The Moderating Effect of Culture on the Negativity Effect in Online Reviews

(1)

5 December 2016

The Moderating Effect of Culture on the Negativity

Effect in Online Reviews

By Joanne van Straalen

(B5068809 / S2939079) Under supervision of

R. Filieri and M. Wilhelm

(2)

2

Abstract

As electronic word-of-mouth (eWOM) and online reviews has been growing in its impact on marketing in the past years, it has become more important to understand how it can impact consumer decision-making process. This study aims to shed more light on the negativity effect of online reviews, i.e. when negative reviews influence consumers’ purchase intentions more than positive ones, by investigating the moderating role of culture. Despite the growing research into the moderating factors on the negativity effect of online reviews, culture has only seldom been considered. The moderating effect of culture on purchase intention has therefore been examined by conducting both an experiment and questionnaire simultaneously among individuals from Brazil and the Netherlands, focusing on the cultural dimensions of uncertainty avoidance, power distance and individualism on a cross-country and an individual level.

Findings show that among the sample, the negativity effect is visible and has a significant effect on purchase intention. Also, this effect is significantly moderated by the espoused cultural dimensions of uncertainty avoidance and individualism. There has not been found a significant moderating effect of power distance on the relationship between review valence and purchase intention. Still, the study provides initial evidence for taking culture into account when analysing the negativity bias. Lastly, different outcomes by the different research designs provide support for the importance of taking individually espoused cultural values into account when doing a cross-country analysis.

Acknowledgements

(3)

3

List of tables and figures

Tables

Table 4.1. Demographics by national culture ... 23

Table 4.2. Purchase intention scores (estimated marginal means) ... 24

Table 4.4. Convergent validity and reliability test ... 28

Table 4.6. Results of a multi-group, cross-country analysis ... 30

Table 4.7 Individually espoused cultural differences ... 31

Table 4.9 Moderating effect test of individually espoused cultural values ... 33

Table 7.1. Review measurements ... 48

Table 7.2. Culture measurements ... 49

Table 7.3. Repeated measures ANOVA - Tests of significance. ... 50

Table 7.4. Factor loadings for the main constructs. ... 50

Table 7.5. Correlation matrix ... 51

Table 7.6. Results cross-country regression analysis . ... 51

Table 7.7. Factor loadings of the constructs on espoused culture. ... 52

Table 7.8. Reliability analysis fo the constructs on espoused culture ... 52

Table 7.9. Moderating effect test of individually espoused cultural values without IND_3. .. 52

Figures Figure 2.1. Conceptual Model ... 17

Figure 3.1. Comparing the cultural dimensions among the Netherlands and Brazil ... 19

Figure 4.3. Effecs of eWOM, expertise and quantity on purchase intention. ... 25

Figure 4.5. National-level estimates for the Dutch and Brazilian sample. ... 29

Figure 4.8. Graphical representation of individually espoused culture ... 31

(5)

5

1. Introduction

Word-of-mouth, the act of exchanging marketing information among consumers, is one of the oldest forms of advertising (Lee, Rodgers and Kim, 2009). However, when always acknowledged as a powerful source, over the last decades word-of-mouth (WOM) marketing has undergone a meteoric rise (Mcconnell, 2007). More and more marketers realize that WOM communication is more persuasive and desired than traditional forms of advertising (Lee, Rodgers and Kim, 2009). Recent studies have shown that consumers trust peer consumers more than any other form of advertising, as peers have no interest in selling the goods (Nielsen, 2012; Koo, 2016; Sen and Lerman, 2007). Consequently, WOM has become a central driver of marketing effectiveness and sales (Mcconnell, 2007).

The impact of WOM advertising has grown even more since the development of network-technology and the Internet (Hennig-Thurau, et al., 2004). Thanks to the Internet, the nature and power of WOM advertising has been changing and shifted control from marketers to the proactive online consumer (Lee, Rodgers and Kim, 2009). This changing marketing environment is refleced by the growing use of websites such as blogs, Wikipedia, Facebook and YouTube (Valcke and Lenaerts, 2010). It has transformed traditional WOM into computer-mediated or electronic WOM communication (Koo, 2016). Electronic WOM, or eWOM, is defined as ‘any positive or negative statement made by a potential, actual or former customer about a product or company, which is made available to a multitude of people and institutions via the Internet (Hennig-Thurau et al., 2004, p. 39). As the internet is a medium that is used worldwide, eWOM is argued to be even more influential than traditional WOM, due to its ability to reach a larger number of individuals instantly and on a global scale (Christodoulides, Michealidou, and Argyriou, 2012; Hennig-Turau, et al., 2004). The otherwise word-of-mouth communication flow between a few friends, has been transformed into enduring messages, visible to the entire world (Duan, Gu, and Winston, 2008).

(6)

6 and Lee, 2009). As many consumers search for online reviews as the first step in online shopping, these reviews have been found to play an important role in forecasting the sales of many products (Dellacoras, Zhang and Awad, 2007).

One of the most discussed topics around online reviews is their valence, which is the evaluative tone of a review varying from very positive to very negative (Purnawirawan, Dens and Pelsmacker, 2012). Most of these studies rely on the belief that “bad is stronger than good” and that negative reviews influence consumers more than positive ones (Ahluwalia, 2002). However, over time quite some studies did not find any significant effect of research valence (Charlett, Garland, and Marr, 1995; Kimmel and Kitchen, 2014; Wu, 2013).

As a response to these inconsistent findings, various authors started to investigate wether there were any special conditions in order to find a significant effect of eWOM or review valence on consumer behavior. For example, Sen and Lerman (2007) showed that the existence of a negativity bias, where consumers pay more attention to negative information, depends on the product type. Specifically, their research shows that this bias only holds for utilitarian products compared to hedonic goods (Sen and Lerman, 2007). Another research by Ketelaar et al. (2015) claims that the inconsistency in previous results is the consequence of the moderating influence of the expertise of the eWOM receiver on the relationship between review valence and cosnumers purchase intentions.

