Talk:Opinion polling for the next United Kingdom general election


Local Regression Graphical Presentation Not Providing An Accurate Up To Date Representation of the Current Polls

The graph currently shows a notable down tick in the polling fortunes of Reform to the benefit of both Labour and the Tories whereas recent polls over the past two weeks indicate the complete opposite and whilst what is shown in the graph might have been true 2 weeks ago it is now no longer the case. For example the latest 'Find Out Now' poll (which came closest to predicting the national projection at the time of the May elections and therefore is arguably the current' gold standard') has just reported Reform's equal highest polling figure and biggest lead and Labour's and Tories lowest polling figure for 5 years. If you run a simple 7 poll running average for each party you will see that Reform are currently on an average of 29%, Labour an average of 22% and the Tories 18% which net over the last couple of weeks implies changes of Reform +2 Labour -2 Conservative No Change but the local regression graph here suggests something completely different is happening.

That the change in direction of the polls has not yet been picked up by the graph tells me that at least one of the controlling variables (eg timespan of the samples under consideration) is not sensitive enough to pick up the immediacy of voter opinion. Anyway I would appreciate the thoughts of those involved on how to make the graph more accurate because its clear the graph is not representative given the contradiction the graph currently offers. 90.215.252.146 (talk) 17:13, 25 July 2025 (UTC)[reply]

The graph will always have a bit of a delay to it just due to it being a rolling average of sorts as you've mentioned. If the recent trend continues another week, what you'll probably find is it undoes a bit of the drop it currently reports and then straightens out slightly.
Honestly, I've felt that it can be too sensitive and sometimes reacts to some very short term margin of error changes but you're right that at the moment, it seems to be a bit slow.
That said, while the average has gone up, some pollsters haven't shown much of an improvement and polling is still somewhat lower for Reform than it's peak hence the drop. Either way, it's a waiting game, the graph won't take more than a month to react to any changes which was a problem when the page was first created (the "span" which dictates how reactive it is was 4 at first and has been reduced dramatically to 0.56 to find a nicer balance). Kirky03 (talk) 19:16, 25 July 2025 (UTC)[reply]
These sorts of smoothed curves often struggle at the ends. One approach might be to not run the curve to the end. The line could stop a week or fortnight before the latest poll. That doesn't solve the problem, but may prevent misinterpretation. Bondegezou (talk) 22:07, 25 July 2025 (UTC)[reply]
Hi Kirky, i didn't want to start a new topic but you seem to know how the main graph works. Just a query really, are GB polling and UK polling results adjusted/balanced on the graph? The reason i ask is that at the last GE Labour polled 33.7% nationally(UK) but 34.7% in GB, thus i assume for each of the major parties GB polling is in fact slightly higher than the national UK figure. For example, Labour's current 21% GB average is in reality c.20% for the UK. It's semantics really if all parties are affected equally, i just wanted to know if it's adjusted or not. However, overstated projections could affect seat predictions where models are based on all 650 seats including Northern Ireland. Scorpius Dog (talk) 19:53, 6 October 2025 (UTC)[reply]
I believe, no, the graph does not adjust for UK v GB polling. Bondegezou (talk) 20:08, 7 October 2025 (UTC)[reply]
I know the page as a whole has a "This page was last edited on" line at the bottom; is it possible, or do people think it's desirable, to have one on the graph as well? My thinking is that although you can see the dates of the polls in the table of poll results, you can't really tell whether the latest few polls have been incorporated into the graph. Nick Barnett (talk) 11:40, 16 December 2025 (UTC)[reply]
Apologies if this isn't the right place to raise this (please let me know) but i think on the main graph Labour have been entered in as 23% for a poll from early November that doesn't exist in the list. So, either the poll is missing or the 23% entry is incorrect. Thanks. Scorpius Dog (talk) 17:04, 7 November 2025 (UTC)[reply]
It's probably this one from JL Partners, which does not seem to have been included in the table. Impru20talk 17:16, 7 November 2025 (UTC)[reply]
I know the page as a whole has a "This page was last edited on" line at the bottom; is it possible, or do people think it's desirable, to have one on the graph as well? My thinking is that although you can see the dates of the polls in the table of poll results, you can't really tell whether the latest few polls have been incorporated into the graph. OR . . . does the graph get drawn according to what is in the table, so if you add a poll result, you get an updated graph? Nick Barnett (talk) 16:20, 15 December 2025 (UTC)[reply]
The graph is updated by a user once every few days, sometimes daily, depending on the polls. The dots aren't added automatically. Eastwood Park and strabane (talk) 19:12, 15 December 2025 (UTC)[reply]
Phew! Then I think I'll edit my previous words to delete from "OR" onwards. Nick Barnett (talk) 11:38, 16 December 2025 (UTC)[reply]
What I suspect may be more reflective than LOESS or moving averages, of the VIs of the electorate, and therefore, more satisfactory, is the changes (it−it−1) in percentages each poll shows, as opposed to the percentages themselves. I suspect that looking at the different pollsters' versions of the changes and finding a way of combining them would be quite revealing -- though it wouldn't, of course, tell you what the actual VIs were!
It wouldn't be straightforward to work out, since some polling is weekly, some is monthly, some is fortnightly, and some is irregular; some is GB, some, UK. Does this answer others' reservations about the current graph? If not, I won't spend my time on working something out. I also don't want it to get labelled as external research (or whatever the jargon is). Nick Barnett (talk) 13:34, 16 December 2025 (UTC)[reply]

Re-arrange the parties in the opinion poll column

I propose re-arranging how the parties are ordered in the opinion polls column. Simply put, it's customary that parties are arranged by seat count in parliament. Hence, the order of the parties should be Labour, Conservatives, Lib Dems, SNP, Reform, Greens/Plaid Cymru. Since virtually every other electoral opinion poll article is arranged like this (See Moldova's recent election, Hungary's upcoming election, or even Germany's 2029 election), it'd be nice if we had a similar arrangement for this article, to be in line with other articles.

It's not a strictly necessary change, obviously, as there's no rule that stipulates that they have to be in order of seats in parliament, but I think it's just a useful change to make - you can instantly discern which parties were the biggest/smallest in the last election. (Although I suppose this brings up the question of whether or not to include NI parties, because those parties have a larger number of MPs in parliament than Reform) Reaperz 2020 (talk) 19:11, 8 November 2025 (UTC)[reply]

it's customary that parties are arranged by seat count in parliament It is customary where, exactly? In 2 out of the 3 cases you mention (Germany and Hungary) parties are arranged by vote share (but this is not difficult, since arranging by seat count would give no change). In the third one (Moldova) it is not clear how the parties are arranged (it is not by seat count, but neither by vote share). I have checked a vast number of opinion polling articles and I see that customary practice is the contrary: arrange parties by vote share. By far. This is true even in Canada 2021 and 2025, where the first place party in seat count was the LPC, but the one shown in first is the CPC. Your proposed change would actually bring this article out of sync with other articles.
Also, since these tables list vote share results, it just makes sense that parties are arranged by vote share. There is a separate table for seat projections, and results there ara arranged by seat count. Placing the SNP ahead of Reform and the Greens in the main table does not look useful at all. Impru20talk 22:24, 8 November 2025 (UTC)[reply]
My mistake - I confused vote share with seat count, I guess. The UK's opinion poll table is perfectly fine. Reaperz 2020 (talk) 22:49, 8 November 2025 (UTC)[reply]
I see the argument but since the table shows vote share, I think it makes sense to order by vote share personally. Kirky03 (talk) 10:44, 9 November 2025 (UTC)[reply]
Precedent is to have them in order of national vote share at the previous election, not number of seats. Please look back at previous articles (2017, 2019 and 2024) when the Lib Dems were significantly outnumbered by the SNP but remained in the 3rd column based on vote share. A headline Opinion Poll figure reflects projected National Vote Share, not seats. The suggested change is not necessary, consistent or appropriate. The order should remain as is and then revised in the next article after the next election, to reflect the national vote share of that election. WestminsterWhistleblower (talk) 11:40, 6 January 2026 (UTC)[reply]
I agree - what I wrote before was something I wrote while seemingly being confused. Although this has already been concluded with a previous reply. Reaperz 2020 (talk) 11:42, 6 January 2026 (UTC)[reply]

Fix the graph

This could just be me but for some reason on the wikipedia app on mobile, it seems to blank out the newest 1 month on the graph, could this be fixed? ~2025-34197-47 (talk) 08:06, 17 November 2025 (UTC)[reply]

I have a similar issue, if I view the page on my mobile the graph is up to date when I view it in the page, however if I click on it to make it bigger it reverts to a graph from about mid-June Saxmund (talk) 11:40, 18 November 2025 (UTC)[reply]
@Saxmund Mine reverts to the end of october ~2025-34197-47 (talk) 07:55, 19 November 2025 (UTC)[reply]
@Saxmund I did not know what it was but it seemed to go away when I got a new phone ~2025-34197-47 (talk) 00:32, 1 January 2026 (UTC)[reply]
@~2025-34197-47 I've got a new tablet and it is showing the graph to the end of December, even zoomed in. So it seems to be picking up an old cached version on my phone. I could clear the cache I suppose. Saxmund (talk) 06:25, 1 January 2026 (UTC)[reply]
@Saxmund Mine does the same ~2025-34197-47 (talk) 09:36, 1 January 2026 (UTC)[reply]

Your Party section needs updating

Conference has happened and a few more polls have happened with YP included that haven't been added on there. Also one poll with Your Party and a hypothetical Rupert Lowe party. I'm not really sure how these should be added in but thought I'd raise it.


One from Freshwater strategy and the YP + Lowe one from More in Common.


I hope someone can act on this :) LevatorScapulaeSyndrome (talk) 11:45, 4 December 2025 (UTC)[reply]

@LevatorScapulaeSyndrome I was going to add the More in Common/YP one but can't find the data tables. It doesn't seem to be included in the most recent standard VI poll Saxmund (talk) 16:20, 4 December 2025 (UTC)[reply]

Where do we put Your Party?

