I'm not a stats guy, I have absolutely zero knowledge of how exactly advanced stats work, but I'm curious... I just had two questions.
First I was wondering whether there existed a statistic that basically computed "points created" by a player in the following way:
Total points
+ offensive rebounds * avg pts scored off offensive rebounds
+ defensive rebounds * avg pts scored off defensive rebounds
+ points off assists (a pass leading to free throws with at least one free throw made counts as an assist)
+ steals * avg pts scored off steals
+ charges drawn * avg pts scored off charges drawn
+ blocks * avg pts scored off blocks
- 2 pt misses * avg pts scored by opposing team off 2 pt misses
- 3 pt misses * avg pts scored by opposing team off 3 pt misses
- turnovers * avg pts scored by opposing team off turnovers
- pts scored off shooting foul induced free throws
You could then compute points created per possession.
If it doesn't exist, do you think it would be useful and why? Can it be improved or is it a hopeless case?
Obviously from the looks of it it's a boxscore-heavy stat, but it seems like it could assess a player's productivity fairly decently.
Another question is, are there stats that assess the value of screens (at the very least the most obvious ones ie successful pick and roll screens) or do you think no number can accurately represent that?
A few questions
Moderator: Doctor MJ
Re: A few questions
-
Chicago76
- Rookie
- Posts: 1,134
- And1: 228
- Joined: Jan 08, 2006
Re: A few questions
What you've come up with is a linear weights statistic that uses box score statistics, weighting them according to how they compare in terms of pts per possession. This is very similar to PER, so the big question is, "Would this be an improvement over PER?"
The biggest weakness your statistic has vs. PER is that it doesn't account for Pace. If a random player creates 30 "stat points" in your formula, this will obviously be more valuable at an 85 possession game pace than it would at 100.
Something that your system appears to do well is tie the occurence of an event to a greater or lower than average probability of pts being generated from the event. For example: a steal has been shown to generate a much higher probability of pts scored than league average pts/poss of 1.04 to 1.09 (
which is the value of a possession over the past 30 years or so, depending upon the year). An offensive rebound also tends to generate points higher than the leage average possession rate as well. At first glance, this would seem to be an improvement over Hollinger, where the value of pts created from something is always the average.
The problem with using different weights for efficiency after a steal/block/etc is that these types of events have penalties for unsuccessful attempts that can't be captured in the box score. With FG made, we can always balance that against misses. For a steal, there isn't a corresponding unsuccessful steal attempt. So if league average value of a possession (VOP) = 1.05 and a VOP after a steal is 1.25, this type of system would give more value to steals. The problem comes in because guys who tend to have a lot of steals also tend to gamble on D. An unsuccessful steal attempt may lead to the other team's VOP increasing to 1.2 from 1.05. In other words, the unsuccessful attempts tend to bring the value of a successful steal back down to league average VOP anyway.
Offensive rebounds are similar. If you aggressively crash the glass, you sacrifice transition defense, which increases points on the other end.
I would encourage you to take a very good look at PER, because this stat has done what you've proposed in a way that greatly reduces the problems in what you're proposing. It's not perfect, but as far as box score stats go, it's preferrable to this.
For screens:
There are sites and game log data out there that take a look at situational efficiencies like PNR, spot up 2s, spot up 3s, etc. The problem you're going to have is in being able to credit the PNR efficiency to the actual act of screening. Just looking at a PNR duo, there are a lot of variables that go into point efficiency:
-screen setting
-ball handling/decision making of the guy starting with the ball.
-ability to exploit the screen by the ball handler
-spot up shooting ability of screener if he pops out to receive and shoot
-hands/athleticism is he rolls to the basket
-finishing ability of the screener rolling to basket
-passing ability of the ball handler
-quality of defense
Someone recently did an analysis over at APBR metrics which looked at various scoring scenarios across teams in terms of efficiency and volume. What he found is that volume of PNR doesn't seem to impact efficiency very much. In other words, teams with an athletic/decent shooting big and a good ballhandler/passer might run more PNR, but they don't do it at a more efficient rate. This is because there are so many decision points along the way where decisions will be different based upon the personnel involved. If you have an inferior shooter, you're less likely to pass to the roller popping out for the jumper. If the roller is less athletic, he's less likely to receive a pass cutting to the basket, and so on. If he does, his chance to score is abt equal to a superior player in the same scenario. What this means is that inferior duos tend to more frequently reset off a PNR that results in nothing, so volume is lower, but the actual success rate when the duo can exploit the defense (resulting in a FGA) really isn't all that different.
The biggest weakness your statistic has vs. PER is that it doesn't account for Pace. If a random player creates 30 "stat points" in your formula, this will obviously be more valuable at an 85 possession game pace than it would at 100.
Something that your system appears to do well is tie the occurence of an event to a greater or lower than average probability of pts being generated from the event. For example: a steal has been shown to generate a much higher probability of pts scored than league average pts/poss of 1.04 to 1.09 (
which is the value of a possession over the past 30 years or so, depending upon the year). An offensive rebound also tends to generate points higher than the leage average possession rate as well. At first glance, this would seem to be an improvement over Hollinger, where the value of pts created from something is always the average.
The problem with using different weights for efficiency after a steal/block/etc is that these types of events have penalties for unsuccessful attempts that can't be captured in the box score. With FG made, we can always balance that against misses. For a steal, there isn't a corresponding unsuccessful steal attempt. So if league average value of a possession (VOP) = 1.05 and a VOP after a steal is 1.25, this type of system would give more value to steals. The problem comes in because guys who tend to have a lot of steals also tend to gamble on D. An unsuccessful steal attempt may lead to the other team's VOP increasing to 1.2 from 1.05. In other words, the unsuccessful attempts tend to bring the value of a successful steal back down to league average VOP anyway.
