Projecting backward


Alice Funke challenges the seat projection phenomenon.

The inherent bias in seat prediction methodologies to favour previous election results means they tend to overly favour parties set to lose seats, such as the Liberals and Bloc Québécois in the last election. They also tend to miss the likelihood of parties on the rise to gain seats, such as was the case with the Conservatives in the last election. Only the NDP, whose vote intention numbers showed little gain by the end of the 2008 campaign, saw seat count predictions on both sides of its eventual total.

Another problem for the seat projection methodologies is that they are backward-looking. They’re using days-old polling data at a time of incredible movement in the polls, and laying that on top of results from the last election when incumbency was a factor for some political parties’ votes that is no longer at play.


Projecting backward

  1. Funkie analysis: rightly taking the Science out of PoliSci.

  2. One of my favourite seat projections currently doesn't show a single seat changing hands in Ontario. While I greatly value the rigour behind how they arrived at those numbers, I think we can all agree that this is probably not going to be the case.

    The real question, of course, is how else could anyone do it? I think the majority of ridings will be dictated by historical voting patterns, so it would be possible to get a 90 or 95% accuracy rate with this method, most elections. That 5-10% that's wrong, however, is going to make the entire approach seem wrong to people who want it to be 100% accurate.

    If anyone has suggestions on how to reach 100% accuracy, I'm sure some pollsters would love to hear from you.

    • I think it's true that most of the time, encumbents that have held a riding more than once are very very likely to be re-elected.

      For example, David McGuinty has held my riding for years, and the chance of him losing this time is practically zero in my opinion, especially given that the other parties are all running low profile candidates against him.

      It could even be that because parties don't like to risk their high profile candidates against other high profile candidates, it's a self fulfilling prophecy, but in any case it's an obvious pattern one doesn't abandon without reason.

      • My experience is that personal incumbency only matters in a very small handful of cases. Ridings that are rock solid for one party can run a rotting log and expect a positive result. In swing ridings excellent constituency-oriented, popular incumbents frequently get turfed in national or regional shifts in voter preference, often to rotting logs.

    • ThreeHundredEight has the Bloc getting more seats the the NDP in Quebec, despite being at about 26% to the NDP's ~40%. Even if the NDP vote collapses before election day, that seems unlikely. Sure, the Bloc have a very developed Get Out the Vote apparatus – but that does them very little good when a third or more of their voters have defected.

      • Yes, but ThreeHundredEight is designed to smooth out bumpy polling, not react swiftly to a changing situation. Eric will post two projections on Sunday: the model output based on all weights, and an output based on that day's polls. That should show two different outcomes, I'd guess.

        • Given that any interesting election has swiftly changing situations, it doesn't strike me as being a very good model.

          • There hasn't been a shift like this in decades. Eric's model would have worked very well in a boring election like 2008 or 2006; even in 2004 and 2000 it would have done well. It does have trouble with elections which have a monumental shift in support in a single region, as it breaks all historic voting patterns. The fact that Duceppe could be kicked out of his own riding after holding it for 20 years? How in the world could you predict that accurately in any model?

  3. There are many models out there that look at seat projection from different angles, and clearly each firm would LOVE to be the one that "got it right", so the criticism seems silly to me.

    All in all, it's just not that easy a thing to do, and while I'm sure some people wish pollsters didn't do it at all, the public interest in these projections is just too persistent to expect it to stop.

    Does it influence votes? Probably a little, but why shouldn't people have a sense of how the public at large is leaning?

    • Group-think is one of the most dangerous traps in our decision making.

      • I'm not sure why you would say that. Given our electoral system, it would seem wise for progressive voters to be on the same page so that they don't split the vote and elect a third party that doesn't really represent their interests.

        Besides, there is nothing dangerous about information. It's what people do with it that matters.

        • False. Like any technology, information has a bias. I point you toward the campaign against dihydrogen monoxide.

          • Sure information has a bias. Sources are obviously something one should take into account, as partial information can be worse than no information at all.

            That said, I would suggest making decisions in ignorance of any facts is ultimately more foolish.

            We have built our modern civilisation using observable facts afterall.

