Talk:Opinion polling for the 2017 French presidential election/Archive 3

Archive 1

Archive 2

Archive 3

The Le Monde-Cevipof poll, published 4 April 2017

Can someone enter the details for this, please. Details are behind a paywall at some sites, but I think it will be in the printed version of Le Monde. The number of people surveyed was 14300. Greengreen2 (talk) 11:31, 4 April 2017 (UTC).

Details are now in the table and doubtless Mélencron will update the graphics later.

Please note that this gives this poll a lot of weight (6 to 8 times the weight of other polls) on the 14 day weighted graphical representations just ahead of the first round of voting so you can expect some "anchoring" of the graphics, even if opinion shifts rapidly in the next couple of weeks. B A Thuriaux (talk) 15:12, 4 April 2017 (UTC)

For what it's worth, I believe that the Cevipof barometer will increase in frequency over the next few weeks – this is wave 12bis, there were 20 originally planned (I'm not sure if they're counting this one and the additional March wave, that would still leave 6 remaining). Mélencron (talk) 15:23, 4 April 2017 (UTC)

Regarding the recent poll with Mélenchon at 19.5%

Followers of French presidential polling may have noticed this recent poll by Le Terrain/Scan Research with some surprising results. I've personally decided to hold off from including it on the basis that it doesn't appear to have been checked by the French polling commission at this time (i.e., veracity unclear), not to mention that CATI polling in general is of questionable reliability, as noted by the commission. It also doesn't appear to have carried out voting intention surveys previously; even the Dedicated Research poll was carried out by a Belgian firm which regularly fields surveys in its own country (and has its own panel in France which was used primarily for market research before they were commissioned by La Libre/RTBF to carry out a survey in France compliant with French regulations). The alarm bell for me here is that it isn't covered in the French press at all despite having been published for a full four days now. Mélencron (talk) 12:49, 3 April 2017 (UTC) (@Impru20:)

Sorry, I was editing to add further second round results from that poll before I saw this, and I rv them as soon as I found out that this could be controversial. I included it because I saw it in being reported in Electograph Twitter (here). There seems to be an ongoing discussion on the French wikipedia article on this poll (here). I think that, in case of doubts as we have here, we should limit ourselves to mirror what they choose to do on this issue, as I think the same arguments would be applied here. For now, it looks like the issue is same as here: that someone else does indeed report them before adding it (as per WP:PRIMARY). I don't know if Electograph counts for that. Impru20 (talk) 16:06, 3 April 2017 (UTC)

French Wikipedia has now decided to include this poll, so we should probably include it too. Chessrat ^{(talk, contributions)} 23:53, 5 April 2017 (UTC)

French Wikipedia does a lot of stupid things. That doesn't me we should do the same. (Seriously, listing polls of specific demographic groups? Completely beside the point.) Searching for results for sondage Terrain "Scan Research" on Google produces literally just five results, of which two are self-published, one is Wikipedia, one is reddit, and one is a pro-Mélenchon blog. There is no indication of credibility, this firm hasn't conducted any public voting intention polls before, and the polling commission hasn't published any notice on this poll. I'm assuming this firm has zero credibility until somehow reported in the press. Mélencron (talk) 00:02, 6 April 2017 (UTC)

Well, actually, the Financial Times does report it. And I wouldn't think something is "stupid" just because I don't like it. The reasoning they give for its inclusion in the French Wikipedia seems convincing, and indeed, if the FT does report it I think it has some notability. The polling institute does seem to be legit, and it's a member of SYNTEC Etudes just as the other polling firms reported here (check). I also can't see anywhere in Wikipedia policies or anything that, for pollsters to be considered, they have to "have conducted public voting intention polls before" (which would lead to a paradox that would essentially bar new pollsters from being considered—despite them actually conducting the polling—just because they didn't conduct voting intention polling before) or that "the polling commission must publishe a notice on the poll" (there are countries where "polling commissions" not even exist, much less publish anything on polls). I'm in favour of its addition, in light of new evidence. Impru20 (talk) 00:34, 6 April 2017 (UTC)

