Dienstag, 21. Oktober 2014

A new look at football players

Abstract: We found a strong Relative Age Effect in youth football. Yet many of the players that benefited from their relative age seem to fail to move to senior football because they lack their advantage among grown-ups. Football club could increase the efficiency of their youth programs substantially, by selecting player according to their talent rather than relative age. The minimum Goalimpact a player needs to achieve to have a realistic chance to become professional football player is 80. Yet older players often stay in business slightly below that age to.

Since the introduction of the football aging curve, the Goalimpact is a direct function of age. Therefore, it should come as a surprise if we see a clear relationship between Goalimpact and age in a xy-plot. However, there are still some noteworthy feature to report.

Each dot represents a non-goalkeeping football player. The x-axis is the players
age at the 1st of July 2014 and the y-axis his Goalimpact at the same time.
To make it easier to describe some of the findings, we highlighted some areas in the chart. The underlying picture is the same. Notice that we restricted the y-axis to 20 to 160. There are only few players outside that range and including them would make the rest less visible.
The top orange region contains very few players. Player in this area are needed to
compete for the UEFA Champions League and hence ambitious teams will need
to pay scarcity prices. The blue area contains youth players. These exhibit a strong
Relative Age effect. There are nearly no players in the green area. These play
in uncovered leagues. In the red area players gradually drop out the covered
leagues as they get too old.
As the aging curve, the Goalimpact distribution peaks around the age of 26. Most players are in a narrow band of ca. 20 Goalimpact points, independent of the age. There are more outliers above the band as there are below. We assume, this results from a selection bias. the best players will always play in a covered league and hence are included. In contrast, senior players below a certain quality do not play in covered leagues. We marked the area green which contains very few players, because presumably players in that range below a Goalimpact of 80 didn't make it to become a professional football player, at least not in a covered league. Some youth players would be in that range if they'd follow the aging curve, so apparently a selection takes place in the transition between youth and senior football - which makes entirely sense. This should enable us to create a probability score that a youth players will eventually play senior, but we leave this to another post.

Among the youth players (blue area) we see three thick vertical lines. These are caused by the Relative Age Effect (RAE). Football teams favor players born early in the year, because these are relative old compared to their team mates and opponents. This makes them have a physical advantage. As this advantage will disappear once the players are fully grown, it is a short sighted and ineffective way of player development. Rather than focusing on developing the most talented players, teams seem to focus on winning the current youth competition and play predominantly players that incidentally were born in the early months of the year. They do so, despite the fact that, according to Goalimpact, the players in the later months often are more talented. We didn't investigate it in detail, but from the chart it looks like that it were mostly the youth players that drove with the RAE ticket that didn't make it to the senior football. It makes intuitively sense as their physical play no longer works in adult football.
The Relative Age Effect is strongly visible until the age of 19. Thereafter, however,
it is much less pronounced. We conclude, that many players that were favored by
their relative age in the youth league, don't make it to senior football.
But given that many of the players born in January to June drop out at a later stage, much resources are wasted on players with a bad prognosis to make it to senior football anyway. We therefore propose to select the players by prognosis rather than current physical development, because this will improve the resource allocation. One way of doing so would be selecting players by PeakGI.

Most of the players in the database will never be good enough to really compete for the UEFA Champions League (UCL), A Bundesliga club aiming for UCL qualification will need an average Goalimpact of 110+ in the team. But to pass the round of last 16, this will not be good enough. The top teams will aim for players that are at least 120+. The area with those players is marked orange. Only few players are in that range and hence the clubs can expect to pay scarcity prices. This is especially true for players that are expected to stay in that range for many years, either because they are young (Thomas Müller) or because they are so extremely good that they are expected to stay above the limit for many years to come despite aging (Xabi Alonso).

