This is Google's cache of http://www.sonicscentral.com/apbrmetrics/viewtopic.php?t=333&postdays=0&postorder=asc&start=43. It is a snapshot of the page as it appeared on Mar 29, 2011 23:39:50 GMT. The current page could have changed in the meantime. Learn more

Text-only version
These search terms are highlighted: usage vs efficiency  
APBRmetrics :: View topic - Usage vs. Efficiency
APBRmetrics Forum Index APBRmetrics
The statistical revolution will not be televised.
 
 FAQFAQ   SearchSearch   MemberlistMemberlist   UsergroupsUsergroups   RegisterRegister 
 ProfileProfile   Log in to check your private messagesLog in to check your private messages   Log inLog in 

Usage vs. Efficiency
Goto page Previous  1, 2, 3, 4  Next
 
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion
View previous topic :: View next topic  
Author Message
HoopStudies



Joined: 30 Dec 2004
Posts: 705
Location: Near Philadelphia, PA

PostPosted: Fri Sep 02, 2005 4:28 pm    Post subject: Reply with quote

bchaikin wrote:
If this inverse relationship holds, if it is significant at 95%, will anyone care?

it would be significant information even if it wasn't significant at 95%...



OK. I can run this.

Following the procedure I laid out above and getting the usage percentage average based on weighted average by minutes played (though it doesn't matter really as long as the individual averages are approximate), I find the following:

- An inverse relationship shows up between individual offensive rating and percentage of possessions used.
- For every increase of 1% in usage, offensive rating drops by about 0.6 over all the players.
- This coefficient is significant at well over 99%. There is roughly a 1 in 10 to the power of 38 that it's mere luck. According to my old quantum phys professor, that would be roughly the odds of banging your head against the wall and none of the molecules in your head actually hitting any of the molecules in the wall, thus avoiding pain.
- The regression has an r2 of 0.01, a case of low fit to the data being actually a meaningful result. If I were to re-add in the player averages, the r2 goes up to about 0.1.
- If you run the regression on different segments of the data -- high use games or low use games, you get essentially the same results.
- If you run the regression on players averaging high or low use, you see that the low use players (I used 18% as a cut off) are more sensitive to increases than high use players (23%). But both are still very significant. This implies that increasing Hoiberg's possessions 5% causes a bigger decline (about twice the size) than a similar increase in, say, Kevin Garnett. Or, from an optimization perspective, taking Garnett's possessions (who increases in efficiency only a little) and giving them to Hoiberg (who declines in efficiency a lot) has a pretty big cost in even this crude analysis.

This is much cruder than several of the studies I've done, but it generally reproduces the observations I've had in other proprietary studies as well as what I showed in Basketball on Paper. Clearly there is a lot of noise in this kind of analysis, which is why Bob can find examples where things just look to fluctuate around some average. But fluctuating around some average is always a zeroth order assumption. Using this elementary principle not only improves predictions but fits with a very reasonable model of the game of basketball.

What shooting percent do you use, Bob, to make projections for the coming year for, say, Allen Iverson? His career average? Last year's value? Some running average over a few years? Something more sophisticated? Because we know that the value you choose matters in your simulation and guys definitely fluctuate.
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.
Back to top
View user's profile Send private message Visit poster's website
NickS



Joined: 30 Dec 2004
Posts: 384

PostPosted: Fri Sep 02, 2005 4:35 pm    Post subject: Reply with quote

HoopStudies wrote:
- An inverse relationship shows up between individual offensive rating and percentage of possessions used.
- For every increase of 1% in usage, offensive rating drops by about 0.6 over all the players.
[...]
- If you run the regression on players averaging high or low use, you see that the low use players (I used 18% as a cut off) are more sensitive to increases than high use players (23%). But both are still very significant. This implies that increasing Hoiberg's possessions 5% causes a bigger decline (about twice the size) than a similar increase in, say, Kevin Garnett.


Thanks for running this. Very interesting. I was expecting that there would be a correlation, but I'm slightly surprised it's that large.

