Talk:Opinion polling for the 2019 United Kingdom general election/Archive 2

Regarding Leaving Party Leaders
There seems to be some edit-warring going on, so I figured we should talk about it. Until a consensus can be reached, I propose that the current state (entering leaders shown; leaving leaders not). However, I think it's quite reasonable that we should include leaving party leaders as an event only until a new leader from that party enters. I think it's significant enough that a party leader is leaving to warrant one line, but not two. For example, Henry Bolton's no-confidence vote should be included for now, but removed once UKIP selects a new leader. UpperJeans (talk) 20:08, 5 March 2018 (UTC)
 * No. Consensus has generally been not to include leaving leaders. Most polling articles don't even include incoming leaders. This is a list of opinion poll data: it is not a timeline of political events. Bondegezou (talk) 10:09, 6 March 2018 (UTC)
 * No. I think we should keep the article very simple (considering some good faith contributors already get confused) a new policy of adding and removing information, might result in more misleading disputes between editors. -- BOD --  11:39, 6 March 2018 (UTC)

Polls of individual constituencies ... Bath
I wonder if single poll of one unique small city is a bit misleading on a page about national political opinion polls. London and the regions are more logical subsections. -- BOD --  11:45, 6 March 2018 (UTC)
 * It appears to be a valid poll, so I would say it warrants inclusion. More individual constituency polling may emerge in time. Until then, it's a bit odd having it all on its own there... but I can't think of any better place for it. Bondegezou (talk) 12:08, 6 March 2018 (UTC)
 * There are two ways this has been dealt with previously — for the last election, constituency polls were included in the article that then held this name. For the 2015 election there was the Opinion polling in United Kingdom constituencies, 2010–15 article. There were a lot more constituency polls for that Parliament (because Ashcroft was still doing polls; if you look at how well those polls performed it's easy to guess why he hasn't published constituency polling for a while), so it makes sense that they have their own article. I'd be in favour of keeping constituency level polls in this article for the time being at least, given that it's 'Opinion polling for the next UK general election' rather than 'National opinion polling for the next United Kingdom general election'. Ralbegen (talk) 21:03, 6 March 2018 (UTC)


 * Aye, I included it on the basis that for the 2017 election (when few constituency polls were conducted) we included constituency polls on the main page. If the number of polls reaches high levels I'm totally up for moving it to its own article like the leadership approval polling FriendlyDataNerdV2 (talk) 17:35, 8 March 2018 (UTC)

Mistake in Lib Dem value
On the polling graph why is there a Lib Dem orange dot above 35 percent? could that be fixed? there is no poll on the polling chart saying they are that high — Preceding unsigned comment added by 2001:8000:1AEF:DF00:B02F:9CA9:D58E:929D (talk) 05:17, 23 March 2018 (UTC) There was a mistake in the Lib Dem value which was spotted by FriendlyDataNerdV2 and I have now fixed the graph. Absolutelypuremilk (talk) 13:21, 23 March 2018 (UTC)


 * There is also another error in the graph - the Survation poll from the 7th March has been put on the graph as 37%/44% instead of 38%/45% for Tory/Labour. — Preceding unsigned comment added by 129.215.4.161 (talk) 10:14, 29 March 2018 (UTC)
 * Thanks, now fixed. Absolutelypuremilk (talk) 12:52, 29 March 2018 (UTC)

Scotland Graph
The graph should only show polls following the 2017 general election as this page is for polling for the next election, not previous ones UpperJeans (talk) 19:22, 4 April 2018 (UTC)

Ditto for Wales if we have to have Wales and Scotland graphs. I would personally argue we dont have enough data post 2017 for graphs yet. Paulharding150 (talk) 10:18, 5 April 2018 (UTC)
 * It's not the end of the world if the figure includes earlier data. A bit of context isn't a problem. I'd like smoothed lines though. Bondegezou (talk) 12:07, 5 April 2018 (UTC)
 * I would agree that the graphs should only have post-2017 election polls. That's what this article is about: polls for the next UK election, not the last one, after all! FriendlyDataNerdV2 (talk) 13:25, 7 April 2018 (UTC)

Cabinet Changes
Should polling include the movements of cabinet? Reshuffles, scandals, removals? Could potentially effect public polling and opinion if such a high office (Chancellor, Home Sec., Foreign Sec.) resigns? Could sway Brexit voters or drive others to support other parties? Opinions? TomPumpkin69 (talk) 11:25, 30 April 2018 (UTC)
 * No. That would be WP:OR. Consensus to date has not been to include such things. In practice, such things tend to have very little impact on polling. Bondegezou (talk) 12:53, 30 April 2018 (UTC)

Table
Anyone know why the table suddenly expanded and the party names are suddenly left-aligned? No-one's changed it on this page, and it seems to have happened on other pages too. It looks ugly, can we change it back? FriendlyDataNerdV2 (talk) 21:29, 11 May 2018 (UTC)
 * See WP:VPT. Mélencron (talk) 21:57, 11 May 2018 (UTC)


 * Whoever fixed the table, nice job - looks much cleaner and easier to read, well done. FriendlyDataNerdV2 (talk) 09:54, 30 May 2018 (UTC)

NI
Just to make people aware - I see that people keep reverting edits to Survation polls. Just to make clear, the changes being made are not "errors". We remove Northern Irish respondents to bring Survation polls in line with the majority of other pollsters so that the numbers are comparable with a) other polls and b) The GB-wide election result. We have discussed this before and settled on this. It's not an error, it's deliberate. FriendlyDataNerdV2 (talk) 10:38, 14 May 2018 (UTC)


 * - Read this for an explanation, please stop reverting my edits. FriendlyDataNerdV2 (talk) 12:42, 14 May 2018 (UTC)


 * - Just letting you know about this talk page so we can discuss this here and not in the edit section. This is more conducive to polite discussion. FriendlyDataNerdV2 (talk) 14:24, 14 May 2018 (UTC)

Dienumbers answering: Frankly, I am amazed that anyone on here should attempt to adjust the figures provided by any polling organisation. Each organisation has their own methodologies, biases, sampling errors etc and some (or perhaps all of them) have attempted to make adjustments to their own input/output and that might improve the accuracy but Wikipedia is meant to be an independent information portal. Any attempt to adjust survey figures or inconsistencies (however well meant or justified) can only muddy the waters further. I noticed that the correct "CON 41%, LAB 40% - CON lead 1%" had been entered initially. Then someone wrongly adjusted this to 40/40 and level which has already led to some misreporting of this "tie" by some of the media. Survation's source figures are correct and there is not even a rounding error. There is a clear 1% CON lead. So, I attempted to change them back to the correct figures but there doesn't seem to be an editing tool for the background colours, as I would have completed the editing properly. I hope someone else will do this now. — Preceding unsigned comment added by Dienumbers (talk • contribs) 15:06, 14 May 2018 (UTC)
 * As explained, we have already established a consensus that we remove Northern Irish respondents. Survation is the only organisation that includes them, and even they remove them when they compare themselves to other polling organisations. The numbers in the Survation poll are, to 2 decimal points, Con 40.7%, Lab 40.2%. When the 15 Northern Irish respondents (all of whom vote 'Other') are removed, the Great Britain figures are Con 41.3, Lab 40.7 - which rounds to Con 41%, Lab 41%. It's not adjusting the figures, just removing redudant data that gives a misleading picture. If the majority of pollsters included Northern Ireland, we wouldn't bother with this - it's only because Survation are the exception that we make this change. FriendlyDataNerdV2 (talk) 15:27, 14 May 2018 (UTC)


 * As I said above (in July 2017): "And the point I'm trying to make is that we are including the Great Britain results in the table...and then inviting readers to compare them to the latest poll, which might not be a GB-only poll. If we're going to put polls covering different geographic areas in the table, without adjusting them, we can't have that set of results because people will assume they represent changes from that GB result when they simply don't." I feel this sums up my position. Also, the Con lead is 0.5% (6 respondents) - "clear" isn't the word I would use. FriendlyDataNerdV2 (talk) 15:29, 14 May 2018 (UTC)


 * - hey folks, you were involved in the original discussion to remove Northern Irish respondents so I thought I would tag you in (I hope that's okay). This is hardly the biggest issue in the world so I'm happy to let it go if the consensus is to stop removing NI respondents, but it seems to be one user who keeps reverting my edits, so I'd rather not change consensus based solely on that. FriendlyDataNerdV2 (talk) 16:18, 14 May 2018 (UTC)
 * Polls including Northern Ireland and polls just of Great Britain are not directly comparable. We thus have two options: (1) list Survation polls separately to all other (GB) polls; (2) re-calculate Survation poll numbers to exclude Northern Ireland. Consensus to date has been to do the latter, which I believe can be done accurately given the data in the published polling tables. As long as we explain all that, that still seems the best option to me. I wouldn't object to switching to the former option (listing Survation separately). Bondegezou (talk) 16:36, 14 May 2018 (UTC)

