December 08, 2008


So, about that 62% majority...

Will they be changing their name to the "49% minority?"

(h/t SDA)

Labels: , ,

August 23, 2007


Dogs know what to do with (media reporting on) polls: Alberta edition

There's been some media coverage of a recent poll by Cameron Strategy (anyone heard of these guys before?) which has the Stelmach-led PCs falling to 32% support, down from 54% seven months ago. According to Daveberta, the full results (with January number in parenthesis) are:

PC – 32% (54%)
Liberal – 16% (16%)
NDP – 11% (9%)
Alliance – 5% (3%)
Unsure/Won’t vote – 36% (18%)

Sound really bad for Stelmach... but wait a minute, this looks strangely familiar. Why, it's one of my favourite poll-torquing techniques: leaving in the undecideds, so as to drag down the apparent support for the party you're gunning for!

Here's the real support levels, for both August and January:

PC - 50% (66%)
Liberal - 25% (20%)
NDP - 17% (11%)
Alliance - 8% (4%)

So there's still a significant drop, but nowhere near as bad as it's being made out to be. In fact, Stelmach is slightly above what Klein got in the last election.

Labels: , , ,

November 03, 2006


The least worthless poll so far

If you've been following the Alberta Conservative leadership race, you probably know that the publicly-released polls consistently show Jim Dinning in the lead, with Lyle Oberg a fairly close second, and everyone else well behind.

Sounds like a two-way race, right? Wrong. Even a casual examination of the polling methodology reveals that they are simply asking Alberta residents in general who they like best, when the race will be decided entirely by the small fraction of Albertans who bother to buy a five dollar party membership in order to vote.

Polls like the one above essentially measure nothing but name recognition. Candidates like Dinning (who has been campaigning for the Premier's chair for close to a decade) and Oberg (who gets himself lots of free media via the Garth Turner method) do well in them, simply because the general public has actually heard their names.

A good example of how this method fails is the current federal Liberal leadership race (which happens to end on the same day as the Alberta PC one). Shortly before the delegate selection meetings, a similar poll showed Ken Dryden in the lead with 19% of the vote, Michael Ignatieff a distant third with 10%, and Gerard Kennedy waaay behind with 4%. Dryden, of course, was a famous NHL goalie, while Ignatieff was a university prof who has been out of the country for most of the last 30 years and Kennedy was an Ontario cabinet minister with little exposure outside of that province. However, when the ballots were counted, Dryden got less than 5% of party members' support, while Ignatieff received 30% and Kennedy finished third with 17%.

The Next Alberta Premier blog has an excellent analysis of what's wrong with these polls and the various factors which will make the actual results quite different.

So, what would be a good poll? So far, there's been only one poll that's even remotely worth considering. It was conducted by the Progressive Group for Independent Business, and was restricted to only those respondents who were PC Party members, and thus actually eligible to vote. The results were as follows:
Lyle Oberg: 17.2%
Ted Morton: 17.2%
Jim Dinning: 16.0%
Mark Norris: 8.1%
Dave Hancock: 7.5%
Ed Stelmach: 3.0%
Victor Doerksen: 1.3%
Gary McPherson: 0.7%
Alana Delong: 0.2%
Undecided: 16.1%
Wouldn't Say: 12.7%
Of course, this poll still has significant problems. For starters, it did not include any respondents from northern Alberta. More importantly, a leadership race means thousands upon thousands of membership cards will be sold, and so all the existing members at the beginning of the race will only be a minority of the total by the time it ends. As the media is starting to acknowledge, the race will be won by the campaign that sells the most memberships and gets them out to vote.

Nonetheless, by limiting their poll to those Albertans who are actually party members, the PGIB eliminated the single biggest problem with leadership race polling. This earns their poll the honour of being... The Least Worthless Poll So Far (tm).

Labels: , , ,

January 21, 2006


Dogs know what to do with (blatant manipulation of) polls

Andrew Coyne has commented on how the Globe & Mail buried their poll results when the Conservatives had a big lead, but made them the front-page headline when that lead dropped sharply. Now that The Strategic Counsel has released the full details, the integrity bar drops even lower.

Click here, scroll down to page nine, and look at the "N equals" row, where the sample size is listed. Notice how it's 1,500 people all throughout the month of January... except for the most recent day, when the sample suddenly drops to 1,000.

Why did they do this?

Well, remember that each SC daily number with a sample size of 1,500 is the sum of the last three days of polling, with 500 people each. If the last poll is only two days (January 17 and 18), it means that they threw out the January 16 result a day earlier. And by coincidence, the Conservatives' big 18-point lead happened on... January 16. This means that the polling day which put the Tories way up was conveniently dropped a day earlier than it should have been, which conveniently allowed the Globe to loudly proclaim they had a big drop in support.


See also:
Dogs know what to do with (media reporting on) polls, part 1
Dogs know what to do with (media reporting on) polls, part 2

Labels: , ,

December 13, 2005


Dogs know what to do with (media reporting on) polls, part 2

Even if the mainstream media doesn't have a liberal bias, it definitely has a Liberal one. Sinister Thoughts brought to my attention a Globe & Mail article which makes it look like the NDP are doomed to lose their Ontario support to the Libs.

The article carries the sub-heading "New Democrats' support level drops to single digits" and gives the numbers as Liberals 40%, Conservatives 24%, and NDP 9%. However, you may have noticed that those numbers only add up to 73%. This means that (unless the Green Party has suddenly jumped into second place with around 27%) the undecideds weren't factored out, which is very unusual for this kind of poll. It looks to me like the Globe left the undecideds in just so they could claim that the NDP was in "single digits."

Don't worry, Dippers, we Tories know how to empathize with those who are victims of shoddy poll reporting. All the more reason why we should work together to get rid of the Grits...

Update: Apparently, the 40-24-9 figure was from the "Which Party has the most momentum towards a federal election?" question, with the undecideds left in. So it looks like this probably isn't an example of the Globe & Mail being biased... just incompetent.

Labels: , ,

September 19, 2005


Dogs know what to do with (media reporting on) polls

Two new polls came out this weekend:

Leger Marketing: Lib 40%, Con 24%, NDP 15%, Bloc 13%
Strategic Counsel: Lib 35%, Con 28%, NDP 17%, Bloc 13%

Guess which one is getting all the media attention?

(Hat tip: Bound By Gravity)

Labels: ,

This page is powered by Blogger. Isn't yours?