The exhaustion of the herding effect in peer-to-peer accommodation

Abstract

A significant reason for the concentration of demand in a subset of the supply in the peer-to-peer market for tourist accommodation is herding behavior, by which the decisions of the first guests are imitated by those who follow. This article proposes a profit- and utility-maximization microeconomic model and implements it with data of Airbnb listings corresponding to 10 European cities. Results show that the influence of each additional review is positive but decreasing, inducing a more balanced distribution of demand among offered accommodation and thus dampening the herding effect. Moreover, reservation policy—specifically, enabling the instant booking option—is a key to explain the initial push that accommodations need to be demanded now and, hence, to increase their possibilities of being demanded in the future.

Keywords

Airbnb convergence demand herding behavior reservation policy

Introduction

Despite the fact that research on the peer-to-peer (p2p) market for tourist accommodation has reached the threshold necessary to justify some review articles (Andreu et al., 2020; Belarmino and Koh, 2020; Dann et al., 2019; Dolnicar, 2019; Guttentag, 2019; Prayag and Ozanne, 2018; Sainaghi, 2020), there are still aspects that continue to present provisional results or that have not been sufficiently analyzed (Sainaghi et al., 2020). In particular, although asymmetric information has been a problem affecting traditional platforms that facilitate exchanges over the Internet (Lewis, 2011), the emergence of new platforms that facilitate p2p exchanges in the field of tourism, such as Airbnb, raises new issues.

Unlike conventional accommodation, such as hotels, which have a star-rate classification system that helps reduce the adverse effects caused by asymmetric information in the hospitality industry (Martin-Fuentes, 2016), in p2p tourist accommodation market, there is no objective rating system in place to resolve uncertainty. Instead, the system centers on user reviews (subjective criteria), which has been shown to suffer from abundant biases (Zervas et al., 2015). One of them is herding behavior, which, by making later consumers’ decisions rely on previous ones’, generates inefficiencies in the aggregate despite agents’ behavior being rational at an individual level (Banerjee, 1992).

The simple model of herd behavior originally proposed by Banerjee was not tied to any specific sector, although he used in his exposition the well-known example of the two restaurants that, despite their geographical proximity, present different dynamics due to the imitative and cumulative behavior of potential clients. In principle, a large share of the theoretical and empirical developments have been carried out in financial markets (Spyrou, 2013) or to explain biases in investment decision-making (Kumar and Goyal, 2015). Applications can also be found in areas related to tourism such as restaurants (Ha et al., 2016) or hotels (Park et al., 2017). However, the expansion of the Internet markets and, more specifically, p2p markets (Einav et al., 2016) added a new dimension to this phenomenon. In particular, the impact of electronic word of mouth (eWOM) on the hospitality and tourism industry has reinforced the influence that those who have already tried a product have on potential consumers (Litvin et al., 2018).

In the p2p market for tourist accommodation, herding behavior has been used as an argument to explain Airbnb’s preponderance over other platforms (Volgger et al., 2019), the upward bias of valuations (Zervas et al., 2015) and the increase in the number of customer purchases when hosts opt for sales history disclosure (Xie et al., 2019). However, as far as we are concerned, there is no specific study of the effects that, at the microeconomic level, herding behavior causes in the said market and of the variables that can attenuate—or strengthen—its influence. On that basis, our main aim is to fathom the impact of prior visits (i.e. of consumption decisions by earlier decision-makers) on future ones; that is, how the decision history affects the purchases of subsequent potential customers, while taking into account, other elements that might influence the demand for that product (which may or may not be under the control of suppliers). Given the uncontested leadership of Airbnb among facilitators of p2p accommodation trading (Hajibaba and Dolnicar, 2018), we use data from the platform on listings in 10 important cities of Europe.

Our research differs from previous studies as we frame our empirical work in an innovative profit- and utility-maximization microeconomic framework adapted to p2p tourist accommodation trading, grounded in conventional economic assumptions but introducing asymmetric information considerations and specificities to this market, such as the effects of instant booking, of cancellation policy, and of reviews. This can both aid the creation of further work for markets with similar dynamics and, as previously mention, guide accommodation managers, both private and professional ones.

The article is organized as follows. First, we survey the literature on reviews and herding behavior. Then, we devote a theory section to present the microeconomic framework. The sources of information as well as the econometric model and technique are defined in the “Data and methodology” section. The ensuing results are presented and discussed afterward. Finally, we conclude.

Reviews and herding behavior

Even though there is no worldwide standard for official hotel classification, the five-star system provides a guide, to both consumers and intermediaries, about the quality standards of certain accommodations (UNWTO, 2015). So even if higher star ratings do not always mean higher quality (Fernández and Bedia, 2004), they have a significant effect on guest satisfaction (Rajaguru and Hassanli, 2018).

Objective classification has always coexisted with subjective valuations by WOM. And now, the Internet has made these opinions much more quickly and easily accessible (Ladhari and Michaud, 2015): The so-called eWOM has significantly reduced search costs, albeit the cognitive costs associated with the processing of relevant information have increased (Liu and Park, 2015). The tourism industry has considerably been affected by the generalization of online reviews (Cantallops and Salvi, 2014). It is even expected that, in the future, conventional rating systems based on objective criteria will be integrated within the platforms that publish guest-authored reviews and that those that do not integrate will end disappearing (Hensens, 2015).

As for accommodations offered in the p2p market through platforms such as Airbnb, reputation has, from the outset, fundamentally been based on subjective criteria (Cheng and Jin, 2019). Even though the characteristics of the accommodation can be partly learned through descriptions and images, potential guests may have doubts about how reliable what the host advertises or the photographs are. It is always possible to establish a mechanism that certifies the authenticity of the photos (as Airbnb offers to owners) and so alleviate the issue of asymmetric information to a degree. Nevertheless, ratings and reviews have consolidated themselves over time as a means of generating reputation and trust (Ter Huurne et al., 2017). This mechanism has worked relatively well and has prevented the proliferation of fraudulent behaviors. However, it has also aroused some suspicions, and some imperfections have been revealed from the academic point of view (Einav et al., 2016).

