Tuesday, 3 June 2014

Stick Shifts and Voting

Anyway, I've hinted at or maybe directly stated that I mostly don't bother voting. I'm hardly alone in this, voter turnout is falling every election and has been for a long time. Canada has a rotten election system and the stink of it seeps into you eventually. It is hard to feel like your vote matters in a system where there are very good odds that it doesn't.

Basically if you are not voting for one of the top two candidates in your riding, you are voting for nothing. Deeper than that, if the difference between the top two candidates is more than just a little, your vote is worth far less than it should be, but if the difference is small it is worth far more than it should be. Because of this, many people believe in strategic voting - that is, you want your vote to count, so you limit it to the top two candidates.

But strategic voting doesn't work. The theory runs into a pretty big snag when it hits the reality of figuring out which candidates are the top two in any given riding. If good polls were conducted at the riding level then we'd have some idea, but good polls are expensive and take time to conduct, and there is necessarily a lag between them and election day in which public opinion can shift. Strategic voting sites used the results of the previous election in 2011, but it turns out the results of the previous election are very bad predictors of the next election.


This chart summarizes the win percentage of incumbents. The horizontal axis shows the percentage of the vote the incumbent got in the previous election, the blue bars show their win percentage in the following election. The red line shows the win percentage of all incumbents at or above the value on the horizontal axis, so of incumbents who won with 42% of the vote, about 85% will won the next election.

This data is Canadian elections from 2004 to 2011. Since there were elections in 2006 and 2008, that means it's three full elections and follow-up election pairs. Anyway, that's not actually a lot of data points, especially not since I'm not considering candidates who didn't run in an election following a win. As you can see, the data looks a little bit skewed by one point up in the 70s.

This is really the reason why strategic voting doesn't work. Of 811 data points, one was Todd Russel. In 2008 he won Labrador with 70.3% of the vote. In 2011 he lost with 39.1% of the vote to Peter Penashue who got 39.8% of the vote. The actual difference was 79 votes. That's less than 4% of the people who voted for the third place candidate.

Of course you never would have known this if you looked at a strategic voting site that based it's recommendations on the previous election. Todd Russel was a lock, it didn't matter who you voted for in that riding.  It turned out to be one of the closest ridings in the country.

So the problem is that the future does not sufficiently resemble the past to do this kind of reasoning.  If each person going to the polls in Labrador had a 70% chance to vote for Todd Russel then his not winning would have been essentially impossible.  But because people change their minds between elections, it happened more than 0.1% of the time.  Because of the small number of data points we don't know if it's 0.1% or if this was a real fluke or if the low number of major turnarounds was a fluke.  It's also important to note that none of these elections had a change in government, so the incumbency rate was probably actually a little high.

So I guess voting isn't a complete waste of time when your riding seems thoroughly decided.  It's more like wearing a seatbelt.  Sure, most of the time it does nothing, but you really don't know when it will do something and when it won't.  The only poll that counts, as they say, is election day.

No comments:

Post a Comment