Jump to content

Talk:Opinion polling for the 2015 Canadian federal election

Page contents not supported in other languages.
From Wikipedia, the free encyclopedia

Leadership polls

[edit]

Started this page now because I thought it was relevant to add issues like leadership polls which would only clutter the original page. Krazytea(talk) 00:31, 26 April 2013 (UTC)[reply]

We are missing a "pre campaign" period poll by Legar that came out February 2 — Preceding unsigned comment added by 174.4.54.249 (talk) 20:06, 6 February 2015 (UTC)[reply]

It is there now. - Ahunt (talk) 12:47, 7 February 2015 (UTC)[reply]

Approval polls

[edit]

Do we want to add approval polls? Maybe in a collapsable form or something? There are definitely more approval poll ratings than Best PM polls. However on the other hand it would probably exactly equal the amount of voting intention polls which could great an extremely long page. Thoughts? Krazytea(talk) 16:27, 25 May 2013 (UTC)[reply]

My main concern is that the methodologies differ so much on these types of polls (many polling companies don't even do them) that the level of comparability would be even lower than the "party preference" polls. That said I am not opposed to it, if you want to go ahead and put a table together. - Ahunt (talk) 16:55, 25 May 2013 (UTC)[reply]

I think that approval ratings should be included, as "leadership polls" (meaning preferred PM) ask a very specific question, and really none of the three main candidates have had much of a convincing lead on that question. Approval ratings, such as the ones conducted by Nanos, do give a better impression of how Canadians view the various leaders. Putting it together would be time consuming though. — Preceding unsigned comment added by 58.65.134.39 (talk) 10:37, 21 May 2015 (UTC)[reply]

Undecided voters

[edit]

In Abacus Data's latest poll, they've dropped the whole "decided voters" numbers that we've been using (due to "the volatile nature of the electorate and the importance of understanding the motivations and size of undecided voters in Canada", or in short, their thoughts on how decided people were caused them to drop the ball on the BC elections) and now report only numbers including another number for undecided voters. Should we be considering changing the format to add an "undecided" column or do we stop using their numbers and just continue with the other polling firms? Grandmartin11 (talk) 05:15, 26 June 2013 (UTC)[reply]

My guess is that we have to work with what we have, as it is likely that other polling companies will follow that lead. I think we just have to note the change somehow in the table, or at least in a footnote. There is some interesting background on this decision by David Coletto, CEO of Abacus Data in this article. - Ahunt (talk) 10:43, 26 June 2013 (UTC)[reply]
Due to popular demand, Coletto has now released the "conventional" results (decided + leaners), which I've added to the table. My understanding is that he'll continue to release them in the future as well as continue to emphasize the proportion of undecided voters, as I imagine other pollsters will too. I had noticed the last few CBC/Nanos polls made special mention of undecided voters, so it looks like the trend is starting to set in. Frankly, I'm surprised it took this long; you'd have thought pollsters would've immediately learned following the Alberta election. -Undermedia (talk) 14:26, 26 June 2013 (UTC)[reply]
Okay, well whatever you put in the table I will add to the graph. It is worth noting that the less comparable the results between polling companies, though, the more the graph will jump around. - Ahunt (talk) 15:14, 26 June 2013 (UTC)[reply]
Just to be clear, those Abacus numbers are indeed "standard" results (i.e. leaners factored in, undecideds factored out, numbers add up to ~100%), and thus technically comparable to the other polling companies. Aside from real shifts in support and margins of error, much of the jumping around appears to result from certain seemingly inherent party "biases" that vary among the pollsters, presumably due to methodological differences. Several of these become apparent if you sort the table by support for each party and note which pollsters tend to be more towards the top vs. the bottom. Abacus for example seems to have a general tendency to find below-average Liberal support. Nanos, and recently Forum, conversely tend to find above-average Liberal support. -Undermedia (talk) 16:44, 26 June 2013 (UTC)[reply]
You make some good points there, I have noticed the same thing. I lot of it hinges on the questions asked and the order in which they are asked. - Ahunt (talk) 17:34, 26 June 2013 (UTC)[reply]
The latest Mainstreet poll numbers are for decided voters only. The Forum poll that was in the field only a day earlier, and which reported widely different numbers, included both leaning undecided and decided voters. Is there a way to signal this discrepancy? It looks like Mainstreeet includes leaning undecided voters in their "Momentum Tracker" scores, but this also includes a weighted adjustment for second choice, so once again it's not comparable to the other polls listed here. Walkinxyz (talk) 18:18, 26 July 2015 (UTC)[reply]
I'm afraid this is probably an inconsistency we simply can't overcome, and the best we can do is just go with the pollster's published results. I believe Mainstreet isn't the only pollster that doesn't follow up the initial ballot question with a "leaning" question; Abacus Data and Insights West don't either if memory serves well. In any case, the leaning question rarely makes much of a difference at all: consider that according to Mainstreet's data, among the 20% of respondents that initially reported being undecided, 60% of them were still undecided upon being asked the leaning question, and according to my quick calculations factoring in the leaners would've merely resulted in the NDP and Liberals picking up one point each and the Conservatives dropping two. Cheers, Undermedia (talk) 15:19, 28 July 2015 (UTC)[reply]

Election poll chart

[edit]

The chart is misleading as it suggests that each poll is directly comparable with each other. Only polls from the same company can be compared due to house effects. It would be better as a scatterplot with a moving average line. Charmed88 (talk) 14:15, 17 July 2013 (UTC)[reply]

I take it you are referring to the graph? The points you make are valid in that regard, but that sort of graph is much more complex to create and produces a much more difficult to interpret graph. This was done in this graph File:ElectionPollingGraphCanada2011.png for the 2011 election period and you can see it probably creates more problems than it solves. - Ahunt (talk) 14:29, 17 July 2013 (UTC)[reply]
The polling graph is a useful way to visualize the evolution of voting intentions over time, and as such I've added some graphs of my own (inspired by Ahunt's) to a few provincial election wiki pages. I've also struggled with figuring out the best way to represent the data, and it's true that simply directly linking the successive data points with a continuous line leads to a fairly jagged representation and a somewhat false impression that voting intentions are constantly jumping all over the place. Take for example my current graph of Ontario voting intentions since the last election, which is modeled after Ahunt's federal graph:
Evolution of voting intentions since the 2011 election
Now consider this alternative representation I just whipped up, which I believe is the sort of thing Charmed88 was suggesting:
Alternative graph
The individual polls are represented by simple points (scatterplot) to which I've added a 3-period moving average trendline for each party. I think a 3-period moving average is appropriate because 3 data points is the minimum required to statistically establish a trend, but anything more may not properly capture the occasional rapid swings in public opinion. The advantage compared to the original graph is that the line is far less jagged and better represents the overall trends, while the individual polls are still represented as points. Thoughts? Undermedia (talk) 16:31, 18 July 2013 (UTC)[reply]
I like the look of this representation. It's certainly less chaotic than the one we've been using. Grandmartin11 (talk) 20:07, 18 July 2013 (UTC)[reply]
Personally I don't mind if we switch to this format from the graph I have been doing as long as you can commit to maintaining it until the next federal election. I guess I could keep the existing one going as a "just-in-case-back-up". The only concern I have is the statistical complexity involved in the new graph, but as long as you explain what the graph is really showing readers, perhaps in the Commons' page description, then that would address my concerns in that regard. - Ahunt (talk) 20:14, 18 July 2013 (UTC)[reply]
Here's what the federal graph would look like using the same format:
Evolution of voting intentions since the 41st Canadian federal election of May 2, 2011. Points represent results of individual polls. Trend lines represent three-poll moving averages.
I've also tried it with polynomial trend lines, which are perfectly smooth, but even using the highest degree Excel allows (6), they still aren't responsive enough to those rapid swings (too much lag). Let's wait to hear back from Charmed88 and maybe a couple of others before deciding. Is the description adequately explanatory? Undermedia (talk) 21:14, 18 July 2013 (UTC)[reply]

I find the shadows on the scatterplot points a bit distractive. Can they be removed? 117Avenue (talk) 22:20, 21 July 2013 (UTC)[reply]

Good call. See updated graph with shadows removed. Undermedia (talk) 23:24, 21 July 2013 (UTC)[reply]
Does anyone else care to comment on this? Looks like we've got a few people who seem favourable to the new graph and no one explicitly against it, though we haven't heard back from the original commenter. If we decide to go with the new graph, I can commit to keeping it updated; I already keep on top of the polls as it is. Undermedia (talk) 12:44, 26 July 2013 (UTC)[reply]
I think you have support for that here, so I suggest that you go ahead with it. For my part I will keep my graph up to date as a back-up, in case there are problems with the other graph or you have to move to the Ross Ice Shelf unexpectedly or similar . It can't hurt to have a back up regardless. - Ahunt (talk) 12:51, 26 July 2013 (UTC)[reply]

