User-Generated Star Ratings Are Not Inherently Comparable 

With Nick Reinholtz

User-generated ratings—often elicited and presented as “star ratings”—are among the most influential sources of information for consumers online, but their ability to lead consumers to make utility-maximizing decisions is uncertain. 

This paper identifies an inherent, structural problem with star ratings: Ratings are created for single alternatives in isolation, but are used to make comparisons. As a result, ratings are affected by differences in raters’ frames of reference, which are meaningless to consumers. 

Through nearly 1 million quarterly observations of 343,327 Airbnb listings, we demonstrate one consequence of this—ratings vary over time due to variation in context, not quality. We then experimentally demonstrate this structural misalignment, showing that objectively superior alternatives can receive lower ratings than inferior competitors when the superior alternative engenders higher expectations. 

Link to paper

Code, data, and materials on OSF

Quality in Context: Evidence that Consumption Context Influences User-Generated Product Ratings.

With Nick Reinholtz

Using 218,918 ratings scraped from, we find that recent unseasonably cold weather at a reviewer's location leads them to rate cold-weather gear lower than during seasonable or warm temperatures. Other product ratings are unaffected by recent temperature: For example, ratings for bicycles are not impacted by weather. 

This suggests that ratings do not communicate objective quality alone-they communicate experienced quality, which is contextually influenced. We posit that this effect is akin to the correspondence bias for products, in that consumers fail to attribute variation in experience to situational factors. 

Further analysis of our REI data suggests that negative information becomes more prevalent with increased variation in consumption context. Through simulation, we find evidence that variation in context depresses ratings due to a ceiling effect, where the majority of ratings are near their cap. Finally, in an experiment, customer search is impacted by the seemingly noisy effect of bad weather on ratings.

Link to paper

Code, data, and materials on OSF

The OSF folder contains the Python and R code used to scrape reviews from REI, merge them with weather data, and clean them. It is a bit messy, but I hope it can be of use to someone. Please cite the working paper if you use the data or code. 

Is a (Money) Problem Shared, a Problem Halved? Investigating the Impact of Communication on Financial Anxiety

With Joe J. Gladstone and Emily N. Garbinsky

Money is a taboo subject in much of the Western world. But would encouraging consumers to more freely communicate about their money issues improve how they feel about their finances?  We hypothesized that when people more frequently discuss and share their financial problems they will experience a reduction in the anxiety they feels towards their finances. We support this hypothesis using multiple data sources, including the application of automated textual analysis on posts scraped from two online forums (N = 343,786 and 561,061), two surveys (N = 101,844 and 711), and a longitudinal diary study (N = 533, Nobs = 2,519) where we experimentally manipulate how participants communicate about money. Results indicate that talking about money benefits consumers by reducing feelings of stress and anxiety towards their finances. The diary study explores potential process explanations, such as the impact of receiving amounts of specific advice and/or emotional support from communications partners. Supplementary analyses find preliminary evidence for a moderating role for financial hardship, with those in greatest hardship benefiting most from talking about their finances.

Please email me for paper

Link to code/data/materials on OSF