument and its influence on helpfulness: As mentioned before, when central merits of products can be depicted by pictures, highly related picture can function as an issue-relevant argument. Verbal messages are always considered high related with strong argument. Compared with pictorial messages, verbal messages need customers to spend more time on processing and are less likely for customers to recall the content of messages after 30s (Keith Rayner and Caren M. Rotello ,2001). Considering the both process of making decision in online shopping, as strong arguments, pictorial messages might decrease the set of alternatives in first stages of process. Because compared with verbal reviews, pictorial messages are more easily to be extracted their information from pictures. The least effort principle states that people naturally choose the path which engage their least ‘effort’ (Guillaume Ferrero, 1894). We assume that objects which is easy to be extracted their information would take less effort. Then since pictorial messages provide customers a least effort path, they will spend less time on picking up messages they need and will also efficiently and effectively compare alternatives as pictorial cues which will last longer in memory. Based on theory mentioned above, it means that pictures increase the quality of

decision making and decrease the time they spend on selecting alternative sets. In other words, pictorial messages increase the helpfulness of online reviews.

Picture as a peripheral cue and its influence on helpfulness: when central merits of products cannot be expressed by picture, high related picture will serve as a peripheral cue. Because peripheral cues would exert more effects when customers think less deliberately, we assume that when customers process information in peripheral routes, picture which act as peripheral cue will engender greater helpfulness . As stated before, on the first stage of decision making, customers always filter information among reviews’ pool and they care more about peripheral cues during this stage. This means reviews with peripheral cues have greater possibility of attracting customers’ attention. The more peripheral cues present in this stage, the higher diagnosticity of information adoption. In addition to that, pictures can serve as cues to help customer filtering information especially when customers process information in peripheral route. That means with these peripheral cues, customer will spend less time on information filtering. The decreased cost for information searching represent increased helpfulness. Thus we predict that for product cannot be depicted entirely its central merit by picture, high related picture still can increase helpfulness of online review.

We mentioned two conditions about pictures in high relevance, one is as a central cues and another one is as peripheral cues. We predict pictures will exert both positive influence on helpfulness in these two conditions. However, we cannot predict the difference of valance on helpfulness between these two conditions. In experiment, we will test the difference of valance by applying two type of products as two treatments.

As mentioned before, no matter what level of relevance between picture and product, there are reviewers pose no related pictures on their online reviews such as posing a dog’s picture when writing reviews about laptop. In this case, picture might be distraction of customer which will engender unfavorable to the reviews and detrimental to the formation of positive attitudes (Yong Zhang, George M. Zinkhan, 2013). In addition to that, Schnotz & Kürschner (2007) mentioned that combination of textual and visual data should be implemented carefully, as it might cause interference during users’ learning process because of inappropriate graphics. Therefore, we predict that no related pictures will generate negative effect on helpfulness of online review.

Medium related picture and its influence on helpfulness: It is possible that reviewers pose medium related pictures such as brand logo on their online reviews. On the one hand, this kind of messages cannot help customers to know the quality of product as strong argument to increase helpfulness. On the other hand, when customers see medium related pictures, they might attribute this behavior of posing picture as customers’ disposition instead of generating negative attitudes to online review (Junyong Kim, Pranjal Gupta, 2010). Thus we predict that medium related picture will not influence helpfulness of online reviews.

H1: If pictorial messages highly related with product, the presence of pictorial messages will increase helpfulness of online reviews. In contrast, if pictorial messages are medium or no related with product, helpfulness will not change or even backfire.

2.5 trustworthiness

As discussed above, a helpful online review increase diagnosticity of information in information adoption process. However, with prevalence of online shopping, when customers

making purchasing decision and accepting information from online reviews, they not only look for helpful information that can help them make decision, but also consider whether the online reviews are trustworthy (Ardion Beldad, Menno de Jong, Michaël Steehouder, 2007). Considering risk existing in online shopping, online trust is defined as ‘an attitude of confident expectation in an online situation of risk that one’s vulnerabilities will not be exploited’ (Corritore, Kracher, & Wiedenbeck, 2003, p740). Thus when customers chose to regardless the risk of believing information from online reviews, that means customers fell trustworthiness in online reviews. Online trust is also defined as that customers would like to rely on one object and information provided by this object (Shankar, Urban, & Sultan, 2002). In online reviews, this ‘object’ represents reviewers and this definition means customer shows reliance on information provided by a specific reviewer and trust on this reviewer. Then trust can be used to explain why consumers adopt one online review instead of other ones.

Many previous researchers have made some contributions on the relationship between trust and information adoption. Horst, Kuttschreuter, & Gutteling, (2007); Welch & Hinnant (2005) convinced that trust is a vital factor in adoption of e-government. Sillence, Briggs, Fishwick, & Harris (2004) proved the positive relationship between trust and adoption of e-health. Except so many researches about e-government and e-health, Ardion Beldad, Menno de Jong, Michaël Steehouder (2010) conducted a meta-analysis about how trust influence information adoption in e-commerce and they stated that trust is an effective prediction for customers’ adoption of electronic information.

Then what factors could make an online review trustworthy? Previous researches investigated various antecedents of trust in online commerce by empirical studies. For

instance, Gefen and Straub (2004) found social presence is an important antecedents of trust; Bart Shankar, Sultan, and Urban (2005) used empirical experiments to prove that privacy protection guarantee and security assurances lead trust of online transaction. In this research, we will use some of these useful antecedents to predict trustworthiness of pictorial messages. Information quality: information quality refers to usefulness, accuracy and completeness of information offered (Liao, Palvia, and Lin,2010). As predicted before, highly related pictures can increase helpfulness of online reviews no matter whether main merits of products can be depicted by pictures, then it is possible that trustworthiness will increase when there are high related pictures. In addition to that, compared with verbal only online reviews, reviews with pictures can provide more information which improve completeness of online reviews and that means trustworthiness will also be improved because of presence of pictures. However, if pictures posed by reviewers are medium related or no related with products, the usefulness is nor increased and pictures in this situation will not enhance affluence of information. Not only this, no related information might decrease accuracy of the whole online reviews which will decrease trustworthiness in reverse. Thus we predict that medium and no related pictures will not increase trustworthiness of online reviews and might decrease trustworthiness of reviews.

H2: If pictorial messages highly related with product, no matter whether picture can reveal central merits of product, pictorial messages will increase trustworthiness of online reviews. In contrast, if pictorial messages are medium or no related with product, trustworthiness will not change or even backfire.

Individual propensity to trust: When shopping online, different customers show various degree of trust to exchange partners (Mayer et al., 1995). To be specific, some customers prefer to trust anything and anyone in the process of online shopping. However, other people will show their trust to limited information and reviewers (Salam, Iyer, Palvia, & Singh, 2005). Gefen (2000) found that customers displayed greater propensity to trust are more easily to show trust in online shopping. Therefore, individual propensity to trust might influence the effect caused by pictorial messages on trustworthiness. For instance, when propensity to trust is high, we predict that the change based on whether there is pictorial message in online review is less because of the high trust on original review and these people will trust most of the staffs no matter how trustworthy they are.

Proficiency in internet usage: Individual proficiency in internet usage based on experience of shopping online. Customers with high proficiency is perceived experienced internet user and these people always perceive internet shopping as low level risk and vice versa (Metzger, 2006). Thus, experienced internet users display high trust on online shopping with low risk perception (Corbitt, Thanasankit, and Yi, 2003). Therefore, we predict that individual proficiency in internet usage would decrease the effect of pictorial messages on trustworthiness to online reviews.

H3 Individual propensity to trust and proficiency in internet usage would moderate effect of pictorial messages exerted on trustworthiness of online reviews.

2.6 involvement

Many researchers make definition on involvement. Sherif defined customer involvement to be a state originated from ego involvement which is determined by how much

people conscious themselves with an object (Sherif, C.W., Sherif, M. and Nebergall, R.E. ,1965). It means that when people holds high involvement to a product, they think this product is associate with themselves which evokes strong feelings of self-identification. Thus, involvement can be described as to which extent people feel the association with object. In order to investigate the involvement, it is needed to know when and what can make people feel personal relationship with objects, in other words, the antecedents of involvement. When it comes to relationship, it means that it concerns with two entities and their environment. So we can roughly divide antecedents into three parts, one part is product related, second part is individual related and the other one is environment related.

Classification of customer involvement

In some previous researches, customer involvement is only for product involvement which is based on product choice behavior theory (Flynn and Goldsmith, 1993; Kapferer and Laurent, 1993; Laurent and Kapferer, 1985; Mittal and Lee, 1989). With development of involvement theory, now other type of involvement has been measured and stated. These are brand decision involvement and purchasing involvement which are based on shopping type choice (Lawrence S Lockshin, Anthony L Spawton, Gerrard Macintosh, 1997). Brand decision involvement is stated by Mittal and Lee (1989). It is ‘the interest taken in making the brand selection’ and it has antecedent that the situation under low product involvement. This means that when customers know little about product, they will consider to buy a product in a brand they like. Purchasing involvement is ‘an activity that some consumers are much more involved in than other consumers’ (Slama and Tashchian, 1985). Purchasing involvement is highly related to shopping effort (Slama and Tashchian, 1983) and response to direct

marketing appeals. When purchasing involvement is high, consumers would like to pay much more effort on purchasing and are more sensitive in direct marketing appeals such as price appeal in order to obtain the best value. Several demographic characteristics and channel that customer choose would influence purchasing involvement. For example, people show high purchasing involvement when purchasing online since compared with shopping in store, online shopping are less trustful so that customers should spend more effort in making purchasing decision.

Product involvement is based on an individual level to determine the relationship between a specific product and consumers themselves. In my research, we will not investigate all these three involvements because it is complicated when considering customer association with brand. Thus we will focus on describing product involvement and investigating customer purchasing involvement in online shopping generally.

Antecedents of product involvement—customer involvement profile

Product involvement is defined as the extent of customer making decision and searching information. It is also considered as an important variable to influence the helpfulness of online reviews. Then it is significant to state a standard method to measure the level of involvement. Firstly, it is required to know the antecedents of product involvement to set criteria. The concept of customer involvement profiles is identified by Kapferer and Laurent in 1985 for the sake of creating convenient criteria to distinguish customers for any product class. This profile is made up by five antecedents of involvement: interest, pleasure, sign, risk importance and probability of error. And these antecedents can be measured by

asking customers valuing each item with 5 levels (1=strongly disagree, 5=strongly agree) towards some statements which are related with the identified five antecedents.

Interest is whether customers are interested with the product they are going to purchase. Pleasure is the enjoyment when customer searching and purchasing online. Sign value is attributed by customer to the product, its purchase or its consumption for avoiding psychosocial risk. Risk importance is defined as when customer purchase wrongly, to what extent they will feel upset, annoyed. Probability of error means the uncertainty feelings when purchasing products.

Influence of involvement level on consumer behavior

Involvement and helpfulness:

Previous research has proved that customers with high involvement will spend more time on purchasing process (Lastovicka and Gardner, 1979). When considering ELM, high involvement represents processing information in central route or, conversely, in peripheral route. This means that when customers are highly involved with product and make purchase decision deliberately, central cues will be helpful; and when customers are less involvement and with low level of deliberate thinking, peripheral cues will be helpful.

To be specific, highly involved customers engage in more cognitive thinking and put more effort on decision making. They are more likely to find some central cues to assist them. Especially when customer shopping online, if they are highly involved with the product, they will focus on information which can reflect the main attribution of product. As discussed before, picture could serve as various functions in online reviews. When picture function as issue-relevant argument, it is a central cue. Then we can predict that these cues will become

more helpful when customers hold high involvement. These issue-relevant arguments can be processed more deliberately in this situation. However, when picture serve as a peripheral cue, its influence on helpfulness might reduce when customers have high involvement because they prefer to look for some strong argument for cognitive thinking and ignore these secondary related pictures.

In terms of low involvement, previous research proved that people in low involvement engage in no cognitive effort. Instead, they exert affective effort in making decisions which will lead to customers make some heuristic decisions because of liking or disliking towards products. In this situation, customers would focus on some peripheral cues. Thus in this case, pictures served as peripheral cues would be more helpful because of less deliberate thinking. However, as pictorial messages are easily for readers to get its meaning than verbal message which will satisfy customers who do not want to think deliberately. Even though picture can serve as central arguments, it still exerts high effect on helpfulness. Then we predict that when picture functions as peripheral cue, people with low involvement will find it more helpful when reading online reviews.

H4a: Involvement moderates the effect of pictorial messages on helpfulness. Under high involvement, helpfulness will increase when pictures served as issue-relevant argument and will decrease when pictures served as a peripheral cue. However, under low involvement, helpfulness will increase when pictures served as a peripheral cue, but no changes with helpfulness when picture served as issue-relevant argument.

Less previous researches investigated how involvement moderate effects of online reviews antecedents on trustworthiness. Sonia San Martín, Carmen Camarero, Rebeca San José (2008) state that effect of involvement on relationship between antecedents and attitude change is still not so clear. In online reviews, trust is customer’s attitude to reviewers about whether to believe information provided by them. With ELM, customers’ attitude might change when they process central cues in central route under high involvement and process peripheral cues in peripheral cues in peripheral route under low involvement. If ELM can explain the attitude change in settings of online reviews, trust, as an important attitude in decision making process, can also be predict with ELM (Burke, R. R, 2002). That is under high involvement, pictures served as issue-relevant argument will positively influence attitudes; under low involvement, pictures served as peripheral cues will positively influence attitudes. Therefore, pictures as strong argumentation will generate more trust under high involvement and picture as peripheral cue will generate more trust under low involvement. In addition to that, as predict before, information quality is an important antecedent for online trust, and it includes usefulness, accuracy and completeness (Liao, Palvia, and Lin, 2006). We assume that if information usefulness increase, information quality would increase and its trustworthiness increase accordingly. Usefulness and helpfulness are two concepts which can replace with each other (Russo, Meloy, and Medvec, 1998). With prediction of involvement’s moderate role in pictorial messages and helpfulness, under high involvement, pictures served in issue-relevant argument will make reviews more helpful. And under low involvement, pictures served in peripheral cues will make reviews more helpful. Then we can postulate that involvement will also moderate relationship between pictorial messages and

information trustworthiness because of the prediction that helpfulness is one antecedents of trustworthiness.

H4b: Involvement moderates the effect of pictorial messages on helpfulness. Under high involvement, helpfulness will increase when pictures served as issue-relevant argument. However, under low involvement, helpfulness will increase when pictures served as a peripheral cue.

2.6 conceptual model

Figure 1. conceptual framework

Pictorial messages relevance helpfulness

trustworthiness Propensity and proficiency

To internet

H1

H2

H3

H4a

H4b

3. method

3.1 Experiment design

Two hundred and fifty students from both under-graduate and graduate school in university of Amsterdam participated in the study. Only 161 participants finished the questionnaire and after delated all the individual data with missing data or invalid data, there were 157 participants’ data analyzed in the following test. All of the recipients were randomly assigned to cells with 4 different level of relevance manipulated groups (high relevance, medium relevance, no relevance and control group without picture). Every questionnaire contains two types of products, one is clothes and the other one is electronic device.

For product choose, in order to avoid influence from other factors, we should manipulate every factors which could influence participants’ perceptions. Firstly, we Choose reviews from same website, www.amzon.com with two of these products which have popularity in sales and have lots of reviews. One specific clothing and one specific brand of laptop keyboard have been chosen. For clothes, we choose ‘Kensie Women's Wool-Blend Cocoon Coat’, which has 23 customer reviews with 3.9 out of 5 stars in average ratings. From all the reviews, we pick one reviews with picture below which presented high relevant pictorial information here as our independent variable, as well as other four reviews with same rating as the one with picture. And change pictorial messages only of that reviews to three other versions: one is medium relevance picture which we put brand name of that coat under the verbal reviews, one we put a dog’s picture and no picture on the third one.

For electronic devices, we choose one keyboard which has more or less the same ratings as the clothes above with 4.2 out of 5 stars in average rating. It is ‘Logi CREATE

Backlit Keyboard Case with Smart Connector for iPad Pro (920-007824)’,we choose one review with pictorial messages presenting his or her new keyboard and other four reviews without picture but share the same ratings for product. Same as the other three different version of clothes, we also got three other version of pictorial messages of this keyboard, one is only having the brand logo of that keyboard, one is a ‘dog’, same dog as that in clothes’ manipulation, and one without picture.

Independent variable

Relevance

In this study, three levels of relevance were manipulated by changing content of pictorial messages and subjects were randomly assigned into four groups including one control group without picture presented in online reviews. Each group of subjects were involved in two categories of products which are clothes and Ipad keyboard. Participants who finished picking the helpful and trustworthy reviews would be asked to score the relevance between online products and pictorial messages by using 3-items on a five-point Likert scale with reference from McKinney, Vicki, Kanghyun Yoon, and Fatmeh "Mariam" Zahedi (2002). These items are ‘This picture in online reviews is informative to me.’; ‘This picture in online reviews do matter to me.’; ‘This picture in online reviews is valuable to me.’ For each type of product, items were reliable. Items in clothes has high reliability for relevance with Cronbach’s α=0.842; items keyboard has Cronbach’s α=0.864 and were all over 0.7 which means high reliability to scales. And all the scales range from -4 to 4.

