Is Westbrook Exposing BPM ?

[feyki]

Or really he's having the greatest season ever ?


Also , one more question .. Do you guys think BPM needs a couple changes in the formula ?

[bchaikin]

Or really he's having the greatest season ever ?

the worst 2pt shooting team in the league now shoots only 47.0% on 2s. westbrook is shooting an even lower 45% on 2s. so he misses alot of shots (over 200 more than any other thunder player has even attempted). on average 3/4 of all missed FGA are rebounded by the opponent...

he is also committing 5.6 TO/g, the most/worst ever by a player (other than harden also this same year)...

that's a ton of zero point team possessions he is personally responsible for, about 16-17 per game (he plays 35 min/g). OKC averages about 99 poss/48min, i.e. so about 1 out of every 6 thunder team possessions they do not score because of westbrook...

does BPM take this into account?...

[DSMok1]

Or really he's having the greatest season ever ?


Also , one more question .. Do you guys think BPM needs a couple changes in the formula ?

Yes, Westbrook is exposing BPM.

He's having a crazy outlier season; I think we can all agree on that point. What this means is that good linear metrics will do better with him than nonlinear metrics. It so happens that the couple of nonlinear terms that BPM includes are the specific terms that Westbrook is an outlier on--specifically, the USG x AST and the sqrt(AST x REB) interaction terms. So probably DRE will do better with him that BPM.

Here's players anywhere near that AST x USG level: http://bkref.com/tiny/hkvcR

It's sort of a trade off in developing a metric. Making a metric more accurate for the majority of the populace often increases the maximum error possible on a select few "edge" cases.

Westbrook is an edge case.

As for improvements, I'd love to hear ideas. Perhaps it should be a sqrt(USG x AST) instead of its current form? I have begun work on a new version of BPM to address a few issues I've seen and to take advantage of better long-term RAPM.

Or really he's having the greatest season ever ?

the worst 2pt shooting team in the league now shoots only 47.0% on 2s. westbrook is shooting an even lower 45% on 2s. so he misses alot of shots. on average 3/4 of all missed FGA are rebounded by the opponent. he is also committing 5.6 TO/g, the most/worst ever by a player (other than harden also this same year)...

that's a ton of zero point team possessions he is personally responsible for, about 16-17 per game (he plays 35 min/g). OKC averages about 99 poss/48min, i.e. so about 1 out of every 6 thunder team possessions they do not score because of westbrook...

does BPM take this into account?...

The full current model is written up at http://www.basketball-reference.com/about/bpm.html

Westbrook does have a TS% of 0.546, which is about average. BPM only looks at TS% in terms of efficiency. So it sees ridiculous volume at league-average efficiency.

Given his level of usage, the turnovers are reasonable. The team averages fewer turnovers with him on the floor than off. ( http://nbawowy.com/#/l9ioib5uhlp )

[Nathan]

I think this is exactly the right explanation. I've fiddled around with BPM some, and last year Curry blew up my model that was nonlinear in 3PM for the same reason

[feyki]

DsMok;

I also think usage is the main problem here . All of metrics are mainly based on scoring . And high volume scoring is like atom of advanced metrics(box score) . If we want to create a statistical RAPM/RPM type metric , we must cut that high volume scoring cancer . Assists could be a factor here . If we put lots of value to assists , high volume scoring or usage wouldn't be a problem anymore . Not just field goals , also free throws coming from assists(usually %60 since 80's) , but statistics don't show us .

Also , rebound+assist stuff is an issue here too . Rebounding and playmaking world different things to each others . Formula shouldn't be an all around detector . It's all about impact .

[bchaikin]

Given his level of usage, the turnovers are reasonable.

oh?...

if you look at the SportVu tracking data at stats.nba.com under Player Tracking Touches, you'll find westbrook with 344 TOs in 6169 touches (5242 frontcourt touches). that's 5.6% TO/touch (or using just frontcourt touches since few TOs occur in the backcourt 6.6 TO/frontcourttouch)...

for comparison isaiah thomas is at 159 TO in 4873 touches (4491 frontcourt touches). that's 3.3% TO/touch (or 3.5 TO/frontcourttouch)...

chris paul is even better at 2.7 TO/touches and 3.3 TO/frontcourttouch...

that's a huge difference over the span of a season...

[Crow]

Understanding that rates and team performance matter too / more, fyi Westbrook was on pace to beat all-time league record for total TOs for season when I checked a couple weeks ago.

