Evaluation of the Differential Response System in Connecticut: A Propensity-Score Approach Analysis

Background: In 2012, Connecticut’s Department of Children and Families (DCF) launched a statewide differential response system (DRS) and established two tracks for families reported for child maltreatment: traditional investigation for families with allegations involving high risk reports and safety concerns; and Family Assessment Response (FAR) for families with a low to moderate risk reports and no or low safety concerns. In the absence of a randomized clinical design, the primary aim for this study is to apply a propensity-score approach to estimate the effect of FAR, by comparing the rate of subsequent child maltreatment reports between the FAR group and the propensity-score matched investigation group.

Methods: Using a quasi-experimental design with a propensity score approach using CT DCF administrative data, we compared the investigation and FAR tracks for subsequent (SR) and subsequent substantiated (SSR) reports. For families receiving a child maltreatment report in 2015, the first report in 2015 was the index report and used to indicate group. In total, 42 covariates were used in the logistic regression to estimate the propensity score. These included: primary caregivers’ demographic information (i.e., age, race/ethnicity, primary language, and family constellation), and DCF’s report information (i.e., maltreatment report history in the past 24 months, index report risk assessment indicators, and who reported the allegation). Because some variables (e.g., Race/ethnicity) from the administrative dataset had more than 10% of missing data, multiple imputation was applied. A total of 10 different data sets with complete data were generated, and separate propensity scores were estimated on each. Results were averaged to derive the final propensity score. After the propensity scores were estimated, survival analysis with Cox regressions were conducted to analyze time from the index report to the first SR and the first SSR within a 12 month follow-up window.

Results: Before propensity score matching, the FAR group (n=9897) was significantly different from Investigation group (n=9942) on most of the covariates, with the FAR group more likely to be non-Hispanic White (44.5% vs 38%), two parent family (39.7% vs 36.1%), older age (33.29 vs 30.25), less likely to have hospital/physicians/healthcare workers as reporters (10.4% vs 14.8%), and had lower percentages for almost all risk assessment indictors. After propensity score matching, the survival analysis indicated that the FAR group had a significantly lower rate of any SSR in 12 months (OR=0.87, P=0.035) than the investigation group. There was no difference in the 12-month survival time to the 1^st SR between the two groups (OR=1.05, p=0.184).

Conclusions and Implications: Findings from our study suggest that FAR is having a positive effect on preventing substantiated maltreatment reports. The non-statistically significant findings for SRs indicates that FAR is as effective as investigation in preventing repeat maltreatment.  Others have found that the DR track is most effective with low risk families.  We will discuss the challenges of evaluating DRS with a non-randomized design, and limitations of the propensity score approach, including that it does not adjust for unmeasured differences between FAR and Investigation groups.