Sunday, August 23, 2009

Shockingly, the Iqaluit typo hasn't proven to be a game changer for the Liberals

Ipsos shows the biggest Tory lead since the coalition days:

CPC 39%
Lib 28%
NDP 14%
Green 10%
BQ 7%

It bears noting that every other poll published over the past month has had the parties neck-and-neck.



  • Welcome to that mystical "1 time out of 20"

    By Blogger Glen, at 1:03 a.m.  

  • Ever notice that the "1 time out of 20" always corresponds to a poll whose result we don't like?

    By Anonymous The Invisible Hand, at 1:13 a.m.  

  • Polls in the last couple of weeks have Cons up 3 points, this one is up 5.
    Not that far out of whack.
    Within the margin of error, could be 36:31

    By Blogger wilson, at 2:00 a.m.  

  • Actually, much to my chagrin, it also seems to correspond to the ones that place the Liberals far out in the lead.

    By Blogger Glen, at 2:01 a.m.  

  • I guess you could say they're wiping our ass with these poll results.

    By Anonymous Iqualuit, at 8:52 a.m.  

  • Well considering its still the summer ... the Conservatives are the only ones campaigning ... and no one cares right now ... I'd say thats about right.

    By Anonymous Anonymous, at 9:08 a.m.  

  • Welcome to the "we don't like the poll results" Liberal spin.

    1 time out of 20
    it's the summer
    it's not awful in the margin of error

    Anyone want to take "the only poll that matters is election day"?

    By Anonymous Deb, at 9:11 a.m.  

  • The line we that hasn't been slapped up yet:

    I'm waiting for the Nanos numbers before I decide anything.

    By Blogger Tarkwell Robotico, at 9:29 a.m.  

  • Michael Ignatieff has to be on his knees praying that this is an outlier.

    We'll find out Thursday with Ekos.

    I haven't taken Nanos polls seriously for some time.

    By Anonymous Drake, at 9:35 a.m.  

  • Na-na-na-NA-na!

    Sorry. I'm having an Eric Cartman moment.

    By Blogger Dr.Dawg, at 9:59 a.m.  

  • RE: Welcome to that mystical "1 time out of 20"

    The "1 time out of 20" is kind of a misnomer. It doesn't mean that 1 out of 20 polls will produce ridiculous results, it means that Ipsos or whoever bases their margin of error on a 95% confidence interval.

    So there is some chance that the results lie beyond the margin of error. However, probability distributions look like a bell curve. Of the possible results outside the stated margin of error, the vast majority are only just barely beyond it (eg. the sample mean might be 3.2 points higher than the actual population mean).

    In other words, it would be extremely unlikely (assuming Ipsos uses a decent methodology, etc.) for a result showing an 11 point Tory lead if reality (what you would get if you polled everybody who will vote in the next election) was actually a tie.

    By Blogger french wedding cat, at 11:06 a.m.  

  • "Shockingly, the Iqaluit typo hasn't proven to be a game changer for the Liberals"

    And neither has Iggy.

    He's your boat anchor, so enjoy.

    By Anonymous Anonymous, at 1:59 p.m.  

  • H2h, how does this relate to the notion of a "rogue" poll, which I confess is what I thought the 1 out of 20 indicates; too long since my stats courses, I guess!

    If I remember things correctly (uhh, it's like Woodstock...if you can remember it, you weren't there!), there's a relationship between standard deviations from the mean and degree of confidance, and I beleive that 2 s.d. typically covers 95% of the sample...and (theoretically, at least) the population. 3 s.d's is about 98.7 %, leading to a higher degree of confidance. But one never sees a poll described as 32 out of 33 times, or whatever that works out to...why not?

    By Blogger Party of One, at 6:14 p.m.  

  • A rogue poll is simply a poll that shows different results than all the other polls coming out around the same time (eg. every poll Zogby does). It may do so for a variety of reasons - though it is possible (albeit unlikely) by random chance.

    The reason we never see polls using 99% confidence levels is that 95% is something of an industry standard. The margin of error would be too large to be all that instructive if you used a 99% standard (not every field has as low standards as politics though).

    For instance, lets say a poll put the Tories at 35% with a margin of error of 3 points at the 95% significance. You could also say that there was a margin of error of 4 points at 99% confidence. In other words, they are somewhere between Harper 2004 and a majority government.

    The only way to get more conclusive results would be to poll even more people. This is expensive, and doesn't really pay off. The only poll that matters for a pollster's credibility is the one they run before election day (Ekos polled ~45,000 people just to get it close - and Angus Reid still beat them).

    Newspapers (who often commission polls) don't really want super accuracy anyway. Nor really do the readers of polls. We lament it, but at the end of the day, we love horse-race coverage. A margin of error of 3 points (plus the myriad of different polling techniques) or so is perfect for this - just enough variation to talk about change (even when there isn't any), and just enough accuracy that people can make semi-conclusive statements like "_____'s policy is really flying with elderly women".

    Everybody wins (well except people who wanted issue coverage, but they are probably a bunch of cranky whiners anyhow).

    By Blogger french wedding cat, at 11:12 p.m.  

Post a Comment

<< Home