[Crow]

Sqrt(USG x AST) seems worth a look. I'd probably favor but haven't seen the analysis for the change vs current choice or the player ratings with that change vs. current. If you wanted / were willing, perhaps you could also share some of the other tweaks and the change vs. current side by side comparisons for perhaps a select group of high achievers, average players and lowest tier guys (5-10 of each?).

Any consideration of adding a scoring versatility factor?

Any thought about maybe making the sqrt(AST x REB) factor somewhat position or superposition (pg, wing, big) / role sensitive? Versatility may be more / less common and / or valuable depending where it is provided?

Any consideration underway for changes to the BPM Team Adjustment? Would you be willing to look at top 5 most used lineups for each player to calculate a BPM Team Adjustment appropriate for those lineups and then maybe use that for a minutes weighted or standard amount of the BPM Team Adjustment and the current method for the rest? Or in some other way use actually on the court team or lineup level defense or shot defense data that is available these days from various sources? Adding more realistic shot defense to an otherwise boxscore metric is one of the big need for metric advancement imo.

Any consideration of including a RPM adjustment (at a modest share of total a la original Rosenbaum blend) to the final rating (or to the BPM Team Adjustment)?

[Mike G]

With Westbrook on the court, OKC has eFG% of .505, without him .476. ORtg drops from 110.3 to 100.7
Turnover% rises from 15.0 to 16.3 when he sits; turnover differential worsens from 0.9 to 2.9
Thunder ORtg-minus-DRtg is 13.8 pts/100 better with RW than without him.
http://www.basketball-reference.com/tea ... 17/on-off/

[v-zero]

As for improvements, I'd love to hear ideas. Perhaps it should be a sqrt(USG x AST) instead of its current form? I have begun work on a new version of BPM to address a few issues I've seen and to take advantage of better long-term RAPM.

In my experience the only way to achieve a stable player model when considering interaction terms is to consider all possible interaction terms and apply randomised model selection and aggregation to the problem. See this paper for an example of what I mean: https://labs.genetics.ucla.edu/horvath/ ... lkRGLM.pdf

[Mike G]

More on Westbrook's effect on OKC numbers.
This page shows differentials, that is team-minus-opponent rates, per 100 possessions:
http://www.basketball-reference.com/tea ... eups_2-man_

Of the top 20 two-man combinations, there are 14 with no mention of Westbrook.
Of these 14, 10 also appear in the 3-man combinations with Westbrook.

Code: Select all

. running mates     Min   Pts   eFG%   TO     w RW   Pts   eFG%   TO
Adams    Roberson  1758   2.5   .001   .7     1465   2.5  -.006   .2
Grant    Oladipo    560  -3.2  -.025   .8      313   0.7  -.007  1.1
Kanter   Roberson   519   9.0   .031  2.2      402  10.9   .045  2.5
Grant    Roberson   442  -1.8  -.012   .2      366  -3.6  -.016   .6
Oladipo  Sabonis    825   2.0   .018   .8      755   2.8   .017   .7
Adams    Grant      474   0.8  -.024   .9      410   2.2  -.007   .4
Adams    Oladipo   1127   6.0   .017   .5     1068   7.3   .021   .5
Oladipo  Roberson  1098   3.3   .014   .8     1057   3.5   .019  1.1
Roberson Sabonis   1006  -1.4  -.005  1.6      986   0.3  -.001  1.4
Adams    Sabonis    976   2.5   .007   .5      964   2.9   .009   .4

Toward the bottom, we see player pairs who seldom are on the court without RW. In each case, RW is usually on the court with these pairs.
So we can derive the without-Westbrook rates and compare to the with-RW rates:

Code: Select all

2man    without RW per100         with-wo per 100      net total
.    Min    Pts   eFG%    TO     Pts+   eFG%+   TO+     Pts+ TO+
A-R  293    2.5   .036   3.2       .0  -.042   -3.0       0  -18
G-O  247   -8.1  -.048    .4      8.8   .041     .7      46    4
K-R  117    2.5  -.017   1.2      8.4   .062    1.3      21    3
G-R   76    6.8   .007  -1.7    -10.4  -.023    2.3     -17    4
O-S   70   -6.6   .029   1.9      9.4  -.012   -1.2      14   -2
A-G   64  -20.2  -.134   4.1     22.4   .127   -3.7      30   -5
A-O   59  -17.6  -.056    .5     24.9   .077     .0      31    0
O-R   41   -1.9  -.115  -6.9      5.4   .134    8.0       5    7
R-S   20  -54.8  -.199  11.3     54.5   .198   -9.9      23   -4
A-S   12  -29.4  -.152   8.5     32.3   .161   -8.1       8   -2

Totals from last 2 columns are a loss of 159 points and another 14 turnovers due to no Westbrook on the floor.
In these 999 minutes, the team shot .026 worse eFG%
In 3 cases the team shoots worse with Westbrook, but in 2 of these the turnover differential more than makes up for it.

