PDA

View Full Version : The Official Hot Hand Thread



MChambers
02-28-2014, 07:06 AM
I've seen the existence of the hot hand debated here many times, but could not find a thread dedicated to it. There's a new paper to be presented very shortly at the Sloan Sports Conference that purports to find some evidence of the hot hand. I haven't read the paper yet, but it sounds interesting.

http://www.sloansportsconference.com/wp-content/uploads/2014/02/2014_SSAC_The-Hot-Hand-A-New-Approach.pdf

Have at it!

cspan37421
02-28-2014, 07:17 AM
I was totally prepared to groan at yet another "not it doesn't!" "Yes it does" back and forth, but this paper actually looks like it moves the thought process forward.

But really, academic research sponsored by ESPN and Ticketmaster, complete with logo on every page? I guess this didn't make it into the peer-reviewed literature yet? I have but a BS in math to bring to this, enough to appreciate it to some degree, but not enough to professionally/thoroughly critique it. I'm going to read through it at some point, though.

MCFinARL
02-28-2014, 07:37 AM
I was totally prepared to groan at yet another "not it doesn't!" "Yes it does" back and forth, but this paper actually looks like it moves the thought process forward.

But really, academic research sponsored by ESPN and Ticketmaster, complete with logo on every page? I guess this didn't make it into the peer-reviewed literature yet? I have but a BS in math to bring to this, enough to appreciate it to some degree, but not enough to professionally/thoroughly critique it. I'm going to read through it at some point, though.

Admittedly, that is weird, but increasingly, academic research is sponsored by somebody--drug companies, technology companies, the government, foundations--and the researchers go where the resources are. The one bright side for this one--the sponsors have an interest in finding the answer to this question, but they probably don't care what the answer is, thus are not especially likely to bias the research.

cspan37421
02-28-2014, 07:41 AM
The one bright side for this one--the sponsors have an interest in finding the answer to this question, but they probably don't care what the answer is, thus are not especially likely to bias the research.

Just curious why you think that. I could easily think of reasons why ESPN might like a "hot hand" to be a real thing - it drives viewership. "LBJ is on fire tonight - you gotta turn it on!" The "no hot hand" people seem to be viewed as a wet blanket/party pooper/Debbie Downer.

MCFinARL
02-28-2014, 07:50 AM
Just curious why you think that. I could easily think of reasons why ESPN might like a "hot hand" to be a real thing - it drives viewership. "LBJ is on fire tonight - you gotta turn it on!" The "no hot hand" people seem to be viewed as a wet blanket/party pooper/Debbie Downer.

Well, that is a fair point. I was just thinking of it from a stats geek point of view--something they could throw into the data mix. But you are likely right that the hot hand is more marketable.

superdave
02-28-2014, 09:57 AM
Check out Andre's game log for this season: http://espn.go.com/mens-college-basketball/player/gamelog/_/id/46255/andre-dawkins

In games where Andre attempts three or fewer 3-point attempts he shoots .375 (6-16). In games where Andre shoots more than three 3-point attempts, he is .457 for the season (53-116).

For the season, Andre is .447 (59-132). You could either call this a "hot hand" or you could consider it good 3-point defense vs bad 3-point defense. I suspect there is some of both going on here.

Here's another way to look at it that is far more telling: In games where Andre shoots .500 or better, he shoots .636 (42-66). In games where Andre shoots under .500, he shoots .258 (17-66).

So when he's hot, Andre is really hot.

Super "I have no idea if this means anything" Dave

gus
02-28-2014, 10:21 AM
It's certainly an interesting paper -- it was published a few months ago (i think).

I'm glad they acknowledge that the lynchpin of the analysis is pretty suspect.


There are several drawbacks to the two stage empirical strategy we employ here of first estimating a Shot Difficulty regression and then using those results to test for the Hot Hand. The principle worry is errors-in-variables: if our Pˆ’s are measured too imprecisely, they may produce a biased estimator. There is reason to suspect that, despite our best efforts and the extraordinary dataset at our disposal, our Shot Difficulty model does not control for or does not correctly specify individual shot difficulties. Specific concerns include: not being able to track appendages, not specifying relationships between variables (i.e., distance from the basket, defender distance, etc.) correctly, and that player fixed effects are not precise enough to accurately estimate Pˆ for individual players. Player fixed effects change the intercept of the probability that a shot goes in for each player, but they do not account for the fact that certain players “specialize” in certain shots - in other words, they do not adjust the slopes.

I strongly suspect that their "1.2 to 2.4" p.p. increasein liklihood of a make (adjusted for difficulty to account for the real and explainable drop in actual liklihood) is more a function of how they chose to correct for difficulty of shot.

I also think people do no consider "hot hand" to be such a small increase in liklihood. For context, for Dawkins - an increase of 2.4pp on subsequent shots would translate to one more made three about every 14 games.

Regardless, it's still an interesting article.

MCFinARL
02-28-2014, 10:38 AM
Check out Andre's game log for this season: http://espn.go.com/mens-college-basketball/player/gamelog/_/id/46255/andre-dawkins

In games where Andre attempts three or fewer 3-point attempts he shoots .375 (6-16). In games where Andre shoots more than three 3-point attempts, he is .457 for the season (53-116).

For the season, Andre is .447 (59-132). You could either call this a "hot hand" or you could consider it good 3-point defense vs bad 3-point defense. I suspect there is some of both going on here.

Here's another way to look at it that is far more telling: In games where Andre shoots .500 or better, he shoots .636 (42-66). In games where Andre shoots under .500, he shoots .258 (17-66).

So when he's hot, Andre is really hot.

Super "I have no idea if this means anything" Dave

Let me just start by saying I have no idea either. But--your second stat seems more suggestive of a hot hand than the first.

First, there is the chicken-egg question: does Andre hit fewer shots because he shoots fewer shots, or does he shoot fewer shots because he hits fewer shots? It's possible, of course, that when Andre is "hot" he gets more opportunities to take good shots because his teammates look for him more, and thus makes more. It's also possible that sometimes Andre does better as the game goes on--so if he takes more shots, he hits a higher percentage of them. And it's possible that, if Andre doesn't hit his first couple, his teammates stop looking for him (or he gets fewer minutes), so we don't really know whether he would have hit a higher percentage in that particular game if he had taken additional shots or not.

As you pointed out, 3-point defense plays a role as well--we have seen some games (like the one at UNC, IIRC) where defenders are all over Andre and others where he has more room. And I suppose his perceived role on the team may play a part as well--Coach K looks to Andre for an offensive spark of the bench and for zone busting, so if that doesn't come in the first couple of shots, sometimes a starter will be subbed back in sooner. Again--not clear whether Andre really isn't hot that game or not or whether he just didn't hit his first couple of shots and would have hit more with more time.

I know from time to time people on DBR have advanced the theory that if Andre doesn't hit his first couple of shots, he will be cold all night--but I think your data suggest that assumption is based on insufficient data, because unless someone else is in foul trouble or ailing, Andre often doesn't get enough minutes or enough shooting opportunities to test whether he is hot if he doesn't hit on those first couple shots.

Also--I suspect people who know statistical analysis (which would not be me) might find the sample size for games with 3 or fewer 3-point attempts a bit small.

But this was interesting and I am glad you looked at these figures. And it is very interesting to see the big difference between his average percentage in better shooting games and that in worse shooting games.

gus
02-28-2014, 10:44 AM
Check out Andre's game log for this season: http://espn.go.com/mens-college-basketball/player/gamelog/_/id/46255/andre-dawkins

In games where Andre attempts three or fewer 3-point attempts he shoots .375 (6-16). In games where Andre shoots more than three 3-point attempts, he is .457 for the season (53-116).

For the season, Andre is .447 (59-132). You could either call this a "hot hand" or you could consider it good 3-point defense vs bad 3-point defense. I suspect there is some of both going on here.

Here's another way to look at it that is far more telling: In games where Andre shoots .500 or better, he shoots .636 (42-66). In games where Andre shoots under .500, he shoots .258 (17-66).

So when he's hot, Andre is really hot.

Super "I have no idea if this means anything" Dave

I think defense is the answer, as it limits his shots.

If you turn one miss into a make in the first group (3 shots or fewer), and one make to a miss in the second (more than three shots), the percentages are 43.8% and 44.8%. Not sure I'd use that as an argument that "when he's hot, he's really hot". Your second analysis is the definition of sampling bias.

brevity
02-28-2014, 12:48 PM
But really, academic research sponsored by ESPN and Ticketmaster, complete with logo on every page?

I can accept that a 10-page research study would have corporate sponsors listed in the header and footer. So I started reading it. But then I gave up when I got to page 5 and only saw this.

3973

hurleyfor3
02-28-2014, 12:56 PM
Those of us who read Bill James back in the 80s collectively yawn.

hurleyfor3
02-28-2014, 12:58 PM
academic research sponsored by ESPN and Ticketmaster

Aren't they competitors? One wants you to go to the game, the other wants you to sit at home and watch it on teevee.

gus
02-28-2014, 01:06 PM
I decided to break out a random number generator again, and ran 29 games with a 45% shooter who took a random number of shots, from 1-7.



In games where Andre attempts three or fewer 3-point attempts he shoots .375 (6-16). In games where Andre shoots more than three 3-point attempts, he is .457 for the season (53-116).

I ran the simulation a few times -- here are the first 10 results of this random "shooter":


<=3 >3
1 45.45% 51.92%
2 45.00% 45.63%
3 48.39% 54.22%
4 46.15% 51.72%
5 51.52% 52.33%
6 44.83% 56.25%
7 50.00% 37.50%
8 42.11% 56.52%
9 40.00% 59.21%
10 52.00% 57.32%
There's often a substantial departure between the percentage.


Here's another way to look at it that is far more telling: In games where Andre shoots .500 or better, he shoots .636 (42-66). In games where Andre shoots under .500, he shoots .258 (17-66).

This analysis is a very clear cut case of sampling bias. You're really not saying anything more than "games where he shoots well, he shoots well."

To illustrate that, here is my random "shooter":

<50% >=50%
1 15.38% 60.00%
2 14.81% 62.39%
3 5.88% 63.37%
4 10.71% 62.07%
5 10.00% 62.37%
6 7.14% 61.47%
7 0.00% 60.87%
8 8.70% 60.53%
9 8.33% 61.29%
10 7.69% 64.20%

Note, these are 20 different simulations, so the two outputs don't correspond.

Listen to Quants
02-28-2014, 01:14 PM
It's certainly an interesting paper -- it was published a few months ago (i think).

I'm glad they acknowledge that the lynchpin of the analysis is pretty suspect.



I strongly suspect that their "1.2 to 2.4" p.p. increasein liklihood of a make (adjusted for difficulty to account for the real and explainable drop in actual liklihood) is more a function of how they chose to correct for difficulty of shot.

I also think people do no consider "hot hand" to be such a small increase in liklihood. For context, for Dawkins - an increase of 2.4pp on subsequent shots would translate to one more made three about every 14 games.

Regardless, it's still an interesting article.


This point deserves emphasis. While stuff like defensive adjustment etc. cloud the efforts to exactly measure its existence, the evidence is that 'hot hand' if it exists is materially insignificant. That conclusion has important ramifications for a coach, e.g., do not base substitution on "hotness" :) and emphasize to players that even when feeling 'it' a shooter should still take only good shots.

superdave
02-28-2014, 01:15 PM
I guess the best way to understand what is going on with Andre is sometimes the defense really gets up on him and his looks are bad or his teammates just cannot find him with much daylight. But on those nights where our opponent does the illogical - giving Andre room - he's pretty awesome.

There's no real statistical story here. But these numbers do indicate how good or bad our opponent's defense of Andre specifically is.

Either way, I do think it points to us needing to get Andre more shots and more minutes.

PS: Thanks to those of you who know stats. There's always a good explanation for these things, and this board finds it quickly.

Kedsy
02-28-2014, 01:54 PM
It's possible, of course, that when Andre is "hot" he gets more opportunities to take good shots because his teammates look for him more, and thus makes more. It's also possible that sometimes Andre does better as the game goes on--so if he takes more shots, he hits a higher percentage of them. And it's possible that, if Andre doesn't hit his first couple, his teammates stop looking for him (or he gets fewer minutes), so we don't really know whether he would have hit a higher percentage in that particular game if he had taken additional shots or not.

As always when discussing this subject, the hardest parts to quantify are the psychological components. You've hit on several of these components, namely how hard his teammates try to get him open and how often they look to pass him the ball and (parenthetically) how many minutes he plays (or doesn't play) because the coaching staff thinks he's hot (or cold). Also, if the shooter himself feels hot, he probably also tries harder to move without the ball to get in good position for shots, and if he feels cold he probably doesn't work as hard for his own shot.

The scoffers can scoff as they often do in these discussions but when I play, all the above are true (except the coaching because at my paltry level there are no coaches). If I feel a teammate is hot I'm looking for him and pass him the ball both more often and in better positions where he can get off a good shot. If I feel hot myself, I work harder to get open, my teammates seem more likely to pass to me, and I think shot first instead of pass first. If I feel cold, I set more screens and when I get the ball I think pass first and hesitate before shooting myself (and hesitation almost always throws off my shooting form). All those things most likely have a very real effect on how many shots I take and how high a percentage of them are makes.

Of course, if a shooter feels super-hot, he might take a "heat check" shot that is far less likely to go in (almost every player I know who thinks he's hot does this at least once in awhile), and that could possibly improperly skew the results of a study in the "hot doesn't exist" direction.

I additionally believe that a shooter's state of mind can affect his form. Sometimes I feel good and I know my shots will be shot correctly, and this leads to the feeling of being hot and ultimately to more made shots. Other times, something feels off and I can't get properly into my shooting motion and I can barely hit the backboard.

Put another way, I don't think there's such a thing as "a 45% shooter," as gus used in his random number experiment. I think there are shooters who shoot x% when their form is as good as it can be and y% when their form is off (for whatever reason) and z% when neither of those things are true. Possibly it's even more complicated than that.

Is it possible to measure the psychological components of shooting in any legitimate study? I doubt it. Which is why this debate will likely go on and on and on, until the game is no longer played.

MCFinARL
02-28-2014, 03:48 PM
Is it possible to measure the psychological components of shooting in any legitimate study? I doubt it. Which is why this debate will likely go on and on and on, until the game is no longer played.

And thank heaven for that, or we would all have to think of something else to do with our time. :D

gus
02-28-2014, 04:22 PM
Put another way, I don't think there's such a thing as "a 45% shooter," as gus used in his random number experiment. I think there are shooters who shoot x% when their form is as good as it can be and y% when their form is off (for whatever reason) and z% when neither of those things are true. Possibly it's even more complicated than that.

I only use a "45% 'shooter'" to illustrate how much randomness explains the trends people attribute to things like "hot hand", not because I think all people are perfectly consistent machines.

Kedsy
02-28-2014, 05:45 PM
I only use a "45% 'shooter'" to illustrate how much randomness explains the trends people attribute to things like "hot hand", not because I think all people are perfectly consistent machines.

I get that, and I think I understand your point of view (after I don't know how many debates). I even agree that randomness does explain it to a certain extent.

But my point is if, for example, someone generally shoots 45% (with random variation) when he feels comfortable and generally shoots 30% (with random variation) when he feels uncomfortable, then the idea of the hot hand does in fact have credence.