I've had a few people asking how my AFL visualisations are created. The video below demonstrates the Shiny app built in R to collect the data required and produce the visuals. The complete code is available on Github:
#rstats
#performanceanalysis
Network plots for each AFL team looking at the players that pass the ball to each other (min. 10 passes). Some obvious "hubs" within each team and some interesting connections as well. Data from the play by play dataset. Ade, Bris, Car, Coll here, remainder to follow.
Played around with the AFL GPS data made available by
@AFLLab
over the weekend, combined with the event data from
@databyjosh
to insert the ball (syncing required). Here I wanted to get an idea of the set ups more than 5m ahead and behind the ball.
I haven’t tweeted much AFL stuff this season due to coaching at local level. I’d seen the AFL Stats Pro page has timestamped videos for each player, so I thought I’d grab the tags and turn them into a play by play which is available here: (~35mb) 1/n
Introducing the first iteration of
#bounceR
, a cricket
#rstats
package that extracts Hawkeye stats from the ICC website for Men's and Women's Test, ODI and IPL matches where Hawkeye has provided tracking. Download from
#AFLTigersBlues
Both teams can feel unlucky for different reasons with any drawn game. Carlton in this case because they were leading with 30s to go, Richmond because they "outplayed" Carlton for most of the game but couldn't get results on the scoreboard. 1/5
With AFL Free Agency underway and Trade Week kicking off I've built a tool to find similar player comparisons and to help project player performance into the future. The app is available at and an example of the output for Zac Williams is below
#AFLTrades
New
#AFL
list depth and form analysis page
Provides a look at each team's depth across positions and the current form of each player. Clicking on the ... in the More column brings up an in depth view. Not formatted for mobile although it can be used. 1/2
Ahead of tonight's
#AFLDraft
some analysis of drafting performance in the 2008-2017 ND compared to winning % in the 2011-2020 seasons. A little bit surprised about the strength of the association between the 2 variables given the simplicity of the model.
Most AFL previews look at the quality of the teams and how that will impact the match. Trying something different here, looking at contest style, defensive style & offence style, with only a small judgement on the respective qualities. First up are
#AFLEaglesTigers
Those calling to ban marks from backwards kicks to increase scoring haven’t really thought it through. Limiting attacking options makes it much easier for the defence to set up. It was also trialled in the VFL in mid 2000s as a “solution” to flooding. Had the opposite effect.
Trialling a new visualisation of disposals from set (Mark or Free Kick) position in the defensive half. Circle has a 40m radius and the data is from 2nd half of
#AFLFreoHawks
. Feedback and suggestions are welcome.
There's 3 possible outcomes for a possession chain in the AFL. The outcomes are score, turnover or stoppage. The plots below show 2021 results (For and Against) with stoppage results split by clearance percentage. Port's attack is the outlier among the top 4.
#AFLPortDogs
style preview. Added a stoppage exit style from mid arc throw ins. Only 30 or so of these a game but there's a lot that happens tactically, both around the stoppage and defensively behind it that I think makes it worthwhile to look at.
After 4 seasons full time and 3 part time with the
@essendonfc
I started today with the
@VicInstSport
Looking forward to the challenge of working with multiple sports in a high performance environment.
Was hoping to do this before the Round started but wanted to take a look at how ball movement has changed in 2020 compared to the last 2 years. Ball movement speed has slowed down noticeably this year. But why? 1/n
A longer form preview of the
#AflDeesDogs
#AFLGF
I've tried to dig into the data with some non-typical metrics across key areas of the game. Data is from the
@databyjosh
dataset, my AFL play by play dataset and footywire. My tip is Melbourne, but really I just hope it's close.
Offensive/Defensive Efficiency R9 update. What stands out?
1. Even after a slip on the weekend, Freo's defence is 🔥
2. Port's have jumped into the ➕on the back of defence
3. Essendon's defence 🤢
4. Richmond have their offensive mojo back 👀
The ability to move the ball into scoring position in the
#AFL
changes based on where and how the ball was won. Simplifying where into zones and how into intercept/clearance/kick in we get a pretty good overview of how teams are going with their ball movement and defending it.
Offensive/Defensive Efficiency R6 update. What stands out?
1. The spread between teams grows with the top 4 all pulling away and North continuing to fall
2. Geelong's offence 👀👀👀
3. Melbourne & Freo's defence post turnover 👍
4. Richmond's post clearance defence 😢
Little bit of analysis from the AFL play by play dataset. Looking at the ability of teams in offensive vs defensive contests. 3 distinct clusters. 1st where off/def ability correlate fairly well, 2nd with poor defence, 3rd with⬆️off. performance but variable defence.
I thought I'd picked the dud Saturday night game to watch again but it turned out to be a heart stopper for
@CarltonFC
fans as they just hung on. Some post match analysis
#AFLCatsBlues
Hope you don't mind me using your great kick to handball analysis in this
@MatterOfStats
Presuming you can get it there safely, the closer to the 50m line the better on inside 50 kicks. The deeper you get the ball the more likely you are to score (oval overlay is MCG so "hotspot" probably moves to top of the square). Deeper is also easier to defend on turnover.
