成对比较数据和排序数据的处理:模型分析的方法

Paired-comparison and ranking data:a method of modeling analysis

对近年来研究者所提出的瑟斯顿模型和BTL模型两大类模型进行综述,总结了目前研究者提出的2种数据分析和处理方法视角:多层线性模型(HLM)和结构方程模型(SEM).基于模型的方法不仅充分利用了成对比较和排序数据的信息,而且模型更具拓展性,如为考察协变量、潜类别等的影响提供可能,也为迫选测验等的开发提供指导意义.本文不涉及复杂的统计知识,运用心理学研究中被熟知的模型对处理成对比较数据和排序数据的模型及其模型估计进行说明,并通过Mplus软件示例,以期模型分析的方法在对成对比较数据和排序数据的处理上得到广泛的应用.

Modeling of paired-comparison and ranking data plays an important role in measuring attitudes, preferences, and values. However, since model estimation is complex and requires special knowledge of statistics, model application analysis is restricted, information of items is lost. Thurstone and BTL models, two main models proposed in recent years, were reviewed. Two perspectives of data analysis were summed up. hierarchical linear model （HLM） and structural equation model （SEM）. Model-based methods not only can make full use of data information, models can also he extended. Effects of covariates and latent class could be investigated, guidance for development of tests including forced choice test could be provided. Known models and model estimation were used to deal with paired comparison and ranking data, without involving sophisticated statistical knowledge. Mplus software examples are provided for further empirical studies. It is hoped that modeling analysis of paired comparison and ranking data can be widely used.