My methodology is described in a blog post here.
Based on 801 polls conducted between 2015-11-04 and 2021-09-19, I performed a best-fit solve of 38200 linear equations in 33573 unknowns to determine that, as of 2021-09-19:
Pollster house effects and other statistics arising out of this analysis for the 5 most frequent pollsters are as follows:
Pollster | LPC | CPC | NDP | BLQ | GPC | PPC | Weight | # Polls | Max effective sample |
---|---|---|---|---|---|---|---|---|---|
Abacus Data | +0.29% | -0.62% | +0.58% | -0.17% | +0.14% | +0.05% | 8.98% | 90 | 5607 |
EKOS | -0.59% | -0.25% | -2.11% | -0.79% | +1.19% | +1.52% | 6.52% | 73 | 1414 |
Leger | +0.30% | -0.60% | +0.81% | +0.20% | -0.04% | -0.64% | 6.62% | 79 | 1454 |
Mainstreet Research | +1.67% | +0.94% | -2.84% | -0.61% | -0.51% | +0.71% | 5.94% | 97 | 1136 |
Nanos Research | +0.82% | -0.49% | +0.23% | -0.38% | +0.30% | -0.49% | 6.38% | 226 | 1657 |
Full data in CSV format: polls, daily polling averages, pollster statistics.