Talk:Opinion polling for the 2019 Canadian federal election/Archive 1

Archive 1

Archive 2

Requested move 16 January 2016

The following is a closed discussion of a requested move. Please do not modify it. Subsequent comments should be made in a new section on the talk page. Editors desiring to contest the closing decision should consider a move review. No further edits should be made to this section.

The result of the move request was: redundant. Whatever the consensus is at Talk:43rd Canadian federal election, this article will follow that. Jenks24 (talk) 05:16, 24 January 2016 (UTC)

---

Opinion polling in the 43rd Canadian federal election → Opinion polling for the next Canadian federal election – WP:COMMONNAME - most sources will refer to the "next election" rather than the 43rd. 43rd is unnecessarily confusing and unrecognisable for average readers. It is not precise nor concise. It is also uncommon, certainly not a "convention" see here AusLondonder (talk) 07:56, 16 January 2016 (UTC)

• NOTE a related move request is open at Talk:43rd Canadian federal election -- 70.51.44.60 (talk) 11:07, 16 January 2016 (UTC)

Survey

Oppose: What is needed is consistency with past practice and current style. Currently we have the article 43rd Canadian federal election. Most new or casual readers accessing Opinion polling in the 43rd Canadian federal election will do so via the parent page. Therefore we should have a consistency in terminology. In the past we have described future Canadian (federal and provincial) elections by their number until such a point comes when we can describe them by their year. Currently 8 of the 9 future provincial elections are described by their number. There has been much discussion in the past about this subject and, I think correctly, that a consensus has been maintained around the number description. Graemp (talk) 08:22, 16 January 2016 (UTC)

• Oppose "next" is a highly fluid concept. I find no reason to constantly flip-flop names around. Just retarget the "next" to the next numerical election the day after an election is held, instead of moving articles around willy-nilly. The redirect exists, so readers will end up at the right article. -- 70.51.44.60 (talk) 11:03, 16 January 2016 (UTC)
• Oppose per above. Ribbet32 (talk) 17:42, 16 January 2016 (UTC)
• Oppose IP editor has expressed this adequately. #redirect is fine Krazytea(talk) 23:48, 16 January 2016 (UTC)
• Oppose per above and as related article does not appear to be headed towards a renaming. Alexander's Hood (talk)
  • Although according to the nominator we are now "the worst of Wikipedia." Move aside, vandals! Ribbet32 (talk) 13:41, 18 January 2016 (UTC)

---

The above discussion is preserved as an archive of a requested move. Please do not modify it. Subsequent comments should be made in a new section on this talk page or in a move review. No further edits should be made to this section.

Campaign research poll

Do we want to include the Campaign research poll here? Two of the main principles are Conservative pollsters (one being Nick Kouvalis). Do we want to include polls run by Conservative campaign managers? Mikemikem (talk) 20:08, 9 February 2017 (UTC)

I think it's fairly common for most pollsters to work for political parties or policitians here and there, mostly to conduct so-called "internal polls" that are not publicly released. This particular poll was not commissioned by a political party though, and just because the pollster has tended to work for conservative campaigns isn't sufficient grounds to assume that their results are biased or untrustworthy. They aren't that far off from other recent polls which have all shown a tightening race. I would advocate including it. If it turns out to be an outlier, so be it; I think the aim of the poll table should generally be to be complete and inclusive. Cheers, Undermedia (talk) 20:28, 9 February 2017 (UTC)

As a non-Conservative myself, I think it is great to include it. People have similar thoughts and feelings about Abacus but I think what they add to the polling pool is great. It's different than some kind of a straight trolling poll, as long as they have some legitimate polling methods, the more the merrier in my opinion. Krazytea(talk) 20:36, 9 February 2017 (UTC)

Alright that sounds good, I'll throw in the poll Mikemikem (talk) 23:38, 9 February 2017 (UTC)

Campaign research poll (again)

The website is down. Do we want to change the links to the HTML article in hill times? Or keep the PDF link from C.R which may come back but may not? Site has been down for 5 days Mikemikem (talk) 17:53, 9 April 2017 (UTC)

Maybe link directly to [1] and [2]? -- Asclepias (talk) 01:15, 10 April 2017 (UTC)

Yes that would be much better, I didn't know where to find the links. Mikemikem (talk) 02:39, 10 April 2017 (UTC)

"Lead" in opinion polling

When I click "lead" on the opinion polls to rank the polls by largest lead to smallest, it orders "9" before "23" because "9" comes before "2." Only the first digit is being used to rank the order.

Is there a way to change this? Mikemikem (talk) 19:38, 29 April 2017 (UTC)

Yes, use "data-sort-value" to explicitly define the values in the column for the purposes of sorting, or use {{sort}}. Mélencron (talk) 22:07, 29 April 2017 (UTC)

I'm not wiki savvy enough to know how to do it.

Do you mind to fix it at some point if it's a quick fix? Mikemikem (talk) 22:17, 30 April 2017 (UTC)

I changed the "tie" text to 0, so that the table knwos the column is numbers and automatically sorts properly. galneweinhaw (talk) 05:49, 3 May 2017 (UTC)

New leaders

MHdy all,

Just curious if there has been a discussion to add an "event" type bar between the polls when a new leader is elected (similar to UK opinion poll pages, although they do it for other major events as well).

Bernier will most likely be the new CPC leader on May 27. Does that warrant a break in the page saying "May 27: Bernier is elected leader of the Conservative Party of Canada" (similar to how we do for the leadership polls)?

If this has already been discussed on rejected, my apologies.

Mikemikem (talk) 21:35, 10 May 2017 (UTC)mikemikem Mikemikem (talk) 21:35, 10 May 2017 (UTC)

Selecting various "events" to highlight throughout polling chronologies is pretty subjective, though changes in party leadership are one thing that usually do cause voting intentions to shift. If anything though, I might suggest that it would be more effective to highlight them in the graph(s) (like this, for example) than in the table, plus I'm concerned that altering the table by splitting it up or adding rows with inconsistent content/formatting will break the new graph that reads the polling data directly from the table. Cheers, Undermedia (talk) 11:55, 11 May 2017 (UTC)

Leaders are probably the only thing I would highlight in the graph, I like the changes Undermedia recommends. I wonder if we highlight the PQ leadership changes though? I guess we have to if we do for the others? Krazytea(talk) 15:44, 11 May 2017 (UTC)

https://en.wikipedia.org/wiki/Opinion_polling_for_the_United_Kingdom_general_election,_2017

The graph from the UK page doesn't seem messed up, and they have a ton of events added (new leaders, every single by election, Brexit vote), I could 17, not counting starting a new graph for each year (it looks like we are one of the only pages that does not have breaks for a new year as well, which is fine as not really necessary).

How come it would mess up our graphical summary? Is it coded differently?

I think including every by election is quite silly, and we could even restrict it to parties with official party status (12 seats) if we felt changing Greens and BQ was too much.

But new leaders of the official opposition and a third party with official party status are definitely major events, I would think worthy of a break.

I agree the only way to avoid subjectivity of "major event" is to restrict it solely to party leaders (and maybe even those with official party status). Mikemikem (talk) 16:15, 11 May 2017 (UTC)

Agreed that we could probably objectively limit it to leader changes for parties with official status, i.e. LPC, CPC and NDP. I just think that marking leader changes directly in the graph (as in the example I shared above) will be more effective for visualizing any impact they may have on voting intentions than burying them in the table. And yes, putting them in the table could very likely mess up the graph because its code expects the table to be laid out in a very specific way. The code could likely be updated to accommodate it, but frankly I'm starting to feel bad about making galneweinhaw work so incessantly to refine the code. I've figured out the basics of it, but I wouldn't know how to make this sort of change, for example. Cheers, Undermedia (talk) 16:37, 11 May 2017 (UTC)

Yeah if galneweinhaw has been working hard to refine the code, not fair to expect him to do more.

How about this? If galneweinhaw confirms that the break could be added to the chart, and it would not affect the graph in any way (with the code exactly as it is now), then we consider adding it (or maybe informal vote type of thing)?

But if adding a break (which would be solely for parties with official party status) would mess up the graph (with code as is), then we don't add it. That way galneweinhaw is not required to do any additional work.

And if it does not work out, then potentially a line directly on the graph?

Mikemikem (talk) 16:46, 11 May 2017 (UTC)Mikemikem Mikemikem (talk) 16:46, 11 May 2017 (UTC)

Current code will ignore blank rows, but I'm not sure what will happen if there is text in one cell of the row, or if the row is split into a different number of columns. However, if we figured out a consistent way to add these annotations into the table, I might be able to add the annotations to the plot! I enjoy doing this and contributing, so I don't mind at all. galneweinhaw (talk) 20:57, 11 May 2017 (UTC)

Well if it is not too much trouble for you (again, if you've been working a lot on the code, no need to contribute any more than wanted!), and it is indeed possible to do, I would be in favour of the addition.

I think we can keep it consistent by only allowing the inclusion for party leaders, and setting the bar as either 1) parties with a seat in the HoC, or 2) parties with official party status.

That way there could be no complaints that some were added and some were not.

I think Party leadership is one of those things that really change the ballot question. Polling has showed only 5-10% of Canadians vote primarily for their local candidate, whereas the large majority vote for party/leader, so leadership changes for major parties are monumental shifts, uncomparable to any other events (like by elections, etc.) Mikemikem (talk) 21:49, 11 May 2017 (UTC)

Also, if it was an issue of continuity, I would have no problem (if it were just a matter of copying and pasting the individual line, as it's the only way I'd know how) to go back to the past 3 or 4 federal opinion poll pages and add those in.

Justin Trudeau became leader on April 14, 2013 and there was a clear difference in the polling numbers after.

I have a feeling we may see a similar shift after Bernier

Mikemikem (talk) 04:24, 12 May 2017 (UTC)Mikemikem Mikemikem (talk) 04:24, 12 May 2017 (UTC)

So I copied and posted this: |18 Apr |colspan="10" |Prime Minister Theresa May announces her intention to seek a general election to be held on 8 June 2017 |-

From the UK graph in to the Canadian chart, and it had no impact on the chart. So unless I missed something, it looks like this could be done without issue.

Mikemikem (talk) 20:56, 12 May 2017 (UTC)Mikemikem

Ouellet added. Let me know what you think.

Graph is still all good Mikemikem (talk) 04:16, 13 May 2017 (UTC)

Actually, the graph had not been updated yet since the table was changed, but I just ran the code with current table and it still works. This graph is unique in that it reads the data directly from the table on webpage; most other graphs are compiled independently (e.g. in an Excel spreadsheet, which is how I used to do them), so they wouldn't be affected by a change to table formatting. I won't put the new version of the graph up though till we incorporate the new formula to control the sensitivity of the trendines. Stay tuned. Undermedia (talk) 00:12, 14 May 2017 (UTC)

Ok, one of the graph stuff is lost on me, but you are knowledgeable of it so whenever you have the time to make whatever updates are needed, that is all much appreciated.

So if we keep the chart information solely to party leaders being elected (no by elections, no interim leaders, no major broken promises [I saw another discussion where major breaks, such as electoral reform, were requested put in], no referendums, etc.), would you be OK with that?

Throughout 4 years the only changes will be Ouellet, Bernier, and new NDP leader, so chart won't be clogged at all.

And seeing as leadership elections are unambiguously major events beyond all else, the "floodgates" won't be opened. Mikemikem (talk) 00:43, 14 May 2017 (UTC)

I don't know why you keep mentioning Bernier. Clearly Andrew Saxton is going to take the leadership. ;) Undermedia (talk) 01:14, 14 May 2017 (UTC)

Deepak 2019!!!!

In all serious, I think it would be a massive upset if Bernier didn't win. We'll find out in two weeks, and I suppose it shouldn't be taken for granted, but money and polling points to "Mad Max." Mikemikem (talk) 01:33, 14 May 2017 (UTC)

Plot

I had a request to provide a plot similar to those created for the past few elections. It took me a while to parse the rolling poll info "1000 (1/4)". Obviously the x-axis isn't complete. Please let me know if there is interest in using this, if so I will finish it up! --galneweinhaw (talk) 07:25, 30 April 2017 (UTC)

Looks great, galneweinhaw. Thanks once again for putting in the effort to figure out the code; the automatic sample size adjustment is a great way to deal with the "rolling" polls. Personally, I think these continuous local regression graphs are the most effective and aesthetically pleasing for visualizing polling trends, and would be in favour of implementing this one for the 43rd Canadian election opinion polling page. I would even be happy to update it as new polls get added to the table if your time is limited, however for some reason I was never able to get the R-Fiddle tool to run properly on my end for the 2015 election campaign polling graph (I seem to be able to 'Run' the code, but then when I press 'Graphs', the window that pops up is simply blank), so you might have to help me figure out what I'm doing wrong... In any case, I'm in favour of using it. Cheers, Undermedia (talk) 14:56, 30 April 2017 (UTC)

Dear editors: Thanks to the hard work of galneweinhaw, we propose this alternative graph of polling trends similar to the ones used to illustrate campaign-period polling for the 2015 and 2011 federal elections. It is generated by an R code that reads all the necessary data directly from the polling table on main article page. The trend lines represent local regressions, sort of like a continuous moving average, as opposed to an average over a fixed period of time or of a fixed number of most recent polls. The ribbons on either side of the trend lines represent 95% confidence intervals. In addition to weighting polls by date in the ongoing calculation of the trend lines, polls are also weighted by margin of error, which is calculated from the sample size. This is especially handy for preventing disproportionate influence on the trend lines of the weekly Nanos rolling polls, for which each one only contains 250 new respondents compared the previous week's poll. Thus, the code for the graph reads the sample size statement of "1,000 (1/4)" given in the table for each of these Nanos polls, and automatically assigns them a weight equivalent to that of a poll of 250 respondents rather than 1,000. Note that the weighting is ultimately based on the margin of error, which has a logarithmic relationship to sample size, so it therefore doesn't overly reward very-large sample size polls compared to smaller or average sample size polls. Feedback is welcome. Cheers, Undermedia (talk) 16:50, 3 May 2017 (UTC)

I think that using this method the nanos polls are being discounted too much. It would be better if there were a way to just include every 4th nanos poll with a full sample size and MOE, because ultimately now it looks like it has a very diminished weight, and we end up with a graph that shows the top 2 parties 5% apart, which is not a fair representation of the polls. Of the 3 pollsters in the field this month, Campaign research shows a 9% spread, Nanos a 13.4% spread, and Forum a tie, but because nanos is so severely discounted, the graph comes out to a 5% spread as if end of April, which is not really a representation of what these polls are showing. If the nanos poll were given full weight, every 4 weeks, the last portion of the graph would show a spread of 6.7% (splitting the difference between the forum tie and 13.4% nanos gap), which would be more representative of the polling.

But my heart isn't set on it, and either way, as long as there is a graph I'm fine with it. Mikemikem (talk) 17:49, 3 May 2017 (UTC)

Thanks for the feedback, Mikemikem. Since the weighting of polls is ultimately by margin of error, I believe 4 polls each with an adjusted sample size of 250 and resulting margin of error of 1/sqrt(250) = 6.32 pp will technically have more weight in total than a single poll with a sample size of 1000 and resulting margin of error of 1/sqrt(1000) = 3.16 pp. However, since the 4 small polls are spread over 4 weeks, when factoring in the time decay, I suspect it might actually work out to pretty much the same total amount of influence on the trend line for any given 4-week period as having a single large poll once every 4 weeks; galneweinhaw might be able to confirm. Also note that the most recent Nanos poll (13.4 point lead) has not yet been included in the graph, so it will probably widen the LPC-CPC spread a bit once it is. The other thing to keep in mind about the Nanos polls is that each one contains unusually old data—going back 4 weeks!—compared to ordinary polls which are typically conducted over the course of just a few days, so it is even debatable how “current” the numbers in any given Nanos poll truly are in relation to the last date of polling; they will inherently lag behind ordinary polls in terms of revealing shifting trends in public opinion. To be sure, there’s probably no perfect way to manage all these variables, but in the past this sort of graph has seemed to provide the overall best formula that I’ve seen. Cheers, Undermedia (talk) 22:04, 3 May 2017 (UTC)

Oh I was unaware that the most recent Nanos poll had not been included. I am not a statician so admittedly this is not my strong point.

However, intuitively (as a law student, more my strong point), when we look at the month of April we have polling from 3 pollsters. One pollster left the field on April 11, showing a Liberal lead of 9%. One pollster left the field on April 24, showing a tie. And a third pollster left the field on April 28, showing a Liberal lead of 13.4%.

Even with a lot of the Nanos data being older, a 5% lead as an "average" of this information does not pass the smell test.

Again though, I have no idea how quickly older information is discounted (it appears the 9% lead in the Campaign Research poll that predates the Forum poll by only 13 days must be almost entirely discounted, along with most the first 3 weeks of the Nanos poll showing a 13.4% lead, for the one Forum poll showing a tie to weigh the polling average down to a 5% spread).

I feel, on its face, the numbers from these 3 pollsters for the month of April (9.0%, tie, 13.4%) do not suggest a horserace separated by only 5 points.

Again though, not a statician, and if people more familiar with this area believe the formula being used is appropriate, and a 5 point spread is an accurate representation of the polling (9%, tie, 13.4%), I would defer to them. Mikemikem (talk) 00:44, 4 May 2017 (UTC)

And again I completely missed that the most recent Nanos poll is not in the graph, this is my mistake.

Everything I'm saying could be irrelevant depending on how the addition of that poll impacts the graph. Mikemikem (talk) 00:47, 4 May 2017 (UTC)

Good day, Mikemikem. After some discussion with galneweinhaw about the underlying code and statistical formulas involved in generating the graph, I think we may have come up with a solution to address your concern about what can be described in simple terms as the trendlines' "oversensitivity" to a recent poll showing a tie for the lead when several other recent polls showed a significant gap. Simply put, the trendlines' "alpha" value, currently set to 0.35 as stated in the graph's caption, represents the proportion of the total number of polls in the table that is factored in when calculating the y-axis position of the trendlines at any given point along the x axis. So say there are currently 87 polls in the table, the current positions of the trendlines as of the most recent poll are taking into account the 87 x 0.35 ≈ 30 most recent polls. Of those 30 polls though, the most recent ones are much more heavily weighted than the most outdated ones, as is the nature of the local regression. In addition, polls with smaller margins of error (e.g. the Forum one) are more heavily weighted than polls with larger margins of error (e.g. the Nanos ones). With the addition of each new poll, it isn't only the leading edges of the trendlines that are updated, but indeed the entire span of the trendlines, with any given point along them recalculated based on the nearest polls on either side. This means that if several more polls were to come out that are at odds with the Forum poll and show a significant lead for the LPC, the conspicuous convergence of the LPC and CPC trendlines that we currently see would eventually smooth out, perhaps even disappear almost completely, and we may then be able to statistically conclude that the Forum poll was an outlier. That being said, I agree with you that the leading edges of the trendlines are perhaps currently a bit too sensitive to the very most recent polls. This can be simply corrected by increasing the alpha value so that a larger proportion of the total number of polls in the table is factored into the calculation of the trendlines, thus "diluting" the weight of the most recent polls. However, if the alpha value remains a constant proportion (as it currently is at 0.35), then as more and more polls are added between now and next election, an ever larger number of polls will be used to calculate each point along the trendlines, meaning that the trendlines will become less sensitive to local fluctuations in polling numbers over time. It may instead be most logical to maintain a constant sensitivity of the trendlines over time. This could be achieved by setting alpha to a fixed number of polls rather than a fixed proportion of total polls in the table. So right now the proportion 0.35 is causing the nearest 30 polls to be used to calculate points along the trendlines, but once the number of polls in the table has doubled, that number will have risen to 60! Instead, we could set alpha to a fixed number—say 35 polls to make the trendlines a bit less sensitive than they currently are—and then that would always be the number of polls used to calculate local points along the trendlines regardless of how many total polls are in the table. Does that make sense? Cheers, Undermedia (talk) 14:01, 9 May 2017 (UTC)

My thoughts on it: I don't think the Forum poll is overweighted, and I don't think Nanos is underweighted. They are weighted according to their sample size (with a time decay), so unless we have some reason to count some pollsters more than others, I don't see a problem with it. I think this is more meaningful than just counting all polls the same regardless of sample size/margin of error. I also don't think the end points are an issue. The large increase in the confidence bands (the trumpet shaped endings) reflect less confidence in the trendline at the ends. The two graphs use different methods for estimating a trendline, I wouldn't expect them to give the results. galneweinhaw (talk) 20:21, 9 May 2017 (UTC)

I agree with all of the above. As I previously stated, I think the sample size/margin or error weighting scheme used in this plot is the best way to deal with the Nanos rolling polls. Clearly, treating them as independent 1,000-respondent polls coming out on a weekly basis would be giving them undue weight since each new poll is comprised of 75% of the same data as the previous poll, 50% of the same data as 2 polls back, and 25% of the same data as 3 polls back. However, the value of alpha, or the formula used to calculate its value, is ultimately arbitrary, so it's legitimate to have a discussion about what we should set it to. Personally, I think it makes more sense to use a formula based on a fixed number of polls rather than a constant proportion of the total number of polls in the table, so that the trendlines maintain a consistent "sensitivity" over time. With the current alpha value and total number of polls in the table, about 30 polls are being used to calculate each local regression point along the trendlines. If some feel that's making the trendlines look too "sensitive", I would be open to using a formula that would fix the number of polls to something slightly higher, say 35. I don't feel too strongly about the number either way. But the underlying formulas that generate the trendlines are beyond reproach; in fact, this plot is a most elegant and robust implementation of statistical principles! Cheers, Undermedia (talk) 20:58, 9 May 2017 (UTC)

Popping in with an unrelated query – is the source code used to generate this graph publicly available? (In addition, would it be possible to adapt it to read comma-separated values instead?) Mélencron (talk) 20:47, 9 May 2017 (UTC)

Yes, it's here, using the R statistical programming language, but I'll put it on github so it's more public (and stays up to date), and CC licensed, then we can link to the code in the image's description. And yes, could easily be adapted to read a csv file, and I'd be happy to help you out with it if you need help. galneweinhaw (talk) 21:10, 9 May 2017 (UTC)

Yeah that all sounds good. Thanks for explaining! Mikemikem (talk) 23:48, 9 May 2017 (UTC)

95% confidence ribbons?

Thank you for the graph, it truly is informative and aesthetically pleasing. However, I do have a request to make: remove the confidence ribbons. I find these ribbons potentially misleading: they do not represent 95% confidence interval of the values of the real-world election, but rather the confidence in the "true" polling average. The typical reader is not going to understand this point, and will assume that the true election results reside within the interval, despite plausible systemic bias.Rami R 12:39, 10 June 2017 (UTC)

Hmmm. I understand your point, but am reluctant to endorse the change, since the graph is based on rigorous statistical theory (the "local regression"), and any rigorous expression of statistical results is accompanied by an expression of the uncertainty associated with those results. Call me biased (I've personally done a lot of stats in the context of ecology), but I find the general public tends to lack proper understanding that statistical results are inherently uncertain/approximations—and consequently unfairly declares pollsters, for example, useless and incompetent when their predictions are a few points off (admittedly there have been larger "misses" than that, but not very frequently in the grand scheme of things), even though polls clearly state that they are within a certain margin of error, 19 times out of 20 (meaning that it's statistically expected that 5% of polls on average will be major outliers)—and if anything needs to be reminded at every opportunity about the concepts of statistical error, uncertainty, confidence, etc. Just my two cents, anyway. What if we more clearly explained the meaning of the confidence ribbons and emphasized your points in the caption, and perhaps displayed the graph in "thumbnail" mode so that the caption appears right underneath it on the page? Undermedia (talk) 14:20, 10 June 2017 (UTC)

I can accept the addition of a caption (perhaps even added to the legend?). As an aside - polls' margins of error do not inform about systemic bias, and as such are not really applicable here. Rami R 15:27, 10 June 2017 (UTC)

"Lead" in opinion polling (again)

When I try to sort the polls by "lead" it is messing up again. Putting "3" before "20" because of the first digit.

It did this before and was fixed. Does anyone know how to fix it again?

I'm hoping the addition of "Scheer" did not mess it up, but it was working fine with "Oullette" included before. Mikemikem (talk) 05:14, 2 June 2017 (UTC)

The problem before was that one of the column entries was text instead of numeric (used to say "tie" instead of 0). When text is detected, it sorts the column alphabetically (11 < 3) rather than numerically (3 < 11) . It's now doing that again, but a quick glance and I can't see which entry is causing it to revert to alphabetical sorting. galneweinhaw (talk) 18:49, 11 June 2017 (UTC)

Ugh that is unfortunate. Thanks for letting me know the problem though. Its unfortunate we can not fix it Mikemikem (talk) 19:09, 13 June 2017 (UTC)

So none of them say "tie" it is all zero. Could it be the addition of "leader elected" having words across the whole row would make the difference? Mikemikem (talk) 19:13, 13 June 2017 (UTC)

Yes, very likely. The text messes not only the lead column but all the columns with numbers. One can tell by looking at the NDP column, which is the other column with both double-digit entries and at least one single-digit entry (there's a "9" there somewhere). The problem is not immediately apparent in the other columns because at present each of them has only one type of entries (either double-digit or single-digit), but they would be affected also. -- Asclepias (talk) 20:05, 13 June 2017 (UTC)

See my new "talk" created just after this. I tested it out and it is correct, so we need to decide what to do Mikemikem (talk) 21:07, 13 June 2017 (UTC)

You don't need to create new versions to test. You can use the old versions from the history. For example, take the version from 22:04, 12 May 2017 (before text was added), sort the polls by the column "NDP", and see that the "9" falls in the right place. Then take the version from 05:00, 13 May 2017 (after the addition of one line of text), sort by the column "NDP" and note that the "9" still falls in the right place. Then take the version from 12:52, 1 June 2017 (after the addition of a second line of text), sort by the column "NDP" and note that the "9" ends up in the wrong place. -- Asclepias (talk) 22:12, 13 June 2017 (UTC)

Ok, so could you comment your opinion in the next talk page so we can get a base opinion of what to do. Because it seems we need to decide to either get rid of the leader elected rows, or have a messed up "lead" column Mikemikem (talk) 00:12, 14 June 2017 (UTC)

"lead in opinion polling" and "party elected leader"

So it looks like the 2 lines "oulette elected leader" and "scheer elected leader" are throwing off the "lead" column to the far right of the opinion polling column. It orders by the first digit, so "11" goes ahead of "2" and "21" goes ahead of "3"

Is there any way to fix this ordering while keeping the election of the leaders in the chart?

And if not, what do we feel is more important?

I quite enjoy having the election of leaders in the chart because party leaders play such a large role in our system, so a new leader is a drastic change unparalleled by other events. However, being able to order the polls by "lead" seems important too.

If these two can not be reconciled what do we want to do?

And if anyone does know how to format it so that the leader elections can remain in the graph AND the "lead" ordering remains correct, that would be HUGELY appreciated. Although I realize this may not be possible. Mikemikem (talk) 21:02, 13 June 2017 (UTC)

Unfortunately I don't know how to fix the table's formatting so that numerical sorting would work properly with the new party leader rows. If it turns out to be impossible and we value the ability to properly sort the columns, I suppose one option would be to reconsider the past suggestion to show the new leaders in the graph (in the form of vertical lines, perhaps matching the graph's party colours, and labelled with the leaders' names) instead of in the table... if that, in turn, is even possible (we'd have to enlist galneweinhaw again to wrestle with the code some more). Cheers, Undermedia (talk) 02:23, 14 June 2017 (UTC)

Hmmm ok. Well, I was the one who pushed to have "X is elected leader of X party" in the table, but since it messes up the chart I'm fine either way if people feel it should be removed to fix the chart (or keep it, whichever people want).

What is your preference? Mikemikem (talk) 02:48, 14 June 2017 (UTC)

What I find most strange is that the sorting continued to work after the addition of Ouellet and only broke when Scheer was added. Yet the formatting for both of those rows is identical as far as I can tell. There must be some explanation for why the one had an effect but not the other, and hopefully some way to fix it, but I'm honestly at a loss. Undermedia (talk) 15:36, 14 June 2017 (UTC)

Could it be due to "invisible unicode characters"? You may want to replace the offending text with text cleaned in a text editor and try again. - Ahunt (talk) 15:43, 14 June 2017 (UTC)

There's a logic to it, when we think about it. A single line of text does not need to be sorted as text in relation to other lines of text, which do not exist. It can just get placed at the top or at the bottom of the table, and the rest of the table can still be sorted by numbers. But two or more lines of text must be sorted as text in relation to each other. Since sorting by text takes priority over sorting by numbers, this priority applies when sorting by text becomes necessary, i.e. when there are two or more lines of text. -- Asclepias (talk) 16:52, 14 June 2017 (UTC)

I think you've probably nailed it. Undermedia (talk) 16:55, 14 June 2017 (UTC)

Yeah that does make sense then why the one addition would not impact it but the second did.

So what should we do? Keep them in and sacrifice the ordering? Or get rid of them? Mikemikem (talk) 17:21, 14 June 2017 (UTC)

Personally, I'd lean towards having the columns sort properly, and hopefully there'd be a way to show the new leaders in the graph instead. Undermedia (talk) 18:16, 14 June 2017 (UTC)

Okay I would be supportive of that as well. If we hear from one or two more people who agree, we can take them out (and hopefully try to find a way to put it in the graph) Mikemikem (talk) 18:55, 14 June 2017 (UTC)

I'm going to make the change so we can properly look at "lead" and other columns.

But I don't want to unilaterally change, so if anyone doesn't want the change and wants the leader elected columns back we can discuss it here Mikemikem (talk) 21:54, 14 June 2017 (UTC)

Digging through the docs I found the solution and it appears to be working. https://en.wikipedia.org/wiki/Help:Sorting#Forcing_a_column_to_have_a_particular_data_type galneweinhaw (talk) 22:42, 14 June 2017 (UTC)

Thanks Galneweinhaw!!!! Very happy you found that, now we can keep both! Mikemikem (talk) 23:37, 14 June 2017 (UTC)

Obscure polls

Thanks Mikemikem, for going through the effort of digging up those "buried" Angus Reid polls. I would caution though that obtaining accurate details for such polls can get sketchy. The recent one from 31 May 2017 seems to check out, but I'm not too sure about the older ones from May 2016 and November 2015 that are shown in a graph from the 12 June 2017 poll. Where exactly did you find the precise last dates of polling for those, as I don't see them anywhere in that report? And what about the sample sizes? It looks like you may have taken the May 2016 sample size from another table on the last page of the report, but we can't be entirely sure that's the same poll, and I don't see a sample size for a November 2015 poll anywhere (in fact, the numbers LPC 40 - CPC 32 - NDP 20 look suspiciously like the actual results of last election). Finally, I wonder if at a minimum there should be numbers for all 5 parties with seats in HoC for a poll to be included in the table. Cheers, Undermedia (talk) 14:15, 16 June 2017 (UTC)

Yeah I was a bit hesitant to add them because the data was only as part of a "tracking" graph. With regards to the other 2 parties not being included, I think I have seen polls before with only the 3 major parties added (although it appears later releases included them and they were subsequently added).

As for the polling dates and sample sizes, I made the assumption Angus had asked the question as part of a federal poll they actually released (as they wouldn't bother doing a federal poll and not releasing something. Obviously I wouldn't add them based on this assumption though, so I emailed them and that was confirmed. This is the email:

"Hey Mike,

Here are the releases related to the vote intention information for November 2015 and May 2016. We did not report on vote intention in these releases but the numbers in the most recent release were taken from the respective surveys - so you can use the fielding dates and sample sizes (methodology and full report including tables is located at the end of each release). Hope this helps:

November 2015 - fielded November 23-27/ 5294 sample size: http://angusreid.org/trudeau-first-month

May 2016 - fielded May 6-13/5150 sample size: http://angusreid.org/federal-issues-may2016/

Cheers,

Lucas Neufeld Research Intern Angus Reid Institute"

However, the polling dates and sample sizes for the November and May polls in the fracking graph are not publicly listed, so if you only want a poll included that individually has its own sample size and dates included in the release (as opposed to just horse race data), you could take them down. Mikemikem (talk) 16:23, 16 June 2017 (UTC)

Again, your efforts to track down all the details are appreciated, and it sounds like they check out. However, I think the ultimate issue here is that personal communications like the above are generally not considered admissible sources of information on Wikipedia because they fail to meet Wikipedia:Verifiability, in the sense that any random reader who visits the page would not be able to verify those details based on the links provided for the polls. Personally, I would suggest drawing the line at needing to patch together poll details from multiple sources and/or personally communicating with the pollsters to obtain missing details; i.e. it would be best if the link provided for each poll in the table contained all the necessary details. For the graph, the minimum required is a last date of polling and a sample size, in addition to party support. It will actually still work if the numbers are missing for some parties (e.g. Green and Bloc), but by and large just about all proper poll reports give the numbers for all five parties. Cheers, Undermedia (talk) 18:53, 16 June 2017 (UTC)

Yeah that is a good point all of the info is not in the same place for those polls.

Seems a shame to let their previous tracking go to waste but the data is incomplete, so makes sense to remove them. Mikemikem (talk) 19:06, 16 June 2017 (UTC)

Given the May 31st poll you found though, looks like it's worth digging through the detailed tables whenever they publish a nation-wide poll on any topic. Undermedia (talk) 23:06, 16 June 2017 (UTC)

Yeah I think especially after the ndp leadership, and from then on. This is what they said:

"Unfortunately, we do not. We don't typically report on vote intention in the two years after an election (based on its relevance with no election pending - the most recent releases being an exception, given that the Conservative leadership race served as an interesting opportunity to look back into it).

We might have a release in the near future that also reports on vote intention after the NDP leadership has been announced, however. Sorry I could not be of more assistance!

Lucas"

So my guess is after October (2 years from election, and also ndp leadership) they might start sneaking them in there Mikemikem (talk) 00:25, 17 June 2017 (UTC)

I'm not sure the most recent Ekos poll meets the criteria....it doesn't have field dates and only lists two parties....

Hopefully more info is released Mikemikem (talk) 01:47, 24 June 2017 (UTC)

User:Undermedia Mikemikem (talk) 01:50, 24 June 2017 (UTC)

Hopefully EKOS will publish it on their website, or else it'll have to be removed from the table; it doesn't even have a proper date. Undermedia (talk) 20:08, 25 June 2017 (UTC)

http://globalnews.ca/news/3554435/northern-populism-canada/

Looks like field dates are June 1-19, from this report on "populism" from ekos with exact same sample size.

But can't find a link that has both the horserace and sample together Mikemikem (talk) 21:53, 25 June 2017 (UTC)

Looks like it's all there now. Good stuff! Mikemikem (talk) 22:56, 25 June 2017 (UTC)

Angus Reid poll Sep 2017: eligible or not?

Angus Reid just released a poll on federal politics with a bunch of typical questions like who would make the best PM, do you approve of the current PM, which leader would be best to deal with this or that issue, etc., and then there's what might be interpreted as a "ballot/voting intention" question, though instead of the usual wording down the lines of "If an election were held today, which party would you support/vote for?", it's worded "Regardless of your individual choice for best Prime Minister – tell us which you think is the best party to form government". See page 13 of the PDF report. How do folks feel about whether or not this should be eligible for addition to this opinion polling page? The results are relatively similar to the recent Forum poll, but then again there may be grounds to believe that the unusual wording of the question — in particular the instruction to disregard one's preference among the party leaders — could have caused respondents to answer differently than had they been asked a standard ballot question. Thoughts? Cheers, Undermedia (talk) 12:39, 28 September 2017 (UTC)

I think it is eligible, some organizations can be a bit more push poll-y, I still think the law of averages apply to all the polls as a whole though. I would include it and one can always compare the different methods. The question is close enough for my liking. Krazytea(talk) 13:49, 28 September 2017 (UTC)

Yeah, I guess the only other point I'd make concerning the wording of the question is that I suspect it may have caused inflated NDP and Green numbers, since as we all know with the phenomenon of strategic voting in our electoral system, asking someone generally which party they think would be the best to form government is different from asking which party they'd actually vote for if an election were held. Similarly, I suspect the 3% support for the BQ may be biased low, as surely some respondents who may be inclined to vote for them recognized that they literally cannot form government. Undermedia (talk) 20:07, 28 September 2017 (UTC)

This poll should not be included at all... "who would be best to form government" is not at all the same as "who do you intend to vote for"

Imagine if they asked this question in the US. Green and Libertarian support would be way up as people on either ends of the spectrum may feel who is "best" to form government is not Democrats or Republicans, but the reality of casting a vote is very different.

In this situation the Bloc can not form government, so the Question is even less comparable to a ballot question in Quebec.

This question completely ignores strategic voting.

Even polling guru Eric Grenier's on twitter has said this is different than a ballot question.

Those 7% who say the Green would be best to form government? Many of those in close 2-way races, despite this, may shift to the ndp or even LPC.

It should be disregarded from our voting intention tracker, as it simply does NOT ask voting intention. Mikemikem (talk) 20:51, 28 September 2017 (UTC)

Look to the media for guidance. https://globalnews.ca/news/3774532/justin-trudeaus-liberals-would-beat-conservatives-ndp-in-election-ipsos-poll/ In the article they reference the AR poll, but only in regards to government approval rating and whether or not it's time for change in government. No horserace mention, as it would be totally inappropriate. AR didn't ask for voting intentions. Including that poll in the tracker would literally be making things up.

I agree with whoever wrote above (but did not sign). Attributing ballot answers to "best to form government" is quite a stretch. Mikemikem (talk) 23:24, 28 September 2017 (UTC)

What caused the drastic change in November 2016 ?

Dr. Universe (talk) 04:16, 2 October 2017 (UTC)

What do you mean? Mikemikem (talk) 19:25, 2 October 2017 (UTC)

Angus reid polls

So Angus reid has now, a week after their Jagmeet Singh release, released a horse race poll with the horse race as the lead.

The Jagmeet Singh release included a horse race question that we have included in our table, but it was question 16.

I spoke (on twitter) with Eric Grenier (polling expert who runs the CBC poll tracker) and he said he would not include the horse race poll that was question 16 because 15 questions being asked before can influence it (and the questions were not randomized like other omnibus surveys).

The recent Angus poll that is being promoted as a horse race poll asked the horserace question first.

Should we include pllls that ask the horse race question way down the list as an after thought and do not have the question order randomized?

Mikemikem (talk) 06:45, 13 October 2017 (UTC)

I would be inclined to keep that one, as well as the other similar one from May 31st. Note that the new poll does not in fact ask the horserace question first, but fifth, after asking respondents to rate the most important issues facing Canada as well as rate each of the party leaders, so I think it would be tricky to try to draw a clear line on this. Cheers, Undermedia (talk) 14:57, 13 October 2017 (UTC)

I did not realize that actually. That makes sense then the line would be tough to draw. I thought for this release it was first question. So makes sense to keep both.

Mikemikem (talk) 18:16, 13 October 2017 (UTC)Mikemikem Mikemikem (talk) 18:16, 13 October 2017 (UTC)

Errors Nanos Polling Numbers

Numbers for Nanos polls should be double checked. I have found quite a few errors. Nanos produces a monthly poll with a sample size of roughly 4,000. They often release the poll as weekly segments with a sample size of 1,000. Along with their party preference numbers they produce a "party power index" that is meant to quantify "momentum as well as popularity" but, is not generally speaking an "opinion poll" for while the sampling may be random, the criterion that make up the PP index are subjective. I believe this may be the cause of the errors.

I'm afraid you're misunderstanding how the Nanos rolling polls work. As per the description of the methodology in any of the PDF reports linked to in the poll table, "The weekly tracking figures are based on a four-week rolling sample comprised of 1,000 interviews. To update the tracking a new week of 250 interviews is added and the oldest week dropped." That's why the sample size for each weekly poll in the table is listed as "1,000 (1/4)", because each set of results is based on a total of 1,000 interviews, but only a quarter of them are new interviews in relation to those that made up the previous week's poll. Nanos indeed also publishes weekly "Party Power Index" numbers based on a propriety set of indicators, but it's only the standard "ballot" question results that are entered into the poll table. Feel free to double-check any of the numbers in the table against those in the linked PDFs, but I'm fairly certain they've been properly entered; there are several of us regularly overseeing quality control on this page. Cheers, Undermedia (talk) 22:26, 23 April 2018 (UTC)

Decided versus decided+leaning

Some of the included polls (e.g., Innovative Research for May 14th and 2nd) refer to decided voters whereas other polls (e.g., Forum for May 16th) refer to decided+leaning voters. These are not directly comparable.

Moreover, the two aforementioned Innovative Research polls include decided+leaning totals.

For a more accurate chart, I suggest choosing one (decided or decided+leaning) as either a single standard or default standard, but with some type of clear communication of what standing is being used. Alternatively, two charts can be created - one for decided voters, one for decided+leaning voters.