Dienumbers again: Well I see that someone else has just corrected the figures to CON 41% LAB 40% = 1% CON lead, which is as they "must" be. Whatever decision was taken before brings the whole question of Wikipedia's independence into the frame. Quite simply, the figures should not be altered from those that the survey company provided. They have already made their own adjustments. To tamper with those who voted "other" is wholly wrong. You cannot make any assumptions and should never make adjustments like this. It smacks of a political bias which, whether it does or does not occur, can only blemish Wikipedia's reputation. The whole point of having a table representing "all" the main polling companies is to make comparisons, accepting the different survey and processing techniques used. Most of the polling companies made some adjustments before the General Election to try to eliminate the overstating of the LAB vote - now several of them feel that they took these adjustments too far. For Wikipedia to further play around with these figures is totally inappropriate. Has anyone spoken to Survation about this misrepresentation of their figures? How do they adjust their own figures etc? There is some information on this on their own website. This practise of adjusting survey results must stop immediately. At the very most, you could put a note under the table about the inclusion of NI responses in Survation's polling figures but even then I would be very concerned that you're misleading all your readers. As I said, whenever we see "other" you simply cannot guess on how that breaks down - in fact the unusual circumstances without a Government being formed there, probably makes their public's voting intentions even more uncertain but if a polling company includes the NI opinions in their surveys, then we "must" accept them in the format that they process their results. Quite obviously, a number of Wikipedia users feel the same way about this as I do. So, please do the sensible thing and do not interfere with this data in any way. If you query any figures, then take it up with the polling company. I hope that my point has been made clearly. It is not an opinion, it is a question of presenting the facts accurately. — Preceding unsigned comment added by Dienumbers (talk • contribs) 16:44, 14 May 2018 (UTC)  Dienumbers (talk) 07:34, 15 May 2018 (UTC)


 * "This practise of adjusting survey results must stop immediately."
 * Wikipedia doesn't work like that. You can't just order us to stop doing something that has a clear rationale and logic to it. You have to convince people. As I've said, if there's a strong broad feeling against this, I'm happy to let it go. It's hardly the end of the world. But I think it helps inform people about the Great Britain result and provide some consistency, and thus far, people have agreed with me. I'm willing to listen to a fair argument, but please assume good faith here and don't accuse people of political bias just because they differ in their view of the most accessible way of presenting data. FriendlyDataNerdV2 (talk) 16:55, 14 May 2018 (UTC)


 * Furthermore, I should say that you don't seem to be reading what I've typed. You write: "As I said, whenever we see "other" you simply cannot guess on how that breaks down" - but that's not what we're doing. I'm going into Survation's own tables, which they have published, and simply subtracting the 15 'other' responses that specifically come from the Northern Ireland sub-sample. Then I recalculate the percentages and round them to the nearest decimal point. It's not guesswork, nor is it 'adjusting' a poll - it's simply restricting the polls in the table to Great Britain results only. I feel I should make that clear so that everyone is on the same page. FriendlyDataNerdV2 (talk) 16:57, 14 May 2018 (UTC)

Dienumbers - further reply: I'm becoming increasingly concerned about your responses. So far you have claimed that my approach is "not the way you do things on here" which is extremely unhelpful to say the least. Then you say that I'm accusing you (or others) of making this political when I actually chose my words very carefully. I actually said: "It smacks of a political bias which, whether it does or does not occur, can only blemish Wikipedia's reputation". I was referring to the impression that readers would get. Then you tag some users who supported the earlier decision, which I have been most strongly questioning - which will hardly present a balance of views and is pretty undemocratic! Then you say that I'm the only one who objects to this decision and yet at least one other user has been changing the figures - I only did that just once 9which drew your comments) but it's happened at least twice since then and also beforehand. So a number of times and I do not know whether these were all done by another single user or by several users? Then you say that unless I make an effective argument you see no reason to change.

So this time I shall make some further points that "should" persuade all the doubters. I shall "assume" that we all have the common goal of wishing to endeavour to present accurate and unbiased figures and I shall give you the benefit of the doubt that you decided to make the changes in the interests of achieving those goals. This is surely one of the basic principles, ethics and values of Wikipedia maintaining an independent and non political approach? If I'm wrong in any way about this, then please say so.

Now my main point. The purpose of these polls is an attempt to reflect how the electorate might vote if there was a General Election right now. In a General Election, everyone who is qualified to and registered to vote can vote. That includes England, Wales, Scotland and Northern Ireland. If you exclude any one of those countries, you are not reflecting a balance of voting intention across the whole nation. The fact that NI has a different set of parties is not the key matter here. There have a very approximate parallel. Yes, in a poll which states CON, LAB etc they can only realistically vote for "other" but I dare say a few might still state their favoured UK party (outside of NI). Quite apart from my argument that it is wholly wrong to adjust the figures at all (which you simply do not seem to accept), we now have a case of the correction leading to a false adjustment. In the current set up (going by the 2017 results) you are not only removing the 6.6% DUP lead (with their 36% share) but also the UUP (10.3% share) where most of the votes would go to CON. Now SF never take up their seats in Westminster - so their 29.4% vote share overstates the more left intentions. The SDLP are the closest party to a LAB equivalent but only 11.7%. Alliance may be closer to a LIB DEM vote and others could be anywhere on the spectrum. Perhaps on the balance of total votes (if SF counted) we may be looking at around a 50/50 split between left and right. Given the first past the post system that we have, these minor parties/votes would be less relevant and we know the NI make up of their share of Westminster seats (before subtracting SF from the equation). Now, if I've lost anyone along the way, that merely proves what a lottery this all is and how dangerous it is to tamper with the figures. The main point is that by removing the NI figures from the Survation statistics will always slightly reduce the CON share and slightly increase the LAB share (based on 2017 proportions).

If we're taking all of our own political interests out of this (as we should be) then surely you can see that your adjustment produces a permanent error, albeit a modest one. It looks worse right now, given the difference between a tie and 1% CON lead. There is actually a very compelling argument that Survation have got this one right and that the other polling organisations are wrong to leave out NI altogether. Then I return to my original points. We simply do not know exactly what bias corrections and adjustments all the polling firms have done. Yes, they explain some of it on their websites but not all of it (and not detailed figures for any adjustments) and you may be distorting or exaggerating any biases by making your own adjustments. You never answered my questions about whether you had queried Survation about their reasons for including NI voting intentions but I bet they're very much along the lines that I've suggested. I can only assume that you have not contacted Survation (again please correct me if you have and elaborate). By maintaining this most unfortunate decision to adjust the Survation figures, you are guaranteeing an unrealistic result and it is that which I passionately believe is against Wikipedia's most basic principles.

So, there it is - I've given up quite a bit of time to clarify my points. Please now re-set the figures to the true ones. Thank you in anticipation. — Preceding unsigned comment added by Dienumbers (talk • contribs) 20:59, 14 May 2018 (UTC)  Dienumbers (talk) 07:34, 15 May 2018 (UTC)


 * Thank you for giving me the benefit of the doubt, as yes, I am also motivated by a desire for good and clear information. Your passion is very clear and your arguments fair, thank you for your contribution. You make good points and I'm happy to leave the numbers (i.e. 41-40 etc) as they are, in line with your suggestion. Thank you for engaging in the discussion and explaining your view - I appreciate it. Kind regards, FriendlyDataNerdV2 (talk) 22:15, 14 May 2018 (UTC)
 * , your argument does not make sense to me. We're not guessing how NI voters would vote in GB. We are simply removing them and presenting GB-only results. We do this in order to allow the comparison with other pollsters, who only poll the GB. If we leave the Survation poll being of the whole UK while other polls are for just GB, then we can't have them in the same table. They are not measuring the same things. Bondegezou (talk) 22:48, 14 May 2018 (UTC)

As in the previous discussion, I support changing the Survation polls. FriendlyDataNerdV2 and Bondegezou have summarised my reasons fairly well. Absolutelypuremilk (talk) 06:37, 15 May 2018 (UTC)

Thank you @FriendlyDataNerdV2 for your kind words and I am delighted that you now agree to leave the Survation figures unadjusted. Perhaps I hadn't made my points clear enough initially. Kind regards Dienumbers (talk) 07:14, 15 May 2018 (UTC)

Well @Bondegezou please re-read what I said in my long explanation as you must be missing the main points. Surely you must realise that there are all sorts of minor and not so minor variations between how all the different pollsters and many of these lead to variations in the final figures which are larger than the NI adjustments - so I agree that "they are not measuring the same things" but that applies to "all" of them and not just Survation. The attempts at the pollsters correcting their own biases has been swinging like a pendulum. The table of "all" the main polls throws up some of these variations. Any attempts by Wikipedia members to try to adjust any set of figures must be wrong. We must leave that up to the pollsters themselves. They know what they are doing, they know that they come in for a lot of scrutiny and criticism and they are constantly tweaking their methods to iron out biases in their strive for greater accuracy. When a pollster regularly churns out figures that are outside the general consensus, one starts to query them but in fact, just going by the 2017 General Election results, Survation's last few polls were closer than most of the others. This is yet another reason for leaving all the figures as they are. I hope that you now agree. Dienumbers (talk) 07:14, 15 May 2018 (UTC)

I see that we have the reverse problem from yesterday @FriendlyDataNerdV2. After you changed it back to 41/40 somebody in the last hour (0730+) has changed it back again to 41/41! Dienumbers (talk) 07:34, 15 May 2018 (UTC)
 * I have removed all the Survation polls from the article while they are under dispute. You can't have the most recent poll not being adjusted, but all the others being adjusted. We need a decision one way or the other that applies to all of the Survation polls.
 * If we want unadjusted Survation polls, they can go in a separate UK polling table as opposed to a GB polling table. The UK is not the same as Great Britain. They are not trying to be the same thing. We should not pretend that they are the same thing.
 * Yes, obviously pollsters all take slightly different approaches, but they are trying to measure the same thing in the case of the GB polls, i.e. the voting intentions of the GB electorate. Survation are trying to measure something different (the voting intentions of the UK electorate). Given the numbers Survation make available, we can adjust them to be comparable. argues that, "Any attempts by Wikipedia members to try to adjust any set of figures must be wrong." That statement wrong. We have policy on that, which is WP:CALC, which is what we previously agreed applied. However, if we don't have consensus on that now, we don't have to do it. Bondegezou (talk) 09:32, 15 May 2018 (UTC)

I'm sorry @Bondegezou but I couldn't disagree more with your argument (which has no real substance to it). You completely ignore the key points that I made. Now, if I correctly read what you just said above, it seems that this practice of adjusting pollsters' figures is occurring more generally and not just for Survation. This makes a complete mockery of reporting the different pollsters' surveys. I cannot believe that this is happening. If this is "Wikipedia policy" then it completely stinks and the public have a right to know that Wikipedia is tampering with carefully constructed surveys and figures and not providing independent and accurate information. @FriendlyDataNerdV2, please reassure me that this is not general practice. How can anyone outside of a polling company know how to make tweaks and adjustments to those figures - you do not have the data, the bias corrections that have already been applied or any reasoning behind this. To remove the recent Survation data is completely out of order. We may as well remove the whole table and close down this service (which we should not of course) - what an appalling situation. — Preceding unsigned comment added by Dienumbers (talk • contribs) 10:00, 15 May 2018 (UTC)

This controversy illustrates why it is a really bad idea to publish different figures for a poll than the originating organisation. Eventually, you are going to hurt someone's sensitivities by showing a tie rather than a lead for one party - or even reversing a lead - and probably because of a microscopic change which is magnified by rounding. To what good end? Can I opine that WP:OR is there for a very good reason, and you have just banged your head against it. There is a perfectly satisfactory way of dealing with this, which is to tabulate GB+NI polls separately to GB polls.That makes the posting of all data easily verifiable against the source, and makes it easier for anyone to add a poll without any initiation rites. Yes, there was a consensus, but that doesn't make it right. If the Survation polling methodology is out on a limb, then it is fair and right that they should be tabulated out on a limb. — Preceding unsigned comment added by 88.106.194.185 (talk) 10:21, 15 May 2018 (UTC)
 * Uninvolved editor here. I have big problems with WP:SYNTH and WP:NOR for us adjusting pollsters' figures, though I totally understand why you're doing it. As I see it, either exclude them or have a different chart for polls that include NI. --Dweller (talk) Become old fashioned! 11:25, 15 May 2018 (UTC)


 * May I suggest a compromise? Let's put UK-wide polls in a seperate table, with the UK-wide figures at the bottom of the table. This should help satisfy both sides of this discussion. FriendlyDataNerdV2 (talk) 12:02, 15 May 2018 (UTC)


 * I'm not very involved as an editor these days either and my main interest in this page is as a scraper. However, I do not see this as a matter of WP:SYNTH and WP:NOR. This is simply making a statement of fact for which there is a reliable source, exactly what editing an encyclopaedia is about. It is a mechanistic redaction just like extracting any content from a reliable source. If we can't do that we are just a gazette of published content. I don't see it as editorialising as there is no subjective content or opinion. If it is research then it certainly has no claims to originality. The difficulty arises because this is a report of published polls so, to some extent, does act as a gazette. That all said I am neutral as to outcome.Cutler (talk) 12:11, 15 May 2018 (UTC)

@ 88.106.194.185 - Thank you and I'm "broadly" in agreement. There is still a problem as to where do we "draw the line". With the Survation poll we are dealing with a fairly minor variation in geographical spread (all of the UK as opposed to GB excluding NI) but that is actually very slightly more representative of a General Election vote than shown by the other pollsters. We could just as easily say (by way of an example that may not be accurate but just to make the point) that we should put ICM/Guardian into a separate table as they do not appear to do the same bias corrections as YouGov or IPSOS/MORI. We simply do not know to what extent they do this even though YouGov (and also ComRes for that matter) show the most details regarding re-weightings of their results, if you search their websites thoroughly. The extent of any bias correction is likely to make a rather greater difference than the NI or no NI figures do (I've seen +/- 2% to 3% quoted previously, so far great than the 0.5% or so for the NI exclusion)). So, we could just as easily make some quite reasonable arguments for and against the various inclusions, exclusions and adjustments but this merely illustrates the absurdness of even attempting to embark on this exercise.

We could contact all the pollsters about these matters and ask them to provide their reasons for why they include/exclude certain details and to what extent they make any adjustments. I suspect, with their wishing to justify their own approaches and methodologies that we would get a mix of answers that may make things even more confusing! Please can someone point me to the adjustments that are made by Wikipedia to the "other" pollsters' output. This is a major concern to me, even more than the Survation matter. Whilst, I "could" agree to putting the Survation figures into a separate "UK" table and renaming the other table "GB excl NI" I most certainly do not believe that this would resolve the matter permanently and it would rear its ugly head again quite soon - perhaps immediately as some other members would query the justification for separating the tables.

Surely the best solution is to not interfere at all with the figures and then put a note above the table with a link to another page which summarises the different methodologies by all the pollsters. In fact as the pollsters would probably want to ensure that their figures are not manipulated and recognise the value (and publicity) of an unbiased table on Wikipedia, we could invite them to provide some of the input for that page. We could list each pollster in alphabetic order and provide the details. If any pollster does not provide this info it would show up clearly as a gap on the page and would surely provoke them to fall into line by providing what we need. Obviously this will take time to evolve and needs to be thoroughly thought through by those of us who might contribute towards this exercise. We would need to draft a standard approach document with perhaps 90% of it being the same questions. At the end we can ask certain questions relating to any specific variations (like Survation including NI opinions in their polling). In fact, I have seen such an array of info on each pollster's websites, that a Wikipedia comparative summary would be extremely useful anyway. In the meantime, please put the Survation data back into the table (without adjustments). Please treat them as innocent until proven guilty and not the other way around. Although I'm extremely busy running an online business, I am checking emails all day and will often respond quite quickly. I'll be happy to do a share of the work load, if you agree to my suggestions above. Dienumbers (talk) 12:08, 15 May 2018 (UTC)


 * I'm happy to leave this policy discussion to others, but at the moment most Survation polls (all but 1) aren't even in the table at all. So we should decide on a solution soon; leaving them out completely isn't ideal. FriendlyDataNerdV2 (talk) 14:09, 15 May 2018 (UTC)

@FriendlyDatanerdV2 - well it's a shame if you leave it up to others as it doesn't look like a consensus will be reached any time soon. You are right to raise concerns about Survation being left out. Whoever did that, needs to put them all back pronto. At the very least they need to remain there (unaltered) pending a longer term decision. I came up with not only a workable solution but also a very positive way forward. I hope that you and others comment on that. Dienumbers (talk) 15:30, 15 May 2018 (UTC)


 * I've gone ahead and restored the Survation polls as well as chart. Regardless of the eventual decision on weighting, they are still polls and should not have been deleted.  PaperKooper (talk) 15:48, 15 May 2018 (UTC)
 * I agree with and past consensus: this is a straightforward calculation (allowed under WP:CALC) that allows for a direct comparison. (, no, we do not adjust any other polls. I'm unclear how you got to that conclusion.) If enough others chime in in favour of the old consensus, we can stick with that.
 * We cannot have a table where some Survation figures are adjusted and some are not. That is cherry-picking and misleading.
 * Editors above, principally Dienumbers, argue that the adjusted figures are wrong. If they're wrong, we definitely can't have them in a table and most of our Survation figures are adjusted (all but the recent one). I think Dienumbers is wrong on this point, but, Dienumbers, if you think the adjustment is wrong, then you must agree the figures can't stay as they are? If you want to go through and change all the Survation figures to unadjusted, go for it. We could work from there.
 * The suggestion to contact pollsters is a non-starter. That's not how Wikipedia works.
 * The UK is not GB. If we're not adjusting figures, we need a separate table, as suggests. That could be done easily and quickly, once Dienumbers or someone else has put in the unadjusted Survation figures. Bondegezou (talk) 17:08, 15 May 2018 (UTC)

Just to confuse things, if we decide the adjustment isn't valid, then I would support keeping them both in the same table, with a note about the different methodologies. Absolutelypuremilk (talk) 20:22, 15 May 2018 (UTC)

Summary
To try and summarise the above in a useful way...

Survation polls are of the whole UK. Other polls are just GB. What do we do about this?

Past consensus has been to take Survation data tables and remove the NI component and present adjusted figures (with a note explaining what we've done). This is done under a WP:CALC argument. This approach is supported by me and, and partially by. supports the argument, but has declared neutrality as to the outcome.

However, other editors feel that adjusting the poll figures is inappropriate (WP:OR): this is the position of, 88.106.194.185 and (I think).

If the adjustments are invalid, what's the alternative? We could list UK polls in a separate table. FriendlyDataNerdV2 and 88.106.194.185 support this. Dweller and me are OK with this too. I think Dienumbers just wants everything in one table (with a note on the different methodologies).

