Abstract

In this article, we applied three different approaches to estimate pseudo-weights for a self-selected sample of public schools in Brazil. The goal was to produce reliable estimates for Internet quality measurements collected by sensors installed in participating schools. The primary challenge was the size of the database, which contained over 100,000 records and more than 250 variables. These variables, whose values are known for the target population, were used to model and validate the results.

We describe the models applied, the statistics used to compare modeling results, and the challenges associated with each approach. Despite finding a "good" model, some bias still remains in the final results.