Homegrown analyses of KO's play (2014-2015)

[tarheelsxxiii]

In a moment of extreme dorkiness, I decided to sample some basic stats of KO's play from this season and last (89 games, excluding games missed due to injury). The purpose was to better understand his inconsistency.

Variables
Basic per game inclusion of minutes, points, turnovers, fouls, and field goal %age (2% and 3%)

Analyses
First, a regression to test relation between minutes and points. This was intended to be a crude way to determine if his minutes predicted his offensive production (which seems to be most in flux).

Second (if the presumed relation is there), tests of moderation to identify variables influencing this relation. For example, we've all observed that he's foul prone... he does make some boneheaded turnovers.... his shooting seems all over the place -- are any of these significantly influencing the relation between minutes and points?

Results
Unsurprisingly, minutes and points were significantly related. Effect size was fairly large -- his playing time accounts for about 50% of the variance in his scoring.

Variables that shed further light on the relation? Turnovers did not, so he doesn't appear to be [dang]ing himself with reckless abandon on offense. Fouls? Nope... the relation was somewhat trending to significance (p=.18), but didn't even account for .01% of the minutes-points relation. So his hacking doesn't seem to be limiting his production on the court (when he is on the court)--he seems no less aggressive when in foul trouble.

FG %age? Yes... but not his 3P %age (not even close). It is inconsistency in 2P %age that seems to be the major influence on his minutes--points relation.  Makes sense, and accounts for about 5% of the variability in his on-court production (fairly considerable given sample size and crudeness of analyses).

Interpretation
Variability in his shooting is affecting his on-court production, offensively, over and above minutes played per game. Fouls, turnovers, and 3P %age, however, are not.

This is a crude enough look that I almost considered not sharing. I am, though, in case others may find this interesting and, more, have suggestions for how to taker a closer look at his inconsistency. There's an obvious chicken-egg issue here: Are his minutes lower when he's not hitting shots? Or is he not hitting shots because his minutes are low/inconsistent. No real way to test directionality here with cross-sectional data though. Other thoughts for improvement may be using RPM (offense and defense) rather than Points to better assess production, but in reality, it seems that fluctuations in his scoring are of greatest interest. D-League reports an Efficiency statistic that could be useful, but it's cumulative and to my knowledge not reported for individual games... nor have I found this stat for NBA games.

[footey]

TP, although I only understood half of it. 

[tarheelsxxiii]

TP, although I only understood half of it.

Thanks, back at ya. The confusion is my fault, I'll take some time to think about how to explain it more clearly/make it more accessible. Happy to answer any questions you may have in the meantime.

[hwangjini_1]

tp, and i understood all of it. 

key variables to also consider would be who is on the court with him. does his scoring drop with outside shooters on the floor? which celtics at center/low post affect his scoring, if any?

[danglertx]

Awesome that you are interested enough in the Celtics and KO to look into it. 

From my non-analysis of statistics and just observational, KO becomes tentative if he misses his first couple shots.  If he isn't going to take his shots, he doesn't need to be out there because he doesn't do anything else so well he deserves to be out there if he isn't shooting.

So if he misses a few shots, then passes a few open looks up, I think CBS pulls him.  Simple as that.

On a side note, I really think he would benefit from starting.  I think he has a better chance of getting an easy basket and into the flow early than coming in cold off the bench.  A guy like Amir probably doesn't matter because his looks are coming in the paint and he does other stuff anyway.

[tarheelsxxiii]

Regression is similar to correlation, just more powerful. For example, you can control for covariates (stuff that could have an undue influence on results; e.g., sex, gender, other hidden variables could explain results... so you want to account for that, making sure you "control" as much as you can to draw the conclusions you want.

Second, regression allows you to directly test the influence of moderators (third variables)...same as covariates, really, but instead of saying I want to test the relation between A and B (while controlling for undue influence of C), the C is a variable of considerable interest (e.g., does C significantly influence the result of A and B?)

I chose minutes and points because they are the 2 variables that seem most inconsistent for KO. The basic question was "How much variance in his scoring can be explain by variance in his minutes?" Basically, if MPG predicted 100% of his point production, then KO's ups and downs with re: to scoring could be entirely explained by ups and downs in minutes.

Results indicated a moderate relationship... just 50% of variance in scoring is attributed to minutes. That's a lot, sure... but only half. So then the question is "What else is accounting for his ups and downs in scoring?" Hence the moderation analyses... which are intended to determine variables that significantly influence his ups and downs in scoring, importantly, over and above playing time (you're controlling for playing time, here... we wanna see if anything contributes beyond it in order to shed further light on the story).

So TOs, fouls, 3P % didn't do it... but 2P % did. It accounted for 5% more of his ups and downs in point production, beyond just minutes played. So that's interesting because it suggests his offensive production (limited to just points here, unfortunately) is related to ups and downs in shooting, even after taking into account the variability in his playing time. While it's not really possible to account for all the variability in his scoring (or any component of his game, really), there still remains 45% of the variance in his scoring left unaccounted for.