          • I would suggest that group-think is one of the worst forms of "partial information"

          • There is simply no way to remove bias from information. Information is generated by humans, not by machines. And even if some of it is generated by machines, those machines were generated by humans. To complain about information being dangerous because it is biased is to assume that some other form of unbiased information exists. It does not. As Nassim Taleb is fond of saying: Trust the science, hate the scientist. No, he does not hate scientists. He is just warning that all information can be tainted by human perceptions, even amongst those individuals, like scientists, whose practices are supposed to eliminate said biases.

          • Is dihydrogen monoxide a "think or thwim" issue? Is it splashy enough?

          • Oh go soak your head. :)

          • But it's been the cause of so many deaths, and it's in every Canadian household! You monster!

          • Clearly he has no sympathy for 2HO victims. None whatsoever.

  4. Understand this argument, however it is crucial to make seat-by-seat predictions. That's how our Parliament works. You win a riding not a popular vote on a national level.

    • however it is crucial to make seat-by-seat predictions

      Why? Just research the platforms of the parties, vote, and await the outcome Monday.

      • It's be nice if we had an electoral system that was immune to strategic voting. It might get rid of some of the nattering about vote-splitting, etc.

        I like a straight-up preferential ballot for simplicity. STV would probably be even better as it allows better representativeness/proportionality.

        • Preferential systems always run into problems where you can strategically vote for a less favored candidate in order to eliminate a least favored one.

          In addition, they're harder to count, and can lead to confusion for people. (Sure, it doesn't seem that hard to rank your preferences, but think back to hanging chads)

          The Percentage Proportional System is immune to strategic voting since it looks at percentages in each riding and at the national level. Thus if you hate candidate A, voting for B, C, or D, won't make any difference as to whether A gets in, since that is determined based only on A's percentage of the vote total, not in comparison to any other particular candidate.

          It has the added benefit that the vote can be run exactly as it is today, with the proportional part all done in the calculations thereafter.

          • Is there evidence that that kind of strategic voting has any significant impact on election outcomes?

            There is usually a trade-off in terms of other desirable characteristics of an electoral system to get a system that is less susceptible to gaming. Local representativeness is one of these characteristics. I think there is also something to be said for encouraging at least somewhat broad parties.

          • No idea, but if the system permits it, then you can't really say it's better at dealing with strategic voting than another system which permits it. STV gets more information from the voter, but boil it down and it's simply an improved FPTP system. The only real change is that the post is raised.

            As such, it has many of the same problems, particularly strategic voting and the wasting of votes. They're just a little harder to see.

          • I couldn't find any detail on the system you mentioned, but it presumably leads to similar problems with wasted votes. It also overly-entrenches parties in the political process. Independents would have a very difficult time getting elected. You might end up with an MP that has very little local support, especially when the second-ranked candidate is preferred overall.

            Any system that ignores preferences beyond first has a hard time being locally representative. Perhaps % prop is immune to strategic voting, but whether it has output legitimacy or outcomes most voters are happy with is another matter. Are there any jurisdictions that have implemented it?

          • Besides which, I don't see here, nor have I seen, any truly convincing arguments to suggest that strategic voting isn't a sensible option in our current multi-party dog-race system.

          • Strategic voting is beyond idiotic in any system. The likelihood that your vote will influence the outcome of an election is so infinitely small it can barely be counted – as far as I know, no significant race has ever come down to 1 vote (and even then, for it to really matter, it would have to be a seat that could make or break a majority). Nor does your vote impact the votes of others.

            Since your vote doesn't impact outcomes, the only reason to vote is that it makes you feel good. Voting for an outcome you don't want makes very little sense in that kind of a world.

          • Couldn't find detail because it's my baby. Only exists in my head, and whoever else's head I can get it into :)

            Works as follows. Votes are cast as they are now. You vote for your single preferred candidate in your riding.

            Votes are counted. Any candidate with over 50% of the vote total of their riding gets their seat. –> Preserving popular candidates first and foremost.

            We then look at the percentage of the total vote a person would need to claim a federal seat. So if there were 306 seats left after the popular candidate, each seat would be worth 0.32% of the vote. Divide a party's total vote percentage by that to get the number of seats that party is assigned. Round down to the nearest seat.

            To determine which seats, we look first to the parties running the least number of candidates (to prevent regional parties from being locked out of their seats) and the percentage of the riding vote each candidate received. If they are to receive N seats, the top N candidates by riding percentage are the ones that get their seats.

