You are on page 1of 19

JID: RETAIL

ARTICLE IN PRESS [m5+;September 25, 2023;11:20]

Available online at www.sciencedirect.com

Journal of Retailing xxx (xxxx) xxx


www.elsevier.com/locate/jretai

The effect of review images on review helpfulness: A contingency


approach
Raoul V. Kübler a,∗, Lara Lobschat b, Lina Welke a, Hugo van der Meij c
a ESSEC Business School, Cergy, France
b University of Maastricht University, the Netherlands
c Free Researcher

Available online xxx

Abstract
Online retailing is still dominated by information asymmetries, as it often remains difficult for consumers to fully judge the quality of a
product online. Reviews written by customers help to reduce this asymmetry. Helpful reviews have thus become an important tool to drive
online sales. Beside textual information, reviews nowadays also often include images that can further help consumers to better judge products
or services. While online retailers need to invest substantial resources in hosting and incentivizing review images, it remains unclear under
which conditions review images drive (or reduce) review helpfulness and how review image content affects review helpfulness. We rely on
a set of more than 97,000 reviews from Amazon to investigate the contingencies under which review images increase review helpfulness.
Furthermore, we rely on more than 6,000 images in our data set to explore how review image content (i.e., image focus and context fit)
drives review helpfulness. Our results show that online retailers should especially motivate consumers to include images in a review when
the overall rating is extremely positive, when the reviewer has a high reputation, and when the review addresses a hedonic or experience
product. Our image content analysis further shows that images help to increase helpfulness when they show the product in application. This
effect is especially strong in the case of longer reviews.
© 2023 The Authors. Published by Elsevier Inc. on behalf of New York University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Keywords: User-generated content; Electronic word of mouth; Online product reviews; Image analysis.

Online product reviews have seen a substantial increase e.g., Srivastava and Kalro 2019). Consequently, an increasing
in interest within the last decade. As a valid tool to help proportion of online reviews is posted with user-generated im-
consumers reduce information asymmetries while shopping ages, as e-commerce platforms like Alibaba or Amazon start
online (Swaminathan et al. 2020) and make better-informed to leverage the potential of latest developments in smartphone
decisions (Forman, Ghose, and Wiesenfeld 2008), they have cameras and provide incentives to users for adding images to
been shown to be an effective tool to positively affect their product reviews.
consumer’s decision making (Kübler et al. 2018), drive e- While initial research has established strong empirical ev-
commerce (Babić Rosario et al. 2016) and improve an online idence for the average positive impact of review images (and
retailer’s bottom line profits (Chevalier and Mayzlin 2006). thus subsequently on buying behavior and customer satisfac-
A review’s positive influence on buying behavior depends on tion; see e.g., Ceylan, Diehl, and Proserpio 2023), it remains
its helpfulness (Topaloglu and Dass 2021). Recent research unclear in which cases review images are especially help-
points out that review helpfulness is driven by non-textual el- ful and if review images may in some cases even decrease
ements, such as images and videos attached to a review (see helpfulness. The impact of review images on the helpfulness
of the review may thus be contingent on the capability of
these images to reduce information asymmetries by reducing
∗ Corresponding author. a customer’s uncertainty about a product, online retailer, or
E-mail address: kubler@essec.edu (R.V. Kübler). service.

https://doi.org/10.1016/j.jretai.2023.09.001
0022-4359/© 2023 The Authors. Published by Elsevier Inc. on behalf of New York University. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Please cite this article as: R.V. Kübler, L. Lobschat, L. Welke et al., The effect of review images on review helpfulness: A contingency approach, Journal
of Retailing, https:// doi.org/ 10.1016/ j.jretai.2023.09.001
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

The magnitude of information asymmetry in an online re- reviews and that reviews for hedonic and experience prod-
tail setting may depend on various factors. For example, the ucts profit more (than utilitarian and search products) from
fit of a fashion piece is harder to judge for a consumer on- review images. Furthermore, online retailers should educate
line than offline. Review images may thus be more helpful consumers to show the reviewed product in application (in-
for clothing than for other products when it comes to reduc- stead of only showing the plain product) when producing an
ing information asymmetry (in comparison to e.g., technical image for a review. Our results further indicate that low con-
products with more objective product characteristics). Simi- text fit plays a special role in the case of more negative re-
larly, the consumption purpose may further increase informa- views, while the application focus is more important in case
tion asymmetry and thus affect the impact of review images. of longer reviews with more text.
In case of a hedonic product, where consumers are more sub-
jective (compared to utilitarian consumption), consumers may Information asymmetry and the role of images in reviews
need more cues and aids when assessing a product and its re-
view. Here again, images may be more helpful than in other In online shopping, the geographical and temporal barri-
settings to reduce information asymmetry and increase review ers between retailers and shoppers are lifted as consumers
helpfulness. In order to grasp the various boundary and con- can buy products and services via the Internet anytime and
tingency conditions associated with review images, it is im- anywhere (Dimoka, Hong, and Pavlou 2012). However, even
perative to cultivate a deeper comprehension of the dynamics though research argues that online (vs. offline) shopping al-
between images and other relevant factors in the review pro- lows consumers to easily access information about products
cess. These factors encompass review valence, review length, and services (Atefi et al. 2020), it is still also characterized by
reviewer reputation, and product-specific characteristics, all of information asymmetry – i.e., (online) retailers holding private
which contribute to the influence on review helpfulness. information on products and/or services which consumers are
Furthermore, the impact of review images on review help- not aware of (Ahearne et al. 2022) – that increases uncer-
fulness may not only be driven by the context, but also by tainty and thus may prevent consumers from buying a prod-
what the image displays. To be able to not only advise com- uct online. Possible information asymmetry arises from the
panies on when to incentivize image inclusion, but also to retailer being unknown (Kim and Krishnan 2015), its ambigu-
provide guidance on what to display, we build on previous re- ous level of reliability and credibility, and/or uncertainty about
search and investigate how image focus (application vs. prod- the product itself (Littler and Melanthiou 2006) or its char-
uct focus) and image context fit further affect review helpful- acteristics (e.g., quality) (Weathers, Sharma, and Wood 2007;
ness and how this is again moderated by review, reviewer, Dimoka, Hong, and Pavlou 2012). In today’s hyperconnected
and product characteristics. world, online reviews – as a main source of consumer infor-
We first test our contingency framework with the help of mation – serve as important information signals which help
a unique data set containing more than 97,000 Amazon re- to overcome information asymmetry and reduce uncertainty
views from 46 products to investigate how review-, reviewer-, about various product characteristics, such as product quality
and product-related characteristics affect the impact of images (Swaminathan et al. 2020). Prior research reveals that through
on review helpfulness. Subsequently, we analyze more than providing product reviews, online retailers enable consumers
6,000 review images posted within our sample and investigate to obtain information on experience attributes of products
how image focus and image context fit – in interaction with which are more difficult to convey via product descriptions
review-, reviewer-, and product-related factors – affect review (Hong and Pavlou 2014). Also, given that the information in
helpfulness. reviews comes from a third-party rather than directly from
In doing so, we make three important contributions to the the retailer, consumers will likely perceive the information
stream of online review research: First, we investigate in a to be more credible (Weathers, Sharma, and Wood 2007).
real e-commerce setting how the impact of review images on Through evaluating other consumers’ product experiences as
review helpfulness is contingent on review valence, review described in reviews, a consumer is provided with a more
length, and reviewer reputation. Second, we investigate, how realistic expectation of the product which helps in assessing
the impact of review images differs for search vs. experience the product’s fit with the consumer’s preferences (Hong and
and utilitarian vs. hedonic products as well as products with Pavlou 2014). However, the extent to which online reviews
varying brand strength. Third, to the best of our knowledge, help mitigate information asymmetry depends on the type of
our study is the first to explore how image content-specific information provided online which includes the review itself
factors affect review helpfulness and how this influence is (Dimoka, Hong, and Pavlou 2012) and the type of product
again contingent on review-, reviewer-, and product-related (Manes and Tchetchik 2018).
factors. Online reviews are commonly understood as a specific
Our analyses deliver important implications for managers. form of user-generated content often also categorized as
We find that online retailers should especially motivate (or in- eWOM (Hennig-Thurau et al. 2003). Many studies have
centivize) the upload of images with reviews that are overly shown the positive impact of reviews on online consump-
positive about a product. Similarly, we find that online re- tion (for an overview, see Karimi and Wang 2017), thereby
tailers should include specific incentives for reviewers with fostering research interest in what makes reviews more or
a high reputation to include one or multiple images in their less impactful within consumers’ decision-making process.

