The interplay of service experience and online consumer ratings
Romain CADARIO
IESEG School of Management
[email protected]
Jonas HOLMQVIST
KEDGE Business School
[email protected]
Abstract
In this article, we examine the effects of service experience on both antecedents and
consequences of online consumer ratings (OCR). Focusing on perceptions after the
actual service, our framework is grounded in the expectation-disconfirmation theory,
in which we distinguish two concepts: the valence of situational service experience
(i.e., the positive/negative experience with a given service) and the level of
accumulated service experience (i.e., knowledge from past experiences). In study 1,
we analyze data from Yelp.com to show that the effects of positive and negative
situational experience on OCR are exacerbated (weakened) when the level of
accumulated experience is low (high). Study 2, consisting of a controlled experiment
on services experiences, confirm the findings of Study 1 by finding that the effect of
OCR on repurchase intentions is stronger when the valence (level) of service
experience is negative (low) rather than positive (high).
Key-words
Online consumer ratings, service experience, valence, hotel booking.
The interplay of service experience and online consumer ratings
Introduction
In the last years, the development of online consumer ratings (OCRs) has radically
changed customer access to word-of-mouth, as consumers now can access reviews
of thousands of other customers as sites such as Amazon.com, TripAdvisor.com or
Hotels.com. As the opinions of other customers have a very high credibility when
consumers form their expectations about services (Grönroos, 2007, building on
Duncan and Moriarty, 1997), the rapid development and extensive availability of
other customers’ online ratings have direct consequences for consumers'
expectations and perceptions of services. Electronic word-of-mouth (eWOM) has
become one of the most trusted sources of recommendations: 70% of consumers
say that they trust online recommendations from virtual strangers (Nielsen, 2009).
Although the service experience might play a fundamental role in service perceptions
after the actual encounter (see Grönroos and Voima 2013), the extant literature pays
little attention to the interplay of service experience and online consumer ratings.
The service experience is the most important antecedent of online ratings; a
customer who has a positive (negative) experience will leave a good (bad) online
rating. However, two consumers who share the exact same service experience may
leave different ratings. In this paper, we argue that the level of accumulated service
experience plays an important direct and moderating role. The literature on the
interplay between OCR and the level of service experience is scarce with the
exception of Sridhar and Srinisavan (2012) who examined the effects of the level of
service experience, measured as the number of previews reviews, but failed to find
any significant direct or moderating effects. We believe that this result may be an
artifact of data collection, due to independent booking website in which the average
number of previous reviews and the standard deviation were quite low (1.9 and 3.8)
and may not reflect the actual variability in the actual levels of accumulated service
experience.
The extant literature largely studies the direct impact of online consumer ratings on
value and loyalty (Gruen et al., 2005), persuasiveness (Zhang et al., 2010),
evaluations (Kim & Gupta, 2012) and sales (Chevalier & Mayzlin, 2006; Yang & Mai,
2010). Moreover, prior literature has investigated the boundary conditions associated
with the effects of OCRs. Studies have shown the existence of various moderating
factors such as popularity (Zhu & Zhang, 2010), search vs. experience (Park & Lee,
2009), life cycle (Liu, 2006; Cadario, 2014), emotional expressions (Kim & Gupta,
2012), consumption goals (Zhang et al., 2010) or even installed base (Yang & Mai,
2010). However, the literature is still limited in terms of service experience, with the
exception of Zhu and Zhang (2010) who studied the related - yet different - concept
of Internet experience. The authors found that the impact of OCR on sales is stronger
when targeting consumers with greater Internet experience. Hence, one might expect
a similar result such that the effect of COR on purchase decision is stronger with a
greater level of accumulated service experience. However, we actually posit and find
the opposite effect, that is, the effect of OCR on purchase decisions is stronger for
consumers with a low level of service experience.
We examine the effects of service experience on both online ratings (study 1) and
repurchase intentions (study 2). Our theoretical contributions are two-fold. First, we
conceptually distinguish two concepts: the valence of situational service experience
and the level of accumulated service experience from past experiences. Second,
given the mixed results and gaps in the literature, we provide a comprehensive two-
stage theoretical framework, in which we examine the effects of service experience
on the antecedents (stage 1) and the consequences (stage 2) of OCR. More
precisely, we find in study 1 that the effects of positive and negative situational
experience on OCR are exacerbated (weakened) when the level of accumulated
experience is low (high). Finally, we consistently find in study 2 that the effect of OCR
on repurchase intentions is stronger when the valence of service experience is
negative rather than positive.
Theoretical background
Consumers continue to evaluate the service after the actual interaction (Grönroos
and Voima 2013) and our conceptual framework is rooted within the expectationdisconfirmation framework (Oliver, 1980) to acknowledge this situation, which
Grönroos and Voima (2013) term the ‘customer sphere’. Building on this situation, we
view satisfaction as a function of expectations and expectancy disconfirmation. In
other words, evaluations after the service interaction depend on the actual service
experience exceeding or equaling the expected service (Grönroos, 1984). Hence, we
believe that the interplay of service experience and online consumer ratings depends
on consumer expectations. In the next sections, we present our focal concepts of
service experience and online consumer ratings, before building hypotheses through
the lenses of consumer expectations. This leads to a two-stage comprehensive
framework about the effects of service experience on the antecedents (stage 1) and
consequences (stage 2) of OCR.
