Political betting is not normally in the same league as sports betting, but the 2014 Scottish Independence Referendum was Paddy Power’s biggest ever non-sports event. The liquidity was impressive enough to create a strong betting line.
There was, however, a clear disparity between the betting line and the forecasting determined by pollsters.
The difference was money.
The bookmaker was accepting huge sums for No to independence. All of the chunkier bets were being wagered on that side. The Yes betting was not insubstantial (with a slightly higher Yes percentage on average from Scottish punters), but the weight and power of the cash for No forced us to keep it odds on all the way. ‘No’ was only ever priced as generously as 4/9 (representing about a 67% chance), occurring during the two week run up to the vote.
Who was going to end up looking sheepish by the time the votes were in and counted? The bookies, or the pollsters?
The betting line was not a representation of answers people gave to questions about voting intention or past political preferences.
It was a line of opinion created by investments. Punters put their money where their mouth was and gave a different story than was given to the pollsters.
There are a number of possible reasons why the pollsters struggled to get to grips with Scotland 2014.
- 16 & 17-year-olds were permitted to vote. This age group had never voted in an election before. It was not known how many would come out and vote, or how they would vote.
- There was a huge uptake in registration of those from every age group that had never voted before, and therefore had never been polled before.
- The older vote is normally critical because they tend to vote in disproportionately high numbers – everyone knows this, it’s why politicians bow down to older voters. Pollsters have relied on this constant and it is built in to normal polling rules and methods – but those rules went out of the window for this referendum.
- The pattern of behaviour was different, meaning that it was difficult for any classic polling model to deal with.
Not a strictly party political issue
- Pollsters normally use information on previous voting records to make predictions about elections – this was not as useful as usual in this referendum.
- Ahead of the vote, it was estimated that 40% of those that voted Labour in the 2010 general election were going to vote YES in the referendum, for all kinds of different reasons – the pollsters’ models were not designed to take that kind of differentiation into account.
- By the beginning of September, the voter registration was high and the turnout was expected to exceed 80% (the 1995 Quebec independence referendum turnout was 93.5%).
- The Scottish turnout for the 2010 general election was 64%, and the turnout for the 2011 Scottish elections was 50%.
- Pollsters are not experienced at dealing with such high turnout.
- A host of people who normally just didn’t vote were about to contribute to the independence referendum result – it was not a normal election, and the normal predictive methods clearly struggled to make sense of the results they were receiving.
- The turnout issue made it difficult for pollsters to get valuable information out of their sample, because the shape of the vote was considerably altered.
So, what can be surmised is that the tried and tested models for prediction fell somewhat short. And the old adage rang true – follow the money.