Thursday, April 26, 2007

Mean Value

With Decima and the Strategic Counsel releasing new polls yesterday, I figured I'd average them in with the most recent Ipsos and SES numbers:

CPC 35.25%
Lib 30.25%
NDP 15.75%
BQ 9%
Greens 9%

Hard for anyone, outside of maybe the Greens, to be overly excited about those numbers.

Labels:

29 Comments:

  • Looks like the Green party has not yet been decimated in support, as its outgoing strategist, Dan Baril, predicted in his swan song, kiss and tell blogs, after the Dion-May joint announcement.

    By the way, if you offer a bunch of advice, and no-one takes it, can you still call yourself the "former GP strategist"?.

    By Blogger The Anonymous Green, at 1:17 PM  

  • Maybe we should start doing this will all of the polls. Just average them out and come up with what the numbers pretty much always seem to be at the end of the day.

    By Blogger Red Tory, at 2:35 PM  

  • Is it my paranoid imagination, or do the polls comissioned by Canwest-Global consistently show a higher level of Conservative support than the other polls?

    Scrutineer

    By Blogger grogey, at 3:44 PM  

  • No it's not, and it's true both with their old company Compas and now with Ipsos-Reid. Canwest also managed to commission the only poll since 1999 that didn't have more support than opposition to same-sex marriage.

    Polls done for Canwest are great for rallying the troops, but as an indicator of public opinion they're pretty much worthless. There's a good reason, I'm sure, why Ipsos has released two polls this month showing Harper within reach of a majority and no election was called. He's not stupid and he certainly isn't relying on their results.

    Nodice has a list of all polls since the last election. http://www.nodice.ca/elections/canada/polls.php

    I averaged each pollster's results (of course some have conducted many more than others). The results are, from left to right, for BQ, Con, Green, Liberal and NDP

    Decima Research 9.6 34.3 9.5 29.9 16.1

    Ipsos-Reid 9.1 37.1 6.5 31.1 15.4

    SES Research 10.3 35.0 6.0 31.7 17.3

    Strategic Counsel 10.1 35.6 8.8 30.1 15.8

    Decima has conducted 27 polls since the election. Ipsos 16, SES 6 and Strategic Counsel 14.

    If you look at 2007 polls only, it's:

    Decima Research 8.7 33.7 10.1 30.6 14.3

    Ipsos-Reid 8.6 37.3 7.9 31.6 14.3

    SES Research 10.0 34.5 6.5 33.0 16.5

    Strategic Counsel 9.6 35.2 10.2 31.2 14.0

    By Blogger Reality Bites, at 4:52 PM  

  • The average of all their 2007 polls, by the way, is

    9.0 35.1 9.2 31.2 14.4

    Based on that, Strategic Council is closest to the average. This doesn't mean, of course, that they are right, or that anyone is. The most any poll can ever be is an accurate reflection of how all Canadians would say answer the same question asked by the pollster. If you conduct a large enough sample, with an unbiased question, right before an election it is likely to be pretty close to the results. Polls conducted now are just a fun game.

    By Blogger Reality Bites, at 5:06 PM  

  • Polls mean nothing right now. There will not be an election for at least several months.

    If you take into account the margin of error, other than a few blips here and there (budget, grit love-in), the polls haven't changed at all.

    By Blogger Andrew Smith, at 9:05 PM  

  • Bah, I can't take polls seriously - sure, they do indicate how people feel about a topical issue, granted. But in terms of who they like vote-wise? There's only one poll, and that's on election day.

    Well... they are fun, though. I don't mean to rain on any parades.

    By Blogger Jason Bo Green, at 12:16 AM  

  • There's one thing we can all agree on. The Liberals are at 30%.

    By Blogger Jason Cherniak, at 8:34 AM  

  • David Suzuki calls Baird's Green Plan a scam. Shouldn't Canadians have a referendum on Kyoto?

    What about the fumbling incompetence of the prisoner-transfer scandal?

    Shouldn't Canadians have an opportunity to censure harper and his crew? Just as they censured the old guard of the Liberals?

    By Blogger JimTan, at 11:49 AM  

  • Looks like the Green party has not yet been decimated in support,

    That's because most of the pollsters prompt respondents for Green.

    By Blogger WJM, at 11:50 AM  

  • That's because most of the pollsters prompt respondents for Green.

    I was speaking relative to prior polling periods.

    Presumedly the methodology has remained the same. So, comparisons would be valid over time.

    Your point is that the absolute numbers may be incorrect - but that is a different issue.

