Predictive Risk Modelling; How does it work? Can it work?

by Hitendra Patel

In Aotearoa New Zealand, a politically motivated ‘social investment‘ strategy determines social policy. Adopting a residual welfare approach social investment is dependent on the use of data and evidence to inform and target early interventions to mitigate against future liabilities. Within this regime, predictive risk modelling (PRM) methodologies have been signalled for use in child protection (Ministry of Social Development [MSD], 2015; Oak, 2016). Researchers utilising anonymised MSD beneficiary data linked with data from the now defunct Child, Youth and Family Services (CFYS) created an algorithm to detect the risk of child maltreatment. The algorithm calculates over 200 variables in generating a risk score to assess the probability of child maltreatment (Vaithianathan et al., 2012). The variables include socio-economic status, educational qualifications, interactions with criminal justice system, whether a parent has suffered abuse, and prior notifications of child neglect. The algorithm accurately predicted 75% of substantiated child maltreatment findings that occurred in the population sample of 57,986 babies born between 2003-2006 (Vaithianathan et al., 2012).

While there is enthusiasm for PRM, concerns about its methodology, implementation, and ethical use have been raised. Gillingham suggests that flaws in PRM methodology will see greater numbers of children deemed at risk than are in danger (2015). Among the predictor variables in use for PRM are substantiated cases of maltreatment. Over eighty percent of child maltreatment findings are not always clear-cut recognisable cases of sexual or physical abuse, and findings of neglect can be a contentious arena(Gillingham, 2015). Substantiations can vary because of ideological shifts, cultural factors, resource availability and practitioner discretion but their use as a ‘value neutral’ variable within PRM’s can indicate higher probabilities of child maltreatment when in fact the risk may be low.

The social investment agenda dictates ‘at risk’ children are targeted for intervention at the earliest possible juncture. Under this jurisdiction, once a risk score indicating child maltreatment is calculated, families can be targeted for intervention even in the absence of evidence that maltreatment has taken place (Oak, 2105). The use of PRM scores to motivate interventions has consequences for families found to be  ‘at risk’, that include intrusion, upheaval, and the stigma of maltreatment risk scores attached to their MSD client files. The use of welfare recipient data reinforces existing social prejudices and increases the potential for beneficiaries to be subject to greater levels of surveillance and punitive actions for failure to comply with MSD’s PRM requirements. Within this population, Māori are disproportionately represented and face a very real risk of state sanctioned discrimination through targeted interventions.

Along with privacy concerns, there are implications in obtaining informed consent from MSD clients (Keddell, 2015). Families who are in desperate need of MSD services may feel obliged to give consent to receive services, or conversely, families in hardship may be reluctant in seeking much-needed support for fear of possible intervention from child protection services.

The use of PRM within an ‘evidence-driven’ social investment agenda requires further critique. The creators of PRM acknowledge the weight of research evidence linking poverty to child maltreatment (Vaianathan et al., 2012) however, comparable evidence is absent in the government’s social investment strategy.

References

Gillingham, P. (2015). Predictive risk modelling to prevent child maltreatment and other adverse outcomes for service users: Inside the ‘Black Box’of machine learning. British Journal of Social Work, 46 (4): 1044-1058bcv031.

Keddell, E. (2015). The ethics of predictive risk modelling in the Aotearoa/New Zealand child welfare context: Child abuse prevention or neo-liberal tool?Critical Social Policy35(1), 69-88.

Ministry of Social Development (2015). Expert Panel Final Report: Investing in New Zealand’s Children and their Families. Wellington, NZ: New Zealand Government.

Oak, E. (2016). A minority report for social work? The Predictive Risk Model (PRM) and the Tuituia Assessment Framework in addressing the needs of New Zealand’s vulnerable children. British Journal of Social Work46(5), 1208-1223.

Vaithianathan, R. (2012). Can administrative data be used to identify children at risk of adverse outcomes? Auckland, New Zealand: Business School, Department of Economics, University of Auckland.

 

Advertisements

About socialworknz

I'm a social work researcher in Aotearoa New Zealand
This entry was posted in Uncategorized and tagged , , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s