One of the most discussed aspects is the apparent upward bias in the valuations of internet transactions. Hu et al. (2009) refer to the plot of the valuations’ frequency on their scale as a “J-shape,” as they are concentrated in the high end. Zervas et al. (2015) analyze more than 600,000 accommodations advertised through Airbnb and found their score to be between 4.5 and 5 points (the maximum) in nearly 95% of cases. Besides, a qualitative analysis of the comments also revealed a tendency towards positive valuations (Bridges and Vásquez, 2018). And more recently, a large-scale study confirmed that accommodations obtaining a poor valuation were relatively rare (Ke, 2017).

Multiple reasons can explain the high percentage of positive ratings (Zervas et al., 2015). First, negative valuations may be underrepresented to avoid retaliation on platforms that allow bilateral valuations. In addition to that, it is possible that negative evaluations harm not only the recipient but also the grantor. Whoever receives a negative valuation experiences a decrease in their expected utility as they will have fewer possibilities of establishing future transactions. In turn, the expected utility of who makes a bad valuation may also decrease as potential sellers or customers—as the case may be—will be less likely to trade with them to avoid negative valuations. In one experiment, Fradkin et al. (2015) altered Airbnb’s simultaneous disclosure system, allowing the actors to send valuations once the other parties have been published. Their results confirmed the existence of a bias in fear of possible retaliation, the effective existence of retaliation for negative reviews, and reciprocity for positive reviews. The authors point that it is the social interaction that takes place between hosts and guests what tends to prevent negative valuations. In this way, a tacit pact to exchange positive valuations would be made.

A second origin of the upward bias in valuations might be that it is precisely the most satisfied customers who are most likely to leave a review. In another experiment, the same authors mentioned in the previous paragraph encouraged a group of guests to leave a valuation after using an Airbnb accommodation, offering them a $ 25 discount voucher for their next trip. That resulted in not only an increase in participation but also in a decrease in the percentage of highest ratings. This supports the thesis that a system such as that of Airbnb, which does not compensate for rating your stay, may cause a bias to the extent that some experiences are underrepresented.

Thirdly, we should not discard possible manipulation mechanisms that certain agents set in motion to strengthen their reputation. For example, Mayzlin et al. (2014) compared the ratings of Expedia, where only those who have confirmed the reservation of a hotel room can leave their valuation, with those of TripAdvisor, where anyone can leave a valuation, and found evidence of manipulation in the latter. Various strategies of manipulation by the managers of conventional accommodation have been documented (Gössling et al., 2018; Magno et al., 2018). Still, as far as we know, none has been described for the case of accommodation in the p2p market. At a more general level, this fact raises a debate on the design of the most reliable valuation system, including elements of discussion such as who can make an assessment, anonymity, the incentives to leave a comment, or the convenience of a unilateral system (only one party can make a valuation) versus bilateral (both parties, buyer and seller, can).

Finally, the upward bias in valuations might be explained by herding-type models, in which each decision-maker is influenced by the decisions made by previous decision-makers with respect to a given product, as explained by Banerjee (1992). His model depicts a “winning” action under three possible information states (“right”, “wrong” and “no information”). The information set is a continuous closed set of values and several tiebreaking rules are imposed: when a decision-maker has no signal and everyone else has chosen the lowest point in the set, he/she imitates them; when he weights his own signal against someone else’s individual choice, he chooses the former; and when he is indifferent between more than one previous choices, he/she picks the one that is highest in the set. In this setting, if the first two players choose the same option, the rest will follow ignoring private information, creating an irrational cascade that can result in a non-Pareto optimum if the chosen action was not the winning one. These assumptions simplify the results but condition the conclusions.

In the case of Airbnb or online accommodation booking in general, potential guest can only observe whatever information is listed on the ad (including textual and photographic cues), but other elements are only known to the host—who might find no credible way to signal their value ex ante. Through this reasoning, the initial number of reviews may be considered an important determinant variable for the subsequent growth in reviews.

It has been argued that having a single review, as opposed to none, makes an enormous difference (Hill, 2015). With a sample of 4178 room data from the cities of New York, Chicago, Los Angeles, San Francisco, and Seattle, Lee et al. (2015) found surprisingly that the quantity of reviews is more important than their associated rating in explaining room sales—in fact, five of the seven ratings (including overall rating) are excluded from their model, implying that they are not critical enough to predict room sales. If we consider the possible herding behavior of the guests, this result is not unexpected. Along these lines, Chen and Chang (2018), with a poll made to learn about the factors that drive Airbnb purchases, assess that rating volume had a significant impact, while ratings did not.

Theoretical foundation

To understand the role of reviews in the sale and purchase of short-term accommodation, we first need to understand the structure of the product being traded. Suppose g is a potential guest of a generic listing, p is the price of the listing, q is the quality of the listing observable ex ante from the ad by the guest but not necessarily fully observable by the econometrician, and $r$ is the accommodation's number of reviews. The ex ante expected utility of the guest g from staying in the advertised accommodation is as follows:

u = v_{g} - p

where

v = r^{1 - α (g)} q^{α (g)}

where v is the guests' gross valuation of the accommodation and $α (g)$ is a parameter in (0,1). This utility is quasilinear in the price p. $τ_{g} \in (0, 1)$ is the probability that guest g cancels. It might vary by guests but cannot be known by hosts; although they know its expectation is $Ε [τ_{g}] = τ$ . c is the compensation that the host receives in case of cancellation, which is detailed in the “Cancellation policy” section.

$I \in {0, 1}$ is an indicator, that is, 1 if the instant booking option is activated and 0 otherwise. $δ \geq 0$ is the disutility of not being able to discriminate. This means that, by enabling instant booking, the host might get certain visits that they would have liked to decline if instant book had not been activated—either because of the requesting guest or because of the timing. This parameter varies by host but not in a way that is necessarily correlated with the value of the accommodation v. The demand function of a listing is D, which depends on some variables in the ways that will be seen below.

Instant booking

D can be written as D(I), assuming that demand depends (non-negatively) on $I : D (1) \geq D (0)$ , as it seems reasonable that guests find it equally desirable or more to have their booking instantly confirmed than having to wait—and possibly have their request declined or expired. Keeping price fixed, the host profit is as follows:

max {D (1) p - δ, D (0) p}

and IB = 1will be set if

D (1) - D (0) \geq \frac{δ}{p}

which means that instant booking will be enabled by the host if the increase in demand from this option is greater than the loss from being unable to discriminate, as a ratio of the price.

