Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

European Journal of Information Systems

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tjis20

Negativity bias in the diagnosticity of online


review content: the effects of consumers’ prior
experience and need for cognition

Hamed Qahri-Saremi & Ali Reza Montazemi

To cite this article: Hamed Qahri-Saremi & Ali Reza Montazemi (2023) Negativity bias
in the diagnosticity of online review content: the effects of consumers’ prior experience
and need for cognition, European Journal of Information Systems, 32:4, 717-734, DOI:
10.1080/0960085X.2022.2041372

To link to this article: https://doi.org/10.1080/0960085X.2022.2041372

Published online: 10 Mar 2022.

Submit your article to this journal

Article views: 1394

View related articles

View Crossmark data

Citing articles: 6 View citing articles

Full Terms & Conditions of access and use can be found at


https://www.tandfonline.com/action/journalInformation?journalCode=tjis20
EUROPEAN JOURNAL OF INFORMATION SYSTEMS
2023, VOL. 32, NO. 4, 717–734
https://doi.org/10.1080/0960085X.2022.2041372

RESEARCH ARTICLE

Negativity bias in the diagnosticity of online review content: the effects of


consumers’ prior experience and need for cognition
a b
Hamed Qahri-Saremi and Ali Reza Montazemi
a
College of Business, Colorado State University, Fort Collins, Colorado, USA; bDeGroote School of Business, McMaster University, Ontario,
Hamilton, Canada

ABSTRACT ARTICLE HISTORY


The importance of online review valence is a subject of debate among scholars. Prior studies Received 12 January 2021
mostly assumed valence as a “peripheral” cue derived from online review surface features (e.g., Accepted 3 February 2022
star ratings). This assumption has important implications as it restricts the negativity bias KEYWORDS
effects to a certain group of consumers who lack pertinent prior experience with the pro­ Negativity bias; online
duct/service domain and the motivation to assess the product/service. Focusing on online reviews; need for cognition;
service context and drawing on an adaptational view to negative information, we investigate prior experience; elaboration
the negativity bias in the effects of the valence of the “content” of online reviews on con­ likelihood model; online
sumers’ attitudes and show that it can be attributed to the higher perceived diagnosticity of service
negative reviews. This is determined by consumers’ in-depth elaborations of reviews’ contents,
which are contingent on their prior experience with the domain of online service and need for
cognition. Our findings provide a new perspective to negativity bias by showing that more
experienced and thoughtful consumers are also influenced by negativity bias when the
content of online reviews is considered. This is a novel account of negativity bias in the effects
of online reviews that underscores the importance of response strategies for reducing their
adverse effects.

1. Introduction (Chen & Lurie, 2013). Indeed, if not addressed effec­


While seeking other consumers’ personal experi­ tively, NRs can spark an “online firestorm” which can
ences with products/services for purchase decisions, spread at an exponential rate reaching thousands or
traditionally known as word of mouth (WoM), is millions of consumers in a short period of time. An
not a new phenomenon, online platforms, such as online firestorm entails a sudden discharge of large
Yelp and TripAdvisor, have amplified the popular­ quantities of negative messages and complaints
ity and importance of WoM for consumers and against a firm on social media platforms (Pfeffer
providers to unprecedented levels by creating the et al., 2014). Therefore, firms must detect and ade­
online version of WoM: “online reviews” (also quately respond to NRs in online platforms to avoid
referred to as “electronic WoM” and “eWoM”). potential public debacles, online rating declines, cus­
These platforms have facilitated ubiquitous and tomer defections, and profit reductions (Curchod
convenient access to a large number of online et al., 2020; Pfeffer et al., 2014). To that end, it is
reviews about almost any product/service. As important for firms to understand how NRs influence
a result, compared to offline WoMs, online reviews their consumers and when their impacts on consu­
are profoundly more influential on consumers’ mers are stronger than those of PRs (i.e., “negativity
decisions. A recent survey found that in 2020, bias”).
87% of consumers consulted with online reviews Despite the relative importance of NR, prior
before choosing a local business, and 79% of con­ research findings about its asymmetric effects on con­
sumers trusted online reviews as much as personal sumers’ opinions and decisions (i.e., negativity bias)
recommendations from their friends and family are mixed (Qahri-Saremi & Montazemi, 2019, 2020).
(Murphy, 2020). While some prior works have found support for the
However, not all online reviews are equal in their negativity bias in the effects of online reviews (e.g.,
impacts on consumers’ decisions. In this vein, an Chen & Lurie, 2013), others have not been able to find
ongoing debate among scholars concerns the “nega­ this effect (e.g., Wu, 2013). In most of these studies,
tivity bias” in the effects of online reviews that happens valence (i.e., positivity and negativity) of an online
when negative reviews (NR) have larger and/or more review has been implicitly or explicitly assumed as
lasting impacts on consumers’ decisions as compared a “peripheral cue” (a heuristic) simply derived from
to positive reviews (PR) of comparative magnitude the surface features of an online review without giving

CONTACT Hamed Qahri-Saremi hamed.qahri-saremi@colostate.edu


© Operational Research Society 2022.
718 H. QAHRI-SAREMI AND A. R. MONTAZEMI

much considerations to the contents of the message firms to increase their sales, but also more influential
(Qahri-Saremi & Montazemi, 2019). Indeed, in these on other potential consumers’ decisions to adopt
studies, “star ratings” has been predominantly used as a firm’s product/service (Boyer & Hult, 2006).
the basis for determining the positivity and negativity Extant online review literature has paid little atten­
of online reviews. As an aggregated measure that tion to the conditions and pertinent factors that can
serves as a heuristic, “star ratings can reduce cognitive influence the negativity bias effects based on the actual
processing effort” (Nazlan et al., 2018, p. 451); there­ contents of online reviews. In this paper, we address
fore, by relying on star ratings, prior studies on online this important matter by extending the concept of
review valence have predominantly assumed negativ­ negativity bias effects beyond simple star ratings and
ity bias as a peripheral effect. On this basis, a common to the valence of the online review content. By masking
explanation in studies that could not detect the nega­ the star rating effects in our study, we assess the
tivity bias effects is that valence of an online review as asymmetric effects of NR (versus PR) reflected in the
a simple peripheral cue loses its significance for contents of the online reviews on consumers’ atti­
a certain group of consumers who pay more attention tudes, in the pre-usage stage of online experience
to the “central” cues related to the content of an online services, when no firsthand use experience with the
review, such as its quality (Wu, 2013). online service is available. To that end, our research
Assuming the valence of an online review as objective is to investigate whether the negativity bias
a peripheral cue has important theoretical and practi­ based on the contents of online reviews is attributed to
cal implications. As explained by dual-process the­ the differences in the perceived diagnosticity of NRs for
ories, such as elaboration likelihood model (ELM; different consumers. For doing so, we take an “adapta­
Petty & Wegener, 1998), only a certain group of con­ tional” view of information as explained in prior IS re-
sumers who lack either pertinent experience with the search (Lee,2010; McKinney Jr. & Yoos II, 2010). We
product/service domain or the motivation to assess contend that negativity bias occurs when the negative
a product/service (that is they are either novice or information within an online review is a “perceived
uninvolved) rely on peripheral cues for decision- ‘difference that makes a difference’ (Bateson, 1973) to
making. Thus, based on this assumption, it can essen­ a subject” (McKinney Jr. & Yoos II,2010, p. 336).
tially be inferred that negativity bias should only Perceived diagnosticity of NRs for different consumers
impact the decisions of novice and uninvolved can be a function of their in-depth elaborations of the
consumers. online reviews’ contents. On this basis and drawing on
Nevertheless, a separate stream of research, outside the ELM and extant negativity bias literature, we
of online review literature, has found that consumers hypothesize that the negativity bias in the effects of
who are more experienced and involved with online reviews content is a function of consumers’
a decision are indeed more affected by negativity bias depth of elaboration of the online review messages,
in the contents of the messages related to their deci­ which is contingent on their prior experience with the
sion (e.g., Ahluwalia, 2002; Lane et al., 2021; Rozin & domain of the product/service and their level of need
Royzman, 2001). One major difference in this stream for cognition. We validate our hypotheses via a ran­
of research as compared to the (majority of) extant domized controlled experiment. Our findings provide
online review research is their focus on the “content” a novel, counter-intuitive, and complementary
of the message, rather than peripheral cues around it account of the negativity bias in the effects of online
(e.g., star ratings in online review context), for inves­ reviews and help to explain some of the inconsistent
tigating the negativity bias effects. Not only are their findings in the extant literature in this regard.
findings antithetical to the predominant assumption
in online review literature about the peripheral nature
of negativity bias, but also markedly change its impor­ 2. Theory and hypotheses
tance for a different and arguably more important
2.1. Negativity bias
group of consumers, who are experienced in and
involved with a product/service. While negative Research in decision-making has shown that people
valence as a peripheral cue only affecting novice and generally give higher weight to negative information
uninvolved consumers may not invoke firms to devise than positive information of comparative magnitude
strategies to respond to the NRs about their products/ and that the former tends to invoke greater informa­
services, negativity as a central cue impacting the tion processing and lasts longer in the memory (Rozin
decisions of experienced and involved consumers is & Royzman, 2001). This effect, called negativity bias, is
sufficiently crucial for firms to devise clear response inherently a comparative notion as it is the function of
strategies to contain the adverse consequences of NRs the weight and importance of negative information in
(Casado-Díaz et al., 2020; Wei et al., 2013; Yu et al., the formation of attitudes and judgements relative to
2021). The reason is that experienced and involved those of positive information (Rozin & Royzman,
consumers are not only more important targets for
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 719

2001). Prior studies in psychology (e.g., Ito et al., 1998) messages. This happens because consumers often
found that the human brain responds more strongly to describe negative experiences in greater details
negative than positive information, and retains the than positive experiences. Moreover, the risk averse
memory of negative information longer, even when nature of most humans leads them to pay closer
the behavioural response to the stimuli is over. attention to negative messages to avoid and miti­
Negativity bias can arise due to negative infor­ gate the risks associated with their decisions. As
mation’s greater veracity, specificity, diagnosticity, a result, they find negative information that inform
and contagiousness than positive information them of the risks associated with their decisions as
(Rozin & Royzman, 2001). A common theoretical more diagnostic than positive information, which
explanation for the negativity bias effect is based on highlights the benefits.
the “category diagnosticity approach”, which Extrapolating the foregoing arguments to the
explains that consumers’ judgements and attitudes context of online reviews, we expect that the
towards objects are based on placing them into lower frequency of NRs as well as their specificity
different categories using existing informational and focus on risks (rather than benefits) of consu­
cues about the objects (Skowronski & Carlston, mers’ decisions lead consumers to perceive them as
1989). It further explains that some informational more diagnostic information for assessing pro­
cues are perceived as more diagnostic than others in ducts/services than their more common counter­
distinguishing between different categories and that parts, PRs. This higher diagnosticity can lead
more diagnostic informational cues will have consumers to devote more attention and informa­
a stronger influence on consumers’ judgements tion-processing resources and assign more weight
and leave more lasting traces in their memory to NRs than to PRs in judgements and impression
than less diagnostic information (Skowronski & formation. Therefore, the asymmetrical effect of
Carlston, 1989). NRs (i.e., the negativity bias) is a function of their
Drawing on the category diagnosticity approach, perceived diagnosticity.
prior research shows that negative information can The diagnosticity judgement is a subjective assess­
be perceived as more diagnostic than positive infor­ ment contingent on contextual as well as individual
mation for categorising objects into evaluative cate­ factors (Ahluwalia, 2002; Ketelaar et al., 2015). A piece
gories because social norms lead consumers to provide of information that is perceived as diagnostic for one
positive information about products/services more consumer in a specific context may not be perceived as
frequently than negative information, making the for­ such by another consumer or in a different context. To
mer more prevalent and the latter rarer (Chen & Lurie, this end, several studies (e.g., Lane et al. (2021) and
2013; Ketelaar et al., 2015). Therefore, “negative Ahluwalia (2002)) have shown that when decision
reviews may have greater ‘surprise’ value to readers” makers critically process messages – i.e., “high infor­
(Yin et al., 2012, p. 2). This is also consistent with the mation elaboration”, negative messages are perceived
“Pollyanna Hypothesis” that “there is a universal as more diagnostic than the comparative positive
human tendency to use evaluatively positive (E+) information, leading to stronger negativity bias effects.
words more frequently, diversely and facilely than For example, Lane et al. (2021) show that experts are
evaluatively negative (E-) words. Put even more sim­ more likely to engage in deep processing and elabora­
ply, humans tend to ‘look on (and talk about) the tion of information when they are presented with
bright side of life’” (Boucher & Osgood, 1969, p. 1). negative information as compared to positive infor­
In other words, “as most products, whether low-, mation. Moreover, Ahluwalia (2002) shows that con­
average-, or high-quality, have positive attributes, sumers who desire to hold attitudes and beliefs that are
negative ones are much rarer. Hence, negative product objectively valid and care about the accuracy of infor­
attributes are considered to be more characteristic of mation find the negative messages as more diagnostic
poor-quality products than positive attributes are for than the positive messages, leading to stronger nega­
high-quality products” (Ketelaar et al., 2015, p. 653). tivity bias effects on their judgements as compared to
This position has been further corroborated by the fact other consumers. These findings are consistent with
that a positivity bias is observed in limited occasions Taylor’s (1991) view that negative information stimu­
where positive events are rarer than negative events lates a special set of cognitive processes designed to
(e.g., manifestations of exceptional human ability; cope with the threat that involves elaborate informa­
Rozin & Royzman, 2001). tion processing. Similarly, Peeters and Czapinski
In addition, prior findings in psychology litera­ (1990) argue that, compared to positive information,
ture (Feldman, 1999; Skowronski & Carlston, 1987) negative information can have greater veracity which
explain that negative messages are conceived as would require more cognitive elaboration to be eli­
more diagnostic because they are typically more cited. Against this backdrop, in this paper, we investi­
specific and convey more information than positive gate the role of information elaboration in negativity
720 H. QAHRI-SAREMI AND A. R. MONTAZEMI

bias in the effects of online reviews. We draw on the However, negativity bias is attenuated, eliminated,
ELM to theorise about the contextual and individual or even reversed when consumers only rely on the
factors that can alter the negativity bias in the effects of surface features of a message and its source in deci­
online reviews. sion-making and the perceived diagnosticity of the
message content is less important in formation of
their judgements (Ahluwalia, 2002; Skowronski &
2.2. Elaboration likelihood model and negativity
Carlston, 1987). This occurs when conditions favour
bias
the dominant use of the peripheral route of informa­
ELM explains that perceptions, impressions, and jud­ tion processing for decision making (Petty &
gements can be formed via two conceptually distinct Wegener, 1998). As a result, in the peripheral route
information processing routes representing the two of information processing, the asymmetrical effect of
ends of the elaboration likelihood continuum: central negative information – i.e., the negativity bias effect –
and peripheral information processing (Petty & is less evident.
Wegener, 1998). When the elaboration likelihood is Given the importance of central versus peripheral
high, people process and evaluate the object-relevant route of message processing for negativity bias, the
information based on their prior experience with the factors that can determine consumers’ route of mes­
sage processing can arguably be important determi­
domain of the object (i.e., “attribute-oriented
nants of negativity bias. What directs consumers
thoughts” (Sujan, 1985)) which would result in the
towards the central versus the peripheral route of
formation of perceptions about the object that are
information processing is contingent on contextual
well articulated and bolstered by their prior experi­
as well as individual factors that shape consumers’
ence. This is referred to as the central route of informa­
ability and motivation to elaborate on the content of
tion processing in ELM. However, when the a given message (Petty & Wegener, 1998). In particu­
elaboration likelihood is low, the extent of processing lar, the ELM, like other dual-process theories, argues
and scrutiny of information is significantly reduced that because human cognitive resources are limited,
and people’s perceptions about the focal object are people cannot process the details of every single mes­
formed as a result of less resource-demanding heur­ sage they receive in-depth.
istics (i.e., “simple evaluative thoughts” (Sujan, 1985)) Generally, the amount of information processing
that do not require effortful processing of the object- effort expended is a function of an individual’s ability
relevant information. This is referred to as the periph­ and motivation to process a given message (Petty &
eral route of information processing in ELM (Petty & Wegener, 1998). Consumers with higher ability and
Wegener, 1998). motivation to process a given message are more likely
Prior studies on the effects of message valence to follow the central route of processing (i.e., higher
explain that when the elaboration likelihood is high elaboration likelihood). Considering their need for
(i.e., central route of information processing in ELM), more diagnostic information, these consumers are
negative messages are perceived more diagnostic and more likely to perceive negative information as diag­
bear more influence on consumers’ decisions than nostic and persuasive, given the less common and
positive messages (Ahluwalia, 2002; Ketelaar et al., more novel, specific, and risk-focus nature of negative
2015; Lane et al., 2021; Peeters & Czapinski, 1990). information as compared to positive information,
This is because, in the central route of information leading to more pronounced negativity bias effects
processing, consumers base their judgements on (Ahluwalia, 2002; Lane et al., 2021). Therefore, higher
effortful processing, detailed scrutiny, and the per­ consumers’ ability and motivation to process a given
ceived diagnosticity of the message content (Petty & message can increase the likelihood of negativity bias
Wegener, 1998). Due to the non-normative nature of effects in their message processing. Against this back­
negative information in most instances and its higher drop, we contend that consumers’ prior experience
perceived specificity and risk-focus, it tends to be with the context of an online review and their need
viewed as more novel and diagnostic, than more fre­ for cognition are two factors that can increase the
quent positive information (Ahluwalia, 2002; likelihood of negativity bias in the effects of online
Skowronski & Carlston, 1989), causing central route reviews on consumers’ attitudes towards online
processors to assign it greater weights during impres­ services.
sion formation and decision-making, as compared to
positive information. Consequently, when conditions
2.3. Consumers’ prior experience with the online
favour the dominant use of the central route of infor­
service domain and negativity bias
mation processing, negative messages are expected to
be perceived as more diagnostic and produce greater Consumers’ prior experience with the context of
persuasion, as compared to positive messages, leading a given message increases their ability to process
to a negativity bias effect. the contents of the message (Qahri-Saremi &
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 721

Montazemi, 2016; Sujan, 1985). Prior experience elaboration (Petty & Cacioppo, 1986). Prior research
with the context of a message refers to consumers’ has shown that consumers’ level of need for cognition is
experience and familiarity with other products/ser­ an important determinant of consumers’ motivation to
vices that are within a similar context/domain as elaborate on a message content (Cacioppo et al., 1996;
the ones presented in the message. This provides Srivastava & Sharma, 2012). Need for cognition is
the consumers with the “cognitive schema” to scru­ a well-established individual characteristic manifested
tinise the message content (Peracchio & Tybout, in people’s disposition to engage in and enjoy thinking
1996). This schema-based scrutiny and inference (Cacioppo et al., 1996; Gupta & Harris, 2010; Wood &
allows the consumers to engage in an in-depth Swait, 2002). In the persuasion literature, need for
elaboration and assessment of the diagnosticity of cognition is conceptualised as the proclivity to process
the information within a message, manifested in information, which, distinct from intellectual ability,
the central route of information processing. predicts consumers’ motivation to evaluate products/
However, without the prior experience to shape services (Gupta & Harris, 2010; Srivastava & Sharma,
their cognitive schema, consumers will simply lack 2012; Wood & Swait, 2002).
the reference point for assessing a message content ELM explains that both ability and motivation to
(Peracchio & Tybout, 1996). As such, they are more process a given message and elaborate on its content
limited in the depth of their elaboration, mani­ are necessary conditions for a consumer to engage in
fested in the peripheral route of information pro­ the central route of processing (Petty & Cacioppo,
cessing. In other words, the cognitive schema 1986). Without high motivation, consumers are less
determines the consumers’ ability for engaging in likely to engage in effortful processing of the message
central route of information processing. Those who content, even when they have high prior experience
are less able to engage in central route of proces­ with its domain (Gupta & Harris, 2010). Against this
sing a message are likely to follow a peripheral backdrop and given the higher likelihood of negativity
route in assessing the message. As a result, they bias effects in the central route of processing, we con­
rely on the surface features of a message and its
tend that higher levels of need for cognition can
source in decision-making and the perceived diag­
strengthen the effects of prior experience with the
nosticity of the message content is less influential
online service domain on the likelihood of negativity
in the formation of their judgements.
bias in the effects of online reviews. To this end, we
In the case of our study, consumers’ prior experi­
ence with the domain of the online service that is postulate a second-order (three-way) moderation
presented in the online review provide a cognitive hypothesis.
schema for the consumers to scrutinise and evaluate
the content of the review and its diagnosticity (i.e., H2: Consumers’ level of need for cognition positively
central route). Therefore, we contend that the atti­ moderates the effect of their prior experience with the
tudes of consumers with higher prior experience online service domain on the negativity bias in the
with the online service domain are more likely to be effects of online review content.
influenced by the diagnosticity of review content. As
explained before, negative information is conceived to
be more diagnostic than positive information. Thus, 3. Methodology
the negativity bias effects are expected to be more
pronounced in higher elaboration likelihood condi­ 3.1. Study context
tions (i.e., central route). On this basis, we propose In this study, we focus on the negativity bias in the
the following hypothesis: effects of online reviews’ contents on consumers’
attitudes before using online experience services.
H1: The negativity bias in the effects of online review Experience services are characterised by attributes
content on consumers’ attitudes towards an online service that cannot be objectively evaluated, rather need to
is stronger when consumers have higher prior experience be experienced and subjectively evaluated by the
with the online service domain. consumers before usage (Xiao & Benbasat, 2007).
Examples of online experience services include
online music services and online healthcare ser­
vices. Given the subjective and personal nature of
2.4. Consumers’ need for cognition and
online reviews, they are an important source for
negativity bias
reducing consumers’ uncertainty towards evaluat­
Motivation to process a given message is another ing online experience services. Therefore, we focus
important contingency factor for high elaboration like­ on the online experience service context (called
lihood even among consumers with high ability for “online service” in this paper).
722 H. QAHRI-SAREMI AND A. R. MONTAZEMI

Consumers’ usage of an online service is a process that As a result, we used a no-review condition to ensure
occurs over time consisting of two main stages – pre- the true neutrality of the control group in our
usage and post-usage – with different decisions and experiment.
actions occurring at each stage (Montazemi & Qahri- Moreover, as will be presented later, we used
Saremi, 2015). Because, unlike the post-usage stage, con­ the control group for two objectives, namely vali­
sumers in the pre-usage stage have no experience with dating the experimental manipulations for PR and
using the focal online service. Thus, external sources of NR and conducting a post-hoc analysis. After
information about the focal online service, such as online validating our PR and NR manipulations, the con­
reviews, play an important role in determining their trol group was removed from our data for the
attitude towards using the online service in this stage. main analysis of the negativity bias effects. The
To this end, we focus on the effects of online reviews at reason is that for testing the negativity bias effects,
the pre-usage stage. the absolute magnitude of changes in subjects’
attitudes attributed to NR had to be compared
with the absolute magnitude of attitudinal changes
3.2. Subjects
attributed to PR. Therefore, a direct comparison
Subjects in our experiment were 450 (43% women, of the absolute magnitude of attitudinal changes
57% men) undergraduate (freshman, sophomore, between NR and PR was sufficient for testing our
and junior) commerce students in a business school hypotheses, without a need to include the control
in North America, with an average age of 19.6 group in the main analysis. 1 This reduced our
(SD = 1.11). Student samples were used in prior ser­ sample size for the main analysis to 332, split
vice adoption and online review studies (e.g., Yin et al., between NR and PR treatments.
2014). Students’ participation in our study was volun­
tary and was compensated by a 2% bonus course
3.5. Experimental manipulations
credit.
3.5.1. Higher/lower prior experience with the online
service domain
3.3. Experiment design
To manipulate the higher/lower consumers’ prior
We conducted a controlled laboratory experiment to experience of the online service domain, we
assess the two hypotheses. We randomly assigned selected two different domains of the online experi­
subjects to six conditions based on a 3 × 2 between- ence services with which we expected our subjects
subject factorial design, as depicted in Table 1: three to have different levels of prior experience: online
different conditions (negative, positive, and no music service to represent higher prior experience
reviews), in conjunction with two different conditions domain (e.g., eMusic.com), and online healthcare
to represent higher/lower prior experience with the service to represent lower prior experience domain
online service domain were manipulated between (e.g., WebMD.com). Students are frequent users of
subjects. online music services (Kennedy et al., 2008). Thus,
we expected our subjects to have higher prior
experience of browsing for and listening to their
3.4. Role and design of the control group
favourite music using online music services (web­
Consistent with prior studies (e.g., Ketelaar et al., sites). However, considering that our subjects were
2015; Lane et al., 2021; Sundaram & Webster, 1999), undergraduate commerce students, we expected
we used a “no-review” condition as a neutral control them to have lower prior experience of using
group for validating the PR and NR manipulations. online healthcare services (websites), relative to
The reason is that due to possibly asymmetrical and online music services. To assess the validity of
subjective effects of negative versus positive informa­ these manipulations, we asked our subjects about
tion on a consumer’s attitude, it was impossible to the extent of their prior experience with the music/
ensure that any two-sided review would be perceived healthcare online service (depending on which
as equally neutral by all our subjects. Indeed, this, in online service domain was randomly assigned to
itself, would have required a separate validation check. them). We used one question with a 5-point
Likert scale before the start of the experiment.
The results of ANOVA tests showed that, as
Table 1. Experiment Design.
Higher Prior Lower Prior
expected, our subjects had significantly higher
Experience Experience prior experience with the online music domain
(Online Music) (Online Healthcare) than the online healthcare domain (Mean
Negative Experimental Treatment Experimental Treatment
Review Music = 2.92, SD Music = 1.05; Mean Health = 1.34,
Positive Review Experimental Treatment Experimental Treatment SD Health = 0.63; F = 318.762, p < 0.001). This
No review Control Condition Control Condition result validated our experimental manipulation for
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 723

higher/lower prior experience with the online ser­ significantly higher PU (ΔMean = 1.43, p = 0.02),
vice domain. Furthermore, to avoid the confound­ PEOU (ΔMean = 1.83, p = 0.001), and trust
ing effects of major variations in online service (ΔMean = 1.84, p = 0.001) towards using the
functionalities on our results, we selected online online service compared to the control group.
services with comparable functionalities, as Moreover, as expected, subjects who received
explained in Appendix A. NRs had significantly lower PU (ΔMean = −3.93,
p = 0.000), PEOU (ΔMean = −2.74, p = 0.000),
3.5.2. Negative and positive reviews’ contents and trust (ΔMean = −3.12, p = 0.000) compared to
(online review valence manipulations) the control group. These tests validated the
To manipulate the valence of online review con­ expected effects of the manipulated valence of
tent in our experiment, we developed four ficti­ online reviews’ contents on consumers’ percep­
tious NRs and four fictitious PRs for the tions. Therefore, our valence manipulation was
healthcare/music online services to nudge our sub­ validated.
jects’ perceptions against or for using the online
service (see samples in Appendix B). Prior
3.6. Measures and factors
research has shown the importance of volume of
online reviews on consumers’ decisions (Duan The measurement instruments for our factors are
et al., 2008; Montazemi & Qahri-Saremi, 2014; provided in Appendix C. The dependent factor in
Ren & Nickerson, 2019). Therefore, controlling our study is consumers’ attitudes towards using an
the number of online reviews (to four online online service, which represent users’ overall orien­
reviews) across our experimental treatments tation towards an online service. To measure this
allowed us to control this effect. We randomly factor, we adapted the measures from Pavlou and
selected eight online music services and eight Fygenson (2006) (e.g., “using [this website] is
online healthcare services. Each fictitious online a good idea”). Furthermore, for measuring PU,
review discussed an online healthcare/music ser­ PEOU, and trust that we used in our manipulation
vice (website) in terms of its usefulness, ease of check, we adapted scales from Pavlou and
use, and trustworthiness, where NRs criticised Fygenson (2006) that are based on Ajzen and
them and PRs endorsed them. Fishbein (1980)’s expectancy-value formulation
To control the potential confounding effects of for measuring attitudinal beliefs. According to
other factors, such as source credibility and online expectancy-value formulation, these belief factors
review length and quality, all subjects were informed were measured as the product term 2 of belief
that the reviews were from a credible source with strength (e.g., “[healthcare/music services] would
verified experience with the online service (for con­ enhance my effectiveness in getting useful infor­
trolling source credibility). Furthermore, the fictitious mation [healthcare/music]”), and the importance
positive/negative reviews had the same length (210 of its outcome for the respondent (e.g., “For me,
words). In addition, the level of justifications in each getting useful information about [healthcare/
fictitious online review was controlled to avoid any music] from [healthcare/music website] is impor­
confounding effects caused by the variations in the tant”; Ajzen & Fishbein, 1980; Pavlou & Fygenson,
quality of the online reviews. To that end, a pilot 2006). Furthermore, measures for the need for
study involving 35 Ph.D. students was conducted cognition were adapted from Wood and Swait
where we asked the participants to assess the quality (2002).
of the content of the positive/negative online review,
which was randomly assigned to them, in terms of its
3.7 Experimental procedures
relevance to the focal online service and the clarity of
the arguments in it, using a single question with At the beginning of the experiment, all subjects
5-point Likert scale. No significant difference between were asked to complete an online questionnaire
the four experimental treatments was found in terms that collected their demographics and the extent
of the review content quality (p= 0.71). of their prior experience with the online music/
To check the validity of our manipulations for healthcare domain. Next, to avoid possible con­
PRs and NRs, we measured our subjects’ perceived founding effects of subjects’ prior firsthand use
usefulness (PU), perceived ease of use (PEOU), experience with a particular online service while
and trust in the online service after reading the ensuring their sufficient involvement with the
reviews (i.e., the three online service attributes online service, we adapted the two criteria sug­
that we discussed in our fictitious reviews), and gested by Sundaram and Webster (1999) and Lim
statistically compared them with those of the con­ and Lee (2019) as follows. First, a randomly
trol group. The results of ANOVA tests showed selected name of an online service (website) from
that as expected, subjects who received PRs had either a music or a healthcare list was displayed
724 H. QAHRI-SAREMI AND A. R. MONTAZEMI

on the computer screen for each subject. The assigned online service and by offering the same
subject was asked whether he/she had used it description to all subjects within each domain, we
before. If he/she had used the online service, maintained a common frame of reference for all sub­
then another randomly selected name from the jects (in control and treatment conditions), against
list was presented on the screen until it reached which we then captured the effects of our experimental
a name that the subject had not used before, treatments: NRs and PRs.
which was then assigned to the subject. This
allowed us to avoid the confounding effect of
subjects’ prior firsthand use experience of the 3.8 Preliminary analysis
assigned online service. There was no incident in Before testing the hypotheses, we conducted a series of
which the subject had used all of the eight online preliminary analyses to ensure that statistical and
services in the list. methodological artefacts did not affect our measure­
Second, all subjects (in control and treatment ments and analyses. Results demonstrated high relia­
groups) received a general and neutral description bility and good construct validity scores for the
of the assigned online service (see Appendix A). factors, no serious skewness and kurtosis were
The description explained what the online service detected, and square roots of average variance
was about (music or healthcare services) and extracted were larger than 0.7 and the corresponding
described the features that it offered to its users, correlations for each factor which indicated acceptable
without any positive or negative information about validity of the factors (see Appendix D for details).
the quality of those features. The descriptions were
the same within each domain (music or healthcare)
except for the name of the assigned online service. 4. Analysis and results
To ensure that students were sufficiently involved 4.1. Model specification
in the online services, in the description for the
online music service, they were told that they can The full regression model for testing our two hypoth­
search for their favourite music songs, albums, and eses is depicted in equation (1).
singers in its directories and stream them, and in
the description for the healthcare domain, subjects
were told that it offers healthcare contents includ­
ing information about stress, its symptoms and
effects, and stress management techniques. These
topics are considered to be relevant to student ATT, the dependent variable, is the magnitude of
subjects (Kennedy et al., 2008; Misra & McKean, effects on consumers’ attitudes towards the online
2000). For verification, regarding relevance of stress service. The effects of PRs and NRs on consumers’
management, we asked subjects in the healthcare attitudes differ in terms of both the direction of the
condition about the importance of stress manage­ effect and the magnitude of the effect. The direction of
ment for them (see Appendix C), using a 5-point the effects alters because an effective PR is expected to
Likert scale. Their ratings (Mean = 4.02, SD = 0.23) increase the attitude, while an effective NR is expected
were significantly larger than the midpoint of 3 to decrease the attitude. The magnitude of the effect
(p < 0.001), which verified the relevance of stress (ATT in our model) is the absolute extent of change in
management to them. the attitude as a result of PRs and NRs, regardless of
Next, subjects in the treatment conditions were their direction (increase or decrease). The negativity
shown four distinct PRs or NRs regarding their bias occurs when NRs have a significantly larger mag­
assigned online services (see Appendix B). This step nitude of effect on attitude, as compared to the PRs,
was skipped for the control group (i.e., no-reviews when the direction of effect is removed. Therefore, to
condition). Next, subjects were asked to complete an accurately estimate the extent of negativity bias, we
online survey regarding their evaluations of PU, need to estimate the magnitude of the effect of NRs on
PEOU, trust, and attitude towards using the assigned attitude as compared to the magnitude of the effect of
healthcare/music online service. We used these data PRs on attitude, after removing the direction of the
for manipulation checks, and to measure respondents’ effects (decrease and increase). To do so, we reversed
attitudes towards the online service. the attitude scores for NRs to compare their magni­
This design of experiment is in accordance with tudes with those of PRs.
Helson’s (1964) adaptation-level theory that consu­ NR in our model is a dummy variable (1: NR; 0: PR)
mers’ judgements are based on their prior experiences, to estimate the extent of the effects of negativity bias
context, and a stimulus (i.e., our experimental treat­ (b3 coefficient) on ATT (it estimates the unique mag­
ments: NRs and PRs). In particular, by ensuring sub­ nitude of the effects of NR on ATT as compared to
jects’ no prior firsthand use experience with the those of PR). As mentioned earlier, we excluded the
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 725

no-reviews (control) group from our data for estimat­ significantly larger in high prior experience condition
ing our model to enable the direct comparison of the (i.e., online music service) as compared to low prior
magnitude of the effects of NRs with those of PRs. experience (i.e., online healthcare service).
EXP is a dummy variable for prior experience with
the service domain (1: high; 0: low). The coefficient for
its first-order interaction term with NR (i.e., NR × 4.3 Post-hoc analysis: individual effects of NR and
EXP), b5, indicates the moderation effect of prior PR
experience on the negativity bias in the effects of online
Moreover, as expected, our results show that higher
review content. This result is used for testing H1.
levels of need for cognition significantly strengthen the
NFG is a continuous variable with scores for the
moderation effects of prior experience with the online
need for cognition. The coefficient for its second-order
service domain on the negativity bias in the effects of
interaction term (i.e., NR × EXP × NFG), b9, indicates
online review content on consumers’ attitude (b9
the effect of need for cognition on the moderating
= 0.82, p = 0.000; see Model 4 in Table 2). Therefore,
effect of prior experience on negativity bias in the
H2 is also supported. In particular, as depicted in
effects of online review content. Therefore, it is used
Figure 2, when need for cognition is high, the level
for testing H2. Age and gender (Female dummy vari­
of prior experience with online service domain exerts
able in equation (1)) are used as control variables.
a larger impact on the negativity bias in online review
Given the higher-order interaction terms in equa­
effects compared to when need cognition is low. These
tion (1) which can change the interpretation of lower
results fully support our hypothesis that higher prior
order terms, we used hierarchical (fixed-order) regres­
experience with the online service domain and higher
sion approach (Cohen et al., 2003; De Jong, 1999) to
levels of need for cognition disproportionately
add the independent variables to our regression ana­
increase the magnitude of the effects of NRs (as com­
lysis step-wise: Step 1 only includes the control vari­
pared to the effects of PRs) on consumers’ attitudes
ables (i.e., Model 1); Step 2 adds the main effect of
towards an online service.
negativity bias (i.e., NR) to Model 1 (i.e., Model 2);
So far, the results provide strong support for the
Step 3 adds EXP and its first-order interaction term to
strengthening effects of users’ prior experience with
Model 2 (i.e., Model 3), which is used for testing H1.
the online service domain and need for cognition on
Finally, Step 4 adds NFG and its first-order
the negativity bias. Nevertheless, negativity bias is
and second-order interaction terms to Model 3 (i.e.,
a relative concept reflecting the comparative effect of
Model 4). The coefficient of second-order interaction
NR relative to those of PR on consumers’ attitudes. To
term in Model 4 is used for testing H2.
understand the individual effects of NR and PR at
different levels of prior experience and need for cogni­
tion, we conducted a post-hoc analysis. To that end,
4.2 Results
we split our sample into eight subsamples based on 2
As depicted in Table 2, higher prior experience with (high/low prior experience) x 2 (high/low mean-split
the online service domain significantly increases the need for cognition) x 2 (NR + control groups/PR +
negativity bias in the effects of online review content control groups). Then, we conducted eight separate
on consumers’ attitude towards an online service (b5 OLS regressions (one for each subsample) to test the
= 0.66, p = 0.000; see Model 3 in Table 2). Therefore, individual effects of NR and PR on the users’ attitude
H1 is supported. Figure 1 provides a more nuanced as compared to the control group in the same condi­
view of this moderation effect. While the negativity tion. We controlled for the effects of users’ age and
bias exists under both low and high prior experience gender in all eight regressions. As presented in Table 3,
conditions, its magnitude of effects on attitude is the results show that when prior experience with the

Table 2. Results of Hierarchical Multiple Regression.


Model 1 Model 2 Model 3 Model 4
Constant (b0) 3.85 (0.000) 3.55 (0.000) 3.68 (0.000) 3.55 (0.000)
Age (b1) −0.02 (0.549) −0.02 (0.564) −0.02 (0.612) −0.01 (0.935)
Gender (b2) 0.08 (0.33) 0.06 (0.408) 0.03 (0.64) 0.04 (0.518)
NR (b3) 0.53 (0.000) 0.20 (0.042) 0.07 (0.464)
Prior Experience (EXP) (b4) −0.32 (0.001) −0.43 (0.000)
NR × EXP (b5) 0.66 (0.000) 0.72 (0.000)
Need for Cognition (NFG) (b6) 0.42 (0.000)
NR × NFG (b7) −0.37 (0.027)
EXP × NFG (b8) −0.57 (0.000)
NR × EXP × NFG (b9) 0.82 (0.000)
R2 0.004 0.149 0.205 0.253
ΔR2 (p-value) 0.004 (0.507) 0.144 (0.000) 0.056 (0.000) 0.048 (0.000)
Coefficients are unstandardised; p-values are in the parentheses
726 H. QAHRI-SAREMI AND A. R. MONTAZEMI

Low EXP High EXP

4.4

p = 0.000
4.2

Manitude of Effect on Attitude


4
p = 0.037
3.8

3.6

3.4

3.2

3
PositiveReview NegativeReview

Figure 1. First-order (two-way) interaction effect of prior experience (EXP) on negativity bias in the effects of online reviews on
attitude (H1).

(1) High EXP, High NFG (2) High EXP, Low NFG
(3) Low EXP, High NFG (4) Low EXP, Low NFG
4.3

(1) p = 0.000
4.1
Magnitude of Effect on Attitude

3.9

(2) p = 0.006
3.7
(3) p = 0.320

(4) p = 0.010
3.5

3.3

3.1

2.9
Positive Review Negative Review

Figure 2. Second-order (three-way) interaction effect of need for cognition (NFG) and prior experience (EXP) on negativity bias in
the effects of online reviews on attitude (H2).

Table 3. OLS Regression Results for the Effects of PR and NR and their Comparisons based on Cumming (2009).
B 95% CI for
Prior Need for Online (absolute Absolute POL of Absolute CIs for NR and p-value for
Experience Cognition Reviews N Beta p-value value) Values of B PR POL
Low Low NR 72 −0.398 0.001 0.530 0.236 0.824 −0.01 < 0.010
PR 87 0.208 0.056 0.115 −0.003 0.233
Low High NR 76 −0.453 <0.001 0.745 0.407 1.082 0.24 < 0.050
PR 53 0.469 0.001 0.328 0.151 0.506
High Low NR 50 −0.460 0.003 0.644 0.310 0.978 −0.10 < 0.010
PR 56 0.151 0.275 0.090 −0.074 0.255
High High NR 86 −0.720 <0.001 1.259 0.996 1.522 −0.70 < 0.001
PR 88 −0.103 0.328 −0.069 −0.210 0.071
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 727

online service domain is low, both PR and NR signifi­ generated reviews” (Wu, 2013, p. 978). The consensus
cantly influence consumers’ attitudes towards the based on these findings is that negativity bias does not
online service, almost regardless of their level of need influence experienced and thoughtful consumers’ atti­
for cognition. However, when prior experience with tudes towards the products/services, because as per
the online service domain is high, PR loses its potency dual-process theories (e.g., ELM), these consumers
in influencing consumers’ attitudes, while NR main­ are expected to rely on the central route of informa­
tains its significant effects. tion processing where perceived diagnosticity of infor­
Moreover, following Cumming (2009) approach, mation is the main driver of influence.
we calculated the proportion overlap (POL) between In this study, we have directly tested this idea by
confidence intervals for absolute values of coefficients focusing on the valence of the content of online reviews.
for NR and PR in the regressions for each of the eight Contrary to the implied consensus discussed above, our
conditions. POL value between 14% and 50% is findings show that the asymmetrical effects of NR con­
approximately equivalent of p < 0.05, POL value tents are more pronounced among the experienced and
between −37% and 14% is approximately equivalent thoughtful consumers. Our post-hoc findings show that
of p < 0.01, and POL value less than −37% is approxi­ NRs are highly influential for both groups of consumers
mately equivalent of p < 0.001 (Cumming, 2009). (novice and experienced) and comparing their effects
As presented in Table 3, when prior experience is with the effects of PRs reveals the disproportionate
high, the difference between the effects of NR and PR effects of NRs on the experienced and thoughtful con­
grows larger as users’ level of need for cognition sumers. While both PRs and NRs can move the attitu­
increases; so much that, when both prior experience dinal needle for novice consumers, only the NRs can do
with the online service domain and need for cognition so for experienced and thoughtful consumers. As such,
are high, the PR has the smallest effect across all our results present the negative information as an ele­
conditions, while the NR has the largest effect, and ment of message diagnosticity in online review content
consequently the p-value of difference between their and a central cue that disproportionately influence the
coefficients is the smallest. Therefore, the moderation experienced and thoughtful consumers’ attitudes. Our
effects that we observed on the negativity bias can be findings provide a new perspective on the importance of
attributed to both weakening of PR effects and strength­ NRs and the negativity bias in the effects of online
ening of NR effects, at the same time, as prior experi­ reviews.
ence with the online service domain and need for Another important issue that our findings illuminate
cognition increase. is the importance of the operationalisation of message
valence in the online review studies. Prior studies have
largely operationalised the valence based on the star
5. Discussion ratings attached to the online reviews, with a one-star
rating representing a very negative and a five-star rating
5.1. Contributions to research
representing a very positive valence (Chevalier &
Leidner (2020) describes a theoretical contribution as Mayzlin, 2006; Wu, 2013). In this paper, we implemen­
“ . . . insights that are unexpected, surprising, or coun­ ted and tested the effects of valence of the content of the
terintuitive (e.g., nonobvious) . . . ”. One counterintui­ online reviews, while masking the effects of different star
tive contribution of this study is showing that the ratings (i.e., star ratings were not presented to the sub­
asymmetrical effects of NR contents are more pro­ jects). As a result, our findings show that NRs can be
nounced among the experienced and thoughtful con­ perceived as more diagnostic by experienced and
sumers. However, the consensus in the prior online thoughtful consumers, which brings to question the
review research has implicitly or explicitly been that predominant assumption in the extant literature that
online review valence is a peripheral cue (a heuristic) negativity bias is a function of peripheral processing.
that predominantly influences novice consumers, as Nevertheless, scholars should be cautious in extending
compared to experienced and engaged consumers these findings to contexts where valence is operationa­
(Qahri-Saremi & Montazemi, 2019; Wu, 2013). NRs lised via star ratings only (e.g., Chevalier et al., 2018).
have often been attributed to the negative emotional This is because experienced and thoughtful consumers
state (e.g., anger) of the reviewers, rather than the predominantly rely on the elaboration and scrutiny of
product/service attributes (Kim & Gupta, 2012; Qahri- the online review contents, while the star rating is
Saremi & Montazemi, 2019; Wu, 2013). On this basis, a simple heuristic that is disproportionally attractive to
the prior research on online review valence has argued consumers who do not engage in the scrutiny of the
that “negative information might grab the attention online review contents (Nazlan et al., 2018). NRs tend to
more easily, but attention alone does not guarantee the include more concrete and detailed information (Yin
value of the information” (Wu, 2013, p. 977), and has et al., 2012), but such informational value could only
concluded that “bad is not necessarily stronger than be realised by the consumers who read the NR content
good in relation to the perceived value of consumer- (text) rather than those who only rely on the star ratings.
728 H. QAHRI-SAREMI AND A. R. MONTAZEMI

This emphasises the importance of how valence is oper­ strategies, such as “Channel Change” or
ationalised in the online review studies, which in turn “Censorship” of NRs (Howell et al., 2012), can have
determines the type of consumers (e.g., experienced vs. adverse consequences by making them even more
novice) who are more likely to be affected by the nega­ scarce and thereby more diagnostic for this group of
tivity bias. arguably influential consumers. This can give the NRs
even more potential for sparking an online firestorm
(Herhausen et al., 2019).
5.2. Implications for practice
Recently, the U.S. Federal Trade Commission (FTC)
5.3. Limitations and avenues for future research
fined a company, Fashion Nova, a popular online
fashion retailer, $4.2 million for concealing NRs We acknowledge several limitations in our study
from being posted to its website. “Deceptive review that can open avenues for future research. First,
practices cheat consumers, undercut honest busi­ we assessed negativity bias using fictitious one-
nesses, and pollute online commerce”, said Samuel sided reviews. A one-sided review contains only
Levine, director of the FTC’s Bureau of Consumer positive or negative content, not both. Therefore,
Protection in a statement. “Fashion Nova is being an extension of our study is assessing negativity
held accountable for these practices, and other firms bias within the two-sided reviews, which include
should take note” (D’Innocenzio, 2022). Thus, con­ both positive and negative content. Second, while
cealing NRs is not a viable long-term strategy to avoid we controlled the volume of online reviews (four
their adverse effects on firms. Instead, firms should opt reviews) in our experimental treatments and
to understand the adverse effects of NRs and devise masked the potential effects of aggregated star
appropriate response strategies to mitigate these nega­ ratings (by not presenting it to our subjects) in
tive effects. In this paper, we show that the effects of this study, extant research shows the importance
NRs are stronger and more consequential for online of these factors affecting consumers’ judgements
service providers as compared to PRs’ because more towards products/services. Therefore, one promis­
experienced and thoughtful consumers are more likely ing area for the future studies is looking into the
to engage with and be influenced by NR contents. importance of the variability of these two factors
Moreover, the opinions of this group of consumers in addition to the valence of online review content
are usually more influential on others, given their for the effect of negativity bias. Third, while the
prior experience and depth of scrutiny. Therefore, two online service domains (music and health
attending to NR in a timely manner and offering care) differentiate in the level of our subjects’
effective responses to consumers’ concerns in these prior experience with them, there might be other
reviews in a way that addresses the core arguments differences between them (e.g., utilitarian versus
in them can be important to manage their impacts. If hedonic use). Future studies can investigate these
a NR is not responded to or resolved in a timely additional differences and their role in negativity
fashion, it can influence some of the most important bias in the effects of online review content.
consumers and sparks “online firestorms” (Herhausen Fourth, similar to prior experimental studies in
et al., 2019). Indeed, the timeliness of a firm’s IS literature (e.g., Yin et al., 2014), the subjects in
responses to NRs improves the firm’s performance, our study were students, not actual consumers.
whereas responses to PRs do not have such effects We ensured their involvement with the online
(Chung et al., 2020). services in our experiment via a neutral descrip­
In this vein, prior findings that the investors seek tion that we provided to them, as outlined in the
information about the firm’s performance from online Methodology section. While we believe this mini­
review contents are noteworthy (Tirunillai & Tellis, mised possible adverse effect of student sample on
2012). Investors often have higher prior experience our results, future study can validate our findings
with the service domain (they have industry focus) using non-student consumers of online services.
and have high motivation to elaborate on online Fifth, we focused on the “consumption” side of
reviews as informational cues for their financial deci­ the effects of NRs that may present only a partial
sions (Aggarwal et al., 2012). As such, they fit the picture of the negativity bias phenomenon.
profile of experienced and thoughtful readers of online A complementary perspective on the effect of
reviews that can be significantly affected by negativity negativity is its influence on the “production”
bias (Aggarwal et al., 2012). This further emphasises side. Prior research in social psychology has
the importance of effective management of NRs. shown that negativity can improve the quality
Furthermore, the fact that the scarcity and infre­ and effectiveness of persuasive messages and social
quency of NRs (as compared to PRs) render them influence strategies (Forgas, 2007). In the online
more diagnostic for experienced and thoughtful con­ review context, this would mean that the consu­
sumers suggest that some organisational response mers with negative experiences about products/
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 729

services are likely to post higher quality online References


reviews. While we controlled for the effects of
Aggarwal, R., Gopal, R., Gupta, A., & Singh, H. (2012).
message quality in our experiments, this means Putting money where the mouths are: the relation
that NRs may not be only perceived as more between venture financing and electronic word-of-
diagnostic by receivers, but they may also be of mouth. Information Systems Research, 23(3–part–2),
higher quality than PRs. The production effect of 976–992. https://doi.org/10.1287/isre.1110.0402
negativity on online review quality is an important Ahluwalia, R. (2002). How prevalent is the negativity effect
in consumer environments?. Journal of Consumer
area for future research.
Research, 29(2), 270–279. https://doi.org/10.1086/341576
Finally, while we focused on the valence of online Ajzen, I., & Fishbein, M. (1980). understanding attitudes and
review content (NR and PR) to investigate the negativity predicting social behavior. Prentice-Hall Englewood Cliffs .
bias effects in this paper, there are additional ways in Bateson, G. (1973). Steps to an ecology of mind. London:
which the valence framing can be discussed. For exam­ Granada Publishing.
Boucher, J., & Osgood, C. E. (1969). The pollyanna
ple, Levin et al. (1998) proposed a typology of three hypothesis. Journal of Verbal Learning and Verbal
different kinds of valence framing effects: (1) “attribute Behavior, 8(1), 1–8. https://doi.org/10.1016/S0022-
framing”, where a single attribute of a product/service is 5371(69)80002-2
the subject of the valence manipulation, and the depen­ Boyer, K. K., & Hult, G. T. M. (2006). Customer behavioral
intentions for online purchases: An examination of ful­
dent measure of interest is the extent of favourability or fillment method and customer experience level. Journal of
unfavourability of the product/service; (2) “risky choice Operations Management, 24(2), 124–147. https://doi.org/
frames”, where discrete choices between a risky option 10.1016/j.jom.2005.04.002
and a riskless option of equal expected value depend on Cacioppo, J. T., Petty, R. E., Feinstein, J. A., &
Jarvis, W. B. G. (1996). Dispositional differences in cog­
whether the options are described in positive (i.e., gains)
nitive motivation: The life and times of individuals vary­
or in negative (i.e., loss) terms; and (3) “goal framing”, ing in need for cognition. Psychological Bulletin, 119(2),
where the issue is framed to focus on the potential to 197–253. https://doi.org/10.1037/0033-2909.119.2.197
obtain gain (positive frame) or the potential to avoid loss Casado-Díaz, A. B., Andreu, L., Beckmann, S. C., & Miller, C.
(negative frame), while both frames can enhance the (2020). Negative online reviews and webcare strategies in
social media: Effects on hotel attitude and booking
favourability of the issue. Prior research has shown that intentions. Current Issues in Tourism, 23(4), 418–422.
the asymmetrical effects of negative versus positive can https://doi.org/10.1080/13683500.2018.1546675
appear in all three kinds of framing effects (Levin et al., Chen, Z., & Lurie, N. H. (2013). Temporal contiguity and
1998). However, to the best of our knowledge, the nega­ negativity bias in the impact of online word of mouth.
Journal of Marketing Research, 50(4), 463–476. https://
tivity bias research in the online review context has been doi.org/10.1509/jmr.12.0063
predominantly focused on the “valence”, rendering the Chevalier, J. A., Dover, Y., & Mayzlin, D. (2018). Channels
need for further investigation into the effects of the other of impact: User reviews when quality is dynamic and
types of framing in online review literature. managers respond. Marketing Science, 37(5), 688–709.
https://doi.org/10.1287/mksc.2018.1090
Chevalier, J. A., & Mayzlin, D. (2006). The effect of word of
mouth on sales: online book reviews. Journal of
Endnotes Marketing Research, 43(3), 345–354. https://doi.org/10.
1509/jmkr.43.3.345
1. Nevertheless, our post-hoc analysis presents an alter­ Chung, S., Animesh, A., Han, K., & Pinsonneault, A. (2020).
native, more complex way of testing the hypotheses Financial returns to firms’ communication actions on
that involves the control group. The post-hoc findings firm-initiated social media: evidence from facebook busi­
are consistent with our main results. ness pages. Information Systems Research, 31(1), 258–285.
2. The product term is the product of two factors mea­ https://doi.org/10.1287/isre.2019.0884
sured using 5-point Likert scales. Therefore, the Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003).
resulted scale is [1, 25]. Applied multiple regression/correlation analysis for the
behavioral sciences (Third ed.). Erlbaum.
Cumming, G. (2009). Inference by eye: Reading the overlap
Disclosure statement of independent confidence intervals. Statistics in Medicine,
28(2), 205–220. https://doi.org/10.1002/sim.3471
No potential conflict of interest was reported by the Curchod, C., Patriotta, G., Cohen, L., & Neysen, N. (2020).
author(s). Working for an algorithm: power asymmetries and
agency in online work settings. Administrative Science
Quarterly, 65(3), 644–676. https://doi.org/10.1177/
ORCID 0001839219867024
D’Innocenzio, A. (2022). Fashion retailer pays $4.2M in settle­
Hamed Qahri-Saremi http://orcid.org/0000-0002-4933- ment with FTC. The Associated Press. Accessed on January
834X 26, 2022. https://apnews.com/article/technology-business-
Ali Reza Montazemi http://orcid.org/0000-0002-4069- lifestyle-federal-trade-commission-
2844 75f313a380b020618ce633a90ac55d1f
730 H. QAHRI-SAREMI AND A. R. MONTAZEMI

De Jong, P. F. (1999). Hierarchical regression analysis in Levin, I. P., Schneider, S. L., & Gaeth, G. J. (1998). All frames
structural equation modeling. Structural Equation are not created equal: a typology and critical analysis of
Modeling: A Multidisciplinary Journal, 6(2), 198–211. framing effects. Organizational Behavior and Human
https://doi.org/10.1080/10705519909540128 Decision Processes, 76(2), 149–188. https://doi.org/10.
Duan, W., Gu, B., & Whinston, A. B. (2008). Do online 1006/obhd.1998.2804
reviews matter? – an empirical investigation of panel Lim, J., & Lee, Y. (2019). Does a mixed online consumer
data. Decision Support Systems, 45(4), 1007–1016. review lead to neutral or ambivalent evaluation? Journal
https://doi.org/10.1016/j.dss.2008.04.001 of Marketing Management, 7(1), 83–93. http://jmm-net.
Feldman, J. M. (1999). Four questions about human social com/journals/jmm/Vol_7_No_1_June_2019/7.pdf.
behavior: the social cognitive approach to culture and McKinney, J., E, H., Yoos, I. I., & C, J. (2010).
psychology. In J. Adamopoulos & Y. Kashima (Eds.), Information about information: a taxonomy of
Social psychology and cultural context (pp. 43–62). views. MIS Quarterly, 34(2), 329–344. https://doi.
SAGE Publications, Inc. https://doi.org/10.4135/ org/10.2307/20721430
9781452220550.n4 Misra, R., & McKean, M. (2000). College students’ academic
Forgas, J. P. (2007). When sad is better than happy: Negative stress and its relation to their anxiety, time management,
affect can improve the quality and effectiveness of per­ and leisure satisfaction. American Journal of Health
suasive messages and social influence strategies. Journal Studies, 16(1), 41–51. https://web.s.ebscohost.com/ehost/
of Experimental Social Psychology, 43(4), 513–528. https:// pdfviewer/pdfviewer?vid=2&sid=f432b719-03e4-470b-
doi.org/10.1016/j.jesp.2006.05.006 9d5d-4dc1bf574d19%40redis.
Gupta, P., & Harris, J. (2010). How e-WOM recommenda­ Montazemi, A. R., & Qahri-Saremi, H. (2014). The
tions influence product consideration and quality of effectiveness of electronic word of mouth on consu­
choice: A motivation to process information mers’ perceptions of adopting products/services –
perspective. Journal of Business Research, 63(9–10), a literature review. 10th international conference on
1041–1049. https://doi.org/10.1016/j.jbusres.2009.01.015 web information systems and technologies (webist
Helson, H. (1964). Adaption-level theory: An experimental 2014), Barcelona, Spain. doi:10.5220/000485060
and systematic approach to behavior. Harper & Row. 3240331.
Herhausen, D., Ludwig, S., Grewal, D., Wulf, J., & Montazemi, A. R., & Qahri-Saremi, H. (2015). Factors
Schoegel, M. (2019). Detecting, preventing, and mitigat­ Affecting Adoption of Online Banking: A Meta-Analytic
ing online firestorms in brand communities. Journal of Structural Equation Modeling Study. Information &
Marketing, 83(3), 1–21. https://doi.org/10.1177/ Management, 52(2), 210–226. https://doi.org/10.1016/j.
0022242918822300 im.2014.11.002
Ito, T., Larsen, J., Smith, N., & Cacioppo, J. (1998). Negative Murphy, R. (2020). Local consumer review survey 2020.
information weighs more heavily on the brain: The nega­ https://www.brightlocal.com/research/local-consumer-
tivity bias in evaluative categorizations. Journal of review-survey-2020/.
Personality and Social Psychology, 75(4), 887–900. Nazlan, N. H., Tanford, S., & Montgomery, R. (2018). The
https://doi.org/10.1037/0022-3514.75.4.887 effect of availability heuristics in online consumer
Kennedy, G. E., Judd, T. S., Churchward, A., Gray, K., & reviews. Journal of Consumer Behaviour, 17(5), 449–460.
Krause, K. (2008). First year students’ experiences with https://doi.org/10.1002/cb.1731
technology: Are they really digital natives. Australasian Pavlou, P. A., & Fygenson, M. (2006). Understanding and
Journal of Educational Technology, 24(1), 108–122. predicting electronic commerce adoption: An extension
https://doi.org/10.14742/ajet.1233 of the theory of planned behavior. MIS Quarterly, 30(1),
Ketelaar, P. E., Willemsen, L. M., Sleven, L., & Kerkhof, P. 115–143. https://doi.org/10.2307/25148720
(2015). The good, the bad, and the expert: how consumer Peeters, G., & Czapinski, J. (1990). Positive-negative asym­
expertise affects review valence effects on purchase inten­ metry in evaluations: the distinction between affective
tions in online product reviews. Journal of Computer- and informational negativity effects. European Review of
Mediated Communication, 20(6), 649–666. https://doi. Social Psychology, 1(1), 33–60. https://doi.org/10.1080/
org/10.1111/jcc4.12139 14792779108401856
Kim, J., & Gupta, P. (2012). Emotional expressions in Peracchio, L. A., & Tybout, A. M. (1996). The moderating
online user reviews: How they influence consumers’ role of prior knowledge in schema-based product
product evaluations. Journal of Business Research, 65 evaluation. Journal of Consumer Research, 23(3),
(7), 985–992. https://doi.org/10.1016/j.jbusres.2011. 177–192. https://doi.org/10.1086/209476
04.013 Petty, R. E., & Cacioppo, J. T. (1986). The elaboration like­
Lane, J. N., Teplitskiy, M., Gray, G., Ranu, H., Menietti, M., lihood model of persuasion. Advances in Experimental
Guinan, E., & Lakhani, K. R. (2021). Conservatism gets Social Psychology, 19(1), 123–205. https://doi.org/10.
funded? a field experiment on the role of negative infor­ 1016/S0065-2601(08)60214-2.
mation in novel project evaluation. Management Science, Petty, R. E., & Wegener, D. T. (1998). Attitude change:
Forthcoming, 1–8. https://doi.org/10.1287/mnsc.2021. Multiple roles for persuasion variables. In D. Gilbert,
4107 S. Fiske, & G. Lindzey (Eds.), The handbook of social
Lee, A. S. (2010). Retrospect and prospect: information psychology (Vol. 1, 4th ed., pp. 323–390). McGraw-
systems research in the last and next 25 years. Journal of Hill.
Information Technology, 25(4), 336–348. https://doi.org/ Pfeffer, J., Zorbach, T., & Carley, K. (2014). Understanding
10.1057/jit.2010.24 online firestorms: Negative word-of-mouth dynamics in
Leidner, D. E. (2020). What’s in a contribution? Journal of social media networks. Journal of Marketing
the Association for Information Systems, 21(1), 238–245. Communications, 20(1–2), 117–128. https://doi.org/10.
https://doi.org/10.17705/1jais.00598 1080/13527266.2013.797778
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 731

Qahri-Saremi, H., & Montazemi, A. (2016). On the Taylor, S. E. (1991). Asymmetrical effects of positive and
Effectiveness of eWOM Communications: the negative events: the mobilization-minimization
Moderating Effect of Consumers' Prior Experience. hypothesis. Psychological Bulletin, 110(1), 67. https://
Americas' Conference on Information Systems, San doi.org/10.1037/0033-2909.110.1.67
Diego, CA, The Association for Information Systems Thomas, J. B., Peters, C. O., Howell, E. G., & Robbins, K.
Qahri-Saremi, H., & Montazemi, A. R. (2019). Factors (2012). Social media and negative word of mouth: strate­
affecting the adoption of an electronic word of mouth gies for handing unexpecting comments. Atlantic
message: A meta-analysis. Journal of Management Marketing Journal, 1(2), 87–108. https://digitalcom
Information Systems, 36(3), 969–1001. https://doi.org/ mons.kennesaw.edu/amj/vol1/iss2/7.
10.1080/07421222.2019.1628936 Tirunillai, S., & Tellis, G. J. (2012). Does chatter really
Qahri-Saremi, H., & Montazemi, A. R. (2020). Negativity Bias matter? Dynamics of user-generated content and stock
in the Effects of eWoM Reviews: An Elaboration Likelihood performance. Marketing Science, 31(2), 198–215. https://
Perspective in Online Service Adoption Context. European doi.org/10.1287/mksc.1110.0682
Conference on Information Systems, A Virtual AIS Wei, W., Miao, L., & Huang, Z. (2013). Customer engage­
Conference; The Association for Information Systems. ment behaviors and hotel responses. International
Ren, J., & Nickerson, J. V. (2019). Arousal, valence, and Journal of Hospitality Management, 33(June 2013),
volume: How the influence of online review characteristics 316–330. https://doi.org/10.1016/j.ijhm.2012.10.002
differs with respect to utilitarian and hedonic products.
Wood, S. L., & Swait, J. (2002). Psychological indicators of
European Journal of Information Systems, 28(3), 272–290.
innovation adoption: Cross-classification based need for
https://doi.org/10.1080/0960085X.2018.1524419
cognition and need for change. Journal of Consumer
Rozin, P., & Royzman, E. B. (2001). Negativity bias, nega­
tivity dominance, and contagion. Personality and Social Psychology, 12(1), 1–13. https://doi.org/10.1207/
Psychology Review, 5(4), 296–320. https://doi.org/10. S15327663JCP1201_01
1207/S15327957PSPR0504_2 Wu, P. F. (2013). In search of negativity bias: An empirical
Skowronski, J. J., & Carlston, D. E. (1987). Social judg­ study of perceived helpfulness of online reviews.
ment and social memory: The role of cue diagnosti­ Psychology & Marketing, 30(11), 971–984. https://doi.
city in negativity, positivity, and extremity biases. org/10.1002/mar.20660
Journal of Personality and Social Psychology, 52(4), Xiao, B., & Benbasat, I. (2007). E-commerce product
689–699. https://doi.org/10.1037/0022-3514.52.4.689 recommendation agents: Use, characteristics, and
Skowronski, J. J., & Carlston, D. E. (1989). Negativity and impact. MIS Quarterly, 31(1), 137–209. https://doi.
extremity biases in impression formation: A review of org/10.2307/25148784
explanations. Psychological Bulletin, 105(1), 131–142. Yin, D., Bond, S., & Zhang, H. (2014). Anxious or angry?
https://doi.org/10.1037/0033-2909.105.1.131 effects of discrete emotions on the perceived helpfulness
Srivastava, K., & Sharma, N. K. (2012). Consumer attitude of online reviews. MIS Quarterly, 38(2), 539–560. https://
towards brand-extension incongruity: The moderating doi.org/10.25300/MISQ/2014/38.2.10
role of need for cognition and need for change. Journal Yin, D., Mitra, S., & Zhang, H. (2012). Mechanisms of
of Marketing Management, 28(5–6), 652–675. https://doi. negativity bias: an empirical exploration of app reviews
org/10.1080/0267257X.2011.558383 in apple’s app store. Thirty third international conference
Sujan, M. (1985). Consumer knowledge: Effects on evaluation on information systems, Orlando, FL, USA, The
strategies mediating consumer judgments. Journal of Association for Information Systems (AIS). https://aisel.
Consumer Research, 12(1), 31–46. https://doi.org/10.1086/ aisnet.org/icis2012/proceedings/
209033 HumanComputerInteractions/5/.
Sundaram, D., & Webster, C. (1999). The role of brand Yu, I. Y., Wan, L. C., & Yi, X. S. (2021). Managerial response
familiarity on the impact of word-of-mouth communica­ to negative online reviews in the service industry: a
tion on brand evaluations. Advances in Consumer tactic-based and culture-based model. Journal of Global
Research, 26(1), 664–670. https://www.acrwebsite.org/ Marketing, 34(3), 238–248. https://doi.org/10.1080/
volumes/8342/volumes/v26/NA-26/full. 08911762.2021.1888393
732 H. QAHRI-SAREMI AND A. R. MONTAZEMI