What about extending the x-axis to October 2015 rather than just 3 months out? - Thewatcher2015 (talk) 17:23, 14 May 2015 (UTC)[reply]

That would just add more empty space to the graph. - Ahunt (talk) 17:29, 14 May 2015 (UTC)[reply]
Ahunt is correct. I was planning to maintain the 3-month buffer until it runs up against the anticipated election date of 19 October (aka in just a couple of months from now) and then stop extending it. Once the writ is dropped, what's been done for several previous provincial and federal elections is that the current graph stops being updated and becomes the "pre-writ" graph, and a new graph is created spanning only the election campaign, since polling tends to become very fast and furious during campaigns. Sound good? Undermedia (talk) 18:24, 14 May 2015 (UTC)[reply]

I figured out how to automatically scrape data from wikipedia to produce charts, so I tested it out on the inter-election period info. I haven't tidied it up yet but it's cool that it worked! galneweinhaw (talk) 08:29, 7 August 2015 (UTC)[reply]

I like the example provided but the current graph on the main page doesn't look nearly as good, simply because the parties are so close together and all the error bars are grey, it is hard to see what is going on. maybe somebody could try color coding the error bars also?23.91.145.254 (talk) 02:15, 28 August 2015 (UTC)[reply]

Polling methods, margins of error and sample sizes

[edit]

Now that there's a separate page dedicated to polling, perhaps we can discuss possible new ways to address some recurring debates. As we previously discussed, it's unwarranted to disparage online polls beyond the statement to the effect that they technically can't have a margin of error. This recent article again highlights how online polls are becoming the norm and have a very respectable track record compared to telephone and IVR polls. Also, the methodological boundaries are becoming blurred. In Nanos' recent polls, it's stated that they recruited participants "randomly" by phone and then asked them to go complete an online survey. This online poll by EKOS claims to have used a unique probability-based method which supports margin of error estimates. And there are a handful of other examples of so-called hybrid telephone/online polls (Ipsos has done these). What I might suggest is that we add a "Method" column similar to the BC election page so that's it's easy to tell what's what.

Regarding margin of error, I've actually previously argued for the all-out removal of that column on the basis that MOE reporting is inconsistent among pollsters (hence the note). Some pollsters only report the MOE for the total sample size while others report it for the reduced sample of decided/leaning voters, leading to the false impression that the latter are less accurate than the former. Forum seems to round their MOE to the nearest whole number. And then there's the whole debate about whether online polls should have a MOE at all. Finally, Bern99 makes a good point about how the practice of most pollsters of rounding party results to the nearest whole number makes it impossible to know the precise MOE interval. Although the MOE seems like a useful and convenient metric to display, I would argue that it ends up giving an erroneous impression that polls with lower MOEs are inherently more accurate. Take for example the final two polls conducted during the 2011 federal election, both with relatively small MOEs of 1.6pp and 1.8pp respectively. Yet both of them were outside the MOE compared to the actual results of several parties and performed altogether worse than several other pollsters whose final polls had larger MOEs. Given these issues, but recognizing the value of providing some idea of the size of polls, I'll suggest once again that we replace the MOE column with a simple "Total sample size" column, also similar to the BC election page. Total sample size is consistently reported for just about every poll. Undermedia (talk) 13:39, 31 July 2013 (UTC)[reply]

You make some good arguments there. I would support adding "sample size" and perhaps replacing MOE with that after a trial period showing both. - Ahunt (talk) 15:17, 31 July 2013 (UTC)[reply]

changing the title of this thing

[edit]

It should read "opinion polling for the Canadian Federal Election, 2015. capriciously repealing the current elections law and delaying it until next year or later is out of the question for all involved.Ericl (talk) 14:00, 12 May 2015 (UTC)[reply]

We agreed here to wait until the writ is dropped to change the name of this page. I don't see anything you have presented here as being a reason to change that. - Ahunt (talk) 17:47, 12 May 2015 (UTC)[reply]
That's funny. Everyone in the Government has said that the election will come no later than this fall. Lookie here: The Calender clearly states that the Commons get next week off, then there's the summer recess on June 24th. This lasts two months. There's only a couple of weeks after that before the law says that Parliament is dissolved and the writs are dropped. With all that money already spent on advertising, the summer organizing back in the ridings and all, the Tories would just say we're going to change the law and cancel the election this year and hold them sometime in the future? Besides, do you really think cancelling the election, which needs a change in law, wouldn't have leaked out by now? Do you really think the NDP or the Grits or both would STAND for something like that?Ericl (talk) 00:03, 14 May 2015 (UTC)[reply]
You know what? It's not funny. You may have interesting points to make on some other forum but please see WP:NOT HERE. We're not here to chat about stuff or use article space for edits like the one you added. Sorry. Please also see WP:OR and WP:TONE. I've reverted your edit. thanks, Shawn in Montreal (talk) 00:12, 14 May 2015 (UTC)[reply]
The current article title is still accurate. You haven't presented any convincing reason to change it before the writ is dropped. This is solution in search of a problem that doesn't exist. - Ahunt (talk) 00:31, 14 May 2015 (UTC)[reply]
No it is NOT. The year is important. The use of the year means we know what year it was or will be. The use of "the next" or the "42nd federal election" gives the false impression that the year of the next is uncertain. It is not. The only way, as I have said ad nauseum, is through some calamity you would find offensive if I mentioned it. While the title was fine a year ago, it's ugly and misleading so late in the game. The reason the year is not listed on the title is based on a false assumption.Ericl (talk) 00:43, 14 May 2015 (UTC)[reply]
The main article 42nd Canadian federal election begins that it "is scheduled to occur on October 19, 2015."[1] (I won't use reflist here). If that's all you'd like added, I have no problem with that. It's in the main article. Shawn in Montreal (talk) 00:50, 14 May 2015 (UTC)[reply]
Governments are elected for five year terms so there's nothing stopping the Conservative government from holding the election in May 2016. The chances of it not happening this October are very rare, but it's still a possibility. Anything could happen that could see the election postponed till 2016. Newfoundlander&Labradorian (talk) 02:22, 14 May 2015 (UTC)[reply]
There is. it's called the Fixed date election act. In order to postpone the election, it would have to be repealed. In ordedr for it to be repealed, it would have to have a first reading sometime in the next two weeks, what with the time off next week and for two months in the summer. There is no way that can physically happen.Ericl (talk) 21:33, 14 May 2015 (UTC)[reply]
Since this law was passed no federal election has been held on the date required. The current law has no penalties for ignoring it and the election is not constitutionally required until 2016. You still haven't put forward any convincing arguments as to why we need to change the name of this article, from the correct current title, at this point in time. I think you need to explain why it is so important to change the name now instead of waiting until the writ is dropped. - Ahunt (talk) 10:17, 15 May 2015 (UTC)[reply]
Esthetics, mostly. It makes the articles look bad. If every single reference on the subject in the Canadian press says that something is going to happen by a certain date, it's going to happen by a certain date. It could happen earlier, but it can't happen later unless (yes I know that the zombie invasion reference is somewhat insulting, but that's the level it would take to get it to '16 or later) something earthshattering happens. Otherwise it looks like Wikipedians are taking the un-encyclopedic position that the Prime Minister is a fascist liar who's going to end elections in Canada forever. The consensus in the real world of Politics is that Canada will have an election in 2015. July or October it doesn't matter. The year is the only thing in the title.Ericl (talk) 20:07, 16 May 2015 (UTC)[reply]
I don't see anyone here taking any position remotely like that, so please stop insulting other editors here. You aren't making any kind of case as to why this needs changing now, why the big hurry and why this can't wait until the election is called. The current title remains factually correct. - Ahunt (talk) 20:32, 16 May 2015 (UTC)[reply]
  • Of course. And main articles matter (that's putting in mildly). Not a chance of renaming this without the main article changing, though again, if someone's upset I don't think anyone'd care much if we added the referenced expected date in the lead.... Shawn in Montreal (talk) 03:00, 14 May 2015 (UTC)[reply]

Regional federal polls

[edit]

Is there any provision for including/noting federal polls that have been conducted on a regional level? For instance, CROP has polled Quebec to determine their federal preferences, and Insights West has doen the same in BC. Is there a way of including such polls in the list and chart? Pinkville (talk) 14:03, 27 May 2015 (UTC)[reply]

If you're willing to dig up and compile the provincial-level polls of federal party support (lots of them have been conducted since the 2011 election, and I'm afraid I can't commit the time myself), I would suggest putting them in a separate table with an additional "province" column, otherwise it will clutter up and introduce inconsistency in the current table. Cheers. Undermedia (talk) 15:59, 27 May 2015 (UTC)[reply]
I agree, to avoid creating graph issues, it would have to be a separate table. - Ahunt (talk) 21:40, 27 May 2015 (UTC)[reply]
It could get pretty messy though once you factor in regional polling done at the national level, etc. My thoughts right now are...I dunno. Krazytea(talk) 03:27, 28 May 2015 (UTC)[reply]

Column shading

[edit]

This has now been removed by two editors, including me and reverted by two. Let's see if we can come to a consensus for or against. My thought is that it adds nothing in addition to the existing bolding of lead number and makes the chart messy and harder to read. Why is this needed? Please make your case. - Ahunt (talk) 12:41, 9 June 2015 (UTC)[reply]

As a weekly visitor to this page I find that it greatly improves the aesthetics of the table. CanadianChemEng (talk) 12:48, 9 June 2015 (UTC)[reply]

  • As another regular reader of the page -- I go to it more than 308.com or tooclosetoocall -- I find the colours greatly helps readability for me because the trend line graph atop is a rather messy, with its almost random-seeming spray of dots. So there's not much that I get from that chart. Whereas the colouring of the column blocks is a) not distracting, for me, because one needs to scroll down to see it and b) an improvement, because it's more effective than just the bold at indicating, at a glance, who the leading party is. You can see patterns and trends at a glance -- which again, are not as readily apparent from that trend line on top. Thanks, Shawn in Montreal (talk) 12:53, 9 June 2015 (UTC)[reply]
I'm not opposed to the cell shading, and I think it does make it easier than the bolding to assess at a glance which party(ies) led any given poll, though my suggestion would be to make it several shades lighter/duller than the shading of the column titles, which I think would be more pleasing to the eyes. See Opinion polling for the 2015 United Kingdom general election for a good example of what I mean. Nevermore27, would you be willing to undertake this modification? Cheers, Undermedia (talk) 14:09, 9 June 2015 (UTC)[reply]
I think the much lighter shading would be a great improvement! - Ahunt (talk) 14:35, 9 June 2015 (UTC)[reply]
Lighter shades would be fine for me in principle -- just not sure how that would work for the BQ, which is already a lighter shade of blue than CPC and would now have to get lighter. That could be the stumbling block, here. Shawn in Montreal (talk) 15:49, 9 June 2015 (UTC)[reply]
Oh, wait, no, of course it wouldn't: the BQ will never lead in a national poll and so will never be shaded. Yep, lighter shades for the blocks is fine with me. Shawn in Montreal (talk) 16:00, 9 June 2015 (UTC)[reply]
Never say never! In an epic vote split among all 5 major parties where the BQ takes 100% of voting intentions in Quebec, they could plausibly lead all parties with roughly 24% support nationally!... ;) ... In all seriousness though, I agree with you and say that we go with lighter shading. I would do it myself but I honestly don't know how, which is why I invited the original editor to do it. I would suggest doing some test runs on a small sample of the table—ideally right here in the Talk page—until everyone's happy with the colours, before changing all 200+ polls in the table. Cheers, Undermedia (talk) 16:41, 9 June 2015 (UTC)[reply]
Lighter shading is absolutely doable, I was just doing this quick-and-dirty last night so it's just a simple matter of updating the template. Thanks for having this discussion, I think shading helps a lot for people who are just looking for polling trends at a glance. Nevermore27 (talk) 21:24, 9 June 2015 (UTC)[reply]
Just did it. Undermedia Ahunt Shawn in Montreal Thoughts? Nevermore27 (talk) 04:55, 10 June 2015 (UTC)[reply]
It's a definite improvement! - Ahunt (talk) 12:22, 10 June 2015 (UTC)[reply]
Fine with me. I also liked it in the original colours but this does the trick, too. thank you, Shawn in Montreal (talk) 13:15, 10 June 2015 (UTC)[reply]
I think the lighter colours are an improvement. Thanks for making the change. One last question: Since in the case of polls where 2 parties were tied for the lead, we had been bolding both parties' numbers, should we not also shade both parties rather than shading neither? In any case, that would be my personal preference. Cheers, Undermedia (talk) 13:25, 10 June 2015 (UTC)[reply]
Oh yes in the case of tie, please do shade both/all. Shawn in Montreal (talk) 13:37, 10 June 2015 (UTC)[reply]
Done Nevermore27 (talk) 21:16, 10 June 2015 (UTC)[reply]

Likely vs. Eligible Voters

[edit]

Since the pollsters use different (and sometimes secret) methods for determining likely voters, and not all pollsters even release likely voter adjustments, I am inclined to continue listing the poll results based on eligible voters rather than likely voters. There seems to be the threat of a small edit war over this question, so I think it prudent to discuss it rather than trying to force our preferences on the article through repeated edits. --Llewdor (talk) 17:44, 6 July 2015 (UTC)[reply]

  • Agreed As I mention in my edit summary —for what is at this point the current version—it isn't a question of whether or not one polling firm's proprietary so-called "likely" voter model is a useful tool or not. I have no strong view on that one way or another (though it seems to me to introduce another level of extrapolation and therefore more possibility for error imo). The problem with entering that data here is that everyone else is measuring eligible voters. We need to compare apples to apples. Shawn in Montreal (talk) 20:54, 6 July 2015 (UTC)[reply]
  • Although I'm a maniac for consistency (and therefore tend to agree with you guys), others have also made the perfectly valid point that it's most intuitive to simply report the pollsters' top-line results, regardless of whether they're based on eligible or likely voters. You have to admit it's a bit counter-intuitive, for example, to put 31-36-23-5-5 in the Wiki table for the recent Angus Reid poll, only to click on the link to the report and see 33-36-23-4-4 front and centre and have to dig way down into the report's detailed data tables to find the eligible voter results that I entered in the Wiki table. Also, I've heard that other pollsters may be publishing likely voter results without explicitly saying so. If we were to go strictly with top-line results, whatever they may be, I would suggest perhaps adding "LV" or "EV" superscript linking to a footnote at the bottom of the table for any polls that explicitly provide 2 sets of results, in order to clearly identify what's being shown in the table. The superscript labels could maybe be added after the name of the polling firm or something, e.g. Angus ReidLV. Cheers, Undermedia (talk) 21:28, 6 July 2015 (UTC)[reply]
  • Oh, well, I'd completely misunderstood your position. If "top-line results regardless of methodological differences" is the consensus, and if other pollsters are playing with the numbers in other fun and wondrous ways, then I won't stand in the way. But I should point out you don't report top-line results for the enduring mystery that is "The Nanos Party Power"--because, of course, not even pollsters can figure out what the hell that measures. Paul Wells has been most amusing on that point. In that odd case, you do drill down and pull the sensible number that fits. Anyway, I thank you all for maintaining this page. Shawn in Montreal (talk) 02:17, 7 July 2015 (UTC)[reply]
  • Remember, I'm the one who switched the Angus Reid poll from top-line likely voter results to buried-down eligible voter results in the first place. I was just stating the opposing argument in this debate which has been raised by other editors, and does have some merit, in an attempt to get a discussion going... Cheers, Undermedia (talk) 03:05, 7 July 2015 (UTC)[reply]
  • I don't think intuitiveness matters. Shawn makes the most relevant point, I think, that we want the table to be consistent in what it reports so as to allow apples to apples comparisons. If every pollster were reporting likely voters, then perhaps we could use likely voters, but since some pollsters only release eligible voters numbers, we need to report eligible voters numbers for everyone in order for the data to make sense. --Llewdor (talk) 14:05, 7 July 2015 (UTC)[reply]
  • Yeah, I seem to have people's POV's flipped but apples⇔apples is best imo. Let's see what consensus is. Shawn in Montreal (talk) 14:13, 7 July 2015 (UTC)[reply]
  • Like I said, I tend to agree with you guys, so I guess we leave the results in the table as is (eligible voters) for now, and wait and see if someone holding an opposing view jumps into this discussion. Cheers, Undermedia (talk) 17:44, 7 July 2015 (UTC)[reply]

