Applying Structural Equation Modelling to Tested Measurement Model of Psychological Measure

This study tested measurement model of psychological attributes through structural equation modelling (SEM). Classical test theory divided measurement model into four: parallel, tau equivalent, congeneric and error-correlated. Parallel and tau equivalent model assumed that each of items in scale to measure had an equal amount of measure and error, whereas congeneric model assumed that each item has a different amount of both properties. Most measurements in psychological testing employ parallel and tau equivalent model since Cronbach’s alpha coefficient was used to estimate psychometric properties of measurement. Presumptions of parallel and tau equivalent are difficult to be met since items to measure represent a dynamic and complex human behavior. Since its represents a dynamic and complex human behavior, items will be more likely to have different precisely to measure. Authors propose a hypothesis that congeneric and error correlated model fit with data than parallel and tau equivalent model. The instrument was used in this study was (1) Big Five Inventory (BFI-44) which corresponds to personality measurement, (2) Beck Depression Inventory (BDI), which represents the clinical measurements, and (3) Attitudes toward policy changes organizations scale that represents the attitude measurement. BFI-44 and BDI are completed by students (N = 185 and 184) while the attitude scale was completed by employees. Using confirmatory factor analysis techniques through LISREL 8.30 suggest that congeneric and error-correlated model have better goodness of fit indices than parallel and tau-equivalent. In other words, the error correlated and congeneric models were better in representing complexity of item measure than parallel and tau equivalent model. These findings suggest to researchers who develop psychological measurements to: (1) identify a potential differences amount and precise of measures among items in a scale measure with factor analysis before estimating psychometrics properties of measurement (2) using the psychometrics properties that relevant to the measurement model of scale to get accurate estimation