Are the new rules ‘serving’ up another mode of footy in 2019?

Embed from Getty Images


The new “6-6-6 rule” might have been a headline grabber in the early weeks of the 2019 AFL season, but what I was thinking more about leading into the 2019 season is more akin to a “6-4 6-4 6-4” scoreline.

That’s right, thinking about footy more like tennis.

Wait, what?

For a few years now I’ve noticed a few similarities between the team dynamic of football and individual dynamic of tennis. In both sports, teams/players exchange ‘plays’ and essentially the aim to transport the ball past their opponent/s to hit the scoreboard.

In Australian football these individual plays takes the form of possession chains, with teams trading successive sequences of possessions essentially up and down the ground until one is able to ‘pass’ the other’s defence and a score is recorded. In tennis, players trade successive shots up and back across the net until finally one is able ‘pass’ the other’s defence for a winner (or, enforce an error) to register a point.

alt text
Map of possession chains in Australian football | Map of shots in tennis
Sources: Figuring Footy:

I also like to think of the strategies and strengths of footy teams in the style of tennis player types. Here I think of a team’s ‘first serve’ as its first play out of the centre bounce when it has won the clearance. Like on the court, in football not only can this be a point scorer on its own, but it can heavily dictate the rest of the play until the next score. Then a team’s ‘shots’ are their possession chains, in which they are successively trying to gain territorial advantage (‘court position’) in order to eventually overwhelm the opposition and score. Here may be some analogies of teams’ strategies with regard to commonly accepted types of tennis players:

  • The big first server – These teams have their biggest strength in the initial play from the restarts at the centre bounce, both in quantity (number of clearance wins relative to the opposition) and quality (how much damage they can do from them). Through their centre clearances they either serve a lot of ‘aces’ (centre clearance leading to direct scores), or they set up a lot of the rest of their play via putting their opposition heavily on the back foot through winning strong field position via a forward press.
  • The aggressive baseliner – These teams are less likely to worry about damaging the opposition with their first serve (centre clearance), but are really successful at grinding down their opponent around the ground (‘around the court’). With each possession chain typically a little more damaging than the next one coming back the other way, they are able to score ‘winners’ from all over the ground. They typically prevail by being a little bit better for a little bit longer.
  • The counter-puncher – These teams are strong in defence and are effective in soaking up what ‘shots’ their opponent is hitting their way. They wait until their opponent has found themselves out of position after unsuccessfully attempting a few ‘winners’, and are able to cause damage on the turnover. These teams generate a higher percentage of scores from defensive intercepts, where they exploit their out-of-position opposition to convert defence into attack and score quickly the other way.

First serve wins?

In recent years in the AFL, I think it’s fair to say that ‘court speed has slowed’ and as a result there are ‘longer rallies’ between scores. The consistent and persistent underlying conditions of play have been horde of midfielders surrounding the ball and one or two additional defenders outnumbering a couple of forwards a kick or so either side of the play.

One of my hypotheses heading into this season was that the “6-6-6 rule” (now to be known in this article as the ‘restart positions’ rule) was going to change the initial mode of play for some period of time following each restart. As restricted restart positions would, for the first time, significantly alter this status quo for at least some period after each restart, I was interested in both the first order and potential second order effects of these changes. I was expecting that the ‘first serve’ would become more significant as teams could either potentially score with more ‘aces’ or at least set up to be in more dangerous field position from the first possession chain following a centre clearance. I thought teams would be able to gain more ground from centre clearances (as there would be fewer players closing into the middle), both taking the opportunity to exploit the opposition backline and score directly as well as moving the ball as far away from their own equally-numbered defence, in turn finding scoring easier with quicker and cleaner entries.

Going into 2019, I was wondering to what extent the court would ‘speed up’ in the initial play following a centre bounce, counteracting the overall trend towards slower plays and longer rallies. I was interested in how far the needle would swing back towards favouring the ‘big first server’ and away from the ‘aggressive baseliner’ and/or ‘counter-puncher’.

The approach

There are some limitations in the publicly-available data in trying to address my hypothesis. The first is that play-by-play data is not available in any sense. Score source aggregates by game are available through some channels, but to my knowledge that requires manual collation. The other issue with that data is that it does not contain a temporal aspect so we cannot see the timeliness of the original starting position effects before players are once again able to more freely over the ground.

Instead, my approach was to look to score progression data from AFL Tables and infer the ability to score based on each phase following a centre bounce.

I defined a ‘phase’ as the time it takes to realise an event (goal, behind or siren) following a previous event (goal, behind, or start of the quarter). For each restart from the centre bounce (either at the beginning of a quarter or following a goal within a quarter), I considered what was the following event and how long it would take for that event to occur. Because the data is recorded in ‘count up’ format and doesn’t account for stoppages in play, I used a manual adjustment based on sight of the data (it turned out to be about 50 seconds) to line up the plays following a goal with those from the start of a quarter (as best as I could). This time more or less accounts for the time taken for television broadcasters to play their ads and the ball to be returned back to the middle (where there is a rare double-goal from an immediate free-kick this time is reduced accordingly).

