Efficient Statistical Algorithms for Dropout Data

说明(申请人提供)：本研究将用S+语言开发一个辍学数据统计软件库。数据缺失和丢失是纵向研究中常见的现象。当辍学过程与结果过程相关时，就会在分析这类数据时产生巨大的挑战。目前还没有商业软件在处理信息性或非随机辍学时考虑辍学机制。因此，结果是有偏见和误导性的。这项研究的最终目标是开发一个统计软件库，用于使用模式混合和选择模型两种方法分析辍学数据。这些方法将线性模型、广义线性混合效应模型或广义线性混合效应模型应用于响应过程，并使用IOGIT、概率位或阻塞对数连接的回归来描述退出过程。该库将包括参数估计、灵敏度分析、图形分析和模型选择的方法。发展起来的参数估计算法包括随机EM法、似然极大法和插值法。将开发图形工具，用于显示辍学数据、监测参数收敛和诊断拟合值。基于解析法和图解法的灵敏度分析在检验模型假设的有效性方面是有用的。综合的案例研究和模拟将显示本调查结果的优势和适用性。

DESCRIPTION (provided by applicant): This research will develop a statistical software library in S-PLUS for dropout data. Missing and dropout data are common nature in longitudinal studies. When the dropout process is related to the outcome process, it creates tremendous challenges in analyzing such data. No commercial software currently considers the dropout mechanisms in dealing with informative or non-random dropout. Consequently, the results are biased and misleading. The ultimate objective of this research is the development of a statistical software library for analyzing dropout data using both pattern mixture and selection model approaches. The approaches apply linear models, generalized linear mixed-effects models or GEE models for the response process and a regression using a Iogit, a probit or a Clog-log link for the dropout process. This library will include methods for parameter estimation, sensitivity analysis, graphical analysis, and model selection. The algorithms developing for parameter estimation include stochastic EM, likelihood maximization and imputation methods. Graphical tools will be developed for displaying dropout data, monitoring parameter convergence and diagnosing fitted values. Sensitivity analysis based on analytic and graphic methods are useful on testing the validity of the modeling assumptions. Comprehensive case studies and simulations will show the advantage and the applicability of the results of this investigation.