So Angus Reid has just released a new poll, and like their last one they've put the results among "likely voters" front and centre, whereas you actually have to perform a bit of arithmetic on the numbers in their detailed tables at the end of the report to figure out the results among "eligible voters" (i.e. manually remove the undecided/non-voters and redistribute the decided voters to 100%). According to my calculations it works out to CPC 30% NDP 37% LPC 23% BQ 4% GPC 4%. This is obviously a bit of a workaround in an effort to achieve results that we believe are more comparable to the other polls. I'm all for consistency, but I again predict this will become contentious among Wiki readers/editors if we seemingly fiddle with the results of each Angus Reid poll that comes out between now and voting day without any explicit explanation or footnote. On the page for the last Ontario provincial election we created a secondary dedicated table for likely voter results, and displayed eligible voter results in the main table and in the graph. Otherwise, like I suggested above, we might alternatively just put an "LV" superscript linking to a footnote below the table for polls that emphasize results among likely voters. Thoughts? Cheers, Undermedia (talk) 12:26, 26 August 2015 (UTC)[reply]

The idea of having to do calculations to "fix" these results sounds too much like WP:OR to me and introduces the possibility of errors, not to mention additional discussions. Notwithstanding my earlier comment, perhaps we have to report what the polls say and footnote them as you have done in this case. - Ahunt (talk) 17:28, 26 August 2015 (UTC)[reply]
Yeah, at the end of the day it's really up to the pollsters themselves to decide what numbers they want to put forward and stand behind, and if they want to take risks with various "unproven" turnout adjustments and what not, that's their prerogative. Who are we to "correct" them in the name of consistency? If the turnout adjustments prove to be faulty, then the pollsters pay with their reputations, so I agree we should probably just let them be. Cheers, Undermedia (talk) 02:28, 27 August 2015 (UTC)[reply]

Adding selected events

[edit]

I noticed that in the poll section for the 2015 British General Election, they had events interspaced throught the list of polls in order to give the reader a better sense of what is going on. This would work very well here, as right after the Alberta election, the NDP surged into first place. Also, there's a national leaders' debate on the sixth, and that might change everything as well. So what do you think?Ericl (talk) 17:11, 14 July 2015 (UTC)[reply]

According to this there will be ten leaders' debates of some kind. If we even just put those on the chart it will cause an awful clutter. Then there will be dozens of other "events" that at least someone will think are significant. So I would say "no", a list of opinion polls should be a list of opinion polls and nothing else. - Ahunt (talk) 17:42, 14 July 2015 (UTC)[reply]
I agree it could be messy but commenting where there could be movement would be significant in providing context for shifts in the polls such as the Alberta General Election result. The question is how does one determine the movements and shifts, who are the deciders and why, ie. This is an open source webpage and could be hard to determine. I do like adding some context to the poll list. Krazytea(talk) 19:20, 14 July 2015 (UTC)[reply]
This may seem like a nice idea in theory, but I agree with Ahunt in the sense that I can easily see this turning into constantly recurring fights over which events can be considered "significant" enough to include in the table, as there's a seemingly endless variety of things that could potentially influence voting intentions. In addition to all the leaders debates, should we insert a note whenever a party gets a new leader (or if a popular leader or former minister dies, for that matter)?; whenever there's a provincial election that may or may not influence federal voting intentions?; whenever there's a major political gaffe or scandal that could affect voting intentions?; whenever there's a major downturn or upturn in the economy or fiscal outlook, or every time a new budget is introduced for that matter?; whenever the government passes very popular or very unpopular legislation?; and how about every throne speech?; or if there's an attack on Parliament Hill?... See what I mean? For this reason I'm inclined to leave the table strictly to its main raison d'être, i.e. a list of opinion polls. Cheers, Undermedia (talk) 20:41, 14 July 2015 (UTC)[reply]
Agreed, a nice idea in theory but the current confusion over election debates makes this unworkable. Eric Grenier has in the past only flagged it when there has been either an election or a new leader -- and those should be the only events indicated in any line graph imo. Shawn in Montreal (talk) 20:45, 14 July 2015 (UTC)[reply]

Campaign period polling graph

[edit]

Dear fellow editors: I intend to update the current polling graph up until the writ is officially dropped, then I propose a dedicated campaign period graph be created as has been done in multiple past federal and provincial elections. The main reason for this separate graph is that the rate at which polls are conducted becomes much more fast and furious during the campaign period and these data would show up all scrunched together on the current >4-year-long graph, which would be of little value. However, I'm afraid I might not have enough free time to sufficiently fast and furiously update a graph throughout the campaign, so... any takers? Some really nice campaign period graphs have been created in the past, my personal favourite in terms of aesthetics and sophistication being this one. I also in principle like the graphs with separate lines linking successive polls by each pollster, which worked fairly well, for example, in the last Quebec provincial election, but turned out pretty messy in the last Ontario provincial election which was marked by highly inconsistent results among different pollsters. Cheers, Undermedia (talk) 19:06, 18 July 2015 (UTC)[reply]

Update: Given that no one has yet volunteered to take over graph duties, I have just created and added a campaign-period graph which will mimic the style of the pre-campaign-period graph. I will do my best to update it at least daily as polls are released. Cheers, Undermedia (talk) 12:42, 2 August 2015 (UTC)[reply]
Thanks Undermedia! What about adding the 2011 election results as the first data point, as was done in the 2011 graph? Foreen (talk) 12:57, 2 August 2015 (UTC)[reply]
I'll give it a test as soon as polls start coming out. As it stands I plan to carry the current trend lines from the pre-campaign graph over onto the new campaign graph (you'll see what I mean as soon as the first poll is added), which might look confusing in combination with showing last election's results at the start of the graph. Cheers, Undermedia (talk) 13:27, 2 August 2015 (UTC)[reply]
Thanks for taking that on. There will be lots of polling to record! - Ahunt (talk) 15:18, 2 August 2015 (UTC)[reply]
Hi! If there is interest, I am able to produce the a plot in the style of previous elections like this one. I can do this myself, or I would be happy to show someone else how to run my script that creates the plots. galneweinhaw (talk) 00:16, 5 August 2015 (UTC)[reply]

A combination of election period opinion polls during the 2011 Canadian general election. The trend lines are Local Regressions with α = 0.45 and 95% confidence interval ribbons. The point sizes and trend lines are weighted according to the confidence intervals of each individual poll. galneweinhaw (talk) 00:25, 5 August 2015 (UTC)[reply]

I think that graph looks slick but my main suggestion would actually be to not factor in poll sample sizes for weighting the trend lines, i.e. weight each poll equally. Sample size is generally overrated and interpreted far too superficially/simplistically. Take IVR polls for instance: the typical demographic composition of people who are willing to participate in a completely robotic poll over the phone is so badly skewed that by the time they're done weighting the data to reflect the actual population according to the census, they've sacrificed a significant chunk of the effective sample size. Good old fashion live telephone polls by comparison tend to have smaller sample sizes, but also tend to collect a more representative sample of the population from the get-go, and thus require less weighting. Cheers, Undermedia (talk) 04:08, 5 August 2015 (UTC)[reply]
I believe in the past the issue we had was that the rolling polls (originally just Nanos) were having way too much influence, because his polling method did small samples daily (say 300 samples) and then he used the last 3 days, but they were being displayed and reported as entire polls every day (say 900 samples), so one could not tell the difference between a 300 vs a 3000 sample poll. Also, the weighting is based of CI, which is prop to sqrt(sample size), which is a smaller effect. I mention this just for context to how they got added in the first place. However, I can provide both for now (and any other changes, ideas, improvements) and allow the community to choose which, if any, they would like to use! I take requests! I just like making pretty graphs and its an excuse to learn R =D galneweinhaw (talk) 15:53, 5 August 2015 (UTC)[reply]
Good points. Well I say we tentatively go with your graph, same formula as last election, cause it's frankly much cooler than mine. I invite you to go ahead and make the replacement. Cheers, Undermedia (talk) 00:34, 6 August 2015 (UTC)[reply]
Actually if it is not too much trouble I say we keep both graphs, like we did in 2011 and just show them side-by-side on the page. They are both of value. - Ahunt (talk) 00:37, 6 August 2015 (UTC)[reply]
Sure, I can keep mine going too. I was considering laterally compressing it a bit, so that would actually work out better for displaying two graphs side-by-side. Cheers, Undermedia (talk) 00:48, 6 August 2015 (UTC)[reply]
Actually keeping both graphs also provides some redundancy in case one of you becomes unable to keep up your own graph for some reason, so I think it is a good idea. I'll let you figure out how to display them, whether to stack them vertically, reduced side-by-side, or as thumbnails. - Ahunt (talk) 13:02, 6 August 2015 (UTC)[reply]
Doesn't look like there's interest in using this so I'm not going to bother posting updates. If anyone would like it just let me know. I enjoyed doing them while they were useful! If anyone wants the script, let me know. galneweinhaw (talk) 05:03, 18 August 2015 (UTC)[reply]
I thought the consensus was to put both graphs up? I just laterally compressed mine some more, so side-by-side should work if you compress yours as well. And where are the trendlines? Cheers, Undermedia (talk) 18:25, 18 August 2015 (UTC)[reply]
Just for the record I am still in favour of using both graphs, if you both will maintain them until the election in October. - Ahunt (talk) 18:32, 18 August 2015 (UTC)[reply]
An auto-generated plot that scrapes this page for polling data: http://www.r-fiddle.org/#/fiddle?id=0UNj2rov&version=26. Click the green "Run Code" button at the bottom right, then resize the plot (grab corner and drag) up to 480 x 480px. Right click to save image. galneweinhaw (talk) 21:11, 22 August 2015 (UTC)[reply]
Nice! I've added it to the main article page. Cheers, Undermedia (talk) 23:03, 22 August 2015 (UTC)[reply]
The trendlines at the beginning of this graph are ridiculous because of the one initial poll and then a big gap before any others came out. I'm going to try and copy what you did by including earlier data to make it more accurate. galneweinhaw (talk) 02:41, 27 August 2015 (UTC)[reply]

