This article is rated List-class on Wikipedia's
content assessment scale. It is of interest to the following WikiProjects: | |||||||||||||||||||||
|
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)
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)
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)
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)
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)
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)
I find the shadows on the scatterplot points a bit distractive. Can they be removed? 117Avenue ( talk) 22:20, 21 July 2013 (UTC)
What about extending the x-axis to October 2015 rather than just 3 months out? - Thewatcher2015 ( talk) 17:23, 14 May 2015 (UTC)
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)
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)
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)
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)
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)
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)
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)
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)
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)
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)
galneweinhaw ( talk) 00:25, 5 August 2015 (UTC)
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)
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)
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)
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)
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)
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 ( talk • contribs) 00:40, 12 August 2015 (UTC)
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.
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)
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)
We can't graph this without the Green Party and Bloc numbers. - Ahunt ( talk) 12:29, 12 September 2015 (UTC)
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)
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)
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)
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)
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)
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)
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)
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)
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)
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)
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)
This article is rated List-class on Wikipedia's
content assessment scale. It is of interest to the following WikiProjects: | |||||||||||||||||||||
|
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)
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)
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)
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)
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)
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)
I find the shadows on the scatterplot points a bit distractive. Can they be removed? 117Avenue ( talk) 22:20, 21 July 2013 (UTC)
What about extending the x-axis to October 2015 rather than just 3 months out? - Thewatcher2015 ( talk) 17:23, 14 May 2015 (UTC)
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)
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)
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)
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)
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)
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)
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)
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)
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)
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)
galneweinhaw ( talk) 00:25, 5 August 2015 (UTC)
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)
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)
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)
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)
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)
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 ( talk • contribs) 00:40, 12 August 2015 (UTC)
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.
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)
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)
We can't graph this without the Green Party and Bloc numbers. - Ahunt ( talk) 12:29, 12 September 2015 (UTC)
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)
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)
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)
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)
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)
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)
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)
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)
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)
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)
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)