We show that in general correlation between two variables does not imply a causal effect. The reason is that confounding variables can completely distort the observed correlation. We illustrate this phenomenon with simulated and real-world datasets, and also show that randomization is able to neutralize confounders.
0 Comments