As we all know, attackers don't put up stats by themselves. Their goal and assist totals depend a great deal on the players around them, yet when we look at the numbers, we don't have an easy way to take that into account. In college soccer, this is a particular concern because teams vary widely ... some play low-scoring, defensive soccer; others are much more open. A useful question to answer is how much a player's stats depend on his team, but that's also a hard question*. As a result, we might consider an easier one: how much do a team's stats depend on a player? A simple first pass at this is the percentage of a team's goals in which a player is involved: Load = (Player Goals + Player Assists - Player PKs)/(Team Goals - Team PKs) A bit of anecdotal evidence suggests that this can be a useful stat to know. For instance, in 2002/03 Chris Goos was the runaway national scoring leader, with 20 goals and 20 assists for UNC-Greensboro; meanwhile, Pat Noonan had a respectable 14 goals and 9 assists but wasn't among the nation's top 25. All the same, Noonan had a hand in 59.0% of IU's goals, while Goos was involved in 52.6% of UNC-G's. Once you account for their respective team offenses, Noonan suddenly looks great, while Goos is no longer so exceptional. This stat might be particularly appropriate for identifying strong performers on weak teams. For example, over his last two seasons at UNLV, Rod Dyachenko totaled only 14 goals and 3 assists, but that was still a Load of 54.8%, which may explain why he was his conference's MVP and a third-round MLS draft pick. Similarly, Temple's Tony Donatelli had only 6 goals and 5 assists as a senior, but this accounted for a remarkably high 68.8% Load. (After being selected in this year's MLS Supplemental Draft, Donatelli joined Vancouver, for whom he scored on Friday.) Note that all of the above Loads are very high, above those of forwards like Jason Garey (43.9%) and Calen Carr (44.7%), both MLS first-rounders. Also, these figures may be useful for non-forwards, too ... among 2006 draftees, Sacha Kljestan had one of the highest Loads (47.1%), and so did Matt Groenwald (42.9%). On the other hand, Mehdi Ballouchy's Load was only 20.6%, so like all stats, it's not a silver bullet, just something else to keep in mind ... any thoughts? (For starters, is there a better name for this than "Load"?) *One possibility is to look at the number of shots a player attempts, but that has the drawback of not factoring in the quality of those shots, so it may not be a great option.
Interesting idea. Thanks. I'd be interested in seeing a broader application. I feel like it would possibly undervalue players on good teams (despite the Noonan example). For instance, Maryland had a few games last year where their subs ran up some goals on overmatched teams. That would drop Jason Garey (as you showed), through no fault of his own. I've always kind of that any perfect college soccer stat would have to have a strength of opponent component. I guess you would also want to normalize it for position, although defining positions gets tricky too. I looked at Stanford, who scored only 10 non-PK goals last year in 18 games. And nobody on that team had a 'load' above 30%. I guess I can conclude Stanford needs to recruit better attackers, not just improve their tactics. numerista, your PM box is full.
I had a quick look around at some players, and could not find any easy correlations. I had thought that maybe this would favor players on low-scoring teams (with the idea that attacking teams could create goals in more different ways), but I don't think that is true. Duke is an attacking team with a wealth of offensive talent, and their best player in 05 was Spencer Wadsworth at only 38.1%. But then South Florida is an attacking team with a wealth of offensive talent and Jordan Seabrook had 57.9%. Fort Lewis scored gobs of goals, and John Cunliffe still managed to have a load of 53.2%, while Lindsey Wilson scored even more goals but their highest scorer Mkhokheli Dube had only 27.2%. I suppose I could look at this as meaning that Duke and Lindsey Wilson have better supporting casts while Seabrook and Cunliffe are more individual stars. Perhaps the best pairs I found were Penn State's Jason Yeisley and Simon Omekanda (both 53.6%) and Cincinnati's Omar Cummings (53.8%) and Kenny Anaba (46.2%). Willy Guadarrama of Campbell was the single highest I found at 61.0%. A few others on my lists from a variety of school strengths: Kamani Hill (UCLA) 55.6% Kiki Willis (Elon) 46.4% Anthony Hamilton (UC Irvine) 46.9% Tommy Krizanovic (Jacksonville) 46.7% Carlos Diaz (Saint Mary's) 52.0% Joshua Okoampa (Wisconsin-Green Bay) 44.4% Charles Altchek (Harvard) 52.4%
Quick note that I forgot to include in my initial post ... even though I think it's a good idea to remove PKs, I didn't have PK data handy, so my post above leaves them in. As a result, the numbers will all be off a bit ... e.g. Jason Garey's Load was actually 45.9, not 43.9. Removing all the goals from Maryland's 6-0 and 7-0 wins raises Garey's (pk-free) Load to 50.0%. That's not a huge difference, although I think you're right that the stronger a team is, the harder it will be to register a high Load. While I agree that that's a concern, there are a few reasons that I'm not too troubled by it ... 1. Speaking intuitively, this is a measure of a player's importance to a team, so it's natural that a player who has strong teammates may play a less pivotal role. 2. The usual stats favor players on strong teams, so this provides a counterbalance. After all, nobody's likely to dismiss Jason Garey's numbers just because he contributed to fewer than half of Maryland's goals. 3. Players on very good teams can still be nationally competitive. As a senior in the fall of 2004, Scott Sealy had 16 non-PK goals and 10 assists for the #1-seeded team in the NCAA tournament, and he had a Load of 65.0%. One thing that's been tried is Weighted Scoring, which adjusts the value of each goal according to (roughly) the probability it affects a game's outcome. In general, I suspect that its weights tend to add more noise than information, although if a team plays a lot of blowouts, it could become useful. http://soccer-europe.com/Data/Goals/MarginNew.html https://www.bigsoccer.com/forum/showthread.php?t=120580 I've done a little fiddling with MLS data here ... https://www.bigsoccer.com/forum/showthread.php?t=374322 ... looks like you've now done as much with college numbers as I have, although I've cleared out a little PM space, so I'd be happy to pass along what I've collected ... not that much, I'm afraid.
