Risk, Race/Ethnicity and Subsequent Reports in a State's Differential Response System

Background:

Racial disproportionality in child welfare-involved populations is often explained by bias theory, that institutional and individual racial bias results in different treatment, and risk theory, that risk factors for child maltreatment, like poverty, are more prevalent in families of color, resulting in higher maltreatment rates.  Connecticut is an interesting state to test these theories because of the a significant proportion of Black (11%) and Latino (22%) children allowing racial/ethnic group comparisons, and the state has one of the highest disproportionality ratios in the country for Black/White and Latino/White risk factors for maltreatment including child poverty and teen mothers. 

Method:

This study used administrative data from five years of Connecticut’s differential response system (DRS) (n=33,773 families) with a primary outcome of 12-month subsequent maltreatment report. Of the families, 49% were White, 18% were Black, 27% were Hispanic and 5% were Other races/ethnicities.  Families assigned to the DRS receive up to 45days of services and are then assessed for Additional Services, No Services, or Other Follow-up. The 12-month subsequent-report window began at the end of DRS services. Other variables include: any prior maltreatment report since 1996 (32%); administrative/geographic region; and a four-category structured decision-making-tool risk assessment (28% were Very Low Risk, 55% were Low Risk). Cox proportional hazard modeling was used to analyze subsequent reports. All analyses were conducted in SPSS.

Results:

Subsequent report rates were similar for White, Black and Latino families (24.3%, 25.5%, and 25.6% respectively) but only 17.3% for families of Other races/ethnicities (χ²(3)=64.5,p<.00).  At the bivariate level, race/ethnicity was associated with: disposition (χ²(9)=267.0,p<.00) with more Latino families (21%) being assigned to additional services than Black (17%) or White  (14%) families; assessed risk (χ²(9)=1292.1,p<.00); prior reports (χ²(3)=177.9,p<.00), although the percent of families with prior reports was similar for Black (37%), White (31%) and Latino (33%) families, Other races/ethnicities were lower (21%). Post-DRS assessment was significantly associated with subsequent reports (χ²(3)=725.9,p<.00). 

Using hazard modelling, disposition, region, risk assessment and prior report were all significantly associated with experiencing a subsequent report faster.  Race/ethnicity was not associated with an increased risk. Being assessed for Additional Services (Exp(B)=1.9,p<.000) and the Other Follow-up category  (Exp(B)=1.6,p<.000); being in two of the six regions (Exp(B)=1.2,p<.01 and Exp(B)=1.2,p=.02); a Low (Exp(B)=1.5,p<.000), Moderate (Exp(B)=2.2,p<.000.) and High (Exp(B)=2.7,p<.000) risk assessments with Very Low as the reference group; and any prior report (Exp(B)=1.8,p<.000) were associated with a faster subsequent report. In a sensitivity test using block entry, the interaction between race and prior reports did not produce a significant change in the model. 

Conclusion and Implications:

Assessed risk and prior reports remain strong and significant predictors for future reports, but within the differential response system, families with similar risk patterns had similar outcomes, regardless of race/ethnicity. Given the amount of discretion involved in the differential response system, the lack of an effect for race suggests support for the risk model of racial disproportionality rather than the bias model. The issues identified in the standardized risk assessment and the variation in the regions are possible targets for reducing subsequent reports.