Seat Projections

Pre-Debate Seat Projections

Posted on by CalgaryGrit in 2011 Federal Election, Federal Politics, Seat Projections | Leave a comment

Up until now, I’ve taken the approach that the polls are rather meaningless, as the electorate still isn’t tuned in to this campaign. There’s no need to curl up in a fetal position in the shower just because your favourite party is down 10 points in Atlantic Canada.

But consider this the last free pass. Voters traditionally tune in after the debates, and there’s no reason to think things will be any different this time. If the polls aren’t moving by this time next week, it will be time to start worrying.

As the above trend-line shows, there has been little movement in the polls since last week:

Nanos (April 8-10, n = 982 phone): CPC 41%, Lib 30%, NDP 15%, BQ 8%
Decima (April 7-10, n = 1018 phone): CPC 40%, Lib 28%, NDP 15%, BQ 8%
Ipsos (April 5-7, n = 1001 phone): CPC 41%, Lib 26%, NDP 19%, BQ 9%
Ekos (April 4-7, n = 2555 autodial): CPC 36%, Lib 28%, NDP 17%, BQ 8%
Angus Reid (April 4-5, n=2031 online): CPC 38%, Lib 27%, NDP 21%, BQ 8%
Environics (Mar 30-April 5, n = 968 phone): CPC 38%, Lib 25%, NDP 20%, BQ 8%

Running Average: CPC 38.8%, Lib 27.4%, NDP 17.7%, Bloc 8.8%, Green 6.2%

This marks a slight narrowing of the gap from last week, but the emphasis is on slight. At the rate we’re going, it would take a 4-month campaign for the Liberals to win.

At these numbers, if we were voting today, the Tories would have a 45.5% chance at a majority (down from 56.5% last week). The Liberals are up 6.4 seats from last week, with the Conservatives down 1.6, the NDP down 2.1, and the Bloc down 2.5. Despite this, the Liberal seat range places them right around their 2008 totals.

Here are the tables – for the methodology, check out last week’s update. To clarify, “safe seats” are seats the party has a 95% chance of winning given current poll numbers and “seats in play” are seats they have at least a 5% chance of taking – it’s based on current levels of support, it’s not a prediction.

Updated Seat Projections

Posted on by CalgaryGrit in 2011 Federal Election, Federal Politics, Seat Projections | Leave a comment

Despite some mid-week adventures by Nanos, the poll numbers stayed relatively flat in Week 1 of the campaign. In effect, outside of our daily dose of Nik, there wasn’t a lot to report on until Ekos and Leger rolled in over the weekend. (We also got new numbers from Decima today, but only after I ran my seat projections)

The end result, is a vote line as flat as Saskatchewan.

As a bit of context, the graph above gives a daily popular vote average from on all publicly released polls, based on:

-How recently the polls were conducted. Each poll is given a one-week half-life which meaning a poll released today is worth twice as much as one released a week ago. I consider a poll “added” to the sample on the mean day it was in field (i.e. night 2 of a 3 night field).
-Poll sample sizes
-The accuracy of each company’s election polls over the past 5 years

From this, I ran my seat projection model. For the lengthy methodology, click here. The key thing to keep in mind is this model simulates the election 10,000 times in every riding – in each simulation, I include 3 sources of variance:

1) The variance between regional shifts and riding shifts observed in the 2004, 2006, and 2008 elections (i.e. if the Liberals are up 5 points in Ontario, how can we expect their vote to swing in a given riding?)
2) Sample variance, based on the sample size of the polls being used
3) “Pollster variance”, based on how much polls have missed the mark in recent campaigns, beyond the sample variance

The model also takes by elections and incumbency into account. And unlike other models, it includes not only the 2008 election as the baseline, but some 2004 and 2006 election results, as well as a riding projection based on demographics. This helps “smooth” out anomalous results from the 2008 campaign, to ensure we don’t project the Greens winning Central Nova or anything like that.

What I want to stress is that this model is completely data-driven. At no point have I “tweaked” the results, or given boosts to individual candidates. Once you start doing that, you might as well just consult the Election Prediction Project.

Also, when calculating variance numbers, by election effects, and incumbency effects, the numbers I’m using are based 100% on research I’ve conducted (using 2004 and 2006 data to simulate the 2008 election). No factor (outside of the 7-day polling half life) has been arbitrarily picked.

OK – I guess that short explanation turned into a long one. I’ll keep it brief on further updates. One final disclaimer – this is not a prediction of how the election will turn out. It’s based on the poll numbers released as of today…poll numbers that reflect a relatively unengaged electorate.

