Latest national poll median date: October 20
Projections reflect recent polling graciously made publicly available by pollsters and media organizations. I am not a pollster, and derive no income from this blog.

Monday, October 21, 2019

FINAL 2019 Projection: \o/

The following polls have been added to the model:
Mainstreet's 10/19-20 national poll (current weight among national polls: 30%)
Research Co's 10/18-20 national poll (current weight among national polls: 9%)
EKOS' 10/17-20 national poll (current weight among national polls: 11%)
Forum's 10/18 national poll (current weight among national polls: 3%)
Nanos' 10/19 and 10/20 national polls (national results only; current weight among national polls: 6% and 18% respectively)
I have also updated ON regional adjustments using the breakdown from Campaign Research's 10/16-20 national poll, as well as Atlantic Canada and BC regional adjustments using the breakdown from Innovative Research's 10/15-17 national polls.
For a full list of included polls, see previous Projection posts.

And I'm finally done! This projection update includes polls that are favourable to the Liberals (EKOS, Forum) and polls that are negative for them (Mainstreet, Nanos). Because the latter have more weight thanks to their recency and sample sizes, the projection moves against the Liberals.

As you can see, the final projection gives an insignificant edge to the Liberals, which is flipped from the "quickie" projection I posted on Twitter just after the final poll was released. The Liberals gained roughly 1 seat, at the expense of the NDP. What accounts for the difference?
- In order to avoid Nanos and Mainstreet being the only polls with a midpoint date after Oct. 19, I used the Oct. 19-20 roll-up from the EKOS poll, and inferred Oct. 17-18 results, which were included as a separate poll. This slightly increases the weight on EKOS, which helps the Liberals and hurts the Tories. (As it turns out, the inferred results were given a weight below 0.5% by the model, so it doesn't matter if they're slightly off.)
- Updates to regional adjustments for BC, ON and Atlantic Canada help the Tories and Liberals and hurt the NDP and Greens.

Projection as of the latest national poll (midpoint: October 20)
LIB - 133.2 (32.1%)
CON - 132.7 (33.6%)
BQ - 35.3 (7.1%)
NDP - 34.5 (18.1%)
GRN - 1.2 (5.9%)
IND - 0.6
PPC - 0.5
If you're new to my 2019 projections, view key interpretive information here.

Without the turnout adjustment, the projection would be:
LIB - 134.2 (31.6%)
CON - 126.0 (32.1%)
NDP - 39.1 (19.1%)
BQ - 36.1 (7.1%)
GRN - 1.4 (6.9%)
IND - 0.7
PPC - 0.6

I very crudely estimate the confidence intervals of the projection as follows:
80% confidence intervals
LIB 95-175
CON 95-170
BQ 20-50
NDP 15-50

95% confidence intervals
LIB 85-190
CON 85-185
BQ 15-55
NDP 12-55

Seats changing hands since the last projection:
- In NL, LIBs get St. John's South--Mount Pearl back from the NDP.
- In NL, CONs regain Avalon from LIBs.
- In PE, CONs regain Egmont from LIBs.
- In QC, the BQ regains Beauport--Côte-de-Beaupré--Île d'Orléans--Charlevoix from CONs.
- In QC, the BQ regains Saint-Maurice--Champlain and Argenteuil--La Petite-Nation from LIBs.
- In ON, CONs regain Peterborough--Kawartha from LIBs.
- In ON, the NDP regains Oshawa from CONs.
- In MB, the NDP gets Elmwood--Transcona back from CONs.
- In BC, CONs regain Cloverdale--Langley City from LIBs.
- In BC, the NDP retakes Nanaimo--Ladysmith from GRNs.

Below is a map showing the seat-by-seat leading party. Please keep in mind that my projection uses riding polls only if they are not behind a paywall - i.e. very few riding polls outside QC are used. Moreover, I only take into account factors described in my methodology description. So I am tying one hand behind my back - no "the NDP isn't winning Acadie--Bathurst because of their organizational problems in NB and because this year's candidate does not share a last name with long-time MP Yvon Godin."

As you can see, the Liberals have a 5-seat lead in ridings projected ahead. The sums of probabilities are roughly equal because the Liberals win more tight races. Below are the 75 ridings where the gap between the top two parties is less than 5%:

A gap of less than 5% means that there is at least a 30% chance that the projected winner won't actually win. I expect to get almost half of these ridings (say, 30-35 out of 75) wrong.

Next, here are the 55 ridings where the gap is between 5% and 10%:

In "standard" two-way races, these projections should be correct 70-85% of the time. However, in three-way races, the probability of error significantly increases. This is also the case in special situations, such as the Territories, for which we don't have any polling. Therefore, I would expect to get 15-20 of these wrong.

Of the remaining 208 ridings, the standings are:
91 CON
82 LIB
20 BQ
14 NDP
  1 GRN
Of these:
- 56 have a gap between 10 and 16 points (85-95% chance of being correct in a standard two-way race),
- 58 have a gap between 16 and 22 points (95-99% chance of being correct), and
- 94 have a gap above 22 points.
Therefore, I would expect to get 10-15 of these wrong. Adding up the expected numbers of mistakes, there should be 55-70 of them; in other words, I expect 79-84% accuracy on an average night. If the polling averages are right, accuracy will likely be slightly higher, while if there is a significant polling error, then accuracy could be slightly lower.

Here's a potential pre-Speaker seating chart of the House of Commons:

No comments: