With no clear leader in sight, the outcome of the UK's general election is hard to model.
On Thursday, the UK goes to the polls for the 2015 General Election, and it couldn't be a much closer—or more confusing—race. And that makes predicting the outcome a difficult task.
The problem is that in this election, the two main parties, the Conservatives and Labour, are pretty much level pegging. There are 650 seats in the House of Commons and theoretically a party needs to win more than half to have a majority. But that doesn't look like it's going to happen.
"How do you make sense of the most unpredictable election ever?" reads the introduction to one forecasting site, May2015.
As it's very unlikely that either party will win a majority on its own, we're probably looking at another coalition situation, where parties join together after the vote to form a workable government. So for those trying to make predictions about the actual government we'll end up with, it's no longer as simple as picking a winner.
James Kanagasooriam works with political analytics at Populus, a polling and research firm that has been running weekly simulations through its proprietary computer model, with results reported in the Financial Times. He explained that more simple ways of modelling elections, which have been used in the past, wouldn't cut it this time around.
"Over the last 20 years, people's way of forecasting any election—regional, national, it doesn't matter which country—would be traditionally using uniform swing," he said.
'Uniform swing' basically looks at the average movement between two main parties since the last election and applies that change to each constituency.
In the past, it was much more of a two-party system in the UK, which meant votes for Labour and the Conservatives were effectively inversely correlated. But that's not so much the case now. Instead of "swing", we have "churn" among the parties. For instance, Labour could gain votes from the Liberal Democrats but lose votes to the Greens, as well as losing and gaining supporters from the Conservatives.
"You're able to understand the almost-infinite permutations there could be on election night."
Kanagasooriam said simulations are a good way to try to capture these dynamics. Populus uses a Monte Carlo-type model, which takes in the changing data regarding support for the parties and runs simulations for hundreds of different scenarios.
"What you get is, you're able to understand the almost-infinite permutations there could be on election night," said Kanagasooriam. He's referring to the many different coalitions that could arise as parties battle it out to work together.
Simulations are a good approach, Kanagasooriam added, because these things are "predictably unpredictable"; you might not be able to predict each seat but you know that the overall outcome is going to fall somewhere on a distribution curve. Tracking the probability of different outcomes is better than taking one set of figures and trying to come up with a static solution.
In their last 'Populus Predictor' report, published last week, Managing Editor Rick Nye wrote that, "In only one out of every 200 simulations run do either Labour or the Conservatives win enough seats to govern by themselves."
There are other models out there that suggest probabilities for different groupings after the election, and they don't all offer the same results. Scenarios suggested by Electionsetc, run by Oxford University Politics Fellow Steve Fisher, for instance, paint a different picture. But Electionsetc also puts the chances of a hung parliament at 91 percent.
Other forecasts offer a similar conclusion on that front. May2015's model uses a two-step approach by first figuring out who will win each seat according to national polling information, and then separately factoring in results from polls conducted by Lord Ashcroft, who polls the most marginal seats (Ashcroft is a Conservative, but his polling work is widely considered reliable). The site's latest results show a hung parliament with a slight Conservative lead.
One group of academics has even turned to Twitter-harvesting algorithms to try to make sense of public opinion. The University of Warwick team combines insights gleaned from political tweets with more conventional polling data in its forecasting model. Guess what? Looks like the two main parties are pretty neck-and-neck.
Even in predictive models, some things come out looking pretty certain.