Wednesday, January 18, 2006
I can't stand it
I knew the race would tighten. My guess is Monday will show something like 36 - 33 in favour of the Tories. But it's still agony watching it happen. It's agony because I have no trust in the political judgement of my Learned Friends in Ontario. None. Zero. I wouldn't trust Ontario voters to choose the prize bull at a county fair, let alone a g**d*** Prime Minister.
I know Occam and others think polls are stupid, but they're not. They're just data. Sure, the data may be biased or inconclusive or misinterpreted by eager partisans... but they're still just data. They're generally true... or, at least, true-ish. As Terry Pratchett might say, they're true for a given value of true.
That being said, something looks wrong with these data:
Decima: 37 - 27 for the Tories, +/- 3.1%, 19 times out of 20
SES: 37 - 30 for the Tories, +/- 2.9%, 19/20
CTV: 42 - 24 for the Tories, +/- 2.5%, 19/20
Is it different voter support on different polling days? Which is to say, is voter support shifting quickly from one day to the next? Decima is Jan. 12-15, SES is Jan. 14-16, and CTV is Jan. 14-16. Given that SES and CTV are the same dates, the answer here is "no".
Or are these just different points along the "plus-or-minus" line? Maybe we can make these make sense. Look just at Tory support.
Decima's Tory range: 33.9 - 40.1
SES' Tory range: 34.1 - 39.9
CTV's Tory range: 39.5 - 44.5
And the Grits:
Decima: 23.9 - 30.1
SES: 27.1 - 32.9
CTV: 21.5 - 26.5
It's possible to make the Tory number work out--something in the 39.5 to 39.9 range fits all 3 polls. The Libs don't quite fit: with only Decima and SES considered, we can make the Liberal number anything from 27.1 to 30.1, but the CTV number is incompatible. The highest CTV number (26.5) and the lowest SES number (27.1) don't mesh.
Either one of these polls is the "mythical" 20th poll, or there are local complications (e.g. is one province sampled differently in SES and CTV polls?) or different ways of handling the undecideds. We could split the difference, and say "the mid point between SES and CTV is 26.8% for the Liberals", but that's pretty fudgey and likely wouldn't pass muster with most statisticians.
Regrettably I am no statistician. If one is out there, please help me understand these polls...
I know Occam and others think polls are stupid, but they're not. They're just data. Sure, the data may be biased or inconclusive or misinterpreted by eager partisans... but they're still just data. They're generally true... or, at least, true-ish. As Terry Pratchett might say, they're true for a given value of true.
That being said, something looks wrong with these data:
Decima: 37 - 27 for the Tories, +/- 3.1%, 19 times out of 20
SES: 37 - 30 for the Tories, +/- 2.9%, 19/20
CTV: 42 - 24 for the Tories, +/- 2.5%, 19/20
Is it different voter support on different polling days? Which is to say, is voter support shifting quickly from one day to the next? Decima is Jan. 12-15, SES is Jan. 14-16, and CTV is Jan. 14-16. Given that SES and CTV are the same dates, the answer here is "no".
Or are these just different points along the "plus-or-minus" line? Maybe we can make these make sense. Look just at Tory support.
Decima's Tory range: 33.9 - 40.1
SES' Tory range: 34.1 - 39.9
CTV's Tory range: 39.5 - 44.5
And the Grits:
Decima: 23.9 - 30.1
SES: 27.1 - 32.9
CTV: 21.5 - 26.5
It's possible to make the Tory number work out--something in the 39.5 to 39.9 range fits all 3 polls. The Libs don't quite fit: with only Decima and SES considered, we can make the Liberal number anything from 27.1 to 30.1, but the CTV number is incompatible. The highest CTV number (26.5) and the lowest SES number (27.1) don't mesh.
Either one of these polls is the "mythical" 20th poll, or there are local complications (e.g. is one province sampled differently in SES and CTV polls?) or different ways of handling the undecideds. We could split the difference, and say "the mid point between SES and CTV is 26.8% for the Liberals", but that's pretty fudgey and likely wouldn't pass muster with most statisticians.
Regrettably I am no statistician. If one is out there, please help me understand these polls...
