Talk:Opinion polling for the 2017 United Kingdom general election/Archive 4

Archive 1

Archive 2

Archive 3

Archive 4

Qriously poll

Does anyone know much about this poll and how it compares to others in the table? They do their polls by "sending short questionnaires via mobile phone", which I assume is an SMS/text but it isn't clear. Here's the article about the poll - Wired article. At the very least it seems different, in terms of method and result. ZeroRPM (talk) 23:25, 7 June 2017 (UTC)

I believe that they conduct polls with in-app surveys; they've also polled other elections and referendums (2016 UK, 2016 U.S., 2016 Italian, 2017 Dutch, 2017 Turkish) with varying degrees of success (hit, several misses, hit, miss, extreme miss). Not a member of the BPC. Mélencron 23:29, 7 June 2017 (UTC)

"Not a member of the BPC" is a fact but not relevant to whether the poll is included in the list. The main criterion should be that it isn't a "voodoo poll", i.e. statistical techniques have been appropriately applied in the sampling and the weighting. If a poll is presented as non-voodoo I think Wikipedia should list it until proven otherwise, albeit with a note that the pollster has not provided any evidence. Dadge (talk) 00:30, 8 June 2017 (UTC)

Random popups on smartphone apps. Yes I'm sure that will give you a true representative sample for over 65s. Ha this page has gone from respectable to messy to downright propaganda/advertising. 88.98.211.79 (talk) 00:34, 8 June 2017 (UTC)

Tom Melencon to the rescue. https://twitter.com/TomRugbyFan/status/872597379949764608 88.98.211.79 (talk) 00:38, 8 June 2017 (UTC)

Guys guys "Programmammaticic Sampling" which mean they asked the Saudi Orb from last week. https://pbs.twimg.com/media/DBwXmt2XoAAWIhd.jpg Shame they can't spell Methodology right. Rolls eyes 88.98.211.79 (talk) 00:41, 8 June 2017 (UTC)

"A new terms" for this page: Joke 88.98.211.79 (talk) 00:58, 8 June 2017 (UTC)

Don't you have a better use of your time than to endlessly complain on talk pages all day? (If you're going to do that, the least you could do would be to remain civil and engage constructively rather than badgering other editors and constantly insinuating that those arguing against you are doing so because they're politically biased.) Mélencron 01:05, 8 June 2017 (UTC)

Don't you have a better use of your time than to endlessly be the first to defend any spurious poll as long as it shows a left wing boost? And if you read my comments you will see that I attacked nobody. I was taking the piss though as this page is nowhere near the respectable, logical debate required hence things like Spuriously being included no questions asked. I'm sure you'll be the first to reply as usual. 88.98.211.79 (talk) 01:15, 8 June 2017 (UTC)

It's on my watchlist. I've got 2,285 articles on my watchlist and I check every single edit on all of them. I'll see your comments. Your insinuation that I'm defending every single poll that gives a "left wing boost" is obviously incorrect (and irrelevant to this discussion – I should expect that my personal political views should not an influence on others' judgment regarding the inclusion of specific polls or pollsters), since I've only defended the inclusion of a single poll that might be described as such (YouGov), and your constant insinuations about my username (it's a portmanteau which I chose because I found it absurd and sounded rather nice, not because I adhered to either of their views, and I'm not sure if you're even seeing the "r" within it). I'm not sure if you read my comment above correctly, but it should be clear enough that I'm against the inclusion of the Qriously poll, as indicated by my mentioned of the fact that they're not a member of the BPC. (Norstat and SurveyMonkey, by extension, should also be removed or separated.) This was the eventual solution used on the EU referendum polling article at one point. (Norstat and SurveyMonkey are both legitimate pollsters but not members of the BPC and, along with Qriously, don't release their tables; Norstat polls in Norway, naturally, while SurveyMonkey is a current member of AAPOR, the main U.S. polling organisation). Mélencron 01:24, 8 June 2017 (UTC)

It's not clear to me that we should include polls from pollsters that aren't part of the BPC. The current EU referendum polling article merely mentions them in prose saying that none of them are members of the BPC; they aren't included in the table, and I don't see a good reason not to be consistent with that here. Mélencron 01:24, 8 June 2017 (UTC)

Please point to me where you explicitly said Qriously should be removed? Actually, do it right here right now below me. REMOVE not a member of the BPC or commissioned by a paper of record/reputable media. I'm happy for non BPC polls to be put in another table but certainly not Qriously anywhere. There should be a basic entry level requirement of random sampling or quota sampling. If new fangled systems are proven to be correct in the long run, they could be included in the future but certainly not now. 88.98.211.79 (talk) 01:40, 8 June 2017 (UTC)

No objection to throwing out Qriously; an employee added their poll on the Dutch polling article some months back and I immediately junked it for obvious reasons (five major polling agencies in the Netherlands, Qriously isn't one them and its methodology is obviously dissimilar from the rest). Mélencron 02:16, 8 June 2017 (UTC)

Er, why, @Melencron: should you just junk it just like that? If you read further below you'll see that Qriously were the only ones to get the Dutch election right. Just because their methodology is dissimilar? Seems to me like a very good argument for their inclusion! Boscaswell talk 08:31, 8 June 2017 (UTC)

Where are you getting the obviously self-promotional claim that they got the Dutch election right? On balance, by my calculations, they were beaten by Kantar Public (86.7% accurate), Ipsos (89.3% accurate), Peil.nl (84.0% accurate), and I&O Research (85.3% accurate), and themselves were only 82.7% accurate. Claims are obviously selective; they massively missed the Turkish referendum result (16 points off and had "yes" leading in Istanbul and the Kurdish southeast, which is utter garbage). Mélencron 12:06, 8 June 2017 (UTC)

Well, why don't you go ahead and remove it? I fail to see how a senior user who watches this polling page like a hawk feels the need to ask for every little change. Just edit the page. On the other side, we have that user who put in the YouGov unilaterally and left it there for a week just based on his opinion. There is enough weight here, on previous pages as you say and simply on the sheer methodolgy (sic) to remove them. Just do it. 88.98.211.79 (talk) 02:29, 8 June 2017 (UTC)

The fact that Qriously's methodology is so different from other polls, that they are not part of the BPC, and that they have been very hit and miss would seem to me to be grounds to not include it in this list. schetm (talk) 03:47, 8 June 2017 (UTC)

If we're excluding Qriously on the basis that we don't like the methodology and it isn't a member of the British Polling Council, then surely we also have to remove the SurveyMonkey poll for the same reasons? There's a SurveyMonkey poll on this table, and on the 2015 election tables. There's no consistent rule to remove one and keep the other. Zcbeaton (talk) 07:29, 8 June 2017 (UTC)

No consensus was reached on SurveyMonkey. I argued against its inclusion initially but yesterday reinstated it after its deletion from the table. I see no reason to delete Qriously. If they correctly predicted Brexit (as they claim) then who am I to argue against it's inclusion? The famous BPC members all got that wrong, so.......a firm Keep. Boscaswell talk 07:41, 8 June 2017 (UTC)

That poll does not belong on the polling list. If we allow this one, any other random online poll will have to be included. Stick to the recognised and respected polling organisations, the only exception to the rule appears to be SurveyMonkey who polled previously and were fairly in tune with the result. and have had legitimate media coverage. This specific poll does not. It does not belong in this article. Obs2017 (talk) 07:46, 8 June 2017 (UTC)

we have a single source for this poll, and wired would not be considered a legitimate news source anyway. At least the survey monkey is reported by a news source and picked up by other news media. If this one was covered by other news sources i would not oppose its inclusion. but as far as i can see its got just ONE wired article about it and that is it. that is not reliable enough for inclusion on this list. Surely the threshold should be multiple news sources reporting the figure or it is by a member of the British polling council. one website with an article on it is not enough! Obs2017 (talk) 07:52, 8 June 2017 (UTC)

Who says that Wired is less reputable than The Sun? Pischaw! Obs2017, what exactly is it that makes you dislike the inclusion of the Qriously poll? I hope it's not that its summary findings are a lead for Corbyn. They are not BPC members. In my book, they must almost by definition be more reliable than those who are, because BPC members have been shown to be rather inaccurate, in the 2015 GE and the BREXIT referendum. But Qriously were not inaccurate over BREXIT. Therefore if we were to exclude their poll, even though it - shockingly - predicts a majority of Labour votes, we would be failing in our encyclopaedic duties, in a kind of ostrich-with-head-in-sand manner, n'est-ce pas? Boscaswell talk 08:09, 8 June 2017 (UTC)

Keep it, they were pretty spot on in brexit, dutch elections, turkish elections, being the most accurate pollster in the dutch election. They proved their reliability as a legitimate pollster. --Ohartro (talk) 07:54, 8 June 2017 (UTC)

If this unreliable crap poll which is wildly out of tune from ALL other polls, and which has not a single reliable news source is to remain, can someone at least put it in the correct positioning on the table in line with polling dates. By putting it at the top of the table when it started polling on the 4th and there are several there that started after the 5th, it makes it even more misleading as though its the most recent poll. But i repeat again. this poll has no place on this polling table. we may as well include a survey of a dozen people asked down the local pub. Obs2017 (talk) 08:02, 8 June 2017 (UTC) If the "poll" is to remain, it belongs between ComRes and Panelbase, not right at the top. Obs2017 (talk) 08:08, 8 June 2017 (UTC)

Keep - this poll was published by Wired, a reliable source with a large audience. This, in my mind, is the only reason to add or remove a poll. Rami R 08:27, 8 June 2017 (UTC)

Keep for now - To avoid us being accused of political bias I think we should keep the poll for today. Yes I am aware this is a non scientific poll done via click bait links with no discernable methodology and that the history of these polls by Qriously is pretty bad at predicting anything like an accurate result. So on June the 9th I'll be changing my vote to Delete, but for now, as I said Keep in the interests of political neutrality. - Galloglass 08:53, 8 June 2017 (UTC)

Er... @Galloglass: Qriously say that they got Brexit right and another editor reported (further up in this discussion) that their prediction of the Dutch election result was the best there was. That doesn't sound to me to be in line with "Qriously is pretty bad at predicting anything like an accurate result." Maybe they should be at the top of our list of those to be included? Because as far as I am aware, none of the supposedly reputable polling org's fared well over Brexit. Boscaswell talk 10:34, 8 June 2017 (UTC)

Even a stopped clock is correct two times a day dear boy. - Galloglass 10:40, 8 June 2017 (UTC)

And if by some chance they are correct and all the other pollsters are wrong I'll be happy to leave my vote as Keep. - Galloglass 10:51, 8 June 2017 (UTC)

Keep - for the reasons exposed above, as well as my own reasons for keeping the SurveyMonkey poll here. Impru20 (talk) 09:07, 8 June 2017 (UTC)

Keep - The results are not that different (within 2 points) of the Survation poll, so it is not massively out there, it has a record of polling and has been reported in a news outlet. Removing it simply because it shows a Labour lead is making a call about which results 'seem' right, which is not our job. Notably they've also been acknowledged as a real poll by Nate Silver of 538 https://twitter.com/NateSilver538/status/872593361861058560

The Tory % in Survation on 3rd June was 40%. In this poll it's 39%. The Labour vote has hit 40% in 3 polls in the last week. In this poll it's 41%. This is not "wildly out of tune from ALL other polls" it's well within the margin of error of other pollsters. Just because it shows an outcome that is not widely predicted is not a reason to exclude it. That leads to herding. FriendlyDataNerdV2 (talk) 09:39, 8 June 2017 (UTC)

FriendlyDataNerdV2 - Nate Silver is just posting a copy of this pages latest polls, not an endorsement of this poll. - Galloglass 10:38, 8 June 2017 (UTC)

Keep It was printed in a reliable source. While it may not be a traditional method, i.e. calling landlines, such newer techniques are becoming more common. Published in an RS is good enough for me, as long as info about the sample is published, and we have that. Dbrodbeck (talk) 12:38, 8 June 2017 (UTC)

I recommend that anyone doubting the pedigree of young upstarts Quriously read the 3rd para of this blog page of theirs... http://www.qriously.com/blog/brexit-trump-qriously-accurately-calls-italian-referendum/# Boscaswell talk 12:41, 8 June 2017 (UTC)

Keep Assuming the info published is in line with other polls I'm happy with it there, seems to be reasonable in terms of sample size and approach. For me it will be quite interesting to see how accurately their method predicts the results, especially considering YouGov revised their methodology recently which seemingly brought their projections in line with other polls, it's quite interesting to watch the fight of the methodologies pan out. ZeroRPM (talk) 17:28, 8 June 2017 (UTC)

Survation - Keep or Delete?

