Estimating diffusion and adoption parameters in networks: New estimation approaches for the latent-diffusion-observed-adoption model

This thesis investigates parameter estimation in a widely applicable model of interaction in social networks. The econometric challenge arises from the fact that this interaction is not fully observable. Three distinct estimation methods are proposed to tackle this problem. The “Two-Period Estimator” resolves the dimensionality issue by limiting the time horizon modelled. The “Trimming Estimator” achieves a dimension reduction by restricting the set of considered network interaction scenarios to a manageable size. Both of these estimators use the Maximum Likelihood estimation method. The “Moment-based Estimators” on the other hand make use of individual-specific moment conditions that are easily calculated using shorthand formulas. In an over-identified system, other moments, which are more complicated to evaluated, are not necessary as long as the moments considered are sufficient to identify the model parameters. The properties of the estimators are investigated by means of analytical considerations and Monte Carlo experiments. Furthermore, the estimators are applied to a concrete setting using publicly available data. Recent years have seen an increase in the availability of data on social networks and the various activities mediated through them. This has generated a need for econometricians to develop or adjust estimation methods to fit the particular requirements of social network models. This thesis hopes to make a contribution to this new strand of research