The FT includes it...because those poll aggregators copy off of Wikipedia, as a very quick glance should show (they're using the French version of the article). Note that the French article uses only non-overlapping results for Ifop-Fiducial polls and weekly results for the OpinionWay ones, per a lengthy talk page discussion there. The FT is hardly "reporting" it, in any case; though publications like it (Reuters and Politico Europe) often report on French polling, they've ignored this one entirely, as they should. I'll admit that it's beside the point that the firm hasn't conducted polls before, but that doesn't exactly lend any additional credibility to this poll, which is none. (By the way, there's precedent for this on Opinion polling for the United Kingdom European Union membership referendum; strangely, the Qriously poll used to be listed but is now excluded entirely.) Mélencron (talk) 00:40, 6 April 2017 (UTC)

It's not for us to say why they do include it or not. They've considered it as relevant enough to include it, and you sought one such a notable source. If they're indeed using the French version of the article (rather than the English version of the article) then maybe they think the French one is more complete and they use it over this one. We surely can't now, but the fact is that they report it, and we're here to report what sources say, not to discuss why they say it. Ok, some others may not include it, but that's not what you asked for before (you literally said I'm assuming this firm has zero credibility until somehow reported in the press. The FT reports it, and now you add additional requisites for you to consider it valid). Further, you don't respond to anything I've laid down to you in my previous comment. Is this an invalid pollster? Again, that you think this poll should not be here doesn't mean it shouldn't be here. Some sources indeed report it, and there is no proof that Le Terrain is not a valid pollster by facts (to the contrary, actually). Impru20 (talk) 10:06, 6 April 2017 (UTC)

Is there some mechanism whereby the poll reports its methodology and a detailed breakdown of its findings, in line with good practice? If they're refusing to disclose the full details of how the poll was carried out or can't provide that they used a representative sample then I don't think it should be included. If they are largely publishing a complete set of data somewhere (maybe there's a good .pdf to link to) then I think it should be. I don't want to see polls shown when the sample was self-selected participants on a website that leans heavily to one political persuasion, for instance. Maswimelleu (talk) 10:42, 6 April 2017 (UTC)

They do indeed. The methodology seems to have been telephonic-based. There doesn't seem to be anything wrong with the poll, aside from giving slightly higher results for Mélenchon than other polls (and lower for Fillon), but could be explained as a result of the poll's margin of error (we've indeed seen some Mélenchon's strong growth recently, so this would be in line with it). Impru20 (talk) 10:51, 6 April 2017 (UTC)

Looks good enough to include, at least in my opinion. I agree that the poll looks like it's inline with a trend and could have spoken to people on an especially good day for Mélenchon in the media. I can see no harm in including it provided that it has an appropriate weighting. Maswimelleu (talk) 11:19, 6 April 2017 (UTC)

Are you not at all concerned by the fact that a Google search for this poll ("Scan Research" Terrain sondage) (edit: within the past 7 days) brings up literally just 7 results (of which none are not self-published, social media, or the French Wikipedia)? This poll hasn't been reported on elsewhere at all (and no, the FT doesn't count – they're just riffing off of the French Wikipedia). Mélencron (talk) 14:20, 6 April 2017 (UTC)

I don't think that should be a concern. The poll clearly uses a reliable methodology, and FT is a reliable source regardless of where they get their information from. Besides, the Terrain polling company has an article on French Wikipedia; it's not just some obscure organization with no credibility.Chessrat ^{(talk, contributions)} 16:56, 6 April 2017 (UTC)

I agree with this completely. Impru20 (talk) 17:10, 6 April 2017 (UTC)

