ABSTRACT

After completing work on confirmatory factor analysis and analysis of covariance structures around the end of 1968, Karl Jöreskog had laid the groundwork for a methodological development that was to have considerable impact on the behavioral and social sciences in the years to follow. His interests turned to structural equation modeling, a topic that had its first origins in the field of genetics in the work of Sewell Wright (1921, 1931, 1934) on path analysis, but which had been extended by biometricians (Turner and Stephens, 1959) and sociologists (e.g., Blalock, 1964, 1969; Duncan, 1966; Land, 1969; Heise, 1969; Costner, 1969) and approached independently from the theory of regression by econometricians (e.g., Klein, 1953, 1969; Wold and Jureen, 1953; Koopmans and Hood, 1953; Goldberger, 1964; Fisher, 1966). Jöreskog’s contribution to this literature was to merge the latent variable idea of common factor analysis with the traditional theory of systems of linear structural equations of measured variables, to produce a new and more general model along with an efficient algorithm for the estimation of its parameters. By the fall of 1970, he had completed a paper on his new system and presented it at a conference cosponsored by the Social Science Research Council and the Social Systems Research Institute of the University of Wisconsin at Madison, Wisconsin, November 12-16, 1970. This paper, published as Jöreskog (1973), was the first of many papers from his and others’ laboratories that was to be concerned with the new field of structural equation modeling with latent

with

variables. The conference also brought key figures concerned with path analysis, structural equations modeling, and causal modeling with correlational data from many fields and gave new impetus to the development of this methodological area. Because of the generality of Jöreskog’s model and the practicality of the computing packages that he made available for its implementation, much of the direction of this field in the years to follow were to be in terms of this model. Amajor consequence of Jöreskog’s model and the computer programs available for its implementation was an awakened interest of researchers who traditionally conducted correlational research in the issues of causality and the study of causal models, because the new methodology made this seemingly possible.