Separately to all this, there's the question of what do we show while we work this all out. I say we remove anything contentious until it's sorted. and Dienumbers want everything left in while we're sorting it.

I hope I have summarised everyone's position appropriately. Right now, separate UK and GB tables seems the most agreeable compromise. I will seek input from the relevant WikiProject. Bondegezou (talk) 17:28, 15 May 2018 (UTC)
 * I've been trying to find the past Talk discussion that came up with the Survation adjustment, but without success. If anyone can find it, please give link here. Thanks. As far as I remember, it pre-dates this article. Bondegezou (talk) 18:51, 15 May 2018 (UTC)


 * Comment I don't see a problem with adjusting the Survation polls (it's not WP:OR as simple calculations are allowed). However, if they are really disputed, then the only option I see is to omit them from the list, as it's not comparing like-with-like. Number   5  7  20:08, 15 May 2018 (UTC)


 * Comment The original discussion was here: https://en.wikipedia.org/wiki/Talk:Opinion_polling_for_the_next_United_Kingdom_general_election#Including_NI_parties_in_main_table I'm swayed by both sides, but would rather have adjusted Survation figures in the table than have none at all! I can see the argument of Dienumbers and think they make fair points but if the choice is between 'adjusted Survation' and 'no Survation' I'd choose adjusted. However I think yes, the clearest compromise is separate tables. FriendlyDataNerdV2 (talk) 20:30, 15 May 2018 (UTC)


 * Comment I think you should follow reliable sources, of which the main one for this subject, ukpollingreport, just reports the figures that're given by pollsters and does not adjust for methodology. I have only seen "reliable [newspapers]" sources make these adjustments when it's convenient for them, the actual RS don't seem to. Wiki is supposed to (regardless of it not actually) report stuff from "reliable 3rd party sources with a reputation for fact checking and accuracy", going into the prime source tables & making adjustments yourselves seems quite wrong. I know WP-synthesis (orwhatever its called) allows for such mathematical alterations, but I don't agree with it. Why not try to even out all the other methodogical differences too? Report what the RSs say. 92.3.156.60 (talk) 00:19, 16 May 2018 (UTC)
 * I've already made my position clear on this matter and the IP above also puts it well – it's ridiculous not to simply report the figures as-is (just as they are in the press). Just add a footnote or inline note for the Survation polls rather than recalculating the figures every time when no other source does. Mélencron (talk) 03:53, 16 May 2018 (UTC)
 * Comment I can follow the contrary argument, but definitly prefer standing consensus to adjust UK polls to match GB polls. The change is sound by policy per Bondegezou’s argument and methodologically sound as a large sub-sample. If a compromise is required, I’d be inclined to keep both types of poll in the same table with UK polls adjusted, but with adjusted polls made more distinct. Perhaps with italics. Separating UK from GB polls seems like it would introduce clutter and make the page less useful, and would not eliminate the problem of well intentioned but unfamiliar users making inappropriate edits to the page. Keeping both types of poll in the same table with a qualifying column explaining the geography of each poll would be my least preferred option of those advocated. Ralbegen (talk) 05:44, 16 May 2018 (UTC)


 * Comment First thing, Keep the Survation Poll in the article, it was the closest polling organisation to the last election result. The current system one table has worked well and is most clear to the reader, but relies on folks knowing how the adjustments are made. I agree with making the adjusted figures more visible is a good idea. The Survation poll ought to have remain in the article until a change is made. -- BOD --  09:00, 16 May 2018 (UTC)


 * Comment It seems pretty clear cut to me that if this is a table for GB polls, we don't include UK polls in this table, and instead have a separate table for them. Adjusting polls without making it extremely clear what we are doing is WP:OR, and even if some very clear way of labeling is found, it still doesn't sit well. We wouldn't include the subset of each poll for Scotland in the Polls of Scotland section, as that is not the published figure; just because the GB value is more statistically significant doesn't mean we're not doing WP:OR. The demographics of the UK and GB are not the same, and the Survation figures will have been weighted on UK values; if we're not adjusting for the fact that Northern Ireland has, for example, fewer ethnic minorities than GB then our figures will be wrong, and if we are doing such adjustments it's definitely not covered by WP:CALC.   On a personal note, I've used this page for a long time as a source for statistical modelling and I'm very concerned to learn that figures have been adjusted. Barnowldance (talk) 09:15, 16 May 2018 (UTC)


 * Comment Thanks everyone for a rapid accumulation of thoughtful and respectful comments! Unfortunately, we don't appear much closer to a consensus. What we do all agree on is that we need to clearly mark/label what's going on. If we have adjusted figures in a big table, they should be clearly indicated to be adjusted. If we have unadjusted figures in a big table, they should be clearly indicated to be UK polls versus GB polls. (If we have UK polls in a separate table to GB polls, then that will be clear in itself.) Personally, I am somewhat swayed by 92.3.156.60 and 's point. If others covering polls are happy to include unadjusted Survation figures in with GB polls, then maybe we should be to. Thus, I would be OK with a single table, with unadjusted Survation figures, with some clear marking to note they are UK polls. Bondegezou (talk) 11:30, 16 May 2018 (UTC)


 * Comment May I suggest the following style of table? We can have both the UK and GB results at the bottom, and a box beside each poll indicating whether it is UK-wide or GB-wide. Then people make comparisons between the result and the poll for themselves.


 * I like it. Bondegezou (talk) 12:48, 16 May 2018 (UTC)


 * Comment Well, I thought that I would await a few more comments from other readers prior to commenting again and I like most of what I have read and especially the reformulated table! This is an excellent solution - the revised table now shows that the "other pollsters" (for GB only) are actually out on a limb as Survation are the only one to reflect the actual "General Election" voting intention for the whole nation as I pointed out before. It's really clear and, most importantly of all, there are no controversial adjustments to the figures provided by each pollster. So, this way forward definitely gets my vote and full support. I still think that my other idea of providing additional information on the pollsters on a link page (behind the tables page) would be extremely useful and very welcome by many readers. This is simply providing additional facts on the methodologies in an easy to read summary for each pollster. The contacting the pollsters was to ensure accuracy and to take on board any particular points that we might have missed. For someone to once again say that "this is not how Wikipedia works" just couldn't be more wrong. Almost all of Wikipedia is built on providing independent information and where we can provide useful additional information, that must always be encouraged. Dienumbers (talk) 15:14, 16 May 2018 (UTC)
 * Comment I would like to make a proposal then that we adopt the table format above, reporting the headline figures as Survation presents them without adjustment, but with that additional column indicating the geographic area covered and a footnote explaining what it means. Does this sound acceptable to people? FriendlyDataNerdV2 (talk) 17:20, 16 May 2018 (UTC)
 * Support Personally I still prefer adjustment to permit like-for-like comparison stylistically, but the case for changing to FDN2's proposed approach is strong, so I support the proposed change. (Perhaps a thin column headed 'area' with data either UK or GB.) Ralbegen (talk) 17:37, 16 May 2018 (UTC)


 * How's this ? FriendlyDataNerdV2 (talk) 18:07, 16 May 2018 (UTC)


 * Looks good to me! Thanks for mocking it up. Ralbegen (talk) 19:54, 16 May 2018 (UTC)
 * Oppose Any table that juxtaposes unadjusted results inevitably implies an equivalence. Unless the Survation numbers are adjusted they do not have equivalence, being calculated with a larger population.  A separate table for UK-wide polls would be acceptable, but not unadjusted pears in a table of apples, even with a note that they are really pears.  DrArsenal (talk) 08:39, 17 May 2018 (UTC)


 * Support Though I would prefer a separate table, I think this is a reasonable compromise at least for now, and certainly better than editing figures. 82.33.221.20 (talk) 12:28, 17 May 2018 (UTC)


 * Comment I would like to see a decision reached on this question soon - at the moment we're excluding the most accurate 2017 pollster from the tables. The balance of opinion seems mostly in favour of the compromise table, so I'm minded to put the Survation polls back in, unedited, with the above table's style allowing people to make UK or GB comparisons and know what area each poll covers. I'm happy to do that, but I'd like to do it soon as this is a page read by a number of people. If anybody has strong objections please raise them . FriendlyDataNerdV2 (talk) 14:21, 17 May 2018 (UTC)
 * I'm OK with that in the short term. I'm OK with that in the long term too, actually. Bondegezou (talk) 14:45, 17 May 2018 (UTC)

I think incorporating Survation, marked as UK versus others marked as GB is an excellent suggestion, and the mocked up tables look good. FWIW, thinking some more about the adjustment calculations, I think it would generally be quite accidental if the Survation polls were appropriately balanced demographically with the NI data chopped out. They don't set out to achieve that, and I suspect it is unlikely that they do. Just adds to my feeling that the adjustments unwise, not only through being potentially controversial. I think it's important that Survation is reported in a way which is visible in the 'main line', because they did better than most at the last election, IIRC. — Preceding unsigned comment added by 88.106.194.185 (talk) 17:01, 17 May 2018 (UTC)
 * Discussion has rather ground to a halt. We don't have a clear consensus. I think the position with most support is for a single table, with a column indicating UK vs GB, including unadjusted Survation figures, and, likewise, for a graph including unadjusted Survation polls, plus some text explaining this. Let's do that and see what people think...? Bondegezou (talk) 11:28, 20 May 2018 (UTC)