Well, I actually obtain 433 results by searching for "Scan Research" Terrain sondage [1]. And 187,000 if I search for it without the quotes [2]. And, if I search for it with "opinion poll" instead of "sondage" (that is, in English rather than in French) I obtain 531,000 results. So, I think the search results depend mostly on what you search for; obviously, if you search for a very restricted version of the term, you'd obtain few results. I don't think any other pollster has been subject to such a scrutiny or have been required to comply with such restrictive requisites in order to be considered. That you yourself chose to dismiss the FT source is your call, but I think that if the FT itself has chosen to reflect the French version of this article rather than the one in the English wikipedia, it could be because it finds it more complete or accurate or whatever... and such a fact can't be just dismissed because of it. The poll's methodology seems legit; the pollster seems to comply with legality and is a member of SYNTEC Etudes as the other pollsters; it has been reported by some notable sources; arguments for its inclusion are strong and no argument against it can be brought other than personal opinions. I acknowledge that it seems to give slightly more pro-Mélenchon results than others, but that seems in line with Mélenchon's recent surge and could be justified by the poll's margin of error or just because it happened to poll on a specially good time for Mélenchon (no other pollsters' validity has been discussed on the basis of the results they give). The presentation also seems ok. Also, given that we're constantly bombed with the day-to-day trackings of other polls, I believe that this one can hardly have any weight influence compared to others. So... I say that we add it. Impru20 (talk) 17:09, 6 April 2017 (UTC)

The commission des sondages has published a notice for this poll, so I think it's fine to include now. Mélencron (talk) 17:15, 6 April 2017 (UTC)

Thank you for adding that poll to all of the tables. :) It looks like that poll still isn't included in the graph, though. Chessrat ^{(talk, contributions)} 04:13, 8 April 2017 (UTC)

Cheminade as far-right

Apologies for editing twice this morning in short succession, the issue wasn't clear on preview. Ultimately, is there a reason Cheminade is the furthest right candidate, in a chart that is otherwise going left to right politically, from Arthaud to Le Pen? In this Vote Compass he's clearly with the left-wing candidates. —William Quill (talk) 17:47, 10 April 2017 (UTC)

I'm using the same order as that given in the pollsters' files. Almost all of them have him on the farthest right/down position, hence his position in the table. (The current table order is based on the original ordering of 11 candidates in the Ifop-Fiducial rolling polls).

Pollster	Order
Odoxa	NDA	FA	JC	MLP
Sofres	NDA	FA	MLP	JC
Elabe	FA	NDA	MLP	JC
Ifop	NDA	MLP	FA	JC
Harris	NDA	JC	FA	MLP
BVA	NDA	MLP	FA	JC
Ipsos	NDA	MLP	JC	FA

Mélencron (talk) 18:02, 10 April 2017 (UTC)

Fair enough! That makes sense for good practical reasons. —William Quill (talk) 18:05, 10 April 2017 (UTC)

The two most recent Harris polls

Two polls are listed for Harris: one with data collected from 2812 people on 18-19 April; the other, from 962 people on 18-20 April. What's going on here? Are these polls discrete? They don't seem rolling in the usual way. — Preceding unsigned comment added by Greengreen2 (talk • contribs) 20:42, 20 April 2017 (UTC)

They're discrete polls conducted for two different clients. The prior poll was conducted for France Télévisions (released bi-weekly early Thursday morning); the latter for LCP/indeed (weekly Thursday afternoon). Mélencron (talk) 20:50, 20 April 2017 (UTC)

Greengreen2 you can read them and see for yourself.--Gciriani (talk) 21:07, 20 April 2017 (UTC)

Le Terrain/Scan Research

Because no major media outlet spoke of them, we had a long discussion on these polls admissibility over at fr:Discussion:Liste de sondages sur l'élection présidentielle française de 2017#Scan Research le Terrain 17-04. Our consensus was to wait for citation by credible WP:Secondary sources before including it, even more because it is currently fr.wikipedia most viewed article. So we were happy to have it included by the Financial Times aggregator. But it appear it was already included in this en.wikipedia article, which seems to be the origin of the FT inclusion. It's WP:Circular and lowers the quality standards of the wikipedia project.--Marc Lacoste (talk) 14:01, 21 April 2017 (UTC)

