The BC Conservatives got 2.1% of the vote in 2009. However, they achieved this by running only 24 candidates. So make no mistake, these 24 candidates got in average way more than only 2.1% (in fact they got 7.4%, with one candidate getting as much as 20%). This time around, they are currently polled around 10-12%. Quite a a big increase. But the good news is attenuated by the fact that less than a year ago, this party had a real shot at passing the BC Liberals in the polls where it was standing above the 20% mark.

I don't want to cover or try to explain why the BC Coservatives have dropped so much in the last 10 months. What I want to do is look at the number of seats this party could potentially win. By doing so, I'll compare my projections to other methods.

With the current voting intentions, the BC Conservatives are projected to win 2 seats (Kelowna-Lake-Country and Kelowna-Mission) by small margins. In the best case scenario, the party of John Cummins would get 9 MLAs. If you look at the most recent projections (just click on the picture in the right column), you see that the CP is actually in play in 11 ridings (i.e: where its probability of winning is >0%). But except in the above mentioned ridings in Kelowna, the odds of winning are very low. Still, 9 MLAs would be quite an achievement for this party and would constitute a surprise few saw coming. Of course, the best case scenario mentioned here is possible but unlikely.

How can my model project so many Conservative wins while, at the same time, projecting no more than 1-2 victories for the Green party? The answer is simple: vote concentration (note: I'll talk about the Green very soon, doN,t worry Green fans). With our electoral system, a small party is better concentrated. A big party is better widespread. The model uses past election results to estimate regional coefficients. In the case of the Conservatives, the task wasn't easy since they only ran 24 candidates in 2009 and even less in 2005. Nevertheless, looking at the (few) data, it was clear this party was more succesful in some regions of the province, mostly the interior. I worked hard and came up with some regional coefficients reflecting this fact. For instance, it is assumed that when the Conservatives takes 1% from the Liberals province-wide (as seen in the polls), they will actually take 1.6% in the Okanogan but only 0.3% in Vancouver.

The other piece of the puzzle here is to acknowledge the source of the Conservative growth: The BC Liberals. Just look at the most recent polls by Angus-Reid and you'll see that while only 5% of people who voted NDP in 2009 would currently cast a ballot for the new right wing party, this number is as high as 14% among Liberals. Actually, Christy Clark is losing almost as many votes to the Conservatives as to the NDP! The model takes this fact into account. It assumes that the Conservatives' share of votes is taken proportionally more from the Liberals.

So when the Conservatives are increasing from 2.1% to 12%, this swing is assumed to be more concentrated in the interior than in Vancouver for instance. It is also assumed to come at the expense of the Liberals. So ridings where the Liberals got a lot of votes in 2009 are better for the Conservatives. The "pool" of potential voters is bigger. This is why the model currently projects wins in the two Kelowna ridings, even though the best score in 2009 for the CP was in BoundarySimilkameen. The difference? In the latter, the Liberals only got around 35% of the votes, while they got above 50% in the former.

These are assumptions of course, but they are based on past regional trends. Still, let's compare our projections to other models and methods. Specifically let's compare for the Conservatives (comparing the general outcome is pretty useless as everybody will agree that the BC NDP would win if the election was held today).

Method 1: redistribution matrix from prof. Werner Antweiler (UBC-Saunder School of Business).

This method (and simulator) relies on the principle that voters don't vote for the same party every election. And if one party is experiencing a positive swing, it must come from one or more other parties. While the principle is sound, the application is more tricky as it requires you to either guess the matrix or rely on data not easily available. Indeed, polls usually don't provide us with this information. We simply see thet net, aggregate effects. For instance, we see that the BC NDP is increasing, so are the Green and the Conservatives. But we don't see the exact transfers of votes between these parties and the Liberals for instance. The latest poll from Angus-Reid does provide us with some partial information regarding this (page 6). Using these numbers and playing/guessing the other numbers in order to match the province-wide percentages of the polls, we would find that the Conservatives aren't projected to win any seats. In Kelowna, they would fall short by about 10-points in my scenario, it really depends on your matrix.

Why the differences? Well, as opposed to my model, this matrix doesn't have regional reffects. If you input in the matrix that 14% of the Liberals are now voting Conservatives, it'll use this ratio in every riding. While my model would transpose this 14% differently depending on whether the riding is in downtown Vancouver or in Kelowna.

Method 2: prediction market.

This method doesn't use polls, at least not directly or explicitly. It is based on trading with real money. Just click on the link and read the details. The seat projections are similar to mine except that the Conservatives would get slightly more seats (5) than in my most likely scenario. But given that I have other scenarios where this party would get as many as 9 seats, it seems 5 is a good middle ground.

Why the differences? Well, in this case, the predictions are a lot less different than using method 1. It seems the traders are aware of the concentration of the vote of the Conservatives. Looking at the predicted shares of vote, it seems that traders also believe that the BC Conservatives are underestimated in the polls. Ultimately though, it's hard to directly compare and explain discrepancies as the two methods are widely different. Since historically predictions based on trading have performed relatively well, I see that as an encouraging sign that I'm not completely loss with my current projections for the Conservatives.

Method 3: other projection websites.

I'm not the only one blogging about this. Hey, some even do that for a living despite using very simplistic models. As opposed to my model, the difference is that the other models usually don't include regional effects and simply apply an uniform or proportional swing (read the FAQ for more details). Or if they do include regional effects, it's entirely based on the current polls and not the past election. In theory, you could use the regional breakdowns of the polls. In practice, the sample sizes by region is so small that you are at the mercy of very imprecise estimates. Right now, the polls don't show any concentration of the vote for the Conservatives. Indeed, the AR poll actually has the Conservatives at the same level in the lower mainland and in the interior! To me, this can't possibly be right. But models who use these numbers directly will obviously assume a very uniform swing across the province for the party of John Cummins. This leads to a lot less wins than my model or method 2.

Who is right? Well honestly we won't know until election night. However, in the past, uniform or proportional swing models have performed poorly for small parties. If the Conservatives really has a vote distributed almost evenly everywhere, then this party will likely not win a single seat. However, given the past election results and the ideology of the party, it isn't unreasonable to imagine some sort of concentration of the vote in a couple of key regions. My model probably overestimates this concentration while method 1 and 3 probably underestimate it.

At the end, I'd say that it's safe to assume that the Conservatives will be in play in about 5-10 ridings and could win some of them. Luck, turnout and vote concentration will play key roles in determining whether this party actually gets MLAs or not, despite experience the highest positive swing of all parties in the province.