Does this mean that you're back from your vacation?
Back to top
View user's profile Send private message
HoopStudies



Joined: 30 Dec 2004
Posts: 705
Location: Near Philadelphia, PA

PostPosted: Fri Sep 02, 2005 4:50 pm    Post subject: Reply with quote

NickS wrote:
HoopStudies wrote:
- An inverse relationship shows up between individual offensive rating and percentage of possessions used.
- For every increase of 1% in usage, offensive rating drops by about 0.6 over all the players.
[...]
- If you run the regression on players averaging high or low use, you see that the low use players (I used 18% as a cut off) are more sensitive to increases than high use players (23%). But both are still very significant. This implies that increasing Hoiberg's possessions 5% causes a bigger decline (about twice the size) than a similar increase in, say, Kevin Garnett.


Thanks for running this. Very interesting. I was expecting that there would be a correlation, but I'm slightly surprised it's that large.

Does this mean that you're back from your vacation?


In studies where I actually include more stuff, the slope actually increases.

There is so much noise in this crude way of looking at it that the slope gets closer to 0.

Including what you suggested -- opponent D rating -- would increase that slope.

And, yeah, I'm back from vacation. Made a million phone calls but it looks like everyone is gone for Labor Day.

So I'm stuck putting together pictures, running errands, and doing regressions... Let's see if this Img tag actually works (see FFSBasketball post)...
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.


Last edited by HoopStudies on Fri Sep 02, 2005 5:16 pm; edited 1 time in total
Back to top
View user's profile Send private message Visit poster's website
Ben F.



Joined: 07 Mar 2005
Posts: 391

PostPosted: Fri Sep 02, 2005 5:13 pm    Post subject: Reply with quote

Here, I transferred it over to imageshack.us for you, which allows remote linking.

It's very nice.

Back to top
View user's profile Send private message
bchaikin



Joined: 27 Jan 2005
Posts: 688
Location: cleveland, ohio

PostPosted: Fri Sep 02, 2005 11:07 pm    Post subject: Reply with quote

What shooting percent do you use, Bob, to make projections for the coming year for, say, Allen Iverson? His career average? Last year's value? Some running average over a few years? Something more sophisticated?

for the simulation i always use the most recent regular season data. in the off season that's the final regular season stats of the previous season. during the season its the regular season stats up and through the previous weekend (once about 18-20 games have been played).....

but in addition i also run (simulate) the player's stats from two seasons earlier and three seasons earlier if the data exists, to check for consistency. take joe johnson for example, other than his individual defense (outside of steals, blocks, and def reb), the only major difference between his 04-05 and 03-04 stats is that he shot much much better from 3pt range (with more 3pt FGA) in 04-05, 48% to 31%. that difference between his 04-05 and 03-04 3pt shooting is huge, and results (along with other minor changes in his game) in a difference of close to 4-5 wins per average 82 games (playing him 40 min/g) in favor of his 04-05 stats....

so using his 03-04 stats in any 04-05 simulation if i remember i think he produced wins at the rate of just 4-5 more wins per average 82 games than some of the worst starting SGs in 04-05 (playing 40 min/g). but using his 04-05 stats that gets bumped up to 9-10 more wins per average 82 games than some of the worst starting SGs in 04-05. that's quite a difference (5-6 wins, maximized by him playing 40 min/g)...

so any analyses i run for joe johnson using his 04-05 stats is tempered by also quoting his simulation results using his 03-04 stats (and 02-03 for comparison), saying to the effect of should his 3pt shooting go south like it did in 03-04, expect those wins to drop...

for players who played few minutes in a season, say < 500, i will often add their 04-05 and 03-04 stats together and use that for the simulation...

actually the difference in using allen iverson for 40 min/g using his 04-05 versus his 03-04 stats is also about 5-6 wins per average 82 games in favor of his 04-05 stats...

Because we know that the value you choose matters in your simulation and guys definitely fluctuate.

i don't "choose" values, i only use actual data. so it doesn't matter how much anyone's stats fluctuate from one season to the next because i can run simulations using a player's data from any previous seasons (or the current season if the analysis is done in season). if for example a player was hurt in his most recent season but not the previous two, and he's now healthy, you can use all 3 sets of single season data and quantify the result of the most recent data saying he was hurt...

but in all honesty its been my experience that most players do not fluctuate that much from one season to the next., not upwards of the 5-6 wins difference you see in joe johnson or iverson above. yes you do see the few that do like how shawn marion and amare stoudemire were both much better in 04-05 than in 03-04, but for the most part players do not tend to fluctuate by more than 2-3 wins from one season to the next, and many are less than that when you normalize their minutes played (i.e. in the sim run them for the same amount of minutes using their stats from different seasons regardless of how many min/g they actually played in those different seasons)....

