Everyone always forgets the false positives

This is the sort of thing that Ben Goldacre and Alex Harrowell continually go on about. It\’s all very well having a medical or security check to try and find something, in this case the eeeevil pervs who would monster the kiddies, and there can indeed be some success at finding them. However:

Up to 13,000 people have been wrongly labelled as criminals or accused of more serious offences because of blunders by the police and the Criminal Records Bureau.

There will also always be false positives. Always.

Which leads to something of a problem, for this existence of false positives is something that those designing and proposing such systems are extraordinarily shy of talking about or acknowledging. And the problem is that without revealing, accurately, what this rate is, we can\’t work out whether the system itself is worth having.

Now I know nothing at all about (please do make the "no surprise there" jokes) how many people actually have CRB checks nor how many who do have them are successfully found out and prevented from being eeeevil pervs who would monster our children.

But imagine some numbers (and they are very imaginary). Say there are 100 paedophiles prevented from working with children by this system. Yes, we do think that 1 prevented molestation is more important than 1 person who wouldn\’t do so being prevented from working with children. But more important than 10? 50? 130?

At some point we do have to say that a high rate of false positives is causing more damage than the system is preventing. In logic, at the limit, we might have the entire population falsely identified and thus there is no one to care for the kiddies at all, clearly a worse outcome than whatever small number would be molested in the absence of the system (because, of course, not being cared for by anyone the kiddies would all be dead).

As I say, I don\’t know the real numbers here but the logic still stands. Unless we know the true rate of false positives and unless we can clarify the value we place upon what is being prevented against the costs of those false positives, we cannot know whether the system itself is value adding or value destroying.

I suspect the latter but it is that, a suspicion (and that\’s without taking into account the cost of the system itself. Are we sure that we\’re preventing crimes to hte value of the squiddely millions that people have to pay to get checked?).

4 comments on “Everyone always forgets the false positives

  1. An excellent post by Tim on a very important matter. And for very many government activities, not just CRB checks.

    I have an additional point of scepticism, though I am not certain of its correctness.

    The tail end of the Telegraph article quotes: “Of around 3.3 million checks conducted by the CRB last year, over 99.91 per cent were issued correctly. While any disputes are clearly regrettable, the percentage of disputes upheld has fallen for the last three years.”

    This gives 2,970 wrong CRB results. However, from the experience of a family member, one has to obtain a CRB check on application for many jobs (paid or voluntary), though one will of course only accept one job. Thus the number of persons applying is not the same as the number of reports issued.

    My sceptical query is: does 2,970 represent the number of wronged persons (in one or more reports in the year), or is it the number of wrong CRB reports? Likewise, is the 3.3 million, the number of separate persons who apply for CRB reports in the year, or is it (as quoted) the number of reports issued.

    If one of these measures (2,970 and 3.3 million) is a count of persons and the other is a count of reports, then the given percentage (99.91%) is not an accurate measure of correctness.

    Does anyone know?

    Best regards

  2. It’s worth stressing that these figures suggest that when the CRB say someone is a criminal 12% of the time they are wrong.

    1 in 8.

  3. Still angry about this.

    “Of around 3.3 million checks conducted by the CRB last year, over 99.91 per cent were issued correctly.”

    I hate that kind of weasel defence. It’s the same argument that used to be trotted out on Watchdog by some hapless British Gas spokesman when arguing that most of his customers hadn’t been blown up.

    Hey, nameless spokesman! If you fired all but one of your staff and that person set up an email rule that replied with “Yes, he’s fine” to any email that came in, you would still be able to claim 99.39% success. So let’s not be too satisfied that you only blight the lives of 3,000 people a year, OK?

  4. This is the ploy of the Righteous (copyright Leg Iron and Old Holborn). Can it be right to let even one little kiddie suffer? And of course the right answer is yes. Nothing “bad”can be completely, irrevocably eradicated with absolute certainty, so punishing 10,000 people for the potential sins of 1 person is no way to live.

    Compare and contrast with the argument used against capital punishment (which I do not advocate) that it is better to let 100 criminals go free rather than hang an innocent man.

    This attitude also can have physically harmful effects – the law of unintended consequences applies particularly to the Righteous’ knee-jerk policies. A friend of mine used to walk his dog in the local woods. His doctor said it was good for him (and the dog!). Sadly, the dog died and my friend found that he could not walk through the woods alone, especially during school holidays, such was the social stigma attached to this innocous activity. Who wants to live a life like that? Oh, and the Righteous suggest he goes to the gym. A much better choice in the eyes of the Righteous, than to enjoy the nature that the Righteous want to preserve.

Leave a Reply

Name and email are required. Your email address will not be published.