APBRmetrics

[blb2397]

Ok
I am a newbie to this incredible forum and have been trying to understand Hollingers PER. My question revolves around how Hollinger uses 2 points for a FGM made as his start point for a value. If a 2 point FG is worth two points and a possession is worth nearly 1 then how do we justify an increase of two points of player worth on a possession. I am sorry if I am missing on something obvious but its been kind of bugging me. I created a rating system years ago that I used in a game and I gave one point for a 2 pointer and -.67
for a missed shot (back when offensive rebounds were at a little higher rate) for a difference of 1.67 points between a good shot and a missed shot on a possession. Nothing too original with my system but I am tweaking it for my own site.
Thanksa bunch
_________________
NBA Post Up with Player Report Cards

[asimpkins]

I'm not entirely sure I understand your question, but I'll try to talk you through it.

Because some value is subtracted for the passer, a 2 point field goal is actually valued at 1.7 and a 3 point field goal is valued at 2.7 in PER.

This is simply awarding how many points you generated for your team. If you score 2 points then 2 points gets credited -- minus 0.3 credit which goes to the passer for his assists.

A possession is valued at 1 point because that's approximately how many points an average team will generate with that possession.

PER essentially tries to add up all box score statistics in terms of their value in points. So scoring a basket generates 2 points for your team -- obviously -- but securing possession (through a steal for example) only generates 1 point for your team because that's just an estimate of what a team will score with that possession.

[blb2397]

Thanks for the response. It is helpful. I guess the main problem I have is I am attempting to compare one stats value to another stats value in the formula. So I can understand that a steal might be worth one point less than a field goal, being that a steal grants another possession (average worth about 1 point) So it strikes me awkwardly that a two pointer would be worth a full 2.4 (3 without assist adjustment and rebound adjustment) points more than missed shots when the difference between a missed shot and made shot
is at most 2 points. I'll assume that PER is not trying to compare one stat to another or to a possession type baseline but more rewarding due credits to each individual stats.
I am just trying to come up with something more relative where one stats value compared to another stats value tens to add up. Right now I've been using something similar to
value = 2gm + 3gm*1.5 + ftm-(fta*.5 - .15*missed ft)
+ orb * .7 + drb * .3 + ast*.8 + stl + blk*.8 - to

Thanks again
_________________
NBA Post Up with Player Report Cards

[Mountain]

Somebody else can probably explain this accounting issue more fully but I see your rating centering FG made and misses roughly 1 pt from average value of possession.

Going 1 for 2 would result in a modest positive credit.

Hollinger's PER treats going 1 for 2 much more positively.
I think of it as a "1" base for this action as opposed to "zero" base.

That doesn't remove the question. I think your approach on shooting seems closer to Wins Produced philosophy and credit system. Should average shooting be viewed positively and creditted amply or treated as expected and for no credit gain? Plenty of threads dealing with that debate if you want to read all the commentary.

PER is on other stats value of possession based but things do get modified some.

[asimpkins]

[blb2397]

Thanks
I neglected missed shots on error. It should be -.7 per missed shot along the lines of PER. I understand the Wages of Wins comparison but it is not near the extreme. In this sytem a player who shoots 9 for 20 (45%) from 2 still gains 1.3 credits. I believe that average 2g% should still be rewarded somewhat. So a shot creator gains credits slowly. The rebounding doesn't overpower this like WOW does because like PER it uses .7*oreb and .3*dreb. Also because their is some more deviation in shot attempts the credits created by the extra shots can still compensate for the value of a good rebounder as opposed to an average one. So I guess it ends up being somewhere between PER and WOW. Although not the intention I hope to work in that area.
Basically as long as a player is around average or above in any stat he should be rewarded. I mean shouldn't a 9-20 game with 6 of 8 FT's and no turnovers be considered a good game (24pts in a touch over 20 possessions), not a great game like PER seems to say or a poor game like WOW.