I then aggregated all of these phases following centre bounces each season back to 2018 and looked at three metrics over game duration:

  1. The ability to score in a phase (how frequently are scores recorded?)
  2. The accuracy of scores in a phase (of those scores, how many are goals?)
  3. The scoreboard impact per phase (what are the average points scored?)

The data returns only a mini-break from the usual

1. The ability to score

My first hypothesis was that teams would find it easier to generate shots at goal (resulting in scores) for some period following the centre bounce, while positions are constrained. The data shows that, to date in 2019, any micro story in the uptick from the new ‘restart positions’ rule has been swamped by macro trend ongoing downward pressure on scores. There is some evidence that the rule has retained the ease of scoring from the previous decade for the first minute or so following each centre bounce, but after that there is quite a stark reduction in the ability to score this season (refer 2019 pane).

The average number of scores (per 15-second time interval across the first four minutes) following a restart at the centre bounce. The size of the circles refer to the number of phases in that interval. Note the high rate in 2008 and the gradual drop-off per year until the lowest values in the data set in 2019, countered only by a similar rate within the first minute of a centre clearance.

Within the first minute of game play following the restart, this year’s scoring ability tends to match the long-term trend, which suggests to me that the new rule has temporarily offset the increasing defensive prowess and structures of teams for around 60 seconds. After this time, the average number of scores per phase drops away well under the long-term trend.

Looking across the seasons, it’s interesting to note how much easier it was to generate scores in 2008 than it was in 2009 (where almost a goal per team per game was lost). Of course, we are only looking at a sub-set of all game play in this chart (within the first four minutes of play following each centre bounce), rather than all game play, but these trends should reflect game play and score dynamics following a clean restart (rather than, say, a kick out from one end of the ground).

There is always some risk with binning continuous data that it creates a signal that doesn’t otherwise exist. I played around with displaying this data in a number of different ways (including using smoothing methods) however they tended to have issues fitting to the first few data points (where there can often be a secondary stoppage and rarely a shot within the first few seconds) as well as the overall trend. I hope a set of 15-second intervals both displays the data simply enough to understand but doesn’t create signal where there is none. I think in this case it’s clear to see that until approximately one minute of game play, the 2019 scoring ability matches the long-term trend, whereas after this point it is consistently below the overall average.

2. The accuracy of scores

My second hypothesis was that teams would be able to generate cleaner entries from centre clearances, create shots from better positions and therefore improving conversion from goal kicking. The overall trend has the goal kicking conversion rate a little lower (50.5%) in the first minute than in all post-restart phases (overall 52.7%). The (admittedly a little noisy) data does support this hypothesis, with the goal kicking conversion rate above the long-term trend (52.6%) once again for about the first 60 seconds following a restart. After this point, the conversion rate hasn’t had quite the same drop off as the ability to create scores, however there has been a slight drop off this season (overall 51.4%) in that area too.

The average goal kicking conversion rate, or accuracy, (per 15-second time interval across the first four minutes) following a restart at the centre bounce. The size of the circles refer to the number of shots taken in that interval. Note the relative drop-off in 2019, other than in the first minute or so following a centre bounce, particularly as this period historically has been a little under the stable trend.

If you are interested in an in-depth discussion on factors affecting the reduction in accuracy in the modern game, refer to the ‘Footballistics’ chapter titled Goal Kicking Accuracy which addressed a number of these.

3. The scoreboard impact

My overall hypothesis was that through cleaner entries there would be greater ability to impact the scoreboard in the early moments after a centre bounce. The trend of average points per phase over game duration is effectively a combination of the average scores per phase and goal kicking accuracy.

Overall, there does seem to be some positive impact on the scoreboard within sixty seconds of the centre bounce, particularly in comparison to the overall slow-down of scoreboard impact more generally. The new ‘restart position’ rule looks to have retained a similar potency to the long-term modern trend, while the more macro evolution of the AFL has seen the unconstrained score dynamics dry up much further.

The average points scored (per 15-second time interval across the first four minutes) following a restart at the centre bounce. The size of the circles refer to the number of phases in that interval. Note the high rate in 2008 and the low rates across 2018 and then even moreso in 2019. This suggests the uptick in 2019 following a centre bounce may be significant.

The ace in the pack?

I must admit that during the preseason I had expected the ‘restart position’ rule may have had a bigger impact than what we’ve seen in the first six rounds and that it would have fallen out of this analysis. I thought that we may have seen similar to 2008-style scoring dynamic patterns for potentially 90-120 seconds following a centre bounce. Although this doesn’t seem to be the case, certainly one reason appears to be that any decent uptick in initial scoring conditions has been heavily diluted by the greater overall downward trend. I think I am confident that data shows the mode of football in the first minute post-centre bounce adheres to a different behaviour than the resulting game play from then on.

