Clash of reputation and status in online reviews

Abstract

This study extends the heterogeneous effectiveness of market signals by examining when textual sentiments have the most influence on purchasing decisions. Specifically, we argue that reputation and status, two distinct theoretical constructs, which are difficult to disentangle in practice, may influence the effectiveness of textual sentiments on customers’ decision making process in opposite directions. Reputation refers to the quality trajectory for a product whereas status sets a societal expectation from a product based on the social standing of that product among its peers. In this study, we examine reputation and status as contingencies that affect how electronic word of mouth (e-WoM) is perceived by customers in the context of review platform. To demonstrate the impact of textual sentiments and the moderation effects of reputation and status, we used an online platform to crawl review and reservation data at the same time of everyday over a period of 100 days on 310 hotels located in New York City. We found that customers are more sensitive to the sentiment of textual reviews on hotels of high status but less receptive when reviews are on hotels of high reputation. Our robustness tests and two identification strategies are all consistent with these findings. This research offers a strategic guideline to businesses and platforms in terms of how much they should rely on e-WoM, contingent upon their reputation and status.

Appendices

Appendix A: N-recent textual comments

See Table 8

Table 8 The interaction effect of price with average reviews’ sentiment

.

Appendix B: Machine learning pipeline for subjectivity score

We pre-processed a total of 62,000 reviews collected from Hotels.com. Before training a model, we pre-processed subjectivity training set data with a TF-IDF vector. Through tokenizing, lemmatization, and POS-tagging procedures, we first prepared the subjectivity training set using the movie review data. The training data set includes 5000 subjective review data from the Rotten Tomatoes pages and 5000 objective plot summary data from IMDb (Internet Movie Database). Then we prepared to predict the final subjectivity score by vectorizing the hotel review dataset with the same process. Using the Naïve Bayes classifier, we collected the subjectivity score at each review level. Our procedure of text-mining is graphically displayed in Fig. 9, and summarized statistics of text-mined variables are available with those of other hotel information in Table 1.

Fig. 9

Text-mining procedure of textual comments

Appendix C: The detailed examples of magnitude and sentiment score

See Table 9

Table 9 Example sentence of subjectivity score

.

Appendix D: Price effect with reputation, status, and sentiment

In addition to the opposite moderating roles of reputation and status on the impact of the textual comments on the sales, we also included the moderation of price with organizational attributes (price*status, price*reputation) and sentiment (price*average sentiment) in our models. We found that, after controlling the confounding effects of the price, our main findings, i.e., review sentiment effect and the moderation effects of reputation and status on the same, are qualitatively identical and even more clearly identified. In other words, with the consideration of the price, it is confirmed that reputation and status take opposite roles in moderating the effect of text review on the sales. We present our findings in Table 10.

Table 10 Additional interaction of price with status, reputation, and average reviews’ sentiment

Appendix E: Sub-sampled analysis for hotel groups

In this section, we conducted an additional two-by-two sub-sampled analysis by dividing the hotels into four groups depending on (1) whether a hotel belongs to a franchise or it is an independent hotel with a single brand and (2) whether the hotel’s star (reputation) level is above and below a certain range (3.5). As a result of the sub-sample analysis, we presented our results in Table 11. We used negative binomial regression with the same control variables that we used in our main models. The result shows that, out of the four groups, the impact of the average sentiment on the sales is maximum for the group of franchised hotels with hotel reputation less than 3.5 stars. On the other hand, the impact of average sentiment on the sales becomes minimal for the group of franchised hotels with higher than 3.5 star ratings. For the group of hotels with an independent single brand, the average sentiment increases the sales for both low and high star groups, although the effect is stronger for the low-ranking group. The results indicate that the sentiment of textual reviews plays a critical role in increasing the sales for independent hotels with a single brand. However, for franchises, the sentiment in textual reviews plays a role for only the hotels with less than 3.5 stars.

Table 11 Sub-sampled groups depending on hotel star-rating and chain-affiliation

Appendix F: Rating effect with reputation and status

In this section, we further examined how ratings independently work on sales in our dataset.The analysis results, which are presented in Table 12 show consistency as the effect of textual reviews. However, considering the ratings as static information accumulated and aggregated during certain periods of time, the rating information can not fully synchronize with the dynamics of the sales data which is changeable daily. In addition, the ratings are significantly correlated with reputation information(0.59), causing multicollinearity problems in the model. Therefore, we did not include ratings as a control variable in our main model for to prevent potential endogeneity issue. We further investigate the correlation between sentiment(average sentiment, the recent sentiment both) and reputation, and the correlation shows 0.25 and 0.14. As we examine the unique impact of reviews excluding rating information, we identify that both the base model and the moderation show more robust results.

Table 12 The impact of rating with status and reputation