UK DNA database consultation

By Ashok Argent-Katwala

The UK's Home Office has been running a consultation, entitled Keeping the right people on the DNA database.

I'm gravely sceptical about the entire episode and, throughout, the document tilts heavily towards keeping DNA for a long time because that will – supposedly – make us safer.

David Mery has had some choice words and a very thorough response to the Home Office's proposal. I am not so thorough, and kept my contribution to the section of which Ben Goldacre rightly asked 'Is this a joke?'.

The consultation closed yesterday, here is my contribution, written from the vantage point of my academic high horse.

---

 From: Ashok Argent-Katwala
 To: DNAconsultation@homeoffice.gsi.gov.uk
 Date: Fri, 7 Aug 2009 16:03:31 +0100 (BST)
 Subject: DNA consultation

Dear Mr Brown,

I am a computer scientist and engineer by training. This naturally involves a proper understanding of logic and statistics. I feel that the consultation document circulated falls short in both these regards.

I must begin by noting that the language that the consultation document uses is quite frequently careless. It uses the word 're-offending' to describe individuals arrested but never convicted, nor guilt admitted. In common parlance I believe that is 'innocent', so the 're-' is quite inappropriate.

The headline suggestion that six years is a proportionate response to someone being accused, arrested but, finally, innocent, could only possibly be supported by very strong evidence that the person is, in fact, likely to be guilty of a future crime. I note with some dismay that the section dealing with this, Annex C, seeks to prove something far weaker than that and even fails to show that satisfactorily.

I will focus here on Annex C as I find the lack of care and attention in there quite remarkable. In particular, the annex compares (p30) the 'subsequent criminality of those arrested with no further action' to that of 'those cautioned or proceeding to court where the result was a non-custodial sanction'. This is a tremendously skewed comparison of dubious logical merit. It goes on to 'exclude those with earlier arrests or convictions'. In short, the author picks a slice from the latter population that is as non-criminal as possible.

What follows is even less careful. The report is at least honest about this (again p30):

These figures include subsequent contacts where the outcome was also NFA so we need also to looking at the proportion of those with subsequent contacts at which guilt was admitted or established.

It does not, however, go on to do this. As such, it seems very likely to me that you will be over-counting individuals who just happen to be more likely to be arrested by the police for whatever reason, even if they are routinely and always set free, innocent. In particular on the matter of ethnic background, the text on page 31 is hopeful at best, but really just details how lacking this annex is in any substance. The 'larger scale exercise' it recommends is very probably a good idea, but it needs also to be better run, and more precise in its definitions to avoid the problems that are quite evident in the annex as it stands.

The presumption throughout Annex C that holding more samples for longer will mean more good, clean matches is itself doubtful. A serious analysis looking to 'maximise profile value in detecting future crime' (p37) would also look to minimise the number of false positives. As the number of held samples increases, and given that the distribution of DNA sequences in the population is quite obviously strongly clustered (by relation) then the possibilities for false matches will grow.

The conclusion of Annex C omits even the serious flaws recognised in the text itself. This section, in particular, is hardly supported by the material that precedes it:

Analysis of a sample of such cases from the Metropolitan Police force area suggests that the level of subsequent criminality over three years is on a par with the subsequent criminality of those given non-custodial sentences or cautioned.

All together, I believe the merit of that whole section is highly dubious, and with it the underlying merit of the argument that innocent people on the database are just as likely to be future criminals as those on the database as a whole. I am quite sure that in any half-way rigorous scientific review process Annex C would be thrown back very quickly indeed.

Yours sincerely,

Dr Ashok Argent-Katwala