In summary, we are a quarter of the way through the 2019 AFL season and there is some evidence for a little extra dominance by teams on their ‘first serve’, however at the same times they ‘court has slowed’ significantly during the rallies. It may be slightly more advantageous to have a big first serve this year, but to be a successful side you would also need the ability to either wear your opponent down ‘from the baseline’ during gameplay or ‘counter-punch’ swiftly when the opportunity arises!

I have a few ideas for future analysis looking at this data set and analytical approach. It would be worth addressing the score dynamics at a team-level to see which clubs are stronger or weaker from the centre bounce or in general gameplay. Further, adding in phases following kickouts can add an extra dynamic and can potentially call out the counter-punch ability of sides. This way we could infer and categorise teams into different styles, perhaps equivalent to the tennis analogies I provided above.

I am very open to feedback on this or any of my analysis. If I have missed something, made an error, or if you have any suggestions or ideas, please feel free to comment below, shoot me a direct message or hit me up on Twitter.

All data provided for analysis on this page comes thanks to AFL Tables.

Buddy 900 – and benchmarking the greatest goal kickers in VFL/AFL history

If you’re anything like me, you’re already missing footy. But the fake news-fest that is the AFL trade period doesn’t fill my tank. The ratio of actual news to number of topics and conversations recorded must be tending towards zero. As an alternative, I have a few items in the wings that are best discussed at the end of a given season, as part of what I am calling my 2018 ‘Summer Series’.

This marks the first such post.

Buddy and Plugger

Back in Round 17 against North Melbourne, Lance Franklin belted through his 900th AFL goal – somewhat appropriately from the forward flank outside the 50. It seems as though Buddy has kicked half his career tally from that spot, although that is probably not quite the case.

A total of 900 hundred goals is indeed a significant milestone, as Franklin became just the ninth player to reach the mark in 122 VFL/AFL seasons. But what is perhaps more telling that the all time league record sits more than another 50% higher again than this mark.

When Tony Lockett wobbled a drop punt through for goal against Collingwood at the SCG in 1999, he became the first man in VFL/AFL history to reach 1300 career goals. He would play on, and finish his career with 1360 goals from 281 senior games. Lockett is the most prolific goal kicker in 122 years of the VFL/AFL – but can we say he is the best ever?

The spearheads

Lockett surpassed a mark set by legendary Collingwood spearhead Gordon Coventry which had stood untouched for 62 years. A generation earlier, the Magpies boasted the league’s first great full-forward in Dick Lee who had debuted when the league record was only 144 career goals but by the time he retired he had pushed it way out to 707 goals. Following World War II, Essendon’s John Coleman kicked 537 goals in just 98 games before a serious knee injury prematurely ended his career at just 25. Likewise, two decades later an injured knee also interrupted the career of Peter Hudson however the Hawthorn superstar still averaged 5.6 goals per game in an golden era for full-forwards. Which takes us to the modern era, where Lance Franklin has kicked more goals than any other this century and now has over 900 goals in a period where team scoring has fallen to 50-year lows.

Embed from Getty ImagesLockett and Franklin have booted 2,277 VFL/AFL goals between them

It is hard to compare like-with-like on raw numbers alone, as over more than 120 seasons of the VFL/AFL scoring trends have evolved continuously like a living organism. Yes, no player has kicked more goals than Lockett, but Lockett played in an era where team scores were high and the full-forward thrived. In other eras, defence has been king or a team approach to scoring has been in vogue. In this post I look to benchmark the records of the VFL/AFL’s greatest goal kickers across multiple eras and propose, once and for all, the best we’ve seen.

Goal kicking trends

There are two major trends in scoring over 122 years of the VFL/AFL that have had major impacts on goal kicking tallies:

  1. The number of goals typically kicked in a game
  2. The typical spread of goals across players within a team

Scoring spreads

The ‘Footballistics’ chapter titled Goal Kicking Accuracy, a number of factors which have impacted conversion rates to varying extents over time were assessed. One such factor was the changing nature of the ‘scoring spread’ of teams over time, and this effect was two-fold:

  1. Players who tend to have more shots tend to be more accurate, and
  2. Over time, the proportion of shots/goals taken by the more predominant goal kickers for a team in a given game has tended to fall consistently over time

You can read much more about this and other trends which have impacted goal kicking conversion rates in that chapter.

It is not just the proportion of goals kicked within games that have changed. Even adjusting for the proportion of goals kicked by predominant forwards within a game, this doesn’t full account for changes as players have become more flexible and roles have become more blurred. The leading goal kickers of clubs within seasons are also more likely to have games where they are not one of the leading goal kickers for their team in a game.

For benchmarking purposes, I want to introduce the concept of ‘era-adjusted goals’ (EAGs). The goals of each player will be adjusted accordingly for their rank of goals within a game so that the average proportions of goals in each season is equivalent (1). Then, the season tallies of each player will be adjusted accordingly for their rank of goals within a year, so that the average proportion of season goals in each year is equivalent (2).

