“Comments are turned off”: the effects of eWOM valence and absence on consumers’ trust, brand attitudes, and purchase intentions

“Comments are turned off”.

The effects of eWOM valence and absence on consumers’ trust, brand attitudes, and purchase intentions.

Cedric Bratzke, 12345776

Master’s Thesis, Graduate School of Communication Persuasive Communication, MSc Communication Science Dr. Hilde Voorveld

Abstract

Social networking sites such as YouTube allow users including brands to disable comments which may contain product reviews but little is known about advantages and disadvantages in comparison to enabling them. The present study approaches this understudied phenomena by testing the effects of eWOM valence (positive eWOM and negative eWOM) and eWOM absence (non-referential comments and disabled comments) underneath YouTube brand content on consumers’ brand trust, brand attitude, and purchase intention. An online experiment with a 1 x 4 between-subjects design (N = 207) revealed that disabling the comment section is only beneficial to brands when negative eWOM is anticipated and not beneficial when positive eWOM is anticipated. When the comment section was disabled, participants had higher brand attitudes and purchase intentions in comparison to negative eWOM, and they had lower purchase intentions in comparison to positive eWOM. There was no difference in consumer responses when comparing comments which did not refer to the brand or product in question (non-referential) with positive eWOM and in comparison with disabled comments. Consumers’ trust in the brand mediated the relationship between eWOM absence and valence with brand attitude and purchase intention when comparing disabled comments with positive eWOM and negative eWOM.

Introduction

Social networking sites (SNS) enable the exchange of firm-created social media communication (FCSMC) and user-generated social media communication (UGSMC) such as in the form of brand content and user comments, respectively (Jha, 2019). UGSMC can be used to provide feedback and give recommendations regarding brands and products also known as electronic word-of-mouth (eWOM) (King, Racherla & Bush, 2014). EWOM enables consumers to gain intrinsic insight into individuals’ opinions and has been shown to have effects on consumer responses (Roy, Datta & Mukherjee, 2019). YouTube hosts such an exchange of FGSMC and UGSMC including eWOM. On YouTube, brands can benefit from reaching the largest audience of any online video sharing platform which coincides with a saturated brand presence (Saunders, 2016a; Saunders, 2016b). Consumers may follow brand accounts on YouTube to derive more information on brands and their products from FGSMC and UGSMC (Saunders, 2016a; Saunders, 2016b; Statista, 2019a). Despite this exchange, YouTube allows content creators to disable the comment section which therefore limits any UGSMC including eWOM (YouTube, n.d.). Apple, one of the world’s most valuable brands, uses this feature whilst publishing brand content for over 10 million subscribers (Apple, n.d.; Forbes, 2019).

EWOM absence has thus far largely been ignored within the academic realm. EWOM absence may entail two categories; firstly, non-referential UGSMC which are user comments that differ from eWOM as they do not provide feedback or give recommendations in reference to brands and their products (Jha, 2019; King, Racherla & Bush, 2014). Secondly, disabled UGSMC where the comment section is deactivated and therefore does not allow individuals to produce or consume UGSMC including eWOM (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015; Jha, 2019). Past eWOM research has mostly focussed on the

impact of eWOM valence which encompasses positive and negative eWOM (López & Sicilia, 2013). For instance, eWOM with positive and mixed sentiments have been found to have a positive effect on both brand attitude and purchase intention (Jha, 2019; Roy, Datta & Mukherjee, 2019). Negative eWOM has been linked to an opposite and sometimes stronger effect on both brand attitude and purchase intention (Basuory, Chatterjee & Ravid, 2003; Ladhari & Michaud, 2015; Roy, Datta & Mukherjee, 2019). There is only a limited collection of academic explorations which have investigated eWOM absence (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). The first study compared the effects of disabled UGSMC with positive eWOM and found a positive effect for positive eWOM, and a negative effect for disabled UGSMC on credibility, and thereafter on brand attitude and purchase intention (Hayes & Carr, 2015). The second study did not incorporate positive eWOM but compared the effects of negative eWOM with disabled UGSMC and found in part more positive levels of trust for disabled UGSMC (Bühler, Murawski & Bick, 2017). Despite these findings, the two studies focussing on disabled UGSMC did not compare all levels of eWOM valence and absence in one singular study and did not incorporate non-referential UGSMC (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). However, it may first has to be known whether non-referential and disabled UGSMC differ in their effect on consumer responses in comparison to the other conditions. A difference would suggest that it is not mere eWOM absence driving the effect and that there may be advantages and disadvantages in disabling the comment section. This points to a research gap which bears the opportunity for the present study to investigate all four states combined, positive eWOM and negative eWOM (eWOM valence), and non-referential UGSMC and disabled UGSMC (eWOM absence), for a complete comparison.

Lastly, previous research suggest brand attitude and purchase intention as suitable measures of consumer responses for both eWOM valence and eWOM absence (Hayes & Carr, 2015; Ladhari & Michaud, 2015; Jha, 2019). Nevertheless, the construct trust has also been suggested by studies incorporating disabled eWOM as the underlying mechanism explaining the effects of eWOM valence and disabled eWOM on both brand attitude and purchase intention (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). The study comparing positive eWOM with disabled eWOM incorporated expertise and credibility as potential mediators but in reflection of their measures instead propose trust as a more suitable measure (Hayes & Carr, 2015). The other study comparing negative with disabled eWOM exclusively measured the effects on trust (Bühler, Murawski & Bick, 2017). In reflection of this, trust has been proposed as a potential factor explaining the effects of eWOM valence and disabled eWOM on brand attitude and purchase intention but has not been tested thus far. This shall therefore be addressed in the present study.

RQ: What are the effects for eWOM valence (positive and negative) and eWOM absence (non-referential and disabled) on consumers’ brand attitude and purchase intentions, and does consumers’ perceived trust mediate this relationship?

Through exploring this research question the present study makes the following four main contributions. Firstly, the present study redefines eWOM absence by differentiating between non-referential UGSMC and disabled UGSMC to observe potential differences. Secondly, through testing an inclusive model which recognises the importance of different levels of eWOM valence and eWOM absence, as well as testing the mediating effect of trust, this research is able to compliment current knowledge by capturing an extensive comparison.

Thirdly, the present study will compliment the eWOM absence stream in terms of modality by testing the effects on the online video sharing platform YouTube. Previous studies within this stream have exclusively used static text and imagery and not dynamic visuals as majorly present on YouTube (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015, YouTube, n.d.). Lastly, this study may have professional implications for social media marketeers in shining a light on the prominent question whether as well as when and why disabling the comment section may be effective or ineffective.

Theoretical Background and Hypotheses

YouTube, EWOM and Involvement in the Consumer Decision Journey

The present study will test the effects of eWOM valence and absence under high product involvement and high situational involvement based on suggestions that YouTube may be a particularly attractive source of information under those circumstances. The consumption of brand content (FGSMC) and of eWOM is suggested to majorly occur in the information search and evaluation phases of the consumer decision journey when the consumer still considers an array of competitors and requires information to make informed comparisons (Batra & Keller, 2016; Court, Elzinga, Mulder & Vetvik, 2009; Chen, Nguyen, Klaus & Wu, 2015). Consuming FGSMC and eWOM prior to purchase are types of digital consumer engagement practices which have been linked to heightened cognitive processing (Eigenraam, Eelen, Van Lin & Verlegh, 2018; Muntinga, Moorman & Smit, 2011). The degree of cognitive processing effort consumers direct towards the consumption of FGSMC and eWOM in the information search and evaluation phase of the consumer decision journey

depends on the consumers’ level of involvement (Batra & Keller, 2016; Muehling, Laczniak & Stoltman, 1991). High-involvement products are associated with higher risks such as the monetary value of the product and are therefore purchased based on thorough deliberation of the information provided by both the brand and by other consumers (Gu, Park & Konana, 2019). Besides product involvement, high situational involvement may also contribute to the motivation to process information thoroughly. Situational involvement defines a state of heightened interest and thus involvement in a particular moment such as certain stages in the consumer decision journey (Laurent & Kapferer, 1985; Zaichowsky, 1994). YouTube may be a destination for consumers who are highly involved and are hence inclined to actively search for information on high involvement products as users are able to find both information provided by the brand (FGSMC) and information provided by other consumers (eWOM) on one platform (Saunders, 2016a).

YouTube’s Structural and Functional Characteristics

The present study tests the effects of eWOM valence and eWOM absence on YouTube which has different structural and functional characteristics than the platforms used in previous eWOM absence studies (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). Therefore, findings may expand current academic knowledge on the effects of eWOM valence and absence on platforms consumers use as their source of information prior to making a purchase decision. Retrieval media such as YouTube allow the audience to externally pace content through pausing and resuming (Dijkstra, Buijtels & Raaijj, 2005). Retrieval media enhance information processing because consumers may “process the information at their own pace” (Dijkstra, Buijtels & Raaijj, 2005, p. 378). Research into creative formats has suggested that dynamic content outperforms static content regarding user

engagement which may suggest YouTube’s potential to highly involve consumers (Bruce, Murthi & Rao, 2017). It is suggested that user engagement with the platform is particularly driven by the entertaining characteristics of YouTube (Voorveld, Van Noort, Muntinga, & Bronner, 2018). Based on the media richness theory, YouTube hosts rich media as the interactive platform allows feedback, enables personalisation, uses an array of communication cues and simplifies complex messages (Brunelle & Lapierre, 2008; Daft & Lengel, 1986; Pavlou & Stewart, 2013; Saunders, 2016a; Saunders, 2016b ). Lastly, rich media have been found to reduce cognitive information processing effort which enables consumers to process complex information more easily (Maity, Dass & Kumar, 2018). Therefore, Youtube can be considered a dynamic and rich retrieval media source which differs in comparison to static and lean retrieval as employed in previous eWOM absence studies (Bühler, Murawski & Bick, 2017; Dijkstra, Buijtels & Raaijj, 2005; Hayes & Carr, 2015; Maity, Dass & Kumar, 2018).

The Effect of Perceived EWOM Valence on Consumer Responses

Literature states that a key difference which causes differential consumer responses for eWOM is eWOM valence, the perception of eWOM being either more positive or negative (López & Sicilia, 2013). This notion is based on information integration theory which posits that consumers do not see positive or negative statements singularly but rather aggregate all assigned values into one average score (López & Sicilia, 2013). Nevertheless, negative statements may be assigned an overall larger and more impactful value known as the negativity bias which posits that individuals, “give greater weight to negative entities” (Rozin & Royzman, 2001, p. 296). In general, eWOM valence may be more influential when product involvement and situational involvement is high as more cognitive resources may be directed

towards the actual content of eWOM and thereby also its overall sentiment (Cacioppo & Petty, 1983; Jalilvand, Esfahani & Samiei, 2011; Krishnamurthy & Kumar, 2018). Information integration theory has been tested and confirmed in the field of eWOM research in a meta-analysis which tested predictors of purchase intention for eWOM and found perceived valence as one of the strongest predictors (Ismagilova, Slade, Rana & Dwivedi, 2019). Additionally, eWOM research has established a link between perceived eWOM valence and brand attitude (Lee, Rodgers & Kim, 2009). Previous studies inspecting positive and negative eWOM have found differential effects in which positive eWOM had a more positive effect, and negative eWOM had a more negative effect on consumer responses (Ismagilova, Slade, Rana & Dwivedi, 2019; Jha, 2019; King, Racherla & Bush, 2014; Ladhari & Michaud, 2015; Lee, Rodgers & Kim, 2009; Roy, Datta & Mukherjee, 2019). Another study also found a sometimes stronger effect for negative eWOM on consumer responses than positive eWOM, potentially due to the negativity bias (Basuory, Chatterjee & Ravid, 2003; Rozin & Royzman, 2001). It is suggested that these findings may apply to any product where consumers cannot fully judge the quality of the product without physically using or experiencing it such as in an online setting (Basuory, Chatterjee & Ravid, 2003). The negativity bias could not be confirmed in previous eWOM absence studies as they solely compared disabled UGSMC with either positive eWOM or negative eWOM, but not positive eWOM and negative eWOM (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). The studies only found more positive consumers responses for positive eWOM than for disabled UGSMC and the opposite for negative eWOM (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). Therefore, the present study expects positive and negative eWOM to have opposing effects on brand attitude and purchase intention.

H1a: EWOM valence (positive eWOM and negative eWOM) of comments underneath

YouTube brand content influences brand attitude and purchase intention. Both brand attitude and purchase intention will be more positive for positive eWOM, and more negative for negative eWOM.

The Effect of EWOM Absence on Consumer Responses

As previously noted, eWOM absence can be categorised into non-referential UGSMC and disabled UGSMC. In the case of disabled UGSMC, one may apply uncertainty reduction theory to predict consumer responses when they cannot consume any UGSMC including eWOM and non-referential UGSMC. The theory has its roots in interpersonal communication and individuals’ desire to reduce uncertainty when interacting with others during initial encounters which may reduce unpleasant feelings and thus increase attraction (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger & Calabrese, 1975). It is proposed that when individuals increasingly engage in verbal communication and information seeking, uncertainty may be reduced (Berger & Calabrese, 1975). Uncertainty reduction theory has been applied in computer-mediated communication where researchers found no difference in increased attraction between face-to-face and text-only communication, therefore confirming the possibility to reduce uncertainty merely through text-based online communication (Antheunis, Schouten, Valkenburg & Peter, 2012). Other findings suggest that in computer-mediated domains and eWOM communication users highly value interpersonal communication as they are more satisfied when they are able to engage with the opinion of others (Sarmiento Guede, Esteban Curiel & Antonovica, 2018). Therefore, consumers may reduce uncertainty on YouTube through consuming UGSMC including eWOM underneath brand content. Reducing uncertainty may be particularly imperative for high involvement

products which are associated with higher risks (Gu, Park & Konana, 2019). It is suggested that consumers actively search for information from both the brand and third party sources in their decision journey with the intent to reduce risk or uncertainty (Batra & Keller, 2016; Pavlou & Stewart, 2013). As eWOM is considered a more trustworthy source than FGSMC, it is often applied by consumers as a risk reduction strategy (Gu, Park & Konana, 2019; Jha, 2019; King, Racherla & Bush, 2014). Lastly, in a consumer-driven realm such as YouTube a power struggle between brands and consumers exist in which brands are expected to act according to the rules and expectations set out by consumers. This power imbalance results in brands having to hand over control in order to succeed in the Web 2.0 (Fournier & Avery, 2011).

