What we can learn from selected, unmatched data: Measuring internet inequality in Chicago

By integrating a “big” dataset of Internet Speedtest® measurements from Ookla® with data on household incomes from the American Community Survey (ACS), we attempt to measure Internet speeds across income tiers. In the Ookla data, each measurement is technically rigorous but the sample frame is unknown. The ACS provides necessary information on income and Internet access from a known sample frame. Our likelihood combines these data and endogenizes selection effects to identify Internet speed distributions by income tier. We credibly identify the speed distribution for middle and high-income households. However, because the participation rate of low-income households in the Speedtest data is so limited, the speed estimates for these households are not identified. •Internet access is necessary for equal participation in modern society.•Existing data are coarse and do not capture notions of performance or quality.•Ookla® crowdsources rigorous Internet performance measurements worldwide.•These data are self-selected, so we develop a model endogenize this selection.•Our model makes precise what can and cannot be constrained with the data.