I can see that these daily Nanos polls based on a rolling 3-day average could become problematic for the graphs, mine in particular. Unless 2 more polls per day get published on average from now on, the Nanos polls will dominate my 3-poll trendlines. How would folks feel if I only added every 3rd Nanos poll to my graph, thus tempering their pull on the trendlines and also making it such that there are no overlapping data from one poll to the next? Cheers, Undermedia (talk) 14:05, 9 September 2015 (UTC)[reply]

This is one problem with graph averaging over just using raw data, in that more frequent polls affect the lines more than less frequent polls. I am not sure we can pick and choose which polls to graph, though. - Ahunt (talk) 14:41, 9 September 2015 (UTC)[reply]
I would argue that systematically including every 3rd poll doesn't quite amount to "picking and choosing" (which has a bit of a "cherry-picking" connotation to it), but yeah, it's a tricky situation. I think it's going to greatly diminish the value of my graph if I include each daily poll. I don't mind retiring it entirely in favour of only showing galneweinhaw's more sophisticated graph, although I am also curious to hear how galneweinhaw intends to graph these polls given the data overlap, which is somewhat problematic. One option would be to consider each poll as having a sample size of only 400. I'm sure soulscanner would be keen to weigh in here as well. Cheers, Undermedia (talk) 15:18, 9 September 2015 (UTC)[reply]
Including all daily Nanos reports, particularly in graphs, severely overweights the Nanos data by a factor of 3. It would be best to include the Nanos data every 3 days. For example, the first such poll was taken Sept 4, 5, 6. The next poll is compiled using data from Sept. 5, 6, 8. A better way would be to wait for the Sept. 8, 9, 10 data to come out. It's probably not so bad in the table as the links to Nanos reports usually include other useful analysis. But I think in the graphs it would be good to remain mindful of this. Otherwise, you'd be, for example, weighting in Nanos's Sept 6 poll three times in three days. That's really important in regression curves. --soulscanner (talk) 05:36, 10 September 2015 (UTC)[reply]
That's what I'm thinking of doing in my simple 3-poll moving average graph if there's concensus. For galneweinhaw's graph which also factors in sample size, it's perhaps worth considering what Éric Grenier from threehundredeight.com and CBC's Poll Tracker is doing, in his own words: "I'm treating the newest poll as normal, but reducing earlier polls to 33% of their normal weight to represent that it only contains one day of polling that isn't in the newest set of numbers." So in other words at any given time you give the most recent Nanos poll a sample size value of 1,200 and all the older ones a value of 400. Cheers, Undermedia (talk) 12:05, 10 September 2015 (UTC)[reply]
In previous years, I reduced the weight of the Nanos polls by 3 (i.e. divide sample size by 3). You can see the effect here, Nanos polls result it tiny dots, though a lot of them! galneweinhaw (talk) 06:38, 12 September 2015 (UTC) — Preceding unsigned comment added by Galneweinhaw (talkcontribs) 06:35, 12 September 2015 (UTC)[reply]
Yes, I think that makes sense. I say go ahead with it. You might consider giving the first one (Sep 4–6) full sample size, and 33% sample size to each one thereafter. You might further consider similarly reducing the weight of the 4-week rolling average polls released throughout August before Nanos started their daily polling. For my part, should I simply graph every 3rd Nanos poll to avoid duplicate data skewing the trendlines? I think it's the only sensible way to do it. If people are opposed to this though, I'll just retire my graph for good and suggest you stretch yours laterally to take up more of the freed-up space. Cheers, Undermedia (talk) 13:11, 12 September 2015 (UTC)[reply]
I don't think it is ideal, but perhaps the cleanest way is to just graph every third Nanos poll, as that way each graph point will represent wholly new data. - Ahunt (talk) 14:31, 12 September 2015 (UTC)[reply]
Honestly, while my simple graph format did the job during the pre-campaign period, I think galneweinhaw's is proving to be much better suited for the rapid-fire polling during the campaign. It's already been substituted for mine on the main election 2015 page, and truthfully it's a bit redundant to have both. So unless there are strong objections, I will be retiring mine. As mentioned above, I suggest that galneweinhaw's graph be stretched laterally (maybe 1.5x the width), and also see my Nanos poll sample size tweaking suggestions above. Cheers, Undermedia (talk) 12:40, 13 September 2015 (UTC)[reply]
Undermedia, thanks for your efforts, much appreciated! - Ahunt (talk) 14:09, 13 September 2015 (UTC)[reply]
Thanks Undermedia, and by the way, if you're interested in dabbling with R, I'd be happy to walk you through generating these types of plots via pm or email. I will adjust my plot as you recommended about 1.5x galneweinhaw (talk) 21:35, 13 September 2015 (UTC)[reply]
Actually, if you were interested in generating and updating this plot for the campaign, I'd totally be all over that ;), or we could collaborate after the election and create a script to spruce up the next inter-election plot? galneweinhaw (talk) 21:38, 13 September 2015 (UTC)[reply]

Rename

[edit]

Should the title of this entry be changed to Opinion polling for the Canadian federal election, 2015, to reflect the fact the the writs have been dropped, and to match the main article? 58.153.97.134 (talk) 16:48, 2 August 2015 (UTC)[reply]

Ericl, do you want to do the honours? :) -Undermedia (talk) 22:17, 2 August 2015 (UTC)[reply]
Since the 2011 article was called Opinion polling in the Canadian federal election, 2011, shouldn't this be be moved to Opinion polling in the Canadian federal election, 2015 (currently a redirect) instead of Opinion polling for the Canadian federal election, 2015, for consistency? - Ahunt (talk) 00:47, 3 August 2015 (UTC)[reply]
Agreed with Ahunt, also because "for" is improper syntax. Cheers, Undermedia (talk) 01:01, 3 August 2015 (UTC)[reply]
Agreed! galneweinhaw (talk) 06:43, 6 August 2015 (UTC)[reply]
and it has been moved. - Ahunt (talk) 11:09, 21 August 2015 (UTC)[reply]

Seat projections

[edit]

Since governments in Canada are elected riding by riding, seat projections such as those at threehundredeight.com allow a more accurate prediction of which party will form a government, and whether it will be a majority or minority.

Is there any way to do something similar here? Pmerriam (talk) 03:11, 4 August 2015 (UTC)[reply]

These are essentially riding by riding polls and have very large margins of error due to small "n". In the past they have not proven very accurate, so I am not sure they are worth reporting here. - Ahunt (talk) 10:19, 4 August 2015 (UTC)[reply]
Sorry, but I think that's a terrible idea. Yes, Eric Grenier does seat projections, so does Bryan over at tooclosetocall.ca, and David Akin does his "Predictinator," for pete's sake. I think we're better off sticking to polling numbers and leaving aside the projectors, with all their pet formulas. Shawn in Montreal (talk) 14:16, 4 August 2015 (UTC)[reply]
That is a whole other issue brought up - that each organization uses totally different methods for seat projections, making them not at all comparable. - Ahunt (talk) 16:47, 4 August 2015 (UTC)[reply]

On the other hand, with the opinion polls largely pointing to a three-way contest, Canadians familiar with their voting system may be able to predict the likely outcome, but a seat projection chart might prove extremely useful for foreign observers. Culloty82 (talk) 15:16, 19 August 2015 (UTC)[reply]

(over)Accuracy of Margins of Error?

