BC projections: close race but the odds favour the Liberals

Here it is, the BC model for the upcoming 2017 provincial election. It took longer than expected and this is very much a version 1.0, but it is here! And since we got a new Mainstreet poll yesterday (just before the BC budget), the timing couldn't be better.

In order: Voting intentions; Seat projections with 95% confidence intervals; Chances of winning the most seats


Since this is the first projections, I'll spend more time talking about the model itself than the projections or the political situation in BC. We are 3 months away from the election anyway. I'll go with the bullet points approach.

- The model uses the new electoral map that added two ridings (one in Richmond and one in Surrey), bringing the total to 87. It was a giant pain in the butt to transpose the 2013 results myself. Election BC decided not to provide such a transposition, an odd decision since every other election agency in this country does it. They told me that because a significant portion of the votes were only counted at the riding level (as opposed to the vote section; this is the case for advance voting for instance) that they couldn't be confident enough in the transposition. Fair enough but it meant that I had a lot more works. BC Election is right in that a perfect transposition isn't possible, but that's the best I could do. Honesty, the new map didn't change much in most electoral districts. It does add a level of uncertainty though and it could make a difference in some close races.

- The model includes regional coefficients. BC isn't the best province to estimate those coefficients, for a couple of reasons. First of all, recent elections have seen very small swings for the main parties, except in 2005. Little variation means it's harder to estimate things (a basic statistical fact). On top of that, you don't see the same strong and consistent regional patterns as you'd in Quebec. Let me explain. In la Belle Province, you systematically observe that the main parties experience larger swings in the Montreal suburbs and around Quebec city. In BC however, you often see a party going up in one region while going down in another. And there is little consistently across elections. All that to say that the model does include regional coefficients based on very general trends, but people need to realize that projections are more difficult to produce in BC than in Quebec.

- As I mentioned before, this is very much a v1.0 model. I did account for the fact that independent MLA Vicky Huntington decided not to run again in Delta South (a move advantaging the Liberals). But over the next 3 months, I'll need to make a lot of little adjustments. One of those will be to include the effect of losing a long term incumbent, a phenomenon that has hurt the Liberals in the past. 
It's also very possible and likely that there are a couple of errors. If you spot something, please let me know.

- As it stands, the projections assume that the BC Green and BC Conservatives will both run a full slate of candidates. This is highly unlikely to happen. And this could be important. In 2013 for instance, my estimations show that a big determinant of the observed swings was whether the Green party stopped running a candidate (they had a candidate in 2009 in this riding but not in 2013). And surprisingly, my estimations consistently show that the BC Liberals were the one benefiting from that, not the NDP. As for the BC Conservatives, they ran a lot more candidates in 2013 and in some regions, it really paid off (surprisingly on Vancouver Island). So, as we'll learn more about how many candidates (and where) these parties will run, I'll be able to make the necessary adjustments. It is also important because a party polled at 10% but running candidates in only half the ridings is most likely overestimated by these polls and we need to take that into account.

- The probabilities are obtained by doing 10,000 simulations where I account for the uncertainty at the province level (is the BC NDP at 35% or 39%?) as well as the regional distribution of the vote (two simulations with the NDP at 35% will not give the exact same results). I also have within-region correlation (if the NDP is higher than expected in one riding in Surrey, this party will likely be higher in the other Surrey ridings). There is a lot of uncertainty and randomization. So much that the model, as it stands, has not made a mistake when calling a riding at 100% in the last couple of elections (federal in 2015, Alberta 2015, Quebec in 2014, Ontario 2014). It doesn't mean it doesn't make mistake, but it  has represented the uncertainty very well so far.

Ok, enough about the model, what about the projections? First of all, this is obviously a close race. Liberals and NDP are tied at 37%. Polls in recent months have not been very consistent. With Ipsos and Innovative putting the BC Liberals ahead while Mainstreet or Insight West had the NDP leading. For the projections here, I used the Mainstreet numbers only. Why? Because they were the most recent. The caveat obviously is an increased uncertainty. And we all know how wrong the polls were in BC 4 years ago.

This is why I believe it's better to focus on the probabilities rather than the pure numbers. And these probabilities show that while it is indeed close, the Liberals have the edge. One of the main reasons is because the NDP is currently more at risk with the Green Party. This one is polling incredibly high and would be a major player on the island (thus hurting the NDP). Of course, the real question is whether the Green party can really get over 15% of the votes in May. Only time will tell, but as it stands, the Green party would elect 3 MLAs (obviously keeping the riding its leader Andrew Weaver won in a by-election). They would also have a chance in 12 others ridings, although the chances are just above 0% in many. A more conservative approach would conclude that the Green party currently has a chance in 8 ridings at best. This would be absolutely major and would likely cause a minority government.

Speaking of which, the odds are around 55% for a Liberal majority, 6% for a Liberal minority, 28% for a NDP majority, 5% for a NDP minority and around 4% for a tie (which would in theory allow Christy Clark to remain Premier, but let's discuss about this another time). The fact that both main parties can technically win a majority really shows how uncertain an election tomorrow would be. I think the merit of these projections is to show that there is indeed tremendous uncertainty but that the Liberals are favoured (the odds are quite close to the final odds of FiveThirtyEight for Clinton vs Trump). There isn't a strong bias of the electoral map but the Liberals do see slightly more efficient with their votes, at least in average.

As for the BC Conservatives, they are also polling well above their results of 2013 and many will be skeptical as to whether this party can really remain that high during the election. As it stands, they are still too low to be projected ahead in a riding but they have non-zero chances in 304, in particular in Peace-River-South. Be careful not to think that the BC Conservative only take their votes from the Liberals. The second choice provided by the Mainstreet poll show that this is far from true.

Personally, as far as having an interesting campaign, I'd love for the Green and Conservative to remain that high. But I have my doubts.

If you want the detailed, riding by riding projections, you can find them here.

That's all for now. If you want to use the model with your own numbers, you can use the BC simulator here.