I've got a beta version of the stats ready to go (as in so beta, that it's alpha). So those of you interested, go take a look.
However, PLEASE read the rest of this post before you click on over to it.
Couple of important things to know before you go crazy.
1) Like I said, this is not finished by any means. I've gotten the basic functionality there, and I would like to hear from you ideas on what to add to this. The only thing I've really thought about alot but haven't implemented yet is a way to look at the schedule of games and then a page for each individual game, with the two teams and the players involved. Past that, I'm wide open to hear what you guys want to see. Also, if you notice any bugs, things that don't work right, missing players, etc, please let me know.
2) Right now this is just on my own personal site. Shared hosting, shared database, the works. I'd like to move it off of this. Preferably soon. However, to keep the wife happy it has to pay for itself. So you'll notice a few ads. I've tried to keep them out of the way so as to not interfere with usability, and hopefully they won't be too bothersome. If you appreciate the work being done, that's one way to show it. And as soon as the page produces enough to purchase a domain name and a few months of hosting, I'll move it on over.
3) Not that I'm too worried about it, but lets not publish this too widely just yet. Tell a few friends or whatever, but don't post us on Digg just yet.
4) All comments can come back here to the blog. That way I can keep it all organized 'n' such.
5) Oh yeah, you'll notice that some of the numbers are a little bit weird. Mainly that the WP48's don't match the studies done over at the Wages of Wins blog. I've used the math from this post to estimate WP48. The issue is with the assigning of positions. When DBerri and company calculate Wins Produced, they manually go over a team and assign minutes based on who is playing what position. This is a somewhat complicated process, but it ends up with a good approximation of what position each individual player plays. Since I don't have a good automated imitation of this process, I've just pulled the players' positions from the ESPN.com site and used that. This affects PAWS, PAWS/min and WP48. You'll notice large changes in some players, especially those listed as "forwards" (PF/SF) and "guard forwards" (SF/SG), since those respective positions have greatly different position adjustments. I'm not sure how to get around this issue. If any of you have any ideas of a way to automatically calculate this, it would be appreciated.
Thanks for all the supportive comments thus far. Hopefully this will turn into a useful tool for everyone.
Here is the link. Enjoy.
Subscribe to:
Post Comments (Atom)
21 comments:
sensational! really.
two quick things:
1) maybe you should list per game or per minute averages as opposed to totals. That way, people can start to get an intuitive sense for the connection between per-game stats and wins score.
2) one stat this isn't tracked on most stat sites is points-per-shot. I know berri occassionally refers to this. Might be interesting to also track that.
minor points. site is great and look forward to visiting it often.
thanks
Just great. Thanks so much for making this happen. I clicked through some of the ads. :-)
As you said, it would be great if you could look on a game by game basis. I have been thinking about posting win scores for games over at k-blogger, but have only done it twice because it takes some time to calculate. If I could somehow take it directly from here that would be really great.
But I love love love the site, again, thanks...
oh, one other thing that might be useful:
a way to do multiple sorts. Maybe you could have a box that you check if you wanted the sequence of sorts you do to be preserved. That way, for example, you could do things like sort on win-score and position to see who the most productive PGs are. As far as I know, no stat site offers the ability to sort on multiple columns.
Awesome!
Ironic, isn't it, that Isiah Thomas gets an extension even though he's not even starting the most productive player in the game?
I'd love to see "Age" as one of the sortable columns. Great work.
Thanks
FGA per 48 would be interesting to me but I guess i can save to excel and do stuff like that
What I would love to see is a standard deviation and/or a percentile applied to PAWS/min so that it is easier to get a relative idea of how good players are.
Unbelievably great - can't thank you enough for creating this database.
Quick question: how frequently will you update it? Is there a certain time each day you'll run the previous night's numbers? Is there a way to automate it, or do you have to enter each stat manually?
Thanks for all the kind comments. Some great ideas there on some really interesting stats to look at. I'll be looking at adding a bunch of those once I get the basics moving well. As for the question on updating, right now it's kind of a manual automatic thing. That is, I have to manually set my scripts to run each morning. Part of the reason I want to move this to another domain and server is so I can automate a lot of this stuff to run at 4 am or whatever. So that should be forthcoming, once I've got enough to move it.
Again, thanks for all the suggestions. I'll keep you updated on progress.
JChan,
This is exactly what I envisioned when I suggested a sortable database. I'm glad that you have the interest in WoW and the ability with databases to make this happen.
I think this will get you about 100x more site visits than just about anything else you could have done.
Congrats.
Matt Barnes a better player than Gilbert Arenas?
Tyson Chandler better than Dwyane Wade?
Are you kidding me? Who takes these stats seriously?
It looks like you are using the WP48 approximation that Dave Berri suggests on his blog. That approximation has no team adjustment.
I would recommend multiplying your WP48 values times minutes played and adding them up team by team. I bet you will not be close to explaining 95% of team wins when you do.
(Square the correlation coefficient of team wins and your added up WP48 to get an approximation of how much of team wins you are explaining.)
I would bet you will be close to explaining something like 75%. And when that happens, don't assume that you have made a mistake. But do ask Dave Berri what is going on and why, without a team adjustment, you can't explain anywhere near 95% of team wins.
And then when you get done asking that question, ask Dave Berri why this had to be a surprise. Why didn't he make it clear to his readers months ago that the team adjustment is critical in explaining team wins?
