The formula that reveals this season's AFL ladder

Prior to the season I made a little algorithm to try and predict the ladder.

I didn’t manage to tap out an article in time and many other people put their own predictions up, which led me to think such an article might be redundant.

However, I like it as a tool.

The results over a long data sample are fairly impressive.

Having said that, based purely on the ladder it generated, the results from Round 1 forced me to look at it again.

Just like many of us, Round 1 did not play out in line with anticipated form.

Now, of course, it’s only one game, but Round 1 got me thinking. How much stock should I place in one game?

What I’ve tried to add to the ladder algorithm is a form adjustment supplement to try and show how much my expectations should change on the basis of one game.

The predictor
First to explain the method. In short, I’ve performed a multi-linear regression to produce an algorithm that estimates how many games a team should win based on net points from the previous season, change in net points from the season before last to the previous season, and change in total wins from the season before last to the previous season.

To give an example what I’m looking at, if we take Essendon as a random example.

Points for minus points against in 2018: 94. Points for minus points against in 2018 less points for minus points against in 2017: -37. And 2018 wins less 2017 wins: 0.

(Photo by Mark Metcalfe/AFL Photos/Getty Images)

So on the basis of this model, my ladder came out like this.

It’s worth mentioning that this works purely on the correlation between the three aforementioned variables and historical win production.

I’m not taking into consideration number of injuries, who is injured, players added or deleted from the list or any other number of variables that might affect performance. It doesn’t worry about who the team has to play or strength of schedule.

In terms of performance, I applied this formula to 20 years of ladders and team wins, and the results were as follows.

Only once was my model out by more than four games in 20 years of ladders (0.35 per cent).

Only 8.83 per cent of the time it was out by between 2.5 and 3.9 games, while 44.17 per cent of the time it was out by between 1 and 2.4 games, and 46.64 per cent of the time it was out by 0 to 0.9 of a game.

In short, it was rarely very wrong and roughly 91 per cent of the time it was within 2.4 games.

My initial reaction was that this was pretty good but that’s open to speculation.

If you look at this predicted ladder, 2.4 games would make a pretty big difference in terms of ladder position for lots of teams. Positions four to 12 are separated only by 2.59 games as it is.

If we take the predicted 12th team and the predicted fourth team and apply a 2.4 game error in favour of the lower placed team, we would see a huge shift in ladder position.

Indeed, holding all things equal, such a change would take West Coast down to 11.9 wins and 11th on the ladder and boost Adelaide up to 14.18 wins, which would be good for fifth.

(Photo by Jono Searle/AFL Photos/Getty Images)

The fact that there is quite a lot of parity from fourth to 12th makes sense, and to that extent, I suppose the model isn’t ideal for predicting ladder position so much as it for predicting total wins.

Also, because teams can’t have fractions of wins, it would be necessary to round to whole figures.

Easy enough to round in line with standard ‘rounding rules’ but you could use a little of your own discretion if you had a feeling a team was better than predicted and round up from a fraction smaller than 0.5 or vice versa if you were low on that team.

Round 1 debacle
So, like many people, my tipping wasn’t great in Round 1.

I tipped in line with my expected ladder positions, which is not necessarily the way you should go about it.

It did lead me to pinch Hawthorn and Geelong against the bookies but also meant I lost big on Melbourne, West Coast, North and Sydney.

Now then, as sports media has a tendency to do where unexpected results are concerned, the reactions to these results have been pronounced.

Despite qualifying every statement with the “it’s only one game” caveat, everyone seems to have formed definite positions on how right or wrong they were only two weeks ago.

What I have tried to do, then, is apply a one-game adjustment based on the first week’s result, and Round 2 for Collingwood and Richmond.

Effectively, I have applied the difference in net points from this season and a new shift in net points based on this result. The new win totals and change in expected wins is as follows.

Again, because this system doesn’t take into account the quality of opponent, I can understand some subjective disagreement.

It’s entirely plausible that one could view a one-point victory by St Kilda over Gold Coast as detrimental to form not beneficial.

More or less, the shags in win expectation are simply in line with the size of the teams win.

Nevertheless it’s interesting to try and quantify the difference a big win or a little win makes in this regard and also to think about how that one solitary victory alters our long term projection of that team’s success or failure for the season.

If we take Fremantle, for example, my model projected about 6.5 wins. If we take my standard error this would give them a win range between about 4 and 9. Not enough to make the finals even at best and potentially a wooden spoon in the worst case scenario.

Their win on the weekend seemed to elicit responses from the general public that maybe they were wrong about Fremantle altogether.

I tend to be more conservative in this respect. That is, I tend to stick fat with the prediction, especially if I have quantitative evidence until I’m proven wrong.

Of course, this means at times I stick stubbornly to my guns while all the smart people have adjusted accordingly, but it also means I’m occasionally vindicated for staying strong while everyone reacts impulsively.

This article is not intended as an attack on Fremantle per se, they just happen to have had an unexpected result and so they are worth mentioning.

Does one win against a team many expect to be average at best mean everyone was really wrong about them?

We have all sorts of ways of justifying the position. They played with more flair and attack, or they scored more than usual.

The flip side of this is to imagine, against other more highly rated opposition, away from home no less, would we expect Darcy Tucker to have 24 touches and a goal again?

What about Cam McCarthy? Are five-goal hauls the new standard? Or is it more likely that come September we are looking back at that game as one of his few good games for the season?

And so it is with other teams. Essendon are one who seem to have evaporated any small amount of confidence anyone had in them.

My model projected Essendon as basically a 12-win team. Their hiding at the weekend makes that projection look a little closer to 11. Either way this is fairly in line with the consensus opinion on that team.

So what’s my point here? I was motivated to write something for one pertinent reason, aside from the fact that I like my little model.

AFL media is some of the most reactionary media you can consume.

One win and a team is unbeatable, one loss and the team is going to fold permanently.

The reality is, especially after the first round of the year, that these games by themselves probably don’t mean all that much. Not as it relates to the likely long run outcome anyway.

Form can often be a somewhat intangible concept and we are hardwired to place stock in the thing we saw most recently.

To that end, I only urge a little perspective and patience.

Fremantle and Brisbane aren’t playing off for the flag just yet and Essendon will probably hit a target or two more over the next six months than they did last week.