Yougov Now Says Hung Parliament

[scooping]

Blimey

[douglas9401]

I now say hanging's too good for them.

[maggiebee]

Ha ha Douglas; you took the words right out of my mouth!

[naomi24]

They're probably using the same sources that they used to predict an Ed Miliband victory.

[jim360]

Actually not; it's meant to be an updated model using more advanced statistical techniques (and a sample size around 50 times larger than typical polls).

But never mind. It's still an outlier at this point; there hasn't been a poll since April last year that had Labour ahead (and only three total since the 2015 election). Eye-catching but hardly definitive.

[naomi24]

//there hasn't been a poll since April last year that had Labour ahead (and only three total since the 2015 election).//

Curiouser and curiouser.

[jim360]

You can look at it a few ways. Labour don't need to be ahead to stop the Tories winning, just "close enough". I believe this particular poll has a national four-point lead for the Tories. Also the current constituency boundaries favour Labour (or, at least, don't favour the Tories so much), so that again helps stop a narrow Tory win from leading to enough seats.

Finally, as YouGov has pointed out, this headline figure is central and comes with some error. I am not sure exactly how they'd define their confidence bands but assuming the usual 68% then it's about a 30-40 seat swing either way. So 310 seats for the Tories means a 70% change of between 270 and 350 seats, ie a majority is still quite reasonable based on this poll.

[naomi24]

Conjecture ('cos that's what it is) is best ignored - in my opinion.

[jim360]

Fair enough.

Electoral Calculus puts it best, I think:

"Electoral Calculus salutes YouGov for their modern approach of combining polls and models to get richer and more insightful predictions. We will know next week whether their approach has got it exactly right this time. If it's right, then they will be justly celebrated. But even if it isn't, it is still the right thing to do and the method can be refined in future years to be more accurate. One day, maybe all polling will be like this."

[naomi24]

//But even if it isn't [right], it is still the right thing to do //

Why? If it isn't right, it's misleading. I can't see how misleading the public is the 'right thing to do'.

[hereIam]

Blimey indeed !

[jim360]

Because how else do you make progress? A technique has to be tried and tested in order to be evaluated and improved.

As I have stated, YouGov has provided some information about the techniques and associated errors behind this poll. If they don't get reported more widely, that isn't misleading on their own part. And maybe next time the errors will be understood better, and the accuracy will be improved. Progress

Oh, and -- again -- the prediction for 2015 was for a Hung Parliament. This was still wrong, but it wasn't a Miliband victory as you keep saying every time this comes up.

[Apariah]

is that the poll that predicted Hilary Clinton would win the US election?

[mikey4444]

Apariah....Hillary got many more votes than Trump, but the system won in the end.

[naomi24]

Jim, you take it seriously if you like. I prefer to wait and see.

[jim360]

I prefer to wait and see as well. But I think you are confusing taking the method seriously with taking the results seriously. I've said several times that this is interesting but hardly conclusive -- that doesn't detract from the methodology. But because the headline "hung parliament" result is still attached to an uncertainty then I wouldn't see this as being that we will (or even probably will) end up with a Hung Parliament next week.

Still, the fact remains that, two weeks ago, polls indicated a huge Tory lead, and now that lead is gradually vanishing. Since these polls will all carry the same flaws and same systematic vulnerabilities (eg
overestimated Labour support), then taking any one seriously is mistaken but they still can form a coherent picture, because they'll be wrong in the same way as each other.

So that leaves us with a very likely truth: The Tory lead has shrunk. Has it shrunk enough to stop a Tory win, or just changed the scale of it? I'd tend towards thinking that this is only the difference between a majority of 150-odd and a majority of 50-odd.

It's worth paying attention either way, as the advice has to be not to take the result for granted and go out and vote!

[Apariah]

Hihilary lost the election under the rules, you can't change the rules to what you'd like them to be.

[naomi24]

Jim, to be perfectly honest I’m not really interested in the methodology if the results aren’t accurate.

[jim360]

You never did much statistics, did you? Certainly your definition of "accurate" seems to me to mean something like "spot on or worthless", which seems incredibly restrictive.

Apariah - the final prediction for the 2016 US election, from fivethirtyeight.com, was that Hillary Clinton would win 48.5% of the popular vote, and Trump 44.9%. In the event, Clinton got 48.2% and Trump 46.1%. By most standards that's not a bad prediction. Unfortunately, the small miss had huge consequences in terms of the Electoral College, as Trumps "extra" votes were concentrated in a few states -- notably Wisconsin, Pennsylvania and Michigan.

As misses go, a small error has had huge consequences. But such is the nature of probabilistic methods. There is always the chance that you'll end up wrong, but that doesn't mean that your approach was flawed.

[naomi24]

Jim, //You never did much statistics, did you?//

Why would I do statistics?