When they (likely) start getting included in headline polls, where would we put them? I propose between SNP and Plaid Cymru would work. Maybe between the Greens and SNP. Lukt64 (talk) 17:27, 4 December 2025 (UTC)[reply]

For the rest of the year, I think we could put them as a footnote or hidden note in the Others column if they are included in any headline polls.
And if multiple pollsters begin to include Your Party, it could be added as a column in the 2026 table, probably between Plaid Cymru and Others. Pretzel Quetzal (talk) 18:08, 4 December 2025 (UTC)[reply]
Columns are in descending order of party vote share at the most recent (2024) general election. As Your Party did not exist in 2024, it had no vote share, so it should come after Plaid, just as it does it does in the existing hypothetical table. For seat projection tables the number of seats, rather than vote share, determines party order. ~2025-38695-49 (talk) 01:41, 6 December 2025 (UTC)[reply]
Hi, I am wondering if you have a reference for that being Wiki policy? Not doubting, just want to know for future reference myself. I'd intuitively have put YP after GRN to leave the regional parties together at the end, but what you say seems sensible. LevatorScapulaeSyndrome (talk) 16:07, 8 December 2025 (UTC)[reply]
There was a fruitful discussion in the talkpage for the 2024 election polling regarding editors moves to try and include UKIP in the polling columns at the time. Over the past few elections, there were several editors who wanted to place undue weight on UKIP/Brexit Party/Reform's supposed popularity in the article, perhaps as a means of promotion for the party itself, but that leans into WP:ASPERSIONS territory. I would be concerned if we added YP now, that it would be the same result. I would be more comfortable with either the status quo, or a footnote in the Others column as Pretzel Quetzal suggested. Bkissin (talk) 17:57, 11 December 2025 (UTC)[reply]
It's worth noting that no headline poll yet has included Your Party, so it would not be possible to add it to the main table now. I strongly support putting Your Party as a footnote in Others if it is included in any headline polls, and then holding a discussion about adding a separate column if a considerable portion of pollsters include the party.
However, I was wondering what the practice is for including events about minor parties. At what point is Your Party notable enough to add events related to the party, in this case the registration of the party to the electoral commission in September, to the main table? Pretzel Quetzal (talk) 23:22, 11 December 2025 (UTC)[reply]

Adding Your Party

Shouldn't (at least) the More in Common poll including YP now be included under Others in the main table? Probably too soon to make a whole column for YP but since the party has been founded and had its conference it's no longer "hypothetical polling". Eastwood Park and strabane (talk) 15:18, 6 December 2025 (UTC)[reply]

As far as I can tell, the headline 28 Dec - 1 Nov More in Common poll doesn't include Your Party, so the tweet cited for the Your Party poll must be from something separate – I'll have to see again when More in Common publish the tables. My understanding is that the main table is for headline polls, which (as far as I know) there are none yet including Your Party, whereas most polls including Your Party so far have been hypothetical, or asking whether voters would "consider" voting for it.
But yes, I think if there is a headline poll including Your Party, it should be included under Others. Pretzel Quetzal (talk) 15:53, 6 December 2025 (UTC)[reply]
I think the "consider" question is a separate poll, YouGov did one recently. Such polls are not included in this article, as they are not prompting for other parties. The polls in the hypothetical table are prompting for their usual parties and additonally for Your Party (or an unnamed left-wing party, e.g. led by Corbyn - the wording varies as in the footnotes). As far as the More In Common poll goes, the tweet refers to it as "hypothetical" and appears to frame Your Party as a "new left-wing party ... led by Zarah Sultana and Jeremy Corbyn with other independent MPs". This enables the comparison with their July poll, which used similar wording (I have corrected the footnote which was referencing the different wording from their June poll). Given the comparsion with the earlier poll and also the now out of date leadership, unless the data table says something different, I think it should still be treated as hypothetical. ~2025-38695-49 (talk) 17:16, 6 December 2025 (UTC)[reply]
I can agree with this. Eastwood Park and strabane (talk) 18:26, 6 December 2025 (UTC)[reply]
The recent Scottish poll is the first to include YP in the headline figures. If other pollsters begin to do this, I'd support moving them into their own column. Eastwood Park and strabane (talk) 12:35, 30 December 2025 (UTC)[reply]
Agreed, although personally I think that, in addition to multiple pollsters polling on Your Party in headline voter intention polls, there should be multiple WP:reliable sources reporting on Your Party's polling, which may take some time. I'm largely new to Wikipedia so I'm curious to know if there is a previously agreed-upon criteria for including parties that could be used to assess Your Party.
I've put Your Party in the Others column of the Scottish poll for now, as there were no objections when I suggested it earlier, and I think we should continue to put it in the Others column until there is a future discussion if/when Your Party gets significant polling coverage. Pretzel Quetzal (talk) 13:24, 30 December 2025 (UTC)[reply]
There was some debate over this issue around a year ago on Talk:Opinion polling for the 2026 Scottish Parliament election#Moving Alba to Others on regional vote, where a (now banned) user had moved the small Alba Party into Others, similar to where YP is today, where it was sort of agreed to include as many parties as are being significantly prompted for, someone linked some examples that this is the standard elsewhere e.g. Germany, Italy, Austria, the most parties I've seen is Israel. I'd wait until at least a few more polls come out with YP prompted, but I think it's probably not far away that we will want to add a column. Eastwood Park and strabane (talk) 14:37, 30 December 2025 (UTC)[reply]
I think it also makes sense to see how pollsters classify YP themselves. In this article from earlier this month, YouGov explains why they place YP under others. —Profzed! 18:22, 30 December 2025 (UTC)[reply]
Just as an aside, was there any discussion about adding Alba to the Scotland table in this article? The party is prompted for, by 3 pollsters out of 7, in 11 of the current 22 polls, although the latest Ipsos poll (3 Dec 2025) doesn't report a vote share for it, as the base was too small (2 respondents out of 742 = 0.27%). The three prompting pollsters are Norstat (6 polls), Ipsos (2, 1 reported) and Survation (3 out of 6, depends on client). Would this count as "significantly prompted for"? Also, in the latest Ipsos poll, as Alba was prompted for, but not reported, it is not included in the "Other" 1% share (albeit, due to rounding, adding Alba would not increase it above that). So, if Alba has its own column, how should its vote share in the Ipsos poll be shown? Consensus seems to be that figures should be whole integers, so 0%? Has, say, "<1%" ever been considered? Or a general note somewhere to the effect that 0% usually (if not always) means "less than 1%". If there is no new column, then should the Ipsos poll Others be a show box with "Alba 0% Other 1%". Is it OR to add that or OR to omit Alba altogether? ~2025-41479-54 (talk) 11:33, 31 December 2025 (UTC)[reply]
I'd personally just round down to 0%, but yes I'd support including a column for Alba, given that half the polls prompted for it, in fact I'm going to WP:BEBOLD and add a column - don't think I realised there wasn't one yet Eastwood Park and strabane (talk) 14:06, 31 December 2025 (UTC)[reply]
Given that a significant of pollsters find Alba significant enough to poll, I think it's probably safer to give it a column.
It seems that the separate Opinion polling in Scotland for the next United Kingdom general election article has included an Alba column since its creation. But it's also not clear why that article exists, as it is largely the same as the Scotland section of this article, the only unique section being a table for sub-samples of UK-wide polls. Pretzel Quetzal (talk) 18:32, 31 December 2025 (UTC)[reply]
I've put Your Party in the other column for the Electoral Calculus MRP released today, which is the first inclusion of the party in the main national table. I think it should included this because the party is reported in the poll's headline voting intention and seat count, because they were polled as "Your Party" and not "Your Party led by Jeremy Corbyn" as previously, and because the party has had its inaugural conference, registered with the Electoral Commission, and now has a permanent name.
Obviously this doesn't warrant its own separate column, but I think we should continue to include Your Party whenever the pollster reports it in their headline VI.
I've also noticed that Electoral Calculus abbreviated the party as "YP" and not "Your" as we have been doing. I think that's more natural, so I'm going to WP:BEBOLD and change it throughout the article. Pretzel Quetzal (talk) 19:29, 13 January 2026 (UTC)[reply]

LOESS Graphical Summary

The graphical summary obviously invites a lot of attention at the top of the article. Some people might not even scroll under it, so it does matter if there is a bias in it.

The graph has been showing a steep Green rise for several months at the end of the line suggesting momentum. Weirdly a history of previous versions of the picture is unavailable, even though it is updated almost every day. However I do remember that a few weeks ago the whole line was higher. It seems that the (almost nameless) Wikimedia Commons editor ("Icantthinkofausernames") would do anything to display the Greens as rising dynamically.

Currently the Greens have stalled on teen percentages for a month, as you can see in the chart. Still the graph is showing extremely dymanic rise throughout november. How is that not a deliberate attempt at propaganda? ~2025-39443-78 (talk) 17:48, 8 December 2025 (UTC)[reply]