    By Blogger The Anonymous Green, at 12:40 PM  

  • That's because most of the pollsters prompt respondents for Green.

    Source?

    By Blogger James Bow, at 12:43 PM  

  • Source?

    The scripts that have been posted on some of the pollsters' own web sites, and various email discussions.

    By Blogger WJM, at 1:02 PM  

  • I was speaking relative to prior polling periods.

    How far back are you taking that?

    They were polling in the 10% range prior to the last election, too.

    By Blogger WJM, at 1:03 PM  

  • wjm

    "That's because most of the pollsters prompt respondents for Green."

    You had better have a credible source when you question the integrity of others.

    By Blogger JimTan, at 1:11 PM  

  • You had better have a credible source when you question the integrity of others.

    I'm not questioning their integrity, I'm questioning the soundness of their methodology.

    I have no doubt they are reporting the statistics accurately, as they have found them. It's the method they use to obtain those statistics that's problematic. That says nothing about their integrity at all.

    Two examples of what I would contend are bad method:

    If the federal election was being held tomorrow, do you think you’d be supporting the (ROTATE LIST) Liberal candidate in your area, Conservative candidate in your area, the NDP candidate in your area, or the Green Party candidate in your area or (QUEBEC ONLY) Bloc Quebecois candidate in your area?

    If a Canadian federal election were held today, which one of the following parties would you vote for [ROTATE PARTIES] the Liberal Party, the Conservative Party, the New Democratic Party, [Quebec Only] the Bloc Québécois, the Green Party, or another party?


    I would contend that prompting for Green inflates the Green polling vote-intent above the real level of support the party enjoys. If they prompted for "Rhinoceros Party" or "Marxist-Leninist", they'd find "support" out there for those parties, too.

    In fact, prompting for party AT ALL is probably unwise. SES, for example, seems to not prompt, leaving the vote-intent question open-ended.

    And the sources I've discussed this issue with are very credible, yes.

    By Blogger WJM, at 1:38 PM  

  • How far back are you taking that?

    Within the context of my original post, it would be safe to assume I was speaking about prior to the Dion-May announcement, April 12th or thereabouts, which caused all of the media and blogging interest, and doomsday scenarios.

    It doesn't appear to me that they have gone down for the GP since pre April 12th. I don't have the averaged surveys, similar to the methodology used here, to compare specifically, however.

    By Blogger The Anonymous Green, at 2:58 PM  

  • wjm

    "If a Canadian federal election were held today, which one of the following parties would you vote for [ROTATE PARTIES] the Liberal Party, the Conservative Party, the New Democratic Party, [Quebec Only] the Bloc Québécois, the Green Party, or another party?"

    What prompting? The reader is given 6 choices. The Green Party is listed as 5th while the 'other' category is the last choice.

    This seems normal. The reader does not have to choose one of the five named parties.

    The only improvement I can see is to list the five parties in alphabetical order. Nonetheless, I don't think that there is any trick here.

    What source are you talking about? Did hey actually say that only the GP was being prompted. Or, was it the LPC (1st name)?

    By Blogger JimTan, at 4:04 PM  

  • What prompting? The reader is given 6 choices. The Green Party is listed as 5th while the 'other' category is the last choice.

    No, the choices are arranged randomly. Not so long ago, the Green Party was not mentioned explicitly in any national poll, and strictly fell under the "other" category.

    This seems normal. The reader does not have to choose one of the five named parties.

    Except that support for "other" parties is not reported at all. The Greens have consistently polled above their actual vote totals - it's happened in the last two federal (and two BC) elections.

    The only improvement I can see is to list the five parties in alphabetical order. Nonetheless, I don't think that there is any trick here.

    No, the appropriate methodology is not to mention any party by name, and hence let the respondent indicate her preferred party without prompting. Open questions are less subject to potential bias, since the pollster does not give a predetermined range of choices to the respondent.

    What source are you talking about? Did hey actually say that only the GP was being prompted. Or, was it the LPC (1st name)?

    Huh? All of the major parties are prompted because they are explicitly mentioned by the pollster. The "prompting" is clear since it's mentioned in those polls with publicly released questionnaires.

    By Blogger Josh Gould, at 4:18 PM  

  • What prompting? The reader is given 6 choices. The Green Party is listed as 5th while the 'other' category is the last choice.

    That is what the industry calls "prompting", as opposed to the "unprompted" type of question SES, and possibly others, use.

    When you lay out the choices, people are more likely to pick one of them.

