Measuring Clutch Play

[boooeee]

New post up: Measuring Clutch Play in the NBA

Here's a summary:

I use my win probability model to split the play by play data into two categories: "clutch" and "garbage time". Clutch time is when the magnitude of a play's Win Probability Added WPA is greater than average. For example, Ray Allen's game tying three from Game 6 of the Finals was worth +35% WPA, but on average a three pointer adds about 5% in WPA, so this is an example of clutch time (obvious in this case, I know). Garbage time is when the magnitude of a play's WPA is below average. For example, if a player makes a two point field goal with 5 minutes left in the game and his team up by 20, the actual WPA of that play is +0%. Since an average two pointer is worth +2.2%, this is a garbage time play.

With these definitions, I can break down a player's Win Probability Added (WPA) into three components:

• expected WPA (eWPA): This is the box score version of WPA, based on the average WPA contribution of each type of play (made shot, missed shot, turnover, etc.)
• clutch WPA (clWPA): For a clutch time play, the amount by which the play exceeded (either positively or negatively) the corresponding eWPA
• garbage time WPA (gbWPA): For a garbage time play, the amount by which the play's eWPA exceeded the actual WPA

WPA can then be expressed as follows:

WPA = eWPA + clWPA - gbWPA

Note the sign on gbWPA. It's negative so that a large positive value of gbWPA would indicative of a "stat padder": someone that has high eWPA, but low actual WPA.

On a volume basis, James Harden has the highest clutch WPA at +3.09. This means that his actual WPA exceeds his expected WPA (based on his box score stats) by more than 3 wins. Lebron is a close second in clutch WPA with +2.80.

Dwight Howard is the worst clutch free throw shooter (once again, on a volume basis). Oddly enough, Kevin Love is #2 in negative clutch free throw WPA. Missing three free throws with 2 seconds left on the clock and your team trailing by 2 will do that.

[Mike G]

Wow, a lot going on in that article. I definitely think you are on a good track here, both in defining 'clutch' and 'garbage' production.

It bothers me some that your limitation is still to shots, rebounds, and turnovers. You know someone got the ball to Ray Allen before his big shot; that would seem to have a big WPA.

Short of giving equal credit to all players on the floor (ala plus-minus), I don't know where one would stop in assigning WPA to boxscore stats. Maybe the whole enchilada is easier to cook than trying to recreate from the ingredients?

Lastly, apologies if this has already been covered, but you give this list of normal WPA for these events:

Code: Select all

. play type           eWPA
turnover             -2.0%
missed field goal    -1.5%
missed free throw    -1.6%
made free throw      +0.5%
getting 1 foul shot  +1.6%
getting 2 foul shots +1.5%
getting 3 foul shots +3.3%
make 2 point shot    +2.2%
make 3 point shot    +4.6%
defensive rebound    +0.8%
offensive rebound    +1.5%

You're treating the act of getting to the line as an event, separate from making or missing the FTA. This is interesting and probably necessary.
How is it that an OReb is roughly twice as important as a DReb? -- The DReb denies your opponent the OReb.

[boooeee]

Wow, a lot going on in that article. I definitely think you are on a good track here, both in defining 'clutch' and 'garbage' production.

It bothers me some that your limitation is still to shots, rebounds, and turnovers. You know someone got the ball to Ray Allen before his big shot; that would seem to have a big WPA.

Short of giving equal credit to all players on the floor (ala plus-minus), I don't know where one would stop in assigning WPA to boxscore stats. Maybe the whole enchilada is easier to cook than trying to recreate from the ingredients?

Lastly, apologies if this has already been covered, but you give this list of normal WPA for these events:

Code: Select all

. play type           eWPA
turnover             -2.0%
missed field goal    -1.5%
missed free throw    -1.6%
made free throw      +0.5%
getting 1 foul shot  +1.6%
getting 2 foul shots +1.5%
getting 3 foul shots +3.3%
make 2 point shot    +2.2%
make 3 point shot    +4.6%
defensive rebound    +0.8%
offensive rebound    +1.5%

You're treating the act of getting to the line as an event, separate from making or missing the FTA. This is interesting and probably necessary.
How is it that an OReb is roughly twice as important as a DReb? -- The DReb denies your opponent the OReb.

Thanks Mike. Once I have this all operationalized, I think I can do clutch for the other stats like rebounds, assists, blocks, and steals.

My main motivation for splitting out getting to the line and taking the foul shots is that then I could focus on clutch free throw shooting.

The Oreb vs. Dreb is a bit tricky. The difference in WPA is due to Dreb being the more likely outcome. It is true that, in the end, possession is possession. But after a shot is missed, you have a 64% chance of losing possession (Dreb) and a 34% chance of retaining possession (Oreb). So, an Oreb adds, effectively, an expected +0.64 possessions. I pretty much had to build it this way so that I could properly assign win probability to missed shots (if a player missed a shot, they shouldn't get credit if one of their teammates happens to rebound the miss).