RESEARCH ARTICLE
DOI:10.25300/MISQ/2022/15666
MIS Quarterly Vol. 46 No. 3 pp. 1573-1602 / September 2022
1573
COMPETING WITH THE SHARING ECONOMY:
INCUMBENTS’ REACTION ON REVIEW MANIPULATION
1
Cheng Nie
Ivy College of Business, Iowa State University,
Ames, IA, U.S.A. {cheng@chengnie.com}
Zhiqiang (Eric) Zheng, Sumit Sarkar
Naveen Jindal School of Management, University of Texas at Dallas,
Richardson, TX, U.S.A. {ericz@utdallas.edu} {sumit@utdallas.edu}
Introduction
The sharing economy presents a new economic system that
uses technology-mediated platforms to match customers with
service providers for fee-based exchanges such as short-term
apartment rentals, car rides, or household tasks (Slee, 2016, p.
9). These platforms provide a convenient and inexpensive way
for owners to make (potentially underutilized) goods or
services available to consumers. Recent years have witnessed
rapid growth in sharing economy companies. Prominent
examples include Uber and Lyft in the transportation industry
and Airbnb in the hospitality industry.
The emergence of such companies has the potential to
significantly disrupt incumbents in traditional markets. Such
companies represent a different type of competitor compared
to traditional firms, requiring firms to revisit models of
1
T. Ravichandran was the accepting senior editor for this paper. Jui
Ramaprasad served as the associate editor.
competition to account for the novelty of this expanded
competitive landscape (Eckhardt et al., 2019). Ways in which
traditional firms may need to adjust include pricing decisions
(Li & Srinivasan, 2019; Zervas et al., 2017), branding (Bardhi
& Eckhardt, 2012), product variety (Hughes, 2017), capacity
management (Cramer & Krueger, 2016), regulatory reforms
(Kemp, 2017), distribution channels (Tian & Jiang, 2018), and
even creating their own sharing platforms (Wallenstein &
Shelat, 2017). Given the plethora of possible responses,
Eckhardt et al. (2019, p. 16) observe that “further research is
needed to more fully assess the impact of sharing platform
entry.” They go on to note that the “response by traditional
firms may vary across different types of product or service
categories as well as by a firm’s standing in an industry.”
In our work, we investigate a nascent competition strategy—
consumer opinion manipulation via online review
Nie et al. / Competing with the Sharing Economy
1574
MIS Quarterly Vol. 46 No. 3 / September 2022
platforms—in the lodging sector of the hospitality industry.
This represents an ideal sector for examining the impact of the
sharing economy because of the advent of Airbnb. Airbnb’s
platform allows hosts to list their properties for rent to guests
at a price set by the hosts and accrues revenues by charging
service fees from both hosts and guests. Airbnb has
experienced rapid growth since its launch in 2008. As of June
2021, Airbnb hosts over 5.6 million listings in more than
100,000 cities.
2
As of October 2021, Airbnb has a market cap
of more than $100 billion, exceeding several major hotel
chains such as Marriott and Hilton.
3
Online reviews have become an influential source of
information for helping customers make purchase decisions
(e.g., Chevalier & Mayzlin, 2006; Dellarocas et al., 2007;
Zhu & Zhang, 2010), especially in the hospitality industry
(Ye et al., 2011). In fact, customer reviews are generally
considered to be more credible than promotional campaigns
(Bickart & Schindler, 2001; X. Lu et al., 2013). Online
reviews are especially critical when competing with the
sharing economy; such technology-mediated platforms
enable strangers to transact with each other, adding
importance for online reviews as a means to establish trust
between customers and service providers.
Recognizing the importance of online reviews in competing
for potential customers, many firms have resorted to
manipulating reviews (e.g., He et al., in press; Tibken, 2013).
In the context of the conventional lodging business, Mayzlin
et al. (2014) have demonstrated that hotels with neighboring
competing hotels tend to demote each other more. It has also
been established that firms are more likely to engage in
manipulation when competition intensifies (Luca & Zervas,
2016; Mayzlin et al., 2014). The success of Airbnb is taking a
toll on the incumbents in the disrupted lodging business, who
are forced to constantly respond to the competition. For
example, Zervas et al. (2017) document that hotels have
responded to the entrance of Airbnb by reducing prices.
Since Airbnb continues to intensify the competition in the
lodging business, and research has shown that competition
motivates hotels to engage in review manipulation, we ask
whether hotels change their review manipulation strategies in
response to the emergence of Airbnb in their market and
question whether this is exacerbated in response to Airbnb
growth. We draw on strategic group theory (e.g., Cool &
Schendel, 1987; Fiegenbaum & Thomas, 1995; Mas‐Ruiz et
al., 2014) to frame our questions. The theory helps explain
how the nature of competition between Airbnb properties and
hotels differs from that among hotels themselves, e.g., in terms
2
https://www.airbnb.com/about/about-us (accessed December 11, 2021).
3
https://companiesmarketcap.com/hotels/largest-hotel-companies-by-
market-cap/ (accessed December 11, 2021)
of product or service types, promotional channels, locations,
and review channels. These differences have important
implications for the review manipulation strategies used by
incumbent hotels. Does the different nature of competition
posed by the sharing economy lead incumbents to change their
review manipulation activities? Specifically, the research
question we address is whether and how conventional hotels
change their review manipulation behaviors in response to the
emergence and growth of Airbnb in their market. We examine
two types of manipulation strategies, self-promotion and
demotion. We further investigate whether changes in
manipulation behaviors are heterogeneous across different
categories of hotels (i.e., high-end and low-end hotels).
We empirically examine the review manipulation level that
a hotel engages in as a function of the supply of its nearby
Airbnb listings. We do this by analyzing a unique panel data
of 2,188 hotels extracted from six different sources—
Airbnb, TripAdvisor, Expedia, AirDNA, the Texas
Comptroller’s Office, and Smith Travel Research (STR).
4
We find that the supply of nearby Airbnb listings indeed
exerts a significant impact on hotels’ review manipulation
strategies and that such impacts are not homogeneous across
different types of hotels. Our main results reveal that high-
end hotels promote themselves more after Airbnb penetrates
their market; surprisingly, they demote each other less
following the entry of Airbnb. Low-end hotels do not
increase their self-promotion or demotion behaviors as
Airbnb becomes more popular. Our findings suggest that the
disruptive innovations from sharing economy companies
change the dynamics of competition among the incumbents
in unexpected ways. These findings are aligned with
predictions from strategic group theory.
Our work contributes to the growing literature on the impact
of the sharing economy on traditional incumbents by
demonstrating how the emergence of Airbnb changes hotels’
review manipulation strategies. Our findings have interesting
implications both for review-hosting platforms like
TripAdvisor as well as their end-users. With the
understanding that increased Airbnb supply drives high-end
hotels to self-promote more while reducing their demotion
activities, review platforms could potentially adjust their fake-
review filtering algorithms to account for both Airbnb supply
near a hotel and the type of the hotel. Our results suggest that
customers frequenting high-end hotels should take extra care
when using reviews to help them decide where to stay, e.g.,
by discounting overly positive reviews for these hotels.
4
We thank AirDNA, Smith Travel Research (STR) and the Texas
Comptroller’s Office for graciously sharing their data with us.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1575
Theoretical Framework
We outline the main empirical questions addressed in the
paper and discuss the theoretical rationale underpinning each
question.
Incentives to Manipulate Reviews
Customer reviews have been found to be important drivers of
consumers’ purchase decisions (e.g., Chevalier & Mayzlin,
2006; Dellarocas et al., 2007; Ye et al., 2011; Zhu & Zhang,
2010). As a result, reviews play an important role in shaping
customer opinions, and firms routinely monitor and manage
them as an integral part of their marketing strategy
(Dellarocas, 2006; Yang et al., 2019). Because of the
importance of online reviews for businesses, firms may even
attempt to manipulate customer opinions by disguising self-
promotion as customer recommendations (Mayzlin, 2006).
5
Research has shown that many firms resort to manipulating
reviews (e.g., He et al., in press; Tibken, 2013). Consequently,
it is not uncommon to encounter fake (manipulated) reviews
on review websites like TripAdvisor and Yelp (Lappas et al.,
2016). For example, more than 50,000 Chinese retailing
accounts were suspended from Amazon in 2021 because of
review manipulations, costing these third-party merchants an
estimated $15.4 billion (Einhorn et al., 2021). Despite the
commitment to combating fraud using filtering algorithms and
legal actions (Marinova, 2016), a significant portion of online
reviews have been found to be fake. The percentage of fake
reviews is estimated at around 15% to 30% despite the best
efforts of review platforms to combat fake reviews (Belton,
2015; Lappas et al., 2016; Luca & Zervas, 2016).
Review manipulation occurs in two different ways: self-
promotion and demotion. Firms promote themselves by
posting fake positive reviews, a phenomenon that has been
scrutinized in the popular press. For example, in February
2004, due to a software error, Amazon.com’s Canadian site
mistakenly revealed the identities of book reviewers and a
number of these reviews were found to be written by the
books’ own publishers and authors. For example, it was found
that Dave Eggers, the author of A Heartbreaking Work of
Staggering Genius, posted positive reviews for his own book
to inflate its rating (Harmon, 2004). Similarly, Luca and
Zervas (2016) found that a restaurant is more likely to self-
promote after receiving negative reviews. Mayzlin et al.
(2014) observe that independent hotels tend to promote
themselves more than chain-affiliated hotels.
5
Mayzlin (2006) refers to such behavior as promotional chat.
Besides promoting themselves, firms may demote their
competitors by posting negative reviews. Consumers are
found to respond more to negative reviews than to positive
reviews (Chevalier & Mayzlin, 2006). Therefore, firms may
be inclined to post disingenuous negative reviews for its
competitors, especially when their products are strong
substitutes. For example, Luca and Zervas (2016) found that a
restaurant is more likely to demote other restaurants when
facing increased competition. In the lodging business, before
Airbnb became popular, Mayzlin et al. (2014) documented
that hotels with nearby competitors are more likely to receive
fake unfavorable reviews than those without competitors in
the neighborhood.
The incentives to manipulate reviews may also depend on the
firm’s own quality as demonstrated in two analytical
modeling papers. Mayzlin (2006) argued that producers with
lower quality will expend more resources on manipulating
reviews. On the other hand, Dellarocas (2006) showed that an
equilibrium exists where the high-quality producer will invest
more resources in review manipulation. Both papers assume
that posting negative reviews about one’s competitor is
qualitatively equivalent to posting positive reviews about
oneself, without differentiating self-promotion from
demotion.
However, self-promotion and demotion may impact
consumers in different ways. Leading review platforms such
as Yelp and TripAdvisor consider both the quality and
quantity of reviews when ranking search results. Lappas et al.
(2016) use simulations to show that the relative effectiveness
of self-promotion and demotion in influencing a hotel’s
visibility varies depending on how consumers form their
consideration sets. They posit that self-promotions are more
effective when consumers consider only the top few choices
in ranked results to form their consideration sets, while
demotions are more effective when customers are willing to
search further down the list.
In summary, hotels have incentives to manipulate reviews;
further, it has been established that they engage in such
practices and that they manipulate more in the face of fiercer
competition. Further, hotels may utilize self-promotion and
demotion differently, and manipulation actions may be
heterogeneous across hotel types.
Identification of Review Manipulation
Identifying manipulative reviews is nontrivial. Several
algorithms based on text mining have been developed to
detect fake reviews (Jindal & Liu, 2008; Kumar et al., 2018;
Nie et al. / Competing with the Sharing Economy
1576
MIS Quarterly Vol. 46 No. 3 / September 2022
Li et al., 2014) with varying degrees of success. Review
websites like Yelp screen reviews using their proprietary
algorithms.
6
Despite such endeavors, it remains a challenging
task for a website to accurately detect fake reviews since
manipulated reviews are designed to mimic truthful reviews.
As mentioned earlier, a substantial number of fake reviews
show up on review platforms despite the use of filtering
algorithms. Since we lack ground truth regarding which
reviews are genuine and which ones are fake, we do not use
text mining approaches to classify reviews as fake or not.
Instead, we use the identification approach proposed by
Mayzlin et al. (2014). Their approach exploits the different
regulations for users to post reviews between the two travel
websites—TripAdvisor and Expedia. TripAdvisor allows any
user, whether a real customer or not, to post reviews. Expedia,
on the other hand, only allows customers who have stayed in
a hotel (either booked through Expedia or its partners) to post
reviews for the hotel on its site. Therefore, review
manipulations will be more likely to occur on TripAdvisor’s
site than Expedia’s because of the lower costs associated with
fabricating a fake review on TripAdvisor. Thus, for hotels, the
difference in review ratings between the two review platforms
offers an indication of the level of review manipulation.
Mayzlin et al. (2014) consider both the self-promotion type of
review manipulation as well as demotion by competitors. Self-
promotion activities are identified by examining the
difference in the proportion of reviews with high ratings (5-
star ratings) on TripAdvisor as opposed to Expedia, while
demotion levels are identified using the net proportion of
reviews with low ratings (1-star and 2-star ratings).
Strategic Group Lens
As of September 2021, Airbnb had hosted over 1 billion guest
arrivals.
7
Similar to the way in which the sharing economy has
changed car owners to Hertz competitors (Stein, 2015, p. 34),
Airbnb has transitioned certain homeowners from persons
with homes to hotel competitors. The encroachment of Airbnb
listings into the territory of traditional hotels is clearly taking
a toll on incumbent hotels. While the impact of Airbnb on
hotel prices has been previously examined (Li & Srinivasan,
2019; Zervas et al., 2017), the competitive reactions of
incumbent hotels in terms of review manipulation have not.
