See, it does happen
A special in response to the criticism you sometimes see that “no one I know ever gets called by a pollster”. Remember there are only a couple of hundred voting intention polls carried out each year, and they include a one or maybe two thousand people. The chance of any individual being polled in any given year therefore is only around 1 in 200. Anyway, just to demonstrate that it does happen Mark Pack over at Lib Dem home got called up by ICM yesterday. I suspect he said he would be voting Lib Dem!
That interesting first question that Mark writes about is “Some people have said they would not vote in a new General Election, while others would go and vote at their polling station. I would like to know how certain it is that you would actually go and vote in a general election?” The bit of bumpf there are the start about some people not voting and some going to polling stations is to try and get over the social desirability bias and letting people know that it is alright to admit that you might not both to vote.
I have been polled twice for a national Yougov poll and once some 30 years ago in a face to face poll from Gallup or NOP ( too long ago to remember which ) .
Anthony
The comment about likelihood to vote gives me the excuse to raise an issue I’ve been wondering about and which Luke Akehurst has raised on his blog: what difference will it make if there is an increased turnout at the next election?
The polling companies all use one methodology or other to deal with the problem of Labour supporters who don’t actually vote: either by screening out those not certain to vote or by weighting to past voting (revealed rather than stated) behaviour.
But what if we assume that the low turnouts since 1997 are the aberration because everyone, Labour supporters included, believed that Labour was going to walk it, and that with a closer contest turnout goes back towards the 70+ percentage norm?
If we look at the IPSOS Mori data, and we count not just the 55% or so who score 10 on certain to vote , but also the 8s and 9s, we see that this would tend to favour Labour. However I don’t know how robust this is I haven’t seen all the data. Also with the polls that use weighting you’d need to adjust the weighting formula not to the 2005 turnout but to a higher figure.
Interested in your thoughts…
YouGov pay you though
I’m puzzled why ICM don’t have a “or vote by post” subclause in their openning question. Can you shed any light on this?
Chris – I did a long post on it here. Incidentally, in there only 1% of non-voters said the reason they didn’t vote was because the winner was obvious, the top answer was 19% who said they couldn’t be bothered. (Lots of those “couldn’t be bothered” could easily have not been bothered because the winner was a foregone conclusion though)
Generally speaking a higher turnout should help Labour, since there are more Labour supporters who say they might not vote than there are unsure Conservative supporters. It makes the assumption though that all non-voters are equally likely to be spurred into action, and we really don’t know that – if turnout did rise it might rise more amongst supporters of one party than another, there could easily be a skewed rise.
The polls should take into account any rise in turnout anyway – none of them assume a certain level of turnout, MORI take the 10/10s whether that is 55% of the sample or 75%, ICM take 7+/10 however many they are, Populus weight 10/10s to 1.0, 9/10 to 0.9 and so on regardless. If turnout does rise (I haven’t looked at the figures for a while to see if there is any sign of a trend, it’s something I should really do soon) it should already be taken into account in the polls. If turnout rises in the future, we can’t tell who these newly enthused voters are, so we can’t say what the effect will be – it depends entirely who they are.
A final thing worth considering is that a lot of what people call “bias” in the electoral system is down to low levels of Labour turnout in their safe seats – it’s probably not co-incidence that lower turnout and the perception that the system is increasing biased against the Tories have happened at the same time, it’s cause and effect. If all those Labour voters who didn’t bother in 2001 and 2005 suddenly decided to vote, they would disproportionately be in safe Labour seats, so an increased turnout could increase Labour’s headline support, but whittle away the pro-Labour “bias” in the system.
Mark – I can’t speak for ICM, they might just not have considered it when writing the question. On the other hand it could be another subtle encouragement to get people to admit to not being bothered to vote, by reminding them it’s likely that voting will involve them leaving the house and going all the way to the polling station.
You know how tricky it is knocking up come 8pm on polling day when people have got home from work and settled down on the sofa – ICM probably want to gently remind people that this voting malarky actually involves getting up off your arse, and maybe they aren’t as certain to bother doing it as they think they are.
I remember reading somewhere (sorry, forgotten where) that turnout in 2005 was up to 10% greater in the most marginal seats than in the least marginal.
If that is so, then there is a good chance that turn-out will be greater in 2009/10. Also, the increasing popularity of postal voting could increase turn-out.
I’ve completed a few YouGov voting intention surveys in the past.