Williamefwilson (talk) 18:32, 16 August 2018 (UTC)

Polls released on Twitter

I'm wondering if we should draw the line when it comes to including polls released only on Twitter. I've been waiting to see if EKOS releases a proper report on their website for the July 2018 polling results Frank Graves tweeted, but at this point it seems no such report is forthcoming. Generally, Twitter is not considered to be a reliable source for information on Wikipedia, although Frank Graves is certifiably the president of EKOS and can personally be considered a reliable source, so perhaps an exception could be argued. All the necessary poll details are there, although the "4%" for the BQ is likely rounded, as EKOS often does for the lesser parties on their graphs to save space. Undermedia (talk) 11:12, 7 August 2018 (UTC)

If Graves posted it, it would be reliable to cite, but it would be nice to have more details in a proper report. - Ahunt (talk) 13:12, 7 August 2018 (UTC)

From Graves' explanations in the Twitter thread, it seems that those results are from a first batch, for the period of July 20 to 26, which will be integrated in the final report into a larger sample spanning a longer period. Previous EKOS polls spanned about 15-20 days, so this one might perhaps be about July 20 to August xx (8?, 10?, 12?, ...) and then add a few more days until the release of the report. However, the fact that Graves publicly released separately the results of the July 20-26 batch means that EKOS considers that it is a valid sample in itself and that it could stand alone as a poll. In the Wikipedia list, my personal preference right now would be to remove the line about this batch, and wait for the complete report. Then, depending on the information available in the report, Wikipedia editors can decide if there is one unseparable poll or if it can be considered more accurately as an aggregation of two or more polls which could be listed on separate lines (e.g. a July 20 to 26 poll and a July 27 to August xx poll). -- Asclepias (talk) 16:38, 7 August 2018 (UTC)

Ok, that makes sense. I guess we can leave it as is for now, and I suspect when EKOS comes out with something more official it'll end up as a single, larger poll ending on a later date. Undermedia (talk) 18:04, 7 August 2018 (UTC)

I agree: go with what we have now and then update when the final report comes out, with whatever details can be used from that. - Ahunt (talk) 18:11, 7 August 2018 (UTC)

In the specific case of the latest Ekos poll, Frank Graves tweeted the results from his personal account and they include a sample size, margin of error, and date range, so I think they're perfectly fine. You can compare this to a recent tweet from Darrell Bricker (Ipsos), where he vaguely refers to the NDP being at "around 20%" nationally - I wouldn't include this, even if he provided numbers for all of the major parties, unless he added the extra bits of information that Graves provided.

Williamefwilson (talk) 18:36, 16 August 2018 (UTC)

Removing floor crossers

Hello all,

I've removed two references to today's floor crossing by Leona Alleslev because we haven't been keeping track of other seat minor changes to date. Thoughts?

Unlike leadership changes, seat changes don't have a clear or obvious effect on party polling performance. The polling list will also get more cluttered if we add these changes.

Williamefwilson (talk) 21:32, 17 September 2018 (UTC)

While an intriguing and important event, it does not belong in this polling list. Krazytea(talk) 21:41, 17 September 2018 (UTC)

This was discussed in a previous talk called “New Leaders.”

We agreed to include new, permanent (not interim) party leaders because Canadian politics are so leader focused, so having a new permanent leader can really shift the political landscape.

Note that the only five additions to the table involve major party leaders.

In my opinion we shouldn’t actually include Oullet resigning. We didn’t include Harper resigning, or Mulcair resigning, or Duceppe resigning, so I’d be in favour of removing that.

As for party switching and other minor things like that, my vote is definitely no. Way too much clutter.

Permanent party leaders should be it, in my opinion (this includes taking out the Oullet resignation). Mikemikem (talk) 21:42, 17 September 2018 (UTC)

PPC/QD/CCF

I noticed that someone added Bernier's new party to the polling page. Shouldn't Quebec Debout (more members) and the CCF (older) also be added? — Preceding unsigned comment added by 162.221.126.231 (talk) 13:08, 17 September 2018 (UTC)

Personally I think it's a bit premature to add Bernier's party to the table; it remains to be seen whether any pollsters (let alone most/all) will actually measure/report PPC support from this point on. As for Québec Debout, it's being reported just this morning that those members have decided to rejoin the Bloc. Cheers, Undermedia (talk) 13:17, 17 September 2018 (UTC)

I agree with undermedia on this one. I think it depends on which parties the pollsters decide to ask about/measure/report. We have one poll including PPC. If that continues and becomes the trend, then perhaps we can add them. I think for now, we add a note saying that says (includes 13% for a proposed Maxime Bernier party) or something along those lines. Bkissin (talk) 13:23, 17 September 2018 (UTC)

No, we don't have a usable poll including PPC. There's only that poll about a then non-existent party and where the question about voting intentions was introduced with an explanation of only that party's possible platform. Introducing a voting intention question with an explanation of the platform of only one of the parties and not the others obviously biases the result. It's not a normal poll. I understand why the pollster did that poll and it may present an interest if not taken out of context. But that's definitely not comparable to normal voting opinion polls. That's comparing apples and oranges. That poll should not be included in this article in any shape or form. -- Asclepias (talk) 14:48, 17 September 2018 (UTC)

Agreed. From Abacus Data's press release: "After describing Mr. Bernier’s criticism of the Conservatives and informing respondents about his intention to form a new party which would reduce immigration, end supply management and avoids retaliatory tariffs against the US, we asked voting intentions again." Without a doubt that "preamble" would've biased subsequent responses to the voting intention question. Cheers, Undermedia (talk) 15:54, 17 September 2018 (UTC)

I would love to include them for curiosity's sake, I don't think we have enough polling history or information on whether they are organized and substantive enough to be included. Also while I would consider a floor crosser notable, I don't think it is large enough to include as a bolded event in the polling list. Krazytea(talk) 19:30, 17 September 2018 (UTC)

Putting 2 versions of the same poll in the table is not going to work from the graphing perspective. Need to find another way to do this. Given that Abacus's PPC results only came by tweet after the poll was released, I'd say it's still looking uncertain as to whether polling for the PPC will become a trend among the pollsters. 7 is a fairly significant level of support though. Cheers, Undermedia (talk) 14:06, 20 September 2018 (UTC)

7 points is about where the Reform Party of Canada was polling prior to the 1993 election. I would hope pollsters make PPC results available. --Llewdor (talk) 23:44, 20 September 2018 (UTC)

Should "other" results be included?

Just as the title says: should support for "other" be included? It's the case in Opinion polling for the Quebec general election, 2018, and it gives a broader sense of the uncertainty of the electorate heading into the election, and how that uncertainty varies over time. WhyIsItWereHere22 (talk) 10:43, 25 September 2018 (UTC)

Party Abbreviations

In this equivalent page for the 2015 election polling, the following abbreviations was used: Liberal, Cons., NDP, BQ, Green

In this page, the abbreviations are changed to: LPC, CPC, NDP, BQ, GPC

Is there a reason why there has to be inconsistency for the way information is presented between articles?

Granted, the abbreviations from the 2008 elections differs from 2015, but that's no reason not to begin streamlining (to some degree) the reading experience for readers. One possible solution is to change the abbreviations from the 2015 polling to match the abbreviations used in this article, or vice-versa. Whatever consensus reached could also be applied to the opinion columns from the 38th 39th, 40th, and 41st federal election articles.

Thoughts? Aquitoba (talk) 20:13, 29 September 2018 (UTC)

I actually prefer the LPC, CPC, NDP, BQ, GPC method, it is a consistent way of indicating each party even though some like GPC might not be as common. I am not totally invested in this debate as I do feel it is a small issue. The only thing I would say is that I hate Gre as an indicator for green, it is really ambiguous and though we know it is for the Greens I feel it is a poor indicator for that party. Krazytea(talk) 20:34, 29 September 2018 (UTC)

I'm also not terribly invested in this, but for what it's worth, I second Krazytea on LPC, CPC, NDP, BQ, GPC: makes for consistent abbreviating of all the party names, and for nice narrow columns of consistent width among the parties. Cheers, Undermedia (talk) 21:03, 29 September 2018 (UTC)

Lib, Con, NDP, BQ and Grn also create narrow columns. Onetwothreeip (talk) 04:35, 1 October 2018 (UTC)

Add latest Nanos PDF when available

I've added the new Nanos numbers for Oct. 5, linking to their data portal. However, the PDF should be linked to instead when it becomes available.

Williamefwilson (talk) 22:59, 8 October 2018 (UTC)

Sample size for latest Innovative Research Group poll

Hello everyone,

The raw sample size for the latest Innovative Research Group poll is 2,410 whereas the weighted sample size is 1,200.

According to the methodology description on Slide 2, the reported results reflect the weighted sample size, so is this not what we should be reporting here as opposed to the raw sample size, and is this something that we have been consistently checking for all polls?

Thanks!

Williamefwilson (talk) 16:47, 4 October 2018 (UTC)

I confirmed with Innovative Research Group - it should be the weighted numbers. I'm going to correct the chart accordingly.

Williamefwilson (talk) 17:10, 4 October 2018 (UTC)

Just to clarify, the results of all polls are actually based on a "weighted" sample; weighting by demographic parameters is a standard step in the processing of polling data. Furthermore, the headlining "horserace" results reported by pollsters are by standard practice based on only "decided and leaning" respondents (hence they add up to ~100%), which represents a smaller chunk of the sample still. But since pollsters regularly don't actually report what the weighted sample size was, nor what the sample size of decided and leaning voters was, we've been going with the raw sample size, which all pollsters consistently report. Cheers, Undermedia (talk) 19:32, 4 October 2018 (UTC)

Somewhat off topic (but still about this poll), it looks like the link for this Innovative poll is no longer working Mikemikem (talk) 15:33, 5 October 2018 (UTC)

They re-released the PDF with regional numbers added. That may be why the original link no longer works. In any event, the link works now, so it's apparently no longer an issue.

Williamefwilson (talk) 23:02, 8 October 2018 (UTC)

As a heads up, Innovative poll DOES ask for people’s party without any special prompting.

They are at 3%.

So they could be added for this poll Mikemikem (talk) 18:06, 9 October 2018 (UTC)

Maxime Bernier polls. Do we want to include?

When polls ask “what if Maxime Bernier’s party ran a candidate in your riding? How would you vote if an election were held today?”

Do we want to include these preliminary hypothetical polls?

I’m open-minded, but it is a pretty big addition to the table that I feel warrants discussion Mikemikem (talk) 14:43, 20 September 2018 (UTC)

I think they should only be considered for inclusion if the question is asked in a standard "unprefaced" manner, i.e. straight up "Who would you vote for: Liberal, Conservative, NDP, BQ, Green, PPC or Other?", not if it's a leading question like "If Maxime Bernier's People's Party of Canada ran a candidate in your riding, who would you vote for?". Furthermore, as I mentioned in the discussion two topics above, putting two versions of the same poll in the table isn't going to work from the graphing perspective. We'll need to decide shortly whether we consistently go with one or the other, I suggest once we start to see whether or not most of the regular pollsters begin measuring and reporting PPC support in a standard fashion from now on. In my view, Abacus's supplementary tweet of PPC results doesn't really count towards this. Cheers, Undermedia (talk) 18:00, 20 September 2018 (UTC)

Alright I am in agreement with that.

Should we remove PPC from the table then?

Until they start getting included as a regular party and not an afterthought as a second question, that is Mikemikem (talk) 20:13, 20 September 2018 (UTC)

They should be included completely, even in hypothetical questions. They can be placed additionally. Onetwothreeip (talk) 08:03, 21 September 2018 (UTC)

From a previous talk, it doesn’t work for the graph to have two entries for the same poll.

And why should a hypothetical be included? That isn’t a neutral question Mikemikem (talk) 11:10, 21 September 2018 (UTC)

I would argue they should be included even when not directly listed as the People's Party of Canada, as the most similar analogue, Francois Legault and the CAQ, is included in the polling table for Quebec 2012 even when listed as "New Party led by François Legault". Tholden28 (talk) 19:00, 21 September 2018 (UTC)

CAQ is being asked in the first horserace question.

People’s Party is not. The horserace question being asked if LPC/CPC/NDP/Bloc/Green/Other. Then as a follow up People’s Party is asked.

When they are included in the original horserace question and not given special attention as an afterthought in a second follow up horserace question, then they’ll be included. Mikemikem (talk) 21:31, 21 September 2018 (UTC)

It's standard practice to include multiple results from the same poll if they are analogous to results. Any caveats are easily handled by including a note next to the polling value for the given party, or next to the polling detail itself. Onetwothreeip (talk) 22:23, 21 September 2018 (UTC)

Well, the graph is going to have to go on hold until this gets sorted out. As I've mentioned twice already, having 2 versions of the same poll in the table isn't going to work in the graph. In order for me to update it with new polls at this point, I would need to adjust the script to account for the addition of a new column in the table and consequent shifting of the sample size column, but I hesitate to do so because it still looks rather uncertain whether the PPC column will stay: we're still yet to see a single poll that has gauged PPC support in a standard, non-leading way, though it's only been a week since the party was officially created. Cheers, Undermedia (talk) 01:04, 22 September 2018 (UTC)

I say include because the pollsters are asking about support for his party and we have some early numbers. It provides a broader sense of the overall context heading into the next election.

Williamefwilson (talk) 12:59, 24 September 2018 (UTC)

Just a quick note that the latest Innovative Research Group poll includes new numbers for the People's Party of Canada (3%), but they have been collapsed into the overall figures for "Other Parties" as explained on Slide 11. As more and more pollsters begin polling for the PPC support, I think we should start to include the numbers even though the party hasn't been officially registered yet.

Williamefwilson (talk) 16:53, 4 October 2018 (UTC)

Absolutely. 3% is certainly enough. We have listed polling results for Bloc Quebecois and Green lower than that. Onetwothreeip (talk) 22:25, 9 October 2018 (UTC)

Adding a column for People’s Party

I think now is the time to consider adding a column for People’s Party.

Innovative Research put the PPC at 3% in the headline question (unlike Abacus asking as a second question), Pollara put PPC at 2% in the headline question, and the Mainstreet poll coming in a few weeks will include PPC as well.

I do not know how to add a party to the table, but believe now may be the time.

Thoughts? Mikemikem (talk) 15:49, 13 October 2018 (UTC)

I agree, the polling is recording them and they are going to have an impact on the next election, it is time to add them in. - Ahunt (talk) 15:58, 13 October 2018 (UTC)

If you know how to add them, I’d say do it whenever you have time, and those who disagree can comment in here to discuss a removal Mikemikem (talk) 17:25, 13 October 2018 (UTC)

It only 1 poll that used PPC as an option. It make sense to add an extra column if more polling companies uses PPC as an option. Until then, it looks premature at this point to add another column. Aquitoba (talk) 01:59, 14 October 2018 (UTC)

They are at 3% in Innovative poll and 2% in Pollara. Both asked in the original horse race question.

That’s two.

How many do you need? Mikemikem (talk) 03:59, 14 October 2018 (UTC)

I'd say once we have a 3rd pollster reporting PPC numbers, that should seal the deal. Clearly the party is having a noticeable impact, as even pollsters who aren't yet reporting PPC results (like Nanos) are showing unusually high support for "other parties" of late (about ~5% in the latest Nanos poll), presumably due to support for the PPC. Cheers, Undermedia (talk) 14:20, 15 October 2018 (UTC)

I would also add that the "other" support is even more noticeable provincially: the same Nanos polls show it at about 14% in QC. That alone could be a huge game-changer. Hopefully Nanos starts reporting PPC numbers soon to show a more accurate picture of things. Given the 4-week rolling sample they use, that should be this week. WhyIsItWereHere22 (talk) 15:26, 15 October 2018 (UTC)

That was a good prediction! Nanos included PPC this week Mikemikem (talk) 23:59, 15 October 2018 (UTC)

Man I'm good. Kind of odd that it still has "other" at such a high number in QC, though. But then "other" was abnormally high a year out before the QC election too and dwindled down to 2-4%, so it must be a Quebec thing. WhyIsItWereHere22 (talk) 00:43, 16 October 2018 (UTC)

I had thought 12% or so of that 14% “other” was PPC, seeing as Bernier is from Quebec and all. Shocked to see them barely registering there Mikemikem (talk) 01:00, 16 October 2018 (UTC)

Yes that's what I assumed as well. Living in QC I haven't heard much talk about him at all, though the Montreal area isn't exactly the best indicator of Quebec's mood (see the results of this year's election here). Although I must say I'm more amazed that he's doing best in Atlantic Canada, or that the Greens have surpassed the NDP there. WhyIsItWereHere22 (talk) 02:18, 16 October 2018 (UTC)

Just an FYI (in case people start wondering) that I won't be able to add the PPC to the graph until more polls showing PPC support are released (maybe at least ~half a dozen in total), as it takes a certain minimum number of data points to be able to properly compute a trendline. Cheers, Undermedia (talk) 12:53, 17 October 2018 (UTC)

Exclude Forum?

This will likely generate a lot of pushback, but is there any interest in excluding Forum polls outside of an immediate election context?

They're a consistent outlier and Eric Grenier applies a penalty to them in the Poll Tracker.

It just seems like Forum is a consistent outlier, and this distorts the overall picture. (In a way, it's kind of like asking if you should keep Rasmussen in a tracker of American polls.)

There are some methodological issues with the latest Forum poll, too. Over half of the total sample is 65+ and they surveyed more men than women, and although the numbers are supposedly redistributed based on the latest census data, they don't actually provide real numbers for this (compare to Ipsos).

Anyway, I'm just tossing this out there. Feel free to say "No!"

Williamefwilson (talk) 18:28, 16 August 2018 (UTC)

I think the most impartial policy to follow here is to accept all polls by legitimate/established pollsters (i.e. those that are commonly cited in the media), which includes Forum. As much as Forum polls admittedly have a tendency to seem like outliers, for what it's worth they actually have a pretty strong record of calling elections, including the 2015 federal election, the 2017 BC election, the 2018 ON election, and a bunch of others. Overrepresentation of the senior age group in the raw sample is an inherent issue with all IVR polls (which EKOS and Mainstreet also do) because seniors evidently tend to be more inclined than younger age groups to respond to a "robo-poll" over the phone. No polling methodology is perfect! Cheers, Undermedia (talk) 19:32, 16 August 2018 (UTC)

I would definitely include it, even outliers can predict something other elections can't, we've seen this in the UK. I know that Sun Media is now using Forum as their primary polling firm and may have an agenda as well. All polling companies can have certain agendas that we don't know about. As Undermedia has said, no polling method is perfect, and I hate the idea of censoring certain polling firms, or major media platforms, whether I agree with them or not. Krazytea(talk) 21:26, 16 August 2018 (UTC)

I just want to vent, but another outlier from Forum has landed... From an industry perspective, I don't think this is good... They're consistent outliers, generating big but misleading headlines... Sigh. Williamefwilson (talk) 14:46, 17 October 2018 (UTC)

Outside their methodological issues, there seems to be some evidence that their results are tainted by some political agenda. Be reminded that polling can create its own political momentum. - Ahunt (talk) 16:09, 17 October 2018 (UTC)

name change

With it now physically impossible to have the election this year, isn't it time to change the name of the articles to Opinion polling in the Canadian federal election, 2019?Arglebargle79 (talk) 22:17, 3 November 2018 (UTC)

This discussion should really stay at Talk:43rd Canadian federal election, since the name of this article is dependent on the name of 43rd Canadian federal election. We aren't going to change this article name unless that one is changed first. I see you have already started this same discussion there anyway, so let's discuss there. - Ahunt (talk) 22:41, 3 November 2018 (UTC)

Polls by provinces

I added polls by provinces. Some polling firms do not poll the entire federation, and thereby cannot be added to the main table. I have added 10 MQO Research polls, bbut they have 2 more coming out soon, and more made in April 2017 and October 2016. These will also be added latter on. 173.209.113.109 (talk) 21:59, 10 November 2018 (UTC)

Leger website

Leger has updated their website and the links in our table are all broken.

Has anyone been able to find the section of the website that has the polls? Mikemikem (talk) 22:52, 13 December 2018 (UTC)

Polls completed on the same day

Where two polls were completed on the same day, how do we want to arrange them?

By release date? By start date? Alphabetically?

Thanks!

Williamefwilson (talk) 18:30, 2 January 2019 (UTC)

Put the one that started latest on top, and in the event that the two polls have the exact same field dates, I think we've typically sorted them by release date (i.e. latest release on top). Cheers, Undermedia (talk) 20:05, 2 January 2019 (UTC)

Proposal for Party Section within table

I feel that there are some problems with the layout of the table; mainly the fact that the text has the party colour in the background. To me, I don't know about others, this can sometimes make it difficult to read, as in the case with the PPC. I have a proposal in which the table follows a similar format such as the Opinion polling for the next German federal election, Opinion polling for the next United Kingdom general election &c. Below is how it would look...

Feel free to discuss any criticisms, suggestions &c below.

Polling firm	Last date of polling	Link	LPC	CPC	NDP	BQ	GPC	PPC	Margin of error[1]	Sample size[2]	Polling method[3]	Lead

JDuggan101 ^{talk. | Cont.} 20:32, 16 January 2019 (UTC)

Alternatively dark text on a dark background can easily be changed to white text on a dark background. - Ahunt (talk) 20:38, 16 January 2019 (UTC)

I was thinking about that but when one of the party's text is white it looks out of place and, with the bold font, looks rather ugly. JDuggan101 ^{talk. | Cont.} 20:42, 16 January 2019 (UTC)

I agree with JDuggan101's opinion. One party should not be made to stand out from the others. -- Asclepias (talk) 21:25, 16 January 2019 (UTC)

JDuggan101, I like the look of your layout above. IMO, it looks better than the current one. There might be one problem: in the German and UK poll tables, the party columns are not sortable. Me, I don't mind, but from previous discussions you can see that there are users who want sortable party columns. If you can make them sortable, it would be better. If not, I still prefer your layout, but you might hear objections from other users. -- Asclepias (talk) 21:25, 16 January 2019 (UTC)

It might be the most intuitive, but on all the polling articles I maintain (German/Dutch/French/Swedish/etc.), the columns are indeed sortable – you click the color bar (because of MediaWiki issues, it's not on the column itself as it messes up with the width and also makes some already-wide tables excessively so). Mélencron (talk) 21:32, 16 January 2019 (UTC)

Oh, that's right, they are sortable, it's just not indicated, so you have to know that they are and where to click. Thanks for the info. Problem solved. -- Asclepias (talk) 23:26, 16 January 2019 (UTC)

Not sure if those "split" cells would throw off the graph-generating script (which extracts the polling data directly from the table), but if so I could probably figure out a way to modify the code to accommodate it. Otherwise it looks nice, provided the columns are sortable. Are all the columns sortable, or just the parties? Sometimes people like to sort by polling firm, for example. Cheers, Undermedia (talk) 16:38, 17 January 2019 (UTC)

Judging by the example of the article about Germany, all columns are sortable. My question was only about the party columns, because they were the only columns whose sortability was not indicated by the little triangles. -- Asclepias (talk) 17:34, 17 January 2019 (UTC)

All Nanos links are broken

Does anyone know why all our Nanos links are broken? And if there is an easy fix to this? Mikemikem (talk) 00:13, 29 January 2019 (UTC)

Not all. The link works if you link to the partly blingual version instead of a single-language version. Example 1: Instead of "www.nanos.co/wp-content/uploads/2019/01/Political-Package-2019-01-18-Slides.pdf", link to "www.nanos.co/wp-content/uploads/2019/01/Political-Package-2019-01-18-FR.pdf". Example 2: Instead of "www.nanos.co/wp-content/uploads/2019/01/Political-Package-2018-12-28.pdf", link to "www.nanos.co/wp-content/uploads/2019/01/Political-Package-2018-12-28-FR.pdf". Etc. -- Asclepias (talk) 01:02, 29 January 2019 (UTC)

It's a general mess. All the older ones from before their recent switch from nanosresearch.com to nanos.co are also broken, I assume. Undermedia (talk) 15:50, 29 January 2019 (UTC)

Many older polls not listed at "www.nanos.co/our-insight/" may have been removed from the Nanos website. User:InternetArchiveBot linked [3] to the pages that had been copied to web.archive.org, but not all polls were copied there. -- Asclepias (talk) 16:52, 29 January 2019 (UTC)

More generally, one can try to rewrite the url by adding, removing or replacing parts of the url such as "Slides" or "FR" or by switching the date and the text or a combination of all those methods, until finding a url that works. Example 3: Instead of "www.nanos.co/wp-content/uploads/2019/01/2019-01-11-Political-Package-Slides.pdf", link to "www.nanos.co/wp-content/uploads/2019/01/Political-Package-2019-01-11-FR.pdf". Their list at "www.nanos.co/our-insight/" seems to have the links that currently work. -- Asclepias (talk) 16:23, 29 January 2019 (UTC)

Change to structure of preferred PM

I propose that we change the structure of the preferred PM numbers to reflect how the national polls are listed.

I understand that the current structure is divided between the arrival of each new leader and that this is meant to signify the change and isolate the findings from the previous period, but in fact, there remains overlap between periods and the current layout isn't always easy to follow/intuitive. (In other words, the timing of polls that actually include the new leader don't always align with his or her elevation to leader of the party.)

For example, not everyone has begun polling for Bernier/the PPC (e.g., Abacus). If we follow the current layout, these findings should still be listed under the Bernier sub-heading even though there are no numbers for him, or we should place them out of chronological order and under the previous period.

Similar to how the national polls are currently listed, we could indicate in a general list when a new party leader was chosen/formed a new party and just list whatever numbers are included in a particular poll, linking the leader to his or her party.

I did a major update to the preferred PM numbers a few weeks ago following the formation of the PPC/Bernier becoming leader of the PPC, and I first encountered the problem then. Now it's happened again with Blanchet taking over the BQ.

The one major con, I guess, is that where the leaders' names currently appear, we would have to change them over the corresponding party names, but this would allow us to create a static list that doesn't require a new one every time a party leader changes, and it would allow us to capture the overlap between periods.

Thoughts? #Controversial

Williamefwilson (talk) 21:16, 29 January 2019 (UTC)

I agree that all the subdivisions of the best PM section are unsightly. Not sure quite what to suggest. Maybe just list the parties in the top row, then add a "current leader" row just below that, and then insert special rows throughout the table to indicate the various leadership changes along the way, same as the main polling table. Would that work? I think for the sake of simplicity, we can't make special accommodations for the 4-week Nanos polls that overlap leadership changes; just go with the last date of polling. Undermedia (talk) 22:02, 29 January 2019 (UTC)

That sounds like a good approach to me, anyone else? Williamefwilson (talk) 05:53, 30 January 2019 (UTC)

I agree, let's give that a try. - Ahunt (talk) 14:28, 30 January 2019 (UTC)

Yes I agree, that sounds good Mikemikem (talk) 14:19, 20 February 2019 (UTC)

Sorting table by party vote %

Does anyone know why we can’t sort the table by party vote % any more? In the 2015 table, and until very recently in this table, there used to be two little arrows on top of “CPC” “LPC” etc. And if you clicked them it would sort all the polls from that party’s highest % to their lowest %.

But looks like we can’t anymore. Is there a way to fix this? And does anyone know why the option was removed? Mikemikem (talk) 14:18, 20 February 2019 (UTC)

You can still do it, just click on the colored bar. Mélencron (talk) 14:43, 20 February 2019 (UTC)

See the discussion a few sections above, in the section "Proposal for Party Section within table". -- Asclepias (talk) 15:58, 20 February 2019 (UTC)

You are correct. Some laziness on my part.

