|
APBRmetrics The statistical revolution will not be televised.
|
View previous topic :: View next topic |
Author |
Message |
Eli W
Joined: 01 Feb 2005 Posts: 402
|
Posted: Tue Nov 13, 2007 9:06 am Post subject: Pomeroy's rates vs. Hollinger's rates |
|
|
I decided to start a separate thread on this so as not to take over the discussion of updates at Basketball-Reference.
davis21wylie2121 wrote: | I strongly favor using Pomeroy's definition of usage, assist, and TO rate, etc. They are both attempting to answer the same question (ex: on what % of his poss. did a player turn the ball over?), but Ken's formulas do it in a much more refined (and therefore accurate) way. Hollinger's definitions are overly simplistic and needn't be anything more than a stopgap if you don't have all the necessary stats to calc Ken's. Since we have those stats at our disposal, why not go with the better method? |
First, at least with assists, I don't think they are trying to answer the same question at all - I talked more about that here. For a quick example, consider a player that gets few touches on offense but gets an assist on a high percentage of those touches. Such a player would rate low in Pomeroy's Assist Rate but high in Hollinger's Assist Ratio. Neither one of these is "right," they're just measuring different things.
Second, what are Pomeroy's formulas for usage and turnover rate, and what extra stats are they using that Hollinger doesn't? I don't see the formulas listed on this page (is he using DeanO's formulas?):
http://kenpom.com/blog/index.php/weblog/individual_stats/ _________________ Eli W. (formerly John Quincy)
CountTheBasket.com |
|
Back to top |
|
|
asimpkins
Joined: 30 Apr 2006 Posts: 245 Location: Pleasanton, CA
|
Posted: Tue Nov 13, 2007 11:21 am Post subject: |
|
|
Yes, their assist measures are certainly measuring two different things. In an ideal world, I'd like both of them to be offered, but if I can only have one then I think Pomeroy's answers a more important question.
My suggestion would be to keep Hollinger's Assist Rate also -- which shows what a player is likely to do when get gets the ball -- and group it together with a bunch of similar rates: how likely is he to go to the foul line? to shoot a three? to turn it over? etc. You could get as detailed as you like with this.
As for usage and turnover. My impression was that the difference was that Pomeroy's percentages didn't count assists as using a possession -- only shooting, free throws, or turnovers. This keeps everything neat because you can't have two people, the shooter and passer, credited for using the same possession.
Again, just an alternate way of doing things, but I like it a bit more because of the dubious natures of assists. Again, I'd be all for expanding the Advanced Stats page to another table and offering more instead of less. |
|
Back to top |
|
|
Neil Paine
Joined: 13 Oct 2005 Posts: 774 Location: Atlanta, GA
|
Posted: Tue Nov 13, 2007 12:40 pm Post subject: |
|
|
Yeah, I should say that only "usage" and "TO rate" are technically trying to measure the same thing -- although both "assist rates" are trying to tell you about passing tendencies for a player. But even if you used the Hollinger concept for assist rate (Ast/Poss), you need a better possessions estimator in the denominator. What Hollinger does in all three of his rates is apply some variation on the team possession formula to individual players (and for usage it's not even the proper formula, he multiplies assists by 1/3 to "avoid overvaluing PGs"), which is entirely inappropriate. In the denominator, we need actual individual possessions, which is why Ken is correct to use Dean's formula from BoP. (Incidentally, that formula does include assists as a partial possession consumed, but it splits out possession credit between the passer and the shooter.)
As for assist ratio, don't you think that the % of teammate FG a player assists while on the floor is a better indicator of passing than the % of his own possessions he devoted to assists? Both speak to tendencies, but the latter says much less about "passing skill". You can devote a large % of your poss to assists and still be a bad passer if you never shoot and never do anything with the ball. However, under Ken's definition, you have to be actively engaged in the offense as a passer to score highly. To me, that definition says a lot more about passing than Hollinger's.
