How do valence, types of attributes mentioned and
recency influence the credibility and perceived
helpfulness of online reviews?
The moderating role of consumer skepticism
Alexander Robertson
Author: Alexander Robertson1
First Supervisor: Dr. J.A. Voerman2
Second Supervisor: Dr. J.C. Hoekstra3
Word count: 15447
Date: 22nd
June 2015
1 Alexander Robertson is an Msc Marketing student at the Faculty of Economics and Business,
2 Dr. Liane Voerman is a Senior Lecturer in the fields of Business and Marketing, Faculty of
Economics and Business, University of Groningen, The Netherlands
3 Dr. Janny Hoekstra is a Senior Lecturer in the fields of Business and Marketing, Faculty of
Economics and Business, University of Groningen, The Netherlands
MANAGEMENT SUMMARY
In this research two different studies are conducted. In the first study, the dependent variable is credibility of the online reviews. In the second study, the dependent variable is perceived helpfulness. The aim of this empirical research is to understand whether there is a potential relationship between three independent variables on these dependent variables. The independent variables used are recency, valence and the attributes mentioned in the online consumer review (OCR). In addition to this, consumer skepticism is added as a moderator of this effect.
These variables are tested, by implementing a 2 (positive vs. negative) x 2 (outdated vs. recent) x 2(search vs. experience) between-participants design. Overall, 222 respondents are collected and allowed to participate in an online survey. This survey asks questions to measure both credibility and helpfulness of the OCR and also background questions concerning gender, age and nationality as well as degree of usage of OCR’s.
This study presents some interesting results. Firstly, the OCR attributes are found to have a significant effect on the degree of credibility and helpfulness of the OCR. On the other hand, the valence of the review did not present any significant effect on the credibility or helpfulness of the OCR. Furthermore, the effect of the recency of the review was also tested and was also not significant for the credibility and helpfulness. Lastly, the level of consumer skepticism of consumer skepticism was only a moderator of the effect of OCR-attributes. For the valence and the recency results are not significant.
ACKNOWLEDGEMENTS
Finally the time has come. This thesis represents the last hurdle of my academic career and brings my life as a student to an end. The extensive period spent studying has allowed to me to grow both as a person and in my capabilities. I learnt not only that I should be confident in myself but also that there is always a solution to any problem that may arise. My university career commenced 5 years ago in a small city in the North of Italy and ends here in Groningen. I will forever be proud to have finished my studies not only in such a high quality University but also in such a beautiful city.
The first people I must thank are undoubtedly my family, for their relentless support and for always believing in me even when I did not. In addition to this, they allowed me to pursue this fantastic opportunity of studying at the University of Groningen for which I will be eternally grateful. Even though they are far away I felt like they were with me everyday in my journey. Throughout this experience I have also had the pleasure of meeting some fantastic people who will always be part of my life. For this reason, I would also like to thank my friends for getting me through the more stressful periods and allowing me to achieve great results.
Last but not least, my last acknowledgement must go to Dr. Liane Voerman who supervised my thesis and without which this would not have been possible. Her help and invaluable support during our numerous meetings were critical for me in reaching this goal. She also managed to transmit her endless enthusiasm to me and had me walking away from our meetings with a smile even during the most demanding times. I will always be thankful for the way she taught me to think in a positive way and to never give up at the first difficulty I encountered.
Now, it is time for a new and exciting chapter of my life to begin.
.
To my family and friends,
TABLE OF CONTENTS 1. INTRODUCTION
1.1 EWOM vs. WOM and Online Consumer reviews………...7
1.2 Online Consumer reviews………..7
1.3 Why are credibility and helpfulness of an OCR important?...9
1.4 Antecedents of OCR credibility and perceived helpfulness………...11
1.4.1 Types of Attributes as an antecedent of credibility and helpfulness…………11
1.4.2 Valence as an antecedent of credibility and helpfulness……….11
1.4.3 Recency as an antecedent of credibility and helpfulness………12
1.4.4 Consumer Skepticism as a moderator ….. ……….12
1.5 Research Question………. 13
1.6 Theoretical relevance and uniqueness of the thesis……….. 14
1.7 Structure of the thesis……….15
2. THEORETICAL BACKGROUND 2.1 Effects of attributes in OCR’s on credibility and helpfulness………... 16
2.2 Effect of valence on credibility and helpfulness of OCR’s………... 16
2.3 Effect of recency on credibility and helpfulness of OCR’s……….. 17
2.4 Moderator: Consumer Skepticism………..21
2.5 Conceptual Model………. 24
3. METHODOLOGY 3.1 Research design and number of participants………..26
3.2 Sample and procedure………... 26
3.3 Operationalization and reliability of measures……….. 27
3.3.1 Operationalization……….. 29
3.3.2 Reliability of scales……… 29
3.4 Manipulation of independent variables………..32
3.5 Plan of analysis……….. 34
4. RESULTS 4.1 Descriptives of the sample……… 36
4.2 Model Selection: Study 1……….. 39
4.3 Hypothesis validation: Study 1………. 45
5. CONCLUSION 5.1 Discussion………. 47
5.2 Limitations and future research……… 48
References………... 50
1.INTRODUCTION
In recent times, simultaneously occurring with the rapid explosion of the Internet, searching for products information online has emerged as a popular endeavor amongst consumers (Horrigan, 2008). For consumers to make a rational and good choice as possible they often exert a lot of time and effort in searching for information about products and services they may want to acquire in the future (Goldsmith & Horowitz, 2006). Specifically, online consumer reviews (OCR’s) are developing as a crucial source of information for latent product purchasers (Dellarocas, 2003). Online consumer reviews are frequently defined as “peer-generated product evaluation[s] posted on company or third-party web sites” (Mudambi & Schuff, 2010, p. 186). Extensive research has demonstrated that every fourth customer first analyses other customers’ views about a product prior to purchasing it (comScore 2010; Wagner and Wiehenbrauk 2014). At this moment in time, Amazon.com has approximately 10 million consumer reviews covering all its product categories and this number is rising exponentially (Park and Han, 2007). According to a recent report by market research firm Nielsen (2012), 70% of consumers confirmed they have confidence and trust in online product reviews. However, the credibility and helpfulness of these reviews remains debatable and a major talking point and thus will be analyzed in more depth in this thesis.