Service experience
Consistent with prior literature on OCRs (Zhu and Zhang, 2010; Zhang et al., 2010;
Yang and Li, 2010), we focus on services, and the case of hotels. Services are
difficult to evaluate before consumption (cf. Parasuraman et al., 1985) and
understanding how consumers perceive value in service experiences is challenge for
managers (Helkkula, Kelleher and Pihlström 2012; Holmqvist, Guest and Grönroos
2015). The intangible nature of services can increase the difficulty of evaluating the
service before consumption (Crosby et al., 1990; Laroche et al., 2004), which in turn
can have a negative influence on customers (Conchar et al., 2004; Grewal et al.,
1994). To avoid this kind of negative influence, the front line service personnel needs
to interact with customers in a way that renders the service interaction enjoyable
(Bitner, Booms and Mohr 1994; Gwinner et al. 2005).
We distinguish two concepts of service experience. First, the valence of situational
service experience may be defined as the positive, negative or mixed feeling that a
consumer may experience after the consumption of a given service. Second, the
level of accumulated service experience is a concept close to domain-specific form of
consumer familiarity, or purchase-frequency. More precisely, it may be defined as the
number of purchase-related service experiences.
Antecedents and consequences of online ratings
Whereas eWOM involves informal electronic communications among consumers
(Liu, 2006), OCRs may be defined as a specific form of eWOM that consists of (a) a
personal communication, (b) in the form of a qualitative comment and/or a
quantitative rating, (c) from an online rating system. Although the textual reviews are
usually not mandatory, online rating systems systematically include both individual
and aggregated quantitative reviews. In this research, we focus mainly on the ratings
rather than the number of reviews, although we control for the number reviews in our
design and analyses when necessary. Moreover, we use textual reviews as a way to
measure the valence of service experience.
We develop a two-stage conceptual framework. In the first stage, we focus on the
antecedents of OCR, examining the effect of a consumer’s service experience on the
subsequent online rating, and the moderating role of the consumer’s level of service
experience. In the second stage, we examine the impact of service experience on the
consequences of OCR. Hence, we examine the effects of other consumers’ online
ratings (i.e. the average rating from previous reviews) on subsequent purchase
decisions, and the moderating roles of the valence and the level of service
experience. Figure 1 presents an overview of our conceptual framework and the
hypotheses that we further develop in the next two sections.
[Insert Figure 1 about here]
Stage 1: The effects of service experience on the antecedents of reviewer’s OPR
First, although we do not develop specific hypotheses for previously established
results, we expect that positive valence of service experience positively influences
the reviewer’s rating (Mano and Olivier, 1993; Boulding et al. 1993; Anderson et al.,
1994; Sridhar and Srinisavan, 2012). Similarly, we expect that other consumers’
ratings will influence the influence the reviewer’s rating.
Second, the consumer’s level of service experience may have a direct effect on
the rating. In fact, previous research has shown that satisfaction is relative to the
difference between the consumer’s expectations and the actual experience. Zeithaml
et al. (1993) suggest that past experience is an antecedent of consumer expectations
of service such that higher levels of experience lead to higher expectations. In turn,
experienced consumers with higher expectations will be more likely to leave more
negative ratings. Hence, we posit:
H1: The level of accumulated experience has a negative influence on the
reviewer’s online rating
Third, if the valence of service experience influences the reviewer’s rating, two
consumers who shared the exact same experience may leave different online
reviews. Our hypothesis suggests that this positive or negative effect will be stronger
when the level of service experience is low rather than high. In other words,
compared to experienced consumers, unexperienced consumers will leave a
relatively higher rating after a positive experience, but they will leave a relatively
lower rating after a negative experience. This hypothesis may be explained through a
difference in expectations. Experienced consumers such as recurrent hotel guests
have clear and concrete expectations, they know what a good or bad service is like.
They have high expectations for good services and low expectations for bad
services. Unexperienced users, on the other hand, have no specific expectations
since they do not yet differentiate good and bad services. Thus, experienced
consumers may not be surprised by a positive (vs. negative) experience, which just
meets or slightly exceed (vs. fall behind) their concrete expectations. However,
unexperienced consumers may be surprised by a positive (vs. negative), which
considerably exceed (vs. fall behind) their vague expectations. In turn, when the
service experience is positive (vs. negative), unexperienced consumers will leave a
relatively higher (vs. lower) rating than experienced consumers. Finally, we posit the
following:
H2: The effect of the valence of service experience on the reviewer’s rating is
stronger when the level of experience is low rather than high
Stage 2: Effects of service experience on consequences of other consumers’ OPR
As the perceived service quality depends on the actual service experience exceeding
or equaling the expected service (cf. Grönroos 1984), we further posit that the impact
of OCRs on the decision to engage a second time is higher when the service
experience is negative. If the service experience is positive, the customer is likely to
perceive added value and thus more likely to return regardless of what online
reviewers had said. In the case of a negative service experience, however, a highly
positive OCR could still indicate that the service usually is better than the one the
customer experiences, thus making the customer more likely to give the service
provider a second change even after a negative service experience. We thus
hypothesize:
H3: The effect of OCR on repurchase intentions is stronger when the
valence of service experience is negative rather than positive.
Similarly, we expect that the level of service experience moderates the relationship
between OCR and purchase decision in the same direction – controlling for the
valence of service experience. Zhu and Zhang (2010) showed that the effect of OCR
on sales is stronger when targeting consumers with greater Internet experience. This
result occurs because greater Internet experience supposedly reduces search costs
and enables convenient comparison between alternatives. On the contrary, we
expect that the effect of OCR on repurchase intentions is stronger for consumers with
a low level of accumulated service experience. By definition, unexperienced
consumers have no clear idea about what makes a good or a bad service. Hence,
their purchase decision may be more influenced by OCR compared to experienced
consumers.