[Crow]

Not everyone will even give much attention to the offensive / defensive splits of RPM or BPM but it could could be useful to consider / argue with or against the 4 factors of RPM or RAPM if they were available again and if BPM was also broken out that way. PER2.0 probably could be as well (and perhaps Winshares / 48). Want to more productively discuss the why & where of metrics giving players credit or grief? Have factor level breakdowns of the overall rating.

[Dr Positivity]

Runs into the trouble of measuring defense with boxscore stats. OBPM is probably a little high as well, but Westbrook's 4.3 DBPM is the real egregious one. Kidd's career high was 2.8, Payton's 1.4, Jordan 2.7... RWB is rating as a defensive god by guard standards.

[Crow]

We are told than BPM's use of offensive stats for defensive ratings helps the model or overall model but Westbrook is a case where this choice probably hurts the credibility of his DRPM. I don't know how rare it is that it matters much but from the outside I worry that it is a generalized problem.

[huevonkiller]

Or really he's having the greatest season ever ?

the worst 2pt shooting team in the league now shoots only 47.0% on 2s. westbrook is shooting an even lower 45% on 2s. so he misses alot of shots (over 200 more than any other thunder player has even attempted). on average 3/4 of all missed FGA are rebounded by the opponent...

he is also committing 5.6 TO/g, the most/worst ever by a player (other than harden also this same year)...

that's a ton of zero point team possessions he is personally responsible for, about 16-17 per game (he plays 35 min/g). OKC averages about 99 poss/48min, i.e. so about 1 out of every 6 thunder team possessions they do not score because of westbrook...

does BPM take this into account?...

Understanding that rates and team performance matter too / more, fyi Westbrook was on pace to beat all-time league record for total TOs for season when I checked a couple weeks ago.

Man, (most of) you guys are funny. The only thing Westbrook is exposing, is the low energy players like James Harden and LeBron James have. I won't fault West for not being overweight and lacking stamina like those aforementioned individuals, Harden can hardly dunk anymore by the way. Yeah he's not the most efficient player, but he's scoring at a better rate than 06 Kobe per minute (see below), so he doesn't have to be. People talking about FG% and Turnovers this year are really annoying also. Seriously, the number of shots someone takes has to be taken into account. The Hollinger player stats are not kind to Westbrook and they still have his turnover rate as better than LeBron, really it is a non-issue. I can also tell most of you guys are clearly not paying attention to Thunder games, if they had floor spacing they'd be an elite team. No, Andre Roberson doesn't count, nor does a failed rookie starter who can't do anything.

Not only is Westbrook's current season in the discussion for greatest of all time, at worst he's producing at a rate that seems significantly better than Prime Kobe Bryant, which is not what I think of when I think of LeBron this year, or James Harden this year.

Let's compare Westbrook and Kobe Bryant's best season, per 100 possessions.
44.9/14.8/14.3/7.6/2.3/0.6/54.7 TS% vs 45.6/6.8/5.8/4.0/2.4/0.5/55.9 TS%.

Now if you just take into account Westbrook's extra Offensive Rebounds, he's more efficient than Kobe once you adjust TS% for that. In addition Westbrook's massive assist rate more than compensates for the turnovers (his turnover rate is LeBron-like, stop complaining).

Stated differently, Westbrook averages almost as much as Prime Kobe Bryant scoring-wise, shoots a higher percentage adjusted for offensive rebounds, and takes on additional possessions assist-wise that Kobe simply didn't have the skills or energy to do. I haven't even gotten into the fact that the Thunder rebound better when Westbrook plays, or that his playoff stats suggest he's the opposite of a choker. Since I watched pretty much every game Kobe played as a Lakers fan, I have a high appreciation for what Westbrook is doing (He's basically a rich man's Kobe). His value is pretty easy to see, if anything the only problem I have with BPM is how much it doesn't take into account the futility of comparing different, less athletic eras with the current one. That's another topic though. I'm sure James Harden will win the MVP, but I'll always remember how much Westbrook produced this year, it is simply remarkable. And he doesn't play with shooters around him like low-energy Eastern Conference loving LeBron, so I won't fault him for that either.

What a fun year. I think the Warriors will recover in time and win a title, or a surprise team will otherwise take it, imo. Anyway, the MVP is a joke award, it doesn't matter who wins that in the end.