Monday, May 17, 2010

Turnout, unemployment, safe seats and marginals

Two things that everyone knows: poorer areas have lower voting turnouts than richer areas, and safer seats have lower turnouts than marginals.

I’ve looked at these effects in this general election: they both exist, but one is much bigger than the other. I used data on 2010 election turnout, 2005 election closeness and the April 2010 Jobseeker’s Allowance claimant count for each constituency in Britain.

The JSA numbers are hardly a perfect guide to unemployment, but they’re all I can find available on a constituency basis, and the result is striking:

The dots are mostly grouped pretty closely around a very clear line; the correlation between unemployment and turnout is -0.76. So some of those most affected by government policy are least likely to vote.

As for the safeness or marginality of a seat, here the correlation with turnout is still significant, but a much more modest -0.45:

There’s still a clear tendency, but the dots are mostly a lot farther from the line of best fit than they are in the unemployment chart. (The correlation between turnout and 2010 majority is a even feebler -0.13.)

So to all those electoral reformers who say that abolishing ultra-safe seats would boost turnout: on these figures, abolishing pockets of high unemployment would boost it even more. And it might even be a good thing in itself.


chris said...

Great work. But we should ask: what (if any) is the causal link between unemployment and the tendency to vote?
It could be that unemployment causes low turnout, if say the unemployed believe that no party cares about them.
But it could be that both are due to similar character traits. People who feel a sense of hopelessness are likely to be unmotivated to either vote or to look hard for work. Or people who feel no obligations to others are likely to not vote and to be sacked.
And so on.

chris said...

No sooner do I ask than I come across this:
"Personal joblessness experience translates into negative opinions about the effectiveness of democracy."

Tom Freeman said...

Very true, correlation doesn't imply causation etc. Although obviously there is some hefty causation going on somewhere in this territory. It's probably a lot of things.

For instance, I'd expect that poor education lowers the likelihood both of voting and of being in work.

Also, note that the fall in turnout is sharper than the rise in unemployment: the line shows roughly a 5% drop in turnout for every 1% rise in claimants. So I think this is tapping into a more general set of deprivation/social exclusion issues.

And I'd guess the phenomenon operates at least as much on a community as an individual level: someone on a sink estate who moves from long-term unemployment to being a toilet attendant is still less likely to vote than a banker who gets made redundant.

Glenn said...

you could look at other indicators such as ward deprivation indices (index of multiple deprivation) to see if findings are replicated...

Tom Freeman said...

And what could be very telling is looking at correlations between changes in turnout since 2005 and changes in unemployment (or other stats). Alas, boundary changes make that impossible.

Ole Phat Stu said...

Significant at what level?

Tom Freeman said...

Er, dunno exactly. But there are 631 data pairs, so I'd expect it do be pretty damn significant. This calculator reckons it passes the p<.0001 level efforltessly.