Stimulated by Jacques A. Hagenaars' comments, I have given further thought to the issue of indirect effects, and I no longer subscribe to the opinion stated on page 165 of CAUSALITY (last paragraph of Section 5.4.2) that indirect effects lack intrinsic operational meaning because they cannot be isolated using the *do(**x*) operator. I have found that it is possible to define partial effects, confined to any given set of paths without controlling variables on the remaining paths. This permits the assessment of indirect effects in nonparametric models, as well as in nonlinear parametric models (e.g., the logit variety). Details can be viewed in R-273-U [ps] [pdf].
Best wishes,

========Judea Pearl

]]>