Top 20 per 48 minutes as of a week ago
---------------------------------------------
LeBron James
Chris Paul
Amare Stoudemire
Tim Duncan
Kevin Garnett
Kobe Bryant
Carlos Boozer
Dirk Nowitzki
Al Jefferson
Marcus Camby
Chris Bosh
Dwight Howard
Manu Ginobili
Baron Davis
Yao Ming
Carmelo Anthony
Steve Nash
Antawn Jamison
Josh Smith
Shawn Marion
Chauncey Billups

Thanks for the insight I will look into the other discussions more closely.

corrected:
value = 2gm + 3gm*1.5 + ftm-(fta*.5 - .15*missed ft) - .7*missedfg
+ orb * .7 + drb * .3 + ast*.8 + stl + blk*.8 - to
_________________
NBA Post Up with Player Report Cards

[Mountain]

Somewhere between PER and WP is a good zone to be in overall.

That top 20 includes a lot of names you'd expect to be there at first cut.



The right assist weight is much debated.

Taking credit from the shooter to give to passer to prevent or reduce overcrediting the basket (I think it is reduce but am not sure) is an aspect of PER that most simple linear weight sets don't handle. In your formula that is extra credit given to the assister beyond the full amount awarded for a made basket to the shooter but of course assisted baskets do not count more than unassisted so this extra award breaks with strict court value accounting. A common past practice but something to be aware of.

[asimpkins]

Mountain has it pretty much right. Compared to PER, you devalue scoring in relation to everything else, most importantly rebounding, but you don't go quite as far as WoW. So it stands somewhere in between.

You also do something different with foul shots. A player that goes 2-4 from the foul line will get 1.2 credits in PER, while they will get 0.3 credits in your system. So PER gives an extra reward for generating shots from the foul line and you treat it all the same.

You also have no penalty for personal fouls, and slight increases for assists and blocks.

I haven't seen a good argument for how scoring should be definitively weighted to rebounding, so it's all pretty much a subjective judgment call. WoW goes so far to one extreme that it strikes me as pretty silly -- average shooting is valued at 0, but average rebounding is valued very high -- but there's a large area in the middle where any weights can be reasonable.

Your list is pretty similar to the top PER players. The top 3 are the same. Ginobili and Billups drop way down. Dwight Howard and Bosh end up below Boozer and Al Jefferson. Camby and Josh Smith make big jumps up. Missing are players like Kevin Martin, Jose Calderon, and Deron Williams.

[blb2397]

I'm having trouble grasping the idea of strict value accounting. I guess I'm not alone. How can we share credits for on stat like a made field goal but not for another stat. If the passer shares in the field goal than he would perhaps need to share some blame for a missed shot. Poor passing inevitably leads to poor shots that players are often times forced to take.
As mentioned I am trying more to compare one stats worth to another. Thats another reason I use 1.5 multiplier for 3pm as its 50% more valuable in terms of production.
As for assists I came up with .8 because I figured this made assists comparable to rebounds. I cant clearly say if passing is less or more valuable than rebounding so If I assume that they are equal than I come up with a .8 multiplier.
How?
.7*league average orb + .3*league average drb = .8*league average ast.
At last check they equaled bout 17
Also the sharing between assists and shots made does come into play when I calculate my wins produced. With 2pm = 1 and assists = .8 I multiply both by 5/9 so 2pm = 5/9 and assists = 4/9. I do this with all the stats (multiply by 5/9) This makes orb for example = 7/18. The remaining "value" of orb stays uncredited but thats ok because I am trying to get a total value for a player that compares well to another player by just looking at his credits. i.e Kobe comes out at 3.95 (per 40 above my average value of 8.5) Grant Hill at 0.4 above average. Kobe therefore adds about 2.55 points to score differential for every 40 minutes played. Using Oliver's basic pythagorean method this gives Kobe +12.62 wins and Hill +1.26 wins. Kobe (stats wise anyway at 40 minutes per) makes a 41 win team a 53.6 win team and Hill gives a minor increase (42.26)
Hope this makes some sense. Of course this leaves a lot out as well. No pace adjustments and using approximations like .15 for missed ft's and .7 and .3 for reb ( I used to use 2/3 and 1/3 back in the day of higher orb rates)
Here is the top 50 by wins list using player minutes per game instead of per 40 and assuming 82 games played