H4: The effect of OCR on repurchase intentions is stronger when the level
of service experience is low rather than high
Study 1
Previous literature has established that the valence of service experience is the main
antecedent of OCR. However, little is known about the potentially moderating effect
of the level of accumulated service experience on this relationship. The aim of study
1 is to demonstrate that the effects of positive and negative situational experience on
reviewer’s online ratings are exacerbated (weakened) when the level of accumulated
experience is low (high).
Method
We collected data on the website yelp.com using a web scrapping algorithm
developed by the authors on Python. We selected 4 large American cities (San
Francisco, Boston, Chicago and Seattle) in which we randomly selected 50 hotels out
of the 100 first hotels that appeared on the search engine when specifying no search
options. For each hotel, we collected up to 40 most recent reviews posted on the
website, or less if the total number of reviews for the hotel was lower than 40 (this
situation happened for 22 hotels in our sample, that is 11%). The average number of
observations per hotel is 38.7 (min=12, max=40), with a total sample of 7739 reviews
for 200 hotels in 4 cities.
We collected several focal and control variables, as presented in Table 1 (See
Appendices A and B for descriptive statistics and correlations). First, the reviewer’s
rating on a 5-point scale is our dependent variable. Second, the valence of service
experience was measured according to the textual comments attached to the review.
We performed a sentiment analysis using a naïve Bayesian classifier on R1. Trained
on Wilson et al. (2005)’s subjectivity lexicon, the classifier computes the likelihood
that the review expresses a positive and a negative service experience based on
word occurrences. Third, the reviewer’s number of previous reviews posted on Yelp
was used as a proxy for the level of accumulated service experience. Note that we
cannot track if those previous reviews were for hotels of for other services. However,
we can easily suppose that the percentage of hotel vs. other reviews is stable across
reviewers. Hence, the number of previous reviews still appear as a good proxy for
the level of service experience.
We also collected control variables at the reviewer level, such as (a) sociability
measured with the number of friends on Yelp, (b) the fact that the reviewer lives in
the city from which he reviewed the hotel and (c) reviewer’s gender. Since Yelp only
displays the name of the reviewer, we use a gender prediction classifier based on the
US Social Security Administration, which is said to be about 82% accurate2. Last, we
control for variables at the hotel level, such as the average rating for the hotel, the
total number of reviews for the hotel, as well as the price category.
[Insert Table 1 about here]
Empirical model
Consistent with the structure of our data, we use a hierarchical linear model with
three levels: city (level 3), hotel (level 2) and the reviewer (level 1). Using a variance
component model, we are able to estimate the variability in ratings accounted for by
each of the three hierarchical levels. The model can be written as:
𝑅𝐴𝑇𝐼𝑁𝐺𝑖𝑗𝑘 = 𝛽0 + ∑ 𝛽𝑛 𝑋𝑛𝑖𝑗𝑘 + 𝑢𝑖 + 𝑣𝑖𝑗 + 𝑒𝑖𝑗𝑘
𝑛
where RATINGijk is the dependent variable for review k, about hotel j in city i. First,
the fixed part of the model includes the n independent variables Xn. Second, the
random part of the model includes a city specific effect (ui) as well as a hotel specific
effect (vij), while eijk represents the reviewer level residual error. ui, vij, and eijk are
assumed independent of one another and normally distributed with zero means and
constant variances.
The intraclass correlation coefficient (ICC) measures the relatedness of the
reviewers within a group, such as a city or hotel. It is the ratio of the variance
component due to cities or hotels to the total variance for individual reviewers. In
model HLM1, the ICC for the reviewers within hotels is .167, so 16.7% of the
variance may be attributed to hotel traits. As the ICC shows a fair amount of variation
across hotels, model HLM2 adds three hotel-level variables, and greatly improves the
estimation, such that the ICC drops to .024.
[Insert Table 2 about here]
Results
The estimation results are displayed in Table 2. Models HLM1 and HLM2 allow us to
verify the appropriateness of the HLM estimation technique given the structure of our
data. Model HLM3 allows the estimation of direct effects for the focal variables, and
HLM4 allows the estimation of the hypothesized interactions.
First, consistent with previous literature, other’s consumers’ ratings, measured as
the average hotel rating, have a strong influence on reviewer’s ratings (HLM3:  =
1
2
Timothy Jurka’s “sentiment” package: http://cran.r-project.org/src/contrib/Archive/sentiment/
Stephen Holiday’s “genderPredictor” code: https://github.com/sholiday/genderPredictor
.67, t = 22.14, p < .01). Moreover, the positive (HLM3:  = .01, t = 34.96, p < .01) and
negative (HLM3:  = -.01, t = -26.90, p < .01) valence of service experience are also
important drivers of reviewer’s ratings, with the greater t-values. However, while
reviewing experience significantly impacts future ratings (HLM3:  = .01, t = 3.49, p <
.05), this link is positive in contradiction with the negative hypothesized effect. H1 is
not validated.
Second, we find that reviewing experience moderates the strong effect of positive
(HLM4:  = -.01, t = -8.39, p < .01) and negative (HLM4:  = .01, t = 8.91, p < .01)
service experience on reviewer’s ratings. Hence, the effects of positive and negative
experience are exacerbated when reviewing experience is low, and weakened when
reviewing experience is high. H2 is supported.
