Matthias Kullowatz

Data scientist @datarobot. Co-founder and @analysisevolved. @mattyanselmo.bsky.social #rstats he/him

[Matthias Kullowatz @MattyAnselmo]

Continued some research on measuring how well individual soccer players play defense. Last time it was in open-play phases, and this time I tried to allocate defensive value on set piece plays:

https://t.co/hqS4JC5ePq

[American Soccer Analysis @AnalysisEvolved]

Excited to announce our partnership with @MNUFC

https://t.co/pK8xrnM0xs

[NWSLPA @nwsl_players]

Today, the Players are proud to announce the creation of the Om Arvind Gift in honor of the life & legacy of soccer journalist and women’s soccer advocate Om Arvind.
Read more at ⬇️
https://t.co/y3P8YkERmA

[Matthias Kullowatz @MattyAnselmo]

If you are actually looking for metrics to complement xGD in team predictions, consider other flavors of EPV metrics. Something like the average of the maximum EPV in each possession, or the average starting EPV of each possession.

[Tiotal Football @TiotalFootball]

Mostly I want to make the point that xG and EPV type metrics are doing the same thing/ using a similar framework (goal probability) just at a different level of accounting. It’s why that 2008 @soccermetrics post imagines “an expected goal value” broadly defined (not just shots)

[Matthias Kullowatz @MattyAnselmo]

I was able to explain nearly 90% of the difference between xGD and g+Diff (i.e. R2 = 0.9) with these two things:
1. How much passing value is lost to imaginary receivers on incomplete passes.
2. That g+ is built on a two-possession outlook, rather than just current possession.

[Tiotal Football @TiotalFootball]

hey @AnalysisEvolved, remind me what would be the major sources of difference above or below this trend line for team g+ diff vs xGD?

Each dribble/pass turnover hands EPV over to the other team via interrupting g+ right? Is it just the mix of dead ball vs open play turnovers?

[Matthias Kullowatz @MattyAnselmo]

@joeclowery Personally I would have left the Sounders off the list altogether

[Matthias Kullowatz @MattyAnselmo]

@ProfessorSides @isureppin12 @SethWalder @minakimes I had the same estimate, FWIW.

[Matthias Kullowatz @MattyAnselmo]

@Stat_Ron @DW__Analytics @jaseziv @AnalysisEvolved You can get a lot of ASA stuff with the itscalledsoccer packages!
Python: https://t.co/PrU0VwMrhV
R: https://t.co/LFKAcopxLV

[Matthias Kullowatz @MattyAnselmo]

@VictorPontes070 @STLCinderace @AnalysisEvolved g+ has been calculated before on European leagues, though it is not published. Between about 2015-2022, I believe Messi had 5 of the top 10 seasons. Ronaldo had one or two seasons in the top 10 over that time period.

[American Soccer Analysis @AnalysisEvolved]

It's that time of the year! Here is ASA's NWSL g+ Best XI! Read more about how and why g+ loves these players from @KierDoyle.

https://t.co/c6p6gajcVB

[American Soccer Analysis @AnalysisEvolved]

A new year means a new playoff format . With a bigger emphasis on penalties this season, knowing which clubs have the best takers could be more important than ever. @etmckinley has you covered.

https://t.co/r1JDPkdZTK

[Matthias Kullowatz @MattyAnselmo]

@STLCinderace @AnalysisEvolved Here's what model sees:
* STL lineup has players with historically bad g+
* As a team, STL has produced below-average g+ and xG diffs

Check out numbers here: https://t.co/rosIt5biaC

[Matthias Kullowatz @MattyAnselmo]

@ShipTheJustice @AnalysisEvolved I like your rhyming

[Matthias Kullowatz @MattyAnselmo]

@colinalber10 @AnalysisEvolved Note that Western Conf teams are extremely likely to go on the road for the final, based on total points, and all their conditional cup final probabilities are thus <50% because of the huge home-field advantage we've observed in MLS for years. (e.g. STL is 4/13 to win cup final).

[Matthias Kullowatz @MattyAnselmo]

@HellbenderSTL @AnalysisEvolved 2 is definitely arbitrary, just trying to automatically capture likely lineups. To clarify, the player g+ itself averages the player's entire history in our database, with more weight to recent games. 2-game cutoff is just for determining likely lineup.

[American Soccer Analysis @AnalysisEvolved]

Decision Day is nearly here!

@MattyAnselmo simulated MLS playoff outcome probabilities based on the new playoff format & the strength of teams, as measured by both team-level metrics and player-level metrics according to the best combined lineup of each team's last two games.

[American Soccer Analysis @AnalysisEvolved]

Tomorrow is the big day. Before the World Cup starts, @EvanDavisSports wrote about the the best NWSL players to never make a World Cup roster.
https://t.co/Aq2AJAZzuu

[Naveen Elliott @NaveenElliott]

And here are the five worst teams in the final third, according to goals added from @AnalysisEvolved. Atlanta United is the surprise team for me here! #InterMiamiCF #NYCFC #TFCLive #ForTheCrown #RCTID #HoldItDown #CFMTL #cf97 #wearethea #DTID

[Backheeled @Backheeled]

𝑻𝒉𝒂𝒏𝒌 𝒚𝒐𝒖!

All of our subscription revenue from June is headed to TOPSoccer, a program for children with disabilities!

We raised enough money to help purchase new equipment for local TOPSoccer clubs.

Read more ⤵️
https://t.co/tbuMmsy3Je

[Matthias Kullowatz @MattyAnselmo]

@AnalysisEvolved @bbmikej Again, roughly, I think the 1.5 goal adjustment covers that.

[Matthias Kullowatz @MattyAnselmo]

@AnalysisEvolved @bbmikej Back of the envelope: typical MLS club scores like 1.5 goals per game. In predictive models, 60% of a team’s g+ over average is stable to future games. So if that 3.56 is like 2g+ over average, then this team would be predicted to score about 1.5+0.6x2=2.7 goals per game.