Despite the groing research into the moderating factors on the negativity bias of online reviews, one possible moderator has only seldom been considered: culture (Luo et al., 2014). Hofstede (1980) explains culture as a phenomenon which can be used to explain cognitions and behaviors among different nations. Research has indicated that cultural values also affect consumers’ expectations and perceptions of products or services, and therefore, their purchase behavior (Kueh and Voon, 2007). As the Internet and social netwoks have put the power in the hands of the consumers and as culture influences the way consumers process information, it is likely that culture would also play a significant role in the way consumers percieve online reviews. If online reviews influence purchasing behavior in a way that negative reviews affect consumer decision-making behavior more than positive reviews, this effect might be different across nations and cultures. Therefore, the research question of this study is:

(7)

7 Only limited research has been conducted into the moderating role of culture with respect to review valence. By researching this topic more in depth, this study aims to bring consistency into the questions in literature around the effect of review valence on consumer purchasing intentions. Doing so, this study focuses on the impact of three different cultural dimensions on the relation between review valence and purchase intentions: individualism-collectivism orientation, power distance and uncertainty avoidance. These effects are tested in the Netherlands and Brasil, as these countries represent contrasting positions on these three cultural dimensions (Hofstede, 2010). However, culture is not only taken as a national-level phenomenon, but also applied to an individual level since individuals, even within a single nation, often espouse cultural values in different degrees (Straub et al., 2002). The research that has been conducted on the relation between culture and eWOM effectiveness until now, only looked at culture as an nation-level aggregate (Christodoulides, Michealidou and Argyriou, 2012), or only focused on one aspect of a national culture (Luo et al., 2014). This study therefore fulfills an important gap in the literature on culture as well as the literature on review valence and has scientific relevance.

From a managerial perspective, it is important to investigate the moderating effect of culture on consumer reactions on online reviews as Internet usage is growing rapidly and making international business possible for almost any organization. With this growing competitive environment, organizations also have greater chances to become exposed to negative events that may threaten customer-brand relationship. As eWOM alows customers to obtain information from all over the world, crisis management becomes increasingly challenging (Aihwa, Hsieh and Tseng, 2013). Especially since more and more consumers rely on reviews for product information, a thorough understanding of how people within different countries are influenced by online reviews is central for marketers (Ketelaar et al., 2015).

(8)

8

2. Theoretical foundations and hypotheses development

2.1. EWOM Valence and the negativity bias

In the literature on the effects of online reviews, review valence is probably the most studied variable (Cheung and Thadani, 2012). Review valence refers to the tone of with which products or services are being discussed in online reviews, varying from very positive to very negative (Cheung et al., 2009). This valence is often seen as a recommendation that can inform consumers’ purchase decisions (Bickart and Schindler, 2001). The general conclusion of most studies is that reviews influence purchase behaviour in the way that positive information encourages consumers to purchase while negative information discourages them (Ketelaar et al., 2015). Nevertheless, there are also studies hat have found no effects for review valence (Duan, Gu, and Winston, 2008; Liu, 2006).

Besides the mixed findings in literature on the impact of positive or negative reviews on consumer decision-making behavior, studies also report different opinions on how much weight is put on either positive or negative online information (Ketelaar et al., 2015). In the literature on psychology, marketing and sociology, the vast majority has shown that when consumers have to judge an object based on an equal amount of positive and negative information, negative information is given greater weight (Lalwani, 2006). This relatively stronger impact of negative information on behavior or perceptions is called the negativity effect or negativity bias (Dougherty, Ward and Hoffman, 2013).

There are various explanations of this negativity effect, which can be summarized into four theories. The first theory, the expectancy-contrast theory, relies on the idea that people have a psychological anchor based on which they judge, and that this anchor is on the positive end of the scale (Singh and Teoh, 2000). As negative information deviates more from this positive anchor than positive information, it will stand out and therefore receive greater weight in judgements (Simpson and Ostrom, 1976). This theory is complemented by the frequency-weight theory which also claims that because people view the world as a positive place, negative information is seen as different and unexpected and therefore receives relatively more attention (Fiske, 1980). The first two theories both thus predict that more positive consumers will find negative information relatively more surprising and thus give it more weight in their evaluations (Lalwani, 2006).

(9)

9 information, it carries relatively more weight (Birnbaum, 1972). Lastly, the category-diagnosticity theory suggests that negative information is more helpful in discriminating between two categorizations (more diagnostic) as they are seen as characteristics primarily of negative attributes whereas for positive information this is not that clear (Skowronski and Carlston, 1987). The last two theories thus predict that less ambiguous information is likely to be given more weight into evaluations (Lalwani, 2006).

All four theories around the negativity effect imply that negative information has a greater impact than positive information. However, these theories are mostly based on impression formation used to form an impression of a person (Ketelaar et al., 2015). Still, various authors have shown that the negativiy effect also holds for product and brand evaluations, product choice and reviews (Sen and Lerman, 2007; Xue and Zhou, 2010; Aihwa, Hsieh, and Tseng, 2013). As many researches have shown that online reviews have an impact on purchasing behavior (Argo, Dahl, and Morales, 2008; Pookulangara and Koester, 2011; Schindel and Bickart, 2002) and assuming that the negativity effect holds for online reviews, the first hypothesis is:

H1: Negatively valenced online reviews have a greater effect on purchase intention than positive valenced online reviews.

2.2. The determinants of the negativity effect

Even though the vast majority of the literature supports the idea of the negativity effect, there are also quite some studies that did not find such an effect (e.g. Charlett, Garland, and Marr, 1995; Kimmel and Kitchen, 2014; Wu, 2013). Other studies even found a positivity effect, where positive information carries more weight than negaive information (Ahluwalia and Klein, 2004; East, Hammond, and Wright, 2007). These contradicting findings show that there must be some conditions in order for the negativiy bias to hold. In the literature, various authors have investigated the determinants of the negativity effect. These findings can be summarized in three different main determinants: review attributes, product attributes and consumer attributes.

(10)

10 hold anymore and an opposite effect becomes visible. It can thus be assumed that in order for the negativity effect to hold, the negative information has to be at least equal in diagnosticity compared to the positive information.

Considering product attributes, research has shown that product type moderates the effect of review valence. A study by Sen and Lerman (2007) showed that the negativity effect only holds for utilitarian products, which are instrumental and functional goods, compared to hedonic goods, that mainly satisfy emotional wants. This effect can be explained by the idea that negative reviews for hedonic goods are seen as more subjective (less diagnostic) and therefore less useful (Sen and Lerman, 2007).

Even though the source, product type and message factors have received sufficient attention in exploring the effects of online reviews (Chaiken and Eagly, 1976), less attention has been given to the receiver: the consumer reading the reviews. Research has shown that whether information is judged as being diagnostic is subjective to individual factors and characteristics (Ahluwalia, 2002). However, the literature has only recently begun to examine individual-level differences in online review susceptibility. Recent studies did show that consumer goals have a significant moderating influence on the negativity effect (Ahluwalia, 2002; Ketelaar et al., 2015; Lalwani, 2006). Also prior experiences of consumers with the brand or attitude towards the brand matter. Studies by Ahluwalia (2002; 2004) and Chang, Hsieh and Tseng (2013) showed that consumers highly committed to a brand do not act conform the negativity bias but even show the opposite: a positivity bias. Other studies again considered dimensions such as internet experience (Zhu and Zhang, 2010) or trust disposition (Utz, Kerkhof and Bos, 2012). These studies indicate that the impact of the valence of online reviews is related to individual factors. Therefore, it is needed to examine these individual factors and their effect on consumer responsiveness to review valence more in-depth.

