these large regional margins of error suggest that sudden swings, especially when they slice across the spectrum, are mostly due to sampling error. but, that doesn't give their earlier results greater stability. i look at some of the trendlines in these regional polls and see absolutely nothing whatsoever of value out of them; they're flying around all over the place by 10-15%. i really wish the firms would get meaningful sample sizes if they're going to do weekly polls. the result does make the ndp numbers seem the most volatile, but that has to be interpreted in the context of how the ndp's immediately apparent volatility affects the numbers for the other parties. for example, the last nanos poll gave the conservatives a 10% bump on the ndp in ontario in a way that is inconsistent with the other polling. on first glance, the conservative numbers look roughly stable while the ndp numbers look volatile. but, that can't actually be true - it's actually impossible.

the abacus polling was another internet poll. it seems like the internet polls are consistently polling the ndp a little higher. i suspect this is a consequence of bias in the sample, due to how the poll is being taken. the forum polls seem to continue to be understating their margins.

the phone polling is mostly placing them tightly together, but that may also be technology dependent because younger people are going to be harder to reach over the phone. it's well understood that phone polling tends to overweight older people. this has tended to balance out in recent elections with higher conservative voter turnout. the fact that the conservatives are having a hard time getting over 32 in phone polling - and with questionable regional polling - suggests they ought to be extremely concerned.

www.cbc.ca/news/politics/canada-election-2015-grenier-polltracker-aug19-1.3195182