I don't personally see the unexpected rise mentioned - the chart only goes up at a maximum rate, so if the greens stagnate another month or two it will flatten off - the chart doesn't really deal with sudden jumps in party support well. Reform are down a little just now but also show some lag in the numbers - the individual points appear to drop faster than the line does, I think it's just part of the trade off of smoothness of the line vs total accuracy month to month.
I am personally able to access the history of this svg, so I'll link a couple here.
1 October 2025: https://upload.wikimedia.org/wikipedia/commons/archive/6/6e/20251002111955%21Opinion_polling_graph_for_the_next_United_Kingdom_general_election_%28post-2024%29.svg
1 November 2025: https://upload.wikimedia.org/wikipedia/commons/archive/6/6e/20251105102249%21Opinion_polling_graph_for_the_next_United_Kingdom_general_election_%28post-2024%29.svg
18 November 2025: https://upload.wikimedia.org/wikipedia/commons/archive/6/6e/20251118105933%21Opinion_polling_graph_for_the_next_United_Kingdom_general_election_%28post-2024%29.svg
Current version: https://upload.wikimedia.org/wikipedia/commons/6/6e/Opinion_polling_graph_for_the_next_United_Kingdom_general_election_%28post-2024%29.svg
Comparing them it's clear the green's line has smoothed out since mid-November, as you would expect for a party not rising at speed any more. Eastwood Park and strabane (talk) 19:44, 8 December 2025 (UTC)[reply]
No propaganda. Changes take a while to come through. They will need to be flatlined for a couple more weeks before the graph reacts. Just the way of the algorithm. Kirky03 (talk) 21:15, 8 December 2025 (UTC)[reply]
LOESS is a feature, not a bug. Certainly not manipulation by another editor. Hope that helps. CNC (talk) 21:28, 8 December 2025 (UTC)[reply]
LOESS isn't really appropriate for this type of graph. Poll ratings do not change in a smooth curve. They change suddenly due to actual political events in the real world. So a LOESS smoothed curve gives a misleading picture of what is happening. In addition, if a party's support flatlines after a rise, the graph will actually turn down.
I don't know who decided to try to be "clever" by using LOESS for the graph, but it gives a much more misleading graph than a simple moving average would. ~2025-40257-97 (talk) 00:52, 14 December 2025 (UTC)[reply]
As a statistician in a former life, this comment is correct. LOESS is not appropriate here. The most appropriate would be EWMA, but whatever factor we use to decay the weights would maybe constitute original research? For that matter I wonder why the LOESS calculation isn't considered original research actually... Best would definitely be EWMA but if that's not possible for whatever reason then just moving average and call it a day. — Preceding unsigned comment added by ~2025-40334-47 (talk) 11:19, 14 December 2025 (UTC)[reply]
To be more explanatory, there are two big issues I see with using LOESS. Firstly, it uses future data to smooth past points. This results in a trend line which can appear to "see the future" with points appearing higher or lower than they really were on a particular date because of a shift in sentiment which hadn't happened yet but was about to. I think this completely rules out LOESS from ever being acceptable in this context. The second problem for wikipedia is that somewhere we will be specifying how many points contribute to the local regressions - maybe this parameter is reverting to a default somewhere and so editors are unaware that a subjective judgment call is being made, but there is not one canonical "best" choice of parameter, and so the resulting calculation constitutes to a degree original research. EWMA would be most appropriate, but also has this possible issue of original research. Plain old moving average has neither issue. — Preceding unsigned comment added by ~2025-40334-47 (talk) 11:45, 14 December 2025 (UTC)[reply]
Even a plain moving average would require some subjective decision over how many points to include in the average (and whether that represented a fixed number of points or a fixed timespan), wouldn't it? But certainly there's been a lot of discussion in previous months/years about the "right" LOESS parameters to allow the graph to react to changes at all without being so sensitive it picks up who ran the last N polls more than anything else.
I'd prefer to have there be no averaging line at all, and so have the mess of coloured points emphasise that there is a good chance that averaging two wrong numbers doesn't give a right number. ~2025-40936-44 (talk) 15:57, 15 December 2025 (UTC)[reply]
I think a no line approach would be a bit hard to read but you're right that any approach has potential issues, and a moving average would face many of the same issues as LOESS. Perhaps there's an argument it would be better though, albeit not perfect but either are better than nothing in my opinion.
From memory, I believe the code for this came from the German voting intention page so this is not the only page using it and like I say, I think it is fine to just give an idea of a trendline. The issues arise when people expect immediate changes on the graph. We did discuss and finetune the span parameter a fair bit to get it to a level that was smooth but responsive. At current, I think it is 0.56 (from memory, I would need to check) and was 4 when we first started. This is obviously a bit subjective but we weren't making any analysis of the data, just working out the best way to present it so I'd say it wasn't quite OR.
But yes, maybe we should discuss changing to a type of moving average but I'm not going to be the one to code it. Kirky03 (talk) 09:48, 18 December 2025 (UTC)[reply]
I don't see the issue with LOESS using data-points after the position being considered in this case. Normally that objection is raised in circumstances where the point of the work is to predict future events. Wikipedia isn't in the gambling or fortune-telling game. There's no issue with a given point reflecting what is now known to have happened afterwards any more than it reflecting the past.
As an analogy, it's not wrong for someone to say in an article here that something happened "on the eve of the First World War". You could even argue it would be wrong to treat the past and (at the time) future asymmetrically. Prolepsis is fine, and different to predictive anticipation.
And the tuning parameter issue issues for all methods. ~2026-36172-7 (talk) 13:30, 17 January 2026 (UTC)[reply]
The intuitive understanding is surely that the trend line represents political opinion at the given point in time. LOESS doesn't do that. ~2026-37718-7 (talk) 09:50, 18 January 2026 (UTC)[reply]
The point raised as problematic in the comment I was replying to (other than the need for tuning) was that LOESS uses future data in its calculation of a value. I don't see why that's problematic in this context. Nearby future data is no less reliable at refining likely opinion at a given moment in time than recent past data is, if it's real data and not hypothetical. ~2026-36172-7 (talk) 18:01, 18 January 2026 (UTC)[reply]
It is problematic because, by using future data points in the calculation, the trend line can appear to anticipate changes in sentiment that had not yet occurred. A reader’s intuitive understanding of such a graph is that it answers the question “what was average public sentiment at that time?”, not “what value lies on a mathematically smooth curve fitted retrospectively to the entire dataset?”. LOESS answers the latter rather than the former, whereas one-sided smoothers answer the former but not the latter. ~2025-40334-47 (talk) 06:54, 19 January 2026 (UTC)[reply]
Each poll finding is only approximate, affected by sampling error. Information from polls both earlier and later can help correct for that sampling error. Thus, a two-sided smoother makes sense.
Whether we should be using any sort of smoothing algorithm is a harder question. WP:CALC applies. Bondegezou (talk) 10:13, 19 January 2026 (UTC)[reply]
It’s true that using data on both sides would reduce sampling noise. However, future data points also contain genuine information about opinion at that later time, not just noise around the earlier value. Incorporating them therefore does not merely average out error, but blends in information about changes that had not yet occurred. In a time-ordered context like opinion polling, that future signal is precisely what should be excluded when the aim is to summarise sentiment as it stood at a given moment, rather than to construct a retrospective best-fit curve. ~2025-40334-47 (talk) 11:51, 19 January 2026 (UTC)[reply]
On the issue of smoothing, I notice that the graph on this page is dramatically more smoothed than the one for opinion polling for the 2024 general election: https://en.wikipedia.org/wiki/Opinion_polling_for_the_2024_United_Kingdom_general_election
Have different parameters been used? If so that is confusing because some people are likely to compare the two periods and the presentation and description are the same. Personally I think the greater degree of smoothing is unhelpful as it hides fluctuations caused by real events. ~2026-10809-75 (talk) 08:23, 18 February 2026 (UTC)[reply]
Impossible to know without seeing their code, but I would say almost definitely yes, that looks like an alpha of ~0.4 whereas the current one uses ~0.6. I think the code for all of these charts should really be required to be made open source. Flobberz (talk) 17:19, 24 February 2026 (UTC)[reply]
You can access the code here:
https://gitlab.com/gbuvn1/opinion-polling-graph
which is available if you click on the graph, click on “Details”, then scroll down to “Author” and click on “code”. Sammy zammy1 (talk) 13:00, 25 February 2026 (UTC)[reply]
I know, but the code in that repo is not the same as the code used to make the current graphs and I don't believe is the same as what was used to make that one either, despite being in the descriptions of both. That code uses an alpha of 0.25, which would be less smoothed than that. Flobberz (talk) 13:16, 25 February 2026 (UTC)[reply]

Inaccurate Graph on Wiki Poll Tracker

The graph is showing the Reform parting heading downwards, yet its membership is climbing rapidly. it is now over 270,000 and, if it keeps growing at that rate, it won't be long before it is 300,000 members. Yet you show the Green Party growing rapidly with only 180,000 members. That surely can't be right, and therefore, it seems to be true that there is a bias in Wikipedia. ~2025-40745-52 (talk) 13:08, 14 December 2025 (UTC)[reply]

Apples and oranges. Kirky03 (talk) 09:33, 18 December 2025 (UTC)[reply]
i'm a Reform member but party membership is irrelevant when it comes to opinion polling. Firstly different parties have different fees and rules. Also party membership is not independently scrutinised. However as discussed in another thread the LOESS graphs seem very bad at reflecting steep climbs and tail offs. The Politico graph clearly show Greens flat since October, whereas here the graph implies a continuous but softening trend upwards which is statistically incorrect. It did the same for Reform's climb in the spring, showing them still climbing when they had levelled off. And then there is the issue of regular pollsters (YouGov, FON) giving false weight over monthly or quarterly polls, and implying trends that don't exist. Scorpius Dog (talk) 10:19, 18 December 2025 (UTC)[reply]
@Scorpius Dog the problem is that any attempt to weight the polls for recency is OR. Maybe it would be better to link to the Politico poll as a reliable 3rd party source Saxmund (talk) 09:45, 19 December 2025 (UTC)[reply]
I would disagree on it being statistically incorrect - you could feasibly have a very jumpy line that went up and down at the slightest change, or a smoother one that takes longer to smooth out after an increase - neither is inherently right or wrong, but the current chart does lean towards the smoother option. Eastwood Park and strabane (talk) 16:46, 19 December 2025 (UTC)[reply]

Query on trend line shown in polling graph

I’d like to query the way the current trend line in the polling graph is being drawn.

At present, the graphic appears to show support for Reform falling. However, looking at the polling figures over the past several weeks, support seems to have been broadly flat rather than in sustained decline. The underlying data clearly doesn’t support the downward trend suggested by the line.

I'd suggest that the smoothing or weighting settings being used are giving a misleading impression, perhaps by over-emphasising short-term variation. Either way, the graph is not accurately reflecting the stability seen in recent polls.

I suggest it's worth reviewing how the trend line is generated to ensure the visual presentation matches the data more closely. Happy to hear from anyone familiar with how the graph is currently produced.

The fact is, in most polls that showed the biggest previous drops in reform support (e.g. YouGov), the party has essentially recovered in those polls. — Preceding unsigned comment added by ~2025-42093-58 (talk) 12:42, 20 December 2025 (UTC)[reply]

