Latent Variable Models and Methods

描述(由申请人提供)：潜变量是不能直接观察到的现象，而必须用一些问题或项目清单来测量。抑郁、饮食失调、家庭关系和酒精问题的严重性就是一些例子。这项建议是为潜在变量(具体目标如下)开发新的方法和模型，这些方法和模型可以非常普遍地应用。这些方法的用处将通过两个现有的数据集来展示，这些数据集来自公共卫生中的行为健康研究项目。这两个项目在主题和数据收集方面有很大不同，但有共同的需要对变量进行建模，这些变量由于其概念现象的性质而不能直接观察到(即潜在变量)。第一个研究项目涉及对明尼苏达州4746名青少年的221项调查，评估一系列与测量结果潜在相关的社会环境、个人和行为因素，包括营养健康、肥胖以及危害健康的行为，如吸毒和酗酒。第二个研究项目从管理数据库中收集了一个数据集，该数据库涵盖了1993-2000年间被诊断为有酒精问题的32,985名成年人的保险索赔。这个纵向数据库还与对5683名患者进行酒精中毒治疗时的心理社会调查问卷相关联。其具体目标是：1.为了弥合分类和连续潜变量模型之间的差距，特别注意对连续和分类潜变量之间的结构关系进行建模；2.发展连续和分类潜变量之间的非线性关系的建模方法；3.发展用连续和分类潜变量模型对多变量纵向数据进行建模的方法。这项提案的最后一个最重要的目标是使用开发的方法和模型来回答在青少年健康危害行为和成年酗酒者的医疗后果等特定领域与公共健康相关的研究问题。

DESCRIPTION (provided by applicant): Latent variables are phenomena that cannot be observed directly and instead must be measured using, for example, some battery of questions or check list of items. Depression, eating disorders, family connectedness and severity of alcohol problems are some examples. This proposal is to develop new methods and models for latent variables (specific aims below) that can be applied very generally. The usefulness of these methods will be shown with two existing data sets originating from behavioral health research projects in public health. The two projects are very different in subject matter and data collection, but share in common the need to model variables that because of their nature as conceptual phenomena cannot be observed directly (i.e. latent variables). The first research project involves a 221-item survey of 4746 adolescents in Minnesota assessing a range of socio-environmental, personal, and behavioral factors of potential relevance to the outcomes measured, including nutritional health, obesity, and health compromising behaviors such as drug and alcohol use. The second research project assembled a dataset from an administrative database of insurance claims for 32,985 adults diagnosed to have had an alcohol problem between 1993-2000. This longitudinal database is also linked with a psychosocial questionnaire administered to a subset of 5683 patients at the time they presented for alcoholism treatment. The specific aims are 1. To bridge the gap between categorical and continuous latent variable models with specific attention put on modeling the structural relationship between continuous and categorical latent variables 2. To develop methods for modeling nonlinear relationships between both continuous and categorical latent variables 3. To develop methods for modeling multivariate longitudinal data with both continuous and categorical latent variable models. A final, overarching aim of this proposal is to use the methods and models developed to answer research questions relevant for public health in the specific areas of adolescent health compromising behavior, and medical outcomes for adult alcoholics.