[edit]

I noticed that some of the Margins of Error listed in the article imply a greater accuracy (in the MoE) than actually exists or is reported. For example, the Forum Research MoEs are listed as +/-3.0% when it is actually reported as +/-3% This makes sense as most of them are likely +/-2.7% rounded up (e.g. in the last one 1/sqrt(1399) = 2.67%). I realize it looks nicer to have them all similar, but by doing so we are changing the meaning of the numbers. Thoughts? galneweinhaw (talk) 07:01, 6 August 2015 (UTC)[reply]

I agree. At some point Forum stopped including a decimal in the MoE presented in their poll reports. I think we should present the MoE's as presented by the pollsters themselves. This would involve editing a whole bunch of Forum polls in the pre-campaign table, but I think that's all. Cheers, Undermedia (talk) 13:06, 6 August 2015 (UTC)[reply]
I agree with User:Undermedia, we should report what the polls report, decimals or not. - Ahunt (talk) 13:08, 6 August 2015 (UTC)[reply]

MoE of internet polls

[edit]

Internet polls are not random samples so a margin of error does not apply to them. See http://www.cbc.ca/news2/interactives/poll-tracker/2015/index.html#fn4 Should we replace with "n/a"? — Preceding unsigned comment added by Foreen (talkcontribs) 00:40, 12 August 2015 (UTC)[reply]

Nanos/EKOS poll dates

[edit]

Great work everyone. This is what Wikipedia should be. I've been using this page to compile my own little spreadsheet using google sheets on google drive. It lets you plot some neat regression curves relatively simply.

I pay particular attention to the dates of polling in the methodology section of the poll reports. Everyone seems aware that Nanos reports a 4-week rolling poll. This is an excellent methodology for Nanos to use, but it does mean that the data is most representative of the date at the midpoint of that 4 week period. The date used should be 14 days earlier than the end date of the latest polling report. For example, the most recent Nanos poll date (August 21) should in fact be August 7. The following is a summary of other changes that should be made for Nanos poll dates.

  • Aug 21 -> Aug 7
  • Aug 14 -> Jul 31
  • Aug 7 -> Jul 24
  • Jul 31 -> Jul 17
  • Jul 17 -> Jul 3

Less critical are the EKOS poll dates. EKOS does its polling over 7-day periods. The latest EKOS poll was August 19-25. That means the ideal date of the poll would be Aug. 22, not Aug. 25 as reported on the page.

The Nanos error actually affects the representativeness of the table on the page. The Conservative leads shown in the last 3 weeks for Nanos should actually be pushed back to early August and late July when the Conservative and NDP seemed to be neck and neck. The NDP seems to have opened up a lead since then as indicated by galneweinhaw's graph.

I'll leave it to sonmeone more familiar with the table format to judge which of these whether these changes are worth making. --soulscanner (talk) 05:13, 1 September 2015 (UTC)[reply]

This is definitely a tricky issue. I know threehundredeight.com's Éric Grenier used to use the median polling date of polls for his aggregates—as you are suggesting—but actually stopped doing so because he found that it gave excessive weight to polls by firms that tend to poll over just 1–2 days such as Forum and Mainstreet. The 4-week Nanos polls are definitely problematic at this stage, but apparently Nanos in on the verge of switching to much shorter polling periods going forward in the campaign, so the problem should soon disappear. IMO the 1-week EKOS polls aren't nearly as problematic, and they too are likely to adopt shorter polling periods as we get closer to voting day. Cheers, Undermedia (talk) 14:14, 1 September 2015 (UTC)[reply]
I agree that we should leave it the way we have it, with the date used the last date of the polling done. It isn't perfect, but, all things considered, it is the least-worst method. - Ahunt (talk) 17:37, 1 September 2015 (UTC)[reply]
That's a good point, but Grenier's objective is to get the most recent "snapshot" of current opinion using the most recent polls. To do that he weights the most recent polls more heavily. In his case, it makes sense to use the last date because otherwise the Nanos numbers (and other rolling poll data) would be almost entirely eliminated, and he'd lose that information. It's a justifiable compromise in his case because it effectively expands the database.
On this page's data table, there is no composite "snapshot" of the most recent data and no need for compromise. Indeed, the objective of displaying a timeseries of raw data like this is to infer trends in party support. Rolling polls are in no danger of being omitted or underweighted in a time series by using the median date. It is therefore crucial to use the median date for the reasons outlined in the original post. It's actually misleading to used the end date of the polling period. The degree of Conservative support, for example, is underestimated for the end of July and is over estimated for the end of August as a result. The July upward trend and August downward trend in Conservative support are therefore smoothed out because the Nanos data is out of phase. The same holds for data used in regression curves. Adding the Nanos data earlier in the tie series actually makes the data more useful in building an accurate regression curve. --soulscanner (talk) 18:44, 1 September 2015 (UTC)[reply]
You make some good points, but using the mid-date for the Nanos data makes it act like a snap shot from two weeks ago, which is not what it is. Sure it includes old data, but it also includes new data, too. It isn't accurate to just backdate it two weeks, because the newer data averaged in means it isn't representative of two weeks ago either. That is why I called using the poll closing date "least-worst method"; it is not ideal, but is at least consistent. Overall here on Wikipedia we are writing history, not news, so what we are doing is showing how the issue developed over time, retrospectively. - Ahunt (talk) 18:59, 1 September 2015 (UTC)[reply]
Yeah, there's clearly no perfect solution to this, so I'm also still leaning towards keeping it the way it is for convenience and consistency (i.e. changing the way we do it would require someone to dig into the links for all 267 polls published since last election to figure out the median polling date for each!). The only truly problematic polls are the Nanos ones, which alone don't manage skew the trendlines all that much considering my graph uses a 3-poll moving average and galneweinhaw's graph's trendlines are derived from an even larger dataset when polls are being released as frequently as they are now. Cheers, Undermedia (talk) 19:21, 1 September 2015 (UTC)[reply]
In consideration of the graph producers, it would also mean changing all the dates on the graphs, too. I am really not convinced this is worth doing. - Ahunt (talk) 19:54, 1 September 2015 (UTC)[reply]
Sounds reasonable. Now is obviously not the time (in the middle of an election campaign) to make this change. I will point out though that more accurate graphing is the motivation for doing this (I'm using the regression curves from Google Sheets to produce these). Using the 4-week average of the Nanos poll every week skews the graph considerably owing to a) the two-week shift and b) the overweighting of the Nanos data (weekly sample is around 300 as opposed to 3000 for EKOS). This could be crucial when discerning 2-week trends at the end of the polling period. Perhaps this could be discussed after the election. --soulscanner (talk) 16:22, 7 September 2015 (UTC)[reply]
I added a column for first date of polling. I think this is a useful data point and may help put the Nanos polls in context, but am happy to remove if folks disagree. Foreen (talk) 22:05, 10 September 2015 (UTC)[reply]
Thanks for your effort, Foreen, but if anything I think this would be more economically displayed in a single column simply showing the range of polling dates, e.g. "September 4–8, 2015". This single-column date range format has been used in various provincial-level polling tables on Wiki. That being said though, I think I ultimately agree with Ahunt and soulscanner that this is perhaps not the time to implement these sorts of changes to a page that's been maintained in a largely consistent format for over 4 years. Why don't we discuss all these points again in about 6 weeks when launching the page for Opinion polling in 43rd Canadian federal election? Cheers, Undermedia (talk) 02:14, 11 September 2015 (UTC)[reply]
I agree, the new extra column just adds more clutter. - Ahunt (talk) 13:12, 11 September 2015 (UTC)[reply]

ABOUT THE NANOS DAILY ROLLING AVG POLLS - Shouldn't you only be posting it every third day so the data is fresh each time? Also, what happened to that nice chart that was used before? I loved it and miss it. — Preceding unsigned comment added by 66.209.53.195 (talk)

See above. In the current graph the daily Nanos polls are being assigned a sample size weight of only 400 instead of the full 1,200 to reflect the fact that only one third of the data in each new poll is independent of the previous poll. For my part, I decided to retire my graph because the rather crude 3-poll moving average trendlines were not proving to be terribly well suited for the rapid pace at which polls are being released during the campaign. Cheers, Undermedia (talk) 16:02, 16 September 2015 (UTC)[reply]
Yes I for one prefer the new graph and the fact that this page is updated as soon as a new poll appears, be it Nanos' daily poll or any other. thanks, Shawn in Montreal (talk) 16:20, 16 September 2015 (UTC)[reply]

Innovative Research poll September 10, 2015

[edit]

We can't graph this without the Green Party and Bloc numbers. - Ahunt (talk) 12:29, 12 September 2015 (UTC)[reply]

Nanos?