Cancellation policy

Because of the possibility of a cancellation, selling accommodation is a lottery. If the visit is not cancelled, the host gets p. If the visit is cancelled, the host gets c such that $c \leq p$ . Because guests prefer paying less, $\partial D (c) / \partial c < 0$ . To have a closed-form solution, we chose one demand function that meets this criterion: $D (c) = a - b c$ , with ${a > 0, b > 0}_{}$ . Then, the host’s problem is as follows:

max {(a - b c) * [(1 - τ) p + τ c]}

and the first-order condition is as follows:

a τ - (b p (1 - τ) + 2 c τ) = 0

which implies the solution is:

c^{*} = \frac{a τ - b p (1 - τ)}{2 b τ}

where p = c implies strict cancellation. Intuitively, the less reactive demand is to paying this compensation, the lower the b; hence the higher the $c^{*}$ and the more likely that the cancellation policy is strict.

A possible concern is that guests could self-select since $τ_{g}$ varies by guest, and those with higher values of that parameter prefer more flexible cancellation policies. Hosts then do not have to pay c with probability $τ$ but with a higher probability. This is not modelled in our example but it could be done making τ depend on c such that $\partial τ (c) / \partial c \leq 0$ . Without knowing the exact distribution of $τ_{g}$ , there is no way for hosts to determine the optimal strategy except by trial and error or by learning from other hosts. Anyhow, this will result in a negative force from flexible cancellation against (effective) visits. This is reinforced by the moral hazard problem that we explain in the following paragraph.

Sometimes the reason for cancelling makes it impossible to go to the destination—such as when it is injury, illness of the visitor or other unexpected events of greater importance. In other situations—such as finding a better alternative—the guest might find it profitable to cancel if the policy is flexible but not if it is strict. This, given the quasilinear utility function that we chose, u = v − p, can be translated into the appearance of an unexpected outside option posterior to booking, s, such that $v - p \in (s - c, s)$ . That is, when nothing has to be paid for the cancellation, the outside option yields $s > υ - p$ , so the guest prefers to cancel. On the contrary, when cancellation is costly, c has to be paid, and because $s - c < υ - p$ , the guest will attend the trip as planned.

Reviews

As argued in the “Reviews and herding behavior” section, review content may be uninformative, being their quantity what eventually matters. The logic behind this reasoning can be better understood with an example. Suppose there are N observationally identical accommodations (listing the same price and tangible and intangible attributes). There might still be some uncertainty about factors experienced ex post that cannot be signaled ex ante. The probability of a bad experience with these factors can be represented by a value ϵ ∈ (0, 1). Once an initial consumer has a sufficiently high individual valuation for the stay to afford choosing at random, some accommodation among the N will be selected and will receive a review that is positive with probability 1 − ϵ. The choice of the consumer that chooses later will be clear: one of the N accommodation is certainly good in terms of the ex ante unobservables, while the remaining N − 1 could still be bad with a probability ϵ > 0, so the former will be chosen again. This is the so-called herding effect and will occasion a strong concentration of reviews in a few listings that will only increase over time.

We have implemented this positive effect of reviews in our model. Given the form of the value in (2), the consumer likes both observable quality $q$ $q$ and reviews $r$ (the latter as an ex ante signal of accuracy and safety), although in a balanced manner. The marginal influence of reviews on overall value is as follows:

\frac{\partial v}{\partial r} = α {(\frac{q}{r})}^{1 - α}

which is positive but decreasing. Because it is possible to approximate a function of the form of (8) with a quadratic polynomial, we will include not only initial reviews but initial reviews squared in our estimation.

Our theoretical framework predicts that, while price and instant booking are unequivocally affecting demand (negatively and positively, respectively), we will find an ambiguous relationship of review growth with flexible cancellation and of review growth with previous reviews. The aim of our empirical approximation is to discover the dominant sign of these relationships as well as to approximate their magnitude and thus the importance of each of these variables in obtaining visits.

Data and methodology

Data on p2p markets are not gathered in official statistics, so surveys, isolated experiments, or information scraped from some platforms are normally used to analyze them. There have been some proposals to estimate the demand for short-term rental housing. Airbnb offers aggregate data on the average stay of the guests for some cities (Airbnb, 2019). For New York, it is even disaggregated by districts (Airbnb, 2017). But since stays are not public at the listing level, the number of reviews has often been used as a proxy for it. In Airbnb, only confirmed guests can leave a review; so the number of reviews is a lower bound on demand. The same variable has also been used to proxy hotel and restaurant sales (Schuckert et al., 2015), and our research would not be the first to use it for the minimum number of room reservations over time in Airbnb (Lee et al., 2015; Marqusee, 2015; San Francisco Budget and Legislative Analyst, 2015).

We use data from InsideAirbnb, a website that periodically publishes information on the listings offered through Airbnb in different cities of the world (InsideAirbnb, 2018). We have selected the 10 European cities with the most listings, which are naturally cultural capitals and popular tourist destinations. Scraped information was available for 2018 around mid-April and for the same days in 2019, which we have taken as starting (t) and ending dates (t + 1). A year-long time difference minimizes the effects attributable to seasonality and eases the comparison between cities. Based on the results of previous studies (Benítez-Aurioles, 2018b; Gibbs et al., 2018; Gunter and Önder, 2018; Wang and Nicolau, 2017) and in our theoretical model, we include the explanatory variables listed in Table 1, which we detail in this section. The variables correspond to dates t or to the change between t and t + 1. Without the indication of a date or subscript, the variable will correspond to a change or to a value that should not change over the period (e.g. number of bathrooms or type of accommodation).

Table 1.

Description of variables.

Name	Description
Reviews (²)	Initial number of reviews in t (squared)
Δ Reviews	Review growth between t and t + 1 (normalized by days in the period)
Accommodates	Maximum beddable guests
Bathrooms	Number of bathrooms
Entire home	1 if the room is entire home and 0 if private or shared room
Distance	Euclidean distance, in kilometers, from the accommodation to the City Council.
Superhost	1 if owner is a superhost and 0 otherwise
Listings	Number of listings owned by the same host
Price	Price in dollars between t and t + 1
Δ Price	Change in price between t and t + 1
Flexible	1 if the cancellation policy is flexible in t and 0 if stricter
Disable flexible	1 if flexible policy is removed over the period and 0 otherwise
Enable flexible	1 if flexible policy is added over the period and 0 otherwise
Inter × flexible	Interaction of flexible cancellation policy dummy with reviews
Instant	1 if the instant booking option is enabled and 0 if not
Disable instant	1 if instant booking policy is removed over the period and 0 otherwise
Enable instant	1 if instant booking policy is added over the period and 0 otherwise
Inter × instant	Interaction of instant booking policy dummy with reviews
Amenities	A vector of dummy variables that take value 1 if the item is present.

