AFL Prediction & Analysis

Some Useful Things I Learned About Football

I got into football stats because I wanted to understand why my team was so bad for so long and was there any hope for the future or should I just die. That was about 20 years ago. Along the way, I learned that plenty of common football wisdom is probably wrong. Stats people can show it’s wrong, but everyone just keeps repeating it anyway.

So to save you 20 years, if it’s not already too late, these are some of the most useful things I’ve picked up.

Scoring Shots Matter

It’s better to be a team that kicks 6.18 (54) than 9.0 (54). People will criticize the former and say bad kicking is bad football, but the truth is that there’s a lot of luck – or at least unpredictability – in goalkicking, no matter who you are. It’s harder to fluke 24 passages of play that move the ball into a scoring position than 9 accurate kicks, so you want to be the team that’s generating lots of chances, even if it’s not taking them. Because, all things being equal, that team will perform better in the future.

AAP Image/Julian Smith

Momentum isn’t Really a Thing

Story is a huge part of why we enjoy football, so there’s lots of talk of momentum: of power shifts, and causality, where this thing caused that thing to happen. But momentum is a weak effect at best, and might not exist at all. When a team goes on a run and scores five or six goals in a row, that’s not a dam breaking, where if only the first goal had been stopped, none of the others would have followed. It’s more like flipping a coin a bunch of times and getting a random string of heads or tails.

Close Games are a Toss-Up

When a team puts together a string of close wins (or losses), here come the hot takes on what it means for their culture or mental toughness. And while good teams do win more close games, they also win more not-close games. So, more accurately, they just win more.

And when the streak of close wins is broken, it vanishes without a trace – again, just as if it were the result of random chance.

There is No Rack

Common wisdom says teams put the cue in the rack once a game is won, so might let through a few junk time goals that don’t really mean anything. A similar piece of common wisdom says teams on the wrong end of a belting drop their heads and so also let through late goals that don’t mean much. These ideas contradict each other.

In fact, nothing special seems to happen in junk time at all. Late scores contribute just as much information about future team performance as any other. So, in reality, better teams win by greater margins no matter what the clock shows when they score.

There isn’t Much Wood, Either

There’s that one team. The bunny. No matter how badly things are going, you always beat them. At least, at a particular venue. You can count the streak going back years and years, even to when none of the players or coaches were the same.

There might be times when one team matches up particularly well on another, or when psychological factors are in play, but with 18 teams and 30 distinct venues, you would expect to find long win streaks even if every match were randomly decided.

And once you permit some wiggle room – e.g. “X have beaten Y in 4 out of the last 5 games at venue Z” – it’s even easier to find an interesting but meaningless stat.

Home Advantage is Probably About Crowds

Home advantage is definitely a thing. But people attribute it to a lot of different factors – especially travel distance and ground shape – that there isn’t much evidence for. But there is reason to believe that if you dominate a stadium with your team’s fans, it will exhibit home ground advantage, no matter who travelled where. The main vector is likely to be social pressure on umpiring decisions.

Travel is a burden, to be sure, and generates physical and logistical challenges that teams have to manage. But its effect on game-day performance seems small enough to be hard to detect at all. This is probably why a 50-minute drive to Geelong depresses the performance of Melbourne-based teams more than a flight to Queensland.

Also Home Advantage is Not Everything

At its strongest, home advantage is probably worth somewhere around two goals. Which is nothing to sneeze at. But usually it’s less. And the world’s best football tipper will struggle to reliably get within four goals of a game’s actual margin, which gives you some idea of the room for error here. So home grounds aren’t quite the unbreachable fortresses that people like to think.

Also, across a season, the home advantage games of teams are so well balanced by their away disadvantage games that the difference is rarely worth getting worked up about. With the exception of 2020, you wouldn’t normally expect any team to win or lose even one whole extra game based on its fixtured home advantage.

When considering who to tip between two evenly-matched teams, it’s logical to lean toward the home team, because they really do win more often. But that doesn’t make it anything like a sure thing.

Beware the Bye, Not the Short Break

It makes intuitive sense that a team with fewer days’ break between games than their opposition is at a disadvantage. And, like travel, it’s obviously something you’d avoid if you could. But there isn’t much evidence that a day or two’s less rest than your opponent makes you win any less.

There is, however, a reasonable case to be made for the Killer Bye – the increased likelihood of poor performance following a much longer break than your opponent. And not just if you’re wearing blue hoops.

There is a Lot of Unpredictability in Football

Firstly, people aren’t great at understanding probability:

Actual ChanceWhat People Hear, Apparently
50%50%
50.1%60%
60%90%
90%100%

Also, good tippers get somewhere around 6/9 correct in the long term. So do average tippers. For many people, the number of tips you get right in a year depends on how predictable that year turns out to be, how lucky you are, and how much you know about football, in that order.

So while some people are better tippers than others – and the market as a whole is better than all of them – the difference can be hard to detect, especially within a single season.

What this really means, though, is that there’s always a chance. No matter how terrible your team and how numerous or confident the people tipping against them, there is always a pretty good chance that those people are wrong. And thank God, you know, because who wants football to be that predictable?

The Squigglies 2020: Reviewing Preseason Ladders

It was a rough year for preseason ladder forecasts, with not one of the 56 experts, websites and models tracked by Squiggle managing better than a C+ grade. So, to be honest, no-one should be that proud. Then again, 2020 has presented unique challenges for forecasters, so perhaps it’s surprising that anyone got as close as they did.

Every Expert Preseason Ladder Rated

Best Ladder: Sam McClure

All year, Nick Dal Santo seemed to have this award on lock, but in the final round, McClure got him by the barest of margins.

McClure’s ladder is, at first glance, not great: He has GWS to win the minor premiership and Fremantle the wooden spoon. But it’s deceptive, because between these bookends, every other team is not far off, with two-thirds within three rungs of their actual position. The two misses in his Final Eight (Geelong and St Kilda) were tipped for 9th and 10th. So it’s not flashy, but it adds up.

Dal Santo’s ladder is ever so slightly weaker across the board, but boasts only one miss in the Final Eight (GWS instead of St Kilda), and three teams within a single rung of their actual finish, including the Crows at 17th.

Runner-Up: Nick Dal Santo

Best Ladder by a Model: Squiggle (20th overall)

It was a particularly bad year for models, who ply their craft by analyzing a fixture that, as it turned out, went up in smoke after Round 1. Eleven of the worst 14 ladders were by models, who were commonly stung by North Melbourne, West Coast and Hawthorn. Squiggle escaped the pit mainly by tipping a bottom-2 finish for the Crows.

Honourable Mention: The Flag (34th overall)

Lifetime Achievement Award: Nat Edwards

Anyone can fluke a good year; true oracles have staying power. Of the 24 forecasters tracked in both years so far, Nat Edwards has the best long-term record, following a 7th-placed finish in 2019 with 6th this year.

Honourable Mention: Mitch Cleary (14th in 2019, 5th in 2020).

Worst Ladder: PlusSixOne

There are things to like about this ladder: Geelong in the Top 4, which many missed, Port Adelaide up at 5th, and Essendon bang on at 13th. But around the middle it’s a disaster area, with Hawthorn and Adelaide making finals just ahead of Sydney, while the Saints sit way down at 16th. There but for the grace of God go any one of us, and there were plenty of wonky ladders in 2020, but someone has to be last, and this year it’s this one.

Introducing s10

There’s a new entry on tip pages named s10. This isn’t a model; instead, it’s the average of tips by the 10 top-performing models of the previous year, as measured by MAE.

The idea here is that in the future, we can allow more models to join the Squiggle platform without worrying so much about whether they’ll turn out to be any good. Because although they might affect the performance of Aggregate, they can’t throw out s10.

Aggregate is the average of all models, including Punters.

s10 is the average of the 10 models with the lowest season MAE the year before.

Note that if a model qualifies for s10 but does not participate, it is omitted and not replaced.

Years10 Models (in alphabetical order)
2017
2018The Arc, FMI, GRAFT Ratings, Matter of Stats, plusSixOne, Squiggle
2019The Arc, FMI, GRAFT Ratings, Live Ladders, Massey Ratings, Matter of Stats, plusSixOne, Squiggle, Stattraction, Swinburne
2020Aflalytics, AFL Lab, The Arc, GRAFT Ratings, Live Ladders, Matter of Stats, plusSixOne, Squiggle, Swinburne

Hawthorn are Virtually Season 2020 Premiers

In a classic contest, Hawthorn kicked the last three goals to defeat Port Adelaide by 6 points in the inaugural virtual Grand Final.

The Hawks led – barely – at quarter and half time, before the Power turned it on in the third, kicking away to a 13-point break at three-quarter time. But in the final stanza, Hawthorn surged back to tie it up at 79 points apiece with only minutes remaining on the clock. Mitch Lewis delivered the winning goal, with the Hawks resisting a final Port Adelaide surge to capture the premiership.

The Hawks finished 5th after the 11-game regular season, and won four finals en route to claiming the Virtually Season 2020 flag.

Port Adelaide’s Grand Final defeat was their only loss of the season.

Thanks to everyone who followed along with the season – it was a lot of fun during the cold, football-free weeks. Special thanks to the trusty model authors who fronted up each week with their simulations: Stattraction, Live Ladders, Aflalytics, and The Flag, who modeled the Grand Final.

Virtually Season 2020: Home & Away Wrap-up

With the home & away part of the season virtually complete, here’s the final ladder!

Yes, an undefeated season for Port Adelaide, and a winless one for Gold Coast. Other highlights:

  • Sydney finished 6th (and were top 4 with a round to go) with a sub-100 percentage, thanks to a series of close wins.
  • Fremantle also surprised and delighted, in what may be a more sustainable way, climbing steadily as the season progressed.
  • Similarly Adelaide jumped into the Eight in the final round, defeating GWS.
  • Richmond seem to be peaking at the right time, with a dominant display in the final two rounds.
  • West Coast lost five in a row early but seemed to be getting back on track before running out of time.
  • The comp was featured on AFL.com.au.

Thanks to Stattraction, we have Coleman Medal results:

Yes, that’s a shared Coleman. Also a low-scoring Coleman, thanks to a simulation bug that afflicted the first half of the season and stole goals from each team’s main scorer. Also scores in general were down, since we were simulating 16-minute quarters.

Finals begin Wednesday June 3rd at 7:50pm EST and continue almost every night for the next week.

View Virtually Season 2020 here.

Finals Are Virtually Here

Actual football involving real humans is just around the corner! Which means we need to wrap up this virtual stuff!

The last week of Home & Away will be Round 11 by the original fixture, i.e. the one with Port Adelaide v St Kilda playing in China.

Then we’re straight into finals! We’ll play the regular Final Eight system, but compressed into a single week, running from Wednesday June 3rd to Wednesday June 10th.

Virtually Season 2020

Since there is no actual football, we will do the next best thing: simulate it.

With the support of the world’s best football computer models, Squiggle will play out each and every cancelled game in real-time, as if it were really happening.

Goals. Behinds. Score worms. Quarter time breaks. They will all unfold here at the exact same time the match is supposed to be played.

It will look like this:

This will continue all season long, game by game, until actual games resume. We will track a virtual Ladder and Top Eight. If, God help us, we don’t get real football back by September, we will hold virtual Finals and award a virtual Premier.

This Thursday night at 7:25pm Eastern Time, Collingwood will play Richmond in the first virtual match in real-time, here on this site. You can check in and see it happen.

Why Tho

I believe Australia needs football. I need football. Or, in the absence of the real thing, a simulated version from computer models.

How It Works

Usually models make predictions about the most likely outcome of a game (e.g. “Collingwood by 4 pts”). But they can also generate batches of simulations, where if Collingwood is a 60% win chance over Richmond, then in 100 sims, Collingwood will win 60 of them. In the other 40, Richmond will win. (In a few, unusual things might happen, like a team scoring over 120 points.)

Participating models supply Squiggle with their sims. At match time, Squiggle randomly plucks one out and unspools it in real-time. No-one knows in advance which sim it will be.

