Antony Green is well known as the ABC’s Election Data Specialist, and he generously shared his time and expertise in a wide ranging conversation about the statistics of elections, how stats are misused, and what he wishes everyone knew about data. Turns out there’s a vast amount of preparation that goes into those fascinating election night broadcasts.
“I don’t have a formal statistics qualification, but I have two years of economic statistics in the background. To be honest, for a lot of what I do I’m doing much more descriptive statistics. I don’t get much beyond variance, bias, and medians and means.”
“As the votes come in they have statistical bias and statistical variance. The bias in Australia is caused by the fact that smaller polling places report first, because they’re easy to count, and smaller polling places tend to be rural booths, they tend to be anti labor, so the statistical bias is that the Labor party starts low and rises through the night.”
“The average is only one way of measuring something about data. People talk about an average swing, and I think an average swing is nearly a useless figure to use for a by election, because the variance is enormous.”
If this leaves you wanting more, check out Antony’s keynote at linux conf: https://www.youtube.com/watch?v=h_0bRylRZg0
“If you ask someone to draw a hundred dots randomly on a chart, a human will always draw something with a pattern, because they can’t really do random.”
“People draw the wrong conclusions from data. Race statistics is a classic example, crime statistics. There’s lots of statistics that people draw a very simple conclusion from where the conclusion is more complex.”