http://www.theglobeandmail.com/news/politics/much-ventured-for-only-slight-conservative-polling-gain-in-2010/article1844239/
Perhaps Éric Grenier could engage in an in depth discussion of the accuracy of his numbers.
A review of the actually polling results will show maybe 2 - 4 % accuracy (19times out of 20, of course). However, they are notoriously inaccurate even by the time you get to the Provincial level.
The Ekos results Nationally for the last number of poll | ||||||||
| Date | Con | Lib | NDP | Grn | Blk | Other | Size | Margin of Error |
| 16 Dec: | 32.0% | 26.5% | 17.1% | 10.9% | 10.6% | 3.1% | 2135 | 2.1/dt> |
| 9 Dec.: | 33.7% | 29.2% | 14.4% | 10.4% | 9.8% | 2.5% | 2153 | 2.1 |
| 25 Nov.: | 33.3% | 27.1% | 16.6% | 9.5% | 9.5% | 4.0% | 1696 | 2.4 |
| 14 Nov.: | 29.4% | 28.6% | 19.3% | 10.7% | 9.3% | 2.7% | 1587 | 2.5 |
Although it is hard to read Grenier has:
| NAT | Con | Lib | NDP | Grn | Blk | Other | Size | Margin of Error |
| 34.8% | 29.2% | 15.6% | 8.7% | 10.2% | ? | ? | ? |
These are outside Ekos margin of error for the Cons, Libs, Grns and there is no number for "other". For the parties his numbers add up to 98.5, leaving no more than 1.5 for other - a review of Ekos suggests that this is questionable by about 1.5, but then how do you adjust the other numbers, e.g if you take it off the Con's you get 33.3 which is exactly inline with Ekos.
In order to 'seat project' he must have numbers at the riding level, even the Regional level is not sufficient. However, it is very difficult to do predictions even at the Prov level with the general National polls released:
The recent Ekos Poll, 9 Dec.'10, has the following accuracy:
REGION
British Columbia: +/- 7.6
Alberta: +/- 7.8
Saskatchewan/Manitoba: +/- 8.4
Ontario: +/- 3.6
Quebec: +/- 3.5
Atlantic Canada: +/- 8.1
Nationally the accuracy was: +/- 2.1
The above is typical of all such Ekos polls and it would be very surprising if they were out of line for all the other national polls by reputable pollsters.
It would be very interesting to see just exactly how Éric Grenier gets from those type of margins of error at the Prov level to seat projections.
excerpt: Lloyd MacILquham cicblog.com/comments.html
