Canada

Those were individual public opinion polls. Basically the raw data of a questionnaire which may have been targeting specific areas or demographics.

The 338 data is the best guess at who would win if the election were held today. It takes aggregated polls, ridings, demographics and historical data into account. So it is a much more sophisticated prediction.

However, it notably doesn't project into the future. That means the odds are not taking into account current trends, so Poilievre's drop from 100 to 75 is extremely significant. His odds will likely continue declining for some time. The question is where will it taper off and will there be a rebound before the actual election.

Here’s an overview of how these models work:

Polling companies release polls, 338 aggregates them and assigns rankings to polls based on previous election outcomes. The idea is pollers who were more accurate get higher weighting, and less accurate polls get lower weightings.

Once polls are aggregated they’re used to predict the riding level outcomes.

They simulate dozens/hundreds of potential outcomes based on the polling probabilities.

High polling might not result in a favourable split in ridings in FPTP.