Estimating Diffusion Models Using Repeated Cross-sections: Quantifying the Digital Divide∗

This paper shows that a broad class of dynamic models of diffusion of new technology (usually estimated using duration models) can be estimated using data with extreme censoring i.e. when only discrete choice data is available in the form of repeated cross-sections. This finding should have broad relevance across numerous fields where such models are currently used, since it opens up new potential sources of data. For example expenditure surveys are commonly available for most countries in this format. We present Monte Carlo evidence to show that the maximum likelihood estimator is consistent and efficient in this context. We also apply this methodology to estimate a diffusion model of Internet services at home for U.S. consumers. We use publicly available Current Population Survey (CPS) data to estimate these models. Consequently we are able to test for the existence of and, estimate the dimensions of the so-called Digital Divide. These estimates also allow us to make forecasts of Internet penetration in the United States across various levels of aggregation (geographic and demographic). ∗I would like to thank Patrick Bayer, Hanming Fang, Martin Pesendorfer, Gustav Ranis and Subrata Sen for many helpful discussions, seminar participants at Yale and also other IO folks at Yale who have all provided valuable suggestions and encouragement at different times. †Corresponding Address: Department of Economics, Yale University. 37 Hillhouse Avenue, New Haven. CT 06520, e-mail:[email protected]