One important consumer-related factor, for example, that until now has almost not received any attention in studies around the negativity bias, is culture (Luo et al., 2014). In order to distinguish the effect of culture on the negativity bias, the product and review attributes will be held constant throughout this research.

2.3. Culture and the negativity effect

(11)

11 communicate (Leonard, Scotter and Pakdil, 2009). Therefore it is considered as an important influencer of consumer behaviour, especially online. The Internet is a global space and information can be found on products and by consumers from different countries, making people more interconnected than ever (Pookulangara and Koester, 2011). Besides, it has been found that online consumer behavior is not homogeneous but rather differs across countries (Dobele et al., 2007). Samiee (2001) even claims that culture “is the single most important factors that influences international marketing on the Internet”.

Hofstede’s (1980) cultural dimensions have been determined to be the most influential culture theory in social sciences and has received strong empirical support (Pavlou and Chai, 2002). Hofstede’s theory separates cultures on the basis of 6 main dimensions: (a) individualism-collectivism, (b) power distance, (c) uncertainty avoidance, (d) masculinity-femininity, (e) long-term orientation and (f) indulgence (Hofstede, 2001). Individualism-collectivism refers to whether individuals within a culture are group- or self-oriented (Luo, et al., 2014). Power distance represents the degree to which the less powerful members of institutions and organisations within a country expect and accept that power is distributed unequally (Pavlou and Chai, 2002). Uncertainty avoidance reflects the extent to which the members of a culture feel threatened by ambiguous or unknown situations and try to avoid these (Park and Lee, 2009). Masculine cultures emphasize work, competition and material accomplishments whereas feminine societies put caring for others and quality of life at the forefront. The long-term orientation dimension indicates the extent to which people try to control their desires and impulses while dealing with challenges of the past and future (Hofstede, 1980).

(12)

12 2.3.1. Uncertainty avoidance and the negativity effect

Uncertainty avoidance refers to the extent to which members of a society feel uncomfortable with uncertainty (Hofstede, 1980). Countries low on uncertainty-avoidance work to meet basic needs, are tolerant of variety and feel relatively secure. On the other hand, countries high on uncertainty avoidance actively try to avoid risk (Money et al., 1998). Consumers in high-uncertainty avoidance countries are therefore more likely to find online reviews useful in providing purchase-related information to reduce uncertainty (Park and Lee, 2009). It therefore is expected that consumers from cultures high on uncertainty-avoidance pay more attention to diagnostic reviews and focus more on the accuracy goal of information seeking. This especially holds on the Internet, as online purchases create ambiguity because of the heterogeneity of quality and higher associated risk (Lim and Chung, 2011). In these situations, online recommendations are seen as an important factor to reduce this risk (Koo, 2016). However, when there is inconsistency in reviews, with positive and negative reviews without a clear overall tone, reviews may also create perceived risk rather than reduce it. Nevertheless, when there is consistency in the reviews, it is expected that individuals high on uncertainty avoidance place more value on online reviews than consumers from cultures low on uncertainty-avoidance. This leads to the following hypotheses:

H2a. There is a positive correlation between consistency in online reviews and purchase intention.

H2b. Consistency in online reviews affect purchase intention to a larger extent among individuals from cultures high in uncertainty avoidance (i.e. Brazil) compared to individuals from cultures low in uncertainty avoidance (i.e. the Netherlands).

(13)

13 H2c. Negatively valenced online reviews affect purchase intention to a larger extent than positive reviews among individuals from high uncertainty-avoidance cultures (i.e. Brazil) compared to individuals from low uncertainty-avoidance cultures (i.e. the Netherlands).

2.3.3. Power distance and the negativity effect

According to the Handbook of Social Psychology, power distance is one of the two central cultural dimensions affecting psychological processes (Fiske et al., 1998). Power distance refers to the extent to which less powerful individuals accept inequality in power and consider it normal (Stohl, 1993). In cultures high on power distance, superiors tend to be autocratic and subordinates are willing to do as they are told (Pavlou and Chai, 2002). In these cultures, the idea persists that all men are born unequal (Bond and Smith, 1996). It can thus be said that power distance is closely related to social influence. Past researches indeed confirmed that differences in power distance moderate the effectiveness of leadership, with leaders being more influential in cultures high in power distance (Pasa, 2000). Individuals from cultures high in power distance, such as Brazil, are thus more susceptible to opinions from others that they consider as being more powerful or higher in status than themselves (Kirkman et al., 2009).

(14)

14 H3a. There is a positive correlation between online reviews written by a reviewer ranked high in reputation/expertise and purchase intention.

H3b. Online reviews written by individuals ranked high in reputation/expertise affect purchase intention to a larger extent among individuals from cultures high in power distance (i.e. Brazil) compared to individuals from cultures low in power distance (i.e. the Netherlands).

The dimension of power distance does not only translate in whether individuals place more or less value on status. It also reflects in the way individuals communicate with each other (Merkin, 2006). For example, research has shown that in large power distance cultures, people are more reluctant to verbally express negative emotions (Fernández et al., 2000). The expression of negative emotions is generally viewed as inappropriate in countries with a higher rate of power distance (Matsumoto, 1988). For example, a study by Steil and Hillman (1993) showed that Japanese and Koreans, both countries high in power distance, use less confrontational communication styles and are more concerned with being polite. Individuals from large power distance cultures thus appear to be less likely to express their (negative) opinions directly to maintain the emotional distance (Basabe et al., 2002).

(15)

15 H3c. Negatively valenced online reviews affect purchase intention to a larger extent than positive reviews among individuals from large power distance cultures (i.e. Brazil) compared to individuals from small power distance cultures (i.e. the Netherlands).

2.3.2. Collectivism-individualism and the negativity effect

The collectivism-individualism orientation is often described as the most significant cultural dimension to explain differences across cultures (Sia et al., 2009). This dimension explains the extents to which the society values group norms or individual freedom (Singh, Zhao and Hu, 2005). Individualism can be defined as a preference for a loosely-knit social framework in which individuals are expected to take care only of themselves. As an opposite, collectivistic cultures represent a preference for a tightly-knit framework in which individuals are expected to look after members of a particular group (Hofstede, 1980). Various studies have shown the impact of this cultural dimension on online communications (Long, 2011), Internet usage (Chau et al., 2002), knowledge sharing (Ardichvili et al., 2006) and brand community formation (Nadkarni and Hofmann, 2012). Also, there have been found differences between these two types of cultures in terms of eWOM and in particular the extent of information search before a purchasing decision (Long-Chuang, Rose and Blodgett, 1999). For example, consumers from collectivistic cultures rely relatively more on others in the information search process (Wong and Chan, 1999). Individualistic cultures, in contrast, place more value on independence and therefore rely less on opinions from others in online information seeking (Pablos, 2005). This idea also became visible in a study by Doran (2002), who found that Chinese consumers (a collectivistic culture) were more likely than American consumers (individualistic) to search for and rely on personal sources of information. American consumers, on the other hand, relied more on internal knowledge and personal experience when determining wether to buy a product or not. Where the Chinese were more likely to let reference groups influence choices, the Americans were more likely to make their final decisions alone (Doran, 2002).

