Talk:Opinion polling for the 2008 Canadian federal election

Untitled

if this page is about polling "in the Canadian federal election, 2008" should we not have a separate graph showing only the polls taken "in the Canadian federal election, 2008"? 198.96.35.219 (talk) 11:11, 15 September 2008 (UTC)

Graph

This graph was uploaded on 14 Sept 2008 and appears to not incorporate the last three or perhaps four polls. It really needs to be updated daily and also needs a date on it to show clearly the date when it was last updated. If it isn't going to be updated then it will need to be deleted within a day or two as it will be misleading as the numbers change. - Ahunt (talk) 01:08, 16 September 2008 (UTC)

Graph is misleading now. Shows some parties polling 0, which is not correct. If party not polled there shouldn't be a data point at all. I'm deleting. Nfitz (talk) 08:52, 16 September 2008 (UTC)

I think you are right, while the idea of doing a graph is good, these ones need some work to make them accurate before they are re-instated. - Ahunt (talk) 12:12, 16 September 2008 (UTC)

• Graphs are back ... but I still find them a bit misleading. Why use lines, when there are multiple samples each day? wouldn't just using symbols look better? Nfitz (talk) 21:57, 17 September 2008 (UTC)

I agree - it took me a few minutes looking at the graph to figure out what it is saying - the vertical lines some days are confusing, but I guess they indicate multiple polls with different results on the same day. I am more concerned that the graph will need updating every day or else it will quickly become out of date. It seems in elections these days we are getting polled to death. - Ahunt (talk) 22:07, 17 September 2008 (UTC)

The graphs at Opinion polling for the New Zealand general election, 2008 might be helpful. You'd have to look at the article in edit mode because the graphs are commented out. Alternatively, there was a discussion at that article about the graphs, and at least one of those editors knows how it was made. -Rrius (talk) 23:40, 18 September 2008 (UTC)

This new graph is very confusing. I don't understand it at all. Could someone please make a more understandable graph like the one for the 2006 elections? Esn (talk) 00:41, 19 September 2008 (UTC)

Hear hear! —Preceding unsigned comment added by 134.153.216.129 (talk) 20:06, 19 September 2008 (UTC)

I did the 06 graph and will go for it agian this election. I should be able to update it daily. The coloured bands on the 06 graph bug me because they are qualitative. I have an idea for a more systematic band/range I'll try. Obviously open to suggestions. I'll make sure anything like I do is fully explained here =) galneweinhaw (talk) 21:50, 19 September 2008 (UTC)

Please do this! Anything to replace the current confusing graph. I like the coloured bands in the 2006 graph... they make things clearer. Don't they represent something like the combined margin of error? Esn (talk) 02:41, 20 September 2008 (UTC)

No, they are just visual. If the Uncertainties were listed/gathered somewhere for easy reference I could do that. Actually, I think each of the polling firms are using constant sample sizes and thus have constant uncertainty with their polls. I'll check into it, but updating the plot daily will be enough work for me as it is =P. I chose a weighted moving average, because an unweighted 3 day average just looks silly with so little data and no obvious trends. I think some kind of moving average is needed because otherwise (with a simple daily average) a trend line will be dominated by days with single polls, and if that single poll is an outlier then it looks even worse. galneweinhaw (talk) 04:44, 20 September 2008 (UTC)

I'd say don't worry too much about daily updates; once every few days should be fine. Take your time and do it right, we're in no rush... it's not like the polls are in a state of flux; in fact, nothing much has been happening all week. All of the links mention the methodology of the polling companies, and there aren't too many of them... I really think that turning the margin of error numbers into coloured bands would be a very useful thing to add eventually (I realize that this will take a lot of time probably).