@~2025-42093-58 it's just a bit laggy. Any "smoothing" calculation will do this because it is based on previous figures. If it isn't laggy it will be too jerky. Saxmund (talk) 05:07, 21 December 2025 (UTC)[reply]
The graph shows support for Reform falling, because it has indeed fallen. Their average over the first five polls in October was 32.8%, first five in November was 30.8%, first five in December was 29.2%, and now average over the last 5 polls is 28.4%. The trend lines are bad for several reasons, but this isn't one of them. ~2025-34430-90 (talk) 09:18, 21 December 2025 (UTC)[reply]
You are correct that over-emphasising short-term variation is misleading. Yet, you are asking for exactly that: a short-term variation to be emphasised.
It takes time for changes to come through with LOESS - that's a feature, not a bug (or a conspiracy). ~2025-42318-25 (talk) 00:08, 22 December 2025 (UTC)[reply]
Is it not a bug? Points at the boundary requiring an asymmetric fit, the more recent points we look at on the trend line the further back into the past we are considering as part of the fit. It makes it much harder to accurately depict the most recent points on the trend line, which are likely the points readers are most interested in. Feels like a bug to me. ~2025-40334-47 (talk) 13:34, 22 December 2025 (UTC)[reply]
As I've said before, LOESS is completely inappropriate to track opinion polls. Political opinion does not change smoothly. It changes abruptly in response to real world events. Additionally, LOESS effectively predicts the future by assuming trends continue. Reform support fell a little and then stabilised - the graph shows their support continuing to plunge. Green support rose following the election of their new leader and then stabilised. The graph shows their support continuing to soar.
Carry on using a LOESS graph if you want. But it brings Wikipedia into disrepute. ~2025-40257-97 (talk) 09:28, 8 January 2026 (UTC)[reply]
I struggle to think of any viable solution that won't include smoothing since the point of the trendline is to smooth the trend. Kirky03 (talk) 13:06, 8 January 2026 (UTC)[reply]
This trend line simply is not reflecting the real world. For example the Greens have not risen since mid December but the graph still shows them increasing steeply. Same with Reform - no significant change for weeks but showing decreasing support.
It seems that the trends are too "sticky" - in other words, the graph is not responding to actual events.
Either the parameters are wrong or you need to stop using LOESS. It is not serious or professional as it is. It needs fixing. ~2026-38329-6 (talk) 12:24, 18 January 2026 (UTC)[reply]
By the way, the purpose of the "trend line" is not to smooth the trend. It seems you have misunderstood what opinion polls are. They are NOT exact data points that need smoothing. They are estimates based on samples. Every poll contains an error margin because of this (normally +-2% for each party's figure). So the line should not be smoothing a trend of opinion jumping up and down with each poll.
The purpose of the trend line is to show a better estimate of what is happening to underlying public opinion. This is so that readers can make sense of the jumble of figures coming from the individual poll estimates. This is why a rolling average would do this job much better. ~2026-38329-6 (talk) 12:39, 18 January 2026 (UTC)[reply]
Hi i'm now really confused with the trend lines, Reform have had countless results of 29% or above for the last three or four weeks with a tiny number, mostly YouGov, below 29% (although even YouGov polling is rising). I am baffled therefore why the trend line persists in showing Reform as going down and persists in showing Reform at less than 29%? Ditto the Greens have flatlined for four months, yet their trend is always shown as 'up', with their 'up' angle getting softer, but (conveniently) always up. These trend lines are not being shown correctly in comparison to other political statistics sites. Perhaps it is time to only show straightforward graphs for individual pollsters if the poll-of-polls LOESS graph cannot be managed non-politically, especially given the activism that is clear in these threads. Scorpius Dog (talk) 15:59, 12 February 2026 (UTC)[reply]
The politico poll of polls currently has Reform on 28% Lab 19% Con 17% Green 14% LD 12% SNP 3 % PC 2%, so slightly differernt to here, but not hugely so. Happytiger00 (talk) 16:10, 13 February 2026 (UTC)[reply]
I disagree and it’s not political. If you look at the latest positions of the LOESS regressions, they lie in the middle of the cluster of most recent result plots. This is true of any point on the lines. The gradient of the line at any point represents the change in where this “middle” point lies as you move along the x axis.
What we all have is a personal impression of current political trends and we look to statistics to support or diminish this impression. So I, like others above, have watched Green support being reported as growing, Reform support as not growing and Labour support as dropping. In fact, the LOESS regressions show us that growth in Green support is steady (look at the latest gradient if you disagree - it’s very shallow), Reform support is falling but slowly, and Labour support is now growing (albeit from a low base).
I live near a flood zone and if I was watching river levels that looked like these lines I would predict - if no other factors come into play - that Green support would stabilise at 15%, Reform at 27%, and Labour at 20%. Does this really “feel” so very different to our impression given the latest polls? Of course they could do something completely different because, like opening and closing weirs on the river, there are constantly new inputs being introduced.Btljs (talk) 08:16, 14 February 2026 (UTC)[reply]
Iagree with the point that this isn’t political. LOESS itself doesn’t “favour” any party (it’s just a smoothing method). But I also think the original post reflects genuine concerns about real and well-known properties (and limitations) of LOESS that are visible in these graphs.
LOESS is not “the middle of the recent cluster." If it were simply plotting the midpoint of the most recent batch of polls, that would effectively be a moving average (which is what we should be doing). LOESS is different: at each x-value it fits a locally weighted regression using a window of surrounding data, with weights declining by distance. The fitted value is the result of that regression, not the midpoint of recent dots. Because it is regression-based rather than an averaging window, the line can legitimately sit above or below the centre of the most recent cluster. This is especially visible if the broader window contains earlier higher or lower values which is exactly what we see with Green and Reform.
Endpoint bias and lag are real. LOESS has well-known edge effects. Near the right-hand edge (i.e. “now”), the fit is asymmetric because it can only look backwards in time. That produces lag. If support has recently levelled off after rising, LOESS will continue to slope upward for a while because the local regression window still contains earlier growth. The reverse happens after declines.
This is not political, it’s just how the method works, but I can see why Reform supporters in particular would be unhappy because in this particular case they've drawn the short straw and the graph makes things look worse for them than it is, consistently showing a drop when it's now pretty stable (Politico mentioned above has them on 28% average - it also showed 28% average exactly one month ago).
The gradient is not stable. This is a subtler one, but the gradient at recent points is not fixed. When new data is added, LOESS refits the entire curve. That can and does change the slope of recent sections retrospectively. So saying “the gradient shows X trend” is only true conditional on the current dataset. We do not know what the gradient at the current points will end up being.
So yes it's not "bias" or targeted in any way. When Reform were on the up, they would have "benefited" from the same flaws in the smoothing method that Greens are "benefiting" from now. But we really shouldn't be doing this. We wouldn't have any of these problems if we just swapped to a more case-appropriate smoothing method. ~2026-37718-7 (talk) 09:10, 14 February 2026 (UTC)[reply]
The politically motivated activist decision would surely be to change from the longstanding LOESS chart at this point. Happytiger00 (talk) 15:23, 14 February 2026 (UTC)[reply]
I feel like I’ve just had LOESS mansplained to me ;) I know what LOESS does! It’s not a flaw and no, moving average wouldn’t be better - just different. It would still look backward and would still show lag. What I’m saying is the perception of it being somehow wrong isn’t supported by looking at the chart and removing the political nature of the data. The light blue dots in the most recent 2 months lie between 24 and 32. Halfway is 28 and the light blue line is on 29. If you took a range of 95% it would still be around 29. The previous 2 months had some higher values so would have been a bit higher. So it’s come down. That’s not perception - that’s fact.
We should not be making decisions on content based on somebody misinterpreting or misusing it. Other sources are awash with bad statistics and that’s not our concern. If there is an argument to be had here - it is not what smoothing algorithm should we use but rather - should we be using one at all? It is OR after all. I’d support removing the line completely if that was proposed - it’s more useful to look at individual pollsters change over time anyway and aggregation is dubious at best - a bit like trying to spot the ball by tracking the positions of all the players on a football pitch. Btljs (talk) 06:07, 15 February 2026 (UTC)[reply]
Apologies for explaining LOESS. I can see that would be unpleasant. I picked up on the line "If you look at the latest positions of the LOESS regressions, they lie in the middle of the cluster of most recent result plots. This is true of any point on the lines." This is not true of LOESS. I can imagine my correction of this seemed pedantic or pointless, which I'm sorry for. I didn't mean to come across as doing the equivalent of correcting your spelling. I do think drilling into that specific point is important, because the misunderstanding I think is basically universal. I think it's a completely natural instinct to think that's how a trend line works (because it is usually how a trend line works). Many of the people complaining about LOESS have this same expectation, and it's easy to imagine 100% of the people visiting the page make exactly the same mistake too.
LOESS does represent broader movements correctly, as you rightly say. Compare 2 months to 2 months prior, and we see a decrease for Reform support which is there in the data. Great. One issue here is that LOESS does far worse on the shorter scale. The average of the last 10 polls is 29.6% for Reform. The average of the first ten for this year is 29.1%, actually slightly less. If I broaden the window, I can get the gradient to be flat, or even slightly up, but we are currently depicting it going down and I can't see how that's really defensible.
The discussion over the "political" implications overshadows the fact that our smoothing method is inappropriate. People on one side say "this graph misrepresents Reform" (which it does) and the other side then say "LOESS is politically neutral" (which it is). This discussion wouldn't need to happen at all if we just did a moving average.
I do think we should consider how a graph is interpreted. If we asked 100 visitors "the Lib Dem line shows 13% for October last year, what does that mean?" how many of them do we honestly think would wrongly answer "it shows Lib Dem support was about 13% in October last year"? I would guess every single one. We see that same mistake in interpretation from the people complaining about the graph here, and the same again from the people defending it. At some point surely we should take responsibility and just put up a graph which will be interpreted correctly by everyone first time?
I don't agree that there is anything misguided with showing a trend line - it's perfectly natural to try and pick a trend out of noisy data. We have multiple measurements of the same thing, and so give an indication of the underlying trend. In the football analogy, there's an additional step of saying that average trend indicates something else (the ball position) which doesn't follow logically. In our case, it does follow logically that the average trend indicates average political opinion. That's what the polls are measuring after all. ~2026-37718-7 (talk) 08:18, 15 February 2026 (UTC)[reply]
I apologise - I shouldn't criticise anyone here for explaining things; it is, after all, what we're all trying to do here. All your points are valid. I did a bit of OR myself, and discovered that a monthly moving average for Nov 25 to Feb 26 shows a very static position for the light blue points: medians are 30% for all 4 months, means are 29.5%,29.2%,29.0%,29.1%. So with a squint, you could say there's a decline, but not as much as portrayed by the LOESS. The green points medians are 13%,15%,13%,16% and the means 13.8%,14.2%,13.6%,15.3%. So, a rise but rather a turbulent one. The gradient is at least, the same in both methods for these. The biggest change of using a monthly trailing average is arguably that the red line (shown as below the dark blue line in LOESS) is actually above it 19.0% to 18.4% in February.
I would be happy to change the trend line to a monthly trailing average (either median or mean) which would show more bumps and lumps which might better reflect the impact of events which tend to be more or less immediate, such as conference bounces and by-elections. Btljs (talk) 09:59, 16 February 2026 (UTC)[reply]
“The LOESS graph cannot be managed non-politically”
>proceeds to make political arguments of why it’s wrong
It’s a lagging indicator, and is over a long time period. Therefore it will not pick up small changes. You’ll note that Labour were still falling as of the other week despite other aggregates showing them as having bottomed out and rising a bit. Nothing political about it. Sammy zammy1 (talk) 12:50, 16 February 2026 (UTC)[reply]
I have experimented with using EWMA instead of LOESS but honestly the trends don't look all that different. Not sure where is the best place to share a comparison of the images, but for each party the trendline differences are as follows:
  • LAB - essentially the same but troughs slightly lower and shows a steeper recovery over the last few weeks
  • CON - shows a brief rally in Apr 2025 followed by a sharp cliff-edge fall in May 2025, and a plateau/steady drop more recently
  • RFM - mostly the same, shows a brief drop in Apr 2025 before a sudden, very sharp cliff-edge rise in May 2025
  • LD - practically identical
  • GRN - shows a steeper rise occurring in Oct 2025 and completed by Nov 2025, being essentially flat since then, overall not that different
Other than this all it adds is a lot of spikiness and unnecessary noise, which can be misleading due to the already inconsistent biases of different pollsters (eg. if one pollster doesn't do a poll for a while, this can look like a rise/fall for certain parties). I believe this makes EWMA inappropriate. The above differences can instead be elucidated by using LOESS with a span size of ~0.4 rather than ~0.6 as we currently do, which shows much more granular changes (like sudden but definite rises within the space of a month or so) without adding all the noise.
I do agree that the current LOESS with ~0.6 span size smooths it far too much but would probably favour changing this to ~0.4 rather than switching to EWMA. Perhaps we could use both with a button switcher if people really want. Flobberz (talk) 16:05, 17 February 2026 (UTC)[reply]
Having the line be a bit "noisy" isn't necessarily a bad thing, since it's created by averaging multiple sources with different methodologies (which WP:CALC says "don't", but there seems to be local consensus that "it's okay if they're political opinion polls"), and which individually have quoted errors of +/-4% - it should be a bit noisy for reasons of random variance and pollster frequency and it might get across that it doesn't matter all that much exactly where the line goes on any particular day better than the oversmoothed form does.
(Whether that's done by a shorter LOESS span or by an EWMA with a fairly short timescale I agree is probably not all that relevant) ~2025-40936-44 (talk) 16:03, 19 February 2026 (UTC)[reply]
Maybe our eyes are different, but I don't see any real spikiness or noise. What decay are you using? I used 0.03 and the result looks fine to me.
You're absolutely right that the broad strokes of the trend lines will be the same. We certainly wouldn't expect big differences in general - that would be surprising. But focus on the most recent points on the trend lines. This is where LOESS performs worst because of edge effects. It is also where visitors to the page are most likely to be interested in. If we look through the comments from people who are unhappy with the graph, they are also all talking about issues with the most recent parts of the trend line. Two additional differences here are:
- LOESS shows Conservatives above Labour, while EWMA shows Labour above Conservatives.
- LOESS shows Reform on a downward gradient, while EWMA shows them flat.
In both cases, it's much easier for us to defend what EWMA is doing from the data. And it's also far, far more intuitive to explain and subsequently understand how it got there. How might we defend LOESS currently putting the Conservatives above Labour? I would struggle to mount any case for that, because it's a ghost in the machine, emerging magically from a series of calculations with no clear foundation. The best I could do is shrug and say "computer says so" (and be thankful that Labour supporters aren't as eager as Reform supporters to insist there's a graphical conspiracy against them).
A defense of the EWMA side is nice and easy to make in contrast. We'd just say "look at the data. Most recent polls have Labour ahead, so our line shows Labour ahead." Done. Nothing more to it. No drawn out discussion. No inexplicable regressions. Everyone understands. Everyone can move on. ~2026-37718-7 (talk) 21:47, 19 February 2026 (UTC)[reply]
Here are the graphs I have created with LOESS 0.6 vs LOESS 0.4 vs EWMA 0.3:
From these, we see Labour above the Conservatives in the latter two, but below with LOESS 0.6 (the currently used setting). We see Reform on what I would call a downward gradient in all 3, but much shallower in the latter two (to what you might call flat over the last month or so). They also allow us to see the sharper rises/falls.
The main differences between the EWMA 0.3 and LOESS 0.4 are:
  • Small bumps in EWMA that don't show up in LOESS, most notably the Labour bump around Jan 2025. If we inspect the polls around this time, a key cause is one pollster - Deltapoll - who performed 3-4 polls in that two-month period, and then stopped, only releasing 2 more in the year since then. All 4 of their polls had Labour at 29-30%, indicating a likely polling bias, and driving that bump up. I believe this is a bad thing, and the LOESS 0.4 graph omits it as such.
  • Janky lines on EWMA in the period after the 2024 GE, due to sparse polling, which I think can be misleading as it gives the impression that, for instance, Reform climbed by ~3% in a single day.
  • A downward Green trend at present on LOESS 0.4, which appears flat on the EWMA. I'm not really sure what's driving this on the LOESS, I think it is just an edge effect as you say, and this is probably where EWMA does look better. However, I don't think it's really worth it overall just for this, and we could even lower the LOESS alpha to 0.3 or something to mitigate this further.
The EWMA is also just a lot messier at a glance, especially where lines cross each other and parties are at similar levels. I appreciate LOESS is certainly not perfect but I think with an alpha of 0.3 - 0.4 it is a good balance between statistical accuracy and simplicity.
Either way, we can certainly agree that LOESS 0.6 sucks and should not be used. We should probably bring the other graph editors into this conversation to discuss it further. Jamie Eilat User:DimensionalFusion Flobberz (talk) 21:25, 20 February 2026 (UTC)[reply]
That "the graph responds to systematic differences in polling" and "the graph doesn't attempt to smooth out periods where we have no idea what was going on because there was almost no polling" are cases I think strongly in favour of EWMA.
The point of the line should surely be "to show what the polls were doing", not to try to show "what we WP:OR think the polls would have been doing, if they'd been happening and if Deltapoll hadn't been there".
I agree the EWMA jaggedness isn't aesthetically pleasing, but then, the underlying data isn't aesthetically pleasing, and making clear that there is a lot of noise in it I think is also a positive. I think it represents the effective polling tie between Labour and Conservative, for example, better than the artificial neatness of either LOESS. (I still think "no line at all" is most appropriate, but EWMA is clearly better if there must be one) ~2025-40936-44 (talk) 12:14, 21 February 2026 (UTC)[reply]
I can now see why the worry about noise comes in! Taking alpha=0.3 in EWMA is a very sharp decay, and will lead to the nasty volatility seen there. 0.06 is the standard in finance and NHS stats. I used 0.03 when making my comments above. ~2025-40334-47 (talk) 17:57, 21 February 2026 (UTC)[reply]
As far as I knew 0.2 was generally the starting point, 0.03 creates a very lagged trendline that clearly doesn't pass through the point clusters well, and it doesn't really begin to do so until 0.1 - 0.2 in my testing, particularly on the early region of the chart (up to about mid-2025), which kind of rules these out to me. This is another point to why I think LOESS ~0.4 is more appropriate. I will post some from 0.03 up to 0.2 below:
Flobberz (talk) 17:12, 24 February 2026 (UTC)[reply]
This probably highlights why both this year and the 2019-2024 cycle had repeated "do we need to change the averaging" discussions over the graph: the frequency of polling is not at all constant, so a value which works well when there's multiple polls per week doesn't also work when there's just three in a month.
Is it practical to use a fixed time period for the averaging window instead? (probably some multiple of seven days, to avoid the most obvious pollster frequency effects) ~2025-40936-44 (talk) 14:37, 25 February 2026 (UTC)[reply]
I cannot quite tell exactly what is going wrong with the EWMA calculation here, but it is incorrect. The second point of Labour's line should be averaging two data points (34 and 39) with the second having more weight, and yet the lines depicted do not even go above 35. Perhaps the weights are being misapplied backwards? ~2025-40334-47 (talk) 20:01, 25 February 2026 (UTC)[reply]
Yeah that EWMA is wayyy too jagged. Anything too reactive implies a bit too much trust in each individual poll; i.e. ideally we shouldn't see any movement if polls fluctuate within the margin of error unless it is sustained.
Is it possible to adjust the EWMA parameters to get it less jagged, I'd be interested to see the 0.06 suggested.
I'd also be interested to see 0.5 for the LOESS. Kirky03 (talk) 01:40, 22 February 2026 (UTC)[reply]