Even the best players eventually decline in performance due to age. There is no fix limited when a player will stop playing. Besides the playing strength this is influenced by many other factors such as his health, his wish to go on playing, and the need of his current club. What is notable, though, is that players often drop out at a lower Goalimpact level than the marked at the green area. We assume that this is caused by different factors. Partly, this due to a bias in the aging curve. The aging curve treats all non-goalkeepers equal. In many tactical setups, however, defenders are less prone to age-related decline in performance. E.g., if a team doesn't play a high line, the reduction in speed will not influence the quality of a center back as much as striker that needs speed to counter. Hence, some defenders may actually be better than indicated by their Goalimpact. We may distinguish the aging curve for non-goalkeepers into tactical positions at a later stage, but it is not as straight forward as it seems. Some players, like Kevin Großkreutz, play many positions and it is unclear which aging curve to use for them.

Another reason why players linger around with a Goalimpact below the value that teams would accept for young players may be that older player can be still every effective but can't take 90 minutes at full strength anymore. They then might be still be very useful as substitutions or backups (e.g. Claudio Pizarro).

To summarize, a scatter plot of player age vs Goalimpact reveals that players need to have a Goalimpact above 80 to be professional football player. If they are pro already, they may stay in business a bit longer below that level. Below a level of 70, however, many drop out. Even today, youth football is driven strongly by the Relative Age Effect. The data suggests that many players that benefited from their relative old age do not manage to take the step to senior football where this advantage is gone. Football clubs could improve the efficiency of their youth programs substantially by selecting their players according to talent rather than relative age even though it may cause their youth teams to produce less victories.

Sonntag, 19. Oktober 2014

Which German Goalkeepers should Löw pick?

The German national coach for goalkeepers, Andreas Köpke, recently named the goalkeepers that he has on his list for the national team and, hence, which are competing for participation in the European Championship 2016 (should Germany qualify). Besides the current number one Manuel Neuer and his senior backup Roman Weidenfeller, these are the known contenders Barca's new signing Marc-Andre ter Stegen, and Leverkusen's Bernd Leno. Less known are Schalke's Ralf Fährmann, Kevin Trapp (Frankfurt), Timo Horn (Köln) and Oliver Baumann (Hoffenheim).

We take the opportunity and check if this is the best choice according to Goalimpact, too. If you want to know how our rating works on goalkeepers, we wrote the concept down here. In short, we evaluate the average goal difference in games with the keeper as compared to games without and correct this result for many many factors such as team mate and opponent strength, home field advantage and others. Wait, you will say. My team always plays with the same keeper. There is no such thing as a "game without". In order to circumvent this, there are many precautions taken, most of all we take the information of all games of a players career and not only one season.

If we plot the current Goalimpact on the x-axis and the expected career peak value "PeakGI" on the y-axis, we obtain a chart with all currently excellent keepers on the right-hand side and all very talented keepers on the upper part.

All German professional goalkeepers. Keepers on the short list for the national
team are marked orange. Most of them are either very good and have a Goalimpact
of 140+ or are a prospect of having one later in their career.

As we can see, Manuel Neuer is in the top right corner, basically playing in his own league. Roman Weidenfeller is the second best keeper but already considerably behind Neuer's Goalimpact. Moreover, he is already at his peak and hence is unlikely to improve from here on. He himself declared the European Championship as a "welcome end" to his career in the national team.

The chart enables us to rate the contenders in comparison to the former two. They are all significantly weaker in their current performance, but this is due to their young age. If we compare the PeakGIs, we see that all of them but Ralf Fährmann are in the range spanned by Weidenfeller and Neuer. Trapp and Baumann are very good keepers that are on their way to become as successful as Weidenfeller, but that may not be enough to secure a place in the future national team. The reason is that the other contenders, Leno, ter Stegen and Horn, are more talented and are expected to overtake Trapp and Baumann in Goalimpact eventually. Ralf Fährmann is currently not expected to peak above 140 and hence high enough to join the national team.

All in all, Goalimpact would have selected a very similar set of goalkeepers as a mix of talents and experienced players. Only exception is that we would prefer Jens Grahl over Ralf Fährmann. And then there are some more talents out there that didn't make it to a major league yet and were probably not considered by Köppke and Löw for that reason. However, given their talent they are likely to move to higher leagues soon.

Beside the named players, there are some other even younger goalkeepers leaving the youth development centers that are expected to peak in the relevant range of 140 to 160. Germany, you will keep your tradition of good goalkeeping for at least one more player generation.

Montag, 13. Oktober 2014

Iceland vs Netherlands 2:0

Because of the upset caused by the victory of the apparent underdog Iceland, here the Goalimpact values of all players. Iceland has some decent players, but the Netherlands should have won this. Yet, given the home field advantage the odds based on the starting XI were

Iceland: 32.6% (Home), draw: 30.8%, The Netherlands: 36.6%

Accounting for the fact that the Dutch bench was also much better, substitutions should make the Netherlands an even clearer favorite.

Iceland

PlayerGoalimpactPeak GIAgeClubNo. GamesNo. Minutes
Ari Skúlason96.1100.127.4Odense BK22119185
Ragnar Sigurðsson117.4123.328.3FK Krasnodar24922757
Jón Böðvarsson90.7101.622.3Viking FK1037680
Kári Árnason89.8114.032.0Rotherham United27123280
Birkir Bjarnason88.189.126.3Pescara Calcio21114514
Gylfi Sigurðsson109.6111.825.1Swansea City22915322
Hannes Halldórsson105.0106.430.4Sandnes Ulf14213112
Teddy Bjarnason97.6102.127.6Randers FC17914701
Kolbeinn Sigþórsson110.6113.924.6AFC Ajax1539701
Emil Hallfreðsson85.895.530.3Hellas Verona21116266
Aron Gunnarsson102.8104.125.4Cardiff City29823885
Average99.4105.627.2
Bench
Ólafur Skúlason80.8101.131.5SV Zulte-Waregem18713103
Þórarinn Valdimarsson95.599.124.4AB Vestmannaeyja1159144
Gunnleifur Gunnleifsson95.4106.539.3Breidablik13712599
Ingvar Jónsson106.2128.824.9UMF Stjarnan878118
Rúrik Gíslason95.697.326.6FC Köbenhavn22614872
Helgi Daníelsson82.0112.533.3Os Belenenses21717993
Sölvi Ottesen93.0105.830.6FK Ural16113773
Birkir Sævarsson90.998.429.9SK Brann19716653
Viðar Kjartansson100.8104.224.6Valerenga IF806468
Hallgrímur Jónasson88.894.828.4SönderjyskE13912294
Alfreð Finnbogason111.5112.325.7Real Sociedad16512591
Average94.6105.529.0


The Netherlands

PlayerGoalimpactPeak GIAgeClubNo. GamesNo. Minutes
Nigel de Jong136.0143.429.8AC Milan47037378
Stefan de Vrij123.8133.622.7Lazio Roma17214734
Wesley Sneijder139.7149.930.3Galatasaray47737374
Arjen Robben153.2166.630.7Bayern München50437244
Daley Blind135.1138.524.6Manchester United18315272
Jasper Cillessen115.7134.125.4AFC Ajax1029224
Jeremain Lens138.3140.726.8Dinamo Kiev29820553
Ibrahim Afellay109.8115.928.5Olympiakos Piräus28820400
Bruno Martins Indi117.7127.422.7FC Porto14812718
Robin van Persie131.5148.831.2Manchester United49134717
Gregory van der Wiel151.1152.926.7Paris Saint-Germain27123491
Average132.0141.127.2
Bench
Virgil van Dijk104.8112.323.3Celtic FC12110515
Davy Klaassen114.1128.721.6AFC Ajax755797
Kenneth Vermeer121.4124.728.8Feyenoord18316857
Luciano Narsingh120.2124.724.1PSV Eindhoven1349121
Jeroen Zoet101.4134.423.8PSV Eindhoven13612563
Jeffrey Bruma105.3114.222.9PSV Eindhoven13611257
Leroy Fer102.1105.024.8Queens Park Rangers25119922
Quincy Promes116.7126.122.8Spartak Moskva897737
Jordy Clasie117.9125.323.3Feyenoord17214282
Paul Verhaegh74.791.431.1FC Augsburg32929653
Klaas-Jan Huntelaar142.1159.331.2FC Schalke 0444835867
Joël Veltman116.1125.722.8AFC Ajax614818
Average111.4122.725.0



Freitag, 10. Oktober 2014

Turkey vs Czech Republic: Goalimpact of Lineups

Turkey hosts Czech Republic in an European Championship qualifier. The great times of both teams seem a bit over lately, but the Czech silently improved a bit again. They now have a better team than Turkey, especially if we also account for the bench.

Just based on the starting XI the odds are
Turkey: 44.6%
Draw: 29.8%
Czech Republic: 25.6%

Turkey being a slight favorite due to the home field advantage. A draw shouldn't come at a surprise either.

Turkey

PlayerGoalimpactPeak GIAgeClubNo. GamesNo. Minutes
Selçuk Inan115.1122.429.7Galatasaray31828062
Arda Turan129.5134.227.7Atletico Madrid36127703
Olcay Sahan110.3114.127.3Besiktas25920182
Ozan Tufan90.2119.819.5Bursaspor231727
Gökhan Töre104.0113.622.7Besiktas805201
Mehmet Topal114.4120.628.6Fenerbahce26820779
Tolga Zengin105.1105.931.0Besiktas16915383
Semih Kaya112.7118.823.6Galatasaray12311053
Gökhan Gönül118.9126.329.8Fenerbahce28124977
Umut Bulut94.9115.531.6Galatasaray38328143
Caner Erkin99.999.926.0Fenerbahce25915937
Average108.6117.427.0
Bench
Volkan Babacan99.8112.226.2Istanbul Basaksehir FK978749
Bilal Kisa76.394.731.3Akhisar Belediyespor22113509
Ersan Gülüm101.9105.827.4Besiktas685259
Ümit Kurt90.997.723.4Sivasspor807222
Mert Günok91.5108.725.6Fenerbahce262387
Muhammed Demir103.4112.922.8Gaziantepspor663690
Adem Büyük101.3104.427.1Kasimpasa SK633897
Tarik Camdal103.1109.623.5Galatasaray1289911
O?uzhan Özyakup106.2118.422.0Besiktas684516
Hamit Altintop88.8111.831.8Galatasaray43630462
Olcan Adin100.3107.029.0Galatasaray20516272
Ismail Köybasi101.5103.325.3Besiktas13310986
Average97.1107.226.3


Czech Republic

PlayerGoalimpactPeak GIAgeClubNo. GamesNo. Minutes
Vladimír Darida114.5118.824.2SC Freiburg13410387
Michal Kadlec103.7111.129.8Fenerbahce22919781
Pavel Kaderábek113.4123.922.4Sparta Praha977681
Borek Dockal127.3127.326.0Sparta Praha22117763
Ladislav Krejcí116.8128.122.3Sparta Praha1118943
David Limberský117.7133.631.0Viktoria Plzen24020796
Petr Cech184.9185.532.3Chelsea FC64059494
Tomáš Rosický93.7128.034.0Arsenal FC49237009
Lukáš Vácha121.6123.025.4Sparta Praha18716143
Tomáš Sivok94.2110.631.1Besiktas23219942
David Lafata110.8140.433.0Sparta Praha25720510
Average118.1130.028.3
Bench
Tomáš Vaclík118.7136.525.5FC Basel13312387
Václav Pilar102.2102.326.0Viktoria Plzen977710
Matej Vydra106.6117.322.4Watford FC1076116
Tomáš Necid103.3105.325.2PEC Zwolle18210642
Radim ?ezník114.6115.325.7Viktoria Plzen18014835
Ji?í Pavlenka86.5131.922.5Banik Ostrava302714
Daniel Kolár120.3126.828.9Viktoria Plzen23117910
Jaroslav Plašil87.6115.732.8Girondins Bordeaux47235887
Václav Procházka111.6122.630.4Viktoria Plzen20016986
Daniel Pudil99.2105.929.0Watford FC27022247
Marek Suchý119.1120.626.5FC Basel25622724
Josef Šural101.1105.024.3Slovan Liberec14010018
Average105.9117.126.6



Samstag, 27. September 2014

FC Schalke 04 - Borussia Dortmund: Goalimpact of Lineups

Dortmund has a much better starting XI, but the home field advantage gives Schalke a slight edge. Still, the probability of an draw is rather high. This is in contrast to the betting markets that see Dortmund as a 50% favorite.

It maybe is also a function of the much better bench of Dortmund. With Schmelzer, Kagawa and Jojic, they have three additional 110+ players to bring into play. Schalke only has Kaan Ayhan.

Odds based on starting XI: FC Schalke 04: 43.4% (Home), draw: 30.1%, Borussia Dortmund: 26.6%

FC Schalke 04

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Dennis Aogo95.099.827.7Deutschland29923585
Sidney Sam121.5123.226.6Deutschland25017064
Joel Matip113.2121.323.1Kamerun22318325
Klaas-Jan Huntelaar140.9157.731.1Niederlande44235383
Marco Höger113.3115.724.914911695
Atsuto Uchida117.1118.426.4Japan23320211
Roman Neustädter129.8131.326.526421690
Ralf Fährmann101.1117.325.911810921
Maxim Choupo-Moting98.199.425.4Kamerun19511581
Kevin Prince Boateng106.8111.227.5Ghana29421404
Max Meyer107.8143.818.9Deutschland [U19]936314
Average113.1121.725.8
Bench
Chinedu Obasi108.8114.928.3Nigeria16110922
Tranquillo Barnetta109.8116.829.3Schweiz34724404
Marcel Sobottka91.3113.220.3968515
Christian Wetklo98.298.534.718316799
Christian Clemens98.7106.823.1Deutschland [U21]16811712
Christian Fuchs105.6111.828.4Österreich38131156
Kaan Ayhan116.7142.919.8Türkei [U21]927718
Average104.2115.026.3


Borussia Dortmund

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Sven Bender122.0123.425.3Deutschland25517719
?ukasz Piszczek131.0138.029.3Polen27021691
Erik Durm105.1116.222.3Deutschland [U21]1249634
Pierre-Emerick Aubameyang109.9111.725.2Gabun24017009
Matthias Ginter107.2127.620.6Deutschland [U21]13211967
Roman Weidenfeller142.5143.234.141538371
Adrián Ramos107.8114.228.6Kolumbien22117924
Kevin Großkreutz133.9134.226.1Deutschland [U21]31624589
Mats Hummels157.4158.025.7Deutschland30326560
Ciro Immobile105.2108.524.5Italien1439987
Neven Suboti?151.0151.625.8Serbien28626346
Average124.8129.726.1
Bench
Joseph-Claude Gyau97.2109.721.9966261
Shinji Kagawa122.2123.425.5Japan17112794
Sokratis Papastathopoulos106.7107.426.3Griechenland25320876
Marcel Schmelzer141.9143.726.6Deutschland25121777
Mitsuru Maruoka78.0117.418.7171105
Mitchell Langerak109.9125.226.0605432
Miloš Jojic110.7121.222.4Serbien [U21]764965
Average109.5121.123.9


Donnerstag, 18. September 2014

Everton FC vs VfL Wolfsburg: Lineups

The starting XIs are about equal if only a slight advantage for Wolfsburg. But the home field advantage fives Everton the edge,

Everton FC: 48.9%
Draw: 28.9%
VfL Wolfsburg: 22.3%

Everton FC

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Steven Naismith105.7111.428.0Schottland31521461
Romelu Lukaku106.9123.221.3Belgien20313638
Tim Howard149.8150.935.5USA48244950
Seamus Coleman108.9109.126.1Irland17614190
Gareth Barry115.4147.833.5England66057055
James McCarthy92.397.623.8Irland22919746
Kevin Mirallas114.8117.526.9Belgien32820358
Aiden McGeady111.8118.028.4Irland36726695
Leighton Baines112.6120.029.8England45439678
John Stones100.4122.920.3England [U21]644996
Phil Jagielka111.6136.832.0England50345209
Average111.8123.227.8
Bench
Christian Atsu104.3114.022.7Ghana1006620
Leon Osman93.1124.433.3England40732451
Darron Gibson107.4109.926.8Irland1389281
Antolín Alcaraz91.6117.132.1Paraguay20518477
Muhamed Bešic110.3122.722.0Bosnien-Herzegowina12710964
Joel97.3128.024.2413807
Samuel Eto'o128.7160.933.5Kamerun63353374
Average104.7125.327.8