1.1 EWOM vs. WOM and Online Consumer Reviews
While e-WOM communication possesses a few characteristics jointly with traditional WOM communication, it diverges from traditional WOM in several aspects (Cheung, 2009).
1.2 Online Consumer Reviews
Online consumer reviews (OCRs) are considered by many to be one of the most significant and influential components of e-WOM (Schindler & Bickart, 2005). In stark contrast to traditional WOM communications that are diffused through one's near contacts and strong ties between people, the Internet implies the connection of a large amount of unacquainted users and allows them to express feelings and views without unveiling their true identities. As a result of this, the genuineness and authenticity of an OCR review is unclear, making its believability a critical factor of whether or not the actual review shall be accepted or rejected by the reader (Qiu, Pang & Lim, 2012).
1.3 Why are credibility and helpfulness of an OCR important?
As was mentioned previously, credibility has often been referred to as the extent to which a receiver of information considers it to be believable (Eisend, 2006). Credibility has also been described as the degree to which, a portion of information is perceived as genuine and valid, by the reader (Tseng & Fogg, 1999). Credibility is also a vital factor as perceptions of believability can enhance a receiver’s objective to adjust his or her outlook based solely on the information presented (Hovland, Janis & Kelley, 1953). A conventional theory about uncertainty reduction insinuates that individuals who are exposed to situations of uncertainty or ambiguity will try to implement strategies available that can reduce the uncertainty as much as possible (Berger and Calabrese, 1975). Individuals will endeavor to lessen uncertainty by searching for credible information that can be relevant and useful for them in their decision-making (Jacoby et al. 2014).
think the incoming information is credible, they will have more confidence to adopt the e-WOM comments and use them for making purchase decisions. (Sussman and Siegal, 2003). Multiple researches have already established the positive relationship that exists between information credibility and adoption. One empirical example from McKnight et al‘s (2002) study shows this. They demonstrated the positive effect of receiver’s perceived information believability on inclination to accept the information of a review.
Nevertheless, various advocates have asserted that online information is higher in perceived credibility than information from other more traditional media (Park et al., 2007; Senecal and Nantel, 2004). The reasoning behind this is that experienced individuals usually post information and they can therefore be considered as credible sources of information (Gretzel et al., 2007; Park et al., 2007; Wilson and Sherrell, 1993). Having said this, others can argue that the online reviews can be posted by any individual regardless of their experience and therefore is lower in credibility than other types of information sources (Gretzel et al., 2006; Litvin et al., 2008; Mack et al., 2008; Magnini, 2011). Moreover, OCR’s can also provide a clear opportunity for dishonesty as they can be completed without having to reveal specific author information such as a real identity or photo (Fogg et al., 2001) and are produced by a source that has a minimal relationship with the receivers of the message (Schindler & Bickart, 2005). This absence of cues regarding the source of the review (Jin et al., 2002; Smith et al., 2005) presents readers of reviews with a challenging task. They are challenged to distinguish between deceptive reviews and believable ones.
find the most helpful online review and makes the consumer browsing process much more efficient. For these reasons, the author has chosen credibility and perceived helpfulness of OCR’s as dependent variables in the following empirical analysis.
1.4 Antecedents of OCR Credibility and Perceived Helpfulness
In this thesis, the focus will be on the type of attributes mentioned in the OCR, the valence and the recency of a review as well as the moderation of consumer skepticism.
1.4.1 Type of attributes as an antecedent of Credibility and Helpfulness
The first potential antecedent and independent variable worth mentioning is the attributes of a review. In this paper, the author will focus primarily on the distinction of Nelson (1974) between search and experience attributes. Although many products involve a blend of search and experience attributes, the classification of search and experience goods continues to be applicable and generally accepted (Huang et al. 2009). Darby and Karni (1973) added credence attributes, defining them as those attributes that the consumer cannot verify even after use, however these will not be tackled in this paper. Products therefore, can be described as existing along a continuum from pure search goods to pure experience goods (Mudambi et al. 2010). The reasoning behind the usage of this variable is that all reviews either contains search attributes or experience attributes and thus have should have varying impacts on credibility and helpfulness.
1.4.2 Valence as an antecedent of Credibility and Helpfulness
1.4.3 Recency as an antecedent of Credibility and Helpfulness
The third antecedent of credibility that will be investigated in more depth is the factor of recency. Recency refers to whether the review has been posted recently or whether is it outdated. Electronic word of mouth and OCRs typically all possess a unique time feature that has seldom been fully addressed by academics. In comparison with traditional offline word-of-mouth, the precise posting date of every online review is documented and is readily available to readers to see. However, the empirical work that has studied how the time dimension of online reviews impacts consumer decisions is limited. In particular, by considering the other dimensions like attributes and valence of the online reviews as constant, how would the supplementary information of posting time and date change the impact of the online reviews (Jin, Hu & He, 2014).
While countless online retailers (e.g., Amazon.com), remain under the common supposition that outdated product reviews are less beneficial in shaping consumers’ decisions, the exact nature of the correlation between a review’s posting time and its influence on consumer decision is doubtful (Jin, Hu & He, 2014). For instance, one could read about how other clients who visited a hotel recently (e.g., five days ago) assessed the hotel or could also with very little effort just click the “scroll down” button to compare that with information regarding the hotel from individuals who lodged there many months earlier. Therefore, online shoppers, rivaled against their offline equivalents, can easily search specific outdated product reviews with little extra effort. Subsequently, in comparison with traditional word-of-mouth, the effectiveness of older or outdated OCRs in shaping consumer preferences may reveal itself as stronger in the context of online shopping (Jin, Hu & He, 2014). For all the aforementioned reasons, this thesis will incorporate recency as a factor than influence the credibility and helpfulness of an online review.
1.4.4 Consumer Skepticism as a Moderator
defined as the tendency toward disbelief (Obermiller & Spangenberg, 1998). Skepticism implies that a portion of consumers may not be inclined to believe online reviews in any case, no matter what the posting date, the valence or the attributes contained within the review. Although OCR’s are a great tool for gaining access to large quantities of information about products there are a few drawbacks. One of the most important is that the information, is posted by strangers, whom the consumers have never encountered which will inevitably cast doubt over their authenticity (Sher & Lee, 2009). For this reason, online reviews have been the subject of considerable criticism (Sher & Lee, 2009). In fact, Chatterjee (2001) even goes as far as suggesting that there is a portion of online surfers that get paid to post product reviews. For all the reasons mentioned, consumer skepticism will be employed as a potential moderator of the effect of the independent variables on credibility and perceived helpfulness of the OCR.
