Statistically Speaking: Does Clutch Three-Point Shooting Ability Exist?

[MaNu4Tres]

The most clutch team graded out to be the 2003 Spurs -- which makes sense subjectively. The least clutch team was the 2010 Spurs -- which could explain why that team was the worst team of the bunch.

Pretty interesting, tbh.

"We just need to hit more shots." -

[FuzzyLumpkins]

You cannot make predictions on future performance with sample sizes of 20 or even 100. The variance between values is just too significant.

[timvp]

On one hand, if my goal were to simply prove that Horry is clutch and Bonner is a choker, I should like the Vander Method more than my own because it resulted in an even larger spread. However, for the reasons I listed above, this method doesn't make as much sense to me. But maybe that's just me.




Now that I think even more about it, the fluctuation corresponds with how I thought the numbers would be skewed if we tried to include "dagger" threes. When the Spurs are nursing a late lead, they always run a pick-and-roll late in the shot clock.

1. If it's defended perfectly, the result is a contested three-pointer by the ball-handler. And if you look at that list, the last four players on the list were used primarily as ball-handlers.

2. If the ball-handler is able to penetrate and the other team helps off of a three-point shooter, that three-point shooter will get the pass and get a wide open three. That could explain why the standstill shooters all improved.



Either way, Horry grades out as clutch and Bonner grades out as a choker, so I can go forth believing in clutchness and chokers. But if you trust these numbers more, Ginobili is a massive choker in disguise.

Damn, maybe ducks was right all along

[timvp]

You cannot make predictions on future performance with sample sizes of 20 or even 100. The variance between values is just too significant.

I don't necessarily disagree. Optimally, a quality sample size in basketball is about 800 shots. But since we are dealing with something that happens so infrequently, it's just impossible to reach that number.

That said, if you throw away these numbers completely, you'd have to believe that there's no such thing as a choker. Instead, you'd have to view "chokers" as players who simply haven't been given an adequate sample size to prove themselves.





P.S.

The Vander method resulted in a sample size about 60% as large as the original method, which obviously hurts the reliability.

[Mel_13]

The average STer will prefer the first set of numbers. Data that puts Bonner, SJax, and Manu in the same group won't be well received.

[FuzzyLumpkins]

I don't necessarily disagree. Optimally, a quality sample size in basketball is about 800 shots. But since we are dealing with something that happens so infrequently, it's just impossible to reach that number.

That said, if you throw away these numbers completely, you'd have to believe that there's no such thing as a choker. Instead, you'd have to view "chokers" as players who simply haven't been given an adequate sample size to prove themselves.

When you showed the team stats and that they were within .02 it made me feel that there is little justification that it has an effect either way.

if i throw the numbers away i do not have to believe anything. i will just keep an open mind about it and not try to make something that its not.

i am not saying you are wrong; i am just saying the analysis doesn't prove anything.

[timvp]

The average STer will prefer the first set of numbers. Data that puts Bonner, SJax, and Manu in the same group won't be well received.

True.

Speaking of Bonner, his shots against Memphis are even more impressive now because outside of those two clutch threes, he has a total of only five other clutch threes in his career according to this criteria. And one was that one against the Clippers.

Matt Bonner, saving 28.6% of his clutchest shots for the playoffs

[Mel_13]

Matt Bonner, saving 28.6% of his clutchest shots for the playoffs

and making all of them....

[timvp]

When you showed the team stats and that they were within .02 it made me feel that there is little justification that it has an effect either way.

IMO, a 5% drop in accuracy over a sample size that large is significant. Especially since the drop was consistent year over year.

And, even if it isn't, it doesn't really have an impact regarding whether "clutch" exists. In theory, it's entirely possible that the "clutch" players negate the "chokers" and thus X remains X.

if i throw the numbers away i do not have to believe anything. i will just keep an open mind about it and not try to make something that its not.

I didn't mean you as in you personally. I meant you as in someone looking at this phenomenon mathematically.

i am just saying the analysis doesn't prove anything.

Agreed. Nothing has been proven in this thread. And, tbh, I don't think it's possible to prove anything. Even if you take the entire history of the NBA, one could still point to the sample size still not being large enough to prove whether there are clutch players and chokers.