Now then, here are some margin of error stats:

Ipsos Reid (Sept 18) 3.1 percentage points, 19 times out of 20

EKOS Sept 18 1.6 percentage points, 19 times out of 20

Nanos Sept 18 2.9 or 3.1 percentage points, 19 times out of 20 (methodology section says 2.9, while yellow row under "Canada" says 3.1)

Harris-Decima Sept. 18 2.7 percentage points, 19 times out of 20

I haven't checked yet to see if these are the same on the other dates as well... Esn (talk) 09:03, 20 September 2008 (UTC)

Ok, some more for Sept. 17: Nanos (same as before, 2.9 or 3.1?); Harris-Decima (2.6%); EKOS (1.6% again).

Sept. 16: Nanos (2.9 or 3.2), Harris-Decima (2.6%), EKOS (1.8%)

Sept. 15: Nanos (2.9 or 3.2), Harris-Decima (2.6%), EKOS (1.8%), Segma (2.5%)

Sept. 14: Nanos (2.9 or 3.2), Harris-Decima (2.6%), EKOS (2.1%)

Sept. 13: Nanos (2.9 or 3.2), Harris-Decima (5.5%), Ipsos Reid (3.1%)

Sept. 12: Nanos (2.9 or 3.2), Angus Reid (2.9%)

Sept. 11: Nanos (2.8), Harris-Decima (2.6%), EKOS (1.5%), Ipsos Reid (3.1%)

Sept. 10: Nanos (2.8)

Sept. 9: EKOS (2.0%), Angus Reid (3.1%)

Sept. 8: Harris-Decima (??? - seems to be a dead link now)

Esn (talk) 09:28, 20 September 2008 (UTC)

One other suggestion: The % results of the last election should be on the very left of this graph, like in the 2006 graph. Esn (talk) 09:32, 20 September 2008 (UTC)

Nodice includes the MoE with each poll, but it doesn't seem to have been updated in a few days (though it had been updated daily before). frogg 17:09, 22 September 2008 (UTC)

Alrighty, so, in order for the confidence intervals to have any quantitative meaning, they'll have to be the calculated confidence intervals on the trend. Poll Averaging is most accurately done by weighing the polls based on their sample size (or margin of error... same thing really). I checked out the confidence bands on the New Zealand plot linked to above... and they are not the calculated confidence intervals of the trend, just bands created with a built in a algorithm by the graphing program used (ggplot2 for R)... however I don't think excel can do the bands, so I might have to switch programs if we want those added in. 154.20.9.49 (talk) 21:58, 21 September 2008 (UTC)

One more thing about the graph: would it be possible to show when there are two polls in the same day giving the same % to a party? For example, by having two squares (or circles, etc.) side by side. Esn (talk) 02:55, 23 September 2008 (UTC)

I switched to different software (sorry excel, you just weren't cutting it) to try and include some of the graph requests. Feedback appreciated ont the new plot! I was able to add some trend ribbons and offset overlapping data, but lost the election (should be easy to add back in).galneweinhaw (talk) 03:30, 25 September 2008 (UTC)

I think it looks fantastic! Thanks a lot! I think the vote from the last election should be added back in, and one other thing... would it be possible to write a more detailed description on the image profile page, including what the grey area is exactly and what calculations you used to obtain it? Esn (talk) 04:28, 25 September 2008 (UTC)

I also like this better than before, but a few suggestions, anyway: 1) can you make the graph larger? 2) Can you make the gridlines black? They are hard to see... 3) Not a big deal but having the bands be the same colour as the data points (like in the 2006 graph) would be easier to read, especially with the green/bloc overlap and 4) Can you start the Y-axis at 0? frogg 09:49, 25 September 2008 (UTC)

I think the size of the graph is perfect as it is, but then my screen resolution is 800x600 so I may be untypical. The gridlines are also not hard to see for me... (again, this could be because of my lower screen resolution). I also agree that the graph should start at 0 on the y axis, though. As for having the bands be the same colour, honestly I think it's fine either way... if they're made colour, it could be a bit muddy trying to figure out what colours to use when 2 bands overlap. Esn (talk) 01:48, 26 September 2008 (UTC)

y=0 and 06 election added. galneweinhaw (talk) 14:25, 26 September 2008 (UTC)