Study 2
Previous literature has established the strong consequential effects of previous
customers’ online ratings on potential customers’ purchase decisions. However, little
is known about the potentially moderating effect of service experience on this
relationship. The aim of study 2 is to demonstrate that the effect of OCR on
repurchase intentions is stronger when the valence (level) of service experience is
negative (low) rather than positive (high).
Method
We use a 2 (OCR valence: 2.7 vs. 3.9 on a 5-point rating scale) × 2 (service
experience: positive vs. negative) between-subject experiment. We collect data
through the online panel of a marketing research company, in which respondents
receive either payment or coupons for participation. 403 French participants were
randomly assigned to one of the four conditions of this web-based study, with 51%
women and an average age of 39 years (from 18 to 69).
Setup. Participants were shown an Internet page of a fictional hotel (“Hotel Bella
Fiorentina”, in Florence, Italy) from a fictional website (“www.reservation-hotel.com”).
A photo and a neutral description of the hotel were added to make the offer more
realistic (see Appendix C). In order to make the webpage more realistic along the
lines of existing review sites, we added five written comments from previous
customers, including positive (e.g. “friendly staff”, “good localization”) and negative
features of the hotel (“noisy”, “staff not friendly”). All comments were modeled based
on real comments left by customers on actual review sites. Recent service research
suggests that customers perceive less risk when using their first language (Holmqvist
& Grönroos, 2012), and this is mirrored by most online review sites automatically
displaying reviews in the customer's language first. In line with these practices, all the
displayed comments from previous customers were in French. The respondents were
required to look at the hotel page (and then the experience scenario) for at least one
minute and could stay on the page as long as they wanted before continuing the
questionnaire.
Manipulations. We chose the stimuli of online consumer ratings based on real
ratings for Florence hotels from three existing websites, which varied from 1.5 to 5 on
a 5-point rating scale. We selected less extreme values of OCR (low: 2.6 vs. high:
3.9 on a 5-point rating scale), to reduce the possibility of floor effects, i.e., the stimuli
being too negative or too positive (Elder & Krishna, 2012). The number of reviews
was the same in the two conditions (249 reviews).
The positive and negative valence of situational service experience were
manipulated using two hypothetical textual scenarios (about 500 words each, see
Appendix D). Compared to the positive scenario in which everything was good, the
negative scenario includes several negative hotel features such as waiting at the
reception, smell in the room, and noise in the room that caused sleeping problems.
Manipulation check. We verify our manipulation with a service experience scale (
= .86), composed of the three following 7-point differential scale items “Bad/good”,
“Unfriendly/friendly”, “low-level/high-level”). The manipulation check was significant,
the valence of service experience was rated more positively in the positive than in the
negative scenario (M = 5.86 vs. M = 4.59, F(1,401) = 105.12, p < .01).
Measures. The main dependent variable was the repurchase intentions ( = .84)
measured on a 3-item scales scored on a 7-point Likert format, including “If I have
the opportunity, I would stay in this hotel again”, “As long as the service quality
remains the same, I would rather return to this hotel rather than discover others in the
neighborhood”, and “In the future, I will not come back to this hotel” (r). Moreover, we
measured the level of accumulated experience about the service category adapted
from Söderlund (2002), including the three following items: “I know a lot about
hotels”, “I consider myself to be an experienced hotel guest” and “I know very well
what characterizes good and bad hotels” ( = .86).
Results
We ran an ANOVA on repurchase intentions with OCR and the valence of service
experience as the independent variables. The main effect of OCR is not significant (p
= .13). The main effect of service experience is significant (F(1,399) = 355.33, p <
.01), such that repurchase intentions are stronger when the experience scenario is
positive (M = 5.49) rather than negative (M = 3.17). Moreover, the interaction
between OCR and the valence of service experience is significant (F(1,399) = 4.41, p
< .05), as depicted in Figure 3. We explore this interaction with planned contrasts. In
the positive experience condition, there is no difference in repurchase intentions
between the two valence conditions (p = .69). However, in the negative experience
condition, repurchase intentions are stronger for a high rating (M = 3.39) compared to
a low rating (M = 2.94, F(1,399) = 6.63, p < .05). H3 is thus supported.
[Insert Figure 2 about here]
We performed a regression on repurchase intentions with independent variables (i)
OCR, (ii) the level of accumulated experience with the service category and (iii) their
interaction. Controlling for the manipulated valence of service experience, the results
show a significant two-way interaction between OCR and the level of consumer
experience ( = -.19, t = -2.07, p < .05); Figure 4 depicts this interaction effect. To
decompose this interaction, we use the Johnson-Neyman technique to identity the
range(s) of the level of consumer experience for which the simple effect of OCR is
significant (Spiller et al., 2013). This analysis reveals that there is a significant
positive effect of OCR on repurchase intentions when the level of experience with the
service category was lower than 4.28 (JN = .24, p = .05) but not when the level of
experience with the service category was higher than 4.28, thus supporting H4.
[Insert Figure 3 about here]
General discussion
We contribute to the service literature by showing that the level of service experience
has a positive impact on the reviewer’s ratings. These results contradict our
negatively hypothesized effect based on the assumption that past experiences
increase the level of expectations (Zeithaml et al., 1993). Our explanation for this
result builds on previous research on usage frequency showing how usage frequency
positively impacts consumers’ evaluations (Goodman and Irmak; 2013; Hamilton et
al., 2011). Hamilton et al. (2011) show that frequency cues continue to influence
consumers’ evaluations after the actual interaction through perceived fit. Similarly, we
believe that the level of accumulated experience might lower expectations because
experienced consumers have a better idea about what fits their needs. In other
words, being a less experienced to staying in hotels may lead a consumer to be more
negative, due to some unrealistic expectations. In turn, experienced consumers have
more realistic expectations, and are more prone to leave a good rating.
