Electoral Calculus and pollster Find Out Now ran a couple of predictive polls in the run-up to the local elections in May 2021. How accurate were our predictions?
For the West Midlands Mayor election we ran a poll at the start of April (details). It predicted that Andy Street would win the mayoralty after a two-round contest. We predicted a 7pc lead for Street in the first round (it was actually 9pc) and he would beat Liam Byrne (Labour) in the second round by 52-48. The actual result was 54-46, which is within the margin of error.
Our prediction was not only the first public prediction of this contest, but was also more accurate than other predictions of the final outcome (source: Wikipedia).
That is a very satisfactory performance for this poll.
There were elections to 123 district and unitary councils on 6 May. For these election we ran a larger-scale MRP poll, which interviewed over 11,500 respondents. The MRP method is often a bit more accurate than traditional polling approaches, and also allows estimates of public opinion in small geographic areas, such as district councils and council wards. More details of the MRP approach are available here.
The MRP poll allowed us to estimate the result in every council ward which was up for election. By adding up the wards won, it also predicts which party would win each council, or whether there was 'no overall control' (NOC) of the council.
The final predictions, including the council-by-council predictions, are visible on our website at Districts Poll April 2021.
The poll predicted that the Conservatives would gain 12 council, with Labour making small gains, and the Liberal Democrats gaining one council. That was broadly correct, though the Conservatives did a bit better than expected and Labour did noticeably worse.
The following table shows the number of councils predicted and actually won by each party:
Of the 123 councils, the vast majority (103) were correctly predicted. Highlights include:
It's not possible to correctly predict every council, as some are very finely balanced, and others are driven by specific local factors which are hard to capture even in regional MRP polling. Mispredictions include:
This suggests that the Labour vote share had been slightly overestimated by the poll, or else there was anti-Labour tactical voting. It will need further analysis of the detailed results to identify the precise causes and their relative importance.
The impression is that voters were less inclined to vote Labour. In straight Conservative/Labour contests, Labour votes switched to the Conservatives. In other non-Conservative areas, such as Bristol and Sheffield, voters switched from Labour to other left-wing parties such as the Greens and Lib Dems.
Overall the prediction worked well, and was the first (and only) published prediction showing that the Conservatives were on course to do better than Labour in the district council elections. The MRP method was able to give mostly accurate predictions of over 120 individual councils with a moderate sample size, and a quick delivery time of only a few days of fieldwork and analysis.
Electoral Calculus and pollster Find Out Now successfully predicted the results of the West Midlands Mayor election, and the local elections for district and unitary councils in England.
Using modern polling techniques, we were able to give detailed geographic predictions quickly at low cost.