We draw on the theory of strategic groups, a central construct
in the strategy literature, to examine the competitive behavior
of firms in a market (e.g., Cool & Schendel, 1987;
Fiegenbaum & Thomas, 1995; Short et al., 2007). Porter
(1979) formalizes the notion of a strategic group as a group
of firms in an industry that closely compete against each other
6
http://www.yelp-support.com/article/Why-would-a-review-not-be-
recommended (last accessed December 11, 2021)
and are similar to one another along key strategic dimensions
(e.g., degree of vertical integration and investment in
advertising and R&D). Fiegenbaum and Thomas (1995) note
that a strategic group establishes a reference point for group
members when they make strategic decisions. Strategic group
theory has been widely used to study how firms compete in an
industry (e.g., Mas‐Ruiz et al., 2014; Mas‐Ruiz & Ruiz‐
Moreno, 2011; Short et al., 2007).
Even though competition from Airbnb substitutes the demand
for a hotel (akin to competition from another hotel), this does
not mean that Airbnb’s entry is identical to that of other
competing hotels. Airbnb and hotels are quite different in
terms of their asset bases, cost structures, and other
dimensions of their strategic profiles, which are typically used
to distinguish strategic groups (Peteraf, 1993). As a result, the
competition from Airbnb and the competition from hotels are
quite different.
To begin with, the offerings are quite different across these
two types of providers. For example, hotels often need to
provide amenities like meeting rooms, conference facilities,
shuttle services, and gyms. These are typically not part of
Airbnb listings. On the other hand, Airbnb listings often offer
amenities such as full kitchen and laundry facilities, cozy
living and dining areas, and local knowledge from Airbnb
hosts. There is also a wide variety of Airbnb listings (e.g.,
treehouse, tent, and recreational vehicle) to accommodate
travelers’ unique preferences compared to the relatively
standardized offerings from hotels (Kelleher, 2019).
Importantly, because Airbnb does not own the physical
properties it provides access to, its asset base is very different
from that of traditional hotels. Instead, Airbnb’s key assets are
the underlying information technologies that enable effective
transactions between hosts and guests.
The cost structures of hotels and Airbnb are very different as
well. Hotels cannot quickly change their supply because of the
long lead time needed for construction and employee training.
In contrast, Airbnb can expand supply almost overnight, as
demonstrated in the context of seasonal events like the SXSW
festival in Austin, Texas (Zervas et al., 2017). Thus, Airbnb
can better accommodate volatile demand because of the low
capital costs necessary to add (or remove) marginal capacity
(Li & Srinivasan, 2019). The regulation and compliance costs
that Airbnb and hotels face are also different. Hotels must
comply with a litany of health, safety, and zoning rules, in
addition to registering with local agencies and paying taxes
and other fees (Martineau, 2019). The regulations that Airbnb
faces in different cities, if any, are mostly directed toward
mitigating neighborhood impact rather than creating a level
7
https://news.airbnb.com/about-us/ (last accessed December 11, 2021).
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1577
playing field in the hospitality industry (Nieuwland & van
Melik, 2020). Another driver of the existence of strategic
groups is the presence of features that serve as mobility
barriers across the two groups (Mas‐Ruiz et al., 2014; Porter,
1979). The different regulations that apply, respectively, to
hotels and Airbnb properties make it difficult for an entity to
transition from one group to the other.
Because of these differences, the important strategic
considerations and related actions of hotels and Airbnb are also
quite different. For hotels, long-term strategic decisions include
factors such as physical location, capacity, and the quality of
accommodations; short-term decisions include pricing and
promotional activities (Kim, 2018). Hotels typically locate their
properties in pockets of local density, such as downtown areas.
For pricing decisions, hotels must pay considerable attention to
location, season, day of week, and other factors so that they can
dynamically set prices for their offerings to maximize revenues.
In contrast, an important long-term strategic imperative for
Airbnb is to ensure that its platform attracts healthy
participation from both hosts and guests, thereby generating
strong network effects. To promote this, algorithms matching
potential guests with hosts are an important aspect of Airbnb’s
growth strategy. Also, since guests and hosts typically do not
know each other in advance, building trust between them is a
very important consideration (Edelman & Luca, 2012). The
platform uses a two-sided review system (reputation system) to
alleviate such concerns (Proserpio et al., 2018). As far as
revenues are concerned, Airbnb sets the commission fee rate for
hosts and guests transacting on its system and provides
complete flexibility to hosts in setting prices for their listings.
All these considerations exemplify the different strategic
behaviors of hotels and Airbnb.
Airbnb’s Impact on Hotels’ Review
Manipulations
Strategic group theory asserts that firms within the same group
recognize their mutual dependence and follow similar
strategies in response to market opportunities or threats (e.g.,
Mas‐Ruiz & Ruiz‐Moreno, 2017; Porter, 1979). Cool and
Dierickx (1993) noted that changes in the strategic group
structure could lead to a shift from within-group rivalry to
between-group rivalry. In our context, such a shift to between-
group competition is expected to occur when incumbent
hotels face competition from Airbnb listings.
The issue of central interest here is how competition from
Airbnb led the incumbent group of hotels to adjust their
review manipulation strategies (over and above their reaction
8
We observe that a very small portion of Airbnb listings are cross-listed on
both Airbnb and TripAdvisor. Zervas et al. (2021) found that out of 466,000
to conventional dimensions such as price adjustment and
lobbying). Importantly, the respective review platforms for
hotels and properties listed on Airbnb differ in important
ways. Reviewers evaluate Airbnb listings through Airbnb’s
own website whereas conventional review platforms such as
Expedia, TripAdvisor, and Yelp do not include reviews for
Airbnb listings.
8
Airbnb only allows users who have stayed at
a property to post reviews for that property, and it stipulates a
bilateral review system where the host also reviews guests
(Proserpio et al., 2018). Therefore, reviewers are not
anonymous to Airbnb.com, making reviewers identifiable.
Further, a typical Airbnb listing likely comprises only a few
rooms. The demand implications for a focal hotel from a new
hotel competitor would be equivalent to that from a fairly
large collection of neighboring Airbnb listings. This makes
the economics of faking reviews for nearby Airbnb listings
relative to that for a competing hotel markedly different, with
the costs of manipulating reviews (i.e., by demoting
competitors) an order of magnitude higher when dealing with
competition from Airbnb listings. Thus, incumbent hotels
cannot counter between-group competition from Airbnb
listings by demoting them in the same way as is possible when
dealing with competition from other hotels (i.e., within their
own strategic group).
While recognizing that “strategic group membership is a
predictor of the manner by which firms compete with one
another” (Smith et al., 1997, p. 156), the literature is silent
regarding how the emergence of a new strategic group could
impact rivalry within an existing group. With the emergence
of Airbnb listings, would hotels change their review
manipulation strategies targeted toward other hotels? On the
one hand, when Airbnb takes away a portion of demand from
hotels, the rivalry between hotels should become more intense
and the reaction across hotels may be to “instigate warfare”
(Smith et al., 1997, p. 151). In our context, this implies that
hotels may be incentivized to demote competing hotels more
in order to seize a larger share of the remaining demand. For
instance, it has been shown in both the hotel and restaurant
industries (Luca & Zervas, 2016; Mayzlin et al., 2014) that
demotion activities intensify in response to an increased
number of conventional competitors.
On the other hand, competing hotels now face Airbnb as a
common rival. As an ancient proverb goes: “the enemy of my
enemy is my friend”; thus, competing hotels may find it
beneficial to work together to fight against the common new
enemy. There is considerable evidence that incumbents do team
up to fight against sharing economy competition. For example,
taxi companies have teamed up to fight Uber by taking
collective legal actions against Uber’s lack of regulations
properties on TripAdvisor and 381,000 listings on Airbnb, only around
2,000 properties (0.5%) were cross-listed.
Nie et al. / Competing with the Sharing Economy
1578
MIS Quarterly Vol. 46 No. 3 / September 2022
(Goldstein, 2018). In the lodging business, the American Hotel
and Lodging Association (AHLA), a trade group that oversees
Marriott International, Hilton Worldwide, and Hyatt Hotels, has
not taken Airbnb’s incursions lightly. According to The New
York Times, AHLA has backed efforts by the Federal Trade
Commission and the state of New York to investigate Airbnb’s
impact on local housing prices since 2016 (Benner, 2017). The
AHLA has also launched a campaign to portray Airbnb hosts as
being commercial operators competing illegally with hotels
(AHLA, 2017). Hotels collectively funded an anti-Airbnb
mailer that claims that Airbnb has made local housing less
affordable (Bredderman, 2018).
The literature on strategic groups recognizes that firms
within a group may not always compete intensely with other
firms in their own group. As noted by Porter (2008, p. 17),
“if moves and counter moves escalate, then all firms in the
industry may suffer.” Indeed, it has been suggested that
rivalry in such cases will be lower within a group because
firms are better able to recognize their mutual dependence
and thus cooperate or tacitly collude with one another (Caves
& Porter, 1977; Peteraf, 1993). Tit-for-tat competitive
interactions within the same group can be unstable and
destructive (Smith et al., 1997). In our context, this means
that mutual demotion between incumbent hotels might be
counterproductive for all hotels.
9
Therefore, hotels demoting
competing hotels may be wary of retaliation that could end
up hurting the hotels altogether since consumers are able to
switch to Airbnb alternatives. Thus, strategic group theory
suggests that co-opetition rather than tit-for-tat is a viable
choice for members of the same group when between-group
competition exists. In our context, this amounts to incumbent
hotels becoming less incentivized to demote each other in
the presence of Airbnb competition.
If increasing mutual demotion is not an effective choice in
such cases, what other review manipulation actions can
incumbents take? One possibility is to self-promote more.
According to Porter (2008a, p. 84), “if an industry does not
distance itself from substitutes through product performance,
marketing, or other means, it will suffer in terms of
profitability—and often growth potential.” Therefore, when
competition resulting from Airbnb listings intensifies, hotels
may need to self-promote more to better differentiate
themselves; for example, such self-promotions could
highlight the unique amenities offered by a hotel. Since
Airbnb has shrunk the collective demand for hotels (Zervas et
9
Along similar lines, the balance theory proposed by Heider (1958)
conceptualizes a motive called “cognitive consistency,” which drives the
formation of friend and enemy relationships. Balance theory is aligned with
the tertius iungens strategic orientation where a newcomer facilitates new
al., 2017), a hotel may need to bolster its self-promotion
actions to fight other hotels for the shrinking demand.
The level of self-promotion may also depend on the nature of
the ratings received. Luca and Zervas (2016) found that a firm
receiving more negative reviews tends to self-promote more.
Their finding also suggests that hotels might self-promote less
if hotels receive fewer negative reviews. Two reasons may
contribute to fewer negative reviews. One, as already
discussed, hotels may demote each other less in response to
the emergence of the common rival Airbnb. Second,
customers whose needs are better suited to Airbnb listings
(e.g., the cost of meals and rooms for a family of four may be
significantly higher for hotels compared to Airbnb listings, see
McCool, 2015) would be more likely to prefer such properties
resulting in fewer poor ratings for hotels. In summary, it will
be interesting to explore which of these counteracting forces
dominates in the lodging business in the face of Airbnb
competition.
Differential Impact on Hotel Types
The literature on quality signaling postulates that high-quality
firms are expected to spend more advertising resources than
low-quality firms to promote their products, where advertising
serves as a credible signal of quality (see, e.g., Kihlstrom &
Riordan, 1984). In the lodging business, Hollenbeck (2018)
found that online reviews have emerged as a new type of
quality signal. Due to the importance of online reviews, it has
been documented that firms manipulate online reviews to
realize financial gains (Luca & Zervas, 2016; Mayzlin et al.,
2014). However, it is unclear whether high-end hotels and
low-end hotels would respond similarly to competition from
Airbnb listings.
Vertically differentiated firms (i.e., high-end vs. low-end)
often need to choose different competitive strategies that are
suitable for their needs. Michael Porter (2008b) posited that
firms in the low-cost position are better off utilizing the
“overall cost leadership” strategy to remain profitable after
their competitors have competed away their profits. In
contrast, firms at the higher end may adopt a “differentiation”
strategy by providing unique products or services, enhancing
customers’ brand loyalty, and lowering their price sensitivity.
For hotels, this implies that low-end hotels and high-end
hotels may employ different strategies to counter competition.
coordination between incumbents (Obstfeld, 2005). The predictions from
these theories are essentially the same as the prediction from strategic group
theory in our context.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1579
Regarding the nature of review manipulation in particular,
Lappas et al. (2016) showed that the relative effectiveness of
self-promotion or demoting others depends on the size of
consumers’ consideration sets, which could differ for different
types of hotels. Also, firms have been found to increase their
self-promotion to counter negative reviews (Luca & Zervas,
2016). Since the average ratings for low-end hotels are
generally lower than those for high-end hotels, low-end hotels
may need to self-promote more.
With regard to demotions, firms may increase such activities
when facing intensified competition (Luca & Zervas, 2016;
Mayzlin et al., 2014). That is, if the competition is more
intense for one type of hotel compared to another because of
the advent of Airbnb, then this type of hotel may be more
likely to increase demotion activities. Zervas et al. (2017) have
shown that while all hotels’ revenues are negatively
influenced by Airbnb, low-end hotels suffer more compared
to high-end ones. Accordingly, we may expect low-end hotels
to be more responsive to the challenge posed by Airbnb.
These studies suggest that different types of firms may adopt
different self-promotion and demotion strategies, but how the
quality dimension of firms plays a role remains unresolved.
Because of the different forces leading to conflicting
observations, it is unclear which one would dominate as Airbnb
gains more popularity and whether the outcomes would be the
same for different hotel types. To address this, we investigate
Airbnb’s impact on the change in review manipulation behavior
for low-end and high-end hotels separately.
In summary, we ask whether the emerging competition from
the sharing economy leads incumbents to manipulate reviews
differently. Specifically, we examine whether hotels engage
in more review manipulation (i.e., self-promotion and
demotion) when facing the new type of competition presented
by Airbnb. We also examine whether the change in review
manipulation behavior is homogeneous across different types
of hotels. The rapid emergence of the sharing economy
provides an ideal opportunity to study these new competition
dynamics across incumbent hotels.
Data
We obtained and synthesized data from six different sources:
Airbnb.com, Expedia.com, TripAdvisor.com, AirDNA.co,
Smith Travel Research (STR), and the Texas Comptroller’s
10
https://en.wikipedia.org/wiki/List_of_cities_in_Texas_by_population
(last accessed December 11, 2021).
11
To reflect the different demand across cities, STR categorizes hotel tiers
using different price brackets for different cities. The price brackets used in
Office (comptroller.texas.gov). Each source provides
complementary data items for our analyses.
We conducted analyses for hotels in the state of Texas. All
data collected are from the period between January 2008 (the
year of Airbnb’s inception) to December 2015. We restricted
our attention to hotels in cities with a population of over
50,000, as the number of hotels and Airbnb listings is typically
low in smaller cities. There were 67 cities in Texas meeting
this requirement for the period under consideration.
10
We
wrote Python crawlers to scrape the review data on all hotels
in these cities from the TripAdvisor and Expedia sites. We
obtained the review ratings and review dates from each site.
Expedia provides a link to each hotel’s TripAdvisor page (if it
exists). Therefore, matching the hotels on these two websites
is straightforward. The STR Texas census data include the
name, price tier, and address for each hotel. Based on STR’s
price tier information, we considered two categories of hotels:
low-end (or budget hotels) and high-end (non-budget
hotels).
11
The Texas Comptroller’s Office provides public
records on quarterly hotel tax filing records for all the hotels
in the state of Texas,
12
in addition to hotel names and
addresses. We identified the period of each hotel’s operation
based on tax filing records. We matched all the hotels
identified on TripAdvisor and Expedia with their
corresponding entries in the data provided by STR and by the
Texas Comptroller’s Office using the hotels’ names and
addresses. Because the tax filing data is available on a
quarterly basis, our unit of analysis is the hotel quarter, i.e.,
the quarterly information for hotels in terms of reviews,
competitors, etc.
In order to determine the extent of competition faced by a
hotel from Airbnb, we wrote Python programs to collect all
Airbnb listings from the cities identified (parts of the data,
such as historical prices, are obtained from AirDNA.co). We
recorded the location and the host’s registration information
on 14,922 distinct listings on Airbnb’s website. Following
prior research (e.g., Zervas et al., 2017), we used the host’s
registration date as the time a listing became available.
To match hotels with competing Airbnb listings, we
calculated the distance between them based on the latitude and
longitude of each hotel (available on Expedia.com) and that of
each Airbnb listing (available in the HTML file for each
listing on Airbnb.com). In accordance with Mayzlin et al.
(2014), we identified competing properties (hotels as well as
Airbnb listings) as those within a 1-kilometer radius of a focal
different markets are not available to us. But the results are robust to change
of cutoff points for high-end versus low-end hotels
12
https://comptroller.texas.gov/transparency/open-data/search-datasets/
(last accessed December 11, 2021).
Nie et al. / Competing with the Sharing Economy
1580
MIS Quarterly Vol. 46 No. 3 / September 2022
hotel.
13
To control the level of competition from traditional
hotels, we counted the number of competing hotels of the
same type (i.e., low-end or high-end) for the focal hotel in
each quarter. Likewise, we counted the number of distinct
listings that appeared on Airbnb by that quarter to identify the
level of competition resulting from the sharing economy. Our
data consisted of 2,188 hotels that received reviews on both
the TripAdvisor and Expedia websites. Three cities were
dropped because none of the hotels in these cities received any
reviews, leaving us with hotels from 64 cities. Hotels in 10 of
the 64 cities did not have any competing Airbnb presence.
Overall, 48% of hotels had at least one Airbnb competitor.
Among all the cities included in our data, Houston had the
largest number of hotels with 333, while McKinney had only
one hotel. Table 1 provides summary statistics for the data.
Empirical Strategy
Identifying Review Manipulation
We identified review manipulations following Mayzlin et al.’s
(2014) measure of self-promotion as the difference in the
share of 5-star reviews on TripAdvisor (TA) and Expedia
(EXP), respectively, for hotel i in year-quarter t:
Similarly, they measure demotion as the difference in the share
of 1-star and 2-star reviews on TripAdvisor and Expedia:
One nuance is that in our sample, the average TripAdvisor
(Expedia) rating was 3.0 (3.1) for low-end hotels and 4.0 (4.1)
for high-end hotels. A 4-star rating could still promote an
average low-end hotel but not an average high-end hotel.
Therefore, the aforementioned measures of self-promotion and
demotion are reasonable proxies of review manipulation for
high-end hotels. For low-end hotels, however, we considered
both 4-star and 5-star ratings as potential promotions and 1-star
ratings as demotions. Thus, for a low-end hotel:
13
In addition to using a fixed distance threshold, we also report in Appendix
E the results of a robustness check where we use Gaussian kernels to model
a gradual decrease in competition intensity as distance grows.
Note that this demotion measure popularized by Mayzlin et al.
(2014) only captures how a focal hotel is demoted. This
measure is passive, in that a focal hotel does not directly control
whether or how competing hotels demote it. In contrast, our
interest is in understanding the review manipulation strategy in
which a focal hotel actively engages. We thus propose and
construct a measure to capture the level at which a focal hotel
demotes other competing hotels as follows: (1) identify all the
competitors for a focal hotel from the same category; (2) for
each competitor, calculate the number of demoting reviews it
receives (Demotion
it
×Total Reviews
it
TA
); (3) attribute the
demoting reviews evenly to the competitor’s other competitors
(of which the focal hotel is one);
14
and (4) take the average of
the demoting reviews on competing hotels that are attributed to
the focal hotel. We refer to this measure of review manipulation
as DemotingOthers
it
. Therefore, this measure differs from the
demotion measure in Mayzlin et al. (2014) in that they capture
the level of demotion received by a focal hotel instead of
demotion activities pursued by the focal hotel.
At the aggregate level, TripAdvisor ratings increased while
Expedia average ratings remained relatively flat during our
observation period (Figure A1 in Appendix A). When
differentiating hotels, we found that low-end hotels have
relatively stable proportions of negative and positive ratings
(relatively flat over time), while high-end hotels exhibit
increasing proportions of positive ratings but decreasing
proportions of negative ratings on TripAdvisor (Figure A2
in Appendix A).
Econometrics Analyses
To identify the impact of Airbnb on hotels’ review
manipulation behavior, we exploited the variability in the
number of competing Airbnb listings with respect to each
focal hotel (we refer to the number of competing Airbnb
listings as Airbnb supply).
14
We also conduct a robustness check by attributing the demoting reviews
based on hotels’ proximity (in terms of geographical distance). The results,
discussed in Appendix E, are qualitatively similar.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1581
Table 1. Summary Statistics at the Hotel-Quarter Level
Mean
Standard
deviation
Min
Max
Number of TripAdvisor reviews per quarter
10.89
18.00
1
509
Number of TripAdvisor 1-star reviews per quarter
0.71
1.43
0
31
Number of TripAdvisor 2-star reviews per quarter
0.76
1.51
0
28
Number of TripAdvisor 3-star reviews per quarter
1.47
2.58
0
61
Number of TripAdvisor 4-star reviews per quarter
2.97
4.96
0
98
Number of TripAdvisor 5-star reviews per quarter
4.98
10.91
0
373
Number of Expedia reviews per quarter
19.71
25.98
1
441
Number of Expedia 1-star reviews per quarter
1.16
3.23
0
92
Number of Expedia 2-star reviews per quarter
1.43
2.87
0
65
Number of Expedia 3-star reviews per quarter
2.83
4.70
0
109
Number of Expedia 4-star reviews per quarter
6.08
8.49
0
142
Number of Expedia 5-star reviews per quarter
8.21
12.96
0
287
Number of competing Airbnb listings per quarter
3.31
19.47
0
411
Number of competing hotels per quarter
4.97
7.24
0
48
Total number of hotels
2,188
Total number of hotel-quarter observations
38,759
Specifically, we asked whether the difference in review
distributions between TripAdvisor and Expedia increased (or
decreased) for a hotel, given an increase (or decrease) in the
nearby Airbnb supply (i.e., within a specified radius). We
estimate:
The dependent variable ReviewManipulation
it
represents
either SelfPromotion
it
or DemotingOthers
it
for hotel i in
quarter t. We begin with the simplest specification in Model
1, which only considers the competitive environment of a
hotel by including the log of competing Airbnb supply and
the log of competing hotels, i.e., log(Airbnb
i,t-1
) and
log(CompetingHotels
i,t-1
).
15
It is important to note that our
model specification uses two-way fixed effects that include
both individual hotel-specific and time-specific (year-
quarter) effects, as opposed to the cross-sectional estimation
of Mayzlin et al. (2014). City-specific seasonality might be
a confounding factor since it may impact both the Airbnb
supply and hotels’ review manipulation intensities.
Therefore, we introduced controls for city-specific
seasonality City
i
×Quarter
t
in the model. Incorporating these
controls ensured that seasonal events in different cities, such
15
Since the variables Airbnb and CompetingHotels could have values equal
to zero, we add one to them before the log transformation.
as the South by Southwest festival in Austin in the spring
and the Texas State Fair in Dallas in the fall, did not bias our
estimation. Moreover, we mitigated simultaneity bias in the
panel model by measuring the Airbnb supply prior to the
measurement of self-promotion and demotion behaviors.
Our identification strategy is very similar to the difference-
in-differences (DID) strategies used in Zervas et al. (2017)
and Mayzlin et al. (2014). In our paper, the differences were
implicitly and explicitly taken in three ways. The difference
between TripAdvisor and Expedia is explicit, as in Mayzlin
et al. (2014). This difference allowed us to control for
unobservable qualities or popularity changes in hotels. We
defined treated hotels as those hotels with nearby competing
Airbnb listings, and untreated (control) hotels as those with
no nearby Airbnb listings (as discussed in the Look-Ahead
Propensity Score Matching subsection, an exception is the
LA-PSM analysis, where control hotels eventually get
competing Airbnb listings). The difference between treated
and untreated hotels is measured implicitly because of the
hotel-fixed effects, which account for time-invariant
differences in review manipulation between treated and
untreated hotels. Pre-treatment and post-treatment
differences are also measured implicitly over time using
year-quarter fixed effects, which allow for unobserved time-
varying manipulation differences that are common across
different hotels.
16
16
We validate the pre-treatment parallel-trend assumption in Appendix B.
Nie et al. / Competing with the Sharing Economy
1582
MIS Quarterly Vol. 46 No. 3 / September 2022
Since reviews in the previous period might impact self-
promotion in the current period (Luca & Zervas, 2016), we also
added controls for the reviews received in the previous period.
In Model 2, we introduce controls that include the review ratios
(ratios of 2-, 3-, 4-, and 5-star reviews) and review counts on
TripAdvisor (controlling for review ratios and review counts
on Expedia leads to qualitatively similar results).
We then investigated whether the impact of Airbnb on review
manipulation is moderated by the type of hotel. Model 3
introduces the interaction between Airbnb supply and hotel
types to capture the differential impact that Airbnb supply might
exert on different types of hotels, i.e., high-end vs. low-end.
We should reiterate that the data used in Mayzlin et al. (2014)
are cross-sectional while ours are panel data. It was
unnecessary to include other controls on hotel characteristics
used by Mayzlin et al., such as the official star categorization
and location dummies (airport, interstate, resort, small-
metro/town, suburban, urban), because these time-invariant
features are subsumed by our hotel-specific fixed effects.
We controlled for the number of competing hotels,
log(CompetingHotels) that might influence a focal hotel’s
review manipulation decisions, which are time-varying. The
main coefficient of interest β
1
reflects the change in review
manipulation in response to a change in competing Airbnb
supply, and β
4
indicates the differential impact that Airbnb has
on different types of hotels.
Addressing Endogeneity
There are several sources of endogeneity that could potentially
bias the estimation results—for example, omitted (and
relevant) variables. Although we controlled for two-way fixed
effects and city seasonality, there may still be unobserved
characteristics that could impact the hotel review
manipulation level. The Airbnb supply could become
endogenous if those unobservable characteristics also
correlate with the Airbnb supply. One example is the
17
We thank an anonymous reviewer for suggesting this test to us.
advertising budget of hotels, which was unobservable to us.
Along with review manipulation, hotels might use their
advertising budget to promote themselves and demote their
competitors through advertising on traditional channels. If the
advertising budget correlates with the Airbnb supply, then the
estimation on the impact of Airbnb supply might be biased.
In our context, it would be virtually impossible to tease out
causal effects by running field experiments, because that
would require coordinating the level of Airbnb supply across
thousands of hotels. As a result, we first considered an
instrumental variable (IV) approach. A desirable IV should
be strongly correlated with the endogenous variable but must
not be related to the hotel manipulation level in unobserved
ways (i.e., through the error term). We identified the
following IV for a focal hotel’s competing Airbnb supply—
the level of competing Airbnb supply for the focal hotel’s
competing hotels—which is defined as the distinct number
of Airbnb listings for competing hotels, after excluding the
competing Airbnb listings of the focal hotel. On the one
hand, because the focal hotel’s competing Airbnb listings
were removed when constructing this instrumental variable
for the Airbnb supply, the instrument is unlikely to impact
the focal hotel’s manipulation level. On the other hand, it
should be highly correlated with the competing Airbnb
supply of the focal hotel since they are in close proximity.