VfL Wolfsburg

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Ivica Olic99.8143.535.0Kroatien35624016
Kevin De Bruyne107.1114.823.2Belgien18614173
Sebastian Jung122.7126.824.2Deutschland [U21]24521984
Diego Benaglio116.4119.431.0Schweiz34932271
Daniel Caligiuri111.3113.126.720013303
Luiz Gustavo137.3140.627.1Brasilien25420097
Naldo115.2139.932.0Brasilien33229967
Maximilian Arnold113.2135.620.3Deutschland [U21]1129335
Ricardo Rodríguez105.5117.722.0Schweiz14812843
Robin Knoche119.6130.822.3Deutschland [U21]16214083
Junior Malanda98.6122.520.0Belgien [U21]816488
Average113.3127.725.8
Bench
Nicklas Bendtner124.8126.626.7Dänemark29618325
Josuha Guilavogui109.1113.823.9Frankreich1339784
Timm Klose107.0107.926.3Schweiz14112040
Aaron Hunt98.0103.828.0Deutschland [U21]33223463
Max Grün111.4118.127.411911065
Marcel Schäfer99.5109.230.3Deutschland36130493
Mateusz Klich98.0102.024.3Polen1269787
Average106.8111.626.7



Partizan Belgrad vs Tottenham Hotspur

Tottemham has the better starting XI, but due to the home field advantage Partizan is expected to win.

Partizan: 46.9%
Draw: 29.4%
Tottenham Hotspur: 23.8%

Partizan

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Miroslav Vulicevic125.9132.929.319217479
Vojislav Stankovic118.0120.826.9998175
Milan Lukac111.0116.128.91059707
Lazar Cirkovic93.7105.822.0Serbien [U21]504269
Nikola Drincic87.094.630.0Montenegro21018261
Vladimir Volkov116.4122.528.3Montenegro1179461
Danilo Pantic88.1134.617.8Serbien [U19]21774
Branko Ilic96.8118.031.6Slowenien1078684
Petar Grbic102.1102.326.1Montenegro633536
Danko Lazovic112.0130.931.3Serbien34921788
Saša Ilic88.6151.236.7Serbien35024348
Average103.6120.928.1
Bench
Filip Kljaji?86.3118.324.1585334
Andrija Živkovi?101.0146.418.2Serbien [U19]382583
Nemanja Petrovi?105.4116.222.3Serbien [U21]181323
Nikola Ninkovi?105.6132.919.7Serbien [U21]743918
Ismael Fofana102.4102.426.0231381
Predrag Luka105.0105.826.314310397
Saša Markovi?106.1112.623.5Serbien [U21]1197826
Average101.7119.222.9


Tottenham Hotspur

PlayerGoalimpactPeak GIAgeLast National TeamNo. GamesNo. Minutes
Paulinho108.7109.026.1Brasilien17814143
Ben Davies96.9112.921.3Wales978409
Federico Fazio112.6116.927.5Argentinien Olymp.20816927
Kyle Naughton107.1107.525.8England [U21]22118966
Aaron Lennon113.5117.527.4England40729857
Jan Vertonghen143.6147.627.3Belgien35231315
Hugo Lloris124.0130.427.7Frankreich38435541
Harry Kane102.4119.921.1England [U21]1106335
Andros Townsend96.9104.723.2England15810437
Nabil Bentaleb94.5121.119.8Algerien362653
Benjamin Stambouli96.8101.224.1Frankreich [U21]1249186
Average108.8117.224.7
Bench
Étienne Capoue110.7111.126.2Frankreich21318317
Erik Lamela95.1105.422.5Argentinien1309618
Vlad Chiriches116.9119.624.8Rumänien15413386
Christian Eriksen128.5138.622.6Dänemark21216421
Eric Dier104.0124.320.7England [U21]594589
Soldado114.6121.629.3Spanien31322173
Michel Vorm110.1113.230.8Niederlande27024769
Average111.4119.125.3