[edit]

Is it just me or did someone delete most of the nanos polls? — Preceding unsigned comment added by 144.172.210.122 (talk) 17:49, 25 September 2015 (UTC)[reply]

Yes, he did it. It's an user who is keen on deleting opinion polls just because he seemingly doesn't like trackings (opinion polls in quick succession one after another), not hearing what others say on the issue and just for the sake of removing them. He started doing it in this article, and came up here because I put a link to this article on his talk page as an example of election campaign opinion polling frequently having trackings. Sorry about luring him here, I guess. Impru20 (talk) 18:00, 25 September 2015 (UTC)[reply]
I've issued him a second level warning for content blanking, as we can see from his talk page that he repeatedly cautioned (though not always templated) about it. Reverting without issuing warnings is only half the job, imo. If this continues I do believe he should be blocked. Shawn in Montreal (talk) 18:09, 25 September 2015 (UTC)[reply]
He did it again in the Portuguese legislative election, 2015 article, and said he answered you in his talk page. I have issued a level 3 warning to him, answered him and will definitely report him for illegitimate blanking of content against consensus, and continuous present-and-past disruptive editing in other articles, if he does it again. Just keep an eye on this article in case he strikes again here too. Impru20 (talk) 18:52, 25 September 2015 (UTC)[reply]
Certainly, as I just stated on his user talk page. No worries. Shawn in Montreal (talk) 18:55, 25 September 2015 (UTC)[reply]

References

[edit]

I removed the Reference section for this article. The first reference has a broken link, and the second goes to a home page which does not state the information referenced. The third and fourth references are duplicates of the first and second, respectively. The first piece of information referenced is statistical knowledge, while the second is simply a detail of the methodology of some of the polls included (too many to reference). For these reasons, I found it unnecessary to include references for this information. Aa508186 (talk) 18:36, 27 September 2015 (UTC)[reply]

Agreed. Dbrodbeck (talk) 18:40, 27 September 2015 (UTC)[reply]

"Lead" column?

[edit]

Could someone please present a case for adding this column to the table? For the longest time we would just bold the number for the level of support of the leading party, then it was decided that we would add the background shading to make the leading party really obvious. And now we additionally have a "Lead" column... for readers who aren't very skilled at performing mental arithmetic? Plus it's way over on the far right edge of the table, separated from the numbers used to calculate it. Can't say I'm terribly convinced of the necessity of this at this point. Cheers, Undermedia (talk) 12:40, 28 September 2015 (UTC)[reply]

Yeah I find it redundant. Dbrodbeck (talk) 13:10, 28 September 2015 (UTC)[reply]
Not needed, unnecessary, remove. - Ahunt (talk) 21:38, 28 September 2015 (UTC)[reply]
I concur with removing it. Otherwise, at least make it the more lightly-shaded color to match the middle columns of the table so that it is less of a distraction. Ddcorkum (talk) 00:14, 29 September 2015 (UTC)[reply]
Making the pre election and election poll tables different broke my plot generator, so I won't be able to update the plot until either the column is removed, or it is also added to the pre-election polls! galneweinhaw (talk) 02:25, 29 September 2015 (UTC)[reply]
I'm someone who consults this page at least several times a day, and I was glad to see the column added. And no, I don't think that makes me some kind of math-impaired moron. I can perform the necessary mental arithmetic, thank you. It's just that as a column, I find it allows one to more easily and quickly scan down the line to see the leads. It's one more way of expressing in tabular form what's expressed by the line graph. I rather like it. But if it presents some kind of technical problem, then of course it should go. Shawn in Montreal (talk) 03:00, 29 September 2015 (UTC)[reply]
Unless someone's willing to promptly do the 250-some polls in the pre-campaign table, it looks like the column will have to be removed for technical reasons concerning the graph. Sorry my mental arithmetic joke was a bit heavy-handed! Cheers, Undermedia (talk) 03:57, 29 September 2015 (UTC)[reply]
I can probably still make the graph work with the extra column if it's wanted, it might just take me a few days to figure it out before the graph is updated again. galneweinhaw (talk) 04:01, 29 September 2015 (UTC)[reply]
Nevermind, it was easier than I thought. Plot updated =) galneweinhaw (talk) 04:05, 29 September 2015 (UTC)[reply]
Well, as an addict of the page, I really appreciate all the hard work that all of you have done, Undermedia included. I'm a devotee, no matter what you decide. I pretty much live by what I find here and at 308.com. thank you, Shawn in Montreal (talk) 11:50, 29 September 2015 (UTC)[reply]
What Shawn in Montreal said - I like the Lead column EncycloCanuck (talk) 12:15, 3 October 2015 (UTC)[reply]

Trend line question

[edit]

I know it's as clear as mud, with the most recent polls from EKOS, Forum and Angus Reid sharply disagreeing with Nanos, Innovative and Leger where the CPC is relative to the Liberals. I don't think I've ever seen this before. But with pollsters evenly split, why isn't the trend line for the CPC/LPC in the graph converging? Again, 50% of the polls show the Liberals ahead. thanks, Shawn in Montreal (talk) 14:30, 3 October 2015 (UTC)[reply]

As of this moment it doesn't look like today's Nanos or yesterday's Innovative polls have been added to the graph yet, but then again neither has the new EKOS poll. Cheers, Undermedia (talk) 14:36, 3 October 2015 (UTC)[reply]
Okay, I thought they had. Thank you. Shawn in Montreal (talk) 14:42, 3 October 2015 (UTC)[reply]
Btw, I've asked Galneweinhaw on his or her Commons User talk page - but I'll briefly mention it here - if s/he might use the edit summary updates this to simply mention the polling agency names when adding data -- if that would be possible. It would help someone like me know what's there and what isn't yet. either way, it's a great job. Shawn in Montreal (talk) 15:21, 3 October 2015 (UTC)[reply]
I heard back from Galneweinhaw. He uses an automated script to make all these updates so what I've asked isn't possible. It's fine. I'll just concentrate and try to use The Force to bend the trend lines in my desired direction. That ought to do it. Shawn in Montreal (talk) 16:43, 3 October 2015 (UTC)[reply]
Shawn in Montreal thanks for the prompt. I will start updated the graph more often. I was trying to update once per day, but that's not enough in this crazy race we have going on =D. Is there a way to get an immediate email when someone changes the page? galneweinhaw (talk) 16:48, 3 October 2015 (UTC)[reply]
I don't have email enabled on my Wikipedia account, sorry. I'd really prefer not to. I assume you have a Watch on this page, right? Shawn in Montreal (talk) 16:49, 3 October 2015 (UTC)[reply]
Sorry, I meant automated emails by Wikipedia! galneweinhaw (talk) 17:14, 3 October 2015 (UTC)[reply]
Galneweinhaw, on a somewhat related note, I've been meaning to suggest that you consider reducing the weight of the 4 Nanos polls published in August—same as you're doing with the current Nanos polls—as they were also based on a rolling average (in their case, a 4-week average!) and therefore contain overlapping as well as highly outdated data. Thus, I would suggest assigning each of them a sample size of 1,000 / 4 = 250. Cheers, Undermedia (talk) 17:15, 3 October 2015 (UTC)[reply]
Undermedia easy to do. I'll just give it a day to see if anyone else chimes in. If not I will go ahead with it. galneweinhaw (talk) 00:22, 4 October 2015 (UTC)[reply]
Sounds good, galneweinhaw. There don't seem to be any objections as of yet. In addition to the 4 Nanos polls in August which should have their sample sizes reduced by a factor of 4 (i.e. to 250), I would also suggest reducing the sample size of the July 31st pre-campaign Nanos poll (which I believe influences the trendline, correct?) by a factor of 2 (i.e. to 500) since it is also part of the 4-week rolling average series, with the preceding published poll having come out 2 weeks prior on July 17th. Make sense? Cheers, Undermedia (talk) 13:32, 7 October 2015 (UTC)[reply]

Graphing the rolling-average polls

[edit]

1. We have agreed that for graphing purposes, the sample size of individual "rolling-average" polls such as those released by Nanos and now EKOS should be decreased, otherwise the duplicate data among successive polls unduly influence the trend line.

2. This is being done effectively for the daily Nanos polls based on a 3-day rolling average, for which each day of tracking adds 400 new interviews while dropping the oldest 400 interviews. Thus, with the exception of the very first poll in the series (6 September) which was assigned the full sample size of 1,200, each subsequent poll has been assigned a sample size of 400 to reflect the fact that only 400 of each poll’s 1,200 total interviews are independent of the previous poll’s data.

3. Myself and others have argued that some sort of similar “down-weighting” should also be applied to Nanos’ polls prior to September, which were based on a 4-week rolling average where each new week added 250 interviews while dropping the oldest 250 interviews. I have therefore suggested that the four such Nanos polls released throughout August be assigned a sample size of 250 in the graph; and that the poll released on 31 July (which affects the trend line even though it doesn’t show up on the graph) be assigned a sample size of 500, since the previous poll was released two weeks prior on 17 July as opposed to one week prior.