Thanks for the explanation Mikemikem (talk) 16:34, 20 February 2019 (UTC)

PPC on the graph

Does anyone know how to add the PPC to the graph? They’ve been featured in enough polls I’d say to be included, but I don’t see them. Thoughts? Shin Jun Bak (talk) 14:35, 21 November 2018 (UTC)

Hi. I maintain the graph and addressed this point already in the "Adding a column for People’s Party" above. Basically, a certain minimum number of data points are required for the plot generation algorithm to be able to successfully compute a smooth LOESS trendline, and more points tend to be required when there is little variation among the values of the data points (as is the case with parties that have very low levels of support like the PPC). I'll try again soon to see if it will work. Cheers, Undermedia (talk) 14:57, 21 November 2018 (UTC)

Update: I finally figured out what was truly preventing the PPC from being properly plotted in the graph. In order for the level of sensitivity of the trendlines to remain constant as new polls are added, the graph-generating code is set up so that the proportion of the total available data in the table that's used to compute the trendlines is an ever-adjusting figure calculated by dividing the fixed number of recent polls we want to base the trendlines on (25) by the total number of rows in the table. This currently works out to a very small proportion (<10%) of the data in the table, which works fine for the parties with polling numbers going all the way back to last election; but in the absence of valid numbers for the PPC before their recent creation (blank table cells are simply treated as "no data"), the code was trying to use <10% of the (relatively scarce) available PPC data, which was insufficient to properly compute a trendline. Thus, to fix this I've set PPC support to "0" in all polls prior to its creation to provide the same amount of valid data for the trendline algorithm to work with as for the other parties. Please do not remove these 0's. Cheers, Undermedia (talk) 16:19, 21 February 2019 (UTC)

I removed the zeroes because the People's Party did not poll at 0% for those polls. Once I removed them the graph was not changed. If it's necessary to use the actual table that we present on the article to generate the graph, then that can be done by having a parallel table somewhere else, like on a user page. Clearly we can't actually present results for the People's Party that are completely fabricated. Onetwothreeip (talk) 20:47, 21 February 2019 (UTC)

The chart does not change automatically, you must generate it manually, and without the zeroes the trend line for PPC will just not work. I've coloured the zeroes white so that them remain invisible to the naked eye; this way, the chart code will keep functioning for PPC. Impru20^talk 21:05, 21 February 2019 (UTC)

Unfortunately that won't work either, my eyes are most naked and I can still see them. They're also much clearer when highlighting it, and will be transferred if the data is copied. It's still fabricated information being presented as fact. Onetwothreeip (talk) 21:10, 21 February 2019 (UTC)

@Ahunt and Impru20: It's obvious how the graph is generated and the problems with many null entries for the People's Party, by the edit summary and again from the talk page. That doesn't change the fact that the values are incorrect and should be corrected like another other factual error, regardless of why it's there. As I said already, if it's necessary to use the actual table that we present on the article to generate the graph, then that can be done by having a parallel table somewhere else, like on a user page. It's not necessary to generate the graph from the table that readers of Wikipedia will actually see. Onetwothreeip (talk) 21:08, 21 February 2019 (UTC)

This is merely a technical requirement the chart needs, no one is intending to fabricate any information and present it as fact. Further, you cannot force anyone into having and maintaining a parallel table in their user page. This is really making a mountain out of a molehill... Impru20^talk 21:32, 21 February 2019 (UTC)

The false numbers still appear when they are highlighted, and when they are copied and pasted elsewhere. For the third time now, I understand that nobody intends to mislead people, but that doesn't change that we have fabricated information on the table. If nobody wants to put it on their user page then I can make my own user page for it, but even if nobody would that still wouldn't justify publishing wrong information. This is seriously not the only way to make a graph work. Onetwothreeip (talk) 21:49, 21 February 2019 (UTC)

Goodness, what a passionate debate this has sparked. Personally, I think maintaining a parallel graph elsewhere is rather a bother compared to what I find to be the very clever alternative of colouring the 0's white. I just looked at the table very closely on 2 different computer monitors and truly couldn't make them out at all, and on my phone I can just barely make them, but even there it's unlikely I would've noticed them had I not been looking for them in the first place. So I'm willing to bet that next no one who doesn't have prior knowledge of their existence will ever notice them, and even if they did (by highlighting or copying, for example), it's pretty self-evident in the overt effort to conceal them that they're meant to be disregarded. Cheers, Undermedia (talk) 22:07, 21 February 2019 (UTC)

I wasn't after a debate, I thought I would just remove the false information and we could move on. This really isn't a tenable solution, often when these things are copied it defaults back to black text. Let's not pretend that the only simple solution is to publish wrong information, nowhere else on Wikipedia does this happen. Onetwothreeip (talk) 22:11, 21 February 2019 (UTC)

Abacus Data ?? Some missing??

I've noticed that some of the Abacus data polls are included but not all of them.... https://twitter.com/Colettod/status/1104026347347869696 CBC recently reported today about this poll saying Abacus was the last poll out with a release on Friday night... although I can't see anything on their website I see some of the Abacus polls in that graph shown in the table, but not all... I was going to include them, what do you guys think? — Preceding unsigned comment added by Kflack (talk • contribs) 03:25, 10 March 2019 (UTC)

Strongly encourage you or anybody to include them. Onetwothreeip (talk) 03:44, 10 March 2019 (UTC)

Rolling window (temporarily), methodology/older data available here. Unlike the Nanos releases which are weekly, Abacus Data is publishing daily results, so I don't know how Undermedia would prefer that these be entered. Mélencron (talk) 04:18, 10 March 2019 (UTC)

They also have older preferred PM numbers (before the rolling poll) which aren't in the table, if anyone feels like tracking those down. Mélencron (talk) 13:43, 10 March 2019 (UTC)

I guess you refer to the daily rolling polls conducted between 28 February and 10 March 2019 (when this rolling poll is expected to stop). Currently, in the Wikipedia list, the last Abacus entry, dated 4 March 2019, is the aggregate of the first five days (28 February - 4 March) of this series, as it was reported in the report Abacus released on 6 March. I suppose you can add the polls, if you have all the relevant numbers and sources and if you find a way to fit the polls in the list in a consistent manner. It would probably look like the Nanos rolling polls, except daily instead of weekly. As such, it would almost monopolize the list for a few days, but that can be ok, as the whole daily rolling sequence covers only 11 days in total. (If you choose to leave the 4 March line as it is currently and if you add one line per day from 5 March to 10 March, that would add six lines to the list.) -- Asclepias (talk) 04:22, 10 March 2019 (UTC)

I guess I'd been waiting to see if Abacus releases another official followup report on their website, ideally with multi-day aggregate results again, rather than sourcing a series of daily rolling polls from tweets. In the interim the latter could work though. From my understanding each daily set of results is based on a 3-day sample of 900 respondents with 300 new interviews per day, so in the sample size column the correct notation would be "900 (1/3)". Cheers, Undermedia (talk) 12:40, 10 March 2019 (UTC)

Regional polls?

The regional poll section is a complete mess. Is it supposed to be for regional only polls or regional numbers from national polls? Both? — Preceding unsigned comment added by 160.32.199.46 (talk) 20:13, 2 July 2019 (UTC)

Agreed, it's a total mess. Not to mention that creating an exhaustive list (which is generally the aim of such endeavours on Wikipedia) of federal voting intentions by province would cause the contents of this page to grow exponentially to impractical proportions, as the vast majority of polls in the main table contain provincial breakdowns, and many other "standalone" provincial polls of federal voting intentions have been released since last election. Currently the list isn't remotely exhaustive, and therefore of little objective value. I suggest removing the section. Cheers, Undermedia (talk) 11:25, 3 July 2019 (UTC)

If it isn't going to be complete then it is going to present a slanted set of numbers, so I agree, it aught to go. - Ahunt (talk) 11:33, 3 July 2019 (UTC)

The purpose of the section is to record the sub-national polls, not the regional breakdowns from national polls. Manitoba, Quebec and Atlantic regions do have regular regional polling, in other places, polling is sporadic, but ramps up as the election approaches - last election there were several Toronto-only federal polls during the campaign. — Preceding unsigned comment added by Mw843 (talk • contribs)

Okay I see your changes there. That is an improvement. - Ahunt (talk) 13:59, 4 July 2019 (UTC)

I am okay with the changes that Mw843 and his rational for which polls to include. Once you open it up further though, it would not belong on this page and would need its own page as the content would be massive. Krazytea(talk) 17:34, 8 July 2019 (UTC)

I proposed a page for regional polls, similar to the page for individual ridings, but the idea was rejected; this section was the compromise. Mw843 (talk) 01:18, 9 July 2019 (UTC)

Latest Ekos and Innovation Research numbers from Hill Times article

Just a housekeeping note: Both the latest Ekos and Innovative Research numbers from the Hill Times have been added to the chart, but the data is incomplete because we only have the article as the source. When they become available, we should link to the proper, full releases.

Williamefwilson (talk) 12:29, 16 July 2019 (UTC)

Agreed. Furthermore, the Innovative numbers given in the article can't be the 'decided voters' results, because LPC + CPC + NDP + Green only add up to 79%, and there's no way the BQ and PPC have ~20% combined. I've therefore removed them from the table pending a release of the full report. Cheers, Undermedia (talk) 13:29, 16 July 2019 (UTC)

It could happen ;) Williamefwilson (talk) 16:34, 16 July 2019 (UTC)

Party colour changes?

Did someone change the party colours, specifically for the Liberals and Conservatives? It's ugly.

Williamefwilson (talk) 18:43, 23 July 2019 (UTC)

Some guy is unilaterally conducting changes at Template:Party shading/Conservative (Canada) and Template:Canadian politics/party colours/Conservative. Impru20^talk 18:48, 23 July 2019 (UTC)

Please note that the user is openly refusing to seek any consensus for the change, which is affecting a wide number of Canada elections-related articles. I can't revert him further without violating WP:3RR, but this should possibly be addressed as this is openly disruptive. Impru20^talk 19:04, 23 July 2019 (UTC)

Tried to centralise discussion at Template:Canadian party colour, though looking at the talk page for the other person involved they seem to have a history with revert wars. I tried! Bkissin (talk) 19:41, 23 July 2019 (UTC)

Polls by provinces

Should MQO Research Atlantic provinces polls be added to the regional polls list? This would mean that there would be both an Altantic Canada section, and a sub-section for each of the four provinces. Here are the four latest polls from spring 2019: New Brunswick Nova Scotia Newfoundland and Labrador Prince Edward Island

Yes, as subsections of the Atlantic Canada subsection. Onetwothreeip (talk) 02:43, 26 July 2019 (UTC)

Criteria for poll inclusion

For possible inclusion criteria, how about something like a minimum 400 sample and a link to published results? We could allow Twitter links where a pollster (like Graves) breaks down his results, but an emphasis should be placed on transparency and reproducibility. Williamefwilson (talk) 18:35, 29 July 2019 (UTC)

Warner Group Orleans poll

I don't think it should be allowed. From their website: "The data from (th)is poll is coming from The Warner Group from one poll posted to several Orléans area Facebook groups, including, but not limited to, Orléans Votes and Orléans Politics". It seems like it has more in common with the online non-representative clickbait surveys you see on news websites than scientific polls published by reputable firms. Also a sample size of 64 is ridiculous. And this "Warner Group" is not even a polling company. — Preceding unsigned comment added by 69.166.121.182 (talk)

Yes, this should not be included. Williamefwilson (talk) 18:29, 29 July 2019 (UTC)

I agree with not including it, seems to be not a WP:RS. - Ahunt (talk) 00:40, 30 July 2019 (UTC)

Nanos

So it looks like Nanos is no longer releasing any polling data publicly. The link for their latest poll leads to a 'subscribers only' page. Should they still be included in the index? — Preceding unsigned comment added by 45.74.75.2 (talk) 15:02, 23 July 2019 (UTC)

Yes, they should still be included. Williamefwilson (talk) 18:42, 23 July 2019 (UTC)

Only if there's a publicly accessible source for the results, that is; e.g. the Polling Canada tweet of the July 19 results. Undermedia (talk) 20:23, 23 July 2019 (UTC)

Yes agreed, as long as there is a public source (including a tweet).

