Big-data expert Rhema Vaithianathan hit the headlines this month with her study revealing almost one in four Kiwi kids are reported to the Ministry of Children (Oranga Tamariki). The Auckland University of Technology (AUT) professor has created algorithms which predict which children are most at risk to help call centre staff prioritise calls.
1 You've revealed our hidden statistics on child abuse and neglect by integrating three sets of government data. Were you surprised by the results?
We were surprised to find that almost one in four New Zealand children are notified to child welfare by age 17 but what really shocked us was that maltreatment was substantiated for one in 10 Kiwi kids. That number is incredibly high and the true figure will be even bigger. That's why the Children's Commissioner is very keen that we have a national conversation about it. With this kind of linked data work we have the opportunity to have an informed debate.
2 Can big data help with solutions for child abuse as well?
We've had success in Allegheny County, Pittsburgh, in the United States, developing an algorithm to help call-centre staff better "triage" allegations. When the county reviewed cases of serious harm, they often found calls had been received but were screened out because the individual allegation wasn't serious enough. On the other hand you had families of colour or poverty being investigated because of calls made when their children weren't actually at risk.
3 I've seen this study reported in the New York Times. How does the algorithm work?
The algorithm measures risk by going through all the existing data to look at the pattern of contact with social services and give each piece of information appropriate weight. For example, how many calls have been received? What type of calls? Were they investigated? Has the person previously come into contact with the system either as a perpetrator or victim? Staff members already have access to this information but they often don't have time to sift through it all. The algorithm has been such a success in Allegheny County we've been asked to implement it in Douglas County, Colorado.
4 How did you measure success?
The study's still going so we don't have the final numbers yet but a key reason it's considered a success is because we actually managed to implement it. Most don't get that far for reasons including community concerns about the use of big data. Those have to be addressed and it's a lot of work. We met with more than 20 community groups, judges and the American Civil Liberties Union. The biggest fear is that people will be stigmatised by the number. We had an independent ethics report done and took on its suggestion that the score does not go beyond the call screener. The social worker has to rely on their own clinical judgement.
5 What was the best predictor of a child needing to be placed in care?
A lot of history with child welfare. Many children who end up being placed in care have parents and family members who have been in care too. We're seeing that child welfare has very strong inter-generational transmission. If we want to break this cycle we need to think about putting in a lot more preventative resources because the current system is not working.
6 Has there been any interest in trialling this in New Zealand?
No, we present our findings at an annual workshop in Wellington but there's been no interest so far. Data is much more integrated in America so it's easier to look across all the social services a person has come into contact with there.
7 Do New Zealanders mistrust the use of big data by government agencies?
I was part of a working group set up by the previous Government to ask New Zealanders what their concerns are with data use. We held about 47 workshops up and down the country and found a huge variety of comfort levels. People told us their concerns would be alleviated if the agencies taking their data answered a simple set of questions; How will their data be kept secure? Will it be depersonalised? What is it going to be used for? Who will benefit? We've created a set of guidelines for trusted data use for New Zealand agencies. Our advice is just be honest and transparent and you'll get a long way.
8 MSD's controversial data sharing plan that would have forced NGOs like Women's Refuge to hand over their clients' personal details was quickly scrapped by the new Minister for Social Development, Carmel Sepuloni. Do you think the concerns over privacy and security are valid?
You have to respect people's concerns and address them articulately. What the ministry failed to do was go through and answer all those key questions like; "How will the data be kept secure? Why does it have to be personalised?" You need to be able to assure people that enough checks and balances will be in place.
9 What was your childhood in Mangere like?
Great. I was born in Colombo, Sri Lanka. We moved here when I was 7 because my parents were worried about corruption creeping in. Dad worked at Mangere Hospital with people with developmental delay. Mum studied sociology part-time from home. We already spoke English because Sri Lanka was a British colony. I was really good at maths but I'm very people-oriented so economics was my favourite subject because it brought together society and maths. I felt it could make a difference to people's lives.
10 So being the co-director of AUT's Centre for Social Data Analytics must be your dream job?
It is! We created the centre two years ago because there's a lot of hype around big data with companies like Facebook using it for marketing and profit but there hasn't been a lot of work on how to use big data to improve social justice and equity. It's a challenging area. You can't just helicopter drop the methods used by business into the social services space. There are only a few research centres like ours worldwide which is why we do a lot of international collaborations.
11 You're also doing a study on which children beat the odds and why. How do you go about that?
The Growing Up in New Zealand Study has been following about 6000 Kiwi children and their families since 2009. We've counted the number of "adverse childhood experiences" (ACEs) experienced by each GUiNZ child that are associated with health and social problems as an adult. The next step is to predict the number of ACEs a child is likely to experience and look for protective factors related to children who do not experience the number of ACEs we would expect. The GUiNZ study is particularly strong on resilience factors like parents' levels of optimism for their child and whether they read to their child; the types of things you don't usually find in databases.
12 What other studies do you have on the go?
We've been working to map neighbourhoods in California where the children are most at risk of maltreatment so agencies can put resources in the right places. We've found that the children most at risk of maltreatment are not the same as the children living in poverty. The majority of poor families are not a maltreatment risk. We know there are geographical hotspots for maltreatment but we have yet to answer the question why.