(16)

16 important in collectivistic cultures, as maintaining the ‘we’ consciousness is valued over expressing private opinions (Laroche, Kalamas and Cleveland, 2005). Therefore, it can be assumed that although both collectivistic and individualistic cultures may be influenced by the overall opinion of a group of consumers via online reviews, this influence might be larger in collectivistic cultures. This assumption leads to the following hypotheses:

H4a. There is a positive correlation between online reviews that are consistent in valence among a sufficient amount of reviewers and purchase intention.

H4b. Online reviews that are consistent in valence among a sufficient amount of reviewers affect purchase intention to a larger extent among individuals from collectivistic cultures (i.e. Brazil) compared to individuals individualistic cultures (i.e. the Netherlands).

After having hypothesized that individuals from collectivistic cultures are more likely to be influenced by group reviews than individuals from individualistic cultures, the literature on this cultural dimension and the negativity effect can be interpreted twofold. The majority of research shows that individuals from collectivistic cultures are less likely to put negative reviews online than people from individualistic cultures (Fong and Burton, 2008; Pablos, 2005). The reason is that in collectivistic cultures, maintenance of harmony, respect for hierarchy and preservation of face are key factors (Chen and Pan, 1993). Therefore they will be less likely to express an opinion which may challenge the opinion from others in a group. On the contrary, people from individualsitic cultures like to express their own opinion and are less resistant to hurting others by negative reviews (Laroche, Kalamas and Cleveland, 2005). This is supported by a study by Priester et al. (2004) that showed that consumers high in individualism generate more anger to service failures, which affected their purchase intentions. He explained that in individualistic cultures, people appear to have lower thresholds of tolerance when their expectations conflict with a perceived failure of a service or product (Priester et al., 2004).

(17)

17 Lerman, 2007). Assuming that individuals in general perceive negative reviews as being more helpful, it might be inducted that this negativity effect is even stronger for individuals from collectivistic cultures.

Nevertheless, a recent study by Christodoulides, Michealidou and Argyriou (2012) does not underline this assumption. Based on an experiment that they did among Chinese and British individuals on cross-national differences in eWOM influence, findings showed that British consumers were overall more susceptible to negativity biases induced by eWOM information (Christodoulides, Michealidou and Argyriou, 2012). These results can be perceived as a contradicting argument against the theoretical induction described above. However, the authors do not give a clear explanation behind why the results show a greater negativity bias in individualistic cultures. Therefore, despite these contraditcting theories, the last hypothesis is stated as followed:

H4c. Negatively valenced online reviews affect purchase intention to a larger extent than positive reviews among individuals from collectitivistic cultures (i.e. Brazil) compared to individuals from individualsitic cultures (i.e. the Netherlands).

2.4. Theoretical model

Summarizing all that has been discussed above, Figure 2.1 presents a proposed model of the relationships among the mentioned determinants of the influence of negative online reviews on purchase intention.

(18)

18 As can be seen, this study mainly focusses on the moderating effect of culture on the relation between negatively valenced online reviews and purchase intention. The other determinants of the negativity bias, such as product category (Sen and Lerman, 2007) or prior attitude towards a brand (Chang, Hsieh and Tseng, 2013) are taken as constant in this study.

With an exception of a few studies there has been little research on the impact of culture on the effectivness of online reviews. However, the studies that are available (Christodoulides, Michealidou, and Argyriou, 2012; Park and Lee, 2009; Xue and Zhou, 2010), have been taking culture as a national-level phenomenon. Even though this is in accordance with Hofstede’s (1980) definition of culture, many studies (Luo et al., 2014; Shin, Ishman, and Sanders, 2007; Srite and Karahanna, 2006) believe it can also be applied to an individual level. Most individuals, even within different cultures and nations, often espouse these cultural values to different degrees (Luo et al., 2014). The reason for this is that each person is influenced by their professions, religions, ethnicities and other variables, which influence their individually espoused cultural values greatly (Straub et al., 2002). Because of these theories, this study does not only test the hypotheses using a cross-cultural national-level framework, but also takes individual espoused cultural values into account.

3. Data and Methodology

In order to test the hypotheses and answer the research question, this study uses two different research designs. More specifically, among a sample of Dutch and Brazilian individuals, both an experiment and a survey have been conducted. The experiment has been conducted to obtain initial insight into the model and its variables as it is a relatively easy method to uncover causal effects (Creswell and Clark, 2011). This method is especially useful with respect to the negativity effect, as this is a psychological and subconscious effect, and thus relatively difficult to measure with a questionnaire (Fiske, 1980). For this reason, many studies on review valence have used an experimental design (Charlett, Garland, and Marr, 1995; East, Hammond, and Lomax, 2008; Ketelaar et al., 2015; Koo, 2016).

(19)

19 53 38 80 76 69 38 Uncertainty Avoidance

Power Distance Individualism

The Netherlands Brazil

Both the experiment and survey were sent out via an online software, translated in Dutch for the Dutch individuals and in Portuguese for the Brazilian individuals. By translating the questionnaire, a larger sampe could be reached as often the English knowledge of Brazilians is insufficient (British Council, 2015). Both the Dutch and Portuguese versions have been translated and backtranslated from and to English.

3.1. The sample

Data has been collected from Dutch and Brazilian individuals. There are important differences between the Netherlands and Brazil that can form the context of the needed emperical analysis on eWOM. According to InternetLiveStats, both the Netherlands and Brazil score relatively high on internet access and usage (InternetLiveStats, 2016). More specifically, the Netherlands has an online penetration rate of 93.7 per cent, and as of May 2015 Brazil has been one of the largest internet markets in the world (Statista, 2016).

Nevertheless both countries represent contrasing positions on the cultural dimensions as described in the theoretical framework of this study. As can be seen in Figure 3.1 based on Hofstede’s (1980) Cultural Model, Brazil scores higher on uncertainty avoidance than the Netherlands. This indicates that Brazil has a relatively stronger need to avoid risk and emphasizes rules and systems to a larger extent than the Netherlands (Hofstede, 1980).

