分析方法與交乘項策略組合對潛在變項非線性效果估計之評估

Evaluation of Analysis Methods and Product Strategies Estimation on Latent Nonlinear Effects

陳淑萍(Shu-Ping Chen)

指導教授 ： 鄭中平

共同指導教授 ： 余麗樺(Yu L)

高雄醫學大學/健康科學院/心理學研究所/碩士(2007年)

https://doi.org/10.6832/KMU.2007.00084

摘要

在實徵研究上，非線性關係常為研究者所關切，發展潛在變項間交互作用與非線性效果之估計有其重要性。結構方程模型相較於傳統分析交互作用或非線性效果的方法，透過引進潛在變項與納入測量誤差之考量，估計較為準確，因此具處理此議題之優勢；此外，樣本人數與觀察指標測量品質為評估潛在交互作用或非線性效果影響要素之一。
本研究基於目前多數SEM軟體可實作之方法，包括限制式方法、Jaccard ＆ Wan之多重指標最大概似法、部分限制式方法與未限制式方法，透過模擬，系統性地比較單一配對、兩兩配對與所有配對等三種交乘項策略於不同樣本人數與外生潛在變項負載量下之表現，以評估潛在交互作用與非線性效果。結果顯示，四種方法與不同交乘項策略對估計影響不大，但部分限制式方法於三種配對組合與未限制式方法於兩兩配對和所有配對組合在適當解比率與參數估計評估上表現皆較其他組合差，尤以小樣本或外生潛在變項負載量較低的情況。
基於模擬結果，本研究建議在簡單模型中，研究者可使用限制式方法於三種配對組合，或以未限制式方法搭配單一配對組合；在複雜模型中，則建議使用限制式方法於所有配對組合。最後並討論資料違反常態分配假定與潛在變項所屬觀察指標為異質情況下之研究限制與未來研究方向。

關鍵字

結構方程模型 ； 潛在交互作用效果 ； 非線性效果

並列摘要

In conducting empirical research, the researcher is often interested in nonlinear relationships. Structural Equation Modeling (SEM) differs from the traditional approaches of analysis of interaction or nonlinear effect in that, by including latent variables and measurement error, the estimates are more precise. Thus, SEM is the superior approach in estimating latent interaction and nonlinear effects, of which sample size and reliability of the observed variables are two influential factors.
Most SEM software available today can implement the following four approaches: the constrained approach, the multiple indicators ML approach of Jaccard and Wan, the generalized appended product indicator approach (GAPI), and the unconstrained approach. These four approaches are applied in a simulation study for this research. The simulation study consists of varying the sample size and reliability of the observed variables in order to systematically compare the performances of each of the four approaches when applying the product strategies one pair, matched pairs, and all possible pairs. It is found that all four approaches and the product strategies have little effect on the estimates, but the GAPI approach with three types of indicator products and the unconstrained approach with matched pairs and all possible pairs resulted in fewer fully proper solutions, more bias and large values for observed standard deviations, in particular for the smaller sample size or poor reliability of the observed variables.
On the basis of the result, we recommend the use of the constrained approach with three types of indicator products and the unconstrained approach with one pair for a simple model, and the use of the constrained approach with all pairs for a complex model. Our results are constrained by the particular limitations of the simulation, for example with regard to non-normality of the latent variables and heterogeneous indicators of latent variables. Extensions to some more general cases represent an interesting research topic.

並列關鍵字

structural equation modeling ； atent interaction effects ； nonlinear effects

參考文獻

Algina, J., & Moulder, B. C. (2001). A note on estimating the Jöreskog-Yang model for latent variable interaction using LISREL 8.3. Structural Equation Modeling, 8, 40-52.

Arminger, G., & Muthén, B. O. (1998). A Bayesian approach to nonlinear latent variable models using the Gibbs-Sampler and the Metropolis-Hastings algorithm. Psychometrika, 63, 271-300.

Bollen, K. A. (1989). Structural equations with latent variables. New York: John Wiley & Sons.

Bollen K. A., & Paxton. P. (1998). Two-stage least squares estimation of interaction effects. In R. E. Schumacker & G. A. Marcoulides (Eds.), Interaction and nonlinear effects in structural equation modeling (pp. 125-151). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.

Boomsma, A., & Hoogland, J. J. (2001). The robustness of LISREL modeling revisited. Http://www.ppsw.rug.nl/~boomsma/ssi.pdf