ABSTRACT
98The fixed-effect model is based on the assumption that there is one true effect size which is shared by all the included studies. The random-effects model is based on the assumption that the true effect could vary from study to study. If inferences were to be generalized to a population in which the studies are permitted to have different effects and different characteristics, then a random-effects model would be appropriate. The random-effects model leads to relatively more weights being given to smaller studies and to wider confidence intervals than the fixed effects model. The DerSimonian and Laird method of meta-analysis is based on the random-effects model with the assumption of common effect is relaxed and the effect sizes are assumed to have a normal distribution with mean θ and variance τ2. The maximum likelihood method is considered when variance of the estimator is assumed as known. Restricted maximum likelihood method is an approach to estimation that maximizes the likelihood over a restricted parameter space. Applying Bayesian methods to perform meta-analysis provides a more informative summary of the data than non-Bayesian approaches. The major advantage is the ability to incorporate the uncertainty from our estimates of the true effects in individual studies. The practice of starting with the fixed-effect model and then moving to a random-effect model, if Q is statistically significant should be discouraged. If the logic of the analysis says that we are trying to estimate a range of effects, then the selection of a computational model should be based on the nature of the studies and our goal.