    This seems normal. The reader does not have to choose one of the five named parties.

    No, but when the interviewer does name them, the interviewee is less likely to say "Don't know" or some other non-party response.

    The only improvement I can see is to list the five parties in alphabetical order.

    That would definitely NOT be an improvement. If you have to prompt, and you shouldn't, then the prompting should be rotated or randomized. Otherwise the first and the last items on the list will be "skewed".

    Nonetheless, I don't think that there is any trick here.

    Neither do I.

    What source are you talking about? Did hey actually say that only the GP was being prompted. Or, was it the LPC (1st name)?

    No, the people I know in the industry have talked about the in-built bias problems with prompted questions, and the noticeable effect this has on the Green's poll figures.

    By Blogger WJM, at 4:21 PM  

  • Wjm said

    “No, the people I know in the industry have talked about the in-built bias problems with prompted questions, and the noticeable effect this has on the Green's poll figures.”

    Actually, a list of names is what a voter will see on his ballot on voting day. So, the standard polling method used by Decima, Ipsos and Strategic is a more accurate way of simulating the process of voting. No?

    I would argue that the SES method is actually unfair to he GP. Yes?

    By Blogger JimTan, at 4:33 PM  

  • Actually, a list of names is what a voter will see on his ballot on voting day. So, the standard polling method used by Decima, Ipsos and Strategic is a more accurate way of simulating the process of voting. No?

    What people do on voting day is really quite different from stating their voting intention weeks or months before an election. On what basis are you arguing that those companies employ a more accurate methodology?

    I would argue that the SES method is actually unfair to he GP. Yes?

    Unfair in what respect exactly? The fact that SES reports consistently lower (arguably more "reasonable") support for the Greens indicates, if anything, that their methodology is superior. Is it fair to inflate one party's numbers with question bias?

    By Blogger Josh Gould, at 4:49 PM  

  • For Josh,

    Hang on a minute. I want this to be a learning experience for wjm. I'll come back to you if he doesn't reappear.

    By Blogger JimTan, at 5:01 PM  

  • Actually, a list of names is what a voter will see on his ballot on voting day. So, the standard polling method used by Decima, Ipsos and Strategic is a more accurate way of simulating the process of voting. No?

    No.

    I would argue that the SES method is actually unfair to he GP. Yes?

    If by "fair", then, you mean a method which artificially, and, if you're a GP supporter, cruelly, inflates your apparent popular support.

    A "fair" method is one which improves accuracy.

    By Blogger WJM, at 10:55 PM  

  • Hang on a minute. I want this to be a learning experience for wjm.

    I love learning experiences! Whaddya got?

    By Blogger WJM, at 10:55 PM  

  • Josh said

    “The fact that SES reports consistently lower (arguably more "reasonable") support for the Greens indicates, if anything, that their methodology is superior. Is it fair to inflate one party's numbers with question bias?”

    That’s the classic situation of the answer looking for the question. I am arguing that you haven’t proved that it is not the other way around. That is, SES is deflating the GP %.

    Your negative bias about the GP has a priori led you by the nose.

    Reality Bites has provided 2007 averages for the four companies. Plot the averages.

    There are two groups. Decima and Strategic (Cluster A) are closely grouped together (10.1% and 10.2% average). SES (Cluster B) is grouped together with Ipso (6.5% and 7.9%). Immediately, there is a problem with the hypothesis that different methods matter. You have similar results from two methodologies in Cluster B.

    There are two issues. A) You have to prove which method is more accurate if you think that the difference matters. B) In the first place, is there a significant difference?

    How did your source(?) explain it to you?

    By Blogger JimTan, at 1:12 AM  

  • That’s the classic situation of the answer looking for the question. I am arguing that you haven’t proved that it is not the other way around. That is, SES is deflating the GP %.

    Your negative bias about the GP has a priori led you by the nose.


    No, my negative bias about sampling methodology has led me to prefer open questions to closed ones, since the former involve considerably less "priming" of the answers by the pollster. The best questions are the simplest ones in which the respondents are free to give any response they like.

    In any case, the inherent flaws in all of these polls are the lack of reporting concerning the non-response rate and the number of undecided. Without those numbers, especially the latter, it's difficult to get a clear picture of what's going on.

    By Blogger Josh Gould, at 4:59 PM  

  • I'm happy because liberal did not got the first place.

    By Anonymous satin shirt, at 8:45 AM  

  • By Anonymous Anonymous, at 2:50 PM  

Post a Comment

Links to this post:

Create a Link

<< Home