Collaborative Research on Latent Class Models of Measurement Error

调查中最重要的活动之一是开发和测试调查问题。不幸的是，这个过程在很大程度上仍然是一个定性的努力，它的特点是专家对问题的回顾，少数参与者的焦点小组讨论，以及少量的密集的“认知”访谈。许多研究人员质疑这些方法识别问题项的有效性。此外，这些传统问卷预测技术产生的定性数据与数据要解决的定量标准（如可靠性和有效性）之间存在脱节。该项目将系统地评估一种定量方法的潜力——潜在类分析（LCA）——用于开发和测试调查问题。该项目旨在通过对现有数据进行一系列新的实验和分析，回答有关LCA模型作为评估调查问题的工具的应用的几个具体问题。实验研究将把LCA模型的结果与“金标准”进行比较，在“金标准”中，被评估变量的真实值是已知的。这些研究将比较LCA方法得出的结论与更传统的分析得出的结论。分析研究将LCA模型应用于现有数据集，并使用模拟来评估LCA方法对违反其基本假设的鲁棒性。这个项目将推进关于各种策略的基本知识，包括使用潜在类别模型来开发问卷。它将显示这些模型是否能够在没有外部验证数据（如管理记录）的情况下评估调查项目的度量特征。该项目将把潜在类模型与传统问卷开发技术进行比较，并确定与传统方法相比，它们是否能产生更好的问卷，降低问卷开发成本，或者两者兼而有之。这项研究的结果将对包括联邦统计机构在内的调查界有价值。

One of the most crucial activities in mounting a survey is the development and testing of the survey questions. Unfortunately, this process largely remains a qualitative endeavor, one that features reviews of the questions by experts, focus group discussions with a handful of participants, and small numbers of intensive "cognitive" interviews. Many researchers have questioned the effectiveness of these methods for identifying problem items. In addition, there is a disconnect between the qualitative data produced by these conventional questionnaire pretest techniques and the quantitative standards (such as reliability and validity) that the data are meant to address. This project will systematically assess the potential of a quantitative method -- latent class analysis (LCA) -- for use in developing and testing survey questions. The project seeks to answer several specific questions about the application of LCA models as a tool for evaluating survey questions by conducting a series of new experiments and analyses of existing data. The experimental studies will compare results from the LCA models against "gold standards," where true values for the variables being assessed are known. These studies will compare the conclusions from the LCA method against those from more conventional analyses. The analytic studies will apply LCA models to existing data sets and also use simulations to assess the robustness of the LCA method to violations of its underlying assumptions.This project will advance basic knowledge about various strategies, including the use of latent class models, for questionnaire development. It will show whether these models can assess the measurement characteristics of survey items even in the absence of external validation data (such as administrative records). The project will compare the latent class models to conventional questionnaire development techniques and determine whether they can yield better questionnaires, reduced questionnaire development costs, or both compared to the traditional methods. The results of this research will be of value to the survey community, including the federal statistical agencies.