The methodology given suggests likelihood of voters 18-24 as 82% when the long term historical average, even unweighted, has been around 50% (https://pbs.twimg.com/media/DBFIdapWAAAxVtM.jpg:large). Data given suggests bias and out of sync numbers with other polls bear this out. No information on why it selected it's [sic] sample and the reason behind such a model relying solely on voter's own number likelihood (given it's a young market research company, the people chosen may have been too many young or too many politically active like its employees, which would be a major bias). Only source is a single, morning chat show with no clear political analysis. I push for a swift Delete 90.215.121.192 (talk) 19:37, 30 May 2017 (UTC)

I think a big part of this discrepancy is that Survation's poll says "82% of 18-24s would/already have voted", whereas the Guardian poll says "50% will vote". My interpretation is that 82% probably won't vote this time, but 82% have voted in the past. I could be wrong, but even then, if other sources continue to see Survation as reliable, it should be reported. UpperJeans (talk) 00:18, 31 May 2017 (UTC)

82% turnout modelling for under 25 is unheard of even in states where the dictator falsifies results. Survation are either taking a massive gamble here or will be proven massively wrong come June 8 or have chosen a completely unrepresentative sample of politically motivated under 25s. As ICM mention, even UNWEIGHTED, that number historically is barely over 50%. If Survation are right, the whole polling industry will either need to be scrapped or revolutionised and if they are wrong, surely then they do not merit inclusion in any other future such page on Wikipedia? That's why I would err on the side of caution and remove them here, especially if we are following very stringent criteria for other pollsters. 90.215.121.192 (talk) 01:17, 31 May 2017 (UTC)

Survation's methodology and results are controversial but they are not so unambiguously wrong as to merit exclusion from the table. Survation are a member of the British Polling Council and follow the proper rules of the trade. Respected polling websites report Survation results along with those of other pollsters (e.g. http://ukpollingreport.co.uk/blog/archives/9899 ). And Survation's overall results are hardly out of line with those of some of the other pollsters (such as YouGov). All we can say at this stage (and maybe there should be a section on this) is that there is a big divide between one group of pollsters (YouGov, Survation) who are predicting relatively small Conservative leads, and another group (ICM, ComRes) who have double-digit Conservative leads. Unless they converge by polling day, at least one of these two groups of polling companies will have egg on their face. But at this stage it would be wrong for Wikipedia to take sides and to exclude one of the groups or one of the companies. 86.174.248.121 (talk) 01:37, 31 May 2017 (UTC)

Def Keep. However, maybe the should be a short paragraph explaining the basic methodology of the various pollsters. I read somewhere that the basic raw data figures across the pollsters are essentially the same, but what varies is that since the 2015 election errors some pollsters weigh things like age and social class more heavily in to the voting intentions. I am not sure if these extra weighing strategies have been proven.-- BOD -- 11:30, 31 May 2017 (UTC)

A bit late in the day, but, emphatically, Keep. Survation are a member of the BPC. That they make different assumptions/weightings to, for example, ICM, is not for us to judge. Boscaswell talk 08:41, 7 June 2017 (UTC)

Survation vote share forecast was closest to the actual result. It makes me wonder.......this thread was started by an Anon post. Why the heck do we even bother to take into consideration anything that Anon posters do? One or two of these caused havoc yesterday morning, together with one very angry signed in member, over the inclusion in the table of the Qriously poll. *rolls eyes* Boscaswell talk 09:56, 9 June 2017 (UTC) Boscaswell talk 09:56, 9 June 2017 (UTC)

DELETION of Surveymonkey "polls"

There is a consensus majority against keeping the SurveyMonkey polls with the rest of the standard polling house polls.

Aside from that, the underlying issue is not simply the unusual and larger numbers polled, but the absence of polling methodology or any information that makes these polls reasonably and credibly comparable to traditional polls. This is not a matter of what the polls themselves say, it's again a matter of putting appropriately similar categories of data together. It is not reasonable or credible to permit SurveyMonkey polls to be allowed to be seen in the same light as polling work done by professional psephologists - they don't match the other data either qualitatively or quantitatively, nor do they have any indication of shared characteristics with the other data, and as such they could distort legitimate data. It is observable that they do resemble outliers along the lines of the YouGov rolling models; and the issue here is the bigger picture credibility of polling itself. If opaque and dubious polls such as YouGov model and SurveyMonkey are allowed to gain legitimacy by association with proper polling like ICM and ComRes and such, which use similar sized datasets as well, then the whole perception of polling will suffer. The results of the actual vote seem likely to result in some polling sources losing credibility, and those that are deviating from established norms seem most vulnerable to that outcome. By all means, create a new table for user-generated pop polling, but don't lets mix it up with more rigorous stuff. As an aside, I would also suggest that phone polls are different data and should be separated from the others, simply because they are a different category of data. I believe the consensus is established on this page for the removal of the SurveyMonkey xx,000 sized polls from the rest. I also note that there's a strange YouGTov poll of about 12,740 somewhere in the list that doesn't look like it should be there either. 18:40, 7 June 2017 (UTC)Macdaddy

No, there isn't. Please read WP:CONSENSUS - we don't do simple majority voting here. If you think the discussion bellow has not reached a satisfactory conclusion (or a conclusion, full stop), you're welcomed to pursue further dispute resolution steps, such as an WP:RFC. Rami R 19:00, 7 June 2017 (UTC)

Agreed. There never was a consensus. Boscaswell talk 07:53, 8 June 2017 (UTC)

Disappointed by this decision (if it's been properly arrived at) which seems like censorship. It'd be better if all polls were listed. Those polls without published data could be flagged. But they have been published by a reputable company in reputable newspapers (yes, even the Sun comes under that heading) and are noteworthy inasmuch they are quoted by election experts, so they meet the key criteria for inclusion. Dadge (talk) 19:08, 7 June 2017 (UTC)

I'd like to point out that 212.105.162.130 (the one calling himself "Macdaddy") has just left what seems to be some sort of intimidation message in my user talk page in relation with the previous discussion in this talk page. I comment this here because it's directly connected to a discussion in this talk and because he's making "commands" to me so as to how should I act, as well as because he seems to keep engaging in the same behaviour here as previously. I don't know if I should report him right away (since this could constitute WP:HUSH, and anyway he's clearly going personal now) or I should first seek some advice on the issue. I just saw he did something similar in IMarc89 talk page. Impru20 (talk) 20:16, 7 June 2017 (UTC)

So, let's just summarize: One or two IP users have been carrying out systemic attacks on certain polls on this page. Trying to remove them unilaterally, making false statements as to the consensus among editors and accusing others of crass political bias. Now after the election we can conclude that the specific polling outfits attacked were among the most accurate in their predictions. I am sure this will cause no further reflection for the IP users in question. But I should hope that it will provide a valuable lesson for all other registered editors going forward. Anjoe (talk) 13:11, 9 June 2017 (UTC)

Practical Issue about Moving Averages

A lesser issue worthy of consideration - when multiple polls were listed on 7 June, the rolling average approach fell down as effectively the first few polls made no contribution to the final point, affecting the accuracy of the final figure. Ironically if all the 7 June polls had been averaged the final figure would have been reasonably accurate. Would it be better to graph daily averages, then produce a moving average of these daily figures? Stub Mandrel (talk) 08:02, 9 June 2017 (UTC)

