Utilizing Goal Bayesian Inference to Interpret Election Polls | by Ryan Burn | Oct, 2024

Tips on how to construct a polls-only goal Bayesian mannequin that goes from a state polling result in likelihood of successful the state

With the presidential election approaching, a query I, and I count on many others have, is does a candidate’s polling in a state interprets to their likelihood of successful the state.

On this weblog submit, I wish to discover the query utilizing goal Bayesian inference ([3]) and election outcomes from 2016 and 2020. The purpose will probably be to construct a easy polls-only mannequin that takes a candidate’s state polling lead and produces a posterior distribution for the likelihood of the candidate successful the state

Determine 1: An instance posterior distribution for predicted win likelihood utilizing FiveThirtyEight polling information from 2016 and 2020 ([1, 2]) and a snapshot of polling in Pennsylvania. The determine additionally reveals the 5-th, 50-th, and 95-th percentiles of the prediction posterior distribution. Determine by creator.

the place the posterior distribution measures our perception in how predictive polls are.

For the mannequin, I’ll use logistic regression with a single unknown weight variable, w:

Taking the 2020 and 2016 elections as observations and utilizing an appropriate prior, π, we will then produce a posterior distribution for the unknown weight

the place