Marc Lacoste Actually, it was included here AFTER the FT included it, as we already discussed here. So this is not the origin for the FT inclusion and it's not WP:CIRCULAR. It could've copied it from the fr.wikipedia article, but then, other poll aggregators did not include it back then, so it was actually the FT's choice to include it. Impru20 (talk) 14:29, 21 April 2017 (UTC)

The first one yes, discussed on 3-6 April, but not the second one, added on 17 April--Marc Lacoste (talk) 15:33, 21 April 2017 (UTC)

Marc Lacoste Well, if the first one was accepted, I don't think there's a need for such requirements for subsequent ones if it's proven as a genuine poll from the same pollster, since the validation of the first one would extend to these too. We don't wait for other pollsters to have every one of their polls published in secondary sources to include them, as long as they've been validated before and the data is credited as genuine. Impru20 (talk) 15:43, 21 April 2017 (UTC)

The first one was included in fr.wp after it was cited in a politics TV show (by a commentator, not journalists btw). Every other poll is ordered by a media, so they fulfill the secondary source need. --Marc Lacoste (talk) 15:48, 21 April 2017 (UTC)

Marc Lacoste The first one was regarded as valid. So, there's no reason not to include the other one if it's proven as a genuine poll from the same pollster (the fact that some polls may not be ordered by a media do not make them non-existant, and practice in Wikipedia is to include them anyway unless the pollster is proven to be a fake one). This said, the first one was included by FT in its own right before being included here, and the second one is included there too, so that'd count as a secondary source. And given that FT included the first one before it was included in the en.wikipedia, you can't guess out whether they included the second one only as a result of it being included here or, most probably, just included it anyway given that they had already included the other one earlier. So this is valid enough, I think, being reported by both primary and secondary sources, and not WP:CIRCULAR. Impru20 (talk) 16:01, 21 April 2017 (UTC)

The first one was validated on fr.wp on 3 April 2017, so it could have been brought to the FT attention then, and I hope you're right they kept the address. So we have higher standards than the FT :) --Marc Lacoste (talk) 16:21, 21 April 2017 (UTC)

For clarification, the poll was first included in the FT average after being added to the French Wikipedia article. If you look at how the rolling polls are listed on the FT average, it's obvious that they're using the French Wikipedia article. Mélencron (talk) 19:02, 21 April 2017 (UTC)

Polls in Belgian/Swiss media conducted after polling ban

I saw this discussion circulating on the talk page of the corresponding French article and was wondering about it here as well. Wikipedia is formally under the jurisdiction of American law, but for each language Wikipedia the dissemination of material is generally covered by the corresponding countries for each of these Wikipedias. In the case of the French Wikipedia, that implies it's subject to French law, and therefore the polling ban in France. However, I don't believe this is the case on the English Wikipedia, and therefore I don't believe there is any such restriction. I already know from following User:Impru20's articles on Spanish politics in the past that such a polling ban already exists in Spain in the last week, and yet polls are still published in Andorra and included on the English Wikipedia. I hope that isn't also a problem here? The Tribune de Genève, for instance, will be publishing a poll of French voters in tomorrow's edition. (Side note: regardless of early leaks of results/estimations by foreign press, I think we should respect the usual embargo.) Mélencron (talk) 19:07, 21 April 2017 (UTC)

Mélencron Spanish regulation bans the "publication, dissemination or reproduction of opinion polls by any 'mass media'". However, this must be put on context, given that this has remained unchanged since 1985 when social networks nor Internet existed (and it doesn't prevent particulars from publishing them). Indeed, as of currently, the polling ban is virtually ineffective, as polls are published outside Spain and throughout Twitter and the Internet freely without any consequences. I think the same would apply in France, though I'm not sure of what's the actual wording of such a ban in French legislation. Impru20 (talk) 19:23, 21 April 2017 (UTC)

My gut instinct is to publish it, but place it within a collapsed and clearly headed box for people to expand if they so wish. Maswimelleu (talk) 20:36, 21 April 2017 (UTC)

My personal suggestion: black bar to separate polls conducted after the legal ban on polling in France, light-red background for post-ban polls (same as those on Impru's polling articles), plus a textual note above the table. Mélencron (talk) 20:47, 21 April 2017 (UTC)