I think this version looks better; the current version is too squished, and some of the symbols are weird shapes (it's not clear whether there were 2 polls with the same % or if it was one poll but the symbol was incorrectly drawn). Any possibility of keeping the other changes, but making it wider again? Esn (talk) 20:03, 26 September 2008 (UTC)

I'm not sure how to fix this more than temporarily. As more dates are added it will continue to get compressed. I'll look through some of the other symbols available that might be easier to distinguish, maybe making the points smaller will prevent run in also. The problem with widening the plot is that it automatically gets shrunk to 800x460 or something and gets blurry unless you look at it full-res. Regardless, I'll keep an eye out for the problem because I see exactly what you mean.

One thing that would help would be making that blank area on the left much smaller. How about moving the 2006 election %s to the very left, for example? Like in the 2006 graph. Also, the "legend" (which colour=which party) could be moved below the graph (or the top of the graph, like in 2006) to free up more horizontal space. It does seems that an 800px width is the limit (I would recommend making your image no wider than 800px, so that it doesn't get compressed and look blurry), but it's possible to use that space more efficiently. Never mind. When an image is posted in the main article, its width isn't restricted to 800px. Still, it's a good idea to make more efficient use of horizontal space. Esn (talk) 23:17, 26 September 2008 (UTC)

"would it be possible to write a more detailed description on the image profile page, including what the grey area is exactly and what calculations you used to obtain it?"... done! let me know if your lookiing for something more/different. -galneweinhaw (talk) 20:05, 26 September 2008 (UTC)

I finally figured out how to create an image of constant size =P so I'll be doing it at this size unless someone requests another (just let me know!) At least it's size won't be flying all over the place based on my eyeballing it anymore =P galneweinhaw (talk) 00:22, 3 October 2008 (UTC)

It's a good size. I think "Oct. 3" in the current version is actually "Oct. 2". Esn (talk) 00:40, 3 October 2008 (UTC)

Proposed new graph

Just having some fun learning this new graphing software... any use for a graph like this? File:PollsByResearcher.png galneweinhaw (talk) 05:28, 27 September 2008 (UTC)

Heh, that's pretty interesting. I wouldn't be against including something like that in the article, though as a thumbnail image, not full-size. There are more than 3 pollsters, though... how about Ipsos Reid, Environics, Strategic Council and Segma/Unimarketing? Although the three you included are the only ones giving daily updates. Esn (talk) 09:25, 27 September 2008 (UTC)

By the way, what graphing software is this? Esn (talk) 09:42, 27 September 2008 (UTC)

it's R (a command line statistical analysis program/language) with the ggplot package, which I have found out is pretty awesome =D galneweinhaw (talk) 21:09, 27 September 2008 (UTC)

Ouch

Two opinion polls published at the same day, one putting the liberal support at 31% and another putting it at 23%. If that's correct, the only thing it tells us is not to trust the polls. JdeJ (talk) 12:47, 17 September 2008 (UTC)

Yeah I have been watching those - if you check back and see Nanos vs EKOS they are always far apart on the Liberal numbers, as well as several other party numbers, too. These days, with most people hanging up on pollsters when they call, the polls are a long way from a random sample, as participants are very self-selecting. Retired people have more time for pollsters while working people don't - retired people are more likely to vote conservative in general. Also the questions asked may be quite different between the two firms. The classic example of this is if you ask people if they are in favour of higher taxes to pay for more welfare they will say 'no', whereas if you ask whether the government should provide more financial support to the poorest citizens they will say 'yes'. It is all in how the question is asked. In the end the only poll that counts is on Oct 14th. - Ahunt (talk) 13:38, 17 September 2008 (UTC)

They're not all phone polls anymore, though, so that helps. Online polls can prevent this somewhat by surveying a very large sample and then randomly selecting from that a sample that is more representative of the electorate (so as not to give more weight to the sorts of people who are online). Decima's polling is mostly done online (I've been surveyed by them twice in the last 4 months or so). --Llewdor (talk) 18:51, 17 September 2008 (UTC)

