A pre-diffusion growth model of intentions and purchase

Abstract

In this paper, we investigate whether information on the history of purchase intentions is useful in predicting actual purchase behavior. The research is motivated by two factors. The first factor is the empirical finding in the literature that measuring intentions just prior to purchase provides better predictions of actual purchase as compared to when these intentions are measured earlier. The second factor is the role of the timing of the formation of intentions prior to purchase. While one stream of literature based on preference fluency predicts that early formation of intentions is more likely to lead to actual purchase, the other stream based on the memory-based “recency” effect predicts that formation of intentions just prior to purchase is more likely to lead to actual purchase. Together, these two factors motivate the potential need to account for the entire history of intentions prior to purchase. A canonical example of a market where intention histories are tracked is the movie industry, where “first choice” movie watching intentions are tracked up to (and in some cases beyond) the time of release. Accommodating the history of intentions in an econometric model that predicts actual box office performance is challenging due to the differing numbers of observations for the movies, the large numbers of observations for certain movies, as well as the role of various time-invariant and time-varying covariates influencing intentions. We propose a two-part model where the first part involves a hierarchical growth model that summarizes the trajectories of intentions via “growth factors.” These growth factors also reflect the role of the various covariates. The second part is a regression of the box office performance on the growth factors and other covariates. The models are simultaneously estimated within a Bayesian framework. Consistent with the previous literature, we find that including information on intentions improves our ability to predict behavior, with the recent intentions being the most informative. Importantly, when the history of intentions is accounted for, our results indicate that the data support the “recency” literature—intentions grow over time leading up to purchase, and this growth has a positive impact on opening box office performance. While a linear growth model performs best for most movies, there exists a subset of movies for which the quadratic growth model better captures the “spike” in intentions just prior to purchase. Further, accounting for information on the history of intentions dramatically improves model fit and forecasting performance relative to when only the intentions at one point in time (e.g., the ones just prior to purchase) are accounted for.