The analytic approach is described in Chapter 3. Basically, we allow earlier measured variables to affect subsequently measured ones as well as the ultimate outcome variable of interest.

Table E.1 shows the variables included in this analysis. All variables have two categories. Some variables are naturally dichotomous, such as “is the respondent using birth control at time 2?” Other variables are polytomous, such as number of subsequent children born. For the polytomous variables, we have examined various contrasts and have chosen one that captures the greater impact on the variables being examined, that makes the most sense substantively, and that produces a reasonable distribution between the two categories. Since we have dichotomized all variables, we do not have to model further the functional form of effects. This is a major advantage given the broad audience we are trying to reach – such discussions of functional form can become quite technical and the relevant tests can become quite involved. In addition, our sample is very small, thereby making fine distinctions between various functional forms difficult. Instead, our goal is to identify the important factors leading to variability in the subsequent life course. Subsequent research should replicate our work and extend the analysis to consider the functional form of effects.

We used a commonly used logistic-regression package to estimate the multivariate models in this chapter (i.e., the logistic-regression package in SAS; see Harrell, 1979).