Thursday, April 7, 2011

Some Poll Analysis

These are some broad observations on the polling numbers in this election.

First off, Nanos, the only company releasing daily numbers, is significantly out of sync with the other polling companies.  Of the five pollsters that have released numbers in the last week, only Nanos has the Conservatives ahead of their 2008 election result.   Nanos also has the Greens at about half of the support that other pollsters have for them.   The numbers for the Liberals, NDP and Bloc released by Nanos are comparable to the other companies numbers.

Second, the Liberals are clearly ahead of their 2008 result.   Of the seven pollsters that have released polls held after the election started, only two had the Liberals below their 2008 result and one really is tied because it was rounded to the nearest whole number.   In meaningful regional breakouts - by this I mean polls with more than 300 respondents in a province - the Liberals are marginally up in Ontario and down in Quebec.  In BC I can not be certain of where they are at because there has been only two post writ polls with meaningful BC numbers and they cancel each other out.

Third, the NDP is down from 2008.   Not only are they down, they could be more evenly spread out over the country.   17% instead of 18.2% means the NDP would be below not only their 2008 result but their 2006 result.   In the regional numbers the NDP is clearly down in Ontario and up in Quebec.  That sort of shift in votes will lose them seats in Ontario but not gain seats in Quebec.  This trend could mean the loss of five to seven seats in Ontario and the gain of one seat in Quebec.

Fourth, the Bloc is down from 2008 in national numbers but up in the regional numbers.  How this works I am not sure, but I certain there is a math that makes it make sense.   Because of the conflicting numbers I have no idea what this means for their support.

Updated four hours later

There has been an Environics poll out today that is close to the same results as the Angus Reid poll also out recently.   They have the NDP at 20% and 21% respectively.   This is the first good news for the NDP at the polls.     The Environics Liberal number is low, only 25%, which is markedly lower than others.

The Environics put the Conservatives at 38% which is closing in on Nanos country, but still way outside of what it should be.  

The quoted margin of error on polls is the range in which 95% of the time the results should fall.   The margin of error is a bell curve which means the quoted value is the most likely one and the extremes of the curve on either end are very unlikely.   A 3.1% margin of error at 95% is 2.6% at 90%, 2% at 80% and 1.8% at 75%.   This means the values will fall within +- 1.8 percentage points of the value expressed.   This is three times out of four.

The margin of error also changes based on the result achieved.   For the Greens the margin or error is much narrower, in a 1000 response survey the standard margin of error is +-3.1 percentage points, for the Greens at 8% it is +- 1.7 percentage points.

So the point of all of this is that there is some major problem with the pollsters since they are not narrowly close to each other in their results most of the time.   I have no idea who to believe and take seriously and I can only look at what trends I can tease out of the data.

No comments: