Research Report—The Importance of Specification in Causal Modeling: The Case of End-User Computing Satisfaction

In a survey of IS instruments spanning the years 1973 to 1988 (Zmud and Boynton [Zmud, R. W., A. C. Boynton. 1991. Survey measures and instruments in MIS inventory and appraisal. K. I. Kraemer, ed. The Information Systems Research Challenge Survey Research Methods, Vol. 3. Harvard Business School, Boston, 149–180.]), Doll and Torkzadeh's (Doll, W. J., G. Torkzadeh. 1988. The measurement of end-user computing satisfaction. MIS Quart. (June) 259–274.) 12-item End-User Computing Satisfaction instrument was reported as one of three IS instruments that met conditions to qualify as “well developed.” Recently, Etezadi-Amoli and Farhoomand (Etezadi-Amoli, J., A. F. Farhoomand. 1991. Issues and opinions on end-user computing satisfaction. MIS Quart. (March) 1–5.) questioned the validity of these measures. Part of their critique centered on the poor model fit obtained in a re-analysis using LISREL. While other potentially valid points were raised by Etezadi-Amoli and Farhoomand's critique, this report focuses only on their use of confirmatory factor analysis. In our re-analyses of Doll and Torkzadeh's original covariance measures, we show how model fit is extremely dependent on model specification. While still maintaining the same number of constructs and respective measures, we demonstrate how two alternatives to the original model analyzed by Etezadi-Amoli and Farhoomand can result in models with acceptable fits.