Based on these theoretical suggestions, if a brand decides to disable UGSMC, consumers wanting to reduce uncertainty are according to uncertainty reduction theory unable to do so (Berger & Calabrese, 1975). They may be left in an unpleasant state with high uncertainty and lower attraction for the brand in question (Berger & Calabrese, 1975). They may also not apply eWOM evaluation as a mean to reduce risks and may be left unsatisfied as they cannot communicate interpersonally (Sarmiento Guede, Esteban Curiel & Antonovica, 2018; King, Racherla & Bush, 2014). Lastly, it may appear to the consumer as if the brand acted inappropriately in a consumer-driven online space (Fournier & Avery, 2011). This study therefore expects disabled eWOM to have a negative influence on brand attitude and purchase intention. To understand how this may compare to positive and negative eWOM, one may turn to previous studies testing disabled UGSMC in comparison to either positive or negative eWOM. In part, their results suggested for disabled UGSMC to have a more negative impact than positive eWOM on consumers responses and a more positive

impact than negative eWOM on consumer responses (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). The present study therefore expects a replication of these results.

H1b: EWOM absence (i.e. disabled UGSMC) underneath YouTube brand content

influences brand attitude and purchase intention. Both brand attitude and purchase intention will be more positive for disabled UGSMC in comparison to negative eWOM, and more negative for disabled UGSMC in comparison to positive eWOM.

The second aspect of eWOM absence is non-referential UGSMC. The present study incorporates non-referential UGSMC which is not overly positive or negative to eliminate positive or negative valence as a confounding factor. This will enable the study to make a direct comparison to disabled UGSMC and therefore draw conclusions regarding potential advantages and disadvantages of disabling the comment section. Therefore, in the present study consumers likely cannot assign an overall positive or negative value to non-referential UGSMC as proposed by information integration theory (López & Sicilia, 2012). This also means that consumer responses are expected to be more favourable for positive eWOM or less favourable for negative eWOM than non-referential UGSMC. In comparison to disabled UGSMC, non-referential UGSMC is expected to yield more positive consumer responses as users may still reduce uncertainty through interpersonal communication, as proposed by uncertainty reduction theory (Berger and Calabrese, 1975). Additionally, the brand may not be perceived has having acted out of order in a consumer-driven online space where brands are expected to follow the rules of the consumer (Fournier & Avery, 2011). Based on these two tendencies, the difference in consumer responses between disabled UGSMC and non-referential UGSMC is expected to be present despite consumers not being able to reduce risk

through eWOM consumption in both instances (King, Racherla & Bush, 2014). Therefore, the present study proposes the following hypothesis for consumers’ brand attitude and purchase intention when faced with non-referential UGSMC.

H1c: EWOM absence (i.e. non-referential UGSMC) underneath YouTube brand content

influences brand attitude and purchase intention. Both brand attitude and purchase intention will be more positive for non-referential UGSMC in comparison to both disabled UGSMC, and negative eWOM, and more negative for non-referential UGSMC in comparison to positive eWOM.

The Mediating Effect of Trust for EWOM Valence and Absence

Trust has been suggested as a direct consumer response to eWOM valence and absence as well as an underlying mechanism explaining their effects on brand attitude and purchase intention (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). Researchers have attempted to define trust and what it entails extensively but two intertwined core ideas reoccur in numerous definitions, reliance and expectancy, the subjective belief that another entity can be relied upon to meet set expectations (Blomqvist, 1997). In a marketing context, this may be translated into the consumer’s belief that a brand can be relied upon to meet the expectations generated through promises made in the brand’s marketing communication (Pavlou & Stewart, 2013). To then reduce the risk or uncertainty of a wrongful promise, individuals may rely on additional information other than the information provided by the brand such as eWOM which is considered a more trustworthy source of information (Bhandari & Rogers, 2018; King, Racherla & Bush, 2014). This belief is reflected in warranting theory which proposes that the perceived credibility of a computer-mediated information source is

dependent on whether it can warrant to be free of manipulation by the original sender (Walther & Parks, 2002). Walther and Parks (2002) further differentiate between self-generated information produced by the original sender and other-self-generated information produced by an external source about the original sender. Self-generated information is said to be less trustworthy due to higher chances of manipulation by the original sender (Walther & Parks, 2002). Applying this theory would therefore classify FGSMC as less trustworthy and UGSMC including non-referential UGSMC and eWOM as more trustworthy (Walther & Parks, 2002). Warranting theory has been successfully tested on other social networking sites including Facebook and LinkedIn (Rui, 2018; Walther, Van Der Heide, Hamel & Shulman, 2009). Moreover, warranting theory has been applied in the field of eWOM absence research to explain the mediating effect of trust (Hayes & Carr, 2015). The study compared disabled UGSMC with positive eWOM and tested whether perceived credibility and expertise mediated the effects on brand attitude and purchase intention (Hayes & Carr, 2015). Results suggested that only expertise mediated the relationship but drawing on warranting theory, the researchers noted that trust may be a more indicative measure of the indirect effects (Hayes & Carr, 2015). Another eWOM absence study compared the effects of negative eWOM with disabled UGSMC on trust and found more negative levels of trust for negative eWOM and more positive levels of trust for disabled UGSMC (Bühler, Murawski & Bick, 2017). Therefore, their findings indicated that eWOM absence and valence may have an effect on trust (Bühler, Murawski & Bick, 2017). The suggested mediating effect of trust has been shown in an e-commerce study which found that a decrease in perceived risk first led to an increase in consumers’ perceived trust in the brand and then led to an increase in the intent to buy (Hong & Cha, 2013). The authors suggest that particularly in an online setting, having trust comes with a decrease in uncertainty because users want to be certain that the brand can

be trusted in purchasing from without re-assuring face-to-face interaction (Hong & Cha, 2013). This connects to uncertainty reduction theory and consumers desire to reduce uncertainty through interpersonal communication (Berger & Calabrese, 1975; Hong & Cha, 2013).

