Abstarct
Broadband networks have become a fundamental infrastructure and, as a result, there is a growing need for accurate data on some of their main characteristics. In particular, broadband speed measurements are a key input to several consumer, policy and regulatory decisions. This project reviews the state-of-the-art in broadband measurement platforms, both from a technical and statistical point of view, and applies it to assess whether crowdsourcing Internet data on broadband speed measurements can be used to extract robust inference on real broadband speeds. One hundred million observations from Bredbandskollen, the broadband speed measurement platform of the Internet Foundation in Sweden, are processed using big data methods. The analysis provides strong evidence of a selection bias and an unstable sample composition across years, thus indicating that the sample of test users may not be representative of the whole population of Sweden. Nevertheless, the analysis finds that there is a relative equality of speeds across regions and that there does not seem to be an urban/rural broadband speed divide in Sweden.