ABSTRACT
This chapter considers standard and more advanced meta-analysis methods for the situation that the available data per study are an estimate of the effect measure and its standard error. All methods assume that the estimated study effects and the true study effects all follow normal distributions, while the standard errors are considered fixed and known. All important common-effect and random-effect methods published in the meta-analysis literature, as well as their interrelationships, are extensively discussed and put in a uniform framework. Given any estimate of the between-studies variance, standard and more advanced methods for estimation of the overall effect and for constructing corresponding confidence intervals are explained. Also, the Hartung–Knapp–Sidik–Jonkman and the profile likelihood method are discussed. Much attention is given to different methods for estimation with confidence intervals of the between-studies standard deviation, such as the DerSimonian–Laird method, the Paule–Mandel method, the (generalized) method of moment estimators, and (restricted) likelihood methods. Complementary methods to describe the heterogeneity between studies, such as the I 2 measure and the prediction interval, are also discussed. Furthermore, empirical Bayes methods to obtain improved estimates of the effects of individual studies are considered.
