Estimating Mediation Effects in Prevention Studies

 描述(由申请人提供)：这项竞争性延续拨款提案的目的是开发、评估和应用方法学和统计程序，以确定预防方案如何改变结果变量。这些中介分析评估预防计划所针对的结构上的计划效果和对结果的影响之间的联系。正如许多研究人员和联邦机构指出的那样，调解分析确定了最有效的计划组成部分，并增加了对导致结果变量变化的潜在机制的理解。来自中介分析的信息可以使程序更强大、更高效、更短。这笔赠款的私人投资机构获得了一笔为期一年的NIDA小额赠款和三笔多年期赠款，用于开发和评估预防研究中的调解分析。这项工作导致了许多出版物和创新。拟议的五年延续期侧重于令人振奋的新调解分析统计发展的实际应用。四个统计主题代表了这项研究的下一步，包括分析和模拟研究以及病因学和预防数据的应用。在研究1中，调查了最近发展起来的方法--因果调解和贝叶斯调解分析--应用中的实际问题。因果调解和贝叶斯方法能够极大地提高调解分析的准确性。研究2探讨了预防研究中两个常见问题的解决方案，测量误差和混杂偏差，这两个问题会使中介分析不准确。将制定和评估纠正调解分析的分析以及评估这些问题如何影响结果的方法。研究3开发和评估了新开发的基于因果效应的纵向中介方法，以及对具有多次重复测量的数据的新方法。这些新方法有望更准确地模拟随时间变化的情况。研究4评估了在中介分析中发现亚群的方法，包括因果调解方法和群体和个体的多水平模型。对于每项研究，我们将通过预防数据的调解分析来调查独特的问题，包括在改进调解分析的研究之前获得的信息，纵向调解模式的变化时机，以及小组影响的类型和重要性。研究5将新的统计方法应用于来自NIH资助的几个预防数据集的数据，提供了关于这些方法的有用性的重要反馈。研究6通过我们的网站和其他媒体，通过与研究人员的交流和该项目的出版物，传播关于调解分析的新信息。

 DESCRIPTION (provided by applicant): The purpose of this competing continuation grant proposal is to develop, evaluate and apply methodological and statistical procedures to determine how prevention programs change outcome variables. These mediation analyses assess the link between program effects on the constructs targeted by a prevention program and effects on the outcome. As noted by many researchers and federal agencies, mediation analyses identify the most effective program components and increase understanding of the underlying mechanisms leading to changing outcome variables. Information from mediation analysis can make programs more powerful, more efficient, and shorter. The P. I. of this grant received a one-year NIDA small grant and three multi-year grants to develop and evaluate mediation analysis in prevention research. This work led to many publications and innovations. The proposed five-year continuation focuses on the practical use of exciting new mediation analysis statistical developments. Four statistical topics represent next steps in this research and include analytical and simulation research as well as applications to etiological and prevention data. In Study 1, practical issues in the application of recently developed methods, causal mediation and Bayesian mediation analyses, are investigated. Causal mediation and Bayesian methods have the capability to greatly improve the accuracy of mediation analyses. Study 2 investigates solutions to two common problems in prevention research, measurement error and confounder bias, that can make mediation analysis inaccurate. Analyses to correct mediation analysis as well as methods to assess how these problems could affect results will be developed and evaluated. Study 3 develops and evaluates newly developed longitudinal mediation methods based on causal effects and also new methods for data with many repeated measurements. These new methods promise to more accurately model change over time. Study 4 evaluates methods to uncover subgroups in mediation analysis including causal mediation methods and multilevel models for groups and individuals. For each study, we will investigate unique issues with mediation analysis of prevention data including information obtained prior to a study that improve mediation analysis, the timing of change in longitudinal mediation models, and the types and importance of subgroup effects. Study 5 applies new statistical methods to data from several NIH funded prevention data sets providing important feedback about the usefulness of the methods. Study 6 disseminates new information about mediation analysis through our website and other media, by communication with researchers, and publications from the project.