Why Bother With Past Election Results?

A follow-up to the pieces How Difficult Will It Be For the Federal Liberals to Win an Election? and How Difficult Will It Be For the Federal Conservatives to Win an Election?. An obvious question to ask is:

Who cares what happened in, say, 1911? What relevance does that have today?

The answer is simple:

If we understand where are party's geographic strengths are and why seats change hands, we can better predict the outcome of future elections.

In my view, we can classify seats changing hands for four (not necessarily mutually exclusive) reasons (or factors):

1. Campaigns and Policies. What we all tend to focus on when discussing politics. "Can Harper sell his tax policy to voters?", "Will the electorate warm to Michael Ignatieff?". That sort of thing. This effect can be rather large - think Trudeau in English Canada in 1968 or Mulroney in English Canada in 1984. However, the effects mostly do not persist. I use the modifier 'mostly' due to the 4th factor.



2. Tectonic Shifts. Like the first factor, however these are large shifts in a geographic region that do persist over time. Until the 1980s, these were 'once in a generation' type events. However, I can think of a handful since the late 1970s:
   1. The Liberal collapse in Quebec in the early 1980s.
   2. The collapse of the Mulroney coalition and the rise of the Reform Party and the Bloc Quebecois in the early 1990s.
   3. Re-uniting the right in the early 2000s.
   4. A possible one: The collapse of the Conservatives in Newfoundland prior to the 2008 election.
   The difference between the two factors is primarily determined in retrospect, as shown by the Newfoundland example.
   
   
3. Demographic Shifts. Far less sexy than the previous two factors, but important nonetheless. As people get older they are more likely to vote Conservative, so if the average age of a riding increases, it should help the Conservatives. Changes in the ethnic or religious make-up of a riding can make a difference too (which is one of the reasons political parties use data collected from the long-form census when conducting their streeet level campaigns). We consider urbanization of a riding as a form of demographic change as well.
   
   
4. Incumbency (or lack thereof). It is hard to pin down the value of the incumbency effect. My best guess is that being the incumbent ads two points to the vote share of the incumbent party and subtracts two points from the challenging party. That is, if you had a riding that, without an incumbent, would be 40% Liberal, 40% Conservative, you will get a 42/38 split in favour of the Liberals with a Liberal incumbent and a 38/42 split in favour of the Conservative with a Conservative incumbent.

You can think of other reasons, but I believe these explain the vast majority of cases. Consider the case of Anne McLellan. In 1993, she won an area of Edmonton which had been voting Progressive Conservative consistently since the 1950s thanks to a tectonic change. When the tectonic change reversed itself in the early 2000s, she still managed to eek out a win in 2004 thanks almost entirely due to the incumbency effect. However, that was not enough to hold onto Edmonton Centre in 2006. Now Conservative Laurie Hawn is MP of that traditionally Conservative riding. Barring a tectonic shock or an unbelievably poor Conservative campaign, they should expect to hold that riding for a very long time.

Using these four factors, we can examine the elections of 2006 and 2008 and use them to assist in predicting a few riding level outcomes for the next election.

Comments

[hosertohoosier]

I'm rather surprised that an economist wouldn't seize upon economic voting as a key determinant of electoral behavior. Economic factors come out pretty well in determining elections (for a Canadian example look at Nadeau and Blais' work among others).

Nor is the economy a fundamentally national variable. A key question will be which parts of the country have recovered from the recession most quickly. For instance, over the past year, unemployment in PEI rose 1.6%. I don't think that bodes well for Gail Shea. The fact that employment is down 8% among agricultural workers could be significant in some parts of the country as well.

Some of the factors you cite are not necessarily likely to result in the change of seats, because political parties have agency, and may adjust their positions accordingly. For instance if there was a large increase in the number of religious Canadians, you'd hear a lot more God Bless Canada's from the major party leaders. By contrast, repositioning cannot get around a good/bad economic record (even if it is not reasonable to blame politicians for the state of the economy).

Incumbency is a pretty minor factor (at the local level) in Canada because voters consistently indicate that they vote primarily for a party leader or a party - not their individual member of parliament (and who could blame them, given that MP's have little sway over big decisions).

Moreover, some of your tectonic shifts confuse an outcome (which is what we are trying to explain) with an input. Tectonic shifts do not just happen - for instance the causal chain of Tory decimation in Newfoundland is pretty clear (Atlantic Accord -> angry Williams -> Tory decimation).

[Mike Moffatt]

"Economic factors come out pretty well in determining elections (for a Canadian example look at Nadeau and Blais' work among others). "

You're right. I should add this into the 'campaigns and policy' part. The idea, though, is that the first factor considers relatively short-term phenomena whereas factor 2 and factor 3 are more long-term in nature.

"Some of the factors you cite are not necessarily likely to result in the change of seats, because political parties have agency, and may adjust their positions accordingly."

I disagree (partly) here, partly because I don't think policy matters as much as identity (policy matters, but at the margin).

"Incumbency is a pretty minor factor (at the local level) in Canada because voters consistently indicate that they vote primarily for a party leader or a party"

Agreed, but we're talking about a couple of points here. If, say, 5-10 people in 100 voted primarily based on the local candidate and the 90-95 voted for the party leader/party, we could still see incumbency swings this wide. Though I admit, the evidence isn't as strong here as I'd like, so I could be convinced otherwise.

"Tectonic shifts do not just happen"

Absolutely agreed! I should have made that point clearer.

[Aaron Sheldon]

What happens when the seat is not the seat? Gerrymandering in the form of mal-apportionment, cracking, and packing can significantly shift the outcome.

In the lead up to the disintegration of the Liberals in the early 2000s there was a substantial re-districting of urban, largely Liberal ridings, to favor the Conservatives.

[Winston]

Mike, when asking the question "How Difficult Will It Be For the Federal Liberals to Win an Election?", I think you'd get a more useful answer by looking at archives of voting intention polls since whenever such polls became reasonably good predictors of electoral success.

Let's assume that in the history of frequent and "accurate" opinion polls, no federal political party has ever gone on to form a majority government without polling at least 35% within two weeks of the election (probably not too far of the mark). Let's also assume that the Liberal Party currently stands at 25% in voting intention polls, and that the next general election is three months away (which is reasonable to expect, current analysts say). Then we can use past polls to answer the question, "What is the likelihood that a political party will increase its share of voting intentions by 10 percentage-points within a three-month period?"

I'm not a statistician by any means (as this post might attest to), but the data is certainly available, and judging by the number of polls EKOS has conducted during just the past year, we could have a reasonable prediction based on data from just a few election cycles. Also, the question is much more specific to the current situation of the Liberal Party, so I would think it would yield a more useful answer.

[jad]

While I agree with most of your points, Anne McLellan is not a particularly good example. The main reason she kept on winning was the support, both overt and covert, that she received from Ralph Klein and the provincial PCs. Ralph felt it was important for Alberta to have a voice at the Cabinet table and his people organized and campaigned for her, plus he got to kick sand in the eyes of the Reform/Canadian Alliance. Encumbency had very little to do with it.

[Aaron Sheldon]

Winston, this sounds like an good idea. I would be interested in doing the statistics for it. The major catch will be the geographic resolution of the polls, because the distribution of political support is highly geographically contingent. If the data has poor geographic resolution, then marginalizing over the geography will lead to high variance and low significance. Likewise using standard multi-variate hyper-geometric tests is far to stringent a test because it assumes there are no other contingencies, like geography.