Canadian polling analysis from cdnpo.li, a project by Erik Partridge.
Last updated Wednesday, April 10th, 2019 at 20:06EST. Français.
Model v0.1a-032619. Totals may not add to 100%
NEW: See the Albertan analysis.
The model is developed based on historical data from 2006 to the present day. At the moment, no future forecasting is made. If you wish to use this data as a prediction, please recognize that this accepts the freshness hypothesis, which would assume the latest polls are the best predictors of the future. This may not be the best hypothesis.
The support percentages are based on the exponential weighted moving average of recent polling history. This weights the most recent polls substantially, and the older the poll, the exponentially weaker its impact on the forecast.
The weight for each individual poll if it is the most recent poll is defined as the base 6 logarithm of the sample size, divided by thirty. This effectively means that a 5,000 person poll will have a 33% greater weight than a 500 person poll. This favors fresh polls vs large polls, but also has relatively low weights—so the history of the most recent ten polls or so will be significant.
One ground rule we set is that polls which only contain partially new data (i.e. Nanos Research's) only have the fresh data included. A poll with a sample size of 1,000 with only 333 new responses will be weighted as 333 responses.
This is a passion project—I am a data scientist, but I do not specialize in polling. I provide no warranties about this data, and am thrilled to hear your feedback. All the data used in creating these models can be found on GitHub.