Hmm, stats are my area. Unfortunately a random sample of a non-random sample is still a non-random sample. Polling these days has some distinct limits, but it is all we have to go by. - Ahunt (talk) 12:07, 20 September 2008 (UTC)

It wouldn't be a random sample of a non-random sample. It would be a selected sample from the non-random sample to match the demographic breakdown of a random sample of the general population. --Llewdor (talk) 21:21, 25 September 2008 (UTC)

Wow, and this feat is repeated once again for the September 20 polls. Liberals either at 31 or 23%. Seems like nothing is changing, and the changes in the polls are due entirely to the different methodologies used. Esn (talk) 07:23, 22 September 2008 (UTC)

Do we know which one was closer in the last election? -Rrius (talk) 09:05, 22 September 2008 (UTC)

The one currently showing the Liberals at 31% (Nanos). "The firm received great praise when it correctly predicted the outcome of the 2006 Federal Election to within one tenth of one percentage point for the four major parties - a record in Canadian polling history." Esn (talk) 11:29, 22 September 2008 (UTC)

Coincidentally the Harris-Decima polls show the Liberals doing much worse. It is worth noting that Decima Research was "founded in 1979 by Progressive Conservative Party of Canada strategist Allan Gregg" - Ahunt (talk) 11:48, 22 September 2008 (UTC)

It seems there are two main groups, each giving very different results. Nanos (and to a lesser extent Ipsos Reid) show Liberal support to be much higher and Green support to be much lower than the other pollsters. All of the other polling companies, meanwhile, have very similar numbers showing Liberal support to be low and Green support to be high. Esn (talk) 02:51, 23 September 2008 (UTC)

Nanos uses a methodology different from the others. Nanos does only top-of-head polls - they don't list the options - and that's probably why they show higher Liberal support and lower Green support. If the Greens are a bit of a fringe party, then they simply wouldn't occur to many respondents when searching for answer. Also, Nik Nanos thinks that listing options encourages people who don't have a strong opinion to make up their mind and pick one right now, and he doesn't want to measure those people (since they're not likely voters, no matter what they claim). As the Green Party looks more and more like a mainstream party in this race (debate inclusion, attacks from the other parties), the Nanos numbers should rise to match the others. --Llewdor (talk) 21:19, 25 September 2008 (UTC)

Wow... and once again we've got Nanos showing the Liberals at 8% higher than Harris-Decima, after a week of not being much different than the others. Makes me wonder if there's some unique demographic that Nanos is getting that none of the other pollsters are... Esn (talk) 21:20, 3 October 2008 (UTC)

Remember, the margin of error is only vaid 19 times out of twenty. That Nanos today disagrees badly with Decima, EKOS, and Ipsos suggests they might have stumbled upon some rogue data. But there is the chance their unusual methodology has uncovered something the others are missing; it will be very interesting to see how and if the numbers converge next week. If Nanos did get rogue data, it will take a few days to work its way through their rolling average. Llewdor (talk) 23:57, 3 October 2008 (UTC)

Daily tracking polls vs. others

There are three companies -- Ekos, Harris-Decima, and Ekos -- doing daily tracking polls where each day they report the average results from the past 3 or 4 days; it seems to me that the table might be much clearer if these were separated out into a different table. Thoughts?

24.80.10.198 (talk) 21:39, 25 September 2008 (UTC)

Except, there are only 8 polls so far during the entire campaign that weren't from Ekos, Harris-Decima, or Nanos (which you meant). There's nothing wrong with daily tracking polls; they do tend to remove a lot of volatility from the numbers, but if someone wants more information they can visit the pollsters' sites. Remember, we're not reporting the news, here. --Llewdor (talk) 18:02, 26 September 2008 (UTC)

I removed the "separated by polling firm" section because all that needed to be done to make this information available was make the main table sortable. This allows readers to not only see the results by polling firm, but to see at what dates a certain party had highest or lowest support. Esn (talk) 22:42, 30 September 2008 (UTC)