I would place a textual note within the bar so it's clear why the division is there at a glance. Not everyone will be reading from the top down. Maswimelleu (talk) 20:54, 21 April 2017 (UTC)

@Impru20: FYI, re the poll referred to in the Tribune de Genève I added earlier, the name of the institute seems not to be disclosed for an obvious reason (necessarily, it's an institute operating in France and therefore could face charges under the polling ban law), and yes, it's a unique poll, as has been confirmed by other sources (Politico Europe confirmed the day before that they would be releasing a poll on Saturday). In addition, there's also been a poll published in the Belgian press as well; this one specifically for a campaign, apparently. Mélencron (talk) 22:05, 22 April 2017 (UTC)

It's not that TdG doesn't disclose the institute name, but that the wording it provides implies that it may not be just a poll but a poll average or some sort of amalgamation of last-hour polls (in fact, the very first paragraph here implies that it could be a poll average, but not a single poll, given that it uses the plural form all through the article). I don't know if the TdG intended to be ambiguous so as to prevent breaching French law, but it's actually so ambiguous and provides so little data that we can't say that what it publishes is an actual poll, an average of several of them made by several institutes. And what Politico Europe says it's not enough, since it should be TdG the one that specifies that without having to resort to others' statements assuring that they may publish a poll. This is obviously so off from what we discussed for legal campaign polls...

On the other hand, the one from lalibre.be could be included (with all the due cautions, nonetheless), since the source itself specifies that it's a single stand-alone poll. Impru20 (talk) 22:15, 22 April 2017 (UTC)

Here's the specific item in Politico Europe that explicitly refers to the item published in Swiss press as a "poll". However, I also found that Le Temps referred to the numbers as those of "pollsters", quoting the article as you did. Mélencron (talk) 22:33, 22 April 2017 (UTC)

Politico Europe only says this: "La Tribune de Genève, a Swiss daily, has told POLITICO it hopes to publish the last poll of the campaign on Saturday". This is, a report by Politico Europe that TdG "hoped" that it could (or, alternatively, couldn't) publish a poll on Saturday. Yet there's no indication that it has done so, that the numbers it published do actually constitute the poll it hoped to publish or are, instead, a random poll average that it made in place of such a poll. If this is an error by TdG or is intentional we can't know, but it's ambiguous enough (and sources, as you point, are contradictory) that we are unable to confirm its validity. Impru20 (talk) 22:38, 22 April 2017 (UTC)

La Libre describes the item published by the TdG as a poll as well; my guess is merely that their description of it as an estimate by pollsters is merely an obfuscation by them to attempt to avoid falling foul of the law. Unless they published another piece on Saturday which I didn't see... Mélencron (talk) 22:43, 22 April 2017 (UTC)

Revised estimations

I don't think it makes sense to list every single instance of the estimations within the tables; not only would it be several per pollster, in my view it's like creating a results table for every point at which the percentage of votes for candidates is updated – at 10% counted, 25% counted, 30% counted, etc. – since that's essentially what these are based on in any case. We don't use results tables listing partial results; only the most recent results (while counting is ongoing) or final results (when published). I don't see what's particularly different in this case. Mélencron (talk) 19:20, 23 April 2017 (UTC)

But why it doesn't make sense? They were published. And then they were updated, yeah, but the previous ones were still published. Why should we make as if the previous ones did not exist? Candidates have been conceding the election based on the previous estimations, so they of course are important. Impru20 (talk) 19:24, 23 April 2017 (UTC)

I'm sure the campaigns have data of their own, given the behavior of Macron and Fillon's supporters even before the estimations were announced. In any case, they're not discrete estimates at all... they're updated based on the same sample of polling stations. They existed, yes, but that's a bit beside the point if they're not discrete. Mélencron (talk) 19:27, 23 April 2017 (UTC)