In time for
#AFLTrade
my Player Similarity site has been updated with 2021 data and is ready to project future performance. A couple of tweaks to the algorithm will hopefully increase the accuracy as well. The link is
Possession flow diagram from
#AFLDonsDogs
The blue ovals highlight the issues the Bombers had in getting the ball forward in 2nd & 3rd quarters. The long, wide kick out of D50 was particularly ineffectual in the 3rd quarter.
In the AFL teams can either take the ball directly towards goal or shift it around in an attempt to work through the opposition defence. The differences between teams as defined in the plot are fairly small, but seem to match up with their ball movement styles.
Offensive/Defensive Efficiency R7 update. What stands out?
1. Freo stand on their own as the top defensive team 👍
2. Port's defence is trending ⬆️
3. Gold Coast's offence post clearance 🤢
4. West Coast & North continue their 🛝
A very simple method of looking at which is more important in the AFL, field position or possession. For the top 4 teams in each of the 2012-2020 seasons, I'd class 17 of the 36 teams as field position teams vs only 3 as possession teams.
As
#AFLTrades
continue next week it's worth considering how mid-career players perform in the future compared to the past. At age 26, there's a <50% chance that they maintain their previous 3 years performance over the next 3 years, for all positions other than Rucks.
First and very generic look at the AFL GF GPS data from
@AFLLab
The data is big and synching with play by play is an issue so take these results with a grain of salt. I looked at GBG contests and the number of players from each team within 5m at the time of possession.
Inspired by this post about stabilisation of skills in baseball I investigated various AFL stats.
The article explains the details better than I can and if anyone wants to double check my analysis I'd be happy to correct any mistakes.
If you consider yourself an AFL data analyst, you now have the data to show off the full range of your skills. We're about to see what the community knows about footy and hopefully broaden everyone's understanding of the game. I can't wait to see the impact.
Gold Coast getting plenty of entries but Essendon defending them well so far. Essendon retaining their kicks but struggling to convert into scores
#AFLSunsDons
#AFL
Post R4 Off/Def Efficiency Ratings. What stands out?
1. North's defence is killing them scoring wise and creating horrible starting positions for their offence.
2. Most of Essendon's defensive issues are from intercept.
3. Port's profile is excellent.
4. GWS' offence 🔥
Offensive/Defensive Efficiency R17 update (with "ghosts" from 4 games ago). What stands out?
1. Geelong continue their reign as
#1
2. Essendon's recent wins off the back of offence
3. St Kilda's dramatic defensive 🛝
4. Sydney look better than their W/L record
Watching
#Tokyo2020
I've heard commentators say "If only they'd swum as fast as they did in the heats/semi's" a few times. Interesting to see how often individuals don't manage that. Unintuitively, experience doesn't seem to help much either.
#Swimming
This is what Melbourne's transition from intercepts in the defensive half have looked like during the 2021 AFL Finals. I'd eventually like to cluster them across the season so I'm open to any suggestions on clustering algorithms for paths.
The 2 goal shot that
@NetballAust
has introduced provides a completely dominant strategy. Shots from this area will be worth an expected 1.43 points under the new rule, while shorter distance shots are worth an expected 0.87 points, based on shot data from the last 3 seasons.
Port had 3 periods of repeat entries where they had 5 or more inside 50s in a row to 1 period for Collingwood. Scores from these periods were 7.4 to 0.1
#AFLPiesPort
Ok, I was excited about this before but having quickly checked out the data I'm bloody happy. The
#fitzRoy
R package made accessing "box score"
#AFL
data simple. My play by play data set added some contextual info. This adds so much more.
It's taken six of them I've finally finished a lockdown project - a little website where I'll post various data stuff.
The first is something I think many will like: **R functions for scraping match chains from the AFL app** + a little kick-in analysis.
#AFL
Post Round 4 Offensive/Defensive Efficiency & xChainScore wrap. What stands out so far?
1. GC are allowing their oppo to get brilliant starting position.
2. They might not lead on the 🪜 but Melbourne are the team to catch.
3. Dogs offence still 🥶
4. Essendon offence 🔥
Within the 91 game sample provided, Tyreke Hill (KC) had the highest max speed on a Touchdown reception (10.35 y/sec) in Week 3 2017 against LAC
#BigDataBowl
#NFL
Hawthorn's attacking speed put Melbourne's structure under a bit of pressure and they were able to extend their chains by winning contests. Melbourne responded with good offence of their own and typically good stoppage play.
#AFLDeesHawks
The 10 most "in form" players according to
#AFL
Player Ratings after R10 (min. 20 games played). Heeney's on a tear at the moment. Check out full list depth and form analysis at
Just a quick look at players who generate kicks inside 50 but also get targeted on kicks inside 50. There's a group of key forwards and then another group that spend time as midfielders and forwards. And Max Gawn!
Which players have been the best at taking the ball from inside the contest to outside effectively in 2021? Big difference between the top 10, who all tend to handball out vs guys who kick it out more often. Be great to split into hard & loose balls too but they're not available.
A couple of R1
#AFL
observations. 1. Pressure was down compared to previous seasons. 2. UM haven't had a definite trend over the last few years but were above ave. in R1. 3. Other than 2019, turnovers increased have in importance vs clearances, this looks like it might continue.
Western Bulldogs offensive/defensive/xChainScore over the season. Offence needs work, but solid enough defensively and gave themselves very good starting points.
Looking at last night's
#AFLDeesDogs
game, Melbourne were on top in most areas except at stoppage. Even then, when they won the clearance they were excellent at generating field position. The Dogs were beaten badly in the air at both ends of the ground.
Melbourne offensive/defensive/xChainScore over the season. The risk would be throwing the baby out with the bath water if Melbourne were to try to improve their offensive profile. Figuring out how to increase their connection on kicks I50 might be enough.
Resurrecting this plot type again. The Dogs made the most of their chances in the 2nd Q to get back into it and then controlled most of the early going in the 3rd. Interesting that both of Melbourne's 3 goal runs in the 3rd originally started from the Dogs F50.
#AFLDeesDogs
#AFL
Post Round 6 Offensive/Defensive Efficiency & xChainScore wrap. What stands out so far?
1. It's a race to the bottom defensively for WC, North & Haw.
2. St Kilda's defence is a wall.
3. Coll, WB & now Geelong are generating ✅ field position.
4. Adel leading efficiency atm.
#AFL
Post Round 10 Offensive/Defensive Efficiency & xChainScore wrap. What stands out so far?
1. Field position is dominating the top of the ladder.
2. Essendon's offence 🔥, field position 🤢
3. Carlton's offence is in the 🚽
4. Can anyone reach the heights of previous premiers?
One of the reasons for AFL teams slowing down their Attack Speed from defence is that their defensive structure can get out of position and they get hurt on TO. Geelong, Collingwood, Carlton & St Kilda seem to be the teams this is true for. Others should consider⬆️ Attack Speeds.
An update of a plot I posted earlier in the season. From kicks where the next possession is contested, who is winning those contests on offence and defence? Melbourne and Geelong are the only teams in the top 8 on both sides of the ball.
In lieu of a playing style preview as there's no data to preview for 2024, a look at how the game went last time
#AFLSwansDees
played. Neither team kicked well in front of goal but Melbourne controlled field position, especially on the wings, flanks and pockets forward of centre.
#BBL10
starts today so I built a couple of models to try to analyse player performance. Data was collected from for the last 6 BBL seasons. Logistic regression was used to predict the probability of the batting team winning after each ball. 1/n
Quite possibly trying to show too much info here and it gets a bit messy, but a look at Melbourne's average disposal location for each player in the 2021 GF, combined with passing links (3+ only). The curve originates from the disposal player and straightens at the receiver.
With access to a key dataset being shut off 😢 the stats used in these have reverted to only those found in the excellent
#fitzRoy
package. Still attempting to describe style in the key areas of contest, defence and offence.
#AFLPiesDogs
up first.
Form difference between L50 and L10 games from AFL Player Ratings for top and bottom 3 for tonight's selected teams (min. 50 games played). Thanks to
@fryzigg
for providing such a great dataset
#AFLSwansDogs
Average team age and games played are such poor (and lazy) proxies for potential performance in the AFL. Compare the Rd 2 teams to Grand Finalists of the last decade, how many clubs are outside the “Premiership window”?
#falsepositiveseverywhere
Form difference between L50 and L10 games from AFL Player Ratings for top and bottom 3 for tonight's selected teams (min. 50 games played). I should’ve mentioned that the Player Ratings for this season have been adjusted to reflect the shorter game time.
#AFLGiantsPies
The ladder according to
@championdata
Expected Scores. A few differences compared to the actual ladder. Essendon drop out of the top 8 (but have a game in hand), replaced by the Dogs. Adelaide also have a win on the board.
#AFL
Post Round 2 Offensive/Defensive Efficiency & xChainScore wrap. What stands out so far?
1. Melb, Syd ✅for offence & defence
2. Haw defence in the 🚽
3. Big spread in team differential from stoppage. That will decrease over time.
4. Rich, Coll, Freo ✅for starting positions
@onepercentas
@crow_data_sci
@EmlynBreese
@CapitalCityCody
Bit of a visual for this. Clearly some teams look to come back inside more than others but I don't think any of them particularly try to jam the ball in there. Hoping I got the data cleaning right.
A few visuals on how teams moved the ball by foot following an Uncontested Mark in 2022. The corridor areas tend to highlight how aggressive/risk averse teams are in these situations. From the back pockets teams go wider again, hugging the boundary most times.