Next to this, Brazil also scores lower on power distance. This indicates that, compared to the Netherlands, Brazil can be seen as a relatively large power distance culture, meaning that they are relatively more likely to accept inequality in power and to do as they are told (Pavlou and Chai, 2002). Furthermore Brazil can be perceived as a collectivistic country valuing group-orientation, whereas the Netherlands can be perceived as an individualistic culture, valuing individual freedom and independence, as Brazil scores relatively lower in individualism (Singh, Zhao and Hu, 2005).

(20)

20 Because several studies have shown that culture does not only vary across nations, but also across individuals within countries (Luo et al., 2014; Shin, Ishman, and Sanders, 2007; Srite and Karahanna, 2006). Therefore, also individual cultural levels are taken into account. In order to better explore the differences in culture among the individuals of both countries, several survey-items have been added using a seven-point Likert Scale, based on a study by Yoon (2009) who also supported the reliability of these items. The specific questions can be found in Appendix 7.2 (Table 7.2).

3.2. The experiment

As explained earlier, an experimental study has been conducted in order to explore the model and its hypotheses. For this experiment, a 1 x 2 x 2 factorial design has been used, with review valence as a within-subjects factor, source reputation and review quantity as between-subjects factors and national culture as a blocking factor.

3.2.1. The stimulus product

Based on the study by Sen and Lerman (2007) that has shown that the negativity effect is more likely to hold for utilitarian (functional) products than for hedonic (emotional) goods, it has been determined to use a utilitarian search good for this experimental design. A search product comprises attributes that are possible to evaluate prior to purchase (Vakrastas & Ambler, 1999). It is therefore assumed that the negativity bias is relatively strong for search goods, as they foster less ambiguous reviews (Sen and Lerman, 2007). This is in line with the majority of experimental studies on the negativity effect (Charlett, Garland, and Marr, 1995; Ketelaar et al., 2015; Lee, Rodgers, and Kim, 2009) who used search goods in their experiments.

(21)

21 phones affect purchase intention of that product for a relatively greater amount than for goods like a computer, holiday destination or cafe (East, Hammond and Lomax, 2008).

Individuals buy smartphones for various different reasons (Filieri and Zhibin, 2016). In order to control for a possible influence of brand image, aesthetic, functional, social or cultural features of the phone, there were no product descriptions or images presented in the experiment. Even though this made the experiment less realistic, this made it possible to only focus on the effect of culture on the negativity bias.

3.2.2. Design of the experiment.

After presenting the participants the situation (“please imagine that you are planning to buy a smartphone. After having reviewed all information regarding the specifications, you found the product that fits your needs and suits your budget limits…”), they were shown two reviews on the smartphone: one positively and one negatively valenced review.

In order to determine those reviews, a pretest has been conducted. This pretest involved a selection of eight different eWOM reviews on smartphones, based on actual reviews obtained from www.kieskeurig.nl and www.buscape.com/br. These reviews were then manipulated a little in two steps in order to create an equal extremity in valence. Firstly, it had been made sure all reviews contained descriptive information on at least two similar smartphone attributes (i.e. camera, design and functionality) based on an experiment done by Ketelaar et al. (2015). Next, several words were replaced by bi-polar adjectives that were, in line with a study by Lee, Rodgers and Kim (2009), selected from the hierarchically ordered list of 50 evaluative adjectives commonly used to describe product taste (Meyers and Warner, 1968). Among these adjectives, “great” and “outstanding” were chosen for the positive reviews, and “bad” and “dissappointing” for the negative reviews. The eight reviews were rated by 32 consumers according to the positivity/negativity of the message on a seven-point scale ranging from “extremely positive” to “extremely negative”. This allowed for the selection of two reliable, comparable and non-extreme reviews as a base for the experiment as shown in Appendix 7.1.

(22)

22 out any influence of prior attitude towards the product on the negativity effect (Lalwani, 2006). After the first indication (t1), respondents were randomnly exposed to new information. Here, two groups respondents were told that one of the reviews, either positive or negative, has been written by an expert (“taking a closer look at the reviews, you discover that one of them had been written by an independent expert on smartphones”). The other two groups were exposed to an overall group evaluation of the smartphone, giving either a positive or negative average score. Following the study of Ketelaar et al. (2015), the positive score has been determined above ‘8’ and the negative average score below ‘3’. After this information, all groups were asked to indicate their purchase intention again (t2).

3.3. The questionnaire

In order to further explore and test the relationships described in the theoretical model, also a questionnaire has been conducted among the respondents. In this questionnaire, the value the individual gives to negative online reviews has been measured using three statements on a 7-point agreement scale (1= strongly disagree, 7 = strongly agree), using three statements: “Negative reviews are trustworthy”, “I tend to read negative reviews in order to assess the quality and performance of a product/service” and “negative reviews help me to assess the quality and performance of a product/service”. The 7-point Likert scale has been used for all subsequent items. Furthermore, questions have been constructed to measure consumer susceptibility to reviewer expertise, group evaluations, consistency in online reviews, and the dependent variable purchase intention. According to reseach by Park and Lee (2009), the relation between review valence and purchase intention is mediated by how useful the reviews are perceived. Therefore, also a three-item construct has been created in order to measure review usefulness. All items and their related questions are shown in Appendix 7.2 (Table 7.1)

4. Analysis and results

(23)

23

Table 4.1. Demographics by national culture

The Netherlands (%) Brazil (%)

Age 18- 5 6.3 18-24 45.8 21.3 25-34 23.1 22.5 35-44 7.5 33.1 45-54 11.9 11.9 55-64 7.5 4.4 65+ 1.3 0.6 Gender Male 38.1 38.1 Female 61.9 61.9 Education Primary education 0.6 4.4 Secondary education 38.8 25.6 Higher education 47.5 43.8 Post-graduate 13.1 26.3 Employment Student 32.5 14.4 Intern 1.9 3.1 Unemployed 8.8 10.6 Employed 49.4 52.5 Self-empoyed 7.5 19.4

A comparison of the demographic characteristics of the respondents using ANOVA revealed that the perceived differences between the two samples were not significant (p>0.05) for all characteristics except for age (p = 0.005). However, as for both the experiment and the questinonaire age is not significantly correlated with the model variables, the demographic factors are not represented in subsequent analysis in order to obtain more clarity.

4.1. The experiment

As the negativity effect, i.e. placing relatively more weight on negative information, is a complicated psychological process, it is a difficult concept to measure with a questionnaire (Ahluwalia, 2002). Therefore, in order to obtain initial insight into the research question and the theoretical model, first the experiment is anlysed.

(24)

24 not met (Mauchly’s test of sphericity p=0.000). Therefore, the reported results are based on the Greenhouse-Gesser test. As previously indicated, tests for correlation between any demographic factor and purchase intention showed no significant correlation (p>0.10). However, consumer expertise is taken as a covariate in the model, as it showed a significant positive correlation with purchase intention (p=0.000). Figure 4.3 gives a graphical representation of the outcomes of the model. Please refer to Appendix 7.3 (Table 7.3.) for more details on the results and significance levels.