Confidence Intervals

This may be a dumb question, but why do 95% confidence ribbons on the poll graph contain much less than 95% of the data points? There's been 52 polls since Sept 8 (not sure if all of them are on the graph), and, for example, for Conservatives, 13 of the points lie outside the 95% confidence interval ribbon. That means that the 95% confidence ribbon contains at most 75% of the observations... the difference seems large enough to be statistically significant. Is this an error, or am I misunderstanding the analysis? ikh (talk) 18:32, 27 September 2008 (UTC)

There are a couple reasons. First, they are not the 95% CI's for the combined polls, but instead Local Regression with the Regression's 95% CI. It is a common method for scatter-plots. The trend line is smoothed, so the 95% confidence intervals are on the smoothed data, not the actual data.

Second, even if it was a 95% CI of the combined data, it would not include 95% of the points, By combining polls, you add up the total sample of each of the polls, resulting in a smaller 95% CI = 0.98/SQRT(sample size). If we were plotting the trend of a single series (say EKOS) it would by definition have ~95% of it's points within the error bands (see the plot broken down by company...though again this is actually a regression). However, by combining polls, you are gaining accuracy by increasing the sample size, so your 95% error is smaller. Hope I am making sense =P galneweinhaw (talk) 21:32, 28 September 2008 (UTC)

I've been working on trying to get an actual combined poll line with it's CI (more work and calculations), but I don't think it will be as interesting or be more meaningful.

I added this under the plot as a description: "A combination of election period opinion polls during the 2008 Candian general election. The trend lines are a Local Regressions with α = 0.5 and 95% confidence interval ribbons."galneweinhaw (talk) 20:42, 28 September 2008 (UTC)

Graph position

IMO the graph should be at the top of the page, it provides a concise summary of polling trends. -- Gordon Ecker (talk) 03:56, 4 October 2008 (UTC)

I also prefer it at the top of the page, and believe that the chart is what most people would want to see.. The numbers don't mean much in a table without any indications of averages, trends, or statistical significance. The chart, on the other hand does in excellent job of aggregating the information and showing trends. The chart is better than anything you'll find in newspapers, which only report the results of one polling firm. The chart is the strength of this page and should be shown off. --Arctic Gnome (talk • contribs) 09:50, 4 October 2008 (UTC)

I think during the election, most people want to see the most recent polls. It is incomprehensible to me that people who have never been to this page would click the link on the main election page thinking, "boy, I can't wait to see the graph!" Someone clicking on a link for "opinion polling in the Canadian federal election, 2008" expect to see, well, polling data. People want to see what the parties are polling at in a way that is easy to make out. You cannot tell from the charts exactly what each party is hitting in each poll or, frankly, the exact value of any data point.

We had a perfectly acceptable compromise before, which was to have a thumbnail of the the graph to the right of the table of contents, and a full-sized one below the chart. The chart as it exists now is especially unhelpful in the ongoing race because incorporates such a long period of time. That keeps it from being very sensitive to polls. In other words, the graph won't look too much different in two weeks because the trendlines are so compact. -Rrius (talk) 10:47, 4 October 2008 (UTC)

I also prefer the graph at the top of the page - it gives a good summary of the data in the table. The table of polls itself is very hard to scan down and determine any kind of trend, especially when the different polling firms disagree so much with each other. - Ahunt (talk) 13:07, 4 October 2008 (UTC)

Same here. Makes no sense for such a comprehensive graph to be put at the end of an article, especially when it provides an excellent at-a-glance view of the progress of the campaign.GrahamNoyes (talk) 14:36, 4 October 2008 (UTC)

I also prefer the graph at the top. If you don't like it, one just has to scroll past a small graph. Now to get to the graph, one has to scroll down a very long page. Nfitz (talk) 14:59, 4 October 2008 (UTC)

