Causal Effects of Product Type on Text Review Generation: Evidence from Double Machine Learning in E-Commerce

Online textual reviews provide richer product information than numerical ratings but require greater consumer effort, making it important to understand what drives their generation. This study examines how product type (search, experience, and credence goods) affects the product-level proportion of reviews containing textual feedback and whether these effects vary with product price and average rating. Using data from 890 Amazon products, we apply Double Machine Learning (DML) to estimate average and conditional treatment effects while accounting for high-dimensional product-level covariates. The results show that search goods are associated with a higher level of textual review generation relative to non-search goods, whereas experience goods exhibit a marginally lower level relative to non-experience goods. Heterogeneous treatment effects further provide exploratory indications that the positive effect of search goods tends to be greater among lower-priced and higher-rated products, whereas the negative effect of experience goods tends to become more negative as price increases but less negative as average rating rises. These findings indicate that product evaluability systematically shapes textual review generation and that its influence varies with contextual market cues. By applying a causal machine-learning framework, this study provides a more nuanced understanding of how product characteristics and market conditions jointly shape consumers’ textual contributions in online marketplaces.