|
APBRmetrics The statistical revolution will not be televised.
|
View previous topic :: View next topic |
Author |
Message |
HoopStudies
Joined: 30 Dec 2004 Posts: 80 Location: Bay Area, California
|
Posted: Tue Mar 01, 2005 2:44 am Post subject: possession data |
|
|
I've been meaning to pass this on for someone to work on. I've been going through play-by-plays to count the actual number of possessions used by the two teams in a game. These are Sonic games going back a few months. Feel free to take a stab at improving total possession formulas.
Code: |
Date Opp Actual Meaningless
12/9/2004 dal 189
12/11/2004 bos 196
12/14/2004 lal 182
12/17/2004 pho 191
12/22/2004 den 179
12/27/2004 uta 184
12/28/2004 phi 197
12/30/2004 atl 174
12/31/2004 cha 170
1/3/2005 mia 172
1/5/2005 orl 182
1/6/2005 was 190
1/9/2005 MIA 175
1/11/2005 LAC 179
1/12/2005 lac 175 2
1/14/2005 gsw 165 0
1/16/2005 cle 185 1
1/18/2005 den 203 1
1/21/2005 min 165 1
1/23/2005 uta 182 0
1/25/2005 lal 176 1
1/26/2005 uta 182 1
1/28/2005 gsw 173 1
1/31/2005 san 173 0
2/1/2005 sac 181 1
2/5/2005 cha 180 1
2/8/2005 nor 172 1
2/10/2005 sac 189 2
2/11/2005 pho 184 0
2/13/2005 dal 183 3
2/16/2005 gsw 185 1
2/22/2005 hou 179 1
2/23/2005 nor 171 1
2/25/2005 min 175 0
2/27/2005 mil 182 1
|
The far right column includes situations at the end of quarters where a team has the ball but doesn't take a shot. Some call that a possession, some don't. I didn't start tracking this value until later.
If you work on it, let us know whether you're matching to the possession total or that number minus the number on the right. Such an estimate is ideally unbiased with a minimized error. _________________ Dean Oliver
Consultant to the Seattle Supersonics
Author, Basketball on Paper
http://www.basketballonpaper.com |
|
Back to top |
|
|
mathom
Joined: 16 Feb 2005 Posts: 10
|
Posted: Thu Mar 03, 2005 5:56 pm Post subject: |
|
|
A few questions:
Are these the possessions as you determined them or did you at all attempt to match them up with the final boxscore (if there were any discrepancies)?
What was your method for determining a possession? Are you considering only offensive possessions (ie resulted in a shot or fts) or total possessions ie including turnovers, and how did you categorize technical free throws?
I think the numbers could be quite useful for determining a more exact FT constant. |
|
Back to top |
|
|
HoopStudies
Joined: 30 Dec 2004 Posts: 80 Location: Bay Area, California
|
Posted: Thu Mar 03, 2005 6:21 pm Post subject: |
|
|
mathom wrote: | A few questions:
Are these the possessions as you determined them or did you at all attempt to match them up with the final boxscore (if there were any discrepancies)?
|
Match them up? I went through the play by play and counted possessions. I have a tool for counting them that has the one issue you reminded me of -- Technical foul shots can add an extra possession or two. I should check that.
mathom wrote: |
What was your method for determining a possession? Are you considering only offensive possessions (ie resulted in a shot or fts) or total possessions ie including turnovers, and how did you categorize technical free throws?
I think the numbers could be quite useful for determining a more exact FT constant. |
Basically the fourth column is what is accounting for the difference between total possessions and "offensive possessions", as I think you labeled it. So a 2 at the far right means that there were 2 possessions where no shot was taken at the end of a period. A team got the ball and held it.
Using the format
FGA - a * (oreb/(oreb + oppdreb)) * (fga - fgm) + tov + 0.4 * fta
where a = 1.07, I find that this underpredicts total possessions by an average of about 1.8 per game. Standard error of 3.2. If a = 1.0, it's average error is 0.0, standard error is 3.0. The average number of "meaningless" or purely defensive possessions (I like you calling them offensive possessions) is about 0.9 per game. Accounting for that would mean that a = 1.0 is overpredicting possessions and a=1.07 is still underpredicting. If I go back through and look for cases where technical fouls falsely added 2 possessions, that would make a difference, too, but those don't happen too much.
