Dr Patrick English is Director of Political Analytics at YouGov, and their spokesman on political research.
Andy Burnham’s victory in Makerfield last night will likely set off a chain of events with potentially huge consequences for British politics today, at the next general election, and potentially long beyond.
Almost in anticipation of these seismic consequences, the 2026 Makerfield by-election became one of the most polled single-constituency races ever. Four polling organisations published five polls over three weeks to give us insight into how the race was going.
These constituency polls had a pretty wide range in terms of the margin, but all agreed on one thing – that Makerfield would be a Labour victory. That should be seen as a success for constituency polling, which is notoriously difficult to do. Well done to all involved.
The scale of the projected win according to the polls was anything from 12 percentage points (Convergent) down to five (More in Common, Opinium). That figure of five would have included a Reform win in its probable results intervals given the sorts of margins of error we can expect from small sample sizes and the particular difficulties of constituency polling.
As it was, Burnham went on to win by 20 percentage points in a thumping victory. Adding over 6,500 votes to the total Labour achieved in 2024, and almost 17,000 more than Labour managed at the local elections just six weeks ago.
But why do we see so few constituency polls compared to national polls? And what makes them so difficult as to have produced a miss on Labour’s winning margin?
Firstly, constituency polls cannot realistically be produced using online panel data alone. Online panel data is the cheapest and fastest method we have for political polling, but even with a panel the size of YouGov’s with our hundreds of thousands of sign-ups, we cannot recruit enough people from one specific constituency to fulfil (what ought to be) a minimum acceptable standard sample size for published constituency polls of around 500 people (or more).
Which means anyone wishing to produce them must recruit respondents via other methods such as face to face or phone polling. Both far more expensive and far more time consuming than online polling. This precludes a lot of pollsters from even attempting it.
That said, there will undoubtedly have been constituency polls of Makerfield conducted which the public will never see. Those done by parties, campaigning groups, or those with financial interest in trying to anticipate the result before it comes in. Each will have their own interests and agendas regarding any decision to publish or withhold a constituency poll commissioned by them.
The existence of these private polls creates an additional blocker on how much information we get from pollsters about constituency by-elections – even those polling companies who can and do produce constituency polls, can only do so many. If their capacities are taken up by private polling, it naturally reduces the amount which becomes public.
What we can be sure of is that every pollster putting out these polls is trying their absolute best to correctly capture and report vote intention as they see it, at the time that they see it. All good pollsters have no interest in trying to misrepresent the results or push up or down certain parties or candidates.
At the end of day, our credibility is on the line each and every time we produce these polls. Our numbers are judged against real results, and our reputation rests on our figures being as close as possible to those. It’s in our interests – as individual pollsters and collectively as a polling industry – to get it right.
In this instance, the polls did largely get it right – they each anticipated that Andy Burnham would win, and two (Convergent and Survation) reported double-digit victories. There will be some interesting post-mortem work to be done on what might have caused the polls to understate Burnham’s victory.
One such potential source of error was in fact pointed out by pollsters themselves before election day – a problem known as “false recall”, whereby respondents give incorrect information as to who they voted for at the last election, might have caused problems in making sure the samples were balanced and representative of Labour’s 2024 vote.
The worry, expressed by Damian Lyons-Lowe of Survation on Tuesday, was that many people might incorrectly recall having voted Labour at the General Election in 2024, instead thinking back to the Mayoral vote in that same year in which Andy Burnham won by a much larger margin than that which Labour won Makerfield later that year. Correct past vote information is needed to ‘weight’ the samples collected in polling to make them look closer to the actual voting population (which never happens by chance, for a number of reasons). If that was the case, then, Damian outlined, it could have meant that the measured 2026 Labour vote share could have been suppressed (as past Labour voters were being ‘weighted down’) while Reform UK’s was inflated (as their past voters were being ‘weighted up’ to make up the difference).
False recall and its consequences are just one of many methodological, measurement, and modelling barriers and puzzles which pollsters must overcome to produce their figures.
It is unlikely that we will see a by-election of such wider (potential) consequence again for a very long time. And the people of Makerfield will probably tell you that’s a big relief; at times it felt like the entirety of the nation’s media, political activists, politicians, and indeed pollsters all descended upon the ~100,000 people of Makerfield for six weeks straight, asking them just one question again and again – “who will you vote for in the upcoming by-election?”. There were more than a few stories of some local residents getting rather tired of it all by the end.
Nonetheless, they answered rather emphatically, when the time came.