Loken Eric
Multivariate Behav Res. 2004 Oct 1;39(4):625-52. doi: 10.1207/s15327906mbr3904_3.
Mixture models are appropriate for data that arise from a set of qualitatively different subpopulations. In this study, latent class analysis was applied to observational data from a laboratory assessment of infant temperament at four months of age. The EM algorithm was used to fit the models, and the Bayesian method of posterior predictive checks was used for model selection. Results show at least three types of infant temperament, with patterns consistent with those identified by previous researchers who classified the infants using a theoretically based system. Multiple imputation of group memberships is proposed as an alternative to assigning subjects to the latent class with maximum posterior probability in order to reflect variance due to uncertainty in the parameter estimation. Latent class membership at four months of age predicted longitudinal outcomes at four years of age. The example illustrates issues relevant to all mixture models, including estimation, multi-modality, model selection, and comparisons based on the latent group indicators.
混合模型适用于源自一组性质不同的亚群体的数据。在本研究中，潜在类别分析应用于对四个月大婴儿气质进行实验室评估的观察数据。使用期望最大化（EM）算法拟合模型，并使用后验预测检验的贝叶斯方法进行模型选择。结果显示至少有三种类型的婴儿气质，其模式与先前使用基于理论的系统对婴儿进行分类的研究人员所确定的模式一致。建议对群体成员身份进行多重插补，以替代将受试者分配到后验概率最大的潜在类别，从而反映参数估计不确定性导致的方差。四个月大时的潜在类别成员身份预测了四岁时的纵向结果。该示例说明了与所有混合模型相关的问题，包括估计、多模态、模型选择以及基于潜在群体指标的比较。
