Win Probability Graphs and Box Scores for the 2013 Season

[boooeee]

For those that are interested, I now have Win Probability Graphs and Box Scores (Win Probability Added at a player level) for every game of the current NBA season (through Christmas).

Graphs and Box Scores
Introductory Post

Any feedback on presentation, features, or methodology would be welcome. I plan on keeping this updated throughout the season. A longer term project will be to add previous seasons (I believe play by play data is available all the way back to the 1996-97 season).

[Mike G]

This is interesting. Visually appealing, too.
When I checked the link, the game at top was the GSW 105-103 win over LAC. Here are those Clippers, as you listed them:

Code: Select all

Player       Pos   WPA   FGA  eFG%  shWPA   TO  toWPA   FTA   FT%  ftWPA
B. Griffin   FC    0.22   16   50%   0.23    4  -0.01    6    67%   0.01
J. Crawford   G    0.15   21   45%   0.22    2  -0.07    0     0%   0.00
D. Collison  PG    0.14    6   75%   0.14    0   0.00    2   100%   0.01
D. Jordan     C    0.10    6  100%   0.13    3  -0.03    2    50%  -0.04
W. Green     SG    0.08    2   75%   0.08    0   0.00    2   100%   0.02
A. Jamison   PF    0.03    2   50%   0.03    0   0.00    2   100%   0.01
R. Hollins    C   -0.02    0    0%  -0.02    0   0.00    2    50%  -0.02
J. Dudley    GF   -0.03    5   40%  -0.03    0   0.00    0     0%   0.00
M. Barnes    SF   -0.05    3    0%  -0.03    1  -0.02    0     0%   0.00
C. Paul      PG   -0.06   20   47%   0.01    3  -0.07    7   100%   0.10

You say Griffin was mvp of the game, and Paul was lvp (least valuable).
Paul had 26 Pts (.563 ts%), 11 Ast, 2 Stl, 3 TO. But last in WPA

In an analysis of another game (Nets), you wrote:

Paul Pierce was the MVP (in a losing effort), having added an impressive 78% in Win Probability on some very efficient shooting (9 field goal attempts, with a 100% eFG%). Nearly half of Pierce's WPA came on a single shot: a three pointer with 16 seconds left in overtime that put the Nets up by one point.

Some large part of me wants to say a shot made in the first quarter improves one's chances of winning the same amount as a shot at the buzzer. It's good to reward clutch play -- and discount garbage time play -- but it doesn't seem right to treat early contributions as equivalent to garbage time, in win probability added.

Suppose a player attempts 2 shots in a game. If he missed his earlier shot and hit the game winner, that's huge in wpa. If he'd hit the earlier one, the 2nd one might be a lot less important -- in which case, wpa rewards him for his earlier miss.

[Crow]

It would take work but it would be interesting to summarize available data for certain circumstances such as when team win probability reaches 25% or 10% by halftime or end of third quarter. What happens and by when for teams that reverse this negative outlook and come back and win? How much does the strategy (frequency of 3 pt shots, fouls given, etc.) vary from what the losers do?

[boooeee]

Some large part of me wants to say a shot made in the first quarter improves one's chances of winning the same amount as a shot at the buzzer. It's good to reward clutch play -- and discount garbage time play -- but it doesn't seem right to treat early contributions as equivalent to garbage time, in win probability added.

Suppose a player attempts 2 shots in a game. If he missed his earlier shot and hit the game winner, that's huge in wpa. If he'd hit the earlier one, the 2nd one might be a lot less important -- in which case, wpa rewards him for his earlier miss.

I understand the rationale of thinking that "all points are equal", but statistically, that is certainly not the case. Winning a basketball game is not just about scoring more points, it's about having more points at a specific point in time (i.e. end of Q4/OT). All that being said, I doubt that WPA has much value from a predictive point of view. Players that score well probably do so regardless of context, and variations in WPA can most likely be chalked up to random variation. But that is an empirical question that I hope to address (do certain players consistently over/under perform in the clutch?). In a nutshell, WPA is more likely a "fan" stat, rather than a "front office" stat.