An initial within-subject analysis shows that for both the Dutch and the Brazilian respondents, purchase intention decreases significantly after being exposed to a positive and negative review of equal extremity compared to the initial value of ‘4’ (MNLt1 = 3.37 vs. Mt0 = 4, p = 0.000; MBt1 = 2.54 vs. Mt0 = 4, p=0.000). This finding indicates that the respondents indeed seem to place more value on the negative rather than positive information, when indicating purchase intention, which is in line with hypothesis 1. As the reviews are presented in an inconsistent manner, this observation also provides initial support for hypothesis 2a.

Table 4.2 presents an overview of the estimated marginal mean purchase intentions before and after being exposed to the reviews and manipulations for both countries.

Table 4.2. Purchase intention scores (estimated marginal means)

Time

0 1 2

Manipulation National culture Mean SD Mean SD Mean SD

No Manipulation The Netherlands (N=160) 4 0.00 3.37 - -

Brazil (N=160) 4 0.00 3.54 - -

Positive Group evaluation The Netherlands (N=40) 4 0.00 3.43 1.19 4.60 1.11 Brazil (N=40) 4 0.00 3.44 1.83 4.24 1.40

Positive expert review The Netherlands (N=40) 4 0.00 3.49 1.12 4.15 1.99

Brazil (N=40) 4 0.00 3.17 1.52 3.81 1.69 Negative group evaluation The Netherlands (N=40) 4 0.00 3.48 1.33 2.03 1.05 Brazil (N=40) 4 0.00 3.62 1.75 2.69 1.47

Negative expert review The Netherlands (N=40) 4 0.00 3.22 1.39 2.66 1.48

Brazil (N=40) 4 0.00 3.79 1.56 3.08 1.23 Estimated marginal means based on a constant, average expertise, gender and age.

(25)

25

Figure 4.3. Effecs of positive/negative eWOM, expertise and quanity on purchase intention for both countries.

(26)

26 In order to estimate the moderating effect of culture on the effect of both a positive/negative group evaluation and expert review on purchase intention, means have been compared using a Wilcoxon Signed Ranked test (tests for normality: p>0.05). The main outcomes of these tests are shortly explained below.

In the case where Dutch respondents are exposed to a positively valenced group review, their purchase intentions increase significantly (Mt1 = 3.43 vs. Mt2 = 4.60, mean difference 1.17, p<0.001). Brazilian respondents’ purchase intentions also increase significantly from t1 to t2 when they are exposed to a positively valenced group review, but to a lesser extent (Mt1 = 3.44 vs. Mt2 = 4.24, mean difference 0.80, p<0.001). Still, both the Dutch and Brazilian respondents’ purchase intentions at t2 and the magnitude of increase at both countries, were not significantly different (p>0.10). On the other end, when Dutch respondents are exposed to a negatively valenced group review, their purchase intentsions decrease significantly (Mt1 = 3.48 vs. Mt2 = 2.03, mean difference 1.45, p<0.001). The purchase intention of the Brazilian respondents decreases as well, but again to a smaller extent (Mt1 = 3.62 vs. Mt2 = 2.69, mean difference 0.93, p<0.001). In this case, Dutch and Brazilian respondents’ purchase intentions at t2, as well as the difference in magnitude, were significantly different (p<0.05). According to these results, purchase intentions significantly increase (decrease) after seeing a positive (negative) group evaluation for both countries, which provides initial support for hypothesis 4a. However, Brazilians seem to be significantly less susceptible to negatively valenced group evaluations than Dutch respondents when determining purchase intention. This finding is contrary to what has been predicted and does therefore not support hypothesis 4b.

(27)

27 Brazilian respondents would be relatively more susceptible to expert reviews when indicating purchase intentions.

When aggregating the Dutch and Brazilian respondents, it can be observed that a negatively valenced expert review lowers purchase intention on average by 17 per cent, whereas a negative group evaluation lowers purchase intention significantly more, with an average of 30 per cent (F(2,145.3) = 10.84, p=0.000). On the other end, purchase intention increased with an average of 29.9 per cent for a positively valenced expert review, compared with an increase of 43.8 per cent for a positive group evaluation (F(2,146.58) = 0.833, p=0.187). The pooled sample therefore thus reveals that negatively valenced group evaluations have a significantly greater effect on purchase intention than a negatively valenced expert review.

4.2. The questionnaire

Even though using an experimental design is an often used method for determining the negativity effect (Charlett, Garland, and Marr, 1995; East, Hammond, and Lomax, 2008; Ketelaar, et al., 2015; Koo, 2016), there are some disadvantages to it. The outcomes of an experiment often only apply to a particular situation and can often be explained by many different factors (Cooper and Schindler, 2014). As consumer expertise on smartphones has a significant effect in the estimated experimental model (p=0.001), it is likely that the outcomes of the experiment are influenced by the product type. In order to further explore the results of the experiment, the questionnaire has been analysed.

(28)

28 The validity of the measures in this questionnaire was assessed by adopting a Confirmatory Factor Analysis for each of the two countries. The validity has been implied by the strength of the factor loading of each observed measure on its proposed construct. As can be seen in Appendix 7.3 (Table 7.4), all the factor loadings apart from two scored above 0.50, so the results support convergent validity of the measures (Osborne and Costello, 2009). Having established the reliability of the measurements for both countries using Cronbach’s alpha, the items have been converted into single construct measurements for further analysis. An overview of these variables and their reliability values can be seen in Table 4.4.

Table 4.4. Convergent validity and reliability test

The Netherlands Brazil

Variable #items Mean SD α* AVE Mean SD α* AVE

Review valence 3 4.260 1.004 .681 .465 4.885 1.322 .671 .429 Review quantity 5 4.671 .958 .779 .424 5.229 .826 .781 .441 Review consistency 3 5.265 .840 .695 .444 5.367 .664 .636 .405 Source reputation 3 4.117 1.107 .780 .579 4.244 1.154 .750 .556 Review helpfulness 3 4.771 1.078 .792 .562 5.429 1.050 .873 .701 Purchase Intention 3 4.692 1.004 .796 .568 5.181 .928 .791 .566 α* = Cronbach’s alpha

Discrement validity of the measurements has been established as the square root of the Average Variance Explained (AVE) of each construct is larger than its correlations with other constructs (Chin, 1998). The correlation matrix and AVE’s of the main constructs for both countries are reported in Appendix 7.3 (Table 7.5)

4.2.1. A cross-country analysis

Using the outcomes of the questionnaire, the stuctural model has been estimated independently for the Dutch and the Brazilian sample. As explained before, research has indicated that the relation between review valence and purchase intention is mediated by the perceived usefulness of the reviews (Park and Lee, 2009). Using a three step analysis as explained by Baron and Kenny (1986), it indeed appeared that review usefulness plays a mediating role in the relation between review valence and purchase intention (p=0,000). As this is in line with Park and Lee (2009), review usefulness has been added to the theoretical model for the subsequent analyses.

