Mike Smithson on why Lib Dem poll ratings vary so much

by Stephen Tall on October 24, 2008

Mike Smithson, founding editor of PoliticalBetting.com and occasional Lib Dem Voice contributor, has written an interesting post on his site analysing why the Lib Dems’ poll ratings vary so much between the different pollsters – this is a question that LDV has also looked at a few times, including this past week, with two polls (by YouGov and ICM respectively) placing the party at 14% and 21%.

Mike notes:

All the pollsters have different ways of processing the data and the ICM mathematical approach is probably the most friendly to Nick Clegg’s party. But that is not enough to explain the often quite substantial differences that we see so often. I have looked at this before and am now even more convinced that it is down to the way the questions are put and the actual words that are used.

Mike then compares the different wording used by four of the main polling companies, ICM, Mori, Populus and ComRes (though not YouGov) to ask voting intentions – the differences are subtle, but it’s certainly plausible that they magnify the differences which we’re seeing in the polls at the moment. And as LDV has long argued, which firm you believe to be most accurate will probably largely depend on which firm you would like to be most accurate.

Two other related points I’d note. First, all the polling firms ask the question using the phraseology, “if there were a General Election tomorrow…”, yet the public knows there won’t be an election tomorrow. It’s a false hypothesis, and will inevitably distort the figures to some extent or another. Secondly, there is one time when it’s not a false hypothesis – on the eve of a general election: and what we normally see during general election campaigns is the polls beginning to converge, certainly to within the margin of error of each other.

And of course election campaigns are important to Lib Dems, both because of the increased media exposure we gain, and also because the public is reminded – in a way that’s often forgotten between elections – that in over 200 constituencies up and down the country, the Lib Dems are either defending the seat, or are the main challengers to the incumbent.

Enjoy reading this? Please like and share:

No comments

I find it difficult to believe that asking the questions in a subtly different way could account for the Lib Dem rating in the latest ICM poll being 50% higher than it is in polls by other companies.

I suspect this poll will turn out to be freakishly high, for reasons discussed elsewhere.

But on the reason why ICM’s ratings are consistently favourable to the Lib Dems, I thought this comment by Anthony Wells on the other thread was interesting:

“YouGov, MORI and ComRes all dismiss don’t knows – making the assumption that would either be unlikely to vote, or that their votes would split in roughly the same proportion as other people (or that it isn’t a pollsters place to second guess what people will do).

ICM and Populus both reallocate a proportion of the don’t knows on the assumption that they will vote in the same way as they say they did in 2005. Populus reallocates 50% of former Conservative and Labour voters, and 30% of former Lib Dem voters (which seems unfair, but is on the back of how interviewees behaved in Populus’s 2005 polling). ICM reallocate 50% of former Con, Lab and LDem voters.”

If you’re an optimist, you could say retains some of the boost the previous election campaign gave us (though presumably that implies there’s less of a campaign boost still to come). If you’re a pessimist, you could say that it’s simply disguising the fact that our support has fallen since 2005.

by Clegg's Candid Friend on October 24, 2008 at 10:21 am. Reply #

Reallocation of don’t knows is a factor but not as much so as weighting for past vote .
Compare the recent Mori and ICM polls
Mori sample ………… 1004
No of LD’s ………….. 136
No who say voted LD 2005. 106
No adjustment for past weight voting in Mori

ICM sample …………. 1007
No of LD’s ………….. 125
No who say voted LD 2005. 111
ICM weight by past vote to a figure of 13% LD’s on 2005 therefore their last 2 figures adjusted by weighting become
No of LD’s 138 and No who say they voted LD 2005 129
Note that despite the final headline figures in the 2 polls the Mori LD raw data was in fact marginally better for LD’s than ICM .
Another useful comparison is to compare the changes for how people say they will vote now compared to how they say they voted in 2005 .
Mori have 30 people changing from Lab to LD with 6 from LD to Lab
ICM have 17 people changing from Lab to LD and 6 from LD to Lab
Mori have 15 people changing from LD to Con and 7 from Con to LD
ICM have 7 people changing from LD to Con and 8 from Con to LD
Although the ICM figures for LD/Con switchers are more favourable to the LD’s than you would perhaps expect overall the raw Mori data is rather better for LD’s than the ICM data and is it not strange that raw data that indicates a strong movement of voters from Labour to LD with a lesser movement from LD to Con can end up with a headline Mori figure indicating a substantial loss of support – something does not compute !!!!
We can do the same analysis with the recent Comres poll The figures are :-
Sample size 1005
No LD’s 111
No LD’s 2005 112
No LD’s after weighting 115
Changes from how people voted in 2005
LD’s gain 18 from Labour lose 9
LD’s lose 13 to Conservative lose 3
The figures again show that LD support is relatively unchanged losses to Conservatives being almost matched by gains from Labour .
My own view is that the detailed data tells us far more than the headline figures quoted in these polls .

by Mark Senior on October 24, 2008 at 11:47 am. Reply #

When working in Cowley St 10 years ago someone told me that “ICM polls us high but accurately” and I’ve generally found that a pretty good rule of thumb since then.

by Jeremy Hargreaves on October 24, 2008 at 11:48 am. Reply #

Mark Senior

Thanks for posting those figures, but (as you say) it’s difficult to make sense of them.

In particular, the MORI numbers show 30 more LDs than in 2005, but the party-by-party switching figures imply an increase of only 16. I don’t understand that. Is it down to people who didn’t vote, or voted for Others? In contrast, the discrepancy in the ICM figures is only 2.

by Clegg's Candid Friend on October 24, 2008 at 12:24 pm. Reply #

Looking at these figures

Is it possible that ICM “factor in” the increase in vote that we get in the campaign through there questioning methodology. That would mean the other companies would show us increasing during the campaign but ICM would have us static.

Interesting what Jeremy says. IIRC going back to the late 80s/early 90s it was MORI who gave us the higher ratings.

by Hywel Morgan on October 24, 2008 at 12:31 pm. Reply #

CCF Other factors are as you say switchers into don’t knows won’t votes or died and in from did not vote or too young in 2005 . A bit annoyingly the way the tables are presented show switchers between LD and others only one way but probably usually would show a small net loss .
Re your comment on Mori , the discrepancy you mention would not have been as high as 30 if they had weighted for past vote as the 106 would have increased rather more than the 136 .

by Mark Senior on October 24, 2008 at 12:52 pm. Reply #

Leave your comment


Required. Not published.

If you have one.