The effect of calorie posting regulation on consumer opinion: A flexible latent Dirichlet allocation model with informative priors

摘要

Due to rising obesity, restaurants across United States must prominently display calories for every item on the menu. Previous research shows that regulations pertaining to health claims on food labels affect consumer search and behavior in various ways. The article focuses on consumers' post-consumption opinions of the product. The data used are 761,962 reviews of 9,805 chain and standalone restaurants in New York City, posted on a leading restaurant review site in an eight-year period from the website's inception. Based on the data, an automated and scalable probabilistic model that summarizes the large volume of free, unsolicited, rich user-generated reviews into a few interpretable topics is proposed. The model belongs to a class of probabilistic topic models termed latent Dirichlet allocation models to analyze words in large sets of original text to discover themes or topics within. Based on the summary of a large collection of reviews into a few representative latent topics, the topics were characterized by a probability distribution over all words in reviews. For the position of each word in a review, a topic is first chosen. This process continues until the review adequately represents the topics of interest of the writer. Each review is composed of a random mixture of several topics. The model isolates the causal effects of the regulation on consumer opinions in chain restaurants, by controlling for differences in characteristics between chain and standalone restaurants, for differences in reviews of all restaurants between the two time periods (before and after the regulation) and for geographical differences in topic proportions. Several additional tests for robustness of causal inference were also done. Results show that the mandatory calorie posting regulation can be seen as a success in making health more salient in the minds and voices of consumers. It was also found that textual content posted in online consumer reviews affects subsequent demand. And increased discussion of health in online reviews can serve as a basis for changes in consumer buying behavior. Analysis also showed that widespread health discussion can provide insights into consumer segments that might influence restaurant choices of the population. (38 refs.)