(29)

29 culture later on (Luo et al., 2014). Figure 4.5 below displays the outcomes of the model. A more complete overview of the coefficients and their siginificance levels can be found in Appendix 7.3 (Table 7.6).

Figure 4.5. National-level estimates for the Dutch and Brazilian sample.

The overall fit of the model was good, with a goodness of fit index (GFI) of 0.934. As can be seen in Figure 4.5, the majority of path coefficients were statistically significant (p<0.05) for the combined sample. Both the paths from quantity and consistency to perceived usefulness of reviews show significantly positive coefficents, indicating that respondents who indicated that they found consistency and quantitiy in reviews relatively more important, are also relatively more likely to purchase (not purchase) a positively (negatively) reviewed product/service. This provides further support for hypothesis 2a and 4a. The path from source reputation to perceived review usefulness is not significant for all samples, which does not support hypothesis 3a.

(30)

30 Having established the strucutral relationships among the constructs in each country, a multigroup analysis is performed to test the similarities and differrences across the structural relationships across the two countries (Park and Lee, 2009). The outcomes of this comparison show that the structural model was significantly different between the Netherlands and Brazil (p=0.029). Table 4.6 illustrates al the results of the multigroup model comparison.

Table 4.6. Results of a multi-group, cross-country analysis Path Hypothesized

effect

Results multigroup comparison

Chi-square difference test Hypothesis tesing result H2b RC  PU B > NL B > NL X2 d(1) = 4.567 p = .033* Supported H3b SR  PU B > NL B > NL X2 d(1) = .231 p = .631 (n.s) Not supported H4b RQ  PU B > NL B < NL X2 d(1) = 0.343 p = .558 (n.s) Not supported H2c, 3c & 4c RV  PU B > NL B > NL X2 d(1) = .025 p = .874 (n.s) Not supported No Hyp PU  PI B > NL B > NL X2 d(1) = 6.406 p = .011*

RV = Review Valence, PU = Perceived usefulness, SR = Source Reputation, RC = Review consistency, RQ = Review quantity, PI = Purchase Intention, n.s. = not significant, * p<0.05

As shown in Figure 4.5 and Table 4.6, the relation between source reputation and perceived usefulness of reviews was stronger in Brazil in comparison to the Netherlands. However, this difference in strength was not significantly different (p=0.874). Therefore, hypothesis 3b can not be supported. In the same manner, the model does not provide support for hypothesis 4b. However, the relationship between review consistency and perceived usefulness of reviews is significantly stronger for Brazilian respondents in comparison to the Dutch (p=0.033). This provides support for hypothesis 2b.

Even though the relationship between review valence and perceived usefulness of reviews was relatively stronger in Brazil, this difference was not significantly different across the two countries. This therefore does not provide support for hypothesis 2c, 3c and 4c. However, the relationship between perceived usefulness of reviews and purchase intention is significantly stronger in Brazil, compared to the Netherlands (p=0.011).

4.2.2. Within-country analyisis on individually espoused culture

(31)

31 investigated, as individuals within single nations often espouse cultural values to different degrees (Straub et al., 2002). Since the questionnaire presents a more convenient way of testing individually espoused cultures, and as it was obtained before the experiment, the outcomes of the questionnaire have been used as a base for this analysis.

The mean difference in each item used to measure individually espoused cultural dimensions are represented in Table 4.7. As the items were not normally distributed, a Mann Whitney U test has been used to compare the obtained means per country. Figure 4.8 gives graphical representation of the mean individually espoused cultural values for the two countries.

Table 4.7 Individually espoused cultural differences per sub-item between the Netherlands and Brazil.

The Netherlands Brazil Mann-Whitney U test results Mean SD Mean SD

Uncertainty avoidance UAV_1 2.94 1.472 4.20 1.700 p = .000

UAV_2 3.27 1.651 3.83 1.865 p = .025

UAV_3 3.24 1.556 3.67 1.747 p = .000

3.15 1.347 3.90 1.465 p = .000

Power distance PDI_1 3.81 1.433 3.83 1.643 p = .094

PDI_2 4.17 1.571 3.53 1.621 p = .000 PDI_3 3.40 1.455 3.01 1.582 p = .000 3.79 1.108 3.46 1.305 p = .005 Individualism IND_1 2.69 1.116 2.61 1.534 p = .552 IND_2 3.14 1.137 2.88 1.527 p = .000 IND_3 3.88 1.305 3.93 1.666 p = .015 3.24 1.347 3.14 1.006 p = .371

Figure 4.8. Graphical representation of individually espoused culture in the Netherlands and Brazil

0 1 2 3 4 5

UAV_1 UAV_2 UAV_3 PDI_1 PDI_2 PDI_3 IND_1 IND_2 IND_3

(32)

32 As can be seen from the table and the figure, the differences do not represent the expected results for all cultural values as induced from Hofstede’s Cultural Model (Figure 3.1, page 19). As the model predicted Brazil to be more susceptible to power distances, analysis showed that Brazilian respondents scored significantly lower on power distance in comparison to the Netherlands (p=0.000). Furthermore, Brazilian and Dutch respondents did not differ significantly in individualism (p=0.371). The Brazilian respondents did however, score significantly higher values on all items of uncertainty avoidance (p=0.000), which was in line with expectations. Interestingly enough, this was also the only cultural dimension that showed a significantly stronger correlation with purchase intention through perceived usefulness of reviews in Brazil in comparison to the Netherlands (see Table 4.6, page 30 for reference). These findings therefore provide initial support for the theory that individual espoused culture should be taken into account when doing cross-country analysis (Straub et al., 2002).

The validity of the items on cultural dimensions were assessed by Factor Analysis; the reversed scale item (IND_2) was reversed before the statistical analysis. As can be seen in Appendix 7.3 (Table 7.7), the measures on power distance and uncertainty avoidance have relatively high factor loadings. However, the three statements on individualism have low factor loadings, which was confirmed by the low outcomes of reliability tests (Cronbach’s Alpha = 0.101). Item reliability testing showed that reliability would improve by removing the item IND_3 (see Table 7.8 in Appendix 7.3). However, as this improvement was so small, all measures were kept in the model, in order to stay in line with previous research by Yoon (2009).