– Price. Without any qualification, this variable refers to the nightly price that would be paid by the first guest for a listing on the first date of our sample. We add the cleaning fee, which is a mandatory lump-sum payment per trip. There exists some evidence, albeit nonsignificant, of a positive relationship between price and reviews (Ert et al., 2016). Yet, most of the significant evidence points to price and reviews moving in opposite directions (Benítez-Aurioles, 2018b; Gibbs et al., 2018; Teubner et al., 2017; Wang and Nicolau, 2017).

– Room type (entire). Airbnb classifies accommodation into three categories: entire houses/apartments, private rooms, and shared rooms. Shared rooms represent a small proportion of total supply, so the idiosyncrasy of rooms in these categories weights substantially and can bias the results. For example, when comparing different types of Airbnb accommodation in various tests with respect to price, it is clear that entire homes are typically considered to be a “more luxurious” product than the other two; but the rank is not so clear when we compare private to shared accommodation, in part because of the latter’s small size. Therefore, we have grouped private and shared rooms into one base category. We expect the effect of entire homes on demand to vary across cities, depending on travelers’ preferences for having a more private accommodation but with less assistance from hosts and social experience.

– Number of bathrooms and accommodates. These two variables are directly related to the physical capacity of an accommodation. The latter one refers to the highest amount of guests that can be bedded in it. Despite being often included, we have omitted the number of beds and bedrooms as it is highly correlated with accommodates (unlike bathrooms) and, after controlling for bedrooms, often display a confusing negative correlation. We expect accommodates to be correlated with more reviews.

– Superhost. Airbnb grants this distinction to hosts that have at least 4.8 of 5 stars in overall rating, respond in 24 h at least 90% of the time, have hosted a minimum of 10 guests in a year, and do not make any reservation cancellation save for exceptional reasons. This mechanism is meant to reward hosts with a distinction that certifies them and differentiates them from other hosts in front of consumers. Nevertheless, empirical evidence on the badge’s effect on occupancy rates is limited, with the exception of Liang et al. (2017), who report, based on 2015 Hong Kong data, that the negative association of price and review volume can be positively moderated by the superhost badge and that the hosts’ effort to obtain the qualification may translate into higher prices through increased willingness-to-pay of guests for superhosts’ accommodations. And recently, Gunter (2018), using data on Airbnb listings from San Francisco and the Bay Area, revealed that excellent ratings are, by far, the most important criteria—followed by reliable cancellation behavior of the host, host responsiveness, and sufficient Airbnb demand. We expect the superhost distinction to attract visits.

– Distance. Tourists’ taste for locating close to places that they visit has long been used to explain hotels’ location (Arbel and Pizan, 1977). More recently, location patterns of Airbnb accommodations have been discussed as well. For the case of Barcelona, Gutiérrez et al. (2017) found that Airbnb was able to locate better than hotels in terms of proximity to cities’ touristic attractions. Benítez-Aurioles (2018a) contributes a model backed by empirical evidence to explain Airbnb accommodations’ tendency to locate in the center of cities. In our case, following the suggestion of Gibbs et al. (2018), we have considered the distance from the city council of each city, assuming that this institution is usually located somewhere in the center and near some tourist attraction. We expect distance from it to be negatively associated with the number of reviews.

– Booking policy. Two binary variables are incorporated, for having a flexible cancellation policy and for instant booking. Flexible cancellation allows a guest to cancel up to 24 h before their scheduled check-in without penalty; and if later, the first night is not reimbursed. For the reasons that we described, laxer policies might encourage cancellations and, therefore, have an adverse effect on reviews. Instant booking, instead, is expected to increase reviews. We also included variables representing the change during the period: Disable flexible and Disable instant for those who had previously activated the policy but removed it and Enable flexible and Enable instant for the opposite change to capture possible asymmetry. The interaction of flexible and instant with initial reviews is also defined.

– Number of reviews (Reviews) and number of reviews squared (Reviews²). As argued in the “Reviews and herding behavior” section and postulated in our model, we expect the amount of existing reviews to reinforce future transactions. Although not commonly included, we add Reviews² as a regressor to capture the possible decreasing or increasing marginal returns of this variable in terms of producing newer reviews, giving flexibility to the function.

– The Amenities vector contains amenities related to safety (carbon monoxide detector, first aid kit, smoke detector, and safety card), appliances (coffee maker, essentials, shampoo, washer, and microwave), privacy (lock on bedroom door and private entrance), comfort (air conditioning, extra pillows, and blankets), practicality (paid parking off premises, luggage drop-off allowed, self-check-in, and 24-h check-in), connection (Wi-Fi, Internet, and pocket Wi-Fi), and audience (family/kid-friendly and smoking allowed).

We incorporate the aforementioned variables into the following relationships:

Δ {Reviews}_{i} = α_{c} + β_{c}^{'} X_{i} + F ({Reviews}_{i t}) + u_{i}

where $α_{c}$ is a constant that varies by city, $β_{c}$ is a vector of coefficients associated to X_i, the vector of all the variables that do not include reviews, u_i is an error term at the listing level, and

F ({Reviews}_{i t}) = a * {Reviews}_{i t}^{2} + b * {Reviews}_{i t}^{2}

We estimate the coefficient vectors $β_{c}$ (9) and a and b in (10) through ordinary least squares with heteroscedasticity-robust standard errors. The results are presented and commented in the following section. In any case, we have to acknowledge that the number of reviews in the present is possibly conditioned by the number of reviews in the past and these two variables enter the equation via the independent and dependent sides in a way that might affect the estimation results.

Results and discussion

The results of the estimation (9) are presented in Table 2 and correspond in general terms to the expected results. Among the first observations, we can make is that the R² coefficients are remarkably large: the model specified accounts for nearly 50% of the data variation in most cases. There is also notable consistency in the signs and sizes of the parameters estimated within variables (especially of significant ones), which we review in the paragraphs to come.

Table 2.

Regression results for (9).

