The NDP is the wild card. Who is the most affected by an over/under-estimation of this party?
Coup d'oeil sur cet article
Last day of this Ontarian campaign. I'll have my final projections up tonight after Abacus, Ekos and Ipsos have released their last numbers (and probably another Angus-Reid). For now, and before we move on, here are the most up-to-date projections.
The situation remains quite stable. It's close in term of votes, even though the Liberals clearly have a small edge. It's not that close in term of seats and the chances of winning reflect that. I understand the media and parties will try to spin it as a 50-50 race, but so far it's not really the case (and hasn't been for the most part of this campaign). There is however enough uncertainty to make a clear call difficult. The Green party also seems to be falling quickly and it benefits the Liberals. I already adjust the Green down because I believe the polls overestimate this party by allocating undecided voters to it. But with the recent decline, it seems this party won't be experiencing such a big positive swing as some might have expected. It's more likley the result of a tight race (or at least a race seen as close) where voters don't want to "waste" their votes on a small party.
The polls keep fluctuating. Actually, that's only partially true. Look at the graph below representing all the polls to date (note: Ekos has a 3-day tracking, I thus add a new Ekos polls every three days). It includes the last Forum poll of yesterday where the Liberals increased their lead. But it's a small sample size and remember that Forum polls tend to have more volatility (which isn't a bad thing, it probably shows they have true random samples).
First of all, as I mentionned it very often, the trend has been remarkably stable. On the graph above, I'm using a simple linear trend, but it wouldn't change much if I was using a polynomial. And we do observe a convergence for the Liberals and the PC, at least among decided voters (the last Forum poll being the exception of course). Firms with a likely voters model such as Ipsos, Abacus and Ekos still disagree very much once they make their adjustments. But it has more to do with their own models than the polls themselves. However, the NDP now seems to be the wild card. The average is stable (with a slight negative trend) but the polls are now in two camps. One group (Ipsos, Abacus; both online) sees the NDP at 24-26% while the other (Forum, Ekos; both IVR) has this party more at 17-19%. Our trend-average is really between the points with none directly on it. It's quite frankly puzzling, especially since a smaller party such as the NDP has statistically smaller margins of error than the OLP or PC (so we should actually see less variation). The graph above shows that the NDP is more volatile at the end of this campaign than at the beginning.
So who is right? Maybe the new polls of today will show the NDP at 22% and convince us that this is its most likely result. Nevertheless, there remains the possibility that one group of pollsters is right while the other is wrong. And between a NDP at 18% or at 25%, it's a very different situation. Both for the NDP and potentially the other parties. The question is therefore: which of the Grits or the Tories is the most affected by the NDP? Let's take a look.
As usual, I ran my 5000 simulations with the NDP at its average (22%). I thus have half of the simulations above this and the other half below it. I then average the results whether the NDP is closed to 18% (or between 17 and 19% to be exact) or at 25%. Here are some of the key results:
- The difference in term of seats for the NDP is 7 seats in average. It's around a seat per percentage point. It's not bad but it shows the NDP is far from the "zone payante", the zone where each point translates into many seats thanks to our old electoral system.
- The Liberals are down 4 seats between a situation where the NDP is at 18% and one where this party gets 25% of the vote.
- The PC is down 3 seats. Regression analysis confirm that it's affecting the Liberals more significantly. But not by much. Remember, here, when the NDP loses 7 points in my simulations, these points go to the PC and the Liberals evenly (at least in average out of the 5000 simulations). I'm not talking about whether a last minute shift of NDP voters is more likely to turn to the Grits or the Tories. That is simply not the objective of this post. We could argue that it's more likely to see soft-NDP voters switch to the Liberals in order to prevent a PC government. But I'm simply not having this discussion here.
- The chances of a majority for the Liberals decrease from 25% to 15%.
- The chances of winning for the Liberals don't change significantly and remain stable (at 77-78%) whether the NDP is at 18% or 25%. Again, don't be mistaken. If you have a massive shift of NDP voters from the NDP to the Liberals, Kathleen Wynne would benefit from it. Here I'm looking in average, thus the NDP voters sometimes shift to the PC as well.
All in all, while it could (and will) make a big difference for the NDP whether this party gets 18% or 25% of the vote, the effects on the other parties are pretty even. The Liberals seem slightly more affected but the difference with the PC is small. The main effect would be to decrease significantly the chances of a majority for Kathleen Wynne. But this part is obvious: in order for a minority to occur, you need a strong third party getting a lot of seats. So the NDP is indeed hard to predict based on the polls, but let's not exaggerate how important this party could be at deciding the winner.
Mes projections selon la moyenne des sondages ci-haut:
45 OLP 42 PC 20 NDP
Selon mon propre modèle, le NDP fait beaucoup plus mal au OLP qu'au PC; le OLP a un avantage inhérent dans le mode de scrutin actuel, mais plus le NPD devient fort, moins cet avantage est évident, au point tel que lorsque les intentions de votes deviennent 34% OLP, 34% PC et 26% NDP (tel dans le dernier likely-voters de Abacus), le PC gagne l'élection par un siège (43-42-22).