Scoring rates

Average scores per game have roller-coasted over the league’s history. From the inaugural breakaway year of 1897, scores bottomed out only two years later when teams managed an average of just 5.01 goals per game. There was then a steady rise for the next four or so decades, peaking at 13.3 goals per game in 1941, before another dip to just 9.7 goals per game in 1952. Once again scores were on the climb for the next 30 seasons, with an all-time league high reached in 1982 with teams kicking an average of 16.2 goals per game. It has been well reported that scoring rates have fallen in more recent times, but this has typically been a gradual regression in the past 36 seasons to 12.0 goals per game this year.

For benchmarking purposes, each game will be scaled proportionally so that the average number of goals per game is equivalent in each and every season (3).

Benchmarking spearheads

To account for the evolution of the league, three separate adjustments were carried out to benchmark the conditions for all goal kickers over time. In order, these were:

  1. On a game-by-game basis, standardise the proportions of goals kicked by each team’s ‘goal rank’ player in each given match
  2. On a season-by-season basis, standardise the proportions of these new scaled goals kicked by each team’s ‘goal rank’ player across each given year
  3. Finally then scale each game goal tally to standardise the average number of goals per game across each season

The methodologies are a little clunky to explain without losing my entire audience, so instead I’ve chosen four examples from different eras to articulate how the benchmarking played out across some famous performances:

Jim McShane’s bag of 11 goals for Geelong in 1899 is scaled up to 17.4 era-adjusted goals (+6.4):

  1. First scaled down because in 1899 a typical team’s top goal kickers in a match kicked a higher proportion of goals in a game than the 122-season average
  2. Then further scaled down because in 1899 a typical team’s top goal kickers in a season kicked a higher proportion of team goals in a year than the 122-season average
  3. Then finally scaled up because in 1899 games featured much fewer goals in a game than 122-season average

Fred Fanning’s bag of 18 goals for Melbourne in 1947 is scaled down to 15.9 era-adjusted goals (-2.1):

  1. First scaled down because in 1947 a typical team’s top goal kickers in a match kicked a higher proportion of goals in a game than the 122-season average
  2. Then further scaled down because in 1947 a typical team’s top goal kickers in a season kicked a higher proportion of team goals in a year than the 122-season average
  3. Then finally scaled down because in 1947 games featured slightly more goals in a game than 122-season average

Lance Franklin’s bag of 13 goals for Hawthorn in 2012 is scaled up to 14.9 era-adjusted goals (+1.9):

  1. First scaled up because in 2012 a typical team’s top goal kickers in a match kicked a lower proportion of goals in a game than the 122-season average
  2. Then further scaled up  because in 2012 a typical team’s top goal kickers in a season kicked a lower proportion of team goals in a year than the 122-season average
  3. Then finally scaled down because in 2012 games featured slightly more goals in a game than 122-season average

Jack Riewoldt’s bag of 10 goals for Richmond in 2018 is scaled up to 12.5 era-adjusted goals (+2.5):

  1. First scaled up because in 2018 a typical team’s top goal kickers in a match kicked a lower proportion of goals in a game than the 122-season average
  2. Then further scaled up  because in 2018 a typical team’s top goal kickers in a season kicked a lower proportion of team goals in a year than the 122-season average
  3. Then finally scaled up because in 2018 games featured slightly fewer goals in a game than 122-season average

Every such goal tally from every player in every game from 122 years has received the exact same treatment based on the scoring characteristics of the league in that season. We therefore end with both actual goals and era-adjusted goals tallies for all players, in every match, season and career.

The greatest of them all

The best careers

This analytical approach to benchmarking the greatest goal kickers in VFL/AFL history presents… Tony Lockett as the most prolific sharpshooter ever! Was that a surprise? I’m not sure. However as a result of the era in which Plugger played, his tally is adjusted significantly downwards to 1209 EAGs (-151 on his actual tally). Jason Dunstall (1103 EAGs, also -151) hops into second position on the overall adjusted tally, jumping Gordon Coventry (1082 EAGs, -217) who slips into third.

It is significant that Lance Franklin’s career record is well respected by the analysis, jumping from eighth to fourth on the tally (1034 EAGs, +117). As big as Buddy has been in the modern era, does the current media and footy pundit still underrate the imprint left by him on this league? Only two others fair better in additional adjusted goals than Franklin, both from the early decades of the league – namely Dick Lee (863 EAGs, +157) and Jack Leith (279, +117) who in played in such a dour era he is awarded an additional 72% of his actual tally as a result.

The top 20 goal kicking performances across a career by total era-adjusted goals

With regards to averages, it probably comes as no surprise that the brightest star shining is that of John Coleman. The player whose name is enshrined on the annual medal for the season’s leading goal kicker stands above all others, average 5.26 EAGs per game (-0.22 on his actual average), leapfrogging Peter Hudson (4.99 EAGs per game, -0.64). Dick Lee (3.75 EAGs per game, +0.68) and Lance Franklin (3.56, +0.40) are the big winners on this measure, with many of the prolific goal kickers in history more harshly punished by the high-scoring eras they played in.

