i don't agree with this analysis. the mainstreet poll....it asked a different question. you've got differences in methodology across the different polling firms, making this kind of aggregate...it's useful, but you need to be careful. ivr, telephone & internet seem to be the three types and they have very different biases to correct for.
i don't have access to numbers to crunch. but it seems to be that the forum polls, specifically, are underestimating their margin of error. the most recent data is comparable to the one three polls ago, where the two in the middle are the same. and, it's similar when you move backwards a little. the electorate really doesn't seem to be that volatile right now. i mean, the undecided questions are consistently demonstrating that harper has almost no room to grow, yet the polling is consistently wavering around 5-6%. if he really had that much of a swing, it would come out in the undecided polling. rather, it just looks like more error exists than is being acknowledged. there are some reasons that i'm skeptical about ivr polling. but, the ekos polls are also ivr and haven't change much relative to themselves, within the margin of error - which i think is more along the lines of what a pollster ought to expect in july with no major reasons to shift voting intentions. at the least, you'd expect these to demonstrate the same trends. that they're not indicates something wrong with one of them. again, i think forum is underestimating their error.
net polling, i think, is even worse. you can't tell what is sampling error and what is a change in intentions. you can't even tell if people are filling out the results multiple times.
i don't see any clear trend at all over the last two months.
www.cbc.ca/news/politics/poll-tracker-harper-mulcair-in-close-race-as-campaign-begins-1.3176101
James in Kanata
@Jessica Murray
But the inescapable facts are:
* The Mainstream poll is the only one to reveal what questions were asked. Depending on the question, the results can be skewed, which is the case when an institution like the Star sponsors a polls and says "make sure the Conservatives are behind". Its called propaganda and the Soviets were masters of it
* The margin of error was the lowest because the sample base was 5 times the others. A Nanos poll showing each party the same with a 5 point margin of error actually legitimizes the Mainstream poll. Add and subtract the Nanos margin of error and you have the Mainstream one but the reverse is not true
Bottom line, there's no escaping the merit of the Mainstream poll and that company has an excellent record, unlike the others which predict
"Ignatieff surges ahead"
"Dix to be Premier of BC"
Jessica Murray
@James in Kanata
the mainstream poll was likely accurate in terms of what it reported. it reported 38% conservative amongst decided voters, which was about 80% of the sample. however, to translate that into results that are comparable to the other polls you need to scale it down by a factor of 0.80 to account for the 20% undecided. that takes the conservatives down to 30.4% of all voters in the mainstream poll.
it also suggested that almost all of the undecided vote is wavering between the ndp and the liberals.
in totality, there is nothing inconsistent about the poll. it's not an outlier.
now, i'd actually argue that the question they asked is better, if you're looking for a sort of a quarks & gluons type analysis. but, just dumping it into the other "leaning" type polls without accounting for it being a fundamentally different type of poll is bad methodology and it's going to inaccurately skew the results of the aggregate. the way he does this is in such a way that it fades out over time, so it will go away. but, he shouldn't repeat that error.
the nanos polling i've seen has put the conservatives in the 30-32% range - which is probably about right. a citation would be useful. it might be the same issue.
to be concrete...
the mainstreet poll reported:
cons: 38%
ndp: 27%
libs: 25%
grn: 6%
bloc: 4%.
...of the 80% of people that were decided.
the actual results were approximately:
cons: 30.4%
ndp: 21.6%
undecided: 20%
libs: 20%
grn: 4.8%
bloc: 3.2%
...based on napkin math that assume the whole number results that they reported.
generally, a statistician would want to distribute that 20% relatively evenly. but, we have plenty of polling on the topic, and we know that we can't do that - you couldn't reasonably give the conservatives more than a percent or two of it.
again: this is probably the most accurate poll because it measured undecideds rather than forcing them to pick an option they're not really decided on. i actually agree with that. it just wasn't reported well. and grenier is making an error with it that he really shouldn't be making.
i'm actually a good example of the results of the poll. if you forced me to pick, i'd pick the ndp. but, it's not really an accurate reflection - the truth is i'm not decided and could vote liberal in the end, if they make an effort to stop pushing fiscal conservatism and come out against the tpp. that would inflate the ndp's numbers a little at the expense of the liberals. i think that's what we're seeing - people on the left swing are leaning ndp, but we're just a tad uncomfortable about diving in for whatever reason.
if the ndp can get 12-15% of that, they win a majority. if it splits evenly, it's another four years of harper. and that's going to be his strategy for the election.