The debut and popularization of BPM
Re: The debut and popularization of BPM
dsmok, Have you ever looked into trying to correlate BPM with xRAPM instead of RAPM? xRAPM is more reliable and predictive. I know there are issues with that approach, but it would be interesting to see how the results would differ if the stat used xRAPM rather than RAPM.
Re: The debut and popularization of BPM
Yeah, big issues with that approach. xRAPM already is significantly box score based, so I'd sort of be back-calculating J.E.'s values for his box score metric.colts18 wrote:dsmok, Have you ever looked into trying to correlate BPM with xRAPM instead of RAPM? xRAPM is more reliable and predictive. I know there are issues with that approach, but it would be interesting to see how the results would differ if the stat used xRAPM rather than RAPM.
14 year RAPM should be more accurate than xRAPM, for what we're trying to measure.
Re: The debut and popularization of BPM
it probably has a lot to do with how basketball was somewhat different then.
how was basketball different back then?...
in 82-83 nba teams averaged 8.5 ST/100poss, 5.4 BS/100poss. in 92-93 it was 8.7 and 5.3 per 100poss, in 02-03 it was 8.6 and 5.4 per 100poss, in 12-13 it was 8.3 and 5.5 per 100poss...
rather than risk, say, the regression overweighting blocks and steals to try to account for good defensive teams. If most of the defense is other things, and blocks and steals don't correlate that well with actual defensive quality, I don't want to overweight them.
steals have a definitive value, a defensive stop that is a zero point possession as defined by the average pts/poss for that season. a blocked shot has the value of a defensive stop for those retrieved as defensive rebounds...
how do you account for individual player defense outside of steals and blocks? such as for moses malone in 82-83 when he was named all-D 1st team?...
how was basketball different back then?...
in 82-83 nba teams averaged 8.5 ST/100poss, 5.4 BS/100poss. in 92-93 it was 8.7 and 5.3 per 100poss, in 02-03 it was 8.6 and 5.4 per 100poss, in 12-13 it was 8.3 and 5.5 per 100poss...
rather than risk, say, the regression overweighting blocks and steals to try to account for good defensive teams. If most of the defense is other things, and blocks and steals don't correlate that well with actual defensive quality, I don't want to overweight them.
steals have a definitive value, a defensive stop that is a zero point possession as defined by the average pts/poss for that season. a blocked shot has the value of a defensive stop for those retrieved as defensive rebounds...
how do you account for individual player defense outside of steals and blocks? such as for moses malone in 82-83 when he was named all-D 1st team?...
Re: The debut and popularization of BPM
Are you, in your simulations, giving full credit for a defensive stop to the player that got the rebound of a missed shot? And similar for a steal? Or how are you doing it?bchaikin wrote:rather than risk, say, the regression overweighting blocks and steals to try to account for good defensive teams. If most of the defense is other things, and blocks and steals don't correlate that well with actual defensive quality, I don't want to overweight them.
steals have a definitive value, a defensive stop that is a zero point possession as defined by the average pts/poss for that season. a blocked shot has the value of a defensive stop for those retrieved as defensive rebounds...
how do you account for individual player defense outside of steals and blocks? such as for moses malone in 82-83 when he was named all-D 1st team?...
My approach is to figure out how much weight should be given to the player who gets the rebound, and how much should be given to the other 4 players on the court, through the regression. Same for all other stats.
Re: The debut and popularization of BPM
Are you, in your simulations, giving full credit for a defensive stop to the player that got the rebound of a missed shot?
there is no "credit" given for anything - a player on offense makes/misses a shot based on his FG%s (2pt, 3pt), the defender's defFG% allowed, and the shot blocking abilities of the 5 defenders on the floor...
And similar for a steal?
a player/defender who gets a steal gets the ball, and the simulation continues with the ball in that player's possession...
My approach is to figure out how much weight should be given to the player who gets the rebound, and how much should be given to the other 4 players on the court,
for a missed shot the simulation looks at the rebounding ratings for all 10 players on the floor (5 off ratings, 5 def ratings), adds them up, assigns a range for each, picks a random number in those ranges, then assigns a rebounder. if it's an offensive rebound, the simulation continues with that player having the ball. if it's a defensive rebound, the team possession for that player's team begins...
there is no "credit" given for anything - a player on offense makes/misses a shot based on his FG%s (2pt, 3pt), the defender's defFG% allowed, and the shot blocking abilities of the 5 defenders on the floor...
And similar for a steal?
a player/defender who gets a steal gets the ball, and the simulation continues with the ball in that player's possession...
My approach is to figure out how much weight should be given to the player who gets the rebound, and how much should be given to the other 4 players on the court,
for a missed shot the simulation looks at the rebounding ratings for all 10 players on the floor (5 off ratings, 5 def ratings), adds them up, assigns a range for each, picks a random number in those ranges, then assigns a rebounder. if it's an offensive rebound, the simulation continues with that player having the ball. if it's a defensive rebound, the team possession for that player's team begins...
Re: The debut and popularization of BPM
Interesting work!
Are you accounting for diminishing returns on rebounds, offensive skill curves (usg vs. eff tradeoff), and such? How long does it take for your simulations to run & results to stabilize?
Are you accounting for diminishing returns on rebounds, offensive skill curves (usg vs. eff tradeoff), and such? How long does it take for your simulations to run & results to stabilize?
Re: The debut and popularization of BPM
Are you accounting for diminishing returns on rebounds
not sure what you are asking here...
offensive skill curves (usg vs. eff tradeoff)
again not sure what you are asking - the sim inputs are actual player and team stats...
and such?
such as?...
How long does it take for your simulations to run
8200 games in 50 sec, and that's on an older pc...
& results to stabilize?
if you run an 8200 game scenario and then repeat that same identical scenario you won't see more than a +/- 2 games up/down (4 game swing) difference in W-L record per 82 games, i.e. if your first run results in a W-L record of say 41-41, your subsequent runs will be between 39-43 and 43-39...
and as i asked before, how do you account for individual player defense outside of steals and blocks? such as for moses malone in 82-83 when he was named all-D 1st team?...
not sure what you are asking here...
offensive skill curves (usg vs. eff tradeoff)
again not sure what you are asking - the sim inputs are actual player and team stats...
and such?
such as?...
How long does it take for your simulations to run
8200 games in 50 sec, and that's on an older pc...
& results to stabilize?
if you run an 8200 game scenario and then repeat that same identical scenario you won't see more than a +/- 2 games up/down (4 game swing) difference in W-L record per 82 games, i.e. if your first run results in a W-L record of say 41-41, your subsequent runs will be between 39-43 and 43-39...
and as i asked before, how do you account for individual player defense outside of steals and blocks? such as for moses malone in 82-83 when he was named all-D 1st team?...
Last edited by bchaikin on Fri Jan 09, 2015 6:58 pm, edited 1 time in total.
Re: The debut and popularization of BPM
For simming an individual play, using 10 players on the court--do you account for the diminishing returns effects Eli Witus found for rebounds? http://www.countthebasket.com/blog/2008 ... g-returns/bchaikin wrote:Are you accounting for diminishing returns on rebounds
not sure what you are asking here...
offensive skill curves (usg vs. eff tradeoff)
again not sure what you are asking - the sim inputs are actual player and team stats...
And for offense? http://www.countthebasket.com/blog/2008 ... fficiency/
In other words, a lineup's effectiveness is not = the sum of its parts?
Re: The debut and popularization of BPM
For simming an individual play, using 10 players on the court--do you account for the diminishing returns effects Eli Witus found for rebounds? http://www.countthebasket.com/blog/2008 ... g-returns/
i don't see any specific examples in that article - care to examine some particular example?...