I like it at the top too. It's like the summary that's on the top of many other pages. Plus, you are mistaken about the trends. As the graph becomes compressed, the trends will become more visually promenent because they are not stretched out across so much space on the screen =). I'm also not sure what you mean about not being sensitive to the polls. If the polls change so will the plot and the trend. If two polls go up and two polls go down though... it's gonna stay in the middle. And I thought that was kind of the point! galneweinhaw (talk) 15:46, 4 October 2008 (UTC)

It is actually not like the summary at the top of other pages, those have words, are above the TOC, and are at least nominally bound by guidelines on length. As to the sensitivity, I think I was looking at a different election's polling graph. It is the disadvantage of following so many at once. -Rrius (talk) 21:38, 5 October 2008 (UTC)

Ya, you're right about the summaries, bad comparison on my part =) Is there anything else about the graph you don't like that I can make better? galneweinhaw (talk) 22:22, 5 October 2008 (UTC)

• Seems we have consensus here - I'll move it. Nfitz (talk) 18:24, 4 October 2008 (UTC)

Belatedly, I'll say that I also support the graph being at the top. Esn (talk) 05:09, 5 October 2008 (UTC)

Graph size

I tried to shrink the graph a little, to 750px. Esn reverted me based on his assertion that it is now blurry in his configuration and his belief that this somehow contravenes the consensus above. I have no understanding of how a low res image like this could become blurry when reduced in size, but could someone throw me a bone here and take it down the size a little? -Rrius (talk) 04:30, 6 October 2008 (UTC)

No prob, I'll make it 750px and we can see how it looks (I've noticed blurriness on shrunk images too) galneweinhaw (talk) 01:01, 7 October 2008 (UTC)

I think the graph should be larger. According to the wikipedia guidelines, graphs should be 1000px on the shortest side. This is consistent with the graph for the 2006 election, for example. I understand some people have smaller resolutions, but others (like myself) have larger resolutions (1920x1200) and the details of graph are difficult to see. frogg (talk) 18:04, 8 October 2008 (UTC)

I'm happy to create the plot at whatever size there is a consensus for. I like a bigger plot than this myself. I was able to get an SVG version this morning, but the file size is huge =( galneweinhaw (talk) 16:14, 10 October 2008 (UTC)

Nanos Research

The Nanos numbers are often very different from all the other firms; are you sure that they are a legitimate polling organization? —Preceding unsigned comment added by 134.153.216.129 (talk) 19:31, 7 October 2008 (UTC)

See the discussion above. Nanos is very well respected and in past elections they have proven more accurate than other firms. - Ahunt (talk) 19:37, 7 October 2008 (UTC)

From the look of Nanos' methodology, they are going to produce statistics that are bias twards the Conservatives and the Liberals, for three reasons:

• by not prompting subjects with party names, they are lessening the similarity between their survey from an election ballot, and eliminating voters who might choose their party at the last minute when they hear or see the name;
• by not prompting subjects with party names, they are going to have fewer NDP and Green responses from people who might be publicly embarrassed by their choice or assume that the poller is only interested in the big two, and would only admit their support for a smaller party after hearing the poller give its name;
• by using phone-based surveys, they are biasing their results against youth voters, who often do not own a phone or own an unlisted cell phone; since youth are less likely to vote for fringe parties and for left-leaning parties, their results are even further biased;
• by asking subjects for their top two party preferences and only reporting the first given, they are changing the subject's interpretation of whether they think they need to give a strategic vote, which could bias the result either way for the left-leaning parties, but not for the Conservatives;
• by statafying results by region, they are biasing their results against urban voters, who tend to be more left-leaning while also not taking into account that it takes more votes in urban ridings to elect a given number of MPs.

Given the nature of their biases, it's no wonder that Nanos is reporting higher numbers for the Liberals and lower numbers for the Greens than any other firm. I don't think that we can take them seriously as a polling firm given their flawed methodology and their apparent bias. They should be moved off the main table and off the graph.— Preceding unsigned comment added by 134.153.216.129 (talk)

Those are all good points, however Nanos' methods are well within the accepted practices of polling in Canada and their results have proven more accurate than other firms in past elections. They are considered one of the leading polling firms in Canada, mostly because of their track record.

