Computer Modeling Strategies for Meta-analysis

Over the past decade, meta-analyses have played an increasingly important role in summarizing clinical information and informing policy. Consequently, a new generation of meta-analysts have come forth and the demand for a streamlined modeling approach has emerged. In particular, compared to linear model estimation, many aspects of selecting and tting models for meta-analysis involve nonstandard data structures and statistical assumptions. For example, each data point in the analysis is associated with its own measure of precision that must be accounted for in the estimation process. Other study variables such as study design covariates, within-study predictors, and other aspects of the studies also play an important role when making inferences. Model-selection therefore can be considerably facilitated by exploratory graphical analyses. In this Chapter, we discuss exploratory methods and a uniied modeling approach to meta-analysis which integrates xed, random and mixed eeects models, as well as Bayesian hierarchical models into one framework. The authors gratefully acknowledge their colleagues with whom they have collaborated in the development of MetaGraphs, c a software package designed to perform meta-analysis using Hierarchical Bayes Linear Models: Richard