Saturday, April 2, 2011

Poll Tracker - April 1st, 2011

There have been three new polls since I last updated the poll tracker. Two Nanos daily tracking polls and an Ekos poll. The biggest difference between the two polls is the level of support for the Green party, which tracks higher in the Ekos polls than it does in the Nanos polls. I suspect that this has something to do with the methodology used; Nanos polls do not prompt the poll takers for a party, whereas the Ekos polls do. I'm not sure which methodology is more accurate since not prompting would probably get a better idea of what parties are actually on voters' minds, but when it comes to the polling station they will be prompted when they see the ballot. Nevertheless, I won't make a judgment call on the accuracy of a polling method and will average the polls together equally (except that the Ekos poll had a larger sample size).

The Conservatives are holding steady around 38% in my averaging and have been for the entire past week, the Liberals has slowly tracked up a few percentage points but this is partly the result of that odd Nanos result three days ago that I mentioned. Tomorrow, that day's results will be factored out and the Liberal support will drop 2-4% while the NDP will rise 2-4%. I trust that the media will make a big fuss out of what essentially is just statistical noise.


A quick note on my averaging technique; in order to reduce the effect of statistical noise on the polling results I use a relatively (for lack of a better word) conservative averaging method. To determine the value for a new day, I determine the average of the polls for that day and all previous days and take a weighted average of them all, weighing the days depending on the number of respondents and how long ago those respondents were polled.

The net effect of this is to reduce the noise, but I'm not entirely comfortable with it because it assumes that the changes in political support from one day to the next will be small or non-existant. In most cases that may be true, but if there were a significant event which shifted support one way or another, my polling trending would miss it at first.

2 comments:

Calgary Junkie said...

Re-posting, as it mostly agrees with what is written here about Nik's numbers:

The Nanos numbers (based on a 3-day rolling average) are hard to predict. For example, when the Libs got their 4 point bump Thursday morning, it could only be because they had a huge result (around 38 % ?) from the Wed night sample of 400. So I fully expected the Friday morning number to show another increase. But it showed a drop !?!

Anyway, Nik's number tomorrow will drop off the Libs excellent Wed night sample of 400, and insert the Sat night sample of 400. So if the Libs do poorly tonight, there should be like a 3 point drop for them.Similarly I expect about a 1 point drop for us.

But don't quote me on these predictions. It could be that Nik uses some kind of formula to smooth out the results.

Eric said...

Yes, I still find it amusing that some columnists continue to hype the Nanos results even though they clearly are the result of a statistical aberration. See Stephen Maher's column in The Chronicle Herald as an example.

Post a Comment