Therefore, based on the notion of trust, warranting theory, and uncertainty reduction theory, the study suggests the following process of effects for the mediating role of trust in the model (Berger & Calabrese, 1975; Pavlou & Stewart, 2013; Walther & Parks, 2002) In general, eWOM valence may have high warranting value as it is produced by others (Walther & Parks, 2002). For positive eWOM, consumer expectations based on the brand’s promises would be confirmed and the brand could be relied upon (Pavlou & Stewart, 2013). This may first result in more positive levels of trust, meaning the user now has less uncertainty, and therefore then has a more positive attitude towards the brand, and is more inclined to purchase (Berger & Calabrese, 1975). For negative eWOM, consumer expectations would not be confirmed and the brand may not be evaluated as reliable (Pavlou & Stewart, 2013). This may first result in more negative trust levels, the user has more uncertainty, which therefore may result in more negative attitudes towards the brand, and more negative purchase intentions (Pavlou & Stewart, 2013; Berger & Calabrese, 1975). Disabled UGSMC may have low warranting value and result in low trust levels because it may be perceived as a deliberate manipulation by the brand to block other-generated information and merely allow less credible self-generated information (Walther & Parks, 2002). In comparison to positive and negative eWOM, disabled UGSMC may not present a complete confirmation and rejection of consumer expectations which therefore may suggest trust, brand attitude, and purchase intention levels to be in between positive and negative eWOM (Pavlou & Stewart, 2013). In comparison to disabled UGSMC, non-referential UGSMC may have higher warranting value

as it is produced by others and may therefore first generate more positive trust levels, and then more positive purchase intentions and brand attitudes as consumers are less uncertain (Berger & Calabrese, 1975; Pavlou & Stewart, 2013; Walther & Parks, 2002) Additionally, trust levels for non-referential UGSMC are expected to be more positive than negative eWOM because of less uncertainty and may therefore result in more positive brand attitudes and purchase intentions (Berger & Calabrese, 1975). The opposite is expected when comparing non-referential UGSMC with positive eWOM as non-referential UGSMC may neither confirm nor reject consumer expectations (Pavlou & Stewart, 2013). In summary, the present study expects trust to mediate the relationship between eWOM valence/absence with brand attitude and purchase intention.

H2: Consumers’ perceived trust mediates the relationship of eWOM valence and

absence with brand attitude, and purchase intention.

Research methodology

An online experiment was utilised to study the effects of eWOM valence and eWOM absence below YouTube brand content on consumers’ perceived trust, brand attitude, and purchase intention. The 1 x 4 between-subjects design incorporated positive eWOM and negative eWOM (eWOM valence) and non-referential UGSMC and disabled UGSMC (eWOM absence). The study aimed to recruit participants similar to the demographics of YouTube’s user base to strengthen external validity. Their user base has a diverse age range

with 15 - 35 being the most penetrated age segment and 56+ the least penetrated segment (Statista, 2019b). The present study employed the convenience sampling method on social media to recruit participants who are active on social media and thus potentially representative of YouTube’s user base. To reduce sampling bias, users were encouraged to further distribute the study similar to the snowballing sampling method. The initial sample (n = 214) was reviewed and seven participants were excluded as two participants did not give their consent, three were not over the age of 18, and two could not display the video, reducing the final sample to N = 207. The final sample included more female participants (n = 121) than male participants (n = 84) , and non-binary participants (n = 2), with age ranging from 18 to 57 (M = 26.87, SD = 6.90). As the sample’s mean age lies in between YouTube’s most penetrated age segment, it may be assumed that the sample’s demographic reflects YouTube’s user demographic. The majority of participants completed a Bachelor’s degree (n = 115), followed by a Master’s degree (n = 46), Vocational degree (n = 19), High school degree (n = 18), PhD/Doctorate degree (n = 5), below High school (n = 1), and three specified other educational achievements. The majority of participants resided in the Netherlands (n = 59), followed by Germany (n = 37), The UK (n = 35), the US (n = 17), and other countries with n < 8 per country. Furthermore, on a scale from 1 (“very bad”) to 7 (“very good”) the English language proficiency of the final sample had a mean of M = 6.37, SD = 1, indicating an overall sufficient level of English proficiency to comprehend the presented material. Only n = 2 had an English language proficiency below average. Lastly, the majority of the final sample used a mobile device to participate in the online experiment (n = 147), followed by desktop (n = 51), tablet (n = 6) and three did not specify any device.

Stimuli

A digital camera was chosen for the YouTube brand/product video to ensure high product involvement. This product is associated with monetary risks and thus longer purchase deliberation, and has been used in previous eWOM studies (Gu, Park & Konana; Hayes & Carr, 2015). The brand Aurora and the digital camera Aurora T1 in the YouTube video were fictional to avoid preexisting brand attitudes amongst participants (App. A.). The YouTube video as well as the user comments were initially uploaded on the platform and embedded via HTML into the questionnaire to mimic YouTube’s user interface and user experience to strengthen ecological validity. Each condition had the same amount of displayed video views and a short informative description underneath the video to highlight the technical specifications of the product.

The comment section was digitally altered across the four conditions to best reflect each condition (App. B). The conditions positive, negative, and non-referential featured 10 comments in total which varied in character length within each condition but were comparable in character length across those conditions and were posted in the same order. These measures were taken to ensure a similar amount of required reading time for each condition. All comments were designed to appear as if they had been posted by real YouTube user profiles which were the same across positive, negative, and non-referential conditions. The positive and negative comment sections featured eWOM which was majorly brand-specific and product-brand-specific, and displayed opposing sentiments such as, “awesome quality” versus “terrible quality” as well as one neutral fill comment for ecological validity. The non-referential UGSMC section did not refer to the brand or product in question and aimed to be majorly neutral in sentiment such as, “check out my channel please”. The disabled condition

displayed the standard YouTube notification when the comment section is disabled, “Comments are turned off”.

Procedure

After instructions and informed consent, participants were asked whether they were above the age of 18, and were asked to indicate their English language proficiency on a seven-point scale ranging rom very bad (1) to very good (7). To ensure participants were highly involved in the situation, participants were asked to imagine they had saved up money for a significant purchase and were actively searching for information to compare brands and their products. Participants had to spend at least 10 seconds reading the information before continuing to brief technical instructions to ensure comprehension. Thereafter, participants were randomly assigned to one of the conditions. The video was one minute and thirty seconds long and participants could only continue after two minutes to encourage interaction with the entire YouTube page. Participants were then asked whether they were able to watch the video to determine potential exclusion from further data collection. The questionnaire continued with the main measures; brand trust, brand attitude, and purchase intention, followed by situational involvement and a manipulation check. The remaining questionnaire measured product category involvement, social media experience, and further demographics. The questionnaire and scales for the dependent variables, mediator, covariates, and manipulation check can be found in the appendix (App. C.). After questionnaire submission, participants were debriefed regarding the purpose of the experiment and notified that the brand, product, and user comments were fictional.

