To estimate trends in the probability of being a custodial parent, we use logit models. The model takes the form:

log(p/(1-p)) = a + b₁x₁ + b₂x₂

where p is the probability of being a custodial parent, x₁ is a vector of dummy variables for seven age groups and x₂ is a vector of dummy variables for years 1988 through 1999. Because the CPS sample sizes are large, we are able to develop separate models by gender, race and ethnicity, and marital status. The coefficients for the time dummies are then used to develop projections for each gender, race and ethnic group, and marital status.

The model for nonparent custodians is similar, except fewer age groups and fewer race and ethnic groups are included in the vectors of the dummy variables.