Debate over U.S. broadband policy has largely shifted away from mere availability, toward the degree of competition among service providers, and effects on price and quality, in distinct regional markets. By far the largest sustained effort to increase broadband provision in the United States has been the so-called "e-Rate" schools and library subsidy system, funded by the telephone rate payer through the Universal Service Fund, and administered by a private industry consortium. In December 2014, the FCC issued an e-Rate modernization order that increased the annual funding limit for the e-Rate program by 63%, to $3.9 billion annually, and shifting if further from supporting legacy telecommunications systems, and toward provision of high speed broadband within schools, including in particular an emphasis on support for internal wireless internet infrastructure within schools. One rationale for expansion of the e-Rate subsidy program voiced by advocates has been that it would also create spillover benefits for neighborhoods served by recipient schools, by enabling scale economies in the provision of broadband services to neighborhoods by new providers. This paper undertakes a rigorous empirical evaluation of this argument using a rich panel data set assembled from multiple sources. Using these data to control for a variety of economic, social, and demographic factors that might shift demand, as well as local factors that might shift the costs of network and service provision, I examine whether the pre-2014 e-Rate program had an identifiable and statistically significant impact on the evolution of broadband competition at the individual U.S. zip code level over the period 2005-2008. The impact of a much smaller but much more focused program also funded by the Universal Service Fund, the rural health center program, can also be evaluated, and provides a useful counterpoint in assessing the impact of these subsidy programs on the competition issue. One statistical problem faced in undertaking this evaluation is that FCC statistics on broadband provision by geographical locale censor reported data on the number of competitors when small numbers of providers are present. If observations with censored outcome variables are simply dropped, it is well understood that estimates of effects in a regression model will be biased. Using simulations, I test a multiple imputation method I have devised that exploits aggregate state level information on the distribution of the censored provider outcomes, and find that it performs well in removing censoring bias. Employing these methods with an econometric model, I conclude that the small but highly focused USF rural health center funding has had a statistically and economically significant impact on numbers of local broadband service providers, while the e-Rate program generally did not in most areas. However, in the very poorest or most rural areas, there is some evidence that the e-Rate program had a small, but statistically significant impact in stimulating greater competition in broadband service provision
