Sufficient Conditions of Equivalence in Measurement Model

• Main Authors: Lin, Weisheng, 林威聖
• Other Authors: 鄭中平
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/50588948495624118852

碩士 === 國立中正大學 === 心理學研究所 === 100 === Structural equation modeling is a common method used by social and behavioral researches to test the hypothetic model, and fitting indexes provided are frequently used to judge the fitness between model and data. Fitting indices of a theoretical model and its equivalent model are identical, so the model cannot be selected statistically. Comparing to the studies of equivalent model which mostly focused on equivalent structural part in the past, equivalence of measurement model had only few examples. However, it has all shown that the equivalent measurement models might threat the inference of researches on number of latent variables and quality of questionnaire. This research discussed the equivalence issue of measurement models. With the first study, the required constraints and rules to generate equivalent measurement model are derived under five situations, and verified with simulated data. These five situations include: the replacement of correlation between latent variables with covariance of measurement errors, the replacement of correlation between latent variables with factor loadings, the replacement of factor loadings with covariance of measurement errors, the extra latent variables generated from covariance of measurement errors, and the replacement of different covariance structure of measurement errors. The above five rules can be properly applied under either one-factor or two-factor conditions; however, the last two may not be able to generate equivalent models if extra factors are added. In the second study, five empirical researches are taken as examples to demonstrate the generation of equivalent models according to the rules derived in the first study. We also discussed the plausibility of the equivalent models We found all the equivelant models generated from the five empirical researches could all be plausible. The equivalent measurement models are indeed possibly threatening the inference based on measurement models in empirical researches. With these two studies, we found equivalent measurement models have the possibility to influence the validity of research conclusion; several suggestions are also provided, including to create possible equivalent models before data collection and assessing possible influence, to avoid the confusion of result explanation on equivalent models with research design or theory.