Measures

Trust. The hypothesised mediator Trust was measured with five items on a seven-point

Likert scale ranging from ‘Stongly disagree’ (1) to ‘Strongly agree’ (7) introduced by Spark and Browning (2011) and has since been employed in another eWOM study (Ladhari & Michaud, 2015). Example items included, e.g.: ”I have trust in this digital camera” and, “I believe this brand reliable”. The hypothesised mediator Trust (M = 4.36 , SD = 1.35 ) was computed with the mean of the five items, Cronbach’s alpha = .93.

Brand attitude. The hypothesised dependent variable Brand attitude was measured

with a three item, seven-point semantic differential scale adopted from Coyle and Thorsen (2001). The statements were adapted for the purpose of the study, e.g.: “My attitude toward the brand Aurora is unfavourable (1) / favourable (7)” and, “The brand Aurora is bad (1) / good (7)”. The dependent variable Brand attitude (M = 4.54 , SD = 1.23) was computed using the mean of the three items indicating the level of brand attitude ranging from “low” (1) to “high” (7), Cronbach’s Alpha = .88.

Purchase intention. The hypothesised dependent variable Purchase intention was

measured with a three item, seven-point Likert scale ranging from “Extremely unlikely” (1) to “Extremely likely” (7). The scale was previously employed in an eWOM absence study for the same product category (Hayes & Carr, 2015). Example items include, “How likely would you be to purchase this product?” and, “How likely you would you consider the use of this product?”. The hypothesised dependent variable Purchase intention (M = 3.90, SD = 1.56) was computed using the mean of the three items, Cronbach’s Alpha = .93.

Covariates

Involvement. Involvement is a multifaceted construct which may be categorised into

product involvement and situational involvement (Petty, Cacioppo & Schumann, 1983; Kapferer & Laurent, 1993; Zaichkowksy, 1994). Relevant to the present study are high levels of product involvement and situational involvement which have been associated with a strong desire of risk reduction and may therefore result in more extreme consumer responses (Gu, Park & Konana, 2019; Petty, Cacioppo & Schumann, 1983).

Product involvement. The covariate Product involvement was measured using the Consumer Involvement Profile (CIP) established by Laurent and Kapferer (1985). The 15 item, seven-point Likert scale ranging from “Strongly disagree” (1) to “Strongly agree” (7) was adapted to fit the product category digital cameras, e.g.: “Choosing a digital camera is rather difficult” and, “It’s very irritating to buy a digital camera which isn’t right”. Participants scores were constructed using the mean of the 15 items indicating the level of Product involvement ranging from “low” (1) to “high” (7), M = 4.63, SD = .95, Cronbach’s alpha = .87.

Situational involvement. The covariate Situational involvement was measured using the Personal Involvement Inventory (PII) (Zaichkowsky, 1994). The 10 item, seven-point semantic differential scale asked participants to evaluate the YouTube video page with adjectives such as, “unimportant” (1) / “important” (7) and “boring” (1) / “interesting” (7). Participants’ scores were computed using the mean of the 10 items indicating the level of Situational involvement ranging from “low” (1) to “high” (7), M = 4.40, SD = 1.18, Cronbach’s alpha = .92.

Social media experience. Another covariate which may influence consumer responses

regarding eWOM valence and absence is Social media experience. Depending on their level of social media expertise, users show varying degrees of cognitive processing and critical reflection of eWOM. Users with low social media experience may use more cognitive effort to process eWOM and may evaluate it less critically, which therefore may increase the influence of eWOM on consumers responses (Bühler, Murawski & Bick, 2017; López & Sicilia, 2014). Social media experience and internet experience have often been operationalised by asking respondents regarding their weekly usage in hours (Balabanis & Reynolds, 2001; Liu & Shrum, 2009; López & Sicilia, 2014). The present study applied the same measure with higher weekly usage indicating higher levels of Social media experience and lower weekly usage indicating lower levels of Social media experience. Usage per week ranged from 1 hour to 100 hours (M = 18.22, SD = 14.05).

Results

Randomisation, Manipulation, and Covariate Analyses

Randomisation checks. Firstly, to check whether gender and device was comparable

between conditions, two chi-square analyses with eWOM valence/absence as independent variable and gender, and device as dependent variables was conducted. Randomisation regarding gender and device was successful as results showed a non-significant association between eWOM valence/absence and gender, chi-square (6) = 3.10, p = .797, and a non-significant association between eWOM valence/absence and device, chi-square (9) = 10.78, p = .291.

Secondly, to test whether participants’ age was successfully randomised, a one-way ANOVA with eWOM valence/absence as independent variable and age as dependent variable was conducted. The results showed no statistically significant effect of eWOM valence/ absence on age, F (3, 203) = .09, p =.965. Therefore, randomisation regarding age was successful.

Manipulation check. A manipulation check asked participants whether the comment

section below the brand video was mostly positive, negative, non-referential or disabled towards the brand and product. Table 1 shows the distribution of answers according to the condition. In total, 181 participants answered correctly according to their condition, whereas 26 participants did not. The manipulation of eWOM valence/absence was successful with a clear majority of correct answers. It was decided to include all participants in all further analyses to strengthen external validity.

Covariates. Pearson correlation analyses with the three hypothesised covariates, the

covariate should be included in further analyses. Firstly, there were statistically significant, positive, weak to moderate associations between the control variable situational involvement with brand attitude (r (205) = .26 , p < .001), purchase intention (r (205) = .36, p < .001), and trust (r (205) = .23 , p = .001). Therefore, the control variable situational involvement will be included as a covariate in the main analyses.

Secondly, there were both statistically non-significant as well as statistically significant, positive, weak association between the control variable product involvement with brand attitude (r (205) = .12 , p = .087), purchase intention (r (205) = .16, p = .024), and trust (r (205) = .01 , p = .164). As the control variable product involvement is at least significantly associated with purchase intention it will be included as a covariate in the main analyses.

Lastly, there were no statistically significant associations between the control variable social media experience with brand attitude (r (205) = .07 , p = .339), purchase intention (r (205) = .05, p = .483), and trust (r (205) = -.02 , p = .751). Thus, social media experience will not be controlled for in the main analyses.

Main Analyses

Differential effects of eWOM valence/absence on brand attitude and purchase

intention. In order to test whether eWOM valence/absence had an effect on both brand

attitude and purchase intention (H1), a 1 x 4 between-subjects MANCOVA was performed.

The MANCOVA showed a main effect of eWOM valence/absence on the combined dependent variables, F (6, 400) = 18.52, p < .001, np2 = .22, with Wilk’s lambda as a criterion, after controlling for situational involvement and product involvement. This effect was moderate in size with the factor explaining 22% of the variance in the combined dependent variables. Situational involvement was a significant adjustor of the combined

dependent variables, whereas product involvement was not (F (2, 200) = 17.02, p < .001 , np2 = .15; F (2, 200) = .11, p = .899, np2 < .01).

