Three PA-based methods (Horn’s PA, PAFA, and MRFA) were compared to each other in simulation study. Another point of choosing the type of correlation was also considered, that is, Pearson and polychoric correlation. The method of generating random data was compared as well. In simulation study, a linear factor model was used for simulated data. It includes major factor and minor factor. The results show that the PA-MRFA outperforms Horn’s PA and PA-PAFA. The performance of PA-PAFA is worst. It is noted that the use of permutation or random data from MVN is of any difference. Although it is shown that the polychoric correlation is promising, it is likely to suffer from nonconvergence. It is recommended that Polychoric 95% threshold PA-MRFA should be adopted first. If convergence fails, use Pearson mean threshold PA-MRFA then.
1. This study is constrained in linear factor model for simulated data. Other nonlinear factor model for unfolding model could be used as well, e.g., polynomial factor model.
2. Other IRT model should be consider as well. It is not obvious how the performance of PA-MRFA on IRT model.