Also, just to clarify, early game points are not equivalent to garbage time. In that LAC-GSW game, Chris Paul made a layup with 3:47 to go in the first quarter to put the Clippers up by 9. That was worth +0.8% in WPA. Contrast that to a true garbage time situation, like 4th quarter of the Xmas OKC-NYK game. The win probability graph had flatlined, so Derek Fisher's three pointer with 5:57 to go was worth exactly zero in WPA.

[boooeee]

It would take work but it would be interesting to summarize available data for certain circumstances such as when team win probability reaches 25% or 10% by halftime or end of quarter. What happens and by when for teams that reverse this negative outlook and come back and win? How much does the strategy (frequency of 3 pt shots, fouls given, etc.) vary from what the losers do?

That is definitely my goal. I hope to have a robust, multi-season play-by-play dataset with a win probability attached to each play, which you can then slice and dice at will. It will probably take some time though (for pre-2012 season data, I'll probably have to parse the rather messy NBA play by play feed).

[mtamada]

Cool, although the lack of ability to apportion responsibility for defense and rebounds, and also assists, is a limitation.

The ratings for "excitement" and "comeback" are good innovations, filling a major gap that even Fangraphs' WPA game graphs fail to address. However we readers need more guidance: what is the mean or median excitement rating for games (either for this season so far, or for last season)? And what is the mean or median comeback rating?

[deepak]

Thanks for doing this.

What do these stats tells us about the "importance" of a given possession in a basketball game? I ask, because I find it often the case that people tend to advance the notion that all possessions, whether they come in the first or fourth quarter, are equally important. Consider a recent tweet from Rockets GM Daryl Morey:

"New Years wish -- we understand a possession in the 1st half is just as important as the 2nd half"

I understand that we'd like our basketball players to play every possession with equal intensity and focus, ideally, but as a matter of basic logic it never made much sense to me treat all points as equally valuable or important. Your graphs illustrate the point well. A point that comes late in a close game can dramatically alter the probability of one team winning the game. The curves are flatter, less volatile, in the first quarter. Isn't it undeniable that certain moments in the game are more critical than others?

[xkonk]

I've seen a lot of relevant discussion at advancednflstats on the distinction between Brian's Expected Points Added (EPA) and Win Probability Added (WPA) stats. That WPA seems to be awfully similar to this version for the NBA; EPA is similar but based on how many points teams have tended to score given their field position and down-to-go. The critical difference is that EPA is calculated entirely from 'normal' game states (http://www.advancednflstats.com/2010/01 ... oints.html).

The idea for the NBA would presumably be the same. If a player hits a game-winning shot, then he won the game. That's pretty valuable. Obviously another player could have made the shot or that same player may miss that shot sometimes, but what happened happened and it gave his team a win. If that same player hits the same shot in the first quarter, it can only contribute so much to winning that particular game because there's so much time left to determine the winner. Both shots would be equal in some manner of NBA EPA, though. Most of the NBA player summary stats everyone knows about here (APM, RAPM, etc) are expressed as points added or something that can be converted to points added, and as long as they don't give extra weight to clutch time should be the rough equivalent of EPA, I would think.

The general interpretation is that EPA is more a measure of a player's ability, and thus is more relevant in predictive situations, and WPA is a measure of what a player has actually contributed, and thus is more of a summary stat (Brian at ANFLS calls it a 'story telling' stat). Empirically speaking, at least for quarterbacks, EPA is more stable than WPA, which makes sense because WPA can swing wildly on the outcome of a few plays.

[Mike G]

A player who goes 0-10 before hitting the game-winner surely understands that if he'd hit his normal 5-10, the 'game winner' would have been just another shot with ~zero impact on the game. His teammates know it, including the one who hit 10 of 10 before missing one near the end.

The guy who went 10-11 probably got a huge negative WPA for the late miss, while the 1-11 guy was probably the "mvp".
Of course all points count the same. Players who don't believe it are a menace to team chemistry and function.

A couple of years ago on this topic, someone mentioned that for people who bet on games, the 'value' of a shot (or other play) may be all about whether it beats the spread. From that perspective, win/lose is irrelevant; and the 'importance' of any play is arbitrarily defined by one's definition of "win".

[xkonk]

A player who goes 0-10 before hitting the game-winner surely understands that if he'd hit his normal 5-10, the 'game winner' would have been just another shot with ~zero impact on the game. His teammates know it, including the one who hit 10 of 10 before missing one near the end.

