在不同資料型態下使用混合模型進行分析之程序探討

本研究之目的為，探討在｢多因素潛在類別｣和｢多向度潛在類別｣情境下，分析連續和二元資料型態的正確程序。本研究以混合模型為基礎，提出不同分析程序，並在因素個數 / 向度個數、因素相關/向度相關、潛在類別個數、潛在類別分離程度各個自變項之下，檢視其對於模式選擇正確率、參數估計和分類結果正確性的影響，並將分析程序相互對照比較，選出最佳程序提供未來分析實徵資料之參考。結果發現，當各個潛在類別的因素不變性程度為強假設，並搭配訊息量指標作為模式判斷依據時，程序一：｢先判斷因素結構 / 向度結構，再判斷潛在類別個數，逐步決定最佳結果｣和程序三：｢假設不同潛在類別個數和不同因素結構 / 向度結構的組合，再判斷最佳適配模型｣的表現優於程序二：｢先判斷潛在類別個數，再判斷因素結構 / 向度結構，逐步決定最佳結果｣。此外，因素相關 / 向度相關和潛在類別分離程度是影響模式判斷正確率的重要變項。最後，作者針對未來研究和實務的應用提出相關建議。

關鍵字

多因素潛在類別；多向度潛在類別；因素混合模型；試題反應理論混合模型

並列摘要

Multi-factor and multi-dimension structure exist in many test conditions along with continuous and binary responses respectively. And, population heterogeneity exits, too. Thus, in this study, we proposed three analysis procedures based on factor mixture model and item response theory mixture model to analyze data in context of the multi-factor / multi-dimension latent class. Simulations were also manipulated with different levels of factor numbers / dimension numbers, factor correlation / dimension correlation, numbers of latent class and class separation. The result showed that the procedures of “factor / dimension structure first then class number”(procedure 1) and “factor / dimension structure and class number considered simultaneously”(procedure 3) can mostly select the correct model using information criterion, and yielded precise parameter estimation and classification accuracy. It would be appropriate to choose procedure 1 and 3 when strong measurement invariance was assumed while using information criterion. Finally, study limitations and suggestions for future investigations were provided.