Friday, September 25, 2015

Model Update: Regional Breakdowns and Riding Polls

The election is approaching, and data is accumulating - there have now been at least 85 riding polls in 66 ridings since the start of the campaign (excluding the two polls leaked by the NDP). Furthermore, some regional breakdowns have been published. (In 2011, the Tories were up more in the Toronto area than elsewhere in ON, and since there were more tight races in the GTA, noticing this fact helped make my projection on this blog more accurate.)

It is therefore time to fine tune the seat projection model! Below, I review the evidence that swings within the country's six polling areas (Atlantic, QC, ON, MB/SK, AB, BC) are not uniform. For each polling area, Part 1 discusses sub-regional breakdowns, and Part 2 discusses riding polls. The changes that I am making to the projection model are in bold.

Here are some principles that guide my use of these polls:
- I compare the data to the projection based on national and provincial polls conducted during roughly the same period.
- In the case of regional breakdowns, I make no adjustment if all parties are within 3% of the projected vote share.
- In the case of riding polls, I make no adjustment if all parties are within 10% of the projected vote share. That is, I'm trying to be parsimonious, with the idea that a projection model isn't trying to correctly project every riding, but rather that errors in different directions should roughly cancel out.
- I also make no adjustment if an adjustment won't matter unless there is an extreme swing. For example, if the projection in a riding is 45-40-15, and the poll says 50-45-5, there is no adjustment since party #3 is unlikely to win anyway.
- Adjustments do not make the projection fully match the poll - they only go part of the way. How far depends on whether there is corroborating evidence, and on whether I know of a good reason for the deviation (e.g. star candidate, by-election, etc.).
- Any regional adjustment must be cancelled by an equivalent adjustment elsewhere in the same polling area. For example, to balance out adding 5 points for the Greens on Vancouver Island, I subtract 1 point for them elsewhere in BC.

I will issue a projection with the new model shortly.

Atlantic Canada

1. CRA published a poll of about 1,500 Atlantic Canada respondents with a breakdown for the four provinces, conducted during August.
a) The notable deviations occurred in PEI, where the Liberal change from 2011 was 13.7 points worse than in Atlantic Canada as a whole. Of course, PEI sample sizes are small, but the CRA poll in May also showed a Liberal change 9.7 points worse in PEI than in Atlantic Canada. The Tories also slightly underperform in PEI (by an average of 4.8 in the two polls, which is statistically insignificant), so the gainers are the NDP and the Greens. Due to the small sample sizes, the adjustment won't go as far as the poll is suggesting: in PEI, it's -9 for Liberals, -2 for Tories, +7 for NDP, +4 for Greens. This cuts the Liberal/NDP gap in half in Charlottetown (but the Liberals still have a big lead there), and makes Egmont a really tight Liberal/Conservative race.
b) No big deviation in NB relative to Atlantic Canada as a whole, but for several elections, NB has tracked ON more closely than it has tracked other Atlantic provinces. And indeed, assigning the ON changes to NB would have yielded even closer numbers to what CRA shows for NB. However, demographically, NB is much more similar to its Atlantic brethren than to ON. What to do? This is where I use another source of information: riding polls.

2. I know of 8 ridings polls in 5 ridings here.
- Three riding polls in two ridings in NB: St. John--Rothesay and Fredericton. In both cases, the Conservative-Liberal difference is about 5% better for the Tories than what the Atlantic uniform swing would yield. At the same time, using the ON uniform swing would yield a 16% difference in favour of Tories. Therefore, to address both these polls and point 1b above, for NB, I will use a weighted average of 2/3 Atlantic swing, 1/3 ON swing.
- In Avalon, ex-Liberal independent incumbent Scott Andrews appears to draw from all three main parties, and is only at 19%. Thus, no adjustment is necessary.
- In Cumberland--Colchester, two polls in the past week both have ex-Tory Liberal candidate Bill Casey with about 50% of the vote. This is way above the uniform swing projection, and is almost entirely at the expense of the NDP. I am going with the poll that's closer to the uniform swing projection: in Cumberland--Colchester, +10 for Liberals, -10 for NDP.
- In Central Nova, Peter MacKay's riding, two polls conducted 2-3 days apart tell radically different stories. Mainstreet has Central Nova swing like the rest of Atlantic Canada, leaving the Tories ahead by 6 points. However, Environics has the Liberals with a 27-point lead! Here, I am giving roughly 1/3 weight to Environics: in Central Nova, +5 for Liberals, -5 for Conservatives. Indeed, if Bill Casey is worth 10 points to the Liberals, then it seems reasonable that Peter MacKay would be worth at least 5 points to the Tories in a neighboring riding.

