A Solution to the Ecological Inference Problem:Reconstructing Individual Behavior from Aggregate Data