For completeness, with each forecast we have included links to the source data used; we do not support nor do we condone betting. We recognise that some people have gambling addictions. If this is you then please click here to seek help.
Our model is based on the individual constituency odds published by various betting companies, as quoted alongside each prediction. We have made the assumption throughout that the bookies are not gambling and whatever the result will make the same profit margin on the total value wagered by the public. This should mean that the odds are proportional and representative of the true probability of each candidate getting elected, however it is known that human psychology favours the underdog making the odds slightly too high for the least favourites and not high enough for the favourites. The impact of this was studied by Matthew Wall et. al. at the 2010 General Election and it was found that the relationship between the implied probability from the odds and the actual probability of getting elected can be approximated to that shown below.
We thank Chris Hanretty (www.electionforecast.co.uk) for introducing us to Matthew's paper, and the above relationship was used to remove the bias towards to underdog for forecasts based on odds downloaded on 5th March 2015 onwards.
Using the probabilities within each single constituency and a random number generator, we select a candidate to win each seat; modelling a single possible result of the forthcoming election. We do this 100,000 times; a process called Monte-Carlo modelling and consider the number of times each party wins each possible number of seats.
Why not just use the most seats betting odds?
We cannot simply use the odds for the overall winner because of the assumption that the bookies are running a business and not gambling. That means that they will make the same profit margin while paying out the same amount of money in winnings regardless of the result. Therefore the bookies are taking account of public opinion representative more of a proportional representation electoral system rather than the first past the post electoral system.
Potential bias with our approach
Since our model is based purely on betting odds, we are assuming that the general public are putting money down at the bookies in the same proportions as they will vote for at the General Election within each constituency.
It is assumed that sufficient people are betting that the odds are representative of true public opinion. If very few people are betting then the odds will be representative of that very small group and swayed towards those individuals’ opinions rather than the general public as a whole. Therefore as we get closer to the election and greater quantities of money are placed on the result, the forecast should get stronger and more accurate.
It is possible that people may bet on who they think may win, for financial gain, rather than for who they support and will actually vote for. If sufficient numbers of people do this, and in an unbalanced fashion towards a single party, then the forecast will have a bias towards that party.
It is also possible that some party’s may attract greater numbers of supporters willing to bet, and/or with greater quantities of money. If this happens then those party’s will show a better result in our forecast than they would in the General Election.
Our approach compared with others
There are many different approaches to forecasting the election, each with their merits and drawbacks. None of them can be sure of the result; that is down to the British public on 7 May 2015.
Most forecasts use previous election results for each constituency and apply a swing to those results based on opinion polls. This means there is significant inertia in their forecast due to the relatively stable historical dual-party politics of the UK. Some authors of these other forecasting methods have commented they believe they are under forecasting the smaller parties that have experienced great growth over recent years due to this inertia and little electoral success in previous General Elections. In contrast, our forecast is based purely on this election.
Other forecasts we are aware of include:
We will align the terminology used in our analysis with that used by . The definitions are below and based on “Quantitative meanings of verbal probability expressions” by Reagan, Mosteller and Youtz.
0-10% Very unlikely
26-40% moderately unlikely
75-89% very likely
90-100% almost certain