Page 3 of 4
Re: Possible Steps to improve SPM
Posted: Wed Oct 22, 2014 5:08 pm
by schtevie
A bit late to this conversation, but a couple of points.
First, Jeremias, I am trying to understand the empirical dimension of the over and underrating. If Dirk Nowitzki and Tony Parker (their lineups) are the most offensively and defensively underrated (in absolute terms) to the tune of about 1000 points since 2002 (inclusive of the 2001-02 season?) I calculate this "error" in terms of points per 100 possessions to be less than 0.02 in both instances? That is, it appears to be rounding error. Is there a math error underlying these calculations, or is this the dimension of what we're talking about here?
Second, as a possible positive suggestion. I have always thought that there would be value in adding anthropometric and position data to the xRAPM framework (height most obviously). Is there a compelling reason for not doing so?
Informing this belief is the apparent fact that positional averages are not zero, and a pretty simple and coherent story can be told that height/size and court "geography" explaining the observed trends, what are roughly: increasing (decreasing) average defensive (offensive) value attending positions 1 to 5.
Edit/P.S. Oops. Off by a factor of one hundred, I think, on the estimate. So, if I (re)do the math for Nowitzki, the underestimate is 1.04 points per game. That's not nothing.
Re: Possible Steps to improve SPM
Posted: Wed Oct 22, 2014 8:26 pm
by J.E.
schtevie wrote:Second, as a possible positive suggestion. I have always thought that there would be value in adding anthropometric and position data to the xRAPM framework (height most obviously). Is there a compelling reason for not doing so?
Height (and weight) are variables in SPM, which gets fed into xRAPM. I also have a height*3PA interaction term. Many other things, like "size of hands", are hard to come by
Running seperate regressions for the 5 positions is something I can, again, look into. The last time I tried it I didn't get improved retrodiction results, though. I think it's because you're then working with a smaller sample size (1/5). Although I can imagine a hybrid of standard SPM and 5-position-SPM might be able outperform standard SPM
Re: Possible Steps to improve SPM
Posted: Thu Oct 23, 2014 12:31 am
by schtevie
I am supposing that the basic story is that, all else equal, above-average positional height/length (and perhaps interactions with weight, for some positions) helps to explain above-average positional performance. And if separate positional regressions were run for offense and defense (as missing variables make all else not equal) I can imagine that that this would improve results.
Out of curiosity, is there a reference for the specification of the "positional" regressions that didn't yield improved retrogression results?
Furthermore, if this basic conjecture is true, one might expect, for players playing few minutes, that the variability of the anthropometric characteristics of counterparts would be a relevant factor, with players facing dissimilar-sized counterparts. From the PbP data, it would, at least, be possible to "estimate" the average height/weight of counterpart players, no (assuming, for example, that player counterparts are based upon assumed relative height match-ups)?
Re: Possible Steps to improve SPM
Posted: Mon Oct 27, 2014 3:14 am
by AcrossTheCourt
I learned a few things from doing one this summer. Once I post it sometime later (November?) I'll discuss some things.
One that thing that could be an improvement is having some database somewhere with all the miscellaneous stats like bad pass turnovers, charges, goaltending, etc. I don't have all the stats I want.
Re: Possible Steps to improve SPM
Posted: Mon Oct 27, 2014 3:30 pm
by Crow
Bad pass turnovers are listed at 82games.
Re: Possible Steps to improve SPM
Posted: Mon Oct 27, 2014 5:57 pm
by J.E.
Mike G wrote:If you don't correct for home scorekeeping bias, you can overrate certain players. Ben Wallace is a prime example.
In 2002-03, he blocked 140 shots at home and 90 on the road. Was he "given" an extra 50 or so blocks at home?
In '04 again an "extra" 52 blocks at home, in equal minutes as away.
In 2005, he had 58 blocks in away games and 118 in home games -- more than twice as many!
For his career, he avg'd 2.68 Blk/36 at home and 2.10 on the road -- 2.39 overall.
If you correct by the factor (2.10/2.39), then his career Blk% may be not 5.0 but 4.4
How does that affect his SPM ?
I just looked at my code and turns out I'm actually already doing this, in some way. I actually don't count any home blocks. Might be a little harsh, maybe multiplying home blocks with a factor <1 is better. The question, to me, is whether home scorekeeping bias affects all players (of one team) equally. I could imagine that some players have more home scorekeeping bias in their stats than others, from the same team. Think Rondo and his assists in Boston. I think there was at least one instance where the league took one of his assists away, after the game
When it comes to modern day home scorekeeping bias with blocks: Anthony Davis had 123 home blocks and 66 away blocks last season
Re: Possible Steps to improve SPM
Posted: Mon Oct 27, 2014 9:57 pm
by AcrossTheCourt
Crow wrote:Bad pass turnovers are listed at 82games.
I know. And b-ref.
It'd just be nice to have one source for all these miscellaneous stats instead of trying to scrape things from multiple sites. And charges can only go back to 2006, right? But even just for a person who's curious and wants to look up the leaders in random stats ... that's not completely possible now. Stats.NBA.com has some stuff, but not everything.
