Find the signals

A covariate is a measured variable used as a possible input to a prediction. Rainfall, market prices, road access, population movement, and satellite observations can all be covariates.

Some signals are direct. Others are proxies: indirect clues that often move with the thing we care about. A proxy can help a model without explaining the cause. PPE searches repositories such as Data Commons and Google Earth Engine, and can look for additional signals in government and academic sources.

The design question is provenance: who measured the signal, when, at what scale, and would it have been available before the decision? A useful proxy is not automatically a fair or stable one.