This paper mainly focused on a question: before implementing Rasch analysis, do we appropriately assign each item to one of dimensions when item pool probably included different clusters. This study proposed a method using the scale construction procedure with the R statistic for determining the number of clusters. It was compared with a traditional PCA method. For assessing the effect of this new approach, It calculated the rate of correct reconstructions over 10000 replications of each simulation condition. The result presented the new method performed better than PCA in some specific conditions.
My questions are:
1. The procedure of defining item clusters is difficult to understand to me. Such as the meaning of initial subset A3 and the meaning of f function reaching maximum.
2. Dose this scale reconstruction issue belong to a kind of exploratory factor analysis? If we have a clear theory or concept of what we want to assess in the assessment instrument, would we need to assign items to clusters before Rasch analysis?