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