The problem with polling averages
Welcome to the 208th edition of The Week in Polls (TWIP), which takes a look at the problem with averaging different polls and why I tend to avoid doing so.
This is followed by a summary of the latest national voting intention polls, seat projections from MRPs, and the most recent party leader ratings.
Subscribers who pay for the full edition can also read ten insights from the latest polling and analysis, including:
Shifting public views over the military conflict with Iran;
A bundle of polling ahead of the Scottish Parliament and Welsh Senedd elections; and
How AI chatbots may be better than social media.
If you are not yet a paid subscriber, you can read all ten by starting a free trial now:
For more polling news as it comes in before next week’s edition, you can also follow The Week in Polls on Bluesky.
A quick update on the raised eyebrows last time at The Independent over its reporting of an opinion poll. A further week on, the paper has neither replied to my suggestion that it correct the story nor has it changed the story.
A reminder too that my page of local council election predictions is continuing to be updated, with new figures added this week.
One other preliminary note: I have switched to using the online Keeper password manager, which makes it easy to create secure, unique passwords for all your logins, while also making them easy to use. You can get a free trial and then 30% off your first year with this affiliate link.
And with that, on with the show.
Why I only calculate provocatively crude polling averages
This is an updated version of a piece I first ran in TWIP #101 back in 2024. With voting intention figures often varying significantly between pollsters, now is a good time to revisit the problems with averaging them together.
I like averages.
I use them, I calculate them and I can bore you with the pros and cons of mean versus median versus mode.
Averages sound like they should be useful for understanding opinion polls. After all, Understanding Opinion Polls 101 is ‘don’t be dazzled by just one poll, look at the trends across multiple polls instead’. That sounds like a job averages can help with.
Averages also sound useful if you understand a bit about the random vagaries of opinion poll sampling, about how a poll can be unlucky in having a sample skewed a bit one way and another poll could have a sample tilted the other way. Even though very few polls these days rely on genuinely random sampling (with quotas and weighting instead playing a bigger role), averaging out different polls sounds like a good way to filter out such variations and see the underlying reality.
Plus of course averages of the latest opinion polls are a staple of coverage of polling. Whether it’s the mainstream media or high-profile social media accounts dedicated to polling, it’s common to see averages.
And yet…
You may have noticed that in both my weekly newsletter, The Week in Polls, and in my regularly updated web page of latest polls, I avoid giving averages.
What’s more, although there are monthly averages in my PollBase database of voting intention polls all the way back to 1939, those averages are almost provocatively crude. A simple averaging of all the polls in each month whose fieldwork dates commenced in that month. No adjustments for even basics such as how many polls came out from different companies. Those averages are unashamedly rudimentary.
Moreover, when I do my pieces about leadership ratings (such as on Farage and Polanski), I prefer complicated graphs which show each pollster separately to simple graphs with fewer lines, just showing the averages.
That’s because the wise approach to polling averages is to treat them with a deep ambivalence.
Averaging is sometimes a good way to head towards the truth. Yet on other occasions, it obscures the truth and is purpose-designed to give the wrong answer.
Imagine you are the first time visitor to a building you’ve not been to before and you’re sat in reception. The person on reception is busy at the moment and you need to head to the toilet. You ask a couple of other people sitting there if they know where the toilets are. One says, “I think they’re down that corridor and then through the door on the left”. The other says, “Yeah, down that corridor but actually, it’s the door on the right”.
A true believer in averages would average out one left recommendation and one right recommendation, then stride confidently along the corridor and into the wall at the end instead. Neither left nor right but average. And wrong.
Sometimes you have to choose, not average.
So it can be with polling.
Take the current mix of pollsters and methodologies that we have. There are consistent differences between pollsters, as is often the case. Back when I first wrote this piece, the obvious example was Opinium with its distinctive way of treating people who say ‘don’t know’ consistently giving lower Labour leads than other pollsters reported. (Various pollster methodologies have changed since then.) In 2026, the obvious example is YouGov with its lower ratings for Reform.
Now, you might think YouGov is right, or that they’re wrong.1 If they’re right, then averaging their figures in with those from other wrong pollsters just drags you away from the truth. And if they’re wrong, polluting the average of those other pollsters with YouGov’s figures makes the average worse.
The point is you have to choose. Averaging just guarantees being wrong.2
So should you swear off all averaging?
Not quite.
Averaging does also guard against getting overly excited or depressed by any single poll. (At least in theory. Given the widespread presence of both over-excitement/depression and of polling averages, it’s a good theory that does not hold up particularly well in practice.)
Averaging can give you a simple broad brush picture, which is why I take that deliberately simplistic approach in PollBase.
The problems with averaging, however, are why a table of the latest polls from each pollster plus some super simple averages (or better yet, a graph) is generally my preferred analytical view of the data. Averages obscure, and tables of data reveal, the consistent differences between pollsters which indicate the issues to think about.
Averages are like water wings, useful for getting from incompetence to basic competence in understanding the polls, but then rather annoying encumbrances which get in the way when you’re moving from basic competence to expertise.
But before my colleagues Jack and Luke get too cross with me, I should add that there’s also a role for super-smart averaging, of the sort they used to produce PollBase Pro a few years back.
This sort of analysis brings many advantages, such as making use of the points of truth - general election results - and factoring in expected levels of random noise. It does though still have as a weak spot about that YouGov choice as it ducks the choice. If YouGov is the outlier and is also right - as some previous outliers have been in other Parliaments - then even PollBase Pro’s cleverness will not be clever enough.
Which is why the best solution is to keep an eye on that table of latest polls, keep an eye on graphs for trends and, yes, graze lightly on averages. But only lightly.
Elsewhere from me…
There are many examples of secondary legislation moving slowly. What, though, is the bigger picture tying together the nerdy examples I often write about and share?
If you spot a factual error…
Borrowing an idea from Stuart Ritchie and others, if you spot a factual error in an edition of this newsletter, I will give you a free 12-month subscription to the paid-for version (or an extra 12 months on your existing one). Factual errors qualify rather than grammatical errors or disagreements over interpretation, with my decision being final. Messages about both of those are always very welcome, though.
Voting intention and leadership ratings
Here are the latest voting intention polls from different pollsters. Once again, the figures show that Find Out Now’s days as an outlier on Reform support are (temporarily?) over, and its figures for the party are in line with the pack of pollsters.
The table is also online here, along with explanations about what is included. It is updated regularly through the week with new polls.
Next, the seat projections from MRPs and similar models, also sorted by fieldwork dates. After quite a gap we have a new one - from More in Common for The Sunday Times.
Finally, a summary of the latest leadership ratings, sorted by pollster name. For more on the trends behind these, see last week’s edition.
The table is also online here with additional details and is updated regularly through the week as new data comes out.
For the historic figures for both voting intention and leader approval, including Parliamentary by-election polls, see PollBase.
Catch up on the previous two editions
My privacy policy and other legal information are available here. Links to purchase books or other items are usually affiliate links which pay a commission for each sale.
Quotes from social media messages are occasionally lightly edited for punctuation and clarity.
If you are subscribed to other email lists of mine, please note that unsubscribing from this one will not automatically remove you from the others. If you wish to be removed from all my lists, simply reply to this email to let me know.
Public opinion shifts over military conflict with Iran, and 9 other polling insights you shouldn’t miss
The following 10 findings from the most recent polls and analysis are only for subscribers who pay for the full edition, but you can sign up for a free trial to read them straight away.
Ipsos has new polling out showing changing public attitudes towards the military conflict with Iran.
