Yet another election prediction

You know what this election needs? More nerdy predictions on what is going to happen in just a couple of days’ time. In my defence, I am a massive geek and doing this brings me huge satisfaction, it is you, the reader, that I’m worried about. With that said, what follows is my barely educated guess on what I think might happen on Thursday.

Following the polling miss of 2015, pollsters have employed a variety of fixes to ensure that they get more accurate results this time. These tend to centre on the following;

· Fix the samples

· Adjust for turnout

· Reallocate don’t knows

All pollsters worth their salt have attempted to fix their samples though some have done more on this than others. In my heavily biased view, YouGov have done the best work on fixing their samples; improving the political attention breakdowns, sampling and weighting by education and splitting out vote by region (all these changes helped in the EU Referendum where YouGov had more Leave leads than Remain). In theory, and in an ideal world, fixing your samples alone should fix the problems of 2015.; some pollsters agree, those showing lower Tory leads, and some disagree, those showing higher leads.

There is also a significant divergence on the second change pollsters have employed, that of turnout. One group of pollsters, again those with lower Tory leads, have focussed on self-reported likelihood to vote, whilst others have made their samples look more like the electorate in 2015. Stephen Bush has written excellently on this so I won’t rehash his arguments.

At the last election, this would have been a much better way to go but it can cause significant problems if the makeup of the electorate changes from one election to another. This often happens in the US and embarrassed a couple of UK pollsters at the EU referendum.

The final change is down to manipulating your data based on how you think the electorate will behave. Note, in this context, the word “manipulation” should be viewed as a positive thing given pollsters are honest and upfront about what they’re doing and are simply trying to get accurate results. ICM, long considered the gold-standard pollsters, have been doing this since 1992, reallocating don’t knows which, usually though not always, boosts the Tory score.

So, what does this all mean? Some pollsters have decided to try and make their samples more representative and, largely, leave it at that whilst others, ComRes and ICM chiefly, have fixed their samples, adjusted for turnout and manipulated the data. In my view, I think it is quite possible that some pollsters have gone too far with their changes whilst other have not gone far enough. Given that there are about the same number in both camps, I have a suspicion that the polling average will prove to be more accurate than any individual company this time.

Though there is a sizeable caveat to this (or me hedging my bets if you prefer) in that it is just as likely that one group of pollsters is right and another wrong. Historically, you would assume that being high on the Tory number is a good place to be which happens to also tie in much more closely with what MPs are reportedly seeing on the doorstep. So, there is scope for my estimates to be off and, if they are, it is likely I have understated the Tories.

Now onto the numbers. My, very basic, rule of thumb has been to take three away from Labour’s highest score and add three onto the Conservative’s lowest (about how badly the polls erred in 2015) or take the polling average, whichever is better for the Tories. I’ve excluded Survation in this analysis as they are very clearly the outlier here and there are significant issues with their sample for their most recent poll. This might be a wise choice or it might be stupid but only time will tell.

Based on the above, my best guess is that current voting intention looks like this (this is not a prediction for Thursday, I will update this post once the final polls are known):

Plugging these numbers into Electoral Calculus, we get a projection of a Conservative majority of 66. However, there are reasons to doubt uniform swing as an accurate measure this time. Before the Labour surge, much of the discussion was on how demographic change and variable swing was going to help the Tories and hurt Labour and I don’t think this has gone away making it likely the Tories will overperform UNS this election. With that in mind, my seat projection, if the election were held today, is as follows:

So, there you have it, my prediction with a couple of days of campaigning left, is for a Conservative majority of 80. For a whole host of reasons, not least because I am tribally loyal to my former colleagues at YouGov (who are the smartest people in polling regardless of the result this week), I hope that I’m wrong.

Regardless of the result, my one hope is that people will cut the pollsters some slack. Polling should be a science but is increasingly an art and despite their flaws, our political weather is still best understood by pollsters than by listening to the few pundits who shout the loudest.


Update following the final polls.

Conservative majority of 90