The numbers say that the Premier League (not the Russian one) is the most difficult league to play in. (See y-axis.) Numbers never lie, so of course this is correct.
Please harass
@canzhiye
if you disagree.
João Cancelo has committed the most "rage" fouls in the Premier League so far this season (tied with Emmanuel Dennis at 4). Does he get salty when he isn't saucy? 🤔🤨 🤡
what even is an "expected point"? i wrote blog posts showing how to:
1⃣ compute expected points (xPts) from expected goals (xG) without simulation
2⃣ compare xG sources using {worldfootballR}
3⃣ quantify how unlikely team placings were (🚨 2020/21 Brighton trigger warning 🚨)
✍️New blog post on "meta-metrics" in soccer, replicating part of a 2017 JQAS article from Franks, D'Amour, Cervone, and Bornn.
⚖️Does a metric reliably differentiate between players?
📈Does a metric measure a quantity that is stable over time?
(see reply for link)
I scraped all of
@xGPhilosophy
's end-of-match xG tweets and quantified hwo much Brighton's habit of "winning the xG" but losing the game has driven Twitter engagement.
Formation matchups are a common point of discussion in soccer. Voronoi tessellations indicate that a 4-3-3 with a false 9 generally is the "best" against other structures.
Data (and idea):
@SamGoldberg1882
CC:
@accidental__aRt
#rstats
Fine, since no one has been brave enough to do it, I will. I'm calling out Opta for stat-padding for attackers, giving them more credit than StatsBomb for aerial duel success.
Expected goal difference by game state in the EPL this season. What stands out?
1⃣ Newcastle have been most dominant when leading.
2⃣ Man City have been at their best when tied or trailing.
3⃣ Spurs have been mediocre when leading, but otherwise very good.
#rstats
Want to see an example of a start-to-end analysis? I wrote a 5-part series of posts breaking down academic competition results in Texas, starting here:
Pitch control is cool, but how much does it matter which model we use? I compared
@JaviOnData
et al.'s 2018 model,
@the_spearman
's 2017 model, and a naïve "nearest" model (only using player position) to try to find out.
#rstats
1/n
We say "player x is outscoring their xG" all the time, but how often is it actually significant over-performance?
Not that often, although Kane and Lukaku are coming in to today's matchup over-performing at significant rates. (Kane is benefiting some from recent UEFA matches.)
#rstats
peeps -- I analyzed my google search history using tidy text principles. See my post at . Thanks so much to
@juliasilge
and
@drob
for the awesome
#tidytext
package and the book.
New
#rstats
post✍️ with soccer ⚽️ data
- Translated
@devinpleuler
's python tutorial on decomposition to R to learn a new-ish concept hands-on
- Integrated "non-tidy" functions into a tidy workflow
- Made some charts along the way
To add to Jon's thread, this chart also shows % of time in each game state.
West Ham are good at getting a lead and holding on to it (4th most time leading), so a negative xGD when ahead may simply reflect a more passive play style when leading.
West Ham are 6th in the Premier League table but putting up relegation-worthy underlying numbers to get there. Here's a little thread to explain what's going on:
We can use a similar approach, swapping out xG for VAEP, to learn something more specific. For example, a young FW moving from the Bundesliga to the EPL *cough* Haaland *cough* might expect their xG/90 rate to drop by 0.07 (22% relative to the median).
I compared the club rankings from 538 and The Analyst (Opta) and found some interesting differences.
1. 538 seems to think more highly of the top Brazilian league.
2. Opta seems to rate top teams from non-Big 5 leagues more favorably.
1/3
La Liga and EPL titans face off today. Given how each league's referees call the game differently (EPL refs seem to allow more pHySiCaLiTy), one side is bound to be frustrated by the end of the match.
#UCL
#Chelsea
#RealMadrid
many of the top attacking players get about half of their on-ball value from shots, although it varies. some get more value out of purely passes and pass receptions, e.g. Bruno
I scraped over 10k ⚽️ matches and found that
1. Italy's Serie B and other Tier 2 leagues make more "bad" fouls resulting in yellow and red cards
2. The US's Tier 2 league, the USL, is most likely to have cards described as "violent"
Tier 2 leagues play harder, not smarter?
✍️ New blog post: Calibrating model predictions with the {probably} R package (link in reply)
⚽️ Case study: women's football match forecasts (RIP 538)
yes, the home keepers behavior certainly changes based on game state.
didn't see much of a difference for away keepers, or more generally, when not accounting for home/away status. with throw-ins, the time wasting was also more profound for the home team
Cooked up a new blog post on ⚽️ players underperforming xG. Contents include:
🔴Unnecessary LaTeX usage
🔴Double-digit footnotes
🔴An appendix no one asked for
🟢Sick plots
🟡Questionable insights
been a while since we've had some
#EPL
matches. recent league form based on in-game xG difference:
📈 Arsenal, Brighton, Aston Villa
📉 Leicester, Everton, Chelsea
caveat: xG difference is influenced by the actual score (e.g. playing more conservatively when having the lead)
Does Major League "Golazos Only" Soccer really have a higher number of banger goals than other leagues? Yes, but only on direct free kick shots.
#DecisionDay
Every need to read a long, arduous report and want to identify just the most important parts? You can use
#rstats
(and
#text2vec
) for that! Check out my write-up here: .
i had some requests to describe the methodology in more details, so i put together a blog post:
@thecomeonman
had an interesting (better) take on the same topic:
some are comparing our collab to xavi and iniesta 😉
The numbers say that the Premier League (not the Russian one) is the most difficult league to play in. (See y-axis.) Numbers never lie, so of course this is correct.
Please harass
@canzhiye
if you disagree.
tried to come up with an
#absolutelysick
stat based on ⚽️ pass networks. found that it's no more descriptive than just counting passes.
i wrote about it anyways. 👁️👄👁️
🔗
20 -
@OptaJoe
's most common stat-line finisher is "opener", used in 20 separate tweets over the past few months. Overall, the account's most common one-word summaries describe in-game events, player achievements, and team activities. Interesting.
2 - Harry Wilson has scored in consecutive Wales appearances for the first time in his career, while Gareth Bale, who provided the assist, has now assisted in each of his last three games for his country. Opener.
✍️ ⚽️Are all 20 yard passes judged equally by possession value (PV) models? Can you still get positive PV from incomplete passes?
Answers to these questions and more in my new blog post. (L I N K I N R E P L Y.)
Season-ending xGD by game state for the EPL
Some non-top 4 observations:
🔴Brentford were average at worst in all game states
🔵 Everton did best when drawing
⚪️ Nottingham Fores did best when trailing
🟣West Ham and Man U were outperformed in all game states
@johnspacemuller
last chart, mostly for my own sanity... the difference between being ahead and behind becomes slightly more exaggerated for the home team as the match winds down
Today in
@SciReports
, the graph imputer, a new model that can predict where players will run, on and off-screen, using past and future available observations of subsets of players:
#AISports
1/
#CONCACAF
has developed such a strong reputation for gamesmanship and aggressive play that it's become a verb. ("The
#USMNT
got CONCACAF'd and still won.")
But Brazil and the rest of the CONMEBOL have truly set the precedent for chaos.
quick maths show that border town Whataburgers are the "best". (are out-of-state visitors jealous?!?)
what i did:
1⃣ scraped yelp
2⃣ emperical bayes adjustment for low review counts (h/t
@drob
)
3⃣ matched with McDonalds locations to account for regional confounding
4⃣ regression
back again at 6 PM ET after a week off
agenda:
1⃣ do nerd stuff to find where the best Whataburgers are in Texas
2⃣ discuss my
#SLICED
finals hot takes
3⃣ ???
4⃣ profit
How exciting have this year's
#NBA
playoffs been? So far, it's been the 3rd most exciting since 97-98.
Last night's
#FearTheDeer
win over Miami was the 2nd most exciting game this playoffs.