Offensive rebounds are similar. If you aggressively crash the glass, you sacrifice transition defense, which increases points on the other end.
I would encourage you to take a very good look at PER, because this stat has done what you've proposed in a way that greatly reduces the problems in what you're proposing. It's not perfect, but as far as box score stats go, it's preferrable to this.
For screens:
There are sites and game log data out there that take a look at situational efficiencies like PNR, spot up 2s, spot up 3s, etc. The problem you're going to have is in being able to credit the PNR efficiency to the actual act of screening. Just looking at a PNR duo, there are a lot of variables that go into point efficiency:
-screen setting
-ball handling/decision making of the guy starting with the ball.
-ability to exploit the screen by the ball handler
-spot up shooting ability of screener if he pops out to receive and shoot
-hands/athleticism is he rolls to the basket
-finishing ability of the screener rolling to basket
-passing ability of the ball handler
-quality of defense
Someone recently did an analysis over at APBR metrics which looked at various scoring scenarios across teams in terms of efficiency and volume. What he found is that volume of PNR doesn't seem to impact efficiency very much. In other words, teams with an athletic/decent shooting big and a good ballhandler/passer might run more PNR, but they don't do it at a more efficient rate. This is because there are so many decision points along the way where decisions will be different based upon the personnel involved. If you have an inferior shooter, you're less likely to pass to the roller popping out for the jumper. If the roller is less athletic, he's less likely to receive a pass cutting to the basket, and so on. If he does, his chance to score is abt equal to a superior player in the same scenario. What this means is that inferior duos tend to more frequently reset off a PNR that results in nothing, so volume is lower, but the actual success rate when the duo can exploit the defense (resulting in a FGA) really isn't all that different.
Re: A few questions
-
BoutPractice
- Senior
- Posts: 666
- And1: 540
- Joined: Oct 31, 2011
Re: A few questions
Thanks a lot for the insightful reply!
I thought it might be similar to PER (I knew the gist of it, though not the exact formula, I will confess), but was wondering whether it could be an improvement due to the use of weights tied to actual probabilities. Pace is indeed extremely important, so you've convinced me that PER is the better alternative.
Your point about steals and blocks is also very true. That's the problem with most statistics - what you don't know or can't compute has a lot of consequences that can negate the effect of what you know and can compute.
You can't say with certainty that missed steals bring the VOP down to the average without numbers, but your idea stands.
I would also suspect that the negative effect of "gambling" for steals is greater than the effect of "missed blocks". Missed blocks often result in shooting fouls anyway. In addition they can either have a positive impact (when you challenge a shot without blocking it) or a negative one (biting on a pump fake). So it seems like the problem would mainly be about steals, possibly offensive rebounds.
Regading offensive rebounds, it's hard to know. I personally want my center to crash the offensive glass - centers generally aren't the biggest threat in transition, though there are exceptions. With a swingman, it's a different story.
Very interesting thoughts on PNR. It's true that many more variables go into play than just the quality of the screen. The quality of defense, in particular, has a lot to do with the result (many interior players simply don't know how to hedge correctly). It would be just as true of off-the-ball screens - it takes a good screener, a good off-the-ball player and/or a confused defender to get an open position.
It seems like stats can't tell us much about the importance of setting and taking screens, other than a detailed play-by-play qualitative analysis. Certainly can't give us a meaningful aggregate, which is a shame given how crucial screens are to team success.
Once again, thanks for your analysis.
I thought it might be similar to PER (I knew the gist of it, though not the exact formula, I will confess), but was wondering whether it could be an improvement due to the use of weights tied to actual probabilities. Pace is indeed extremely important, so you've convinced me that PER is the better alternative.
Your point about steals and blocks is also very true. That's the problem with most statistics - what you don't know or can't compute has a lot of consequences that can negate the effect of what you know and can compute.
You can't say with certainty that missed steals bring the VOP down to the average without numbers, but your idea stands.
I would also suspect that the negative effect of "gambling" for steals is greater than the effect of "missed blocks". Missed blocks often result in shooting fouls anyway. In addition they can either have a positive impact (when you challenge a shot without blocking it) or a negative one (biting on a pump fake). So it seems like the problem would mainly be about steals, possibly offensive rebounds.
Regading offensive rebounds, it's hard to know. I personally want my center to crash the offensive glass - centers generally aren't the biggest threat in transition, though there are exceptions. With a swingman, it's a different story.
Very interesting thoughts on PNR. It's true that many more variables go into play than just the quality of the screen. The quality of defense, in particular, has a lot to do with the result (many interior players simply don't know how to hedge correctly). It would be just as true of off-the-ball screens - it takes a good screener, a good off-the-ball player and/or a confused defender to get an open position.
It seems like stats can't tell us much about the importance of setting and taking screens, other than a detailed play-by-play qualitative analysis. Certainly can't give us a meaningful aggregate, which is a shame given how crucial screens are to team success.
Once again, thanks for your analysis.
Re: A few questions
-
DSMok1
- Sophomore
- Posts: 118
- And1: 112
- Joined: Jul 26, 2010
- Location: Maine
- Contact:
-
Re: A few questions
Your stat is very similar to Evan Zamir's EZPM, over at http://thecity2.com/2011/01/31/ezpm-v-2 ... he-biggie/
He also incorporated counterpart numbers for defense from the PbP data.
He also incorporated counterpart numbers for defense from the PbP data.
Return to Statistical Analysis