Reexamined previous statistical comparisons of two models for adolescent drug abuse using new statistical estimation methods in causal modeling not requiring assumptions about normally distributed variables. An asymptotically distribution free method shows that the models fit even better than assumed in the initial work. (Author/JAC)