I was a little concerned to hear today that:

Sex offenders will be made to take lie detector tests as part of probation conditions when they are freed from prison, the Ministry of Justice says.

BBC News: Lie tests tried on sex offenders

Firstly, before I explain, I will qualify. I am not against people on probation having checks to reduce the chance of them re-offending. This is even more relevant when it comes to sex offenders as some of these groups can be very likely to re-offend.

Anything which makes them less likely to re-offend and/or enable the probation service to keep a closer eye on them should be considered very seriously. Indeed, the BBC noted that in 2003, voluntary tests had prompted admissions in 80% of cases. The very idea that someone is to take a a polygraph test makes them less likely to lie.

However, it’s the creeping introduction of polygraph testing into our society, and particularly our justice system, which worries me.

I was recently reading a book called Would I Lie to You?: Deception Detection in Relationships at Work and in Life, which tells you about the ‘tells’ that people make when they are lying. Interestingly, I discovered that, for the most part, these were the sorts of things I was looking for anyway, which possibly means that I am good at telling when someone is lying, or possibly also that I am more likely to be a good liar myself.

However, there was a section about polygraphs and similar ‘tests’, where it discussed the accuracy of them, problems with them and so on. The suggestion was made that they can be accurate in no more than around 70% of cases — not the 97% rates claimed on the human bear-pit of places like the Jeremy bleeding Kyle show — and, perhaps more worryingly, that they will, more than 10% of the time, give the impression someone is lying when they are in fact telling the truth. It also suggests that wherever these are referred to as ‘lie detectors’, whoever has written that story does not understand them well enough.

Now in the case of the sex offenders, this is simply one of a measure of packages.

Pam Hibbert, assistant director of policy at children’s charity Barnardos, said the tests would “increase public confidence” that sex offenders were complying with supervision, staying away from schools and playgrounds and living and sleeping where they are supposed to.

“It is important however that this is used as part of a package of measures including greater use of satellite tracking.”

BBC News: Lie tests tried on sex offenders

So maybe in this particular instance, it’s not a bad thing. But I don’t like the idea of something being introduced into the criminal justice system, which 10% of the time will imply people are lying when they were actually telling the truth. I’d certainly not like to see it used for people who have not yet been found guilty of anything — if one in ten people interviewed in relation to a crime sound “guilty” to the polygraph, then there will be a tendency for the police to assume that they are guilty, instead of continuing to check for other suspects, particularly if the actual perpetrator is one of the 30% who falsely passes the polygraph test.

A 1997 survey of 421 psychologists estimated the test’s average accuracy at about 61%, a little better than chance

Wikipedia: Polygraph Reliability

We’ve seen this before — “Benefit cheats face lie detector”. I particularly liked the assumption of guilt for this headline. Presumably if they already knew that people were benefit cheats there would be no need to run the polygraph test in the first place, so “benefit claimants face lie detector” test is more accurate.

And the voice-stress analysis test isn’t ideal either (it’s not as reliable as the polygraph as it’s over the phone, and you have far less opportunity to get a ‘balance’ reading). In fact…

Lie detector technology to identify false benefits claims will no longer be used at job centre branches, after a Nottingham office pilot reported just 37% of those initially identified as lying actually were.

Personnel Today

I think this is referring to Tony McNulty MP’s statement in the House of Commons about Voice Risk Analysis trials by local authorities and JobCentre plus. These make quite alarming reading.

Firstly, the trial has cost at least £1.5 million to operate. Secondly, if you look at people identified as “high risk” by the JobCentre plus voice risk analysis (not quite the same as stating they were lying, but the implication is there), only 37% of them had their benefits changed as a result of the investigation. In other words, 63% of the people — almost two thirds — identified as ‘high risk’ by the VRA were giving correct information, and the cost of all of these extra investigations found nothing.

On the other hand, 29% of people identified as ‘low risk’ by the machine did have their benefits changed upon investigation. In other words, the VRA was next to useless. If it just picked 30% of the people at random and said that they were “high risk”, you would expect the results to differ little from this statistic.

In fairness, some of the Local Authority results were better than the JobCentre plus ones (Birmingham, Derwentside and Edinburgh showed a 30% differential between ‘low risk’ fails and ‘high risk’ fails, as opposed to the measly 8% in JobCentre plus). However, it is important to understand that even where these results were used, they were not being used to definitively state that benefits claimants were lying; merely that a higher proportion of these cases would be investigated in more detail, just to be sure…

However, as the Guardian points out, none of these trials give any actual numbers, just the percentages, so you can’t tell whether greater percentage differences could be anomalies with a small sample size, and nor do they use any control group (how many would have been identified as “high risk” or “low risk” by job centre operators on their own without VRA? would they have in fact done better?)

In other words, if something claims to be an accurate lie detector, particularly if they don’t give you any accurate statistics to measure it with, you might want to check that your bullshit detector is not going off.

So I am quite happy for polygraphs and VRA to be used to make the people taking them feel that they might well get caught in lying, and encourage them to be more likely to tell the truth, but considering that they are relatively ineffective, I wouldn’t want to see much credence given to the results of these tests in the benefits or criminal justice fields…

…although for the purposes of the Jeremy Kyle show, I really couldn’t care less.