In order to make it easier to analyze the relationship between independent variable and dependent variable, we coded relevance to dummy variable. Firstly, we needed to check if there was significant difference between three level of relevance by using ONE-WAY ANOVA. Scale means were computed for different items, and were coded as RVcTOT for clothes and as RVeTOT for keyboard. In terms of clothes, we found that there was a statistically significant effect of different relevance manipulation on respondents’ perception and on score to relevance, F(2,144)=67.62, p<0.005. Each group has significantly difference in mean, (group 1= 3.77, group 2=2.43, group 3=1.97). For electronic device, there was also a statistically significant effect of different relevance manipulation on respondents’ perception and on score to relevance, F(2,144)=45.88, p<0.005. Each group has significantly difference in mean, (group 1= 3.74, group 2=2.89, group 3=2.02).

As there is significant difference between each group of manipulation to relevance, we recode independent variables to dummy variables (Table 1):

Table 1. Dummy variables:

High relevance (group 1) 3

Low relevance (group 2) 2

No relevance (group 3) 1

No picture (group 4) 0

Dependent variable

After respondents were tested their involvement, they would see five online reviews followed be product’s details and are asked to choose one of the reviews they think more helpful and trustworthy. One of five reviews contains our manipulated variable, whether there is a picture contained in review. Each product followed by two question with five answers. In

order to investigate whether manipulated independent variable has influence on dependent variable, we code reviews had pictorial messages as ‘1’ and the other reviews were coded as ‘0’.

Moderating variable

Involvement

Participants were asked to score each item for testing their involvement to specific product. in this part, we use consumer involvement profile based on five antecedents of involvement. (Jean-noel Kapferer and Gilles Laurant, 1993). In addition to that, we used 5-point Likert-type response format to test this variable. Five items are 1. (interest): ‘I’m really interested in clothing/ electronic devices.’ 2. (pleasure): ‘I really enjoy buying clothes/ electronic devices.’ 3. (sign): ‘the clothes/ electronic devices that a person buys, says something about who they are.’ 4. (risk importance): ‘it doesn’t matter too much if one makes a mistake buying clothes/ electronic devices.’ 5. (error): ‘Choosing clothes/ electronic devices is rather difficult.’

However, for question 4, higher scale represents lower involvement which is inverted with other item, counter-indicative items namely. Then we recoded scales meanings in reverse-coded.

Propensity and proficiency to online shopping

In order to test hypothesis three, we tested respondents’ propensity to trust online shoppers as well as their proficiency to shop online. For each variable, we used three items to test with 5 Likert-type response format. Questions for propensity is based on article from

Gefen, David, and Paul A. Pavlou (2012). These three items are: 1. ‘I usually trust online sellers unless they give me a reason not to trust them.’ 2. ‘I generally give online sellers the benefits of the doubt.’ 3. ‘My typical approach is to trust online sellers until they proveweshould not trust them.’ Questions for proficiency in online shopping is based on article from (Kane, Gerald C., and Stephen P. Borgatti, 2011). These three items are: 1. ‘I find online shopping easy for me to shop.’ 2. ‘I am skillful at using internet to shopping.’ 3. ‘Using online shopping takes too much time from my normal duties.’

Based on Cronbach’s α, we confirmed high reliability for scale of propensity (Cronbach’s α=0.734, >0.7). But items for proficiency only got Cronbach’s α=0.543. After reading the statistics about ‘Cronbach’s α if item deleted’, we found that item was the largest reason for low reliability. If item three deleted, Cronbach’s α would be 0.759. So in next study, we deleted item three and got the mean based on other two items for this variable, proficiency namely.

3.2 Procedure

Respondents would receive one survey randomly from four different versions. The survey was designed by using Qualtrics in English. Firstly, they will read a thanks letter in order to increase respondents’ motivation to finish questions. Then there are some general questions about the frequency to buy products and online shopping. Next, they will score different items with instructions. After they start filling the survey, they cannot return back to previous question which they already finished. In addition to that, every question cannot be skipped and once quit, it cannot be entered again.

The pilot tests

Before send out large amount of questionnaires, we did a pilot tests to send the finished questionnaires to 30 subjects. As we had four group of respondents and then we had four versions of surveys, in pilot tests, we send out 6 surveys for each version. In the end, all of 30 surveys returned back with feedback by asking respondents about 1. Whether they could understand all of the questions they read including errors shown in survey; 2. Whether they feel questions are too complicated and annoying to finish it; 3. Whether they think the instructions is easy to follow (van Teijlingen & Hundley, 2002).

