ComRes have a new poll of Rochester and Strood out tonight that shows UKIP with a solid lead. As far as I can recall it’s the first ComRes by-election poll this Parliament. Like all constituency polls it was done by telephone, and with a healthy sample size by constituency polling standards of 1500.

Topline figures are CON 30%, LAB 21%, LDEM 3%, UKIP 43%, GREEN 3%. The only previous Rochester & Strood poll was by Survation at the start of the month – that showed a nine point lead for UKIP. Obviously one has to be careful about direct comparisons between polls from different pollsters using different methodologies, so it would be wrong to draw too many conclusions about how opinion might have moved between the two polls (differences could be down to methods), but it certainly doesn’t show any obvious sign of the Conservatives eating into UKIP’s early lead.


As I write there is still one council to declare, but the maths mean that the overall result is going to be Yes 45%, No 55%. So, how did the polls do?

The final pre-election polls had all tightly converged around the same figures – Yes 48%, No 52%, with every company was within one point of this. In fact the level of No support was three points higher than this. For a single poll a three point error would be within the margin of error, but every poll being off in the same direction suggests some systemic error.

A possibility is the shy noes/enthusiastic yesses we discussed before the referendum, but on the face of it a simpler explanation is just late swing. The YouGov recontact survey on the day, going back to the same people they interviewed in their final survey found enough movement between final survey to actually voting to take the figures to YES 46%, NO 54%, one point from the actual result and enough to explain the apparent divergence. From that it looks as though no was going to win anyway, but there was a further movement from yes to no when people actually got to the polling station.


It’s been a long journey, but we’ve finally arrived at the eve-of-referendum polls. For a lot of the Scottish referendum campaign the discussion about polls was one of right or wrong – we had lots of polls showing the same trend (flatlining!), but showing different absolute figures. Companies like MORI, TNS and YouGov were showing big NO leads; companies like Panelbase and Survation were showing a tight race. Then we had a period of some companies showing a strong movement towards YES, some not, and we have ended up with everyone showing much the same figures (what was the true picture earlier in the campaign we will never know for sure – by definition you can check eve-of-election results against reality, but never mid-term ones). With one MORI poll still to come, here are the YES shares in the latest polls from each company (taking the online and telephone methodologies seperately for those companies who have done both):

Ipsos MORI (phone) 49%
ICM (phone) 49%
TNS (face to face) 49%
YouGov (online) 48%
Panelbase (online) 48%
ICM (online) 48%
Opinium (online) 48%
Survation (online) 48%
Survation (phone) 47%%

Essentially everyone is predicting the same result, the margin of error on most of the polls is around plus/minus 3%, every poll is within two percentage points of the others. This isn’t going to be a case of individual pollsters getting it right or wrong, they’ll either all be around about right or all be horribly out.

There’s a temptation when the polls are like this to say YES and NO are within the margin of error, that it’s “too close to call”. It doesn’t really work like that – these polls are showing NO ahead. The margin of error is on each individual poll, and it’s equally likely to happen in both directions. Hence if the “true” balance of public opinion in Scotland was 50/50 we’d expect to see a random scattering of results around that point, some polls showing yes, some polls showing no. We’re not seeing that. We’re seeing polls randomly scattered around the 48/52 mark, suggesting that’s most likely where public opinion is – a very small lead for the NO campaign.

It’s possible there will be a very late swing, that people will have changed their minds in the last few hours or in the polling station itself. In most polls there really aren’t that many don’t knows left to make their minds up though.

The alternative route to an upset is if the polls are wrong, if there is some systemic issue above and beyond normal random sampling error that affects polls from all the companies. I wrote yesterday about what the potential risks are – the main challenges in my view are first whether people who are on the fringes of society and normally play little part in politics don’t get picked up in polls but do vote; and secondly whether there has been an issue of differential response rate, have the obviously more enthusiastic yes voters been more willing to take part in polling that no voters are?

Personally I’m a little more worried about the latter – I think there’s more chance of the polls ending up underestimating the NO vote than the YES vote, but there comes a time when you just have to trust the data. The polls say the result will be around about YES 48%, NO 52%. We will see on Friday morning.


We’ve had a couple of weeks to digest the European elections now, and it looks as if the polls since then have been showing a slightly increased Labour lead. I’ve done a graph below of the last few months of polls – in order to remove any variation from irregularly or infrequently published polls the graph below shows voting intention in just YouGov and Populus polls – the two most regularly published polls. I’ve used a seven poll average, as it means (bank holidays aside) every data point is made up of 5 YouGov polls and 2 Populus polls giving us a nice steady figure.

Looking at the parties one by one, there’s an obvious downwards trend in Labour support, interrupted by the aftermath of the European elections, since when they seem to have enjoyed a minor lift in support. Conservative support wiggles about a bit, but there is no definite trend. UKIP support rises in the run up to the European election, but fades as they actually approach (perhaps the impact of the widespread accusations of racism, perhaps just random variation) before increasing in the aftermath of the elections. The Liberal Democrats are steady, but have perhaps faltered since the European campaign. Finally there is an obvious upwards trend in the previously steady level of support for the Green party.

The question of course is whether any of this will have any long term impact whatsoever, or is the merely the impact of the publicity and campaigning around the European elections. After the local elections last year we got a spike in UKIP support, which by July settled down again. We need to wait a couple of weeks and see if all the apparent changes in party support over the last couple of weeks revert back to the trends they were showing before May, or if they’ve had any longer lasting impact.

Meanwhile tonight’s YouGov poll for the Sun has topline figures of CON 35%, LAB 37%, LD 8%, UKIP 12%. A lower Labour lead than the last few polls, but nothing yet that couldn’t be normal sample variation.


I’m having a nice rest after the election, but a brief update to add the BBC’s projected national vote – CON 29%, LAB 31%, LD 13%, UKIP 17%.

So in relation to my previous comments on the local results, Labour’s lead is indeed only modest, very much in line with their position in the national polls. And rather than UKIP doing pretty much the same as they did in last year’s local elections, they’ve actually done significantly worse – 17% as opposed to the 23% they got last year.

I should also comment on what the Projected National Share is. It’s not a sum of actual votes cast, it’s a projection of what the results would be if the whole country was voting and the main *three* parties were contesting all seats (it doesn’t assume a UKIP candidate in every seat, though the process of taking only seats where Lab, Con and LD stood means that it does increase the effective level of UKIP contestation). As regular readers will know, there is a cycle of local elections and in some years the councils voting are more Toryish or more Labourish – so for example, last year’s locals were mostly in shire councils, this year’s elections were mostly in metropolitan councils. The PNS attempts to smooth out those differences so you can compare one election to the next – so even if there are some teething problems in accounting for a new party in the PNS, the year to year comparisons should be valid.