[roycrikside]

I think the idea of limiting to just three point shooting is in itself subjective and flawed. I'd prefer the question of "Who's a clutch player, period?"

And there are various ways to find this... what does a guy score in the last six minutes of the 4th quarter in a game that is +/- 5 point differential per 48 minutes compared to how he scores overall per 48 minutes?

What's his PER in those situations vs. regular?

What's his overall shooting percentage and FT shooting percentage?

What's his +/-?

I think all those are more significant, but I understand that it's problematic to research all that stuff.

[DAF86]

Manu in the clutch goes to the rim, tbh.

[timvp]

I think the idea of limiting to just three point shooting is in itself subjective and flawed.

My original goal was to figure out whether clutch three-point shooters exist, hence the le of the thread. Whether clutch players exist is another question entirely.

I'd prefer the question of "Who's a clutch player, period?"

...

I think all those are more significant, but I understand that it's problematic to research all that stuff.

That's an interesting question too. It'd obviously be a lot more difficult to research but I might try tackling it.

[timvp]

Manu in the clutch goes to the rim, tbh.

Manu has more than twice as many three-point attempts as anyone else, tbh.

[FuzzyLumpkins]

IMO, a 5% drop in accuracy over a sample size that large is significant. Especially since the drop was consistent year over year.

And, even if it isn't, it doesn't really have an impact regarding whether "clutch" exists. In theory, it's entirely possible that the "clutch" players negate the "chokers" and thus X remains X.

I didn't mean you as in you personally. I meant you as in someone looking at this phenomenon mathematically.

Agreed. Nothing has been proven in this thread. And, tbh, I don't think it's possible to prove anything. Even if you take the entire history of the NBA, one could still point to the sample size still not being large enough to prove whether there are clutch players and chokers.

if you were to take the vander method and apply it to the league as a whole over a couple of years and compare that to the league averages overall during that time and there is a significant difference i think you will have gone a long way in showing that the phenomenon exists.

[Brazil]

What does com stand for?

really ?

CoM CoP

[Brazil]

F/C Matt Bonner has no illusions of what his job is in San Antonio. He takes three-pointers and hustles. That's about it. "It's just a matter of roles," Bonner said. "My role is to stretch the court, shoot the ball when I'm open."
(Yahoo! Sports)

[angelbelow]

really ?

CoM CoP

I have no idea what either mean.

[Mel_13]

I have no idea what either mean.

Church of Manu, Church of Parker.

[angelbelow]

Church of Manu, Church of Parker.

Ahh, okay thanks.

[gospursgojas]

Awesome thread Timvp.... Your stat threads are way more relevant and make much more sense than anything Hollinger tries to do.

Props on RTB and bringing ST back to life.

[DAF86]

Manu has more than twice as many three-point attempts as anyone else, tbh.

That's 'cause he probably has played more than twice as many games as most of those guys, tbh.

[Drz]

Statistician here. I hate to be a thread-crapper, but the analysis is extremely flawed.

My original goal was to figure out whether clutch three-point shooters exist, hence the le of the thread. Whether clutch players exist is another question entirely.

If that was truly your goal, you went about it wrong. Like FuzzyLumpkins said, an analysis with thousands of three-point shots, at a bare minimum, would be needed to get results with a high degree of confidence. I suspect you limited your analysis to Spurs players due 1) a lack of knowledge of how to achieve your goal, or 2) because you wanted some numbers that made Horry look good and Bonner look bad. Based on your other posts, I'm leaning towards the latter.

Here is the crux of the problem: Shots made WILL have a distribution, and roughly half the players will perform better than expected, and half worse. Next point bolded for emphasis -- Your analyis makes the (very) faulty assumption that if a player has missed more than their expectation, they must not be "clutch." In reality, that a player missed more than their expectation does not necessarily tell you one single thing about their future performance. Literally, nothing. Not one iota.

Your analysis is nothing more than counting buckets, then pointing to the bad performers and saying "not clutch!" and pointing to the people who had a good count and saying "clutch!" Of course you're going to find "evidence" of clutchness if you use that method. Half will be clutch, half won't.

[timvp]

Statistician here. I hate to be a thread-crapper, but the analysis is extremely flawed.