1.5 Research Question
The following paper aspires to answer the ensuing research question:
How do valence of a review, the recency and the type of attributes mentioned in the OCR affect the credibility and the perceived helpfulness? And do consumer characteristics such as consumer skepticism moderate this effect?
In particular, this research aims to answer the following subset of questions:
o Does the credibility or helpfulness of the OCR change for search and experience attributes?
o Does the effect on the credibility or helpfulness differ if the review is positive versus negative?
o Does consumer skepticism strengthen or lessen the impact of these variables on credibility or helpfulness?
1.6 Theoretical relevance and uniqueness of the thesis
This paper contributes to extending the existing research on the broad topic of OCR’s. This research aims to provide new insights into limited research regarding the effects of recency on the credibility and perceived helpfulness of OCR’s and whether the effect could differ when considering search attributes or experience attributes. Furthermore, the valence of the consumer review will be taken into account. Moreover, characteristics of the consumer, namely consumer skepticism will be implemented as a potential moderator of this effect. Currently, online product reviews have emerged as an imperative tool for consumers and consequently also for online retailers who want to appeal to customers and for consumer retention (Grewal and Levy, 2007). Therefore, it is vital that online retailers do not turn a blind eye to the undeniable paybacks of product reviews, and it is therefore of utmost importance to fully comprehend the factors that determine how consumers react to product and service reviews and how they influence their behavior (Jin, Hu & He 2014). Moreover, worth mentioning is that certain types of e-WOM messages i.e. online consumer reviews in Amazon.com are also manageable because marketers can choose whether to permit consumer reviews to be displayed or not. If the decision is made to show the reviews marketers can then proceed to suggesting a precise review format in order to direct and encourage consumers to share their opinions in the way they desire. Therefore, it considerably easier for marketers to implement marketing strategies for e-WOM than it is for traditional WOM (Park & Kim, 2008). In the following the thesis, the industry that will be examined empirically is the travel industry.
1.7 Structure of the Thesis
2. THEORETICAL BACKGROUND
In the following chapter a thorough literature review will be performed on all of the variables included in this thesis. From this literature, the hypothesis will be developed and finally two conceptual models will be graphically represented.
2.1 Effects of Attributes in OCR’s on Credibility and Helpfulness
Search attribute information can easily be obtained from secondhand sources such as promotions and word of mouth from acquaintances. Thus, it is not necessary to purchase or try the product beforehand to know about it (Wright & Lynch, 1995). For a better understanding some examples of search attributes could range from the color of a table or the number of calories per portion of food and price of a hotel room. On the contrary, experience attributes can be confirmed only by use of the product albeit limited, as they are a matter of personal experience. Examples of experience attributes can be anything from the friendliness of staff in a hotel to the exact taste of a chocolate bar (Wright & Lynch, 1995).
Finally, it could also be debated that consumers' way of thinking about the attributes of an electronic consumer review could potentially be a crucial factor that could affect the review's perceived credibility (Qiu, Pang and Kim, 2012). Specifically, when an e-WOM review happens to be accredited to non-product-related motives, then consumers will have considerably less confidence in the reviewer's integrity and honesty (Qiu, Pang and Kim, 2012). Therefore, based on the aforementioned literature the author develops the following hypotheses:
H1A: The presence of search attribute information in an OCR produces higher perceived credibility than the presence of merely experience attributes.
When considering search goods, extreme and moderate reviews can be equally helpful. This is in stark contrast to experience goods where only moderate reviews are perceived as helpful (Mudambi & Schuff, 2010). As was mentioned previously, reviews containing search attributes are more likely to consider tangible aspects of the product or service. As objective claims about attributes are more easily validated claims for search goods can be perceived as more helpful (Mudambi & Schuff, 2010). Thus, the following assumption can be generated:
H1B: The presence of search attribute information in an OCR produces higher perceived helpfulness, than the presence of merely experience attributes.
2.2 Effect of Valence on Credibility and Helpfulness of OCR’s
into a product of much lower quality more easily than encouraging online reviews help them label the offering into a higher quality product (Bone, 1995; Herr et al., 1991). Furthermore, in the environment of consumer reviews placed on the Internet, Chiou and Chang (2003) have proven that negative consumer evaluations regarding a brand did as a matter of fact, impact brand valuation negatively with respect to neutral reviews, while on the other hand positive and encouraging reviews did not affect brand evaluation whatsoever. Overall, results strongly reinforce the fact that negative (compared to positive) electronic word of mouth will have a larger positive effect on attitude toward the brand and the website and also lead to a higher level of credibility (Lee, Rodgers and Kim, 2009). Therefore, based on the literature we can posit the following:
H2A: An online consumer review with negative information will have a higher degree of perceived credibility compared to one containing solely positive information
The preceding studies of valence of the review on its perceived helpfulness have produced mixed results (Schindler & Bickert, 2005). On one hand, various academics have found that OCR’s that contain a majority of positive evaluations could be regarded as more helpful to the reader in the sense that it will show the product will have to be considered further. Having said this, Sen and Lerman (2007) in their study found that negative reviews could be perceived as more accurate and helpful compared to a positive one. As mentioned previously, extensive research has found that negative information has more value to the reader in comparison with positive information and thus is perceived as more helpful in decision-making (Skowronski & Carlston, 1987). Due to the fact one’s social environment is mainly filled by a greater number of positive cues, negative cues are perceived as counter normative (Feldman, 1966). For all the reasons the following is hypothesized:
H2B: An online consumer review with negative information will have a higher degree of perceived helpfulness compared to one containing solely positive information
and Koo (2012) demonstrated that when the online reviews were of negative valence and they contained objective information (in our case search attributes) then there was a positive impact on credibility and hence also adoption of the online review (Lee & Koo, 2012). As was stated previously, the assumption made is that online consumer reviews containing purely search attributes will have a higher degree of credibility than those with experience attributes. Therefore, now we can consider this when the review is negative if this statement still holds or if there are any differences compared to when the valence was not taken into consideration. Thus, the following hypothesis has been elaborated:
H3A: The positive effect of search attributes over experience attributes on credibility will be stronger if the valence is negative compared to positive
As was stated earlier search attribute information will have a higher degree of perceived helpfulness compared to experience attribute information. Also well-established is the fact that negative information is more helpful than positive information due to the “negativity bias” (Skowronski & Carlston, 1985). Inevitably, customers are also much more interested in what is negative than by positive arguments. So, the following can be hypothesized:
H3B: The positive effect of search attributes over experiences attributes on perceived helpfulness will be stronger if the valence is negative compared to positive
2.3 Effect of Recency on Credibility and Helpfulness of OCR’s
today. Having said this, online consumer reviews, do not rely on recipient memory as the information about the product is ever present and online buyers can simply browse through innumerable product reviews posted by other consumers at different times (Jin, Hu & He, 2014).
