Linking Child Welfare Records to Population Based Data: Results From California

Background and Purpose: An inherent limitation of administrative data is the scope of the information contained in any one database. However, increasingly sophisticated methods for linking records allow individual-level information captured in one database to be extended with information captured in other databases. This paper presents findings from linkages established between California's administrative child welfare data and statewide vital birth records. Although prior research has examined the predictive value of birth indicators for infant maltreatment (Wu, et al., 2003) and infant entries to foster care (Needell & Barth, 1998), this is the first study to 1) examine the birth characteristics of all children who were referred, regardless of the substantiation or removal decision made, and 2) examine associations for children referred after infancy.

Methods: This study matched child welfare records from California's administrative CWS/CMS database to the state's vital birth records. The records of all children who were born in 2002 and referred to child protective services (CPS) for possible maltreatment before age 5 (n=88,052) were extracted from CWS/CMS and matched to the state's 2002 birth file (n=531,035). Records linkages were completed using probabilistic matching software which established linkages based on a combination of unique and non-unique child and parent identifiers common to both data sources. 84% of child welfare records were successfully linked to a birth record. The linked dataset was analyzed as a prospective birth cohort study with a referral to CPS serving as the dependent variable of interest and various biomedical and sociodemographic variables included as independent variables. The relative risk of CPS contact was estimated using generalized linear models with a Poisson error distribution, log link, and robust standard error adjustment (Zou, 2004). This method allowed for the log of the probability of CPS contact to be modeled as a linear function of the independent variables.

Results: 14% (74,374) of California's 2002 birth cohort were referred to CPS for possible maltreatment before reaching their fifth birthday; just over one third had a substantiated allegation. Risk of CPS contact rose in a linear fashion with increasing family size and each additional month that passed before prenatal care began. Prior abortion history, a birth abnormality, and low birthweight all emerged as significant biomedical markers of future CPS contact, although the magnitude of the risk was less than in prior studies examining only infants. Several sociodemographic factors were notable for their association with risk of CPS contact: the children of immigrant mothers were referred at less than half the rate of US-born mothers; children for whom there was no father information entered in the birth record were almost twice as likely to be referred.

Conclusions and Implications: Increasingly sophisticated probabilistic matching methodologies coupled with the growing availability of administrative data position record linkage as a powerful tool for generating new knowledge. This paper presents findings from an ongoing effort to link California's administrative CPS records with population based data.