Predictive Risk Modelling
A predictive risk model is an automated risk scoring tool which generates a risk score for the occurrence of an adverse event through the use of an algorithm and large administrative data set. The AUT Centre for Social Data Analytics aims to incorporate these tools in their analysis of data in order to identify opportunities for social intervention and improve individual and community outcomes.
- For an example of the use of predictive risk modelling in health care, see here.
- For an example of the use of predictive risk modelling in a child and family setting, see here.
For more information about the application of predictive risk models in different research areas, see our presentations.
Presentations on Predictive Risk Modelling
Literature review: Predictive analytics in human services - PDF
This literature review was prepared by Dr. Thomas Packard for the Southern California Consortium of Human Services. Packard is a Professor in the School of Social Work at San Diego State University. The review considers the opportunity for using predictive analytics to gain optimal information and insights from big data. It includes a review of the existing use of predictive analytics in Allegheny County, Pennsylvania; Tennessee; Florida; and New Zealand. Packard also discusses the ethical issues that arise from the use of predictive analytics and predictive risk modeling, and provides suggestions as to how these issues may be addressed.
Implementing a Data Visualisation Tool for the Frontline in Allegheny County, Pennsylvania - PDF
This presentation covers the use of a reactive predictive risk model in child welfare referrals in Allegheny County. These slides were presented at a CSDA workshop on integrated data use in February 2016.
Predictive Risk Modelling Adverse Outcomes for Children: Case Studies from New Zealand
Berkeley, California, 2014 – PDF
University of Southern California Convening, 2014 – PDF
This presentation gives an outline of the use of predictive risk modelling in the fields of maltreatment and hospitalisation. It describes the development of a model for predicting child maltreatment in New Zealand.
Vulnerable Children: Can administrative data be used to identify children at risk of adverse outcomes? – PDF
This presentation provides and overview of the use of predictive risk modelling to identify children at risk of maltreatment and the development of an algorithm to do so. It was presented to the New Zealand Government Select Committee in 2013.
Predictive Risk Modelling: Options for New Zealand – PDF
This presentation outlines the development of a model to identify patients at risk of hospital admission, which could be used to allocate preventive care.
Predicting and Preventing ‘Triple Fail’ Events in Health Care Systems – PDF
This presentation describes the use of predictive risk models in the health care system to identify people to whom preventive care would be most effective. This was presented to Singapore Management University.
Predictive Risk Modelling: Theory, Practice and Prospects for Japan – PDF
This presentation covers the use of predictive risk models in providing preventive care, as well as identifying potential uses for this concept in Japan. These slides were presented to the Japan Cabinet Office in 2012.
Using predictive modelling to identify students at risk of poor university outcomes – PDF
This presentation outlines the use of predictive risk modeling to predict university course non-completion in the first year and non-retention in the second.
Applications of Predictive Analytics
Using Data to Predict Police Misconduct - Article
The White House Police Data Initiative was launched in 2015 in response to the need for better use of data and technology in community policing. As a part of this initiative, researchers from the University of Chicago partnered with the Charlotte-Mecklenburg Police Department to construct a predictive, early warning system to prevent negative interactions between the police and public.
The Police Department provided the researchers with data they had gathered, including information on arrests, traffic stop, dispatches and discipline. Using this data, the researchers identified predictive factors of misconduct and developed an algorithm that can foresee adverse interactions between officers and civilians, and preventative measures. This system enables the Department to target scarce resources more effectively by identifying officers that have a significant risk of adverse reactions and would most benefit from additional training and support.
This is an interesting issue for the Centre for Social Data Analytics, who have been working for a long time in pushing that ethical issues need to be addressed along with the technical details of prediction. Tim Dare (an affiliate for CSDA) has written one of the most comprehensive ethical reviews of the use of PRM in social setting. See the review here.
Using Predictive Analytics to Assess Risk for Child Maltreatment - CSDA Co-Director Tim Maloney
Podcast by University of Wisconsin-Madison's Institute for Research on Poverty