Did he not realize it was so important or was he simply trying to mislead his readers?
As I get deeper into studying this formula there are more questions to be answered, but even so I still find it to be much better than anything else I've seen in evaluating players. And it is
important to remember that Wins Produced doesn't tell us who the best basketball player is, it tells us who is helping their team win the most. Those are two different things. Arenas is one of the best basketball players in the world. Just watch him, you can see that. But there are things he is doing that are not helping his team win. Look at the top seven shot-takers on the Wizards. Arenas has the worst shooting percentage of any of them. When you compare Barnes and Arenas you see that Barnes has 63 steals and 91 turnovers (-28 poss) and Arenas has 116 steals and 212 turnovers (-96 poss).
Now it's easy to argue any of these things. Well Arenas has the ball more often than Barnes, so he's
going to have more turnovers. OK, well Arenas guards the man with the ball more often, so he should have
more chances for steals. Well, Arenas has to take bad shots sometimes because he's the go-to-guy, or he's the only one that can "create" his own shot. If Arenas passed 100 of his shots over to Caron Butler, would the Wizards have a better chance to win? The problem is that we don't know. We don't know any of these things. All we know for sure is the stats. We know that Arenas took X number of shots and made X% of those. And we have to try to extrapolate from the stats who will help a team win.
The other issue is that it takes all kinds of players to win a game. Tyson Chandler can never give a
team what Dwyane Wade does and vice versa. What we can see from these numbers is that a team of Chandler,
Wade, Arenas, and Barnes, plus an average PF, should be a pretty good team. All of those players are above average.
I doubt there will ever be a perfect way to evaluate NBA players. But this formula is the most logical I've seen, and I plan on continuing to use it and watch how it progresses. As for Mr. Berri, I believe he encourages these types of discussions and I encourage you to make comments on his blog. Concerning the team adjustments, he is much more qualified than I am to discuss it.
But I think one of the reasons many have found Wins Produced to be "logical" is because it explains 95% of team wins. But that fact is completely meaningless, because ANY metric with a team adjustment will explain 95% of team wins.
And Berri has had a whole book and more than 100 blog entries to explain this to his readers, but he has chosen to hide this fact, plus the fact that without a team adjustment Wins Produced explains only around 75% of team wins.
I guess the question is this. What empirical evidence do we have that Wins Produced is any better than any other random metric? If the explaining of 95% of team wins is irrelevant, what evidence is there?
Jason,
This is excellent! I am excited that you have such an interest in our work, and that everyone can see these metrics. While others claim to have better systems (and who knows they might) this one is out there with your help. Secretive (and/or anecdotal) methods and methodologies do not enlighten, nor do they allow us to get closer to correctly evaluating player performance. Your work really is a significant step to bring this reasearch to a broader community. Hopefully someone at ESPN or some other sports network will pick up on your database. Bravo!
Isn't Wins Produced still a secretive method? Jason is just producing pseudo-Wins Produced. Where is that long-promised paper describing how Wins Produced is calculated? Is that secret? Or anecdotal?
And what evidence is there that it is better than ANY other metric?
the point is now people can evaluate the predictive power of "pseudo"-wins produced. So even you, anonymous hater, can examine how well win-score does over the course of a season and determine whether it's something you're willing to embrace or not.
I am really eager to see what folks find when they look into the predictive power of Wins Produced.
It might be useful to compare it to Points per Game (as a measure of per-minute productivity) with a team adjustment (PPGTA). Just to make it simple, assume all the rookies and players who play less than 250 minutes a season are predicted perfectly.
Then see how well PPGTA predicts future wins compared to Wins Produced. I think there might be a few years where Wins Produced does better, but PPGTA, I think, does better more often.
But don't trust me. Try it yourself. And remember, since PPGTA is simpler than Wins Produced, we probably should prefer it if the two are similar in predictive power. I think that is an argument that Berri has made.
it's funny. you keep harping on the team adjustment. note: THERE IS NO TEAM ADJUSTMENT IN WIN-SCORE.
I know there is in Wins-Produced, but JChan is mainly tracking Position Adjusted Win Score. And Win-Score isn't dramatically different from the formulas Hollinger and the APBR community come up with (in terms of how easy it is to use and manipulate - no need for regression analysis).
After all, all you folks in the APBR community are trying to do the same thing: come up with models that predict or explain player performance. Should we just debate their merits without exploring their predictive power or actually test them to see whether they work or not? I think we should test them. But maybe you're right, we shouldn't and we should trust hacks like Dan Rosenbaum's whose model has obviously done wonders for the cleveland cavaliers
Take Win Score or Wins Produced without a team adjustment. Add them up for a given team. Then see how much of team wins they explain.
Bet it is way less than 95%.
All of these posts are simply asking folks to examine how well Wins Produced without a team adjustment explains team wins.
Or how well Wins Produced (with the team adjustment) predicts team wins compared to another clearly limited metric, such as Points per Game with a team adjustment (PPGTA)?
What is so unreasonable about that? I have stated simple hypotheses and predicted what will happen. If I am wrong, I am wrong. But maybe I am not.
I wonder what kind of technology you're using. I see .php pages but want to know what kind of database and if is there another language used in the scripts.
Congratulations for the great work.
Post a Comment