and guess what....after multiple polls in excess of 29% for Reform over a month, in comes a YouGov at 24% and the LOESS goes down from 28.5 to 28.0. This page is a joke. Scorpius Dog (talk) 16:14, 18 February 2026 (UTC)[reply]
You can think it’s a joke all you want, but you’re trying to argue with statistics…
Since Monday, both a YouGov -3 and a More in Common -2 have been added. JLP showed a +2 increase but over a 3-month period so won’t have much effect. The trend line dropping by about -0.4 therefore seems reasonably appropriate.
Even judging by eye, the spread of Reform’s numbers is drifting down.
The fact YouGov on average shows lower Reform numbers than other pollsters makes no difference because there’s a YouGov poll every week. For every new YouGov poll that “comes into” the equation, an old one will “drop out” of it. The only thing that will affect the trendline is the change in YouGov’s numbers, which as mentioned previously, was -3.
Worth remembering that LOESS is a curve fitting model. It can only work off the data it has, but its parameters make it slow to react to changes in the trend, until more data comes in to confirm it - this in effect drags the curve down, which is what you’re seeing. You’ll note that Labour’s line continued to go down long after they had clearly plateaued, but no Labour supporters were complaining. Sammy zammy1 (talk) 19:11, 19 February 2026 (UTC)[reply]
The most recent 10 Reform percentages are:
28, 28, 24, 28, 30, 31, 29, 30, 27, 31
One month ago, the most recent 10 Reform percentages were:
24, 26, 31, 25, 28, 28, 29, 31, 24, 33
The average is actually slightly higher compared to this time last month, but our graph shows a full percentage decline. Surely we can see that's an issue, regardless of who we personally support politically? You're right to mention that this kind of misrepresentation has also impacted Labour, so there's a sense in which it's "fair", but we would be better off just having an accurate trend line instead of "equally inaccurate for all." ~2026-37718-7 (talk) 11:39, 20 February 2026 (UTC)[reply]
I think you’ve made some good points there. Of course the answer is that the graph is calculating over a much longer period than 10 surveys. However I do agree that a shorter interval would provide better results which react to changes in trends more quickly.
You’re right that it’s better to be accurate than equally inaccurate for all!
I don’t think this was the point the person I replied to was trying to make, though, given they called out YouGov specifically. They might notice that the trend line is closer to e.g. FindOutNow’s Reform numbers, given YouGov is the outlier in terms of Reform numbers at the moment. Sammy zammy1 (talk) 18:09, 20 February 2026 (UTC)[reply]

Smoothing

The issue is not LOESS but excessive smoothing— File:UK opinion polling 2010-2015.svg had less smoothing and looked fine. I think we need to reduce the smoothing here and have been saying so for a while. Chessrat (talk, contributions) 19:15, 18 February 2026 (UTC)[reply]

The issue is LOESS. It is inappropriate to use future data points to smooth past points. Edge effects mean most recent points on the line (which are of most interest to visitors) behave in unexpected ways. It is repeatedly misinterpreted. It is very difficult and counter-intuitive to explain and defend. The graph linked does indeed look lovely, but we should aim for graphs which are accurate, intuitive, and defensible. ~2026-37718-7 (talk) 21:55, 19 February 2026 (UTC)[reply]

Deletion of Lord Ashcroft Poll

@Jakefolds2 Is there a reason for your repeated deletion of the recent Lord Ashcroft Poll? Please provide a justification here rather than repeatedly reverting, which may be in breach of WP:3RR. Pretzel Quetzal (talk) 18:34, 21 December 2025 (UTC)[reply]