APPENDICES
APPENDIX A: HEALTHCARE AND MUSIC ONLINE
SERVICES
We randomly selected eight in healthcare and music
domains with similar functionalities in each category,
as outlined in Table (A1). Subjects were randomly
assigned an online music/healthcare service that they
had not used before and received a general (neutral)
description of the service. The descriptions were the
same for all online services within each domain (music
or healthcare) (Table (A2)).

Table A1. Criteria Used in Selecting our Healthcare and Music Online Services.
Criteria for Healthcare Websites Criteria for Music Websites
The website has information for a directory of healthcare contents, in The website has a search functionality to explore favourite singers, music
particular stress management techniques. albums, and song tracks.
The website provides reviews on medicines and supplements. The website has music streaming functionality.
The website has information about eating/diet/fitness. The website has online purchase functionality for purchasing favourite
song tracks and albums.
The website has symptom check functionality. The website has the functionality of users’ forums and reviews about
singers, music albums, and song tracks.
The website does not require a paid subscription. The website does not require a paid subscription.

Table A2. Descriptions Provided for Online Healthcare and Music Services.
Description for Healthcare Online Services (90 words)
[WebMD.com] is an online healthcare service that offers medical information and news about different health problems and their possible treatments. It
has information about a directory of different healthcare related contents. This includes contents on different types of stress, their symptoms and
effects, and techniques for managing stress. [WebMD.com] provides users’ forums and reviews to discuss different treatments, medicines and
supplements, and information about eating/ diet/ fitness. It has a symptom check functionality that suggests possible causes for the symptoms. It
does not require a paid subscription for using it.
Description for Music Online Services (90 words)
[eMusic.com] is an online music service that offers a directory of song tracks and albums from different singers and across different genres. It offers
a search functionality for all users to find their favourite singers, music albums and songs in its directory. [eMusic.com] offers a music streaming
functionality for users to listen to songs online. It provides different users’ forums and reviews to discuss different singers, music albums, and song
tracks. It offers albums and individual song track for purchase and does not require a paid subscription for using it.
EUROPEAN JOURNAL OF INFORMATION SYSTEMS 733

APPENDIX B: SAMPLES OF FICTITIOUS


REVIEWS

An Example of a Fictitious Negative Review (210 words)


The very first thing to know about asking medical questions on [WebMD.com], is that you cannot be examined over the internet. For that reason alone,
asking specific questions are unrealistic. The second thing to know about asking questions on [WebMD.com] is that anyone can provide you with
incorrect information. Suggestions and/or diagnoses made on the internet can hurt you, should you follow the advice. On [WebMD.com],
I encountered several broken links. For example, clicking on the “bursitis” in the list of Symptom Checker conditions generates an error. I was also
somewhat disappointed with the small area of text used to describe the conditions. It needs to be larger to avoid a lot of scrolling for viewing its
content. Another concern for me is that [WebMD.com] makes money from ads on its site. An online depression-screening test ad on page for stress
management directs visitors to Eli Lilly’s antidepressant Cymbalta. My concern is that most people who read that page are probably not depressed, or
even consider themselves depressed, but these ads presented by [WebMD.com] can lure them into possibly considering taking an antidepressant!
This is a major trust issue for me. Also, for running these ads they probably collect my health information, which is a major privacy concern for me.
An Example of a Fictitious Positive Review (210 words)

As doctors become busier and more expensive, I regularly use [WebMD.com] for educating myself about healthcare issues that matter to me. This has
also helped me to choose the right health care providers and doctors more effectively, explore lifestyle behaviours that keep me fit and healthy, and
research medications and therapeutic options. [WebMD.com] is supported by the medical community. The language is purposefully non-technical so
anyone can understand. [WebMD.com]’s interactive tools are an easy way to figure out more about your health – and they’re organised from A to
Z by category for easy access. These little quizzes and fun tools make it easy to evaluate various health concerns, from weight to environment and
everything in between. [WebMD.com] has stellar reputation. It covers a vast array of healthcare topics with information about symptoms, causes,
cures, medications, and prevention options available in general medical practice. It is run by a team of Physicians with top medical degrees. [WebMD.
com] has been awarded TRUSTe’s Privacy Seal signifying that its privacy policy and practices have been reviewed by TRUSTe for compliance with
requirements including transparency, accountability and choice for collection and use of your personal information. Privacy of my healthcare
information and data is very important to me, so this helped me trust [WebMD.com].

APPENDIX C: MEASURES

Table C1. Measurement Instruments (1: Strongly Disagree – 5: Strongly Agree).


Attitude Towards Using Using [this website] is a good idea.
[This website] makes [health management/finding my favourite music] more interesting.
Using [this website] is fun.
I like working with [this website].
Need for Cognition I would rather do something that requires little thought than something that is sure to challenge my thinking abilities
(reversed)
I try to anticipate and avoid situations where there is a likely chance, I’ll have to think in depth about something
(reversed)
I only think as hard as I have to (reversed)
The idea of relying on thought to get my way to the top does not appeal to me (reversed)
The notion of thinking abstractly is not appealing to me (reversed)
Perceived Usefulness [This website] would be useful for getting valuable information about [music/healthcare].
For me, getting valuable information about [music/healthcare] from [this website] is important.
[This website] would enhance my effectiveness in getting useful information about [music/healthcare].
For me, getting useful information about [music/healthcare] from a [this website] is important.
Perceived ease of Use My interaction with the [this website] is clear and understandable.
I believe that it is easy to get the [this website] to do what I want It to do.
Overall, I believe that the [this website] is easy to use,
Learning to operate the [this website] is easy for me.
Trust [This website] would be competent in providing objective information about [music/healthcare].
For me, getting objective [music/healthcare] information from [this website] is important.
[This website] would be honest in providing accurate information about [music/healthcare].
For me, getting accurate [music/healthcare] information from [this website] is important.
Stress Management Stress management is an important issue for me.
Importance I have been affected by stress, lately.
Stress management has been on my mind, lately.
734 H. QAHRI-SAREMI AND A. R. MONTAZEMI

APPENDIX D: PRELIMINARY ANALYSIS

Table D1. Descriptive Statistics, Correlations, & Square Root of AVE (diagonal).
Mean (SD) Skew (Kur) Alpha 1 2 3 4 5 6
1 NFC 3.67 (0.64) −0.25 (−0.35) 0.71 0.71
2 Attitude 2.82 (0.78) −0.36 (−0.39) 0.87 −0.09 0.92
3 PU 10.09 (4.98) 0.28 (−0.71) 0.87 −0.03 0.63 0.92
4 PEOU 12.01 (4.84) 0.10 (−0.29) 0.83 −0.01 0.52 0.59 0.91
5 Trust 10.27 (4.60) 0.24 (−0.43) 0.77 0.01 0.55 0.77 0.54 0.90
6 Age 19.64 (1.11) 2.17 (5.69) 1.00 −0.14 0.12 0.06 −0.01 0.03 1.00
7 Gendera 0.43 (0.50) 0.29 (−1.93) 1.00 −0.17 0.06 0.11 0.13 0.03 −0.04
a
0: men, 1: women; PU: Perceived Usefulness; PEOU: Perceived Ease of Use; NFC: Need for Cognition; Skew: Skewness; Kur: Kurtosis.

You might also like