Similar Hausman-type of IVs have been used to address
endogeneity concerns (see, e.g., Bardhan et al., 2015).
We discern the strength of this IV based on the first stage least
squares regression of the two-stage least square analysis (2SLS)
using the Kleibergen-Paap (KP) F-statistic (Kleibergen & Paap,
2006). The KP F-statistic is 796.51, which is greater than the
critical value (16.38) for the Stock-Yogo weak identification
test at the 10% maximal IV relative bias (Stock & Yogo, 2005).
While the exclusion restriction for the IV is untestable, there are
some tests that can be performed to indirectly test the validity
of the IV. In the Validity of the Instrument subsection, we
provide further support for the instrument by running the test
proposed by Barron et al. (2020). The test ensures that there is
no correlation between the IV and the dependent variable in
locations without Airbnb.
17
Another relevant source of potential endogeneity could be self-
selection: the treated and untreated hotels might be intrinsically
different if they self-select into review manipulation levels. To
alleviate this concern, we used look-ahead propensity score
matching (LA-PSM) (Bapna et al., 2018) to balance the treated
and untreated hotels. The details for how LA-PSM was
implemented and the corresponding results are provided in the
Look-Ahead Propensity Score Matching subsection.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1583
We further address endogeneity by constructing synthetic
controls for each treated hotel using generalized synthetic
control (GSC) estimators (Xu, 2017). GSC applies to
situations with multiple treated units and differential treatment
timing, and it works even when the parallel-trend assumption
is violated in difference-in-differences settings. The results of
GSC analyses are reported in a later subsection.
Simultaneity could also cause endogeneity if Airbnb hosts
were to base their decision to list their properties on hotels’
review manipulation levels. This is unlikely for two reasons.
First, it is very difficult for Airbnb hosts and guests to observe
the review manipulation levels of neighboring hotels (short of
conducting analyses, as in this paper). Second, we used a one-
period lag in the Airbnb supply in all estimation models to
capture the potentially delayed impact of Airbnb. Therefore,
simultaneity is less of a concern in our context. Nevertheless,
as discussed above, our IV estimation approach further
alleviates this concern.
Results
We present the estimation results of review manipulation in
response to the number of Airbnb listings. Specifically, we
analyze Airbnb’s impact on hotels’ review manipulations in
terms of self-promotion and demotion activities.
18
Self-Promotion
Table 2 provides the results of 2SLS estimations for self-
promotion. Since we used the panel data for estimating our
specification, serial correlation may be present. To account for
potential autocorrelation of review manipulation across time,
we followed standard practice in clustering standard errors at
the hotel level (Bertrand et al., 2004; Sun & Zhu, 2013) for all
analyses.
Model 1 in Table 2 shows that estimated β
1
is 0.021, meaning
that a 1% increase in nearby Airbnb listings is associated with a
statistically significant increase of 0.021 percentage points (p <
0.001) in self-promotion. It implies that the impact of Airbnb
over the five-year period of 2011-2015 would lead to an
increase of 5.6 percentage points difference in the share of
positive reviews across TripAdvisor and Expedia. This
calculation is based on the increase of Airbnb supply from an
average of 0.554 competing listings in Quarter 1 of 2011 to an
average of 9.187 listings in Quarter 4 of 2015, which implies an
18
SUR (seemingly unrelated regressions) models may seem more
appropriate at first glance for the estimation since we have one equation for
self-promotion and one equation for demotion activities. However, as noted
increase in magnitude of log(9.187/0.554) × 0.021 = 5.6%.
Model 2 incorporates the additional control variables review
ratios and review counts. The estimate for the impact of Airbnb
remains qualitatively unchanged. Thus, our results suggest that
the competition from Airbnb offerings drives hotels to self-
promote more in general.
A follow-up question is whether there are heterogeneous
impacts of Airbnb supply on different hotel categories.
Revenues of low-end hotels have been found to decrease more
than those of high-end hotels after the entry of Airbnb (Zervas
et al., 2017); therefore, low-end hotels would be expected to
engage in more self-promotion because of the intensified
across-group competition from Airbnb. However, this was not
the case in our study. Column 3 provides estimates for Model
3 using the high-end hotel category as the reference level. A
1% increase in Airbnb competition drives high-end hotels to
increase self-promotion by 0.020 percentage points (estimated
β
1
is 0.020 in Table 2 Model 3), and the impact is statistically
significant. What is interesting is that Airbnb’s impacts on
low-end hotels and high-end hotels are indeed different. As
shown in Model 3, even though Airbnb supply drives high-
end hotels to increase self-promotion, its impact on low-end
hotels is weaker, as indicated by the significantly negative
estimate on the interaction term (estimated β
4
is -0.054). To
verify whether Airbnb has a significant net impact on low-end
hotels, we also estimated Model 3 considering the low-end
hotel category as the reference level; the results show that the
net impact is statistically insignificant. It indicates that Airbnb
supply does not drive low-end hotels to increase self-
promotion. Taken together, with an increase in Airbnb supply,
high-end hotels engage in significantly more self-promotion
while low-end hotels do not increase self-promotion.
Our finding that high-end hotels engage in more self-promotion
appears to be counterintuitive at first glance. However, this
finding may not be surprising if we consider that online reviews
may be of greater strategic importance for high-end hotels than
low-end hotels. Lewis and Zervas (2019) show that the average
rating of online reviews is more important for upscale to luxury
hotels. This is also reflected in how hotels respond to reviews:
hotel management teams frequently utilize management
responses on online review platforms to directly respond to
online reviews, especially the negative ones (Proserpio &
Zervas, 2017; Yang et al., 2019). By looking at the percentages
of reviews receiving responses from the hotels, we found that
low-end hotels respond to only 4.1% of Expedia reviews while
high-end hotels respond to 6.8% of such reviews (the difference
is significant at the 1% level).
in Greene (2008, p. 257), an SUR model is equivalent to equation by
equation regressions when the equations have identical explanatory
variables as is the case in our specifications.
Nie et al. / Competing with the Sharing Economy
1584
MIS Quarterly Vol. 46 No. 3 / September 2022
Table 2. Self-Promotion with 2SLS
Model 1
Model 2
Model 3
log(Airbnb)
0.021
***
0.020
**
0.020
***
(0.006)
(0.006)
(0.006)
log(CompetingHotels)
0.050
0.049
0.043
(0.031)
(0.030)
(0.030)
Log(ReviewCount)
0.005
0.004
(0.006)
(0.006)
log(Airbnb) × Low-end
-0.054
*
(0.022)
ReviewRatios
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
Time-fixed effects
YES
YES
YES
Observations
32,122
32,122
32,122
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at the hotel level.)
Further, we found that low-end hotels receive significantly
fewer reviews, on average, on both Expedia and TripAdvisor,
compared to high-end hotels. Along these lines, Lewis and
Zervas (2019) argue that the importance of online review
ratings is larger for high-end hotels because their customers
have higher expectations. Our finding suggests that high-end
hotels tend to utilize the “differentiation” strategy through
increasing self-promotion activities in the face of across-
group competition from Airbnb.
Demoting Competitors
We next turn to the specification where the dependent variable
is DemotingOthers. We conduct 2SLS with the same
instrumental variable described earlier and present the results
in Table 3.
Models 1 and 2 in Table 3 show that when the competing
Airbnb supply increases, hotels reduce demotion activities.
Model 3 includes the interaction terms for hotel categories,
showing the differential impact of Airbnb using high-end
hotels as the reference level. The levels of demotion behaviors
for high-end hotels are qualitatively similar to the previous
two models. A 10% increase in Airbnb supply would drive
high-end hotels to significantly decrease demotion activities
by 0.0035 (estimated β
1
is 0.035 in Model 3). Considering that
the mean of demotion activities is 0.189 per year-quarter, the
decrease in the demotion activities that high-end hotels engage
in against competing hotels amounts to about 1.85% each
quarter because of the impact of Airbnb. Since the estimated
coefficient for the interaction term log(Airbnb) × Low-end is
insignificant, it shows that Airbnb influences high-end and
low-end hotels similarly. Therefore, this suggests that both
high-end and low-end hotels decrease their demotion
behaviors as Airbnb supply increases.
Mayzlin et al. (2014) found that the presence of more
competing hotels leads the focal hotel to receive more fake
demoting reviews on TripAdvisor. Also, Luca and Zervas
(2016) reported that more competition (from other restaurants)
leads to more fake negative reviews in the restaurant industry as
well. One may similarly expect that the intensified competition
resulting from Airbnb supply would lead hotels to demote their
competitors even more. But contrary to those findings, our
results show that intensified competition from Airbnb actually
led to fewer demoting reviews for high-end hotels. This
contrasting finding lends support to our earlier argument that
the nature of sharing economy competition is different. Our
findings align with the predictions of strategic group theory
(Caves & Porter, 1977): rivalry is lower within a group because
firms are better able to recognize their mutual dependence and
thus cooperate or tacitly collude with one another when facing
competition from another strategic group or a common enemy
like Airbnb, particularly because the tit-for-tat competitive
interactions can be destructive for the whole group (Smith et al.,
1997). Accordingly, it may not help hotels to demote each other
in platforms like TripAdvisor given that customers have the
option of Airbnb rentals. This finding indirectly echoes the
“differentiation” strategy that may be attributed to self-
promotion: by decreasing demotion activities, high-end hotels
can indirectly boost their online ratings in a relative manner as
a result of potentially reduced retaliation.
Mutual Demotion across Hotel Groups
Our main analyses used the behavior of individual hotels as the
unit of analysis. In this section, we investigate whether the mutual
demotion activities across groups of hotels decreased as Airbnb
gained popularity, along the lines predicted by strategic group
theory. Specifically, we look at hotel pairs and triplets. We
present the analyses for hotel pairs here; the results for hotel
triplets, presented in Appendix C, are qualitatively the same.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1585
Table 3. Demotion Behavior with 2SLS
Model 1
Model 2
Model 3
log(Airbnb)
-0.035
***
-0.035
***
-0.035
***
(0.009)
(0.009)
(0.009)
log(CompetingHotels)
-0.297
***
-0.294
***
-0.295
***
(0.044)
(0.044)
(0.044)
Log(ReviewCount)
-0.013
-0.013
(0.010)
(0.010)
log(Airbnb) × Low-end
-0.006
(0.049)
ReviewRatios
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
Time-fixed effects
YES
YES
YES
Observations
24,713
24,713
24,713
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
The Identifying Review Manipulation subsection introduced
the procedure to attribute demoting reviews to competing
hotels. For each pair of hotels, we calculated the intensity of
mutual demotion activities by computing the number of
demoting reviews that the two hotels likely contributed to
each other. For the hotel pair (A, B) the mutual demotion level
is calculated as: MutualDemoting(A,B) = Demoting(A→B) +
Demoting(B→A).
When the unit of analysis is a hotel pair (two hotels) instead
of an individual hotel, we needed to aggregate competing
Airbnb supply and competing hotels for the hotel pair in a
consistent manner. For example, to measure the intensity of
Airbnb competition faced by a hotel pair (A, B), we
considered the intersection of the Airbnb listings competing
with Hotel A and the listings competing with Hotel B. The
cardinality of the intersection is used as a measure of the
intensity of Airbnb competition for the hotel pair. Similarly,
we measured the intensity of competition arising from
traditional hotels as the cardinality of the intersection of the
two sets of competing hotels for A and B, respectively.
19
The results are reported in Table 4. We found that high-end
hotels decrease mutual demotion activities, given an increase
in competing Airbnb supply, while low-end hotels do not
significantly change their demotion behaviors. For high-end
hotels, both Table 3 and Table 4 show that the reductions in
demotion activities are consistent. However, for low-end
hotels, Table 3 shows they would reduce demotion activities
while Table 4 suggests they would not significantly change
19
The results are similar if we use the union instead of the intersection to
aggregate the set of competing Airbnb listings and competing hotels.
demotion activities. A conservative interpretation, then, is that
low-end hotels do not increase demotion activities in response
to Airbnb competition. The difference between high-end and
low-end hotels might be explained by the fact that the reviews
are of more strategic importance for high-end hotels as
discussed in the Self-Promotion subsection.
There may be a concern that our findings could be the result
of a customer shift, e.g., some hotel customers who used to
give low ratings may have shifted to Airbnb listings. This
putative shift in customer types could then lead to decreased
demotion activities for hotels. However, it is unlikely that this
alternative explanation is driving our results. First, our
operationalization of review manipulation involves
computing the difference in ratings between the two platforms
Expedia and TripAdvisor. There is no reason to believe that
the reduction in poor ratings from a shift in customer
preferences would occur on one platform but not the other.
Second, we examined whether the travel types of customers
changed after the emergence of Airbnb and did not find any
significant change during our observation period. This
analysis is discussed in Appendix D.
Robustness Checks
We reinforced the causality inferences and robustness of our
results by conducting analyses using several alternative
approaches.