This is a big loss to not have the weekly release public Mikemikem (talk) 01:40, 24 July 2019 (UTC)

Hopefully we can still source those rolling polls. They are very useful. - Ahunt (talk) 02:08, 24 July 2019 (UTC)

Agreed. A constant, steady release is extremely useful Mikemikem (talk) 03:25, 25 July 2019 (UTC)

I don't think you need a publicly accessible source. There's a lot of sourcing in articles to books and such that aren't available easily. You can also link to soruces behind paywall, with the appropriate flags in Template:Cite web such as those listed at Template:Cite web#Subscription or registration required. You need a source. Nfitz (talk) 23:38, 9 August 2019 (UTC)

Although you're correct about the provision allowing paywalled content on the web to be cited on Wikipedia, I have concerns about using it in this particular case. Whereas I imagine this provision is most typically used in cases of a "one-off" citation of a paywalled news article, scientific paper, etc., the Nanos rolling polls are published weekly on an ongoing basis. So for this to work we would need a reliable editor with a subscription to the Nanos tracker to be able to commit to consistently adding the data every week going forward, otherwise I don't think it's acceptable to have the polls added sporadically/opportunistically here and there. I think it needs to be all or nothing. Cheers, Undermedia (talk) 11:08, 10 August 2019 (UTC)

Paywalled sources can be used as they make WP:RS, but reports of paywalled source results by random people in tweets, etc, are not reliable. They make also violate WP:COPYLINK. - Ahunt (talk) 23:50, 10 August 2019 (UTC)

I don't see what's wrong with reporting random tweets, if the source seems reasonable, and is corroborated or partially corroborated. I'm not sure why User:Undermedia is removing Nanos data, when the consensus here appears to be otherwise. If WP:COPYLINK is really an issue, one can still reference the tweet, but not link to it. Nfitz (talk) 04:33, 11 August 2019 (UTC)

I've been one of the primary contributors to the federal election polling pages going back to 2011 and never have we supported the inclusion of polls with such incomplete data (the Reuters article is missing the numbers for 3 of the parties and doesn't include the last date of polling or sample size, so I can't even add it to the graph) or iffy sources (judgement of whether the source of a random tweet "seems reasonable" is completely subjective), so I don't see why we would suddenly start now. Clearly there isn't in fact a consensus on this based on my comments and those of Ahunt (also a very long-time contributor), so I will continue to advocate for removing the poll from the table, though I will refrain from edit warring at this point. Cheers, Undermedia (talk) 11:28, 11 August 2019 (UTC)

I'm not sure how why the data being incomplete (not including all the third parties), or not being able to add to the graph, would preclude it; the information is there for the two primary parties, and meets WP:V. The complete data for all tracked national parties was there, but you objected that some of it came from a tweet (even though the later news article confirmed most of it - do you think the tweet is not reliable for the third parties? I don't see any opinion expressed from User:Ahunt over the media reported Nanos data - only the tweeted data, which was already removed. Nfitz (talk) 17:26, 11 August 2019 (UTC)

Just for the record, I am just concerned that we can trust the source cited as being accurate, as per WP:RS. - Ahunt (talk) 17:36, 11 August 2019 (UTC)

I'm surprised User:Ahunt. What's the concern of trusting Reuters as a RS? They are explicitly listed at WP:NEWSORG alongside BBC News, Interfax, AFP, UPI, etc. Those are some of our most reliable sources. Nfitz (talk) 18:11, 11 August 2019 (UTC)

Reuters is fine as a source, what I am concerned about is quoting random twitter users, who may or may not be quoting polling data accurately. That is why I noted WP:RS, the sources we use here have to be reliable. - Ahunt (talk) 02:55, 12 August 2019 (UTC)

The page should be recording facts, not rumours. If a source (twitter or a secondary news source) is just skimming facts from another source, and not including all parties, and key information about the poll (date, size, MoE), it's just a rumour. On Twitter, @CanadianPolling (Polling Canada) does seem to consistently pass the test for completeness and key information, as do news sources reporting on polls that they have commissioned. Mw843 (talk) 01:47, 12 August 2019 (UTC)

Agreed - though very RS such as Reuters seem fine as secondary sources to me. WP:V and WP:PSTS note a preference for secondary sources over primary sources. Nfitz (talk) 03:47, 12 August 2019 (UTC)

I think things are going too far here. For the longest time we weren’t even including polls that only had a link to twitter. We started doing it for Ekos polls being tweeted out by the owner Frank Graves who gave full data including field dates and sample size.

But frankly including data from random twitter users is ridiculous. This is not a reliable source and can’t be verified.

Also, Ahunt and Undermedia have been long time contributors to Canadian opinion polling pages. Let’s not be so hasty to just override opinions of people who have put their blood sweat and tears in to this for many years.

And I happen to think Nanos is the single best, most reliable pollster we have in Canada. Publishing deal with CTV, live interview telephone polls, expensive and the gold standard. So this isn’t about not including Nanos or anything like that, it’s about not including incomplete polls (needs at least sample size, field dates, and top 2 parties %) or polls from random twitter sources. Mikemikem (talk) 05:29, 12 August 2019 (UTC)

Surely polls that don't list all the third parties, from very reliable secondary sources (such as the Reuters article) shouldn't be excluded. I'm not sure how the history of blood, sweat, and tears is relevant - doesn't than violate WP:OWN - we need to apply policies to decisions - not who has worked on this the most ... or longest. The problem with the most recent Nanos poll, is we have a perfectly reliable source (Reuters) for 3 of the national parties, with the same data as a tweet, that has 2 more national parties. And somehow the result for the regional BQ keep appearing as well, despite not being in either source - which implies that someone is looking behind the paywall (which is also a reliable source, if someone were to reference it ). Nfitz (talk) 05:45, 12 August 2019 (UTC)

It looks like BQ support was added because the twitter user actually posted a screenshot of the Nanos report in a reply to his own tweet, which I would think raises WP:COPYLINK concerns even further. I agree with everyone that it's frustrating to suddenly lose access to these weekly Nanos polls. The amount of time I've been contributing to these pages is indeed irrelevant to this debate; the reason I mentioned it was only to emphasize that over that time I've been through a lot of precedent for debates like this, and this is definitely the messiest attempt to add a poll that I've seen thus far. With the screenshot of copyrighted, paywalled material, I suspect referencing the tweet is now completely off limits. That leaves us with only the Reuters article, which only gives numbers for 3 out of the 6 parties, and doesn't actually state the last date of polling or sample size (even though "we" know what those parameters are). So it still seems awfully iffy to me. Cheers, Undermedia (talk) 10:45, 12 August 2019 (UTC)

I hadn't seen the screenshots. Referencing the tweet might be problematic (ironically, I have the same email from Nanos, but my subscription is ASCII, and it is blanked). But the answer is simply referencing the paywalled Nanos site instead. There's no doubt the numbers are correct, and WP:NOTCENSORED. Nfitz (talk) 15:08, 12 August 2019 (UTC)

What's with Opinion polling for the 43rd Canadian federal election by constituency

I was about to just redirect it to Opinion polling for the 2019 Canadian federal election#Constituency polls - but then I noticed in addition to a couple of polls that are there, but not here - that the numbers are quite different for some of the polls there, that are here? I haven't dug into the details though. Nfitz (talk) 23:35, 9 August 2019 (UTC)

I see that the article "Opinion polling for the 43rd Canadian federal election by constituency" was published on 10 June 2019, before someone began to add constituency polls in "Opinion polling for the 2019 Canadian federal election" on 24 July 2019. Since there's an article for the polling by constituency, I suppose those polls should go there. Regarding the differences in numbers, I looked only at the May 2019 poll for Markham-Stouffville: the numbers in "Opinion polling for the 43rd Canadian federal election by constituency" are the numbers of decided and leaners after repartition of the undecided respondents, while the numbers in "Opinion polling for the 2019 Canadian federal election#Constituency polls" are the numbers for decided respondents, before leaners and before repartition of the undecided respondents, rounded, and with inversion of the NDP and Green numbers. -- Asclepias (talk) 03:15, 10 August 2019 (UTC)

I think we could move those polls over to this article as a temporary solution until it is clear whether or not this should be a seperate article. Username6892 (talk) 16:31, 10 August 2019 (UTC)

It wasn't until I checked the talk page after being away for a few weeks that I realized that Constituency Poll section had been added without any discussion. I think the information should be in the separate article, as in 2015. It's quite likely that by the end of the campaign, over 100 ridings will have published polls, which will make this page unnecessarily unwieldy. Also, I think one of the important uses of this page is to record results over time, and most of the constituency polls are one-offs, never to be repeated. (Note that this is not the case for the regional polls, where the historical data has been lost, as opposed to never existing ... and in the case of the 2015 Election, lost in their entirety.) Mw843 (talk) 13:19, 11 August 2019 (UTC)

All moved now (thanks). But back to the "which result" question. Looking at just one result (May 30 for Vancouver Granville), this article used the results of the poll, when they listed the local candidates, and the Opinion polling for the 2019 Canadian federal election, by constituency lists the results when they asked by just party name. So which should be used? I can't see a 2015 discussion. The results shift by up to 4%. Nfitz (talk) 05:50, 12 August 2019 (UTC)

I hope you're not really considering using the second question of the poll with the fake "candidates". This second question with the names of non-candidates was more an experiment by the pollster, but it's hardly useful alone. If something from that poll could be used at all, it may be the first question. Even then, a question mixing some options without names and one option with a name is disputable. -- Asclepias (talk) 11:50, 12 August 2019 (UTC)

I was considering asking questions to find out more information. I'm not familiar with the riding - I assumed that they were the candidates. There were other differences between the pages ... I haven't dug any deeper. Nfitz (talk) 06:35, 13 August 2019 (UTC)

Nanos Poll Results

I've noticed that the Nanos Poll results are no longer being posted or, I presume, being factored into the results you develop. Is there a reason for this?Kenhardie (talk) 13:22, 14 August 2019 (UTC)

See the Nanos section further up. No reason not to add, if there's a comprehensive reliable source, or one has access to the paywalled data - other than that, there's not clear consensus yet. Nfitz (talk) 13:27, 14 August 2019 (UTC)

Nanos Results

Given that your graph includes the top line results and given that those results are still available from Nanos at no cost, you should resume including their weekly numbers. Those who want to do a deeper dive into Nanos' data can pay the $4 per month to do so. Not including Nanos top lines significantly impairs the value of your chart. Kenhardie (talk) 12:07, 18 August 2019 (UTC)

• Where have you seen the top line results for voting intention available for free User:Kenhardie? The topline results I've seen at the Nanos website are for preferred Prime Minister and a "Power Index", neither of which are tracked on this page. Nfitz (talk) 20:53, 18 August 2019 (UTC)

Ah, I see that Nanos has changed something...when they first put up their paywall, they showed you the main page with the overall results, but I couldn't access the cross-tabs until I paid the $48. So, a couple of options...you can pony up the $48 to get the results, I can pay it for you if you're strapped or I can relay the top line results. What do you think? Kenhardie (talk) 13:24, 19 August 2019 (UTC)

If I'm not mistaken, if you have access to the paywalled data you can add it. JonathanScotty (talk) 22:05, 26 August 2019 (UTC)

EKOS 'Twitter' Polls - Should exclude Tweet-only results like Nanos

Should we really be posting the "EKOS" poll results that are only shared via Frank Graves' Twitter feed on this page? I get that he is the president of EKOS, but his tweets do not include the full data set, tables, methodologies, etc. or even links to this information. If we're not going to post the Nanos results by linking to data shared on Twitter, I think we should apply this standard across the board and exclude poll info that is only available on Twitter posts. Without full poll breakdowns we have no way to verify methodology, possible sampling issues, etc. Some of these polls have tiny sample sizes and huge MOEs, so there are concerns about these subsets of data when we cannot see the full table breakdowns. Even the CBC Poll Tracker excludes these results. 338 only includes ones where the full data sets are provided to P.J. Fournier. - Matticus333 (talk) 14:01, 30 August 2019 (UTC)

I'm inclined to agree that tweets are bit too iffy/informal to source polls from, full stop, but there appear to be strong feelings on either side of this debate. The consensus right now seems to be that if a tweet is from an "authoritative" source (e.g. Frank Graves, 338Canada), then it's acceptable. Cheers, Undermedia (talk) 12:19, 1 September 2019 (UTC)

I tend to agree: if the numbers in a tweet are complete and it comes from a WP:RS person then we should go ahead and use it. You may want to ensure the tweet is archived on archive.org however... - Ahunt (talk) 13:33, 1 September 2019 (UTC)

Adding debates to chart

Should the date of debates be added to the chart? Or is that too much clutter do we think? Mikemikem (talk) 13:07, 13 September 2019 (UTC)

Probably too much clutter and given the limited participation and the large number of debates may not have much polling-discernible impact anyway. - Ahunt (talk) 13:13, 13 September 2019 (UTC)

Polls that do not give results with the undecideds filtered out - any way to incorporate them?

https://ipolitics.ca/2019/09/17/liberals-tories-deadlocked-as-first-week-of-campaign-comes-to-a-close-mainstreet-poll/

Talking specifically about this poll, but also in general Mikemikem (talk) 23:03, 17 September 2019 (UTC)

Looks like someone added the poll...but it looks odd as the numbers are so much lower with the undecideds Mikemikem (talk) 00:00, 18 September 2019 (UTC)

Looks like someone added the poll...but it looks odd as the numbers are so much lower with the undecideds Mikemikem (talk) 00:00, 18 September 2019 (UTC)

This could be problematic to include polls that includes and excludes undecided voters. This is mixing orange and apple juice and calling it apple juice. Simply look at the undecided % that is higher than both the NDP and Greens. Plus, the % for the major parties in the rows only add up to nearly 90%. This is misleading when other rows adds to nearly 100%.

One potential way to accommodate both decided and undecided % is to include an additional column for undecided, but I’m not sure how it would look on the graph. Aquitoba (talk) 03:21, 18 September 2019 (UTC)

Agreed with Aquitoba. It's comparing apples to oranges, and will also have the effect of misleadingly pulling party support downward in the graph. Cheers, Undermedia (talk) 13:47, 18 September 2019 (UTC)

How would showing the percentage of undecided voters pull party support down? The fact that there are more undecided voters doesn't change the percentage of decided voters down. Walter Görlitz (talk) 14:55, 18 September 2019 (UTC)

We're talking specifically about poll results that don't include the usual "final step" of removing undecideds from the mix and subsequently recalculating the percentage support for the parties as if the undecideds didn't exist, so yes they will have unusually low numbers for the parties, typically collectively adding up to ~75–90% (depending on the proportion of undecideds) instead of ~100% (subject to minor rounding errors) as in standard top-line poll results. In any case, in practice it's actually very rare that the "standard" results don't eventually become available (for example, they have since been unearthed for the Mainstreet poll that prompted the creation of this section in the first placed), so it's probably not even worth having a protracted discussion about this. Cheers, Undermedia (talk) 16:22, 18 September 2019 (UTC)

On a separate note, I actually do think that the proportion of undecided voters ought to be given as much emphasis as the standard top-line party support numbers in reporting polls, and it boggles my mind that the media still hasn't learned to do this on a consistent basis (or that pollsters aren't strictly insisting that the media does so to better safeguard their own reputation). Pollsters have taken a lot of flack in recent years for seemingly failing to accurately predict elections, but there's absolutely nothing pollsters can do to confidently predict what might happen when you have upwards of 15–20% of voters who remain undecided until election day, which is what tends to happen in elections where none of the options are particularly well-liked by the electorate, and those are precisely the elections that tend to get quoted as "embarrassing" misses for the pollsters (think Alberta 2012, BC 2013, and US presidential 2016). Aside from that, differential turnout at the ballot box among supporters of different parties is the other wild card that's very difficult for pollsters to deal with. All that to say that I would entertain creating an additional column in the table for reporting the percentage of undecided voters whenever available. Cheers, Undermedia (talk) 16:43, 18 September 2019 (UTC)

I would very much support adding a column for "undecided", as it shows a certain measure of uncertainty. Many undecided voters eventually don't vote, but many do and make up their minds on election day, often in the polling booth, so that measure shows a degree of volatility. - Ahunt (talk) 16:49, 18 September 2019 (UTC)

I agree undecided provides an important context. I believe a lot of pollsters don’t actually report that data but a lot do.

