5050ball

[5050ball]

Tim Howard had a rough year last year. I took players seasons that totaled > 500 minutes, ranked keepers' Save Rate (i.e. saves/min), which reflects the amount of pressure they're under, and Save pct. (i.e. saves/shots on target), which reflects shot stopping. Generally speaking, the bottom left quadrant is a bad season, the top right is a good one; the top left and bottom right are more ambiguous. "Darkness" of the name shows how many minutes they'd played.

[5050ball]

Not really sure why I did this. I don't really have a problem with any of these players so labeling people as "evil" is tough. I guess I went with "evil" meaning "fiery," which is lame and conflated a little bit with "chaotic" but whatever. This is not to be taken seriously.

#MLS #philadelphia union #philadelphia #soccer #alignment

[5050ball]

A year off. I guess I ought to try to do some stuff here occasionally.

[5050ball]

Trying to identify "x factors" of soccer, and found the following surprising relationship: goal rate and offside rate are inversely related. Would have guessed no relationship, or perhaps a slightly positive relationship (attacking teams get caught offside more often).

The inverse relationship is very strong. For every additional offside call per game, a team can be expected to score 1/3 fewer goals. I'm not suggesting a causal relationship, but the strength of this simple relationship is surprising.

#soccer #offside #four factors #analytics #statistics #football

[5050ball]

Turns out that finishing is the curse of the young striker (at least a little bit: _r² = 0.04_). EDIT: Added names.

#soccer #analytics #visualization #dataviz #football

[5050ball]

[5050ball]

[as always, click to enlarge]

On draft day, some people were suggesting that the Union should trade Danny Mwanga. Those people are crazy.

The following graph includes all MLS players under 23 in either 2010 or 2011, who scored more than one goal.

X axis is pct of shots that are on target.

Y axis is pct of shots on target that turn into goals.

Size is proportional to goalscoring rate (goals per 90 minutes).

Yellow is young (17), navy is old (22)

More minutes = bolder (fewer minutes, more transparent)

One could argue from these data that Mwanga is one of the two best young strikers in MLS, along with Teal Bunbury (nods to Juan Agudelo, who only has one season under his belt, and to Brek Shea, whose position is not as clearly defined). All of those players are regularly considered for the senior US squad. Mwanga is right there with them.

Last team Age G90m MINS Accuracy Conversion goals year Adu PHI 22 0.2955665 609 0.2 0.6666667 2 2011 Gonzalez LA 22 0.07142857 2520 0.3076923 0.5 2 2010 Opara SJ 21 0.29540481 914 0.5 0.6 3 2010 Gavin CHV 22 0.19722425 1369 0.3333333 0.6 3 2010 Nazarit CHI 21 0.28846154 624 0.25 0.4 2 2011 Araujo Jr. RSL 22 0.38461538 468 0.5454545 0.3333333 2 2011 Alexander DAL 22 0.17077799 1054 0.4 0.3333333 2 2010 Torres PHI 20 0.29001074 931 0.25 0.5 3 2011 Gil RSL 18 0.13975155 1288 0.3333333 0.2857143 2 2011 Cruz HOU 22 0.15625 1152 0.32 0.25 2 2011 Duka CLB 22 0.12866333 1399 0.3076923 0.25 2 2011 Castillo DAL 19 0.13284133 1355 0.2857143 0.25 2 2011 Mansally NE 21 0.29094828 928 0.3913043 0.3333333 3 2010 Cruz HOU 21 0.13215859 1362 0.3461538 0.2222222 2 2010 Barouch CHI 20 0.28846154 624 0.3214286 0.2222222 2 2011 Juninho LA 21 0.08973081 2006 0.3 0.2222222 2 2010 Nagbe POR 21 0.10902483 1651 0.3333333 0.2 2 2011 Bowen LA 19 0.20501139 878 0.5 0.1818182 2 2010 Sainey Nyassi NE 21 0.11754462 2297 0.2820513 0.2727273 3 2010 Sanna Nyassi SEA 21 0.14504432 1241 0.4137931 0.1666667 2 2010 Najar DC 17 0.2269289 1983 0.3611111 0.3846154 5 2010 Juninho LA 22 0.14084507 2556 0.3333333 0.3076923 4 2011 Mwanga PHI 19 0.4312115 1461 0.5185185 0.5 7 2010 Agudelo NY 19 0.39589443 1364 0.3863636 0.3529412 6 2011 Plata TOR 19 0.10268112 1753 0.3695652 0.1764706 3 2011 Shea DAL 20 0.25041736 1797 0.4871795 0.2631579 5 2010 Mwanga PHI 20 0.29315961 1535 0.5405405 0.25 5 2011 Najar DC 18 0.17842982 2522 0.3508772 0.25 5 2011 Zakuani SEA 22 0.39387309 2285 0.5121951 0.4761905 10 2010 Bunbury KC 21 0.46444954 1744 0.3384615 0.4090909 9 2011 Bunbury KC 20 0.31228314 1441 0.5609756 0.2173913 5 2010 Bruin HOU 22 0.26580035 1693 0.4259259 0.2173913 5 2011 Shea DAL 21 0.37400831 2647 0.3066667 0.4782609 11 2011 Sanna Nyassi COL 22 0.28373266 1586 0.3934426 0.2083333 5 2011

