Selection bias in observational studies can create spurious negative correlations between variables.
If a hospital admits people for either of two unrelated conditions, the two conditions will appear negatively correlated among inpatients. It is a standing hazard in any study drawn from a selected population.
In practice
- Among people already in hospital, smokers appear to have fewer of a second condition — because you need to be sicker in other ways to be admitted as a non-smoker.
- A firm finds its most technically able hires are its worst communicators, because it hired anyone strong on either.
- Standing
- The birth-weight paradox and the obesity paradox are the best known medical cases.
- Further reading
- Wikipedia
- Subject
- Science & Epistemology
- Essay
- Paradoxes in the data, and sayings about knowing