I've admitted as much in just about every post in this thread, tbh.

If that was truly your goal, you went about it wrong. Like FuzzyLumpkins said, an analysis with thousands of three-point shots, at a bare minimum, would be needed to get results with a high degree of confidence.

NBA players simply don't shoot "thousands of three-point shots" during clutch moments of games throughout their careers. Hence the sample size issues discussed throughout the thread.

I suspect you limited your analysis to Spurs players due 1) a lack of knowledge of how to achieve your goal, or 2) because you wanted some numbers that made Horry look good and Bonner look bad. Based on your other posts, I'm leaning towards the latter.

I limited it to the Spurs because this is a Spurs forum. No one asked for a look at other players around the league. I was planning on checking it out -- and probably will -- but I doubt many in here will care about the stats of other players.

And if you had read the thread, you would see that the criteria I selected made Bonner and Horry appear closer than the criteria suggested by others. If my goal was simply to make Bonner look bad and Horry look good, I could have concocted a much more "creative" formula.

Bonner, by any measure, has hit less than about a dozen "clutch" threes in his Spurs career. It'd be quite easy to negate just about all of them if that were my goal.

Your analyis makes the (very) faulty assumption that if a player has missed more than their expectation, they must not be "clutch." In reality, that a player missed more than their expectation does not necessarily tell you one single thing about their future performance. Literally, nothing. Not one iota.

That makes no sense. If the sample sizes were large enough, we could absolutely use such data to predict future outcomes. For example, if we had the thousands of shots necessary to create a reliable sample size you mentioned earlier in your post and I mentioned earlier in the thread, the resulting percentages would be just as reliable as any other percentages used in the game of basketball.

Of course you're going to find "evidence" of clutchness if you use that method. Half will be clutch, half won't.

Again, I was the first to say it's flawed. But did you look at the numbers? The first set ended up with 4-of-13 players rating out as "clutch". The second set had 4-of-12. In both cases, players who otherwise shot about 38% on three-pointers shot approximately 33% in "clutch" situations. That's hardly a redistribution of the original percentages and then a half and half split, tbh.

[bugmenot]

My 1-5. My favorites are 99% based on personality, and 1% on playing ability.

1. Matt Bonner -- Funny. Smart. Unique.
2. Tim Duncan -- The hidden wizard/nerd/paintball side is awesome.
3. Manu Ginobili -- For the obvious reasons.
4. Richard Jefferson -- I've been impressed with how he's handled his role on the Spurs, and how he handles the negative attention.
5. Gary Neal -- Liked him ever since watching his ESPN OTL video.

I almost ran into Matt Bonner and his wife at the Boerne Stage Road HEB maybe 30 minutes ago. I was flying down the center aisle as his wife, who I didn't recognize, came out from the right, and we stopped in a near-collision. She said "watch out, Matt" and I expected a little boy (her son) to come from the aisle. But instead, out whooshes Matt Bonner, deftly pushing a stroller. What made it especially fun was that he physically said the word "whoosh!" as he drove by. He then promptly drained a 3.

Highlight of my life.

Help me understand -- he only shot 27 threes in the playoffs and made 10 for a .370 %. Given the small sample size and the not unrespectable 3P%, why is everyone saying he's a playoff choke artist?

Statistician here.

: This person only adds to the discussion to talk about his love for Bonner; and now he is a statistician with knowledge in data mining and analysis when Bonners clutch is questioned!

[timvp]

Some pretty interesting numbers...

For the last four seasons, I found the league-wide "clutch" three-point shooting percentage using the original criteria. I then compared those percentages to the percentage of three-pointers shot in all other cir stances.

Clutch Three-Point Shooting
2011: 829-2585 (32.07%)
2010: 832-2556 (32.55%)
2009: 933-2824 (33.04%)
2009: 813-2451 (33.17%)

Other Three-Point Shooting
2011: 15057-41728 (36.08%)
2010: 14990-42066 (35.63%)
2009: 15419-41759 (36.92%)
2008: 15311-42093 (36.37%)

Effect of Clutch Situation on Normal Accuracy
2011: -11%
2010: -9%
2009: -10%
2008: -9%


That is a shockingly consistent negative effect, tbh