The Chamberlain News

Who’ll be the big West Midlands losers: Con, Lab or the pollsters?

After two terrorist attacks and the Whitsun holiday, Chris Game gets us under way again in the run up to Thursday’s General Election with a look at the widely ranging polls – and in particular what YouGov “estimates” for the West Midlands.

Remember the 2015 election – and just how comprehensively the final opinion polls screwed up?  Of the 11 polling companies, five had the Conservatives and Labour level, and two actually put Labour in the lead.

The result, in case you’ve forgotten, was Conservatives 38%, Labour 31%. Only four polls got even the winning party right, and none had the Conservatives more than 1% ahead.

If anything, the full story was probably worse still. The chances are that all the polls were consistently wrong throughout the campaign. And last year’s EU referendum was virtually a repeat, with six of the eight companies’ final polls putting ‘Remain’ in the lead by up to 10%.

However, as you may have observed from the appearance of a new poll pretty well every day, these industrial scale disasters haven’t discouraged the pollsters, any more than the similarly comprehensive cock-ups in the 1970 and 1992 elections did.

They’ve had their inquests, reckon they know what went wrong, have made the required methodological adjustments, and will get it right, or at least righter, this time. And in one way at least that’s probably true.

What went wrong, it’s generally agreed, was much more than the attractive instant excuses of ultra-late swing, or ‘shy’ or ‘lying’ Tories – intending Conservative voters reluctant to reveal their embarrassing polling booth intentions, particularly to pollsters face-to-face.

The common fault almost certainly was that, notwithstanding their very different sampling and weighting methods, all the pollsters were interviewing just too many Labour voters.

This isn’t a methodological blog, but the important point is that no commercial election poll – whether using face-to-face, phone or online surveys – can possibly undertake the true random probability sampling that statistical methodology ideally requires, with every British adult having an equal chance of being selected and participating.

Even if you had a random sampling frame and could somehow contact everyone within it, you couldn’t force them to answer your questions, never mind truthfully. Whatever the method, then, a degree of self-selection is involved – reflecting the respondent’s interest in politics and readiness to share their views.

Two effects of which, in modern-day Britain, seem to be that you end up with too many Labour voters and too many young people who, whatever their politics, are happy to chat but less inclined actually to vote, or even register.

Judging from the evidence, the adjustments the pollsters are now making to cope with such problems vary considerably. For, writing as I am before the weekend (and before the latest terrorist attack, Ed.), the 10 most recent polls have recorded Conservative leads of between 14% and 4% – a spread that would seem to increase the chances of at least someone getting a bit closer on Thursday than anyone managed last time. We just won’t know which until it’s too late to be useful.

The remainder of this blog focuses on just one of these polls, the one showing (on June 1) the lead of 4%. The main reason, though, is because the YouGov authors of this poll have opted for an open data policy so open that, if conceived by Yes, Minister’s James Hacker, would undoubtedly have drawn Sir Humphrey Appleby’s congratulations for an extremely courageous decision – ‘courageous’ here as in rash bordering on deranged.

It’s rash for several reasons, but most obviously because of its ambition and scale. YouGov have pledged to publish, between June 1 and Election Day, based on their aggregated statistical election model, .

YouGov has grown in just over 15 years into a huge, international, entirely internet-based market research firm, all of whose online surveys are conducted using a carefully recruited panel of over 800,000 UK adults.

For this quite unprecedented election exercise, they’re interviewing roughly 7,000 panellists each day (50,000 per week) about their voting intentions, then statistically weighting these panel data to the national profile of all adults (including those without internet access) by age, gender, social class etc., but also by previous election and EU referendum votes and level of political interest.

By conventional polling standards, these are huge numbers, but, spread across 650 constituencies, still only about 75 each – much too small for reliable voting estimates. Which is where the seriously tricky MRP bit comes in (sadly, not ‘mature’ role play, but Multi-level Regression and Post-stratification): applying and scaling up the week’s data to “produce a fairly accurate estimate of the number of voters in each constituency intending to vote for a party on each day”.

Obviously, there are all the standard polling qualifications. These are snapshots, not forecasts; even huge samples have margins of error; people change their minds; plus, as a national model, it can’t allow for any specific constituency-level factors.

On the other side of a very unbalanced scale, but as they’re understandably keen to emphasise: “we publicly tested the model during last year’s EU campaign and it always had ‘Leave’ ahead”.

Fellow pollsters are mostly sceptical, as they generally are of any kind of seat forecasting. Personally, however, I’m hugely admiring of both their methodology, which of course I don’t come near to understanding, and their sheer chutzpah, and I felt Files readers should share the fun.

Hence my table, summarising their initial set of ‘estimations’ for Birmingham constituencies and a selection of others around the region.

 

 

 

 

 

 

 

 

The table should be fairly self-explanatory, but I should add one additional point, and perhaps emphasise thereby that these people aren’t completely off the wall. For each constituency they actually give a range of estimates for each party, covering for the big parties perhaps 10-12%, and the figures given in the table are those chosen by them from obviously around the centre of that spread.

Some recent context may also be useful. Theresa May announced this election on May 18, immediately following which the .

‘Wipeout’ was perhaps a touch OTT, but, based on “an average of ten recent polls”, the paper suggested the Conservatives (this was before they became the May Party) could get an overall Commons majority of 106, with Labour losing a net 47 seats.

In the West Midlands these would comprise all those in the table with 2015 Labour majorities of under 12% and therefore requiring Labour-Conservative swings of no more than 6%. In Birmingham, therefore, it would be Edgbaston and Northfield, and SEVEN (Mail’s emphasis, not mine) in the rest of the West Midlands: Coventry S and NW, Dudley N, Walsall N, and Wolverhampton SW.

A week or so later a chap going by the surname of Dale was reported in these columns as predicting that not two, but three, seats were set to switch from Labour to Conservative, the third being Erdington; and, on top of the Mail’s list of seven, Wolverhampton NE.

No, not that Dale. Whether Paul would have been as rash we’ll possibly never know, but this was Iain Dale – daily political blogger and commentator, and author of numerous books on politics, including a quick-off-the-mark, . A busy man, obviously, but perhaps not that frequent a visitor to Kingstanding and Stockland Green.

There’ll be winners and losers aplenty come Friday, and completely regardless (I promise you) of the content and political direction of their ‘estimates’, I really hope this YouGov initiative doesn’t end up getting completely trashed. It deserves better than that.