the only time i will actually change a player's stats from their real life numbers to something else is when running what-if scenarios, i.e. how much better or worse would a team be if for example player A shot 50% from 2pt range rather than his current 45%, or player B got 25% more rebounds, or player C had a much better or worse individual defensive FG% rating...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
gabefarkas



Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC

PostPosted: Thu Sep 08, 2005 9:05 am    Post subject: Reply with quote

Dan Rosenbaum wrote:
I think the instrumental variables approach that I am suggesting could work at the game-level or the season-level. It would be interesting to do both.

I had an idea that might be fun. There is the Journal of Quantitative Analysis in Sports that DeanO and Roland are helping start. Done right, I think this would be a nice article for that journal.

http://www.bepress.com/jqas/

I am very busy this Fall and probably only have time to serve as the organizer of this, but I was wondering who might be interested in being part of this project. We probably can't have 10 co-authors, but I suspect that even four or five would not be completely unworkable. Here is what I think needs to be done.

(1) putting together the season-level data
(2) putting together the game-level data
(3) devising the empirical strategy
(4) running the regressions
(5) putting together the tables
(6) writing up the results/writing the paper
(7) revising the paper
(Cool organizing the process
(9) handling the submission

With the limited time I have this Fall, I think that I am best employed doing (3), (4), part of (5), some guidance on (6), (7), (Cool, and (9). If there were folks who were interested in helping with (1), (2), (5), (6), and (7), that would be great.

Let me know either through e-mail, PM, or in this thread if you might be interested. If we dozens of people interested in helping out, we may need to make some choices about whose name goes on the paper, but it would be good to get many people involved in this. (But it would look bad to have a list of co-authors longer than the article itself - although I have seen that done with medical journals.)


Having worked on papers for medical journals (day job), I could probably help some with #6-#9, and I would LOVE to be involved in it.
Back to top
View user's profile Send private message Send e-mail AIM Address
kjb



Joined: 03 Jan 2005
Posts: 865
Location: Washington, DC

PostPosted: Thu Sep 08, 2005 3:08 pm    Post subject: Reply with quote

I can help with the writing/revision process. I won't have time to help with the data side of the project, however -- plus, I'm a better writer than mathemetician.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Dan Rosenbaum



Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina

PostPosted: Thu Sep 08, 2005 3:41 pm    Post subject: Reply with quote

I have been swamped with schoolwork over the past few weeks, as my posts here and at my blog has dwindled to almost nothing. I owe e-mails to lots of people this weekend.

This weekend I plan to send out an e-mail to those of you who have expressed interest. We probably have about 10 people that have expressed interest in some fashion or another and 10 co-authors probably is too many. But what I was hoping to do is have maybe half those folks be co-authors and the other half be involved in the process and get an acknowledgment in the credits. On the other hand, we could work it like some medical journal articles where we have 10 co-authors for a 10 sentence paper. Very Happy
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
gabefarkas



Joined: 31 Dec 2004
Posts: 1313
Location: Durham, NC

PostPosted: Fri Sep 09, 2005 8:21 am    Post subject: Reply with quote

happy to help in whatever way -- gfarkas@gmail.com
Back to top
View user's profile Send private message Send e-mail AIM Address
Mountain



Joined: 13 Mar 2007
Posts: 1527

PostPosted: Mon Dec 17, 2007 9:32 pm    Post subject: Reply with quote

Is the research project / paper suggested in this thread still a possibility sometime in future? A good next step after past APBR member papers groundwork? Would running it as suggested then still be the approach favored today or modify?
Back to top
View user's profile Send private message
Dan Rosenbaum



Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina

PostPosted: Tue Dec 18, 2007 12:06 am    Post subject: Reply with quote

Mountain wrote:
Is the research project / paper suggested in this thread still a possibility sometime in future? A good next step after past APBR member papers groundwork? Would running it as suggested then still be the approach favored today or modify?

Probably not by me. I just don't have the time right now.
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
Display posts from previous:   
Post new topic   Reply to topic    APBRmetrics Forum Index -> General discussion All times are GMT - 5 Hours
Goto page Previous  1, 2, 3, 4  Next
Page 3 of 4

 
Jump to:  
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