I'm interested in this too. Those final polls just make the rolling average line go up and down, which does look a bit messy. I also didn't know how best to represent the date range of the poll, as I think this is significant, especially when the poll was done over a week long period (I used the mid-point date of the poll). In a way I think the rolling average isn't really a great addition, it pushes a time lag on the data and you need to somewhat arbitrarily choose a period to smooth over, also there is no representation of the percentage error associated with each poll. Ideally I think we'd do some kind of statistical analysis to give a most likely value and an error range around that line - all calculated from inputting all poll values, date ranges and percentage error (would that constitute original research though?) Generally I think it should look something like this: Opinion polling in the Canadian federal election, 2015 ZeroRPM (talk) 13:10, 9 June 2017 (UTC)

Certainly worth looking at. I see that if you click on the image you get a link to the source code. Stub Mandrel (talk) 17:22, 9 June 2017 (UTC)

An alternative approach would be to plot a kernel smoother. Bondegezou (talk) 21:34, 9 June 2017 (UTC)

Headline Parties for Next General Election Polls Page

Both UKIP and Greens earned under 2% of the over all popular vote in the 2017 election. I think this is sufficient to move them into the 'other' collumn as they are not polling major number.

If they are going to stay though, I would suggest adding Plaid Cymru as the only 'major' political party that is not already included (keeping in mind most polls are GB only and don't include Northern Ireland)2A01:388:27F:350:0:0:1:2C (talk) 22:54, 9 June 2017 (UTC)

Seconded. Both parties saw huge wipeouts and the Greens have elected only 1 MP for the 3rd election in a row. Neither are major parties. FriendlyDataNerdV2 (talk) 14:05, 10 June 2017 (UTC)

We follow reliable sources -- that's how Wikipedia works. If future polling regularly includes a party, so do we. If they don't, we don't. That's how the decision is made: not by us assessing how the parties did in this election. Bondegezou (talk) 14:55, 10 June 2017 (UTC)

But polls regularly included Greens/UKIP from 2010-13...they polled 1-2% and were not included in tables. We're not talking about editing this page, but how we format the next one. FriendlyDataNerdV2 (talk) 21:24, 10 June 2017 (UTC)

If UKIP and Greens are included on a survey they need to be recorded in their column. - Galloglass 11:17, 11 June 2017 (UTC)

What Can We Learn From This?

Regardless of the various arguments each way, it appears that, by and large, the most accurate polls were those whose inclusion in this page were fiercely contested by many editors. There is surely a lesson to learn that polls making use of new media to gather their data may be more valid than traditional polls. Yes, this may be a reflection of the high turnout of Millennial voters, but the turnout was high across the board. So, do we continue to have the arguments rehearsed above, or do we learn and move on? Stub Mandrel (talk) 07:58, 9 June 2017 (UTC)

I certainly think that in the future we should avoiding removing polls that don't align with the rest. It exacerbates the problem of herding. We should include any poll that's been widely discussed by people or reported on by a publication with any shred of credibility. It's not our duty to choose which polls are reliable before the result is even known yet. --Ohartro (talk) 08:31, 9 June 2017 (UTC)

Quite. There were several editors who seemed determined to artificially herd the polls and much of the reasoning seemed to lie in the fact that they were different from the rest - which is not our job to determine. In the end, Labour received 41% of the vote, exactly what the much-disputed Qriously poll said. There's something to learn from that. FriendlyDataNerdV2 (talk) 14:03, 10 June 2017 (UTC)

The problem being, including surveys done by non pollsters is very questionable. Thats why in the past we have only included BPC polls. - Galloglass 11:21, 11 June 2017 (UTC)

What's a "non-pollster"? Nearly all political polls are done by firms who spend most of their time doing market research. Just because a company is new does not mean their polls are automatically invalid. Survation was new (and untested) 6 years ago, now they've emerged as the most accurate pollsters of 2017. We can't pick and choose who to include just because the results seem unlikely. FriendlyDataNerdV2 (talk) 09:25, 16 June 2017 (UTC)

I would suggest that the poll itself needs to be published/available in order to be included. If we can't see their data and we don't have any information about how they conducted the poll, then it's not possible to evaluate it and so we should err on the side of caution. This isn't a problem with the Qriously poll, because they put their full dataset on their website, but the Survey Monkey polls seem to have been basically private polls commissioned by the Sun, with no data or methodology publicly available. Formerip (talk) 11:31, 17 June 2017 (UTC)