And with that, here’s what’s projected:

Based on the simulations, Harper sits at a 56.5% chance of a majority.

Mind you, this actually reflects a drop in support from him from my final pre-election simulation, which pegged him at between 148 to 178 seats. So the gap did narrow a bit in week 1, but it’s still a big gap.

Then, as now, Harper’s gains have come almost exclusively in Ontario. The lower end of that 95% confidence interval for Ontario is 51 seats – that’s the toal he got last election, and he could gain up to 19 more. In every other province, Harper’s expect seat total is within 1 or 2 of his total last election.

Again, it’s too early to consider this anything more than a fun statistical exercise. But, based on these numbers, it’s a lot more fun for Conservatives than Liberals.

Pre-Christmas Seat Projections

Posted on by CalgaryGrit in Seat Projections | Leave a comment

Ho ho ho! Look what Santa has brought Stephen Harper for Christmas. Why, it’s a 17.6% chance at a majority.

For the first time since I started running these projections in July, a Conservative majority is a real possibility, abeit still a low probability outcome. Of course, these projections are based on where the ball is now – and Harper’s numbers always seem to bounce back down whenever he gets close to that elusive majority.




Since the last update, Harper’s largest gains have come in Ontario (up 6 seats on average) and BC (up 5 seats on average) – his support numbers have not shifted noticeably anywhere else in the country.

Does this mean he’ll try for an early 2011 election? Unless he thinks he can nab that majority, there’s no real incentive to. He’s in power, his job is secure, and the opposition parties aren’t exactly chomping at his heels (as these numbers show).

For a long methodology description, click here. For a shorter one, you can just read the first few lines of last month’s update.

November Seat Projections

Posted on by CalgaryGrit in Seat Projections | Leave a comment

Yesterday, I gave an overview of the polling numbers – today, an updated seat projection.

The long explanation of how I came up with these numbers is here. The short of it is the model simulates an election 10,000 times, taking the following into account:

-Publicly released polling data
-2004, 2006, and 2008 election results
-Riding demographics
-The historical variance in riding results, compared to regional results
-Accuracy of Canadian pollsters in predicting recent provincial and federal elections
-Incumbency
-By election results

The benefit of this model over other projections is that this gives you a robust prediction that smooths out some of the blips you get when you only use the last election as your benchmark, it’s data driven, and it takes “election day swings” when the polls are all off into account.

It’s NOT a prediction of the next election, but reflects what we could expect if the election were held today.

So, with that, here are the updated projections:



If you compare these to the October sim…well, the results aren’t that interesting, with no party moving by more than 2 seats on average. I know “Nothing Happened” isn’t an eye-catching headline but, more often than not that’s the reality of the situation, as excited as well all get over every mini-scandal and ministerial resignation.

Now, as then, we’re on a crash course for a Tory minority, with only a 0.1% chance of a Liberal government and a 0.3% chance of a Harper majority.

Still, that would mean gains for the Liberals from 2008, most notably in Ontario (+8.1), but also in BC (+2.8), Quebec (+1.7), and Atlantic Canada (+1.4).

Seat Projections

Posted on by CalgaryGrit in Seat Projections | Leave a comment

Below are my updated seat projections, I first posted in July. For the long methodology explanation, click here. To summarize it, the model projects the probability of each party winning every seat in the country, taking the following into account:

-Publicly released polling data
-2004, 2006, and 2008 election results
-Riding demographics
-The historical variance in riding results, compared to regional results
-Accuracy of Canadian pollsters in predicting recent provincial and federal elections
-Incumbency
-By election results

The benefit of this model over other projections is that this gives you a robust prediction that smooths out some of the blips you get when you only use the last election as your benchmark (like the Greens projected to win a seat in Atlantic Canada), it’s completely objective, and it takes “election day swings” when the polls are all off into account.

New this wave, I’ve assigned a two week “half life” to polls, which means a poll released this week is worth twice as much as a poll released a fortnight ago (assuming the sample size is the same and the “pollster accuracy” rating is the same).

It’s NOT a prediction of the next election, but reflects what we could expect if the election were held today.

So, with that, here are the updated projections:



On average, the Liberals are up 13.5 seats from July, with the Tories down 4.2, the NDP down 7.2, and the Bloc down 2.2. The largest gains for the Liberals have come in Ontario (up 9.4) and in BC (up 2.9). The Liberals “best case” scenario (upper end of 95% confidence interval) in Ontario has increased from 47 seats in July to 59 seats now.

The odds of a Tory majority have dropped from 1.6% down to 0.4%.



