Simpson's Paradox
Simpson's Paradox refers to the situation where a correlation between two variables changes sign if we consider a third variable.
See Judea Pearl: Simpson’s Paradox: An Anatomy; http://bayes.cs.ucla.edu/R264.pdf
Two main points where Bayesian networks can help solve this paradox:
- Confounders (in the example: sex) influence the outcome. So we should condition on it to remove this confounding effect.
- Mediators (in the example: blood pressure): it's part of the causal effect we are trying to find, so if we condition on this we may obscure the relation we're interested in.
This means we need to hypothesize whether we have a confounder or mediator and act accordingly. Using these assumptions (and some more) we can drawn up a network.
Simpson's Paradox is therefore an interesting example that data alone is often not sufficient for answering interesting questions, even when it is a simple question concerning correlation.