However, findings suggest that any answers regarding the effectiveness of recent versus outdated reviews in influencing consumer decisions require a more in depth analysis. When considering the traditional word-of-mouth, an individual who receives a message must rely on his or her memory to remember the content of the message stemming from an earlier period, which in turn can lead to problems such as message decay. In electronic word of mouth however the same does not occur. This is mainly due to the fact that out dated and recent online reviews, are processed by the reader virtually at the same time. Thus, memory and newness have very little time to deteriorate (Jin, Hu & He, 2014). When considering online reviews, individuals are more likely to believe a more recent review than an outdated one. Based on the literature the author develops the following hypothesis:
H4A: Online consumer reviews that have been posted recently will have a higher degree of perceived credibility compared to reviews that are outdated
Reviews that are more recent will have a higher degree of perceived helpfulness because even if customers can easily scroll down to see older reviews and memory does not have time to decay consumers will make the assumption that the recent review is more helpful because it is based on current information that is unlikely to have had the time change (Jin. Hu & He, 2014). Thus, the following is postulated: H4B: Online consumer reviews that have been posted recently will have a higher degree of perceived helpfulness compared to reviews that are outdated
As was stated previously, reviews containing search attribute information will produce a higher degree of perceived credibility of the review as the review is more objective compared to a review containing experience attributes (Mudambi et al. 2010). Thus, when considering a possible interaction between recency and the attributes of the OCR the following hypothesis is formed:
information will have a higher level of credibility than older reviews containing the same information
Previously, it was stated that OCR’s that contain search attributes have a higher degree of helpfulness than those containing solely search attributes due to their objectiveness. Hence, when considering an OCR that is both recent and contains search attribute information it should have a higher degree of helpfulness due to the fact that the objective information that has been posted recently will be more helpful for the consumer to make the association that this is unlikely to have changed. Therefore. The following is hypothesized:
H5B: Online consumer reviews that are recent and contain search attribute information will have a higher level of perceived helpfulness than older reviews containing the same information
It is also worth analyzing a potential interaction between the valence of the review and the posting date. Extensive research has already proved the fact that negative information is more credibly due to the negativity effect (Lee, Rodgers & Kim, 2009). However, having said this, it may be interesting to know whether this statement can still hold when comparing outdated reviews and recent ones. To this end, the following hypothesis was generated:
H6A: Online consumer reviews that are negative and have been posted recently will have a higher level of credibility than negative reviews that are outdated
As was established earlier, negative reviews have a higher degree of perceived helpfulness compared to positive ones. However, online reviews that are recent and negative will be perceived as more helpful because they are more current and based on experience from a short time ago (Jin, Hu & He, 2014).
H6B: Online consumer reviews that are negative and have been posted recently will have a higher level of perceived helpfulness than negative reviews that are outdated
2.4 Moderator: Consumer Skepticism
research that individuals might have distinctive levels of skepticism. In other words, the tendency to be skeptical may depend on factors like product type and attributes of the OCR. Consumer skepticism reflects a conviction held strongly by a consumer. This tendency to be suspicious about information is a firm belief, which reproduces a vision of consumer’s implicit views regarding the functioning of the marketplace (Sher & Lee, 2009). Furthermore, Nelson's notorious (1974) framework indicates that when consumers do not possess prior knowledge regarding a product or service they will be considerably more skeptical and unconvinced when presented with claims about experience attributes than claims about search attributes (Nelson, 1974). In other words, consumers are more likely to be skeptical when they are looking at subjective information than objective information. Following this, the following hypothesis can be developed:
H7A: An online consumer review containing solely search attributes information leads to a higher degree of perceived credibility than experience attributes but the positive impact is lessened when the consumer is skeptical about online reviews
An OCR containing search attributes will be considered far more helpful than one containing solely experience attributes. However, this well established positive effect would be lessened due to the fact that skeptical consumers are impossible to persuade no matter what is the content of the review (Sher & Lee, 2009). Thus, the following can be hypothesized:
H7B: An online consumer review containing solely search attributes information leads to a higher degree of perceived helpfulness than experience attributes but the positive impact is lessened when the consumer is skeptical about online reviews
H8A: A negative online consumer review will have a higher degree of credibility than a positive one but the effect is lessened when the consumer is skeptical about consumer reviews
A negative review will have a higher degree of perceived helpfulness compared to a positive one (Sen & Lermann, 2007). Nevertheless, a skeptical consumer may be unwilling to accept the information regardless of whether it is positive or negative. Hence, the following can be hypothesized:
H8B: A negative online consumer review will have a higher degree of perceived helpfulness than a positive one but the effect is lessened when the consumer is skeptical about consumer reviews
As was mentioned previously, a skeptical consumer will be unlikely to be convinced no matter what the posting date of the review (Sher & Lee, 2009). Thus, the following assumption can be made. The positive effect a recent review should have on credibility compared to an outdated one will be considerably less when the consumer is skeptical and does not give any importance to such factors. So the following hypothesis was developed:
H9A: A recent review has a higher degree of credibility than an outdated one but the impact is lessened when the consumer is skeptical about consumer reviews
As was aforementioned, it can be assumed that recent reviews have a higher degree of helpfulness compared to outdated ones. However, it can also be said that this positive effect on helpfulness will decrease when the reader of the review is skeptical about OCR’s in case. For these reasons the following hypothesis was developed:
2.5 Conceptual Model
On the basis of the theoretical framework mentioned above, the following conceptual models for Study 1 with credibility as a dependent variable and Study 2 with perceived helpfulness as a dependent variable are developed. In these conceptual models, the three explanatory variables that are expected to have an effect on credibility and perceived helpfulness are graphically represented as well as the moderator of consumer skepticism.