	Amsterdam	Barcelona	Berlin	Copenhagen	Lisbon	London	Madrid	Milan	Paris	Rome
Accommodates	0.437***	0.161	0.653***	0.156**	0.153*	0.256***	0.609***	0.519***	0.237***	−0.190**
Bathrooms	−1.225***	−0.462	−1.151***	−0.370*	−0.439***	−0.0749	−0.746*	0.0933	0.256**	0.428*
Entire	−7.599***	0.186	−1.196***	−1.153***	2.476***	−2.435***	1.753**	1.103**	−1.034***	4.228***
Distance	−0.229**	−0.817***	−0.116**	−0.173***	−0.154***	−0.0907***	−0.0634	−0.467***	−0.0959*	−0.312***
Superhost	1.300***	4.600***	4.788***	2.412***	6.409***	2.758***	5.922***	5.088***	2.136***	5.084***
Listings	−0.0232***	0.00149	0.0151	−0.0439***	−0.0251***	−0.00199***	−0.0246***	−0.00824***	−0.00606***	−0.0142***
Price	−0.00273***	−0.00290**	−0.00369**	0.00278**	−0.00535***	−0.00320***	−0.00909	−0.0131***	−0.00296***	−0.00589***
Δ Price	−0.00249	0.000178	−0.0012	0.00948*	−0.00422**	−0.00078	−0.000544	−0.0104***	−0.0000475	−0.00382**
Flexible	−2.521***	−3.308***	−1.348***	−0.653***	−1.935***	−2.405***	−2.491***	0.0994	−1.711***	−0.274
Disable flexible	5.869***	1.52***	5.686***	3.761***	5.335***	6.062***	6.391***	4.184***	5.926***	3.428***
Enable flexible	4.520*	0.0753	1.049	1.904*	0.32	0.387	−5.532***	3.600*	1.126	1.225
Flex. × reviews	0.0644***	0.0380*	−0.0182	0.0252	0.0519***	0.117***	0.0835***	0.0300*	0.0329*	0.0215**
Instant	3.286***	4.482***	2.683***	0.987***	5.799***	2.394***	6.175***	6.002***	2.547***	5.220***
Disable instant	−2.036**	−0.258	−1.134*	0.252	−4.922***	−1.303***	−4.596***	−5.247***	−1.360***	−4.137***
Enable instant	3.937***	4.581***	3.054***	2.520***	1.181*	1.854***	2.512***	2.882***	3.494***	1.582***
Instant × reviews	0.0396***	−0.00327	0.0514***	0.112***	−0.0162	0.0659***	0.0348**	0.0543***	0.0452***	0.0255***
Reviews	0.268***	0.327***	0.368***	0.238***	0.334***	0.285***	0.383***	0.259***	0.283***	0.281***
Reviews²	−0.000289***	−0.000407***	−0.000676***	−0.000275***	−0.000421***	−0.000510***	−0.000694***	−0.000217***	−0.000363***	−0.000352***
CO detector	−0.00187	−0.579	0.167	0.892**	0.745	−0.0732	−5.398***	−0.931	0.0366	0.445
First aid kit	0.00213	−0.193	0.846**	−0.0192	0.441	0.106	−0.880*	0.603	0.667**	−0.119
Smoke detector	0.192	−0.0531	1.278***	−0.154	−0.662	−0.0279	3.495***	1.15	0.192	0.327
Safety card	0.31	2.413***	0.0507	−0.101	−0.924**	−0.341	2.920***	−1.386**	−0.389*	−0.903***
Paid parking	2.348***	1.609***	3.641***	2.189***	1.776***	2.478***	4.099***	3.673***	3.866***	1.776***
Coffee maker	2.110***	1.259	2.940***	0.973**	0.367	−1.107***	−0.454	3.844***	0.723*	2.315***
Essentials	1.246***	2.580***	0.118	0.558***	2.160***	1.271***	3.270***	0.362	1.028***	1.747***
Shampoo	1.201***	1.785***	0.731***	0.269*	1.733***	1.006***	1.851***	1.133***	0.672***	0.801***
Washer	−2.029***	−1.555***	−2.330***	−0.133	−1.650***	−1.520***	−4.450***	−0.598	−1.393***	−0.729**
Microwave	−1.279*	0.331	−0.0998	−0.0652	1.519**	2.294***	4.048***	.872	1.544***	−0.206
Bedroom l.	4.288***	4.565***	2.031***	1.279***	2.373***	3.383***	3.274***	1.855***	1.739***	2.524***
Private entrance	1.392***	1.079	3.048***	1.694***	1.634***	1.139***	3.782***	5.117***	2.725***	2.673***
Air conditioner	0.0297	0.116	−1.098	0.682	0.22	−0.273	0.133	−0.019	1.398***	−0.491*
Extra pillows	1.882***	3.524***	0.65	1.556***	1.304**	0.820**	−0.186	0.741	0.542	1.071**
Luggage	1.816***	0.433	1.356**	1.330***	−0.385	1.542***	−0.329	0.406	0.391	−0.367
Self-check-in	2.437**	1.602*	4.338***	2.447***	1.583**	3.063***	3.313***	2.179**	4.510***	3.015***
24-h check-in	−1.437***	−2.217***	−1.866***	−1.025***	−2.608***	−1.834***	−4.518***	−2.919***	−1.150***	−2.414***
Wi-Fi	−1.143*	2.568***	1.415***	0.451*	1.569***	0.996***	2.695***	0.977**	0.972***	0.907**
Internet	−2.430***	−3.914***	−2.455***	−1.604***	−4.005***	−2.858***	−6.284***	−3.580***	−2.313***	−3.516***
Pocket Wi-Fi	−0.761	−0.944	1.683	1.505	−0.0503	−0.381	3.114**	−0.254	2.508***	1.506
Family/kid	−0.828	−0.907*	−0.607**	−0.212	−1.073***	−1.471***	−0.647	−0.638	−1.300***	−0.0148
Smoking	−0.0278	−1.170**	−0.364	−0.658***	−2.534***	−0.760***	−2.219***	−1.388***	−0.421***	−2.020***
Constant	1.55***	4.443***	2.826***	1.638***	3.398***	3.556***	2.435*	0.621	2.302***	1.080*
N	13,221	9102	14,809	18,745	13,987	43,667	10,078	11,987	37,374	20,453
Adj. R²	0.571	0.531	0.563	0.449	0.548	0.446	0.504	0.526	0.502	0.552

Note: Standard errors are heteroscedasticity robust.

