Noisy signals: Do ratings' volatility depend on the length of the consumption span?

Boto-García, David; Leoni, Veronica Working Paper Noisy signals: Do ratings' volatility depend on the length of the consumption span? Quaderni - Working Paper DSE, No. 1183 Provided in Cooperation with: University of Bologna, Department of Economics Suggested Citation: Boto-García, David; Leoni, Veronica (2023) : Noisy signals: Do ratings' volatility depend on the length of the consumption span?, Quaderni - Working Paper DSE, No. 1183, Alma Mater Studiorum - Università di Bologna, Dipartimento di Scienze Economiche (DSE), Bologna, https://doi.org/10.6092/unibo/amsacta/7230 This Version is available at: https://hdl.handle.net/10419/282305 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc/4.0/ ISSN 2282-6483 Noisy signals: do ratings’ volatility depend on the length of the consumption span? David Boto-García Veronica Leoni Quaderni - Working Paper DSE N°1183 1 Noisy signals: do ratings’ volatility depend on the length of the consumption span? David Boto-García1 Veronica Leoni*2,3 Abstract: This paper investigates the informational content of online reviews. For the case of hotels, we model how the length of the stay shapes the variance of review scores. Grounded on violations of temporal monotonicity, errors in recall and hedonic adaptation theories, we first present a characterization of how the consumption span affects the non-deterministic component of consumer satisfaction. Next, we conduct an empirical analysis using more than 525,000 individual reviews from Booking.com in 5 major European cities. Under a heteroskedastic framework, we document that individual ratings’ volatility decreases with the length of the stay. This implies that online ratings from short stayers (short consumption episodes) are noisy signals of the underlying hotel quality. Furthermore, we show that greater volatility in hotel ratings translates into a lower share of useful reviews for subsequent consumers. Our findings offer relevant insights for platform design operators about the sources of ratings’ volatility and how this affects social learning. Keywords: online reviews; ratings’ variance; length of stay; quality uncertainty; heteroskedasticity; Booking.com JEL codes: D12; D83; Z30. * Corresponding author: e-mail: veronica.leo[email protected]. 1 Department of Economics, University of Oviedo (Spain) 2 Center for Advanced Studies in Tourism, University of Bologna (Italy) 3 Department of Applied Economics, University of the Balearic Islands (Spain) 2 Non-technical summary While online user-generated content has become a valuable tool for ex-ante assessment of service quality, the volatility of scores that different consumers assign to each hotel can affect the usefulness of average scores. This article discusses how the length of a hotel stay can affect the variance of individual ratings. Hotel scores with low variance offer a more consistent and credible signal about expected quality, but the retrospective assessment of hedonic episodes is subject to cognitive and psychological biases. The duration of the episode plays a relevant role, making some dimensions more salient to memory than others. The paper evaluates the role of guests' length of stay on the variance of individual hotel ratings. The study uses data from 525,000 individual reviews for 1,233 hotels in five major European cities, and the findings suggest that the relative importance of the random component of overall scores (variance) decreases with length of stay. This finding implies greater consistency and lower polarization in ratings among long-stayers. The study also examines heterogeneity by consumer profile and relates the predicted variance from the model to the share of reviews that other consumers deem as 'useful'. The polarization of reviews is economically meaningful for several reasons. A higher rating variance can reduce demand and sales, and small amounts of misperceptions through polarization can produce important breakdowns when aggregating information in contexts of social learning. Understanding the sources of ratings' polarization thus has important implications for social learning dynamics in online platforms since user learning strongly depends on the size and congruity of information. This study expands existing literature in two important ways. Firstly, it characterizes how the length of stay affects the non-deterministic component of rating scores. Secondly, it adds to an emerging literature on review helpfulness. The study provides empirical evidence that long-stayers assign less polarized scores, which could impact how hotels should be rated and marketed to potential guests. Overall, this study highlights the importance of considering the length of stay when evaluating hotel ratings and quality assessments. The findings suggest that hotels may want to consider ways to incentivize longer stays, as they may lead to more consistent and reliable ratings. Furthermore, the study could provide insights into how to design better review platforms that facilitate social learning and decision-making. 3 1. INTRODUCTION The provision of experiential services, such as a hotel stay, is inherently characterized by a degree of uncertainty regarding the expected quality. In addition to official hotel star ratings, the proliferation of online platforms such as Booking.com has further facilitated the reduction of information asymmetries and the ‘market for lemons’ problem (Akerlof, 1970). In this sense, online user-generated content is nowadays one of the most relevant informational tools for product search and ex-ante assessment of service quality (Magnani, 2020). However, the effectiveness of online reviews and ratings as information sources crucially depend on the volatility in the scores that different consumers assign to each hotel. The variance reflects the extent to which individual ratings diverge from the average and can be understood as a measure of consistency that shapes ex-ante expectations. If the variance in ratings for a hotel is high (greater polarization), it may be more difficult for consumers to accurately gauge its overall quality. That is, the usefulness of average scores is diminished when consumers’ report both good and bad experiences. On the contrary, for the same mean rating, hotel scores with low variance offer a more consistent and credible signal about expected quality. When rating a hotel stay, tourists consider both objective and subjective (affective) dimensions and compare the quality of the service with prior expectations (Oliver, 1980; Engler et al., 2015). However, the retrospective assessment of hedonic episodes (remembered utility) is subject to cognitive and psychological biases (Kahneman et al., 1997). When evaluating temporarily extended outcomes, the duration of the episode plays a relevant role as it makes some dimensions more salient to memory than others. In the context of hotels, ratings made by long-stayers are based on a deeper knowledge about the services provided and therefore more reflective of the underlying hotel quality. At the same time, when consuming heterogeneous goods, people form summary evaluations giving comparatively more weight to positive and negative deviations from expectations during short stays because these shocks (i) tend to dissipate during long consumption spans through hedonic treadmill (Rayo and Becker, 2007), and (ii) are comparatively more salient to memory (Mullainathan, 2002). 