Tangential discussion
, I agree with you about the value in adding more poll methodology detail. However, while editors are free to do whatever they want, it would not be the usual approach of Wikipedia to contact pollsters directly. See WP:NOT and WP:OR for an explanation. Bondegezou (talk) 09:55, 17 May 2018 (UTC)

Providing Additional Information on Pollster Methodologies:

@bondegezou - well, I'm delighted that the gap between our rather differing views has been steadily narrowing - it shows that the robust debate that some of us participated in has been very productive. I do apologise for coming across at the outset in what may have appeared to be a somewhat aggressive manner - that wasn't my intention and it was my passion for accuracy that drove me together with a slight initial misunderstanding of several other views expressed. I am, believe it or not, pretty flexible on the majority of issues. I far prefer to be constructive and often like to think outside of the box. I'm pleased to see you support my suggestion of providing greater details on pollster methodologies. I can see that it's not up to "Wikipedia" to directly contact organisations and in this case the various pollsters but any one of us can contact them as part of our own research and report back with our "personal findings". In fact, I imagine that this is the way that the vast majority of information is gathered on almost any topic. What I propose is that some of us start to gather this useful information and outline it on here for others to see, comment on and correct if they believe that any facts are not entirely accurate (or are misleading). May be we can set up a "draft" page for this for editors' and contributors' use? Forgive me for not being familiar with how Wikipedia usually goes about this type of group exercise. I feel that to appear to be fair to the pollsters, we should spend some time putting this together before we release it for public viewing, when we should have at least some information on the methodologies used by almost all of the pollsters. The more facts that we can show, the easier it will be to compare the varying procedures adopted. On the one this will provide some very useful additional information and greater clarity but it might also throw up some differences which make the GB (excl NI) and UK (incl NI) approaches seem quite minor by comparison! At least we, together with all the readers, will have far more facts and information available to us to understand these differences.

I am extremely tied up with my business commitments for the next few days but I will try to find some time towards the end of next week to start the ball rolling. In the meantime, this is open to anyone to come forward with any information in order to contribute to this combined exercise or start it off. Dienumbers (talk) 11:17, 17 May 2018 (UTC)
 * Thank you for your friendly reply. Your enthusiasm is apparent and welcome. I can see you are newer to Wikipedia and some subtle details of how Wikipedia works can be confusing at first.
 * No, most information on Wikipedia is not gathered as you suggest. Mostly, Wikipedians read reliable secondary sources online, in books or journals. Wikipedians may dig out obscure sources. They may encourage other organisations to release materials into the public domain. But we are cautious about using primary sources (see WP:PRIMARY) and, say, private correspondence between an editor and an organisation could only be used in very limited circumstances. I believe that the methodological differences between pollsters under discussion are sufficiently covered by material in the public domain and there is no need to contact pollsters.
 * Everyone with an account gets a sandbox (linked top right, between "Talk" and "Preferences") where people can individually or collaboratively work on a draft. However, generally most editing just takes place on the live page, under the principles at WP:BOLD. Or one can also draft text here in the Talk page. Bondegezou (talk) 14:44, 17 May 2018 (UTC)

Graphical Summary
Any reason why it was removed? Masterpha (talk) 18:45, 17 May 2018 (UTC)
 * I removed it because of the argument above. The graph uses adjusted Survation figures. Many above think those figures are wrong and against Wikipedia policy. Ergo, a graph using them is wrong. I'm all for returning the graph when we have some consensus on what figures to use. Bondegezou (talk) 18:47, 17 May 2018 (UTC)
 * I'm not sure what the future of that graph will be? It seems like it'd be a bad idea to include both UK and GB figures in a graph, and if adjusted figures are no longer used in the article it'd be inconsistent to use them in graph. Excluding Survation figures from the graph altogether doesn't seem like a great prospect either. And omitting the graph entirely seems the worst option of all... Ralbegen (talk) 19:01, 17 May 2018 (UTC)
 * We have three options. (A) The graph uses GB polls and adjusted Survation figures. (B) The graph uses GB and UK polls (i.e. unadjusted Survation figures). (C) We have separate graphs for GB and UK polls. This mirrors the debate in the previous section, the difference being that a compromise around marking UK polls in some way is harder to achieve. I suppose some compromise is possible, e.g. a main figure prominently displayed showing GB and UK polls together, and then another GB only graph lower down in a section on methodologies, or vice versa.
 * I have become swayed by the argument that RS often just stick GB and UK polls together, so I'd be OK with (B). Bondegezou (talk) 19:29, 17 May 2018 (UTC)
 * While we move towards a consensus, if someone would like to re-do the graph with unadjusted Survation figures (for (B)), a graph for GB polls (i.e. no Survation) and a graph for UK polls (i.e. just Survation polls), that would be very useful. Bondegezou (talk) 19:46, 17 May 2018 (UTC)
 * Not a perfect imitation, not least because I'm not sure how to convince Excel to do a 10-point moving average that starts at the beginning, but:
 * Survation on its own looks a little silly, in my view. Ralbegen (talk) 20:18, 17 May 2018 (UTC)
 * Survation on its own looks a little silly, in my view. Ralbegen (talk) 20:18, 17 May 2018 (UTC)

Comment I have no opinion on the controversy, but I do hope the graph returns. I use this page not as an editor, but as a reader, and the graph is an extremely useful visual aid for me. I imagine many other readers feel the same way. schetm (talk) 23:42, 17 May 2018 (UTC)
 * I think we are all absolutely, 100% in favour of the graph. However, it's gotta have the correct numbers in it, and we can't quite agree on what those are. Bondegezou (talk) 08:52, 18 May 2018 (UTC)

It may be a little clunky, but Excel certainly has the technology to use different markers for, e.g. Labour (GB) say red dot, and Labour (UK) say hollow red circle. In the past I've made charts where the dot size also carries information: e.g. where its area was the sample size. If we are busy making this thing strictly accurate, the positioning of the Excel moving average is pants. The date of the current moving average should be the (possibly sample weighted) average of the (midpoint) dates of the constituent polls - at least that seems right to me. I'm still trying to figure out how to get my hands on the data to contribute an example. — Preceding unsigned comment added by 88.106.194.185 (talk) 10:41, 18 May 2018 (UTC)
 * I copied the results from Britain Elects's table having verified it against this page's record to graph them. That's a much easier way to get them than scraping from this page. The graphs I put together above were purely illustrative about what difference different approaches makes rather than a proposal for a graph to actually display! Being wary of the conversation spirally out of focus: the decision that needs to be made here is about inclusion and adjustment criteria for a graph rather than the design choices. Ralbegen (talk) 19:01, 18 May 2018 (UTC)


 * I disagree about using the midpoint - the remit of the page is simply to report the published opinion polls, not to attempt most accurately reproduce the voting intention of the public. We should use the date the poll was published on. Anything else would be an inference. 92.17.58.165 (talk) 16:53, 18 May 2018 (UTC)

I hadn't thought of using the mid-point for the date of a datapoint, I tend to agree that is confusing. However, for the moving average, the final publication date really does not represent the correct date for the collection of polls which the moving average covers. Some kind of average of the poll dates is preferable, in my view. The table already covers the sample dates, they are not inference.

My preference is just to include all n0n-adjusted numbers in the graph (B above). That is the only accurate and complete graphical representation we can have, and I think is most useful for the reader - include a short sentence to explain this. The other alternative is simply to describe the graph as summarising GB polls and include the first graph in C only (and simply do not include Survation polls at all). Including more than one graph is not a good change and will be confusing to readers. 92.17.58.165 (talk) 16:48, 18 May 2018 (UTC)


 * I am for including UK and GB polls in the same graph, perhaps with a footnote. In population terms, GB is 97-98% of the UK so the effect will be tiny. And the effect will be the same for all major parties as, either they do not stand in NI, or (Greens and UKIP) they receive little support there. --Wavehunter (talk) 19:47, 18 May 2018 (UTC)


 * I have always appreciated the visually clear Graphical summary, I agree with Wavehunter I think we should include both UK and GB polls on one graph with a small and clear footnote to highlight the relatively minor differences. --  BOD --  14:29, 20 May 2018 (UTC)

I agree with BOD (@Bodney) as this will make it consistent with the table. Dienumbers (talk) 15:09, 20 May 2018 (UTC)

It looks like there's consensus for a graph including both UK and GB polls. I've put together this as a neater version of the UK/GB combined graph above; more consistent with the previous graph:

It could potentially be used as a stop-gap until Absolutelypuremilk updates their graph? Ralbegen (talk) 20:32, 21 May 2018 (UTC)
 * I am away but updated my graph earlier today, is it missing some polls? I'm not sure how to easily go about calculating the mid point of the last ten polls, but I will give it a think. Absolutelypuremilk (talk) 22:08, 21 May 2018 (UTC)
 * The consensus seems to be for a graph with unadjusted figures in line with the changes to the article agreed in the section above — I'd assumed that your graph is using the previous adjusted figures? Ralbegen (talk) 17:00, 23 May 2018 (UTC)
 * Yes it did, I've updated it now and re-added the graph to the article. Absolutelypuremilk (talk) 21:14, 23 May 2018 (UTC)
 * Could I ask if it would be possible to make two charts. One showing just Con and Lab and the other showing Lib Dem, Green and UKIP given it being very unlikely that all five will be within a few percent of each other Harry Hayfield (talk) 19:43, 8 June 2018 (UTC)