The main problem with asking people to pick a party or candidate from a list is that the results become distorted by the psychological effect of "primacy" whereby people will most often pick the first name mentioned, especially when they are undecided. I would argue for giving less weight to the other firms who use lists.

While it is true that phone surveys will not create a truly random sample because of cell phone owners and those without phones, internet surveys also suffer the same problems, they don't reach those without computers. Since older voters are less likely to have computers and younger voters are less likely to have home phones I would think that Nanos' numbers would generate a Conservative bias, not a Liberal bias.

As far as the other firms are concerned Decima Research has its own issues as it started out as a Conservative party polling organization.

In the overall picture polls are just that, polls. They may give some forecast of the election, but they don't determine who forms the government - that is decided by the election. This article is about election polling and Nanos is doing election polling. I don't see any reason to exclude companies just because their methodologies differ from other companies, especially when their track record is solid. - Ahunt (talk) 00:51, 8 October 2008 (UTC)

The arguments you make, 134.153.216.129, do make sense. But arguments pased on polling techniques will not cause a polling company to be included or excluded here: if Nanos is widely regarded as a reputable polling organization, then it ought to be included.

In any case, it is well established that Nanos (under its former name of SES Research) was the most accurate predictor of electoral results in 2004 and possibly 2006 as well. --Saforrest (talk) 17:56, 8 October 2008 (UTC)

The "asking canadians" poll

I realize that its inclusion may be controversial, so I'll just mention that at least one Toronto Star reporter called it a "market research firm", so it would seem to fit the definition for inclusion (even though its numbers seem rather out of whack). Esn (talk) 11:42, 14 October 2008 (UTC)

In reading through their website the reason the numbers look odd are twofold: 1. They include 18% undecided in their numbers, whereas the rest of the polls are of "decided voters" and 2. their about page says:

“ AskingCanadiansTM is a unique online community that rewards you for sharing your opinions. When you become a panellist, you join thousands of Canadians who have influenced as many as 70 companies and over 50 leading brands.

Join the panel today and start participating in online surveys that shape the products and services you know and love (or maybe hate).

For every online survey that you participate in, you'll automatically earn your choice of either Hbc Rewards bonus points or Aeroplan Miles. Or you can choose to be rewarded in other ways, like entries into contests and draws.

Participation is free and absolutely confidential.

”

To me this says that this is not a random sample poll, but a self-selected sample poll and therefore is not representative of the Canadian population, but only representative of the people who sign up with them. I don't think it belongs here with the other polls as the methodology is not the same. - Ahunt (talk) 11:57, 14 October 2008 (UTC)

They do however have this in their "FAQ": "Q. Why am I not always qualified to complete a survey? A. We are generally looking for panelists with specific characteristics for particular surveys (e.g. specific age group, gender, location, preferences, etc). This means that you will qualify for some surveys but will not qualify for others." So it does seem as if they try to sift through their available pool of people. I still think it should be included, because none of the other polling companies has identical methodology to any other; each one of them is different in some way. Most of the polling companies (except Nanos) don't call cellphone users, and they each ask somewhat different questions. Esn (talk) 12:16, 14 October 2008 (UTC)

If the 16% (not 18%) undecided voters are removed from their numbers, this is what you get: Conservatives 38.0%, Liberals & NDP 22.6%, BQ 9.5%, Green 7.1%. Those numbers are a lot closer to what some of the other polling companies are getting. Perhaps those edited numbers should be added back in, with an asterisk explaining that the 16% undecided number was removed? Esn (talk) 12:31, 14 October 2008 (UTC)

Some more info. "Established in 2005, AskingCanadians influential online survey currently includes panel of more than 130,000 members across Canada." It's owned by Delvinia Data Collection. Seems reputable enough for inclusion to me. Esn (talk) 12:37, 14 October 2008 (UTC)

So for now, I've added it back in minus the 16% undecideds. Esn (talk) 12:48, 14 October 2008 (UTC)

