Parallel analysis (PA) is a method for investigating the dimensionality of a set of variables. This study introduced three PA methods, including the original approach of PA (Horn's PA), PA based on principle axes factor analysis (PA-PAFA), and PA based on minimum rank factor analysis (PA-MRFA), respectively, and compared their efficiency to accommodate ordinal variables in many conditions by incorporating different settings. These settings were considered: sampling technique for eigenvalues (normal or permutation), extraction method (95% quantile or mean), and computation of correlation matrix (Pearson or polychoric). All simulations were carried out by using FACTOR, a stand-alone program for Widows. The results suggested that, in general, Horn’s PA and PA-MRFA had a similar performance in contrast to PA-PAFA, where PA-PAFA tended to overextract factors.
1. It is an interesting finding that, as shown in Table 1, all employed methods yielded nearly no power on "Total factors correct" across all conditions. What does it imply?
2. In this study using the polychoric method sometimes leaded to nonconvergence results. Due to the finding, does it mean that the polychoric based PA is not recommended in empirical analyses?
3. Since former studies were proposed to explore the number of major factors for dominance models, it never guarantees the feasibility of applying the same methods to unfolding models.
good questions.
1. note that the precision rate is decreasing when sample size large. it may imply that the detectability of PA has increased for minor factors, not the case in small sample size.
2. we can give it a try first. If it fails, use Pearson correlation instead.
2. they are linear model, so never appropriate for unfolding model.
1. note that the precision rate is decreasing when sample size large. it may imply that the detectability of PA has increased for minor factors, not the case in small sample size.
2. we can give it a try first. If it fails, use Pearson correlation instead.
2. they are linear model, so never appropriate for unfolding model.