Scoring Anomalies – Round 15

In last week’s edition, I flagged a desire to make this column less speculative and more scientific. To that end, I have been working on a program that takes what is publicly available – both in terms of raw data, and what we know about the scoring system – to try and simulate what a player’s ‘correct’ SuperCoach score ought to be. For the time being, I am referring to this system as SyntheticCoach, although at some point, I may want to start calling something a bit more catchy… I rather like ‘The Machine™’. Or perhaps ‘RoboCoach’?

As I discussed last week, there are a couple of stats that go into the publicly-available section of the SuperCoach formula to which we do not have access: gathers from hitouts (2 points), and handball receives (1.5 points). The stats we do have access to seem to get close enough in most cases to the official scores that I don’t think these two stats are hugely important factors, but it’s still a blindspot to be aware of. I had initially considered using clearances as a proxy for ‘gathers from hitouts’, but it dawned on me that not all clearances come directly from a hitout, and not all gathers from hitouts end up being cleared; thus, it would be an unreliable proxy stat. However, if a player racks up a high number of clearances, chances are that a good chunk will have been gathered from a hitout, so where there are sizeable discrepancies between the SyntheticCoach score and the official SuperCoach score, the clearance count may be a good place to look.

Likely the biggest cause of these discrepancies, however, is scaling. As you probably know, the sum of all scores in each game is scaled to 3300 points (give or take a point or two). You will be pleased to know that SyntheticCoach takes this into account, scaling games to 3300 points, just like Champion Data. Unlike Champion Data, however, my system is doing this linearly. The exact way that Champion Data scales scores is kept secret, although we know that they reward actions taken when the game is on the line. I also strongly suspect that this is where score involvements and metres gained – both of which are fairly well correlated with SuperCoach scores – come into it: not in the base scoring formula, but in the scaling. But that’s enough speculation for now.

In future columns, I intend to focus on fewer scores, but probe deeper into how and why they might have come about. But as I am pressed for time this week, I will write briefly about just a couple, and then give a list of potentially anomalous scores that SyntheticCoach picked up.

Firstly, Dane wanted to know on Monday whether Carlton’s Marc Murphy (125 points) had been short-changed on the weekend. SyntheticCoach gave him 120 points (123 unscaled), so that score is probably about right, although he did kick the winning goal in the dying stages of the match, so you could certainly argue that he should have got more favourable scaling.

Moving on, Steven May was awarded 94 points on the weekend for an effort that SyntheticCoach thought was worth just 77 (81 pre-scaling). He did, however, have 10 spoils, which I am fairly certain is factored into the base scoring system, but the official SuperCoach terms and conditions don’t tell us about it. If anybody knows (and has sources to back it up) how much spoils are worth, I would love to work them into the base SyntheticCoach formula. Ditto with any other stat, too, so long as you have actual proof (i.e. confirmation from Champion Data, or somebody who works there, that they are worth x number of points).

Finally, here are some other scores that the system picked out as being a bit off (anyone who captained Grundy is hereby entitled to feel slightly ripped-off; ditto with anyone who captained Lycett 😉 – or Fyfe). If you would like a full breakdown of any of these scores, let me know in the comments.

Player name, Linear-scaled SyntheticCoach score, (Raw score), Actual score

Tim Kelly, 143, (153), 128

Joel Selwood, 124, (133), 105

Hugh Greenwood, 105, (113), 82

Jordan Dawson, 101, (104), 120

Isaac Heeney, 138, (142), 117

Jarrod Witts, 137, (141), 110

Anthony Miles, 136, (140), 118

Scott Lycett, 182, (185), 165

Jack Macrae, 166, (169), 142

Josh Dunkley, 93, (95), 111

Brodie Grundy, 167, (170), 133

Chris Mayne, 107, (109), 87

Todd Goldstein, 148, (151), 131 (as an aside, he also only scored 64 Dreamteam points – I know they’re very different systems, but the difference between the two isn’t normally quite that great. Wow!)

Seb Ross, 81, (88), 94

Hunter Clark, 124, (134), 105

Nat Fyfe, 163, (166), 147

Ed Curnow, 119, (121), 107

Lachie Neale, 156, (163), 137

Stefan Martin, 130, (136), 116

Were there any scores that seemed off to you on the weekend? Let us know in the comments, and I’ll let you know what SyntheticCoach had to say about it!

---

Leave a comment / Scroll to bottom