I haven't tried to mess with the factor of 0.4. I haven't tried the
FGA - OR + tov + b * fta
form yet. At one time in the past, I found that the previous form had smaller errors than this one. I still believe that, but, eh, whatevah.
John Maxwell says, and I believe, that team rebounds are critical to making the estimates better. But see what we can do without them first. _________________ Dean Oliver
Consultant to the Seattle Supersonics
Author, Basketball on Paper
http://www.basketballonpaper.com |
|
Back to top |
|
|
HoopStudies
Joined: 30 Dec 2004 Posts: 80 Location: Bay Area, California
|
Posted: Thu Mar 03, 2005 6:50 pm Post subject: |
|
|
HoopStudies wrote: | mathom wrote: | A few questions:
Are these the possessions as you determined them or did you at all attempt to match them up with the final boxscore (if there were any discrepancies)?
|
Match them up? I went through the play by play and counted possessions. I have a tool for counting them that has the one issue you reminded me of -- Technical foul shots can add an extra possession or two. I should check that.
|
I checked. It looks like my possession calculator doesn't have a problem with technicals interrupting a possession. If Iverson is bringing the ball up the court, Marc Jackson gets called for a T, Ray makes the free throw, then Philly resumes with the ball, it's all still Philly possession. There might be an extremely rare situation where it doesn't work (like in between offensive rebounds), but those are rare enough to not affect any estimation.
HoopStudies wrote: |
mathom wrote: |
What was your method for determining a possession? Are you considering only offensive possessions (ie resulted in a shot or fts) or total possessions ie including turnovers, and how did you categorize technical free throws?
I think the numbers could be quite useful for determining a more exact FT constant. |
Basically the fourth column is what is accounting for the difference between total possessions and "offensive possessions", as I think you labeled it. So a 2 at the far right means that there were 2 possessions where no shot was taken at the end of a period. A team got the ball and held it.
|
Having looked back, the number that should be matched is the Possession number minus the value on the right. In other words, offensive possessions. I realize this because if a player commits a turnover at the end of a quarter and the other team gets it, does nothing with it, the possession calculator doesn't see the new possession. So, for consistency, the number we should be matching is the poss - meaningless. And this means that the form with a = 1.0 yields something too high.
Good questions. _________________ Dean Oliver
Consultant to the Seattle Supersonics
Author, Basketball on Paper
http://www.basketballonpaper.com |
|
Back to top |
|
|
mathom
Joined: 16 Feb 2005 Posts: 10
|
Posted: Thu Mar 03, 2005 8:16 pm Post subject: |
|
|
Quote: | Match them up? I went through the play by play and counted possessions. I have a tool for counting them that has the one issue you reminded me of -- Technical foul shots can add an extra possession or two. I should check that. |
I wasn't sure if you were counting them by hand as you watched the game/tape of the game or not. It appears you're using some sort of program to parse game logs from the official scorer, which obviously would mean there were no discrepancies.
Quote: | Basically the fourth column is what is accounting for the difference between total possessions and "offensive possessions", as I think you labeled it. |
Actually, I meant an offensive possession to mean a possession which ended in a FGA or FTs, ie not in a turnover or end of the quarter. In other words, I was asking if you were counting turnovers in your possession total, which it sounds like you are.
I'm not sure why you are using the term: a * (oreb/(oreb + oppdreb)) * (fga - fgm). Unless I'm missing something, this should be equal to a * OR, as fga - fgm is = to oreb + oppdreb, assuming oppdreb is including team rebounds (if it's not, it's just introducing error). Assuming I'm not forgetting something, all that does is weight OR, which I see no reason to do.. |
|
Back to top |
|
|
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
Powered by phpBB © 2001, 2005 phpBB Group
|