bio photo

A data-driven model for Australian elections.

Emma Chisit

Emma Chisit is a data-driven model for predicting and analysing Australian federal and state elections that I’m currently developing.

Why the name?

Emma Chisit is Australian slang for “how much is it?”. In an Australian accent, both expressions sound exactly the same. The phrase has its origins here. I’m using this name because the model aims to give a probabilistic, clearly quantified measure of “how much” the current government is leading by in the polls, or “how many” seats they would win in an election.

What does it aim to do?

The aim of the model is twofold:

  • outside election campaigns, provide a sophisticated poll aggregator that takes into account pollster accuracy, and provides a probabilistic “nowcast” for each seat in the legislature
  • within an election campaign, providing the above, together with a trend-based forecast

Ideally, I’d like to have the model be fully automated, including at the level of scraping the web to update itself to include new polls. I’d like to provide some analysis of its results when reported. It’s worth noting that, as far as I am aware, this would be a completely different approach to any that currently exist.

How is it different from existing models?

Emma is based on the assumption that Australian elections are won at polling places. In particular, if we can obtain an estimate of the swing on a polling-place by polling-place basis, we can aggregate everything to obtain estimates of individual seat swings, and finally a full national picture of the election. This is done by clustering seats and polling places together according to demographic variables which then provide local adjustments to a national swing obtained by a robust poll aggregator. Everything is done probabilistically.

What’s the current state of the model?

The model made some initial predictions for the 2016 election and I am currently in the process of making some adjustments to make predictions available for the 2019 federal election.

Where can I see the source code?

I’m keeping everything open source. It’s all available on the Github repo for this project.

When will it be finished?

My ultimate goal is to have the model fully ready for the next federal election in 2019. There are plenty of steps to complete in the meantime. I’ll keep the “what is the current state” section above updated as the model progresses.