In order to examine the impact of individually espoused cultural values, a widely used moderated multiple-regression model was built (Luo et al., 2014). In order to perform this regression, the Brazilian and Dutch respondents are pooled together as one sample. As explained before, all items have been standardized (z-score). In order to test the moderating effects, the constructs on culture have been multiplied by the relating independent factors separately. Adding the cultural dimensions and the interaction terms subsequently ino the regression model, three different models were obtained. The results of the moderating test are demonstrated in Table 4.9.

(33)

33 avoidance indeed moderates the effect of review valence on perceived usefulness of reviews on an individual level. This provides support for hypothesis 2c. However, including individually espoused cultural dimensions still does not create support for hypothesis 3a, b and c as source reputation in reviews nor power distance seem to have a significant effect on the perceied usefulness of reviews. Also, contrary to hypothesis 4c, it seems as though individualism increases the negative effect of review valence on perceived usefulness of reviews (B=0.103, p=0.05). This finding is specifically interesting as the model also shows an overall effect of individualism on purcase influence of reviews which is significantly negative.

Table 4.9 Moderating effect test of individually espoused cultural values

Model 1 Model 2 Model 3

B SE Sig. B SE Sig B SE Sig

RV .216 .054 .000*** .202 .058 .001*** .214 .058 .000*** RC .181 .056 .001*** .153 .057 .008** .150 .057 .000*** SR .047 .053 .378 .064 .054 .232 .064 .053 .377 RQ .207 .057 .000*** .217 .057 .000*** .219 ,057 .009** UAV (culture) .021 .055 .709 -.003 ,056 .792 PDI (culture) -.096 .050 .057 -.095 .050 .073 IND (cutlure) -.075 .050 .137 -.112 .051 .028** RC*UAV .037 .056 .483 SR*PDI -.024 .049 .651 RQ*IND .089 .050 .090 RV*UAV .114 .051 .032* RV*PDI -.039 .048 .438 RV*IND .103 .049 .050* ΔF (p-value) 22.627 (.000) 2.025 (.000) 2.832 (.000) Adjusted R2 _(p) _.213 _.221 _.247

RV = Review Valence, PU = Perceived usefulness, SR = Source Reputation, RC = Review consistency, RQ = Review quantity,

Dependent variable: Perceived usefulness of online reviews. * p<.05, ** p<.010, *** p<.001

(34)

34

Figure 4.10. Results of multiple-regression analysis

5. Discussion and conclusion

5.1. Discussion of the findings

This study contributes to the understanding of the negativity effect and the antecedents of purchase influence of online reviews by investigating the negativity effect across different nationalities. In order to do so, this study adopted a questionnaire as well as an experiment among individuals from the Netherlands and Brazil.

Primarly, findings from the experiment confirm that exposure to positively and negatively valenced reviews, significantly affects the purchase intentions for both Dutch and Brazilian respondents. More specifically, for both countries respondents’ purchase intentions were lower after being exposed to a positively and negatively valenced review of equal extremity. This initial indication of the negativity effect in online reviews, was supported by the outcomes of the questionnaire. These results are consistent with hypothesis 1 and complements previous literature that indicated an existence of a negativity effect in online reviews (e.g. Alhuwalia, 2002; Christodoulides, Michealidou, & Argyriou, 2012; Sen and Lerman, 2007).

Next to the valence of reviews, the outcomes of both the questionnaire and the experiment show that the purchase influence of online reviews is also influenced by the consistency and quantity of online reviews, which is in line with hypothesis 2a and 4a. This highlights the importance of having a high amount of reviews and consistency in reviews on a product or service on the Internet.

(35)

35 significant results could be found. Hypothesis 3a can therefore not fully be supported. This result is not in line with research done by Sen and Lerman (2007) who found that reviewer attributes play a significant role in consumer attitudes towards a review. However, as the experiment indicates directional evidence for the effect of reviewer expertise on purchae intention, furher analysis will be needed in order to obtain more consistent results. Furthermore, this study provides an interesting finding in both the experiment and questionnaire, showing that the quantity of negative reviews plays a significantly larger role in determining purchase intention than the reputation/expertise of the reviewer. Further research could further explore this difference in effect.

In order to examine the effect of culture on the negativity effect, both a cross-country analyisis and an analysis on individually espoused culture have been conducted. When looking at individually espoused cultures, the espoused cultural dimensions among the Brazilian and Dutch respondents did not portray the differences in cultural dimensions as indicated by Hofstede’s Cultural Model (1980). This provides support for the belief that culture is not only a national-level phenomenon, but should rather be applied to an individual level (Luo et al., 2014; Shin, Ishman, and Sanders, 2007; Yoon, 2009). The main conclusions of this study are therefore also based on the analysis on the individually espoused cultural dimension.

The results of this analysis did not show a significant moderating effect of uncertainty avoidance on the relationship between review consistency and perceived usefulness of reviews. This does not support hypothesis 2b, which stated that the effect of review consistency on purchase intention would be stronger when there is a high level of uncertainty avoidance. Also hypothesis 3b and 4b could not be supported as the level of power distance did not significantly influence the relationship between source reputation and percieved usefulness of reviews and source reputation and review quantity subsequently.

(36)

36 On the other hand, the cultural dimension of power distance does not affect the relationship between negatively valenced reviews and purchase intention through perceived usefulness of reviews. This finding therefore does not provide support for hypothesis 3c. However, this result is in line with a study by Pavlou and Chai (2002), where, also against expectations, power distance did not have a moderating effect on the relationship between social influence of online reviews and online transactions. However, as the individually espoused levels of power distance across the two countries did not represent Hofstede’s Cultural Model (1980), further research among different or more countries might be needed in order to confirm these results.

Lastly, this study shows a surprising but interesting outcome regarding the moderating effect of individualism on the relaionship between the negativity effect and purchase intention through perceived usefulness of reviews. Where it was hypothesized that negative online reviews would affect purchase intention to a larger degree among individuals with high levels of collectivism compared to individualism, the opposite effect was shown. Even though this finding does not support hypothesis 4c, it does provide support for the results of the research by Christodoulides, Michealidou and Argyriou (2012). In their study, a similar experiment was conducted among Chinese and British individuals, which showed that British consumers, who are relatively more individualistic, were overall more susceptible to the negativity effect. As the research of Christodoulides et al. (2012) also failed to explain the reason why the negativity effect might be stronger for individualistic cultures rather than for collectivistic cultures, further research is needed to explore the underlying reasons for the observed outcomes.

When it comes to the research question “to what extent does culture affect the review valence effect on purchase intentions in online reviews?”, it can thus be argued that this research provides initial evidence that culture does have an impact on the negativity effect on purchase intentions in online reviews. More specifically, the negaitivity effect on purchase intention is stronger for individuals with higher levels of uncertainty avoidance and, contrary to what has been expected, higher levels of individualism.

5.2 Theoretical and practical contribution

The Moderating Effect of Culture on the Negativity Effect in Online Reviews