Nie et al. / Competing with the Sharing Economy
1586
MIS Quarterly Vol. 46 No. 3 / September 2022
Table 4. Mutual Demotion Behavior with 2SLS
Model 1
Model 2
Model 3
log(Airbnb)
-0.107
***
-0.092
***
-0.092
***
(0.008)
(0.008)
(0.008)
log(CompetingHotels)
-0.370
***
-0.357
***
-0.348
***
(0.030)
(0.029)
(0.029)
Log(ReviewCount)
0.050
***
0.050
***
(0.005)
(0.005)
log(Airbnb) × Low-end
0.136
***
(0.040)
ReviewRatios
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
Time-fixed effects
YES
YES
YES
Observations
104,424
104,424
104,424
Note:
*p <
0.05,
**p <
0.01,
***p <
0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
Validity of the Instrument
If the instrument is valid, then it should only correlate with the
review manipulation level through its effect on Airbnb
competitors. Therefore, for hotels with no Airbnb competitors,
we should not see a significant relationship between the IV
and the hotel’s review manipulation level. To test this, we
regressed the review manipulation level on the instrument
directly, using only data for hotels where no Airbnb
competitors were observed. The first two columns of Table 5
report the results of these regressions. They show that
conditional on the two-sided fixed effects and controls, there
is no statistically significant relationship between the IV
(labeled as log(Cmp_Airbnb)) and the review manipulation in
hotels without Airbnb competitors. By contrast, Columns 3
and 4 of Table 5 show that if we regress the review
manipulation level directly on the instrument for hotels with
Airbnb competitors, there is a statistically significant
relationship between the instrument and the self-promotion
(demoting) level.
It may be possible that hotels with versus without Airbnb
competitors may be fundamentally different. We therefore
constructed a third sample of hotels with Airbnb competitors
that are very similar to hotels without Airbnb competitors. To
do so, along the lines proposed by Barron et al. (2020), we
used propensity score matching (PSM) to match the treated
hotels with untreated ones based on observable measures (the
measures are detailed in the Look-Ahead Propensity Score
Matching subsection). Columns 5 and 6 of Table 5 report the
results when the review manipulation level is directly
regressed on the instrument in the propensity score-matched
sample with Airbnb competitors. The direct effect of the
instrument is statistically significant, alleviating concerns that
the null effect of the instrument in the non-Airbnb sample is
only because hotels without Airbnb competitors are
fundamentally different from hotels with Airbnb competitors.
Look-Ahead Propensity Score Matching (LA-
PSM)
Since the treatments of Airbnb supply to hotels may not be
assigned randomly (as in a controlled experiment), our
estimations may be subject to systematic differences between
the treated hotels and untreated ones. To alleviate this concern,
we used the LA-PSM method (Bapna et al., 2018). A limitation
of a standard PSM is that it only accounts for observed
measures. It fails if the control group and the treatment group
are systematically different in unobserved measures. In
contrast, LA-PSM suggests using as the control group those
hotels which are currently untreated but will become treated in
the future. Thus, by construction, LA-PSM matches treated and
untreated groups that share the same unobserved time-constant
characteristics that may cause hotels to become treated. As a
result, LA-PSM can account not just for the observed
characteristics in the matching procedure but also for
unobserved characteristics in our panel data.
We determined the treatment and control groups as follows.
As shown in Figure 1, there are three groups of hotels based
on whether they have Airbnb competitors, and if so when they
start to have such competitors. We divided the time horizon
into two periods. Group A denotes the set of hotels that have
Airbnb competitors in both Periods 1 and 2, Group B hotels
that have no Airbnb competitors in Period 1 but have Airbnb
competitors in Period 2, and Group C hotels that have no
Airbnb competitors in both Periods 1 and 2. In regular PSM,
hotels in either Group A or Group B are considered treated
hotels and are matched with untreated hotels in Group C. In
LA-PSM, only hotels in Group A are considered treated hotels
and are matched with hotels in Group B.
To make the number of samples between treated hotels and
their matched hotels relatively balanced, we divided the 32
quarters into the first two thirds and the last third.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1587
Table 5. IV Validity Check: Correlation Between Instrument and Review Manipulation
Sample: Hotels
w/o Airbnb ever
Sample: Hotels
w/ Airbnb
Sample: PSM sample
w/ Airbnb
(1)
(2)
(3)
(4)
(5)
(6)
Promote
Demote
Promote
Demote
Promote
Demote
log(Cmp_Airbnb)
0.004
-0.009
0.012
**
-0.032
***
0.013
*
-0.038
**
(0.019)
(0.015)
(0.004)
(0.010)
(0.006)
(0.012)
log(CompetingHotels)
0.030
-0.209
***
0.062
-0.357
***
0.043
-0.283
***
(0.044)
(0.059)
(0.041)
(0.064)
(0.049)
(0.069)
Log(ReviewCount)
0.009
0.017
-0.001
-0.033
*
-0.013
-0.037
(0.009)
(0.014)
(0.008)
(0.013)
(0.009)
(0.019)
ReviewRatios
YES
YES
YES
YES
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
15,073
11,279
17,117
13,492
10,772
8,640
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
To calculate the propensity score (predicted probability of
being treated), we used observable characteristics including
price, ownership (independent or chain), whether a hotel has
a restaurant, location type (airport, interstate, resort, small-
metro/town, suburban, urban), and the number of rooms.
Some of the characteristics are categorical variables such as
a hotel’s ownership status while others are continuous-
valued variables such as the price. We used exact matching
on the categorical variables and nearest-neighbor matching
on the continuous ones as it provided better matching quality
compared to using just one method. We grouped hotels that
had an exact match on the categorical variables into subsets,
and then matched the treated hotels with untreated ones that
had the closest propensity score within each subset (without
replacement). We found a good match of control hotels
(untreated in Period 1 and treated in Period 2) for 307 of the
411 hotels which had competing Airbnb listings in Period 1.
Because both the treatment and the matched control groups
ultimately were treated (at different time periods), the
matching procedure accounts not only for the observed
measures using propensity scores, but also for unobserved
time-invariant characteristics influencing the hotel’s
intrinsic propensity to be treated. The results, as presented in
Table 6, remain consistent with the main results in the paper.
In addition, the results are robust to alternative matching
methods (coarsened exact matching, exact matching, nearest
neighbor matching) and to the inclusion of other features for
the matching (such as hotel average rating, standard
deviation of rating, and the number of ratings).
20
20
We also conducted a traditional PSM analysis. The results from using
PSM and LA-PSM are qualitatively the same.
Generalized Synthetic Control
Besides LA-PSM, an alternative method that can help address the
endogeneity concern is generalized synthetic control (GSC) (Xu,
2017). The challenge for causal inference is to come up with a
credible estimate of what the outcome would have been for the
treatment group in the absence of the treatment. This requires
estimating a counterfactual change over time for the treatment
group had the treatment not occurred. Instead of using a single
control unit or a simple average of control units, the synthetic
control approach proposed by Abadie et al. (2015) constructs a
control by using a weighted average of a set of controls (thus the
name synthetic). The synthetic control approach “is arguably the
most important innovation in the policy evaluation literature in
the last 15 years” (Athey & Imbens, 2017, p. 9). The traditional
synthetic control approach applies to the case of one treated unit
with a unique treatment time. GSC extends the traditional
approach by allowing multiple treated units as well as differential
treatment timing as is the case in our data.
GSC had been proposed in the traditional binary treatment
context. Therefore, we needed a cutoff point to dichotomize the
Airbnb supply. At the midpoint of the time span of our study
(2011 Quarter 4), the mean Airbnb supply for hotels that face
Airbnb competition is 5.21. We used this number as the threshold
above which hotels were considered to be treated. This also
ensured that we were left with sufficient observations with at least
ten pre-treatment periods, as suggested by Xu (2017). Because
GSC cannot handle time-invariant variables like hotel tiers, we
conducted subsample analyses (i.e., analyzing high-end hotels
and low-end hotels separately) to investigate how high-end and
low-end hotels manipulate reviews differently.
Nie et al. / Competing with the Sharing Economy
1588
MIS Quarterly Vol. 46 No. 3 / September 2022
Note: The shaded left half oval represents the hotels that start facing Airbnb competition in Period 1 (Group A). The shaded right half oval
represents the hotels that start facing Airbnb competition in Period 2 (Group B). Group C denotes the hotels which do not face Airbnb
competition in either Period 1 or Period 2.
Figure 1. Treatment and Control Groups in LA-PSM
Table 6. LA-PSM-Matched Hotels
Self-promotion
Demoting others
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
log(Airbnb)
0.034
*
0.031
*
0.030
*
-0.084
**
-0.086
**
-0.087
**
(0.014)
(0.014)
(0.014)
(0.028)
(0.029)
(0.029)
log(CompetingHotels)
0.058
0.054
0.044
-0.309
***
-0.303
***
-0.300
***
(0.050)
(0.048)
(0.049)
(0.077)
(0.076)
(0.076)
Log(ReviewCount)
-0.011
-0.012
-0.044
*
-0.044
*
(0.010)
(0.010)
(0.021)
(0.021)
log(Airbnb) × Low-end
-0.044
0.017
(0.023)
(0.058)
ReviewRatios
NO
YES
YES
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
10,759
10,759
10,759
8,630
8,630
8,630
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
As shown in Table 7, the analyses yielded results consistent
with the main results in the paper. High-end hotels
significantly increase self-promotion activities in response to
Airbnb while significantly reducing demotion activities. Low-
end hotels do not show any significant change in review
manipulation activities in response to Airbnb.
Being Demoted
Since we wanted to understand the active role that hotels
engaged in with respect to demotion, we analyzed the
magnitude for demoting others in our analyses. Mayzlin et al.
(2014) used the being-demoted measure to see how much
hotels were demoted by their competitors. Although it is
orthogonal to the focus of our paper, we nevertheless explore
whether our results are consistent with this alternative way of
viewing demotions.
We computed the number of fake reviews that a hotel received
using the formula Demoted
it
= Demotion
it
× Total Reviews
it
TA
. We
use the measure Demoted
it
to evaluate how much a hotel had been
demoted by its competitors. The results of using the Demoted
it
variable as the dependent variable are shown in Table 8.
As shown in Model 3 of Table 8, high-end hotels are demoted
more and the impact on low-end hotels is not statistically
significant. These findings appear to be contradictory at first
sight. Our main analyses and the several robustness checks
show that high-end hotels decrease demotion activities with
an increase in Airbnb supply, while the results in Table 8 show
that high-end hotels are demoted more. Interestingly, as
explained below, there is no contradiction.
Imagine that Hotels A and B are competing with each other.
Further, let us assume that Hotel A has more competing Airbnb
listings than Hotel B. Based on our analysis, Hotel A is likely to
engage in fewer demotion activities compared to Hotel B.
Period 1 Period 2
A
B
C
A
B
C
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1589
Table 7. Generalized Synthetic Control Estimators
Self-promotion
Demoting others
All
High-end
Low-end
All
High-end
Low-end
Airbnb Treatment
0.029
*
0.033
*
-0.003
-0.034
-0.037
*
0.192
(0.013)
(0.015)
(0.073)
(0.022)
(0.019)
(0.175)
log(CompetingHotels)
0.011
0.017
0.450
-0.434
***
-0.436
***
-0.316
(0.028)
(0.028)
(0.533)
(0.047)
(0.045)
(0.500)
Log(ReviewCount)
0.028
***
0.032
***
0.016
-0.005
-0.005
-0.001
(0.009)
(0.008)
(0.026)
(0.010)
(0.009)
(0.047)
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Standard errors in parentheses.
Table 8. Being demoted with 2SLS
Model 1
Model 2
Model 3
log(Airbnb)
0.123
*
0.137
**
0.139
**
(0.048)
(0.048)
(0.048)
log(CompetingHotels)
-0.145
-0.160
-0.176
(0.122)
(0.121)
(0.122)
Log(ReviewCount)
0.171
***
0.170
***
(0.027)
(0.027)
log(Airbnb) × Low-end
-0.161
(0.174)
ReviewRatios
NO
YES
YES
Hotel Fixed Effects
YES
YES
YES
Time Fixed Effects
YES
YES
YES
Observations
32,122
32,122
32,122
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
This, in turn, implies that Hotel B will receive fewer
demotions as compared to Hotel A. This is consistent with the
findings in Table 8. The signs of the estimated coefficients
differ because we are looking at the same phenomenon
through the opposite lens.
Other Robustness Checks
We have conducted several additional analyses to ensure that
our results are robust. We provide a summary of those
analyses here and present the details in the Appendices.
Thus far, we have assumed that the competition between and
across hotels and Airbnb listings is primarily constrained
within a 1-kilometer radius with respect to each focal hotel.
Competing hotels located farther away or nearer may have
varying degrees of incentives to demote their competitors. In
Appendix E, we consider alternatives with different
competition radii and also relax the restrictions of requiring a
fixed competition radius and of considering all competing
hotels equally. We build on the idea of Luca and Zervas
(2016) to model a gradual decrease in competition intensity
due to increased distance. The results are consistent with the
main results.
As mentioned in the Identifying Review Manipulation
subsection, because of the difference in average ratings for
high-end and low-end hotels, we measured self-promotion and
demotion differently for high-end and low-end hotels. We also
replicated our experiments by considering the same measures
for both high-end and low-end hotels following Mayzlin et al.
(2014). The results are presented in Appendix F.
Airbnb listings vary considerably in terms of their quality and
price. In the main analyses, we considered all nearby Airbnb
listings to be competing with a focal hotel regardless of the
nature (i.e., price and quality) of those listings. In Appendix
G, we validated that our estimation results are not sensitive to
this assumption.
In our main analyses, we assumed that hotels are only
competing with other hotels from the same category. In
Appendix H, we relaxed this assumption by considering hotels
as competitors regardless of their categories.