As a citizen of Denmark, it has puzzled me why they don’t ask about nationality. Being an EU citizen, I can and do vote in municipial, Scottish and European elections, but Westminster elections are not open to me.
However, when they ask about my likelihood to vote, they seem to assume I have the right to vote, so when I say I’m unlikely to vote, I imagine I’m classified as not being all that interested in politics, which is completely untrue. Furthermore, because I can probably become a UK citizen fairly quickly if I wanted to, having lived here for more than six years, my likelihood to vote in a UK election is not 0, especially if the next election is far off.
Anthony
Could I press my point on turnout a bit?
I know the polls will pick up changes in likelihood to vote; they may get the absolute level wrong but they should pick up the direction.
My point is a slightly different one. It’s about whether the adjustments that pollsters make to reach their headline figures are overly dependent on turnout remaining the same. Maybe, in effect, they are over-adjusting their pro-labour bias in the past.
Weighting party support to actual voting behaviour at the last election disproportionately affects Labour support so in effect adjusts the raw Labour figure downwards a bit. MORI achieve a similar effect by only counting the certain to vote, not because there’s any reason in principle to do this – some of the 9s, 8s and so on will in fact vote just as some of the 10s won’t – but because it’s a rough proxy for actual turnout last time.
Either approach works fairly well if turnout stays the same from one election to the next, as it did between 2001 and 2005. And indeed the polls turned out to be pretty accurate in 2005.
But am I not right in thinking that the various adjustments pollsters introduced to avoid their pro-Labour bias were in place before 2001? Yet the polls were still wide of the mark in both 1997 and 2001, mainly by of overstating Labour support.
I am hypothesising that part of this was because in 1997 and 2001 they failed to allow for the falling turnout (disproportionately affecting Labour support), which obviously was difficult to predict in advance.
The conclusion if I’m right would be that the current adjustments by pollsters will prove accurate if turnout remains the same as in 2005. But if turnout increases as it could, although I agree that’s a matter of judgement, then we could find in 2009/10 that the polls have been over-adjusting for their pro-Labour bias.
Chris – I think your general hypothesis is wrong, in 2001 a lot of pollsters were “unreformed”. There wasn’t a single move for pollsters to reform their methods after 1992. Most made small adjustments in response to 1992, but the wholesale shifts happened slowly over more than a decade.
ICM adjusted their methods around 1993/1994, and their figures in 1997 and 2001 were pretty good. The reason the polls overall were still wrong in 1997 and 2001 is that ICM were the only ones who had really put their house in order. MORI only added their harsh turnout filter in 2003 and Populus and YouGov had yet to emerge. Gallup were still using the methods that failed in 1992.
ICM have altered their turnout model since 2001 and 2005, but not substantially so (it probably wouldn’t make a whole percentage point of difference) and it is adaptive – it doesn’t assume a low turnout, and it’s accuracy shouldn’t depend upon turnout remaining at a similar level. ICM’s real innovations in 1993 were the topline adjustment and political weighting, that’s what made the difference.
People seem to be very bad at predicting their likelihood of voting, but I can see no reason why there should be a political skew to this, why Tory voters should be better or worse at predicting their own behaviour than Labour ones. Turnout filters and weighting are not intrinsically favourable to the Conservatives. If Labour voters became more likely to vote, I can’t see any reason why they wouldn’t report that in polls and, in turn get included in MORI and ICM polls in greater number (and weighted more highly in Populus polls).
The only one where I think you are right is with MORI, where levels of turnout could make a real difference. The absence of political weighting means their figures would otherwise be far more Labour than other pollsters, but this is cancelled out by the very harsh turnout filter which is far more favourable to the Conservatives than other pollsters. There is no reason for these two things to roughly cancel each other out – it’s just co-incidence.
Unlike other companies where the political weighting is the big thing, MORI depend on the turnout filter to bring their figures into line with actual results. The implication is that, while ICM and Populus believe raw samples are wrong because the sample is skewed and needs correcting, MORI believe that the raw samples are representative, but other companies are including too many non-voters. If things happened with turnout therefore, MORI’s figures could diverge from those of other pollsters – for example, if Labour voters told MORI they were more likely to vote and their turnout filter ceased to be so pro-Tory, we would start to see MORI polls showing much larger Labour leads than other company’s polls.
(Looking back, I think I’ve just spent five paragraphs saying “turnout makes much more difference in MORI polls than other ones”
)