Estimating Telephone Demand with State Decennial Census Data from 1970–1990

A telephone demand model (logit) is estimated with pooled Decennial Census data (1970, 1980 and 1990) for the states. Previous studies using pooled FCC penetration data are suspect due to large standard errors in the sample. Since our model includes data across time this allows the inclusion of long distance price and increases the variation in the standard variables. Time-effect dummy variables control for unobserved shifts in the data. Given that these dummy variables may pick up some of the effect of the long distance price, as well as other unobserved effects, their estimated impact is relatively small. Robust model results lead to the conclusion that elasticities decline through time. Furthermore, while subsidized penetration is more effective for targeted than untargeted programs, the cost per year of adding a household to the network is very high in either case (for 1990 about $5368 for untargeted; and for targeted $191 in 1990 and $1581 in 1998).