What does this site do?

This site produces forecasts of by-election results given information about each seat and current polling. These forecasts are described as “par scores” because they give information on the kind of performance each party might expect, even if their performance will be better or worse than this expectation.

What are these numbers based on?

They are based on a Dirichlet regression model of by-election outcomes in the post-war period. The model is described in the paper by Chris Hanretty, “Forecasting multiparty by-elections using Dirichlet regression.” International Journal of Forecasting 37, no. 4 (2021): 1666-1676.

What goes into these predictions?

The result in the seat in the previous general election; whether or not there is a candidate from each party; the identity of the party in government; which party held the seat; turnout in the previous general election; whether the incumbent MP is running again; polling changes since the last election.

What doesn’t go into these predictions?

Whether or not the previous MP resigned as a result of scandal; the number of “other” candidates; any other factors which might affect the local popularity of a party that aren’t reflected in national polls.

What do you mean, “could be as low as”?

All models have uncertainty in them. When we say that the result could be as low as (say) 6%, we mean that there is a 2.5% chance that the true result will be lower than this number. Conversely, hen we say that the result could be as high as (say) 48%, we mean that there is a 2.5% chance that the true result will be higher than this number. Put differently, the probability that the result will fall in between these two figures is 95% (if the model is right).