The top 20 goal kicking performances across a career by average era-adjusted goals per game (minimum 50 games)

It tends to be that the most rewarded (or, underappreciated by raw goal measures) are for the primary forwards playing in three eras: the early days (until about 1920); an approximate decade between the mid-1950s and mid-1960s, and indeed in the modern era (since about 2000). Perhaps expectedly, those spearheads from those halcyon eras of the 1970s-1990s are more harshly penalised (or, had the benefit of playing in eras that more suited their craft).

The following dashboard contains two dynamic views summarising the top VFL/AFL goal kickers of all time, with the ability to toggle between era-adjusted and actual goals. The default views filter those players with at least 500 actual goals, at least 2.5 actual goals per game, and at least 50 career games (but you can change these if you wish).

The first pane effectively combines the two above charts into one, comparing those goal kickers on both total (horizontal axis) and average (vertical axis) era-adjusted goals per game (first pane, ‘By totals and averages’). Those towards the right of the chart are those who have kicked the most era-adjusted/actual goals, while those towards the top of the chart are those who have averaged the most era-adjusted/actual goals per game.

The second pane provides a view of running tally of career era-adjusted/actual goals by match number. Here you can compare goal kickers like-for-like, particularly in adjusted terms. Note the similarity of the paths of Gordon Coventry and Lance Franklin, for example, which we will refer to again later. Also note how the era-adjusted measures pushes John Coleman’s trajectory slightly above that of Peter Hudson.

The best seasons

Looking across the best performances in a given year, famously both Bob Pratt (1934) and Peter Hudson (1971) managed 150 goals in a season. In terms of EAGs, Jason Dunstall’s 1989 season (132 EAGs, -6) becomes the most impressive tally in a given year, with the 1971 season of Hudson (126 EAGs, -24) dealt with more harshly and demoted into second. Of the best adjusted seasons, Dunstall (1989, 1988) and Hudson (1971, 1970 and 1968) end up with the top five on the list. Pratt’s 1934 (105 EAGs, -45) is not rated anywhere near as rosily, falling even below Lance Franklin’s century topping year of 113 actual goals (110 EAGs, -3) and indeed out of the top 20. Most of the top 20, even on adjusted terms, tend to be scaled down – primarily because they tend to be from the modern era, and that is primarily because those players have tended to play in longer seasons.

The top 20 goal kicking performances across a season by total era-adjusted goals

For season averages (minimum ten games), Tony Lockett’s injury-affected 1989 tops the charts with an astronomical 6.82 EAGs per game from 11 matches (-0.27 goals per game down on the actual), followed by Peter Hudson in 1968 (6.37 EAGs per game, -0.21) and Lockett again in 1991 (6.20, -1.28). Alongside Hudson, John Coleman’s name appears three times, in 1952 (5.94 EAGs, +0.22), 1953 (5.71, +0.32) and 1950 (5.25, -1.06).

The best games

One of the best known records in the VFL/AFL is Fred Fanning’s mark of 18 goals in, remarkably, his final senior game in 1947 before taking up a coaching position in the sticks. But with our approach to benchmarking, this performance is relegated to third with an adjusted mark of 15.9 goals (-2.1 on the actual). It is surpassed by the exploits of Jim McShane (17.4 EAGs, +6.4), who kicked 11 majors for Geelong in 1899 against a hapless St Kilda, and Harold Robertson (17.2, +3.2) who booted 14 also against St Kilda in 1919.

The top 20 goal kicking performances in a game by era-adjusted goals

Thanks to Anthony Hudson, Lance Franklin’s 13 majors against North Melbourne in 2012 won’t be forgotten in a while, and the data shows this is for good reason. It is ranked as the sixth greatest individual era-adjusted goal kicking performance on record.

Buddy beyond

So Lance Franklin is ranked fourth all-time in our era-adjusted goals, behind only Coventry, Dunstall and Lockett. As you can view in the dynamic dashboard view (second pane, ‘By number of games played’), he has tracked alongside Coventry game-by-game in EAGs for his entire career, recently pulling ahead as Buddy heads towards 300 games. Can he continue and overcome all three on relative measures, if not absolute tallies?

Franklin has averaged 3.12 goals per game at Sydney, just under his Hawthorn average of 3.19 per game. Given the slight changes in scoring trends even in the past half-decade making goal kicking even harder for main spearheads, his adjusted average at the Swans (3.78 EAGs per game) even outrates his time at the Hawks (3.44).

Helped by finals appearances, he has averaged 21.6 games per season at the Swans. With four years left on his contract, let’s assume he averages 15 games per season in the twilight of his career, for a total of 60 games left to come. This would take him to 350 total, which would take him to equal-16th on the all-time VFL/AFL games tally list. At his current Sydney goals per game rate, we can expect another 187 goals or another 227 EAGs – taking Buddy to 1,104 goals (leaving him in fourth position, behind the same three players) and 1,261 EAGs (surpassing the lot of them). In fact at his current rates, it would only take another 47 games or so for Franklin to eclipse Lockett, Dunstall and Coventry and be ranked as the number one adjusted goal kicker of all time, holding scoring conditions equal.

