April 17th 2013: The BC election begins. Can the Liberals even win?

The BC election has officially started yesterday. One long month of (most likely dirty) campaign leading to May 14th. As this election begins, polls and analysts would agree on one thing: the BC NDP is the big favorite to win it. Actually, one could wonder if it's even possible for the BC Liberals to win. This is the object of this post.

First, let's look at the situation assuming that the polls aren't completely wrong.

A lot of things can happen between now and the election (after all, there is this thing called a campaign), but as of right now, the odds aren’t good for the BC Liberals. Most recent polls have all shown a lead for the BC NDP around 15-20 points. The gap between the main two parties was closing in fast for a couple of months but this trend has clearly stopped recently. In term of seats, this is what the projections would look like by using the most recent polls (released yesterday) by Angus-Reid.

BC Liberals: 14 seats
BC NDP: 67.
Green: 0.
BC Conservatives: 2
Ind.: 2

The detailed projections (in pdf format) can be found here. You can also find the probability of winning for each riding). If you need the details of how I do that, just read the FAQ. As usual, if you want to make your own projections, use the simulator.

Let’s look at the probability of winning the election. People are relatively familiar with margins of error while looking at polls. However, not everybody really knows how much uncertainty can come from these polls and what it can means in terms of electoral outcomes. On top of that, our electoral system is such that knowing the percentages isn’t enough in order to predict who will win. Therefore, there is actually quite a lot of uncertainty when predicting elections.

The model tries to capture and measure the two main sources of uncertainty: the polls and the electoral system. The first one comes from the margins of errors. When a party is polled at 35%, this party could actually be at 32% or at 38% (or even further away, since margins of error apply 95% of the time). Additionally, our electoral system is such that the same province-wide percentages can lead to very different numbers of seats. So there is this additional uncertainty at the riding-level. Our model takes the polls and runs 1000 simulations, accounting for both sources of uncertainty. It can transpose province-wide percentages into riding-level ones using coefficients estimated from past elections results.

The graph below represents the probability of winning the election (defined as having won the most seats) for the BC NDP, as a function of the party’s lead in the polls. As you can see, as soon as the BC NDP is polled 10-points above the BC Liberals, it’s pretty much game over.


Remember, what it means is that when polled 10-points ahead of the Liberals, the BC NDP is projected to win the most ridings in all 1000 simulations, if the election was held today. Even in the case where the NDP would perform significantly lower than the polls (and the Liberals significantly better), even in the case where the NDP would be “unlucky” in the sense that its vote would be inefficient in getting seats, so even in this worst case scenario, Adrian Dix would still get the most MLAs.

Remember as well, these 1000 scenarios account for the natural statistical variation of polls. Actually, it goes beyond the normal margins of error. Specifically, with the vote intentions given above, there is one scenario where the NDP would be as low as 40%. In another one, the Liberals could get as much as 32.6% of the vote, almost 5-points above the poll. Add to this the uncertainty of the riding level (the second sampling if you prefer), and you realize that the model does account for a lot of uncertainty.

With that said, we are still assuming that the informaiton from the polls isn't completely invalid. In other words: we are assuming that what happened in Alberta will not happen here. Let me remind you of how wrong the polls were. One just needs to look at this picture (from wikipedia) to realize how bad the pollsters did.


I don't want to go over the reasons of this failure. Some pollsters even don't seem to acknowledge the mistake and are saying that people really changed their minds in the last minute. I mean really? around 25% of people switch during the last night? lol This is ridiculous. Sure the last minute poll by Forum did better, but I'm not sure this is really an achievement. It could well be luck. The last minute poll in Quebec by Forum was quite off.

Looking at this graph, the average for the Wildrose alliance was around 40% (let's be hard on this party and accounts for the slight decreasing trend) and 35% for the Conservatives. So at best, we are talking of an underestimation of 9-points for the PC and an overestimation of 5 for the Wildrose. This is beyond what my simulations could predict. I'd need to assume an incredibly large sample size to even have some scenarios as crazy as this one.

So the short answer is no, my model doesn't account for that much uncertainty. And quite honestly, I'm not sure it should. What happened in Alberta was a unique failure, something we haven't really seen before or since. Polls are less accurate now than before (for various reasons), but they are still reliable in average. So, during this campaign, I'll sometimes run a scenario where an Alberta-like failure would occur. If you really think polls are wrong, you can always use the simulator.

So, is the election already over? No, of course not. But as long as the polls will show the NDP with such a big lead, there won't be much suspense regarding the outcome of the election. The silver lining for Christy Clark and the BC Liberals is that the current electoral system seems to favor them. Indeed, look at the first graph. The NDP needs a 3-points lead to have more than 50% chances to win. In other words, the Liberals don't need to pass the NDP in the polls, they just need to get closer to have a chance.

Throughout this campaign, I'll also try to keep you entertain and look at various things. For instance, can the Green win a seat? Can the Conservatives win seats as well? How much damage can they do to the Liberals? etc etc.