Approximate Bayesian approaches for assessing intraclass correlation coefficients in dependent binary data

• Main Authors: Hsin-Ni Tsai, 蔡忻妮
• Other Authors: Miao-Yu Tsai
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/07158093177833882557

碩士 === 國立彰化師範大學 === 統計資訊研究所 === 99 === Medical data in clinical studies are commonly carried out in clustered settings, where the subjects are correlated within clusters. When observations approximately follow a normal distribution, the intraclass correlation coefficient (ICC) is frequently used to assess the similarity within clusters with respect to the particular biological or environmental characteristics. However, for correlated binary data, it is difficult to obtain directly the ICCs by the definition of the proportion of the total variation explained by variation between clusters. In this article, we propose two approximate Bayesian statistical approaches, approximate Taylor Bayesian and empirical Bayesian approaches, to estimate ICCs in multilevel logit models for dependent binary data. To compare with a frequentist approach, we make a comparison between the approximate Taylor Bayesian, the empirical Bayesian, and the ANOVA approaches in simulation studies. The results of comparison studies reveal that the proposed approximate Bayesian approaches provide reliable and stable approaches in estimating ICCs, and parameters of fixed effects and variance components simultaneously. Furthermore, the results indicate that the approximate Bayesian approaches are robust in model misspecification.