|
APBRmetrics The statistical revolution will not be televised.
|
View previous topic :: View next topic |
Author |
Message |
Kevin Pelton Site Admin
Joined: 30 Dec 2004 Posts: 978 Location: Seattle
|
Posted: Thu Nov 06, 2008 11:18 am Post subject: |
|
|
davis21wylie2121 wrote: | In basketball, two numbers are always better than one if one describes a player's role and the other describes his efficiency, because there's a dynamic trade-off between the two metrics. For instance, if I know a player's ORtg and his %Poss, and I know that he's going to have to take on more possessions in the future, I can assume that his efficiency will decrease in response. |
Right, and few people agree with you on that point more than me. However, in my opinion that changes the shape of a player's statistics much more than his overall value. In most non-extreme circumstances, a team should be indifferent between a player with higher efficiency in a smaller role and one with lower efficiency in a bigger role.
The other thing here is that we don't know the slope of players' usage-efficiency curves and can only guess at them at best. |
|
Back to top |
|
|
Mountain
Joined: 13 Mar 2007 Posts: 1527
|
Posted: Thu Nov 06, 2008 11:34 am Post subject: |
|
|
If you sorted game by game usage vs efficiency by player into a distribution you could get some ideas, but limited by how much a player's usage has varied and how much data there is overall and in each band and affected by game situation / opponent strength. |
|
Back to top |
|
|
Neil Paine
Joined: 13 Oct 2005 Posts: 774 Location: Atlanta, GA
|
Posted: Thu Nov 06, 2008 11:41 am Post subject: |
|
|
You know what I think would be really ideal (and really tricky to pull off)? Individual "skill curves" for all 8 factors (4 offensive, 4 defensive). To do it, you'd have to slice up a player's individual possessions at both ends (which gets extra impossible on defense) by the % devoted to shooting + assists, turnovers, rebounding, and free throws, and then determine how much each factor changes when the rate of possessions devoted to each factor changes. Unfortunately, that process isn't even on my to-do list -- it's on my Christmas wish list.
I've always felt that, while an improvement over single-number "holy grail" ratings, ORtg/%Poss still didn't tell the whole story in favor of convenience. 4-factors skill curves for every player would theoretically tell a decision-maker all they needed to know regarding performance and fit, assuming we knew the trade-offs. That's an assumption we'll probably never get to make, but it would be a fantastic set of stats for a GM if we ever get there. |
|
Back to top |
|
|
Mountain
Joined: 13 Mar 2007 Posts: 1527
|
Posted: Thu Nov 06, 2008 12:10 pm Post subject: |
|
|
I've said before that Direct Individual and Team Impact when on the court 4 factors (for both offensive & defensive) (8*2=16 aspects) raw is or ideally the partial adjusteds would be a good stat set to summarize a player. In one of my push the envelope moments I suggested that ideally you'd also want the marginal values for each of these in addition to the averages (hence 32 or even 64 data bits for the evaluating / operating software). |
|
Back to top |
|
|
deepak
Joined: 26 Apr 2006 Posts: 664
|
Posted: Thu Nov 06, 2008 12:40 pm Post subject: |
|
|
Has anyone looked into how volatile the offensive APM is compared to the defensive APM, year-to-year? It seems to me that offensive impact might be more role-dependent compared to defensive-impact. |
|
Back to top |
|
|
Mountain
Joined: 13 Mar 2007 Posts: 1527
|
Posted: Thu Nov 06, 2008 1:06 pm Post subject: |
|
|
Good question / speculation deepak. Might be worth doing by position and Sparks type.
Defensive adjusted might be more stable or might not be. It is easier to hold up your end when others are not screwing up or screwing up badly and context may be just as crucial on defense as offense or maybe even moreso. Don't know. Worth checking further.
On skill curves for 4 factors *2 I think Ed K's previous work would cover at least 5.
Last edited by Mountain on Thu Nov 06, 2008 1:07 pm; edited 1 time in total |
|
Back to top |
|
|
Ryan J. Parker
Joined: 23 Mar 2007 Posts: 708 Location: Raleigh, NC
|
Posted: Thu Nov 06, 2008 1:07 pm Post subject: |
|
|
davis21wylie2121 wrote: | You know what I think would be really ideal (and really tricky to pull off)? Individual "skill curves" for all 8 factors (4 offensive, 4 defensive). To do it, you'd have to slice up a player's individual possessions at both ends (which gets extra impossible on defense) by the % devoted to shooting + assists, turnovers, rebounding, and free throws, and then determine how much each factor changes when the rate of possessions devoted to each factor changes. Unfortunately, that process isn't even on my to-do list -- it's on my Christmas wish list. |
This sounds like something similar I will be diving into soon, although I plan on looking at the fundamental building blocks. For offense, these are for things like shooting percentages, turnovers per possession, and offensive rebounding percentage. For defense, these are things like allowed shooting percentages (by area), forced turnovers per possession, and defensive rebound percentage. These are just to name a few.
Hopefully it will provide some insight into these areas, or at least create a path in that direction. _________________ I am a basketball geek. |
|
Back to top |
|
|
Mountain
Joined: 13 Mar 2007 Posts: 1527
|
Posted: Thu Nov 06, 2008 1:14 pm Post subject: |
|
|
Ryan I read with interest your recent work on starter usage at your site
http://www.basketballgeek.com/
fyi to others this is another site worth checking occasionally
feel free to give us a heads up when your new / best stuff hits. |
|
Back to top |
|
|
Harold Almonte
Joined: 04 Aug 2006 Posts: 616
|
Posted: Thu Nov 06, 2008 3:11 pm Post subject: |
|
|
Interesting curve: starters vs. lead
I had once suggested in another topic a total of 4 starters or less in the game between the two teams combined, and/or a 99.0 win% to identify a garbage time. I should reconsider to 5 starters now. |
|
Back to top |
|
|
Mountain
Joined: 13 Mar 2007 Posts: 1527
|
Posted: Thu Nov 06, 2008 3:20 pm Post subject: |
|
|
As I noted earlier with other datasets Houston got the least or relied the least or both on their #1 star than any of top 12 contenders on this study based on last year roster/minutes. Yao 37th overall. McGrady the #2 and in 65th place overall. Now concentration of credit isn't necessarily a good thing even if it is expected of stars.
Landry was the per minute leader, 13th best in league. Orlando didn't go after him hard and instead gave Pietrus somewhere around twice as much money, either a commitment or overcommitment to a certain style.
Artest a fine pickup at 16th best. Old man Mutumbo was 21st. Pushed down some Hayes 23rd (might be a good trade target for right team). Rockets clearly know the value of rugged role players.
Widely praised, nice guy Roy at 110 at this stage, whatever stage this is- early or already near peak.
Wanted to be paid huge Ben Gordon at 270. David Lee at 215 and Kaman also a surprise at 260. Got paid huge Monta Ellis at 159. Iggy at 126.
Much sought Posey 235, Mo Williams 224 don't strike me as game changers on adjusted +/-.
Up for a new deal soon Odom and and Boozer at near ho-hum 150.
How high will Bynum have to go to be a good signing? Low 20s like Gasol or high 30s like Yao or late 40s like Bogut? Over the hill Shaq at a surprising 12. Shouldn't Bynum ("the new Shaq") eventually get to the level of over the hill Shaq on impact? Will he?
Last edited by Mountain on Fri Nov 07, 2008 2:22 am; edited 4 times in total |
|
Back to top |
|
|
Kevin Pelton Site Admin
Joined: 30 Dec 2004 Posts: 978 Location: Seattle
|
Posted: Thu Nov 06, 2008 3:49 pm Post subject: |
|
|
Mountain wrote: | Over the hill Shaq at a surprising 12. |
That shouldn't be a huge surprise. While the weight is relatively low, we're talking about still having 2003-04 in the dataset, which is a long time ago in Shaq's career.
