CAUSAL MODELS IN ORGANIZATIONAL BEHAVIOR RESEARCH: FROM PATH ANALYSIS TO LISREL AND BEYOND
Larry J. Williams Purdue University
Lawrence R. James University of Tennessee
Researchers in the field of organizational behavior (OB) have shown a long-standing interest in testing causal hypotheses with correlational data. In the early 1970s, a practice was established of representing these hypotheses graphically and referring to the figure as a "causal model," because paths represented in the figure were presumed to represent influences between variables that were consistent with causal effects.
With this approach, the subsequent evaluation of the causal model was achieved by using statistical techniques to obtain estimates of the magnitude and direction of relationship depicted in the model. The resulting popularity of this methodology may have been because of its easy application to the survey data commonly analyzed by many OB researchers: survey research enabled them to conduct their science with reduced concerns about the external validity of their findings.
Alternatively, OB researchers may have recognized and appreciated the benefit that can be provided when complex processes are presented visually, and this may have contributed to the extent to which this methodology was embraced by organizational scholars. Regardless of the reason for its emergence, it is evident that over the past 20 years, applications of causal modeling in the field of OB have increased dramatically.
At the same time that this interest in causal modeling was developing, there was also an evolution in the data analytic techniques used to evaluate causal models. Early in this literature, OB researchers used partial correlations and, in some cases, simple correlations to obtain estimates of the paths included in a causal model. As time progressed, ordinary least squares regression became