these results are only different from the consensus in quebec and bc. in both cases, ekos is polling the conservatives much higher and the liberals much lower. it happens to be that these are also the places that have three or four way races.

passive consumers need to understand that pollsters don’t just collect opinions and send them out. they modify the results to fit the census, and there’s some interpretation in the process. that could be a part of the difference you’re seeing.

i’m also not sure that the blended sample is the best idea, because of the way that undecideds are being measured. and, i’d kind of argue against unlike aggregates for the same reason. aggregate live interview with live interview, and ivr with ivr, but don’t aggregate online samples with ivr and live interviews. if the ivr is truly measuring undecideds in such a way that reduces the sample size and inflates the conservatives, and you take more ivr in the sample, than you would expect inflated conservative numbers in the end result.

it’s not enough to do the “one of these things is not like the others” skit and rule ekos out. but, i think this experiment with ivr and live interviewer is maybe pulling out a bit of a bias, and destroying the ability to adjust for it by mixing the data up. that, i think, is a methodological flaw.

www.ekospolitics.com/index.php/2015/10/and-the-winner-is-we-dont-know/#comment-59737

i just want to add that the actual results of the mainstreet poll are

lib 34
con 30
ndp 19
und 10

and, once again, you see the relationship that shows up with this “dithering progressive” effect over ivr, and it’s subsequent inflation of the conservatives due to a decrease in sample space. and, it again absolutely nails the conservatives at a 30% flatline, which has been true for months and months.

--

angus reid is also an online panel.

the shy tory effect is bunk. but, we seem to be seeing a “dithering progressive” effect in the ivrs – it is absolutely consistent.