Model Selection And Model Mean In Meta Analysis

Meta-analysis is a quantitative method which can combine and evaluate different studies on a specific scientific research problem. By merging the existing research results, this method allows researchers to overcome the difficulties caused by finite sample issue in a single research and improve the reliability of the results of statistical inference, while taking the heterogeneity between different studies into account. When we can create multiple models in Meta analysis, we focus on the question that is how to determine an appropriate model for analysis. In this case the problem belongs to the fie ld of model selection. When a model is selected, all statistical inference based on this model, researchers often treat the selected model as the true data generating model, then, the uncertainty of model selection process will generally be ignored. This leads to mis leading conclusions. In current thesis, we will adopt model averaging to handle the concerns.Within the framework of maximum likelihood estimation method, the thesis will adopt frequentist model averaging(FMA) method to study the model averaging problem in meta-analysis. The performance of the FMA method in meta-analysis will be studied via Monte-Carlo simulation. Simulation results show that when the real model is unknown, model averaging estimation outperforms model selection in the sense that model averaging method leads to smaller prediction loss. Model selection and model averaging in meta-analysis are then applied to analyze the BC G vaccine data. The results demonstrate that in comparison to the model selection, model averaging provides better prediction in general. This further supports the use of model averaging in meta-analysis.