Who Should We Worry About?
A critical question at the heart of every social service agency’s work is, “Who should we be most worried about?” The worry may be related to individual, family, or community safety; recurrence of a particular event; or other possible negative outcomes, such as dropping out of school or becoming homeless. Social service agencies have limited resources to address these worries, so targeting resources efficiently and effectively is of the utmost importance. But without knowing “who” to be most worried about, there is no way to know if resources are being directed appropriately. What can be done to help answer the critical question of “who?”
Social service agencies take various approaches to answer this question. On one end of the spectrum, the question may be answered using anecdotal evidence or “gut feelings.” This approach opens the door for biases and inconsistent decision making, and has low accuracy. There is also a strong possibility that the individuals or families who we should be most worried about will go unnoticed.
On the other end of the spectrum, the question might not be answered at all. This approach may result in no one receiving a service or intervention, or everyone receiving a service or intervention. The implications of no intervention are obvious; however, there are also implications of over-intervening with individuals and families. It is more costly and may result in wasted resources. In some instances, intervention may actually be detrimental for individuals and families.
Another, more reasonable approach hinges on using known information about individuals and families to answer the question. In other words, agencies can use data to identify “who” they should be most worried about. One way to leverage data is by using predictive analytics.
Predictive analytics, as discussed here on the blog this month, is a body of statistical methods related to data mining, modeling, and machine learning. These methods use known information to make a prediction about what will happen in the future. Dr. Jesse Russell explains some of the basic concepts of predictive analytics in this blog post.
Here’s one example of how NCCD has worked with a child welfare agency using predictive analytics to answer the “who” question. This child welfare agency was interested in examining ways to better identify children who are likely to suffer from mental health conditions to ensure that the agency could help children get prompt treatment and support. NCCD used predictive analytics to explore how the agency could identify which children to worry about the most.
Using information stored in the agency’s Statewide Automated Child Welfare Information System (SACWIS) and additional information provided by a behavioral health agency, NCCD built a dataset of children who had been served by the child welfare agency in the past. This data set included characteristics about the child and the child’s involvement with the child welfare agency—for example, the type of maltreatment leading to child welfare intervention; how long the child had been involved with the agency; and information from safety, risk, and needs assessments. The dataset also included outcome variables that might indicate a mental health concern: for example, identification of an emotional or behavioral need or involvement with the behavioral health agency.
A predictive analytics model, Classification and Regression Tree (CART) analysis, was applied to the dataset to identify the factors most strongly associated with the outcome. This approach allowed NCCD to partition the sample of children into groups based on their chances of experiencing the outcome given certain characteristics. The analysis helped the child welfare agency to identify the set or sets of characteristics that had the strongest relationship to the outcomes. The agency is doing more work to see if these factors can serve as a way to pre-screen children for mental health concerns and connect them to the right resources as quickly as possible.
While predictive analytics works great for answering the questions around “who,” the method is not well suited for answering all questions that a social service agency encounters. After discovering who we should be most worried about, the next set of questions related to practice, including “What should we do?” and “How should we do it?”, require additional considerations to ensure effective and equitable approaches. Additionally, understanding why certain relationships or patterns emerge is key to reducing and eliminating inequities and biases. In this way, the intersection of analytics, social awareness, and strong practice knowledge will help us to achieve the best outcomes for individuals, families, and communities.