Tuesday, September 20, 2011

Possible seat ranges tighten in Ontario

Yesterday, the QMI Agency's network of newspapers released the results of a new Léger Marketing poll for Ontario, with one glaring error.

QMI reported Progressive Conservative support at 36%, Liberal support at 33%, and NDP support at 29%. That was wrong - the New Democrats were actually at 26% in the poll, with the Greens at 5% (rather than 2%).

As this poll echoed yesterday's projection almost exactly, there have been few changes. So let's take a look at this poll in depth before getting to today's projection.
For the first time in the campaign, we are starting to see real consistency in the polls. Aside from having the Liberals up one point, Léger's poll is exactly the same as Angus-Reid's. Both were taken over the same days (with Léger stretching back one day further) and both used an online panel. I'm looking forward to a new telephone poll from Nanos or Ipsos-Reid this week to see how it lines up.

This is the same statistically insignificant lead that we've now seen in four of the last five provincial polls. The race is neck-and-neck between the Liberals and the Tories, but the New Democrats are slowly creeping upwards.

Regionally, it is difficult to take much from this poll. No pollsters divide Ontario in the same way, and Léger uses the inscrutable "Western Ontario" and "Southern Ontario" regions. I imagine Western Ontario is generally what the other polls calls Central Ontario, because of the high Green result, while Southern Ontario is the standard Southwestern Ontario. How the boundaries compare, however, I cannot say.

Some of the results are odd, like the NDP leading in Eastern Ontario and a three-way race in Northern Ontario (though, if you include the area down to Lake Simcoe in Northern Ontario, it isn't so odd). What the regionals do show, however, is just how close the race is. In all but Southern and Western Ontario, the three parties are in a statistical tie.

On leadership, the advantage lies with the Liberals. Dalton McGuinty topped the poll as the first choice for premier with 28%, followed closely by Tim Hudak at 26%. The NDP is still having some trouble getting Andrea Horwath out there, as she stands at 14%. This is the opposite of what we have seen at the federal level, where Jack Layton always outpaced support for his own party.
With this poll now input into the model, the Progressive Conservatives are projected to win 36.2% of the vote, with the Liberals three points behind at 33.2%. The New Democrats stand at 24.5%, while the Greens are at 4.8%.

That represents a drop of 0.3 points for the Tories and 0.4 points for the Liberals. The NDP has gained 0.6 points since yesterday. The Greens are unchanged.

There have been no seat changes. The Liberals are still projected to win a minority with 50 seats, with 36 going to the PCs and 21 to the New Democrats.

The seat ranges have shifted slightly. The Tories are unchanged, but the upper limit of the Liberals has dropped while the New Democrats could now win as many as 24 seats.

Both the PCs and Liberals could win as much as a majority government, but with an upper limit of 55 and 56 seats, respectively, and the minimum needed for a majority being 54 seats, both parties are unlikely to win more than half the seats in the province. The New Democrats are moving up in the seat ranges, but are not yet close to moving into second place in Ontario.
Over the weekend, I tested the vote projection model on the 2007 Ontario election. The results were positive, with the average error per party being lower than that of the average of the 2007's campaign last week of polling. I was happy to see it perform well, so as long as the pollsters are as good this fall as they were in 2007, the model should give an accurate result.

Manitoba is still gnawing at me, however. The last poll is three months old and the projection as it stands is horribly out of date. The closest race going into the series of fall elections has inexplicably yet to be polled. If nothing comes out of the province and, unlike PEI and Newfoundland & Labrador, I haven't heard a thing about an upcoming poll, I'll have to remove the projection for lack of information. We shall see.

Blogger has changed the way images are viewed. For the most part, I think it is an improvement but only if you don't need to magnify the image, as you must with the Riding Projections. One solution to this problem is to open the image in a new window (you can right-click to choose that option), but the best solution is to click on the Riding Projections image at the top of the right-hand column. That is always up to date.

26 comments:

  1. Please correct the distorted X axis on your graph. It should start at ZERO, not at 15 or so, as it does. Your graph gives the very false impression that the NDP's seat total is less than half what it actually is, because it omits most of the range below where the NDP is polling.

    ReplyDelete
  2. Hudak hasn't a hope !!

    ReplyDelete
  3. Mr. Danforth,

    To make the information on the chart easier to read, the X axis only ranges one below the lowest seat range for any party and one above. Stretching it from 0 to 107 seats makes it much more difficult to read.

    I don't expect many people read the chart and get the impression that the NDP could win 0 seats and the Liberals all 107. Clicking on the chart shows the seat totals on the X axis.