It isn't a matter of reputable or not - the polling is not scientific in that the people polled are self-selected, not randomly selected. I don't think it belongs here because the methodology isn't comparable to scientific polls. - Ahunt (talk) 13:17, 14 October 2008 (UTC)

I have removed this poll on the basis that it is non-scientific as the participants are self-selected. Statistically it lacks external validity and therefore does not represent the Canadian voter, it only represents the people who selected themselves for the poll - it is not generalizable. Removing the "undecided" doesn't help, as the ref does not support the numbers that were entered in the table then. I think we have to be careful here in what gets put on this page, especially today, as it can affect the outcome of the election. It is hard to justify non-scientific data in an article like this. I would request that rather than revert again that any disagreement be discussed here to reach a consensus before this suspect data is reintroduced. - Ahunt (talk) 14:15, 14 October 2008 (UTC)

I agree that the poll should not be included, as it is not scientific Uniqueuponhim (talk) 15:24, 14 October 2008 (UTC)

• Surely any poll that relies exclusively on on-line polling should be removed. There's a huge and obvious bias there. Not only does it require literacy, but it requires a computer. 68.179.114.217 (talk) 18:09, 14 October 2008 (UTC)

In that case, as per this article, all Angus Reid polls should be removed, since Angus Reid relies on online polling in the same way as the "asking canadians" poll does. Let's be consistent. Esn (talk) 09:14, 3 December 2008 (UTC)

The article makes it clear however that Angus Reid still works to ensure that they have a representative sample, while the "Asking Canadians" one uses strictly a self-selected sample. The trouble is that the other polling companies use phone calls, which, as your article explains, results in a very self-selected sample, as upwards of 80% of people called refuse to participate. Essentially this means that none of the polling done is totally scientifically valid and all of it should be disregarded. However the results speak for themselves, in general they turned out to be pretty close to the election results +/- the margins of error involved. You will notice that the "Asking Canadians" poll missed the winner's results by almost 6%. It wasn't very accurate. - Ahunt (talk) 13:14, 3 December 2008 (UTC)

My two posts from 14 October 2008 refute both of your points. "asking canadians" also uses a representative sample AND they were only 0.4% away from the actual Conservative election results. Esn (talk) 21:42, 3 December 2008 (UTC)

The election is long over and so the article is "historical" in nature now. If you want to put it back in, do so, but you shouldn't recalculate their numbers, that is too much WP:OR. - Ahunt (talk) 21:57, 3 December 2008 (UTC)

How is that WP:OR? It's just allowing their numbers to be compared with the others, which do not include undecided voters (the others pre-modify their numbers to take the undecided voters out, though they still give the % of undecideds in tiny print). Anyone can come to the same numbers by simply dividing by 0.84 (i.e. taking out the 16% undecideds). This is simple, basic math, and does not constitute WP:OR. Esn (talk) 00:37, 5 December 2008 (UTC)

Graph style

I think we should should switch to the style of the [[Image:CombinedPollsVotes-39Cdnelxn.png |graph]] on the 2006 page. -- Gordon Ecker (talk) 04:45, 15 October 2008 (UTC)

Accuracy of pollsters

There's an interesting table here about the accuracy of pollsters' final projections. I'll just add the AskingCanadians poll and remove the "poll of polls", and this is what we get:

(% represents total error from final result)

1. Angus Reid (Oct. 11) - 5.8%
2. Harris/Decima (Oct. 12) 6.2%
3. UBC Election Stock Market - 7.8%
4. EKOS (Oct. 13) - 8.4%
5. AskingCanadians (Oct. 10-13) - 9.2%
6 (tied). Strategic Counsel (Oct. 11) - 9.6%
6 (tied). Ipsos Reid (Oct. 11) - 9.6%
8. Nanos (Oct. 10-12) - 10.2%

So the most accurate pollster in the 2006 election was the least accurate this time around. Esn (talk) 01:06, 5 December 2008 (UTC)