A simulation study of model misspecification omitting nonlinear effects and method variances in structural equation modeling

• Main Authors: Li-Chung Lin, 林立中
• Other Authors: Li-Chiao Huang
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/06669449276714720053

碩士 === 中原大學 === 心理學研究所 === 99 === In the past, the research of model misspecification focused on linear effect, as the result, we knew very little about the impact of nonlinear effect misspecification. When we studied nonlinear effect, we didn’t tack into consideration of possible bias effect caused by method variance. However, these two model misspecifications may affect model estimation simultaneously. The main goal of this study is to discuss how nonlinear effect and method variance misspecification impact the estimation of predictive validity, trait correlation, and its standard error, in addition to the power of fit indices. I used the Monte Carlo simulation method to establish seven types of models and manipulate design factors such as the intercorrelations among traits, covariances among method variances, and the amount of variance explained by method variances. The results demonstrated that power of fit indices could be change with design factors and I found that χ2, SRMR, GAMMA, TLI, CFI, and RNI would detect model misspecification successfully while RMSEA had the worst performance and couldn’t detect it. Regarding the impact of parameter estimation, misspecification of method variance underestimated structural parameters of linear effect and the bias changed with different design factors with a smaller influenced on nonlinear effect. Misspecification of nonlinear effect overestimated in predict validity of quadratic effect or interaction effect with a smaller influenced on linear effect. When hypothesized model misspecified method variance and nonlinear effect simultaneously, it could change the estimation of structural parameters of linear effect and nonlinear effect. Howerever, it didn’t aggravate the bias of nonlinear effect. Furthermore, model misspecification could underestimate the standard error of structural parameters and impact the hypothesis test which caused a bias conclusion. Through my research, it can clarify the common impact of these two model misspecification of parameter estimation, and provides researchers with which fit indices that can detect model misspecification more effectively.