Shooting Streaks: Do they exist?

[Siro]

I was bored last night, and I decided to do some original research on whether players get hot and go on shooting streaks. I put some work into this, so I'm not going to post it in General Sports for nobody to see.

Methodology:

I collected data on the last 500 shots for 20 players picked at random (total sample size of 10000 shots). I recorded a made shot as "1" and a miss as a zero. I then wrote the following Matlab algorithm and ran it for each data set, then averaged them out:

function answer = streak(shotlist)

a= 0;
b=0;
c=0;
d=0;

for i = 1:length(shotlist)
if(shotlist(i)==1)
shotlist1(i)=shotlist(i+1);
else
a = a+1;
end
if(shotlist(i)==1 &&shotlist(i+1)==1)
shotlist2(i)=shotlist(i+2);
else
b = b+1;
end
if(shotlist(i)==1 && shotlist(i+1) ==1 && shotlist(i+2)==1)
shotlist3(i)=shotlist(i+3);
else
c = c+1;
end
if(shotlist(i)==1 && shotlist(i+1) == 1 && shotlist(i+2) ==1 && shotlist(i+3)==1)
shotlist4(i)=shotlist(i+4);
else
d = d+1;
end
end

mean(shotlist)

m1 = sum(shotlist1)/(length(shotlist)-a)
m2 = sum(shotlist2)/(length(shotlist)-b)
m3 = sum(shotlist3)/(length(shotlist)-c)
m4 = sum(shotlist4)/(length(shotlist)-d)


end

I'm giving you the code because I'm not much of a coder, and I may have made a mistake in the algorithm. But my intention is to calculate the following:

Percentage on next shot after hitting 1, 2, 3, 4, or 5 shots in a row. Base shooting percentage for all 10k shots was 46.9%

The graph below summarizes the results:

It is important to note that the last set (5 shots in a row) shows a significantly lower percent probably due to the small sample size (it happened 38 times).

Conclusion:

There is no statistically significant difference between a player's shooting average on each shot regardless of how many shots in a row had been made. Shooting streaks do not exist outside of the dictates of probability .

 

[Jamezz]

I was bored last night, and I decided to do some original research on whether players get hot and go on shooting streaks. I put some work into this, so I'm not going to post it in General Sports for nobody to see.

Methodology:

I collected data on the last 500 shots for 20 players picked at random (total sample size of 10000 shots). I recorded a made shot as "1" and a miss as a zero. I then wrote the following Matlab algorithm and ran it for each data set, then averaged them out:



I'm giving you the code because I'm not much of a coder, and I may have made a mistake in the algorithm. But my intention is to calculate the following:

Percentage on next shot after hitting 1, 2, 3, 4, or 5 shots in a row. Base shooting percentage for all 10k shots was 46.9%

The graph below summarizes the results:

It is important to note that the last set (5 shots in a row) shows a significantly lower percent probably due to the small sample size (it happened 38 times).

Conclusion:

There is no statistically significant difference between a player's shooting average on each shot regardless of how many shots in a row had been made. Shooting streaks do not exist.

I guess I'm missing something then. You said you picked 20 players at random. How do players known to have shooting streaks compare to players who are not known for this? Elaborate please.

 

[Siro]

I guess I'm missing something then. You said you picked 20 players at random. How do players known to have shooting streaks compare to players who are not known for this? Elaborate please.

I don't. This is about the idea that players can get "hot". We can conclusively say that isn't happening. The shooting percentage of your average NBA player does not improve the more shots they make. Since applying the code to specific players is easy, I'd be willing to take requests to see if a certain player bucks the trend.

 

[fishonjazz]

. The shooting percentage of your average NBA player does not improve the more shots they make.

Did you mean take rather than make in your sentence?

 

[sahlensguy]

The Hot Hand fallacy.

https://en.m.wikipedia.org/wiki/Hot-hand_fallacy

Data does not support the hot hand hypotheses exept at the free throw line.

 

[Siro]

Did you mean take rather than make in your sentence?

No, make. The idea is that players can get hot and should be fed since they're making their shots. But each shot has the same percentage of going on, regardless of how "hot" a player happened to be.

 

[olden_undercover]

I guess I'm missing something then. You said you picked 20 players at random. How do players known to have shooting streaks compare to players who are not known for this? Elaborate please.

Great topic, Siro... A lot of this depends on your definition of "streak shooter", I think. In your mini-study, you're defining it very strictly-- as someone who shoots 100% of his shots up to and beyond 5 shots in a row. I can see that being the definition of a streak, and I don't disagree with it, but what happens if the player misses his sixth shot, then hits three more in a row, and misses one more? Is the streak defined by consecutive makes, or is a streak shooter someone who gets hot in a game and shoots a crazy percentage (80%, in my example)? They aren't mutually exclusive (streak vs. streak shooter), but quantifying their relationship is tough, maybe.

Jamezz raises a good point. What players did you use?

 

[Siro]

The Hot Hand fallacy.

https://en.m.wikipedia.org/wiki/Hot-hand_fallacy

Data does not support the hot hand hypotheses exept at the free throw line.

Well then, consider this more confirmation of the invalidity of the hypothesis!

 

[fishonjazz]

No, make. The idea is that players can get hot and should be fed since they're making their shots. But each shot has the same percentage of going on, regardless of how "hot" a player happened to be.

Oh I see. The sentence just didn't seem to make sense.
When players make shots thier shooting percentage does not increase. Confusing as a stand alone sentence. In the context of the thread it makes a little sense

 

[Jazz Spazz]

Was Trey Burke part of this Methlab calculation?

 

[Siro]

Great topic, Siro... A lot of this depends on your definition of "streak shooter", I think. In your mini-study, you're defining it very strictly-- as someone who shoots 100% of his shots up to and beyond 5 shots in a row. I can see that being the definition of a streak, and I don't disagree with it, but what happens if the player misses his sixth shot, then hits three more in a row, and misses one more? Is the streak defined by consecutive makes, or is a streak shooter someone who gets hot in a game and shoots a crazy percentage (80%, in my example)? They aren't mutually exclusive (streak vs. streak shooter), but quantifying their relationship is tough, maybe.

Jamezz raises a good point. What players did you use?

How would you measure that? Statistically, a player can have games where he shoots a lot higher or lower than his average. If a player happened to average 80% one game, then it means he must have made a lot of consecutive shots. This isn't about whether players can have good shooting games. Of course they can. Sometimes they will hit 5 shots in a row, miss 2, then hit 4 more shots. But other times they'll hit 5 shots in a row, then miss the next 6. Since the statistics show that their next shot percentage is the same, it means all scenarios average out. It is no more likely for a player to hit the next any number of shots regardless of how many shots they've previously hit or missed. It will always go back to the player's average.

I picked the players at random. I only remember Westbrook and Curry.

 

[gregbroncs]

I was bored last night, and I decided to do some original research on whether players get hot and go on shooting streaks. I put some work into this, so I'm not going to post it in General Sports for nobody to see.

Methodology:

I collected data on the last 500 shots for 20 players picked at random (total sample size of 10000 shots). I recorded a made shot as "1" and a miss as a zero. I then wrote the following Matlab algorithm and ran it for each data set, then averaged them out:



I'm giving you the code because I'm not much of a coder, and I may have made a mistake in the algorithm. But my intention is to calculate the following:

Percentage on next shot after hitting 1, 2, 3, 4, or 5 shots in a row. Base shooting percentage for all 10k shots was 46.9%

The graph below summarizes the results:

It is important to note that the last set (5 shots in a row) shows a significantly lower percent probably due to the small sample size (it happened 38 times).

Conclusion:

There is no statistically significant difference between a player's shooting average on each shot regardless of how many shots in a row had been made. Shooting streaks do not exist outside of the dictates of probability .

Interesting take. I'd be just as interested if shooting slumps exist. Mostly because I believe missing a shot hurts the mindset going forward more than making a shot helps. I'm terrible at this kind of stuff though, so I won't even be making what would be a failed attempt.

 

[silesian]

nice work

cool analysis, although I may quibble that your conclusion is too broad given the analytical approach.

A better conclusion could be "the average NBA player is not a streak shooter" since it says nothing about any individual. You could hypothetically have a group of "streak shooters" (those having an positive slope on your graph) that is offset by others who are less likely to make the next shot after making one (negative slope, perhaps a "fatigue shooter" who is makes lower % as a game progresses?).


You could also get around the issue of only accounting for "pure streaks" -- a series of makes -- by correlating the % made on the nth shot to the percent made of the previous m shots. For example, correlate % made on 6th shot to % made on shots 1-5. May or may not add insights but might be interesting.

Again, thanks for crunching the numbers, your analysis is very interesting.

 

[zbone]

Did you filter the set for jump shots only?

 

[Siro]

Okay, so I decided to run the numbers for Jamaal Crawford, since he is known as one of the streakiest shooters. Sample size 500 shots, which is pretty small given we're trying to measure streaks of shots. But here are the results:

His overall FG% is 39.5%. He shoots 4% better after hitting 2 shots in a row, and a little better with 3 shots in a row. Sample is too small for 4 shots in a row to give meaningful results. I think it is likely that the trend continues with bigger sample, but we're still talking about an overall increase of only a couple of percentage points. Interesting nonetheless!

 

[Siro]

Did you filter the set for jump shots only?

I didn't. I thought about it, but I think you have to consider floaters, hook shots, and such, in streaks. Since big men shoot a higher percentage anyway, I decided to keep the sample to PGs, SGs, and SFs.

 

[Siro]

cool analysis, although I may quibble that your conclusion is too broad given the analytical approach.

A better conclusion could be "the average NBA player is not a streak shooter" since it says nothing about any individual. You could hypothetically have a group of "streak shooters" (those having an positive slope on your graph) that is offset by others who are less likely to make the next shot after making one (negative slope, perhaps a "fatigue shooter" who is makes lower % as a game progresses?).


You could also get around the issue of only accounting for "pure streaks" -- a series of makes -- by correlating the % made on the nth shot to the percent made of the previous m shots. For example, correlate % made on 6th shot to % made on shots 1-5. May or may not add insights but might be interesting.

Again, thanks for crunching the numbers, your analysis is very interesting.

Excellent points. This would require a lot more coding, and I'm extremely lazy. However, I will try to implement some of that sometime in the future.

 

[Siro]

Interesting take. I'd be just as interested if shooting slumps exist. Mostly because I believe missing a shot hurts the mindset going forward more than making a shot helps. I'm terrible at this kind of stuff though, so I won't even be making what would be a failed attempt.

This would require a different approach, but I don't think it would be difficult to come up with a meaningful way to measure slumps. I will probably give it a go when I'm free.

 

[zbone]

I didn't. I thought about it, but I think you have to consider floaters, hook shots, and such, in streaks. Since big men shoot a higher percentage anyway, I decided to keep the sample to PGs, SGs, and SFs.

I would bet layups will add avoidable noise to your data though.

 

[kabenac]

Was Trey Burke part of this Methlab calculation?

That's exactly what I was going to say - J.R. Smith comes to mind also. Some guys shoot either hot or cold and the swing is wide.