Measuring the Digital Divide: Structural Estimation of the Demand for Personal Computers

This paper estimates a structural model of demand for the personal computer (PC) in order to evaluate the drivers of the “Digital Divide.” Taking advantage of a large dataset on household-level PC purchases, the econometric model incorporates dynamic optimization, observed and unobserved heterogeneity, and the presence of a (sunk) learning cost incurred by first-time purchasers. The model therefore provides estimates of the differences in the marginal value for quality across different demographic groups, different consumer types, and an estimate of the difference in costs faced by upgraders and those who have not yet purchased a PC. These estimates allow for counterfactual evaluation of how demand would shift in response to a change in the rate of PC quality improvement, and allows for an assessment of the impact of policies designed to close the Digital Divide, such as subsidies for first-time buyers. The main findings indicate that the marginal value of PC quality varies significantly across income, education, age, and household size – the value of an extra unit of quality (measured as 200 MHz) ranges from $34 to $392 in 1999 and $0 to $142 in 2001. The “learning” cost of buying a PC is estimated as $2938 in 1999 and $2234 in 2001. Further, PC owners are less sensitive to price and more sensitive to changes in the rate of PC quality improvement compared to non-owners. Finally, a short-term (one year only) subsidy of $200 for first-time PC purchasers is estimated to increase non-owner demand (i.e. first time purchases) by approximately 60% while a long-term subsidy of the same magnitude will increase non-owner demand by approximately 10%. The evidence suggests that the Digital Divide results from the interaction between learning costs, persistent consumer heterogeneity and dynamic technological change in the personal computer industry. 1 Northwestern University. I would like to thank Shane Greenstein, Rob Porter, David Barth, Michael Coates, Eugene Orlov, and especially Scott Stern for great comments and support. I would also like to thank Forrester Research for the data and excellent support. Email: [email protected]