First of all, when calculating accuracy for polls you have to check if it falls within the margin of error. For example: Forum’s final poll on Oct 25th (53-43) was in the +-4 MoE and the final result of 54-41. So what’s the margin of error and why does it matter, then? In 95% of cases if you ran the poll again, it would produce a result within 4 points each way. The MoE shrinks with increased sample size. Simply put, you can say with 95% confidence that the true result is contained in that space. If the actual result fell outside of the MoE at 95% confidence level it would be ~2 standard deviations away from the mean. Build Canada’s October 16th poll (Bradford +4.7) had an MoE of +-3.1. The polling miss of 17.1 points (for the margin of victory) meant that the actual result was nearly 5.62 standard deviations away from the poll’s mean. If this was just because of pure random sampling luck, this has a 1 in 51 million chance of occurring. The poll did not reflect the voting public that showed up on October 26th. To attribute this miss solely to voter preferences swinging double digits in a week is an excuse for poor methodology.