Saturday, September 5, 2015

First Projection of 2015: Tories Still Lead, but...

Finally, it's time for the campaign's first projection! I have the Tories marginally ahead of the NDP in the projected seat count:

CON - 126 (30.1%)
NDP - 122 (32.0%)
LIB - 89 (27.9%)
GRN - 1 (5.1%)
BQ - 0 (3.9%)

The Conservative lead, however, is just about as fragile as it gets:
1. The 4-seat margin is extremely slim.
2. The NDP actually leads in the popular vote.
3. The Tories have the edge only because I am adjusting vote shares, as described below, in order to account for voter turnout. Without this adjustment, the NDP would have 126 seats, and the Conservatives would be on 119; the Liberals would tie the Tories for second place in the vote share.
4. Ontario, an almost perfect three-way tie during much of August, is showing movement toward the Liberals: all five polls from this week put the Liberals at least 7 points ahead of the NDP in Ontario. If Ontario's anti-Conservative vote coalesces, things would get really ugly really fast for the Tories. (Not Ignatieff-ugly, since the Tories have a solid base in the rural West, but potentially Dion-ugly.)

A few methodological notes about projections for this campaign.

- I am not taking regional numbers into account until later in the campaign. Regional numbers are important: in 2011, the observation that the GTA was moving more strongly away from the Liberals toward the Tories than the rest of Ontario helped me put the Tories closer to a majority than most other projection blogs. (Update: This means that there's a decent chance that the Liberals are up in the GTA more than elsewhere in Ontario this time. If so, they would win a few more seats than projected.)

- As mentioned above, I am adjusting poll numbers (except for polls explicitly applying a likely voter model) to account for the greater propensity of Conservative supporters to vote. Roughly speaking, I assume that these electors are 10% more likely than supporters of other parties to cast a ballot. (Update: I should note that this is roughly in line with the results of Angus Reid's Institute's likely voter model according to their most recent poll.) This boosts the Conservative vote share by about 1.5%, and depresses the Liberal and NDP vote shares by about 0.5% each. It is important to note that this adjustment may be a bad idea (Conservatives may be demoralized this time), or it may not be sufficient (an even greater adjustment would have been required in 2011).

- The way in which I'm weighing polls has changed slightly. The base weight of a poll is still the square root of its sample size. Any polling day within 4 days of the most recent polling day of any poll is given full weight, and the 9 previous polling days are given weights from 0.9 to 0.1. For example, since the current most recent day of polling is September 2, the most recent EKOS/iPolitics poll, conducted from August 26 to September 1, is discounted by a factor 6.4/7: the days from August 29 to September 1 have full weight, while the days from August 26 to 28 have weights 0.7, 0.8 and 0.9.
The formula above (which might be changed if the frequency of polling increases) implies that there is a two-week window during which polls count. If the same polling firm has multiple such polls, only the most recent one is given full weight; the second most recent one gets weight sqrt(2)-1, the third gets weight sqrt(3)-sqrt(2), etc.

- The change in the background colour is not a blog redesign! That colour was solid blue due to the Conservative majority, and was close to solid blue throughout the 2011 campaign because the Tories were on the cusp of a majority during the whole campaign. The colour is now white with a very slight tinge of blue because we're much closer to a tie than to a Conservative majority. I'm guessing the colour will stay close to white for a while, but it could be red, orange or blue (or still white) at the end of the campaign. When the background is almost white, as it is now, the title of the blog will be in the colour of the leading party; if the background gets dark enough, the title will return to being white.

Finally, a note on the frequency of updates: it looks like I'll still be quite busy during the rest of the campaign, so I may not post several updates a day as was the case in 2011, and I will typically not be commenting on individual polls. I might also not get around to doing a mapped projection until the end of the campaign. Nevertheless, I intend to post at least once (and probably 2-3 times) a week in September, and every two days or so in October.

No comments: