Tuesday 20 January 2015

#GE2015 Forecast by UK ELECTION TREND

Some change in the model's prediction, as the SNP's gains may be cooling off, but the 2015 election still looks on pace to return a left of centre government.

There's concern in the Labour camp that a rise in Green vote share due to publicity from the debate dispute might split the left. However, it's my gut feeling that won't be the massive effect it's feared. While the Conservatives are being damaged by UKIP, it took a massive rise in their support to do so, and based on multiple election cycles of them getting there. The Greens are not in the same position yet. They might get there, but I doubt they have time to do so before May. Greens are special cased in the model along with SNP, because they are not expected to contest nation wide, and so it's hard to extrapolate what's happening with them.

At the moment, it still looks like a continuation of "A close VI result, but a clear advantage to a left of centre Government forming".

As promised I regenerated the history for the model, and while I was doing so I looked at the history of vote share to see how the assumption of a revert to mean was doing. All other things being equal, the vote share prediction should be completely flat as time goes on. After all, if the model is correct, and polling doesn't shift entirely unexpectedly, the prediction shouldn't change. However, as we can see, there has been a drift. This is most noticeable in the third parties, because they have the most movement when a revert to mean is applied. The drift seems somewhat constant across the entire period, and not caused by changes to the economic confidence portion of the model.

It's important at this point to explain how the model works in this regard. A naive "swing back" model views the election as a simple contest between the government and the opposition, and that support from the opposition will move back to the government. Imagine if you will a tennis ball rolling along a line between the two parties, "swing back" says the ball will tend to roll back towards the governing party. Obviously, that's not happening in this election. It's tightening, but there's no clear "swing back". That's because it never really was such a simple thing, and the rise of multiple parties having a strong effect on the result is just highlighting this.

My model instead views the election result as a point in space, that is being pulled upon by the various parties. Now the tennis ball isn't rolling along a line, but moving around in the three dimensions* of the tennis court. The fun part being that it's at night and they don't turn on the lights until the game is won, so you can't quite know where the ball is. But you know the ball is getting pushed around by the forces governing the election, public sentiment, campaigning, and so on. And that it is acting against a slowly diminishing drag of reverting to it's previous state at the last election. You can see glimpses of where it might be in the dark from polling, so you can try to judge where it will be on the day of the election...

I'd applied an assumption that revert to mean would slowly diminish by a percentage for each day closer to the election, increased or decreased by economic factors in the form of consumer-confidence that moderate this. Looking at the history, I've under-estimated how much it would diminish by a small amount. Correcting for this is moot now, because that amount is now bellow the noise margin this close to the election.

But it does let me know the following things. Revert to mean has apparently had an effect on this election. "Swing back" was an incorrect assumption, and applying it would give you an incorrect projection of how polling moved. Revert to mean has a relatively lower effect than previously estimated, even when it is not being hugely moderated by consumer-confidence.

originally posted here >>> http://ukelectiontrend.blogspot.co.uk/

No comments:

Post a Comment

Comment is open to all feel free to link to this blog.