Zhang Q, Ip E H
Department of Biostatistical Sciences, Wake Forest School of Medicine, Winston Salem, NC, USA.
Comput Stat Data Anal. 2014 Sep 1;77:146-156. doi: 10.1016/j.csda.2014.02.017.
The latent class model provides an important platform for jointly modeling mixed-mode data - i.e., discrete and continuous data with various parametric distributions. Multiple mixed-mode variables are used to cluster subjects into latent classes. While the mixed-mode latent class analysis is a powerful tool for statisticians, few studies are focused on assessing the contribution of mixed-mode variables in discriminating latent classes. Novel measures are derived for assessing both absolute and relative impacts of mixed-mode variables in latent class analysis. Specifically, the expected posterior gradient and the Kolmogorov variation of the posterior distribution, as well as related properties are studied. Numerical results are presented to illustrate the measures.
潜在类别模型为联合建模混合模式数据（即具有各种参数分布的离散和连续数据）提供了一个重要平台。多个混合模式变量用于将受试者聚类到潜在类别中。虽然混合模式潜在类别分析是统计学家的一个强大工具，但很少有研究关注评估混合模式变量在区分潜在类别中的贡献。本文推导了用于评估潜在类别分析中混合模式变量的绝对和相对影响的新度量。具体而言，研究了后验分布的期望梯度和柯尔莫哥洛夫变差以及相关性质。给出了数值结果以说明这些度量。
