General guidance for custom-built structural equation models

Abstract

Structural Equation Modelling (SEM) represents a quantitative methodology for specifying and evaluating causal network hypotheses. The application of SEM typically involves the use of specialised software packages that implement estimation procedures and automate model checking and the output of summary results. There are times when the specification details an investigator wishes to implement to represent their data relationships are not supported by available SEM packages. In such cases, it may be desirable to develop and evaluate SE models “by hand”, using specialised regression tools. In this paper, I demonstrate a general approach to custom-built applications of SEM. The approach illustrated can be used for a wide array of specialised applications of non-linear, multi-level and other custom specifications in SE models.