To further reinforce our findings, we used an alternative
approach to investigate how Airbnb listings influence hotels’
review manipulation behaviors differently. In Appendix I, we
demonstrate that the growth of Airbnb did change how
incumbent hotels respond to competing hotels. In sum, the
Nie et al. / Competing with the Sharing Economy
1590
MIS Quarterly Vol. 46 No. 3 / September 2022
main results presented are robust to many alternative methods
of testing them and Airbnb competition does indeed change
hotels’ review manipulation behavior.
Discussion and Conclusions
An interesting outcome of the rise of the digital economy is
the emergence of sharing economy firms that use technology-
mediated platforms to compete with traditional firms. We use
strategic group theory, a central construct in the strategy
literature, to examine how the competition from such novel
digital enterprises affects the strategic actions of incumbent
firms. Because such firms differ fundamentally from
traditional firms in terms of their asset bases and business
models, they do not constitute rivals that can be considered
part of the strategic groups of incumbents. This raises an
interesting question—namely, how the emergence of a new
strategic group can impact rivalry across firms within existing
traditional groups.
We specifically examined how one type of firms’ strategic
communications—review manipulations—is impacted by the
emergence of sharing economy firms. Previous literature has
found that the emergence of Airbnb has intensified the
competition for customers among hotels, and that hotels
manipulate reviews in response to increased competition. Based
on these findings, one might conclude that the problem of online
review manipulation would worsen across the board after
Airbnb gains popularity. Our work provides some surprising
findings regarding incumbent firms’ manipulation strategies in
response to competition from sharing economy firms. We show
that increased Airbnb supply leads to significantly more self-
promotion activities for high-end hotels but does not lead to any
increase in self-promotion behavior for low-end hotels.
Regarding demotions, we find that increased Airbnb supply
leads to significantly less demotion behavior for high-end
hotels. Low-end hotels do not increase their demotion behavior
with increased Airbnb supply. The findings regarding
demotions are particularly surprising given the findings of
Mayzlin et al. (2014) in the conventional lodging business and
those of Luca and Zervas (2016) in the restaurant industry, both
of which suggest that intensified competition leads firms to
demote each other more. These works do not factor in the new
type of competition coming from the sharing economy.
Consistent with the prediction of strategic group theory, we find
that the impact of the sharing economy on the group of
incumbent hotels leads hotels to demote their competing hotels
less in response to Airbnb.
Our findings contribute to the literature on the sharing
economy by showing that the disruptive innovations from
sharing economy firms are changing the landscape of
competition among incumbents in unexpected ways. Our
work also contributes to the strategic group literature by
demonstrating how the entry of a new strategic group (i.e.,
Airbnb) can change the rivalry across firms within an extant
group. In our context, we show that incumbent hotels cannot
use demotion to counter the competition from Airbnb listings
as they typically do to manage competition arising from other
hotels (i.e., within their own strategic group). When this new
group enters the market, intensifying competitive interactions
within the extant group can quickly become destructive. We
find that the within-group rivalry across the hotels decreases
when a new group (Airbnb) joins the industry, as reflected by
the decrease in demotion. Thus, when between-group
competition emerges, co-opetition rather than tit-for-tat
becomes a preferable option among members within the
extant group.
Our findings have important implications for both review
hosting platforms and customers. Review-hosting platforms
like TripAdvisor can benefit from knowing that increased
Airbnb supply drives high-end hotels to increase self-
promotion while engaging in fewer demotion activities.
Consequently, these platforms can potentially adjust their
filtering algorithms to account for both the magnitude of
Airbnb supply near a hotel and the type of the hotel.
Customers need to be more circumspect in their choices of
high-end hotels in areas with high Airbnb penetration, keeping
in mind that the online review ratings may have been inflated
because of abundant Airbnb listings in the area.
A limitation of our study is that the GSC implementation used
here considers only a binary treatment scenario and we had to
dichotomize our continuous treatment to apply GSC.
Extending GSC to the setting of continuous treatment
variables would be desirable. Our work also opens up other
interesting avenues for future research. We analyzed an
important strategy, review manipulation, which incumbent
firms utilize when dealing with new forms of competition. It
would be useful to examine the joint impact of such
manipulations along with other strategies that incumbent
firms may use, such as pricing, advertising, quality, and
capacity management.
Acknowledgments
We greatly appreciate the constructive feedback from the senior
editor, the associate editor, and the three anonymous reviewers
over the several rounds of review. We also thank the conference
participants at WITS 2016, CIST 2017, and POMS 2018 for many
helpful suggestions and encouragement. The paper is
substantially improved as a result. Eric Zheng acknowledges
partial grant support from the National Natural Foundation of
China (NSFC) [Grant 71831006].
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1591
References
Abadie, A., Diamond, A., & Hainmueller, J. (2015). Comparative
politics and the synthetic control method. American Journal of
Political Science, 59(2), 495-510.
AHLA. (2017, March 9). New Study Shatters Airbnb Homesharing
Myth. https://www.ahla.com/press-release/new-study-shatters-
airbnb-homesharing-myth
Athey, S., & Imbens, G. W. (2017). The state of applied
econometrics: causality and policy evaluation. Journal of
Economic Perspectives, 31(2), 3-32.
Autor, D. H. (2003). Outsourcing at will: The contribution of unjust
dismissal doctrine to the growth of employment outsourcing.
Journal of Labor Economics, 21(1), 1-42.
Bapna, R., Ramaprasad, J., & Umyarov, A. (2018). Monetizing
freemium communities: does paying for premium increase
social engagement? MIS Quarterly, 42(3), 719-735.
Bardhan, I., Oh, J. (Cath), Zheng, Z. (Eric), & Kirksey, K. (2015).
Predictive analytics for readmission of patients with congestive
heart failure. Information Systems Research, 26(1), 19-39.
Bardhi, F., & Eckhardt, G. M. (2012). Access-based consumption:
the case of car sharing. Journal of Consumer Research, 39(4),
881-898.
Barron, K., Kung, E., & Proserpio, D. (2020). The effect of home-
sharing on house prices and rents: Evidence from Airbnb.
Marketing Science, 40(1), 23-47.
Belton, P. (2015). Navigating the potentially murky world of online
reviews. BBC News. https://www.bbc.com
/news/business-33205905
Benner, K. (2017). Inside the hotel industry’s plan to combat
Airbnb. The New York Times. https://www.
nytimes.com/2017/04/16/technology/inside-the-hotel-
industrys-plan-to-combat-airbnb.html
Bertrand, M., Duflo, E., & Mullainathan, S. (2004). How much
should we trust differences-in-differences estimates? The
Quarterly Journal of Economics, 119(1), 249-275.
Bickart, B., & Schindler, R. M. (2001). Internet forums as
influential sources of consumer information. Journal of
Interactive Marketing, 15(3), 31-40.
Brandenburger, A. M., & Nalebuff, B. J. (2011). Co-opetition.
Currency.
Bredderman, W. (2018, April 6). Hotels target Airbnb in mail and
ad blitz. Crain’s New York Business.
https://www.crainsnewyork.com/article/20180406/REAL_ES
TATE/180409933/hotels-target-airbnb-in-mail-and-ad-blitz
Caves, R. E., & Porter, M. E. (1977). From entry barriers to
mobility barriers: conjectural decisions and contrived
deterrence to new competition. The Quarterly Journal of
Economics, 91(2), 241.
Chan, J., Mojumder, P., & Ghose, A. (2019). The digital sin city:
An empirical study of Craigslist’s impact on prostitution
trends. Information Systems Research, 30(1), 219-238.
Chevalier, J. A., & Mayzlin, D. (2006). The effect of word of mouth
on sales: Online book reviews. Journal of Marketing Research,
43(3), 345-354.
Cool, K., & Dierickx, I. (1993). Rivalry, strategic groups and firm
profitability. Strategic Management Journal, 14(1), 47-59.
Cool, K., & Schendel, D. (1987). Strategic group formation and
performance: The case of the U.S. pharmaceutical industry,
1963-1982. Management Science, 33(9), 1102-1124.
Cramer, J., & Krueger, A. B. (2016). Disruptive change in the taxi
business: The case of Uber. American Economic Review,
106(5), 177-182.
Dellarocas, C. (2006). Strategic manipulation of internet opinion
forums: Implications for consumers and firms. Management
Science, 52(10), 1577-1593.
Dellarocas, C., Zhang, X. (Michael), & Awad, N. F. (2007).
Exploring the value of online product reviews in forecasting
sales: The case of motion pictures. Journal of Interactive
Marketing, 21(4), 23-45.
Eckhardt, G. M., Houston, M. B., Jiang, B., Lamberton, C.,
Rindfleisch, A., & Zervas, G. (2019). Marketing in the sharing
economy. Journal of Marketing, 83(5), 5-27.
Edelman, B., & Luca, M. (2012). Airbnb (A). Harvard Business
School Case 912-019. https://www.hbs.edu/faculty/Pages/
item.aspx?num=41242
Einhorn, B., Wei, D., Day, M., & Soper, S. (2021). Amazon hits
Chinese sellers with crackdown on fake reviews.
Bloomberg.Com. https://www.bloomberg.com/news/articles/
2021-08-18/amazon-amzn-cracks-down-on-fake-reviews-
hitting-chinese-retailers
Fiegenbaum, A., & Thomas, H. (1995). Strategic groups as
reference groups: Theory, modeling and empirical examination
of industry and competitive strategy. Strategic Management
Journal, 16(6), 461-476.
Goldstein, M. (2018). Dislocation and its discontents: Ride-
sharing’s impact on the taxi industry. Forbes.
https://www.forbes.com/sites/michaelgoldstein/2018/06/08/ub
er-lyft-taxi-drivers/
Greene, W. H. (2008). Econometric analysis (6th Edition). Pearson.
Greenwood, B. N., Ganju, K. K., & Angst, C. M. (2019). How does
the implementation of enterprise information systems affect a
professional’s mobility? An empirical study. Information
Systems Research, 30(2), 563-594.
Harmon, A. (2004). Amazon glitch unmasks war of reviewers. The
New York Times. http://www.nytimes.com/2004/02/14/us/
amazon-glitch-unmasks-war-of-reviewers.html
He, S., Hollenbeck, B., & Proserpio, D. (in press). The market for
fake reviews. Marketing Science. Available at https://doi.org/
10.1287/mksc.2022.1353
Heider, F. (1958). The psychology of interpersonal relations.
Psychology Press.
Hollenbeck, B. (2018). Online reputation mechanisms and the
decreasing value of chain affiliation. Journal of Marketing
Research, 55(5), 636-654.
Hughes, T. (2017). The effect of ride-sharing on the auto industry.
Moody’s Analytics. https://www.moodysanalytics.com/risk-
perspectives-magazine/managing-disruption/op-ed/the-effect-
of-ride-sharing-on-the-auto-industry
Jackman, S. (2009). Bayesian analysis for the social sciences. John
Wiley & Sons.
Jindal, N., & Liu, B. (2008). Opinion spam and analysis. In
Proceedings of the 2008 International Conference on Web
Search and Data Mining (pp. 219-230).
Nie et al. / Competing with the Sharing Economy
1592
MIS Quarterly Vol. 46 No. 3 / September 2022
Kelleher, S. R. (2019). The “Airbnb for campsites” just made it a
lot easier to get outside this summer. Forbes.
https://www.forbes.com/sites/suzannerowankelleher/2019/05/
08/the-airbnb-for-campsites-just-made-it-a-lot-easier-to-get-
outside-this-summer/
Kemp, D. (2017). Don’t regulate Uber, deregulate regular taxis.
Newsweek. https://www.newsweek.com/dont-regulate-uber-
deregulate-regular-taxis-673548
Kihlstrom, R. E., & Riordan, M. H. (1984). Advertising as a signal.
Journal of Political Economy, 92(3), 427-450.
Kim, M. (2018). : run strategic decisions of hotels-shortand -Long
ricingpDifferentiation and [unpublished doctoral dissertation].
Temple University. http://dx.doi.org/10.34944/dspace/3102
Kleibergen, F., & Paap, R. (2006). Generalized reduced rank tests
using the singular value decomposition. Journal of
Econometrics, 133(1), 97-126.
Kumar, N., Venugopal, D., Qiu, L., & Kumar, S. (2018). Detecting
review manipulation on online platforms with hierarchical
supervised learning. Journal of Management Information
Systems, 35(1), 350-380.
Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., & Shin, Y. (1992).
Testing the null hypothesis of stationarity against the
alternative of a unit root: How sure are we that economic time
series have a unit root? Journal of Econometrics, 54(1), 159-
178.
Lappas, T., Sabnis, G., & Valkanas, G. (2016). The impact of fake
reviews on online visibility: A vulnerability assessment of the
hotel industry. Information Systems Research, 27(4), 940-961.
Lewis, G., & Zervas, G. (2019). The supply and demand effects of
review platforms. Available at https://papers.ssrn.com/sol3/
papers.cfm?abstract_id=3468278
Li, H., Chen, Z., Liu, B., Wei, X., & Shao, J. (2014). Spotting fake
reviews via collective positive-unlabeled learning. In
Proceedings of the 2014 IEEE International Conference on
Data Mining (pp. 899-904).
Li, H., & Srinivasan, K. (2019). Competitive dynamics in the
sharing economy: an analysis in the context of Airbnb and
hotels. Marketing Science, 38(3), 365-391.
Lu, X., Ba, S., Huang, L., & Feng, Y. (2013). Promotional
marketing or word-of-mouth? Evidence from online restaurant
reviews. Information Systems Research, 24(3), 596-612.
Lu, Y., Gupta, A., Ketter, W., & Heck, E. van. (2019). Information
transparency in business-to-business auction markets: The role
of winner identity disclosure. Management Science, 65(9),
4261-4279.
