SYNOPTIC ABSTRACT
Several models of meta-analysis are considered for combining estimates from different samples in order to judge a common overall effect. Suitable methods of constructing confidence intervals or tests on the parameter of interest are presented for each model. The models under consideration are the classical common mean problem or homogeneous fixed effects model, the fixed effects model with treatment by sample interactions, the random effects model with random interactions, extensions of the random effects model including the case that no within-sample variance estimates are available, and, finally, basic approaches to prospective meta-analysis. Real-data examples illustrate the performance of the various procedures.