The Book of Why: The New Science of Cause and Effect

by Judea Pearl

The reason for the difficulty is that confounding is not a statistical notion. It stands for the discrepancy between what we want to assess (the causal effect) and what we actually do assess using statistical methods.