Power and sample size for generalized linear models

广义线性模型的功效和样本量

• 批准号: 10680962
• 负责人: Paul Joseph Rathouz
• 金额: $ 24.49万
• 依托单位: UNIVERSITY OF TEXAS AT AUSTIN
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-06-15 至 2025-05-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10680962
• 关键词: Age; Applications Grants; Area; Attention; Chi-Square Distribution; Computer software; Data; Distal; Ensure; Formulation; Foundations; Freedom; Future; Goals; Investigation; Joints; Left; Linear Models; Logistic Regressions; Manuscripts; Mathematics; Measures; Mental Health; Methods; Modeling; Modernization; Nature; Outcome; Paper; Population; Probability; Procedures; Process; Public Health; Research; Research Design; Research Personnel; Resources; Sample Size; Specific qualifier value; Statistical Data Interpretation; Testing; Time; Work; analog; base; expectation; experience; in silico; indexing; interest; novel; novel strategies; randomized, clinical trials; response; sex; simulation; software development; usability

PROJECT SUMMARY / ABSTRACT This project responds to FOA PA-21-235. A variety of well-characterized and valid methods for power and sam- ple size (PSS) estimation in generalized linear models (GLM) have been developed. These are primarily in the area of linear models for continuous data and logistic regression for binary data. By and large, methods that are more general require a prior specification of several study-specific details in order to be implemented in a par- ticular application, and this can pose challenges for applied statisticians and non-statistical collaborators. These challenges have impeded the study design of translational mental health research, randomized clinical trials in psychiatric populations, and other investigations that involve GLM in ways that linear models do not. To overcome these challenges, the overarching goal of this project is to develop methods for estimating power, needed sample size, or minimally-detectable effect size in study designs involving GLMs. To be broadly useful, such methods should be accurate, interpretable, and, importantly, easily-specified. This project offers, in the framework of GLMs, a general formulation with the aim to recapture, to close approx- imation, features of the common approach to PSS for linear models, namely the use of partial multiple R-squared as a general measure of effect size. This is accomplished by introducing two GLM analogues of R-squared. Local and more distal alternative hypotheses are considered, the latter requiring more attention to yield accurate re- sults as the alternative hypothesis moves further from the null. Both Wald and score tests (which coincide under linear models) are also considered. This project has three specific aims: Using novel GLM analogues of multiple partial R-squared for linear models, develop approaches to estimate power, needed sample size, or minimally detectable effect size for Wald (Aim 1) and score (Aim 2) tests to be conducted in the framework of GLMs. Aim 3 is to develop, test, document and disseminate software implementing the new methods. The developed methods are guided by and applied to two collaborative projects in translational mental health. The expectation is a new and general suite of applicable and usable approaches to power and sample size es- timation for the “bread-and-butter” class of generalized linear models forming the foundation of so many analysis methods in modern biomedical and public health investigation.

项目总结/摘要 本项目响应FOA PA-21-235。各种良好的表征和有效的方法，功率和sam- 研究了广义线性模型（GLM）中的单尺度（PSS）估计问题。这些主要是在 连续数据的线性模型区域和二进制数据的逻辑回归。总的来说， 更一般地，需要事先指定几个研究特定的细节，以便在部分中实施， 这可能给应用统计人员和非统计合作者带来挑战。这些 挑战阻碍了转化心理健康研究的研究设计， 精神病人群，以及其他以线性模型所不涉及的方式涉及GLM的研究。克服 这些挑战，这个项目的首要目标是开发方法，估计功率，所需的样本 大小，或在涉及GLM的研究设计中的最小可检测效应量。为了广泛使用，这些方法 应该是准确的、可解释的，而且重要的是，易于指定艾德。 该项目在全球景观管理系统的框架内提供了一个总体方案，目的是重新获得、关闭 仿真，线性模型PSS的常用方法的特征，即使用部分多重R平方 作为效应大小的一般度量。这是通过引入两个R平方的GLM类似物来实现的。当地 和更远的替代假设被认为是，后者需要更多的关注，以产生准确的重新， 随着备择假设进一步远离零，结果。Wald和Score检验（符合 线性模型）也被考虑。该项目有三个具体目标：使用多种新的GLM类似物， 线性模型的部分R平方，开发方法来估计功效，所需的样本量，或最小 在GLM框架内进行Wald（目标1）和评分（目标2）检验的可检测效应量。目标3 是开发、测试、记录和传播实施新方法的软件。开发的方法 的指导下，并应用于两个合作项目的转化心理健康。 期望是一个新的和一般的适用和可用的方法套件的权力和样本量es- 对“面包和黄油”类广义线性模型的估计，形成了许多分析的基础 现代生物医学和公共卫生调查方法。