Lord Ashcroft is a Conservative donor and the former deputy chairman of the Conservative party, which would explain why this poll is an outlier compared to other polls and which is why I don't think this is a credible poll (all political polls should be unbiased). His polling company is also ranked at the bottom of the UK Election Data Vault's 'Pollster Ratings' page, which further adds to my belief that this poll is unreliable. Jakefolds2 (talk) 19:24, 21 December 2025 (UTC)[reply]
Firstly, thank you for giving an explanation, and I'd appreciate it if you'd stop reverting edits before a consensus is reached.
Although the points you have raised are valid, I think that the Lord Ashcroft polls should be included. There is already a note explaining that the pollster is not a member of the British Polling Council, and it is stated that the poll was for the Mail on Sunday. Lord Ashcroft polls also is transparent and releases their full tables.
But I think the primary factor should be WP:RS, and as there has been considerable media coverage of their polls in the past, eg. BBC, The Guardian, I think the Ashcroft polls should be included. I also don't think its proper to re-evaluate a pollster only when a poll is released that shows "surprising" results, in this case a large Tory increase and the Green's highest ever vote share. Pretzel Quetzal (talk) 19:53, 21 December 2025 (UTC)[reply]
Not all that transparent - the tables are as clear as mud. It isn't very clear how the VI values were arrived at. What was the question? What is the methodology? What are the RS? The ones you quote are years old. The Mail is unreliable (deprecated). Mark Pack is just another list of polls (did he get the figures from somewhere or calculate them himself, like here?). That just leaves Ashcroft himself. Like all the pollsters, it's a Primary source, but is it reliable? ~2025-41479-54 (talk) 08:53, 23 December 2025 (UTC)[reply]
I agree (although for different reasons) this pollster is not reliable enough to be included BUT this has been long since discussed and the general consensus was to include them so we have to go with that. Kirky03 (talk) 19:54, 21 December 2025 (UTC)[reply]
This particular issue has been discussed at quite significant length in the past, linking below for context
Talk:Opinion polling for the next United Kingdom general election/Archive 1#BPC Polls
Talk:Opinion polling for the next United Kingdom general election/Archive 3#BPC only polls included
Talk:Opinion polling for the next United Kingdom general election/Archive 6#RfC on exclusion of non British Polling Council members
Talk:Opinion polling for the next United Kingdom general election/Archive 7#Proposal for new section on BPC membership
I'd argue the fact Lord Ashcroft is included in UK Election Data Vault is actually a sign in favour of inclusion here - it shows the polls are backed by a credible source, even if covered negatively.
The UK Election Data Vault also shows five pollsters included in this article with poorer accuracy than Ashcroft at the last election. https://electiondatavault.co.uk/election-results-2024/pollsterratings2024/, although Ashcroft does rank last on the last three https://electiondatavault.co.uk/tables/polling/pollster-ratings/
I'd personally agree with including this poll, noting the above consensus on including pollsters when reliable sources cover them, even if not a member of BPC. Eastwood Park and strabane (talk) 19:57, 21 December 2025 (UTC)[reply]
Hi, thanks for your reply.
I'm just very concerned about other people being mislead over this poll. The fact that this polling company hasn't been approved by the BPC really concerns me. Also, the fact that the Conservative party is afraid of the rise of Reform (especially at the moment due to the ongoing row about the Conservatives and Labour colluding to cancel the local elections), it wouldn't surprise me if a former Conservative deputy chairman would put out a result like this to make out the 2 parties are tighter than they actually are. Can we potentially consider removing this poll from the page on this 1 occasion? Jakefolds2 (talk) 20:12, 21 December 2025 (UTC)[reply]
The poll is an outlier but not extremely so - YouGov had Reform on 26% a few weeks ago, only 1% higher than Ashcroft. Verian, who are the top ranked pollster by the Election Data Vault had Conservative on 21%, only 1% lower than Ashcroft. In theory it's possible these numbers are deliberately thrown off, but to be honest this seems very unlikely.
Outlier polls happen all the time, and there's no reason as far as I can see to believe this is anything else - other polls have also shown Reform are down, not quite as much, but a single outlier poll is unlikely to make a difference to public voting patterns. I'd also doubt a Conservative party member would want to make the Greens appear strong. I think if a poll was to be removed, it would have to be all Ashcroft Polls, to make sure the page isn't biased against good polls for the Conservatives/Greens. Eastwood Park and strabane (talk) 20:30, 21 December 2025 (UTC)[reply]
I'd also note that YouGov was co-founded by Nadhim Zahawi, who later became a prominent Conservative politician - briefly Chancellor in 2022 - it's not too uncommon for people involved in politics to also be involved in polling. Eastwood Park and strabane (talk) 20:34, 21 December 2025 (UTC)[reply]
Regarding BPC membership, the council itself said in 2015 that Lord Ashcroft polls is not eligible to be a member because it "does not do work for multiple clients", but that "it publishes full details of its polls in much the same way as a BPC member would be expected to do".
As for speculating on possible bias within the pollster, I think a reliable source would have to accuse Lord Ashcroft of deliberately manipulating polls before we consider removing polls from the article on those grounds. Pretzel Quetzal (talk) 20:42, 21 December 2025 (UTC)[reply]
I would say removing it runs the real risk of misleading people. Happytiger00 (talk) 22:15, 21 December 2025 (UTC)[reply]
Election Data Vault's "pollster ratings" shouldn't be treated as authoritative in any way. It's a completely arbitrary measure done as a hobby project by one guy. Wikipedia should not factor this into a decision one way or the other. ~2025-40334-47 (talk) 21:43, 21 December 2025 (UTC)[reply]
I left wikipedia a long time ago due to the bias of allowed sources but I couldn't resist adding something. This poll does appear to be garbage / cooked. If you want to see why, look in the data tables. The published numbers are just cooked up with a load of people polled just discarded. If you want to see the actual results that closest mirror a general election scroll down in the dataset on the percentages tab and find the section entitled '"If you had to choose, which of the following parties would be your first choice to be in government after the next election, whether alone or in a coalition?"' Those results are as follows:
Labour 18
Conservatives 19
Reform UK 26
Liberal Democrats 13
Green Party 20
Scottish National Party 3
Plaid Cymru 1
Happy to be corrected. ~2025-40576-55 (talk) 02:42, 23 December 2025 (UTC)[reply]
Polls ask several questions and try to tease out what the relationship between them means. A common one is "who would you want as prime minister?" which always gets slightly different results from voting intention. In this case, you've just spotted something even more subtle. The question "if you had to choose, which party would you want to be in government?" is not the voting intention question (that's on row 110 of the data tables). The results are different because people might want one party to be in government but still intend to vote for a different one. ~2025-40334-47 (talk) 06:50, 23 December 2025 (UTC)[reply]
That isn't a question. The only question is on line 11. ~2025-41479-54 (talk) 08:27, 23 December 2025 (UTC)[reply]
I meant row 110 isn't a question, obviously. And looking at the corresponding table for "first choice government" (at row 210) in the Results (Values) tab, the party totals together comprise the entire sample - there are no DK/WNVs. Same for second choice party, except that magically, the aggregate total is sample +1. So all the DKs and even all the WNVs (as defined at row 110) have a preference for which party they would want to be in government? Smells a bit fishy, at the very least. But I suppose anything's possible. ~2025-41479-54 (talk) 08:43, 23 December 2025 (UTC)[reply]
DK and WNV were not available options for those questions. This is common in polling (they're sometimes called "squeeze questions") and is reflected in the wording of the question - "if you had to choose..." ~2025-40334-47 (talk) 16:59, 23 December 2025 (UTC)[reply]
As I understand it a "squeeze question" is usually a follow-up to a voting intention (VI) question and is used to make adjustments. Even then, I think there are still some who refuse to answer, at least in some polls. This appears to be a stand-alone question which elicited a 100% response. Apart from being a comparison with the previous table, what other purpose does this question have? There is no VI question as such. The initial question is about likelihood to vote (LTV) for each party; respondents rate each party out of 100 and the highest score >=50 is taken to be their VI. Thus the VI is actually inferred from the LTV. If all their responses are <50, they are classed as DK/WNV, which itself is a misnomer - they are not two separate groups, they could all be classed as "unlikely to vote" (given the turnout at recent elections, the 36% in this poll is probably not too wide of the mark). Other pollsters typically adjust VI for LTV (for any party), Ashcroft appears to estimate VI solely according to LTV (for each party). This is an atypical methodology (as far as can be seen) to say the least. ~2025-41479-54 (talk) 21:47, 23 December 2025 (UTC)[reply]

It does look a trifle suspicious, but then yougov constantly lowballs too. It should be included unless there's a source calling into question it's reliability? Ashcroft publishes his data like the others? Halbared (talk) 21:44, 21 December 2025 (UTC)[reply]

Yes, the full tables are linked at the bottom of Lord Ashcroft's article. Pretzel Quetzal (talk) 22:04, 21 December 2025 (UTC)[reply]
Every FiN poll looks questionable to me, but they should clearly be included because this is a site that shows national opinion polls. Lord Ashcrofts polls are widely quoted in other poll amalgam sites, such as Mark Pack's respected site - https://www.markpack.org.uk/155623/voting-intention-opinion-poll-scorecard/ and they should obviously be included here. Happytiger00 (talk) 22:13, 21 December 2025 (UTC)[reply]
Usual things to note: 1) polls being outliers from the 'consensus' of polling companies doesn't mean that they're wrong. In the 2017 election all polls massively underestimated Labour (the polling 'average' was about 5% out) and Survation were the outlier predicting 40%. Labour got 41%. In the 2015 election the error was the other way round: there was a strong polling consensus that the result would be 34:34 Con:Lab and it was actually 37:31. Probably we won't know which of YouGov+Ashcroft, "consensus", FindOutNow and "some other thing entirely" turns out to have had the right idea for another few years. 2) The purpose of this page is not to only list the accurate polls (it would have to remain in draft until after the election if so), but to list the polls which were reported in the media. No-one goes back to the 2017 polling page and starts suggesting removing all the non-Survation polls because they turned out to be wrong. If you want to make your own polling average where you discount the polls you think are wrong then all the data is here for you to do that. ~2025-40936-44 (talk) 09:09, 23 December 2025 (UTC)[reply]
100% agreed. As soon as we are removing polls because we disagree with the numbers, we are hitting original research.
There's an argument to be made that them not being a BPC member makes them less accurate but we've long since reached the compromise of including non-BPC polls if they are given a note to point this out and I don't think rewinding to past arguments does any good.
Ultimately, as you pointed out, we can't say whether this poll is wrong or the only one that's right and our one job is to present all the data and let people interpret that as they wish. Kirky03 (talk) 10:17, 23 December 2025 (UTC)[reply]

Unnecessary protection

One user (not a TA) does half a dozen reverts on one day and is blocked for 48 hours, but the page gets protected (from TAs) for a whole year? Ridiculous. Unexplained or questionable edits on this page have been at low and manageable levels for some time. TAs/IPs are no more responsible for them than are other accounts. ~2025-41479-54 (talk) 09:00, 23 December 2025 (UTC)[reply]

Pinging @ToBeFree —Profzed! 11:32, 23 December 2025 (UTC)[reply]
Temporary accounts and new accounts such as Verstappen08, Bear3424, ~2025-35226-55 and ~2025-40943-86, appear to be significantly more likely to be the cause of disruption in this specific article than elsewhere. The "one user (not a TA)" you refer to is Jakefolds2, who also would have been affected by the semi-protection as they hadn't made 10 edits yet. If there's something specific you would like to change, to create an edit request or click here to create an account which will be able to edit the page after 4 days and 10 contributions to other articles. ~ ToBeFree (talk) 12:52, 23 December 2025 (UTC)[reply]
~2025-40943-86 (15 Dec) was a good faith edit reverted by a bot, then unreverted by Eastwood Park and strabane.
~2025-35226-55 (20 Nov), who may have been the same as ~2025-35156-06; between them they added and re-edited a sub-sample when the article includes only full samples. So, although the motive may have been political bias, they could be given the benefit of the doubt. In any case their edits were manually reverted by a different TA, with an explanation.
Verstappen08 (15 April). Over 8 months ago.
Bear3424 (6 March). Nearly 10 months ago.
Do you have any stats to back up your claim that TA/IP/new accounts are "significantly" more likely to be disruptive? Compared with their positive contributions? I've seen a stat somewhere that estimated that 60% of Wikipedia has been written by newcomers, but I suppose it all was to start with. ~2025-41479-54 (talk) 23:01, 23 December 2025 (UTC)[reply]
There is no justification for the continued protection of this article. A year long protection in response to an isolated case of a handful of repeated reverts on a single day is an over-reaction. Two of the examples of "disruption" quoted above are several months old, another was a false positive and the other was a borderline case which did not involve repeated or retaliatory reverts. Dubious or suspicious edits in recent months have been low-level and swiftly reverted where necessary (including by TAs), with little or no definitive evidence of vandalism. Oh, and by the way, I see that the user at the centre of the recent incident is back editing this article. ~2025-41479-54 (talk) 21:22, 6 January 2026 (UTC)[reply]

