Luck, Skill, and Circumstance

Written by Father Dougal on March 16 2019

Rather obviously, one of the key parts of SuperCoach is figuring out which players to take. It matters both before and during the season, and Lord knows I need to get better at decision-making during the season. So, thinking happened about player performance in a general sense. My thinking was aided by a lot of very smart people who have already looked at that problem for MLB baseball, and then wrote about it. And, I can read! I’m not gonna go over that background because baseball is boring; we’re here for AFL SuperCoach!

There are three factors that go into a player’s SuperCoach performance, meaning scores. Luck, skill & circumstance. (I bet you saw that coming.)  

Skill is the main factor; the base of performance that luck and circumstance modify. Every player has what I call a “True Level of Ability” which after this I am calling TLA.  That tells you what they would score if there was no luck and their circumstances never changed. When trying to figure out who is and is not going to be good, we are mostly trying to figure out their TLA.

Circumstance is the most obvious of the factors, at least if we have access to the information that exists. What position does someone play, are they healthy, and how old are they are the biggest factors. If their Coach likes them is sort of part of position, since bench and VFL are also positions from the circumstance point of view. If someone is playing in the VFL, we can multiply their TLA by zero to get their score quite reliably!

Age is a huge circumstance factor, and it really affects TLA. A good argument could be made that it is its own fourth factor, which is fine and not worth our time as long as we account for it. Athletes in professional team sports tend to follow the same pattern age wise as far as performance goes.

At birth, they are well below replacement level and have zero value. Every year they get better, until they eventually become good enough to get onto the pitch. Some players are good enough at 18, others take longer. They gain experience for the rest of their careers, which makes them better. They also start aging, which when they are young makes them better, but around their mid-to-late twenties makes them worse.  To some degree added experience can compensate for a decline due to age, but eventually old bodies are no longer capable of competing with young bodies, and once again value drops to zero, hopefully on retirement and not before!

So players first start playing, get better, plateau, then decline. And we know around when breakouts happen and when declines start.

But, growth does not start from the same place, meaning the same TLA. I’m gonna make some numbers up for fake players here as example of how that matters and plays out. Let’s take two players, Fred and Charlie. Both of them average 75,90, and 100 in their first three seasons. Just knowing that, they look about the same. But, when we add their ages:


Age Fred Charlie
18 75
19 90
20 100
21 110
22 75 120
23 90 130
24 100 135
25 110 135
26 110 135
27 110 135
28 105 130
29 100 125
30 95 120
31 80 115
32 110
33 105
34 100
35 95

Fred is a nice decent player. Charlie should win a Brownlow or two.

The sooner a player plays and the better he does, the higher the arc of his career will be at his plateau. Coniglio at 25 is not likely to get too much better than last season’s 108 for very long because he is too old to do that. (Excluding luck and such.) Clayton Oliver, on the other hand is 4 years younger, and has several more years of growth available. He could get a lot better, adding another 15-25 points to his 115 from last season. Players often break out at 23-25, which for Oliver is 2020-2022!

Note that the rates of growth vary a lot, with some players, like Oliver, shooting up sooner and others having slower growth, like Dangerfield. There is no one size fits all formula, but there is an arc, if outside forces don’t come into play, which they always do.

Position matters. Part of growth for mids is becoming a midfielder! Team and coach matter. Tom Mitchell looks to have been delayed by playing for the Swans and broke out when he went to the Hawks. Some guys just get hurt. Who knows what a healthy Tom Liberatore might have done? Zorko had a weird year when he started being tagged. Merritt got concussed and had an off season. And don’t get me started on Bevo and his “death to SuperCoach midfielders” style. It’s like he wants to win matches rather than have players be valuable in Supercoach!


Luck is the least visible, and most ignored factor. It plays a bigger part in performance then we usually give it. Humans are natural pattern finders, and we are soooo good at pattern finding that we find patterns where none exist, and then attribute the effects of random chance to things other than random chance. If you don’t believe me, you can look it up. There are books about this, and as I mentioned before, I can read. You can too – I know ‘cause you are reading this!

I don’t want to spend too long on statistics stuff, but I have to a bit for some of the rest of this to make sense. So, sample size. The more times you try something, the less effect luck has on the outcome. Say you flip a coin once and it comes up heads. Wow, you have a coin that comes up heads Every Time! After all every time we flipped that coin it came up heads, right? Right, it did, for true. But, just maybe, a single trial is not enough to go on. So we flip the coin again and it comes up heads again! See, more data and we now know the coin comes up heads every time. Clearly that coin is a lock! Right into the team. It had two trials, just like in the JLT, where after two trials we change all of our plans, because in no way could any of those players just have gotten lucky two practice matches in a row….