Nor should it be surprising if young players rank lower than their reputation. |
|
Back to top |
|
|
Mountain
Joined: 13 Mar 2007 Posts: 1527
|
Posted: Thu Nov 06, 2008 4:01 pm Post subject: |
|
|
The weight on that 2003-4 data was quite low at just under 8% of the total weight, but point taken, the prior years combined played a role in producing that 12th place finish. Last season's 44th place finish got a 63.5% weight. |
|
Back to top |
|
|
Mountain
Joined: 13 Mar 2007 Posts: 1527
|
Posted: Fri Nov 07, 2008 2:26 am Post subject: |
|
|
basketballvalue wrote: |
These weights were chosen to trade off emphasizing 2007-2008 performance with noise reduction. Others certainly could have chosen different weights. I wouldn't say we conducted an exhaustive search before settling on these weights.
Aaron |
Ok, I'll take that into consideration from here. |
|
Back to top |
|
|
basketballvalue
Joined: 07 Mar 2006 Posts: 208
|
Posted: Fri Nov 07, 2008 9:51 am Post subject: |
|
|
Mountain wrote: | Ryan I read with interest your recent work on starter usage at your site
http://www.basketballgeek.com/
fyi to others this is another site worth checking occasionally
feel free to give us a heads up when your new / best stuff hits. |
Yes, that is in an interesting chart. On the x-axis, the lead is for both teams, right? So when the home team is up by 10, the home data point is +10 but the away team is at the -10 point, right? Might be interesting to plot them both as +10 so the two points are on the same x-value.
Among your future directions, it might be interesting to compare the # of starters to the win probability of the game that we've seen posted elsewhere.....
Keep up the good work. I noticed on your 2007-2008 data post you mentioned that you're missing data on 47 games. Presumably those are the more challenging games in terms of personnel with the same last name. I imagine you have already, but if not please leverage my matchup data to help resolve those last few games. If you notice any errors in my parsing or the manual resolution of those names (e.g. Ray and Tony Allen, or the Davis triplets on the Clippers this season, let me know.
Thanks,
Aaron _________________ www.basketballvalue.com
Follow on Twitter |
|
Back to top |
|
|
Ryan J. Parker
Joined: 23 Mar 2007 Posts: 708 Location: Raleigh, NC
|
Posted: Fri Nov 07, 2008 10:03 am Post subject: |
|
|
Quote: | Yes, that is in an interesting chart. On the x-axis, the lead is for both teams, right? So when the home team is up by 10, the home data point is +10 but the away team is at the -10 point, right? Might be interesting to plot them both as +10 so the two points are on the same x-value. |
Actually it's done as you described. When either team is leading by 10 points they are both at +10 on the x-axis. Unless I messed something up, of course. That's the intention at least.
Quote: | I noticed on your 2007-2008 data post you mentioned that you're missing data on 47 games. Presumably those are the more challenging games in terms of personnel with the same last name. I imagine you have already, but if not please leverage my matchup data to help resolve those last few games. If you notice any errors in my parsing or the manual resolution of those names (e.g. Ray and Tony Allen, or the Davis triplets on the Clippers this season, let me know. |
When the data is available, the name resolution has worked out very well, actually. Roughly 1% of the games I don't have this data at all, while the rest stem from some players having no events (so I have no idea if they're on the court or not). Some of this is a data issue, I think, but also some of it is players not having events in overtime games. I'm working on data collection for this year, so I'm hoping I can resolve those things and then work them into last year's code base.
At some point I'll use your matchups and see if I can get everything to come together. Since we're on the topic, how have you overcome the issue of players having no events in overtime? I was thinking I might need to use the box scores to figure out who is missing minutes from OT. _________________ I am a basketball geek. |
|
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
|