Every day we are surrounded by circumstances that affect the decisions we make.
Many of these are completely outside our control. As one example, working with state child support enforcement agencies, we have found that older noncustodial parents are more likely to make their monthly
child support payments than younger parents.
1 However, with child welfare data we have examined, older
parents appeared to be less likely to be successfully reunified with their children. In both examples, while
the parent’s age is correlated with the outcome, it’s completely outside the control of a caseworker.
Drawing from an example from a very different industry,
if you own a home in a flood zone below the National Flood
Insurance Program’s Base Flood Elevation (BFE), the cost of
purchasing flood insurance can be quite high. But unless you
have a magic wand to raise your house above the BFE, affecting the premiums you pay is largely outside your control.
No matter the industry, we can observe the impact of
factors outside of our control on outcomes, but to maximize
impact, we need to figure out how to leverage those factors
within our control. Identifying and acting on those factors
represents our biggest opportunity for change.
For child welfare agencies looking to improve their
performance on permanency metrics such as the timeliness and stability of returning children home safely to their
families, measured by looking at reunification and re-entry
rates, caseworkers need to understand the full spectrum of
uncontrollable and controllable variables that are most likely to provide warning signs and signals of an unsuccessful
outcome. Nationally, there are more than 400,000 children
in foster care and over 100,000 children currently waiting
to be adopted, with an estimated 255,000 children entering foster care and 238,000 exiting foster care every year.
Almost every unsuccessful case in the database represents
an unwanted outcome for a child. But we can learn from
the current and historical cases, both successful and unsuccessful. With these data we can explore, mine, and identify
critical signals and potential opportunities for intervention
throughout the life cycle of a child welfare case.
This article will highlight exploratory work Deloitte
Consulting conducted in collaboration with the child
welfare agency in the District of Columbia, describing predictive models developed using a mix of controllable and
uncontrollable factors, which, combined with behavioral
nudges and customized interventions, can help caseworkers increase the likelihood of achieving timely and stable
For modeling purposes, the authors define a timely
and stable reunification as a child who is reunified with
family within 365 days after removal (i.e., timely) and
is not removed again within 365 days after reunification (i.e., stable).
Setting the Stage
As more and more child welfare systems across the nation
find themselves described in the news as “troubled,” we
have to consider whether they have a “line-of-sight” prob-
lem before we rush forward to prescribe and implement a
set of interventions and solutions—even one as potential-
ly helpful as predictive analytics. While the conversation
about using predictive models to improve child welfare
practice is one that has legs nationally, the question is not
simply “Can you predict something?”, but rather, “If you
knew in advance that a reunification had a greater likeli-
hood to be successful or unsuccessful, what could you do
help them get
How Predictive Modeling Can Improve
Outcomes in Child Welfare Cases
By Kevin Bingham, B.J. Walker, Brenda Donald, and Beryl Washington