And for offense? http://www.countthebasket.com/blog/2008 ... fficiency/
here he states:
This means that lineups made up of lower-usage players fail to score at the rate projected from the players’ season ORtg’s, suggesting they can’t maintain their efficiency levels as they are forced to increase their usage... In general, for every 1% that a lineup has to increase its usage, it’s efficiency decreases by 0.25 points per 100 possessions, and vice versa.
this statement may be true for ORtg's, but i've addressed the usage vs. efficiency debate numerous times. the idea that there is some definitive decrease in player efficiency as his touches increase is simply untrue. there are many players who have similar efficiency or an increase in efficiency with an increase in touches...
demarcus cousins has increased his touches pretty much every season, and also his offensive efficiency...
chris bosh had the identical rate of offensive effciency (pts/0ptposs) his last 5 years in toronto as he's had his 4+ years in miami, despite significantly reduced touches (over 20% drop) for the latter...
steve blake had his 2nd highest rate of offensive efficiency (05-06) when he got his most touches, one of his lowest when he got his fewest (10-11)...
dale ellis in going from dallas (3 years) to seattle (first 3 years) saw a large increase in his touches and a large increase in his offensive efficiency...
how many examples does one need to show before a supposed tenet gets downgraded to a "happens sometimes"?...
i don't see any specific examples in that article - care to examine some particular example?...
And for offense? http://www.countthebasket.com/blog/2008 ... fficiency/
here he states:
This means that lineups made up of lower-usage players fail to score at the rate projected from the players’ season ORtg’s, suggesting they can’t maintain their efficiency levels as they are forced to increase their usage... In general, for every 1% that a lineup has to increase its usage, it’s efficiency decreases by 0.25 points per 100 possessions, and vice versa.
this statement may be true for ORtg's, but i've addressed the usage vs. efficiency debate numerous times. the idea that there is some definitive decrease in player efficiency as his touches increase is simply untrue. there are many players who have similar efficiency or an increase in efficiency with an increase in touches...
demarcus cousins has increased his touches pretty much every season, and also his offensive efficiency...
chris bosh had the identical rate of offensive effciency (pts/0ptposs) his last 5 years in toronto as he's had his 4+ years in miami, despite significantly reduced touches (over 20% drop) for the latter...
steve blake had his 2nd highest rate of offensive efficiency (05-06) when he got his most touches, one of his lowest when he got his fewest (10-11)...
dale ellis in going from dallas (3 years) to seattle (first 3 years) saw a large increase in his touches and a large increase in his offensive efficiency...
how many examples does one need to show before a supposed tenet gets downgraded to a "happens sometimes"?...
Re: The debut and popularization of BPM
Well, Witus found that offensive rebounds stack 80%--switch from a player getting 5% of ORB to one getting 10% of ORB, you can expect the lineup to increase ORB% by only 4%, rather than 5% you might expect. And defensive rebounds only stack at 30%: if you switch from a player getting 5% of DRB to one getting 10% of DRB, the lineup only on average increases by 1.5%.bchaikin wrote:For simming an individual play, using 10 players on the court--do you account for the diminishing returns effects Eli Witus found for rebounds? http://www.countthebasket.com/blog/2008 ... g-returns/
i don't see any specific examples in that article - care to examine some particular example?...
And for offense? http://www.countthebasket.com/blog/2008 ... fficiency/
here he states:
This means that lineups made up of lower-usage players fail to score at the rate projected from the players’ season ORtg’s, suggesting they can’t maintain their efficiency levels as they are forced to increase their usage... In general, for every 1% that a lineup has to increase its usage, it’s efficiency decreases by 0.25 points per 100 possessions, and vice versa.
this statement may be true for ORtg's, but i've addressed the usage vs. efficiency debate numerous times. the idea that there is some definitive decrease in player efficiency as his touches increase is simply untrue. there are many players who have similar efficiency or an increase in efficiency with an increase in touches...
demarcus cousins has increased his touches pretty much every season, and also his offensive efficiency...
chris bosh had the identical rate of offensive effciency (pts/0ptposs) his last 5 years in toronto as he's had his 4+ years in miami, despite significantly reduced touches (over 20% drop) for the latter...
steve blake had his 2nd highest rate of offensive efficiency (05-06) when he got his most touches, one of his lowest when he got his fewest (10-11)...
dale ellis in going from dallas (3 years) to seattle (first 3 years) saw a large increase in his touches and a large increase in his offensive efficiency...
how many examples does one need to show before a supposed tenet gets downgraded to a "happens sometimes"?...
As for the usage/efficiency--what Witus found was the average. "In general, for every 1% that a lineup has to increase its usage, it’s efficiency decreases by 0.25 points per 100 possessions, and vice versa.". That's the average situation--half of the time you get less of an effect than that, half of the time you get more. For a sim, unless you're actually simulating how shot distributions interact and assisted/unassisted shot patterns, it seems like you'd go with the average effect to be most accurate. Or handle it some other way.
---
I really appreciate the insight into what you're doing, Bob; I've seen you mention sims a lot over the years and always wondered how they worked! Interesting stuff, and impressive how fast everything runs.
Re: The debut and popularization of BPM
Well, Witus found that offensive rebounds stack 80%--switch from a player getting 5% of ORB to one getting 10% of ORB, you can expect the lineup to increase ORB% by only 4%, rather than 5% you might expect. And defensive rebounds only stack at 30%: if you switch from a player getting 5% of DRB to one getting 10% of DRB, the lineup only on average increases by 1.5%.
these are generalizations - how about a real world example?...
from 09-10 to 12-13 blazers starting PF lamarcus aldridge averaged just 7.3 defreb/48min, starting SF nicholas batum only 5.2 defreb/48min. both were average at best defensive rebounders for their respective positions...
their C those years were the likes of marcus camby, joel przybilla, and j.j. hickson, all very good to excellent defensive rebounders. then they trade for robin lopez, an excellent offensive rebounder for a C, but a worse than average defensive rebounder, and he plays major minutes (28-32 min/g). in 13-14 and 14-15 aldridge and batum see 32%-34% increases in their per minute defensive rebounding rates (up to 11.1 and 7.6 defreb/48min), but not other key blazers players like wes matthews...
so how would you model that?...
For a sim, unless you're actually simulating how shot distributions interact and assisted/unassisted shot patterns,
how would one do this? how do shot distributions "interact"? and what assisted/unassisted shot patterns are you referring to? last time i checked all assisted shots go in, only unassisted shots are missed...
it seems like you'd go with the average effect to be most accurate. Or handle it some other way.
like what other way?...
these are generalizations - how about a real world example?...
from 09-10 to 12-13 blazers starting PF lamarcus aldridge averaged just 7.3 defreb/48min, starting SF nicholas batum only 5.2 defreb/48min. both were average at best defensive rebounders for their respective positions...
their C those years were the likes of marcus camby, joel przybilla, and j.j. hickson, all very good to excellent defensive rebounders. then they trade for robin lopez, an excellent offensive rebounder for a C, but a worse than average defensive rebounder, and he plays major minutes (28-32 min/g). in 13-14 and 14-15 aldridge and batum see 32%-34% increases in their per minute defensive rebounding rates (up to 11.1 and 7.6 defreb/48min), but not other key blazers players like wes matthews...
so how would you model that?...
For a sim, unless you're actually simulating how shot distributions interact and assisted/unassisted shot patterns,
how would one do this? how do shot distributions "interact"? and what assisted/unassisted shot patterns are you referring to? last time i checked all assisted shots go in, only unassisted shots are missed...
it seems like you'd go with the average effect to be most accurate. Or handle it some other way.
like what other way?...
-
AcrossTheCourt
- Posts: 237
- Joined: Sat Feb 16, 2013 11:56 am
Re: The debut and popularization of BPM
Because you're missing key information: rebounding numbers for individuals don't always translate to the team level. If we had boxout stats, we could have more a more accurate prediction.
Namely, JJ Hickson is the most notorious "rebound stealer" out there who rarely boxes out to help his teammates grab the board while Lopez is the opposite and boxed out for Aldridge.
Namely, JJ Hickson is the most notorious "rebound stealer" out there who rarely boxes out to help his teammates grab the board while Lopez is the opposite and boxed out for Aldridge.
Re: The debut and popularization of BPM
Because you're missing key information: rebounding numbers for individuals don't always translate to the team level.
player rebounds plus team rebounds (missed shots going out of bound, missed 1st FTs, etc) don't add up to the team level? since when?...
Namely, JJ Hickson is the most notorious "rebound stealer" out there who rarely boxes out to help his teammates grab the board while Lopez is the opposite and boxed out for Aldridge.
robin lopez has always been a poor defensive rebounder for a C. if what you are saying is true, have his teams always been good defensive rebounding teams due to his ability to box out? i ask as for example last year in 13-14 portland was just an average defensive rebounding team, grabbing 74.7% of available rebounds as defensive, ranking 13th (they ranked high in defreb/100poss with 35.1, ranking 6th in the league, but this is high due to them forcing the fewest TOs of any team in the league with just 12.4 forced TO/100poss, i.e. there were more team possessions with rebounds available)...
player rebounds plus team rebounds (missed shots going out of bound, missed 1st FTs, etc) don't add up to the team level? since when?...
Namely, JJ Hickson is the most notorious "rebound stealer" out there who rarely boxes out to help his teammates grab the board while Lopez is the opposite and boxed out for Aldridge.
robin lopez has always been a poor defensive rebounder for a C. if what you are saying is true, have his teams always been good defensive rebounding teams due to his ability to box out? i ask as for example last year in 13-14 portland was just an average defensive rebounding team, grabbing 74.7% of available rebounds as defensive, ranking 13th (they ranked high in defreb/100poss with 35.1, ranking 6th in the league, but this is high due to them forcing the fewest TOs of any team in the league with just 12.4 forced TO/100poss, i.e. there were more team possessions with rebounds available)...
Re: The debut and popularization of BPM
Yes, it's been measured, and the effect is huge. Over the past 3 years, here's the data:bchaikin wrote:Because you're missing key information: rebounding numbers for individuals don't always translate to the team level.
player rebounds plus team rebounds (missed shots going out of bound, missed 1st FTs, etc) don't add up to the team level? since when?...
Namely, JJ Hickson is the most notorious "rebound stealer" out there who rarely boxes out to help his teammates grab the board while Lopez is the opposite and boxed out for Aldridge.
robin lopez has always been a poor defensive rebounder for a C. if what you are saying is true, have his teams always been good defensive rebounding teams due to his ability to box out? i ask as for example last year in 13-14 portland was just an average defensive rebounding team, grabbing 74.7% of available rebounds as defensive, ranking 13th (they ranked high in defreb/100poss with 35.1, ranking 6th in the league, but this is high due to them forcing the fewest TOs of any team in the league with just 12.4 forced TO/100poss, i.e. there were more team possessions with rebounds available)...
J.J. Hickson: 23 DRB/100 Opps, but teammates are -7 DRB/100 Opps.
Robin Lopez: 14 DRB/100 Opps, but teammates are +5 DRB/100 Opps.
So overall, Hickson's lineups are actually 3 DRB/100 Opps WORSE than Robin Lopez's, controlling for who the other players are.
Source: http://stats-for-the-nba.appspot.com/ra ... _2015.html
Re: The debut and popularization of BPM
Yes, it's been measured, and the effect is huge.
so - again - how would you model this?...
are their currently players who are good defensive rebounders that also box out as well as robin lopez?...
controlling for who the other players are.
how is this done? i ask as dallas is worst in the league right now in defensive rebounding percentage at just 71.8%, but have one of the league's best defensive rebounders in tyson chandler. does this mean chandler is poor at boxing out like hickson?...
is there a team that has great boxing out players but that are poor defensive rebounding teams?...
i see at that list you link to players like andre drummond, deandre jordan, and reggie evans with a negative number worse than hickson. what does that mean?...
so - again - how would you model this?...
are their currently players who are good defensive rebounders that also box out as well as robin lopez?...
controlling for who the other players are.
how is this done? i ask as dallas is worst in the league right now in defensive rebounding percentage at just 71.8%, but have one of the league's best defensive rebounders in tyson chandler. does this mean chandler is poor at boxing out like hickson?...
is there a team that has great boxing out players but that are poor defensive rebounding teams?...
i see at that list you link to players like andre drummond, deandre jordan, and reggie evans with a negative number worse than hickson. what does that mean?...