The Forest from the Trees: A post for the stats nerds

Stringer goes forward after centre bounces, he hardly attends stoppages unless they are in the centre of the ground or inside forward 50.

3 Likes

Looks good. System seems to categorise players contributions in each role pretty well.

What are the meanings of A_L_UC etc

Anyway, it seems like youā€™re tring to figure how to produce a WAR type value for AFL players. Have you worked out the beginnings of an algorithm to work out how to tease out the contribution of all of the roles to a victory?

Thereā€™s so many moving parts that itā€™s so hard to even make head or tail of it. This at least allows you to begin to stack up what an entire AFL list looks like from the perspective of contribution and quality of each role, and then you can compare that to team results (either win/loss or more granular (like quality of shots on goal given up or something as some sort of granular defensive measure)) and see how teams look compared to the most successful teams (and perhaps allow for a comparison of gamestyle and quality etc etc

My head is reeling with the possibilities of a meaningful measure of contribution and quality of players across different roles, when compared to the sort of stuff we see in the media (Essendon has no players ranked ā€˜eliteā€™ for their position). This stuff seems to allow for a more meaningful quantification across the wide variety of sub-roles that players will play, even across individual games. I look forward to seeing what you do with it all.

On the technicalities of the SOM, my memory is a bit hazy on how it work. So you feed stats into it, and then it groups players based on their stats and then you name the groups?

P l
B0

Well known model aggregator ā€œSquiggleā€ has done its starting points and final table predictions very early.

Having played a final, we get a tougher draw. So they reckon we will win the same number of games as last year. We only need to win 2 more games to get the double chance.

Here it is.

image

Iā€™m pretty lid off ATM. I think if our midfield stays together and builds on the progress of last year and if Ridley has a big year, we can go past geelong, tigers, lions and port

4 Likes

Agree that our 3 names are elite, wonā€™t go into other clubs, let the champion data ratings bashing begin.

Adam Saad is elite at chasing the dollars.

Jeremy Finlayson is rated elite yet GWS traded him for a 3rd round pick. Meanwhile Zac Merrett became a dual all-Aussie and isnā€™t elite. :thinking: I could go on.

Yep CDā€™s rating system make complete sense.

I canā€™t understand how Grundy is rated elite but Gawn is not given their respective form since Grundy signed his big contract.

Also, how come Hobbs is not elite?

3 Likes

I am sorry, this was a great reply and I went totally AWOL on blitz after the season ended.

I definitely needed more info.

A - A Grade
L - Large
UC - Uncontested

I have not worked out the next step which is obviously what you say. I think a lower hanging fruit might be gap analysis and depth analysis, and then work towards seeing what you need to win.

A SOM basically is unsupervised analysis, but the outputs are mapped to the node map (Which you can cluster), but you can essentially easily visualise similarity. The nodes are obviously not representative of the actual ā€œdistanceā€ between the nodes, or ruckmen and Jed Bews would be outliers. It has some limitations, but I like the visualisation to be honest.

Adam Saad and Zac Williams elite for Farkcarlton.

Iā€™ll make no comment.

2 Likes

For the second season in a row, CD algorithms take in all the stats and conclude Nick Naitanui is the best player in the comp.

OK, so is it the GIGO principle at fault here. ( Garbage In Garbage Out )
Or is the algorithm garbage ?

Factor this in: Naitanui is pronounced elite, nay, the best player in the comp but cannot play for more than half a game, cannot mark and cannot run, while Gawn is pronounced just ā€œabove averageā€ so, are they normalising output against time on ground?

1 Like

I think this is the worst CD list yet, doesnā€™t pass the eye test at all

Gawn is clearly the best ruckman in the comp, just look at his preliminary final for evidence

Saad, Weitering and Williams were key members of a backline that was the 2nd worst team from a points against perspective in the comp.

Jamie Elliot?
Jack Scrimshaw and Dylan Moore?
Jeremy Finlayson?

No Zach Merrett
No Fritsch?
No Langdon / Salem?

Youā€™d think by spitting this ā– ā– ā– ā–  out theyā€™d be going back to look at the algorithm

2 Likes

I think you may have found something with the normalising. Over the past 2 seasons Gawn has spent ~20% more time on the ground than Natanui. Looking at the relative stats, Gawn comes out ahead in most categories. However, if you normalise Natanuiā€™s numbers to account for the 20%ā€¦

The one category that Natanui is decently ahead is clearances. Perhaps CD place a lot of value here.

Iā€™d love to know if CD ever change the algorithm. Every year there is some bizarre outcomes.

CD shouldnā€™t be normalise for time on ground unless a player is injured or subbed in. I would have thought a key component to being an elite player is the ability to go for longer than your peers. That is where the Nic Nat elite rating falls down big time.

Agree with this. Fundamentally, a player who gets 30 touches in a game is more valuable than one that gets 16 touches in only half a game.

1 Like

We do not see the hiot outs to advantage, stat so easily, but I think Naitanui would only be slightly ahead of Gawn on this stat.

There is another issue. Even in his 50% game time, Naitanui has trouble covering the ground. This results in low numbers of marks and contested marks, when he ought to be at either end competing for incoming high balls the way Gawn does. Sure Naitanui wins 2.5 more clearances per game than Gawn

So, you know, the Brownlow is a midfielders medal. In 2021 Gawn 16 brownlow Votes against Naitanui 5.

Its pretty clear cut. Gawn is all round a better player than Naitanui, based on the same algorithm.

1 Like

AFL player ratings seem to have been removed from AFL website this year.

However the twitter account is still showing summaries after each match.

Anyone know if overall player ratings are acccessible anywhere?

aflplayerratings come out on tweets, but they still have them under ā€œstatsā€ on the afl site, i think.

I can only see fantasy points on AFL site, not player ratings

@The_New_Coleman this is from phone.
Stats, Then thereā€™s a stats box that appears at bottom you can click on to change category.