Fig. 2From: On the Q statistic with constant weights in meta-analysis of binary outcomesEmpirical levels at nominal level of significance .05, vs \(\theta\), for four approximations to the null distribution of \(Q_F\) for log-odds-ratio and for two approximations to the null distribution of \(Q_{IV}\), when \(p_{iC} = .1\). First three rows: equal sample sizes \(n = 40,\;100\) and 250; fourth row: unequal sample sizes, \(\bar{n}=160\)Back to article page