Improving Validity Measures for Alcohol-Related Models

分类回归模型被广泛用于识别酒精相关症状的模式，评估管理策略，并确定医疗和精神状况。然而，许多这样的模型不可避免地被错误指定，影响模型的有效性，并可能导致不正确的统计推断。此外，目前可用的统计软件忽略了潜在的存在模型误指定，从而使用户没有自动化的手段来容易地评估模型的有效性。改进的统计措施或测试，以评估分类回归模型的有效性将是非常宝贵的酒精相关的研究和整个卫生保健研究社区。Martingale Research将开发先进的统计拟合优度检验，以评估分类回归模型是否存在模型误设。这些检验将利用渐近统计理论来确定是否存在模型误设（即，拟合优度）。拟议的第一阶段研究将使用代表现有NIAAA数据库的数据库来证明拟合优度测试：1）足够敏感，可以检测和量化模型误指定的存在，并且可以2）对模型进行可靠的评估模型拟合。在第一阶段研究中获得的结果将证明在第二阶段期间进一步调查模型有效性研究所需的基本技术可行性，并为开发商用软件提供基础。拟议的商业应用：Martingale Research Corporation打算将拟议的拟合优度检验纳入一个先进的分类回归软件包，该软件包利用渐近理论来正确处理模型误设。这将为在以下过程中遇到的各种分类问题提供更好的解决方案：预测与酒精有关的结果、医疗诊断、财务预测以及信息数据库解释和管理。

Categorical regression models are used extensively to identify patterns of alcohol-related symptoms, assess administrative strategies, and identify medical and psychiatric conditions. However, many such models are inevitably misspecified which affects model validity and may lead to incorrect statistical inferences. Furthermore, currently available statistical software ignores the potential presence of model misspecification thus leaving the user with no automated means to readily assess model validity. Improved statistical measures or tests for evaluating the validity of categorical regression models would-be invaluable to alcohol-related research and the overall health care research community. Martingale Research will develop advanced statistical goodness-of-fit tests to evaluate the presence of model misspecification for categorical regression models. These tests will utilize asymptotic statistical theory to determine the presence of model misspecification (i.e., goodness-of- fit). The proposed Phase I study will demonstrate using a database representative of pre-existing NIAAA databases, that the goodness-of- fit tests are: 1) sensitive enough to detect and quantify the presence of model misspecification, and can 2) make reliable assessments of model fit. The results obtained in the Phase I study will demonstrate the essential technical feasibility required for further investigations into model validity research during Phase II and provide the foundation for developing commercially available software. PROPOSED COMMERCIAL APPLICATIONS: Martingale Research Corporation intends to incorporate the proposed goodness-of-fit tests into an advanced categorical regression software package that utilizes asymptotic theory to correctly handle model misspecification. This will provide improved solutions to a wide range of categorical problems encountered during: prediction of alcohol- related outcomes, medical diagnosis, financial forecasting, and information database interpretation and management.