Study 1
Study 2
3. METHODOLOGY
In the following chapter, firstly the research design and the description of the participants will be presented. Subsequently, the method used for data collection will be explained. Then, the procedure of operationalization of all of the variables included in this analysis will be clarified. Afterwards, the results of the manipulation check will be illustrated followed by a preparation of the data. Finally, the plan of analysis will be introduced.
3.1 Research design and number of participants
As was aforementioned, the principle objective of this thesis is to gain deeper insights into the potential effects of the independent variables on the credibility and helpfulness of the OCR. With this in mind, an experimental design will be set up using valence (Positive/Negative), attributes of the OCR (search/experience) and recency (recent/outdated) as experimental variables. Furthermore, the potential moderating effect of consumer skepticism will also be assessed.
Table 3.1. Overview of Conditions Positive Negative Search Recent Outdated Recent Outdated Experience Recent Outdated Recent Outdated For all possible conditions, the amount of respondents required for the experiment to be considered as reliable and valid is 30. Thus, 240 people must complete the survey.
3.2 Sample and Procedure
The majority of the respondents collected for this survey were of Dutch or Italian nationality, due to time constraints for gathering data. Moreover, the data was collected with the aid of an online survey. The reason for choosing an online survey was for made for three main reasons. Firstly, this digital format allowed for a more accurate organization of the data. Secondly, this allowed the respondents to feel no time pressure whilst filling in the survey, as they could take their time and even close the survey and start it again at a later time, if needed. Finally, the survey allowed the respondents to save a considerable amount of time for the fulfillment of the entire questionnaire.
The survey was organized as follows. First of all, participants would read a simple welcome message informing them about the topic of this survey and encouraging them to read the text and the questions as carefully as possible. By proceeding through the survey, the respondents then encountered a graphic representation of an OCR regarding a Hotel. The text of the OCR was entirely self-made and was written and displayed in a simplistic fashion with just the date to avoid confusion.
Figure 3.1. Search attributes-Positive-Outdated condition
Figure 3.2. Experience attributes- Negative- Recent condition
3.3 Operationalization and reliability of measures
In order to proceed with the empirical analysis it is critical to operationalize all the concepts mentioned using different scales, which allow their measurement. An overview of this is presented in table 3.2.
3.3.1 Operationalization
The first dependent variable is review credibility, which was operationalized previously in literature by Cheung et al. 2012. The authors proposed a measurement with four items on a 7-point Likert Scale. The four items include the extent to which the review is believable, factual, accurate and credible. This scale is anchored with “strongly agree” and “strongly disagree”.
The second dependent variable is perceived helpfulness. For this analysis the operationalization conceptualized by Wu (2013) was adopted. The author suggests three items measured on a 7-point Likert scale anchored with “not at all” to “very”. Moreover, the three items refer to how informative, useful and helpful the OCR is. The moderator of consumer skepticism was measured using a scale first conceptualized by Obermiller and Spangenberg (1998). This scale encompasses five different items and is measured on a 7-point metric scale anchored with “strongly disagree” and “strongly agree”. The items include statements, which question to what degree customer ratings can be dependable, truthful, informative and a reliable source of information.
The manipulation check for recency was performed by firstly asking respondents a control question to determine whether or not they noticed the date of the review. If the respondent did notice the date then they were allowed to answer the following question: “ how recent do you perceive the recency of the review?”. This was anchored from “not at all” to “very” on a 7-point Likert scale.
and “positive”.
Finally, the manipulation check for the attributes mentioned in the OCR was executed by asking respondents to rate their ability to assess the elements that were included in the OCR presented to them prior to visiting the Hotel. Hence, in particular respondents were asked how price, room size, food quality, behavior of staff and location of the Hotel could be evaluated. This was measured on a 7-point scale that was anchored with “strongly disagree” and “strongly agree”.
Table 3.2 Operationalization table
Concept Item Scale Cronbach
Alpha Dependent Variable:
Review credibility Source: Cheung et al.
2012
1) I think the review is believable
2) I think the review is credible
3) I think the review is factual
4) I think the review is accurate 7 point Likert Scale anchored with: Strongly disagree- Strongly agree 0,902 Dependent Variable: Perceived helpfulness Source: Wu (2013)
1) I think the review is informative
2) I think the review is useful 3) I think the review is
helpful 7 point Likert scale anchored with: Strongly disagree- Strongly agree 0,878 Moderator: Emotional Skepticism Source: Obermiller and
Spangenberg (1998)
1) We can depend on getting the truth in most customer reviews
2) Customer reviews’ aim is to inform the customer 3) I believe customer reviews
are informative
4) Customer reviews are generally truthful
Concept Item Scale Cronbach Alpha Independent
Variable:
Recency Source: Jin, Hu and
He (2014)
How did you perceive the recency?
7-Point Bipolar scale anchored with: Not at all-Very N/A Independent Variable: Valence Source: Koo (2015)
How did you perceive the review?
7-point Bipolar scale anchored with: Negative- Positive N/A Independent variable: Attributes mentioned in the OCR Source: Nelson (1970)
Before visiting a Hotel I am able to evaluate the following aspects:
1) Price 2) Behavior of staff 3) Location of Hotel 4) Food quality 5) Room size 7-point Likert scale anchored with: Strongly disagree-Strongly agree N/A 3.3.2 Reliability of scales
In order to determine whether the constructs used for this analysis were reliable, it was necessary to test unidimensionality by means of a factor analysis. The factor analysis was conducted using a Varimax rotation method. Following this, the tests for internal consistency were carried out.
Before proceeding with a factor analysis, there are a few preliminary conditions that have to be met. In fact, if the Kaiser-Meyer-Olkin (KMO) statistic presents a small number this indicated that the correlations between variables are not explained by other variables. In this case, factor analysis may not be appropriate. In general, if the KMO presents a value greater than 0.5 and the Bartlett’s test of sphericity is significant than a factor analysis can be considered an appropriate technique (Malhotra, 2009).
Following this, a factor analysis was run for each individual concept (see Appendix B, Table 2 for output). At this point, a reliability analysis was conducted.
