Research Watch

Foster care reunification rates: Comparing two types of analyses

Year of Publication
Reviewed By
Kristen Lwin
Citation

Putnam-Hornstein, E. & Shaw, T. (2011). Foster care reunification: An exploration of non-linear hierarchical modeling. Children and Youth Services Review, 33, 705-714.

Summary

The California Child Welfare/Case Management System tracks every child who has had contact with California’s child welfare system This article compares two statistical methods for predicting which children end up being reunified with their caregiver of origin within six months of entering out-of-home care. The first analysis used the child as the first unit of analysis. Since these children were nested within families, and children within a given family were assumed to be more similar than children from different families, the family unit was the second level of analysis (clustering unit).

In the second analysis, a single child was randomly sampled from each family (thus removing the family as a higher level or clustering unit) and the clustering unit became the county in which the child was removed. Child level predictors include race/ethnicity, gender, age at entry, removal reason, placement type, substance use, and number of placed siblings. Family level predictors include drug or alcohol treatment recommended in the case plan, and number of siblings living in out-of-home care. The county level predictors include the out-of-home care child entry rate, the teen birth rate (risk factor strongly associated with child maltreatment), and percentage of the population that was Black.

The findings in the first analysis are consistent with prior reunification research. The odds of being reunified within six months for a child removed for physical or sexual abuse were significantly greater than those for a child removed for neglect. Further, the odds of reunification within six months were significantly lower for children placed with kin than for children placed with non-kin. Additional factors that decreased the odds of reunification were parental substance abuse, larger number of siblings.

The second analysis, which included the county as a clustering level, similar factors as the first analysis remained significant predictors of reunification. However, in the second analysis Black children were found to have significantly reduced odds of reunification compared with Hispanic children, which was a different result than in the first analysis where race was a non-significant coavariate. Although the caseworker, supervising county, or state were found to have some influence on reunification, the greatest contributors to reunification were identified at the child or family level.

Methodological Notes

Hierarchical or multilevel models are becoming more common in the field of social work and specifically child welfare. Most statistical models assume that observations are independent. However, multilevel statistical models consider the nested structure of data and that observations may be dependent upon various micro and macro level factors. This article compares two multilevel models using standard logistic regression and non-linear multilevel models to analyze the odds of reunification of a cohort of children in out-of-home care in California.

Authors note several limitations to the study. The manner in which siblings were used to construct family units is distinctive and there are several other classification schemes that could have been used. Authors suggest that ideally the data would have been analyzed using a three-level hierarchical model, where there is statistical consideration that siblings are nested within families, and families nested within counties. However, the binary nature of the outcome variable, the extremely high correlation within families, and smaller cluster size of family units created mathematical issues.