Reconsidering the odds ratio as a measure of 2 × 2 association in a population