Second, we show that the effects of positive and negative situational experience
on OCR are exacerbated (weakened) when the level of accumulated experience is
low (high). Previous literature has shown that external factors such as ratings by
other consumers’ can weaken or exacerbate the effects of the valence of service
experience on reviewer ratings (Sridhar and Srinisavan, 2012). We contribute to the
literature by demonstrating that the effects of the valence of service experience on
reviewer’s ratings are also contingent on individual characteristics such as the level
of service experience. This finding shows that not all consumers are equal in their
evaluation process because they may have higher or abstract expectations.
Third, we show that the valence of the service experience moderates the
relationship between OCR and repurchase intentions. If the service is good, there is
no difference between high and low OCR. However, in the negative scenario, the
repurchase intentions are significantly higher for a higher OCR. This result
contributes to the service literature by partly contradicting the existing order of
credibility in services messages by Duncan and Moriarty (1997) who posited that
WOM is more credible than the actual service interaction. Our results contradict this
order, as we show that customers who are satisfied with the service they received
tend to trust their own experience and no longer pay much attention to the OCRs
they read before engaging in the service. However, if the customer is dissatisfied,
customers who have read positive reviews are more willing to give the service firm a
second chance, perhaps thinking that their own negative experience was an
exception, while customers who have read negative reviews are reinforced in their
own negative evaluation.
Fourth, we show that the level of service experience moderates the relationship
between OCR and repurchase intentions. On one hand, previous literature has found
that the effect of OCR on sales is stronger when targeting consumers with greater
Internet experience (Zhu and Zhang, 2010) as this experience reduces search costs
and enables convenient comparisons between alternatives. On the other hand, we
contribute to the literature by showing that the effect of OCR on repurchase
intentions increases as the level of service experience decrease. In other words, it
means that compared to an experienced hotel guest, novice consumers are more
sensible to differences in OCR as they do not have specific expectations in terms of
what constitutes a good or bad service.
Managerial Implications
One important implication of Study 1 comes from the fact that consumers may leave
different ratings. First, this result has implications for booking websites. We show that
the valence and the level of service experience interact such that consumers may be
positively or negatively biased, leading to a difference between the overall rating and
the true quality of the service. In fact, booking websites calculate average ratings by
giving the same weight to all reviewers, and that average rating has strong
consequences in terms of service referencing within the website. Our results show
that booking website may over- or under-weight consumers in the calculation of the
average rating for a given service, so that it could more accurately reflect service
quality. Second, this result has implications for e-commerce firms, who can use our
methodology to estimate the impact of service experience on online ratings, isolating
the role of the level of accumulated experience, or any individual relevant individualvariable for that matter. As Sridhar and Srinisavan (2012) suggest, making sense of
thousands of online ratings can be a challenging task. Hence, this procedure would
allow them to better estimate the effects of service experience and OCR, as well as
better track down consumers who are more likely to leave good or bad ratings.
The implications from study 2 are two-fold. On the one hand, this study suggest
that managers should be concerned with having good OCR, but more importantly,
they heavily invest in creating a positive consumer experience. As a matter of fact,
the impact of consumer experience is relatively stronger than the impact of online
ratings. In a way, if the service provider offers a nice consumer experience, it can
outshine medium or even lower online ratings. On the other hand, the study shows
that consumers’ own experience are more important than other consumers’
experiences in determining their evaluation and re-repurchase intentions. This carries
important implications in the sense that service providers need to invest in convincing
customer to “try” the service. In this regard, they might offer promotions or discounts,
especially for targeted customers who are booking a service for the first time.
Limitations and Future Research
The findings of this manuscript present a number of limitations and some possible
avenues for further research. First, although online hotel reviews are among the most
common OCRs for services, focusing on a single type of service still limits the
generalizability of this work. Further research may replicate these findings in other
categories. Second, there is a limitation about the realism in study 3. One might ask
(1) why customers would visit a certain hotel when they know it has a low rating, (2)
how they feel about a low (high) rating and a positive (negative) experience. In order
to answer these questions, we ran an ANOVA on perceived realism (“The situation
described in this scenario might happen in reality”). While the effect of OCR is not
significant (p=.11), the effect of the valence of service experience is significant
(F(1,399) = 28.74, p < .01). However, respondents felt that the negative scenario
(M=6.26) was more realistic than the positive scenario (M=5.72). Moreover, the
interaction effect between OCR and the valence of service experience was not
significant (p=.85). There was no difference in realism within the negative experience
condition (p=.31), or within the positive experience condition (p=.20). Hence, these
results rule out the possibility that current results are more demand effects rather
than effects that could be observed in real life.
Another limitation is that the studies are confined to situations after the actual
service interaction and thus open to hindsight bias. Future research may benefit from
interviewing customers after they engaged in an actual service, provided that data
about their perceptions of OCRs have been collected previous to the service to avoid
the risk that the actual service experience influence how the respondents evaluate
their perceptions of the OCRs with hindsight. Last, given our results embedded in the
expectations-disconfirmation framework, an interesting avenue for future research
would be to examine the impact of OCR on attention bias. Consumers might be
looking for cues in the hotel room that might confirm the negative reviews more
easily. It would maybe be easier to spot of piece of dust in the bathroom after reading
negative reviews. However, consumers might overlook or barely notice small
negative features after reading positive reviews.
