Beware of Polls and Forecasting for Gorton & Denton
Full Fact reported last week that
"an advert shared by the Labour Party features energy secretary Ed Miliband claiming that the “real polling” shows that the upcoming by-election in Gorton and Denton is a “fight between Labour and Reform”. It also includes a bar chart, which shows Reform UK on 30%, Labour on 27% and the Green Party on 17%, underneath the claim “only Labour can beat Reform in Gorton & Denton”. But the poll quoted has a very small sample size, with the figures in the bar chart based on responses from just 62 people, and the organisation which conducted the survey has said that while its findings suggest the vote will be close, they should not be analysed beyond that."
YouGov's Anthony Wells, who originally founded Polling Report, told them "while it was possible a company could produce a more comprehensive constituency poll ahead of the by-election, the demographics of Gorton and Denton make polling in the constituency difficult... It is a constituency with distinct communities that are likely to vote and behave in different ways and care about different issues, so it wouldn’t just be a case of getting age, gender and the basics right—it would also be [necessary] to get the different parts of the seat in the right balance and properly represent the ethnic and religious mix of the seat.”
As always in the febrile dynamics of a by-election parties make claims that are hard to support with confidence based on the data available. In the Gorton & Denton by-election both Labour and the Greens are claiming polls show they are the "only party" that can beat Reform in Gorton & Denton. The only poll of Gorton and Denton published since the election was called is a contentious FindOutNow poll, the headline results of which were based on only around 60 responses, while other claims are being made using statistical models from Electoral Calculus and ourselves.
For the record Electoral Calculus give Reform a 61% chance of winning:
Our default model has Labour holding off Reform:
We don't assign probabilities to outcomes, long experience has however told us that any methodology extrapolating national polling to constituencies will be difficult, particularly in a by-election where the political dynamics are even more complicated to rationalise. In short, it is a bit of a fool's errand to try to do. Our algo has no shame and will still try to do it nevertheless. Incidentally I do not believe Electoral Calculus would deliberately skew their projections for partisan reasons. It is just bloody difficult to forecast with a small sample, even if a correctly weighted large sample could be obtained - what is the correct weighting for a single constituency in the circumstances of highly charged politics during a by-election?
Take all forecasts for by-elections with a pinch of salt, I for one am not convinced Labour can hold Gorton & Denton in the current environment when traditional Labour voters know they can punish the government without changing the government - despite what our algo is projecting. The punters are putting their money down the bookies on the Greens as the hot odds on favourites to win. It could be worth in this instance following the money.
