Development of Specific Regression Models Involving Latent Classes in Social Research

这个项目的目的是为分析社会和个人进程发展三组新的具体回归模型。所有三种回归模型都涉及潜在类别，目的是测量未直接观察到的数量，并反映未观察到的群体异质性的特定方面。一组由多项logit潜在类别回归模型组成，具有相关潜在类别之间的特殊对比，并将用于分析社会流动性的社会内部群体比较。第二组是对两类模型的特例——推动者-滞留者模型的回归扩展，用于民意稳定性和变化的面板数据分析。第三组模型，潜在选择模型，假设一对二分类潜在类变量，代表丈夫和妻子的潜在选择，作为观察到的二分类因变量的协变量，用于使用已婚夫妇的事件历史数据对家庭事件进行离散时间危险率分析。虽然这三组模型在形式和实际应用上有所不同，但它们都是用于使用潜在类别分析社会和个人过程的。对于每个模型，该研究解决了与以下相关的问题：(1)模型的特定参数化，特别是关于潜在类别，关于这些参数化可能在特定社会和个人过程的分析中提供的新见解，(2)参数估计的稳定性，以及(3)将项目中开发的模型的结果与相关现有替代模型的结果进行比较。预期这三组具体的潜在阶级模型将产生实质性的发现，如果不将这些发现用于这些模型所打算用于的社会研究领域，这些发现是不可能得到的，或者是不可能得到的。第一组模型的实质性应用领域是社会流动的社会内部群体比较，例如社会流动模式的差异，基于现有流动模型的分析有严重的局限性。在该项目中开发的模型将使人们对社会机会的种族/民族不平等有新的认识。第二组模型的实质性应用领域是公众舆论变化的面板数据分析。这些模型通过假设一个群体的反应是基于未观察到的异质置信水平，将积极反应和消极反应的预测与这些反应的稳定性的预测分开。一项应用，例如对1990-92年间布什政府支持率的上升和下降的分析，可能会得出这样的发现：对政府行政的明显高支持率是建立在一个薄弱的基础上的，因为支持者的信心水平远低于非支持者的信心水平。第三组模型应用的实质性领域是对家庭过程的分析，其结果是一个家庭事件，如离婚、分娩、住宅搬迁或购买房子——关于丈夫和妻子可能同意或不同意对方。这些模型不仅试图确定潜在的个人选择的预测因子，而且还试图确定未观察到的家庭异质性的预测因子，这些预测因子涉及(1)丈夫和妻子之间在导致事件发生或不发生方面的相对“权力”，以及(2)需要妻子和丈夫之间达成共识的程度才能产生家庭事件的发生。

This project aims at developing three new groups of specific regression models for the analysis of social and individual processes. All the three regression models involve latent classes for the purpose of both the measurement of quantities that are not directly observed and the reflection of specific aspects of unobserved population heterogeneity. One group consists of multinomial logit latent-class regression models with special contrasts among the dependent latent-class categories, and it will be developed for the analysis of intrasocietal group comparisons of social mobility. The second group consists of regression extensions of the mover-stayer model, which is a special case of the two-class model, for the panel data analysis of stability and change in public opinions. The third group of models, latent-choice models, assume a pair of dichotomous latent-class variables that represent husband's and wife's latent choices as covariates of an observed dichotomous dependent variable for the discrete-time hazard-rate analysis of a family event using pairs of event-history data of married couples. Although the three groups of models differ in form and substantive application, they are all intended for the analysis of social and individual processes using latent classes. For each model, the study addresses issues related to (1) specific parameterizations of models, especially about latent classes, regarding possible new insights that such parameterizations may provide in the analysis of particular social and individual processes, (2) the stability of parameter estimates, and (3) comparisons of results from models to be developed in the project with those from related existing alternative models.The three groups of specific latent-class models are expected to yield substantive findings that cannot be made, or cannot be obtained definitively, without using them in the areas of social research for which the models are intended. The substantive area of applications of the first group of models is intrasocietal group comparisons of social mobility, such as differences in social mobility patterns, for which an analysis based on existing mobility models has serous limitations. The models to be developed in the project will shed new light on racial/ethnic inequality of social opportunity. The substantive area of applications of the second group of models is the panel-data analysis of change in public opinion. The models separate the prediction of positive versus negative response from the prediction of the stability of these responses by assuming that responses of a population are based on unobserved heterogeneous confidence levels. An application, such as an analysis of the rise and the fall of support for the first Bush administration through the years 1990-92, may yield such a finding that an apparently high rate of support for the government administration stands on a weak basis because supporters' confidence levels are much lower than nonsupporters' confidence levels. The substantive area of applications of the third group of models is the analysis of a family process having as its outcome a family event, such as divorce, childbirth, a residential move, or the purchase of a house -- regarding which the husband and the wife may or may not agree with each other. The models will try not only to identify the predictors of latent individual choices but also the predictors of unobserved family heterogeneity regarding (1) the relative "power" between husband and wife in causing event occurrence or nonoccurrence, and (2) the extent to which a consensus between wife and husband is required to yield the occurrence of the family event.