Modelling Practices of Election Forecasts

Although election models generally do not have a large impact upon people’s lives, it is a case where the nature and practice of modelling is very much in the public eye, and (almost uniquely) they will be objectively judged on their accuracy after the real results are known. Modelling practice here tends to mirror (or slightly precede) practice in serious policy modelling, which is why trends here are interesting.

The trend towards more explanation of models, what they do and (sometimes) their limitations continues. This year:
* the Economist model’s code is freely available to inspect, but their model explanation/documentation leaves something to be desired
* 538 have a good explanation of their code and answer questions by the public on their model in their podcasts, but do not reveal their code because their financial interest.
* YouGov tell us almost nothing about their statistical model.
* Princeton do not provide their code, but do provide some explanation and a link to an academic paper on the approach (albeit more than a decade old now).

List of US Presidential election models at

Explanation of model:
Model Code not available.

Explanation of model:
Model code:

Explanation of model:
(more links from
Code not provided but paper about the approach available:

No explanation of model other than “Our election model is updated weekly and based upon 76,677 interviews, updated with 4,802 new interviews in the past 24 hours” and that it is a “Multilevel Regression and Post-stratification model“.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s