    ReplyDelete
  4. The distorted x axis does NOT make the chart easier to read. What we want to know when we see the chart is what kind of vote gain - relative to existing support- is needed for Party X to win. Your graph gives a really false impression in that respect.

    ReplyDelete
  5. I'm sorry if you don't understand the chart, but it shows the seat ranges of each party if they win or lose all of the close races. It has nothing to do with what kind of support is needed to reach X-number of seats.

    ReplyDelete
  6. Mr. Danforth if you don't like the chart make your own and start your own website. Eric is providing a valuable resource to all political junkies and we should be grateful for each and every service/graph he provides us with.

    ReplyDelete
  7. The chart really isn't hard to read. It has the projection in big bold numbers. No one would assume that while the space between the highest NDP bar and the highest PC bar is 15, the space between the highest NDP bar and the Y axis is 21. It just doesn't make spacial sense.

    Further, graphs in all kinds of areas collapse the beginning of the X axis if they are dealing with higher numbers along the axis than can be easy displayed in the space available. Most people learn this in high school math. The most viable criticism that could be made is pretty mild - that Eric didn't include the correct symbol on the X axis to mark that there was a gap there.

    On the substance of it, the NDP really is creeping up under the radar, which is vaguely reminiscent of the 1990 election. Not to say that the NDP is going to win by any means, but it could be a very interesting election night.

    ReplyDelete
  8. The chart is not exactly a mathematical graph - it is simply to show the seat ranges in a way that better shows how the ranges overlap.

    ReplyDelete
  9. Do you have a link/location for your "base" riding by riding statistics, which I assume are the last election's results, either raw or adjusted?

    ReplyDelete
  10. In the words of Mark Twain 'There are three kinds of lies: lies, damned lies, and statistics.'

    I'm not suggesting that you are lying, but as the disagreement between Eric and Mr Danforth highlights, the simple representation statistics through graphs can distort the true meaning of the numbers, particularly if, like most people I suspect, the person only has a cursory glance at the table/graph without taking the time to see what scale is being employed.

    As someone who has taught statistics at the post-secondary level, I can only say that I would normally expect a student to offer some sort of delineation at the start of an axis when eliminating portions of the scale for clarity, for the same purpose of clarity.

    ReplyDelete
  11. Anonymous 12:22,

    I can certainly do that for next time.

    Alan Stewart,

    I'm afraid not, though Elections Ontario has riding-by-riding results for the 2007 election. I think even Wikipedia does.

    ReplyDelete
  12. Eric, I think you've probably answered this question before, so perhaps merely pointing me in the direction of where the answer is found is sufficient. But when a new poll comes along, do you just use the topline province-wide figures, or do you input the regionals too? I ask since it's famously unclear how the pollsters define the regionals, so it would be a nightmare to decide which riding should get which data. At the same time, however, it must certainly always be clear which region Toronto-Danforth is in or which reason Kenora-Rainy River is in.

    ReplyDelete
  13. Eric

    For what its worth, there is nothing confusing to me about your chart. Anyway, count me among the grateful supporters, not the angry critics.

    At the same time, unless I am missing something, your table of seat numbers by party by region has not been updated to show the latest totals you are projecting. The PCs only add up to 35, the NDP add up to only 19, and the Liberals add up to three too many at 53.

    Next question, I appreciate how you have made changes in your model to reflect the federal election experience - that's great. Still, I wonder how or whether you can (or should?) make any changes to reflect seat-specific surprises from May. Allow me to give two examples:
    In Scarborough Rouge River, the federal outcome was heavily influenced by a massive swing of the south asian community behind the NDP, a party that had never polled well in that riding, federally or provincially. They are running a very similar campaign with a similar candidagte this time and will achieve likely very similar results (at the least, they will be very close, at the most, they will win it strongly, again). The last time provincial results are much less meaningful, IMHO. Your model does not seem to reflect this.

    Similarly, the same NDP Sikh candidate is running for the NDP in Bramalea-Gore-Malton who came within 500 votes of winning in the federal campaign. Again, a very surprising emergence of a South Asian candidate. Again, no previous history of support federally or provincially. Is it likely he will not be at least close again?

    It is a wonderful job you have taken on for us. Thank you

    (Amateur Psephologist)

    ReplyDelete
  14. Bungle Jerry,

    I only use the topline, provincial numbers. As you said, without a common division of the province there is no base from which to work.

    For example, yes it is clear where Kenora-Rainy River is, but I need to know how a firm defines the North in order to compare how things have changed since the last election.

    ReplyDelete
  15. Amateur Psephologist,

    You're right, I've fixed the regions chart. I forgot to update it.