#mls #danny mwanga #strikers #soccer #analytics #stats #statistics #visualization #dataviz

[5050ball]

A follow up to the post from earlier today: added numbers from 2010. To review, average roster age is on the X axis, weighted roster age is on the Y (e.g. the age of a player that plays 2000 minutes is weighted twice as heavily as one who plays 1000 minutes). 2010 is in red; 2011 in blue. Better goal differentials reflected by larger text.

If a team's roster did not change, and the same players played the same number of minutes each season, that team would move "northeast" from 2010 to 2011. Note a few exceptions: Toronto, Chicago, DC, KC and Houston and Columbus all got decidedly younger in 2011. While San Jose's roster got a lot younger, the players they put on the field did not.

Philadelphia has a very young roster, a strategic decision that seems to be paying dividends. New York is using the opposite approach: the Red Bulls' roster is about as old as it can get.

It's not clear to me that goalkeepers should be part of this kind of analysis, but I don't really have a good rationale to exclude them.

[5050ball]

MLS teams' ages, plotted against ages weighted by minutes played (season goal differentials correspond to font size). Teams above the regression line played their older players more; those below the line played their younger players more. NY is an outlier; very old by either measure. Portland, on the other hand, is very young.

Note clump of high-quality (Western conference) teams in the middle; rosters built and playing players in the prime of their career. Note also the clump of Eastern conference teams with younger rosters. The Union are surprisingly above the curve here, despite seeming to play a lot of young players; this is because of Mondragon, Califf and Le Toux.

EDIT: I had accidentally messed up several ages at the end of the alphabetical list. I've updated the graph to reflect the repaired data.

#mls #soccer #football #stats #statistics #analytics

[5050ball]

My beloved Everton are supposedly giving a trial to Edson Buddle. He's 30 and his career has been very up and down. Want to take a flier on a young MLS striker? Check this out. All MLS forwards under 26 with more than 1000 minutes last season are included. Their shot accuracy is on the X axis, the pct. of their accurate shots that became goals on the Y axis. Their (open-play) goals per 90 minute average is represented by the size of their name; their age by color (younger in blue; older in bright green).

EDIT: I had accidentally included Fabian Espindola. Reposted. I think it's right now. Here's the data.

