Family network data may identify at-risk children

Details collected from family trees could help government agencies identify at-risk families in potential child protection cases.

But using big data comes with a warning – incorrectly, the insight could exacerbate bias and inequality.

Researchers from Te Pu¯ naha Matatini have been working with data collected for a tool which Oranga Tamariki could use to better predict the risk of child maltreatment when a case is first flagged.

Using a network of about 5 million relationships collected by social workers between 1996 and 2016, Shaun Hendy and a team of researchers tested the family networks – specifically, pulling information from birth records to determine the close family ties.

It was ‘‘common sense’’, Hendy said. If a child had a family connection to someone who was violent or abusive, then that child was more at risk.

Where a child had been abused, the tool took out the guesswork in identifying potential risks and sped up, or spotted, suitable interventions. It could also allow social workers to better allot their time with families.

‘‘It is much easier to identify the causality. It shifts the focus from the child to identifying the abuser, the person who is most likely to cause the harm.

‘‘That means you might actually design a different intervention. So, maybe rather than removing the child completely from that family environment, maybe you simply act to ensure ... there is some monitoring of that relationship, or heightened attention paid.’’

But there were risks that statistical models and machine learning could entrench or amplify prejudice during an evaluation. ‘‘One of the risks would be that a social worker just uses this to reinforce their own biases. You would want to monitor the use of the model to make sure it was being used appropriately.’’

The research comes at a time when governments increasingly turn to data to build on and implement policies. Hendy and his team recommended independent monitoring of algorithms used by government agencies to mitigate potential ethical issues. ‘‘Our model has increased transparency, so it is easier for people to see what is going on.’’

Children’s Commissioner Andrew Becroft said big data could be helpful in identifying needs and the focus should be on early support and individualised care. But data could never replace personalised social work and detailed assessments.

‘‘I would not like to see already vulnerable children, families and their communities being stigmatised and pre-judged on the basis of an algorithmic approach.

‘‘What families need is effective, individualised assistance and support provided by government and community groups so that when they are struggling, they feel supported to seek out the help they need,’’ Becroft said.

The report, called Using family network data in child protection services, also said: ‘‘Combining predictive models with social worker expertise . . . has the potential to improve decision-making but only with much thought and care can their use be ethically sound and socially beneficial.’’