4. Down-weighting must now also be applied to the daily rolling-average EKOS polls. As with the daily Nanos polls, the very first one (5 October) should be assigned full sample size, and each subsequent poll appropriately down-weighted. However, this could be a bit trickier with the EKOS polls since the number of interviews conducted daily appears to be inconsistent (e.g. the total sample size of the 5 Oct poll was 1,658 whereas that of the 6 Oct poll was 1,788) and the reports don’t indicate exactly how many interviews were conducted on each of the 3 polling days. Assuming the sample sizes won’t vary exceedingly, I suppose a rough way to handle it would be to simply divide the sample size of each poll by 3. Galneweinhaw, would that be doable?

Any comments/reactions? Cheers, Undermedia (talk) 18:40, 8 October 2015 (UTC)[reply]

That all makes sense to me, with the caveat that we are to some extent convoluting a process that already has major convolutions due to the inability of any polls to get truly random samples. I would suggest that after the election results are known that we should have a debriefing discussion here and see if we have any recommendations for how to do it better next election. - Ahunt (talk) 19:18, 8 October 2015 (UTC)[reply]
Done! Pre Aug 31st Nanos are sample size ÷ 4 (Haven't dealt with 31 July, it gets ÷4 right now too) . Post 5 Oct EKOS sample size ÷ 3. Also added a bit of transparency (80%) so that overlapping polls are visible (hard to notice on the small plot, but as a vector graphic you can magnify an arbitrary amount without loss of resolution). galneweinhaw (talk) 03:25, 9 October 2015 (UTC)[reply]
Good stuff, galneweinhaw. Well done! Undermedia (talk) 14:58, 9 October 2015 (UTC)[reply]
Hi again, galneweinhaw. I realize this must be getting tiresome, but unfortunately I'm a maniac for consistency and can't help myself: regarding the 10 October EKOS poll, I think it should exceptionally be assigned two thirds of its total sample size in the graph instead of one third, since curiously EKOS didn't release a poll for 9 October, and therefore 2 out of the 3 field days comprising the 10 October poll are independent of the preceding poll from 8 October. If my nitpicking is getting troublesome, you suggested a little while back that we both collaborate on maintaining this graph. I'd be keen to figure out and edit the source code myself, though there no longer seems to be a link to it in the graph's details page. In any case, thanks for your great work on this; it's a really slick graph. Cheers, Undermedia (talk) 02:12, 12 October 2015 (UTC)[reply]
Undermedia, I added the source code back to the description. That site is a nice playground for experimenting with the code. You can drag the plot it generates larger, but if you want a custom sized/proportion plot you need to generate it yourself with something like RStudio (free). I'm happy to help if you have any questions. For the future we might want to consider somehow integrating this information into the table, that way we don't need to change the source code to make exceptions (bad practice) every time there is something odd going on (like these exceptions you are raising). For example, if in the Polling method column after "rolling average" there was some indication of how many polls it was averaged over, we could just read that info and the plot would account for it automagically. E.g: "telephone - rolling average (3)". This might be good info to have in the table anyway, independent of the plot? — Preceding unsigned comment added by Galneweinhaw (talkcontribs) 18:26, 12 October 2015 (UTC)[reply]
Galneweinhaw, great idea. Check out what I've done in the Polling Method column. Can you edit the graph's source code to automatically multiply the sample size by those fractions whenever present? If so, it should no longer be necessary to rely on so-called exceptions. Cheers, Undermedia (talk) 22:58, 13 October 2015 (UTC)[reply]
Undermedia, excellent. I'll see what I can do with this! galneweinhaw (talk) 03:46, 14 October 2015 (UTC)[reply]
The latest Nanos poll is going to be another exception (and probably tomorrow as well) since they changed their sample size last night. Are you able to access the "fractions" in the method column? Ddcorkum (talk) 12:38, 17 October 2015 (UTC)[reply]
Not yet. I'll take a stab at it this evening. galneweinhaw (talk) 13:27, 17 October 2015 (UTC)[reply]
Any luck, galneweinhaw? As you can see, there was quite a bit of variation in those rolling poll fractions over the course of the final week or so of the campaign. For example, the sample size of the final Nanos poll composed of data from Sunday only should not be reduced at all. Undermedia (talk) 16:13, 20 October 2015 (UTC)[reply]
Undermedia, No, sorry. I"m not sure when I'll get a chance to look at it =( 142.29.143.200 (talk) 23:01, 20 October 2015 (UTC)[reply]
Hi galneweinhaw. Is there anything I can do to help get this figured out (i.e. incorporating the fractions)? You'll notice a new poll from October 18 has since been added to the table as well. I'd love to learn the code myself and have checked it out a few times, but I confess I've been too intimidated to attempt fiddling with it. If we can get the code finalized I would like to keep using this style of graph going forward for the pre-campaign period of the 43rd federal election. Cheers, Undermedia (talk) 14:19, 2 November 2015 (UTC)[reply]
Sorry Undermedia =( Give me a couple weeks and I'll get it done. Re tinkering with the code, you can edit that code on rfiddle as much as you like, it won't override anything. I tried commenting it decently, but will add more when I get a chance, with the goal of giving enough info that one could edit it to reproduce a new chart from another dataset/wikipedia page. You can PM me if you have any questions. When you want to save any changes, hit save and the version number in the link will change, just bookmark the new link, the old code/version will still be there. Even better would be to download Rstudio, then copy/paste the code into a new project and run it offline. galneweinhaw (talk) 15:28, 2 November 2015 (UTC)[reply]

Abacus Leadership Poll Results October 17

[edit]

The question asked by Abacus appears to have been about which leader performed best on the campaign trail rather than who would make the best PM. Feel free to remove this row from the table if you think it shouldn't be there and you won't offend me :)

EncycloCanuck (talk) 21:11, 18 October 2015 (UTC)[reply]

I had a better idea than just removing. Ribbet32 (talk) 21:59, 18 October 2015 (UTC)[reply]

EKOS October 18 date question

[edit]

The document from EKOS at http://www.ekospolitics.com/wp-content/uploads/final_report_october_18_2015.pdf says

"The field dates for this survey are October 8-12, 2015. In total, a random sample of 1,124 Canadian adults aged 18 and over responded to the survey. The margin of error associated with the total sample is +/- 2.9 percentage points, 19 times out of 20."

So shouldn't this really be given the date of October 12 ?

EncycloCanuck (talk) 11:59, 19 October 2015 (UTC)[reply]

Oh - OK - that was in the section for Election Issues not voting intention so my question is probably moot EncycloCanuck (talk) 12:02, 19 October 2015 (UTC)[reply]

Debriefing

[edit]

Okay the election is done! Thank you to everyone for their work on this page, it was widely used and quoted from during the election! As I noted above I think we should hold a bit of a debrief now to see what was done well and should be done again in 2020, what could be improved upon and what should be skipped next time. Any thoughts? - Ahunt (talk) 12:06, 20 October 2015 (UTC)[reply]

A big thank you to everyone who worked so hard on this thing. Well done, people!!!!YoursT (talk) 13:42, 20 October 2015 (UTC)[reply]

Campaign Period: Final Polls

[edit]

I tried to gather the final polls in to a separate table, since they were scattered throughout. I left out Environics though because their last (and only?) poll was a full month ago so it didn't seem to add much value. Ddcorkum (talk) 20:08, 20 October 2015 (UTC)[reply]

It has been reverted, not a good idea and not necessary. - Ahunt (talk) 00:03, 21 October 2015 (UTC)[reply]

Thank you

[edit]

Well, this election is over, thank god, and I've just ended my watch on this page. I want to thank all the editors who have put so much work into this over these many months. I must say I pretty much based my mood around the results I found here! Thank you so much, this was more than just an encyclopedia page for me and I'm sure many others -- you've done a real public service. Shawn in Montreal (talk) 03:58, 21 October 2015 (UTC)[reply]

A lot of us spent a lot of time here everyday, but now it is over. Thank you everyone for all your work. This page was widely consulted by the public and the media, it is hard to overestimate how useful it was and worth all the effort expended on it. - Ahunt (talk) 11:52, 21 October 2015 (UTC)[reply]
It is sort of weird waking up and not checking Nanos and then seeing if anyone has added it yet...... Dbrodbeck (talk) 13:31, 21 October 2015 (UTC)[reply]
I think we all have a bit of "election withdrawal" here. - Ahunt (talk) 14:02, 21 October 2015 (UTC)[reply]
Thanks for everything all of you have done, see you in various provincial elections until 2019. Krazytea(talk) 15:41, 21 October 2015 (UTC)[reply]