In NL and NS, I will make uniform changes that, combined with the above, will net out to zero (in terms of votes, of course) for each party.

Québec

1. The only regional breakdown is from a CROP poll in mid-August; Mainstreet also conducted a poll of the Island of Montréal in late August. Taken together, they do not offer evidence that an adjustment is needed.

2. Seven riding polls here:
- In Papineau, last week's Mainstreet poll has the Conservatives and NDP outperforming uniform swing, and the Liberals and Bloc underperform it. It's quite hard to see why the Conservatives, who got 5% in 2011, would suddenly be at 11%. It's also surprising that Trudeau would not only miss out on the party leader bump, but actually suffer from a penalty. On the other hand, it's quite easy to see separatists voting NDP to spite Trudeau. In Papineau, +5 for NDP, -5 for Bloc.
- This week's Mainstreet poll found a similar pattern in the nearby riding of Ahuntsic--Cartierville. In Ahuntsic--Cartierville, +5 for NDP, -5 for Bloc.
- In the Conservatives' sole target seat in the Montréal area, Mount Royal, they are performing worse than expected. In Mount Royal, -5 for Conservatives, +5 for Bloc.
- In Richmond--Arthabaska, popular incumbent André Bellavance, elected as a Bloc MP in 2011, is not running. The Bloc vote collapsed according to a Mainstreet poll. In Richmond--Arthabaska, +10 for Conservatives, -10 for Bloc.
- The three ridings of the Saguenay--Lac-St-Jean region were polled by Segma last week. The Tories underperformed their provincial swing across the board. The biggest beneficiaries were the Bloc in Jonquière and Lac-St-Jean, and the spoils were split between the Liberals and NDP in Chicoutimi--Le Fjord. In the three Saguenay--Lac-St-Jean ridings, -5 for Conservatives, +5 for Bloc (except in Chicoutimi--Le Fjord, +2.5 apiece for NDP and Liberals).

Ontario

1. We have four polls with regional level data for ON: Forum's GTA polls on Aug. 7 and Sept. 16-17, which each had 1,000+ respondents, Mainstreet's Sept. 8 poll of the city of Toronto, which had 2451 respondents, and Mainstreet's mega ON poll this week, which had a whopping 4,610 respondents. (Ipsos also released a regional breakdown for ON in its national Sept. 5-7 poll, but the sample sizes are too small to be meaningful.) The four polls all suggest that:
- In the city of Toronto, the Tories are holding up much better, and the Liberals are not up by as much. (This is also consistent with 13 out of the 15 riding polls conducted in Toronto - the evidence for this is very strong.)
- In the 905, the Tories have dropped by slightly more than in ON as a whole, and the NDP is doing slightly better.
Moreover, Mainstreet's mega provincial poll suggests that the NDP lost plenty of support in Northern ON, where the Liberals are up by even more than elsewhere. Taking all this into account, the changes are:
- For the Conservatives, +5 in the city of Toronto, -3 in the 905, and -1 in the rest of southern ON to balance things out.
- For the NDP, +2 in the 905, and -5 in northern ON.
- For the Liberals, -5 in the city of Toronto, +5 in northern ON, and +1 in the rest of ON to balance things out.