One thing it might do is punish a guy for having a star attacker beside him ... Stephen King of Maryland had a respectable 15 goals (12 non-pk) and 8 assists for Maryland, but a Load of only 32.8%, and Robbie Rogers' number was only 19.7%. Certainly, if a guy spends time away from the field -- injured or benched -- his numbers will suffer.
Actually tried to do something like this a while back, and in the end came to the conclusion that it is soooo hard to compute a cleanly consistent qualitative measure from context-free quantitative stats. I had all of the league stats at hand and eventually got to stuff like weighting "first goals" and game-winning goals (plus who scored in a win, theoretically helping ID the bigger-game players) while significantly discounting goals scored when a team was already up by 2 goals. Even then, looking at the relative importance of scoring that first goal and game-winning goal devolved into those two often being the same thing when you have lots of 1-0 games. The system used by a coach or team also matters, big-time. Playing w/ a single striker or w/ a target player and scorer pair usually skews the "load", as does the number of games played. If an injured player returns for the last 1/4 of the season and scores in every game they play (sending their team on to a title, etc.) how much more important were those games & those goals? Goals in soccer are relatively scarce for the length of the game. There usually just aren't enough relatively repeatable scenarios and enough statistics generated in a low-scoring game like soccer to make up a baseball-like page of pithy insights (central defenders rarely get to step up to the plate w/ runners on base, stuff like that). Try as I might, I found that soccer just doesn't lend itself to statistical simplification as easily as some other sports. Not that I've stopped trying, though...
A few PDL notes ... -- Among the league's top scorers, I believe the highest Load is 75%; Ross Mackenzie of Williamsburg has been involved in 9 of their 12 goals. Curiously, his Load% for Old Dominion last fall was only 25.7, as he was mainly a provider to Brian Cvilikas, who himself had 15 goals and 0 assists. -- The league's top two scorers also have very high loads; Cunliffe 67.8%, Moojen 62.5%. -- Another high Load is 58.8% for Pete Marino, an ex-MLSer. -- Having three big scorers, Carolina seems to merit special attention; I calculated the Load% for all three, manually omitting goals from games in which a player was absent and PKs (Roberts & Hunter made one apiece). This was the result: Randi Patterson 53.4 (missed 1 game) Darryl Roberts 43.5 (missed 3 games) Ben Hunter 34.6 (missed 2 games) Patterson is the #3 scorer in the league, and while his Load% isn't as high as the top two, it still appears quite strong for someone who is part of a big offense.
As an aside to this idea numerista, if I can pick that big stats loving brain of yours. I have a sense that assists are worth a little more in terms of predictive professional output than goals are. The theories are fairly straightforward - assists may indicate better field vision, ball skills, and tactical sense, they may indicate more of a team mentality, goals may be a function of a single skill attribute that will not transfer as well at the next level. I don't have anything more than my own intuition to go on. I'm not saying the relationship would be much more than 1-1, so a top goal scorer would still look good. But take a guy like Cvilikas, who you mentioned in your last post. CAA Player of the Year with 15 goals and not a single assist. He did get drafted and has signed with Virginia Beach, but his pro future is still a bit cloudy. I'll be interested to see if some of the guys that fed him fare better.