Folks, there is an error in the current graphic, or at least an inconsistency with the table. The graphic shows a poll with Conservatives at 30%, there is no such poll in the table. — Preceding unsigned comment added by WikiRERT (talk • contribs) 11:01, 16 July 2018 (UTC)

Congratulations
Just want to congratulate the many editors who've contributed here to resolve a difficult issue with the UK-wide polling. This is what consensus editing is all about. Nice work. --Dweller (talk) Become old fashioned! 22:09, 23 May 2018 (UTC)

Plaid on Graph
Any reason why Plaid Cymru is featured on the key to the graphical summary when there is no line for it on the actual graph. It doesn't make much sense to me. Lancashire2789 (talk) 18:46, 25 May 2018 (UTC)
 * No idea, I've removed it. Absolutelypuremilk (talk) 21:06, 25 May 2018 (UTC)

New Poll that I don't have the skill to add.
http://www.bmgresearch.co.uk/bmgs-westminster-voting-intention-results-june-2018/ — Preceding unsigned comment added by Dead Asteroid Miner (talk • contribs) 23:07, 19 June 2018 (UTC)

Well done!
Well done to whoever it was who added the commissioning organisations (media) to the polling organisations’ names. The last time I looked there were none for the last 15 or more polls, but since then they’ve virtually all been edited in. It is important information. Many thanks! Boscaswell  talk  23:01, 28 June 2018 (UTC)

Kantar sub-national polls
The Kantar poll shouldn't be used for England, Wales and Scotland separately. The Wales and Scotland ones contain fewer than 100 people, and I don't think the England poll is weighted to match the English demographics. Absolutelypuremilk (talk) 19:23, 10 July 2018 (UTC)
 * Agreed. Bondegezou (talk) 21:00, 10 July 2018 (UTC)
 * I think the difference from other pollsters is that Kantar explicitly publishes results by nation. But a sample of less than a hundred is silly, and produces a correspondingly silly result in Wales. There's some inconclusive discussion on the Talk archive about the subject of England subsamples, but practice seems to have tended to exclude. I don't see any reason to include any of these subsamples. Ralbegen (talk) 21:26, 10 July 2018 (UTC)

I think polls that have that don't have 100 people questioned shouldn't be used. If there is more than 100 then yes as long as it follows the British Poling council rules should be used in the tables. Anonymous — Preceding unsigned comment added by 86.29.74.26 (talk) 13:11, 11 July 2018 (UTC)
 * I think polls following the British Polling Council rules should be used, but not their subsamples, however they are published. Bondegezou (talk) 15:04, 11 July 2018 (UTC)

Their sub samples should be published if there have 100+ people who where questioned less than 100 would be silly and inaccurate. a sub sample is a legitimate poll in my opinion. Anonymous — Preceding unsigned comment added by 86.29.74.26 (talk) 15:14, 11 July 2018 (UTC)
 * Sub-sample polls have not been demographically matched as required by quota sampling. That may be less of an issue when dealing with an English subsample, England being such a large proportion of the total poll, but consensus here to date has been not to include subsamples. Bondegezou (talk) 15:29, 11 July 2018 (UTC)

I still think that the sub samples should be included in the tables as there added together to make the main opinion polling figures. Anonymous — Preceding unsigned comment added by 86.29.74.26 (talk) 15:47, 11 July 2018 (UTC)
 * A poll of Great Britain is not polls of three countries added together, just as it isn't six polls of different age groups added together. Nation or region is just one factor used to weigh the responses to produce results that are representative of Great Britain as a whole. That's explicitly stated in the PDF of the results: Weights have been applied to the sample to ensure that it is representative of the general

public in Great Britain aged 18+. The sources for the population totals are provided in each table. In my view, and I believe as is the established consensus, it's inappropriate to include subsamples on this page. Ralbegen (talk) 16:36, 11 July 2018 (UTC)


 * Sub samples are not polls, they are not demographically weighted in the same way and have a gigantic margin of error. Reliable sources such as Anthony Wells of UK Polling Report consistently decline to report sub samples as being significant. It's not a question of opinion - sub samples are factually not polls and are not reported as such. We shouldn't include them. FriendlyDataNerdV2 (talk) 13:04, 12 July 2018 (UTC)
 * I have now removed all the sub samples. The previous consensus was not to include them, we need a new discussion before changing that. FriendlyDataNerdV2 (talk) 13:40, 12 July 2018 (UTC)

Mistake in the image
Somehow in the latest image update it looks like there was a poll with the Conservatives at 30% a little bit ago. But there wasn't any such poll and that needs to be corrected. --Blemby (talk) 15:57, 16 July 2018 (UTC)
 * Good spot, I've fixed it now. Absolutelypuremilk (talk) 19:28, 16 July 2018 (UTC)
 * Just a heads-up that the current version includes hovertext. Ralbegen (talk) 14:56, 20 July 2018 (UTC)
 * Thanks, I saw it but unfortunately I am away from my laptop for the weekend so it won't be fixed until Monday. Absolutelypuremilk (talk) 15:26, 20 July 2018 (UTC)

Chequers agreement
I know we dont want to turn this page into a running timeline, but would it be worth acknowledging the chequers arrangement. Already we have seen a stark difference in polling which perhaps deserves an explanation: UKIP have already gone up 6% and Labour are now in the lead. Without explanation for this change in polling it may leave readers (particularly from abroad) confused.TP69 (talk) 11:51, 15 July 2018 (UTC) — Preceding unsigned comment added by 80.43.200.204 (talk)
 * I'd agree with adding this event as it has resulted in a major shift in opinion polling. Lancashire2789 (talk) 11:50, 15 July 2018 (UTC)
 * Please not this again. No events in table. 92.3.159.209 (talk) 02:17, 16 July 2018 (UTC)
 * I'd be against the inclusion of the event. Firstly, because it would go against established consensus in UK opinion polling articles of adding only party leadership changes, election debates and local elections and by-elections (the inclusion of such event would thus require a revision of such consensus). Secondly, because it would be dangerous in terms of WP:NPOV to 1) assume a specific event (and not others) has, by itself, contributed to a major shift in opinion polling; 2) determine what a major shift in opinion polling does mean. One poll? Two polls? Five? How large must the shift be in numerical terms? And for how much time must it endure to consider it significant? All of these are elements which are largely non-neutral and depend mostly on specific points of view, which would mean the concept of "major shift in opinion polling" would not be without bias. Thirdly, because a lot of events of varying relevance may affect opinion polling in some way (indeed, every change in each new opinion poll compared to the previous one would mean something has affected polling. And this without even counting possible outliers). And finally, because this has not been done in previous shifts in opinion polling (i.e. when Labour's lead eroded and gave way to a Tory lead in around March/April 2018, no event is introduced to attempt to justify that.  Impru20 talk 14:00, 15 July 2018 (UTC)
 * Thanks for your contribution. I agree that we perhaps need to revise the consensus if the polling continues to rapidly change. I will be watching UKIP the closest to judge this change. But ultimately we need a new discussion for various examples for polling changes and what causes it. Personally I feel every stark change in polling requires an explanation.TP69 (talk) 18:11, 15 July 2018 (UTC) — Preceding unsigned comment added by 80.43.200.204 (talk)
 * No events in the table. Other election articles don't have them. Significant political events can be covered in the appropriate articles, e.g. on Brexit, Next Conservative Party (UK) leadership election or Next United Kingdom general election, where they can be discussed in a WP:BALANCEd way using WP:RS. Bondegezou (talk) 07:21, 16 July 2018 (UTC)
 * It may therefore be a good idea then to change the 2001, 2005 and 2010 UK general election polling. I know consensus was formed on there to include events. Feel free to delete them.TP69 (talk) 14:58, 16 July 2018 (UTC)
 * I've done a very quick run through and deleted items not matching current consensus. Bondegezou (talk) 14:51, 16 July 2018 (UTC)

I agree with TP69 and Lancashire2789. Dear Bondegezou, I’m not so sure that the ‘current consensus’ for your view actually exists. It’s fairly obvious that the Chequers agreement has been a pivotal event as far as the polls are concerned, and therefore I feel that it deserves a mention in the list. Boscaswell  talk  10:01, 20 July 2018 (UTC)
 * I agree with Bondegezou. Consensus is very clear as on which events should be added as well as on the conflicts deriving as on which events are "pivotal" to opinion polling and which ones are not, as well as to what consistutes a drastic shift in opinion polling and what not. I should note that this is not the first "pivotal event" (nor will it be the last one, surely) that some users want to push it into the table (just as a reminder, the Grenfell Tower fire tragedy was also one of such "pivotal events" which ended up being left out). For example, in this case, how do we know that it was the Chequers agreement in itself affecting opinion polling and not, let's say, Davis and Johnson's resignations (either both of them or just one)? Or criticism to May as a result of her management of the ensuing political crisis? Or the political fallout resulting from Trump's visit to the UK? Or the street protests resulting from such a visit? Or a combination of all of these? Who determines which one is the "pivotal event" that counts? One of the major issues with events in opinion polling tables is exactly this. Either we end up adding basically everything that happens and that may be hypothesized to have an impact in opinion polling (thus essentially turning the polling tables into a massive timeline of events), or, by selecting just a few "pivotal events" unilaterally determined by ourselves, we may end up going against WP:NPOV (by introducing an editorial bias to the reader on which events do matter and which ones do not), WP:OR (by determining ourselves which events do impact opinion polling and which ones do not), WP:UNDUE (by assuming that all "pivotal events" we include have an equivalent impact in opinion polling) and WP:CRYSTAL (because we are essentially assuming that the shift in opinion polling that happens in a period of time will last long enough to constitute a major shift in opinion polling; the definition of what constitutes such would also be a major point of friction).  Impru20 talk 11:15, 20 July 2018 (UTC)