I thought of one area of improvement: adjusting for year. Perhaps some stats are more "important" or indicative of value than they were in, say, 2004. Any adjustment for that would be interesting because we're now at the point where we have a large time-frame to work in and basketball has changed a lot since pbp data is first available.
Re: Possible Steps to improve SPM
Posted: Tue Oct 28, 2014 2:52 pm
by mystic
J.E. wrote:The question, to me, is whether home scorekeeping bias affects all players (of one team) equally.
I think that is hard to answer, because the scorekeeping bias is restricted to basically blocks, assists and somewhat steals (going by the traditional boxscore stats). There also seems to be a question of interpretation involved regarding the decision to call something a steal or a block, depending on the scorekeeper. Last season there was a -0.2 correlation between the homebias for blocks and steals, meaning, teams which have more blocked shots at home than on the road tend to have less steals at home than on the road.
Overall, there is a positive correlation between higher blk% and higher ast% at home than on the road.
So, none of that answers your question, but maybe that helps to see some of the difficulties or may even give some inspiration to answer the question. It may be feasible to check whether certain players are getting numbers from teammates assigned to them, like giving the rebound to player x instead of player y, or the blocked shot or the assist or the steal, but that would probably require heavy video studies and might not be as easily found in the numbers alone.
I tried to adjust for a generally better performance level at home and then normalized the homebias to league average levels. Here are the teams in order from most to least homebias scorekeeping for last season based on that method:
Code: Select all
CLE 3.1
CHA 1.4
SAS 1.2
PHI 1.1
BOS 0.9
NOP 0.8
DET 0.7
IND 0.7
SAC 0.5
NYK 0.4
MEM 0.2
LAL 0.2
DEN 0.1
MIL 0.1
ORL 0.1
OKC 0.0
POR -0.1
HOU -0.2
BRK -0.4
LAC -0.5
ATL -0.6
MIN -0.6
GSW -0.7
CHI -0.9
TOR -1.1
WAS -1.1
PHO -1.2
UTA -1.2
MIA -1.4
DAL -1.7
Cleveland had 3.1 standard deviations increased numbers over expectation at home, Dallas had 1.7 standard deviations below and was the team with the least homebias (in fact, they had essentially a clear "road bias"). I have that bias integrated in my current SPM version, helps a bit with the predictions.
Btw, that the Suns are pretty low is not uncommon. For example, Nash had a higher assist rate on the road than at home, overall he assisted about 44% of the home field goals while being on the court, while assisting 48% of the road field goals. Some of that may be explainable by him being forced to create more for his teammates on the road, but from my perspective the scorekeeper for the Suns is responsible for that (though, they are about league average when it comes to giving out blocked shots).
Re: Possible Steps to improve SPM
Posted: Tue Oct 28, 2014 6:25 pm
by Mike G
There are 2 scorekeeper variations to consider:
- Overall generosity: To give an assist or to not give an assist (or block)
- Home bias: To be more generous to the home team.
Those Nash Suns were especially frugal in giving assists, both to the Suns and to visitors. They still had a substantial home bias, in that they were even stingier toward opponents.
Ranked by 'assist generosity' -- defined as aggregate opponent Ast/FG in 2013-14, also showing home and away Ast% and the ratio:
Code: Select all
tm OpAst% Genr hAst% aAst% H/A tm OpAst% Genr hAst% aAst% H/A
Cle .658 1.13 .629 .515 1.22 Tor .581 1.00 .594 .567 1.05
Phl .653 1.12 .610 .542 1.13 Brk .578 .99 .595 .575 1.03
Det .624 1.07 .568 .509 1.11 Uta .570 .98 .526 .603 .87
LAC .622 1.07 .645 .610 1.06 Hou .570 .98 .565 .560 1.01
Sac .621 1.07 .512 .510 1.00 Den .569 .98 .602 .566 1.06
LAL .614 1.05 .666 .611 1.09 Cha .568 .97 .619 .576 1.07
NOH .604 1.04 .567 .558 1.01 Min .565 .97 .613 .618 .99
Atl .601 1.03 .692 .642 1.08 Chi .559 .96 .681 .627 1.09
Was .600 1.03 .605 .597 1.01 Mem .558 .96 .561 .588 .95
Okl .598 1.03 .585 .538 1.09 GSW .557 .96 .608 .573 1.06
Mil .595 1.02 .602 .590 1.02 NYK .551 .95 .522 .564 .93
Mia .591 1.01 .555 .622 .89 SAS .539 .93 .655 .585 1.12
Orl .591 1.01 .569 .574 .99 Ind .535 .92 .569 .550 1.04
Bos .585 1.00 .597 .554 1.08 Phx .534 .92 .470 .517 .91
Dal .584 1.00 .585 .606 .97 Por .516 .88 .597 .591 1.01
avg .583 .592 .575 1.032
Genr = OpAst%/.583 (league avg)
Note that the most generous scorekeepers also have the biggest home bias.
Maybe these guys are trying to mask their home bias by also giving more than usual assists to visitors?
Re: Possible Steps to improve SPM
Posted: Tue Oct 28, 2014 6:42 pm
by fpliii
Does a bias for rebounding/turnovers exist?
Re: Possible Steps to improve SPM
Posted: Tue Oct 28, 2014 7:22 pm
by mystic
Mike G wrote:
They still had a substantial home bias, in that they were even stingier toward opponents.
That is something really, really hard to gauge from boxscore data alone, because different plays have a different probability of generating an assist. The Suns ran a lot of p&r, which is especially assist-prone just due to the nature of the play.
fpliii, when using data corrected for different playing level at home and on the road I found no statistical significant bias for rebounds and turnovers on the team-level. But that would be the point JEs question would come into play: May it be possible that some players are getting special treatment? If there is a group of players after a missed shot battling for the ball, a scorekeeper could just assign that rebound to the "wrong" player. A similar thing can happen for turnovers, where a passing turnover would actually be assigned to another player and not to the passer. A similar thing can happen for a steal, when a player might touch the ball, but a teammate picks up the lose ball; the scorekeeper basically can decide by himself which player gets the steal.
Theoretically, checking rates for home and road games for individual players in comparison to teammates may reveal such thing.
Re: Possible Steps to improve SPM
Posted: Tue Oct 28, 2014 8:29 pm
by fpliii
mystic wrote:fpliii, when using data corrected for different playing level at home and on the road I found no statistical significant bias for rebounds and turnovers on the team-level. But that would be the point JEs question would come into play: May it be possible that some players are getting special treatment? If there is a group of players after a missed shot battling for the ball, a scorekeeper could just assign that rebound to the "wrong" player. A similar thing can happen for turnovers, where a passing turnover would actually be assigned to another player and not to the passer. A similar thing can happen for a steal, when a player might touch the ball, but a teammate picks up the lose ball; the scorekeeper basically can decide by himself which player gets the steal.
Theoretically, checking rates for home and road games for individual players in comparison to teammates may reveal such thing.
Thanks for the response. I think you're correct, it's the same phenomenon as with steals, and checking rates would hint at their prevalence.
Do you think in general it's safe to use road rates, and just ignore home totals? Or do you believe there are legitimate home/away performance splits in a large sample (I'll reread the thread, apologies if this has already been addressed)?
Re: Possible Steps to improve SPM
Posted: Tue Oct 28, 2014 10:06 pm
by Mike G
Referees may also have a home court bias. And a player who gets more blocks at home may also get fewer fouls at home.
A scorekeeper registering an "extra" assist or block doesn't really affect the game. But a referee certainly does when he calls it a clean play (blocked shot) or a foul. And since most teams have more success at home, it's to be expected that most players will have better stats at home. We don't know if they're actually playing better, or if refs are giving them more calls.
I doubt that there's much to find in the way of certain players getting credited for lots more rebounds that should be given to teammates. Steals, maybe there are inflated home totals or favoring certain players over their teammates, in some cases; I haven't found them to be great.
Before b-r.com was found to have home-away splits, we wondered here if they might somehow play a different style on the road; and whether other teams playing in Phoenix were somehow thriving on unassisted shots, due to the Suns' defense. But it was much simpler than that: the Suns' scorekeepers just didn't give out assists very freely.
Re: Possible Steps to improve SPM
Posted: Wed Oct 29, 2014 5:11 am
by AcrossTheCourt
If you're worried someone like Davis is getting significantly better treatment at home with more stats counted, then test this by creating two versions of Davis: home and away. Since HCA is already accounted for in the model, you should see if there's any real difference.
New Orleans had an issue like this with another superstar: Chris Paul and his assists. I can't find any good articles on this though, but via stats.NBA.com, for example, Paul had a 55.2 assist% in 2009 at home and a 47.4 assist% on the road. This is already correcting for having a worse offense on the road ... it's percentage of teammate field goals assisted.
Inflating stats from scorekeepers is no new thing:
http://deadspin.com/5345287/the-confess ... corekeeper
Here's one idea of fixing this: take a player's road stats and then to fill in all the home games apply a league average home/road difference (average road blocks + league average home/away diff.) Or reduce that league average different a little if you think that's prudent.
Re: Possible Steps to improve SPM
Posted: Wed Oct 29, 2014 8:02 am
by mystic
fpliii wrote:
Do you think in general it's safe to use road rates, and just ignore home totals? Or do you believe there are legitimate home/away performance splits in a large sample (I'll reread the thread, apologies if this has already been addressed)?
The latter, I think there are indeed players being better at home than on the road, and at that more players being that way than the other way around. I make a small adjustment based on those previously presented "homebias" numbers (though, it comes from teamlevel data and I apply that to individual players, which might not be the best way to do such thing) to blocked shots and assists (to a smaller degree). Gave me an 0.02 RMSE improvement in my retrodiction dataset ... though, it is a minor thing in the end. That might look different in an analysis based on the individual player-level; I can imagine that some players are getting special treatment from the home scorekeeper or even in general, as JE suggested.