Reconstructing Box Plus/Minus

Home for all your discussion of basketball statistical analysis.
Crow
Posts: 6251
Joined: Thu Apr 14, 2011 11:10 pm

Re: Reconstructing Box Plus/Minus

Post by Crow » Tue May 07, 2019 6:11 pm

Not likely someone would adjust game data by season standings every game or for every situation but perhaps some simple general curve could be laid over the season, particularly late season games.

Also if RPI matters for the NCAA,tournament and coaches & players care about that, then college players may not behave similar to NBA players in blowouts. And behavior might vary in other leagues too. A similar adjustment may not be appropriate across leagues. The court dimensions and 3pt rules and other rules vary across leagues. Taking BPM across leagues is a start but there are imperfections in the transfer.

Crow
Posts: 6251
Joined: Thu Apr 14, 2011 11:10 pm

Re: Reconstructing Box Plus/Minus

Post by Crow » Tue May 07, 2019 6:22 pm

It has been noted that team net ratings are imperfect predictors of future matchup results. That may be especially true in leagues where the net ratings vary more than others. That is in NCAA compared to NBA and perhaps in other comparisons.

Is referee impact on net scores similar across leagues? Maybe not.

How much is change in player performance in blowout conditions is the player vs. the refereeing? That would be interesting to see analyzed. What specific stats change in blowout conditions and what are the proportions on impact of things that refs influence significantly directly or indirectly vs. things they don't influence or apparently not much?

Crow
Posts: 6251
Joined: Thu Apr 14, 2011 11:10 pm

Re: Reconstructing Box Plus/Minus

Post by Crow » Tue May 07, 2019 10:24 pm

Some of the above posts were written quick n hot but with a genuine desire for dialogue. I'll hope for, look for replies that will clarify what was meant.


Some players & coaches may use play and action sequencing to set up future moves. I dunno how much that has been studied. It might be detectable with the right approach & effort. It is not beyond the ability of outsiders to anticipate / understand even if the behavior hasn't been modeled yet.

Nate
Posts: 131
Joined: Tue Feb 24, 2015 2:35 pm

Re: Reconstructing Box Plus/Minus

Post by Nate » Thu May 09, 2019 4:41 am

Crow wrote:
Tue May 07, 2019 4:56 pm
Nate wrote:
Sat Apr 27, 2019 10:25 pm
I'm not sure it's a great idea, but you could try to do something analogous to Elo ratings for players using minutes played and scoring margin in some kind of Bayesian inference model, then estimate "contribution" as rating * minutes played.

So you'd do game-by-game updates, but only using minutes played and score margin. You'd also have to do some per-league calculations to estimate the variance in scoring margin per game.

This is a very different approach from using a single stat line, but splitting into per-game data should be much richer, and probably still be feasible in excel. (Almost certainly feasible in something like R.)
How is this better than the RAPM approach? Do they avoid relying on trivia? How did you come up with it? Are minutes 100% free of trivia? How, why? Is scoring margin free from trivia and free from trivia at the individual player level? Is this raw scoring margin unadjusted by quality of players on court or amount of time with large leads / deficits?
....
I really don't know if it's better or worse than RAPM in a basketball decision making sense.

From a statistics point of view, you want to use more, not less, data, and per-game information is a larger data set than end of season aggregates. Now, it's possible that you're not going to get any useful improvement from the work of splitting things into 'per-game' data, but it really shouldn't hut, and when computers are doing the calculations, the amount of extra work isn't really that huge.

A bunch of this is personal preference. I find the idea of using methods that explicitly keep track of uncertainty (like Bayesian inference) appealing. I wonder how much the box score gives individual credit for team effort. I want to have a simple model of how basketball works that makes sense to me and then attack it with mathematical sophistication.

As DSMok wrote, this is an approach that's built around modeling basketball games as a series of possessions, so possession-by-possession would be much better.

Nate
Posts: 131
Joined: Tue Feb 24, 2015 2:35 pm

Re: Reconstructing Box Plus/Minus

Post by Nate » Thu May 09, 2019 4:51 am

Crow wrote:
Tue May 07, 2019 4:28 pm
....
Trivia is defined as details, considerations, or pieces of information of little importance or value. Are you saying most or everything in performance metrics is trivia? I'd disagree. What is trivia and what isn't? Who knows? Does RAPM bother with or credit any trivia? I assume you are alleging BPM credits trivia. Which specific items, if you know. If you don't, can you name one player or coach that does? Have they ever spoken of this truth? Where / what is truth in the sea of alleged trivia?
....
I'm saying that it seems like people want numbers and don't really care whether the numbers are "right" or "wrong," and, if they don't care whether the numbers are "right" or "wrong" then the numbers can't be important.

