next up previous contents
Next: Others Up: Path analysis Previous: Linear structural equation models   Contents


Nonlinear models

Stolzenberg (1980) [354] Measuring and decomposing effects in structural equation models with nonlinear components (but models still have an additive error term). Effect is defined as the parital derivative of Y w.r.t. X. Effects are decomposed in a similar way as for linear structural equation models.

Roncek (1991) [317] Minor comments on defining effects of binary (compute difference in predicted probabilities) and continuous (partial derivative of the predicted mean) explanatory variables on binary responses in logit models.

DeMaris (1993) [107] Comments to [317], with further discussion of interpretation of coefficients of logit models. Argues for focusing on odds ratios rather than probabilities. Further comment (passionate and incorrect) in Roncek (1993) [318].

Cox and Wermuth (1996) [100] Comments on the decomposition of regression coefficients in the linear case (p. 69) and the difficulty of doing the same in the nonlinear case (p. 77).

Wermuth and Cox (1998) [383] Various issues in associations implied by independence graphs. Includes discussion of parametric cnacellation and tracing of paths.



Jouni Kuha 2003-07-16