References
Anderson, E. W., Fornell, C., & Lehmann, D. R. (1994). Customer satisfaction,
market share, and profitability: Findings from Sweden. Journal of Marketing, 58(3),
53-66.
Bitner, M.J., Booms, B.H. & Mohr, L.A. (1994). Critical Service Encounters: The
Employee's Viewpoint, Journal of Marketing, 58(4), 95-106.
Boulding, W., Kalra, A., Staelin, R., & Zeithaml, V. A. (1993). A Dynamic Process
Model of Service Quality: From Expectations to Behavioral Intentions. Journal of
Marketing Research, 30 (1), 7-27.
Cadario, R. (2014). The impact of online word-of-mouth on television show
viewership: An inverted U-shaped temporal dynamic. Marketing Letters, DOI
10.1007/s11002-013-9278-6.
Chevalier, J., & Mayzlin, D. (2006). The effect of word of mouth on sales: Online
book reviews. Journal of Marketing Research, 43(3), 345-354.
Conchar, M.P., Zinkhan, G.M., Peters, C. and Olavarrieta, S. (2004). An integrated
framework for the conceptualization of consumers’ perceived risk processing.
Journal of the Academy of Marketing Science, 32 (4), 373–89.
Crosby, L.A., Evans, K.R. and Cowles, D. (1990). Relationship quality in services
selling: An interpersonal influence perspective. Journal of Marketing, 54(3), 68–81.
Duncan, T. & Moriarty, S. (1997) Driving Brand Value. New York: McGraw Hill.
Elder, R. S., & Krishna, A. (2012). The “visual depiction effect” in advertising:
Facilitating embodied mental simulation through product orientation. Journal of
Consumer Research, 38(2), 1-17.
Goodman, J. K., & Irmak, C. (2013). Having versus consuming: failure to estimate
usage frequency makes consumers prefer multifeature products. Journal of
Marketing Research, 50(1), 44-54.
Grewal, D., Jerry Gotlieb, and Howard Marmorstein (1994). The moderating effects
of message framing and source credibility on the price-perceived risk relationship.
Journal of Consumer Research, 21 (1), 145–153.
Grönroos, C. (1984). A service quality model and its marketing implications.
European Journal of Marketing, 18(4), 36-44.
Grönroos, C. (2007). Service Management and Marketing: Customer management in
service competition. John Wiley & Sons: Chichester.
Grönroos, C. and Voima, P. (2013). Critical service logic: making sense of value
creation and co-creation. Journal of the Academy of Marketing Science, 41 (2),
133-150.
Gruen, T. W., Osmonbekov, T., & Czaplewski, A. J. (2006). eWOM: the impact of
customer-to- customer online know-how exchange on customer value and loyalty.
Journal of Business Research, 59(4), 449-56.
Gwinner, K., Bitner, M.J., Brown, S.W. & Kumar, A. (2005). Service Customization
Through Employee Adaptiveness. Journal of Service Research, 8(2), 131-148.
Hamilton, R. W., Ratner, R. K., & Thompson, D. V. (2011). Outpacing others: When
consumers value products based on relative usage frequency. Journal of
Consumer Research, 37(6), 1079-1094.
Helkkula, A., Kelleher, C. & Pihlström, M. (2012). Characterizing value as an
experience: implications for service researchers and managers. Journal of Service
Research, 15(1), 59-75.
Holmqvist, J. & Grönroos, C. (2012). How Does Language Matter for Services?
Challenges and Propositions for Service Research. Journal of Service Research,
15 (4) 430-442.
Holmqvist, J., Guest, D., & Grönroos, C. (2015). The role of psychological distance in
value creation. Management Decision, 53 (7) 1430-1451.
Kim, J., & Gupta, P. (2012). Emotional expressions in online user reviews: How they
influence consumers’ product evaluations. Journal of Business Research, 65(7),
985-992.
Laroche, M., McDougall, G. H., Bergeron, J., & Yang, Z. (2004). Exploring how
intangibility affects perceived risk. Journal of Service Research, 6(4), 373-389.
Liu, Y. (2006). Word of mouth for movies: Its dynamics and impact on box office
revenue. Journal of Marketing, 70(3), 74-89.
Mano, H., & Oliver, R. L. (1993). Assessing the dimensionality and structure of the
consumption experience: evaluation, feeling, and satisfaction. Journal of
Consumer Research, 451-466.
Nielsen (2009). Global advertising: Consumers trust real friends and virtual strangers
the most. Retrieved from http://blog.nielsen.com/nielsenwire/consumer/globaladvertising-consumers-trust-real-friends-and-virtual-strangers-the-most.
Oliver, R. L. (1980). A cognitive model of the antecedents and consequences of
satisfaction decisions. Journal of Marketing Research, 17(4), 460-469.
Parasuraman, A., Zeithaml, V. A., & Berry, L. L. (1985). A conceptual model of
service quality and its implications for future research. Journal of Marketing, 41-50.
Park, C., & Lee, T. M. (2009). Information direction, website reputation and eWom
effect: The moderating role of product type. Journal of Business Research, 62(1),
61-67.
Söderlund, M. (2002). Customer familiarity and its effects on satisfaction and
behavioral intentions. Psychology & Marketing, 19(10), 861-879.
Spiller, S. A., Fitzsimons, G. J., Lynch JR, J. G., & McClelland, G. H. (2013).
