evidoria

← Back to browse

Good practice Imported

Predictive Risk Modelling — New Zealand's Shelved Algorithm for Flagging Children at Risk of Maltreatment

New Zealand · Wellington · See the New Zealand profile · See the Wellington profile

Evidence: Descriptive / self-reported Top 83% 40/100 · Ask Evidence Copilot about this practice

New Zealand's MSD tested a statistical model scoring newborns' risk of substantiated maltreatment by age five (AUC 0.76). A planned 60,000-child live trial was cancelled by the responsible minister in 2015 over ethics and racial-bias concerns.

0.76 AUC
Model predictive accuracy (AUC)
47.8 %
Top-decile children with substantiated maltreatment finding by age five
57986 children
Children included in the study dataset
132 variables
Variables used in the predictive model
60000 children
Planned live-trial cohort of newborns (cancelled) (2015 (cancelled))
Predictive Risk Modelling — New Zealand's Shelved Algorithm for Flagging Children at Risk of Maltreatment

Details

Maturity
Discontinued
Promoter
Ministry of Social Development (MSD), with University of Auckland / AUT researchers
Period
2012–2015
Keywords
social welfare, child protection, government administration

Context

Following a 2012 government white paper on vulnerable children, New Zealand's Ministry of Social Development (MSD) commissioned University of Auckland and AUT researchers to build a predictive risk model (PRM) for identifying children at risk of maltreatment, using linked welfare and child-protection administrative records.

Objectives

The project aimed to test whether administrative data could statistically flag newborns at high risk of substantiated maltreatment by age five, to inform earlier intervention.

Activities

Researchers built the model using 132 variables from records for 57,986 children and published the results in the American Journal of Preventive Medicine in 2013. MSD then commissioned a formal ethical review and proposed a prospective study scoring all 2015 newborns for risk.

Results

The published model achieved an AUC of 0.76, and 47.8% of children in the top decile of predicted risk had a substantiated maltreatment finding by age five.

Conclusions

In 2015, Social Development Minister Anne Tolley rejected the planned 60,000-child live trial on ethical grounds, and MSD's own research page confirms it never took place; a later independent academic analysis also found the model risked disproportionately flagging Māori children, a fairness concern the original ethical review had not resolved.

Implementation
Indicative cost
Low (< €50k)
Time to results
Medium (1–3 years)
Staffing & skills
Ministry of Social Development (MSD), University of Auckland researchers, AUT researchers

Conditions for success

  • Linked welfare and child-protection administrative records across agencies
  • Formal ethical review commissioned before any live trial
  • Peer-reviewed publication of the underlying model

Common failure modes

  • Ministerial rejection of the live trial on ethical grounds ('these are children not lab rats')
  • Independent academic analysis found the model risked disproportionately flagging Māori children
  • Original MSD ethical review did not resolve the racial-bias concern

Where it fits

Governance type
national social welfare ministry
Scale
national
Income level
high

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful