FAQ: The Number Cruncher Politics EU referendum forecast

See also the main NCP EU referendum page.

What is the Number Cruncher Politics EU referendum forecast?

The forecast is a model of the likelihood of outcomes in Britain’s EU membership referendum, that combines polling with analysis of the likely accuracy of polls and data from the past referendums to project a probability distribution of outcomes.

There are three main steps in the process. First, aggregate the polls, as described in the polling average FAQ. Second, estimate current public opinion based on the polling average and analysis of likely polling accuracy (the nowcast).

The third step is to model the way public opinion is likely to evolve as 23rd June approaches. This is challenging because unlike elections held every few years, referendums are rare and usually one-off events. The scientific approach is to look at the patterns in historical data, while considering which historical precedents are most applicable to the current forecast.

How can one outcome be much more likely when the polls say it’s close?

Extrapolating polls forward in a straight line is a common “back-of-the-envelope” approach, which is simple and avoids the difficultly (and sometimes controversy) in estimating the likely degree and direction of poll movement between the forecast date and people actually casting their ballots.

But it is also the wrong approach, because polls are not unbiased predictor of the future – they are a measurement at time zero. Referendums tend to have a status quo bias and more often than not, the “change” option tends to lose ground as polling day approaches.

Why should I pay any attention to this when all the forecasts were wrong last year?

NCP publicly predicted the polling failure the day before. But more importantly, this model addresses a number of the problems that befell most forecasters last year. It includes analysis of likely polling bias (based on current and historical analysis) at its heart. And while referendums are typically less predictable than elections, they have one characteristic that makes our lives a bit easier – they are a simple national popular vote, so translating vote shares into seats is a non-issue.

The error band of the distribution is huge – what’s the point in a forecast with so much uncertainty?

The point of the forecast is to assign probabilities to outcomes based on analysis rather than guesswork. Uncertainty in forecasts is a feature, not a bug – one of the problems that forecasts had at the 2015 general election was not incorporating enough uncertainty around the outcome, particularly (though not exclusively) in relation to the accuracy of polls. In other words, the forecasts were more confident that seemingly unlikely outcomes would not occur than they should have been.

Experience tells us that referendums tend to see considerable volatility in public opinion. They also tend to have a status quo bias and more often than not, the “change” option tends to lose ground as polling day approaches.

See also the main NCP EU referendum page.





About The Author

Related Posts

  • Bill

    I am curious to read more on the “status quo bias” that exists in referendums – do you have a good place to point me to in reference to this by any chance? thanks very much!