Seat Projections

Posted on by CalgaryGrit in Seat Projections | Leave a comment

For those who recall, I experimented with a seat projection model last election.

By and large, it worked as well as any of the other models out there. But the rub is, my model, like every other pundit, seat projection and octopus out there, underestimated the Conservative seat total and overestimated the Liberal one. The reason? The Conservatives over performed their polling numbers at the ballot box.

And that’s the problem with the seat projection game – it’s only as good as the polling data it uses. I think I’ve found a way to work around this, although I’m sure some tweaking will be required. But for now, I’m quite pleased with the model I’ve produced.

To read the big and long methodology behind it, click here. The short version of it is this: the model projects seats, taking the following into account:

-Publicly released polling data
-2004, 2006, and 2008 election results
Riding demographics
-The historical variance in riding results, compared to regional results
-Accuracy of Canadian pollsters in predicting recent provincial and federal elections
Incumbency
By election results

For all these factors, I’ve done quite a bit of trial and error research – usually using the ’04 and ’06 elections to “project” the ’08 one. The end result is a projection system that has a few key advantages:

1. It has built in safeguards against “election day swing”, when certain parties over or under perform all the polls. It’s no use pretending this doesn’t happen, so it should be reflected in projection models, especially ones which rely on simulations. This may not make my “best guess” any more accurate, but it makes the model far more adept at calculating the probability of things like a Liberal win, or a Tory majority.

2. It’s robust. By looking at the past three elections and the riding demographics, this model smoothes out some of the fluky results that tend to plague seat projection models due to “weird stuff” like Elizabeth May running in Central Nova or surprise nudity.

3. It’s entirely objective. I plug the numbers into my excel sheet, run some simulations in R, and get the output.

4. It produces a probability in each riding. Most projection models that show the Liberals up by 1% in a riding will project the riding to go Liberal – in reality, if the Liberals are projected to be up by 1%, it really means you might as well flip a coin to pick the winner.

So, if you want the long explanation, click here. Otherwise, here are my projections, based on polling data released in June and July. Remember, this is what would happen if the election were held today – it’s not a prediction of what would happen if the election were held in a few months. Campaigns matter.





Probability of Conservative Win: 100%
Probability of Conservative Majority: 1.6%

Back in the UK

Posted on by CalgaryGrit in Seat Projections | Leave a comment

The British election is now a week away and the Conservatives seem all but assured of victory – the real question is whether or not they’ll be able to get their majority. Sound familiar?

The real twist is the rise of the Liberal Democrats, which has buggered up most seat projections. But hey, that just makes it more fun. Here’s the skinny on UK polling numbers and projections:

1) The UK Polling Report has all the latest polls.

2) The gambling line at Betfair sets the Conservatives at 318, Labour at 216, the LibDems at 87, and other parties at 29.

3) You can play around yourself with BBC’s swingometer. Just plug in the numbers and, Bob’s your uncle, you get projected seat counts!

4) A more advanced projection model can be found here. My only qualm is that I just don’t think there are enough data points to customize vote patterns on a riding by riding basis. Regardless, it projects the Conservatives at 291 seats, Labour at 209, the LibDems at 120, with 8 seats going to other parties. It sets the odds of a Tory majority at 8%.

5) Electoral Calculus projects seats using a regional swingometer and the betting markets. They have the Cons at 283 seats, Labour at 238, the LibDems at 97, with 32 seats going to other parties.

6) Finally, this brings us to our good friends at 538. Their model is a bit too subjective for my taste, and I’m not a fan of the geometric swing. The problem is, if you assume Labour will only hold 60% of its vote and you extrapolate that to every riding, you’ll be taking the most votes away from their strongholds. For example, if you did this in Canada and had the Liberals down in the polls, their biggest losses would come in Toronto. Which I’m not sure is what would happen in reality.

On the flip side, their model will probably account for the rise of the LibDems better than most. So I guess we’ll have to wait and see. Regardless, they project the Conservatives at 299 seats, Labour at 199, and the LibDems at 120.

Seat Projections

Posted on by CalgaryGrit in Seat Projections | Leave a comment

For a full description of methodology used, please see: Week 1 Projections

The polls have taken a turn for the worse for the Liberals, and my projections now show Harper with a 52.3% chance of a majority…and that’s mainly due to older sample that hasn’t completely decayed in weight yet. As always though, things can change a lot between now and election day.

From a methodology perspective, I’m hoping to see a few more regional breakdowns on some of these polls so that some of the Montreal/Toronto dynamics can be better taken into account.




