View previous topic :: View next topic |
Author |
Message |
KnickerBlogger
Joined: 30 Dec 2004 Posts: 180
|
Posted: Mon Jan 31, 2005 10:51 am Post subject: PER Correlate To Offense/Winning? |
|
|
Does anyone have a link to a study with how PER correlates to either offense or winning? I recall one being done, but I think it was on APBR_analysis, which is for all purposes non-searchable.
Thanks! |
|
Back to top |
|
|
Ed Küpfer
Joined: 30 Dec 2004 Posts: 785 Location: Toronto
|
Posted: Mon Jan 31, 2005 2:12 pm Post subject: Re: PER Correlate To Offense/Winning? |
|
|
KnickerBlogger wrote: | Does anyone have a link to a study with how PER correlates to either offense or winning? I recall one being done, but I think it was on APBR_analysis, which is for all purposes non-searchable. |
I don't remember that, but I'll have a look in my inbox. Does PER even make sense at the team level? _________________ ed |
|
Back to top |
|
|
kjb
Joined: 03 Jan 2005 Posts: 865 Location: Washington, DC
|
Posted: Mon Jan 31, 2005 2:38 pm Post subject: |
|
|
I've used a linear weight measure at the team level, and I know KevinP has tried it as well. The one I used was very simple -- pts + reb + ast + stl + blk - missed FG - missed FT - TOV - PF. Just running the linear measure for the offensive or defensive side didn't correlate well with winning %. BUT, the differential did have a strong correlation with winning %, which makes a lot of sense. |
|
Back to top |
|
|
KnickerBlogger
Joined: 30 Dec 2004 Posts: 180
|
Posted: Mon Jan 31, 2005 3:45 pm Post subject: Re: PER Correlate To Offense/Winning? |
|
|
Ed Küpfer wrote: | KnickerBlogger wrote: | Does anyone have a link to a study with how PER correlates to either offense or winning? I recall one being done, but I think it was on APBR_analysis, which is for all purposes non-searchable. |
I don't remember that, but I'll have a look in my inbox. Does PER even make sense at the team level? |
Good question. I'm at work, so I don't have the equation in front of me. However I think most of the minutes/pace factor stuff would be cancelled out (a multiplier of 1). I vaguely remember someone doing a study like this.
Maybe doing a weighted PER (per minute) could give an approximation of the team's PER? |
|
Back to top |
|
|
jkubatko
Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH
|
Posted: Mon Jan 31, 2005 5:49 pm Post subject: Re: PER Correlate To Offense/Winning? |
|
|
KnickerBlogger wrote: | Maybe doing a weighted PER (per minute) could give an approximation of the team's PER? |
As Ed said, I'm not sure PER even makes sense at the team level. Regardless, I found the team PER by taking a weighted average of the player PERs. Here are the results (all seasons since 1978):
Code: |
Pearson Correlation Coefficients, N = 698
WPct EWPct offEff
PER 0.82461 0.84050 0.81163
|
Note that EWpct is expected (or Pythagorean) winning percentage and offEff is offensive efficiency. _________________ Regards,
Justin Kubatko
Basketball-Reference.com
Last edited by jkubatko on Tue Feb 01, 2005 12:22 am; edited 1 time in total |
|
Back to top |
|
|
gabefarkas
Joined: 31 Dec 2004 Posts: 1313 Location: Durham, NC
|
Posted: Mon Jan 31, 2005 6:18 pm Post subject: |
|
|
does Pearson go from -1 (completely inverse) to 1 (complete correlation)? |
|
Back to top |
|
|
Ed Küpfer
Joined: 30 Dec 2004 Posts: 785 Location: Toronto
|
Posted: Mon Jan 31, 2005 6:29 pm Post subject: |
|
|
gabefarkas wrote: | does Pearson go from -1 (completely inverse) to 1 (complete correlation)? |
Yes. _________________ ed |
|
Back to top |
|
|
KnickerBlogger
Joined: 30 Dec 2004 Posts: 180
|
Posted: Mon Jan 31, 2005 10:29 pm Post subject: Re: PER Correlate To Offense/Winning? |
|
|
jkubatko wrote: | KnickerBlogger wrote: | Maybe doing a weighted PER (per minute) could give an approximation of the team's PER? |
As Ed said, I'm not sure PER even makes sense at the team level. Regardless, I found the team PER by taking a weighted average of the player PERs. Here are the results (all seasons since 1978):
Code: |
Pearson Correlation Coefficients, N = 698
WPct EWPct offEff
PER 0.82461 0.84050 0.81163
|
Note that EWpct is expected (or Pythagorean) winning percentage and offEff is offensive efficiency. |
That is excellent! Now how good is .81-.84? |
|
Back to top |
|
|
Kevin Pelton Site Admin
Joined: 30 Dec 2004 Posts: 979 Location: Seattle
|
Posted: Mon Jan 31, 2005 10:35 pm Post subject: |
|
|
Unless I'm confused about the calculations, I'm having a hard time believing those values that are being reported.
With no disrespect whatsoever to John, I don't think the PER correlation should be much different from the correlation of my efficiency, which was about .55 or so. (Manley credits aka the NBA's Efficiency scored a .62 -- I think that's basically what the other Kevin is reporting.)
What I really find unbelievable is that there would be no stronger correlation between PER and offensive ratings than PER and winning percentage, since obviously defensive statistics are so much more limited. I found much, much higher correlations between my efficiency and the NBA's Efficiency and offense, in the .9 area (mine scored better looking strictly at offense). |
|
Back to top |
|
|
jkubatko
Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH
|
Posted: Tue Feb 01, 2005 12:20 am Post subject: |
|
|
admin wrote: | Unless I'm confused about the calculations, I'm having a hard time believing those values that are being reported.
With no disrespect whatsoever to John, I don't think the PER correlation should be much different from the correlation of my efficiency, which was about .55 or so. (Manley credits aka the NBA's Efficiency scored a .62 -- I think that's basically what the other Kevin is reporting.)
What I really find unbelievable is that there would be no stronger correlation between PER and offensive ratings than PER and winning percentage, since obviously defensive statistics are so much more limited. I found much, much higher correlations between my efficiency and the NBA's Efficiency and offense, in the .9 area (mine scored better looking strictly at offense). |
I re-did this and obtained the same results as before (once again, all seasons since 1978):
Code: |
Pearson Correlation Coefficients, N = 698
PER Pelton offEff defEff WPct
PER 1.00000 0.69109 0.81163 -0.24626 0.82461
Pelton 0.69109 1.00000 0.86152 0.18596 0.54593
offEff 0.81163 0.86152 1.00000 0.18266 0.65023
defEff -0.24626 0.18596 0.18266 1.00000 -0.58158
WPct 0.82461 0.54593 0.65023 -0.58158 1.00000
|
As you said, the correlation between your measure of efficiency (Pelton) and team winning percentage (WPct) is about 0.55. One thing that is interesting to note is that while the correlation between PER and defensive efficiency is negative (as PER increases, points allowed per 100 possessions tends to decrease), the correlation between Pelton and defensive efficiency is positive (as Pelton increases, points allowed per 100 possessions tends to increase). While neither linear relationship is particularly strong, the opposite signs are striking. _________________ Regards,
Justin Kubatko
Basketball-Reference.com |
|
Back to top |
|
|
jkubatko
Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH
|
Posted: Tue Feb 01, 2005 12:28 am Post subject: |
|
|
gabefarkas wrote: | does Pearson go from -1 (completely inverse) to 1 (complete correlation)? |
I would phrase it this way: -1 (perfect negative linear association) to 1 (perfect positive linear association). Correlation is on a scale from -1 to 1. A correlation of 0 means no linear association (note that I said no linear association, not no association). As you move to the extremes (-1 and 1), the strength of the linear association increases. _________________ Regards,
Justin Kubatko
Basketball-Reference.com |
|
Back to top |
|
|
Kevin Pelton Site Admin
Joined: 30 Dec 2004 Posts: 979 Location: Seattle
|
Posted: Tue Feb 01, 2005 1:30 am Post subject: |
|
|
That is certainly both surprising and humbling. I figured any set of linear weights would be fairly highly correlated.
But it does neatly prove KB's argument. |
|
Back to top |
|
|
gabefarkas
Joined: 31 Dec 2004 Posts: 1313 Location: Durham, NC
|
Posted: Tue Feb 01, 2005 7:32 am Post subject: |
|
|
jkubatko wrote: | gabefarkas wrote: | does Pearson go from -1 (completely inverse) to 1 (complete correlation)? |
I would phrase it this way: -1 (perfect negative linear association) to 1 (perfect positive linear association). Correlation is on a scale from -1 to 1. A correlation of 0 means no linear association (note that I said no linear association, not no association). As you move to the extremes (-1 and 1), the strength of the linear association increases. |
right. that's correlation in general. i just wasn't sure if Pearson was set up the same way. |
|
Back to top |
|
|
jkubatko
Joined: 05 Jan 2005 Posts: 702 Location: Columbus, OH
|
Posted: Tue Feb 01, 2005 8:18 am Post subject: |
|
|
gabefarkas wrote: | right. that's correlation in general. i just wasn't sure if Pearson was set up the same way. |
When people generically use the term "correlation", they are almost always referring to the Pearson Product Moment Correlation Coefficient. _________________ Regards,
Justin Kubatko
Basketball-Reference.com |
|
Back to top |
|
|
|