An Algorithm for Testing Unidimensionality and Clustering Items in Rasch Measurement
A new method was proposed to the construction of scales tht show a good fit to the Rasch model. Also, the method is based on a partial hierarchical cluster analysis and it doesn’t need to have specific assumptions regarding the model underlying the analyzed item set. A serious of simulations was conducted that based on the cluster analytical algorithm of proposed method and principal components analysis of conventional method. Moreover, four independent variables and one dependent variable were considered. It was found that the more the size and the smaller the item set size and the correlation between the person parameters, the higher the percentage of correct scale reconstructions for the new and traditional methods. In the end, an empirical data was analyzed by the new method.
Comments, Questions and Future Study
1) From table 2, the PCA had extreme high and low percentage of correct results for some conditions. What causes this unstable performance for the PCA?
2) This paper seems not depicted its theoretical framework and simulation procedure clearly. Maybe the authors thought that it’s too simple to mention the details, however, for me it’s hard to catch the main point and understand results of this paper.
For Question 1: Based on the studies from other researches (Weng & Cheng, 2005; Tran & Formann, 2009), the sample size small than 250 would yield indefinite matrices of tetrachoric correlations which imply that the PCA can not be conducted.
For Question 2: Ya, I agree with you. I have write to the author and ask for more details.
For Question 2: Ya, I agree with you. I have write to the author and ask for more details.