DSMok1
Posts: 905
Joined: Thu Apr 14, 2011 11:18 pm
Location: Maine
Contact:

Re: Reconstructing Box Plus/Minus

Post by DSMok1 » Tue May 14, 2019 8:40 pm

One of the first things I have done with the new BPM framework is to run a fully linear analysis similar to Nathan Walker's SPR and Kevin Ferrigan's DRE. I still am running the "team adjustment" at the end to make it match the team rating.

Here's what I got:



Here are the top 30 regular seasons from the past 30 years using this regression (also showing the old BPM, DRE, and SPR):



Some comments:
- This regression is more similar to DRE than SPR (I'm also not including the Games Started part of SPR)
- This regression isn't as high on CP3 as DRE, but he still shows up very high
- Curry and KD are the biggest beneficiaries vs. DRE at the top. Also Derrick Rose and Allen Iverson.
- Anthony Davis and Kevin Garnett are biggest losers vs. DRE. Also Andris Biedrins, Andre Drummond, Joel Anthony...
- Blocks and Steals are less valued
- Turnovers much less penalty
Developer of Box Plus/Minus
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

Mike G
Posts: 4429
Joined: Fri Apr 15, 2011 12:02 am
Location: Asheville, NC

Re: Reconstructing Box Plus/Minus

Post by Mike G » Tue May 14, 2019 9:19 pm

Is this the whole thing? No "advanced" stats, nor interactive terms like Ast*Reb?
Any idea why 2fga are more costly than 3fga? Because they are less likely to be OReb'd?
Earlier you said fouls were not significant of anything. Now they're almost as bad, as an assist is good?

Westbrook'17 is not even top 20 now? At 15.6, he was miles ahead of everyone else.

DSMok1
Posts: 905
Joined: Thu Apr 14, 2011 11:18 pm
Location: Maine
Contact:

Re: Reconstructing Box Plus/Minus

Post by DSMok1 » Tue May 14, 2019 11:01 pm

Mike G wrote:
Tue May 14, 2019 9:19 pm
Is this the whole thing? No "advanced" stats, nor interactive terms like Ast*Reb?
Any idea why 2fga are more costly than 3fga? Because they are less likely to be OReb'd?
Earlier you said fouls were not significant of anything. Now they're almost as bad, as an assist is good?

Westbrook'17 is not even top 20 now? At 15.6, he was miles ahead of everyone else.
No, this is not the new version of BPM. This is a linear statistic that can be used on very small sample sizes (even down to individual possessions). Like DRE, its best use is for analyzing individual games and other small-scale phenomena.
Developer of Box Plus/Minus
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

Crow
Posts: 6251
Joined: Thu Apr 14, 2011 11:10 pm

Re: Reconstructing Box Plus/Minus

Post by Crow » Wed May 15, 2019 1:28 am

Steal worth 9 times what an offensive rebound is worth in GmBPM2? I'd like to hear the justification of that. Both earn the ball. Twice as much? That might be justified.

Defensive rebounds worth twice as much as offensive rebounds? Since when? I remember the fuss folks made when a metric-maker claimed they were of equal value.

How does a point scored get valued at .74 pts? Is the rest moved to the team adjustment?

DSMok1
Posts: 905
Joined: Thu Apr 14, 2011 11:18 pm
Location: Maine
Contact:

Re: Reconstructing Box Plus/Minus

Post by DSMok1 » Wed May 15, 2019 2:50 am

Crow wrote:
Wed May 15, 2019 1:28 am
Steal worth 9 times what an offensive rebound is worth in GmBPM2? I'd like to hear the justification of that. Both earn the ball. Twice as much? That might be justified.

Defensive rebounds worth twice as much as offensive rebounds? Since when? I remember the fuss folks made when a metric-maker claimed they were of equal value.

