Whenever I give an interview and am asked about my model, I always say the same thing: the most important variable to get right is the popular vote. If you nail the percentages for each party in a province, you are 90% there. People always believe most of my work is from building the model. It's not wrong in that it is where I spend most of my time, but it isn't the most important part. Put it another way: a simple model with the right percentages as inputs will do better than a 'sophisticated' model with demographics and what not but the incorrect percentages.
Why am I talking about this? Because for this Alberta election, the percentages you get depend on how you look at the polls. Let me explain. First of all, it is well known that polls in Alberta tend to underestimate the Conservatives. It was the case in 2019 (see below) and even in 2015 (even though nobody really noticed). It was also the case at the federal level in 2021 (more than 5 points underestimation for the Tories). Part of it is likely caused by turnout where older voters vote more. To deal with this issue, I already allocated a majority of undecided voters to the UCP. In this election, I allocate 60% to the UCP and 40% to the NDP. This alone makes the UCP's lead, province-wide, move from +1.8 to +2.9.
 |
| Performance of polls in Alberta in 2019 |
But there is something weirder this time. Polls, in general, are not internally consistent. When you take the raw polling average of the province-wide numbers, you currently get 47.3% vs 45.5%. But if you instead take the average by regions (Calgary, Edmonton and the rest) and you then average those 3 numbers (with the correct weights to reflect the number of voters in each region), you get 49.4% vs 43.4% (that is without adjusting the undecided, just the raw average of the numbers by regions as given by the pollsters).
It doesn't make sense and something must be off with the weights of some pollsters. The more ridiculous example of this is the most recent poll by Research Co. They have the UCP winning the rest of Alberta 63-31, losing Edmonton 35-61 (so it basically cancels out but the UCP still should have an edge) and then the UCP wins Calgary 52-44. Surely the UCP should be ahead province-wide, right? Average 52, 35 and 63 and that's higher than 44, 31 and 61, right? Nope! Research Co has the NDP winning the province 49-47! It makes no mathematical sense. It just can't be. Research Co is an extreme example but we have the same issue overall it seems.
So I'm now left with a decision: do I use the province-wide numbers as main inputs like I usually like (more accurate). After adjusting for the undecided and the lack of candidates for some parties, that would give me 49-46. Or I use the average of the 3 regions and that gives me (after adjustments), 51-44. (Note: no matter which province-wide numbers I use as input, I still make further adjustments to match the regional numbers; Right now it means increasing the NDP in Calgary as they seem up more than in the rest of the province).
You might think that 49-46 vs 51-44 isn't a big deal. Wrong! That's the difference between 43-44 projections and 50-37! It's really all because of Calgary, a few points can shift 5-6 seats. Since the riding polls are more consistent with the 51-44 scenario, I decided to use do an average and use 50-45 as my input. But yes, it does mean I'm quite far from a raw polling average. But at least I explain my reasoning and steps.
So, with the new numbers, here are the projections. They changed a lot from my recent ones on Twitter and I usually don't like this. But the permanent internal inconsistencies have convinced me I needed to do something. I might be wrong and the popular vote will be much closer. But that's a risk I'm willing to take.
The NDP still has a chance but it'll require the 'Mainstreet scenario' of winning Calgary by like 6-8 points. Other pollsters don't have such a lead. Ultimately 22% chances of winning align well with my subjective reading of the race. I think the NDP has better chances in Calgary Bow (strong candidate) and maybe in Elbow. Then add Lethbridge East and Morinville-St. Albert and that's 41. But right now, it's hard for me to see the NDP go over 41-42 seats unless polls were quite off in Calgary.