Your Party

At what point do we start including "Your Party" polling results as its own grouping in the chart and graph? They are polling in the single digits in the polls they are included in but it looks like they are picking up more support than Plaid Cymru in some polls and the SNP in others. No doubt given its national rather than regional orientation. What are peoples thoughts on including them? Completely Random Guy (talk) 08:04, 6 January 2026 (UTC)[reply]

Have they been featured in any headline VI polls? I'm struggling to find. ~2025-40334-47 (talk) 09:57, 6 January 2026 (UTC)[reply]
I think they ahould be included if they come in one of the pollsters found in the opinion poll table (i.e. More In Common, NewGov, FoundOutNow, etc.), but should probably be relegated to "other" as an addendum, in that case. If they don't appear in those (regular) polls, though, I don't think making a new column for them is necessary just yet.
I think we should revisit this question once they start consistently polling at ~5%, honestly, but for now I think keeping it a seperate polling category is fine. Worst case scenario we can always just retroactively include them in the main table/graph when/if they actually matter as a national party. Reaperz 2020 (talk) 11:49, 6 January 2026 (UTC)[reply]
@Completely Random Guy the latest YouGov poll puts Others on 2% so Your Party support must be fairly negligible at the moment. I saw an article by YouGov saying that if they prompted for it at the moment it would inflate their score, and wouldn't prompt until it got to 8-10%. However I can no longer find the link which I am fairly sure is in one of the earlier discussions on here. PC and SNP do of course poll much higher figures in the country they operate in which presumably means they can be prompted for Saxmund (talk) 13:00, 6 January 2026 (UTC)[reply]
@Completely Random Guy Found it! https://yougov.co.uk/politics/articles/53692-voting-intention-ref-27-lab-19-con-18-7-8-dec-2025 Saxmund (talk) 14:40, 6 January 2026 (UTC)[reply]

New London Poll

There have been a new London-only poll from Savanta: Labour hits new record low in London with Tories pushed into fourth place behind Greens and Reform, poll finds | The Standard Gordonltycall (talk) 17:43, 6 January 2026 (UTC)[reply]

wales graph

it has been quite some time since the election and there have been a few polls conducted in wales already. is it time to give that section a graph like the scotland section? KLazyLass (talk) 11:10, 14 January 2026 (UTC)[reply]

Survation 21 Aug - 1 Sep / Summer 2025 Stack Data Strategy polls

There seems to be dispute over whether these should be included. I would maybe suggest putting them in a separate column as I believe they did have a reported seat projection, but not an MRP reported seat projection Eastwood Park and strabane (talk) 16:12, 20 January 2026 (UTC)[reply]

Pinging @Dochanddorris as they removed these polls in the first place.
I think both should be included in the MRP table. I just found the full data tables from the Stack Data MRP poll (it's linked on their website) so I'll add that. As for the Survation poll, although the data tables linked on the Survation website only includes the voting intention, the only sources that I can find that give the seat projection – an LBC article, a LinkedIn Post by the CEO of 38 Degrees, and a twitter post from Survation – all describe the poll as an MRP. Pretzel Quetzal (talk) 16:52, 20 January 2026 (UTC)[reply]
Many thanks for that, glad you were able to clarify the Stack polling and link to data tables, absolutely accept it is a legit MRP poll and in its correct place.
Not convinced about the Survation poll, clearly the data shows VI but nothing that we might expect from constituency level prediction. I wonder if we are perhaps being too narrow in our understanding of MRP and that it is used both for polling and market research, so perhaps this poll is an MRP in terms of market research but not in polling, yet the X post from Survation does initially seem clear "our MRP results" but goes on to say "The seat totals predicted by our model had:" To be honest I am now fairly uncertain, which i suppose is movement on my part just not as clear as i would have liked. Dochanddorris (talk) 22:39, 20 January 2026 (UTC)[reply]

Poll spamming warping the data

Should polls be limited to 1 entry per polling company per month or over 4 weeks? I note Yougov sourced polls are being added weekly but Yougov is notoriously biased for the way it collects data via online paid polls and is usually skewed so it should really be treated as an outlier. Another idea would be to include the polls but only include a single value for the last 4 weeks which is an average of all their polls e.g. Yougov would get (24+24+26+25)/4 = 24.75 for the last 4 weeks. This way Yougov won't weight the graph incorrectly because it has 4 entries in the current value. Don't know if this is intentional on the part of people using Yougov or whoever is adding the polls but doesn't seem to be reflecting reality when most recent by elections have Reform getting 40% and Labour getting 2-3% of the vote for example. Zonjinn (talk) 15:34, 21 January 2026 (UTC)[reply]

People vote very differently in local elections than in general elections, and even more so in by-elections for local government. Yougov put out a weekly poll the same as Find out now, and More in Common. One could argue that Find out now have been the most out of line whith other pollsters, but I guess it all balances out. Happytiger00 (talk) 15:52, 21 January 2026 (UTC)[reply]
We can't get into doing original research. We can't do anything beyond a simple calculation, as per WP:CALC. Bondegezou (talk) 16:08, 21 January 2026 (UTC)[reply]
I personally wouldn't call YouGov in particular biased, they had one of the best ratings at the last election. [1] I'd be interested in discussion about including too many polls from the same pollsters on the chart, potentially throwing things off, but individual council by-elections have tiny voter numbers leading to odd results sometimes - for a counter-example the recent Barrhead, Liboside and Uplawmoor by-election saw Labour's share increase by 22% [2] Eastwood Park and strabane (talk) 16:13, 21 January 2026 (UTC)[reply]
Yes. YouGov are about as average as "attempting-to-do-it-properly" pollsters get, really: hits in the years everyone was generally right, misses in the same direction as everyone else the rest of the time. This is the first time they've really been an outlier from the pack, which is interesting.
I do think it's true that this page shouldn't be averaging ratings from different pollsters at all - WP:CALC has an explicit "Editors should not compare statistics from sources that use different methodologies." clause which averaging a YouGov poll with an Opinium poll with a FindOutNow poll clearly breaks, and if it gave each pollster its own graph line for each party the polling frequency issue would disappear - as would most of the arguments about averaging frequency and smoothing method, as a simple unsmoothed point-to-point line for each separate pollster would generally be fine. ~2025-40936-44 (talk) 10:00, 22 January 2026 (UTC)[reply]
Nice try… I think you made a valid point that the graph will be more heavily weighted to frequent polls. However you gave yourself away when you started quoting random by-elections as evidence for your point…
Obviously what you describe would be original research and so not allowed. It’s also obviously not an intentional attempt to mislead, it’s just a collection of data. If you were to combine YouGov / MoreInCommon / FindOutNow (because let’s not single out YouGov purely because you don’t like its results) into one monthly data point, you would instead be excessively weighting infrequent polls - which, notably, have less data to verify its reproducibility.
Taking a moving average of recent polls is effectively what the trendline does anyway. It’s already calculated over a long period so this wouldn’t have a significant effect beyond making it even laggier. Sammy zammy1 (talk) 18:11, 21 January 2026 (UTC)[reply]
The trend line currently does not take a moving average of recent polls. It uses an obscure statistical method called LOESS, which attempts to display trends by fitting polynomial lines to the data. In the case of poll results, it produces bizarre results - for example, if a party's share goes up for a while and then levels off for a while, the graph will actually show a downturn. ~2026-38329-6 (talk) 12:29, 22 January 2026 (UTC)[reply]
No.
I think every one else has made enough points as to why not so I refer you to their comments as my justification. Kirky03 (talk) 18:27, 21 January 2026 (UTC)[reply]
"Shall we ignore the results of polls from the country's best known pollster and instead make up numbers ourselves to insert into the encyclopedia that thousands of people read?" aesurias (ping me in your reply, or I won't see it) (talk) 23:38, 21 January 2026 (UTC)[reply]


Opinium poll Feb 2026 - SNP and PC omitted from headline VI

Judging by a random selection of about half a dozen previous polls, this hasn't occurred in the recent past. The percentages for the other parties, including Others, are as per headline VI table, those for SNP and PC are from the table "All who might vote and giving consent", which includes 17% DK (n=300) and 3% WNV (n=50) (difference between the weighted bases is n=347). Calculations for SNP and PC:

  • Excl. DK+WNV / With DK+WNV
  • Weighted base 1437 / 1784
  • SNP 2.37% / 1.9%
  • PC 0.7% / 0.56%

So rounds to 2% and 1% in both cases. Should the efn be applied to just those two parties rather than the poll as a whole; or perhaps a different efn to clarify that they are omitted from the headline VI? ~2026-47618-2 (talk) 18:42, 8 February 2026 (UTC)[reply]

I had noticed that. SNP reported on national UK level but subsample of Scotland shows them on 2% and 2 responses, also noticed that subsample for Scotland only adds up to 70%, so kind of suspect typo. Have let Opinium know of issue Dochanddorris (talk) 16:33, 10 February 2026 (UTC)[reply]
Date(s)
conducted 		Pollster Client Sample
size 		Lab Con LD Grn Ref Others Lead
17–23 Dec 2025 Savanta[1] QMUL 1,006 30% 17% 13% 17% 21% 2% 9

~2026-47618-2 (talk) 19:11, 6 February 2026 (UTC) ~2026-47618-2 (talk) 18:56, 8 February 2026 (UTC)[reply]

May I respectfully suggest that opening a 'talk' topic to announce an update to the data tables is both unnecessary and a misuse of the Talk feature.It says so right at the top of the page
The talk page is already difficult to follow due to it being misused in this manner Dochanddorris (talk) 16:29, 10 February 2026 (UTC)[reply]
Unfortunately this is an inevitability now due to the block on IP users editing the page, see the "Unnecessary protection" section above Eastwood Park and strabane (talk) 16:33, 10 February 2026 (UTC)[reply]
Ahhh, thanks for that.So talk page is not protected that seems a bit of an omission, not sure I agree with protection but if going to do it lets do it right
Its like that game of whack a mole Dochanddorris (talk) 16:40, 10 February 2026 (UTC)[reply]
My mistake. I should have prefaced the changes with something like "This is a suggested edit", which is a perfectly normal use of the Talk page, particularly when the article page is semi-protected. There are errors in the date and some of the party share fields in the current table entry. Also, cite can now link to data tables as above. ~2026-47618-2 (talk) 15:59, 11 February 2026 (UTC)[reply]
I've added this poll, how does it interact with the previous one reported 15-21 December? Eastwood Park and strabane (talk) 16:14, 11 February 2026 (UTC)[reply]
Well, I had assumed that they were the same poll. I have to admit that I hadn't looked at the Standard article, but having done so now, I see that the date and figures in the previous table entry were as reported. There are also differences between the article and Savanta table in the figures for the %s of GE2024 voters switching parties. On the other hand, the figures for "important issues" are identical in both. It seems unlikely that there would be two overlapping polls for the same client, but there's no obvious explanation for the discrepancies. Could the Standard have been given provisional data? The Savanta tables were published 3 days after the article. The Standard even has a quote from a Q. Mary researcher, saying that the Greens are ahead of Cons, when the raw data has Green n=116, Con n=117. Is it conceivable that it's the spreadsheet that's in error? It presents a bit of a dilemma for Wikipedia - or maybe a trilemma - leave the table with two polls, or choose one of them, in which case the choice is between a reliable secondary source and the pollster's own table, which (let's be honest) is a primary source. Perhaps an explanatory efn could be added. Anyway, I shall have to leave it up to you and fellow editors how to proceed! ~2026-47618-2 (talk) 18:13, 11 February 2026 (UTC)[reply]

References

Restore Britain

In June 2025, former Reform MP Rupert Lowe launched a political organisation called "Restore Britain", advocating a harder line on immigration than Reform UK. The organisation has been backed by the billionaire businessman Elon Musk. Find Out Now were asked by the organisation to poll on a hypothetical new party led by Lowe.