2. There have been at least 33 riding polls in 23 ridings in Ontario. They are, on the whole, consistent with the adjustments above. While I could further adjust some ridings, these changes would mostly cancel out in terms of the seat count, so I will leave things alone.

Manitoba/Saskatchewan

1. EKOS' weekly poll provides separate numbers for MB and SK, and large sample polls conducted by Mainstreet and Angus Reid in August did the same. Moreover, Insightrix conducted a SK-only poll, also in August. Taken as a whole, there is no strong evidence that MB and SK are moving differently relative to 2011.

2. Eight polls in six ridings here. The four ridings that were polled just once require an adjustment:
- Winnipeg South Centre: -5 for Liberals, +2.5 apiece for NDP and Conservatives. Liberal lead much smaller in poll than in projection.
- Saint Boniface--Saint Vital: -5 for Liberals, +5 for NDP. Again, Liberal lead smaller than in projection.
- Regina--Lewvan: +5 for Conservatives, -5 for NDP. This reduces the NDP lead; Tories actually ahead in the poll.
- Saskatoon West: +5 for Liberals, -5 for NDP. NDP lead much smaller in poll than in projection.

Alberta

1. Two Insights West poll provided a regional breakdown. No adjustment needed.

2. There were 12 riding polls in 11 ridings: people are very curious where the cracks in the Conservative shield will appear! The ones below caught my attention:
- Edmonton Centre, a three-way race with uniform swing, was in fact a two-way Conservative/NDP race in an Environics poll conducted last week due to Liberal underperformance, which mostly benefits the Tories. In Edmonton Centre, +5 for Conservatives, -5 for Liberals.
- In St. Albert--Edmonton, the ex-Tory independent incumbent Brent Rathgeber is drawing proportionally from all major parties and far from winning. This is a very similar situation to Avalon. No adjustment needed.
- In Lethbridge, where much loathed incumbent Jim Hillyer (who hid from a reporter by running into a bathroom in 2011) decided to run in a neighboring (and safer) riding, the Tories appear to be recovering from his negative effect in 2011. In Lethbridge, +5 for Conservatives, -5 for Liberals.
- In Fort McMurray--Cold Lake, the Liberal candidate that ran a strong second in last year's by-election is running again, and seems to have carried some momentum. In Fort McMurray--Cold Lake, +10 for Liberals, -10 for Conservatives.
- In Calgary Confederation, a particularly strong Green run in 2011 does not seem to be repeating itself, allowing the Liberals to be very competitive this time. In Calgary Confederation, +5 for Liberals, -5 for Greens.

As the above changes roughly cancel out in terms of votes, no adjustment is necessary in the 30 other ridings.

British Columbia

1. Two Insights West polls and a Mainstreet poll agree: swings in Metro Vancouver closely mirror swings in the rest of BC. On Vancouver Island, however, the Greens have increased their support more than elsewhere, mainly at the expense of the NDP. On Vancouver Island, +5 for Greens, -5 for NDP. To balance this out: in the rest of BC, +1 for NDP, -1 for Greens.

2. There have been 16 riding polls in 13 ridings in BC. Something quite odd is going on: the Liberals are underperforming what they would get under uniform swing in ALL of them. There might be a house effect at play here: all 16 polls were conducted either by Environics or by Insights West. The one riding where the difference is especially large is Vancouver Granville. There, while uniform swing puts the Liberals about 20 points ahead, both Environics polls have them behind the NDP (12 points in August, 6 points this week), with the Tories also in the mix. I will go with about 2/3 of the adjustment corresponding to the average of the two polls: in Vancouver Granville, +5 for NDP, +7 for Conservatives, -12 for Liberals. This makes Vancouver Granville a three-way toss-up.

Yukon

The Environics poll this week confirmed the Liberal lead in the Yukon, though it is a bit smaller than with a national uniform swing. The big difference is that a strong Green candidate is not running this time, and his votes seem to all go to the NDP, which is competitive as a result (but still 10 points behind the Liberals). In the Yukon, +15 for NDP, -15 for Greens.

No comments: