APBRmetrics

[Ed Küpfer]

Over the last few weeks I've had the oportunity to exchange data files with members here. Unfortunately for me, I don't really have much webspace available to store these files, and I'm not a big fan of exchaging them by email, since the files aren't then available publically. So I've created a Yahoo Group where files can be stored. I'm not sure if you have to sign up for Yahoo membership or not, but it shouldn't be a hassle. I've turned off all the feaures of the group that I could except for the files section. If you want to share a spreadsheet with this group, you can upload it there, and announce it here.
_________________
ed

[gabefarkas]

I've got some webspace that's basically being unused, and I could use it to host some files, provided that they aren't too big. All in all there's about 10 MB free. I know it's not a lot, but it's a start. Let me know and I can set it up.

[Pinot]

I'm uploading a file that analyzes VORP and replacement value in fantasy basketball.

[davis21wylie2121]

Sweet, I just posted my BoP stat database.

[davis21wylie2121]

Plus a sheet that lets you calculate Ed's Within-Game Win Expectancy.

[davis21wylie2121]

...Plus a goofy regression I did once between raw boxscore stats and DanR/David Lewin's adj. +/-. Some of the results are bizarre (John Stockton looks like a better player than Michael Jordan), but it does explain 93.5% of Adj. +/-, for whatever that's worth.

[anarcholis]

Question about sharing files: Is there any place to get the data that seems to be unique to 82games.com in a neatly tabulated format (like a text file or excel file)?

If not, I have built a Java program that parses some of the relevant data out of that website and compies it in a delimited text file. If there is no other way to get access to that data, I could expand the parser to get everything from the site and share the text file on this yahoo group if anyone is interested.

I'm not sure if this constitutes some sort of copyright violation by stealing their data, so if anyone knows about that, please let me know also.

[davis21wylie2121]

Basketball Value theoretically tracks the same on/off data, but their possession totals seem to be off, which skews the ratings.

[JNichols42887]

[Mountain]

Davis I looked at your regression between raw boxscore stats and DanR/David Lewin's adj. +/-. Change of possession actions got heavy weights and missed free throws very light penalty. Do you have any other observations or commentary about this?

Block, steal and turnover weights several times value of possession so I guess they are carrying the hidden value of other actions not in the boxscore like shot intimidations and the value of passing lane disruptions and other things related to good defense or other offense impairing actions? Missed freethrow penalty so low because driving lane helps in other ways that largely offset?

Study of players (further stat work, but moreso tape) aided by this comparison could help detectives track down more about the hidden values and their distribution. I assume some analysts have poked down this road.

[davis21wylie2121]

It was interesting, because going in, I wasn't expecting the data to have that kind of R^2 at all. It was kind of a lark, and things were compounded by my having to "back my way in" -- to basically find "+/- value" over some replacement level and then regress that on boxscore totals, rather than simply regress adj+/- on per-minute numbers -- because I couldn't get a decent correlation any other way. So I was surprised that it came up with a 0.9353 R squared... but most of the coefficients do kind of make sense, if you assume that there are also a lot of "hidden" indicators. Take the "drawing fouls" hypothesis on missed FT, for example, or the positive +/- value for committing fouls; the idea is that players who do these (negative) things a lot end up doing other (positive) things as well. Dan Rosenbaum did something like this before, and came to similar conclusions. I've also got a spreadsheet that calculates his regressed +/- based on the link above, but it doesn't ever give me results that look like real +/- values (unlike my "goofy regression"). (John Quincy, if you're reading this, how did you get decent-looking results in this thread?)

[davis21wylie2121]

[Mike G]

Wow, over 30 months passed before this thread was resurrected; might be a record.

Yahoo groups don't give a lot of file space: some 20 meg last I checked. If I have any old files there, let me know; I will be happy to remove them. Last call!

Roland may send you files of specific data, if you just ask him.
_________________
40% of all statistics are wrong.

[Mountain]

Pinot, is the consistency scoring in your file game to game or week to week?

[basketballvalue]