The results obtained from the reliability analysis show a Cronbach Alpha of 0,902 for review credibility while deleting the items did not increase the Alpha. Thus, all items can be used for the analysis. The scale for helpfulness gave a Cronbach Alpha of 0,878, which also demonstrates a high reliability and all items could be used for the analysis, as deleting one would reduce the reliability. For the moderator of emotional skepticism, the Cronbach Alpha with all of the items produced a value of 0,759. However, deleting the item named “Customer reviews’ aim is to inform the customer” means that the reliability will increase to 0,792. For this reason, this item will be omitted from the research to obtain as high reliability as possible. For the other concepts, being measured on a one-item scale, a reliability analysis is not applicable.
Once all the measures for internal consistency have been completed using the Cronbach Alpha, the next step entails calculating the average of all of the items contained within one variable in order to be left with one single construct. Once, this has been achieved it is possible to proceed with a check of the manipulation of the independent variables.
3.4 Manipulation of Independent Variables
The next step involves conducting a manipulation check on the independent variables to determine the level of validity of the model. This means that there must be significant differences between the conditions participants have been randomly assigned to. To this end, the table 3.3 below provides a clear overview of the full results obtained having manipulated valence, OCR-attributes and recency.
Mean Standard Dev. t-value p-value
Valence Positive 5,6607 1,47382
16,761 0,000
Recency Outdated 2,7625 2,08213
-10.223 0,000
Recent 5,8889 1,63203
Price Behavior Location Food quality Room size
Price Yes No Yes No
Behavior Yes No Yes
Location Yes No
Food quality Yes
Room size
Table 3.3 Overview of manipulation table
The first variable that was manipulated was valence. This was measured implementing a 7-point bipolar scale, with a minimum of 1 (negative) and a maximum of 7 (positive). In this case, the manipulation check produced a mean of 5,6607 for the positive condition and a mean of 2,1000 for the negative condition. This demonstrates that there is a significant difference between those exposed to the positive condition and those exposed to the negative condition. Both results produced a p-value of 0,000 meaning they are significant. Therefore, the manipulation of this independent variable was successful.
The second variable that was manipulated was recency. This concept was measured on a 7-point bipolar scale. Firstly, as mentioned previously respondents were asked whether they saw the review. It is worth mentioning that 68,5% of the respondents noticed the date of the review. Of the respondents that noticed the date of the review, the mean was 2,7625 for the outdated condition and 5,8889 for the recent condition. This implies that there is a clear difference between the two conditions. The results were significant, thus the manipulation was efficient.
significant differences between the three search attributes (location, price and room size) and the two experience attributes (food quality, behavior of staff). The table 3.3 presented indicates a “yes” when there was a significant difference between the two aspects and “no” when there is no significant difference. Thus, the manipulation of the attributes was also successful. The different means are presented below for the two conditions.
Search attributes: Price: 6,46, Location: 6,38, Room size: 5,34 Experience attributes: Behavior: 3,16 Food quality: 3,09
3.5 Plan of Analysis
First of all, once the factor analysis and the reliability analysis have been conducted, the next step will entail giving an overview of the sample for each condition with a description of the participants. Thereafter, taking into consideration background variables like age, gender and nationality will provide a more meaningful interpretation of the results.
Following this, the means will have to be compared in order to assess differences between the conditions. This implies comparing the means of the two valence conditions, the two OCR-attributes conditions and the two recency conditions separately by using a t-test. Once this has been conducted, a multiple regression will be performed firstly with OCR credibility as a dependent variable then with OCR helpfulness as a dependent variable. This multiple regression will have the following basic equation: Y(Credibility/Helpfulness)=B0+B1Recency+B2Valence+B3OCRattributes+B4Valence* Attributes+B5Recency*Valence+B6Recency*Attributes+B7Valence*Consumer Skepticism+B8Recency*Consumer.Skepticism+B9OCRattributes*Consumer Skepticism +
𝜀
interaction between recency and OCR-attributes will be added. Lastly, all the variables will be computed together for the final model. Therefore, it will be composed of valence, recency, OCR-attributes, interaction between valence and attributes mentioned, interaction between recency and valence and interaction between recency and attributes mentioned also taking into account consumer skepticism.
Another crucial step when estimating the models is checking for multicollinearity. If models present a degree of multicollinearity then those models would have to be interpreted with some caution. Performing a median split on the continuous moderator would then solve this multicollinearity. The median split method is chosen because the dichotomization of the independent variables makes analysis easier to interpret (DeCoster & Gallucci, 2009). Once this has been solved, a more reliable interpretation of the results of the models is plausible. Then, it will be possible to evaluate which is the best model by analyzing the adjusted R2
4. RESULTS
In the following chapter, the final results of the research that was conducted will be presented and analyzed. Firstly, a brief representation of the descriptives of the sample will be depicted. Thereafter, the ANOVA tables will portray the results for all the models considered in this empirical research. Then the multiple regressions will be performed. Finally, there will be an overall assessment of whether the hypothesis that were generated are supported or rejected.
4.1 Descriptives of the sample
In total 238 respondents started the survey, however, of these 238 only 222 participants fully completed the survey. The results highlighted in Table 4.1 show a significant majority of the respondents were female compared to male with values of 55,4% and 44,6% respectively. Moreover, the average age of the sample is 25 ranging from 21 to 75. The participants in the survey were from a wide variety of nations; Italy (45,9%) and the Netherlands (13,1%) constituted the majority (see Appendix C, Table 1 for output).
Variable Results Gender Male 99 (44,6%) Female 123 (55,4%) Age (n=222) Mean 32 Range 21-75 Use of online consumer reviews (n=222) Mean 5.16 Standard Dev. 1,25 t-test 61,22
Table 4.1 Descriptives of the sample
In addition to this, an ANOVA was run in order to evaluate whether there were in fact any significant differences between the conditions in terms of their use of OCR’s. The As was mentioned previously, the overall mean is 5,16. The highest mean per condition produced a value of 5,61 while the lowest produced a value of 4,70. In addition, the F-stat was 1,145 and the p-value was 0,336 (See Appendix C, Table 2 for output).