Spotlights, floodlights, and the magic number zero: Simple effects tests in
moderated regression. Journal of Marketing Research, 50(2), 277-288.
Sridhar, S., & Srinivasan, R. (2012). Social influence effects in online product
ratings. Journal of Marketing, 76(5), 70-88.
Yang, J., & Mai, E. (2010). Experiential goods with network externalities effects: An
empirical study of online rating system. Journal of Business Research, 63(9-10),
1050-1057.
Zeithaml, V. A., Berry, L. L., & Parasuraman, A. (1993). The nature and determinants
of customer expectations of service. Journal of the Academy of Marketing
Science, 21(1), 1-12.
Zhang, J. Q., Craciun, G., & Shin, D. (2010). When does electronic word-of-mouth
matter? A study of consumer product reviews. Journal of Business Research,
63(12), 1336-1341.
Zhu, F., & Zhang, X. (2010). Impact of online consumer reviews on sales: The
moderating role of product and consumer characteristics. Journal of Marketing,
74(2), 133-148.
Wilson, T., Wiebe, J. and Hoffmann, P. (2005). Recognizing Contextual Polarity in
Phrase-Level Sentiment Analysis. Proc. of HLT-EMNLP-2005.
Appendix A – Descriptive statistics (Study 1)
Variable
RATING
POS_EXP
NEG_EXP
PREV_REVIEWS
FROM_CITY
GENDER
CHAR
FRIENDS
HOTEL_AVG
HOTEL_NUM
PRICE_CAT
Obs.
7739
7739
7739
7739
7739
7739
7739
7739
7739
7739
7739
M
3.74
70.88
41.55
193.29
.10
.51
932.42
136.55
3.78
155.59
.51
S.D.
1.23
47.31
38.02
344.95
.30
.50
720.45
446.19
.49
122.86
.51
Min
Max
1
1.03
.44
1
0
0
3
0
2.50
12
0
5
432.32
392.24
4356
1
1
5144
5000
4.50
800
1
Appendix B – Correlation table (Study 1)
Variable
1. RATING
2. POS_EXP
3. NEG_EXP
4. PREV_REVIEWS
5. FROM_CITY
6. GENDER
7. CHAR
8. FRIENDS
9. HOTEL_AVG
10. HOTEL_NUM
11. PRICE_CAT
1
2
3
4
5
6
7
8
9
10
1
.20
-.29
.05
.03
.02
-.09
.05
.38
-.10
.01
1
.55
.14
.03
.04
.78
.16
.10
-.01
.01
1
.10
-.01
.03
.80
.11
-.12
.05
-.03
1
.05
.01
.14
.64
.01
-.03
.03
1
.00
.02
.02
.07
-.03
.02
1
.04
.03
.01
.00
-.03
1
.14
-.02
.01
-.01
1
.02
.01
-.01
1
-.21
.05
1
.39
Note: Correlations above .03 are significant at p<.05 and above.
11
1
Appendix C – Example of a stimulus used in Study 2 (translated from French to English)
Appendix D – Scenarios for the valence of service experience (Study 2)
Negative: It is Friday afternoon and you arrive in
Florence, where you have booked two nights at Bella
Fiorentina via reservation-hotels.com. A taxi from the
airport drops you off at the hotel in the historical centre
of the city. When you arrive at the hotel, you have to
wait a couple of minutes as the receptionist is talking
on the phone. She then checks you in and gives you
the key to your room.
You notice that your room looks exactly like the
pictures on Reservation-Hotels. Both the room and the
bathroom are nice but there is a slight smell in the
bathroom. The room is at the first floor with a view of
the street outside. After unpacking your bags, you
swiftly leave the hotel to have a first look at Florence.
After some hours of sightseeing you return to the hotel.
At the reception, you ask the receptionist for advice
about a restaurant for the evening. She says that there
is a good restaurant at the other side of the river, and
gives you the address of that restaurant.
You decide to take her advice and go to the restaurant,
and you have a nice meal. After the dinner, you walk
around a bit to look at Florence at night before
returning to your hotel. You try to sleep but notice that
there is no sound isolation in the room, and you can
hear both people in street outside the hotel as well as
people talking in the room next to yours. You
eventually fall asleep, but are awaken two times during
the night by sounds from outside.
The next day you have planned several activities. In
the morning, you follow a tour of Tuscany. In the
afternoon, you visit the largest art gallery in Italy, the
Uffizi gallery. It is early evening when you get back to
the hotel, and this time there is another receptionist.
You ask for advice for things to do in the evening. He
recommends a visit to a different restaurant. He says it
is by the river and that it is possible to have drinks
there later in the evening, and he gives you the
address. You go to the restaurant he recommends; the
meal is good and the atmosphere later at night is also
very nice with lots of people relaxing and having
drinks. You get back to your hotel quite late. Just like
the previous night, there is some noise from outside
and you wake up three times during the night.
On Sunday your flight is at 13:30 so you have time for
a final stroll in Florence in the morning before going to
the hotel to check out. The receptionist says that there
is a shuttle close to the hotel going to the airport. You
pay 150€ by credit card for the two nights, the
receptionist then informs you that you also need to pay
6€ in cash, 3€ per night for the city tax on hotels.
Having paid your hotel, you take the shuttle and head
out to the airport to fly back to France.