As a result of this analysis I keep asking myself this question: is the career of Lance Franklin still underrated?

Please comment below or reply to me on Twitter!

The ‘Miracle of the Saints’ and more on Win Probabilities

The ‘Miracle of the Saints’

Embed from Getty Images

Three weeks ago, Carrara played host to a remarkable result, in what may have easily been scripted as an unremarkable match-up between 15th-placed Gold Coast and 16th-placed St Kilda at Metricon Stadium.

The off-Broadway match was one that may have easily slipped into the ether, with two no-hopers of 2018 playing in front of just over 10,000 spectators on the Gold Coast, and hardly demanding a television audience with a Saturday twilight time slot ahead of the Socceroos’ 2018 FIFA World Cup campaign launch against France.

The ScoreWorm as per the AFL Match Centre for 2018 R13 Gold Coast v St Kilda

Journalists were no doubt starting to sharpen the knives to finish off embattled coach Alan Richardson as the Saints trailed by 39 points with 26 minutes played in the third quarter. About 35 minutes – and 6.5 (41) to 0.0 (0) later – the Saints had completed the miracle with a dramatic two-point victory. It was the first match this season where a team had come back to win from a deficit of 30 points or greater.

‘Footballistics’ context

One of the 15 chapters in James Coventry’s new book ‘Footballistics’, named Win Probabilities, assesses various ‘heuristics’ or rules of thumb which are commonly adopted by the footy fan, pundit or commentator as a way to ‘call’ a victory for a particular team during a game. Thinking about it another way, this is when the individual believes that the in-game chance of victory for a particular team approaches close to 100 per cent.

In particular, the chapter considers the winning chances of teams meeting the criteria for three heuristics:

  1. Teams reaching 100 points before to their opposition (which considers only points scored)
  2. Teams with at least a 30-point lead (which considers points scored and points conceded)
  3. Teams leading by more goals than minutes remaining (which considers points scored, points conceded and time remaining – but typically relates only to the dying minutes of a match)

The winning rates of all these situations are discussed in some depth within the chapter.

The natural extension to these themes is the creation of a model which can provide an approximation to a probability of victory given point in a match – taking into consideration both scoring factors as well as time remaining. Left on the cutting room floor of the chapter were a list of the most unlikely comebacks, based on a simple logistic regression model I created at the time.

Coincidentally, the same topic was raised on Twitter over the following weekend following various throwbacks to the Bombers’ victory over the Kangaroos from a 69-point deficit in 2001.

I had dragged together much of this content following the St Kilda win over Gold Coast, but this Twitter conversation was the catalyst to (eventually) finishing the post (two weeks later).

A simple in-game model for win probabilities

Recently I have slightly updated the model to create a number of variations. One simpler but stronger combination, which I’ve named WinProb2, takes into account:

  • Percentage of game time duration remaining
  • In-game running margin of the chosen team
  • In-game running combined score (points) of both teams
  • In-game running combined scoring shots of both teams
  • In-game running difference in scoring shots of the chosen team relative to the opposition
  • Home team

For simplicity, estimate probabilities have been calculated at the time of each score, and a second prior to each score, for all in-game scoring events since 2008. In practice, what this means is that probabilities have been estimated as the ball has effectively carried through for a score but before the score has been officially recorded. It is an important point to note: the model doesn’t know where the play is situated, or what may have happened and is yet to be signalled – it only considers the simple attributes above.

The most unlikely successful comebacks since 2008

Using WinProb2 as the basis for estimate in-game win probabilities, the following lists the top 100 most unlikely successful comebacks across all AFL Premiership Season matches between Round 1 2008 and Round 16 2018 inclusive.

0.3%2013R13325:08Brisbane LionsGeelong-51+5
0.4%2008R4324:13Brisbane LionsPort Adelaide-47+20
0.5%2013R9415:13AdelaideNorth Melbourne-30+1
0.6%2015R6304:31St KildaWestern Bulldogs-55+7
0.7%2011R5329:33Gold CoastPort Adelaide-40+3
0.7%2011R23410:17EssendonPort Adelaide-34+7
0.8%2018R13326:06St KildaGold Coast-39+2
0.8%2008R21401:51CarltonBrisbane Lions-32+6
1.1%2013R23321:30CarltonPort Adelaide-38+1
1.2%2013R5320:00Port AdelaideWest Coast-41+5
1.2%2012R8413:42Port AdelaideNorth Melbourne-32+2
1.5%2008R11319:06CarltonPort Adelaide-38+12
1.8%2017R15405:10Brisbane LionsEssendon-27+8
1.9%2008R17329:46RichmondBrisbane Lions-31+3
2.1%2017R2334:03GeelongNorth Melbourne-31+1
2.2%2009EF405:23Brisbane LionsCarlton-29+7
2.3%2008R17323:44CarltonWestern Bulldogs-31+28
2.4%2013R19422:01Port AdelaideAdelaide-20+4
2.7%2011R20413:48AdelaideBrisbane Lions-22+5
2.7%2014R6414:26CarltonWest Coast-24+3
3.1%2011R23319:53West CoastBrisbane Lions-29+8
3.2%2008R5412:51North MelbourneCollingwood-21+7
3.3%2008R11304:31SydneyWest Coast-37+5
3.3%2013R16325:02Brisbane LionsNorth Melbourne-33+12
3.4%2017R9405:30Greater Western SydneyRichmond-27+3
3.6%2013R2229:50GeelongNorth Melbourne-41+4
3.8%2015R9301:16CollingwoodNorth Melbourne-39+17
3.8%2013R14330:36EssendonWest Coast-23+7
3.9%2016R21423:57West CoastGreater Western Sydney-18+1
3.9%2015R15311:48Western BulldogsGold Coast-37+22
4.1%2009R21131:48Brisbane LionsPort Adelaide-47+15
4.4%2012R16418:31Gold CoastRichmond-18+2
4.5%2008R8414:00Western BulldogsFremantle-18+3
4.5%2012R15219:31West CoastNorth Melbourne-35+2
4.5%2016R12316:36AdelaideWest Coast-26+29
4.6%2014R18406:38EssendonWestern Bulldogs-20+7
4.8%2015R13401:34AdelaideBrisbane Lions-24+13
5.1%2017R14419:08MelbourneWest Coast-16+3
5.3%2012R23415:57St KildaCarlton-15+15
5.4%2008R13319:15Western BulldogsCollingwood-23+10
5.6%2014R21310:35EssendonWest Coast-34+3
5.6%2008R16229:47St KildaHawthorn-34+30
5.6%2010R15210:17CollingwoodPort Adelaide-36+26
5.8%2008R9314:53North MelbourneWestern Bulldogs-24+3
5.9%2016R8416:15CarltonPort Adelaide-18+2
5.9%2011R17200:44Gold CoastRichmond-36+15
6.0%2010R6320:31St KildaWestern Bulldogs-23+3
6.2%2009R6302:34FremantleWest Coast-23+13
6.2%2012R22331:58Brisbane LionsPort Adelaide-19+11
6.2%2016R1328:26MelbourneGreater Western Sydney-22+2
6.2%2009R1210:06Brisbane LionsWest Coast-38+9
6.3%2011R1416:07FremantleBrisbane Lions-17+2
6.3%2017R18404:07CollingwoodWest Coast-24+8
6.4%2018R15327:29AdelaideWest Coast-27+10
6.4%2014R12303:15North MelbourneRichmond-35+28
6.5%2014EF306:05North MelbourneEssendon-33+12
6.7%2008R14209:36St KildaNorth Melbourne-33+15
6.9%2017R4307:17Western BulldogsNorth Melbourne-29+3
7.0%2017R10312:21MelbourneGold Coast-30+35
7.0%2014R4324:01Western BulldogsGreater Western Sydney-20+27
7.0%2016R3401:28HawthornWestern Bulldogs-19+3
7.1%2012R10407:09Brisbane LionsWest Coast-21+2
7.2%2013R21325:09Western BulldogsAdelaide-22+17

The most unlikely victory across ten years was the ‘miracle on grass’ by Brisbane against Geelong once again in 2013. The Lions trailed 5.6 (36) to 13.10 (88) into time-on in the third quarter, with our model giving them just an estimated 0.3 per cent chance of victory. In what really was a remarkable result, Brisbane managed 10.7 (67) to 1.4 (10) in just over a quarter to pinch the game – and even then still needed a kick after a siren, thanks to Ashley McGrath’s effort from about 50 metres in a fairytale 200th game. Clearly an unlikely result, whether you are talking in layman’s terms or mathematically.

It is of interest that even though only three of the top ten most bleak situations occurred in the final quarter, the going was that tight that all but one still only resulted in a winning margin of a seven point or fewer. It suggests that, even with well over a quarter of play remaining in the match, many of the wins were that tight that the trailing team needed every last minute remaining. The only match which ended in comfortable result was the remarkable 2008 match between Brisbane and Port Adelaide. In that game, the Power led by 47 points during time-on in the third quarter, before the Lions somehow found 11.8 (74) to 1.1 (7) from that point onwards to eventually win easily. Given nine of the top 31 results are from that season of 2008, I am conscious that the model perhaps is too well-calibrated to a modern, more-stodgier style of game play and may overstate the chances of a large comeback required in a higher-scoring game ten years ago.

But other comebacks came from particular dire circumstances are found late in final terms. In Round 9 of 2013, the Crows were 30 points down half way through the final quarter against the Kangaroos with an estimated chance of just 0.5 per cent, but managed to win by a point in a stunning result. At 21st position on the table, another game later in that season featured Adelaide finding itself on the other side of the ledger. The Crows this time led the Power in the Showdown by 20 points well into time on, with Port Adelaide given just a 2.4 per cent chance but still able to find a way to win by four points.

And lo and behold, well entrenched inside the top ten was the St Kilda performance against Gold Coast back in Round 13. The Saints may not have had too much to get excited about this season, but that result was one of the most unlikely in over a decade.

The ‘greatest comeback of all time’

Embed from Getty Images

Casting our minds back to the Bombers’ greatest comeback of all time against the Kangaroos, I applied the WinProb2 model according to the scores at the moment prior to Essendon’s goal eating into the then-69-point margin. Intriguingly, given the length of game duration remaining combined with the high scoring rate, the model estimated the Bombers still maintained a 2.4% chance of victory – placing 20 comebacks more unlikely in the past ten years alone. One thing I must note is that match was an outrageously high scoring game for the time, never mind for the dour matches of today. Because the model has only been trained on in-game score data since 2008, the model has had very little chance at being exposed to such extreme scoring , which may weaken the level of certainty around its estimate predictions.

Possible improvements

There were other various parameters I played with but they didn’t greatly improve the model, and I liked how well this model performed against one with many other combination of parameters given its simplicity. For example, I attempted trying to control for a ‘scoring end’ (based on wind or weather conditions), by taking into account the scoring rate at each end and the time teams had played to each end. This perhaps may pick up a stronger signal in a future improved model by factoring for various grounds (say, University of Tasmania Stadium in Launceston which seems to be often dominated by windy conditions).

One other definite improvement would be to understand contextualised chances of victory based on team quality and the pre-game estimate chance of victory. The only factor in the current models which takes into account any team context is a flag for the home team (which proves to be advantageous). This would come at the cost of moving the model away from its current simplicity, but would be a natural progression from considering factors based upon merely scoring, margin and time. To get a much better understanding of the chance of winning for a specific side against a specific opponent, a team- or player-based rating model would be required. I am yet to build one of these, but you will find team ratings at excellent sites like Matter of StatsSquiggle, The Arc and a lovely player-based rating model at HPN.

Another element that needs to be modelled differently is the probability of victory within the last minutes of a match where the margin is small. The model is not well calibrated to deal with predicting results with margins tight in the last five per cent or so of matches. I’ve played with a few combinations but as yet I haven’t clocked this one, so I would hesitate to use the existing model to provide estimate probabilities on these types of matches.

Finally, the bias towards 2008 in the results may suggest that it is not adequately fitted to different modes of play across seasons earlier in the sample.

Further analysis

I am looking to build out these models and add more analysis on this topic in future weeks. To view every ScoreWorm since 2008, feel free to browse and play with my dynamic visualisation.

To read more about footy analysis driven by data, why not support great footy writing and our nation’s broadcaster by purchasing ‘Footballistics: How the data analytics revolution is uncovering footy’s hidden truths’ from the ABC Shop.


I am very open to feedback on this or any of my analysis. I expect many readers may have a deeper grounding in statistics than myself, and love to learn. If I have missed something, made an error, or if you have any suggestions or ideas, please feel free to comment below, shoot me a direct message or hit me up on Twitter.

All data provided for analysis on this page comes thanks to AFL Tables.

Footballistics on sale today!

Over the past 18 months I have had the great opportunity to work with ABC journalist James Coventry to provide statistical analysis and data visualisation for three chapters of his new book ‘Footballistics’. James also utilised a raft of very talented other analysts from other websites such as Figuring Footy, Matter of Stats, Ranking Software and HPN Footy.

This proved to be a lot of fun (and a lot of work!) and I’m looking forward to future questions and further investigation into a number of these areas. I contributed to three chapters, namely:

  1. Goal kicking accuracy
  2. Win probabilities
  3. Australian Football Hall of Fame

In each chapter there was some content that was left on the cutting room floor. In the next little while I will be looking to utilise this ‘extra content’ and fill out some of the analysis that was threaded throughout these chapters.

If you are keen to find out more or if you are interested in a copy, you can read more about Footballistics on the ABC Shop online.


A new beginning

Welcome to the new InsightLane website. This has been a long time coming, two years after I launched my bits-and-pieces Twitter account and launched a temporary website without my own domain.

This website is the hub of my statistics, insights, analytics, and data visualisation across a range of Australian-centric domains. It primarily will cover Australian football (specifically the AFL), with plans for bits of pieces of other sports, as well the odd snippet of weather and politics data.

The past 12 months has been a particularly busy time for myself on multiple fronts. One of these is particularly exciting and I am looking forward to being able to announce some news later on in the year.

The new footy season approaches and this year the AFL analytics community has (seemingly) (for the most part) lost the regular blog posts of pioneers @TheArc and @FiguringFooty. I have some new ideas and will look to take up some of their slack.

Please bear with me as I get to terms with the WordPress functionality and what I can and cannot do with this theme.

Always feel free to add me on Twitter or shoot me through an message with a question, compliment, criticism or idea.