Robust Missing Data Methods for Categorical Regression

描述（由申请人提供）：在存在数据缺失和模型错误规范的情况下，从分类回归模型中获得稳健统计推断的改进方法将是流行病学和卫生保健研究界的宝贵工具。目前，流行病学模型通常用于确定酒精相关症状的模式，确定酒精使用障碍的标准，并评估管制酒精饮料使用和分配的政策。这种模型经常依赖于包含不完整数据的数据集。虽然商业上可用的统计软件提供了一些自动化的缺失值程序（例如，数据输入，期望最大化），但需要进一步的理论和实证研究来开发更可靠的统计方法。在第一阶段可行性研究中，Martingale Research成功开发了鲁棒估计和推理算法，该算法结合了随机估计、渐近统计和广义逻辑回归的最新进展，适用于在数据缺失和模型错误规范存在的情况下对流行病学问题进行分类回归建模。这些结果在模拟研究中得到了验证，并将这些方法应用于酒精相关的研究问题。此外，新的理论研究将缺失数据和模型错误规范统一起来，以支持新的鲁棒缺失数据推理统计的发展。

DESCRIPTION (provided by applicant): Improved methods for obtaining robust statistical inferences from categorical regression models in the presence of missing data and model misspecification would be an invaluable tool to the epidemiological and health care research communities. Presently epidemiological models are typically designed to identify patterns of alcohol-related symptoms, define criteria of alcohol use disorders, and evaluate policies regulating use and distribution of alcoholic beverages. Such models frequently rely on datasets that contain incomplete-data. While commercially available statistical software provides some automated missing value procedures (e.g., data imputation, Expectation-Maximization), further theoretical and empirical research is required to develop more robust statistical methods. In its Phase I feasibility study Martingale Research successfully developed robust estimation and inference algorithms that combine recent advances in stochastic estimation, asymptotic statistics, and generalized logistic regression that are suited to categorical regression modeling for epidemiological problems in the presence of missing data and model misspecification. These results were verified in simulation studies and the methods were applied to an alcohol-related research problem. Additionally, new theoretical research that unifies missing data and model misspecification was developed to support the development of new robust missing data inferential statistics. Phase II research will extend Phase I findings to develop and implement new robust missing data methods for categorical regression modeling in the areas of: i) hypothesis testing on parameter estimates, ii) standard error estimation, iii) model selection criteria, and iv) specification testing. The Phase II experimental design will utilize Monte Carlo simulation bootstrapping methods for the purposes of evaluating the missing data methods using representative alcohol-related databases. Specifically, the simulation studies will empirically characterize the appropriateness of the large sample assumptions for both consistent estimation and statistical inference. These simulation study methodologies in conjunction with the new robust missing data methods will be integrated into a prototype user-friendly standalone software package for the purposes of supporting epidemiological and health related regression modeling. In summary, Phase II research will establish the essential technical foundation for Phase III commercialization with the long-term objective of providing a suite of new missing data handling methods as an advanced statistical tool for recession modeling that improves epidemiological and health-related research.