傳統處理探索性二階因素分析時,先藉由斜交轉軸來得到一階的因子,估計一階樣式矩陣,再將這些因子視為新觀察變項抽出二階因子,估計二階樣式矩陣。上述的二階段方法有可能會造成問題,因為二階因子是被一階因子與其樣式矩陣所決定的,當一階因子抽取有誤,二階因子抽取也就可能不合適。 一個較好的方法是,在估計一階因子的同時,即能考慮到其對二階因子的影響。因此研究目的就是發展一次抽取出兩階因子的技術(即同時轉軸);技術上來說,我們開發了一個可以同時評估一階與二階樣式矩陣的轉軸法,而此方法也把估計過程縮減為一階段方法。最後再以模擬與真實資料來呈現其效果。 為了測試本研究開發的一階段方法之效能,我們進行不同情境的模擬實驗 (分別為情境CS、SC及SS)並分析一筆真實資料,最後比較一階段與二階段方法的結果差異。模擬結果顯示,我們的一階段方法在一階樣式矩陣的型態為複雜時,結果表現最好;然而當二階樣式矩陣的型態為複雜時,表現最差。 在未來研究上,尚有其他的不同定義的轉軸法(本研究只藉用一種轉軸法)能夠應用於我們的一階段方法。此外,本研究皆假設一階與二階因子的數目為已知,但在很多情境下此假設並不為真,因此,在未來研究上,開發一個能夠正確決定兩階因子數目的技術仍是必要的。
Traditionally, in second-order factor analysis, the first-order factors are extracted and such factors are subsequently regarded as new observed variables so second- order factors are extracted. Since the second-order factors are determined by the first-order ones, the conventional two-staged method will cause problems if the preselection for the first-order factors is inappropriate. An intuitive way to solve the problem is, when extracting the first-order factors, to evaluate its effect upon the second -order ones. Specifically, we’d like to define a new rotation criterion to assess the performance for both pattern matrices simultaneously. Such estimation process is hence reduced to a one-staged method. To testify the efficiency of such criterion, we analyzed three simulated and one real datasets, then compared the results between the two-staged and one-staged methods. The results showed that the proposed method performed better when the first-order pattern matrix is complex but worse when so is the second-order pattern matrix. The proposed method can also apply to other rotation criteria. Besides, assumption about the correct numbers of first- and second-order factors is not always the case. A future research for determining the numbers of these factors is needed.