一种用于检验回归混合模型假设的诊断工具。
A DIAGNOSTIC TOOL FOR CHECKING ASSUMPTIONS OF REGRESSION MIXTURE MODELS.
作者信息
Wadsworth Ian, Van Horn M Lee, Jaki Thomas
机构信息
Department of Mathematics and Statistics, Lancaster University, Lancaster, LA1 4YF, United Kingdom.
School of Education, University of New Mexico, Albuquerque, NM 87131, U. S. A.
出版信息
JP J Biostat. 2018 Jan-Jun;15(1):1-20. doi: 10.17654/BS015010001.
Regression mixture models are becoming more widely used in applied research. It has been recognized that these models are quite sensitive to underlying assumptions, yet many of these assumptions are not directly testable. We discuss a diagnostic tool based on reconstructed residuals that can help uncover violations of model assumptions. These residuals are found by using the posterior probability of class membership to assign, based on a multinomial distribution, a class to each observation. Standard residual checks can be applied to these posterior draw residuals to explore violations of the model assumptions. We present several illustrations of the diagnostic tool.
回归混合模型在应用研究中越来越广泛地被使用。人们已经认识到这些模型对潜在假设相当敏感，然而其中许多假设无法直接进行检验。我们讨论一种基于重构残差的诊断工具，它有助于发现模型假设的违背情况。这些残差是通过使用类别归属的后验概率，基于多项分布为每个观测值分配一个类别来找到的。标准的残差检验可以应用于这些后验抽样残差，以探究模型假设的违背情况。我们给出了该诊断工具的几个示例。
Zotero 插件