LeBron James 17.49
Chris Paul 14.20
Amare Stoudemire 12.16
Kobe Bryant 12.11
Tim Duncan 11.62
Kevin Garnett 11.03
Dirk Nowitzki 10.96
Carlos Boozer 10.69
Al Jefferson 10.63
Marcus Camby 10.05
Dwight Howard 9.95
Baron Davis 9.62
Chris Bosh 9.27
Yao Ming 8.96
Carmelo Anthony 8.24
Manu Ginobili 8.07
Antawn Jamison 8.00
Andrew Bynum 7.75
Shawn Marion 7.34
Steve Nash 7.33
Caron Butler 7.30
Allen Iverson 7.26
Pau Gasol 7.22
Josh Smith 7.16
Tracy McGrady 6.92
Chris Kaman 6.91
Chauncey Billups 6.64
Dwyane Wade 6.59
Jason Kidd 6.25
Deron Williams 5.98
David West 5.90
Zydrunas Ilgauskas 5.16
Zach Randolph 5.12
Jose Calderon 4.67
Vince Carter 4.61
Rasheed Wallace 4.55
Brandon Roy 4.53
Jermaine O'Neal 4.43
T.J. Ford 4.37
Brad Miller 4.25
Andris Biedrins 4.22
Paul Pierce 3.88
Michael Redd 3.61
Emeka Okafor 3.59
Gerald Wallace 3.50
Ron Artest 3.47
LaMarcus Aldridge 3.44
Josh Howard 3.35
Corey Maggette 3.30

Kevin Martin comes in at #54
Thanks again for the input
_________________
NBA Post Up with Player Report Cards

[blb2397]

Ok
I found it odd that some of the rebound / block guys were a little higher than I remember from previous rankings over the years so I rechecked and was using drb multiplier .7

The new updated list for those interested. (Yeah yet another weights system)
I'm surprised by how much scoring still matters and Dwight Howard's rank.
I've always ignored fouls and only because I could never come up with a multiplier that I could rationalize

LeBron James 16.85
Chris Paul 15.16
Kobe Bryant 12.05
Baron Davis 10.63
Amare Stoudemire 10.49
Allen Iverson 9.33
Dirk Nowitzki 9.15
Tim Duncan 9.02
Kevin Garnett 8.74
Al Jefferson 8.55
Chris Bosh 8.24
Steve Nash 8.20
Carlos Boozer 8.13
Carmelo Anthony 8.04
Manu Ginobili 8.01
Chauncey Billups 8.00
Dwyane Wade 7.70
Deron Williams 7.54
Caron Butler 7.28
Tracy McGrady 7.06
Yao Ming 6.90
Antawn Jamison 6.20
Pau Gasol 6.10
Josh Smith 6.05
Marcus Camby 5.82
Dwight Howard 5.75
Jose Calderon 5.73
Brandon Roy 5.35
Andrew Bynum 5.32
Vince Carter 5.07
Shawn Marion 5.00
Jason Kidd 4.91
David West 4.69
Michael Redd 4.64
Ron Artest 4.07
Paul Pierce 3.95
Zydrunas Ilgauskas 3.72
Andre Iguodala 3.69
Jermaine O'Neal 3.67
Kevin Martin 3.67
Monta Ellis 3.59
Gerald Wallace 3.40
Chris Kaman 3.29
Corey Maggette 3.28
Jason Richardson 3.27
LaMarcus Aldridge 3.23
Rasheed Wallace 3.23
Andre Miller 3.21
Tony Parker 3.10
Richard Hamilton 2.90
_________________
NBA Post Up with Player Report Cards

[Harold Almonte]