Amateur psephology

Something that always annoys me about election coverage is the amazingly unsophisticated interpretation of the resulting statistics. For instance, in the current 2008 local elections, with something like a 35% turnout, the vote shares are projected nationally for the general election, which had a 61% turnout last time - and dire comparisons are made with seemingly little awareness of how meaningless this projection is. A great example is by Martin Kettle in the Guardian:

"Take, for instance, what has happened in Southampton. The Conservatives didn't just win the local council there this week. They also hoisted their share of the vote to levels that place both the city's Labour MPs - including the universities secretary John Denham - on clear notice of losing their seats. Three years ago, Denham had a 21-point majority over the Tories in Southampton Itchen; it made Itchen Labour's 189th most marginal seat. Yet this week the Tories stacked up a 20-point lead over Labour among the selfsame voters."

Ok, first of all we have to point out the boundary changes in Southampton Itchen, which makes comparisons difficult (this doesn't seem to make much difference to the 20-point Labour lead in the 2005 general election with a 55% turnout overall in the old constituency, which was 70% in 1997). A quick look suggests that in the new Southampton Itchen area the 2005 general election turnout was only 30% so Kettle doesn't have problems of differential turnout to content with,

Looking at the 2008 local election results for the new Itchen wards (Bargate, Bittern, Bittern Park, Harefield, Peartree, Sholing, Woolston) we get an 18-point Conservative lead with a 30% turnout. So pretty much what Kettle said. But local elections are, unsurprisingly, not the same as general elections, and we can demonstrate this by considering what the results were in the local elections just preceding that 20-point Labour lead in the 2005 general election.

In the 2004 local election in the new Itchen wards (obvious caveats about boundary redrawing problems), we find that the Tories actually had an 8-point lead over Labour, on a 31% turnout. So if we take a very naive approach to normalising general and local election results we could suggest that an 18-point Conservative lead in the 2008 local elections in Southampton Itchen predicts that Labour will be reduced to a 10-point lead over the Tories in the general election (not considering what will happen if turnout returns to the 70% high of 1997) i.e. it is all a little more complicated than the overpaid opinion makers of the media like to make out.

Ah, while writing this I notice that Kettle has another piece on Southampton Itchen, making most of the same mistakes:

"It also marks a big shift from the results in last year's local elections in Southampton, Tony Blair's last election as Labour leader. In 2007, these same Itchen wards produced a share as follows: Labour 34.9%, Tories 39.4%, Liberal Democrats 20.7% and others 4.9%. In other words, the swing in 2008 compared with last year was nearly 8%. Labour's change of leader to Gordon Brown has not improved things - it has made them much worse.

Yes, of course, there are caveats to enter. A local election is not a general election. Fewer people vote in locals and people can vote differently in different elections. Nevertheless, in this one constituency, Labour's 189th most marginal seat, Labour trailed by more than 19 points yesterday. If it had been a general election, Denham would have been swept out, a high profile victim of an enormous surge to the Tories. If he is to hold this seat in the next general election there will have to be an immense reversal of Labour fortunes"

Notice that the logic of his 8% swing to the Conservatives suggests the 10% Labour victory I indicate above, rather than the Tory victory he implies.