            We then move to the remaining parties ordered by the lowest percentage of the vote. Again, their N seats are given to the top N candidates, ranked by voting percentage in the remaining ridings.

            At the end will be the most popular party in Canada which gets all the remaining seats (which would include any left over from the rounding).

            The downside of the system is that in ridings without a definite preference, the candidate elected will reflect the national character more than the local — although it will be the most popular candidates of the national character.

            Systems which have a strongly defined preference will tend to get who they want, however. And it manages to preserve regional concerns, popular independant candidates, 100% proportionality without party lists, and the simple, single choice ballot.

            It also makes any pre-election polling virtually useless, since just knowing the percentages of a single riding aren't going to tell you much of use — but I tend to think of this as a pro, not a con.

            Perhaps strongest in its favor, however, is that it also pits candidates of the same party vying against *each other* for total popularity. After all, taking your riding by a hair's breadth isn't going to do any good when your party has a bunch of other candidates who are doing excellent. This will force candidates to pay *very* deep attention to their specific ridings' desires.

          • This system is not immune to gaming. Say in a riding we have three parties. If Party A has 47% support, B has 45 and C has 8%. B and C voters have each other's parties as their second choice and abhor Party A. Party C supporters won't have an MP elected for their local seat, but can put the B party candidate over the top to ensure they have an amenable local representative. They have an incentive to vote contrary to their honest preference.

            You also can skew election results by adding candidates. Say 60% of voters in a riding support party A, with one Party A candidate, that candidate will win, with two, there is a chance that each has only 30% support and neither candidate is elected.

            Seems to me this proposal put a large emphasis on proportionality, but does not assure it.

            What happens if a party wins 50%+1 in every riding? They receive 100% of the seats with only 50% support. Thus, proportionality is not preserved.

          • Oo! Good points.. okay, today's a good day, I'm going to learn something.

            It looks like your first and last argument stem from the over 50% of a riding condition. But I'll point out that you would get the exact same result from FPTP, and IRV. STV is more complex because it relies on more seats per riding, but it also would have one seat in every riding filled by that party, giving it an extremely large majority in the house. So my system isn't perfect in that regard, but it's no worse than any other. The other thing is, if we found that it seemed that condition was causing difficulties in our multi-party system (something I find highly doubtful) it could simply be removed. Thus an independant candidate would only need a seat's worth of the total vote percentage. Depending on the riding, this may or may not be possible. Given that it's a federal election and presumably our represenatives are representing our national character, I'm not terribly concerned about this one either way.

            Second, I'll also point out that typically why strategic voting is done is because there are no consequences to the party the voter is strategically switching away from. That is, C voters can switch knowing that doing so doesn't make things any worse for them, and indeed makes them better. In my system, however, switching makes things worse for them on a national level, where 8% of that ridings vote could result in an additional seat for the C party in another riding. Thus, there is incentive to *not* game the system. Something mostly missing in other systems. (Admitted, STV has some incentives against it in that if too many people strategically vote, it can have the opposite effect of what is intended..)

            For your election skewing, you seem to be making the assumption that anybody can just put on the party hat and say they're running for X party. That doesn't match with reality, and any party that allowed that frankly deserves what they're going to get.

            Still.. good points. Made me think harder about the system, so thanks.

          • Other thoughts: you could end up with representatives you really don't like. For instance, if the threshold for a party to get a seat is 100%/308 ~ 0.3%, then you could end up with a situation where the Communist Party, neo Nazi/nationalist parties, anti-immigrant parties, etc. get that share nationally, and thus someone has to end up with an MP from that party, even if they only had a 2 or 3% vote share in the riding. That's a bit scary, and I don't think people would tolerate it from an output legitimacy perspective. The counting system sounds intuitive, maybe, but I think a lot of people would struggle with ending up with an MP that hardly anyone in the riding voted for or even found remotely palatable. This is not just a theoretical risk, but a probable one.

            It's why I like preferential electoral systems. It tends to avoid swings to the extremes and encourages consensual politics. I don't think perfect proportionality is a worthy goal in itself. It is useful for majorities to be hard to achieve.

          • Yup. As I pointed out, that's the primary downfall of my system. Some ridings will get representatives from the national character rather than the local riding. But as I pointed out, it will be in the riding where that candidate is the most popular.. so preserves the will of the country while pissing off the fewest people possible.

            The goal I had in mind when thinking about the system initially was to waste as few votes as possible, while keeping control of who actually gets in out of the hands of the parties.

            The only votes wasted under this system are those for parties or people that don't meet the minimum percentage of one seat. I think that's better than any other system can claim.

          • Oh okay. I don't agree that avoiding 'wasted votes' is the most important criteria in selecting an electoral system. If local representativeness is not important, why bother tying the candidates to a riding? It seems artificial.

            Besides, once you have your governing coalition, the opposition votes are essentially 'wasted' anyway, as they have no direct influence in the legislature.

          • Only in a situation where the government/coaltion can form a majority.

            And local representativeness *is* important, just not *as* important as ensuring that as many votes as possible are weighed into our system. That's what strikes me as democracy. If you go through and calculate how it works out most of the ridings get the most popular candidates. The ones that change tend to be the ridings that are very divided in the first place — ie, where a majority will be pissed off by the result anyway.

            The reason for ridings is the same as it is now. We're a huge country Different regions will have different concerns. It makes sense to have riding level politics so that candidates can concentrate on, and reflect these concerns in the House of Commons.

            Besides, weren't you in favor of mandatory voting? Here's something better in my mind.. a voting system that gives a *reason* to vote. Even if you're in Stephen Harper's riding, there's still reason to get out and vote, whether for or against Harper — because it helps your party of choice's national average, which can mean extra seats.

          • This system doesn't seem like it actually guarantees proportionality; it seems to, if anything, mainly benefit regional parties.

            Suppose there are 300 seats, of which the Bloc runs 75 in Quebec ridings.

            The Bloc wins 40 seats outright with >50% of the vote in each, and captures 10% of the popular vote in total. Of the remaining 250 seats, the Bloc is owed 10% because of their popular vote, collecting another 25 seats. So with 10% of the popular vote, the Bloc racked up 65 seats, nearly 22% of the total. If I had bothered to juggle the numbers a bit more, I could have easily created a situation where the Bloc is winning seats in ridings where they didn't even run a candidate–in Ontario, say.

            The big problem is that for any candidate that hits the 50%+ mark, all of that candidates voters count twice in the election. First for the 50% bonus. Second for the proportional vote.

          • And as I pointed out above, I'm certainly not married to the 50% cut off.

            However, I found it difficult to reasonably conclude that if a majority of people in a riding prefer a candidate as their first choice, that candidate should not get in. The edge scenarios you point out, while important, are highly unlikely in a multi-party system like ours is. There are similar edge case scenarios for any system.

            Given that it's a national election, however, and that candidates are being chosen to represent us nationally, it's not a point I'd be willing to stake the system on.

          • No, STV (single transferable vote) actually has virtually none of these problems. It still keeps riding in, but there are multiple seats per riding, and it is a ranked system of sorts–it sets a threshold for the minimum number of votes that a candidate needs to get elected in a riding (on the order of number of eligible voters/number of available seats). Once a candidate meets that threshold, they are elected, and all future votes to that person are transferred to the second-place choices of those voters instead. This process continues until all available seats are filled. I prefer instant runoff myself (ranked selections, if no candidate has 50% of the vote, lowest person is dropped and their votes are transferred to their second place choices), but STV looks pretty good to me as well. Both STV and instant runoff are mostly immune to strategic voting.

            Pure proportional, that only counts popular vote and not preferences at all, tends to cause very fractured legislatures where fringe parties with small numbers of seats may hold the balance of power (see: Israel) because single-issue parties are very competitive. It's essentially the opposite result of FTFP: pure proportional optimizes on the maximum number of small parties, whereas FTFP optimizes on two very large parties.

          • Yet if you vote for all the unpopular guys first, your vote is still wasted unless there are fewer than seat x2 candidates. Also, in Malta, 51% of the public voted for the nationalist party as their first choice. The Labour party received majority government. They use STV.