Luca, M., & Zervas, G. (2016). Fake it till you make it: reputation,
competition, and Yelp review fraud. Management Science,
62(12), 3412-3427.
Marinova, P. (2016). Amazon is cracking down on more fake
reviews. Fortune. http://fortune.com/2016/10/27/amazon-
lawsuit-fake-reviews/
Martineau, P. (2019). Inside Airbnb’s “guerrilla war” against
local governments. Wired. https://www.wired.com/story/
inside-airbnbs-guerrilla-war-against-local-governments/
Mas‐Ruiz, F., & Ruiz‐Moreno, F. (2011). Rivalry within strategic
groups and consequences for performance: The firm-size
effects. Strategic Management Journal, 32(12), 1286-1308.
Mas‐Ruiz, F., & Ruiz‐Moreno, F. (2017). How strategic groups act
competitively within and across markets. Managerial and
Decision Economics, 38(7), 1017-1032.
Mas‐Ruiz, F., Ruiz‐Moreno, F., & Martínez, A. L. de G. (2014).
Asymmetric rivalry within and between strategic groups.
Strategic Management Journal, 35(3), 419-439.
Mayzlin, D. (2006). Promotional chat on the internet. Marketing
Science, 25(2), 155-163.
Mayzlin, D., Dover, Y., & Chevalier, J. (2014). Promotional
reviews: An empirical investigation of online review
manipulation. American Economic Review, 104(8), 2421-2455
McCool, C. (2015). 8 great Airbnb advantages. McCool Travel.
http://www.mccooltravel.com/2015/08/8-great-airbnb-
advantages/
Nieuwland, S., & van Melik, R. (2020). Regulating Airbnb: How
cities deal with perceived negative externalities of short-term
rentals. Current Issues in Tourism, 23(7), 811-825.
Obstfeld, D. (2005). Social networks, the tertius iungens
orientation, and involvement in innovation. Administrative
Science Quarterly, 50(1), 100-130.
Peteraf, M. A. (1993). Intra-industry structure and the response
toward rivals. Managerial and Decision Economics, 14(6),
519-528.
Porter, M. E. (1979). The structure within industries and
companies’ performance. The Review of Economics and
Statistics, 61(2), 214-227.
Porter, M. E. (2008a). The five competitive forces that shape
strategy. Harvard Business Review, 86(1), 78-93.
Porter, M. E. (2008b). Competitive strategy: Techniques for
analyzing industries and competitors. Simon and Schuster.
Proserpio, D., Xu, W., & Zervas, G. (2018). You get what you give:
Theory and evidence of reciprocity in the sharing economy.
Quantitative Marketing and Economics, 16(4), 371-407.
Proserpio, D., & Zervas, G. (2017). Online reputation management:
estimating the impact of management responses on consumer
reviews. Marketing Science, 36(5), 645-665.
Short, J. C., Ketchen, D. J., Palmer, T. B., & Hult, G. T. M. (2007).
Firm, strategic group, and industry influences on performance.
Strategic Management Journal, 28(2), 147-167.
Slee, T. (2016). What’s yours is mine: Against the sharing
economy. OR Books.
Smith, K. G., Grimm, C. M., Young, G., & Wally, S. (1997).
Strategic groups and rivalrous firm behavior: Towards a
reconciliation. Strategic Management Journal, 18(2), 149-157.
Stein, J. (2015). Strangers crashed my car, ate my food and wore
my pants: Tales from the sharing economy. Time Magazine,
February 9, 2015, 34-40.
Stock, J. H., & Yogo, M. (2005). Testing for weak instruments in
linear IV regression. Identification and inference for
econometric models: Essays in honor of Thomas Rothenberg.
Available at https://ssrn.com/abstract=1734933
Sun, M., & Zhu, F. (2013). Ad revenue and content
commercialization: Evidence from blogs. Management
Science, 59(10), 2314-2331.
Tian, L., & Jiang, B. (2018). Effects of consumer-to-consumer
product sharing on distribution channel. Production and
Operations Management, 27(2), 350-367.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1593
Tibken, S. (2013, October 24). Taiwan fines Samsung $340,000 for
bashing HTC. CNET. https://www.cnet.com/news/taiwan-
fines-samsung-340000-for-bashing-htc/
Wallenstein, J., & Shelat, U. (2017). Hopping Aboard the Sharing
Economy. BCG Global. https://www.bcg.com/publications/
2017/strategy-accelerating-growth-consumer-products-
hopping-aboard-sharing-economy
Xu, Y. (2017). Generalized synthetic control method: causal
inference with interactive fixed effects models. Political
Analysis, 25(1), 57-76.
Yang, M., Zheng, Z. (Eric), & Mookerjee, V. (2019). Prescribing
response strategies to manage customer opinions: A stochastic
differential equation approach. Information Systems Research,
30(2), 351-374.
Ye, Q., Law, R., Gu, B., & Chen, W. (2011). The influence of user-
generated content on traveler behavior: An empirical
investigation on the effects of e-word-of-mouth to hotel online
bookings. Computers in Human Behavior, 27(2), 634-639.
Zervas, G., Proserpio, D., & Byers, J. W. (2017). The rise of the
sharing economy: Estimating the impact of Airbnb on the hotel
industry. Journal of Marketing Research, 54(5), 687-705.
Zervas, G., Proserpio, D., & Byers, J. W. (2021). A first look at
online reputation on Airbnb, where every stay is above average.
Marketing Letters, 32(1), 1-16.
Zhu, F., & Zhang, X. (Michael). (2010). Impact of online consumer
reviews on sales: The moderating role of product and consumer
characteristics. Journal of Marketing, 74(2), 133-148.
About the Authors
Cheng Nie is an assistant professor of information systems and
business analytics at the Ivy College of Business at Iowa State
University. He received his Ph.D. from the Jindal School of
Management, University of Texas at Dallas. His current research
interests are in sponsored search, sharing economy, and user-
generated content.
Zhiqiang (Eric) Zheng is the Ashbel Smith Professor in
Information Systems at the Jindal School of Management,
University of Texas at Dallas. He received his Ph.D. from the
Wharton School of Business. His current research interests focus
on fintech, blockchain, and healthcare analytics. He currently
serves as a senior editor for Information Systems Research.
Sumit Sarkar is the Charles and Nancy Davidson Chair and
Professor of Information Systems in the Naveen Jindal School of
Management at the University of Texas at Dallas. He received his
Ph.D. from the Simon School of Business at the University of
Rochester. His current research interests are in the sharing
economy, crowdsourcing, recommendation systems, sponsored
search, data privacy, and information quality. He is a Distinguished
Fellow of the Information Systems Society at INFORMS.
Nie et al. / Competing with the Sharing Economy
1594
MIS Quarterly Vol. 46 No. 3 / September 2022
Appendix A
Model-Free Evidence
We examine whether the average ratings of hotels change as Airbnb supply increases. Figure A1 shows that TripAdvisor ratings increase in
the aggregate while Expedia average ratings stay relatively flat. At the aggregate level, hotels appear to increase self-promotion and decrease
demotion (more positive ratings and fewer negative ratings on TripAdvisor relative to Expedia).
We also examine the trends in positive and negative reviews on TripAdvisor. We plot positive reviews (4-star and 5-star for low-end, 5-star
for high-end hotels) and negative reviews (1-star and 2-star for high-end, 1-star for low-end hotels) for high-end and low-end hotels in Figure
A2. We find that low-end hotels have relatively stable proportions of negative and positive ratings (relatively flat over time), while high-end
hotels have increasing proportions of positive ratings and decreasing proportions of negative ratings on TripAdvisor. This is consistent with
our main findings that high-end hotels tend to increase self-promotion and reduce demotion activities while low-end hotels do not increase
manipulation activities. Note that these figures only show the trends in general while ignoring potential confounding factors.
Figure A1. Changes in Average Ratings with Airbnb Supply
Low-end hotels
High-end hotels
Figure A2. Proportions of Negative and Positive Ratings on TripAdvisor
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1595
Appendix B
Validating the Parallel-Trend Assumption
In our context, we do not use a conventional DID setup where there is one binary treatment and a common event that occurs at the same instant of
time for all hotels. Here Airbnb listings can enter at different times; further, there can be repeated entries (we use the count of Airbnb listings as the
main independent variable rather than a binary treatment). This forms a staggered DID setting for which the parallel-trend assumption is not directly
applicable. Nevertheless, by dichotomizing our independent variable into 0 and 1 (referred to as Airbnb_Binary hereafter), we are able to test the
assumption of parallel trends using the relative time model proposed in Autor (2003).
21
The relative time model has been widely used in the literature
to test the parallel-trend assumption (e.g., Chan et al., 2019; Greenwood et al., 2019; Lu et al., 2019). The idea is to add lead and lag treatment
variables into the regression and investigate their coefficients. If the coefficient of any lead variable turns out to be significant, it would indicate there
are pre-treatment trends in the data and so the parallel-trend assumption would be violated. We run the following relative-time model:
,
where j is {-2, -1, 0, +1, +2, +3} following Autor (2003).
22
Note that in this model, both the treated and the control groups change every period as
more hotels face Airbnb competitors over time. As a result, for any hotel that is treated at period t (i.e., starts to face Airbnb competition at period t),
the comparison group would be those hotels that do not face any Airbnb competitors in that period (nor in any previous period). As shown in Table
B1, none of the pre-treatment variables are significant in either the self-promotion or demoting others results. These results suggest that the parallel
trends assumption is fulfilled, and the observed relationship between review manipulation and Airbnb supply is unlikely to arise as an artifact from
events that occur in periods prior to the treatment. Importantly, the coefficients for the treatment period (i.e., Airbnb_Binary
treat(0)
) are significant (the
shaded row in the table). The significant coefficients in this analysis yield consistent signs with the main analyses: i.e., self-promotion is positive and
significant after the treatment, while demoting others is negative and significant after the treatment.
Table B1. Relative Time Model
Self-promotion
Demoting others
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
Airbnb_Binary
pre(-2)
-0.079
-0.071
-0.063
-0.042
-0.043
-0.043
(0.234)
(0.234)
(0.235)
(0.030)
(0.030)
(0.030)
Airbnb_Binary
pre(-1)
-0.035
-0.034
-0.035
0.079
0.080
0.080
(0.025)
(0.025)
(0.025)
(0.043)
(0.043)
(0.043)
Airbnb_Binary
treat(0)
0.060
*
0.058
*
0.063
*
-0.108
*
-0.107
*
-0.107
*
(0.026)
(0.026)
(0.027)
(0.054)
(0.054)
(0.052)
Airbnb_Binary
post(1)
-0.046
-0.044
-0.045
0.050
0.049
0.049
(0.025)
(0.025)
(0.025)
(0.041)
(0.041)
(0.040)
Airbnb_Binary
post(2)
0.010
0.009
0.010
-0.012
-0.011
-0.011
(0.030)
(0.030)
(0.030)
(0.036)
(0.036)
(0.036)
Airbnb_Binary
post(3)
0.040
0.040
0.041
0.096
0.093
0.093
(0.025)
(0.026)
(0.025)
(0.061)
(0.061)
(0.060)
log(CompetingHotels)
0.009
0.009
0.007
-0.254
*
-0.251
*
-0.251
*
(0.046)
(0.047)
(0.047)
(0.106)
(0.106)
(0.106)
Log(ReviewCount)
0.021
0.020
-0.008
-0.008
(0.015)
(0.015)
(0.026)
(0.026)
Airbnb_Binary
treat(0)
× Low-end
-0.042
0.003
(0.058)
(0.223)
ReviewRatios
NO
YES
YES
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
18,372
18,372
18,372
12,319
12,319
12,319
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
21
We use zero as the cutoff point to dichotomize Airbnb treatment. The results are robust to alternative cutoff points.
22
In the paper where the test for the parallel trend assumption was originally proposed (Autor, 2003, p. 24), t-n (n > 0) was used to denote post-treatment periods instead
of pre-treatment periods. However, some other papers (e.g., Greenwood et al., 2019; Lu et al., 2019) used relative time dummies to indicate the relative chronological
distance between time t and the treatment period, using t-n (n > 0) for pre-treatment periods. To eliminate such ambiguity, in Table B1, we adopt the notation
Airbnb_Binary
treat(0)
for the treatment period, Airbnb_Binary
pre(-1)
for the one-period pre-treatment, Airbnb_Binary
post(+1)
for one period post-treatment, etc.
Nie et al. / Competing with the Sharing Economy
1596
MIS Quarterly Vol. 46 No. 3 / September 2022
Appendix C
Mutual Demotion Across Hotel Triplets
In the Mutual Demotion Across Hotel Groups section, we examined whether the mutual demotion across groups of hotels decreased with the
emergence of Airbnb, using hotel pairs as the unit of analysis.
We now analyze how mutual demotion activities would change within a hotel triplet, where the unit of analysis becomes three hotels that are
competing with each other. Denote the three competing hotels as A, B, and C, respectively. Then the mutual demotion among these competing
hotels become: MutualDemoting(A,B,C) = Demoting(A→B) + Demoting(B→A) + Demoting(A→C) + Demoting(C→A) + Demoting(B→C)
+ Demoting(C→B). Similar to the hotel-pair analyses, we measure the intensity of Airbnb competition faced by the hotel triplet as the
cardinality of the intersection of the three sets of Airbnb listings that are competing with hotels A, B, and C, respectively. We measure the
intensity of competition from conventional hotels as the cardinality of the intersection of the three sets of competing hotels for A, B, and C,
respectively.
