DEVELOPMENT OF STATISTICAL METHODS TO ANALYZE CLUSTER SAMPLES

本研究项目将研究统计方法， 分类数据来自聚类样本， 每个聚类内的观测可以是相关的，并且其中 可以以不相等的概率选择观测。 在 特别是对基于人口的聚类样本的分析， 病例对照研究和横断面和纵向健康 调查被检查。 研究集中在开发 对逻辑回归和Mantel-Haenzel和Wolf的修改- Halflipping程序可以解释复杂的样本 设计 计算机模拟用于验证统计 在改进方法的开发中使用的近似值。 这项研究的初步结果表明， 适当分析聚类样本数据的方法 考虑到集群内的相关性结构和 对观察结果的加权不等。 这些方法将 用于分析婴儿喂养研究和重复妊娠 研究家庭构成集群的情况。

This research project will study statistical methods for analyzing categorical data that comes from cluster samples where the observations within each cluster may be correlated and where the observations may be selected with unequal probabilities. In particular, the analysis of cluster samples from population-based case-control studies and cross-sectional and longitudinal health surveys is examined. Research has concentrated on developing modifications to logistic regression and Mantel-Haenzel and Wolf- Haldane procedures that would account for the complex sample design. Computer simulations are used to validate statistical approximations used in the development of modified methods. Preliminary results from this research indicate that the modified methods for analyzing data from cluster samples appropriately take into account the intra-cluster correlation structure and the unequal weighting of the observations. These methods will be useful for analyzing infant feeding studies and repeat pregnancy studies where the family constitutes the cluster.