          • Well, any system that uses seats or ridings is ultimately going to end up with some wasted votes–even in pure proportional, votes going to parties who are below the threshold to get a seat, or due to rounding issues (eg. if a party gets 50.4% of the vote out of 100 seats, then 0.4% of their vote is wasted). STV tries to minimize the amount of wastage by allowing voters to transfer; IRV tries to pick the candidate that is most acceptable to the largest number of people. It's worth pointing out that pure proportionality is not necessarily the optimal outcome in an election. If 50% of the electorate chooses Party A as their first choice and Party B as their second, and 50% of the electorate chooses Party C as their first choice and Party B as their second, Party B is arguably the party that best represents the views of the entire electorate, even though they have 0 first-choice votes.

            Arrow's Impossibility Theorem suggests that no voting system is perfectly "fair". It's just a matter of picking what disadvantages that you'd prefer to live with.

          • On that, we totally agree. My bias is for the opinions of as many people as possible to be represented.. that strikes me as fundamental to democracy. Many of these systems, it seems to me, are more interested in getting more information from fewer people.

          • And the fatal flaw of pure PR is that many candidates will have almost zero attachment to any particular constituency. SVT is the best compromise that allows for proportionality while still attaching each candidate to a seat. I believe the BC proposal – narrowly defeated the first time it was on the ballot – was an ideal solution. It's just too bad BC set the bar so damned high. 60% is an absurd level of support to require for anything.

    • In any case, I think it's obvious that FPTP in a multi-party system doesn't represent voter intent very well.

      Whether it's STV, MMP or some other system that tries to determine the intent of the entire electorate rather than turning elections into a dog race, I'd say we'd be better off.

      Give me a chance to rate my preferences and I'll jump at it.

      • Kenneth Arrow did some work he labeled "Impossibility Theorem", in which he allegedly proved that any electoral system can lead to anomalous results. Can't say I've ever studied it that closely, but it would be interesting to look at some time when I'm bored. When I think of ranking candidates, I think of who Alberta ended up with for premier. Two capable candidates hated each other so much they all selected the inoffensive and incompetent third runner up as their second choice.

  5. As problematic as predictions might be, the "science" of seat projections seems to be quite a bit better than the reporting of same.

    • amen

  6. So basically, we will have no idea until after the election.

    Sound like the same type of science as weather forecasting. Doesn't matter if you are wrong. lol

    • Sounds like CPC governance, actually.

      • Silly comment. No relevance to my post.

        • Really? So you say think you have some idea of what the CPC will do after the election? Tell me, what mystical auguries are you basing this on? I'm pretty sure it's not Mr. Harper's word.. one would have to be an idiot to believe that at this point.

          • Aaron wrote an article about how we really can't trust the seat projections. I commented that that it must be great, doing seat projections, as no one apparently expects you to be accurate. You somehow think that this links to CPC governance? I stated that your post had nothing to do with what I was commenting about. You have no gone into some nonsense about me having some idea of what the CPC will do after the election. . .

            You lost me at 'hello'.

  7. And I agree. However, let's be honest, if we were running STV here, parties would be running multiple candidates to fill the seats, and we'd see the exact same results.. even with 20% of the vote, the Liberals still wouldn't be getting second or third place seating out here, and as such, the opinions of the Liberal voters would be totally wasted. My system at least takes them into account.

    MMP takes the democracy out of the hands of the people and puts it into the hands of the party even moreso than it is now. I don't think anybody thinks that's a good idea.

    • I'm not sure. In a 4 seat STV distict in Alberta, if CPC gets 60% of the vote, they are likely to only get 2 of the 4 seats, while depending on the vote share, Liberals and NDP would likely each get one. You'd see something similar in Toronto. With larger districts you get even closer to proportionality. I wouldn't mind going up to 500 MPs if it helped in this regard. There are other benefits to having too many MPs for all of them to be able to jockey to be in cabinet.

      It sounds to me like you don't quite get how STV works. Yes, parties can have as many candidates on the ballot as there are seats in the district, or even more. Voters still get to decide which of the candidates they like most, allowing new blood to push out old incumbents of the same party that do not serve their constituents well. And just because, say, the CPC were to run as many candidates as there are positions in a Calgary distict does not mean they would take them all. STV would select at least a member or two from another party–the results would indeed be quite different, and roughly proportional to the vote within the district.

Sign in to comment.