Electoral maths: data and the UK general election

Last week’s general election made me think about what was scrutinised more closely in the run-up to the vote: the parties’ policies or the opinion polls?

While it should be the policies that carry greater weight and get the most attention, the reality is opinion polls attract as much, if not more, analysis.

Politicians balance what they’re saying against how those statements seem to be affecting their parties’ chances of success. And commentators temper their reactions to policy announcements based on voters’ preferences.

First-past-the-post or proportional representation?

In the UK, this analysis is complicated by our use of a first-past-the-post (FPTP) system whereby the party with a majority of seats forms the next government. The smaller parties, meanwhile, would prefer to see proportional representation (PR) introduced, so the share of seats a party wins matches the share of votes it receives.

And this complication influences voters too.

Because there’s no formal use of PR in the UK, individual voters may want to vote tactically based on who is more likely to win in their particular seat.

And to do this, voters need indications of how others are going to vote. And so, for them too, opinion polls aren’t just a matter of background interest about who’s in the lead and who’s lagging behind, but a critical factor in deciding how to vote.

All of which means opinion polls, and especially their accuracy, play a key part in UK elections.

And just as the politicians get to find out their fate in the hours after the voting closes, the opinion polls may also come under the spotlight.

How accurate are the opinion polls?

Over the years, the opinion polls have become more accurate, remarkably so given both the varied and changing views of those sampled, and the small sample sizes on which the polls are based.

At first sight, you may think that asking as many people as possible for their opinions would be a way to increase the accuracy of the prediction. However, the flaw here is that if the opinions of the people surveyed are skewed in a particular direction, then the size of the sample is almost irrelevant (I use the word ‘almost’ here, because if the sample is really large, then the skew may not matter).

But what’s really remarkable is just how small the sample size can be. Even a very small number of carefully selected people can accurately reflect the views of a much larger number. For example, a sample as low as 1,000 people can offer a margin of error of plus or minus 3% at a confidence level of 95%.

However, even if a small sample of people can produce accurate results, it’s also important to understand the nature of those results. Some polls will predict who’s going to win a particular seat, some will forecast the number of seats, while others will predict percentages of votes. These results are obviously connected, but the context of the questions voters are asked will determine how their opinions should be used. This context covers not only the voters’ characteristics (age, location, job, etc.), but also the questions asked. For example, “Who do you think will win the election?” will produce different results from, “How will you vote?”

Asking the right people the right questions

In short, accuracy is based on asking the right people specific questions, not by asking a lot of people a variety of questions. And, conversely, if you ask the right people, you don’t have to ask a lot of people – so the task is also easier and quicker.

In this respect, the ways in which opinion polls work closely mirrors the ways in which organisations operate, as well as how voters make their choices. A small number of critical operational metrics can be used to monitor the performance of a wide range of organisational activities. Here too, using the right information is more important than using lots of data.

But that doesn’t mean there’s no need for historical data. Just as on election day prediction performance can be judged against actual votes, operational metrics can be compared with organisational performance periodically. And when discrepancies arise, diverse (big) data is still needed to discover new and unexpected patterns in operations.

But like those low-sample opinion polls, it is possible to use a small, cohesive set of operational metrics to monitor and forecast operational performance.

Help with organising and managing your operational information

At Quantiv, our NumberWorks method makes identifying and defining your operational metrics easy, while our NumberCloud platform includes robust mechanisms to collect and expose contextual information about significant organisational activities and events. In fact, it could even be used to collect opinion poll data!

To find out more about how our services could help you, contact our team today on 0161 927 4000 or email: info@quantiv.com