Last edited by Neil Paine on Tue Nov 13, 2007 1:05 pm; edited 1 time in total |
|
Back to top |
|
|
Neil Paine
Joined: 13 Oct 2005 Posts: 774 Location: Atlanta, GA
|
Posted: Tue Nov 13, 2007 12:48 pm Post subject: |
|
|
Oh, and Re: ASimpkins - I, too, would rather see the KenPom definitions, but coupled with tendency stats like Bob Chaikin's touches. Bob breaks down touches per minute, along with the percentage of touches devoted to shooting, passing, drawing fouls, and committing turnovers. Adding those under advanced stats would allow us to retain the superior passing metric (KenPom's "Assist Rate") while also giving us information on player tendencies (like John's "Assist Rate" does). I know that, roughly, Touches = FGA + TO + (Ast/0.17) + (FTA/1.15). The 1.15 comes from the league FTA/PF, but I don't know how the 0.17 is derived. If we can get the formula straight, it would probably solve this dilemma for everyone. |
|
Back to top |
|
|
kjb
Joined: 03 Jan 2005 Posts: 865 Location: Washington, DC
|
Posted: Tue Nov 13, 2007 1:28 pm Post subject: |
|
|
I like to use an adjusted Hollinger usage stat. One thing that has bugged me about his Usage is that it doesn't add up to the team totals. If you multiply Hollinger's basic formula by .82 totals from individual players will then add up to a total very close to their team total. In my opinion, that's a better approach than talking about a player's usage at rates much higher than the team's possession total.
Hollinger's Usage sorta gets at plays, but not when it includes assists. Unless you're going to put in a deflator for assisted FGs. |
|
Back to top |
|
|
Carlos
Joined: 21 Jan 2005 Posts: 64 Location: Montevideo, Uruguay
|
Posted: Tue Nov 13, 2007 1:47 pm Post subject: |
|
|
davis21wylie2121 said Quote: | both "assist rates" are trying to tell you about passing tendencies for a player. |
Well, I would say that Pomeroy's assist rate is telling something about the role of the player within the offense as much as about the player own tendencies. In that sense, it's basically a product of Assist Rate * Usage Rate (the Hollinger kind), if I understand it right. But it would be interesting to compare Pomeroy's Usage to Hollinger Usage Rate (my guess is that guards will have much lower usage and bigs higher). |
|
Back to top |
|
|
Eli W
Joined: 01 Feb 2005 Posts: 402
|
Posted: Tue Nov 13, 2007 3:35 pm Post subject: |
|
|
davis21wylie2121 wrote: | But even if you used the Hollinger concept for assist rate (Ast/Poss), you need a better possessions estimator in the denominator. What Hollinger does in all three of his rates is apply some variation on the team possession formula to individual players (and for usage it's not even the proper formula, he multiplies assists by 1/3 to "avoid overvaluing PGs"), which is entirely inappropriate. In the denominator, we need actual individual possessions, which is why Ken is correct to use Dean's formula from BoP. (Incidentally, that formula does include assists as a partial possession consumed, but it splits out possession credit between the passer and the shooter.) |
For possession usage, I agree that DeanO's formula is more theoretically sound than Hollinger's. But I don't think using (FGA + .44*FTA + Ast + TO) as the denominator for Hollinger's Assist Ratio is a problem. Assists have to be in the denominator since they're in the numerator - otherwise the rate wouldn't make sense and it wouldn't be a percentage of anything.
davis21wylie2121 wrote: | As for assist ratio, don't you think that the % of teammate FG a player assists while on the floor is a better indicator of passing than the % of his own possessions he devoted to assists? Both speak to tendencies, but the latter says much less about "passing skill". You can devote a large % of your poss to assists and still be a bad passer if you never shoot and never do anything with the ball. However, under Ken's definition, you have to be actively engaged in the offense as a passer to score highly. To me, that definition says a lot more about passing than Hollinger's. |
I disagree. They speak to different things, both of which are aspects of passing skills. You can cherry pick examples where each has its quirks - for instance, with Ken's formula, players that dominate the ball and take a ton of shots can actually rate high just by virtue of the fact that they'll assist on many of the few shots that their teammates get to take when they're on the court. And if you just use Ken's formula, you'll miss the fact that Ben Wallace is actually a pretty good passer. I think each formula gives us useful and different information.
davis21wylie2121 wrote: | Oh, and Re: ASimpkins - I, too, would rather see the KenPom definitions, but coupled with tendency stats like Bob Chaikin's touches. Bob breaks down touches per minute, along with the percentage of touches devoted to shooting, passing, drawing fouls, and committing turnovers. Adding those under advanced stats would allow us to retain the superior passing metric (KenPom's "Assist Rate") while also giving us information on player tendencies (like John's "Assist Rate" does). |
I don't see why you would be opposed to Hollinger's Assist Ratio but in favor of Chaikin's %Pass. They're basically the same thing - assists per touch. Both count shots and turnovers as touches. The only difference is that Hollinger counts assists as touches, while Chaikin counts both assists and passes that aren't assists (which he estimates). Since Chaikin's pass estimator (somewhere around 6*Ast) doesn't vary from teammate to teammate and appears to be very similar from team to team (though perhaps not season to season), players are going to be rated similarly. I like both measures, though I prefer Hollinger's since the formula is publicly available. _________________ Eli W. (formerly John Quincy)
CountTheBasket.com |
|
Back to top |
|
|
Neil Paine
Joined: 13 Oct 2005 Posts: 774 Location: Atlanta, GA
|
Posted: Tue Nov 13, 2007 3:58 pm Post subject: |
|
|
I'm opposed to Hollinger's and not BobC's because Bob never claims that his touches breakdown is measuring anything but tendencies. On the other hand, Hollinger openly uses his formula to measure passing skill, when in fact it's nothing but a measure of passing tendency. I also like the fact that BobC's method can tell you what a player does with all of his touches (the only options are shoot, get fouled, pass, or turn it over).
I was thinking about an ideal "assist rate," though, and you are right when you say:
Quote: | One objection to the previous approach would be to say that it’s not just made shots that could have been assisted - missed shots also could have been assisted FGM if a better pass was thrown that led to the shot being made. This would be an argument for including all teammate shot attempts (or perhaps just misses? or only those shots coming after a pass from the player in question?) among the assist opportunities. |
Perhaps the formula should be adjusted to represent not the % assisted of teammate FGM while a player is on the floor, but rather the % assisted of teammate FGA while a player is on the floor. This would account for pass quality, since the probability of making a shot off a pass is at least somewhat dependent on the quality of that pass. |
|
Back to top |
|
|
Neil Paine
Joined: 13 Oct 2005 Posts: 774 Location: Atlanta, GA
|
Posted: Tue Nov 13, 2007 4:09 pm Post subject: |
|
|
Carlos wrote: | But it would be interesting to compare Pomeroy's Usage to Hollinger Usage Rate (my guess is that guards will have much lower usage and bigs higher). |
My guess is that it doesn't differ that much, but Dean and Ken's Usage Rate (%Poss) is much more theoretically sound than John's version. Dean's is a simple, intuitive concept: when you are on the floor, what percentage of your team's possessions do you consume? Since basketball teams have 5 players on the floor at once, the average is always 20%. We can also use this definition of usage when dealing with "skill curves" (the relationship between usage and efficiency), because all of Dean's research in that area has been focused on the link between %Poss and ORtg.
Meanwhile, Hollinger's usage is basically pace-adjusted "fake possessions" per 40 minutes. I say "fake possessions" because the term FGA + 0.44*FTA + 0.33*AST + TO does not represent the number of real-life possessions a player has used. And the presence of the 1/3 multiplier for assists (so that the results will "look right") only makes the formula even more absurd. If you've got the individual and team data (easily available from Dougstats, etc.), there's no excuse to settle for what Hollinger calls "usage" over %Poss. |
|
Back to top |
|
|
Eli W
Joined: 01 Feb 2005 Posts: 402
|
Posted: Tue Nov 13, 2007 4:18 pm Post subject: |
|
|
davis21wylie2121 wrote: | I'm opposed to Hollinger's and not BobC's because Bob never claims that his touches breakdown is measuring anything but tendencies. On the other hand, Hollinger openly uses his formula to measure passing skill, when in fact it's nothing but a measure of passing tendency. |
I still think you're short-changing Hollinger. Chaikin's formula does just measure passing tendency (I should correct something I said earlier - Chaikin's %Pass is not just Hollinger's Assist Ratio with non-assist passes added to the denominator - it also adds non-assist passes to the numerator), but by limiting things to assists as opposed to all passes, I think Hollinger's formula also measures passing skill (in addition to tendency) to some degree. A player doesn't think, "Hmm, should I take a shot in this situation, or get an assist?" in the same sense that he may think "Should I take a shot in this situation, or make a pass?" Assists themselves have a skill element to them - they are a proxy for "good" passes. They're not a perfect proxy, but if we believe that generally better passes are more likely to lead to made shots than worse passes, then the percent of a player's passes that are assists (something Chaikin's formula doesn't measure but Hollinger's captures indirectly) is a skill.
Some time soon in my blog I'm going to look at all this stuff empirically to see how each different stat applies to players from recent seasons. _________________ Eli W. (formerly John Quincy)
CountTheBasket.com |
|
Back to top |
|
|
bchaikin
Joined: 27 Jan 2005 Posts: 690 Location: cleveland, ohio
|
Posted: Tue Nov 13, 2007 7:28 pm Post subject: |
|
|
for what it's worth i've been charting cavs games this season (i'm in cleveland) and counting touches for lebron james in particular since you know he'll be seeing the ball alot on offense. for example in the CLE/NYK game of 11/2/07 james played 44 minutes and got 95 touches, threw 53 passes, got fouled 14 times, took 28 shots, and committed 2 turnovers. 53 + 14 + 28 + 2 = 97, but he got fouled twice in the act of shooting and hit the shot (i.e. got fouled 1 time and had 1 FGA but just 1 touch). with his 53 passes he got 7 assists...
but in the CLE/DAL game of 10/31/07 in 36 minutes james got 64 touches, threw 43 passes, took 11 shots, got fouled 5 times, and committed 5 turnovers (2 of which were offensive fouls). 43 + 11 + 5 + 5 = 64. but with those 43 passes he got only 4 assists...
if you pick one or two players in a game (that you know will get alot of touches) it's easy to get an idea of their tendencies once you log a number of games. i also charted eddy curry in the CLE/NYK game and in 33 minutes he got 26 touches, with 16 shots, he got fouled 5 times, committed 1 turnover (low for him), but he threw just 4 passes (got 1 assist), one of which was actually a handoff more than a pass and thus really never had a chance to be an assist...
i tend to only count passes that i think have a chance to be an assist (even frontcourt passes on out of bounds plays), i think to myself had the player receiving the pass immediately shot and scored would the passer have been credited with an assist. thus passes in the backcourt when a team brings the ball up the floor when there are no defenders around i don't count as "passes", nor do i usually count a handoff of the ball as a pass...
you do this enough and you'll come up with good approximations for calculating player touches... |
|
Back to top |
|
|
Kevin Pelton Site Admin
Joined: 30 Dec 2004 Posts: 979 Location: Seattle
|
Posted: Wed Nov 14, 2007 3:43 pm Post subject: |
|
|
My perspective is that to relate usage as quantified to the conceptual definition of usage we generally have, it makes more sense to entirely throw out assists and offensive rebounds and use the play definition instead of the possession definition.
That means using simply FGA + (.44*FTA) + TO divided by the same measures at the team level. |
|
Back to top |
|
|
jkubatko
Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH
|
Posted: Fri Nov 16, 2007 3:58 pm Post subject: |
|
|
Kevin Pelton wrote: | My perspective is that to relate usage as quantified to the conceptual definition of usage we generally have, it makes more sense to entirely throw out assists and offensive rebounds and use the play definition instead of the possession definition.
That means using simply FGA + (.44*FTA) + TO divided by the same measures at the team level. |
I actually like that. Simple, yet meaningful. Would you do a similar thing for, say, a turnover percentage? Something like TOV / (FGA + 0.44*FTA + TOV)? _________________ Regards,
Justin Kubatko
Basketball-Reference.com |
|
Back to top |
|
|
Kevin Pelton Site Admin
Joined: 30 Dec 2004 Posts: 979 Location: Seattle
|
Posted: Fri Nov 16, 2007 4:47 pm Post subject: |
|
|
Yep. The nice thing about that in terms of Turnover% is you can know that TO% plus FTA% plus FGA% add to 100. |
|
Back to top |
|
|
gabefarkas
Joined: 31 Dec 2004 Posts: 1313 Location: Durham, NC
|
Posted: Thu Nov 29, 2007 9:38 am Post subject: |
|
|
The thing is, you aren't considering everything that a player does to impact the game. You're just considering shot attempts and turnovers. Most of the time when a player has the ball in his hands, he passes it (or attempts to pass it, resulting in a turnover). Those passes/attempted-passes lead to either turnovers, assists, or nothing.
If you include assists in the measurement, it allows you to consider everything a player does with the ball, including "nothing". Using the literal definition of the word "possession" (instead of the basketball definition), you could calculate:
100 = Assist% + TO% + FGA% + FTA% |
|
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
|