4 This paper evaluates the role of guests’ length of the stay on the variance (dispersion from the average) of individual hotel ratings. Previous works have shown that the duration of tourists’ stay at destination affects post-trip satisfaction because of changes in destination perceptions during the course of the vacation experience (e.g., Vogt and Andereck, 2003). For the case of hotels, length of stay has been shown to matter for the likelihood of posting numerical and written online reviews (Kim and Han, 2022) and to be negatively associated with mean rating values (Brandes and Dover, 2022; Leoni and Moretti, 2023). However, to our knowledge, there is no evidence to date about how the stay duration affects the volatility (and therefore polarization) of hotel rating scores. This paper precisely aims to address this gap. Firstly, we characterize the potential mechanisms through which length of stay allegedly influences the variance of ratings from a theoretical viewpoint. We then conduct an empirical analysis using data from 525,000 individual reviews for 1,233 hotels located in 5 major European cities (Madrid, Barcelona, Lisbon, Rome and Milan) and listed on Booking.com platform. Based on a multiplicative heteroskedastic regression (Harvey, 1976) in which both the conditional mean and variance of ratings are modelled, we show that the relative importance of the random component of overall scores (variance) decreases with length of stay. This finding implies a greater consistency and lower polarization in ratings among long-stayers. In doing so, we also examine heterogeneity by consumer profile. Subsequently, we relate the predicted variance from our model to the share of reviews (at the hotel level) that other consumers deem as ‘useful’. We provide evidence of a clear negative relationship between polarization and usefulness, which offers important insights for theory and practice. The polarization of reviews is economically meaningful for several reasons. Firstly, previous works for the case of books (Sun, 2012), restaurants (Wu et al., 2015) and hotels (Ye et al., 2009) have shown that a higher rating variance reduces demand and sales. Secondly, small amounts of misperceptions through polarization can produce important breakdowns when aggregating information in contexts of social learning (Frick et al., 2020). From this viewpoint, understanding the sources of ratings’ polarization has important implications for social learning dynamics in online platforms since user learning strongly depends on the size and congruity of information (Acemoglu et al., 2022). 5 This work expands existing literature in two important ways. On the one hand, we characterize how the length of the stay affects the non-deterministic component of rating scores. Specifically, our framework posits that length of stay as an indicator of the duration of the consumption episode shapes the variance of ratings through a mixture of errors in recall (Mullainathan, 2002), violations of temporal monotonicity (Kahneman and Thaler, 2006) and potential hedonic treadmill (Rayo and Becker, 2007). Based on a large dataset of individual ratings on Booking.com platform, we provide empirical evidence that long-stayers assign less polarized scores. On the other hand, our work adds to an emerging literature on review helpfulness (Lee et al., 2021; Liu et al., 2023; Mudambi and Schuff, 2010; Zhang et al., 2023; Zhao et al., 2013) by showing how greater ratings’ volatility dampers the informational content of online reviews, making the signal about expected quality noisier and less useful. The remainder of the paper is structured as follows. Section 2 reviews the related literature. Section 3 presents the theoretical framework for the analysis. Section 4 describes the data and presents some summary statistics and preliminary evidence together with the econometric modelling. Section 5 reports and discusses the estimation results together with some robustness checks and extensions. Finally, Section 6 concludes with a summary of findings, implications and limitations. 2. LITERATURE REVIEW 2.1.Online reviews as quality cues Product ratings on online platforms are nowadays a major informational source for consumers when choosing among alternative providers (Wu et al., 2015). Electronic word of mouth reduces search costs and offers up to date information, thereby exerting a strong influence on market demand and revenues for different goods and services (Chevalier and Mayzlin, 2006; Liu, 2006). In presence of asymmetric information, people learn about product quality from both numerical ratings and contextual user-generated comments (Fang, 2022). These ratings can be understood as indicators of remembered experienced utility (hedonic quality) from previous consumers in the sense of Kahneman et al. (1997). Accordingly, agents learn from the public disclosure of 6 information by peers (Amador and Weill, 2012), being the reliance on observational learning more prevalent among infrequent and unexperienced costumers (Cai et al, 2009). Moreover, consumers have been shown to rely more on average ratings than on other quality cues like prices or the number of reviews (de Langue et al., 2016). In the hospitality industry, subjective ratings have displaced objective classification systems as reputation signals, and hotel managers are nowadays monitoring consumer reviews on online platforms (Proserpio and Zervas, 2017). A large body of literature has investigated the relevance and pervasiveness of online reviews to consumers in many different settings and from different viewpoints. Scholars in marketing have focused on consumers’ motives to review whereas economists have been more concerned about the effect of reviews on sales and demand. A review of the state of art can be found in Magnani (2020). A common finding is that reviews and ratings are affected by selection effects since individuals with extreme levels of satisfaction (very low or very high) are more likely to rate products than those with moderate evaluations (Moe and Schweidel, 2012; Schoenmueller et al., 2020). 4 Among them, very satisfied consumers are relatively more prone to review than dissatisfied ones, partly due to manipulation practices (Mayzlin et al., 2014). In this vein, there is some evidence of upward bias in online valuations because scores tend to be J-shaped (Pourfakhimi et al., 2020), with most suppliers receiving very high rates (Zervas et al., 2021). One explanation for the ratings inflation is that, when acting as reviewers, individuals are subject to herding behavior as theoretically conceptualized in Banerjee (1992): they are highly influenced by previously posted ratings by other consumers, with reviewer experience acting as a relevant moderator (Sunder et al., 2019). Moreover, they tend to conform to the average score, which acts as an anchor, leading to the so-called rating bubbles phenomenon (Moe and Trusov, 2011). Importantly, herding behaviour is asymmetric: people are comparatively more influenced by excellent rather than low ratings (Cicognani et al., 2022; Moe and Schweidel, 2012). 4 Another source of selection is differential attrition, by which reviewers with moderate experiences are more likely to exit the pool of active reviewers (Brandes et al., 2022). 7 2.2.Noisy reviews and heteroskedastic ratings Despite online reviews are a tool for disclosing quality information for experience goods, the degree of informativeness of such content strongly depends on their polarization. In a recent paper, Acemoglu et al. (2022) characterize learning dynamics from online reviews, showing that more information does not necessarily lead to faster learning; in fact, user learning would strongly depend on the size and congruity of information. Consumers are predicted to purchase a product/good with greater probability under moderate rather than extreme disagreement due to mismatch costs (Lee et al., 2023). In this vein, high-variance reviews are more dampening for expensive products (Kim and Krishnan, 2015) and exert a different influence on consumption propensity depending on subjective prior expectations about the good (West and Broniarczyk, 1998). Several works have shown that high variance reduces demand because polarization makes the quality signal noisier and, in turn, less helpful (Lee et al., 2021; Mudambi and Schuff, 2010; Park and Park, 2013; Sun, 2012; Ye et al., 2009; Zhao et al., 2013). This stream of literature documents that product type moderates the relationship between review variance and helpfulness, with experience goods being more sensitive to polarization. Most work on this matter has mainly evaluated Amazon products like DVDs, PC video games, cell phones and digital cameras (Lee et al., 2021; Mudambi and Schuff, 2010). For the case of hotels, a growing body of research has analysed the drivers of review usefulness. This literature has shown that informative and readable reviews from consumers with high reputation are deemed as more helpful (Liang et al., 2019; Liu and Park, 2015). Review diversity (different ratings for each item that composes the overall score) improves usefulness for the case of negative reviews (Liu et al., 2023) whereas high arousal makes reviews less helpful (Chatterjee, 2020). Moreover, emoticons enhance usefulness when the review is narrative-based (Huang et al., 2020). Nevertheless, only a few have explicitly considered the role of the variance in ratings scores. Ye et al. (2009) estimate that a 10% increase in review variance decreases the number of bookings by around 2.8%. Lo and Yao (2019) prove experimentally that review credibility is higher when the ratings are consistent with previous reviews. Zhang et al. 14 We define three types of room categories (economy, standard and superior), which represent 5%, 41%, and 10% of the sample, respectively. The remaining 43% collapse into a fourth category labelled as ‘other’. As regards the country of origin, there is high heterogeneity in nationality (see Appendix, Table A1 for the full list), with only 19% of the total reviews being from domestic travellers. Concerning travel party composition, 40% of reviewers travel in couples, 23% with family members and 17% in groups. The rest (19%) are solo travellers. Around 10% of reviews are from anonymous guests, almost 28% of reviews were written on weekends and more than 85% in the same month the stay took place. Because some studies have pointed that the temporal distance between consumption and reviewing affects review positivity (e.g., Brandes and Dover, 2022), we define a dummy variable labelled Temp. contiguity that takes value 1 for reviews written close to the stay (same month). Table 1. Summary statistics of the variables Label Description Mean (%) SD Min. Max. Score Rating score 8.648 1.445 2.5 10 Num.reviews (t-1) Stock of reviews received up to t-1 655.74 928.58 0 7,139 Av.score (t-1) Average score of stock of reviews up to t-1 8.66 0.52 2.5 10 SDscore (t-1) Standard deviation of stock of reviews up to t-1 1.30 0.30 0 3.97 LOS Nights at the hotel (length of stay) 2.62 1.55 1 15 Economy =1 if room=economy 5.29 Standard =1 if room=standard 41.37 Superior =1 if room=superior 10.48 Other =1 if other type of room 42.86 Anonymous =1 if left by anonymous guest 10.18 Temp. contiguity =1 if month stay=month review 85.55 Weekend =1 if review is left on a weekend 27.55 Couple =1 if travel party=couple 40.53 Family =1 if travel party= family 22.24 Group =1 if travel party= group 17.66 Solo =1 if travel party= solo traveler 19.55 Domestic =1 if domestic guest 18.90 Observations 525,437 Based on the individual reviews, we computed the stock of reviews received by each hotel in the sample per period. Next, each individual review in the sample was assigned the corresponding hotel review stock up to one-month before the stay. The same was done for the mean and the variance of overall scores. These variables are relevant to control for in the analysis as they capture the quantity, value, and dispersion of ratings at the time of booking, which could influence postconsumption individual ratings through ex-ante expectations and anchoring. These three variables 15 vary over time depending on the date of the stay (and the booking), aiming at capturing learning dynamics and selection effects based on available information in the spirit of Acemoglu et al. (2022). On average, the stock of reviews one-month before the stay is 655, with an overall average score of 8.6 and a standard deviation of 1.30 points. As a first step, we visually inspect whether there is a link between ratings’ variance and the length of the stay. To this end, we compute the standard deviation of rating scores for each hotel in the sample. Subsequently, we calculate the mean length of stay per hotel. Figure 1 presents a binned scatterplot of the pairwise correlation between the two. As illustrated there, it seems there is a negative association: there is lower volatility in ratings’ scores in hotels whose customers stay for longer. However, because this analysis is done using aggregate data, we cannot disentangle the influence of length of stay on ratings’ variance from hotel quality or other sources of heterogeneity associated with long stays. To offer a more detailed characterization, we move to a formal econometric analysis. Figure 1. Descriptive binned scatterplot of SD score per hotel on mean LOS per hotel 16 Note: the bins are defined based on the 100 quantiles of the distribution of the variables 4.3.Econometric modelling Consistent with the model presented in Section 3, we estimate the following Harvey-type heteroskedastic regression model (Harvey, 1976) at the individual review level: 𝑠𝑐𝑜𝑟𝑒𝑖𝑗𝑡 = 𝛼 + 𝛽1𝑁𝑢𝑚.𝑟𝑒𝑣𝑖𝑒𝑤𝑠𝑗(𝑡−1) + 𝛽2𝐴𝑣.𝑠𝑐𝑜𝑟𝑒𝑗(𝑡−1) + 𝛽3𝑆𝐷𝑠𝑐𝑜𝑟𝑒𝑗(𝑡−1) + 𝛿𝐿𝑂𝑆𝑖𝑗𝑡 +𝛾𝑋𝑖𝑗𝑡 + 𝜔𝑅𝑜𝑜𝑚.𝑡𝑦𝑝𝑒𝑖𝑗𝑡 + 𝐻𝑗+ 𝜏𝑡+ 𝜆1𝑇𝑒𝑚𝑝.𝐶𝑜𝑛𝑡𝑖𝑔𝑢𝑖𝑡𝑦𝑖𝑗𝑡 + 𝜆2𝑊𝑒𝑒𝑘𝑒𝑛𝑑𝑖𝑗𝑡 + 𝜖𝑖𝑗𝑡 (3) where i indexes individual reviews, j the hotel and t the month of the review, 𝑁𝑢𝑚.𝑟𝑒𝑣𝑖𝑒𝑤𝑠𝑗𝑡−1, 𝐴𝑣.𝑠𝑐𝑜𝑟𝑒𝑗𝑡−1 and 𝑆𝐷𝑠𝑐𝑜𝑟𝑒𝑗𝑡−1 capture expected quality through the quantity (stock), level and dispersion of pre-existing ratings (up to one month before the stay); 𝐿𝑂𝑆𝑖𝑗𝑡 is the guest’s length of stay at hotel j in period t; 𝑋𝑖𝑗𝑡 are tourist travel party composition and country of origin fixed effects capturing heterogeneity in preferences associated with cultural traits and travel distance (Litvin, 2019; Mariani and Predvoditeleva, 2019); 𝑅𝑜𝑜𝑚.𝑡𝑦𝑝𝑒𝑖𝑗𝑡 are dummy indicators for the type of room, capturing quality and price differences within the same hotel that might impact consumer satisfaction; 𝐻𝑗 are hotel fixed effects controlling for objective mean quality differences across hotels; 𝜏𝑡 are monthly fixed effects; 𝑇𝑒𝑚𝑝.𝐶𝑜𝑛𝑡𝑖𝑔𝑢𝑖𝑡𝑦𝑖𝑗𝑡 and 𝑊𝑒𝑒𝑘𝑒𝑛𝑑𝑖𝑗𝑡 are the dummies defined before; and 𝜖𝑖𝑗𝑡 is a random error term capturing the emotional component of satisfaction that is allowed to be heteroskedastic as follows: 𝜖𝑖𝑗𝑡~𝑁(0,𝜎𝜖2) 𝜎𝜖2=𝑒𝑥𝑝(𝜂 + 𝜋𝐿𝑂𝑆𝑖𝑗𝑡) (4) 17 The additive structure of remembered utility in (2) implies that the value of 𝜎𝜖2 informs about the relative importance of the stochastic over the deterministic components. In the extreme case that 𝜎𝜖2= 0, ratings would be fully deterministic. As long as 𝜎𝜖2 becomes greater, scores are noisier. The model in (3)-(4) is a linear regression with heteroskedastic errors that is estimated in one step by Maximum Likelihood. The estimates from the two-step Generalized Least Squares (henceforth GLS) estimator proposed by Harvey (1976) are also presented as a robustness check (see subsection 5.3). The exponential transformation for 𝜎𝜖2 ensures the variance is positive for all possible values of 𝜂 + 𝜋𝐿𝑂𝑆𝑖𝑗𝑡. The key parameter of interest is 𝜋, which captures how the variability of scores depends on tourists’ length of stay at the hotel. 6 5. RESULTS 5.1.Main findings Table 2 reports the coefficient estimates for the model in (3)-(4). Standard errors are clustered at the hotel level to capture potential cross-correlation in individual reviews for the same hotel over time. The country-of-origin fixed effects are visually presented in Figures 2 and 3. 7 The monthly and hotel fixed are omitted to save space but are available upon request. 8 In the upper part of Table 2, we report the results for the mean equation. Columns 1-3 consider 𝑁𝑢𝑚.𝑟𝑒𝑣𝑖𝑒𝑤𝑠𝑗𝑡−1, 𝐴𝑣.𝑠𝑐𝑜𝑟𝑒𝑗𝑡−1 and 𝑆𝐷𝑠𝑐𝑜𝑟𝑒𝑗𝑡−1 in the regression separately, whereas Column 4 includes the three together. There is high consistency in the estimates across specifications so what follows focuses on the results from the full specification in Column 4. We document that 6 One could argue that the censored nature of the dependent variable would need a Tobit estimator rather than OLS for the expected value of scores. We prefer to use a linear regression for the mean equation since Tobit is known to suffer from incidental parameter bias under high-dimensional fixed effects (Greene, 2004). Moreover, unlike OLS, Tobit renders inconsistent estimates when the disturbances are non-normal (Arabmazar and Schmidt, 1982). 7 There is a total of 228 origin countries in the data. We include dummies for 83 origin countries, which represent 98.5% of the sample. The reference category considers all the remaining origins, which have few observations each. 8 See Figure A1 in Appendix for a histogram of the hotel fixed effects for Column 4 in Table 2. 18 scores are uncorrelated with the pre-consumption stock volume of reviews. This might be explained by the hotel fixed effects already capturing level differences in reviews across hotels. Nevertheless, we find that, conditional on hotel fixed effects, higher average scores from past guests translate into lower individual ratings. Specifically, a unit increase in the mean score the month before the stay is associated with a 0.112-point decrease in the individual rating. This result can be attributed to feelings of disappointment when expectations are not met (Mitchell et al., 1997) and closely relates to the ‘good news-bad news’ paradigm. When consumers observe that past evaluations for a hotel are relatively high (low), individuals set up high (low) expectations before consumption. Theory predicts that good news (experienced utility is greater than expected) are under-weighted relative to bad news (Eil and Rao, 2011; Nguyen and Claus, 2013). Therefore, unmet expectations through high benchmarks result in lower experienced utility (and vice versa). Interestingly, greater volatility in ratings the month before the stay is associated with higher reported level of satisfaction. In particular, a unit increase in SDScore leads to a 0.09 increase in individual scores, everything else being equal. This result is striking, as we would expect the opposite sign direction. Indeed, the raw data suggests an (unconditional) negative association between pre-consumption ratings’ volatility and post-consumption individual scores. Since the estimate for SDScore is conditional on hotel fixed effects, it needs to be interpreted as follows: for the same hotel and mean score, greater pre-consumption inconsistency among reviews is associated with higher post-consumption ratings. This result could be potentially explained by people paying more attention to negative than to positive reviews in cases of high variability in scores. This sets lower expectations pre-consumption that might result in better valuations post-consumption given quality through anchoring. Concerning the role of stay duration on mean scores, an additional night decreases scores by 0.006 points, ceteris paribus. Nonetheless, although significant, the effect is quantitatively small. This result falls in line with prior works (Brandes and Dover, 2022; Leoni and Moretti, 2023), suggesting that long stayers exhibit lower reported utility. One argument could be that the longer a consumer stays, the more time he/she has to critically evaluate the services offered (Kim and Han, 2022); for instance, comfort levels might decrease over time in rooms that are generally designed for short stays. 19 20 Table 2. Results from heteroskedastic linear regression Clustered standard errors at the hotel level in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Regarding travel party composition, and compared to solo guests, those traveling in groups report higher satisfaction (+0.068). Nonetheless, no significant differences are found for couples and those staying with their family. Interestingly, domestic guests are, on average, less satisfied than (1) (2) (3) (4) Dep. Variable: score Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Num.reviews (t-1) -3.1e-06 -2.4e-06 (1.3e-05) (1.2e-05) Av.Score (t-1) -0.168*** -0.112*** (0.031) (0.041) SDScore (t-1) 0.163*** 0.087** (0.027) (0.037) LOS -0.006** -0.006** -0.006** -0.006** (0.003) (0.003) (0.002) (0.002) Travel party: couple -0.005 -0.005 -0.005 -0.005 (0.014) (0.014) (0.014) (0.014) Travel party: family -0.007 -0.007 -0.007 -0.007 (0.015) (0.015) (0.015) (0.015) Travel party: group 0.068*** 0.068*** 0.068*** 0.068*** (0.013) (0.013) (0.013) (0.013) Domestic 0.034*** 0.034*** 0.034*** 0.034*** (0.012) (0.012) (0.012) (0.012) Room type: economy 0.019 0.018 0.019 0.018 (0.028) (0.028) (0.028) (0.028) Room type: standard -0.012 -0.012 -0.012 -0.012 (0.013) (0.013) (0.013) (0.013) Room type: superior -0.016 -0.016 -0.016 -0.016 (0.016) (0.016) (0.016) (0.016) Anonymous -0.198*** -0.198*** -0.198*** -0.198*** (0.011) (0.011) (0.011) (0.011) Temp.Contiguity 0.058*** 0.058*** 0.058*** 0.058*** (0.008) (0.008) (0.008) (0.008) Weekend 0.002 0.002 0.002 0.002 (0.006) (0.006) (0.006) (0.006) Constant 7.241*** 8.638*** 7.064*** 8.076*** (0.036) (0.259) (0.044) (0.381) Hotel fixed effects YES YES YES YES Country of origin fixed effects YES YES YES YES Monthly fixed effects YES YES YES YES Variance equation LOS -0.021*** -0.021*** -0.021*** -0.021*** (0.004) (0.004) (0.004) (0.004) Constant 0.644*** 0.644*** 0.644*** 0.644*** (0.020) (0.020) (0.020) (0.020) Observations 525,437 525,437 525,437 525,437 21 foreigners (-0.034). In this vein, the country-of-origin fixed effects (Figures 2 and 3) point to important heterogeneity in reported utility by nationality. This falls in line with previous works documenting that cultural dimensions are relevant predictors of differences in rating scores (Litvin, 2019; Mariani and Predvoditeleva, 2019), plausibly through different quality benchmarks. Surprisingly, there are no significant differences in reported satisfaction associated to room quality. This finding indicates that, conditional on hotel fixed effects, experienced utility is unrelated to quality differences within hotels, on average. A potential explanation for this is that, as explained in Section 4, ratings are derived from the average of six items that capture hotel quality as a whole rather than the specific attributes of individual rooms. Figure 2. Country-of-origin fixed effects estimates from Column 4 in Table 2 (I) 22 Figure 3. Country-of-origin fixed effects estimates from Column 4 in Table 2 (II) In line with identity disclosure and deindividualization theories (Deng et al., 2021), we find that anonymous guests tend to provide lower ratings. More precisely, other things being equal, review anonymity is associated with a 0.198-point drop in scores. A plausible mechanism is that hiding their identities allows guests to share their honest opinions without the fear of being identified. Furthermore, the temporal proximity between the stay and the review date (temporal contiguity) is associate with higher scores (+0.058). This suggests that individuals report greater experienced utility for consumption episodes that are temporally close to the time of the review. This relates to Mullainathan (2002) framework of biased memory limitations and rosy retrospection theory (Mitchell et al., 1997): temporal contiguity may contribute to better recall and remembrance of details, which tends to be biased towards good aspects. 