With the nation soon heading to the polls, predictions on the results of this year’s general election are never far away from the headlines. You may have come across a new polling methodology which has become increasingly popular in predicting election results. Multi-level regression and post-stratification modelling (or MRP) has successfully predicted the 2017 UK general election and the 2016 US Presidential election. The same model is being dusted off as both elections loom large – but what is it?
What is MRP?
MRP is a form of modelling which allows pollsters to use large national samples to work out accurate estimates of support for candidates in small geographic areas. In a UK context, ‘large’ samples can range from between 6,000 to 100,000 respondents. MRP models use polling data to work out the relationship between people’s characteristics (e.g. gender, age, past voting record, occupation) and their answer to voting intention questions. Whilst a specific combination of characteristics may be limited to a small number of voters, each individual factor is shared by many. MRP, for example, looks at what women are doing, what people in their sixties are doing, what those who voted Conservative in the last election are doing and so on. The MRP model combines the answers to predict voting intention and seat estimates at a UK wide level. Instead of conducting large polls in every parliamentary constituency, MRP models calculate the probability of a seat changing hands based on patterns found at a national level. These estimates are then applied to all 650 constituencies across the UK.
Who uses it?
MRP models usefully estimate what the range of possible results would be if an election took place on the day of its release. A lot can change in the world of politics and they are by no means claiming to predict the actual result on polling day. Nevertheless, MRP models serve to provide a rough sketch of the UK’s voting intention at a constituency level based on current data. Political parties use MRP models to determine which seat is most vulnerable or, conversely, which ones are up for grabs. Charities, advocacy groups and other organisations can use MRP estimates to tailor their outreach and public affair campaigns to match political trends.
Who carries it out?
Various organisations with the appropriate data use MRP models to estimate voting intention. The Langer Research Associates used the MRP model to successfully estimate voting intention in the 2016 US General Election. In the UK, Lord Ashcroft incorrectly showed a Conservative majority of between 162 to 180 seats in the 2017 – proving that all estimates deserve a degree of caution. YouGov for The Times, however, correctly pointed towards a hung parliament in 2017. Let’s take a look at YouGov’s methodology.
In reference to the 2019 general election, YouGov stated that their MRP poll modelled voting intention ‘based on analysis of key demographics as well as voting behaviour in the 2016 EU referendum, the 2017 general election, and the 2019 European Parliament elections.’ In addition, YouGov conducted approximately 14,000 interviews with registered voters who are shown the parties and candidates running in their constituency. The data was used to assess how voters were making choices across Britain, using information about the voters in each constituency in addition to candidates standing in different seats. With all of this data, the MRP model calculates voting intention and seat estimates.
Different Applications for MRP?
MRP polls are not solely limited to estimating voting intention. The strategic insight agency Opinium used the MRP model to reveal nationwide and cross-party support for the protection of EU-derived workers’ rights. The MRP poll found that 71% of voters supported retaining holiday pay, safe limits on working times and rest breaks. In this case, large national samples were modelled to estimate public support for a specific cause at a constituency level. This support was commented upon by various groups, such as the TUC. Similarly, YouGov used a MRP poll to ascertain the degree to which constituencies were affected by the cost of living crisis. Using a combination of UK wide variables, YouGov estimated whether a constituency was: 1) calm and comfortable; 2) unsettled withstanders; 3) squeezed but coping; 4) cautiously hopeful strugglers; 5) worried and suffering.
How can MRP data be mapped?
The estimates produced by MRP models can be visualised on maps to show noticeable and surprising trends. YouGov’s cost of living MRP showed clear clusters of ‘worried and suffering’ constituencies in the North West. Conversely, the vast majority of Scottish constituencies were categorised as ‘squeezed but coping.’ By visualising the data in map form, regional variations become increasingly apparent. The same can be said in relation to voting intention maps at a constituency level, whereby infamous red and blue ‘walls’ are seen from a birds eye view. Importantly, a per constituency based approach allows us to zoom in even further, granting more granular information to analysts and political pundits.
We can expect countless more MRP polls in the coming year. Only time will tell whether they estimate the national picture correctly. Regardless of the result, MRP polls provide a snapshot into the public psyche and will remain an invaluable asset as the electorate decide on the country’s next parliamentary representatives.