The analysis in Chapter 3 was designed to show the causal paths by which independent variables influenced the outcomes of interest. In some cases a variable's influence was almost totally direct e.g.; there were no significant indirect effects. Other variables had weak direct effects but several strong indirect effects operating through other variables included in the model. Here we are interested in the relative magnitude of effects, both direct and indirect. Which variable has the greatest effect on these outcomes? To answer this question, we evaluated the change in the proportion on welfare (the proportion economically secure or having three or more children) that resulted from altering the characteristics of the population. For instance, given the observed proportion whose parents had low education, what proportion do we expect to be on welfare? If we now assume everyone's parents have low education and then in turn assume no one's parents have low education, we have a series of proportions that indicate the predicted impact of this variable. Such proportions are graphically presented in Chapter 6 for variables included in the analysis.

We chose to show the estimated effects in proportions because we felt they would be more meaningful to a broader audience than changes in odds or log odds. Since our prediction equation estimates individual likelihoods and not sample proportions, we transformed each coefficient by multiplying it by [P ✻ (1 − P)], where P is the mean of the dependent variable (see Hanushek and Jackson, 1977).