# Seat betting markets, 2013 federal election

*(Go to odds by date, or odds by seat.)*

I started collecting individual seat odds by hand in January 2013, very soon after Sportsbet opened their markets. From 20 April I ran a script daily to get the odds from both Sportsbet and Centrebet. Luxbet and the TAB were added later, though only the Luxbet prices are shown on this webpage, since the TAB's listed prices were often not those of a market (one side was often frozen out). The full dataset with code and (most) images can be downloaded here.

I didn't pay too much attention to detail – in particular I only grabbed the odds for the two favourites in each seat, assigning one of them to "LNP" and one to "ALP". In Indi, Cathy McGowan becomes "ALP"; in Denison, Andrew Wilkie is "LNP".

I have graphs of the implied probability of an "ALP" win, defined as (1/ALP) / (1/ALP + 1/LNP). The distribution of probabilities across seats is converted into an implied swing by fitting an error function to a plot of implied probability against margin (see animated examples below, or stills in the pages by date), and seeing how far to the right of zero-margin the fitted curve passes a probability of 50%.

Getting into more theoretically dubious territory, I also calculate an implied swing for each seat, defined as its state's implied swing plus how far to the right of the state's fitted curve the seat is. This gives interesting results when the seat is at least a little in doubt, but I would totally ignore the implied swings for seats where the implied probability of victory is below 10% or above 90%.

The curve only fits the width and horizontal displacement of the error function; the top and bottom levels of it are fixed. The latter work pretty well once the Centrebet odds are included (the early Sportsbet-only data had a smaller bookies' over-round, so "almost certain seats" had implied probabilities closer to 0 or 1 than they subsequently became). The big non-classic-division outliers are ignored when fitting the curves.