Subsequent between-subjects ANCOVAS revealed that EWOM valence/absence showed a significant effect on brand attitude, F (3, 201) = 26.98, p < .001 , np2 = .29, and a significant effect on purchase intention F (3, 201) = 38.82, p < .001 , np2 = .37. The effect of eWOM valence/absence on brand attitude is strong in size and explains 29% of the variance in the dependent variable. The effect on purchase intention is also strong in size and explains 37% of the variance in purchase intention. Situational involvement was a significant adjustor of brand attitude (1, 201) = 15.30, p < .001 , np2 = .07, and of purchase intention F (1, 201) = 34.13, p < .001 , np2 = .15. Product involvement was not a significant adjustor of both brand attitude F (1, 201) = .06, p = .802 , np2 < .01, and of purchase intention F (1, 201) = .21, p = .645 , np2 < .01.

Pairwise comparisons with bonferroni correction provided insight into the mean differences for eWOM valence/absence on brand attitude and purchase intention using the estimated marginal means after estimating out product involvement and situational involvement. Table 2 shows the estimated marginal means and standard errors for both dependent variables per condition and significant pairwise comparisons.

Firstly, those exposed to negative eWOM had more negative brand attitudes and purchase intentions than those exposed to positive eWOM, these mean differences were statistically significant (mdiff = -1.55, p < .001, 95% CI [-2.08, -1.02]; mdiff = -2.28, p < .001, 95% CI [-2.89, -1.67]). Therefore, H1a is confirmed as eWOM valence had an effect on

participants exposed to positive eWOM than for those exposed to negative eWOM, and vice versa.

Secondly, participants in the disabled UGSMC condition had more positive brand attitudes and purchase intentions than those exposed to negative eWOM, these mean differences were both statistically significant (mdiff = 1.22, p < .001, 95% CI [.68, 1.76]; mdiff = 1.63 , p < .001, 95% CI [1.00, 2.51]). Furthermore, those exposed to disabled UGSMC had more negative purchase intentions than those having read positive eWOM, mdiff = -.66, p = .038, 95% CI [-1.29, -.02]. However, there was no statistically significant difference in participants’ brand attitudes between those confronted with disabled UGSMC and those having read positive eWOM, mdiff = -.33, p = .664, 95% CI [-.88, .22]. Therefore, H1b is partially confirmed as eWOM absence had an effect on the combined dependent

variables, with more positive brand attitudes and purchase intentions for disabled UGSMC vs. negative eWOM, but only more negative purchase intentions for disabled UGSMC vs. positive eWOM and not for brand attitude.

Thirdly, there were no statistically significant mean differences in participants’ brand attitudes and purchase intentions between those confronted with non-referential UGSMC and those being confronted with disabled UGSMC, mdiff = .29, p = .950, 95% CI [-.26, .84]. There was also no significant difference for both brand attitude and purchase intention when comparing non-referential UGSMC with positive eWOM, mdiff = -.04, p = 1.000, 95% CI [-.57, .49]. However, those exposed to non-referential UGSMC had more positive brand attitudes and purchase intentions than those having read negative eWOM, these mean differences were statistically significant, mdiff = 1.51 , p < .001 , 95% CI [.99, 2.04]; mdiff = 1.96, p < .001 , 95% CI [1.35, 2.57]. Based on these findings H1c is partially rejected. Although eWOM absence had an effect on the combined dependent variables and despite

those exposed to non-referential UGSMC underneath the video reporting more positive brand attitudes and purchase intentions than those shown negative eWOM, there were no differences for both dependent variables comparing non-referential UGSMC with positive eWOM, and with disabled UGSMC. Figures 1 and 2 show the adjusted marginal means and significant mean differences for brand attitude and purchase intention across conditions plotted against each other.

Trust as a mediator of the relationship between eWOM valence/absence with

brand attitude and purchase intention. In order to test whether trust mediated the

relationship between eWOM valence/absence and brand attitude, and purchase intention (H2 ,), PROCESS macro for SPSS (Hayes, 2019) was employed to run two model 4 tests. The

independent variable was dummy coded with disabled UGSMC as the reference group due to eWOM valence/absence being multi-categorical. Thus, the two tests included three comparison groups, trust as the mediator, brand attitude and purchase intention tested separately as dependent variables, whilst controlling for situational involvement and product involvement. Figure 3 shows the mediation path diagram for the first test with brand attitude as the dependent variable.

Mediation analysis with brand attitude as the dependent variable. The first PROCESS model 4 test used brand attitude as the dependent variable. Comparing disabled UGSMC with positive eWOM (b = .49, t (201) = 2.25, p = .026) and with negative eWOM (b = -1.52, t (201) = -7.12, p < .001 ) had an effect on trust. The coefficients indicated that when comparing disabled UGSMC with positive eWOM, levels of trust are more positive for positive eWOM, and when comparing disabled UGSMC with negative eWOM, levels of trust

are more negative for negative eWOM. There was no effect when comparing disabled UGSMC with non-referential UGSMC, b = .22, t (201) = 1.00, p = .319. The covariate situational involvement significantly predicted trust, b = .26, t (201) = 3.61, p < .001, whereas the covariate product involvement did not, b = -.04, t (201) = -.46, p = .648.

Subsequently, results also showed that trust had a positive effect on brand attitude, b = .72, t (200) = 16.33, p < .001. Therefore, the more positive consumers’ trust in the brand was, the more positive were their attitudes towards the brand, and vice versa. When adding trust to the model, there was no direct effect of eWOM valence and absence when comparing disabled UGSMC with positive eWOM (b = -.02, t (200) = -.14, p = .887), with negative eWOM (b = -.13, t (200) = -.87, p = .384), and with non-referential UGSMC (b = .14, t (200) =1.00, p = .316). The same pattern occurred for situational involvement, b = .08, t (200) = 1.67, p = .096, and product involvement, b = .01, t (200) = .14, p = .885. The subsequent model revealed that the relationship between eWOM valence/absence and brand attitude was mediated by trust only when comparing disabled UGSMC with positive eWOM, ab = .35, SE = .16, 95% CI [.05, .66], as well as when comparing disabled UGSMC with negative eWOM, ab = -1.09, SE = .19, 95% CI [-1.46, -.73]. Mediation did not occur when comparing disabled UGSMC with non-referential UGSMC, ab = .15, SE = .16, 95% CI [-.16, .47].

Mediation analysis with purchase intention as the dependent variable. The second PROCESS model 4 test replaced brand attitude with purchase intention. Figure 4 shows the mediation path diagram for this model.

The effect of eWOM valence and absence on trust was the same as in the first model. The results further showed a positive effect of trust on purchase intention, b = .68, t (200) = 11.04, p < .001. As trust increased, so did the intent to purchase. When adding trust to the model, there was no effect when comparing disabled UGSMC with positive eWOM (b = .33, t (200) = 1.72, p = .087) and with non-referential UGSMC (b = .19, t (200) =1.02, p = .31) on purchase intention. However, the effect persisted when comparing disabled UGSMC with

negative eWOM (b = -.60, t (200) = -2.87, p = .005). For the covariates’ effect on purchase intention when adding trust to the model, only the effect of situational involvement (b = .29, t (200) = 4.44, p < .001) persisted and the effect of product involvement (b = -.02, t (200) = -.23, p = .820) did not persist. Lastly, the relationship between eWOM valence/absence and purchase intention was mediated by trust only when comparing disabled UGSMC with positive eWOM, ab = .33, SE = .15, 95% CI [.04, .64], as well as when comparing disabled UGSMC with negative eWOM, ab = -1.03, SE = .19, 95% CI [-1.44, -.67]. Mediation did not occur when comparing disabled UGSMC with non-referential UGSMC, ab = .15, SE = .16, 95% CI [-.15, .46].

