A random effect model with quality score for meta-analysis

grantor: University of TorontoMeta-analysis is a set of statistical maneuvers to quantitatively summarize multiple related studies. There are two major statistical approaches to meta-analysis. One is fixed effect models, which assume that all studies are governed by a common treatment effect, and take the studies be analyzed as the universe of interest; the other is random effect models, which allow different treatment effects for different studies, and treat these studies as representing a sample from a larger population of possible studies. Because the reliability of data used in the meta-analysis is different, in order to get more accurate results, quality scores can be introduced to model the data reliability. In this thesis, I study a random effect model incorporating quality score for meta-analysis in comparison with three other models. I apply likelihood methods and the bootstrap to estimate the study effect, and the bias and standard errors of the estimators. I apply the model to a real data set, and to simulated data sets. By analyzing systematic computational results, I find that the random effect model with quality score can offer better accuracy, precision and coverage of the estimate of study effect than the other three models.M.Sc