Exploring Nonlinear Path Models via Computer Simulation

Although many causal relationships are nonlinear, there has been little general exploration of the implications of using nonlinear forms in multivariate causal models. A program called MULTPATH ¹ has been developed that enables a researcher to create a path model with up to 6 variables, choose any of 10 curve forms for each of the paths, specify the extent of random error in each of the paths, and generate plots of resulting distributions. Three general conclusions appear warranted, although results are often far from simple: (a) If there is much random error present in such models, it is often impossible to deduce the underlying curve forms from the resultant bivariate plots; (b) the order in which curve forms appear in a single, extended path is critical to the resultant relationship between the origin and the destination of that path; and (c) there is a sense of hierarchical dominance when different curve forms converge on the same dependent variable.