2
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Recently, researchers have started to explore the role of im- Zhao (2023) report that the effect of the amount of review
agery in online reviews, especially given the superiority of images on review helpfulness is more pronounced for expe-
images in capturing consumers’ attention (Pieters and Wedel rience than for search products. Hence, given these scatter
2004) which can be explained by the ease of formation of findings, it is not clear which product type will benefit most
both verbal and imaginal codes for pictures relative to words from the addition of review images. Moreover, both stud-
(see e.g., Unnava and Burnkrant 1991). Previous research ies only consider the differences between search and experi-
exploring the role of user-generated images1 for online re- ence products, whereas previous research also indicates sub-
views (for an overview, see Web Appendix A) has predom- stantial differences in consumers’ search process and need
inately looked at how the mere presence of one (or more) for information for hedonic versus utilitarian products (e.g.,
review image(s) impacts review helpfulness (e.g., Wu, Wu, Huang, Lurie, and Mitra 2009) and products which differ on
and Wang 2021), trust in a review, and purchase intentions brand strength (e.g., Ho-Dac, Carson, and Moore 2013).
(e.g., Zinko et al. 2020). With Li, Chen, and Zhang (2020) and Therefore, especially online retailers still lack knowledge
Topaloglu and Dass (2021) indicating that higher levels of re- about which products and reviews are more prone to bene-
view helpfulness increase consumers’ purchase likelihood and fit from the addition of review images (Wu, Wu, and Wang
thus drive online sales, the majority of research on review im- 2021). This is especially crucial information given that a re-
ages has focused on review helpfulness as the main dependent cent industry study shows that 29% of the respondents looked
variable. In line with Siering, Muntermann, and Rajagopalan for reviews with user-generated images before purchasing and
(2018, p. 2), we define review helpfulness as a “measure of 66% of them reported that they are more likely to buy the
perceived value in the decision-making process [which] re- product if these photos are presented on the product page
sembles the diagnosticity of the online review related to the (Bazaarvoice 2021).
reduction of uncertainty.” In line with signaling theory – which posits that qual-
Using Amazon product reviews for DVD products, ity signaling is a solution for information asymmetry
Lee, Lee, and Baek (2021) find the inclusion of a review im- (Akerlof, 1970) – we expect the role of review images (as
age to positively affect review helpfulness. Subsequent studies information signals) will be more or less pronounced de-
have also examined the effects of the number of images ac- pending on the initial level of information asymmetry asso-
companying a review on its helpfulness (e.g., Cheng and Ho ciated with a product purchase. First, whether review images
2015; Srivastava and Kalro 2019) and find review helpfulness will help with mitigating information asymmetry will depend
to increase as the number of review images increases. on review-related factors (signals from the review content)
However, research provides little insights on when re- and reviewer-related factors (e.g., the reviewer’s reputation)
view images will be most helpful, i.e., which review-related (Siering, Muntermann, and Rajagopalan 2018). Second, we
and reviewer-related factors foster (or hinder) the effective- acknowledge that different products are associated with dif-
ness of review images in mitigating information asymmetry ferent levels of information asymmetry given their charac-
(Bigne, Chatzipanagiotou, and Ruiz 2020; Luo et al. 2022). teristics, such as whether their attributes can be easily eval-
Filieri, Raguseo, and Vitari (2018) reveal that extreme reviews uated before purchase or not (search vs. experience prod-
on TripAdvisor.com are voted as more helpful when they are ucts), whether the brand can signal credibility to consumers
associated with photos of the hotel provided by the reviewer. via brand equity (brand strength), and whether the product is
On the contrary, Wu, Wu, and Wang (2021) find that the ef- rather bought for their specific functionality or for the purpose
fect of moderate reviews for a camera on review helpfulness of pleasure and affective experiences (utilitarian vs. hedonic
is strengthened for reviews with (vs. without) review images, products). Therefore, we explore the moderating effects of es-
whereas this effect is diminished for extreme reviews. Hence, tablished review-related (i.e., valence and review length) and
given these limited and inconclusive findings, practitioners reviewer-related factors (reviewer reputation) as well as dif-
need further guidance on conditions for which the effective- ferent characteristics inherent in the reviewed product (i.e.,
ness of review images will be more pronounced in order to product type, product category, and brand strength) on the re-
put appropriate incentive mechanisms in place. lationship between review images and review helpfulness. For
Moreover, findings regarding which products or services an overview, please see our conceptual framework as depicted
will benefit most from adding review images is still lim- in Fig. 1.
ited and inconclusive. Findings by Xu, Chen, and San-
thanam (2015) indicate that both experience and search prod- Contingency conditions of images in reviews
ucts benefit from the addition of a review image in terms
of review helpfulness with search products benefitting more, Images added to a review provide rich information which
whereas reviews for products on the middle of the search- might help reduce information asymmetry beyond a review’s
experience continuum are perceived as less helpful when a textual content (e.g., Wu, Wu, and Wang 2021). Prior research
review image is added. On the other hand, Yang, Wang, and shows that when an image is added to a review, consumers’
attention to the review shifts to the image at the expense
of the textual content (Bigne, Chatzipanagiotou, and Ruiz
1 In the following, we will refer to user-generated review images as review 2020). This is because visual (vs. textual) content is supe-
images if not indicated differently. rior in capturing consumers’ attention, regardless of the size

3
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Fig. 1. Conceptual model.

of the image, and images can be processed more easily than Moderating role of review valence
textual content, which supports the important role of images The valence of a review in terms of its overall rating pro-
for information processing (Pieters and Wedel 2004), and for vides quick guidance and information to consumers when
forming customer experiences as well as purchase decisions purchasing online. With a review rating, reviewers indicate
(Bleier, Harmeling, and Palmatier 2019). their product experience with a single indicator. For exam-
In the context of online reviews, review images can substi- ple, Amazon uses a 5-star rating (where 5 stars signify the
tute for the lack of haptic information and provide more cer- best). This single indicator summarizes the overall valence
tainty when evaluating the performance of a product before of a review (Wu, Van der Heijden, and Korfiatis 2011).
purchase (Peck and Childer 2003). The vividness of the in- In terms of the distribution of reviews, Hu, Zhang, and
formation should further help in retrieving past performances Pavlou (2009) show that the majority of ratings lie on the two
by similar products which – in turn – also assists in form- extremes of the rating continuum, with Schoenmueller, Net-
ing more accurate expectations about a product’s performance zer, and Stahl (2020) finding that 68% of Amazon reviews
(Weathers, Sharma, and Wood 2007). Moreover, review im- can be considered extreme. Extreme (vs. moderate) reviews
ages have the ability to better transfer emotions of the review provide clear and unambiguous information signals help-
content, which is likely to further increase the informational ing to confirm or eliminate product alternatives and hence
value of a review (Srivastava and Kalro 2019). Prior research reduce information asymmetry (Siering, Muntermann, Ra-
also finds that when a review includes an image, the infor- jagopalan 2018). This notion is also supported by research
mation quality increases, which in turn reduces information showing that extreme reviews are perceived as more helpful
asymmetry and subsequently leads to an increase in overall than moderate reviews (e.g., Choi and Leon 2020; Wu, Wu,
helpfulness of a review (Zinko et al. 2021). Hence, in line and Wang 2021).
with Siering, Muntermann, and Rajagopalan (2018), we pur- Review images are highly objective (Zinko et al. 2020)
pose2 : and hence should be able to increase the overall informa-
H1 : The helpfulness of a review increases with the number tional value of the review by either compensating for in-
of review images. sufficient information of the review or by adding support
for the statements made in the review (Wu, Wu, and Wang
2021). Hence, we expect that the effect of review images
Moderating influence of review-related and reviewer-related will be more pronounced for moderate (vs. extreme) re-
factors on the effect of review images on review helpfulness views as these types of reviews are characterized by a high
amount of ambiguous information which – in turn – leads
With our model, we focus on three main factors identi- to higher information asymmetry. In extreme reviews, on
fied by previous research: review valence, review length, and the other hand, clear and unambiguous statements are al-
reviewer reputation (see e.g., Choi and Leon 2020). ready made which is why review images might not add
much diagnostic value. Also, given consumers’ preference
2 We provide additional evidence for this hypothesis by also testing the for extreme reviews (e.g., Wu, Wu, and Wang, 2021), they
impact of the presence (vs. absence) of a review image (e.g., Lee, Lee, and should consider additional diagnostic cues (such as a re-
Baek 2021) on the review helpfulness. We find similar results. view image) to be less important to evaluate review help-

4
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

fulness due to the confirmation bias (Choi 2022). Thus, we on verbal signals due to higher uncertainty associated with
propose: the visual information (Smith, 1991). This notion is also sup-
H2a: The positive effect of review images on review help- ported by Kim and Lennon (2008) who show that the effect of
fulness will be weakened for extreme (positive and negative) visual information on consumers’ cognitive attitudes and pur-
reviews. chase intentions is only significant when the amount of verbal
Even though consumers prefer extreme over moderate re- information is low and hence initial information asymmetry is
views (irrespective of their valence), prior research also re- high. In the context of online reviews, Zinko et al. (2020) find
veals the existence of a context-dependent negativity bias that trust and purchase intention increase with the addition of
(Wu 2013). More specifically, consumers rate negative (vs. review images for all reviews independent of their information
positive) reviews as more helpful when the overall valence quality (i.e., review length). However, reviews with low tex-
of consumers’ opinions is positive. On the other hand, if the tual information (i.e., short reviews) benefited the most from
overall valence is negative, consumers rate positive reviews the addition of review images. Thus, we suggest that the ef-
as more helpful. This supports the notion that an extreme fect of review images will be more pronounced for shorter
negative review has higher diagnostic value for a consumer reviews which contain only a limited amount of information
and hence helps with reducing information asymmetry if the and hence review images can help with decreasing informa-
majority of other reviews is positive. Schoenmueller, Netzer, tion asymmetry by compensating for the lack of sufficient
and Stahl (2020) show that almost all platforms allowing con- information. On the other hand, review images might have
sumers to publish online reviews exhibit an imbalance towards only marginal or even no diagnostic value for long reviews
positive reviews and that this is even more pronounced for where information asymmetry is already relatively low. On
commercial platforms such as Amazon. the contrary, the combination of extensive verbal and visual
Given this negativity bias, consumers likely perceive an information might even lead to information overload (Hu and
extreme negative review as more credible and are more ac- Krishen 2019). Therefore, we hypothesize:
cepting of the review information which – in turn – will make H3: The positive effect of review images on review help-
diagnostic cues, such as review images, less impactful given fulness is weakened by review length.
the lower information asymmetry. On the other hand, in case
of overly positive reviews, information asymmetry is expected The moderating role of reviewer reputation
to be larger as readers may question the reliability and credi- Some consumers may question the expertise of a reviewer
bility of such reviews. For them, information asymmetry may or give less weight to other people’s experiences as expressed
thus be best reduced by showing visual evidence. Images may in reviews which is associated with higher information asym-
engender confidence and increase the credibility of an ex- metry. Previous research has shown that source credibility
tremely positive review, as they show the product in use in serves as a crucial signal in consumers’ evaluation of review
a real setting. Review images will thus increase trust for ex- helpfulness (e.g., Ravula, Jha, and Biswas 2022) and even-
tremely positive reviews and, in turn, enhance the review’s tually their decision whether to adopt the information pre-
helpfulness. sented in the review (Racherla and Friske 2012) because of
Thus, we expect the moderating effect of review valence its ability to reduce information asymmetry. To judge a re-
to be stronger for extreme positive reviews than for extreme viewer’s assessment and reduce potential uncertainties about
negative reviews. In sum, we propose: his/her credibility, consumers also draw upon social factors
H2b: The moderating effect of extreme reviews is stronger to determine the trustworthiness of the reviewer and subse-
for extreme positive (vs. extreme negative) reviews. quently the helpfulness of a review (Cheng and Ho 2015).
More specifically, a reviewer’s reputation within the online
Moderating role of review length review community as determined by external recognition, i.e.,
Previous research suggests a positive relationship between the amount of reviewer badges received, has also been shown
review length and helpfulness (Forman, Ghose, and Wiesen- to create reviewer credibility and hence positively impact re-
feld 2008; Karimi and Wang 2017; Mudambi and Schuff view helpfulness (Racherla and Friske 2012; for results from
2010). This positive relationship has also been confirmed by a meta-analysis, please see: Hong et al. 2017).
a recent meta-analysis (Hong et al. 2017). Longer reviews However, in the presence of two information signals, i.e.,
include more details, enhance the review content, and hence reviewer badges and review images, how do these types of
help in reducing information asymmetry. information relate to one another? When a less reputable re-
However, research has not provided conclusive evidence viewer (without a contribution badge) publishes a review, this
for the relative importance of visual (here: review images) review will be associated with a higher information asymme-
and verbal (here: review length) information in online reviews try because of the lack of credibility of the reviewer. In this
for the evaluation of review helpfulness. Contrary to popular case, review images may help consumers to assess the subjec-
believe, research has shown that visual information is only tivity or objectivity of the reviewer and the review itself. Such
dominant in cases where the visual information conveys dif- visual cues may thus aid consumers in better interpreting the
ferent messages from the verbal information. When the same specific content of a review (Lenzner, Schnotz, and Müller
message is conveyed via visual and verbal information, the in- 2013) and the reviewer’s expertise (Karimi and Wang 2017)
ferences based on visual signals are weaker than those based in favor of review helpfulness. However, when a reviewer’s

5
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

reputation is relatively high, the information asymmetry asso- watches, fine chocolate) are “products whose consumption is
ciated with the review should be lower which is why review primarily characterized by an affective and sensory experience
images will have lesser diagnostic value for the consumer. of aesthetic or sensual pleasure, fantasy, and fun” (Dhar and
Thus, we expect that the role of review images will decrease Wertenbroch 2000, p. 61). Hence, for the latter category,
with an increase in reviewer reputation. evaluations of products are personal and involve emotions
H4: The positive effect of review images on review help- (Moore 2015). Also, review content for hedonic (vs. utilitar-
fulness will be weakened by reviewer reputation. ian) products is less factual and less objective but rather for-
mulated as personal consumption experiences (Moore (2015).
Moderating influence of product-related characteristics on These differences in evaluation criteria as well as differences
the effect of review images on review helpfulness in review content lead to decreased efficacy in the evaluation
of hedonic products pre-purchase and – in turn – a higher in-
Information asymmetry in online shopping varies across formation asymmetry for these types of products. As a con-
different product categories (Hsieh, Chiu, and Chiang 2005). sequence, reviews for hedonic (vs. utilitarian) products are
Accordingly, the importance of reviews and review images perceived as relatively less helpful in line with a negative
may also vary. Moreover, prior research indicates that the category effect (Pan and Zhang 2011).
role of reviews also differs for strong and weak brands (Ho- Given that review images are superior in conveying af-
Dac, Carson, and Moore 2013), which will likely also extend fective experiences in a vivid manner (Kim and Lennon
to review images. Therefore, we also explore how the role 2008) and help imagine using the product in the future
of review images differs for different products characteristics (Roggeveen et al. 2015), they might be especially suitable
(i.e., search vs. experience products, hedonic vs. utilitarian to complement the information in the written review for he-
products, and strong vs. weak brands). donic products and hence help reducing information asymme-
try. Moreover, review images might be able to help consumers
The moderating role of product type (search vs. experience) evaluate to which degree the presented personal experience is
Previous research indicates that information asymmetry in line with their own expectations (Hong and Pavlou 2014).
largely varies across two main product categories: search In addition, as images can substitute for seeing products in
products and experience products. According to Park and real life, review images are likely to be most helpful for prod-
Park (2013), especially online, pre-purchase uncertainty is ucts that benefit from sensory experiences (Lin, Lu, and Wu
greater for experience products than for search products which 2012). Thus, we argue that the review images strengthen the
contributes to a higher level of information asymmetry for effect of product reviews for hedonic (vs. experience) prod-
experience products. After all, experience products need to ucts on helpfulness. Thus:
be seen, smelled, tasted, tried, or touched to assess their H6: Review images have a stronger positive effect on re-
quality (Peterson, Balasubramanian, and Bronnenberg 1997) view helpfulness for hedonic than utilitarian products.
whereas search products can be evaluated more easily based
on the online information provided by the seller. In support
The moderating role of brand strength
of this, multiple studies have shown that product reviews
Conceptually, strong brands are quality indicators that help
are more helpful for experience than search products (e.g.,
reduce information asymmetry (Hazée, Van Vaerenbergh, and
Huang, Lurie, and Mitra 2009).
Armirotto 2017) and build trust (Benedicktus et al. 2010). In
The different degrees of information asymmetry associated
line with this, Ho-Dac, Carson, and Moore (2013) find that
with search (low) vs. experience (high) products (Hsieh, Chiu,
online reviews matter less for strong brands than for weak
and Chiang 2005) also determine the importance of visual
brands. Thus, a review’s contribution to reducing informa-
information, such as review images. Images help to assess
tion asymmetry is lesser for stronger brands than for weaker
experience attributes of products which are more difficult to
brands. As such, we postulate that weaker brands need to rely
evaluate via textual information which in turn reduce informa-
more on reviews than stronger brands to reduce seller uncer-
tion asymmetry for experience products (Xu, Chen, and San-
tainty given that for these types of brands the initial informa-
thanam 2015). For search products, sensory information (such
tion asymmetry is higher (Manes and Tchetchik 2018), and
as in the form of review images) is less important given that
therefore the impact of images in a review will be stronger.
these types of products can easily be evaluated based on ver-
Thus:
bal information (Weathers, Sharma, and Wood 2007). Hence,
H7: Review images have a stronger positive effect on re-
we propose that:
view helpfulness for weak than strong brands.
H5: Review images have a stronger positive effect on re-
view helpfulness for experience than search products.
Data
The moderating role of product category (utilitarian vs.
hedonic) We test our conceptual model with the help of a unique
Utilitarian products are bought for their functionality data set of online reviews from Amazon. Amazon incorpo-
following predominately cognitive information processing rates more than 250 million reviews from over 213 million
(Pan and Zhang 2011). Hedonic products (e.g., luxury customers for 353 million products (LandingCube 2022), thus

6
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Table 1 of review observations for both strong and weak brands. Web
Choice of product types and categories. Appendix B provides an overview of the final product types
Category Search Experience Hedonic Utilitarian and categories with the brand categorizations, including the
DVD Player X X sample composition.
Digital Photo Frame X X
Smartphone X X Data collection and variable operationalization
Computer Adapter X X
Printer X X
Electric Toothbrush X X
For each product, we obtain all reviews posted on Ama-
Mascara X X zon before April 2019, which results in 97,997 reviews in
Lipstick X X total. For each review, we collect the review text, star rat-
Party Dress X X ing, reviewer badges, number of helpful votes, and all im-
Microwave Oven X X ages included in the review. We also collect the review date
Vacuum Cleaner X X
Razor X X
and include the age of the review into our model to con-
trol for Amazon’s ranking algorithm which displays reviews
that have received more helpful votes more prominently than
providing enough representative reviews and review helpful- other reviews. Hence, newer reviews stay farther down the
ness votes for a variety of products and brands. page and have less chance of receiving votes (Salehan and
We obtain all reviews in our sample with the help of a Kim 2016). Further, a visual inspection of the variable distri-
self-coded R-based web crawler that extracts relevant review butions as well as the corresponding box plots indicates that
content and images via the rvest package (Wickham 2016) both review helpfulness and the number of images face outlier
as described in Yildirim and Kübler (2023, pp. 147-156). To problems. We separate outliers from the data set by relying
identify suitable reviews to test the different contingencies of on the 1.5 interquartile rule of Backhaus et al. (2021) and
our conceptual model, we conduct multiple pretests. drop 50 reviews for which either the number of images or
the number of helpful votes exceeds the rule, resulting in a
Choice of products final data set consisting of 97,947 reviews including 6,060 im-
ages. Table 2 provides an overview of the variables derived
To be able to test the product-related contingencies of from the collected data, their operationalization and descrip-
our conceptual framework, we screen previous marketing re- tive statistics, including correlations across all numerical vari-
search studies for products classified as search vs. experi- ables. The distributions of all variables are available in Web
ence or hedonic vs. utilitarian products. We then conduct Appendix C.
a pretest and ask a sample of 20 undergraduate students to
rate the identified 30 products as hedonic vs. utilitarian and Analysis
search vs. experience on scales provided by Wakefield and
Inman (2003) and Weathers, Sharma, and Wood (2007), re- Our analytical approach is guided by the specific needs of
spectively. Afterwards, we choose the three top classified our data set. In line with previous research on review help-
products per type and category (see Table 1). Validating fulness (e.g., Mudambi and Schuff 2010), we face two major
our choice, previous research has also used these product challenges: (1) data censorship and (2) selection bias.
types and categories to represent search vs. experience (e.g., Censorship origins from the particular format of review
Gao et al. 2012; Huang, Lurie, and Mitra 2009) and hedo- helpfulness. The variable is bound in its range as it cannot fall
nic vs. utilitarian products (e.g., Pallas, Mittal, and Groening below the limit of zero. Although readers might find a review
2014). not helpful at all, they are not able to downvote a review. In
addition, we cannot observe if a review’s lack of helpfulness
Choice of brands and brand strength measurement votes (i.e., a zero score) is due to the characteristics of the
review or other (often unobservable) factors such as time,
To select appropriate brands for our analysis, we use an reader characteristics, or category competition.
exploratory approach. In the first step, we screen Amazon Furthermore, our data may be prone to a selection bias
for brands that offer the selected products. Subsequently, to that origins from the fact that not every reader actively rates
classify the brands as strong vs. weak, we rely on the Best a review as helpful. A selection bias might also arise because
Global Brands report 2018 (the year before data crawling) of the large number of reviews on the platform and because
from Interbrand (2018) which ranks brands by their strength. readers can hardly read all the reviews of a product. As such,
To represent a strong brand in our sample, the brand needs we cannot determine whether a reader not voting a review as
to be in the Top 100 of the report. For all remaining and helpful really finds the review not helpful or whether he or
non-listed brands, we conduct a pretest with 30 undergraduate she has not read it.
students to rate brand strength. Based on the results, we select We address these two modeling challenges by following
the brands rated as the least strong for each product. Since the established research in the field of review helpfulness re-
weak brands have fewer reviews than the strong brands, we search (Srivastava and Kalro 2019) and apply a Tobit regres-
choose multiple weak brands to obtain an equivalent number sion model to test our conceptual model.

7
R.V. Kübler, L. Lobschat, L. Welke et al.
JID: RETAIL
Table 2
Variable operationalization and descriptive statistics.

Variable Type Definition Coding Descriptive statistics Correlations

Min Max Mean SD RL RR RA NI


Review DV Number of times a review has been Numerical 0 480.00 1.74 13.30 0.38∗∗ 0.09∗∗ 0.14∗∗ 0.17∗∗
Helpfulness evaluated as being helpful
(RH) (Mudambi and Schuff 2010)
Review and Reviewer Characteristics
Extreme IV Very high (5/5-star) product rating Categorical (0 = no extreme 0 1 0.57
Positive on Amazon’s rating scale review positivity;
Reviews (Mudambi and Schuff 2010) 1 = extreme review
positivity)

ARTICLE IN PRESS
Extreme IV Very low (1/5-star) product rating Categorical (0 = no extreme 0 1 0.15
Negative on Amazon’s rating scale review negativity;
Reviews (Mudambi and Schuff 2010) 1 = extreme review
negativity)
Review IV Number of words per review Numerical 0 2843.00 48.69 72.24 0.12∗∗ 0.08∗∗ 0.09∗∗
Length (RL) (Singh et al. 2017)
Reviewer IV Number of badges a reviewer has Numerical 0 3.00 0.05 0.23 -0.10∗∗ 0.00
Reputation received (Liu and Park 2015;
8

(RR) Ngo-Ye and Sinha 2014)


Product Characteristics 0 1 0.47
Search vs. IV Search versus experience product Categorical
Experience (0 = experience;
Product 1 = search)
Hedonic vs. IV Hedonic versus utilitarian product Categorical (0 = hedonic; 0 1 0.56
Utilitarian 1 = utilitarian)
Product
Brand IV Brand strength Categorical 0 1 0.47
Strength (0 = weak; 1 = strong)
Review
Images
Number of IV Number of images included in a Numerical 0 9.00 0.06 0.37
Images (NI) review
Controls 0 4050.00 646.60 511.09 -0.04∗∗

Journal of Retailing xxx (xxxx) xxx


Review Age CV Difference in days between the date Numerical

[m5+;September 25, 2023;11:20]


(RA) of posting and the date of scraping
the review (Hong et al. 2017;
Liu and Park 2015)
∗p < .01.
∗∗ p < .0001. Note: DV = Dependent Variable; IV = Independent Variable; CV = Control Variable.
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Following the standard notation of Tobit models The impact of images on review helpfulness
(Greene 2019, pp. 933) we define our censored dependent
variable ReviewH el p f ul nessi∗ as Both of our models find a positive and significant estimate
for the number of review images (ß1 = 11.86, p < 0.001).
ReviewH el p f ul nessi∗ = max (ReviewH el p f ul nessi , ci ) (1)
This suggests that – in line with previous research (e.g., Sri-
where ReviewH el p f ul nessi expresses the observed number vastava and Kalro 2019) – adding one or more images to
of helpfulness votes for review i and ci represents the corre- an average review generally leads to higher levels of review
sponding censoring threshold in review i (which is 0). helpfulness. Thus, we can confirm H1.
We then model the latent variable ReviewH el p f ul nessi∗ as
a linear function of the independent variables
 
ReviewH el p f ul nessi∗ = β∗xi + εi with εi ∼ N 0, σ 2 (2) The moderating effect of review valence
where β is a vector with the regression coefficients of the
In case of extremely negative reviews, our model results
independent variables for observation i assembled in vector
in a significant and negative estimate for the interaction term
xi , and εi is a normally distributed error term with mean 0
with review images (ß11 = -1.92, p < 0.05). Consistent with
and variance σ 2 .
H2a, this indicates that – in contrast to moderate reviews –
Finally, we assume that the observed value of
more negative reviews benefit less from review images.
ReviewH el p f ul nessi is equal to ReviewH el p f ul nessi∗ if
In case of extremely positive reviews, we find a positive
ReviewH el p f ul nessi∗ is larger than or equal to ci (0), and
and highly significant interaction (ß10 = 4.47, p < 0.001), in-
equal to ci (0) if ReviewH el p f ul nessi∗ is smaller than ci (0),
dicating that overly positive reviews benefit more from review
as shown in Eq. (3).
images than moderate reviews. We can thus only partially
ReviewH el p f ul nessi confirm H2a.
 Our results also provide support for H2b. More specifi-
ReviewH el p f ul nessi∗ i f ReviewH el p f ul nessi∗ ≥ 0
= (3) cally, the coefficient of the interaction between the number
0 i f ReviewH el p f ul nessi∗ < 0
of review images and extreme positive reviews (ß10 = 4.47,
We can combine these equations and then finally obtain p < 0.001) is larger than the corresponding coefficient for
our model as depicted in Eq. (4). To enhance the inter- extreme negative reviews (ß11 = -1.92, p < 0.05). Hence, in
pretability of our contingency effects, all numerical indepen- line with our expectations, the moderating effect of review ex-
dent variables are mean centered following the approach of tremity is stronger for extreme positive than extreme negative
Cohen et al. (2003). reviews.
ReviewH el p f ul nessi∗
= β0 + β1 ∗ N umI magesi
The moderating effect of review- and reviewer-related
+ β2 ∗ E xt Posit ivei + β3 ∗ E xt N egat ivei characteristics
+ β4 ∗ RevLengthi + β5 ∗ Reput at ioni
+β6 ∗ E x perienceProdi + β7 ∗ H edonicProdi We obtain robust empirical validation for H3. The asso-
+ β8 ∗ Br andSt rengt hi ciated interaction coefficient concerning the relationship be-
tween the number of review images and the length of the re-
+ β9 ∗ RevAgei + β10 ∗ N umI magesi ∗ E xt Posit ivei view is both significant and negative (ß12 = -.01, p < 0.001).
+ β11 ∗ N umI magesi ∗ E xt N egat ivei This is line with our initial assumption that an increasing re-
+β12 ∗ N umI magesi ∗ RevLengthi view length dampens the positive effect of review images on
+ β13 ∗ N umI magesi ∗ Reput at ioni helpfulness.
In case of H4, where we expected that high reviewer rep-
+ β14 ∗ N umI magesi ∗ E x perienceProdi utation would dampen the effect of review images on review
+ β15 ∗ N umI magesi ∗ H edonicProdi helpfulness, we find a strong and significant interaction. This
+ β16 ∗ N umI magesi ∗ Br andSt rengt hi effect, however, is positive (ß13 = 7.28, p < 0.001) and thus
  opposite to the effect suggested by our conceptual model so
+ εi with εi ∼ N 0, σ 2 (4)
that we have to reject H4. Hence, contrary to our expectations,
a reviewer’s reputation strengthens the effect from review im-
Results ages on review helpfulness. Thus, it seems that even though
review images are associated with a decrease in information
Table 3 shows the results of our study. To demonstrate asymmetry due to their ability to complement review infor-
robustness, we present the results of two model variations. mation and/or add additional diagnostic value, they become
The main model includes only main effects, while the model even more helpful when coming from a reputable reviewer
with interactions presents the results of the model depicted in who also increases the credibility of the images (Pan and
Eq. (4). Zhang 2011).

9
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Table 3
Results - contingency conditions of images in reviews.

Main Model Model with Interactions


Intercept -24.775 (.295)∗∗∗ -24.972 (.297)∗∗∗
Effect of Images
Number of Images 11.480 (.237)∗∗∗ 11.863 (.511)∗∗∗
Contingencies
Extreme Positive Reviews × Number of Images 4.470 (.528)∗∗∗
Extreme Negative Reviews × Number of Images -1.916 (.770)∗
Review Length × Number of Images -.011 (.001)∗∗∗
Reviewer Reputation × Number of Images 7.276 (.608)∗∗∗
Experience Product × Number of Images 1.078 (.532)∗
Hedonic Product × Number of Images 3.034 (.488)∗∗∗
Brand Strength × Number of Images -.989 (.499)∗
Main Effects
Extreme Positive Reviews -2.577 (.251)∗∗∗ -2.765 (.252)∗∗∗
Extreme Negative Reviews 11.531 (.316)∗∗∗ 11.575 (.316)∗∗∗
Review Length .103 (.001)∗∗∗ .104 (.001)∗∗∗
Reviewer Experience 1.280 (.432)∗∗ .638 (.435)
Experience Product -.628 (.217)∗∗ -.766 (.217)∗∗∗
Hedonic Product -3.793 (.222)∗∗∗ -4.041 (.223)∗∗∗
Brand Strength 2.520 (.216)∗∗∗ 2.575 (.216)∗∗∗
Review Age .007 (.000)∗∗∗ .007 (.000)∗∗∗
Log() 3.214 (.004)∗∗∗ 3.212 (.004)∗∗∗
Log Likelihood -153760.816 -153588.139
Number of Observations 97,947 97,947
∗∗∗ p < 0.001.
∗∗ p < 0.01.
∗ p < 0.05.

The moderating effect of product characteristics and highlights conditions in which images are of particular
importance to increase the helpfulness of a review.
For H5, we expected review images to have a stronger pos- Still, beyond the “mere presence effect” (Li and Xie 2020)
itive effect on review helpfulness for experience (vs. search) of review images, prior research has also shown that what
products. We find support for this hypothesis as the interaction is depicted in images influences consumers’ evaluation of
between review images and experience (vs. search) products the respective image and their (behavioral) responses (e.g.,
is positive and significant (ß14 = 1.08, p < 0.05). Biehal, Stephens, and Curio 1992; Zhang and Luo 2023).
In addition, our results provide support for H6, as the inter- Given that user-generated images are, by definition, not taken
action between review images and the product being hedonic by professionals, the chosen review images might differ on
(vs. utilitarian) is significant and positive (ß15 = 3.03, p < the degree to which they present information and content rel-
0.001). Thus, in line with our arguments, although reviews for evant to the reviewed product and hence the review itself.
hedonic (vs. utilitarian) products are in general perceived as Consequently, it is essential to better understand how image
less helpful (ß7 = -4.04, p < 0.001), they benefit more from content affects review helpfulness.
the inclusion of images compared to utilitarian products. For this purpose, we conduct a follow-up study based on
Further, in H7 we address the potential contingency of the the subset of reviews from our data set that contain images.
effect of review images on brand strength. The model reveals The filtered sample contains 6,060 images from 3,784 re-
that the interaction between brand strength and review images views. Next to the variables already included in our main
is negative and significant (ß17 = -.99, p < 0.05), in support study, we additionally create two different variables capturing
of H7. This result shows that especially weaker brands benefit image content, i.e., image focus and context fit.
from the addition of images to reviews and that review images First, image focus indicates whether the image shows the
help mitigate the, on average, fewer review helpfulness votes product itself (e.g., pack shot) or its application. When creat-
reviews for weak brands receive, as indicated by the main ing a review, consumers can decide whether to add an image
effect from brand strength on review helpfulness (ß8 = 2.58, showing only the product or the product in use. We assume
p < 0.001). – given that most online stores already provide consumers
with detailed product images – that those images provide lit-
Follow-up study: the impact of image content on review tle new information compared to images showing the product
helpfulness in application and thus do not help (much) to overcome in-
formation asymmetry. Images with an application focus, by
Up to this point, our study underlines that review helpful- contrast, can also help consumers objectively assess the writ-
ness increases with the number of images added to a review ten content of a review. We construct our image focus vari-

10
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Fig. 2. Examples of images content variables.

able based on the image clustering approach introduced by images have an application focus and the mean context fit
Klostermann et al. (2018) which groups images based on al- measured on a scale from 0 to 1 is 0.51. Web Appendix
gorithmically detected objects which are depicted and their D provides more information about the operationalization of
probabilities. both image content variables.
Second, context fit captures how well the objects displayed In addition, we take into account that some reviews contain
in an image fit the reviewed product. Recent research by multiple images. Therefore, we also control for the confound-
Ceylan, Diehl, and Proserpio (2023) highlights the impor- ing effects arising from other images included in the review,
tance of similarity between textual information in a review coded as the number of all other images contained in the re-
and the associated review image to increase consumers’ ease view next to the observed image. Again, we also control for
of information processing and – in turn – also enhance review review age and estimate a Tobit regression model with mean
helpfulness. In a similar vein, we argue that the higher the fit centered variables.
of the objects depicted in a review image with the reviewed Table 4 provides the results of the image content study. We
product (i.e., higher context fit), the lower the information again report the main effects model together with the model
asymmetry, and the easier it is for consumers to process the including all interaction effects among the content measures
review and hence perceive it as more helpful. We calculate and the contingency factors to demonstrate the robustness of
the context fit based on the average Euclidean distance of our findings. In the following, we discuss the findings of the
the top 20 classified objects in each image to the reviewed model including interactions.
product with the help of a pre-trained Global Vectors (GloVe) First, similar to our study investigating the impact of the
model containing word embeddings (Pennington, Socher, and number of images in a review, we find that adding additional
Manning 2014). Word embeddings are numerical representa- images (ß = 15.75, p < 0.001) increases review helpfulness.
tions of words in a vector space and thus allow to compute
the distance between words. Similar words will have smaller
distances between their embeddings, i.e., higher context fit, Effects from image focus and context fit
while more different words will have larger distances, i.e.,
lower context fit. Fig. 2 shows examples of images with a In line with our assumption, we find a significant and pos-
product (coded as 0) vs. application (coded as 1) focus and itive effect (ß = 23.06, p < 0.01) of image focus on review
with a low vs. high context fit. In our subset, 59% of the helpfulness, indicating that images with an application focus

11
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Table 4
Results – impact of image content on review helpfulness.

Main Model Model with Interactions


Intercept -51.758 (4.853)∗∗∗ -60.005 (6.414)∗∗∗
Effect of Image Content
Image Focus 7.847 (3.862)∗ 23.057 (8.356)∗∗
Context Fit 9.682 (10.535) -12.103 (22.375)
Contingencies
Extreme Positive Reviews × Context Fit -7.284 (22.482)
Extreme Negative Reviews × Context Fit -94.093 (32.666)∗∗
Review Length × Context Fit -.227 (.053)∗∗∗
Experience Product × Context Fit -17.291 (25.284)
Hedonic Product × Context Fit -43.041 (22.153)
Brand Strength × Context Fit 53.809 (20.738)∗∗
Extreme Positive Reviews × Image Focus -11.175 (8.725)
Extreme Negative Reviews × Image Focus -13.333 (12.175)
Review Length × Image Focus .098 (.021)∗∗∗
Experience Product × Image Focus -3.825 (9.237)
Hedonic Product × Image Focus 11.912 (8.411)
Brand Strength × Image Focus -7.583 (7.880)
Other Images × Image Focus -5.152 (1.800)∗∗
Main Effects
Extreme Positive Reviews 19.167 (4.233)∗∗∗ 25.231 (6.618)∗∗∗
Extreme Negative Reviews 52.304 (5.942)∗∗∗ 53.680 (8.092)∗∗∗
Review Length .278 (.011)∗∗∗ .228 (.015)∗∗∗
Reviewer Experience 8.569 (5.787) 9.494 (5.772)
Brand Strength 5.422 (3.864) 8.473 (5.794)
Experience Product 5.960 (4.497) 8.964 (6.326)
Hedonic Product -12.901 (4.153)∗∗ -23.240 (6.036)∗∗∗
Review Age .160 (.005)∗∗∗ .159 (.005)∗∗∗
Other Images 14.266 (.852)∗∗∗ 15.754 (1.043)∗∗∗
Log() 4.830 (.011)∗∗∗ 4.823 (.011)∗∗∗
Log Likelihood -25187.931 -25155.806
Number of Observations 6,060 6,060
∗∗∗ p < 0.001.
∗∗ p < 0.01.
∗ p < 0.05.

receive more helpfulness votes than images which just show We furthermore find a significant and negative (ß = -
the product (e.g., pack shot). .23, p < 0.001) interaction between context fit and review
The results further indicate that context fit does not show a length, indicating that shorter reviews should include im-
significant main effect, indicating that how well an image fits ages which are more context-specific and feature (many) ob-
with the product-related context expectations and associations jects that are tied to the reviewed product. This is in line
of a consumer does, on average, not improve the helpfulness with our earlier argumentation that shorter reviews contain
of a product review. Still, we find significant interactions with only a limited amount of information and therefore do not
review(er) and product characteristics, indicating that context add much to reduce information asymmetry. By contrast,
fit matters in specific conditions. adding context-fitting images strengthens the credibility and
reliability of the review and therefore reduces information
asymmetry.
Content-contingencies for context fit For brand strength, we find a positive and significant
(ß = 53.81, p < 0.01) interaction with context fit which high-
We find three significant interactions between context fit lights that reviews for products of strong brands receive more
and our contingency variables. As shown in Table 4, the in- helpfulness votes in case that the attached images contain
teraction between context fit and extremely negative reviews objects which are closely tied to the product. In this con-
is significant and negative (ß = -94.09, p < 0.01), indicating text, referring back to our argumentation that strong brands
that in case a review’s star rating is very negative, consumers serve as quality signals and thereby reduce information asym-
tend to prefer images low in context fit. This preference might metry, images of products from strong brands embedded in
arise because, in such situations, consumers do not expect a context fitting environment could again underline that the
well-matched images. Instead, they find it more fitting to see product is used in an appropriate and valued manner and
images that align with the negativity of the review and thus enhance the trust in the review(er), decreasing information
reduce information asymmetry. asymmetry.

12
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Content-contingencies for image focus Furthermore, our investigation demonstrates that reviews
with 10 images with an average sentiment (represented by
Our results indicate that consumers perceive multi-image the blue and dashed line in Fig. 3A) yields a prediction of
reviews that contain at least one image with an application 67 helpfulness votes, while only three helpfulness votes are
focus to be less helpful. This is indicated by the interaction predicted in the absence of review images. Thus, it seems
between other images and image focus that is significant and that review images help to decrease information asymmetry
negative (ß = -5.15, p < 0.01), suggesting that too many by enhancing the credibility of extremely positive reviews
images next to an image with an application focus lead to which – without these images – might even be considered fake
information overload that does not reduce information asym- (Meek, Wilk, and Lambert 2021). Hence, urging or providing
metry. incentives for users to attach more images to reviews charac-
Furthermore, our results suggest that longer reviews benefit terized by extremely positive sentiment might be a strategic
more from the addition of images which show the product in implication for platform operators.
application, as we find a significant and positive (ß = 0.10, p Fig. 3B elucidates the discrepancy in the augmentation
< 0.001) interaction between image focus and review length. of helpfulness votes in reviews across distinct levels of re-
It seems that the ability of longer texts to be helpful with tak- viewer reputation, specifically low and high levels. Our find-
ing a buying decision is especially supported when the image ings underscore that the positive impact of review images is
adds more information by highlighting a use case, instead of more pronounced for reviewers with higher reputation levels
only showing the product itself. compared to less reputable reviewers. Once again, the non-
overlapping nature of the confidence intervals along our pre-
Discussion dictive trajectories underscores a significant divergence in the
augmentation of helpfulness votes.
To provide additional insights from our analysis results, we Our simulation outcomes corroborate this observation. For
use the estimates of our two studies to further investigate the instance, for reviewers lacking reputation (characterized by
different contingency scenarios and to predict how different the absence of badges), the predicted helpfulness votes signif-
types of review images drive review helpfulness. icantly increase from three (in the absence of review images)
to 63 (with 10 review images). In contrast, a similar review
authored by a highly reputed reviewer (distinguished by two
Prediction of review helpfulness for selected contingencies badges) experiences a significant surge from three (without
of review images supplementary review images) to 209 predicted helpfulness
votes (with 10 review images).
To highlight how review helpfulness votes differ for differ- Fig. 3C shows the differential effects of images within re-
ent contingency scenarios, we compute helpfulness votes for views pertaining to hedonic (denoted by the blue dotted line)
different numbers of images included in a review3 , ranging and utilitarian (represented by the red line) products. For he-
from 0 to 10 for a selected set of our contingencies. Fig. 3 donic products, our analysis reveals that abstaining from at-
presents the results of our predictions (including the confi- taching images to a review corresponds, on average, to a mod-
dence intervals) for extreme positive (vs. average reviews), est count of two helpfulness votes. However, the same review
high (two badges) vs. low (no badges) reviewer reputation, with 10 review images significantly increases the anticipated
and hedonic vs. utilitarian products. helpfulness votes to 93. Conversely, when examining utilitar-
As illustrated in Fig. 3A, our analysis reveals a notewor- ian products, the absence of images similarly results in only
thy pattern regarding reviews with extremely positive ratings three helpfulness votes. Yet, a review for a utilitarian product
(depicted by the red line). The review images consistently with 10 review images leads only to a projected count of 67
lead to an elevated count of helpfulness votes (compared to helpfulness votes.
a review with an average rating), as evidenced by the distinct Still, while evaluating the confidence intervals for both
non-overlapping confidence intervals. For reviews character- slopes, we find that the predicted outcome of helpfulness
ized by extreme positive ratings, the introduction of 10 images votes differs significantly between the two product categories
corresponds to a prediction of 108 helpfulness votes which is only in cases characterized by five or more review images.
in stark contrast to only two predicted votes in instances with Consequently, platform administrators are advised to ac-
no review images. This observed effect is statistically sig- tively promote the attachment of multiple (ideally numer-
nificant at a level of p < 0.0014 , as established by a slope ous) images in the context of hedonic products. This rec-
analysis following Cohen et al. (2003). ommendation arises from our observation that a substan-
tial increase in helpfulness votes is attained only when a
3 The non-manipulated factors in our simulations are held at their means, remarkably high number of images is appended to a re-
with review length being 49 words, reviewer reputation being 0.05, review view. To incentivize such practices, platforms or reviewer
age being 647 days, and the product being a utilitarian, experience product
communities such as Amazon Vine could consider imple-
form a weak brand.
4 We apply a slope test to all interaction effects reported. In all cases the menting rewards for the inclusion of multiple images or ex-
results indicate that the reported effect is significant at p < .001. We thus plicitly urging reviewers to provide a substantial number of
do not include the p-values in the subsequent reports. images.

13
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Fig. 3. Simulation based interaction plots for number of images.

Prediction of review helpfulness for selected contingencies In addition, to examine the distinct impact of review im-
of review image content ages with a product vs. application focus on review useful-
ness in the context of both short and extensive reviews (as
In order to elucidate the variations in review helpfulness depicted in Fig. 4B), we compute the corresponding estimates
votes across diverse content-specific scenarios of review im- of review helpfulness votes. The cumulative outcomes of our
ages, we calculate helpfulness votes for each of these sce- predictions are presented in Fig. 4.
narios. Our initial step involves the computation of help- Regarding context fit, our analysis highlights that lower
fulness votes for an array of images with different context context fit plays a particularly important role in influencing
fits, spanning from the lowest fit (0) to levels characterized review helpfulness, especially evident in instances of highly
by the mean minus one standard deviation (0.33), the mean negative reviews (depicted in Fig. 4A). In this context, we
(0.51), the mean plus one standard deviation (0.70), and the observe that very low levels of context fitting images may
maximum fit value (1). This analysis investigates the dis- yield up to 54 helpfulness votes, compared to 33 votes when
parities in helpfulness votes for reviews marked by an ex- context fit is really high. This significant rise stands in stark
tremely negative rating versus those with an average rating contrast to reviews with average ratings. For such reviews, we
(Fig. 4A). discern a significant decline in helpfulness votes when context

14
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Fig. 4. Simulation based interaction plots for image content dimensions.

fit shrinks, plummeting from 30 votes (with context fit at 1) promote image uploads solely when the word count of a re-
to 16 votes (with context fit at 0). view is exceedingly low.
Turning to the interplay between the focus of images – be Furthermore, our results indicate that reviews authored by
it product-oriented or application-oriented – and the length of individuals boasting a robust reputation reap greater advan-
reviews (illustrated in Fig. 4B), our findings reveal intriguing tages from the inclusion of images, surpassing the benefits
patterns. In the case of images with a pronounced product experienced by less established reviewers. This observation
focus, the inclusion of images markedly and significantly highlights a strategic avenue where companies could focus
amplifies helpfulness votes, surging from 16 votes for concise their encouragement of image-rich content, catering specifi-
reviews to 54 votes for more extensive reviews. This increase, cally to reviewers with well-established reputations. Hence,
evident through the non-overlapping confidence intervals, platforms are advised to concentrate efforts on incentivizing
diverges notably from the more pronounced augmentation reviewers endowed with elevated reputations to append visual
in helpfulness that occurs with the elongation of reviews content to their reviews. Accomplishing this objective could
when images carry an application-oriented focus. In this sce- involve offering emblematic incentives, such as unique badges
nario, succinct reviews obtain approximately 19 helpfulness for content enriched with images or for prolific reviewers who
votes, while considerably lengthy reviews, accompanied by consistently incorporate visual elements. Alternatively, mon-
application-focused images, amass a substantial 88 votes. etary rewards could be extended to highly reputed reviewers
for their contributions of substantial and high-quality visual
content. Similarly, platforms might contemplate the imple-
Implications for marketing practice mentation of a hierarchical structure of visible recognition
symbols (badges). The introduction of supplementary or ex-
Consistent with established research, our findings under- clusive badges, exclusively accessible to recognized review-
score the significance of promoting image uploads alongside ers, contingent upon their sustained incorporation of images
reviews. A clear trend emerges wherein reviews enriched with into reviews, could offer an added impetus for this cohort
a higher number of images tend to amass a greater count to uphold their commendable reviewing endeavors, ultimately
of helpfulness votes. This, over the long term, could poten- catalyzing the amplification of review helpfulness votes.
tially translate into increased sales and reduced product re- Finally, our model presents a compelling argument for on-
turns (Li, Chen, and Zhang 2020). Given the necessity for line retailers to particularly encourage the inclusion of images
companies to employ incentives to stimulate user engagement, in situations characterized by elevated information asymmetry
our framework of contingencies yields crucial insights into the and consumer uncertainty, often stemming from the intrinsic
opportune moments for companies to particularly emphasize attributes of the product. Our model’s outcomes underscore
and incentivize image inclusion within reviews. this point within two specific scenarios: products associated
Our results distinctly emphasize the influential role of re- with weak brand identities and those of hedonic nature.
view sentiment, particularly in the context of remarkably pos- In the case of weak brands, online retailers are well-
itive reviews where the addition of images substantially ele- advised to incentivize users to upload numerous images, po-
vates the levels of helpfulness. This insight positions online tentially through means like special badges or more substan-
retailers in a strategic position to prompt users to incorporate tial monetary rewards. This strategic approach aids unfamiliar
visual evidence when awarding a five-star rating. consumers in gaining a more comprehensive understanding of
Likewise, our findings draw attention to the interplay with products that lack established reputations. Similarly, for he-
review length. Notably, shorter reviews garner more pro- donic products, online retailers should not only promote the
nounced benefits from review images compared to longer upload of additional images, but could also guide reviewers
ones. Consequently, it would be judicious for companies to towards illustrating how they interact with the product and

15
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

what distinguishes it in their eyes. This guidance empow- light on whether this effect is due to higher quality levels and
ers potential buyers to assess whether the review content and experience in image production or due to other factors. More
accompanying images align with their own preferences and importantly, we demonstrate that the presence of images in
habits, which can significantly diverge across consumers of a review moderates the impact of review valence. Thus, we
hedonic products. add to the canon of research examining the impact of re-
Furthermore, these distinctions in consumer preferences view extremity (e.g., Karaman 2021; Mudambi and Schuff
and usage contexts can be addressed by analyzing image con- 2010), contradicting prior findings regarding the helpfulness
tent and furnishing users with tailored recommendations for of extreme reviews, as we show that review helpfulness only
designing and composing images during the upload process. increases when reviewers add images to extremely positive
Our supplementary follow-up study centered on image content (vs. extremely negative) reviews.
unequivocally highlights the heightened helpfulness of images In addition, we investigate the impact of image content
depicting the product in practical application, as opposed to on review helpfulness and explore how this impact is again
mere depictions of the product itself. This finding is largely moderated by review-, reviewer-, and product-related charac-
supported by the common trend and success of “unboxing” teristics. In doing so, we bridge two growing research streams,
content, where consumers document how they use the product combining insights from image content analysis research (e.g.,
for the first time. Online retailers can proactively offer tuto- Li and Xie 2020) with online review research, and pro-
rials demonstrating how users can craft insightful and useful vide guidance to future research on which image content di-
review images that accentuate the significance of product ap- mensions affect review helpfulness. This may again provide
plication. Our results further indicate that such incentives and ground for new research in the domain. Finally, our image
tutorials should be provided to reviewers who tend to pro- classification approach to identify whether images show only
duce longer reviews. Again, dynamic content controls may be the product or the product in an application scenario may
used to educate reviewers in the moment they upload a longer prove useful for future research in related domains and may
review. Given the rapid developments in (live) object detec- even provide guidance for neighboring fields, such as social
tion algorithms, platforms may even mimic our measurement media or user-generated content research.
approach and control images once they are uploaded to then
dynamically address customers who uploaded a pack shot im- Limitations and future research directions
age and to ask them to change the image content and rather
show how they used the product. Our studies have several limitations, which offer possible
We also find that a reduced context fit especially reduces opportunities for future research. First, while our conceptual
the number of helpfulness votes in case of short reviews. model identifies multiple cues causing information asymme-
We suspect that in these situations consumers expect images try, we do not control for which cues cause more or less.
which reflect the negativity of the review. Hence, they would Future research might thus aim to shed more light on which
not expect images witnessing too much effort and focus on the review-related factors cause more or less information asym-
appropriate context. This implies that online retailers should metry so that online retailers can also try to minimize these
use algorithmic help to screen images (e.g., by mimicking our sources.
context fit based measurement approach) and to alert review- Another limitation is that we rely only on reviews from
ers, if an image does not match the right context fit expecta- a single platform (Amazon) produced and rated by U.S. cus-
tions. tomers. Kübler et al. (2018) note that consumers’ reactions to
online reviews are contingent on cultural background and so-
Implications for marketing theory cioeconomic factors such as income and infrastructure. Thus,
future research might consider replicating our study in a
In this study, we introduce a holistic contingency-based multinational context by using reviews and images from dif-
framework that investigates how review-, reviewer-, and ferent national Amazon websites. Additionally, future research
product-related characteristics influence the impact of review may control if some of our findings are idiosyncratic to Ama-
images on review helpfulness. Our framework aids in fur- zon, by replicating our study with reviews from other online
ther strengthening the understanding of different attributes sources.
that contribute to the helpfulness and adoption of online A final limitation pertains to fake reviews. Recent stud-
consumer reviews and, more importantly, their numerous ies have highlighted an increase of manufactured and paid
interactions. reviews (Luca and Zervas 2016). Although some of our ob-
The exploration of the various contingencies also provides served reviews might have come from such sources, they
opportunities to resolve inconsistent findings from previous also might have received legitimate votes, so we can still
review research. Especially in the case of reviewer reputa- learn from those reviews (e.g., what makes them helpful)
tion (Karimi and Wang 2017), our results indicate that the (Mukherjee, Liu, and Glance 2012). However, future research
inconsistent findings may be due to the higher helpfulness might take this issue into account, and investigate how images
scores that reviewers with higher levels of reputation receive in fake reviews work, by analyzing whether the presence (and
when they add images to their reviews. Future research could character) of images may even be helpful to discern whether
examine this effect in more depth and also try to shed more a review is fabricated.

16
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Acknowledgments Choi, Hoon S. (2022), “Do Extraordinary Claims Require Extraordinary Ev-
idence? Differential Effect of Trust Cues on Helpfulness by Review Ex-
We extend our sincere gratitude to the Editors in Chief, tremity: An Empirical Study Using Big Data,” European Journal of In-
formation Systems, 1–22.
Katrijn Gielens and Anne Roggeveen, for their invaluable Choi, Hoon S. & Leon Steven (2020), “An Empirical Investigation of Online
and most insightful guidance during the development of this Review Helpfulness: A Big Data Perspective,” Decision Support Systems,
manuscript. Additionally, we express our appreciation to the 139, 113403.
anonymous Associate Editor and the three reviewers for their Cohen, Jacob, Cohen Patricia, West Stephen G. & Aiken Leona S. (2003).
Applied Multiple Regression/Correlation Analysis for the Behavioral Sci-
constructive and enriching comments, significantly contribut-
ences (3rd ed.). Lawrence Erlbaum Associates Publishers.
ing to the advancement of this work. Furthermore, we ac- Dhar, Ravi & Wertenbroch Klaus (2000), “Consumer Choice between He-
knowledge and appreciate our numerous colleagues who gen- donic and Utilitarian Goods,” Journal of Marketing Research, 37 (1),
erously shared their feedback during various conferences and 60–71.
workshops, enhancing the quality of this manuscript. Dimoka, Angelika, Hong Yili & Pavlou Paul A. (2012), “On Product Uncer-
tainty in Online Markets: Theory and Evidence,” MIS Quarterly, 395–426.
Filieri, Raffaele, Raguseo Elisabetta & Vitari Claudio (2018), “When are
Supplementary materials Extreme Ratings More Helpful? Empirical Evidence on the Moderating
Effects of Review Characteristics and Product Type,” Computers in Hu-
man Behavior, 88, 134–42.
Supplementary material associated with this article can be Forman, Chris, Ghose Anindya & Wiesenfeld Batia (2008), “Examining the
found, in the online version, at doi:10.1016/j.jretai.2023.09. Relationship Between Reviews and Sales: The Role of Reviewer Identity
001. Disclosure in Electronic Markets,” Information Systems Research, 19 (3),
291–313.
Gao, Jie, Zhang Cheng, Wang Ke & Ba Sulin (2012), “Understanding Online
References Purchase Decision Making: The Effects of Unconscious Thought, Infor-
mation Quality, and Information Quantity,” Decision Support Systems, 53
Ahearne, Michael, Atefi Yashar, Lam Son K. & Pourmasoudi Mohsen (2022), (4), 772–81.
“The Future of Buyer–Seller Interactions: A Conceptual Framework and Greene, William H. (2019). Econometric Analysis (8th ed.). Pearson Interna-
Research Agenda,” Journal of the Academy of Marketing Science, 50, tional.
1–24. Hazée, Simon, Vaerenbergh Yves Van & Armirotto Vincent (2017), “Co-Cre-
Akerlof, G. (1970), “The market for lemons: quality, uncertainty, and the ating Service Recovery After Service Failure: The Role of Brand Equity,”
market mechanism,” Quarterly Journal of Economics, 84, 488–500. Journal of Business Research, 74, 101–9.
Atefi, Yashar, Ahearne Michael, Hohenberg Sebastian, Hall Zachary & Hennig-Thurau, Thorsten, Gwinner Kevin, Walsh Gianfranco & Grem-
Zettelmeyer Florian (2020), “Open Negotiation: The Back-End Benefits ler Dwayne (2003), “Electronic Word of Mouth: Motives for and Conse-
of Salespeople’s Transparency in the Front End,” Journal of Marketing quences of Reading Customer Articulations on the Internet,” International
Research, 57 (6), 1076–94. Journal of Electronic Commerce, 4 (8), 51–74.
Babić Rosario, Ana, Sotgiu Francesca, Valck Kristine de & Bijmolt Tammo Ho-Dac, Nga N., Carson Stephen J. & Moore William L. (2013), “The
H.A. (2016), “The Effect of Electronic Word of Mouth on Sales: A Meta– Effects of Positive and Negative Online Customer Reviews: Do Brand
Analytic Review of Platform, Product, and Metric Factors,” Journal of Strength and Category Maturity Matter?,” Journal of Marketing, 77 (6),
Marketing Research, 53 (3), 297–318. 37–53.
Backhaus, Klaus, Erichson Bernd, Gensler Sonja, Weiber Rolf & Hong, Hong, Xu Di, Alan Wang G. & Fan Weiguo (2017), “Understand-
Weiber Thomas (2021). Multivariate Analysis (16th ed.). Wiesbaden: ing the Determinants of Online Review Helpfulness: A Meta-Analytic
Gabler Verlag. Investigation,” Decision Support Systems, 102, 1–11.
Bazaarvoice (2021), Visual and Social Content Increase Online Sales, Hong, Kevin & Pavlou Paul (2014), “Product Fit Uncertainty in Online Mar-
(accessed May 13, 2023), [available at https://www.bazaarvoice.com/ kets: Nature, Effects and Antecedents,” Information Systems Research, 25
resources/visual- and- social- content- increase- online- sales/] (2), 328–44.
Benedicktus, Ray L., Brady Michael K., Darke Peter R. & Voorhees Clay M. Hsieh, Yi-Ching, Chiu Hung-Chang & Chiang Mei-Yi (2005), “Maintaining
(2010), “Conveying Trustworthiness to Online Consumers: Reactions to a Committed Online Customer: A Study Across Search-Experience-Cre-
Consensus, Physical Store Presence, Brand Familiarity, and Generalized dence Products,” Journal of Retailing, 81 (1), 75–82.
Suspicion,” Journal of Retailing, 86 (4), 322–35. Hu, Han-fen & Krishen Anjala S. (2019), “When is Enough, Enough? In-
Biehal, Gabriel, Stephens Debra & Curio Eleonora (1992), “Attitude Toward vestigating Product Reviews and Information Overload From a Consumer
the Ad and Brand Choice,” Journal of Advertising, 21 (3), 19–36. Empowerment Perspective,” Journal of Business Research, 100, 27–37.
Bigne, Enrique, Chatzipanagiotou Kalliopi & Ruiz Carla (2020), “Pictorial Hu, Nan, Zhang Jie & Pavlou Paul A. (2009), “Overcoming the J-shaped
Content, Sequence of Conflicting Online Reviews and Consumer Deci- Distribution of Product Reviews,” Communications of the ACM, 52 (10),
sion-Making: The Stimulus-Organism-Response Model Revisited,” Jour- 144–7.
nal of Business Research, 115, 403–16. Huang, Peng, Lurie Nicholas H. & Mitra Sabyasachi (2009), “Searching
Bleier, Alexander, Harmeling Colleen M. & Palmatier Robert W. (2019), for Experience on the Web: An Empirical Examination of Consumer
“Creating Effective Online Customer Experiences,” Journal of Marketing, Behavior for Search and Experience Goods,” Journal of Marketing, 73
83 (2), 98–119. (2), 55–69.
Ceylan, Gizem, Diehl Kristin & Proserpio Davide (2023), “EXPRESS: Words Interbrand (2018), Best Global Brands 2018 [Brand Report], (accessed
Meet Photos: When and Why Visual Content Increases Review Helpful- March 5, 2021), [available at https:// www.interbrand.com/ wp-content/
ness,” Journal of Marketing Research. uploads/ 2018/ 10/ Interbrand_Best _Global_Brands_2018.pdf]
Cheng, Yi-Hsiu & Ho Hui-Yi (2015), “Social Influence’s Impact on Reader Karaman, Hülya (2021), “Online Review Solicitations Reduce Extremity Bias
Perceptions of Online Reviews,” Journal of Business Research, 68 (4), in Online Review Distributions and Increase Their Representativeness,”
883–7. Management Science, 67 (7), 4420–45.
Chevalier, Judith A. & Mayzlin Dina (2006), “The Effect of Word of Mouth Karimi, Sahar & Wang Fang (2017), “Online Review Helpfulness: Impact of
on Sales: Online Book Reviews,” Journal of Marketing Research, 43 (3), Reviewer Profile Image,” Decision Support Systems, 96, 39–48.
345–54. Kim, Minjeong & Lennon Sharron (2008), “The Effects of Visual and Verbal

