Wish we had a smart student or recent grad with some experience (and or calculator ;) )

[tjreese]

with the binomial curve. Might get there with an OC curve, but too long ago that it would require a lot of time and a person with recent experience might answer this quickly. Assume 3 pt shooting is not separate populations for each player and each game, but that the games are pretty homogenous...meaning random misses are due to chance and should level out if enough shots are taken. If we assume the average is .4 (our desire) then with some confidence level ...95% Purdue should make a 3 within so many shots if that average is due to random error. What is n or number of shots?

Problem is those people are probably studying while we are watching basketball

 

[Erial_Lion]

If you're assuming that you'll make a shot 40% of the time, the odds of making at least one of your first six is 95.33%...that's where you cross the 95% mark.

 

[Do Dah Day]

If you're assuming that you'll make a shot 40% of the time, the odds of making at least one of your first six is 95.33%...that's where you cross the 95% mark.

but that is shooting .16667 .... ugh

 

[Chi-Boiler]

It was my understanding that there would be no math.

 

[Do Dah Day]

so, it appears to my non-math mind ... that we need to make 7.5 3-pointers/game (don't ask). ... so, how many do we need to shoot to make 7.5 .... 21 at 35% makes.

 

[Purdue Grad in Texas]

It was my understanding that there would be no math.

 

[Chi-Boiler]

I couldn't figure out a way to attach that so I just typed it.

 

[Purdue Grad in Texas]

I couldn't figure out a way to attach that so I just typed it.

They don't call me Concorde for nothing......

"Message for you, Sir....."

 

[tjreese]

If you're assuming that you'll make a shot 40% of the time, the odds of making at least one of your first six is 95.33%...that's where you cross the 95% mark.

@Do Dah Day I really need to pull out some books, what is the sample size needed to shoot in which you make ONE shot if the true average was 40% with some confidence level such as .95. Yeah, usually you are given a sample size and assume "if a process average was x percent" what is your ability to detect it? Here is is how many tries and in all honesty it has been 37 years I used to do those things with the binomial and hypergeometric curve...and so I'm just not fluent. Before posting I did a quick search (and of course I don't believe a single population exists for all games, but...), but trying to work backwards a bit this is a nice video but I would have to go read a bit and I already have a mountain or papers (taxes) laying behind me and a starter making noise in a car that I can't figure out what is going on with it.

Anyway, enjoy.. I never really thought along the lines of finite. This answer would be interesting though to get a better feel for how likely is the team going to hit x% after so many shots. Somethng tells me a little excel sheet with various % assumed makes compared to exisiting shots taken would be interesting. AS I understand the quoted if there was a 40% chance of making the shot and six shots were taken the makes, could be, 1,2,3,4,5 or 6 (out of six shots) 95% of the time. I'm wondering what is the minumum shots that should be taken in order to hit the fist shot if the process average is 40%.

then again Erial and Do Dah are on to something easier perhaps, but if the number was making 1 in the first 6 was expected 95% of the time what is the best estimate if 2 out of 8 were made.in the first 10 and so forth. ..with 95% of the time...then how many shots must be taken for the current percentage to estimate with some confidence level the projected percentage if 18-22 shots were taken? At what point again assuming a homogeneous population (which I don't believe) does the data say continuing to shoot will result in x makes out of y shots 95% of the time that the data become of value?

 

[New Pal Boiler]

@Do Dah Day I really need to pull out some books, what is the sample size needed to shoot in which you make ONE shot if the true average was 40% with some confidence level such as .95. Yeah, usually you are given a sample size and assume "if a process average was x percent" what is your ability to detect it? Here is is how many tries and in all honesty it has been 37 years I used to do those things with the binomial and hypergeometric curve...and so I'm just not fluent. Before posting I did a quick search (and of course I don't believe a single population exists for all games, but...), but trying to work backwards a bit this is a nice video but I would have to go read a bit and I already have a mountain or papers (taxes) laying behind me and a starter making noise in a car that I can't figure out what is going on with it.

