分析方法與交乘項策略組合對潛在交互作用及二次效果估計之評估

Evaluation of Analysis Methods and Product Strategies on Estimation of Latent Interaction and Quadratic Effects

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二次效果等非線性效果之估計有其重要性。本研究以模擬方式，評估中心化限制式方法、Jaccard與Wan程序、部分限制式方法及未限制式方法等四種利用引進交乘項的方法，搭配單一配對、兩兩配對及所有配對型式不同交乘項策略下，對潛在交互作用模型與潛在二次模型估計之影響。結果顯示，四種方法與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、未限制式方法於兩兩配對與所有配對組合，在適當解比率與參數估計評估表現皆較其他組合差，以小樣本或外生潛在變項負載量低時尤為明顯。

In conducting empirical research, researchers are often interested in nonlinear relationships. Thus, accurately estimating latent interaction and quadratic effects is of great importance. This research aims to evaluate latent interaction and quadratic effects by means of the Monte Carlo method. Performances of all possible combinations of four product indicator based analysis methods with three product strategies are examined. The four approaches used in this study include the centered constrained approach, Jaccard and Wan's procedure, the partially constrained approach, and the unconstrained approach, while the three product strategies are one pair, matched pairs, and all pairs. The results indicate that different approaches and product strategies have little effect on the estimates, but the partially constrained approach and the unconstrained approach (except for the one- pair strategy) produce fewer fully proper solutions, more bias and greater root mean square error for smaller sample sizes or poor reliability of the indicators.