17
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

Information on Attitudes and Purchase Intentions in Internet Shopping,” Variance in Product Reviews on Product Evaluation,” Psychology & Mar-
Psychology & Marketing, 25 (2), 146–78. keting, 30 (7), 543–54.
Kim, Youngsoo & Krishnan Ramayya (2015), “On Product-Level Uncertainty Peck, Joann & Childers Terry L. (2003), “To Have and to Hold: The Influence
and Online Purchase Behavior: An Empirical Analysis,” Management Sci- of Haptic Information on Product Judgments,” Journal of Marketing, 67
ence, 61 (10), 2449–67. (2), 35–48.
Klostermann, Jan, Plumeyer Anja, Böger Daniel & Decker Reinhold (2018), Pennington, Jeffrey, Socher Richard & Manning Christopher D. (2014).
“Extracting Brand Information from Social Networks: Integrating Image, “Glove: Global Vectors for Word Representation”, in Proceedings of the
Text, and Social Tagging Data,” International Journal of Research in 2014 Conference on Empirical Methods in Natural Language Processing
Marketing, 35 (4), 538–56. 1532–1154.
Kübler, Raoul, Pauwels Koen, Yildirim Gökhan & Fandrich Thomas (2018), Peterson, Robert A., Balasubramanian Sridhar & Bronnenberg Bart J. (1997),
“App Popularity: Where in the World Are Consumers Most Sensitive to “Exploring the Implications of the Internet for Consumer Marketing,”
Price and User Ratings?,” Journal of Marketing, 82 (5), 20–44. Journal of the Academy of Marketing Science, 25 (4), 329–46.
LandingCube (2022), 57 Amazon Statistics to Know in 2022, (accessed June Pieters, Rik & Wedel Michel (2004), “Attention Capture and Transfer in Ad-
23, 2022), [available at https:// landingcube.com/ amazon-statistics/ ] vertising: Brand, Pictorial, and Text-Size Effects,” Journal of Marketing,
Lee, Soyeon, Lee Saerom & Baek Hyunmi (2021), “Does the Dispersion 68 (2), 36–50.
of Online Review Ratings Affect Review Helpfulness?,” Computers in Racherla, Pradeep & Friske Wesley (2012), “Perceived ‘Usefulness’ of On-
Human Behavior, 117. line Consumer Reviews: An Exploratory Investigation across three Ser-
Lenzner, Alwine, Schnotz Wolfgang & Müller Andreas (2013), “The Role of vices Categories,” Electronic Commerce Research and Applications, 11
Decorative Pictures in Learning,” Instructional Science, 41 (5), 811–31. (6), 548–59.
Li, Kunlin, Chen Yuhan & Zhang Liyi (2020), “Exploring the Influence of Ravula, Prashanth, Jha Subhash & Biswas Abhijit (2022), “Relative Persua-
Online Reviews and Motivating Factors on Sales: A Meta-Analytic Study siveness of Repurchase Intentions Versus Recommendations in Online
and the Moderating Role of Product Category,” Journal of Retailing and Reviews,” Journal of Retailing, 98 (4), 724–40.
Consumer Services, 55, 102107. Roggeveen, Anne L., Grewal Dhruv, Townsend Claudia & Krishnan Ram
Li, Yiyi & Xie Ying (2020), “Is a Picture Worth a Thousand Words? An Em- (2015), “The Impact of Dynamic Presentation Format on Consumer Pref-
pirical Study of Image Content and Social Media Engagement,” Journal erences for Hedonic Products and Services,” Journal of Marketing, 79
of Marketing Research, 57 (1), 1–19. (6), 34–49.
Lin, Tom M., Lu Kuan-Yi & Wu Jia-Jhou (2012), “The Effects of Visual In- Salehan, Mohammad & Kim Dan J. (2016), “Predicting the Performance of
formation in eWOM Communication,” Journal of Research in Interactive Online Consumer Reviews: A Sentiment Mining Approach to Big Data
Marketing, 6 (1), 7–26. Analytics,” Decision Support Systems, 81, 30–40.
Littler, Dale & Melanthiou Demetris (2006), “Consumer Perceptions of Risk Schoenmueller, Verena, Netzer Oded & Stahl Florian (2020), “The Polar-
and Uncertainty and the Implications for Behaviour Towards Innovative ity of Online Reviews: Prevalence, Drivers and Implications,” Journal of
Retail Services: The Case of Internet Banking,” Journal of Retailing and Marketing Research, 57 (5), 853–77.
Consumer Services, 13 (6), 431–43. Siering, Michael, Muntermann Jan & Rajagopalan Balaji (2018), “Explaining
Liu, Zhiwei & Park Sangwon (2015), “What Makes a Useful Online Re- and Predicting Online Review Helpfulness: The Role of Content and
view? Implication for Travel Product Websites,” Tourism Management, Reviewer-Related Signals,” Decision Support Systems, 108, 1–12.
47, 140–51. Singh, Jyoti P., Irani Seda, Rana Nripendra P., Dwivedi Yogesh K.,
Luca, Michael & Zervas Georgios (2016), “Fake It till You Make It: Rep- Saumya Sunil & Roy Pradeep Kumar (2017), “Predicting the ‘Helpful-
utation, Competition, and Yelp Review Fraud,” Management Science, 62 ness’ of Online Consumer Reviews,” Journal of Business Research, 70,
(12), 3412–27. 346–55.
Luo, Hanyang, Zhou Wanhua, Song Wugang & He Xiaofu (2022), “An Em- Smith, Ruth Ann (1991), “The effects of visual and verbal advertising infor-
pirical Study on the Differences between Online Picture Reviews and mation on consumers’ inferences,” Journal of Advertising, 20 (4), 13–24.
Text Reviews,” Information, 13 (7), 344. Srivastava, Vartika & Kalro Arti D. (2019), “Enhancing the Helpfulness of
Manes, Eran & Tchetchik Anat (2018), “The Role of Electronic Word of Online Consumer Reviews: The Role of Latent (Content) Factors,” Jour-
Mouth in Reducing Information Asymmetry: An Empirical Investigation nal of Interactive Marketing, 48, 33–50.
of Online Hotel Booking,” Journal of Business Research, 85, 185–96. Swaminathan, Vanitha, Sorescu Alina, Steenkamp Jan-Benedict E.M.,
Meek, Stephanie, Wilk Violetta & Lambert Claire (2021), “A Big Data Ex- O’Guinn Thomas Clayton Gibson & Schmitt Bernd (2020), “Branding in
ploration of the Informational and Normative Influences on The Helpful- a Hyperconnected World: Refocusing Theories and Rethinking Bound-
ness of Online Restaurant Reviews,” Journal of Business Research, 125, aries,” Journal of Marketing, 84 (2), 24–46.
354–67. Topaloglu, Omer & Dass Mayukh (2021), “The Impact of Online Review
Moore, Sarah G. (2015), “Attitude Predictability and Helpfulness in Online Content and Linguistic Style Matching on New Product Sales: The Mod-
Reviews: The Role of Explained Actions and Reactions,” Journal of Con- erating Role of Review Helpfulness,” Decision Sciences, 52 (3), 749–75.
sumer Research, 42 (1), 30–44. Unnava, H.Rao & Burnkrant Robert E. (1991), “An Imagery-Processing View
Mudambi, Susan M. & Schuff David (2010), “Research Note: What Makes a of The Role of Pictures in Print Advertisements,” Journal of Marketing
Helpful Online Review? A Study of Customer Reviews on Amazon.com,” Research, 28 (2), 226–31.
MIS Quarterly, 34 (1), 185–200. Wakefield, Kirk L. & Inman J.Jeffrey (2003), “Situational Price Sensitivity:
Mukherjee, Arjun, Liu Bing & Glance Natalie (2012). “Spotting Fake Re- The Role of Consumption Occasion, Social Context and Income,” Journal
viewer Groups in Consumer Reviews”. In Proceedings of the 21st Inter- of Retailing, 79 (4), 199–212.
national Conference on World Wide Web (pp. 191–200). Weathers, Danny, Sharma Subhash & Wood Stacy L. (2007), “Effects of On-
Ngo-Ye, Thomas L. & Sinha Atish P. (2014), “The Influence of Reviewer line Communication Practices on Consumer Perceptions of Performance
Engagement Characteristics on Online Review Helpfulness: A Text Re- Uncertainty for Search and Experience Goods,” Journal of Retailing, 83
gression Model,” Decision Support Systems, 61, 47–58. (4), 393–401.
Pallas, Florian, Mittal Vikas & Groening Christopher (2014), “Allocation of Wickham, Hadley (2016), rvest: Easily Harvest (Scrape) Web Pages. [avail-
Resources to Customer Satisfaction and Delight Based on Utilitarian and able at https://rvest.tidyverse.org]
Hedonic Benefits,” Journal of Research in Marketing, 2 (1), 106–12. Wu, Philip F., Heijden Hans Van der & Korfiatis Nikolaos (2011). “The
Pan, Yue & Zhang Jason Q. (2011), “Born Unequal: A Study of the Helpful- Influences of Negativity and Review Quality on the Helpfulness of Online
ness of User-Generated Product Reviews,” Journal of Retailing, 87 (4), Reviews”, in International Conference on Information Systems. Atlanta:
598–612. AIS.
Park, Se-Bum & Park Do-Hyung (2013), “The Effect of Low- versus High– Wu, Philip Fei (2013), “In Search of Negativity Bias: An Empirical Study

18
JID: RETAIL
ARTICLE IN PRESS [m5+;September 25, 2023;11:20]
R.V. Kübler, L. Lobschat, L. Welke et al. Journal of Retailing xxx (xxxx) xxx

of Perceived Helpfulness of Online Reviews,” Psychology & Marketing, Zhang, Mengxia & Luo Lan (2023), “Can Consumer-Posted Photos Serve
30 (11), 971–84. as a Leading Indicator of Restaurant Survival? Evidence From Yelp,”
Wu, Ruijuan, Wu Heng-Hui & Wang Cheng Lu (2021), “Why is a Picture Management Science, 69 (1), 25–50.
‘Worth a Thousand Words’? Pictures as Information in Perceived Help- Zinko, Robert, Furner Christopher P., Burgh-Woodman Helene de, John-
fulness of Online Reviews,” International Journal of Consumer Studies, son Patricia & Sluhan Anne (2021), “The Addition of Images to eWOM
45 (3), 364–78. in the Travel Industry: An Examination of Hotels, Cruise Ships and Fast
Xu, Pei, Chen Liang & Santhanam Radhika (2015), “Will Video be the Next Food Reviews,” Journal of Theoretical and Applied Electronic Commerce
Generation of E-Commerce Product Reviews? Presentation Format and Research, 16 (3), 525–41.
the Role of Product Type,” Decision Support Systems, 73, 85–96. Zinko, Robert, Stolk Paul, Furner Zhan & Almond Brad (2020), “A Picture
Yang, Yang, Wang Yuejun & Zhao Jichang (2023), “Effect of User-Gener- Is Worth a Thousand Words: How Images Influence Information Quality
ated Image on Review Helpfulness: Perspectives from Object Detection,” and Information Load in Online Reviews,” Electronic Markets, 30 (4),
Electronic Commerce Research and Applications, 57. 775–89.
Yildirim, Gökhan & Kübler Raoul V. (2023). Applied Marketing Analytics
Using R (1st ed.). SAGE Publications Ltd.

19

You might also like