METHODS FOR ANALYZING REPEATED CATEGORICAL DATA

描述：(改编自调查员摘要)重复测量 在公共卫生的所有领域都进行了研究。在反复的措施中 研究，基本抽样单位是一组或一群受试者；a 对集群内的每个受试者进行测量。缺少‘答复 和协变量是重复测量研究中常见的现象，并且 大部分提案涉及重复丢失数据的问题 测量研究。首先，我们将评估和扩展现有的方法 纵向研究中遗失关联度的估计方法 回应。其次，似然方法在计算上是可行的 因此，我们提出了一种伪随机性方法来估计参数 具有不可忽略非单调的纵向研究的边际模型 缺少结果。第三，条件Logistic回归常被用于 消除扰民的“固定”集群效应，我们提出了一种改进的 适用于缺失的条件Logistic回归 协变量。第四，由于目前的基因检测技术 突变，研究人员现在对估计赔率比感兴趣 在遗传状态(突变，无突变)和治疗成功之间是的， 否)。不幸的是，并不总是有可能进行完整的遗传 评估以确定某个基因是否发生了突变，从而导致数据丢失。 我们将应用缺失数据方法来分析基因的突变状态 吉恩。我们最终提出的项目不涉及丢失的数据，尽管它 评估一种可能在聚集数据中产生偏向结果的方法 学习。我们将确定集群是否会影响通常的测试统计数据 在一项随机临床试验中没有治疗效果。

DESCRIPTION: (Adapted from investigator's abstract) Repeated measures studies are undertaken in all areas of Public Health. In repeated measures studies, the basic sampling unit is a group or cluster of subjects; a measurement is made on each subject within the cluster. Missing 'responses and covariates are common occurrences in repeated measures studies, and the majority of the proposal relates to missing data problems in repeated measures studies. First, we will evaluate and extend existing methods for estimating measures of association in longitudinal studies with missing responses. Secondly, the likelihood methods can be computationally intensive, so we propose a pseudolikelihood methods to estimate parameters of marginal models for longitudinal studies with nonignorable nonmonotone missing outcomes. Thirdly, conditional logistic regression is often used to eliminate nuisance 'fixed' cluster effects, and we propose a modified conditional logistic regression which is appropriate to use with missing covariates. Fourthly, because of current techniques of determining gene mutation, investigators are now interested in estimating the odds ratio between genetic status (mutation, no mutation) and treatment success yes, no). Unfortunately, it is not always possible to perform a complete genetic evaluation to determine if a gene has mutated, resulting in 'missing data'. We will apply missing data methods to analyze the mutation status of the gene. Our final proposed project does not concern missing data, although it does evaluate a method that can give biased results in clustered data studies. We will determine if clustering affects the usual test statistics of no treatment effect in a randomized clinical trial.