Data:
@inpredict
#rstats
ideally we would have "ball-in-play" time to quantify pacing, but pass volume seems like an ok heuristic.
most extreme cases:
Italy 809 - 372 Albania
Ecuador 275 - 332 Jamaica
is this due to just play styles? or maybe physicality of the matches and officiating?
The best XI is chosen based on VAEP, which is basically a framework for quantifying how every on-ball player action influences the game's outcome.
(Admittedly this doesn't work as well for keepers, although Pope and Martinez ranked 2nd and 3rd, so maybe it's not that bad.)
2/7
Blue dress or yellow dress? Yanny or laurel? Now, for
#dataviz
, the question is bubble grid or choropleth? New
#rstats
post: . Thanks to
@rettungstweet
for the inspiration.
Death, taxes, and diminishing returns are the 3 pillars of life. With ⚽️, the 3rd principle is evident in the trade-off between non-penalty xG per shot (quality) and shots per 90 (quantity).
All kidding aside, it's sort of remarkable how we can use a PV metric (e.g. atomic VAEP) to infer league power rankings that don't seem that off.
Perhaps the coolest thing about all of this—no match results were used 🤯. (We use before-and-after transfer numbers.)
Premier League +
#EURO2020
analysis that no one asked for:
📈 Aside from 🏴, 🇵🇹 and 🏴 have the most Premier League players among Euros lineups
📉 4 countries (🇷🇺, 🇫🇮 , 🇭🇺, 🇦🇹) don't have any representation
1⃣ 3 countries (🇭🇷, 🇨🇿, 🇲🇰) have 1 Premier League player
With under two weeks left before the NBA starts up again, here's a look at projected playoff seedings for the Western Conference. (Eastern Conference is mostly locked in🥱.)
Data from
@JacobEGoldstein
. Inspiration from
@experimental361
.
#rstats
If we used un-adjusted xG/90 rates for players moving from league to league, we would have a table that's more sparse and noisy. Outliers (like the large increase in xG/90 for attackers moving from Brazil's Serie A to Turkey's Süper Lig) show the effect of sampling bias.
Despite missing several matches, Jack Grealish still leads the EPL in fouls drawn (100) by a wide margin. (Second is Sadio Mané with 67). He's taken down most in the left attacking half.
Official Football (Analytics) Twitter Power Rankings 2023™️ is in now.
Can you find yourself here (click to zoom)?
If not, try using the interactive version here:
If
#MUFC
fans (👀
@utdarena
) are upset about the red card for Martial last weekend, just know that the referee for the match (Anthony Taylor) is not known for handing out red cards. That honor belongs to good ol' Mike Dean.
h/t
@FootballSlices
for the colors
Scoring goals⚽️ in the Champions League
#UCL
is harder due to stronger competition, but when you adjust for competition, most attackers actually have better goal scoring rates.
Data courtesy of
@21stClub
+ h/t
@rwohan
and his weekly newsletter.
#rstats
Not only did they not block long-range shots at a historically low clip, they were among the best in limiting such shots.
This has been going on for the past 3 seasons.
We can plot (transfer)market values and VAEP to identify players who may be over- and under-valued. It seems that Kane, Salah, and TAA are still under-valued, despite everyone knowing how good they can be.
4/7
To please xGod (
@rwohan
), we can compare VAEP with DAVIES (another "all-in-one" metric).
There's some correlation, but there are some notable differences and oddities. DAVIES loves Werner and Mané this year (lol), and VAEP thinks Westwood is the English Iniesta (also lol).
5/7
As Messi preps for the World Cup next week, I'm sure he's tuning into the USL Championship final tonight. San Antonio "Sickos" FC will play a physical, counter-attacking style against Louisville City
Brazil v. Serbia on Thursday was the biggest height mismatch of the World Cup Group stages. Will other vertically-challenged kings be able overcome their shortcomings? (see Messi, Lionel)
Here's a visual example for a possession in which Kane is assigned a net negative VAEP. Note that this is not xG—actions can have negative values (especially missed shots)!
YouTube link:
3/7