APBRmetrics

Hey

Here's play by play data for the 2000-2001 season in txt CLICK
.. corresponding matchupfiles
and RAPM

Play by play files should be in the same format as the newer bbv matchupfiles

The play by play files don't have everything in perfect order, and there are some other minor issues. This has an impact on the matchupfiles.
In my estimation there is one possession listed wrong in ~every fifth game (two possessions in a row for the same team), and 1.5% of the matchupdata is missing because the script couldn't figure out who started the quarter. Both of these figures might be higher. There's probably a quick and dirty way to fix the second problem, when I have the time

The matchupfiles have bbr player url next to the name, to avoid issues with players having the same name


The RAPM numbers, obviously, don't come with their fair share of headscratchers:
Shawn Bradley is #4 (best defender). Iverson, who won MVP that year, is ~#25.
Players rated above Iverson include Scott Williams, Derek Anderson, Darius Miles(!!), Scott Pollard, Toni Kukoc and Evan Eschmeyer(who?)
John Stockton is rated as the 3rd best player, at age 38 ........

In some cases it agrees with popular opinion:
Shaq being a beast on offense, followed by Ray Allen, Kobe, Vince Carter
David Robinson, Ben Wallace, Mutombo are rated as bad to mediocre offensive players, but very good defenders

Wonderful!

Something I'd like to see in your output would be "possessions on court" and "possessions off court" or the like to help get an idea of the sample size for each...

So is this non-prior informed RAPM?

Do you have full season RAPM data for 2002? Will you now add prior informed data to it?

It's actually good to see Iverson that high.

DSMok1 wrote:Something I'd like to see in your output would be "possessions on court" and "possessions off court" or the like to help get an idea of the sample size for each...

If a player has played on two teams, would you want the sum of the Off possessions? Otherwise Off possessions is pretty much
82*200-"On possessions" , yes?

So is this non-prior informed RAPM?

Yes

Do you have full season RAPM data for 2002?

Not yet, but should be coming soon.

Will you now add prior informed data to it?

The plan is to release BoxScore+PBP informed RAPM numbers for all the years. The whole process will take a while though

It's actually good to see Iverson that high.

Did you expect him to be rated even worse? "Good" in what way? RAPM informed RAPM thinks his best year was 2008 (+3.1). Not that impressive, overall

J.E. wrote:The plan is to release BoxScore+PBP informed RAPM numbers for all the years. The whole process will take a while though

What speaks against releasing no-prior informed RAPM results for all those seasons?

Anyway, big thanks for the pbp data and the matchupfile. It is very much appreciated!

mystic wrote:What speaks against releasing no-prior informed RAPM results for all those seasons?

Oh I can do that, too. I'd just advise everybody to use the version which makes the best predictions. This will probably not be uninformed RAPM.
What I'll definitely do is release 1year, 2year, 3year, .., 12year RAPM so people can fit their model onto those year by year

J.E. wrote:

DSMok1 wrote:Something I'd like to see in your output would be "possessions on court" and "possessions off court" or the like to help get an idea of the sample size for each...

If a player has played on two teams, would you want the sum of the Off possessions? Otherwise Off possessions is pretty much
82*200-"On possessions" , yes?

Right, I'd just like to see some sort of way to quantify the sample size. If you've got any better ideas (is there a way to measure, say, connectivity or intercorrelation that would be useful?), go with it.

J.E. wrote:

It's actually good to see Iverson that high.

Did you expect him to be rated even worse? "Good" in what way? RAPM informed RAPM thinks his best year was 2008 (+3.1). Not that impressive, overall

Compared to what? James Harden was +3.0 this season according to RAPM and he's pretty good.

Team names would be a nice addition, but I have suggested that before. I know a few guys would have 2 team names but that is a minor issue that can either be overcome or ignored while enhancing the rest of the data set. It can be overcome manually but having it from the start seems easy enough and pretty natural.

I will add names and possessions soon

Here's uninformed RAPM for 2001 'til 2006; and 2012 because it's now updated with playoff data

http://stats-for-the-nba.appspot.com/PBP/2001.html
http://stats-for-the-nba.appspot.com/PBP/2002.html
http://stats-for-the-nba.appspot.com/PBP/2003.html
http://stats-for-the-nba.appspot.com/PBP/2004.html
http://stats-for-the-nba.appspot.com/PBP/2005.html
http://stats-for-the-nba.appspot.com/PBP/2006.html
http://stats-for-the-nba.appspot.com/PBP/2012.html

In all those years, Kobe Bryant is listed as positive on defense just once, even though he won many ALL-D first teams

@Dsmok1, you want your multiyear RAPM without playoffs, correct?

The matchupfiles are now missing ~4 quarters *per season*, so that's pretty good. There are still ~250 instances, per season, where the same team has two possessions in a row, though. I might or might not fix that, as it's a very tedious job

I'll upload all the PBP and matchupfiles tomorrow

Yes, the multi-year is best without playoffs for my purposes, because it is hard to incorporate box-score data from the playoffs. But for just an overall average, the more data the better.

J.E. wrote:I will add names and possessions soon

Here's uninformed RAPM for 2001 'til 2006; and 2012 because it's now updated with playoff data

http://stats-for-the-nba.appspot.com/PBP/2001.html
http://stats-for-the-nba.appspot.com/PBP/2002.html
http://stats-for-the-nba.appspot.com/PBP/2003.html
http://stats-for-the-nba.appspot.com/PBP/2004.html
http://stats-for-the-nba.appspot.com/PBP/2005.html
http://stats-for-the-nba.appspot.com/PBP/2006.html
http://stats-for-the-nba.appspot.com/PBP/2012.html

In all those years, Kobe Bryant is listed as positive on defense just once, even though he won many ALL-D first teams

@Dsmok1, you want your multiyear RAPM without playoffs, correct?

The matchupfiles are now missing ~4 quarters *per season*, so that's pretty good. There are still ~250 instances, per season, where the same team has two possessions in a row, though. I might or might not fix that, as it's a very tedious job

I'll upload all the PBP and matchupfiles tomorrow

Is there a prior-informed RAPM for 2002?

DSMok1 wrote:But for just an overall average, the more data the better.

I myself don't care too much about unweighted multiyear RAPM. I'm just creating those so everyone has an equal starting point when doing retrodiction

Is there a prior-informed RAPM for 2002?

The next prior informed RAPM I'll release will be informed with individual numbers from BoxScore+PBP. This will take a while because I have to grab the data from the PBP first, and then find out the best weights. The uninformed 2002 ratings should be pretty good though, because it includes 99.9% of the season+playoffs

Matchupfiles for 2001 - 2006
http://stats-for-the-nba.appspot.com/PB ... pfiles.rar

For the years after '06 I first want to merge bbv's data with mine


.. and the corresponding PBP files
http://stats-for-the-nba.appspot.com/PBP/2001.rar
http://stats-for-the-nba.appspot.com/PBP/2002.rar
http://stats-for-the-nba.appspot.com/PBP/2003.rar
http://stats-for-the-nba.appspot.com/PBP/2004.rar
http://stats-for-the-nba.appspot.com/PBP/2005.rar
http://stats-for-the-nba.appspot.com/PBP/2006.rar