Yeah, so, it takes lots and lots of trials for luck to even out.  If you look at scoring quarter by quarter, there are huge differences. If you look at scoring game by game, there are usually much less huge differences. The players with the most game to game variation, are, surprise, key forwards, who score big when they kick many goals, and score diddly-squat when they do not kick goals. Kicking goals is one of the most luck dependent and high scoring impact things in SuperCoach. Thus, big scoring fluctuations. If you look at score season by season, it evens out pretty decently. There is still some luck, but not so much we can’t use a previous years scores and a players age to make a decent guesstimate of what they will do next season.

How much of season to season variation is luck? Good and big question. I looked at some numbers and then called one of my best mates, who happens to be seriously good at math, and studies baseball statistics as a hobby.  To sum up, we think that luck easily accounts for about a 5% variation from a players TLA each season, with up to 10% being not unheard of. That is not a lot in sports terms, but is enough to affect how we look at player. It also doesn’t mean a player can’t have an over 5% swing on any given season. After all, given enough tries unlikely things become likely to happen one or twice. Flip enough coins, and you will eventually get ten heads in a row even with a fair coin.

Let’s look at a real player.  Mitch Duncan was injured in 2015 but I don’t think that throws things off. Now the below TLA numbers are my estimates. That means I Made Them Up. This is not my normal “I am trying to back up stuff with numbers” thing, because we cannot know a player’s TLA. We can just estimate from their real averages and what we know happened during their careers.


Year Age Games Average TLA (?) % of TLA
2010 18 8 57.9 60 0.97
2011 19 18 74.8 70 1.07
2012 20 21 80.2 80 1.00
2013 21 22 80.5 90 0.89
2014 22 22 100 95 1.05
2015 23 11 91.4 95 0.96
2016 24 22 92.3 100 0.92
2017 25 21 109.8 105 1.05
2018 26 20 106.2 105 1.01

So assuming my TLA guesses are at least decent, he wandered around both sides of it, which would be normal. He was off by 11% in 2013, and his average only going up 0.3 from 20 to 21 is odd, but the rise in his real skill, which probably did go up, would have been masked by bad luck. Interestingly, the average of his % of TLA is 0.99%

Let’s try someone else just to see. More relevant this season is Dusty.

Year Age Games Average TLA (?) % of TLA
2010 18 21 77.9 80 0.97
2011 19 22 98.9 90 1.10
2012 20 20 88.5 95 0.93
2013 21 22 101.8 100 1.02
2014 22 21 99.5 100 1.00
2015 23 22 105.5 105 1.00
2016 24 22 108.1 105 1.03
2017 25 22 119.3 110 1.08
2018 26 21 103.9 110 0.94

Average % of TLA = 1.01

Again I guess at a normallish TLA progression for him. More variance when he was young, which I am going to guess is normal, and not too much after he turned 20. His big year is just 8% over an 110 estimate and last year he was 6% low. Huge swing in scores, and luck, but both years not too far off a reasonable baseline.  Also, we know he was hurt and sort of lazy last season, which implies his luck factor for 2018 was lower than 6%.

One more. Zach Merrett.

Year Age Games Average TLA (?) % of TLA
2014 18 19 63.7 65 0.98
2015 19 17 88.5 90 0.98
2016 20 22 111.5 110 1.01
2017 21 21 109.2 110 0.99
2018 22 22 100.4 110 0.91

His 2018 looks like a huge luck factor! If you put his concussion down as luck, yes. If you want to say that his TLA was lowered because of his concussion, well, six of one, half dozen of the other. We know his variance was due to concussion. After the Bye, when he was presumably recovered his average was 113.1. That would be either a bit high for a 100 TLA and a bit low for a 115 TLA. Either way very close. That implies is his TLA for 2019 will be about 115….

Well, that was long, but there was a lot to cover. Now you all know how I go about evaluating players. I think, if I managed to explain. Um, let me try to sum that up.

You now know about how player’s careers work in general. Estimate TLA from past performance, age, injury, and anything other info you have.  You should probably assume that a player has not been lucky or been unlucky every year. Using that, you can project into next season. Big changes occur young, when older, changes are smaller –  until the old age cliff.