Mean Cred. (St. Dev.) T-Cred. (Sig.) Mean Help. (St. dev.) T-Help. (Sig.) Recency Outdated 4,09 (1,32) -0.573 4,23 (1,48) 4,27 (1,62) -0,141 (0,888) Recent 4,22 (1,48) (0,573) Valence Positive 4,32 (1,33) 0,285 4,24 (1,49) 4,45 (1,49) -1,034 (0,302) Negative 4,27 (1,33) (0,776) Attributes Search 4,73 (1,04) 4,722 4,76 (1,24) 3,93 (1,60) 4,026 (0,000) Experience 3,92 (1,44) (0,000)
Table 4.3 Effects of IV’s on Credibility and Helpfulness
The results of the table 4.3 show that in fact recency and valence did not affect the perceived credibility or helpfulness of the review. Thus, the fact that the review was recent or outdated does not appear to change the degree of credibility or helpfulness. Also whether the review was positive or negative does not appear to influence the credibility and helpfulness of the review. However, the OCR-attributes produced a highly significant value. Thus, indicating that the attributes mentioned in the OCR do have an effect on the degree of credibility and helpfulness of the review.
Table 4.5 ANOVA for Effects on Credibility and Helpfulness per condition
The results clearly show that the experience conditions produced a lower degree of credibility compared to the search conditions which all produced a mean above 4,6071. For the recency, there is no significant difference between the recent and outdated condition in terms of credibility. Also, for valence the results are inconclusive and mixed.
As for credibility, even in the case in which helpfulness was used as the dependent variable, it is clear that the experience conditions have a significantly lower degree of helpfulness compared to the search conditions. Again, there is no significant difference between the positive and negative conditions and also for the recent and outdated conditions results are mixed.
4.2 Model Selection Study 1: Credibility
At this point, the next step consists of testing the models using a multiple regression analysis. As was mentioned previously, in Study 1 the dependent variable used is review credibility. However, it is also worth mentioning that particular attention should be placed on which of the models present a high degree of multicollinearity. This implies looking at the Variance Inflation Factors (VIF) by selecting the collinearity diagnostic. If the VIF score produces a value between 4 and 10 then the
level of multicollinearity is considered to be moderate. However, if the VIF is greater than 10, then multicollinearity is very strong and the models should be interpreted with some caution (Myers, 1990).
Model 1 2 3 4 5 (Constant) 4,749*** 4,624*** 4,638*** 4,638*** 3,728*** Valence -0,002 n.s. 0,096 n.s. 0,086 n.s. 0,086 n.s. 0,405 n.s. Recency -0,014 n.s. -0,012 n.s. -0,022 n.s. -0,022 n.s. -0,054 n.s. Attributes -0,303*** -0,213** -0,213 ** -0,213** -1,493*** Valence*Attributes -0,165 n.s. -0,165 n.s. -0,165 n.s. -0,102 n.s. Recency*Valence 0,018 n.s. 0,018 n.s. 0,040 n.s. Recency*Attributes 0,000 n.s. 0,010 n.s. Valence*Consumer Skepticism -0,331 n.s. Recency*Consumer Skepticism 0,018 n.s. Attributes*Consumer Skepticism 1,301*** Consumer Skepticism 0,150 n.s. R2 0,092 0,101 0,101 0,101 0,306 R2 Adjusted 0,080 0,084 0,080 0,076 0,273 F (Sig) 7,382(0,000) 6,067(0,000) 4,837(0,000) 4,102(0,000) 9,289(0,000) n.s. not significant *p < .10, **p < .05, ***p < .01
Table 4.8 Regression results for all models
was overall significant producing a p-value of 0,000 and F=7,382. However, of the three variables included only OCR-attributes produced a significant result and with a negative coefficient, which highlights the fact that OCR-attributes have a significant negative effect on OCR credibility.
The second model contained all the same variables as the first but in addition includes the interaction between valence and OCR-attributes. Again this model was overall significant as the p=0,000 and the F=6,067 demonstrate. In this case, as in model 1 the only significant result was the OCR-attributes, which again produced a negative coefficient. All other results were not significant.
As was aforementioned, the third model included the interaction effect between recency and valence to the others. This model was also highly significant as a p-value of 0,000 was obtained. As the previous two models the only significant result was the OCR-attributes, which were highly significant and have a negative coefficient. All the rest of the effects resulted as not significant.
The fourth model included the interaction between recency and OCR-attributes. This model produced a p-value of 0,000 therefore, could be considered as significant. However, the only significant result produced was the OCR-attributes, which was again negative but greatly significant.
The final model consisted of all the variables encompassed into the same model. Thus, all the independent variables and the interactions were included as well as the moderator of consumer skepticism. In this model, again OCR-attributes were highly significant with a negative coefficient. Also the interaction between consumer skepticism and OCR-attributes were highly significant with a positive coefficient. All other results provided no sufficient level of significance.
Having found multicollinearity in the regression model 5 that was presented, it is now crucial to find a solution for a more clear interpretation of the results. The technique used is to perform a three-way ANOVA on all the models. However, before it is possible to proceed with the ANOVA it is necessary to implement a median split for the continuous moderator of consumer skepticism. Once the median split has been performed this will signify that the moderator is continuous and the models can be interpreted without any multicollinearity. Thus, the first step was to calculate the mean of the consumer skepticism, which was 4,5023. Then it was possible to divide the respondents into two categories: those who scored less than 4,5023 are skeptical and those who scored above 4,5023 are not skeptical. The table 4.5 that is depicted below shows the ANOVA table containing all the corrected models after the median split.
Model 1 2 3 4 5
F p-value F p-value F p-value F p-value F p-value
Consumer Skepticism 17,761 0,000*** R2 0,092 0,101 0,101 0,101 0,271 R2 Adjusted 0,080 0,084 0,080 0,076 0,236 F (Sig) 7,382 (0,000) 6,067 (0,000) 4,837 (0,000) 4,012 (0,000) 7,827 (0,000) n.s. not significant *p < .10, **p < .05, ***p < .01
Table 4.9 Regression results for all corrected models
The first model is overall significant with a p-value of 0,000 and an F=7,382. In this case, the only significant result was produced by the OCR-attributes with a p=0,000. As the first model, the second one is also overall significant with a p=0,000 and an F of 6,067. Again, all results were not significant except for the OCR-attributes, which were highly significant.
The third model and fourth models are both completely significant and present identical results. These results being that only OCR-attributes presented a significant result.
The final model is also overall significant. The only significant results are the OCR-attributes and the interaction between OCR-OCR-attributes and consumer skepticism. Also the main effect of consumer skepticism is highly significant with a p=0,000. All other results were not significant (see Appendix E for direction of effects).
