Polls, damn polls and statistics
I don’t get this, I really don’t. Political opinion polls, we’re led to believe, reflect an accurate representation of the views of voters.
Last week, New Labour had a 13 per cent lead according to a IPSOS-MORI poll. Tonight, according to a YouGov poll for Channel 4 News, that lead is now just four per cent.
What’s going on? Are people really that fickle? I’m as disparaging of the British public as the next man (it’s more of an exasperated disappointment, if truth be told), but even I find it hard to believe that nine per cent of the British electorate have changed their minds – in just a week – over who they’re going to vote for. Are we really that malleable?
Somebody explain it, please?
Posted on October 4th, 2007 at 7:04pm under UK politics
| Related posts... • Chuckie Bum Tax Bombshell Go Boom? • It’s Iraq Week, look busy • Bye bye buy-to-let |
• Permalink • Trackback • Subscribe |
|
|
|
• 12 Comments |
Given that Tories and New Labour are indistinguishable, I think it stands to reason that there may be huge and rapid swings.
What I cant understand is why anyone would still doggedly stick to voting Labour. Maybe nobody does. Maybe everyone flips a mental coin when they’re asked how they’ll vote.
Re the 9% change in Labour lead. Only 4.5% of the voters need to change to give a 9% swing:
If 4.5% of Labour voters leave Labour and all of that 4.5% vote for Tory we get
L – 4.5%
C + 4.5%
==> 9% change
Margin of error. It don’t mean a thing if it ain’t got that
swingnumber of people surveyed and margin of error interval.The MORI poll says “Representative sample of 1,964 UK adults aged 18+ interviewed face-to-face between 20-26 September 2007.” The Lab/Con percentages thus have a margin of error of +/-2% (95% confidence levels). Ie. when the poll says “Labour 44%”, what it actually means is “Labour somewhere between 42% and 46% but there’s a 1 in 20 chance it could be outside that”.
So Poll A (MORI):
Lab 42-46%
Con 29-33%
Poll B (YouGov):
Lab 38-42%
Con 34-38%
So if poll A overstated the Lab lead to the max, and poll B underestimated similarly, you only need 1% to switch.
The other polls show leads of 1 and 3 points for Lab, but with rather smaller samples (the Populus has a sample of only 800 which implies +/-3.5% @ 95% conf).
So basically, these polls are meaningless because they mean almost anything?
There’s a lot of literature that suggests that the answers people give to polls are shaped by social factors (what they think the interviewer wants to hear) and framing effects. Also the hypothetical “if there was a general election tomorrow how would you vote” is not necessarily a good predictor for actual party preference. Basically people come out with any old crap.
The head chap of Mori on R4 yesterday when asked if politicians read them (or words to the effect of):
“they say they don’t but then they come out with all the figures. But George Bush said he doesn’t read them, I believe him. He gets Karl Rove to read them to him…and then explain what they mean”
Not meaningless, no, just don’t read too much into day to day percentage point movements rather than broad trends. I think there have been enough polls that are broadly consistent in each case that we can say that Labour has a lead, and that that lead has narrowed over the last week, but to be more accurate than that is spurious.
The “headline” figure is the best estimate, particularly if several polls at about the same time come up with similar answers. But don’t read too much into a few percent either way – and occasionally you get a “rogue” poll.
I hate treatment of polls during election campaigns, where movement of one or two percent are blown up into signs of “momentum” or “slipping”, or plotted on pretty graphs without any sign of error bars. Such small movements are practically meaningless unless sustained by several different polls.
As well as the margin of error, I’m pretty sure there are a larger than usual number of people who remain undecided as to who they’d vote for.
These headline polls are designed to force people into make a straight choice; see the “if there was a general election tomorrow how would you vote?” question. The large number of wavering voters will mostly plump for one party or the other (depending on what’s been happening that day/week) rather than saying that they are undecided. As such, the headline figures are more volatile than usual.
You should also be careful of the ‘margin of error’ whilst the 95% confidence limit may be +- 2% it is much more likely that the error will be between +-1 then -2 -> -1 OR +1 -> +2%
Then we have the well known facts that Polls tend to overstate Labour Support – whether this is because fewer Labour supporters turn out or whether people use their heads rather than their hearts when faced with the choice in private in the polling booth, I do not know. This was shown most famously in 1992 when at 10.00pm BBC and ITV were reporting a hung Parliament on exit polls! but Mr Major secured a majority.
In this country we have a bedrock of Labour supporters and a bedrock of Conservative supporters that will never switch. It is only the undecideds that switch. A swing of 4.5% is a massive swing as only a relatively small percentage will.
Looking at electoralcalculus.co.uk even a 38-38 split would give Labour a working majority of 42 – the major effect would be the collapse of the Lib Dems (Who gained many seats on an anit-war protest last time)
Its all bollocks
Indistinguishable parties but a media saturated in personality and ‘pollsters’ interviewing wifies doon the shops saying ‘do you like Mrs Broon?’
all absolute fucking bollocks
If anyone’s using their heart to vote labour it must be a stone cold one.
Don’t forget all those polls in the spring that said Brown would be even less popular than Blair. They were, er, wrong.
Don’t also rule out the possibility of fieldworkers making up the results in order to meet their submission deadlines. As someone with a background in market research, I can assure you that this is common practice, no matter how reputable the company.