Fighting Child Abuse with Predictive Analytics

Dr. Dyann Daley

by Stephen D. Haner

As a pediatric anesthesiologist in Texas, Dr. Dyann Daley saw far too many victims of child abuse in the OR. One horrible case in particular spurred her to move beyond treatment to thoughts of prevention, with a data-driven approach that should fascinate all Baconistas.

Preventing child abuse is not as simple as preventing the flu or even malnourishment. There are plenty of thoughts on the why of child abuse, including the observation that it can be patterned behavior. The insight Dr. Daley and others behind Predict-Align-Prevent brought to the discussion was to focus on the where. Looking specifically at her home of Fort Worth, the group used spatial risk modeling to track a number of adverse outcomes believed to correlate with mistreatment of kids.

The results are shown in two maps – one predictive and the other showing how the next year’s case statistics matched the predictions. That the two maps lined up so well may not be surprising. But Dr. Daley was surprised that poverty was not the strongest correlation with child abuse. Domestic violence, aggravated and sexual assault and children running away from home – those are the other problems with the strongest correlations.

Dr. Daley was at Lewis Ginter Botanical Garden on Tuesday presenting her work to a child abuse and prevention conference, jointly sponsored by Families Forward Virginia and the state Department of Social Services. Predictive analytics is only worth doing if it then drives decisions on what preventive measures to take, where to focus the efforts and how to measure what is working.

What has been done in Fort Worth is now being done in the City of Richmond, with a similar mapping effort underway down to the half-block level. That goal will be the same, to use the map to encourage the relevant players (social services, police, medical providers and the schools) to cooperate on effective prevention. The data will also locate community assets – schools, parks, services, churches, licensed day care – to see if they have any positive impact and how.

The work is not universally applauded, and people in the audience Tuesday expressed concerns that sections of the city will see additional stigma or stereotyping based on the maps. The researchers reported that getting some of the data they want is not quickly shared, and it is being done inside DSS because at least the child welfare data is already in-house. They also stressed that the data does not deal with causation, only correlation.

Disclaimer: I’m on the board of directors for Families Forward Virginia. It was formed last year through a merger of Prevent Child Abuse Virginia (PCAV) and the Comprehensive Health Investment Project (CHIP) of Virginia, and serves as the state office for three evidence-based home visitation programs all across the Commonwealth. CHIP was a paying client from 2013 through 2017, but now I’m working for free. I guess Jim will chalk me up as another social justice warrior.

I’ll keep my eyes open for more information on the Virginia effort with predictive analytics and pass it on.

There are currently no comments highlighted.

6 responses to “Fighting Child Abuse with Predictive Analytics

  1. Steve, working to prevent child abuse doesn’t make you a Social Justice Warrior. It makes you human. If you want to be a SJW, you’ll have to try a lot harder than that!

  2. So Steve…

    What were the highest correlating factors? I was wading through trying to find them. Very interesting stuff. Danny Kahneman would be proud of this lady. .. and you for passing it along.

  3. I didn’t have a copy of her slides, which were busy and went by fast, and a promise to post them has not been met. Should have held off posting….

    The researchers threw many things into the mix – health outcomes like asthma, criminal reports (police have been using this for a while), those community assets mentioned. It was not too surprising that other forms of violence – domestic, sexual – and signs of family stress like children running away are strongly co-located with child abuse issues. Just a general atmosphere of violence seems to be the overwhelming theme, with many families in these areas simply afraid to go out, adding to the stress.

    As a side issue she also talked about using some of these techniques to find the families most likely to sign up to be foster parents, always is high demand. Turns out they watch alot of HGTV….

  4. I’m a little bit troubled by the basic concept. I understand and agree with the idea of collecting data and being observant of correlations…

    but when you take this data and use it to derive policies of intervention… it’s at a level where you better be sure of what you are doing… and I’m not convinced at this point.

    Folks should remember – WHO is it that is going to intervene and try to remove a child based on something that has not yet happened but the data says it might?

    You’re talking about the govt being the one who does this.

    I’m surprised my friend Crazy seems to like this idea…

    • This is not the Tom Cruise movie – nobody is going to removing children because of some predictive algorithm! The idea is to focus intervention and services where more effective. Resources are limited. The traditional model has been reactive, seeking to prevent child abuse from happening again after the first problem – when the goal is to keep it from happening at all.

      • Hopefully so.. but these days , anytime someone says “correlation not causation” but then turns around and wants policies….

        It’s like designating an area as a “high crime” area then turning around and instituting a broken windows policy that is primarily focused on that area.

        We identify kids individually as “at risk” but we do not categorize , as far as I know , entire neighborhoods at high in “at risk” kids….that then turns into some kind of policy with respect to the entire neighborhood.

        Banks used to be accused of doing this for loans… remember?

        I’m going to back up a notch here and agree that having the data is potentially a good thing but having such data has sometimes proven to be the impetus for not good things…also.. but without the data.. you lack information that could be useful – if used right.

Leave a Reply