For me at least, your comment about Brian Cvilikas helps point out how much more there is to the game and that we're missing some of it in the usual statistics. TIA - I don't know Cvilikas, though I have seen him play a few times. He appeared to be used as a singular target player, and he's a flat-out goal getter. The part about getting assists is harder sometimes, 'cause you need others in the area to get the ball to and you need to be credited with an assist. In at least one game I saw Cvilikas really attract the attention of three defenders and while their heads were on swivels a guy named Elcock swooped in behind them from the left and scored. Assist? No, of course not, but Cvilikas made all that space and that all but made the play. On another goal Cvilikas neatly stepped over a through pass, freezing the defenders and letting it run through to the oncharging wing who blasted it home. Assist? Nope - never touched the ball, but again it made the play. All that said, I still can't see how you can go 15 goals and 0 assists - you mean that no one buried a flick-on the whole year? I don't think we're to the point where we adequately identify and statistically capture the more subtle parts of play. Just on gut feel and just like Emile, I do tend to think that players with high assist counts are likely to be more consistently successful - maybe it is that vision and soccer smarts thing. But until we can better capture the smart plays and quality assists and great goals vs garbage goals - I guess, until there are style points - there still is a qualitative beauty to the game that our numbers can't (yet) measure. But that won't stop us all from still trying. At least, it hasn't stopped me yet...
I agree that goals-without-assists could be a red flag, but might there also be problems indicated by assists-without-goals? I'm not talking so much about Hunter Freeman, Kevin Goldthwaite, or Todd Dunivant, who are wide players and wouldn't be expected to get many shooting opportunities. But what about Memo Arzate, who as a senior at UCSB had 2 non-PK goals and 18 assists? Did his lack of goalscoring indicate a limited upside?
Good question. There is another test case at Wake Forest in Ryan Solle, who is piling up assists without scoring himself (with a season to go he has career stats of 1 goal and 23 assists). Is this a case where he is simply the beneficary of good forwards who make something out of an average pass? Or maybe he's a genuinely good passer but can't shoot? Or he simply plays too deep in the midfield to get chances? Here's a fascinating stat on Solle. He had 12 shots _total_ last year and 13 assists. Crazy. It would be nice to have an MLS Equivalency Statistic for college stats, but as Mr. Hosehead notes, the manner of how these stats are accumulated are too sketchy. At least they don't have the double assist, so nobody is getting credit for having kicked a ball to some guy thirty seconds before the goal was set-up.
I thought Solle's stat was surely unique, but I just looked up A.J. Godbolt, who had 0 goals and 13 assists this past season - and he had only 12 shots all season as well. He is more of a wide player than Solle, but I still find these stats astonishing.
Taking the opposite slant on players with many assists but few goals, two MLSers who fit that description are Mark Lisi and Simon Elliott. (I'm sure there are others, but these two came to mind.) Lisi MLS - 7 goals, 30 assists College - 43 goals, 37 assists for Clemson Elliott MLS - 11 goals, 54 assists College - 13 goals, 12 assists for Stanford (in two seasons, I think) Trying to think of bigger names, I came up with ... Tab Ramos MLS - 8 goals, 28 assists College - 31 goals, 42 assists Eddie Lewis MLS - 9 goals, 35 assists College - 30 goals, 28 assists Even though goalscoring hasn't been part of their job in MLS, they all did a fair bit of it in college.
Maybe I'm missing something, but why are a player's PK goals subtracted? Aren't those goals still contributing to the amount of "Load" an individual player is carrying?
It's certainly debatable, but I'd prefer not to include PKs at all. While I'd agree that penalty-taking tells us something, most teams have several guys who are comparably good at it, so PK goals give a lot of credit to one individual for a task that others could do equally well. What's more, if you credit only the penalty shooter, you're effectively taking credit away from the guy who drew the PK (which is often most of the work). Rather than do that, I'd prefer to omit PKs altogether.
This might be more aptly called Relative Offensive Contribution, or ROC. I suppose if you wanted to assign a measure that weighted the statistic to schedule strength for attacking players, you might just create an index (say Weighted Attacking Index ) that multiplies the ROC by the soccerratings.com schedule strength for a given team. It should be a number in the 700-1000 range, and would do a pretty good job of indexing individual attacking prowess against strength of competition. It's also a better metric for MLS considerations, at least as a rough cut device, to be used in conjunction with PDL performance. It'd be a pretty easy stat to develop at the end of every year, and reasonably discounts players who score or create lots of goals against weaker teams. Your comparison of Chris Goos and Pat Noonan would become a good deal more enlightening in that context. In 2002, IU's SOS was 1642, while UNC-G's was 1486. So, Noonan's WAI would have been (0.590 x 1642) = 969, while Goos' was (0.526 x 1468) = 772. BIG difference.