Comparison of factor retention methods on binary data: A simulation study

Abstract

In this study, the purpose is to compare factor retention methods under simulation conditions. For this purpose, simulations conditions with a number of factors (1, 2 [simple]), sample sizes (250, 1.000, and 3.000), number of items (20, 30), average factor loading (0.50, 0.70), and correlation matrix (Pearson Product Moment [PPM] and Tetrachoric) were investigated. For each condition, 1.000 replications were conducted. Under the scope of this research, performances of the Parallel Analysis, Minimum Average Partial, DETECT, Optimal Coordinate, and Acceleration Factor methods were compared by means of the percentage of correct estimates, and mean difference values. The results of this study indicated that MAP analysis, as applied to both tetrachoric and PPM correlation matrices, demonstrated the best performance. PA showed a good performance with the PPM correlation matrix, however, in smaller samples, the performance of the tetrachoric correlation matrix decreased. The Acceleration Factor method proposed one factor for all simulation conditions. For unidimensional constructs, the DETECT method was affected by both the sample size and average factor loading.

Kategorik Veride Faktör Analizi İçin Kullanılabilecek Alternatif Bir Korelasyon Matrisi: Goodman-Kruskal Gamma

“…Çalışmaya ortalama faktör yükünün 0.70 olan koşulu Hair, Black, Babin ve Anderson, 2009 tarafından bu faktör yükünde yapının iyi tanımlanmış olduğu belirtildiğinden eklenmiştir. Ayrıca bu değer faktör analizi çalışmalarında da sıklıkla kullanılmaktadır (Depaoli ve Scott, 2015;Flora ve Curran, 2004;Kılıç ve Uysal, 2019;Li, 2016a). Diğer taraftan faktör yükünün yükselmesinin araştırma sonuçlarına nasıl etki ettiğinin incelenmesine de imkân sağlamıştır.…”