Analysis of Biochemical Data Using Structural Equation Model

• Main Authors: Tsui-Wen Chang, 張粹文
• Other Authors: 吳宏達
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/73890640326133442392

碩士 === 中國醫藥學院 === 環境醫學研究所 === 91 === The data analyzed in this thesis was collected from a part of a community-based screening program as a cross-sectional study. It offered a premise of a longitudinal study in constructing a structural equation model (SEM), and served as a building block of a future analysis for disease diagnosis, perhaps with a consideration of genetic confounding. Since there was no complete physiological and pathological model a priori, it is not possible to obtain a measurement model, as a part of the entire SEM, from the confirmatory factor analysis. Instead, we borrowed the strength of exploratory factor analysis (EFA) and adopted a ‘two-stage’ algorithm as a hybrid method. That is, to set-up the constructs of structural and measurement components in a two-stage manner, and to estimate the parameters of the entire SEM simultaneously. The paths of structural component were selected according to several rules of significance via ordinary multiple linear regressions. After an acceptable construct was obtained, in terms of goodness-of-fit indices, the inter-disturbance correlations re-induced by the structural equation modeling after EFA, which only permitted independent disturbances, were considered in a stepwise order based on absolute ranks of marginal correlations. Finally, results of different statistical packages with different constraint conditions and estimates were briefly discussed.