Anyway, enjoy.. I never really thought along the lines of finite. This answer would be interesting though to get a better feel for how likely is the team going to hit x% after so many shots. Somethng tells me a little excel sheet with various % assumed makes compared to exisiting shots taken would be interesting. AS I understand the quoted if there was a 40% chance of making the shot and six shots were taken the makes, could be, 1,2,3,4,5 or 6 (out of six shots) 95% of the time. I'm wondering what is the minumum shots that should be taken in order to hit the fist shot if the process average is 40%.

then again Erial and Do Dah are on to something easier perhaps, but if the number was making 1 in the first 6 was expected 95% of the time what is the best estimate if 2 out of 8 were made.in the first 10 and so forth. ..with 95% of the time...then how many shots must be taken for the current percentage to estimate with some confidence level the projected percentage if 18-22 shots were taken? At what point again assuming a homogeneous population (which I don't believe) does the data say continuing to shoot will result in x makes out of y shots 95% of the time that the data become of value?

View embedded media

Or, you can just look up some box scores from the 1990 Loyola Marymount crew…pretty sure Kimble and Co. flattened out that curve.

 

[tjreese]

Or, you can just look up some box scores from the 1990 Loyola Marymount crew…pretty sure Kimble and Co. flattened out that curve.

no misses or no makes would flatten out the curve, but more shots wouldn't

 

[tjreese]

@Do Dah Day I really need to pull out some books, what is the sample size needed to shoot in which you make ONE shot if the true average was 40% with some confidence level such as .95. Yeah, usually you are given a sample size and assume "if a process average was x percent" what is your ability to detect it? Here is is how many tries and in all honesty it has been 37 years I used to do those things with the binomial and hypergeometric curve...and so I'm just not fluent. Before posting I did a quick search (and of course I don't believe a single population exists for all games, but...), but trying to work backwards a bit this is a nice video but I would have to go read a bit and I already have a mountain or papers (taxes) laying behind me and a starter making noise in a car that I can't figure out what is going on with it.

Anyway, enjoy.. I never really thought along the lines of finite. This answer would be interesting though to get a better feel for how likely is the team going to hit x% after so many shots. Somethng tells me a little excel sheet with various % assumed makes compared to exisiting shots taken would be interesting. AS I understand the quoted if there was a 40% chance of making the shot and six shots were taken the makes, could be, 1,2,3,4,5 or 6 (out of six shots) 95% of the time. I'm wondering what is the minumum shots that should be taken in order to hit the fist shot if the process average is 40%.

then again Erial and Do Dah are on to something easier perhaps, but if the number was making 1 in the first 6 was expected 95% of the time what is the best estimate if 2 out of 8 were made.in the first 10 and so forth. ..with 95% of the time...then how many shots must be taken for the current percentage to estimate with some confidence level the projected percentage if 18-22 shots were taken? At what point again assuming a homogeneous population (which I don't believe) does the data say continuing to shoot will result in x makes out of y shots 95% of the time that the data become of value?

View embedded media

@Erial_Lion @Do Dah Day Thought a bit more on this and not sure why my original thoughts were not as straight forward as this other than getting the exact number of trials and then rounding up over the number "exactly" required for 95% was my thoughts...especially already knowing the populations were different. So thank you two for steering me down a simpler approach of practicality.

what if 1 -3pt shot is made on the fifth attempt and so the team misses until the 11th shot or any other number to make the second and so now we have a successive sampling plan up to 20+ shots since I don't think Purdue or most schools would stop shooting if 6 shots were taken and all missed.

This also brings me to what I suspect and that is Matt has a much more refined data base than Kenpom specifically tailored to data on Purdue and of interest to Matt- on many offensive sets, rebounding comparisons on 2 pt shots vs 3, maybe offensive eff when the first shot of a possession is inside the arc versus behind the arc and so forth. That precision and possible grouping of the data in similar approaches where Matt believes he can group games or parts of games together?

 

[Do Dah Day]

@Erial_Lion @Do Dah Day Thought a bit more on this and not sure why my original thoughts were not as straight forward as this other than getting the exact number of trials and then rounding up over the number "exactly" required for 95% was my thoughts...especially already knowing the populations were different. So thank you two for steering me down a simpler approach of practicality.

what if 1 -3pt shot is made on the fifth attempt and so the team misses until the 11th shot or any other number to make the second and so now we have a successive sampling plan up to 20+ shots since I don't think Purdue or most schools would stop shooting if 6 shots were taken and all missed.

