Predictive analytics is a technique that can be used to calculate a future event or outcome by using big data and machine learning. Big data refers to a process of gathering data from various data sources such as mental health data, substance use data, health data, and socioeconomic data. This data mining technique helps to identify trends, patterns, and relationships that are used to predict future outcomes. An algorithm is developed to sort through all of the data gathered. The data are continuously compiled, which better trains the algorithm and leads to even better outcome predictions. For example, Netflix will populate suggestions for y T.V. viewers based on your viewing history. The more Netflix shows viewed, the more the “machine” learns and the better the predictions are for the viewers. Predictive analytic models have become increasingly popular within the child welfare field because of the potential to enhance decision-making. Child welfare professionals are often tasked with making decisions that can have long-lasting effects on children and their families. These decisions are often made quickly and with limited information. Assessment tools have been used to reduce subjective biases and increase consistency among decision-makers; however, these tools are still prone to human error and bias. Predictive risk models are thought to reduce these errors and biases; however, these models are not entirely free from bias. Selection bias is of particular concern, as minority children are disproportionately represented in child welfare services. The algorithm, using biased data, could inadvertently assign a higher risk score for maltreatment among minority children. The Institute, in partnership with the Department of Children and Families, developed an implementation plan for a child welfare predictive analytics project. The aim of this project is to determine to what degree the predictive risk model affects pre-commencement activities such as workload standards, worker efficacy and confidence, and chronic re-maltreatment by caregivers. The Institute, in partnership with Dr. Patricia Babcock in the College of Medicine at Florida State University, is currently working on an evaluation of the predictive analytics project in child welfare. The evaluation has four components: a formative evaluation, which will address if the research has worked; a process evaluation, which will determine if the plan was implemented correctly; an impact evaluation, which will address the changes taking place during the implementation phase; and an outcome evaluation, which will determine if the research questions and hypotheses have been answered. The model roll-out is set to occur in March 2019. Data will be collected over a 12-month period using surveys, semi-structured interviews, and focus groups aimed at determining the effects of a predictive model on child protection activities.
Data and Statistics 101: Key Concepts in the Collection, Analysis, and Application of Child Welfare Data
Using Artificial Intelligence, Machine Learning, and Predictive Analytics in Decision-Making
- Jessica Pryce, Ph.D., MSW, Florida Institute for Child Welfare
- Anna Yelick, Ph.D., MSW, Florida Institute for Child Welfare
- Ying Zhang, Ph.D., Florida Institute for Child Welfare
- Kreig Fields, PMP, ACP, Data Engineer/Scientist