Well, if they were uncomplete or where irrelevant then the main media of the country wouldn't have published them. Surely, the updated ones are more complete and/or accurate, but that doesn't mean the previous ones were not complete on their own. They just revised their estimations after gathering more data, but that doesn't make the previous ones as invalid at the time they were published. I don't know if campaigns have data of their own (that would be a guess, not a fact, I think) but I do know that candidates announced their defeats right after the first estimations came out. And media have given much importance to such estimations until they revised them. I don't think they're somehow invalid given this. Impru20 (talk) 19:31, 23 April 2017 (UTC)

I see no point in publishing here estimations that are only valid during a few minutes. But if you choose to do so, you should rename this article because these estimations are not opinion polls. Seudo (talk) 20:55, 23 April 2017 (UTC)

I'll defer to the outcome of the discussion on the corresponding article on fr.wikipedia. Mélencron (talk) 21:09, 23 April 2017 (UTC)

Well, in that case that would apply to ALL estimations, not just to the unrevised ones, given that these are all estimations. Impru20 (talk) 21:27, 23 April 2017 (UTC)

Swing reported in the Odoxa poll published 28 April

The Odoxa poll published today, 28 April, reports figures of 59%-41% for Macron-Le Pen, from data collected on 26-27 April, which it shows as representing a 4% swing from the figures of 63%-37%, from data collected on 24-25 April.

But no Odoxa poll is listed here for data gathered on 24-25 April. Did they run such a poll? If they did, where is it and why isn't it referenced here? If they didn't, what are they referring to? Greengreen2 (talk) 14:02, 28 April 2017 (UTC)

Added. Mélencron (talk) 14:05, 28 April 2017 (UTC)

Trends

I now fully understand that the graphics use the latest polls (in the last 14 days) for each of the main polling organisations.

I wanted to ask if it would be meaningful to report on within-polling-organisation trends since last poll maybe it could be shown somehow on the table or reflected in some other way? (i.e. as an hypothetical show Oxona: -1% for Macron +1% for Le Pen since last polling; IfopFiducial: -1.5% for Macron +1.5% for Le Pen etc). That would give an idea of whether there are consistent within-poll trends? Probably too late to be useful for the 2017 presidential elections but an idea to bring up for the future?

Generally, can I congratulate everyone who has worked on this site throughout. They have done a great job of providing facts and data for a wide range of communities. B A Thuriaux (talk) 09:10, 2 May 2017 (UTC)

Crucial metadata

Each poll should be marked with whether it counts Registered Voters or Likely Voters. — Preceding unsigned comment added by 2600:8801:0:1530:B916:24A3:6162:91D0 (talk) 06:19, 3 May 2017 (UTC)

Question on "original research" policy

Just curious, how do the many original elements on this page – which, do not get me get me wrong are great – escape the "no original research" policy enforced by all versions of Wikipedia? For instance, the summary graph is original content, with an arbitrary (which does not necessarily mean wrong) choice about the amount of smoothing applied to the summary curves.

Just asking. --Phnk (talk) 16:43, 5 May 2017 (UTC)

I saw that this concern came up on the French Wikipedia as well. I don't really have an answer to that – it's original data, yes, but the curves are also based on the polls listed on this article, and in any case, almost every opinion polling article on the English Wikipedia has had a polling graph as long as I can remember. Mélencron (talk) 16:53, 5 May 2017 (UTC)

The discussion over there is slightly different but related indeed. I'm not really concerned about (good) original content being produced by Wikipedians, here or elsewhere, so this is not really an objection. But I do gather that, in this case, the rule is being slightly bent (which, as far as I understand how WP rules are applied, is fine too). --Phnk (talk) 17:14, 5 May 2017 (UTC)

Not bent; more like liberally interpreted. Wiki-legally speaking, the graph can be said to be permissible per WP:CALC and WP:OI - WP:CALC allows the calculation of a moving average and WP:OI allows its presentation the graph. Also WP:IAR applies. Rami R 18:16, 5 May 2017 (UTC)

Thanks for the precisions, all very helpful. --Phnk (talk) 04:52, 6 May 2017 (UTC)