For those watching the polls, Darrell Bricker, CEO of
Ipsos, and co-author, with John Ibbitson of
The Big Shift (which makes his suspect in
progressive circles) offers these cautions on
Twitter:
1. Follow along on what should be making pollsters nervous. Dealing with 2 big issues that aren't really getting discussed much:
2. First big issue is weighting to gen pop census. All do it (or say they do). But, doesn't match voting pop which has a different profile;
3. Better to weight to electorate but this changes every election. If there's big mismatch census/electorate possible source of error;
4. Second big issue is that all polls assume 100% turnout. Around 40% won't show up. VERY tough to separate these 2 groups;
5. Convention in Canada - but almost no place else - is that we report all eligible voters. Usually works, sometimes doesn't; and
6. If polls miss this time it will be because of meaningful difference between census, electorate; or big difference of turnout by party.
If I understand that, Mr Bricker is saying that most polls may be measuring too many people (mostly young people) who don't vote enough but who do favour the Liberals and NDP, and too few people (mostly older people) who favour the Conservatives. In other words the polls may be
biased, by simple mathematics, towards the Liberals or, at least, against the CPC.
By how much (if at all)? If the
bias is, say, 2 to 5% does that mean the race is a tie?