Therefore, H2 is partially confirmed. Positive eWOM resulted first in more positive

levels of trust and then more positive brand attitudes and purchase intentions than disabled UGSMC. The opposite is true for negative eWOM which first induced more negative levels of trust and thereafter more negative brand attitudes and purchase intentions than disabled UGSMC. When comparing non-referential UGSMC with disabled UGSMC, trust did not mediate the relationship.

Conclusion and Discussion

An online experiment was utilised to study the effects of eWOM valence (positive eWOM and negative eWOM) and eWOM absence (non-referential UGSMC and disabled UGSMC) underneath YouTube brand content on brand attitude, purchase intention as well as on the mediating role of trust. This approach enabled a complete comparison between four

states of eWOM valence and absence and thus a synthesis of separate models used in previous studies (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). As expected, results showed that when participants read negative eWOM in comparison to when the comment section was disabled, they had more negative brand attitudes and purchase intentions due to trusting the brand less. Against expectations, when participants read positive eWOM in comparison to when the comment section was disabled, participants only had more positive purchase intentions but not more positive brand attitudes due to trusting the brand more. Although non-referential UGSMC and disabled UGSMC caused similar responses and more positive responses in comparison to negative eWOM, the study cannot conclude that the mere absence of eWOM has a universal impact on consumer responses. Disabled UGSMC resulted in lower purchase intentions in comparison to positive eWOM, whereas positive eWOM and non-referential UGSMC did not differ for both brand attitude and purchase intention. Therefore, disabling UGSMC has an advantage over negative eWOM and a disadvantage over positive eWOM.

Discussion

Differential effects comparing positive eWOM, negative eWOM, non-referential

UGSMC, and disabled UGSMC. For the comparison of eWOM valence conditions results

showed that negative eWOM resulted in lower brand attitudes and purchase intentions, whereas positive eWOM had the opposite effect. This is a common finding across previous studies investigating both solely eWOM valence (positive vs. negative) and comparing positive eWOM and negative eWOM with disabled UGSMC (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015; Ismagilova, Slade, Rana & Dwivedi, 2019; Jha, 2019; King, Racherla & Bush, 2014; Ladhari & Michaud, 2015; Lee, Rodgers & Kim, 2009; Roy, Datta

& Mukherjee, 2019). In the light of information integration theory, the results support the notion that users may assign an overall positive or negative value to all comments which in turn impacts consumer responses in the same direction as the overall eWOM valence indicates (López & Sicilia, 2013). Another finding revealed that only negative eWOM differed from all other conditions for both brand attitude and purchase intention. This may be explained with the negativity bias which posits that negative information is superior to other information in causing stronger consumer responses (Basuory, Chatterjee & Ravid, 2003; Rozin & Royzman, 2001).

For the comparison of eWOM valence conditions with disabled UGSMC, participants confronted with disabled UGSMC were less inclined to purchase than when confronted with positive eWOM. This was not the case for brand attitude where expected differences could not be validated. In comparison to negative eWOM, participants confronted with disabled UGSMC were more inclined to purchase and had higher brand attitudes. These results only partially align with previous findings by Hayes & Carr (2015) who compared positive eWOM with disabled UGSMC on a blog. Their results showed more positive purchase intentions and brand attitudes for positive eWOM and the opposite for disabled UGSMC (Hayes & Carr, 2015). For purchase intention, both the present and previous findings may align with uncertainty reduction theory when comparing positive eWOM and disabled UGSMC (Berger & Calabrese, 1975). When disabling UGSMC, consumers cannot reduce uncertainty before purchase through engaging in confirmatory interpersonal communication and hence may be less inclined to purchase (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger & Calabrese, 1975; Sarmiento Guede, Esteban Curiel & Antonovica, 2018). Different results for brand attitude in comparison to Hayes & Carr (2015) may be explained by differences in the manipulation. In their study, positive eWOM was combined with a reply

written by a blogger. Thus, positive eWOM was supported through a source external from the brand which thereby may have added to source credibility, a predictive construct for eWOM effects able to yield more favourable consumer responses when the source is deemed highly credible (López & Sicilia, 2013; Ismagilova, Slade, Rana & Dwivedi, 2019; Jalilvand, Esfahani & Samiei, 2011). External sources with no affiliation to a brand are considered to be more credible than brands themselves (King, Racherla & Bush, 2014). Source credibility may have increased the difference significantly between positive eWOM and disabled UGSMC for brand attitude in the previous study and may explain why band attitude did not differ in the present study (Hayes & Carr, 2015).

For the comparison of eWOM absence conditions, non-differential UGSMC and disabled UGSMC, results showed no difference for both brand attitude and purchase intention. This is a novel finding as previous eWOM absence studies did not attempt to include non-referential UGSMC (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). However, its inclusion may be imperative to conclude whether eWOM absence itself causes certain responses or whether there are differences in the effects of disabled UGSMC and non-referential UGSMC. The finding itself may partially be explained with uncertainty reduction theory (Berger & Calabrese, 1975). In the non-referential UGSMC condition, participants were able to look for information in the comments which was not possible when UGSMC was disabled (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger & Calabrese, 1975). However, the information at hand may have not been sufficient to reduce uncertainty entirely as it may have been meaningless to the consumer due to not containing any confirmatory or opposing information regarding the brand and product (King, Racherla & Bush, 2014). This may explain similar consumer responses due to their uncertainty for both non-referential UGSMC and disabled UGSMC (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger &

Calabrese, 1975). Although both eWOM absence conditions resulted in the same levels of brand attitude and purchase intention, and both had more positive consumer responses than negative eWOM, they did cause different consumer responses in comparison to positive eWOM. Disabled UGSMC caused lower purchase intentions than positive eWOM whereas non-referential UGSMC did not. This alludes to a difference in the effects of the two eWOM absence categories and thus indicates an advantage to disable comments over negative eWOM and a disadvantage over positive eWOM.

