After writing an effort post for the mid season I’m going to make this one quick, as it’s been a month since the grand final – which frankly was pretty boring for anyone but Richmond fans. We’ve tended to have pretty entertaining GFs for the last little while, so maybe we were due for a blow-out.
As far GRAFT was concerned, it was a pretty decent season for the prediction side of things, just under 70%, which after the early season shake-up that had everyone guessing, that wasn’t too bad. Finished nearer the top than the bottom of the Models Leader at https://squiggle.com.au/leaderboard/ so didn’t do too badly in a relative sense either.
Anyway, with that out of the way, I’ve spent the last few weeks finally getting onto the Historical section; so actually tackling it reminded me of why I kept putting it off before, as I do want to make it a reasonably useful resource, if only for myself.
In doing that I’m having to put about half-a-dozen different hats as far as working out how to link it all together, make it presentable and interesting, and as usual I’m doing all that in a pretty slapdash fashion. I suppose there could be some interest in comparing how teams across eras, with graphs and the like, so we’ll see. Anyway, it’s coming along.
I’ve already refreshed the A-League pages; for this season which started a couple of weeks ago I’ve added some basic projections. Very basic. Since I’m using an Elo system for that, putting probabilities out of that is pretty simple, however of course the base Elo system doesn’t present probabilities for draws up-front. More homework for me.
After looking at a bunch of charts of the data, while my hypothesis was that games with closely matched opponents should be more likely to yield draws than games where one side is expected to win, I didn’t get a curve fit that really made that stick out. So that’s something that will need to be worked out a bit more.
Since the start of the season was looming, I just decided that historically a quarter of all A-League matches ended in draws, I’d make that the fixed probability, no matter the difference in ratings, with the remain win-loss probability worked out of the Elo equation as used here.
So, yeah, with those very rough assumptions in place, the basic projections are set up, and should give a little more insight in how each A-League team is travelling this year. Western United were given a provisional rating of 200 below the average, with a couple of decent results early on, their number should calibrate to reality about halfway through the season I think.
Getting back to the AFL stuff, the Gamma model (which turns the GRAFT lines into a realistic indication of how games might play out) worked pretty well but I feel I should review the finding and do a bit of recalibration over the offseason to tighten that up a little more.
After covering the first AFLW season, I missed the next two due to “life”, so I will try and bring that back as well. I also kind of want to something with the BBL T20 cricket stuff, but interpreting the historical results into something I can mess around with, could take some time. But we’ll see. I might have to kick that can down the road for another year.