 * Personally I am torn, while i can see the benefit & logic of including on the graph, brief background detail of significant events to maybe explain polling changes. However I am not convinced that all contributors will agree which, where, how, when, to what extent etc an event has had an actual influence or not ~ even more so because as humans every editor has their own natural political bias. The simply is not enough space to add too much content without over extending event boxes, while balancing neutrality. It would likely be a can of worms, which would result in many future arguments on this talk page. So maybe in the end it is I believe safest not to include such detail. --  BOD --  11:29, 20 July 2018 (UTC)


 * Impru20, I don’t believe that your link under the word consensus indicates any consensus, just that no action was taken before. ‘Chequers agreement and subsequent ministerial resignations’ is surely a form of words that we could all agree on.  Unless, of course, we’re being silly about it.  There have, after all, been articles in several newspapers saying that the agreement and subsequent fallout have substantially  affected voting intentions.   Boscaswell   talk  12:50, 20 July 2018 (UTC)
 * There have been discussions on the issue of events ever since 2012 at the very least (and probably earlier), all of them leading to the same result, this one I posted being just a mere example. The most recent consensus (which only mirrors the outcome of previous discussions) is here. You can also take a look at the archives of previous opinion polling articles to check it out yourself. What you believe or fail to believe on those is not of my business, as it is obvious that consensus on this issue throughout the years has been consistent and that the addition of random events into the table will lead to a likely violation of several Wikipedia policies. It is not a matter of how you try to word an event, but the sole addition of such event which poses the actual issue.
 * There have, after all, been articles in several newspapers saying that the agreement and subsequent fallout have substantially affected voting intentions Again, I remind you about the existence of WP:NPOV. That opinion articles in newspapers consider some events to have impacted opinion polling only means that those newspapers say that those events have affected opinion polling, not that such an impact may be considered as an undisputed and unquestionable fact. First rule of NPOV: Avoid stating opinions as facts. This said, if we add a row about an specific event in the table, we would be basically hinting readers that:
 * Such an event has impacted opinion polling in a specific way, and that it happening in such a way is an unquestionable and undisputed fact. Even when such assertions are frequently ascertained from mere opinion (thus, doing such would go against both NPOV and OR).
 * Such an event is relevant enough for that specific period of opinion polling to merit inclusion, as compared to other opinion polling periods when there are shifts in polling yet no events are introduced (also against NPOV).
 * We would also be pre-determining ourselves how opinion polling will evolve in the future in order to determine how relevant an event is right now, thus going against WP:CRYSTAL (this is, there have been only five polls published after the Checquers agreement, yet you are already determining yourself not only that such an event is relevant enough to have caused such a shift in opinion polling, but also that 1) such a shift in opinion polling will be of a notoriety such (in terms of both size and duration) that it would justify the inclusion of an event row; and 2) that such a shift, if it persists, will be unaffected by future events so as to make the Checquers agreement the only cause of it, when this is obviously never the case, as many factors and situations may have an impact in opinion polling each week).
 * The only way to avoid any of these would be to add any and all events that could have a potential for impacting opinion polling, which would probably turn the tables into a timeline of events rather than actual opinion polling tables. And such a solution would have to be surely discarded as this is an article for opinion polling, not a timeline of events in the 57th Parliament of the UK.  Impru20 talk 13:43, 20 July 2018 (UTC)

This comes up a lot. For reference, here are previous discussions of event inclusion from this article and its two predecessors: April 2012, March 2014, April 2014, October 2014, November 2014, April 2015, September 2015, April 2016, July 2016 (twice), October 2016, November 2016, December 2016, January 2017, March 2017 (three times), April 2017, May 2017 (three times), June 2017 and November 2017. There's also some relevant discussion here.

Between those I think there's a fairly clear through-line of consensus towards a minimal approach to events, not just inaction on a particular case. That said, it may be worth holding an explicit and general RFC on inclusion criteria for events in polling tables for UK elections, just so that people's time doesn't keep getting wasted having these discussions again and again. I can appreciate both the desire to keep the tables fairly clean and bare, and to enrich them with events to give context. Having read previous discussions, I suspect that the policy argument strongly favours the former approach. Ralbegen (talk) 14:53, 20 July 2018 (UTC)


 * I agree with those who want to keep it all neat and tidy. Their arguments are valid.  After all, if, for example, I was to be walking along a footpath and after a while trip over a pile of bricks, who is to say that the bricks were crucial?    Boscaswell   talk  23:40, 22 July 2018 (UTC)

Table date column not sortable
The Date(s) column in the tables do not sort correctly as the ranges used (eg "3–5 Aug") are not of an acceptable date format. Sorting on dates is hardly of any use for the pre-sorted first column, as all you could do is reverse the order, so I suggest we simply flag the column as unsortable. I have looked at various sort order templates we could use to make it work, but none are ideal. The easiest would be to hack up another version of Template:Sad (easy) so we could do something like "{sad2|3–|5 Aug}", but that doesn't seem worth the trouble for the very little utility of it. Any objections to marking Date(s) unsortable? NB for the record the other sort templates I looked at were Template:Dts which would not work for ranges, and 3 that could work awkwardly: Template:Hidden sort key, Template:Hid, Template:Sort. Rwendland (talk) 14:53, 10 August 2018 (UTC)

May v Corbyn polling - unsourced content
Why do we have several unsourced entries in the list? --Dweller (talk) Become old fashioned! 09:49, 17 September 2018 (UTC)

September YouGov poll
There is a YouGov poll in the 2018 table with no reference, no date other than 'September' and many missing figures. Does anyone know the source of this poll? Lancashire2789 (talk) 18:04, 22 September 2018 (UTC)
 * I was just going to comment on this. The latest poll on the YouGov website is 12-13 Sep and the numbers look suspiciously similar. That said YouGov/ Times polls are roughly weekly.Cutler (talk) 07:45, 24 September 2018 (UTC)
 * I am now going to boldly delete this as it is unsourced, looks suspiciously close to the 12-13 Sep poll and nobody has commented since Lancashire2789's post on Saturday.Cutler (talk) 12:44, 24 September 2018 (UTC)
 * From the Times source (published on 22 September): "The poll for The Times of 2,509 adults, taken this week, has the Tories keeping a four-point lead over Labour. The Tories are on 40 per cent, Labour is on 36 per cent, Lib Dems on 11, all unchanged, and Ukip on 5, up 1 point." Absolutelypuremilk (talk) 13:11, 24 September 2018 (UTC)

Survation polls sept-oct
There has were two Survation polls done within a short space of time with Westminister, Holyrood and EU polling:
 * The one for the Sunday Post on 28/09-2/10
 * The one for the SNP on 3/10-5/10. Jonjonjohny (talk) 03:52, 12 October 2018 (UTC)

September Lucidtalk poll
I have found this poll which asks people what they think about a United Ireland and with Westminster polling. However, it is really difficult to navigate, for the border poll has an option for a fully-independent Northern Ireland and for Westminster intention has a spoil ballot option which takes a sizeable amount of the vote and with not adjusted figure. Is it useful? Jonjonjohny (talk) 14:01, 7 November 2018 (UTC)

Protection?
There have been repeated attempts to falsify the numbers by IP editors. Is it time for some edit protection? Weburbia (talk) 14:11, 28 November 2018 (UTC)
 * I'm ambivalent on this - IP editors have sometimes spotted mistakes. Absolutelypuremilk (talk) 22:26, 28 November 2018 (UTC)
 * IP editors could still highlight mistakes on this Talk page, as on other semi-protected pages. I feel like the effort it takes to distinguish corrections from introductions of incorrect data is higher on this page than it is on others, so I feel like some kind of protection could be appropriate. Ralbegen (talk) 22:54, 28 November 2018 (UTC)
 * The persistent vandal 185.175.114.146 has recently been blocked for a month. We may find things are quieter on the vandalism front without the need for protection of the page. SpaceHamsterBoo (talk) 10:46, 4 December 2018 (UTC)

A suggestion to improve the consensus on events
I've just tried to add the party conferences, only to discover the consensus against adding events, which you seem to have to remind passerby editors every now and then.

There are good arguments on both sides —yes, some events do explain poll swings, but we may indeed dispute the boundaries between notable and insignficant ones.

So I come up with this proposal.

Wikipedia has pages called 2018 in the United Kingdom, 2017 in the United Kingdom, etc., which list the events from the year, including all the major events in politics, and more.

Why not add a link to those pages, for the avid poll analyst to compare? He or she would have a comprehensive list of what has been talked about in the major news outlets. It wouldn't make this article any longer. And if the wording is informative, it would avoid most of the attempts by non-watchers to add events they deem important.

Regards.