Not really sure how that bit is relevant? Putting aside the fact it isn't sourced, Musk's involvement wasn't stated in the poll. ~2026-10256-39 (talk) 00:21, 15 February 2026 (UTC)[reply]

I was the one who added this, as it was a very prominent part of the Restore Britain section on Lowe's page and I thought (at the time) it was helpful added context. I don't object to its removal now Restore has a separate page. Eastwood Park and strabane (talk) 11:10, 27 February 2026 (UTC)[reply]

Good Growth Foundation poll date discrepancies

Just recently, a voting intention poll conducted by a new poll org, the Good Growth Foundation, was added to the polls table in the article. There is seemingly an issue though. The graph for this poll on the GGF Insights page of the website gives the fieldwork dates as 13–16 February. Meanwhile, the Polling Archive page of the website (and in the data tables), though titling it as a February poll, list the fieldwork dates as 13–16 January! — Jamie Eilat (talk) 21:27, 19 February 2026 (UTC)[reply]

Now, the fact that the poll was published in February definitely suggests that it is indeed a February poll, & that the instances of "January" here are the typos, rather than the other way around, but the fact that "January" is what's given in the data tables leads me to give some pause. — Jamie Eilat (talk) 21:33, 19 February 2026 (UTC)[reply]

Second place shading in election results

When second place shading was introduced, a slightly darker shade of grey was used for second place in the election result rows, which have a slightly lighter shade of grey as background. I've noticed that some of the sub-national tables have not highlighted second place in their election result rows, namely London, Greater Manchester and all the constituency polls. For reference:

  • 2nd place - poll results: #e8e8e8   
  • 2nd place - election results: #d9d9d9   
  • background shading - election results: #e9e9e9   

~2026-47618-2 (talk) 02:48, 21 February 2026 (UTC)[reply]

Actually, it's worse.
London, Greater Manchester: election result rows use #e8e8e8 for 2nd place (should be #d9d9d9).
Gorton and Denton:
poll result rows - 2nd place has #e9e9e9 (should be #e8e8e8)
election result rows - no 2nd place shading (should be #d9d9d9)
Hypothetical (with Burnham):
poll result row - no 2nd place shading (should be #e8e8e8)
election result row - no 2nd place shading (should be #d9d9d9).
Runcorn and Helsby:
poll result rows - no 2nd place shading (should be #e8e8e8)
election result row - no 2nd place shading (should be #d9d9d9). ~2026-13144-75 (talk) 19:05, 27 February 2026 (UTC)[reply]

Omnisis Gorton & Denton Poll

Should we include this poll for upcoming Gorton & Denton by-election… just checked and Omnisis are not members of BPC and most worryingly there’s reports they, or at least those who commissioned the poll, donated £10k to the Greens - very least needs a caveat that not BPC member but personally would remove entirely NewGuy2024 (talk) 08:54, 21 February 2026 (UTC)[reply]

The funder commissioning the poll not being politically neutral is unremarkable (lots of polls are commissioned by newspapers, for example). The pollster not being is less so, but Lord Ashcroft is an obvious counter-example and not in itself grounds for removal if the poll itself has been done competently and neutrally.
The poll itself has been reported on in various secondary sources (https://www.manchestereveningnews.co.uk/news/greater-manchester-news/everything-know-new-election-poll-33464877, https://www.thenational.scot/news/25873405.gorton-denton-by-election-poll-greens-beat-reform-labour/, probably others) which report on it as if it were a reasonable-quality poll (specifically mentioning demographic weighting)
According to Omnisis on their primary source page - https://www.omnisis.co.uk/poll-result/constituency-polling-for-by-election-in-gorton-and-denton/ - "Omnisis are members of the The British Polling Council and abides by it’s rules and regulations for published opinion polling" and that page also includes the standard BPC "error margin" boilerplate. The sample size is small but it's transparent about that and it's still quite a bit larger than the other constituency poll.
I don't see any grounds for removing it, without at least some reliable sourcing for "Omnisis are not members of the BPC" and "Omnisis donated to the Greens". Anyway, we'll know in about a week what the actual results are, so it'll be easy at that point for people to compare them with the poll and make their own judgements about how reliable Omnisis are or aren't. ~2025-40936-44 (talk) 12:02, 21 February 2026 (UTC)[reply]
there’s no mention of Omnisis here as being members of BPC!
https://www.britishpollingcouncil.org/officers-members/ NewGuy2024 (talk) 12:20, 21 February 2026 (UTC)[reply]
There is, it is listed as We Think/Omnisis on the link you provided. It's the fourth from the bottom, if you can't find it. Pretzel Quetzal (talk) 12:35, 21 February 2026 (UTC)[reply]

Semi-protected edit request on 23 February 2026

"JL Partners" in the Pollster column for the 4-12 Feb poll in the 2026 polls table should be changed to "J.L. Partners" for consistency across the page. Althezar (talk) 08:35, 23 February 2026 (UTC)[reply]

 Done, sort of. I have made the name consistent throughout the article with UK styling. Halbared (talk) 10:01, 23 February 2026 (UTC)[reply]

Find Out Now poll, including Restore Britain, 20-21 Feb. 2026

This is here [3] ~2026-91714-6 (talk) 21:45, 26 February 2026 (UTC)[reply]

I'm afraid there's a slight mistake in your link - the pipe symbol would not be needed to link from the word "here", in which case the first "here" is not needed, or aternatively, link from the "[3]", in which case the pipe and second "here" are not needed. But all that is no matter, as the link is already in the table, under sub-heading "With Restore Britain" in the Hypothetical Scenarios section, although there has been some disagreement about which table it should be in (see topic below). ~2026-13144-75 (talk) 19:34, 27 February 2026 (UTC)[reply]

The Most Recent Poll With Restore Britain Is Not "Hypothetical"

To quote the polling company, "The methodology is exactly the same as our (see below), with the only difference being the inclusion of Restore Britain and Your Party as explicit vote options." WriterOfScrolls (talk) 03:25, 27 February 2026 (UTC)[reply]

I think the way that we handled the Your Party polls, which is a similar situation as its also a newly establish party, is that any polls released as one-off polls including Your Party would go in the Hypothetical section. In addition, Restore Britain hasn't yet registered with the Electoral Commission. Only the regular voting intention polls should be included in the main table. Pretzel Quetzal (talk) 03:35, 27 February 2026 (UTC)[reply]
Is there a bit of a grey area here? Different criteria have been mentioned. The argument in favour of including it in the main table is that the prompts were not hypothetical, compared with the case of Your Party, where the prompts were either framed hypothetically or qualified with a potential leader's name. Various reasons are given for treating the poll as hypothetical (and implicitly overriding the foregoing):
  • the poll includes a new party (well, two actually)
  • the party is not registered with the Electoral Commission (allowing it to stand candidates in elections)
  • the poll was commissioned by the same party that is the subject of the alleged hypothetical scenario
  • it is an one-off poll in addition to the pollster's "regular" polling (confirmed in the pollster's own report).
But do any of these alone make the poll hypothetical? For example, does lack of registration alone make it hypothetical, if pollsters simply include the party as an option, without any caveat or qualification?
By the way, the poll is also included in the table under sub-heading "Including Your Party prior to formation" as well as "With Restore Britain". Does it need to be duplicated in this manner? ~2026-13144-75 (talk) 19:26, 27 February 2026 (UTC)[reply]
I think the most important criteria are whether it is registered with the Electoral Commission, and whether it's a one-off poll. If the party isn't registered with the Electoral Commission it's clearly hypothetical as no voter could vote for that party if a general election were held tomorrow. Some Your Party polls were explicitly framed as a hypothetical, but there are many in the table currently that were one-off polls which included Your Party.
I reckon it's fine to have them duplicated, as the poll would fall equally under both sections. Pretzel Quetzal (talk) 19:38, 27 February 2026 (UTC)[reply]
There has been at least one poll which included Your Party and which went in the non-hypothetical polling section, since it wasn't some one-off thing where the prompt was "If there was a party lead by Sarah Zultana and . . . "
Just like in this case, it was just a normal poll conducted by a reputable pollster that included the additional party. WriterOfScrolls (talk) 07:20, 28 February 2026 (UTC)[reply]
The Find Out Now poll is a one-off poll and they have stated it is not one of their regular voting intention polls:
Find Out Now has run a new voting intention survey, with the same methodology as our ongoing voting intention tracker, but with Restore Britain and Your Party included. [4]
Additionally, Restore are not yet registered with the Electoral Commission [5], so any poll taken before they are registered is inherently hypothetical as you physically cannot yet vote for a "Restore Britain" candidate.
I don't believe that the fact Restore Britain commissioned the poll themselves is (by itself) a reason to exclude it from the main table, but it does add to the evidence the poll is not a regular voting intention poll.
The poll including Your Party in the main table that you mentioned is Electoral Calculus's regular quarterly MRP poll [6], a regular poll which happens to include Your Party. It's worth noting that the 2024 Q4 version of this poll [7] also included Your Party, but is included in the Hypothetical section as it was conducted in September before Your Party registered with the Electoral Commission.
Based on this I am removing the poll from the main table. Pretzel Quetzal (talk) 12:40, 28 February 2026 (UTC)[reply]
Agreeing here that Electoral Commission registration should be the line - once this party is registered these polls are no longer "hypothetical". Eastwood Park and strabane (talk) 12:48, 28 February 2026 (UTC)[reply]
allikas: https://en.wikipedia.org/wiki/Talk:Opinion_polling_for_the_next_United_Kingdom_general_election