PSYCHOMETRIC ADVANCES FOR ALCOHOL & DEPRESSION STUDIES

该项目的总体目标是提供更好的心理测量方法 在流行病学研究中。特别的目标是开发新的 心理测量学及其在酒精和抑郁症中的适用性 研究适用于大规模的国家样本，如1988年的国家 健康访谈调查了约50,000名观察者。有特殊利益的人 对症状项集合的多变量分析是否适用于 能够诊断各种疾病，如酒精依赖、酒精 虐待和严重抑郁障碍。该项目将尝试 为描述测量误差制定适当的统计模型 在这种诊断中，并提出改进的方法，将这种障碍与 假设的原因和结果。数据的统计方法 项目借鉴了因素分析和结构分析的总体框架 潜变量关系系统的方程建模 构造。特别是，该项目将利用最近的进展 与绝对的和其他非正常的测量有关的，例如 在首席调查员的LISCOMP计算机程序中运行。 将研究三个方法领域：容易出错的回归分析 二分因变量、分类和其他因子分析 非正态变量和纵向结构建模。重点研究 问题是各种问题的适当的潜变量建模 与饮酒有关的障碍的概念化，包括建模 具有连续性状、潜在类别和 它们的混合物。选择适当的潜变量分布将 以及与诊断有关的分类问题， 包括临床方法和心理测量方法之间的比较。这个 对假设的因果关系的适当建模将在新的 结构模型的类型，包括多变量结构概率 酒精中毒和酒精中毒随时间变化和协变的模型 抑郁症。

This project's general objective is to provide better psychometric methods in epidemiological studies. The particular aims are to develop new psychometrics and determine their applicability in alcohol and depression studies applied to large-scale national samples, such as the 1988 National Health Interview Survey of about 50,000 observations. Of special interest is the multivariate analysis of sets of symptom items administered to enable diagnoses of various disorders such as alcohol dependence, alcohol abuse, and major depressive disorder. The project will attempt to formulate appropriate statistical models for describing measurement errors in such diagnoses and propose improved ways for relating such disorders to hypothesized causes and consequences. The statistical approaches of the project draw on the general framework of factor analysis and structural equation modeling of systems of relationships among latent variable constructs. In particular, the project will utilize recent advances related to categorical and other non-normal measurements, such as operationalized in the Principal Investigator's LISCOMP computer program. Three method areas will be studied: Regression analysis with error-prone dichotomous dependent variables, factor analysis of categorical and other non-normal variables, and longitudinal structural modeling. Key research issues are the proper latent variable modeling of various conceptualizations of disorders related to alcohol use, including modeling with dichotomous indicators of continuous traits, latent classes, and mixtures thereof. The choice of proper latent variable distributions will be studied, as well as classification issues related to diagnoses, including comparisons between clinical and psychometric approaches. The proper modeling of hypothesized causal relationships will be studied in new types of structural models, including multivariate structural probit modeling of change over time and the covariation of alcoholism and depression.