Kahlores (talk) 14:52, 26 September 2018 (UTC)


 * A reasonable good suggestion. The are many non-political events on those pages, but also many events that may well have effects on the polls. ~ BOD ~ TALK 15:23, 26 September 2018 (UTC)


 * It's a good idea. I'd include it in italics perhaps at the top of the table. --Dweller (talk) Become old fashioned! 08:53, 27 September 2018 (UTC)

This idea of having 2018 in the United Kingdom instead of adding events is ridicules. if this is the case we shouldn't add major events then why the HELL do we add English local and mayoral elections as sections. its a bloody joke. there is no logic agreement why they should be there while Mays Vote of Confidence should not be there. ps I suspect that UKIP polling may increase in the polls after the vote of confidence. MroWikipedian (talk) 09:27, 14 December 2018 (UTC)
 * There are numerous articles with political polling data across Wikipedia. Almost none of them have any events indicated. The few that do include notable elections and party leader changes. I can see no reason why this article should be different. Bondegezou (talk) 10:49, 14 December 2018 (UTC)
 * My opinion on this is the same as always: the tables are meant for opinion polls, not as an indiscriminate collection of events. Otherwise, this article would be named of the sort of "Timeline for the next United Kingdom general election", and its topical scope would be different. The controversial nature of adding events directly to the table is perfectly reflected by the sentence that I suspect that UKIP polling may increase in the polls after the vote of confidence. That's outrightly OR-ish (and CRYSTALBALL-ish, btw); an unsourced, personal conclusion which has been made even ahead of any opinion poll that supports such an electoral trend; obviously not a valid basis for supporting the addition of such content. In general, I've always said that trying to connect events to opinion polling is against WP:SYNTH, as if an event is added into the table it is automatically implied that it has a relevance to opinion polling; one that these tables are not meant to justify (obviously, events do have an impact on opinion polling, but it is not the scope of this article to identify which events do influence each opinion poll or group of opinion polls and in which scale). Having a separate timeline of events linked here is nice, though: readers have two separate listings (one for events, other for opinion polls) and they may reach their own conclusions without us pointing them specifically to any one of these. Impru 20  talk 13:32, 14 December 2018 (UTC)

Party logos
I wonder what others think of adding party logos to the tables? I've added them to the 2018 table as an example — Preceding unsigned comment added by Frog7 (talk • contribs) 23:04, 3 December 2018 (UTC)
 * I am not keen. I see parallels with MOS:FLAG, which discourages such iconography. Names are easier to read, so let's not clutter the table. Bondegezou (talk) 23:08, 3 December 2018 (UTC)
 * We would really have to see an example first. Onetwothreeip (talk) 23:16, 3 December 2018 (UTC) They are good but some of them are too wide. They don't need to be wider than the party name below the logo. Onetwothreeip (talk) 23:43, 3 December 2018 (UTC)
 * Don't like this at all. Don't see what it adds. It's cluttered. And it's more work for legitimate scrapers.Cutler (talk) 23:47, 3 December 2018 (UTC)


 * Isn't there an issue with WP:NFTABLE for this? I don't see that this change would bring benefit that justifies the use of several non-free images, as The use of non-free images arranged in a gallery or tabular format is usually unacceptable. Ralbegen (talk) 00:49, 4 December 2018 (UTC)
 * This; also, that naturally leads to inconsistency and in general I've moved towards opposing including images (especially as pure substitutes for text) in polling tables as forms of both excessive formatting and reduced accessibility for users. Mélencron (talk) 06:17, 4 December 2018 (UTC)

Weak Support Other opinion poll articles use it, so I personally dont have a problem with it on the condition that the thumbnail logos aren't used as wikilinks and the actual party name is still used in the top row. The links dont work as well. Another alternative is to use the logos in a party summary table that says what they polled last, amount of seats and other info Jonjonjohny (talk) 07:54, 15 December 2018 (UTC)
 * I don't support this, the names work perfectly well I think and the logos are not particularly well known. Absolutelypuremilk (talk) 18:36, 16 December 2018 (UTC)

Theresa May? vs Jeremy Corbyn or Generic Conservative vs Jeremy Corbyn
Seeing as May will not fight the next election, May v Corbyn polls at the bottom of page is redundant Mangotrue (talk) 17:51, 19 December 2018 (UTC)
 * May said she would not fight an election in 2022, which doesn't rule out her fighting an earlier one. In any case, that's the question the polls are asking. Absolutelypuremilk (talk) 19:38, 19 December 2018 (UTC)
 * We report what RS say, so let's keep doing that. Bondegezou (talk) 20:14, 19 December 2018 (UTC)


 * Also most of the polls are asking who would make a good Prime Minister now, so the polls are still very much valid. 2022 is still over 3 years away --BenBezuidenhout (talk) 11:45, 20 December 2018 (UTC)

YouGov poll 14-15 December 2018
This poll does not include headline voting intention figures. Someone tried to calculate these from an intermediate table by omitting don't know, would not vote and skipped. If you do this with other YouGov polls that do include headline VI, you will see the headline VI figures are nearly always slightly different, presumably due to some final fine-tuning of the weightings. Working tables should NOT be represented as headline VI. I have deleted this poll from the table. 82.9.115.128 (talk) 17:09, 23 December 2018 (UTC)

Latest Ipsos Mori poll
Which figures have been taken showing a dead heat on 38% between Con and Lab?

And is this our usual methodology? --Dweller (talk) Become old fashioned! 11:27, 8 February 2019 (UTC)
 * It seems presented oddly the the data we link to, but it does seem to follow our methodology - perhaps easier to see in their topline presentation page 2 (analysis of those likely to vote). All the parties add up to 100% (in fact 101% due to rounding), with the would not vote/undecided/refused presented outside the 100%. Rwendland (talk) 14:03, 8 February 2019 (UTC)
 * In the source we cite, the first page has a totally different set of results than the ones we're quoting. Why are we using table 3 instead of 1? --Dweller (talk) Become old fashioned! 22:22, 12 February 2019 (UTC)
 * The headline voting intention is what we normally quote, which is what the pollster predicts after all their adjustments for likelihood to vote, which in this case is the table on page 3. Absolutelypuremilk (talk) 23:46, 12 February 2019 (UTC)

YouGov MRP model again
The most recent YouGov "poll" (2-7 Feb) is not a straight poll but the output of their MRP model. Before the last election we segregated these under a separate section. We should do so again, surely.Cutler (talk) 15:41, 12 February 2019 (UTC)
 * Done it boldly Cutler (talk) 12:59, 13 February 2019 (UTC)
 * If you don't mind, could you please expand a little on why it's so different from a "straight" poll that it can't be treated as one? I confess I haven't looked into the details of the MRP model.  It's no doubt using more complicated statistical machinery than the regular polls, but if it's using new data to predict the same overall party support statistic as the regular polls, I don't immediately see why it couldn't be treated as comparable.  Thanks. SpaceHamsterBoo (talk) 19:52, 13 February 2019 (UTC)
 * All polls aren't "straight polls" in that they weight their data based on certain assumptions and adjustments. So we would need to hear what YouGov is doing that is so different. Onetwothreeip (talk) 21:19, 13 February 2019 (UTC)
 * This was debated extensively (though not by me) when it was used for the 2017 election and it was decided then to list separately. That is the format on the Opinion polling for the 2017 UK general election page. I have adopted the same here for consistency. The difference is explained in the "YouGov model" section which is sourced. Whatever happens we need consistency between the 2 pages.Cutler (talk) 22:44, 13 February 2019 (UTC)
 * I think I remember arguing against including it in a separate table at the time, but I think keeping it in a separate table is the right call here, as the national vote is determined based on the modeling of individual constituencies' vote. Mélencron (talk) 23:07, 13 February 2019 (UTC)
 * Well, I don't see why consistency with an older page "whatever happens" should be the overriding priority. Having said that, from what happened in 2017 it looks like we might expect YouGov to make regular releases of this model using a rolling, overlapping window of 50000 samples, only a few samples being new each week.  So subsequent releases of the model wouldn't be suitable for treating as new polls, even if this first one is.  So we'll need this separate table for subsequent editions of the model, and I suppose it would be easier and neater not to treat this first release of the model separately, even if it could be treated as a regular poll on its own. SpaceHamsterBoo (talk) 09:28, 14 February 2019 (UTC)
 * Was there a discussion during the last election about whether to maintain a table of seat predictions? I notice that the page just shows the latest prediction from each group that published one. Ralbegen (talk) 14:26, 14 February 2019 (UTC)

Comment - There is no such thing as a "straight poll". Every poll on this page uses some form of modelling & regional weighting; I don't see that there is a level of increased detail at which a poll ceases to be a poll and becomes another animal. 92.3.150.74 (talk) 13:06, 17 February 2019 (UTC) edit: Having had a scan of the previous arguments about this, I'd not describe those objecting to it in order to avoid NPA. It appears the MRP was regarded as npov pro-left for showing a better result for labour than other polls (turned out other polls were wrong, mrp was right). I would not, personally, put much weight on any of the arguments they made. Theoretically there's a problem with rolling means (although I think other polling articles have no problem with them), but these mrp releases aren't actually those, the concerned users appear to have misunderstood. IMO just put it in the main table. 92.3.150.74 (talk) 13:19, 17 February 2019 (UTC)