For the comparison of non-referential UGSMC with positive eWOM, results did not confirm expectations as participants did not respond differently. This appears to defy the notion of information integration theory as non-referential UGSMC in the experiment was low in valence and therefore should not have been evaluated as majorly positive or negative (López & Sicilia, 2013). It also appears to defy the notion of uncertainty reduction theory because non-referential UGSMC supposedly cannot reduce uncertainty as it does not contain any confirmatory or opposing information (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger & Calabrese, 1975). The reason why consumer responses for disabled UGSMC and non-referential UGSMC as well as non-referential UGSMC and positive eWOM do not differ, but purchase intention does differ for disabled UGSMC and positive eWOM, may be caused by two tendencies. Firstly, previous findings show that in computer-mediated domains consumers are more satisfied when they are able to communicate interpersonally, as it is the case for both positive eWOM and non-referential UGSMC, regardless of valence or content (Sarmiento Guede, Esteban Curiel & Antonovica, 2018). However, the valence of positive eWOM may have led to a slight but insignificant positive difference in consumer responses versus referential UGSMC (López & Sicilia, 2013). Secondly, although both non-referential UGSMC and disabled UGSMC may not contain confirmatory or opposing eWOM

to reduce uncertainty, disabled UGSMC may not satisfy the need for interpersonal communication which may have led to a slight but insignificant negative difference versus non-referential UGSMC (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger & Calabrese, 1975; Sarmiento Guede, Esteban Curiel & Antonovica, 2018). Therefore, the combination of both tendencies may have been large enough for positive eWOM and disabled UGSMC to differ, at least regarding purchase intention. But they may have not been large enough for non-referential UGSMC versus disabled UGSMC, and for non-referential UGSMC versus positive eWOM to differ for both brand attitude and purchase intention.

Differential effects of trust as the mediator. Results indicated that positive eWOM

versus disabled UGSMC first led to more positive versus more negative levels of trust and thereafter more positive versus more negative purchase intentions and brand attitudes, respectively. The opposite was true for negative eWOM versus disabled UGSMC which first led to more negative versus more positive levels of trust and thereafter more negative versus more positive brand attitudes and purchase intentions, respectively. These results align with previous findings by Bühler, Murawski & Bick (2017) who compared negative eWOM with disabled UGSMC and in part found more negative levels of trust for negative eWOM. Additionally, these findings confirm previous suggestions that trust may be a more indicative mediator than credibility for eWOM valence and absence (Hayes & Carr, 2015). These results may also align with empirical and theoretical suggestions that an increase in trust, as in the case of positive eWOM, leads consumers to be less uncertain about their attitude toward the brand and less uncertain about their intent to purchase (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger & Calabrese, 1975; Hong & Cha, 2013). In the case of negative eWOM, a decrease in trust may have led consumers to be more uncertain to

purchase and more uncertain about their attitude regarding the brand as decisions in computer-mediated conditions are generally associated with more risks (Antheunis, Schouten, Valkenburg & Peter, 2012; Berger & Calabrese, 1975; Hong & Cha, 2013). Additionally, these results partially corroborate the notion of warranting theory and the definition of trust (Blomqvist, 1997; Pavlou & Stewart, 2013; Walther & Parks, 2002). Disabled UGSMC may have been lower in warranting value as the brand did not allow any other-generated information which is suggested to be more trustworthy (Walther & Parks, 2002). Therefore, positive eWOM may have been higher in warranting value and would have met two central tenets of trust, reliance and expectancy (Blomqvist, 1997; Pavlou & Stewart, 2013). In the case of positive eWOM, consumers may have trusted the brand more because it could be relied upon to meet the expectations it set out in the video which thereby may have resulted in higher brand attitudes and purchase intentions (Blomqvist, 1997; Pavlou & Stewart, 2013). Although negative eWOM may be higher in warranting value than disabled UGSMC, it may have resulted in less trust because the expectations generated through brand and product promises could not be relied upon which thereby may have caused lower brand attitudes and purchase intentions (Blomqvist, 1997; Pavlou & Stewart, 2013; Walther & Parks, 2002). Additionally, this effect may have been strengthened due to the negativity bias as negative information is suggested to be particularly impactful in influencing consumer responses (Basuory, Chatterjee & Ravid, 2003; Rozin & Royzman, 2001).

Against expectations, trust was not a mediator when comparing non-referential UGSMC with disabled UGSMC. This may be explained similarly to the findings for brand attitude and purchase intention when comparing the two conditions. Although disabled UGSMC may have been lower in warranting value than non-referential UGSMC, both disabled UGSMC and the content of non-referential UGSMC neither confirm nor reject the

expectations as promised in the video because they do not state anything related to the brand or product (Blomqvist, 1997; Pavlou & Stewart, 2013; Walther & Parks, 2002). Therefore, in both conditions it may have been unclear to the participants whether the brand can be relied upon and thus trusted, thereby resulting in similar levels of trust.

Limitations and Future Research

Limitations. It should be noted that the present study has certain limitations. Due to

technical constraints, the study could not allow users to interact with the comment section such as commenting, liking, subscribing or replying which may have limited overall interactivity. When consumers are highly involved, interactivity has been shown to have the ability to enhance or reduce the experience of a highly interactive website and interactive advertisement depending on the level of users’ internet experience (Liu & Shrum, 2009; Rodgers & Thorson, 2013). Liu and Shrum (2009) suggested with the dual-process model of interactivity effects that if users have high internet experience, central processing of the presented information is enhanced which may result in more positive brand attitudes. In comparison, when users have low social media experience they cannot accurately process the information as the interactive features overwhelm them which may lead to more negative brand attitudes (Liu & Shrum, 2009). Therefore, consumer responses may have differed if full interactivity on the YouTube page would have been enabled.

Additionally, the present study used non-referential UGSMC which was majorly neutral in valence so it could be compared with disabled UGSMC to exclude positive or negative valence as a confounding factor. However, non-referential UGSMC may also be positive or negative in an online setting (Jha, 2019). Information integration theory suggests that valence may impact consumer responses, which may cause positive or negative non-referential

UGSMC to have an impact on brand attitude and purchase intention (Ismagilova, Slade, Rana & Dwivedi, 2019; López & Sicilia, 2013). Therefore, the present study did not acknowledge the potential effect of non-referential UGSMC valence on consumer responses.

Future research. To address these limitations, future research should allow full

interactivity whilst controlling for social media experience or incorporating social media experience as a moderator. Such an approach may show different results to the present study as full interactivity is enabled. Furthermore, limitations of the present study may be addressed by incorporating non-referential UGSMC valence (positive, neutral, and negative) and comparing it with eWOM valence as well as with disabled UGSMC. This comparison would offer a complete picture whilst taking into the account the possibility of non-referential UGSMC valence.

Furthermore, this field of research can be enriched by comparing eWOM absence and valence on dynamic and rich media versus on static and lean media in one singular study. Previous studies testing disabled UGSMC majorly focussed on static and lean retrieval media (Bühler, Murawski & Bick, 2017; Hayes & Carr, 2015). By making this comparison, future research may be able to test whether unexpected findings in the present study may be due to differences in functional and structural characteristics of the platforms. Lastly, future research should also employ a within-subjects design which tests the impact of disabling the comment section after positive and negative eWOM has already been read by users to see whether a change in consumer responses may occur. In the present study participants were only exposed to one of the conditions. It is therefore unclear whether consumers may express different reactions if the comment section contained eWOM or UGSMC at first and then was disabled.