After collecting all the surveys in pilot tests, we gathered all the feedbacks and found that there were some problems on the design of surveys: 1. Some questions were opaque, especially items of ‘the propensity of trust online sellers’; 2. The interface for showing 5 reviews is too small to see clearly the words of reviews and only the pictures are clearly shown. This would make respondents get to know the aim of study and choose the one which they think would meet the right hypothesis. Except these two mean problems, there were no other critiques. In order to solve these problems, we changed the ways to express relevance to be more easy to understand but still keep the original meaning of these items. For the second problem, we made picture smaller in order to make it looks not that obvious. In addition to that, we only sent the surveys by mails to look for respondents who could finish the survey on computer with larger screen to see reviews shown in surveys more clearly.

3.3 Results

messages with different relevance exert on helpfulness and trustworthiness, we compare the number of how many people thought the manipulated review with or without pictorial message is helpful or trustworthy and the significance of the different between experiment group and control group by using Chi-Square. For moderation test, we use hieratical regression analysis. As two categories might have influence on the effect of relationships, we tested hypothesis separately based on two categories.

Test of H1: Pictorial messages role to influence perceived helpfulness

Hypothesis 1 postulates that highly related picture could increase helpfulness of online reviews. medium and low relevant pictures will not have positive influence on helpfulness and even backfire. This hypothesis was tested by investigating whether the changes in number of people who perceived manipulated review helpful between experiment group and control group is significant. Six separate Cross-Tab were run with different group number (high, medium, no) as independent variable and helpfulness as dependent variable under different product type conditions.

Group 1 & Group 4

Clothes: The Chi-square test compared with group 1 and group 4 revealed that, There is a

significant difference between control group 4 and group 1 in helpfulness with p=0.00, and if compared in row with percentage, the result showed that the percentage of people who thinks reviews with picture in manipulated is more helpful (48.8%) are larger than that of people who thinks the same review without picture (7.5%).

Electronic device: However, the Chi-Square test for this product type showed low

significance in the difference between group 1 and group 4, p=0.576. This means that even though the reviews with picture had that smaller number of people choose it to be the most helpful one( %within group 1=26.8; %within group 4=32.5), the difference cannot be proved because of the lower possibility to happen. Therefore, hypothesis 1 is rejected in this product type.

Group 2 & Group 4

Clothes: after run Chi-square test, we got that There is a significant difference between

control group and group 2 in helpfulness with p=0.022. From descriptive analysis, we found that the percentage of people who thinks reviews with picture in manipulated is more helpful (27%) are larger than that of people who thinks the same review without picture (7.5%). This result rejected part of hypothesis 1 because the medium related picture also could increase the helpfulness of online review in statistical significance.

Electronic device: The result is similar with that in clothes, it showed significant difference in

helpfulness (% within group 2=59.5%, compared with % within group 4= 32.5%), p=0.018. Therefore, hypothesis 1 partly rejected in this product type.

Group 3 & Group 4

Clothes: From the result, we found that there is also a significant difference between group 3

and group 4 in helpfulness (% within group 3=30.8 % VS % within group 4=7.5 %, p=0.008). The result rejected the prediction that no relevance pictures would backfire the helpfulness of online reviews.

Electronic devices: In this product type, there is no significant difference between two groups

p=0.869, which means that there is no significant difference after adding pictorial messages as independent variable. This result supported our hypothesis that no related pictures might not have influence on perceived helpfulness.

Test of H2: Pictorial messages role to influence perceived trustworthiness

Hypothesis 2 postulates that highly related picture could increase trustworthiness of online reviews. medium and low relevant pictures will not have positive influence on trustworthiness and even backfire. In order to test this hypothesis, we still compared three experiment groups with control group by running Cross-Tab.

Group 1 & Group 4

Clothes: There is a significant difference between control group 4 and group 1 in

trustworthiness with p=0.000, and the percentage of people who thinks reviews with picture in manipulated is more trustworthy (61.0%) are larger than that of people who thinks the same review without picture (12.5%). Therefore, H2 was supported.

Electronic devices: This product type showed inverted result with clothes. Firstly, there was

no big difference of percentage of people who chose manipulated review. Secondly, the significance for the difference is low p=0.530. Then H2 was partly rejected.