My Promise To You

This is as rigorous a process as I can make it, drawing from the work of highly talented football analysts and math wonks who created the world’s best football models.

There will be no bias or fiddling. Just hard maths and cruel random variation.

It’s not the real thing. But it’s virtually season 2020.

Virtually Season 2020: Season Results

Squiggle’s Ladder Prediction for 2020

Here’s Squiggle’s own in-house ladder prediction for 2020 (not to be confused with the Aggregate Ladder, which combines this plus predictions from many other AFL models).

This prediction accounts for:

  • 2019 form
  • Trades, retirements, delistings and returns
  • 2020 preseason form
  • Injuries to players listed as “Season” or “Indefinite”

The league was very even in 2019, so it’s going to be harder than ever to make a good ladder prediction. But this is what I’ve got:

2019 Form

Squiggle’s top teams at the end of 2019 were Richmond, Geelong, Hawthorn, and Collingwood. The Cats were widely lambasted for their post-bye form last year, but it wasn’t actually that bad – it was just clearly less good than their 11-1 start (at a percentage of 151%).

The Hawks had a strong finish, Collingwood were 4 points shy of a Grand Final, and Richmond were, well, Richmond.

Notably absent from Squiggle’s 2019 Top 4 were West Coast (8th), Brisbane (5th), and, despite their late surge, the Bulldogs (6th). Squiggle was bearish on the Eagles throughout 2019, primarily because of their reliance on high goalkicking accuracy to win matches – something that, despite much effort, no team has ever been able to sustain for long.

Trades, Retirements, Delistings and Returns

Squiggle uses AFL Player Ratings to gauge the likely impact of list changes between 2019 and 2020, including the return of players who missed games late last year. This last factor is often the important one, as most clubs put out weakened teams towards the end of 2019.

On this measure, the most upside is in Fremantle (regaining Lobb, Ryan, Wilson, Hogan*, Hill, Pearce, Colyer, and Cox, while recruiting Acres and Aish), GWS (regaining Coniglio, Whitfield, and Ward, while recruiting Sam Jacobs), followed by Gold Coast, Collingwood, and Carlton.

At the other end of the scale, the only club to have gone backwards is Adelaide (losing Greenwood, Jacobs, Douglas, Betts), while Brisbane (losing Hodge), Richmond, and North Melbourne have relatively little to add to their sides in 2020.

There’s been a lot of talk about Tim Kelly, but despite his stellar numbers, his trade doesn’t single-handedly drag Geelong or West Coast out of the pack (in either direction).

2020 Preseason Form

The preseason usually contains a few hints about regular season form, and at this time of the year, we don’t have much else. The best pre-performers in 2020, after accounting for the quality of their opposition, were Gold Coast, GWS, St Kilda, Essendon, Port Adelaide and Melbourne.

The worst were Geelong, Carlton, Hawthorn, Richmond, Adelaide, and Sydney.

Long-Term Injuries

For most of the off-season, Squiggle rated Hawthorn a Top 4 team in 2020. But long-term injuries to Howe, Impey, and Hardwick have sent them tumbling to the lower reaches of the final 8.

Also hampered by long-term injury this year are Fremantle (Hogan, Hamling), Collingwood (Beams, Greenwood, Langdon), and Carlton (Curnow).

Summary

The punditry is big on West Coast this year, with the Eagles a popular flag tip and Top 4 lock. Most computer models, however, are much cooler, placing them no higher than 3rd and as low as 11th.

Models have a pretty good record in situations like this, when there’s a divergence of opinion but not because people know something that models don’t. However it shakes out, it’ll be interesting to watch.

Squiggle is high on the Bulldogs, ranking them 2nd, although only by a slim margin. What’s remarkable about the Dogs is how young they are: They’ve been fielding shockingly young teams for two years. Younger teams lose matches pretty reliably, so the ability of the Dogs to make finals in 2019 despite their age profile speaks to their potential upside.

More than any time since 2000 – perhaps since 1993 – we have a very even field entering the new season, so expect surprises! We could have a very volatile ladder, with teams surging and plummeting on the ladder, and a large middle cluster that sits within one or two games of each other.