Soccer metrics to evaluate players

This is a case study in why data geeks are clueless, but it does perhaps explain why Ralston keeps trotting out Eriksson in the starting lineup. The Quakes' brain-trust believe in this number-mumbo-jumbo instead of their own lying eyes, so if the data says Magnus is super, play him till he drops. Same with Godoy.

Also, the other assessments are hit-and-miss, relative to my eye test, as well. Nick Lima is a good player, but not stellar in all phases of his game, as Etnire asserts. And Etinire, remember, is the guy who said at the beginning of the year the team should have kept Simon Dawkins. Didn't remind us of that while patting himself on the back, did he?

 

It's a poor analysis, I could type eight paragraphs on how bad it is but I'll try to keep it short:

Most of it is just some numbers without context.

How the heck am I supposed to know if Godoy's 1.46 "Dispossessed" is high or low? For all of the data provided the only other reference is a Quakes player, we should at the very least have the numbers for an 'average' player in the same position or better yet multiple players who can give us an idea of what sort of range is 'normal'. This is in particular glaring for the Magnus analysis. Magnus' 1.9 key passes per game is good for 30th in MLS which isn't bad but still lower than Marco Urena(who contextless stat 'geeks' fall for probably this reason).
In addition, key passes are not a good stat for measuring playmaking ability because they do not measure the quality of chance, and can be inflated by a player being on corners or free kicks(If Magnus is taken off we would barely lose anything: when evaluating playmaking ability, open play events should be given higher weighting than set piece events). When sorting by xA per 90 Magnus is ~53rd I believe(just copy pasted from ASA and sorted so there are a few low minute players ahead of him) and 114th by big chances created per game(below a ton of mediocre fullbacks like Mikey Ambrose and Pablo Ruiz), and that still includes set pieces.

Leading the Quakes in multiple categories means absolutely nothing because the 2018 Quakes are one of the worst MLS teams in recent memory. The reason we care about improving our chance creation isn't because we are ignoring Magnus' abilities, it is because it isn't good enough. We shouldn't like a guy's chance creation numbers which are on par with Mikey Ambrose because he's the best compared to our players, if we had a real #10 he would obliterate those numbers so hard.

Keep in mind Colin previously posted articles saying we should draft Shamit Shome with our first round pick in '17 because he wouldn't have a cap hit and shouldn't release Dawkins. I think he write these sort of things to provoke discussion.

 

Opta's analysts track on-the-ball events like where passes, shots, blocks, etc take place. They don't track off the ball positions of players which are vital in calculating things like the IMPECT statistic and part of the reason why the Second Spectrum system got hyped; if you know where all the players positioned during a pass and the pass destination it's easy to calculate how many players a pass has bypassed. Opta is slowly putting out their TRACAB tracking system which is installed in the EPL, La Liga, Bundesliga, Eredivisie which uses cameras to have player locations like what Second Spectrum claims to do.

Ultimately putting in six hours of analysis seems like too much for too little when eventually if TRACAB/Second Spectrum type systems gain a wider acceptance that can calculate this with maybe a few hundred lines of code and has an open door for much more analysis.

As for the metric itself it seems like a decent evaluation for a certain type of midfielder but close to useless for things like forwards: if Wondo chases down a ball over the top of the defense, shields it, and lays it off to someone behind him(with a pass that goes by a few defenders) would he get a negative score for that? That would imply that if he didn't do anything he would have been more useful.

 

Boysinblue should post more often.

 

I'll be charting this important metric: W-L-T.

 

This looks to be a graph comparing expected goals versus expected goals conceded. Maybe I don't get it but why have a graph that does not show any real data but something that is based on some other "expected" parametric.

Why not draw a graph of goal differential for each game, anything above 0 is great. Actually it would be more valuable to see goal differential per half to gauge how well the Coaches and players are adjusting to changes in the match with subs, strategy shifts, fitness....

 

I'm no expert on xG and xG conceded, but the metrics are designed to give you an idea of the quality of the chances created and conceded because the actual results of games (as in what the actual score was) doesn't really have value in terms of projections.

I love to use the Colorado Rapids example where they played lights out defense over the course of an entire season, but scored very few goals, and ended up in the top three in their conference. I haven't done this myself, but I would guess that if you looked at xG and xG conceded for the Rapids, the indication would be that they were converting the majority of what few chances they created, at a rate that was unsustainable. And sure enough, the team imploded after that season, revealing that the results the team were getting did not reflect how good the team actually was.

Soccer seasons simply aren't long enough to give you a great deal of statistical accuracy. Players maybe only have the ball for three or four minutes in an entire game, so we aren't talking about a lot of data points that can be collected per season. It's not like baseball where guys get hundreds of at bats and face thousands of pitches. So projections help paint a picture of what the odds are of particular opportunities being converted into goals, etc.

You always need projection metrics as a kind of guidepost, and then see where the real numbers are to see if your projections were accurate or not, and then you adjust.

 

It sure would be nice of the Quakes looked at a player's speed and quickness before signing them, and maybe once in a while signed some fast players (along with skill and all those other useful metrics that are clearly yielding superior talent for the dollar).

 

Thanks for sharing that. Definitely one of the better explained articles I've read regarding soccer analytics. It also really does a good job of showing just how difficult it is just to decide seemingly simple things, such as "how important is this event?".

 

You can check it on Sofascored. At the bottom of the linked page there's a Player Statistics area, click on the Detailed tab, and then on the Defence tab and there are options for 'Errors Leading to Goal' and 'Errors Leading to Shot'.

 

This always needs to be balanced against those shots that a keeper kept out that some other keepers would not have saved.

 

Yup it seems conservative; I think for an error to qualify it must be an obvious, enormous blunder and the shot/goal has to come immediately afterwards.

 

Obviously ball-watching and being caught out of position because you don't know how to defend is not a qualifier. I bet the coaches had a much different take on it.

 

I'm surprised that they reference the film Moneyball as the place where a lot of people heard about sabermetrics, and not Michael Lewis' book, published 8 years earlier. But then again, the film wikipedia page comes up as the first search result over the book, so I guess people just don't read.

This is an aside, but for people who have neither read the book nor seen the movie, please read the book. It's one of the best works of nonfiction ever written. The film is hot garbage and changes a lot of (arguably small) details for absolutely no reason.

As for fantasy sports, there's definitely a lot of crossover. Unfortunately, from what I've seen, the data used for fantasy sports and the data actually used to make player evaluations are not really the same (or not used in the same way), so it seems like there's a weird disconnect between the player information that fantasy sports players are receiving and what the players they have on their teams are actually doing. In other words, the gambling takes priority over the sport itself. If that's what people enjoy about sports, I guess that's fine. Not being a gambler, fantasy sports seem like a huge waste of time. It would make more sense to me to play a simulation game like Football Manager or something like that.

 

There are probably people who play fantasy sports without the gambling side, but the industry of fantasy sports is very much a gambling industry. When the daily fantasy leagues like Draft Kings and those kinds of sites came up, I recall there being all kinds of questions about their legality.