Abstract
A normal distribution reference to the standard test for the overall treatment efficacy in a one-way random effects ANOVA model and meta-analysis leads to unacceptable liberality. Using a convenient heterogeneity variance estimator, a test procedure (a confidence interval) is proposed and shown to attain type I error rates (confidence coefficients) which are closer to the nominal level compared with the standard procedure. The proposed test is also shown to perform well for data from binary experiments.