This also brings me to what I suspect and that is Matt has a much more refined data base than Kenpom specifically tailored to data on Purdue and of interest to Matt- on many offensive sets, rebounding comparisons on 2 pt shots vs 3, maybe offensive eff when the first shot of a possession is inside the arc versus behind the arc and so forth. That precision and possible grouping of the data in similar approaches where Matt believes he can group games or parts of games together?

Didn't I see that he hired a data analyst over some other assistant coach? Seems like that was a year or two ago.

 

[tjreese]

Didn't I see that he hired a data analyst over some other assistant coach? Seems like that was a year or two ago.

Think you are right...can't recall the name, but a refinement of Kenpom makes a lot of sense. THAT guy (he was a guy and a "he") I'm sure has more specific data where he probably works with a basketball guy to assign more accurate groupings of data. All these questions I had remind me that if you don't use it you lose it or at least "some" of it...

 

[FiveWeight]

Expected outcome is a great way to look at it. My stat professor at Purdue used free throw shooting percentage as a good example of why coaches and the media get overly carried away with performance in a single game. his point was that people put way too much stock into whether somebody is having a “hot” or “cold” night, and rarely is that actually a good predictor of whether their next shot will fall. Just not enough data points delivered in a single game to give anything meaningful most of the time. Short streaks like you see players going on can easily be found through the variation of a random number sequence.

The other consideration is risk tolerance. A three pointer is more risky because you are risking more on whether it goes in. Early in the game that shouldn’t matter much, but consider the following: somebody offers to sell you a lottery ticket, if you win you get all the money in the world. If you lose, you are bankrupt. Only 10 tickets are issued, so your expected value is 10% of all the money in the world. Just looking at the numbers, it should be a no-brainer that everybody should take that risk. But expected value doesn’t take into consideration the negative consequence of losing everything you have. to properly account for that, you have to assign a proxy value to going bankrupt of “negative infinity” which gets more into psychology than statistics.

 

[tjreese]

Expected outcome is a great way to look at it. My stat professor at Purdue used free throw shooting percentage as a good example of why coaches and the media get overly carried away with performance in a single game. his point was that people put way too much stock into whether somebody is having a “hot” or “cold” night, and rarely is that actually a good predictor of whether their next shot will fall. Just not enough data points delivered in a single game to give anything meaningful most of the time. Short streaks like you see players going on can easily be found through the variation of a random number sequence.

The other consideration is risk tolerance. A three pointer is more risky because you are risking more on whether it goes in. Early in the game that shouldn’t matter much, but consider the following: somebody offers to sell you a lottery ticket, if you win you get all the money in the world. If you lose, you are bankrupt. Only 10 tickets are issued, so your expected value is 10% of all the money in the world. Just looking at the numbers, it should be a no-brainer that everybody should take that risk. But expected value doesn’t take into consideration the negative consequence of losing everything you have. to properly account for that, you have to assign a proxy value to going bankrupt of “negative infinity” which gets more into psychology than statistics.

Quite simply if accuracy is crucial then understanding the sources of variation is crucial in determing what factors are in the model and what factors go to the error term that cannot currently be assigned or understood. Do you group all that shot 3s in one data set and only use that? Do you group differently each shooter in each independent data set and use that? Do you group differently each shooter at home and away and use two data sets for each shooter? Can you lump home and away if no significant difference exists in shooting the 3 ball for each player...and can you lump some players together? Is there a line of demarcation where you list each shooter by himself for home and also away groupings? Do you do the same, but break it down further into segments of playing time with the four other players at the time and their respective percent? Now we are getting into that "forward stepwise" I mentioned some place.

Anyway, depending on accuracy desired, more precision of those variables in play need accounted and as you can see the degrees of freedom or sample size needed for a desired accuracy would never happen in a single game. Course if there are no significant differences in the few considerations or others potential sources for variation I never listed, then grouping the data into a single data set wouldn't be as problematic and much more accurate for a single game...Then if you get so refined with so many categories where you only have 1 measure for all teh combinations your model becomes perfect, but totally unuseable due to no error term and so you need repeated data in each source of variation desired to be studied and the associated interactions as well...

and you hit on potentially losing out on gathering fouls of hitting a 2 at a higher rate if a team is capable, but at some point you also need spacing and so you can see although I like data and numbers...I'm not as big on it in basketball, but also know it is the best starting point for a game plan...and of course points made early in a game in use for the total and so each possession early and late are both important