How does a point scored get valued at .74 pts? Is the rest moved to the team adjustment?
Yes... The team adjustment. So if an item is worth more to the team than the regression shows, the remaining credit is split evenly among all 5 players. That actually speaks to your questions in the first two paragraphs as well. Rebounds don't depend much on the player, more on the team as a whole.
Developer of Box Plus/Minus
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

jaynay1
Posts: 1
Joined: Tue Jan 29, 2019 7:29 am

Re: Reconstructing Box Plus/Minus

Post by jaynay1 » Wed May 15, 2019 10:44 am

Crow wrote:
Wed May 15, 2019 1:28 am
Steal worth 9 times what an offensive rebound is worth in GmBPM2? I'd like to hear the justification of that. Both earn the ball. Twice as much? That might be justified.
Offensive rebounding is statistically significant at predicting opponent transition efficiency when you control for overall rebounding skill, meaning that while offensive rebounds may gain the ball, which is valuable, there is a tradeoff which makes a difference make sense.

bbstats
Posts: 224
Joined: Thu Apr 21, 2011 8:25 pm
Location: Boone, NC
Contact:

Re: Reconstructing Box Plus/Minus

Post by bbstats » Wed May 15, 2019 1:18 pm

Friendly reminder to everyone reading the results of this model: feature coefficients do not exist in a vacuum. For example: if you had a regression of *just* turnovers vs RAPM, turnovers would be positive because it's correlated with other successful things (namely, being a high-usage passer).

The model won't assign "credit" to certain statistics more than the *other* statistics have already been credited. That's why fouls look so relatively harmless here - they're somewhat correlated with playing good defense.

DSMok1
Posts: 905
Joined: Thu Apr 14, 2011 11:18 pm
Location: Maine
Contact:

Re: Reconstructing Box Plus/Minus

Post by DSMok1 » Thu May 16, 2019 3:20 pm

Interesting GmBPM (that's the linear BPM) evaluation:

What if we let the points from 3 pointers, 2 pointers, and free throws each have a different value? Remember, this regression attributes value to the player AND ALL LEFTOVER VALUE TO THE REST OF THE TEAM. So--maybe some scoring shows more for the individual player, while other scoring is more generic (anybody could do it). Also, some scoring may be more valuable from a spacing perspective.

Does that make sense?

The results from the same GmBPM linear model:



Interestingly, points from free throws are worth the same as before, as are free throw attempts.

2 pointers have less value and 2 point attempts are less penalized. In other words, this regression is indicating 2 pointers don't matter as much to the player.

Conversely, 3 pointers matter considerably more! Made 3 pointers are worth more and missed 3 pointers are penalized more heavily!

This is very interesting to me. What would it look like if we had the data to split out at-rim from 2 point jumpers?

P.S. The rest of the coefficients in the GmBPM regression did not change significantly at all.
P.P.S. This helps elite shooters the most, depresses the value of 2pt scorers and bad 3 point shooters. In other words, it really helps Stephen Curry.
Developer of Box Plus/Minus
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

DSMok1
Posts: 905
Joined: Thu Apr 14, 2011 11:18 pm
Location: Maine
Contact:

Re: Reconstructing Box Plus/Minus

Post by DSMok1 » Thu May 16, 2019 5:46 pm

A quick look at how this translates to analysis of a single game. Here I am using the version with each shot location treated separately.



This is last night's Game 1 of the 2019 Eastern Conference Finals.
https://www.basketball-reference.com/bo ... 50MIL.html

This is not adjusted for the quality of the teams at all, but is adjusted for home court advantage.

Comparison with some other single-game models:
Jacob Goldstein:

Dean Oliver:
Developer of Box Plus/Minus
APBRmetrics Forum Administrator
GodismyJudgeOK.com/DStats/
Twitter.com/DSMok1

Crow
Posts: 6251
Joined: Thu Apr 14, 2011 11:10 pm

Re: Reconstructing Box Plus/Minus

Post by Crow » Fri May 17, 2019 1:19 am

These are not the classic 4 Factors but it would be good to regularly offer these 4 components (or some other variation) for GmBPM2 and season level BPM2. Maybe BRef could also offer BPM2 on its splits page. More detail, less case that it is a blackbox hard to interpret. If BPM2 components were divided like the 4 factors it would ease comparison to RPM. DRE and SPR could be useful with splits as well, preferably standardized.




What is the difference between contribution and points generated?

In rough terms and on average, what % of total credit is funneling to players via the team adjustment? Any consideration of doing the team adjustment as something more discriminatory than same for anybody (on court and off when things happen)?

Post Reply