Oh, my guesses for those three in 2019:

Duncan: 108, Martin: 110, and Merrett: 115

If those are right then their real averages, assuming no unexpected circumstances, should be not much more than 5% away from what I estimated.  I think the most likely to be off is Merrett, because I think his game could be affected by the 666 rule. (Which is a circumstance change.)

There is another takeaway from this. Even if you correctly ID the players with the highest TLAs, they could easily underperform because of luck. Nothing you can do about that. Luck happens.

Huh, yet another takeaway. The highest scoring ruck every year almost certainly was lucky. Odds are they will not be so lucky again.  But, say Grundy was a big huge 10% lucky last season. Still means his TLA was 118.6. And if he was just 5% lucky it was 124.3. If he is 5% unlucky from those numbers, still not shabby!

Stop me before I think again! Gonna stop now. After the poll

So, about those player evals

View Results

Loading ... Loading ...

Thanks for reading!



Leave a comment / Scroll to bottom

10 thoughts on “Luck, Skill, and Circumstance”

  1. Excellent FD.

    I did something very similar regarding all premos. What I have found is that based on # of seasons played, the degree (steepness of the slope) of the climb and the length of the plateau (time at the top) all vary significantly by position. As does the rate of drop off!
    Of course I could never explain it as clearly and succinctly as you do tho. Looking forward to reading more … 🙂


  2. Awesome. The SCT community is lucky to have you FD.

    Hmm… does this mean Sloane’s 97 ave last year at age 28 (now 29) is the new normal?


    1. No. It means it was a year of ‘luck’ for Sloane with his injured foot. BUT it does mean he is probably on the way down …


  3. Pythagoras, Archimedes, Newton, Galileo, Copernicus et al, and my year 11 maths teacher could not have put it better.

    I failed year 11 maths, but have a business degree….obviously i got lucky !!


    1. Parle-tu français, mon ami?
      Si, FD est encroyable, mais ce n’est pas trop difficile à comprendre, n’est-ce-pas?!?
      Encore, bonne chance!


  4. Great article… was hoping there was a chapter two…
    I’ve just run the rule over 3 players in my current team and 2 that aren’t.. I have now decided to dump Selwood and work out how to get Macrae in…
    Very happy now with my backline as it stands .. and one of those is only in 6% of teams … was considering Hurn.. but now he is not in consideration … glad I don’t have BSmith.. never considered him, and based on this article will not be..


    1. Sounds good mate.

      I’m considering Hurn. The key factors being unquestioning durability and his correlation to team output. I’ve no doubt the Eagles will make the four, and he will also soar. I also have no doubt he is on his way down the far side of the bell-curve, but DEFs plateaux are longer and their down curves are longer. He will drop $60k if he averages the same as last year. BUT if he plays 23 games and is good for D6, you’ve started better than most and saved yourself $70k (net cost of a trade).

      On the other point, I never look at % ownership. It has no bearing, yet if you look at it, it may have some irrelevant influence on your ultimate selection. Why? PODs are good. PTS are better!

      Keen to hear who your POD is. And your rationale, given that POD counts for nothing 😉

      PODs: later in the year, maybe. But not now.


  5. “Lord knows I need to get better at decision-making during the season.”

    You know, I’ve always had rather the opposite problem: I’m great at trading in-season, but I always manage to balls-up my initial selections. Maybe we should team up – you pick the initial squad, I’ll take over once the season starts. We could call it The Dougalmander!

    In all seriousness though, this post was terrific. Your idea about player’s career trajectories reminds me a bit of a mathematical model I toyed with a bit over the summer. It was designed to model how players’ output changes with age and experience, and, in the aggregate, predict the on-field fortunes of teams over the medium to long-term, based on list demographics. Although it wasn’t originally designed for it, it could probably have fantasy applications, too, such as predicting when players are likely to break out or are about to go over the old-age cliff, etc. I never really took the model beyond a basic level (it was mostly just something that I put together one afternoon when I was bored), but your article has got me thinking that I should keep working on it. I think I’m going to have to go and dig up my original notes…

    The part about luck was also interesting. I’ve often suspected that the ‘Cox Curse’ that always plagues the previous years top rucks is plain old regression to the mean. I’m also now even more convinced about starting Merrett than I was before.



Leave a Reply

Your email address will not be published. Required fields are marked *