Group 2 & Group 4

Clothes: There is a significant difference between control group and group 2 in

trustworthiness with p=0.005, and from the percentage analysis, we got that the percentage of people who thinks reviews with picture in manipulated is more trustworthy (40.5%) are larger

than that of people who thinks the same review without picture (12.5%). Therefore, H2 was partly rejected.

Electronic devices: There is still significant difference between different relevance in

trustworthiness (% within group 2= 54.1%, vs % within group 4= 30.0%, p=0.032). This result is seam with that in clothes and it also rejected the hypothesis that medium related picture has low or inverted influence on trustworthiness.

Group 3 & Group 4

Clothes: the result showed that the difference is not significant (p=0.500), which support the

hypothesis that no relevant picture would not increase the trustworthiness of online reviews.

Electronic devices: it is same with the result shown in clothes, with low significance

(p=0.941), and this partly support hypothesis 2.

Test of H4: The moderator role of involvement

Hypothesis 4 suggests that customer’s involvement with products moderates the effect of pictorial messages on helpfulness and trustworthiness. To test this hypothesis, regression analysis was used to investigate the moderation roles of involvement. The regression analysis is for testing whether the coordination of moderator and independent variable could bring significant variance in dependent variable. If the above role can be convinced, the moderator role of involvement could be established (Baron and Kenny, 1986).

As there might be different effect between various categories, we run the test separately on clothes and electronic device.

Firstly, we run the correlation analysis in order to test whether there is linear correlation between different variables and the results are presented in Table 2.

Table 2. correlation matrix (clothes)

M SD 1 2 3 4 5 6 7 8 1.shopping frequency 3.2229 0.91005 1 2.online shopping frequency 2.82803 1.01387 .549** 1 3.helpfulness 0.2866 0.45363 -‐0.03 -‐0.004 1     4.trustworthiness 0.3312 0.47215 -‐0.05 -‐0.041 .452** 1 5.involvement 3.1452 0.64395 .583** .353** -‐0.064 0.005 1 6.propensity 2.8747 0.76926 .223** .230** -‐0.154 -‐0.12 0.101 1 7.proficiency 3.5127 0.96068 .239** .358** -‐0.119 -‐0.15 .194* .226** 1 8.relevance 1.5032 1.13581 0.071 0.148 .303** .404** 0.071 0.055 -‐0.071 1 ** Correlation is significant at the 0.01 level (2-‐tailed).

          * Correlation is significant at the 0.05 level (2-‐tailed).

From correlation matrix, we can predict that trustworthiness and helpfulness,

trustworthiness and relevance, helpfulness and relevance are significantly linear correlated. But involvement, propensity and proficiency are not linear correlated with our dependent variables. In order to test whether there are other effect that these variables might be involved, we run regression analysis to test the moderation effect of these variables.

Secondly, we recode our independent variable to dummy variable for further analysis and the result shown in the following table (Table 3):

Table 3. Dummy variables:

relevance X1 X2 X3

1

1

0

0

2

0

1

0

3

0

0

1

After centralize the independent and moderation variable, we build the hieratical regression model:

y=b1x1+b2x2+b3x3+cm+e 1) y=b1x1+b2x2+b3x3+cm+c1mx1+c2mx2+c3mx3+e 2)

m represents moderator, c means the possible coefficient of moderation effect. We use IeM1, IeM2, IeM3 to represent the coordinated relations.

Involvement and helpfulness: Running the hieratical regression model, we got the result that

for model 1, r2_{=0.108. After adding the moderation variables into the model, r}2_{=0.129. The}

result demonstrated that the model 2 could explain 12.9% of the variance, which is only 2% higher than the original model. However, Regarding the coefficient table, all the moderation effect is low significance, with p=0.727, 0.545 and 0.067 separately, which means the moderation effect does not exist from statistical point of view. Therefore, hypothesis 4 was partly rejected.

Involvement and trustworthiness: After running the regression analysis, the result still showed

low significance for three coefficients with p=0.401, 0.976 and 0.415 separately. Therefore, hypothesis 4 was rejected.

Electronic devices

Involvement and helpfulness: we still use the same model as above and create three new

variables, which are IeM1, IeM2, IeM3. Then we run hieratical analysis and we got that when trustworthiness is dependent variable, the predicted moderating effect is still not significant with p= 0.177, 0.695, 0.675 separately, which means hypothesis 4 was rejected.