    I understand that the federal election will likely have an effect, but conceivably the polls should reflect that. For individual ridings, it is something I cannot model. Will those who voted NDP in the federal election vote NDP in this provincial election in those ridings? Very probably - but how would I reflect that in the projection?

    ReplyDelete
  16. it's great to do all these calculations - the final poll will be election night....remember back in april when similar pre election polls showed that mr harper had no chance at a majority government....toronto municipal pre election polls insisted, for the most part, that rob ford and ol' smitty were neck 'n neck.fact is, a lot of people are pissed at mcguinty but then again there is the old adage "beware of the devil you don't know".....

    ReplyDelete
  17. Late response but...

    @Mr.Danforth, Anonymous 12:22 - There's nothing for Eric to correct. The graph is clearly and correctly labeled.

    Here's an astrophysics paper that was published today: http://arxiv.org/PS_cache/arxiv/pdf/1109/1109.3748v1.pdf If you take a look, you'll notice the axes are scaled to only include the range of data relevant to the model. No party is projected to get more than 57 seats or less than 18, and we don't need a great expanse of white to tell us that. I don't at least.

    ReplyDelete
  18. One interesting aspect from Eric's projections is that the region with the most close riding races (less than 3 points) between the Liberals and PC is the 905 area. I know that this fact is not new but Eric's numbers are so much closer, that it looks like the government will be determined there.

    If the Liberals can hold 5-8 close races in the 905 they are back in office. PC has a bigger mountain to climb to win but, without these 8-10 seats, they won't do it.

    ReplyDelete
  19. I wonder - could we get a situation where someone comes in 3rd in popular vote but wins a majority?

    Boy would that scream that FPTP is a poor system to determine what the voters want, but I can see how it could happen.

    ReplyDelete
  20. John_Northey,

    It is possible. It is extremely improbable, but it could happen. Picture this. There are three parties, parties A, B and C, competing in an election for a legislature composed of 108 seats. Each of these seats has an exactly equal population. Of those seats 55 are perfect three way races, and Party C wins each and every one of those races with, let's say 33.34% of the vote, with the other two parties each getting 33.33%. In the other 53 races, Party C receives 0% of the vote, with the other parties splitting the seats between them. This would result in Party C winning a majority government with just under 17% of the vote cast overall.

    That's the absurdity of first-past-the-post. A party could, in the most extreme possible situation, win 50%+1 of the seats, and 100% of the power, with as little as 16.98% of the vote.

    ReplyDelete
  21. The main reason I started to wonder about 3rd winning is how the Liberals have lead in seats but not in popular vote for awhile here.

    I still see this as the NDP's great opportunity. A Liberal who is unpopular, a PC who is 'meh', a popular NDP leader passing away just before the election, a big surge nationally for the NDP all combine to make it such that if the NDP's leader can do well in the debate things could get _very_ interesting.

    Of course, if the voters do hit a limit and feel the NDP isn't the right one we could see Greens move up into seat territory - but that might just be wishful thinking.

    ReplyDelete
  22. when is the Next Poll in Ontario? NANO Forum Harris ? Is there a list of dates they are to be released

    ReplyDelete
  23. Yeah I agree with Anonymous @ 21:25. We have only not had a poll in Ontario for what, 2 days? Maybe the question is really: When is the next poll in Manitoba? Seriously the last one was in JULY! Where are the pollsters for the province, asleep?! I'm sure Manitobans are getting anxious for some new polls. But it is a bit annoying now that the race is getting really interesting in Ontario and we have 2-3 days in between each poll.

    ReplyDelete
  24. I had heard there would be daily polls in Ontario but alas they are only trickling in - i would HAte to be in a smaller province with even less polling - what's a political junkie to do

    ReplyDelete
  25. There is now a small Viewpoints poll out for Manitoba....done for the Mb Real Estate Assoc.
    CJOB (Corus Radio) article:
    http://www.cjob.com/Landing/Story.aspx?id=1544432
    The "Leaders" debate just finished tonight....probably a wash.... S.L.

    ReplyDelete
  26. That Viewpoints poll is only on whether there needs to be a change or not in Manitoba, it isn't a voting intentions poll.

    ReplyDelete

COMMENT MODERATION POLICY - Please be respectful when commenting. If choosing to remain anonymous, please sign your comment with some sort of pseudonym to avoid confusion. Please do not use any derogatory terms for fellow commenters, parties, or politicians. Inflammatory and overly partisan comments will not be posted. PLEASE KEEP DISCUSSION ON TOPIC.