Jacob Goldenberg

Improving Diffusion Forecasts Using Social Interactions Data

Coauthor(s): Olivier Toubia, Rosanna Garcia.

Abstract:
We propose an approach for using data on social interactions (e.g., number of recommendations received by consumers, number of recommendations given by adopters, number of social ties) in order to improve the forecasts made by extant diffusion models. We extend major extant diffusion models to capture explicitly the generation of social interactions and their impact on adoption. In particular, we extend the discrete-time versions of the Mixed Influence Model (Bass model), the Asymmetric Influence Model, and the Karmeshu-Goswami Model. The extended models may be calibrated using a combination of social interactions data and penetration data. A field study conducted in collaboration with a Consumer Packaged Goods company suggests that the incorporation of social interactions data results in improved diffusion forecasts. The field study also suggests that the benefit of using social interactions data comes in great part from an improved ability to select, based on in-sample fit, the model that will produce the best forecasts.

Close Window