Positive: It is Friday afternoon and you arrive in
Florence, where you have booked two nights at Bella
Fiorentina via reservation-hotels.com. A taxi from the
airport drops you off at the hotel in the historical center
of the city. When you arrive at the hotel, you are
immediately greeted by the receptionist who welcomes
you warmly to the hotel and wishes you a nice stay in
Florence. She checks you in and accompanies you to
your room.
You notice that your room looks exactly like the
pictures on Reservation-Hotels. Both the room and the
bathroom are nice and clean. The room is at the top
floor with a view over the roofs of Florence. After
unpacking your bags, you swiftly leave the hotel to
have a first look at Florence. After some hours of
sightseeing you return to the hotel. At the reception,
you ask the receptionist for advice about a restaurant
for the evening. She smiles and says that her own
favorite restaurant is tucked away in a small alley at
the other side of the Arno river, and gives you a map of
the city where she draws you the way to go.
You decide to take her advice and go to the restaurant,
and you have a nice meal. After the dinner, you walk
around a bit to look at Florence at night before
returning to your hotel. The bed in your room is very
comfortable and you fall asleep almost immediately,
sleeping well until the morning.
The next day you have planned several activities. In
the morning, you follow a tour of Tuscany. In the
afternoon, you visit the largest art gallery in Italy, the
Uffizi gallery. It is early evening when you get back to
the hotel, and this time there is another receptionist.
You ask for advice for things to do in the evening. He
recommends a visit to a different restaurant. He says it
has a nice view over the river and that later at night it
turns into a nice lounge where people come to have
drinks and relax. Just like the other receptionist, he
draws you the route to the restaurant on the map. You
go to the restaurant he recommends; the meal is good
and the atmosphere later at night is also very nice with
lots of people relaxing and having drinks. You get back
to your hotel quite late. Just like the previous night, you
again sleep very well.
On Sunday your flight is at 13:30 so you have time for
a final stroll in Florence in the morning before going to
the hotel to check out. The receptionist asks if you
have enjoyed your stay, and she offers to call you a
taxi but also says that there is a shuttle leaving just
100 meters from the hotel, going directly to the airport.
You pay 150€ by credit card for the two nights and the
receptionist wishes you a nice trip home. Having paid
your hotel, you take the shuttle and head out to the
airport to fly back to France.
Table 1 – List of the variables (Study 1)
Level-1: reviewer
Focal variables
RATING
POSITIVE_EXP
NEGATIVE_EXP
LEVEL_EXP
Control variables
FROM_CITY
GENDER
CHAR
FRIENDS
Level-2: hotel
Control variables
HOTEL_AVG
HOTEL_NUM
PRICE_CAT
Reviewer’s rating on a 5-point scale (DV)
Loglikelihood of the review expressing a positive experience
Loglikelihood of the review expressing a negative experience
Reviewer’s number of previous reviews
Reviewer is from the city where the hotel is: 1 (0 if not)
Gender (female: 1, male: 0)
Number of characters in the textual review
Number of friends on Yelp
Average rating for the hotel
Total number of reviews for the hotel
Price category (0: $,$$; 1: $$$,$$$$)
Table 2 – Estimation results from the hierarchical linear models (Study 1)
Fixed-effects
Intercept
HLM1
3.751***
(79.93)
HLM2
.252
(1.63)
.939***
(24.45)
-.039**
(-2.23)
.005
(.12)
HLM3
1.096***
(9.09)
.668***
(22.14)
-.000
(-1.29)
-.065**
(-2.07)
.007
(.22)
.044**
(1.97)
-.046
(-1.22)
-.030***
(-8.30)
.015**
(3.49)
.014***
(34.96)
-.014***
(-26.90)
HLM4
1.106***
(9.17)
.660***
(21.94)
-.000
(-1.27)
-.059*
(-1.88)
.002
(.64)
.043*
(1.94)
-.050
(-1.32)
-.028***
(-7.64)
.025**
(3.87)
.015***
(35.66)
-.016***
(-28.31)
-.007***
(-8.39)
.009***
(8.91)
.003
.252
1.276
.002
.167
.003
.029
1.276
.002
.024
.000
.013
.970
.000
.013
.000
.013
.959
.000
.013
24292
24320
24019
24068
21957
22054
21922
22034
HOTEL_AVG
HOTEL_NUM
a
PRICE_CAT
a
FRIENDS
GENDER
FROM_CITY
a
CHAR
a
LEVEL_EXP (H1)
POSITIVE_EXP
NEGATIVE_EXP
a
POSITIVE_EXP * LEVEL_EXP (H2a)
a
NEGATIVE_EXP * LEVEL_EXP (H2b)
Random-effects
City: var(ui)
Hotel: var(vij)
Residual: var(eijk)
b
ICC within city
b
ICC within hotel
Model Fit
AIC
BIC
Note: RATING is the dependent variable.
t-statistics between parentheses
*p<.10, **p<.05, ***p<.01
a
Coefficients and standard errors for these variables are multiplied by 100 for readability.
b
ICC: intraclass correlation coefficient
Figure 1. Two-stage theoretical framework and studies overview
Figure 2. Interaction effect between OCR and the valence of service experience (Study 2)
Repurchase intentions
6
5,52 5,46
5
4
3,39
2,94
3
2
Negative
Positive
Valence of service experience
Low OCR (2.6 on 5)
High OCR (3.9 on 5)
Figure 4. Interaction effect between OCR and the level of consumer experience (Study 2)
Repurchase intentions
3,5
3
2,5
2
1
2
3
4
5
Level of service experience
Low OCR (2.6 on 5)
High OCR (3.9 on 5)
6
7