23 Moving to the variance equation, a LR test rejects the null hypothesis of homoscedastic errors (chi2(1)=29.8, p-value<0.001). 9 The coefficient estimate for LOS is negative and statistically significant, indicating that the variance decreases with the length of the stay at the hotel. This result is in line with our theoretical argument, showing that shorter consumption episodes result in more volatile ratings. We interpret it as evidence that ratings by short stayers are more likely to integrate negative and positive shocks. In contrast, evaluations from long-stayers are more deterministic, hence mostly driven by objective dimensions. Accordingly, our findings confirm that the length of stay shapes ratings polarization, with short stays providing noisy signals of the true (objective) hotel quality. Based on the coefficient estimates, we have computed the predicted standard deviation of the error component (𝜎𝜖 ) as the square root of 𝜎𝜖2 , with 𝜎𝜖2 =𝑒𝑥𝑝(𝜂 + 𝜋𝐿𝑂𝑆𝑖𝑡). Table A3 in Appendix presents how 𝜎𝜖  varies with length of stay in the sample. We see that the predicted standard deviation in scores decreases from 1.37 for one-night guests to 1.18 for 15-night guests. 5.2.Heterogeneity by consumer profile The pooled analysis presented before might mask relevant heterogeneity associated with guests’ profile. For instance, the role of length of stay and ex-ante quality signals on individual ratings might vary depending on consumers’ nationality (domestic versus international guests) or whether the individual travels alone, in a couple or with the family/in a group. To examine this, Table 3 presents the results from separate regressions. 10 9 Table A2 in Appendix presents the results from a baseline linear regression under the assumption of homoscedastic errors. A Breusch-Pagan test rejects the null hypothesis of constant variance (chi(1)=32380.8, p-value<0.0001), suggesting the presence of heteroskedastic errors and the suitability of our modelling approach. 10 Figures A2 and A3 Appendix present binned scatterplots of the residualized conditional mean relationship between scores and length of stay by nationality and travel party composition. We use Frisch-Waugh-Lovell theorem to visually present the differences in slopes by subsamples conditional on all the remaining controls, hotel, monthly and country- of-origin fixed effects. 30 service quality tend to vanish throughout longer consumption spans. Moreover, memory errors in recall à la Mullainathan (2002) are likely to play a role, since long stays are more accurately remembered. Our analysis also reveals that higher overall scores from previous guests, which set higher expectations about service quality beforehand, are associated with lower individual ratings. In line with the good news-bad news paradigm and social learning theories, post-consumption individual ratings are highly affected by ex-ante expectations. Unmet expectations induce consumers to ‘punish’ the hotel through negative reported satisfaction. Additionally, we have documented substantial heterogeneity in ratings depending on the travel party composition, the nationality of the guest and the latency between the stay and the review time. Our findings are pretty robust, as they remain consistent under the new review system implemented by Booking.com since September 2019. This study makes two contributions to the existing literature on online reviews. Firstly, we offer the first analysis on how the consumption span shapes ratings’ volatility. Existing studies have primarily paid attention to mean ratings, with only a few analysing the role played by the length of the stay at the hotel (e.g., Brandes and Dover, 2022). In doing so, we have shown that long-stayers are more deterministic in their valuations, whereas the ratings by short-stayers are noisier and potentially more affected by emotional (unobserved) factors. Secondly, this study adds to the growing literature on the instrumental value of online reviews as a social learning tool in contexts of quality uncertainty. Existing studies that have analysed reviews’ usefulness have mainly looked at the number of words used in the textual comment, the presence of pictures, or the reviewers’ expertise. We add novel evidence on this respect, documenting that ratings’ consistency in terms of low dispersion is an important metric for prospective consumers at the time of judging the informativeness of user generated content. Our findings offer some relevant implications, particularly for online platforms. Given the nonneutral effects of online reputation for firms’ performance (Chevalier and Mayzlin, 2006) and rating behaviours (Cicognani et al., 2022), platforms should consider the possibility of assigning different weights to reviews based on the length of stay. This could help to mitigate the effects of 31 polarization and improve the accuracy of online evaluations. Stated differently, the weight allotted to each review in calculating the overall score should be based on the guest's level of exposure to the hotel facilities, as this serves as a proxy for their subjective understanding of the intrinsic quality. As an alternative, another valuable possibility is to incorporate length of stay (short vs long stay) as a criterion for filtering reviews. Platforms typically provide users the option to sort reviews based on different criteria, such as the language, overall score, time of year or the type of travel party. Adding this duration filter may enhance the usefulness of ratings to consumers by allowing them to focus on the valuations made by consumers with similar stay duration, which might hence share more similar needs. Furthermore, the negative relationship between length of stay and rating scores indicates that hotels must adapt their services more carefully to the specific needs and requirements of long stayers, who represent an important market segment. The paper has some limitations that should be acknowledged. First, there is scope for potential bias from reviewers’ self-selection: consumers who submit an online review are not a random sample of the population and tend to be those exhibiting extreme opinions (Schoenmueller et al., 2020). This is a common limitation when working with review data. Nonetheless, we believe our results are still highly informative. Even if not truly representative of the population of consumers, it is existing reviews what matters for social learning and subsequent sales. Second, we lack information about actual rates paid by hotel guests, the time in advance with which the booking was made, and the leisure versus business motive of the stay. Future work should expand our analysis by considering the role played by these factors. 32 REFERENCES Acemoglu, D., Makhdoumi, A., Malekian, A. and Ozdaglar, A. (2022). Learning from reviews: the selection effect and the speed of learning. Econometrica, 90(6), 2857-2899. Akerlof, G.A. (1970). The market for 'lemons': quality uncertainty and the market mechanism. The Quarterly Journal of Economics, 84(3), 488-500. Amador, M. and Weill, P.O. (2012). Learning from private and public observations of others' actions. Journal of Economic Theory, 147, 910-940. Anderson, L.R. and Holt, C.A. (1997). Information cascades in the laboratory. American Economic Review, 87(5), 847-862. Arabmazar, A. and Schmidt, P. (1982). An investigation of the robustness of the Tobit estimator to nonnormality. Econometrica, 50(4), 1055-1063. Banerjee, A.V. (1992). A simple model of herd behavior. The Quarterly Journal of Economics, 107(3), 797- 817. Brandes, L. and Dover, Y. (2022). Offline context affects online reviews: The effect of post-consumption weather. Journal of Consumer Research, 49(4), 595-615. Brandes, L., Godes, D. and Mayzlin, D. (2022). Extremity bias in online reviews: The role of attrition. Journal of Marketing Research, 59(4), 675-695. Brunner-Sperdin, A. and Peters, M. (2009). What influences guests' emotions? The case of high-quality hotels. International Journal of Tourism Research, 11, 171-183. Cai, H., Chen, Y. and Fang, H. (2009). Observational learning: Evidence from a Randomized Natural Field Experiment. American Economic Review, 99(3), 864-882. Chatterjee, S. (2020). Drivers of helpfulness of online hotel reviews: A sentiment and emotion mining approach. International Journal of Hospitality Management, 85, 102356. Chevalier, J.A. and Mayzlin, D. (2006). The effect of word of mouth on sales: online book reviews. Journal of Marketing Research, 43(3), 345-454. Cicognani, S., Figini, P. and Magnani, M. (2022). Social influence bias in ratings: A field experiment in the hospitality sector. Tourism Economics, 28(8), 2197-2218. de Langhe, B., Fernbach, P.M. and Lichtenstein, D.R. (2016). Navigating by the stars: Investigating the actual and perceived validity of online user ratings. Journal of Consumer Research, 42(6), 817-833. Deng, L., Sun, W., Xu, D. and Ye, Q. (2021). Impact of anonymity on consumers' online reviews. Psychology & Marketing, 38(12), 2259-2270. Eil, D. and Rao, J.M. (2011). The good news-bad news effect: asymmetric processing of objective information about yourself. American Economic Journal: Microeconomics, 3(2), 114-138. Engler, T.H., Winter, P. and Schulz, M. (2015). Understanding online product ratings: A customer satisfaction model. Journal of Retailing and Consumer Services, 27, 113-120. 33 Eurostat (2019). Annual data on trips of EU residents. Retrieved from https://ec.europa.eu/eurostat/ on March 9th, 2023. Fang, L. (2022). The effects of online review platforms on restaurant revenue, consumer learning and welfare. Management Science, 68(11), 8116-8143. Figini, P., Vici, L. and Viglia, G. (2020). A comparison of hotel ratings between verified and non-verified online review platforms. International Journal of Culture, Tourism and Hospitality Research, 14(2), 157-171. Frick, M., Iijima, R. and Ishii, Y. (2020). Misinterpreting others and the fragility of social learning. Econometrica, 88(6), 2281-2328. Galiani, S., Gertler, P.J. and Undurraga, R. (2018). The half-life of happiness: hedonic adaptation in the subjective well-being of poor slum dwellers to the satisfaction of basic housing needs. Journal of the European Economic Association, 16(4), 1189-1233. Greene, W. (2004). The behaviour of the maximum likelihood estimator of limited dependent variable models in the presence of fixed effects. Econometrics Journal, 7, 98-119. Guizzardi, A., Mariani, M. M., and Stacchini, A. (2022). A temporal construal theory explanation of the price-quality relationship in online dynamic pricing. Journal of Business Research, 146, 32-44. Harvey, A.C. (1976). Estimating regression models with multiplicative heteroscedasticity. Econometrica, 44(3), 461-465. Huang, S., Chang, C.T., Bilgihan, A. and Okumus, F. (2020). Helpful or harmful? A double-edged sword of emoticons in online review helpfulness. Tourism Management, 81, 104135. Kahneman, D., Fredrickson, B., Schreiber, C.M. and Redelmeir, D. (1993). When more pain is preferred to less: Adding a better end. Psychological Science, 4(6), 401–405. Kahneman, D. and Thaler, R.H. (2006). Utility maximization and experienced utility. Journal of Economic Perspectives, 20(1), 221-234. Kahneman, D., Wakker, P.P. and Sarin, R. (1997). Back to Bentham? Explorations of experienced utility. The Quarterly Journal of Economics, 112(2), 375-406. Kim, J.M. and Han, J. (2022). Impact of the length of stay at hotels on online reviews. International Journal of Contemporary Hospitality Management, 34(4), 1249-1269. Kim, Y. and Krishnan, R. (2015). On product-level uncertainty and online purchase behavior: An empirical analysis. Management Science, 61(10), 2449-2467. Lee, N., Bollinger, B. and Staelin, R. (2023). Vertical versus horizontal variance in online reviews and their impact on demand. Journal of Marketing Research, 60(1), 130-154. Leibenstein, H. (1950). Bandwagon, snob, and Veblen effects in the theory of consumer demand. The Quarterly Journal of Economics, 64(2), 183-207. Leoni, V. and Moretti, A. (2023). Customer satisfaction during COVID-19 phases: the case of the Venetian hospitality system. Current Issues in Tourism, https://doi.org/10.1080/13683500.2022.2164709 34 Liang, S., Schuckert, M. and Law, R. (2019). How to improve the stated helpfulness of hotel reviews? A multilevel approach. International Journal of Contemporary Hospitality Management, 31(2), 953-977. Litvin, S.W. (2019). Hofstede, cultural differences, and TripAdvisor hotel reviews. International Journal of Tourism Research, 21, 712-717. Liu, Z. and Park, S. (2015). What makes a useful online review? Implication for travel product websites. Tourism Management, 47, 140-151. Lo, A.S. and Yao, S.S. (2019). What makes hotel online reviews credible? An investigation of the roles of reviewer expertise, review rating consistency and review valence. International Journal of Contemporary Hospitality Management, 131(1), 41-60. Liu, X., Li, C., Nicolau, J.L. and Han, M. (2023). The value of rating diversity within multidimensional rating system: Evidence from booking platform. International Journal of Hospitality Management, 110, 103434. Liu, Y. (2006). Word of mouth for movies: its dynamic and impact on box office revenue. Journal of Marketing, 70, 74-89. Loewenstein, G. (2000). Emotions in economic theory and economic behavior. American Economic Review, 90(2), 426-432. Magnani, M. (2020). The economic and behavioral consequences of online user reviews. Journal of Economic Surveys, 34(2), 263–292. Mariani, M.M. and Borghi, M. (2018). Effects of the Booking.com rating system: Bringing hotel class into the picture. Tourism Management, 66, 47-52. McFadden, D. (1974). Conditional Logit Analysis of Qualitative Choice Behavior. In Paul Zarembka (ed.). Frontiers Econometrics, New York: Academic Press. Mellinas, J.P. and Martin-Fuentes, E. (2021). Effects of booking. com's new scoring system. Tourism Management, 85, 104280. Mitchell, T.R., Thompson, L., Peterson, E. and Cronk, R. (1997). Temporal adjustments in the evaluation of events: The “rosy view”. Journal of Experimental Social Psychology, 33(4), 421-448. Moe, W.W. and Schweidel, D.A. (2012) Online product opinions: incidence, evaluation, and evolution. Marketing Science, 31(3), 372–386. Moe, W.W. and Trusov, M. (2011). The value of social dynamics in online product ratings forums. Journal of Marketing Research, 48, 444-456. Mariani, M. and Predvoditeleva, M. (2019). How do online reviewers' cultural traits and perceived experience influence hotel online ratings? An empirical analysis of the Muscovite hotel sector. International Journal of Contemporary Hospitality Management, 31(12), 4543-4573. Mayzlin, D., Dover, Y. and Chevalier, J. (2014). Promotional reviews: an empirical investigation of online review manipulation. American Economic Review, 104(8), 2421-2455. Mudambi, S. and Schuff, D. (2010). What makes a helpful online review? A study of customer reviews on amazon.com. MIS Quarterly, 34, 185–200. 