Tuesday, June 09, 2009

Rot about I.Q. and voting
Not just wrong, but stupid 

It's all very well our enemies being wrong and evil, but wouldn't it be great if they were stupid, too?

There's been a 4 way discussion going on over twitter between John Band, Mike Power , Joseph Edwards and Alex Gray regarding voter IQ. Mr. Gray informs us:

@mrpower A study done last year found that BNP voters had an average IQ of 98.4, the lowest. Greens, highest, had 108.3. - [Twitter here]

@mezza1959 Labour voters have an average IQ of 103. Fourth highest (Conservatives are third at 103.7, Lb Dems second at 108.2) - [Twitter here]

Those of us who prefer to have citations for our half-remembered pub-talk factoids will ask Mr. Google and find the actual paper: it is Deary, Batty, and Gale, "Childhood intelligence predicts voter turnout, voting preferences and political involvement in adulthood: The 1970 British Cohort Study", Intelligence 36, 548-555 (2008). If you have institutional access to Intelligence you can get it here, otherwise you can get it off the University of Edinburgh website here
We will also note the correction (institutional access here, although the department is insufficiently proud of this to put it on its website. I wonder why.)

The first thing to note is that this is a study of child IQ (measured at age 10) of members of the 1970 cohort study. They were asked about their voting preferences at age of 34, and how they voted in the 2001 election. The results the paper headlines in the abstract are the intelligence-green party and intelligence-liberal democrats correlation; however, they note that the IQ-Green can be accounted for by occupational social class.

The figures which Mr. Gray quotes are drawn from Appendix Table 1 (p.554) although, alas, the means have been lost their associated standard deviations. (In fairness this press release is probably to blame). Let's reunite them (format mu(SD)):

Voted (2001) 104.0 (14.2)
Didn't vote 99.7 (14.1)

Supported in 2001 election
Con 103.7 (13.5)
Lab 103.0 (14.2)
Lib Dem 108.2 (14.4)
Scot Nat 102.2 (14.2)
Green 108.3 (12.9)
Brit Nat 101.1 (15.7)
UK Ind 99.7 (13.4)
Plaid C 102.5 (16.5)

Intended to vote 2004 election

Con 103.1 (13.9)
Lab 101.6 (14.6)
Lib Dem 106.9 (14.5)
Scot Nat 100.2 (12.8)
Green 107.1 (13.7)
Brit Nat 99.6 (13.5)
UK Ind 97.4 (12.2)
Plaid C 98.7 (17.0)
None 98.1 (13.4)

I invite readers to consider the means and standard deviations of these sub-groups, and make up their own minds as to the significance or otherwise of these result. Comments are particularly solicited from LemmusLemmus and PJ.

Those of us with longer memories will remember the time The Economist recycled a fake IQ/Red State Blue State correlation from an internet news group.

[For the avoidance of doubt: the BNP are fascist scum, certainly evil and probably stupid. However, I do not think this data set can bear the interpretation that Mr. Gray rests upon it. I am also less afraid of stupid fascists than clever fascists.]

6 comments:

John B said...

As the paper says:
1) a few of the effects are significant, notably LD/Green
2) these can probably be ascribed almost entirely to social class
3) ...as can the BNP effect, to the extent that it's real, which is probable but not certain.

Also, Gray is a Very Bad Man for not providing the link whenever he first Tweeted it...

LemmusLemmus said...

Comment solicited - comment received!

1. To put things into perspective, you should have mentioned that the intelligence measure was constructed so that the standard deviation for the whole sample was 15.

2. To be interesting in this context, a difference has to be both statistically significant and substantial. A difference of about half a standard deviation (e.g., between Lib Dem and BNP voters) is certainly notable, a difference of about two points (between Labour and BNP) is not in my book (even if it were statistically significant, which it is not).

3. What you'd really want to do to get an idea of what the results mean is to graph them, which would show lots and lots of overlap between the groups. (In this context I'm almost sure it's fair to assume the independent variable is normally distributed.) Language Log had a series of posts about how this can make interesting-sounding results look utterly unimpressive. Anyone interested may dig them up themselves.

4. I don't think it's fair to put this in the same league as the red state-blue state hoax. Those results would have been much less interesting even if true.

5. If you're looking for commentary by someone who knows his statistics much better than I do, you might want to consider writing Andrew Gelman, a statistician/political scientist at Columbia University, an e-mail. Seriously. He has published on voting (albeit in the US) and he has answered reader questions on his blog in the past. He also appears to be a genuinely nice bloke.

6. What I really liked about your post was this bit:

"If you have institutional access to Intelligence you can get it here, otherwise you can get it off the University of Edinburgh website here."

Institutional access to intelligence! This conjured up vague visions of a Being John Malkovich-type film.

8. So far I've only glanced through the paper. Maybe I'll read it and have some more to say later on. If not that may be because I'm busy writing a screenplay.

Brunellus said...

"It's all very well our enemies being wrong and evil, but wouldn't it be great if they were stupid, too?"

Sadly, the thought that reprobates are guilty of intellectual error is a common background assumption in current moral philosophy.

Political Scientist said...

John,

I broadly agree (altho’ as the authors note that LD/IQ correlation doesn’t disappear with SC)

It is Naughty of Mr. Gray not to have provided a citation, although it may we have been either the press release or an article based on it. What was Very Naughty was to cherry pick the Con, Lab and LibDem results from the 2001 reported vote (suppressing the tory-lab difference and using the higher value for the LD IQ) and the BNP from the 2004 voting intentions, without mentioning they are different things.
[There also what we might call the “clever nazis lie” effect, which might reduce the BNP IQ thru’ under-reporting.]
In fact, I would have thought this difference rather undermined the study - one’s childhood IQ in 1980 is fixed, but voter preferences clearly change.

LemmusLemmus: Thanks, I was particularly interested to read your thoughts. I agree the study is much more interesting that The Economist hoax, in that were dealing with actual people with reported data (voting preference and IQ), rather than “lots of people who live in the state have low IQs” and “a lot of people who live in this state voted Republican” and all the RedState BlueState trouble that Andrew Gelman deals with so well.
Also, of course, the data in Deary et al are real data, collected by credible researchers in a well-known cohort study, analysed by a team of credible researchers in a well-known group published in Journal with an impact factor of 2.896.

In addition, of course, the Economist data was cut’n’pasted from a newsgroup by those rigorous journalists at the Economist, without checking if it was actually true (or, indeed, thinking about what the interstate gaussians looked like, and whether that was remotely plausible)

I seriously admire Gelman - as he’s a physicist who became a real political scientist, I think he’s great. I think graphing the results would be interesting. Also, I noted that the sample size for the sub-group of voted greens in 2001 is small (78) and BNP smaller (27), although the standard deviations aren’t awful (12.8 vs. 15.3). I wish I could get to the Correction online, tho’...

Re; Intelligence, when I read what you wrote, it made me think of “Team america: world police, where the computer was called I.NTELIGENCE. “That was very bad, INTELLIGENCE. Very, very, bad, INTELIGENCE”.

Brunelus wrote: “Sadly, the thought that reprobates are guilty of intellectual error is a common background assumption in current moral philosophy.”
Now, if I were a philosopher, I’d be an Immoral Philosopher. Don’t forget to email me with The Plan.

LemmusLemmus said...

As for sample size, the handy thing about significance tests is that they already take that into account, so you can't attack a result's significance by pointing out the small sample size.

Political Scientist said...

"As for sample size, the handy thing about significance tests is that they already take that into account, so you can't attack a result's significance by pointing out the small sample size."

Fair point.