Monday, 21 February 2011

Pick'n'Mix Science

Today the Mail ran a story titled "Life of crime begins at three for some children, scientists warn as disturbing traits identified". As per usual, the story is 10% fact, 40% publishing unrelated information as part of the same events, 40% hyperbole and 10% bullshit.

I have reproduced the whole story here, and as usual, my comments are in red:

"At the age of three, most children will want to grow up to be a train driver, astronaut or princess.

But according to scientists, some toddlers are already destined for a life of crime. The research does not suggest this at all. It suggests that some toddlers who have a lower 'fear' response may be less afraid of committing criminal acts than others in the future. There is no 'destiny' about it. People still have free will. The study may be found here (PDF).

Disturbing evidence has emerged that the psychological seeds of a criminal career can be seen before they even reach nursery school.

Abnormalities in the parts of the brain that handle emotions, guilt and fear are far more common in criminals than in law-abiding members of society, it shows. While it is implied by the study is that poor fear conditioning is a result of an underdeveloped amygdala, and that poor fear conditioning may also result in a propensity towards criminality, the study about three year olds does not specifically or explicitly investigate the link between the two.

It is unclear whether these abnormalities are genetic, the result of upbringing or both – but they can be measured at a surprisingly tender age.

The finding means youngsters could potentially be screened to see if they are at risk – and then ‘treated’ to prevent criminal behaviour. Uhm - remember this story? No, this research does not mean youngsters could be screened to see if they're at risk of criminality. Could they be screened for their 'fear response'? Yes. Could their brains be scanned to see how well-developed their amygdala is? Yes. Will this ever happen on a national scale? Don't be fucking stupid.

Professor Adrian Raine, a former Home Office criminologist, agreed predictive scans were many years off. Surprising, that.

But the father-of-two added: ‘If you told me my son had an 80 per cent chance of being a psychopath, but that he could be treated for it, I would have him treated. But it has to be a decision made by individuals, not by scientists.’

Professor Raine, who now works at the University of Pennsylvania, studied brain scans of prisoners. Not in the study they were originally talking about, you understand. He did here (PDF) though, in perhaps his most famous study - which dealt with Positron Emisson Tomography of 41 adults in California that pleaded not guilty to murder by reason of insanity. In 1997.

Psychologists have identified key personality traits in childhood which are linked to poor behaviour later in life.

Seven-year-olds with unemotional and ‘callous’ traits were much more likely to be involved in anti-social behaviour at the age of 12, a study by Dr Nathalie Fontaine, criminal justice expert at Indiana University, Bloomington, showed.

Other signs include not having at least one good friend, being unkind to other children and not being helpful if someone is hurt. Notice that this study - again, a completely different one (abstract here - doesn't seem to be available online without subscription) - talks about antisocial behaviour, not committing criminal acts.

The experts stress that not all youngsters with the traits turn into criminals – and not all criminals had the traits as children, but that they increase the risk of a life of crime.

He found that murderers who kill in the heat of the moment are more likely to have a poorly functioning prefrontal cortex – which deals with reasoning and helps suppress base instincts. Not in the study of three year olds though.

Psychopaths who lack remorse, guilt or empathy tend to have smaller amygdalas – a region that handles all three emotions, he told the American Association for the Advancement of Science.

Professor Raine also tested the fear response of three-year-olds by playing them a neutral sound followed by an unpleasant one, until the children learned the nasty sound always followed the neutral tone.

For most, the sound of the first tone was enough to raise their pulse rates and start a sweat. But a few showed no ‘anticipatory fear’ – a possible symptom of an abnormal amygdala, Professor Raine said. Please note the word 'possible' with relation to abnormal amygdalas.

The prospect of scans suggests a serial killer such as Yorkshire Ripper Peter Sutcliffe could be spotted and treated as a child –  but it also poses dilemmas. Peter Sutcliffe was found guilty of manslaughter on the grounds of diminished responsibility after four psychiatrists diagnosed him as a paranoid schizophrenic. They would have been hard pressed to predict that he would kill 13 women as a result of this because he didn't respond to a noise when he was three.

‘It raises the question to what extent should we develop new biological interventions to reduce crime,’ Professor Raine said."
UPDATE: My wonderful and learned friend Lucy had this to say about the study of three year olds, which explains the problems with it better than I ever could:

"When the results say they are significant, it means that there is a small possible chance that these findings are the result of pure chance, but probably isn't.

The results section says that the group-by-stimulus interaction was significant to p = 0.033, meaning that there is a 3.3% chance that the results are actually just the result of blind chance – a 3.3% chance that they were mistaken, and that there is no link between low electrodermal fear response and crime. That's not a bad percentage margin; the cut-off point is usually p=0.05, or 5%. So when they say it is significant to the p=0.033 level, it means that there is a 96.7% chance that there is a link – i.e. that one affects the other. Basically, significant means that the results weren't down to chance – the numbers found accurately represent some link between groups.

It doesn't mean that a group are x times more likely to commit a crime at all; you just can't get that result from these numbers. It just means there is a link between electrodermal fear conditioning and crime committed. The results are correlational (i.e. kids with a low response commit more crimes) but you cannot attach an "it's x times more likely" tag to it without running a linear regression, which they haven't. Even if they had I still wouldn't attach that tag because this is correlational evidence which CANNOT, under any circumstances, imply causation. The link may operate the other way around; it may be that low electrodermal response at age 3 is representative of another problem, not linked to fear conditioning, which causes an increased chance of criminal activity when older; or any number of alternatives.

I would also not trust these results AT ALL. The paper has some serious methodological flaws (e.g. absolutely no gender representation at all); the statistics are hard to fathom and presented weirdly, and I'm not convinced the statistical tests were appropriate because they haven't reported why they chose what they did. Plus, it looks overcomplex. The results section is ludicrously short, ill-reported, uses non-operationalised terms and has had no post-hoc tests reported to check which group interactions were significant. That in itself is very suspicious.

They're also drawing unjustified conclusions from the results section. Just because they've found a correlational link, they cannot say that x causes y. It's one of the things we learned in first year in methodological analysis. They're cutting out the influence of environment (despite controlling for home life). It may be that kids with low electrodermal response are more susceptible to peer pressure, or need more attention from peers, or any number of things.

I didn't like this paper. Maybe it was just down to the limited space they were had to report it, but in particular the results section was extremely poor.

They mention that cutting down on smoking, drinking etc whilst pregnant and eating healthily helps prevent the birth of criminally-prone kids (because, they suggest, it allows better brain formation – without ever telling us what that means), but they're stating the bloody obvious and missing the point. Mothers who do these things whilst pregnant are also more likely to be better mothers in terms of discouraging criminal activity when the kid is young; this could be environmental, not biological. They also pulled out that old chestnut about there being brain deficiencies in people with criminal tendencies, which once again trips straight over the cause-and-effect, environment-or-biology fallacy; that idea's been kicking around for ages. They try to recover the conclusion they've made with the very last line of the paper, but that seems to conflict with what they've been saying up til this point."


  1. Ah, I see.

    So to eliminate crime all we need to do is:
    1. Cobble together some mixed up science that you've skim-read.
    2. Brand large numbers of children as 'future criminals'.
    3. Implement Ludovico technique.
    4. ????
    5. Blame foreigners?

    I'm surprised they didn't manage to link Phrenology in to this story.

  2. Didn't King Herod try this method of benevolent social engineering? How did that go?

  3. I've sent you my analysis of one of the papers used in your post, as requested, including the statistical analysis and methodological flaws. That paper was so FAIL it hurts my brain. How did it ever get through peer-review?