Predictions – How do they work?

A warm welcome to my ultra-nerdy new blog about polling and predictions in the run up to this May’s Scottish Parliament elections. Since I’ll be using my own home-brew predictor, I thought the most sensible opening post would be to explain how my predictions work – more specifically, how polling translates into a prediction. This comes with a big, bold disclaimer that this is a system developed by an amateur psephologist, not the utterances of an oracle.

Universal Swing (“Traditional”)

A common method of translating a national poll into specific predictions is “Universal Swing”. This method simply assumes that whatever change in vote share is predicted in a poll applies all across the country. So if last election a party won 20% of the vote and a poll predicts they will win 25% this time around, a prediction using universal swing assumes their vote share will go up by 5% everywhere. It doesn’t matter if in a given constituency they actually won 10% or 30% last time, the predicted increase will be 5%.

This is a nice, easy way to produce a prediction, and although it’s known that it won’t be an entirely accurate representation of the results on the day, it’s generally assumed that the places it over-estimates a change in vote shares will balance out those where it under-estimates. However, it does mean that it takes no account of regional differences in party support which can have significant effects on seat predictions given the Scottish Parliament’s regional list system. So I opted for something a little bit different.

Regionally Modified Swing (“Experimental”)

This is the highly experimental bit. Firstly, I approximated the Scottish Parliament regions using Westminster constituencies. This isn’t perfect as they don’t entirely line up – for example, Rutherglen is a constituency of the Scottish Parliament in the Glasgow region. However at Westminster, the constituency is Rutherglen and Hamilton West, which I’ve assigned to the Central region – but it’s close enough.

I then worked out the swing each party achieved in each region, compared to their national swing. For example, the SNP’s national swing was about 30%, having gone from 20% in 2010 to 50% of the 2015 vote. However, in Glasgow, they went from 17% to 55% of the vote, a swing of 38%. In relative terms, their swing in Glasgow was about 125% of their national swing. By comparison, in Highlands and Islands, they went from 23% to 47%, a swing of 24% or around 80% relative to their national swing.

That regional swing relative to national swing value is what I’m applying to poll results to make my predictions. To use the SNP values from above, if a poll predicts a national swing for the SNP of 5%, my calculator will make that out to be an increase of 6.25% in Glasgow and 4% in Highlands and Islands.

The tough bits… (“Breaking my own rules”)

As the SNP, Labour, Tories and Lib Dems stood in every constituency in 2015, their regional swing values are pretty reliable. However, no other party stood in all constituencies, meaning there are gaps in the data for working out their regional swing values. The Greens and UKIP both stood in more than half of constituencies, so I’ve worked out regional swing values for them anyway, but my predictor averages that with universal swing. Meanwhile, as the SSP stood in a bare handful of constituencies, for RISE we’re stuck with universal swing.

The Greens, UKIP and even the Tories also have an awkward region each. For the Greens, it’s Central, where they stood no candidates at all. For UKIP and the Tories, Mid Scotland and Fife presents opposing problems – UKIP’s vote share dropped there despite a national increase, whilst the Tory vote went up despite a national decrease. In all cases, this gives a negative value for the relative regional swing, which results in weirdness, so I’ve just inverted the swing for ease. It’s a quick and nasty fix, but, it’ll do!

For clarity, those regional swing figures are in the table below. The oddities are highlighted in red. Remember, these are the values that modify the predicted swing in each region. So a 5% increase for the Tories nationally will only be 1.9% in Glasgow.

Support Spread.PNG

Wait… Can you even use percentages?

I’m sure everyone knows the answer to that question is “Yes”, but just in case – Yes. I once had someone tell me you can’t use percentages to predict Scottish Parliament elections, so I’m covering all bases!

In a given set of data, use of percentages is effectively just a way of rescaling that data so that it all fits in a range between zero and a hundred, to make it easier to understand. Saying “80,760 out of 1,991,051 people voted Green in 2011” isn’t of much help to anyone who doesn’t have incredible skill in mental maths. Describing it as “4.4% of people voted Green in 2011” is much more informative.

So long as it’s referring to the same set of data, the percentage and the actual number are mathematically equivalent. If you have 10,000 votes, you can say 3,000 or 30% – they are exactly the same thing. Although in actual elections the calculations use the number of votes, they could just as easily use the percentages – especially useful for predictions given we don’t know what turnout will be in each constituency and region, but with polling can predict that around X% of people will vote a certain way.

Bear in mind that the swings predicted in each region based on the national result can take the total in a region above 100% – that’s just the nature of polling and predictions.


3 thoughts on “Predictions – How do they work?

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s