Seat Projections

Posted on by CalgaryGrit in Seat Projections | Leave a comment

For a full description of methodology used, please see: Week 1 Projections

With three companies running daily tracking polls, there’s a lot more data available, which has decreased some of the variance in my projection model.

I’ve also tweaked the system slightly by weighting the Ekos polls less – with the demon dialing, the’re getting much larger samples than the other companies and were dominating the model.

No other changes in methodology from last week, so without further adieu, here is where things sit. Again, as a disclaimer, these results are only as good as the publicly released polls and makes no effort to project where things will be at on October 14th.

Simulation Totals

Seat Projections

Posted on by CalgaryGrit in Seat Projections | Leave a comment

Since 538 rocks my world, I figured it might be fun to try something similar for the federal campaign, so I’ve developed a seat projection system that uses a probabalistic approach. Here’s the explanation, in as plain English as I could put it. Feel free to e-mail me or ask for clarification in the comments section – I’m open to suggestions for improvements:

1. I’ve collected all the publicly available polling data released since September 1st (Nanos, Leger, Segma, Environics, Ipsos, Angus, Decima, SC, and CROP). I’ve only included data where regional splits are available.

2. A weighted average of the data is calculated for each region, assuming a 3 day half life for polling data. What that means is that a 3 day old sample of 500 from Quebec is weighted equally to a 6 day old sample of 1000 from Quebec – both would be counted as 250 completes. The “days old” number is based on the middle night of polling.

3. At this point, each seat is projected based on the change in the region. So if the Liberals are up 10% in Alberta (ha ha ha), they get 10% added to each seat. Sort of. In order to make the model more realistic, I’ve weighted the “base” for each seat 3/4 from 2006 and 1/4 from 2004, to reflect any “bizarre” fluctuations that may have happened last election due to local candidates, etc. I’ve also made a correction for incumbents retiring (between 2004 and 2006, the “incumbency factor” was worth 4.1% once regional changes were controlled for). And, in ridings where a by election has been held, I’ve given the by election and last election equal weighting. The important thing to remember is that, even after all this, everything gets projected to the regional numbers. So regardless of the tweakings, the numbers projected in Atlantic Canada will match the polling data.

4. And I could quit at this point and just list the projected wins and loses. But the problem with that is that a projected 2% Liberal win counts as 1 Liberal seat, as does a 30% projected Liberal win. And when you consider all the error associated with these projections, the two are definitely not even. So I decided to go the simulation route.

So I ran 1000 simulated elections. In each one, the regional numbers were simulated based on the sample size for the region (using the half-life discussed above). And then each riding was given a “random shift” based on the “regional to riding variance” observed when I ran this same model on the 2006 election using 2004 data (standard error of about 4% for each riding).

So in each of these 1000 elections I’ve got a winner in every riding. That means I can project a “probability of victory” for each seat, and get an average number of seats won per party.

It should be noted that this is all assuming the election is held today…I’m not predicting future shifts in popular support. It’s also assuming the polling numbers are accurate. And it’s not going to take into account a lot of the “unique” riding dynamics (Bill Casey, Lizzie May, etc) or different shifts that might be occuring between, say, Vancouver and rural BC.

So, based on the simulations, here are the results and graphs:

National
Conservatives: 137.9 (95% CI from 132 to 144)
Liberals: 98.5 (95% CI from 93 to 104)
NDP: 28.8 (95% CI from 25 to 32)
Bloc: 42.0 (95% CI from 39 to 45)
Indepent: 0.8 (Andre Arthur is the only independent who’s going to show up on this model and, regardless of whatever the projections say, he’s pretty much a lock since the Tories aren’t challenging him)

It should be noted that, up until a few days ago when the Decima and Nanos polls rolled in, this model was projecting a high probability of a Tory majority.

Here are the regional breaks:

Atlantic Canada
Liberal: 22.2
Conservative: 6.5
NDP: 3.3

Quebec
Bloc: 42.0
Liberal: 16.0
Conservative: 15.4
NDP: 0.8
Ind: 0.8

Ontario
Conservative: 51.4
Liberal: 44.3
NDP: 10.2

Prairies
Conservative: 19.4
Liberal: 4.7
NDP: 3.9

Alberta
Conservative: 27.7
Liberal: 0.3

BC + Territories
Conservative: 17.4
Liberal: 11.0
NDP: 10.6


See Also: Barry Kay Projections, Hill and Knowlton Projector, UBC Election Stock Market

Plugin from the creators of Brindes Personalizados :: More at Plulz Wordpress Plugins