Wednesday, 4 June 2014

From Third to First

Yesterday I wrote about how often incumbents won and lost the next election based on what percentage of the vote they had. The conclusion is that if someone got around 56% of the vote or more then you could safely assume they would win the next election, unless there is a fluke and the next election is one of the most hotly contested elections in the country where every vote counts. From the data I have we don't know how unlikely such flukes are, but we do know that it is probably more than one in 10,000 and probably less than one in 100.

The instance of such an anomolous event in the data was Todd Russell losing to Peter Penashue despite having taken the previous election with over 70% of the popular vote. Something I didn't mention was that in the 2008 election where Todd Russell won, the Conservatives - the party of Peter Penashue - came in third. Using the previous election's data, there was no way to predict a Conservative win. One would think that if the Liberals were somehow going to tumble from their great height Labrador that it would be the second place NDP who would benefit.

We know that incumbents win most of the time, that incumbents who had very low vote shares win not-as-much as incumbents who had very high vote shares, and that we can't rely on these things. How reliable is the idea that the third place party in the previous election won't win - was Labrador just another fluke in this regard?

Between 2004 and 2006 there were some riding changes, so I lost a few data points by having to ignore those ridings, but not as many as I lost to retirements in the incumbency data. Overall, 882 election results could be compared with previous elections. Of those, 72 resulted in a previously third or lower placed party winning. One of these cases was Bill Casey winning his Nova Scotia riding in 2008 as an independent having won it the previous election at a Conservative, which we can probably call a very strange case that possibly shouldn't be reflected in the data.

Of the remaining 71, 25 were cases of the third place party winning the next election and 46 were cases of the fourth place party winning the next election. More fourths than thirds winning might sound weird unless you know your Canadian election history. In fact, 55 of these third and fourth place winners who went on to win were NDP candidates in Quebec in the 2011 election.

So if we count out both people leaving their parties to run as independents and the Orange Crush, we are left with 15 cases of third place parties winning and 1 of a fourth place party winning. That doesn't sound like a lot, but remember that incumbents win most times. There are actually only 117 cases where the second place party goes on to win the next election. So when things like the Orange Crush aren't happening and an incumbent is overturned, they are overturned by the party that came second in the previous election about 86% of the time. That leaves about one in seven where the person sending the incumbent party packing is from the third or worse place party.

And, of course, you may be wondering why I dismissed the Orange Crush as an anomaly. Well, it was an anomaly, but it was an anomaly that happened in one in four elections I have data on. Unlike the Todd Russel case, we can easily look at history and think of times something similar happened. In Ontario in 1990 the NDP won a stunning victory, gaining 55 of the province's 130 seats. They were in second place after the election previous, but they were not in second place in many of the ridings they went on to win and the fact that they had edged out the PCs in the previous was strange itself. The next election the NDP rocketed back down to third as the third place PCs from 1990 gained 62 seats. In 2007 the CAQ rose from 4 to 41 seats in the Quebec legislature. In 1999 the brand new Saskatchewan party replaced the Conservative party and overtook both the Liberals and the NDP in 25 of 58 seats. I'm sure more examples exist.

Just as you can count on the incumbent to win except when you they don't, you can rely on them - when they lose - to lose to the party that came in second in the previous election, except when that doesn't happen. When there is a large change in political winds it seems to catch everyone by surprise, so you can't really count on anyone to predict it for you, least of all people who are using past results as their sole predictor of the future. Besides which, in the next federal election, would it really be such a surprise if the Orange Crush retreated and Quebec went largely Liberal? We could have another "anomalous" turn around very soon.

When things happen 80% of the time it's pretty easy to look like you know what you are talking about. You make all the safe guesses and then point out that you were mostly right. So bet on incumbents, and if not, bet on the second place party from the previous election, but if you actually care who wins, you can't really bet so heavily as to think your vote can't matter. It turns out that it nearly always can.

No comments:

Post a Comment