As the post GE15 polling debate continues SPIN’s Aidan Nutbrown asks “Are Elections Random?”

As the post GE15 polling debate continues SPIN’s Aidan Nutbrown asks “Are Elections Random?”

The outcome showed why punters should expect the unexpected

Albert Einstein famously said: “God doesn’t play dice”. He was wrong. Everything has a random element to it. I aim to board the 8.08 train each morning but sometimes for unforeseen reasons I miss it. And sometimes for unforeseen reasons it doesn’t appear. This uncertainty is precisely why betting is fun, and why bookies exist.

It is a given that sports events have large degrees of uncertainty to them. A horse can clip a fence and fall, a batsman can misjudge an awkward delivery or an underdog, with a dollop of good fortune can fell a goliath (note Leicester City’s 5-3 victory over Man Utd, September 2014 as just one example).

However there appears to be a widely held belief that Elections don’t follow this pattern – that they are entirely predictable. That with correct methodology expert pollsters should be able to tell us precisely how many seats each party will win. This is a false assumption. Even if their methodology is correct it supposes that nothing happens between the poll and the election itself. That people don’t change their minds, or decide not to vote. Polls are a great guide to the state of play at that moment in time – a snapshot – but as forecasts they are limited.

    One reason is that the media demand dumbed-down answers – headline numbers about exactly how many seats will be won, by whom. In fact forecasters should be providing probability distributions. If we understand that there’s a 10% chance of rain today why can’t we understand that there’s a 6% chance of a Tory majority?

At Sporting Index we like to use a football analogy to explain how the polls are our best guide as to how to calculate and align our forecasts. Polls are like having a dataset of matches that tell us to expect, on average, 2.7 goals in a Premier League match. But any forecast model needs to allow for the uncertainty that may follow. This uncertainty may take many forms – the weather affecting turnout, shy Tories, lazy left-wingers as recently uncovered by Prof LVW, Jungian crowd behaviour or a myriad of other factors we may not have even thought of yet. The levels of this uncertainty may reduce as the date of the election nears, or they may not – news might break that swings the public one way or the other. In short the probability distribution of our forecast has to account for a considerable degree of unpredictability. It has to be wider and flatter – in other words more disperse.

In football we frequently see 0-0 results, or as above (but more rarely) eight-goal bonanzas. Describing how to model football scores or seats distributions is beyond the scope of this post, but it is worth showing Sporting Index’s distribution models for Con & Lab seats that were created in March 2015 and that are aligned to the polls of the time, giving expected seats of Con 284 and Lab 262. By varying the degrees of uncertainty, the chances of Con getting most seats and a majority can be seen to change. In the first graph the chance of a majority is 0%, in the second it is about 10% and clearly describes a far more complex situation than the former, which is much more akin to the straightforward “forecasts” we were provided with in the media in the run up to May 7th.


By allowing elections randomness, and crucially by describing the possible outcomes as a range of chances, it is perfectly feasible that the pollsters in the final days were in fact not necessarily wrong. Rather that they provided an expectation and the actual outcome based on this expectation was improbable but not impossible – much like Leicester City beating Man Utd in an eight-goal thriller.

One thing that the 2015 Election forecasters did get wrong though was to tell us that there was 0% chance of a majority. This bold statement was due to insufficient modelling and a failure to respect the market view, in which a Tory majority was trading at around the 5-10% mark for much of April and May.

Another pointer to this underlying randomness is the relative success of the Exit Poll. Yes it has a much bigger basis, but the clue really is in the name. The Exit Poll is after the fact. The turnout ratios are perfect. There is no time left for uncertainty. The only variation from the result emanates from the limitations in the numbers and locations sampled and having to extrapolate nationwide.

There is another big live televised event that many view as predictable and not subject to uncertainty, and the 2015 running is only days away. It is Eurovision. Like an Election it follows a basic predictable framework – that the Balkan states will vote for their friends and that the UK will do atrociously as we don’t ‘get’ Eurovision. But that view is limited and doesn’t allow any room for chance to play its part. The UK can win it – note Katrina & The Waves in 1997 – and for three years in a row from 2001 to 2003 the winners were 20-1 underdogs!

We should learn to embrace the chaos and expect the unexpected. After all this is why it is fun to bet.

Aidan Nutbrown is Sporting Index’s political trading supremo

Comments are closed.