Modeling the Effect of Images on Product Choices

Conjoint is one of the most popular methods in marketing research, widely used to understand how customers trade-off features of a product. Since product images have a strong influence on customer choice, it is natural to include images in conjoint studies. However, estimating the effect of images on conjoint choices has proven difficult, especially when there are a large number of images and preferences for images are heterogeneous across consumers. This paper proposes a novel specification that accounts for the effect of images on respondents’ choices. Heterogeneity in the appeal of the images is modeled through the covariance structure in a probit model so that images that appeal to the same people have correlated utility and are more substitutable. The covariance structure is informed by a separate task where respondents rate the images, allowing the method to scale to a large number of images. This ratings data can also be replaced with physical measurements of distance between the images. In our application to midsize crossover vehicles, we show that the proposed model makes better predictions about product substitution particularly for ``sister" products that are visually similar. We discuss how this approach could be used to parsimoniously characterize the effect of other product attributes such as sound quality, or taste