And it provides important context for volatility

Also, please check back on the “Campaign period polls and graph” thread Mikemikem (talk) 23:58, 19 September 2019 (UTC)

Mainstreet/Ipolitics

It seems that the Ipolitics articles with Mainstreet topline numbers are only accessible for about a day before going behind a paywall. Should they not be removed/not included in the future? — Preceding unsigned comment added by 168.91.56.55 (talk) 15:17, 19 September 2019 (UTC)

Paywalls are not forbidden, so no. Walter Görlitz (talk) 15:18, 19 September 2019 (UTC)

Both the Sept 17 and Sept 18 searches are still available, at least on mobile... Mikemikem (talk) 18:21, 19 September 2019 (UTC)

Make sure they get saved on archive.org while they are available! - Ahunt (talk) 02:07, 20 September 2019 (UTC)

Is there any reason to think they will disappear?

They are all still functioning Mikemikem (talk) 00:34, 21 September 2019 (UTC)

Most things on the Internet disappear at some point. Sometimes the relocate on the same server and the organizers to not take care to install redirects to the new location, while other times—and this is particularly the case with news—it simply ages-out and is no longer accessible on the site so it can only be seen in archives. Not sure about these two sources, but that's the case with the websites for CBC, CTV, Globe & Mail and National Post. Walter Görlitz (talk) 01:02, 21 September 2019 (UTC)

Mistakes regarding change in %

I simply included the numbers that were reported for the poll. But it seems that the change in % in the new Mainstreet/iPolitics article has some math mistakes. JonathanScotty (talk) 17:12, 22 September 2019 (UTC)

I noticed the same thing in one or two of the previous daily updates as well. I figure it's likely just sloppy reporting. The best we can do is just go with the numbers given in the article associated with each poll. Cheers, Undermedia (talk) 17:27, 22 September 2019 (UTC)

Campaign-period polls & graph

With the writ apparently dropping any day now, I wanted to suggest that we simply add a "special event" row (same as for new party leaders) to the national polls table indicating the cut-off between pre-campaign- and campaign-period polls. This will make it straightforward for me to create a new version of the graph zoomed in on the campaign period, with the trendlines simply carrying over from the current graph. Creating a whole new table would make it more complicated to achieve this continuity between the graphs, which is why I'm suggesting we simply go with a separator row. Cheers, Undermedia (talk) 14:14, 3 September 2019 (UTC)

Good idea. I think it helps readers anyway. - Ahunt (talk) 14:32, 3 September 2019 (UTC)

Yes agreed Mikemikem (talk) 00:29, 11 September 2019 (UTC)

I see a separate graph has been created.

Is it not too much work to have it updated? Mikemikem (talk) 17:05, 12 September 2019 (UTC)

Ya I would create a new section not only as this is what we have historically done but it creates a clean separation. On the other hand I am not sure what kind of work would go into the graphs, etc. Otherwise seems like a clean break to me. Krazytea(talk) 18:37, 12 September 2019 (UTC)

I'm afraid splitting the table in two would mess up the campaign-period graph. Not sure I'd be able to make it work properly that way. Cheers, Undermedia (talk) 19:00, 12 September 2019 (UTC)

I just wanted to check in on this again. How did we manage to have a separate chart and graph in both 2011 and 2015?

Looking through those 2 years and now I actually do think it looks better with a clean break.

Also, it gives the ability to toggle the polls ONLY during the campaign by pollster, party support, etc.

Without a separate chart and graph there is no way to easily see a parties high and low during the campaign as all the polls are mixed together.

Could we not just do some copying and pasting from 2011 and 2015 where we have 2 charts and 2 graphs and it works fine? Mikemikem (talk) 23:53, 19 September 2019 (UTC)

It's because both in 2011 and 2015 the pre-campaign graph and campaign graph were completely different in nature. The pre-campaign graphs were very "rudimentary" graphs made in Excel (Ahunt made the one for 2011 and I made the one for 2015), while the campaign graphs are the more sophisticated type we're using now with the smooth local regression trendlines, which is made using a programming script in R that automatically reads the polling data directly from the table in Wikipedia, which is very efficient and convenient. Back in 2011 and 2015 those were made by another user, who subsequently passed the torch on to me, so to speak, and for the first time I created and maintained one spanning the entire pre-campaign period. A key feature is the ability to "zoom in" on a small portion of the graph (the campaign graph is really nothing more than a zoomed-in version of the pre-campaign graph), and that way the trendlines nicely carry over seamlessly from the pre-campaign graph to the campaign graph. Currently, the only way I know how to do this, script-wise, is by keeping all the polling data in the same table; i.e. if the campaign polls were moved to a separate table, it would basically break the graph (in a nutshell, the number of polls in the campaign period alone is too small to statistically compute the smooth trendlines). There's probably a way to adjust the script to have it read data from two separate tables, but unfortunately I must confess I'm not actually a coding wizard (I only know the basics) and I don't currently know how to do that. That being said, your comments about the practicality of having two separate tables are valid. I can try to make time, maybe this weekend, to see if I can figure something out by playing around with some code, but no guarantees. Cheers, Undermedia (talk) 23:55, 20 September 2019 (UTC)

Yikes. Sounds very complicated.

Alright well I appreciate you are willing to try.

Is there any way to have just one chart but create a “break” when the campaign starts so that when you toggle by party support it would only include the campaign polls above the break, and then include the polls below the break seperately?

The issue is now if you toggle by party support it puts all the “party leader elected” notes and the “campaign starts” note at the top, and then there is nothing to separate pre-campaign and the campaign polls.

I’m not opposed to everything being in one chart, the issue is just being able to toggle by party support or party lead and have them separate.

If there’s no way to possibly do that in one chart, and if you are unable to make it work with two, no worries but I think it would be ideal to have some type of break between the two Mikemikem (talk) 00:09, 21 September 2019 (UTC)

I just ran the graph script, without any changes to the code, using the now-separate table for campaign-period polls, and uploaded the result (you may need to clear your browser cache for the graph to update). The good news is there now seem to be enough campaign-period polls to compute all the party trendlines, so the graph isn't broken. The drawback is we now lose the continuity between the two graphs; the campaign-period graph is now only using the data from the campaign-period poll table and not taking into account the polls conducted in the days/weeks leading up to the writ. Although that's not ideal, I'm actually kind of liking how the trendlines now look more responsive to the apparent shifts we've been seeing since the start of the campaign, whereas I was finding them quite "stiff" in the previous version of the graph. I'm not exactly sure why this has happened, but like I said before, I'm actually not a coding expert! Cheers, Undermedia (talk) 18:07, 22 September 2019 (UTC)

I am glad to hear that it looks like it will work out.

With two daily tracking polls, the campaign period ends up having an incredible amount of data very quickly Mikemikem (talk) 18:47, 22 September 2019 (UTC)

Leger Over-sample

I earlier argued that the Regional Poll section was for stand-alone polls, not regional breakouts of national polls. Now Leger has run a poll where they have over-sampled Quebec by roughly a factor of 4: The total sample was 2,153, but the Quebec subset was 1,028. Effectively, Leger has run two polls as the same time. Should this poll be reported as both National and Regional? Mw843 (talk) 13:32, 25 September 2019 (UTC)

I would say no. Pollsters do sometimes over sample one project or region. And Leger has done this before specifically regarding Quebec Mikemikem (talk) 00:29, 26 September 2019 (UTC)

905 polls/GTA polls

There have been three 905 polls this election (Mainstreet, Nanos & Ipsos), but the only one that has defined 905 was Mainstreet. So, I was wondering if I should add them, since we're not sure if they're even polling the same area? And if I do add them, do I put them in the same section (a 905 section)? JonathanScotty (talk) 17:56, 28 September 2019 (UTC)

So, I was wondering if I could get an answer? JonathanScotty (talk) 14:08, 1 October 2019 (UTC)

I would say no because “905” means different things to different pollsters. Sub-regions within provinces are tough because they aren’t clearly defined like w province is Mikemikem (talk) 21:51, 6 October 2019 (UTC)

No Mainstreet topline during the weekends (except the final one)

I just wanted to mention that according to Maggi & the Mainstreet twitter account, there won't be public toplines during the weekend except the final weekend due to the election being on Monday. JonathanScotty (talk) 17:36, 29 September 2019 (UTC)

Edited:

I found the one for the weekend of October 5th. (Polling October 3-5) for Mainstreet. I've linked the polling data that has been floating around and the info is not only accurate but also reflected in the 338 poll here: http://338canada.com/polls.htm — Preceding unsigned comment added by Angryaudifanatic (talk • contribs) 00:49, 7 October 2019 (UTC)

That's not a unique, permanent link to the poll; it will eventually no longer be found at http://338canada.com/polls.htm as the table on that third-party page gets filled with new polls. And the cell phone screenshot of copyrighted, paywalled content is also not acceptable. Please stop edit warring. Undermedia (talk) 01:24, 7 October 2019 (UTC)

Ok I disagree but will leave it to the other editors to decide. Thx — Preceding unsigned comment added by Angryaudifanatic (talk • contribs) 01:34, 7 October 2019 (UTC)

There was an error on my part regarding the scribes screenshots, I apologize for that.

http://338canada.com/polls.htm

However, why is 338Canada not an appropriate link to a poll? It’s public, it contains all the data necessary to fill out the table, and the polls there date back to early 2018. That link will exist far beyond many dead poll links from pollster websites.

Is the issue that we don’t want to rely on public websites like 338Canada if they got the info from behind a paywall? That seems...pretty ludicrous. Once the data is in the public domain I would think it’s appropriate to post here is it not? Mikemikem (talk) 01:38, 7 October 2019 (UTC)

Yes, we've used 338Canada as a source at least once before, but in that case it was a unique "permanent" page dedicated to the poll, whereas their general poll table page is inherently ephemeral and could even get wiped as soon as a few months from now once polling gets underway again after the election. In any case, I'll say again since this is a 3-day rolling poll we won't technically be missing any data as a result of not having the daily updates on weekends, so the fervour to find a roundabout way of adding this update isn't even really justified. Notice that even 338 only lists every 3rd Mainstreet update. Cheers, Undermedia (talk) 02:03, 7 October 2019 (UTC)

For the record Mike that is what tipped me to come back to edit/add the post above and beyond your support. If Wikipedia (yes I'm new) won't give me information that I can find publicly on other sites like 338 then this page is worthless to me. Because the 338 page is publicly sharing the Mainstreet poll, that is the basis why I strongly feel it should be added here. //angryaudifanatic

I agree with that, but I await the argument from the other side before putting the poll back in.

But AFAIK, we have never said we can’t put in publicly a available data contains all of the necessary info to fill out the chart. 338Canada is certainly a trusted, public website. Mikemikem (talk) 02:02, 7 October 2019 (UTC)

Please see my reply to your previous comment above. And I would kindly suggest everyone use colons (:) to indent their replies on the Talk page as per the usual convention, so that the discussion threads are tidier and easier to follow. Cheers, Undermedia (talk) 02:09, 7 October 2019 (UTC)

Re: Undermedia From: angryaudifanatic Body: The reason that I am trying to get the Mainstreet information into the Wikipedia page is not just that it's available on 338 and fits the criteria for it to be here, but there is absolutely significant relevant information that it presents here that is publicly found on 338 as well. Both Nanos polls this weekend (Oct 5/6) indicate that the Liberals are gaining steam (outside of MOE and consistent), and the Mainstreet polls indicate otherwise. If the information for MS was not something that could be gleaned elsewhere, then IMHO "whatever", but in this case, this is important pertinent information for a poll.

By leaving the MS polls out and leaving the Nanos polls in (both weighted over a 3 day average), that strikes me as being politically biased.

The 338 polling info is public, and is captured on Wayback archive. There is 0 reason that I think it should be left out.

//end message — Preceding unsigned comment added by Angryaudifanatic (talk • contribs) 02:13, 7 October 2019 (UTC)

There’s really no reason to get so agitated over a couple of days of missing updates for a 3-day rolling poll, because today’s update will include all of the data from those days. Like we did last weekend, we can compensate for the missing daily updates by adjusting the fraction by which the sample size of today’s update is multiplied (in this case we can simply remove the “1/3”), and that will effectively triple the weight of the poll in the graph. Having daily updates at 1/3 weight or every third update at full weight works out to the same (note that 338Canada only lists every 3rd update for this same reason; but again, that page is ephemeral and doesn’t constitute a unique, permanent source for specific polls). Cheers, Undermedia (talk) 12:07, 7 October 2019 (UTC)

Not about being 'agitated' as you insinuate. It's about capturing information on Wiki appropriately. Personally I feel strongly about this. Maybe time to enter dispute resolution if your opinion remains steadfast? If according to you it doesn't matter, why not include it since it doesn't matter anyways to you? /angryaudifanatic

I would say this is not worth going to dispute resolution over...we are about to get 5 straight public releases of Mainstreet rolling polls Mikemikem (talk) 14:13, 7 October 2019 (UTC)

Toronto Is A Region

The city of Toronto constitutes a "region" for the purposes of this page; it contains multiple (25) constituencies, and has well-defined, widely agreed upon, boundaries. Mw843 (talk) 13:34, 13 October 2019 (UTC)

Agreed. Cheers, Undermedia (talk) 16:23, 13 October 2019 (UTC)