Technical reports

How did we create these six voter types?

Based on just 12 questions, we have generated a powerful grouping that allows us to engage with six more concrete and more dimensional voter types.
A person working at their desk

Summary

Our classification of British voters into six types was created using machine learning. The method takes people’s answers across a series of questions and looks for groups of voters with similar views. The foundation for our work was our NatCen's left-right and libertarian-authoritarian scales, introduced by our predecessors in 1986.  

But we didn’t stop there. A broad range of more recent divides were explored to assess whether they could simply be explained using the scales, or if other underlying divides in our outlook were at play. We put a series of hypotheses to the test and came up with a very large number of possible clusters of voters. We then assessed how strong these different cluster solutions were, and how much sense they made to us conceptually.  

Ultimately, we settled on a classification that we found to be both robust and useful. Each of the voter types was notably distinct from the others, the classification does a good job of predicting views on a range of current issues, and each of the types represent enough of the electorate to be politically important and analysable.  

Political opinions are complicated. Some issues are ever-present—the distribution of wealth and resources, the power of the state to control how we live—whereas others are specific to a moment in time—Brexit, small boats. But ways of measuring people’s views and opinions have emerged that have proved consistently useful. With Dividing Lines, we build on our NatCen's legacy work on values scales. We introduce something that’s both more dimensional, and hopefully more concrete to engage with: twelve questions, six voter types.  

What are values and how do you measure them?

Values are the underlying beliefs or principles that can help explain people’s views on individual issues. Once we get survey questions on values right, it can be very powerful. A small set of questions can help you understand an awful lot about a person’s views on more specific issues, either by contextualising them or by predicting them. One approach is to measure whether people’s economic views are broadly left wing or right wing. Another is to measure whether people prioritise individual freedom and autonomy, or social order and deference to authority.  

The British Social Attitudes (BSA) survey uses a series of questions to measure both of these. The former is known as the economic left-right scale, whereas the latter is known as the libertarian-authoritarian scale. The left-right scale consists of five questions, while the libertarian-authoritarian scale consists of six, and in both cases we use people’s answers to the questions to assign them an overall position on the scales.  

Both these values scales were developed by our organisation for our British Social Attitudes survey, and we’ve been tracking how the public’s position on them has shifted since they launched back in 1986. These scales are powerful. In fact, we found that they performed better in our analysis of the British electorate today than we had initially expected.  

So what are the dividing lines in British politics?

Britain has evolved over the decades, as have our dividing lines. Among other things, Britain has become more multicultural. Part of people’s underlying feeling about this can be explained by their position on the libertarian–authoritarian scale, but it only takes you so far. To understand how different groups experience politics today, we also need to reflect on how close different people feel to what’s going on in Westminster; a divide that currently appears to be widening.

We wanted to create a way to simply and robustly explain where the British electorate is at. The scales capture relatively abstract ideas, and it can be a challenge to communicate insights to a general audience, who often find it easier to think in more concrete terms. Looking at the scales in isolation can also hide interesting and important interactions between them. Given how well the two scales still explain views on individual issues to this day, we incorporated most of these questions as questions that drive our voter classification.  

Getting technical – what did we actually do with the data then?

We generated hypotheses of combinations of survey questions that would identify dividing lines in public opinion. To get from theory to a clear voter classification, we used an analysis method that takes people’s answers across a series of questions and looks for groups of voters with similar views. More specifically, we used different extensions of the k-medoids clustering algorithm, a machine learning algorithm used to identify distinct groups in multidimensional data. We explored over 7,000 survey question combinations, and more than 30,000 ways to group the British public. We considered letting a larger number of survey questions drive the grouping, and then explored whether we could remove some and still achieve an equally powerful outcome.  

Ultimately, we settled on a set of six voter groups that was both statistically robust and conceptually useful, meaning that each of the groups was notably distinct from the others, and was large enough to be politically important and analysable. We were also able to use look at distinct subgroups within each of our six main groups.

The decisions made at each point in the process are documented. A brief methods paper outlining our approach will be available soon.