Figure 1. OCR-attributes
Figure 2. Consumer Skepticism*OCR-Attributes
Thus, it is now possible to compare the models with each other in order to establish which performs best by looking at the adjusted R2
. This is clearly the fifth model, which produces an adjusted R2
of 0,236.
4.3 Study 2: Helpfulness
At this point, the same procedure that was performed in Study 1 was conducted for Study 2. For this study, the same independent variables and moderator were implemented. The only difference being that now the dependent variable is perceived helpfulness of the OCR.
3.5 3.7 3.9 4.1 4.3 4.5 4.7 4.9 Search Experience Est im at ed M ea n s
Estimated Marginal Means of Credibility
3 3.5 4 4.5 5 5.5
Skeptical Not Skeptical
Est im at ed M ar gi n al M ea n s
Estimated Marginal Means of Credibility
As in Study 1, when checking the VIF scores for Study 2 high multicollinearity was encountered in model 5 (see Appendix D, Table 4 for output). Nevertheless, it is possible to say that model 5 was the best performing model (R2
=0,299). Thereafter, the same procedure as in Study 1 was implemented to solve multicollinearity (see Appendix D, for output). The table 4.10 presents the results for the ANOVA with the corrected models. The Study 2 presented identical results to study 1 thus all the output for Study 2 is presented in the Appendix.
Model 1 2 3 4 5
(Constant) F p-value F p-value F p-value F p-value F p-value
R2 0,079 0,081 0,081 0,082 0,263 R2 Adjusted 0,066 0,064 0,059 0,056 0,228 F (Sig) 6,236(0,000) 4,758(0,001) 3,791(0,000) 3,186(0,005) 7,527(0,000) n.s. not significant *p < .10, **p < .05, ***p < .01
Table 4.10. Regression results for all corrected models Study 2
4.3 Hypothesis validation: Study 1 and Study 2
At this point, the entire hypotheses that were generated in chapter 2 can be summarized in the table 4.6 presented below.
Hypothesis Results
Credibility
Results Helpfulness H1A: The presence of search attribute information in an
OCR produces higher perceived credibility than the presence of merely experience attributes.
Accepted Accepted
H2A: An online consumer review with negative information
will have a higher degree of perceived credibility compared to one containing solely positive information
Rejected (Not sig.)
Rejected (Not Sig.) H3A: The positive effect of search attributes over
experience attributes on credibility will be stronger if the valence is negative compared to positive
Rejected (Not sig.)
Rejected (Not sig.)
H4A: Online consumer reviews that have been posted
recently will have a higher degree of perceived credibility compared to reviews that are outdated
Rejected (Not Sig.)
Rejected (Not Sig.)
H5A: Online consumer reviews that are recent and contain
search attribute information will have a higher level of credibility than older reviews containing the same information
Rejected (Not Sig.)
Rejected (Not Sig.)
H6A: Online consumer reviews that are negative and have
been posted recently will have a higher level of credibility than negative reviews that are outdated
H7A: An online consumer review containing solely search
attributes information leads to a higher degree of perceived credibility than experience attributes but the positive impact is lessened when the consumer is skeptical about online reviews
Accepted Accepted
H8A: A negative online consumer review will have a higher
degree of credibility than a positive one but the effect is lessened when the consumer is skeptical about consumer reviews
Rejected (Not Sig.)
Rejected (Not Sig.)
H9A: A recent review has a higher degree of credibility than
an outdated one but the impact is lessened when the consumer is skeptical about consumer reviews
Rejected (Not Sig.)
Rejected (Not Sig.)
5. CONCLUSION
In the concluding chapter, a general summary and discussion of the most relevant results will be presented. Following this, the hypotheses that were generated will be recapped. Lastly, the final section will introduce conceivable limitations of the research that was conducted and possible improvements that could be made in the future regarding this topic.
5.1 Discussion
The primary objective of this thesis was to establish whether there was indeed any relationship between valence, OCR-attributes and recency on the credibility and helpfulness of the OCR. In addition to this, consumers’ degree of skepticism towards the use of online review was tested as a potential moderator of the effects.
Consistent with existing literature, this study confirms indeed that OCR-attributes have a significant effect on OCR credibility and helpfulness. This is clearly demonstrated by the OCR-attributes being highly significant in all models used. Moreover, from the results it is clear that the reviews containing search attributes had a much greater degree of credibility and helpfulness compared to those containing merely experience attributes. Thus, this study confirms existing literature, which clearly states that when an individual attempts to perceive quality of a search attribute it is usually based on unbiased and objective characteristics. Furthermore, it confirms also that experience attributes are a matter of personal taste and more subjective (Mudambi et al, 2010).
On the other hand, the recency of the review was not significant in any of the models except and the results obtained were mixed. In fact, in the results the outdated review was perceived as more credible and more helpful than the recent one. Thus, these results emphasize the fact that the posting date of the online review does not have an effect on the perception of helpfulness and credibility of the OCR.
fact, although prior literature states clearly that a negative should be more credible and helpful than a positive one the outcome of this study shows that the positive review was perceived as more helpful and credible than the negative one.
Lastly, consumer skepticism was tested as potential moderator of the effect of valence, OCR-attributes and recency on the credibility and helpfulness of the OCR. In fact, the only significant interaction was with OCR-attributes. It does not significantly moderate the effect for valence and recency. The results showed that indeed a consumer perceives an OCR containing search attributes as more credible than one containing experience attributes but the effect is lessened when the consumer is skeptical about OCR’s.
5.2 Limitations and future research
Although the thesis presents some surprising results there are still some limitations that should be taken into account in the future in order to improve the analysis on this topic and draw some interesting conclusions. Firstly, due to the time constraint the majority of the respondents that took part in the survey were from Italy or the Netherlands. Thus, in the future it could be interesting to gain a larger sample and also to generate a comparison between nations that are very heavy users of OCR’s and others that rarely rely on online reviews.
Secondly, different moderators could be implemented such as level of involvement in the review and assess whether the results remain unchanged compared to this thesis where consumer skepticism was used. The reason for this being that some of the respondents may have been more interested in the review if the topic was relevant to them or if they had some level of expertise on the industry.
according to changes in academic background, age and social status of the respondents.
