Thursday, April 7, 2011

Poll Weighting and Projection Roundup

We are already through 1/3 of the campaign (time flies!), and I'm finally done with accelerating my poll depreciation formula. This was done gradually since the start of the campaign, and took longer than expected since polls were less forthcoming than I thought. Polls up to 3 days old instead of 7 receive full weight (where a poll's age is the difference between its midpoint date and that of the most recent poll), and the threshold for a poll to receive some weight has been reduced from 27 to 12. In the last week of the campaign, I may decide to further accelerate the depreciation if the frequency of polls warrants it.

With the current formula, 12 national polls were taken into account for the latest projection (shares don't add up to 100% due to rounding):
Nanos 4/4-6: 12%
Ang R 4/4-5: 16%
EKOS 4/4-5: 12%
Env 3/30-4/5: 11%
Nanos 4/1-3: 6%
H-D 3/31-4/3: 10%
Léger 3/30-4/2: 17%
Nanos 3/29-31: 2%
EKOS 3/28-31: 6%
Forum 3/26-27: 4%
Abacus 3/24-28: 2%
H-D 3/24-27: 1%

As you can see, the polls released added today account for more than half of the weight. It is therefore rather striking that the projection has not changed at all.

At the start of the campaign, I noted six other websites that also offer projections based on an average of polls. (Again, there are links to all of these sites on the left.) All of them have made an update this week, and here are the results. In parentheses are the date of the most recent polling on which the projection is based, and the numbers that I got based on comparable polls.

150 C, 74 L, 51 B, 33 N by Canadian Election Watch (4/6)
155 C, 71 L, 50 B, 32 N by (4/6, 150-74-51-33)
150 C, 76 L, 47 B, 35 N by Too Close To Call (4/5, 150-75-51-32)
150 C, 75 L, 48 B, 35 N by democraticSPACE (4/4, 150-74-51-33)
150 C, 75 L, 51 B, 32 N by Riding by Riding (4/3, 151-73-51-33)
156 C, 69 L, 52 B, 31 N by CalgaryGrit (4/3, 151-73-51-33)
150 C, 74 L, 51 B, 33 N by LISPOP (4/2, 151-73-51-33)

Average: 151.6 C, 73.4 L, 50 B, 33 N

As you can see, there is broad consensus over the numbers, but a few seats' disagreement makes the difference between a majority and a minority.

No comments: