Election Night Analysis – Art or Science?

On 15 January I gave a presentation at the 2022 Linux conference on how Australian elections are counted and reported, and in particular, how I go about calling the winner as part of the ABC election coverage.

I titled the talk “Election Night Analysis – Art or Science?”, and as I explain in the talk, it is all statistics and mathematics with only a hint of art and hunch around the edges.

Here’s a video of the address which you can watch in the embedded version below or link through to the version on YouTube.

If you are concerned about the colour of my face during the presentation, it is a bit of transitory sunspot treatment that should be gone soon.

4 thoughts on “Election Night Analysis – Art or Science?”

  1. Enjoyed that very much. You are an engaging speaker.

    One thing I’ve noticed about US elections is that they don’t appear to get down to the detail as you do in looking at vote changes in individual precints to identify any trends that might affect the ultimate results. Their ‘decision desks’ may but the on screen presenters only mention the % of precints reporting without any contect of where they are in the state and how populous they are or how that reflects on turnout.

    COMMENT: The issue with the US is that they have no public concept of swing. Admittedly, swing was once unreliable in pre-2000 elections when voting was much more about the candidate. But since 2000, voting has become more partisan. No one in the US does it, but since 2000 you can easily put states on an electoral pendulum based on margin. To analyse the result, you just talk about the swing, how the votes in states have changed in the key states.

    For the last decade, all networks have called Virginia when the Republicans were in the lead because they had knowledge of the result origin by county and could apply the sort of swing analysis I described in the speech. Fox News did the same with Ohio in 2012, Dick Cheney Karl Rove famously disagreeing on camera and pointing to the Republicans leading, before Fox took their cameras behind the scenes to one of their analysts and explained the call based on where votes were from. So the networks have that level of analysis, but they do not use it to drive presentation graphics. It must be said, this sort of analysis was more difficult in 2020 because of the surge in postal and early voting, plus the huge increase in turnout.

    We are blessed in Australia by three factors. First, compulsory voting stabilises turnout. Second, full preferential voting turns all seats unto easier to analyse two-candidate races. Third, the quality and granularity of state from Australia’s electoral commissions is world leading.

    I’ll also take some credit for convincing the ABC to use projected 2PP percentages rather than raw percentages back in 1993. How the results were animated caused some initial confusion, but we’ve since ironed out the presentation.

  2. Thanks Antony. You mentioned the impact of the higher rates of early voting towards the end and how it had sometimes affected election night analysis, in that the swing for the early voting cohort can be different than the swing for the the booth results. Can I ask if the election night prediction models now allow for that? Is there enough consistency in the historical data to date to reliably account on election night for the impact of outstanding early votes (ie not so much postal or absentee votes) at an individual seat level? Thank you

    COMMENT: You account for it by altering the formulas so it takes longer to give away the seat. If you are going to get very large totals late in the count, and you cannot be sure it will have the same swing, you use a formula that lifts the downward curving line the track of 2CP must pass before the seat is given away.

  3. A brief correction to your comment Antony. It wasn’t Dick Cheney (on Fox in 2012), it was Karl Rove. Great analysis, as always.

Leave a Reply