Return to MUSA 801 Projects Page
This project was produced as part of the University of Pennsylvania’s Master of Urban Spatial Analytics Spring 2020 Practicum (MUSA 801) taught by Ken Steif, Michael Fichman, and Matt Harris. We would like to thank the team at Guilford County Child Protective Services for providing much of the data used in this analysis.
This document is intended to enable others to replicate a study of social worker assignment to child welfare cases. We first introduce the context of this project, followed by an exploration of the data and predictive modeling procedures. We’ve included hyperlinks throughout the document, plus a closing appendix with details on data transformation and code.
In our use case, we’re working to optimize the assignment of social workers to child welfare cases, based on the predicted difficulty of each case. Our report focuses on social workers in Child Protective Services (CPS) in Guilford County, NC.
Currently, social workers in Guilford County are assigned to new cases alphabetically based on the social worker’s last name. This would result in an equitable distribution of cases only if cases were similar in duration and difficulty. Instead, because cases differ in duration and difficulty, social worker caseloads vary widely.
To improve this condition, we first had to define a difficult caseload. We determined that a difficult caseload is one that includes (1) a disproportionate number of cases and/or (2) cases that ultimately result in the most stressful result for a child: being removed from the home. This result, being removed from the home, is referred to as a “transfer to services”.
With this definition of difficulty in place, we explored the data and determined that the most experienced social workers are most likely to be overburdened by difficult caseloads. To improve this condition, we built a model to predict if a case will result in a transfer to services. To put this predictive modeling to practical use, we developed an app to help social worker managers assign cases with difficulty in mind.
Our ultimate goal is to make it easier for the managers of social workers to assign their team members equitable caseloads. By making this administrative task easier and its result more equitable, we hope to enable social workers to spend more time on their important and highly demanding work with families. To preview our app, click here
In the US, more than 5 million reports of suspected child abuse are made annually. These reports include physical harm, emotional harm, and sexual abuse and exploitation. The potential consequences of abuse are severe - the US Department of Health and Human Services reports that an estimated 1,720 children died from abuse and neglect in fiscal year 2017. Individuals who experience abuse in childhood are more at risk of experiencing or perpetrating abuse as adults, with accompanying personal and financial impacts.
Our use case focuses on Guilford County, NC, the third most populous county in North Carolina. Guilford includes the city of Greensboro, the third most populous city in North Carolina. As of the 2010 census, Guilford County had a population of over 500,000 people or 192,064 households. 30% of households included children under the age of 18.
The Guilford County Department of Social Services recieves more than 4,000 reports of suspected child abuse each year, which represents 3% of the 125,000 reports made annually across all counties in North Carolina. These cases are recieved, investigated, and serviced by a system of intake worker, social worker, and interventionist teams.
To understand the existing condition, we focused our exploratory analysis on four areas:
The sankey chart below illustrates the progression from Stage 1 (intake) to the outcome of the assessment period (transfer to services or close without transfer to services). Although the vast maority of cases are screened in initially, only a small portion ultimately transfer to services.
Cases Assigned by Time Period
New reports are made to Guilford County CPS nearly every day. The total number of cases assigned each month during our study period ranged from 570 to 800.
The number of cases assigned each day varies widely, from 1 to 23. On average, the there are between 6 and 11 cases assigned each day, depending on the month.
Cases Assigned by Social Worker The number of cases on each social worker’s caseload varies. Figure 4b.3 is a heatmap in which cells in red represent months when a social worker has a particularly large caseload- up to 36 cases.
Social workers are arranged by months of employment, with the most experienced social workers at the top. A careful analysis of Figure 4b.3 reveals that the social workers with the highest number of cases in a month tend to be those who were in their roles for all, or nearly all, 31 months of our study period. This leads to the next section of our analysis, a comparison of social workers by duration of employment.
As depicted in Figures 4b.1 and 4b.2, the number of cases assigned varies monthly and daily, making it difficult to determine a set “limit” to the number of cases that should be in a caseload. To determine which social workers are most overburdened, we refined our analysis to compare individual caseloads to team-wide averages.
To do this, we compared two values for every social worker and every month of their employement (1) the number of active cases assigned to the social worker in a month (2) the average number of cases assigned, that same month, across all social workers. The difference between these two values speaks to whether a social worker had a large or small caseload compared to peers.
We found that Group 4 social workers, those with the longest duration of employment, tend to have caseloads larger than the team-wide average. Group 1 social workers, those with the shortest duration of employment tend to have caseloads that fall well below the team average.
The histogram below, Figure 4d.1 displays these deviations for individual social workers, ordered by the social worker’s length of employment.
This difference between experienced and less experienced social workers held when we looked at averages across experience groups (Figure 4d.2). We averaged individual deviations within experience groups and found that Group 4, the most experienced social workers, continued to have caseloads larger than the team-wide average.
Why do the most experienced social workers consistently have the largest caseloads? Our analysis revealed that because cases are assigned alphabetically, without considering existing caseloads, it takes many months for a new social worker to accumulate a significant caseload. As less experienced social workers come and go from Guilford, the most experienced social workers are overburdened by disproportionately large caseloads.