結構方程模型多向度構念題目組合表徵之構念意涵

Abstract

題目組合（item parcels）為題目分數之加總或平均，可於結構方程模型（structural equation modeling）分析中作為構念之指標。Little、Rhemtulla、Gibson、及Schoemann（2013）與Cole、Perkins、及Zelkowitz（2016）皆曾指出，多向度構念題目組合表徵之構念意涵可能不同於研究者原先之假設，致影響分析結果，故使用題目組合之前應先釐清其所反映之構念本質。多向度構念為涵蓋數個向度之構念，Cole等人以及Williams與O’Boyle（2008）皆指出許多心理學研究探討之構念乃為多向度，研究者並建構題目組合作為多向度構念之指標進行結構方程模型分析。本研究因之擴展Sterba與MacCallum（2010）之題目組合共變數矩陣推導，由單向度構念延伸至多向度構念，自多向度構念題目組合之共變數矩陣組成，提供一理論架構使研究者能檢視題目組合所表徵之多向度構念意涵。本研究並進一步以此架構論述不同題目組合策略、向度間相關，以及題目因素結構複雜度之題目組合所反映之多向度構念本質，亦以一模擬資料為例展示推導結果。實徵研究者可透過本研究推演之架構，檢視其所建構之多向度構念題目組合共變數矩陣，藉之釐清其所表徵之多向度構念意涵是否符合原先假設，以避免不當之結構方程模型分析結果。

Item parcels, represented as summation or average over items, can be used as indicators of latent variables in structural equation modeling (SEM). Little et al. (2013) and Cole et al. (2016) indicated that the nature of multidimensional constructs represented by parcels can be different from that assumed by the researchers and thus affects the results of SEM analysis. Researchers should therefore examine the nature of multidimensional constructs represented by parcels prior to the analysis. Cole et al. and Williams and O’Boyle (2008) indicated that many of the constructs investigated in psychological research are multidimensional, consisting of several facets, and item parcels have been frequently adopted to be indicators for these constructs in SEM. The present study therefore extended the algebraic derivation of the covariance matrix of item parcels in Sterba & MacCallum (2010) from unidimensional to multidimensional constructs to provide a theoretical framework for examining the nature of multidimensional constructs implied by parcels. The effects of parceling strategy, correlations among facets, and factorial complexity of items on the nature of multidimensional constructs represented by parcels were discussed using this framework and illustrated by a numerical simulation example. Researchers may apply the framework proposed in this study to clarify the nature of the multidimensional constructs inferred from item parcels through examining the covariance matrix among parcels so as to avoid misleading results from SEM analysis.