Dynamic Weighting of Panel Data

虽然情况并非总是如此，但调查通常是用来用相对较小的样本对较大的人口做出推断。为了做到这一点，调查中的每个受访者都被赋予一个权重，以描述调查中与他们相似的个人相对于人口中的比例之间的关系。这些权重的构建是至关重要的，因为它们确保调查样本可以准确地反映人口。这个项目为用户开发了一个工具来估计政治科学中几个最常用的小组调查的权重。通过使研究人员能够估计适合其个人研究问题的权重，我们增加了使用这些小组调查的研究的可靠性。由于面板加权涉及的问题，我们开发了一个R包，该包基于磨损和项目无响应后获得的子样本提供单个权重变量。创建这样一个统计软件包背后的动机是需要一个工具，它可以根据结果变量、协变量和指定的波动，为研究人员所需的分析提供一组合适的权重。此外，我们还开发了一个程序，在考虑到损耗和项目无反应后，对剩余样本的社会人口统计学特征进行诊断，以标记潜在的并发症，例如属于高损耗组的受访者超重。最后，我们生成了一个易于访问和用户友好的程序，便于分析面板调查数据。无数学者依靠具有全国代表性和可公开获得的小组调查作为数据资源来推断国民意见。这个项目通过为研究人员提供他们需要的工具来适当地衡量他们的数据，从而改进了这些推论。

GeneralAlthough it is not always the case, surveys are generally used to make inferences about a larger population using a relatively small sample. In order to do that, each respondent in a survey is given a weight to describe the relationship between the proportion of individuals like them in the survey relative to those in the population. The construction of these weights is critically important as they ensure the survey sample can accurately reflect the population. This project develops a tool for users to estimate weights for several of the most commonly used panel surveys in political science. By enabling researchers to estimate weights that are appropriate for their individual research questions, we increase the reliability of studies that use these panel surveys. TechnicalBecause of the issues that panel weighting involves, we develop an R package that delivers a single weight variable based on the sub-samples obtained after attrition and item non-response. The motivation behind the creation of such a statistical package is the need for a tool that can deliver a set of weights suitable for the analyses desired by researchers dependent on the outcome variables, covariates and waves specified. Moreover, we develop program to run a diagnosis of the socio-demographic characteristics of the remaining sample after accounting for attrition and item non-response in order to flag potential complications such as over-weight of respondents that belong to groups with high levels of attrition. Finally, we generate an accessible and user-friendly program that facilitates the analysis of panel survey data. Countless scholars rely upon nationally-representative and publicly-available panel surveys as a data resource to draw inferences about national opinion. This project improves those inferences by providing researchers with the tools they need to weight their data appropriately.