K-means、潜在类别模型和混合Rasch模型的比较

The Comparision among K-means,Latent Class Model and Mixture Rasch Model

基于模拟研究比较了K-means方法、潜在类别模型和混合Rasch模型在二分外显变量情境下的聚类效果。结果表明:(1)潜在类别数量、变量数量、样本量、样本平衡和变量间相关对K-means方法、潜在类别模型和混合Rasch模型的分类准确性均有影响且因素间的交互作用存在;(2)除了在2个潜在类别的样本不平衡条件下K-means方法表现较差外,在其他条件下与潜在类别模型和混合Rasch模型的表现相当;(3)混合Rasch模型的分类一致性在2个潜在类别的情境下要好于潜在类别模型,但是在4个潜在类别的情境下要差于潜在类别模型。

Based on Monte Carlo simulation study, this article provides a comparision among K - means, Latent Class Model and Mixture Rasch Model on condition that the observed variables are binary. The results show that（ 1 ） the number of latent classes and variables, the sample size, the balance of proportion and the correlation among the variables have impact on the classification and the interaction among these factors exists ; （2） the classification of K - means is poor when the data generated from two latent classes and unblanced population, otherwise, the performance of K - means is similar with the model based approaches; （3） with two latent classes, the performace of Mixture Rasch Model is better than the Latent Class Model but when the latent classes is four,the later is better than the former.