It can be shown that α<Xe2/X2<1 and σ̂e/σ̂ > 1, when α < 1. In other words, when the effective sample size is smaller than the apparent sample size, the test statistic is smaller (leading to larger p-values), and the 95% CI of OR is wider.
