Chapter 6: Total Unimodularity and Integrality
7.4 Future Work
The design of this study and its results can be used for future studies of the United States foster care system. By selecting just infants from large central-metro counties in Texas, we limited the dataset dramatically to a subset of cases with less variance in environmental conditions. However, this model can be applied to any subset of children in the foster care system across the United States. The model can explore subsets of children in different age groups, regions, risk factors, etc. Figure 10 demonstrates the scope of our project in comparison to all children in foster care as a result of the opioid epidemic and to the foster care system as a whole.
Figure 10: Project Scope in Greater Context
An example of an interesting comparison to our scope could be done within children of different age groups but also within urban counties of Texas Urbanicity. This could aid with determining how services are allocated to everyone from those areas and how your age can influence which services are most effective. Overall, it could better help the child welfare
workers know what each child needs most to benefit them, and what combination of services provide the best results in decreasing the length of stay a child has.
Another comparison could be done on infants from Texas across all urbanicities. This would be beneficial because some services are more relevant for more urban areas than rural areas. Transportation Services, for example, would likely be better suited for children in rural areas where services are not typically within walking distances, as opposed to urban areas where public transportation is more common. By recognizing any common services across all six urbanicities and comparing their usefulness to each other, researchers could determine which will have a more positive impact on certain areas and should be more heavily allocated to those urbanicities than ones where children will not benefit as much.
Rather than solely focusing on age or Texas, future studies could be extended to different infants in Urbanicity 1 locations across the nation. These studies could be beneficial because the child welfare system is different from state to state, and what one state excels at could be another state’s weakness. By comparing a small subset of cases from Urbanicity 1 locations from California, Texas, and New York, for example, the CWS could easily determine how resources are allocated differently and provide constructive feedback on what each state could do better to help children be more quickly returned to normalcy. Something to take into consideration when comparing states to each other is the fact that each has its own jurisdiction within the CWS.
There is the possibility that each has a system so unique from the others that they cannot be compared, rendering a state-by-state comparison interesting but unable to draw conclusions from.
Overall, the data collected by NCANDS and AFCARS as well as the environmental factors incorporated can all be used to slice the data into comparable subsections. Child age, urbanicity, and state are clearly determining factors of the types of services a child receives in care, but the environmental factors are contributors as well. Environmental slices that can be considered are whether the child entered the system due to abuse versus neglect, whether they reside in a high or low poverty area, or if there are a lot of crimes versus no crime in their
neighborhood. These comparisons will provide valuable insights for the child welfare system to use when matching children and services.
Instead of inputting the results from the regression analysis into an optimization model, one could use the results to determine the services that have the greatest impact or those that are in high demand. After talking with Adam Schaffer from the Harvard Kennedy School andDr.
Melinda Gushwa from Simmons University School of Social Work about child welfare service allocation and budgeting we found that a major issue within any state’s child welfare system is inability to properly forecast the demand for a given service each year. We noticed this trend within our when analyzing the coefficients that were derived from our regression analysis. Our model interprets all negative coefficients as a benefit to reducing a child’s length of stay. While all positive coefficients add additional days to a child’s length of stay. This would suggest a state should invest more in the services with negative coefficients. However, from our discussion with Dr. Gushwa, we determined that just because a coefficient is positive, it does not necessarily mean it does not provide a benefit to the child or family. A coefficient could be positive because it is in high demand and the state did not budget enough resources towards it. Thus, the child or family had to wait for the service which increased their length of stay within the system. An extension of forecasting the demand of services to properly budget for them would require more granular data than the dataset we worked with. One would need to know how much of a service was budgeted for in a given year and the number of times that service was distributed in a given year.