Welcome to stats club


This is a blog that tries to make use of available electoral level data to generate insights and predictions about the 2013 Australian election. If you like stats and politics, you might find it of interest.

The analysis and commentary is by Kaighin McColl and me. We were friends at Melbourne University and used to run a lot together. To make the running a bit more fun, we’d talk a lot of shit about things we didn’t understand – running, women and statistics. The more we ran, the more we realized we weren’t going to get better at understanding running or women. But statistics was something we could learn! We got better and K is now using stats in his doctorate at MIT, and I used stats in my doctorate as well.

With the January announcement of the 2013 election date, we thought emulating Nate Silver’s methods would be interesting. His predictions did really well in 2008 and 2012 in the USA, but less well for the 2010 UK election compared to other forecasters. Given the similarities between the UK and Australian system, we wanted to see how our predictions would match up against traditional pollsters.

Polling data in Australia is typically only available at the national level, but elections in Australia are determined at the electoral level. Electoral polling data in Australia is not readily available. This leads to the problem where coverage of electorates is often mixed with an assumption of uniform national swings. This may mean variation at the electoral level is missed. To overcome this limitation,  we’re going to be relying on electoral betting market data.

We hope to use available data and basic statistical techniques to make better election predictions. We acknowledge there will be weaknesses and trade-offs made in our approach. We just hope to be sensible with our statistical reasoning and report the predictions as objectively as possible. Let us know if you see any mistakes.


3 thoughts on “Welcome to stats club

  1. Stats club is legit! Nice work. 🙂

  2. What if I am disillusioned by politics and still having nightmares about uni stats exams? Can I still read your blog?

  3. 1 you should acknowledge your source and inspiration. the idea is comes from an american who has done something very similar.

    2 the idea that betting odds are a better predictor of voting intentions than polls has some merit and obviously worked in the last usa election. but also some problems:
    (a) to what extent are people who bet on elections a representative sample of the population that votes?
    (b) the odds are an integration of the history of bets; they are not a reflection of current bets. think of it like this:
    in time 1, 80 is bet on libs and 20 on labor. then p[labor] is calcuated to be 0.2 [ignoring the margin for profit]
    in time 2, 50 is bet on libs and 50 on labour. now aggregate p [labor] is 70 / 200 = 0.35.
    but of course, time 2 voting intentions are actually p[labor] = 0.5. bookies need to make money over the entire set of bets not over the latest ones.
    (c) suppose that you are going to bet. do you bet on who you think is going to win or on who you are going to vote for? if you are smart, you bet on who you think is going to win.

Leave a Reply

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

WordPress.com Logo

You are commenting using your WordPress.com 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 )

Google+ photo

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

Connecting to %s