Temporal Aggregation in Diffusion Models of First-Time Purchase: Does Choice of Frequency Matter?

摘要

Consistent with work in the advertising response literature, the author addresses the time-interval bias present when estimating innovation models of new product growth and diffusion with discrete time-series data. Specifically, the author explores the theoretical and empirical implications of using varying data frequencies when estimating diffusion models using both nonlinear least squares (NLLS) and ordinary least squares (OLS). Parameter estimates across five consumer durables are obtained using annual, quarterly, and monthly data. The central conclusion is that the information gained and bias minimized by using seasonally adjusted quarterly data results in empirical estimates that are an improvement over those obtained by using annual data. This is true for both the NLLS and OLS estimates. In contrast, the move from quarterly to monthly data produces only marginal statistical improvement.