Understanding Foster Parenting: Using Administrative Data to Explore Retention. Analyses


The study team conducted three types of analyses. First, a descriptive analysis examined the characteristics of foster parents over the years for which data were available. Characteristics of interest included the demographic characteristics described previously and license types.

A second set of analyses described the utilization of licensed homes. These analyses compared foster parents with different characteristics in terms of whether any children were cared for during the time the home was licensed, the average number of children in the home, and an annualized rate representing the number of new placements in the home. Foster parent characteristics were also used to describe the likelihood of providing care for infants, adolescents, and children with special needs.

Finally, the team used longitudinal analysis to model the length of service in the first episode of active foster parenting. These analyses produced measures of time that are less biased than those based on cross-sectional data (Usher, Wildfire, and Gibbs, 1999) because they make use of right-censored data in which the event of interest (in this case, exit from foster parenting) has not yet occurred. Life table analyses estimated the cumulative probability of exiting foster care within specified time periods and examined factors associated with varying lengths of service. Kaplan-Meier analyses provided estimates of median length of service for foster parents.

In addition to bivariate analyses of the relationship between foster parent characteristics and experiences, the study team tested multivariate models using Cox proportional hazard regression (Allison, 1995). These models yield hazard rates, which can be conceptualized as the likelihood of an event  in this case, exit from foster parenting  on any given day. Higher hazard rates indicate an increased probability of exit, hence, a shorter length of service. Because New Mexico data were limited in both years of data available and number of cases, multivariate models used data from Oklahoma and Oregon only.

View full report


"report.pdf" (pdf, 528.13Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®