Better to protect too many children than too few?

When it comes to predicting child abuse and neglect, false positives are generally considered to do less damage than false negatives. In other words, it’s felt to be better incorrectly to flag a child as being at risk of abuse than it is mistakenly to flag him as safe in his parents’ care. When practitioners make the second mistake, more children are maltreated.

In their efforts to prevent abuse and neglect, State governments in the US have relied on research that points to the characteristics of children and families likely to be involved in maltreatment. The idea is that where abuse or neglect has possibly occurred a caseworker rates families on various characteristics (such as whether the parents drink too much, how large the family is, and how well they are coping financially). The more risky traits the family has, the more concerned the caseworker should be.

The approach might seem logical – but there’s a problem: it doesn’t work very well. According to researchers at in Department of Psychology at Wesleyan University [1], caseworkers who use this method struggle to identify children truly at high risk of abuse.

Eve M. Sledjeski and her research team argue that a more accurate solution might lie in looking at the characteristics of a group of families referred to government services for child maltreatment and then identifying subgroups of parents most at risk for continuing abuse or neglect. In this way, researchers might discover constellations of traits that make abuse or neglect more likely.

They tried the approach by examining information about 244 families in Connecticut collected over the course of a year. One example of the subgroups they identified included parents who had previously been accused of maltreatment and whose children were rarely seen by people outside the family. Another included families referred to a government agency for maltreatment for the first time who were not otherwise actively involved in the agency’s services but who had a history of domestic violence.

This new approach to identifying “at-risk” families did a much better job of predicting which parents actually did abuse or neglect their children again. By identifying subgroups, the researchers correctly identified 95% of the total number of cases of repeat maltreatment. The previous method correctly identified only 37% of these cases.

The downside of the subgroup approach was that it also incorrectly categorized 61% of cases in which no maltreatment occurred. The false negatives of the old approach were replaced by false positives.

In the UK, talk of false positives in the context of child protection work brings back memories of the 1987 Cleveland Child Abuse Scandal when children identified as being at risk by a particular diagnostic test were summarily taken from the parents and in some cases held for weeks on end in two children's hospital wards. The repercussions led to a Government Inquiry that precipitated far-reaching changes in child welfare policy and triggered a movement toward need-oriented, evidence based practice still in play in the UK system.

The families in the Wesleyan study were already in the system and the false positives were in relation to a recurrence of abuse as opposed to a first suspicion, but the UK experience is cautionary, all the same. For a brisk, under-punctuated journalistic account of the 1987 furore, see The Cleveland scandal - 20 years on [2]

• Summary of “The Use of Risk Assessment to Predict Recurrent Maltreatment: A Classification and Regression Tree Analysis (CART)” by Eve M. Sledjeski, Lisa C. Dierker, Rebecca Brigham and Eileen Breslin in Prevention Science, Volume 9, Number 1, March, 2008, pp 28-37.