This is true to the extent that you assume strategy doesn't change in-game to account for what has happened earlier.

But in any case, it's a clear example of what we usually do as stats-focused people, which is shift our thinking towards 'what if'. You don't expect first player to go 0-10 or the second player to go 10-10; it happens sometimes but you presumably wouldn't evaluate those players strictly on those numbers going forward.

WPA, on the other hand, doesn't care a lick for 'what if' and cares only about 'what happened'. If I offered you a bet on the outcome of yesterday's game and said you could choose between the team with the guy who hit the game winner and the team with the guy who went 10-10, you would be a fool to take the second team. If your interest is in knowing which players led to a team winning in the particular way the games unfolded, then WPA is that stat you want. If it isn't, then WPA probably isn't of interest in the least.

[Mike G]

If your interest is in knowing which players led to a team winning in the particular way the games unfolded, then WPA is that stat you want. If it isn't, then WPA probably isn't of interest in the least.

Everything in the environment of a win is not of equal import. The player who missed his first 10 shots was not one who " led to a team winning" any more than was the hot dog vendor who had a good night or a bad night.

Missing 10 straight shots puts your team in roughly a 10-point hole, relative to average NBA rates of scoring.
Making 1 shot retrieves a couple of those points, but there's still a deficit. Very few players are 10 points better than average in non-scoring benefits.

The 1-11 shooter had to have some help to still be in the game with seconds left. His teammates had to somehow overcome his terrible shooting.

Hopefully this is a useful example. Suppose the Pistons are tied with 20 seconds to play and the ball.
Andre Drummond is fouled with 0.2 seconds left. He needs to make at least one of two FT.

Well, he's a 36% FT shooter. The probability that he'll miss both is 40%. The Pistons, then, are 60% likely to win in regulation. So about 80% that they'll win either in 48 min. or in overtime.

Then he misses the first FT. Now there's a 36% chance that they'll win in regulation, and 64% that it goes to OT. If the teams are evenly matched, it's therefore (36+32) 68% that they'll win eventually.

The missed FT cost Drummond 12% in wpa
If he makes the 2nd one, Pistons W% increases by 32. The net is +20

If he'd made the first FT, odds increased from 80 to 100 -- also a gain of 20.
The 2nd FT is either worth 20, or zero.

[mtamada]

(Brian at ANFLS calls it a 'story telling' stat). Empirically speaking, at least for quarterbacks, EPA is more stable than WPA, which makes sense because WPA can swing wildly on the outcome of a few plays.

That's exactly the phrase that I was thinking of: "story-telling stat". WPA tells us what actually happened, a way of quantitatively describing who contributed what to the game (subject to the limitation that we don't always have good ways of measuring this contribution, e.g. maybe LaMarcus Aldridge sank the game-winning shot, but was set up by a great pass from Lillard). But in the long run, or for predictive purposes, we can expect WPA to have poor predictive value relative to some measure of EPtsA or for that matter other measures such as RAPM, PER, etc.

[xkonk]

Mike, I think we both understand how WPA works (and your example is very clear). My sense is more that you (and apparently Daryl Morey) fundamentally disagree with the idea of it. Would that be fair?

[mtamada]

Mike, I think we both understand how WPA works (and your example is very clear). My sense is more that you (and apparently Daryl Morey) fundamentally disagree with the idea of it. Would that be fair?

It depends on what WPA is being used for. For evaluating players, predicting the future, etc. it is inferior to measures which try to measure the players' inherent ability, and which are less affected by situation, context, etc.

But for describing what happened in a a game or a season, it's a fine tool. E.g. it could be an element for deciding who should be the MVP; i.e. who accomplished the most, as opposed to deciding who is the NBA's best player. Or for measuring which games were most exciting or had the most stupendous comebacks, as the inpredictable site is trying to do.

[Mike G]

Such as declaring Chris Paul the 'least valuable player' in a game, which is the most erroneous conclusion possible?
It's essentially like watching the game 'highlights' vs watching the whole game. Which really tells more about 'what happened'?

It would be more interesting if it were like a plus-minus summary of all players on the floor, rather than just the one who made/missed the shot. It may well be that equal/more credit should go to the guy who set the pick, or made the pass, or the previous pass, or who got the offensive rebound and kicked it out ...