The results reported in Table C1 are qualitatively the same if we use the union instead of the intersection to aggregate the set of competing
Airbnb listings and competing hotels for the hotel triplets. In sum, the mutual demotion analyses demonstrate that high-end hotels reduce
their demotion activities within the incumbent hotels when they face Airbnb.
Table C1. Hotel Group Demotion Behavior with 2SLS (Hotel Triplet)
Model 1
Model 2
Model 3
log(Airbnb)
-0.259
***
-0.185
***
-0.178
***
(0.011)
(0.010)
(0.011)
log(CompetingHotels)
-0.797
***
-0.661
***
-0.640
***
(0.036)
(0.032)
(0.033)
Log(ReviewCount)
0.141
***
0.141
***
(0.004)
(0.004)
log(Airbnb) × Low-end
1.667
*
(0.798)
ReviewRatios
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
Time-fixed effects
YES
YES
YES
Observations
437,695
437,695
437,695
Note: *p < 0.05, **p < 0.01, ***p <
0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1597
Appendix D
Change in Customer Travel Types
We examine whether the customer travel types changed with the emergence of Airbnb. For example, customers who travel with families or
friends may shift to Airbnb listings. We verify if this is indeed the case from the data. During the observation period, many TripAdvisor
reviewers revealed their travel as belonging to one of five types: as a couple, on business, solo, with family, and with friends; the type is
recorded as “not specified” when no type is selected. The percentages of different types of travel remain relatively stable as shown in Figure
D1.
We use a repeated measures ANOVA to test the null hypothesis that the population means of the ratios of each travel type do not change.
23
The resulting p-value is close to 1, which means we fail to reject the null hypothesis. We also consider an alternative to the ANOVA by
testing if the time series of the six travel types are stationary. Using the KPSS test (Kwiatkowski et al., 1992) developed for this purpose, we
find that all six categories of travel types are stationary. Thus, there is no evidence that hotel customers have changed their preferences during
the observation period.
Figure D1. Changes in Proportions of Various Travel Types
23
In our context, the proportions of each alternative travel type are measured at each quarter. The assumption of one-way ANOVA is violated due to the
repeated measure for each ratio. Instead, we utilize the repeated measure ANOVA (as referred to as within-subject ANOVA) to examine whether the population
means of these groups change (Jackman, 2009, p. 317).
Nie et al. / Competing with the Sharing Economy
1598
MIS Quarterly Vol. 46 No. 3 / September 2022
Appendix E
Quantifying Competition Intensity Using a Smooth Kernel
Thus far, we have assumed that the competition between hotels and Airbnb listings is primarily constrained within a 1-kilometer radius with
respect to each focal hotel. The results are not sensitive to this cutoff distance, and the estimations are qualitatively the same when we use
0.5km or 2km as the competition radius. However, we have not differentiated between the competing hotels within the radius and have
attributed demoting reviews evenly across competitors. Competing hotels located farther or nearer may have varying incentives to demote
their competitors. In this Appendix, we test an alternative where we relax such restrictions that all competing hotels are considered equal.
We build on the idea of Luca and Zervas (2016) to model a gradual decrease in competition intensity due to increased distance. To see if our
results are robust, we use a Gaussian kernel with different values of bandwidths, where bandwidth corresponds to the standard deviation in
the context of z-score calculation. More specifically, let the impact of hotel j on hotel i be
,
where d
ij
is the distance between the two hotels, K is a kernel function, and h is a positive parameter called the kernel bandwidth. Depending
on the choice of K and h, w
ij
provides different ways to capture the relationship between distance and competition. The hard cutoff using the
competition radius h we consider in the Results section is a special case of this kernel weight when using a uniform kernel:
,
where 1
{…}
is the indicator function; i.e., K
U
assigns unit weight to competitors within a distance h, and zero to competitors farther away. The
Gaussian kernel, on the other hand, produces spatially smooth weights that are continuous in u and follow the Gaussian density function:
When the 1-km bandwidth is used for the Gaussian kernel, it means that a competitor (an Airbnb listing or a hotel) at the exact location as
the focal hotel would contribute 1 to the competition intensity, a competitor at a distance of 1 kilometer would contribute 0.61, one at a
distance of 2 kilometers 0.14, and so on. This captures the intuition that competing hotels that are closer may have more incentives to demote
a focal hotel. Therefore, we attribute the demoting reviews proportional to the competition intensity of a nearby hotel. We should point out
that the IV is modified accordingly. As shown in Table E1, our findings using the kernel weights remain consistent with the main analyses.
We test different alternatives of bandwidth choices to see if the results are robust. Both bandwidths of 1 km and 0.5 km (as used in Luca &
Zervas, 2016) generate results consistent with those reported in the main analyses.
Table E1. Quantifying the Competition Intensity Using a Smooth Kernel
Self-promotion
Demoting others
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
log(Airbnb)
0.012
*
0.011
*
0.011
*
-0.006
*
-0.006
*
-0.005
*
(0.005)
(0.005)
(0.005)
(0.002)
(0.002)
(0.002)
log(CompetingHotels)
0.058
0.057
0.054
-0.072
***
-0.073
***
-0.076
***
(0.036)
(0.035)
(0.035)
(0.020)
(0.020)
(0.021)
Log(ReviewCount)
0.004
0.004
0.003
0.003
(0.006)
(0.006)
(0.003)
(0.003)
log(Airbnb) × Low-end
-0.010
-0.011
(0.016)
(0.010)
ReviewRatios
NO
YES
YES
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
32,122
32,122
32,122
24,713
24,713
24,713
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1599
Appendix F
Alternative Operationalization of Self-Promotion and Demotion
As mentioned in the Identifying Review Manipulation subsection, because of the difference in average ratings for high-end and low-end
hotels, we measure self-promotion and demotion differently for high-end and low-end hotels. We also replicate our experiments by
considering the same measures that are used in Mayzlin et al. (2014): for all hotels (both high-end and low-end), the differences in the
proportions of 5-star ratings on TripAdvisor and Expedia are considered as potential self-promotions and the corresponding differences in 1-
star and 2-star ratings as demotions. The 2SLS results presented in Table F1 are consistent with our main results.
Table F1. Alternative Operationalization of Manipulation
Self-promotion
Demoting others
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
log(Airbnb)
0.027
***
0.025
***
0.026
***
-0.036
***
-0.036
***
-0.036
***
(0.006)
(0.006)
(0.006)
(0.009)
(0.009)
(0.009)
log(CompetingHotels)
0.066
*
0.064
*
0.054
-0.289
***
-0.286
***
-0.290
***
(0.031)
(0.030)
(0.030)
(0.044)
(0.044)
(0.044)
Log(ReviewCount)
0.012
*
0.011
-0.014
-0.014
(0.006)
(0.006)
(0.010)
(0.010)
log(Airbnb) × Low-end
-0.095
***
-0.037
(0.020)
(0.046)
ReviewRatios
NO
YES
YES
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
32,122
32,122
32,122
24,713
24,713
24,713
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
Nie et al. / Competing with the Sharing Economy
1600
MIS Quarterly Vol. 46 No. 3 / September 2022
Appendix G
Segmenting Airbnb Competitors
Airbnb listings are quite different in terms of their quality and price. In the main analyses, we considered all nearby Airbnb listings as
competing with a focal hotel regardless of the nature (i.e., price and quality) of those listings. In this Appendix, we evaluate whether our
estimation results are sensitive to this assumption.
24
A common measure of quality is review rating. The user-generated reviews for Airbnb listings have very small variance and thus do not help
to differentiate across those listings. For example, 95.7% of all listings boast an average user-generated rating of either 4.5 or 5 stars (highest);
less than 0.3% of Airbnb listings have less than 3.5 stars. The mean and standard deviation of the user-generated ratings in Airbnb is 4.82
and 0.31 respectively. These statistics in our Airbnb listings sample are very similar to the ones reported in Zervas et al. (2021). Therefore,
we did not use user-generated ratings as a control for the quality of Airbnb listings.
To obtain the price information of Airbnb listings, we obtained data from AirDNA.co that tracks the price information of each Airbnb listing
over time. This enables us to categorize Airbnb listings as high-end or low-end and to examine the competition between similar segments of
hotels and Airbnb listings (i.e., the competition between high-end hotels and high-end Airbnb listings, and the competition between low-end
hotels and low-end Airbnb listings). To segment the Airbnb listings, we calculated the price per person (guest) for each Airbnb listing in our
sample and then categorized the listings based on this price per person. Since 24% of all the hotels in our observations are low-end hotels,
we use the 24th percentile of the Airbnb listing price per person as the cutoff point (which corresponds to $25 per person). The results, as
presented in Table G1, are consistent with the main findings.
25
Table G1. Segmenting Airbnb Competitors
Self-promotion
Demoting others
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
log(Airbnb)
0.020
**
0.019
**
0.018
**
-0.034
***
-0.034
***
-0.033
***
(0.006)
(0.006)
(0.006)
(0.008)
(0.009)
(0.008)
log(CompetingHotels)
0.049
0.048
0.049
-0.296
***
-0.294
***
-0.295
***
(0.031)
(0.030)
(0.030)
(0.044)
(0.043)
(0.044)
Log(ReviewCount)
0.004
0.005
-0.012
-0.012
(0.006)
(0.006)
(0.010)
(0.010)
log(Airbnb) × Low-end
0.003
-0.006
(0.012)
(0.015)
ReviewRatios
NO
YES
YES
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
32,122
32,122
32,122
24,713
24,713
24,713
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
24
We thank an anonymous reviewer for suggesting this analysis.
25
Low-end hotels appear to increase self-promotions based on Model 3 for self-promotion in Table G1. However, such an increase is statistically insignificant
when we use the low-end hotel as the reference level.
Nie et al. / Competing with the Sharing Economy
MIS Quarterly Vol. 46 No. 3 / September 2022
1601
Appendix H
Combining Competitors Across Categories
In our main analyses, we assumed that hotels are only competing with other hotels from the same category; namely, low-end hotels are
competing with low-end hotels while high-end hotels with other high-end hotels only. The rationale behind this is that a low-end hotel is
unlikely to demote a nearby high-end hotel in order to capture a portion of the demand for the high-end hotel and vice versa. Next, we relax
this assumption by considering hotels as competitors regardless of their categories. To incorporate this change, the competitor count, the
count of demotion actions (i.e., negative reviews), and the instrumental variable measures are modified correspondingly. The results, as
presented in Table H1, show consistency with our main findings.
Table H1. All Neighboring Hotels as Competitors
Self-promotion
Demoting others
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
log(Airbnb)
0.022
***
0.020
**
0.021
***
-0.018
*
-0.018
*
-0.017
*
(0.006)
(0.006)
(0.006)
(0.007)
(0.008)
(0.008)
log(CompetingHotels)
0.057
0.055
0.054
-0.245
***
-0.243
***
-0.245
***
(0.031)
(0.030)
(0.030)
(0.038)
(0.038)
(0.038)
Log(ReviewCount)
0.005
0.005
-0.007
-0.007
(0.006)
(0.006)
(0.008)
(0.008)
log(Airbnb) × Low-end
-0.021
-0.027
(0.024)
(0.014)
ReviewRatios
NO
YES
YES
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
32,122
32,122
32,122
26,748
26,748
26,748
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)
1602
MIS Quarterly Vol. 46 No. 3 / September 2022
Appendix I
Difference between Airbnb and Hotel Competition
To further reinforce our findings, we used an alternative approach to investigate how Airbnb listings influence hotels’ review manipulation
behaviors differently.
The number of Airbnb listings started increasing rapidly around the middle of our observation period (i.e., Quarter 16 (2011 Q4) is an
inflection point as shown in Figure A1). Prior to that point, the average number of competing Airbnb listings was 0.31. The average bumped
to 4.62 after the middle point. Therefore, we use 2011 Q4 as the cutoff to split the data into two subsamples. We expect the impact of Airbnb
to be minimal before it gains momentum, but its impact would become stronger as more Airbnb listings appear.
The subsample analyses, presented in Table I1, demonstrate that the growth of Airbnb did change how incumbent hotels respond to competing
hotels. Before Airbnb listings became popular, the impact of competing Airbnb listings and competing hotels are both insignificant. After
the inflection point, an increase in either competing Airbnb listings or competing hotels leads to significantly fewer demotion activities. This
is consistent with the prediction from strategic group theory, that the incumbent group of hotels engaged in less destructive competitive
behavior after Airbnb listings grew substantially.
Table I1. Demotion Behavior before and after the Inflection Point (2011 Quarter 4)
Before 2011 Q4
After 2011 Q4
Model 1
Model 2
Model 3
Model 1
Model 2
Model 3
log(Airbnb)
0.010
0.010
0.005
-0.025
*
-0.026
*
-0.026
*
(0.015)
(0.015)
(0.013)
(0.013)
(0.013)
(0.013)
log(CompetingHotels)
-0.098
-0.097
-0.087
-0.258
***
-0.256
***
-0.256*
**
(0.159)
(0.159)
(0.159)
(0.056)
(0.056)
(0.056)
Log(ReviewCount)
0.010
0.011
-0.013
-0.013
(0.011)
(0.011)
(0.015)
(0.015)
log(Airbnb) × Low-end
0.255
0.001
(0.251)
(0.069)
ReviewRatios
NO
YES
YES
NO
YES
YES
Hotel-fixed effects
YES
YES
YES
YES
YES
YES
Time-fixed effects
YES
YES
YES
YES
YES
YES
Observations
5,305
5,305
5,305
19,226
19,226
19,226
Note: *p < 0.05, **p < 0.01, ***p < 0.001; Robust standard errors are in parentheses (errors clustered at hotel level.)