Under pressure: Investigating how system-level factors shape racial inequality in child welfare outcomes
Whether one looks at reporting for maltreatment, time spent in foster care, or rates of foster care placement following a substantiated report of maltreatment, research shows disparities in the experiences of Black and White foster children and youth.
For our study, we set out to add to this body of work in two distinctive ways:
- First, although much policy attention is focused on congregate care, little of that attention has zeroed in on whether Black youth are more likely to be placed in non-family settings than youth of other races and ethnicities. If, as critics of congregate care suggest, we should rely less on congregate care in favor of family-based care, it is important to ask how this preference plays out in terms of disparity. The same can be said about running away from foster care: Because running away has deleterious effects on the well-being of children, we should know whether Black youth run away more frequently than youth of other races and ethnicities, all else being equal. Again, there is some research that looks at how often young people run away, but little of that research explicitly targets matters of racial disparity.
- Second, it is often said that system structure is one of the reasons why disparities persist. However, there has been relatively little work done to define more precisely what we mean when use the term system structure and whether it accounts for variation in the level of disparity we find. Should it exist, such a connection ought to change how we think about reducing disparity. To that end, we set out to define system structure and to determine whether, when viewed through this lens, our understanding of disparity is more actionable.
At the heart of our study, we were interested in ecological similarity: When context is held constant, how does our view of disparity among Black, Hispanic, and White youth change?
How did we look?
For our first aim, we looked at the intersection of congregate care utilization, running away, and race/ethnicity. Going in, we knew that Black youth are more likely to be placed in congregate care than White or Hispanic youth. We also knew that young people placed in congregate care are more likely to run away. But we didn’t understand yet the connection between the two: Do Black youth run away more frequently because they are more likely to be placed in congregate care? This seemingly straightforward question gets complicated quickly: Youth who enter foster care in urban counties are more likely to be either Black or Hispanic, and both congregate care utilization and running away are strongly correlated with urbanicity. In other words, our measure of disparity had to untangle the effects of race/ethnicity from the joint effects of both congregate care placement and urbanicity.
Our measure of disparity had to untangle the effects of race/ethnicity from the joint effects of both congregate care placement and urbanicity.
For our second aim, we sought to measure system structure and then link the absence/presence of that structure to the level of disparity. For this, we turned to what health economists call supply-induced demand elasticity (SIDE). In simple terms, SIDE refers to the idea that the supply of something affects the demand for that something (Wennberg, Barnes, & Zubkoff, 1982; Delamater et al., 2013; Roemer, 1961). For example, an increasingly persuasive body of evidence shows that the supply of ICU beds dictates their utilization. Similarly, our work on SIDE effects on congregate care utilization points to strong evidence that congregate care tends to be used because we maintain a supply of those beds (Wulczyn & Halloran, 2017).
We thought that the presence of SIDE effects would vary by county, in part because access to congregate care placements differs from one county to the next. To test that idea, we measured the county-level SIDE effect in more than 900 counties using counts of admissions and discharges from a time series of 700-weeks (Wulczyn & Halloran, 2017). With those results, we were able to group counties into two categories: counties where we detected a SIDE effect and those where we did not. We used a threshold of statistical significance to say whether the county was one where we could see a SIDE effect. That county distinction is what went into our final models as a measure of system structure.
What did we find?
In terms of running away and congregate care utilization, the experiences of Black, Hispanic, and White youth differ from one another. That said, there is no easy way to summarize those differences other than to pay close attention to the form of disparity and how it varies with respect to person and place. (Time is important too, but this angle is beyond our current scope of work.)
As to the form of disparity, we measured separately the log odds of either running way and being placed in congregate care at the child-level, with children nested in counties. At the child-level we use age, gender, and race, plus a bit about the young person’s placement history, as covariates. At the county-level, we have information about whether the county is urban and whether we detected a SIDE effect. This latter set of covariates provides a link to system structure. We expected rates of congregate care placement would be higher in counties where we see a SIDE effect. We also expect running away to be higher where the utilization of congregate care is above average. The differential exposure of Black youth to these conditions is an important part of a familiar story.
A simple unconditional model shows that Black and Hispanic youth are both more likely to run away from foster care and be placed in congregate care than White youth. However, as you add layers of nuance to that story, the narrative subtly shifts. Two observations seem especially relevant. The first relates to how we measure disparity. As a ratio of two numbers, there is no presumptive association between the magnitude of the rate and the disparity between them. In the case of younger Black girls and younger White girls, compared to the average, the rate of running away for both groups is much lower, but the disparity between them is larger. At the other end of the age spectrum, rates of running away are much higher among Black and White females, but the disparity between tends to be smaller. That insight leads to the second observation: There is not just one common level of disparity. Instead, how and where we look determines what we find. Others have called this the “third variable problem” (Armistead, 2014; Knight & Winship, 2013; Neil & Winship, 2019; Pearl, 2014).
The differential exposure of Black youth to these conditions is an important part of a familiar story.
With the basic disparity rate in front of us, we asked how disparity rates vary with respect to counties. Are there qualities of counties associated with whether the rate of running away there is above or below the average? Running away is higher in urban counties. It is higher among young people placed in congregate care. Black youth are generally more likely to run away, but the differences are to a large extent explained by where they live—in urban counties with easier access to congregate care. We found that utilization of congregate care, measured as the probability a young person would ever be placed into congregate care, is influenced by the SIDE effect. In counties with a strong SIDE effect, the log odds of spending time in congregate care were about 80 percent higher than in counties with no SIDE effect. We also found that rates of running away are also correlated with the strength of the SIDE effect.
In sum, we found that our understanding of disparity does depend on where and how we look. In counties with a SIDE effect, placement rates are higher regardless of race and ethnicity. From the perspective of the overall disparity rate, the supply effect matters. Where the supply effect is strong, the underlying placement rates of all children are affected. So too are the disparity rates.
What does it mean and what should we do?
In our study, we wanted to examine links between disparity and system structure. To do so, we first measured racial differences in running away and placement in congregate care. Next, we considered SIDE effects (i.e., supply-induced demand), which, in the case of congregate care, arise because of how we fund congregate care. Specifically, the organizations that provide congregate care tend to be private agencies; their stability as organizations requires fiscal stability, and their fiscal stability depends on utilization in fee-for-service systems. The relationship between organizational stability and utilization of services is the structure of interest in that it causes the system to lean toward utilization. Supply induces demand, which means supply is an important element in the decision-making context, even though we tend to think of clinical decisions as being immune from these sorts of subtle pressures.
Young people are more likely to be placed in congregate care if the county where they are from shows strong signs of a SIDE effect. On top of that, young people placed in congregate care are more likely to run away. Lastly, it turns out the Black youth tend to live where SIDE effects are strongest. The three factors together, including the SIDE effect, go a long way toward explaining the disparity we see.
Supply is an important element in the decision-making context, even though we tend to think of clinical decisions as being immune from these sorts of subtle pressures.
Although there are dots that have yet to be linked explicitly, the connection between fee-for-service reimbursement models and supply-induced demand is the sort of explanation one should expect to find when looking for the connection between system structure and disparity. In this case, how we pay for a service may well affect how much of the service is used. If disparity rates are affected by that interplay, then altering these inherent dynamics could have positive effects for everyone involved. For alternative payment models, health care offers a number of potential solutions worth testing, including prospective payment models that are more focused on outcomes and with weaker ties to utilization.
Armistead, T.W. (2014). Resurrecting the third variable: A critique of Pearl’s causal analysis of Simpson’s paradox. The American Statistician, 68(1): 1–7. https://doi.org/10.1080/00031305.2013.807750.
Delamater, P.L., Messina, J.P., Grady, S.C., WinklerPrins, V., & Shortridge, A.M. (2013). Do more hospital beds lead to higher hospitalization rates? A spatial examination of Roemer’s law. In Fielding, R (ed.) PLOS ONE 8(2): e54900-17. https://doi.org/10.1371/journal.pone.0054900.
Knight, C., & Winship, C. (2013). The causal implications of mechanistic thinking: Identification using directed acyclic graphs. In S. Morgan (ed.)Handbook of Causal Analysis for Social Research. Springer.
Neil, R., & Winship, C. (2019). Methodological challenges and opportunities in testing for racial discrimination in policing. Annual Review of Criminology, 2(1): 73–98. https://doi.org/10.1146/annurev-criminol-011518-024731.
Pearl, J. (2014). Comment: Understanding Simpson’s paradox. The American Statistician, 68(1): 8–13. http://ftp.cs.ucla.edu/pub/stat_ser/r414-reprint.pdf.
Roemer, M. (1961). Bed supply and hospital utilization: A natural experiment. Hospitals, 35(November): 36–42.
Wennberg, J.E., Barnes, B.A., & Zubkoff, M. (1982). Professional uncertainty and the problem of supplier-induced demand. Social Science & Medicine, 16(7): 811–24. https://doi.org/10.1016/0277-9536(82)90234-9.
Wulczyn, F., & Halloran, J. (2017). Foster care dynamics and system science: Implications for research and policy. International Journal of Environmental Research and Public Health, 14(10): 1181–12. https://doi.org/10.3390/ijerph14101181.