35 Mullainathan, S. (2002). A memory-based model of bounded rationality. The Quarterly Journal of Economics, 117(3), 735-774. Nelson, P. (1970). Information and consumer behavior. Journal of Political Economy, 78(2), 311-329. Nguyen, V.H. and Claus, E. (2013). Good news, bad news, consumer sentiment and consumption behavior. Journal of Economic Psychology, 39, 426-438. Oliver, R.L. (1980). A cognitive model of the antecedents and consequences of satisfaction decisions. Journal of Marketing Research, 17(4), 460-469. Oswald, A.J. and Powdthavee, N. (2008). Does happiness adapt? A longitudinal study of disability with implications for economists and judges. Journal of Public Economics, 92, 1061-1077. Park, s. and Park, D. (2013). The effect of lowversus high-variance in product reviews on product evaluation. Psychology & Marketing, 30(7), 543-554. Pollak, R.A. and Wales, T.J. (1981). Demographic variables in demand analysis. Econometrica, 49(6), 1533-1551. Pokryshevskaya, E.B. and Antipov, E.A. (2017). Profiling satisfied and dissatisfied hotel visitors using publicly available data from a booking platform. International Journal of Hospitality Management, 67, 1-10. Pourfakhimi, S., Duncan, T. and Coetzee, W.J.L. (2020). Electronic word of mouth in tourism and hospitality consumer behaviour: state of the art. Tourism Review, 75(4), 637-661. Proserpio, D. and Zervas, G. (2017). Online reputation management: estimating the impact of management responses on consumer reviews. Marketing Science, 36(5), 645-665. Rabin, M. and Schrag, J.L. (1999). First impressions matter: a model of confirmatory bias. The Quarterly Journal of Economics, 114(1), 37-82. Rayo, L. and Becker, G.S. (2007). Evolutionary efficiency and happiness. Journal of Political Economy, 115(2), 302-337. San Martín, H. and Rodríguez-del-Bosque, I.A. (2008). Exploring the cognitive-affective nature of destination image and the role of psychological factors in its formation. Tourism Management, 29(2), 263-277. Schoenmueller, V., Netzer, O. and Stahl, F. (2020). The polarity of online reviews: prevalence, drivers and implications. Journal of Marketing Research, 57(5), 853-877. Schreiber, C. A. and Kahneman, D. (2000). Determinants of the remembered utility of aversive sounds. Journal of Experimental Psychology: General, 129(1), 27–42. Sirakaya, E., Petrick, J. and Choi, H. (2004). The role of mood on tourism product evaluations. Annals of Tourism Research, 31(3), 517-539. Sun, M. (2012). How does the variance of product ratings matter? Management Science, 58(4), 696-707. Sunder, S., Kim, K.H. and Yorkston, E.A. (2019). What drives herding behavior in online ratings? The role of rater experience, product portfolio, and diverging opinions. Journal of Marketing, 83(6), 93-112. 36 Thurstone, L.L. (1927). Three psychophysical laws. Psychological Review, 34(6), 424–43 UNWTO (2019). Yearbook of Tourism Statistics. Retrieved from: https://www.unwto.org/tourismstatistics/key-tourism-statistics. Accessed on March 9, 2023. Vogt, C.A. and Andereck, K.L. (2003). Destination perceptions across a vacation. Journal of Travel Research, 41, 348-354. West, P.M. and Broniarczyk, S.M. (1998). Integrating multiple opinions: The role of aspiration level on consumer response to critic consensus. Journal of Consumer Research, 25, 38–51. Wu, C., Che, H., Chan, T.Y. and Lu, X. (2015). The economic value of online reviews. Marketing Science, 34(5), 739-754. Xu, X. (2022). A growing or depreciating love? Linking time with customer satisfaction through online reviews. Information & Management, 59, 103605. Ye, Q., Law, R. and Gu, B. (2009). The impact of online user reviews on hotel room sales. International Journal of Hospitality Management, 28(1), 180–182. Zervas, G., Proserpio, D. and Byers, J.W. (2021). A first look at online reputation on Airbnb, where every stay is above average. Marketing Letters, 32, 1-16. Zhang, X., Zhang, X., Liang, S., Yang, Y. and Law, R. (2023). Infusing new insights: How do review novelty and inconsistency shape the usefulness of online travel reviews? Tourism Management, 96, 104703. Zhao, Y., Yang, S., Narayan, V. and Zhao, Y. (2013). Modeling consumer learning from online product reviews. Marketing Science, 32(1), 153-169. Zimmermann, S., Hernnmann, P., Kundisch, D. and Nault, B.R. (2018). Decomposing the variance of consumer ratings and the impact on price and demand. Information Systems Research, 29(4), 984-1002. 1 ONLINE APPENDIX Table A1. Descriptive statistics of number of reviews per country of origin of the guest Country Obs Abkhazia, Georgia 121 Afghanistan 8 Albania 362 Algeria 1114 Andorra 302 Angola 891 Argentina 15517 Armenia 172 Aruba 13 Australia 10680 Austria 6193 Azerbaijan 322 Bahamas 10 Bahrain 306 Bangladesh 76 Barbados 11 Belarus 723 Belgium 9645 Belize 6 Benin 9 Bermuda 17 Bolivia 180 Bonaire St Eustatius and Saba 2 Bosnia and Herzegovina 144 Botswana 15 Brazil 25076 Brunei Darussalam 13 Bulgaria 1525 Burkina Faso 6 Cambodia 41 Cameroon 9 Canada 5919 Cape Verde 74 Cayman Islands 16 Chad 5 Chile 4226 China 5333 Colombia 3403 Costa Rica 590 Croatia 1137 Cuba 6 Curaçao 17 Cyprus 695 Czech Republic 3500 Côte d'Ivoire 64 Democratic Republic of Congo 16 Denmark 2866 Djibouti 5 Dominican Republic 251 East Timor 9 Ecuador 554 Egypt 1174 El Salvador 142 2 Equatorial Guinea 19 Estonia 792 Ethiopia 16 Faroe Islands 14 Fiji 8 Finland 2807 France 40286 French Guiana 35 French Polynesia 31 French Southern Territories 6 Gabon 24 Georgia 477 Germany 27443 Ghana 34 Gibraltar 95 Greece 4451 Greenland 6 Grenada 6 Guadeloupe 60 Guam 5 Guatemala 305 Guernsey 32 Guinea 9 Guinea-Bissau 19 Haiti 15 Honduras 71 Hong Kong 1200 Hungary 3901 Iceland 466 India 1582 Indonesia 523 Iran 303 Iraq 179 Ireland 6498 Isle of Man 32 Israel 10593 Italy 76829 Jamaica 14 Japan 4638 Jersey 77 Jordan 364 Kazakhstan 726 Kenya 71 Kosovo 38 Kuwait 1477 Kyrgyzstan 42 Laos 11 Latvia 909 Lebanon 1001 Lesotho 7 Libya 169 Liechtenstein 53 Lithuania 1123 Luxembourg 1671 Macao 170 Madagascar 10 Malaysia 626 Maldives 25 Mali 10 Malta 1096 3 Martinique 38 Mauritania 20 Mauritius 92 Mayotte 8 Mexico 3324 Moldova 205 Monaco 141 Mongolia 10 Montenegro 205 Morocco 2654 Mozambique 265 Myanmar 20 Namibia 22 Nepal 23 Netherlands 12914 New Caledonia 65 New Zealand 1536 Nicaragua 19 Nigeria 110 North Macedonia 230 Norway 2873 Oman 313 Pakistan 397 Palestinian Territory 46 Panama 420 Papua New Guinea 5 Paraguay 205 Peru 1237 Philippines 579 Poland 6350 Portugal 23028 Puerto Rico 421 Qatar 1005 Reunion 226 Romania 4299 Russia 22077 Rwanda 10 San Marino 46 Saudi Arabia 4280 Senegal 62 Serbia 825 Seychelles 27 Singapore 1029 Slovakia 1435 Slovenia 585 South Africa 1625 South Korea 4370 Spain 66954 Sri Lanka 112 Sudan 18 Sweden 5657 Switzerland 13768 Syria 23 São Tomé and Príncipe 11 Taiwan 1579 Tajikistan 17 Tanzania 34 Thailand 819 Trinidad and Tobago 25 Tunisia 447 10 Table A6. Coefficient estimates from heteroskedastic linear regression using data for the period September 2019-February 2020 (Booking.com new rating system) Clustered standard errors at the hotel level in parentheses. *** p<0.01, ** p<0.05, * p<0.1 (1) (2) (3) (4) Dep. Variable: score Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Num.reviews (t-1) -2.4e-06 7.2e-06 (3.1e-06) (3.0e-06) Av.Score (t-1) -1.097*** -1.813*** (0.342) (0.284) SDScore (t-1) 0.898*** -0.260 (0.144) (0.168) LOS -0.013*** -0.013*** -0.013*** -0.013*** (0.003) (0.003) (0.003) (0.003) Travel party: couple 0.035*** 0.034*** 0.034*** 0.034*** (0.013) (0.013) (0.013) (0.013) Travel party: family 0.027* 0.026* 0.026* 0.025* (0.015) (0.015) (0.015) (0.015) Travel party: group 0.068*** 0.067*** 0.068*** 0.067*** (0.015) (0.015) (0.015) (0.015) Domestic -0.023 -0.024 -0.024 -0.025 (0.016) (0.016) (0.016) (0.016) Room type: economy 0.005 0.004 0.005 0.004 (0.023) (0.023) (0.023) (0.023) Room type: standard -0.031** -0.031** -0.031** -0.030** (0.014) (0.014) (0.014) (0.014) Room type: superior -0.032 -0.032 -0.032 -0.032 (0.023) (0.022) (0.023) (0.023) Room type: anonymous -0.193*** -0.193*** -0.193*** -0.194*** (0.015) (0.015) (0.015) (0.015) Temp.Contiguity 0.067*** 