*** Significant at 1%; **Significant at 5%; *Significant at 10%.

In general, the sign of the parameters that accompany the control variables is the one that was predicted; although they may need some additional clarification in cerain cases. For example, no special correlation is observed between the number of bathrooms and review growth. This is because apartments with more bathrooms are attractive but are usually not the type of accommodation that is booked by the typical Airbnb guests and might have extra charges and requirements that make them unlikely to be booked by short-term, low-cost guests. Something similar can be said for entire homes. Likewise, for the initial price, the relationship is inverse with review growth in all cases except in Copenhagen and Madrid, alluding to the typical laws of demand and supply by which more expensive accommodation can be purchased by fewer people. The relationship is less clear for ΔPrice, reflecting different reactions by hosts in front of demand movements.

Of greater interest to the objectives that have been set out in this article is to discuss the results relating to the accommodation reservation policy. On this matter, there are theoretical and empirical arguments that support the thesis that owners use the most flexible booking options (flexible cancellation and instant booking) to increase the attractiveness of their accommodation and compensate, to some extent, for the lack of other attributes. This would explain why flexible booking policies are priced negatively (Benítez-Aurioles, 2018b). Our results show, on the one hand, that flexible cancellation policy is inversely correlated with review growth in all cases, except in Italian cities (Milan and Rome) where the coefficients are not statistically significant. As explained in our theoretical model, this might mean that the cancellation force dominates. That is, although the flexible cancellation option increases the demand for accommodation in principle, it is precisely the fact that guests can cancel the reservation what weighs more to explain the inverse relationship between the increase in reviews and this option. While disabling the policy in favor of stricter ones in our period of interest is positive in all cases, it is not so significantly so for enabling it. Perhaps, the kind of accommodation that enables it after having experienced numerous visits while having it disabled are looking to tone down an already elevated demand.

On the other hand, Instant booking is always positively influencing review growth, either in the case of having it at the start or of enabling it during the period. This makes it evident that instant booking makes it easier to obtain guests, both because of the ease for guests to book without having to wait and because of the practicality of the hosts not having to watch to confirm them. In fact, Airbnb (2020) shows a special interest in promoting the instant booking option, stating that they often get double the reservations. Moreover, the type of hosts that enable this option might be more likely to be available, which is in line with our theory of an existing disutility parameter of enabling instant booking that varies across owners.

The estimations regarding the quadratic form of the influence of the initial reviews on review growth point to a concave function, of the form q*Reviews² + l*Reviews where q < 0 and l > 0 for all cities. The function is likely to be strictly increasing in reality, given that estimations of a are small; or, put otherwise, the numbers at which the influence starts to weaken are unrealistically large. Specifically, for the cities in our sample, these values would be (rounded upwards): 464, 401, 273, 433, 397, 280, 260, 597, 390, and 400 for Amsterdam, Barcelona, Berlin, Copenhagen, Lisbon, London, Madrid, Milan, Paris, and Rome, respectively. When we compute the increase that having, for example, 10 reviews would suppose on future reviews in the following 365 days, we obtain surprisingly consistent results across cities: 2.64, 3.25, 3.62, 2.31, 3.26, 2.79, 3.77, 2.58, 2.80, and 2.79, in the same order. We observe the biggest effect in Madrid and the smallest in Copenhagen but all of these values are within a range of 1.5 reviews. The effect is clearly not one-for-one or multiplicative, which alleviates concern about herding or an extreme unequal distribution of demand. The interaction coefficients are to be added to the review effect. That is, the base computations we have made for the joint effect of Reviews and Reviews² correspond to an accommodation without flexible cancellation and without instant booking. When these options are on, the effect of reviews is positively amplified, though by a small amount (the largest boosts being of 0.11 per review over 365 days when flexible is activated in Copenhagen and of 0.12 for instant in London).

Importantly, even after controlling for many important characteristics of a listing, our findings corroborate that the existence of previous reviews—irrespectively of their content—has a positive influence on future visits. Guests seem to prefer accommodations that have been visited before for reasons of reliability. An unvisited accommodation might make the prospective guest suspect that they may get scammed. Although remote communication is supposed to take place before the trip due to the social nature of Airbnb, this being a possible channel for the host to incite trust; it is apparently not so good as to counter the effect of reviews.

At the same time, it has to be accepted that besides herding behavior (captured by the parameter in front of the preexisting number of reviews), there are other vectors that influence the orientation of the demand. Actually, a radical approach of the herding model would imply an ever-increasing concentration of demand in a part of the supply. By contrast, we encounter a reduction in the coefficient of variation—measured as the ratio of the standard deviation to the mean—of reviews in all cities during the period under study (data available upon request to the authors). This means that reviews, at least in the accommodations that are present in both these starting and ending dates, become more equally distributed.

Our results challenge a naïve interpretation of individual behavior where each person mimics others’ behavior ignoring their own information (Eyster and Rabin, 2010). The evidence presented confirms that the demand for accommodation in the p2p market depends, not only on previous decisions but also on the attributes of the accommodation—and, especially, on the reservation policy that can offset the tendency to copy behavior. In addition, the fact that the number of current reviews (as a signal of demand) eventually stops being relevant to explain further review growth suggests the existence of congestion effects, or in general, of the exhaustion of the herding behavior after exceeding a certain threshold.

Conclusions

The reputation and trust needed to enable transactions in p2p markets for tourist accommodation rely, to a large extent, in a review system, like in other Internet-operating markets. That is, they depend on the subjective opinions expressed by participating agents. Nevertheless, this system has important weaknesses, among which we find the upward bias of valuations. Given that this risks making review content useless, their number becomes the relevant variable. As proven by empirical evidence, the number of reviews explains the number of bookings better than ratings do. Herding behavior-based models provide a sound explanation of this phenomenon. As agents do not know with certainty the quality of a given accommodation, they look at decisions previously taken in the market—which is a rational behavior at an individual level but might result in inefficiencies on the aggregate. A simplistic depiction of herding behavior would imply an ever-growing concentration of demand, leading to monopolistic positions and imposing an impenetrable barrier to new hosts that wish to enter the market. This is not what is observed in reality.

We have built a profit- and utility-maximization microeconomic model in which we incorporate a series of variables that, jointly, can dampen or even compensate the herding effect. Our estimation strategy has been executed with data from the Airbnb platform corresponding to listings in the 10 European cities where Airbnb is most popular, and they confirm that there are other variables besides the number of reviews, such as price, physical capacity proxies, distance to the center, the superhost status, and reservation policy that are significant in explaining the increase in demand for some accommodations—with the last category being of particular importance because of its large coefficients. We also observe their analogous relationship to prices, which is not always obvious; and the change in behavior of hosts between the beginning and end of the period. Airbnb gives the opportunity to people with little to no expertise on running an accommodation business to do so, on a reduced scale. Although the freedom in certain management options is offered for their convenience, hosts may make mistakes in choosing what is best for their business.

Another observation is that the number of reviews itself has indeed a positive influence in the future number of bookings, meaning that the herding effect is present. However, the influence of the marginal review decreases as the number of reviews is higher, preventing an ever-increasing concentration of bookings in certain listings. The key takeaway of our findings is that in p2p markets for accommodation, in spite of highly prevalent asymmetric information issues and imperfections in established trust and reputation mechanisms, it is still possible for new entrants to compete with incumbent suppliers. In particular, the instant book option can be the lever needed to abandon the vicious circle of not receiving visits because of not having been visited before and thereby to enter the opposite dynamic, which is also favored by the decreasing marginal effect of reviews as the stock of these grows.

These findings represent a novel contribution to both the tourism literature and the tourism industry. Regarding tourism literature, it illustrates, for the first time, how herding behavior works in the p2p market for tourist accommodation. In this sense, the results of the econometric analysis carried outfit within a model in which the agents choose their actions in function of not just their own information but also the decision of others, as signals. In particular, against the bandwagon effects induced by herding behavior, other variables contribute to define the choice of accommodation. Additionally, it has been found that, once a certain level has been reached, the capacity of these effects to influence demand is exhausted; and, therefore, other attributes of the accommodation (among which the reservation policy plays a special role) determine the final choice of potential guests.

On the other hand, for the tourism industry, in general, and for the owners of accommodation in the p2p market, in particular, the results of this research contain practical implications that can help improve management. The relative influence of herding behavior on demand induces those who have achieved occupation in the past to pay attention to the characteristics of their listing to improve or consolidate their position in the market. Likewise, for accommodation that has not yet been occupied and that, therefore, according to a simplistic version of herding behavior, would have a low probability of being reserved by potential guests, there is an opportunity to become active in the market. In this context, the booking policy and, more specifically, the instant booking option would grant a high potential to new entrants to join the p2p market for tourist accommodation effectively.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Beatriz Benítez-Aurioles

References

Airbnb (2017) June Update on One Host, One Home: New York City. Available at: https://2sqy5r1jf93u30kwzc1smfqt-wpengine.netdna-ssl.com/wp-content/uploads/2017/06/OneHostOneHomeJune2017Update-1.pdf (accessed 7 June 2020).

Airbnb (2019) Airbnb Economic Impact. Available at: https://blog.atairbnb.com/economic-impact-airbnb/ (accessed 7 June 2020).

Airbnb (2020) Business is Better with Instant Book. Available at: https://www.airbnb.ca/host/instant (accessed 7 June 2020).

Andreu

Bigne

Amaro

, et al. (2020) Airbnb research: an analysis in tourism and hospitality journals. International Journal of Culture, Tourism and Hospitality Research 14(1): 2–20.

Arbel

Pizam

(1977) Some determinants of urban hotel location: the tourists’ inclinations. Journal of Travel Research 15(3): 18−22.

Banerjee

(1992) A simple model of herd behavior. Quarterly Journal of Economics 107(3): 797−817.

Belarmino

Koh

(2020) A critical review of research regarding peer-to-peer accommodations. International Journal of Hospitality Management 84: 102315.

Benítez-Aurioles

(2018a) The role of distance in the peer-to-peer market for tourist accommodation. Tourism Economics 24(3): 237−250.

Benítez-Aurioles

(2018b) Why are flexible booking policies priced negatively? Tourism Management 67: 312–325.

10.

Bridges

Vásquez

(2018) If nearly all Airbnb reviews are positive, does that make them meaningless?. Current Issues in Tourism 21(18): 2057–2075.

11.

Cantallops

Salvi

(2014) New consumer behavior: a review of research on eWOM and hotels. International Journal of Hospitality Management 36: 41−51.

12.

Chen

Chang

Y-C

(2018) What drives purchase intention on Airbnb? Perspectives of consumer reviews, information quality, and media richness. Telematics and Informatics 35(5): 1512−1523.

13.

Cheng

Jin

(2019) What do Airbnb users care about? An analysis of online review comments. International Journal of Hospitality Management 76: 58−70.

14.

Dann

Teubner

Weinhardt

(2019) Poster child and guinea pig-insights from a structured literature review on Airbnb. International Journal of Contemporary Hospitality Management 31(1): 427−473.

15.

Dolnicar

(2019) A review of research into paid online peer-to-peer accommodation: launching the annals of tourism research curated collection on peer-to-peer accommodation. Annals of Tourism Research 75: 248–264.

16.

Einav

Farronato

Levin

(2016) Peer-to-peer markets. Annual Review of Economics 8: 615−635.

17.

Ert

Fleischer

Magen

(2016) Trust and reputation in the sharing economy: the role of personal photos in Airbnb. Tourism Management 55: 62−73.

18.

Eyster

Rabin

(2010) Naive herding in rich-information settings. American Economic Journal: Microeconomics 2(4): 221−43.

19.

Fernández

MCL

Bedia

AMS

(2004) Is the hotel classification a good indicator of hotel quality? An application in Spain. Tourism Management 25(6): 771−775.

20.

Fradkin

Grewal

Holtz

, et al. (2015) Bias and reciprocity in online reviews: evidence from field experiments on Airbnb. In: Proceedings of the 18th ACM Conference on Economics and Computation. June 2015, Portland, Oregon, USA, New York: ACM.

21.

Gibbs

Guttentag

Gretzel

, et al. (2018) Pricing in the sharing economy: a hedonic pricing model applied to Airbnb listings. Journal of Travel and Tourism Marketing 35(1): 46−56.

22.

Gössling

Zeiss

Hall

, et al. (2018) A cross-country comparison of accommodation manager perspectives on online review manipulation. Current Issues in Tourism 22(14): 1744−1763.

23.

Gunter

(2018) What makes an Airbnb host a superhost? Empirical evidence from San Francisco and the Bay Area. Tourism Management 66: 26–37.

24.

Gunter

Önder

(2018) Determinants of Airbnb demand in Vienna and their implications for the traditional accommodation industry. Tourism Economics 24(3): 270−293.

25.

Gutiérrez

García-Palomares

Romanillos

, et al. (2017) Airbnb in tourist cities: comparing spatial patterns of hotels and peer-to-peer accommodation. Tourism Management 62: 278−291.

26.

Guttentag

(2019) Progress on Airbnb: a literature review. Journal of Hospitality and Tourism Technology 10(3): 233−263.

27.

Park

(2016) Which restaurant should I choose? Herd behavior in the restaurant industry. Journal of foodservice business research 19(4): 396−412.

28.

Hajibaba

Dolnicar

(2018) Regulatory reactions around the world. In: Dolnicar

(ed), Peer-to-Peer Accommodation Networks: Pushing the Boundaries. Oxford: Goodfellow Publishers, pp. 120−136.

29.

Hensens

(2015) The future of hotel rating. Journal of Tourism Futures 1(1): 69−73.

30.

Hill

(2015) The secret of Airbnb’s pricing algorithm. IEEE Spectrum 52(9): 32−58.

31.

H-HS

Kandampully

Devi

(2009) Relationships and Impacts of Service Quality, Perceived Value, Customer Satisfaction, and Image: An Empirical Study. The Service Industries Journal 29: 111–125. DOI: 10.1080/02642060802292932.

32.

InsideAirbnb (2018) Get the Data. Available at: http://insideairbnb.com/get-the-data.html (accessed 7 June 2020).

33.

(2017) Sharing Means Renting? An Entire-Marketplace Analysis of Airbnb. Available at: https://ssrn.com/abstract=2902840 (accessed 7 June 2020).

34.

Kumar

Goyal

(2015) Behavioural biases in investment decision making—a systematic literature review. Qualitative Research in Financial Markets 7(1): 88−108.

35.

Ladhari

Michaud

(2015) eWOM effects on hotel booking intentions, attitudes, trust, and website perceptions. International Journal of Hospitality Management 46: 36−45.

36.

Lee

Hyun

Ryu

, et al. (2015) An analysis of social features associated with room sales of Airbnb. In: Proceedings of the 18th ACM Conference Companion on Computer Supported Cooperative Work & Social Computing (CSCW’15 Companion), March 2015, Vancouver, BC, Canada, pp. 219−222.

37.

Lewis

(2011) Asymmetric information, adverse selection and online disclosure: the case of eBay motors. American Economic Review 101(4): 1535−46.

38.

Liang

Schuckert

Law

, et al. (2017) Be a ‘Superhost’: the importance of badge systems for peer-to-peer rental accommodations. Tourism Management 60: 454−465.

39.

Litvin

Goldsmith

Pan

(2018) A retrospective view of electronic word-of-mouth in hospitality and tourism management. International Journal of Contemporary Hospitality Management 30(1): 313−325.

40.

Liu

Park

(2015) What makes a useful online review? Implication for travel product websites. Tourism Management 47: 141−51.

41.

Magno

Cassia

Bruni

(2018) Please write a (great) online review for my hotel! Guests’ reactions to solicited reviews. Journal of Vacation Marketing 24(2): 148−158.

42.

Marqusee

(2015) Airbnb and San Francisco: Descriptive statistics and academic research. San Francisco Planning Department.

43.

Martin-Fuentes

(2016) Are guests of the same opinion as the hotel star-rate classification system? Journal of Hospitality and Tourism Management 29: 126−134.

44.

Mayzlin

Dover

Chevalier

(2014) Promotional reviews: an empirical investigation of online review manipulation. American Economic Review 104(8): 2421−2455.

45.

Park

(2017) An experimental investigation on the determinants of online hotel booking intention. Journal of Hospitality Marketing & Management 26(6): 627−643.

46.

Prayag

Ozanne

(2018) A systematic review of peer-to-peer (P2P) accommodation sharing research from 2010 to 2016: progress and prospects from the multi-level perspective. Journal of Hospitality Marketing & Management 27(6): 649−678.

47.

Rajaguru

Hassanli

(2018) The role of trip purpose and hotel star rating on guests’ satisfaction and WOM. International Journal of Contemporary Hospitality Management 30(5): 2268−2286.

48.

Sainaghi

(2020) The current state of academic research into peer-to-peer accommodation platforms. International Journal of Hospitality Management 89: 102555.

49.

Sainaghi

Köseoglu

Mehraliyev

(2020) The intellectual structure of the sharing economy. Tourism Economics. DOI: 10.1177/1354816620920434.

50.

San Francisco Budget and Legislative Analyst (2015) Policy analysis report re: Analysis of the impact of short-term rentals on housing. Memo to San Francisco Supervisor David Campos. Available at: https://www.sfbos.org/Modules/ShowDOcument.aspx?documentid=52601

51.

Schuckert

Liu

Law

(2015) Hospitality and tourism online reviews: recent trends and future directions. Journal of Travel & Tourism Marketing 32(5): 608−621.

52.

Spyrou

(2013) Herding in financial markets: a review of the literature. Review of Behavioral Finance 5(2): 175−194.

53.

Ter Huurne

Ronteltap

Corten

, et al. (2017) Antecedents of trust in the sharing economy: a systematic review. Journal of Consumer Behaviour 16(6): 485−498.

54.

Teubner

Hawlitschek

Dann

(2017) Price determinants on Airbnb: how reputation pays off in the sharing economy. Journal of Self-Governance & Management 5(4): 53−80.

55.

UNWTO (2015) Hotel Classification Systems: Recurrence of Criteria in 4 and 5 Stars Hotels. Available at: https://www.e-unwto.org/doi/pdf/10.18111/9789284416646 (accessed 7 June 2020).

56.

Volgger

Taplin

Pforr

(2019) The evolution of ‘Airbnb-tourism’: demand-side dynamics around international use of peer-to-peer accommodation in Australia. Annals of Tourism Research 75: 322−337.

57.

Wang

Nicolau

(2017) Price determinants of sharing economy based accommodation rental: a study of listings from 33 cities on Airbnb.com. International Journal of Hospitality Management 62: 120−131.

58.

Xie

Mao

(2019) Learning from peers: the effect of sales history disclosure on peer-to-peer short-term rental purchases. International Journal of Hospitality Management 76: 173−183.

59.

Zervas

Proserpio

Byers

(2015) A First Look at Online Reputation on Airbnb, where Every Stay is Above Average. Available at: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2554500 (accessed 7 June 2020).