Close Races Rarely Endure
The following article appears in today's Hill Times, and is reprinted here with kind permission. Notably, the versions of my tables should be a LOT easier to read here than there.
This is my rejoinder to the many "seat targetting" and "seat projection" methodologies that are proliferating, and which ALL accord some magical status to the seats that were only narrowly won last time around.
Here's what you need to know:
- only 12-16% of ridings wind up with margins of 5% or less in any election
- of them, maybe a quarter will still be close the next time
- only 8 ridings have been close three times in a row since 1988
- only 1 riding has been close four times in a row in that time period
- no by-election has even been close after a close general election since 1988
- no general election in a riding has ever been close after a close by-election either
- more of the ridings that switch hands in any election were NOT close in the last election, than those that were
More evidence for my "prime directive" that "past results don't predict future outcomes". Like "Fortress" Vaughan, and "Squeaker" Parry Sound — Muskoka. Inevitably there wind up being other factors in play.
———————————–
So close, but no cigar: close ridings aren't always safe target ridings
In the last 20 years, no close seat in a general election remained close in a subsequent byelection. And no close byelection race had ever been a close race in the previous general election. Not one.
ALICE FUNKE
Published January 31, 2011
OTTAWA—The narrow defeat, the one that got away, the squeaker victory—the drama of all these close races is remembered long after election night, and is a big part of the adrenaline rush that feeds political junkies election after election.
But does what we remember so vividly about those close races distort our ability to assess the next election?
We've all seen barstool pundits pull out a pen, and start a back-of-the-napkin list of target seats with the ones that were close in the last election.
As a quick way to handicap the next election, how reliable a method is that? The answer, sadly, is not very good at all.
In a typical general election over the last 20 years, 12 per cent to 16 per cent of the ridings are close races—that is, races where five per cent or less of the vote separates the first and second-place finishers. The 2004 election came in a bit higher at 19 per cent, and 1988 sported an unusual 24 per cent.
Of those close races, roughly a quarter of them will go on to be close in the next election. That's slightly higher than average, but still means that just three per cent to four per cent of all ridings will be close two elections in a row.
Let's put that another way. Of all the races that weren't close last time, all other things being equal, they stand nearly as good a chance of being close this time as any riding in the last election does, whether it was close last time or not. This will be good news for the Liberals who picked the ridings for leader Michael Ignatieff's current tour. See Table 1.
Table 1. Closeness of Close Races From a Previous Election
[Click on table to open full-sized version.]
Moreover, in the last 20 years, no close seat in a general election remained close in a subsequent byelection. And no close by-election race had ever been a close race in the previous general election. Not one.
The truth is that parties with good opinion research and demographic data, and strong candidate search and local knowledge, don't have to fall back on "which ridings were close last time" as the basis for their target riding lists.
But suppose they did, and the result of their effort was for a previously close seat to change hands and its margin increase beyond five per cent. There are a lot of seats like that, aren't there?
Well, there are some, obviously. In fact a greater proportion of previously close races changed hands in the next election, than did seats that weren't close the last time around.
On the other hand, in every election studied, more of the seats that flipped by party had NOT been close in the previous election than had been. See Table 2.
Table 2. Likelihood of Close Races From a Previous Election to Change Hands in Next Election
[Click on table to open full-sized version.]
Furthermore, while there is an overall relationship between ridings' percentage margin of victory and their previous and later margins of victory, that's not at all true for the close races.See Table 3.
Table 3. Relationship of Margin in Current Election to Margins in Previous and Next Elections
[Click on table to open full-sized version.]
So clearly there are some qualitative factors like local history, local organization, demographics, and candidates that make a riding close in any given election, and some magical combination that make it close in repeated elections.
The Eight Longest-Running Close Races in Canada
Eight ridings across Canada have been close three times in a row, over the past 20 years. These have been the knock-'em-down-drag-'em-out fights – and the perennial nail-biters – and every indication is that many of them will continue that way in the next election.
Let's take a closer look.
[Click on table to open full-sized version.]
[*** UPDATE: Thanks to a reader for reminding us of the precise details in Dr. Martin's case. He left the Alliance to sit as an Independent, and was nominated as a Liberal before declaring as such in the House of Commons.]
Three of the eight are in the Montréal area, two in the north end (Ahuntsic and Papineau) and one just south of the Champlain Bridge (Brossard-La Prairie). All three switched from the Liberals in 2004 to the Bloc Québécois in 2006, with the Liberals regaining two of the three under leader Stéphane Dion in 2008. They clearly signalled, with Leader Michael Ignatieff's recent tour itinerary, that winning Ahuntsic back is one of their campaign priorities, while the Bloc will be running Vivian Barbot again in Papineau.
Meanwhile Brossard-La Prairie, Que., actually moved into four-way race territory in the 2008 election as well, such that other parties cut into the Bloc's support, allowing Ms. Mendès to win in a surprising upset, even as the Liberal vote there fell. All five parties identified or nominated their candidates here long ago.
Sault Ste. Marie in Northern Ontario hovered around the threshold of being a three-way in 2004 and 2006, but became a straight two-way race in 2008 as the Liberal opposition gave way to the Conservatives. The Prime Minister has visited the riding, making a famous speech about the gun registry, and Conservatives held their national caucus there last August. However, sadly their most likely candidate to challenge New Democrat Tony Martin, mayor John Rowswell, lost a short-lived battle to cancer last summer.
The Edmonton entry demonstrates clearly how Anne "Landslide Annie" McLellan came by her nickname. She won her old seat of Edmonton Northwest by just 12 votes in 1993, but redistribution put the Liberals behind the Reform Party on the new boundaries. Nevertheless, she managed to eke out a 3.4 per cent margin of victory over the Reform Party in 1997 and another 1.5 per cent margin in 2000.
The remaining three seats are in British Columbia, one in the Vancouver suburbs (Burnaby-Douglas), and two on Vancouver Island (Esquimalt-Juan de Fuca near Victoria, and Vancouver Island North). Two of the three now have retiring incumbent MPs (New Democrat Bill Siksay in Burnaby-Douglas, and Liberal Keith Martin in Esquimalt-Juan de Fuca), and all three are clearly being targetted by the top two if not three parties from the last race.
NDP Leader Jack Layton in particular just attended the nomination of his returning 2004-06 Esquimalt-Juan de Fuca candidate, Randall Garrison, the week before the House returned, while the Liberals have a contested nomination already under way to replace Keith Martin, and 2006-08 Conservative candidate Troy deSouza has been in place and compaigning for over two years.
Burnaby-Douglas is the only riding I found in the last 20 years to have had four close races in a row (2000-2008), and it was also a three-way race in the first three of those four elections (2000-2006), only moving to a purely two-way race in 2008 as the Liberal vote there dropped. The riding's changing demographics, coupled with its close margin and now incumbent MP Bill Siksay's retirement, make it a prime target for Immigration Minister Jason Kenney's outreach campaign to new Canadians and the ethnic communities.
– 30 –
Alice Funke is the publisher of PunditsGuide.ca, and an election data analyst (alice@punditsguide.ca)
Tags: Close Races, Data Analysis
What’s the implication for projections sites like democratic space or 308 then ?
So much of our political discourse seems based on the idea that the recent past will predict the future, from dismissing Larry Smith’s chances or treating the LGR as a non-issue because a lot of ridings where its unpopular weren’t close last time.
Well, if I were developing a projection model, Shadow, I’d want to validate it by ensuring that it projected earlier, known results using the same methodology. Also, I’d be very limited in my application of the results; for example, I would not say that so-and-so is “doomed”, or such-and-such a seat is a “fortress”. I’d just say that according to the methodology, a certain outcome is predicted, but acknowledging the limits of that methodology.
Numbers don’t have lives on their own. They just represent the sum of all human actions and interactions at that moment in time. You need to have some real life experience to imagine what all those could be, in order to interpret any set of numbers properly, in my opinion.
Do you think that’s fair?
So essentially Alice, you are saying you want to make a predictions model similar to Nick Silver at FiveThirtyEight?
Nate Silver? No, I just don’t believe there’s really enough low-level opinion research data in Canada to do that properly.
I am getting quite interested in documenting lower-level vote swings, though, hence my focus always on describing party vote as a percent of the electorate, and including non-voters in any analysis. To be able to build a model that could describe those historically one day, maybe at the census tract level, would be a real accomplishment.
Silver only has to predict binary outcomes. There was a fascinating methodological discussion between him and some european contemporaries on the best methodology for predicting parliamentary elections with multi-party systems, the lessons of which I haven’t fully absorbed yet, but I think that would have to be a starting point, for sure.
The problem is that we don’t have a complete dataset from a full electoral cycle yet. I would classify a full electoral cycle as being, for example, from 1988 to 2008, but so many data points are not available for the elections prior to 2000. Meanwhile the party system also changed substantially over that time, as has the composition of the population. It seems very difficult to me to tease out the impacts of the electoral cycle from all those other changes, and come up with something that can be sufficiently predictive going forward.
Remember that some elections are incremental affairs, and some are transformative. Nothing about 1988 predicted 1993, and nothing about 2000 predicted 2004. We don’t know if the next transformative election is going to be the 41st general election, but it might well be. In which case any of the seat projections being done now are not terribly predictive.
I’ve seen people try to defend the accuracy of the projection methodologies based on the ability of the final polls before E-Day being able to predict the outcomes. Well, of course they can: first of all they’re polling a population that is finally paying attention to the subject matter, they’ve all bumped up their sample sizes and put themselves on their very best behaviour in terms of (otherwise costly) sampling techniques, because the pollsters know that those are the polls they’re going to be judged on.
But, to cite the most popular statistic from Michael Marzolini’s work for the Liberal caucus: 15% of the public is paying any attention to politics right now. I would like to see opinion research data being plotted the way BC2009.com did for the last provincial election, in terms of the number of days before Election Day. Assume E-Day is 36 days or so after the first likely budget vote: how predictive were polls this far out from other elections of the final seat counts … in 1988, 1993, 2004, 2006, etc. That’s what would convince me a bit more.
OK, I’ve gone on a bit too long on this one, but tell me what you think. I realize this position has made me a bit of a contrarian, but that sort of positioning has never really bothered me.
Fair points i’d say.
Thinking about what an election hinges on (since its not close ridings), I wonder if what’s more important than the numbers themselves is the demographics of the ridings.
One of the things Nate Silver seems to do with his models is analyze the strength of correlations between all kinds of demographic factors and election outcomes.
The performance of the parties between elections and the type of messaging they’re using would appeal to different demographics, potentially scrambing the coalition of voters each party is appealing to and therefore changing their results in every riding.
Thinking about it this way election predictions based on past performance seems pretty sketchy.
Don’t forget, that the demographic data Silver has access to is FREE in the U.S. In Canada, it’s unbelievably costly, below the census tract level, and moreover we won’t have most of it reliably collected in the next “census” either.
Calgary Grit has done some work to try and model the factors that reliably predict different parties’ support using riding-level data (he probably has access to more detailed versions through his workplace). I don’t have a second to fish out that link right now, but it shouldn’t be hard to find with a well-formed Google search.
Anyways, thanks for the feedback Shadow. I thought that post had really gone out into space as quite a clunker for people.
The type of polling needed to properly use demographic information is incredibly expensive too. That Liberal poll had an eye popping price tag.
It was a good investment for them though.
They’re catching up to the CPC which has been using a market research/focus group approach to target voters for ages now.
Its unfortunate that in Canada the only people with the resources to do this kind of analysis is the major parties.
Its one of the drawbacks of being a small country I suppose.
In politics they are, not in business of course. And, you’re right about the price tag probably being high, although the source for that estimate was another (competing) pollster.
Parties who are serious about developing their election strategies will do a large “baseline survey” like this, as a point of comparison for later rolling-tracking-type polling, and from which to draw the demographic characteristics of their likeliest supporters for further research in focus groups.
However, it’s always surprised me the extent to which the Liberals will publicly discuss their internal polling. Other parties are undoubtedly doing the same kind of thing, but you don’t hear them blabbing about it to the Globe and Mail.
Yeah I took that strategic leak as two things:
1) Equivalent of NDP showing off the war room – we’re ready to go.
2) Meaningful pushback against pollsters and pundits who had written them off.
Yeah sure we’re behind in the polls, but those polls only capture 15% of the public. What about the other 85% ? We’ll win a majority with them!
(Of course, considering turnout it’ll only be the other 45%. And what evidence is there that voters who don’t pay attention between elections are somehow more Liberal friendly ?
If they’re less educated it could actually mean the opposite …)
I think the Liberals have just become habitually desperate to spin any good news.
Its not comparable to the NDP showing off the war room. You just dont talk about internal polling. Period.
All parties- provincial and federal have some pretty extensive internal discussions of internal polling that reaches well beyond the inner sanctum. At least dozens, and often 100 people, get a thorough analysis of the numbers and the strategic implications.
Even though the bulk of those people are volunteers rather than career staffers, there are rarely leaks that are any more than vague and general rumours. And those people frequently see things that would be fun to brag about in the media.
some very interesting stats here , i wouldn’t of guessed that so few ridings have been close on a regular basis since 1988 . i know that its not unusual for a close riding to suddenly not be close especially if the mp is high profile but for only 8 ridings to have been close 3 times in a row is surprising to learn .
Alice,
Could you figure out if the NDP are going again with Chief in Win-North. If they are not, and have no candidate prospects…there will likely be NO general election this spring.
Then again maybe a JWL rebirth !
I haven’t heard any names for Winnipeg North, yet, Cynthia. Have you? I believe I read an interview with Judy WL recently however where she said she was not running again.
The NDP will not force an election if they do not have all of their ducks in a row in their most important future +1.
I’ll keep you posted.
I see your point, Cynthia. What makes you believe they aren’t running Chief again? His new baby?
the ground is quiet
Well, given your record for “fore”-sight here, Ms. “fore”-man, I’m inclined to listen to what you have to say :-)
I’ll certainly be keeping my eyes and ears open now, for sure. Thanks for your comments.