0.067*** 0.067*** 0.067*** (0.009) (0.009) (0.009) (0.009) Weekend 0.007 0.007 0.007 0.007 (0.007) (0.007) (0.007) (0.007) Constant 8.504*** 17.549*** 6.860*** 23.932*** (0.034) (2.821) (0.268) (2.581) Hotel fixed effects YES YES YES YES Country of origin fixed effects YES YES YES YES Monthly fixed effects YES YES YES YES Variance equation LOS -0.025*** -0.025*** -0.025*** -0.025*** (0.005) (0.005) (0.005) (0.005) Constant 0.794*** 0.794*** 0.794*** 0.794*** (0.022) (0.022) (0.022) (0.022) Observations 335,470 334,766 334,640 334,640 11 Table A7. Coefficient estimates from heteroskedastic linear regression with sequential addition of covariates Clustered standard errors at the hotel level in parentheses. *** p<0.01, ** p<0.05, * p<0.1 (1) (2) (3) (4) (5) Dep. Variable: score Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Num.reviews (t-1) -8.8e-06 -1.0e-05* -1.3e-05** -2.4e-06 (6.6e-06) (6.0e-06) (5.3e-06) (1.2e-05) Av.Score (t-1) 0.997*** 0.992*** 1.001*** -0.112*** (0.019) (0.019) (0.019) (0.041) SDScore (t-1) 0.179*** 0.177*** 0.192*** 0.087** (0.031) (0.031) (0.030) (0.037) LOS 0.006 -0.009*** -0.010*** -0.009*** -0.006** (0.005) (0.003) (0.003) (0.003) (0.002) Travel party: couple 0.005 -0.005 -0.005 (0.012) (0.012) (0.014) Travel party: family -0.001 -0.014 -0.007 (0.014) (0.014) (0.015) Travel party: group 0.050*** 0.059*** 0.068*** (0.012) (0.012) (0.013) Domestic -0.019** 0.052*** 0.034*** (0.010) (0.012) (0.012) Room type: economy 0.029 0.018 0.018 (0.023) (0.023) (0.028) Room type: standard 0.001 -0.002 -0.012 (0.009) (0.009) (0.013) Room type: superior 0.011 0.016 -0.016 (0.013) (0.012) (0.016) Room type: anonymous -0.200*** -0.192*** -0.198*** (0.010) (0.010) (0.011) Temp.Contiguity 0.042*** 0.060*** 0.058*** (0.008) (0.008) (0.008) Weekend -0.001 0.003 0.002 (0.006) (0.006) (0.006) Constant 8.631*** -0.195 -0.172 -0.384* 8.076*** (0.026) (0.198) (0.204) (0.201) (0.381) Hotel fixed effects NO NO NO NO YES Country of origin fixed effects NO NO NO YES YES Monthly fixed effects NO NO NO YES YES Variance equation LOS -0.020*** -0.019*** -0.019*** -0.020*** -0.021*** (0.004) (0.004) (0.004) (0.004) (0.004) Constant 0.790*** 0.673*** 0.671*** 0.659*** 0.644*** (0.023) (0.020) (0.020) (0.020) (0.020) Observations 526,130 526,130 526,130 525,437 525,437 12 Table A8. OLS regression results considering two-period lags of stock of reviews, average score and standard deviation of score (1) (2) (3) (4) Dep. Variable: score Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Num reviews (t-2) -7.9e-06 -6.1e-07 (1.2e-05) (1.2e-05) Av. Score (t-2) -0.220*** -0.181*** (0.032) (0.043) SD Score (t-2) 0.178*** 0.059 (0.032) (0.041) LOS -0.006** -0.006** -0.006** -0.006** (0.003) (0.003) (0.003) (0.003) Travel party: couple -0.005 -0.006 -0.006 -0.006 (0.014) (0.015) (0.015) (0.015) Travel party: family -0.007 -0.008 -0.008 -0.008 (0.015) (0.016) (0.016) (0.016) Travel party: group 0.068*** 0.069*** 0.069*** 0.069*** (0.013) (0.014) (0.014) (0.014) Domestic 0.034*** 0.032** 0.032** 0.032** (0.012) (0.013) (0.013) (0.013) Room type: economy 0.019 0.022 0.022 0.022 (0.028) (0.027) (0.027) (0.027) Room type: standard -0.012 -0.012 -0.012 -0.012 (0.013) (0.014) (0.014) (0.014) Room type: superior -0.016 -0.019 -0.019 -0.019 (0.016) (0.017) (0.017) (0.017) Room type: anonymous -0.198*** -0.201*** -0.201*** -0.201*** (0.011) (0.012) (0.012) (0.012) Temporal contiguity 0.058*** 0.052*** 0.052*** 0.052*** (0.008) (0.009) (0.009) (0.009) Weekend 0.002 0.001 0.001 0.001 (0.006) (0.006) (0.006) (0.006) Constant 7.237*** 9.129*** 7.191*** 8.766*** (0.035) (0.281) (0.037) (0.382) Hotel fixed effects YES YES YES YES Country of origin fixed effects YES YES YES YES Monthly fixed effects YES YES YES YES Variance equation LOS -0.021*** -0.021*** -0.021*** -0.021*** (0.004) (0.004) (0.004) (0.004) Constant 0.644*** 0.653*** 0.654*** 0.653*** (0.020) (0.021) (0.021) (0.021) Observations 525,437 468,109 467,692 467,692 Clustered standard errors at the hotel level in parentheses. *** p<0.01, ** p<0.05, * p<0.1 13 Table A9. Heteroskedastic regression results per city (1) (2) (3) (4) (5) City Barcelona Madrid Milan Rome Lisbon Dep. Variable: score Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Num reviews (t-1) 4.0e-05*** -5.4e-06 -1.4e-04*** -1.3e-05 2.0e-05 (1.1e-05) (4.4e-05) (4.6e-05) (3.9e-05) (1.3e-05) Av. Score (t-1) -0.213*** 0.022 -0.482*** -0.053 -0.090 (0.081) (0.126) (0.101) (0.056) (0.086) SD Score (t-1) 0.037 0.129 -0.027 0.133*** 0.214** (0.079) (0.179) (0.109) (0.047) (0.092) LOS -0.004 -0.010 -0.020** -0.006 -0.003 (0.005) (0.007) (0.008) (0.004) (0.004) Travel party: couple 0.024 0.037 -0.002 -0.029* -0.057*** (0.042) (0.023) (0.030) (0.017) (0.016) Travel party: family -0.009 0.072** -0.006 -0.026 -0.043** (0.046) (0.031) (0.031) (0.019) (0.018) Travel party: group 0.114*** 0.094*** 0.048 0.003 0.035** (0.039) (0.025) (0.035) (0.018) (0.018) Domestic 0.049 -0.120** 0.157** 0.103** 0.081* (0.053) (0.060) (0.065) (0.048) (0.046) Room type: economy 0.010 0.000 0.094 0.011 0.084* (0.054) (0.041) (0.123) (0.036) (0.046) Room type: standard -0.000 -0.021 -0.016 -0.002 -0.027 (0.035) (0.022) (0.035) (0.022) (0.023) Room type: superior -0.000 0.012 0.014 -0.033 -0.036 (0.053) (0.039) (0.039) (0.024) (0.024) Room type: anonymous -0.204*** -0.170*** -0.125*** -0.251*** -0.189*** (0.023) (0.023) (0.029) (0.021) (0.021) Temporal contiguity 0.032** 0.143*** 0.004 0.077*** 0.056*** (0.014) (0.016) (0.025) (0.018) (0.014) Weekend -0.005 0.013 0.032 0.003 -0.015* (0.012) (0.009) (0.024) (0.011) (0.009) Constant 9.963*** 6.942*** 11.948*** 7.539*** 10.231*** (0.786) (1.131) (0.939) (0.511) (0.881) Hotel fixed effects YES YES YES YES YES Country of origin fixed effects YES YES YES YES YES Monthly fixed effects YES YES YES YES YES Variance equation LOS -0.027*** -0.019* 0.014 -0.020*** -0.031*** (0.007) (0.011) (0.010) (0.006) (0.007) Constant 0.717*** 0.539*** 0.641*** 0.654*** 0.572*** (0.048) (0.047) (0.044) (0.031) (0.037) Observations 157,737 64,616 64,607 121,493 116,984 Clustered standard errors at the hotel level in parentheses. *** p<0.01, ** p<0.05, * p<0.1 14 Table A10. Heteroskedastic regression results from GLS estimator (1) (2) (3) (4) Dep. Variable: score Coef. (SE) Coef. (SE) Coef. (SE) Coef. (SE) Num reviews (t-1) -2.8e-06 -2.2e-06 (4.3e-06) (4.3e-06) Av. Score (t-1) -0.168*** -0.112*** (0.016) (0.021) SD Score (t-1) 0.164*** 0.088*** (0.016) (0.021) LOS -0.006*** -0.006*** -0.006*** -0.006*** (0.001) (0.001) (0.001) (0.001) Travel party: couple -0.005 -0.005 -0.005 -0.005 (0.006) (0.006) (0.006) (0.006) Travel party: family -0.007 -0.007 -0.007 -0.007 (0.006) (0.006) (0.006) (0.006) Travel party: group 0.068*** 0.068*** 0.068*** 0.068*** (0.006) (0.006) (0.006) (0.006) Domestic 0.034*** 0.033*** 0.033*** 0.033*** (0.008) (0.008) (0.008) (0.008) Room type: economy 0.018 0.018 0.018* 0.018 (0.011) (0.011) (0.011) (0.011) Room type: standard -0.012* -0.012* -0.012* -0.012* (0.006) (0.006) (0.006) (0.006) Room type: superior -0.016* -0.016* -0.016* -0.016* (0.009) (0.009) (0.009) (0.009) Room type: anonymous -0.198*** -0.198*** -0.198*** -0.198*** (0.007) (0.007) (0.007) (0.007) Temporal contiguity 0.058*** 0.058*** 0.058*** 0.058*** (0.006) (0.006) (0.006) (0.006) Weekend 0.002 0.002 0.002 0.002 (0.004) (0.004) (0.004) (0.004) Constant 7.268*** 8.664*** 7.090*** 8.097*** (0.945) (0.954) (0.945) (0.963) Hotel fixed effects YES YES YES YES Country of origin fixed effects YES YES YES YES Monthly fixed effects YES YES YES YES Variance equation LOS -0.032*** -0.031*** -0.032*** -0.031*** (0.002) (0.002) (0.002) (0.002) Constant 0.610*** 0.609*** 0.611*** 0.608*** (0.006) (0.006) (0.006) (0.006) Observations 525,437 525,437 525,437 525,437 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 