Last Club Age MINS Shots On.Goal Goals SC. Accuracy Gp90 5 Chaves CHI 25 1699 49 14 6 0.4285714 0.2857143 0.26486168 21 Mondaini CHV 24 1493 28 13 3 0.2307692 0.4642857 0.18084394 22 Braun CHV 24 2055 69 29 8 0.2758621 0.4202899 0.35036496 42 Heinemann CLB 24 1315 45 15 3 0.2000000 0.3333333 0.20532319 45 Rogers CLB 24 2218 36 11 2 0.1818182 0.3055556 0.04057710 81 Castillo DAL 19 1355 28 8 2 0.2500000 0.2857143 0.13284133 98 Davies DC 25 1553 34 19 11 0.5789474 0.5588235 0.40566645 100 Pontius DC 24 2114 53 20 7 0.3500000 0.3773585 0.21286660 101 Najar DC 18 2522 57 20 5 0.2500000 0.3508772 0.17842982 103 Quaranta DC 21 1091 22 5 1 0.2000000 0.2272727 0.08249313 118 Bruin HOU 22 1693 54 23 5 0.2173913 0.4259259 0.26580035 134 Sapong KC 23 2096 53 25 5 0.2000000 0.4716981 0.21469466 136 Bunbury KC 21 1744 65 22 9 0.4090909 0.3384615 0.46444954 172 Schilawski NE 24 1198 24 9 1 0.1111111 0.3750000 0.07512521 175 Mansally NE 22 1140 29 4 0 0.0000000 0.1379310 0.00000000 191 Agudelo NY 19 1364 44 17 6 0.3529412 0.3863636 0.39589443 202 Mwanga PHI 20 1535 37 20 5 0.2500000 0.5405405 0.29315961 223 Zizzo POR 24 1515 22 10 0 0.0000000 0.4545455 0.00000000 262 Montero SEA 24 2304 108 42 12 0.2857143 0.3888889 0.46875000 263 Fucito SEA 25 1045 24 7 2 0.2857143 0.2916667 0.17224880 281 Lenhart SJ 25 1163 34 14 5 0.3571429 0.4117647 0.38693035 293 Martina TOR 24 1389 15 9 2 0.2222222 0.6000000 0.12958963 295 Soolsma TOR 24 1578 27 14 3 0.2142857 0.5185185 0.11406844 296 Plata TOR 19 1753 46 17 3 0.1764706 0.3695652 0.10268112

#mls #strikers #football #soccer #goals #analytics

[5050ball]

Visualizing MLS forwards' shooting accuracy, effectiveness. X is shot accuracy, Y is pct. of accurate shots that become goals. Text size is number of goals. Get thee to the interpretation machine.

#mls #soccer #statistics #shooting #stats #analytics #football #yo mama

[5050ball]

Not too fast, not too slow. Just right.

#mls #stats #soccer #football #analytics #pace

[5050ball]

Possession based metrics

One of the most critical insights in basketball stats has been the recognition of pace as a critical aspect of play. A team that scores 100ppg is often seen as a better offensive team than one that scores 90ppg, but that may just be because it plays at a higher pace (i.e. jacks up more shots).

Soccer statistics are few and far between (and even harder to find), but a measure of pace would be interesting.

A possession ends when a team makes an incomplete pass or attempts a shot. Incompletions are easily measured using "box-score" statistics: incompletions = attempts - completions; possessions can then be computed by adding shots to the number: possessions = incompletions + shots. A team's giveaway rate is essentially the proportion of possessions that have ended via an incomplete pass: giveawayrate = incompletions / possessions, while the shot rate is the proportion of shots that have ended via a shot attempt: shotrate = shots / possessions.

[EDIT: Clarified last paragraph]

#mls #stats #statistics #apbrmetrics #pace

[5050ball]

Cumulative average points per game and goal differential in MLS, 2009-2010. I forgot just how bad NY was in 2009 (and DC in 2010).

#mls #stats #visualization #meaningless graphs #CLT #central limit theorem #regressing to the mean

[5050ball]

No differences between teams in MLS strength of schedule (as measured by opponents' points per game). Any arguments to the contrary have to use different variables.

[5050ball]

I did this longer write up for Philly Soccer Page, which brings together a few of the things I did here earlier this month. As with everything I write, I hate it. In any event, I'll be using this space to post updates on stuff I'm playing with, but I hope to write some longer pieces over there.

[5050ball]

Visualization of last post. Very data dense. Differential is Union minus opponent.

ggplot( union.games, aes(x=Attempt.differential, y=On.target.differential, size=Goal.differential, color=H.A)) + geom_point() + stat_smooth(method=lm, alpha=0.25, aes(fill=factor(union.games$H.A))) + scale_color_discrete(name="", breaks=c("H","A"), labels=c("Home","Away"))+ scale_size_continuous(legend=FALSE) + scale_fill_discrete(legend=FALSE) + theme_bw()

#Union #Philadelphia Union #MLS #soccer #football #analytics #stats #statistics #Philadelphia #Home field advantage