Just above Hwang is making some suggestions that may help further explain the relation between mins and point production for KO. That's really good stuff. Any behavioral observations (stuff you see on the court) you guys may have could potentially be testable (albeit difficult, but I don't care about that)... and shed some more light on his offensive ups and downs (again, beyond just PT).

I hope that is at least somewhat helpful. Again, feel free to ask any questions.

[Denis998]

Something i've been thinking about in terms of statistics is to determine if a players production is significantly better than the play of the previous season. Something like an independent samples T I think should do the job, but would be a pain to do all the calculations by hand lol. Also going off of your points x  minutes correlation, maybe try to run the test with points per 36 x minutes, I think that might account better for his offensive production during the time he is on the floor.

[tarheelsxxiii]

tp, and i understood all of it. 

key variables to also consider would be who is on the court with him. does his scoring drop with outside shooters on the floor? which celtics at center/low post affect his scoring, if any?

Awesome, thanks Back at ya. I believe you or significant other is a psychologist, correct? I'm glad my post wasn't entirely lost on everyone (due entirely to my struggle to distill the info)... I was hoping to spur some discussion.

Good insights here, too. Immediate thought is that I could run analyses separately, based on who he is playing with, and then compare results across line-ups/big main pairings. Any idea where to get that data? May be tricky to incorporate into any model, but I'm down to try if it's available...

KO needs to be figured out!

[tarheelsxxiii]

Awesome that you are interested enough in the Celtics and KO to look into it. 

From my non-analysis of statistics and just observational, KO becomes tentative if he misses his first couple shots.  If he isn't going to take his shots, he doesn't need to be out there because he doesn't do anything else so well he deserves to be out there if he isn't shooting.

So if he misses a few shots, then passes a few open looks up, I think CBS pulls him.  Simple as that.

On a side note, I really think he would benefit from starting.  I think he has a better chance of getting an easy basket and into the flow early than coming in cold off the bench.  A guy like Amir probably doesn't matter because his looks are coming in the paint and he does other stuff anyway.

Thanks. Observations are really helpful... there's often ways to incorporate it into an analysis. This one being so crude, any suggestions are welcome.

And I agree -- he seems to have a pretty strong reaction to his first few shots, then is more likely to get pulled. The first major question I had was how can that somehow be incorporated into the model. If per-quarter stats (minutes, FGA, FGM, etc.) is accessible anywhere, that would be doable and cool to look into. I'm not sure that they are, though.

He certainly did well with the last start. Hope we get to see him start again tonight. After the game, he commented that he was able to get some easy buckets early and that helped him get going.

[tarheelsxxiii]

Something i've been thinking about in terms of statistics is to determine if a players production is significantly better than the play of the previous season. Something like an independent samples T I think should do the job, but would be a pain to do all the calculations by hand lol. Also going off of your points x  minutes correlation, maybe try to run the test with points per 36 x minutes, I think that might account better for his offensive production during the time he is on the floor.

Thanks for your suggestions. T-tests wouldn't be that hard, it's just manually entering all the data in a database that sucks. But interesting question... he has definitely shown changes in his defensive game over the years... wouldn't be far-fetched to think trends have changed for him offensively, too.

[Denis998]

Something i've been thinking about in terms of statistics is to determine if a players production is significantly better than the play of the previous season. Something like an independent samples T I think should do the job, but would be a pain to do all the calculations by hand lol. Also going off of your points x  minutes correlation, maybe try to run the test with points per 36 x minutes, I think that might account better for his offensive production during the time he is on the floor.

Thanks for your suggestions. T-tests wouldn't be that hard, it's just manually entering all the data in a database that sucks. But interesting question... he has definitely shown changes in his defensive game over the years... wouldn't be far-fetched to think trends have changed for him offensively, too.

It would be an interesting project doing a multiple regression to account for projected wins pulling variables from this page http://espn.go.com/nba/hollinger/teamstats
Its quite interesting to apply what I learned in my psychology courses to basketball stats

[Neurotic Guy]

I don't fully understand but really appreciate the effort! TP.

I am wondering about the relationship between points scored (or shots taken) in his first few minutes on the court to subsequent minutes played, subsequent shots, and subsequent points scored.   Wondering about the case for aggressive-Kelly v. non-assertive Kelly

[hwangjini_1]

tp, and i understood all of it. 

key variables to also consider would be who is on the court with him. does his scoring drop with outside shooters on the floor? which celtics at center/low post affect his scoring, if any?

Awesome, thanks Back at ya. I believe you or significant other is a psychologist, correct? I'm glad my post wasn't entirely lost on everyone (due entirely to my struggle to distill the info)... I was hoping to spur some discussion.

Good insights here, too. Immediate thought is that I could run analyses separately, based on who he is playing with, and then compare results across line-ups/big main pairings. Any idea where to get that data? May be tricky to incorporate into any model, but I'm down to try if it's available...

KO needs to be figured out!

yes..
he...
does!

i am sure the celtics front office runs mountains of stats such as this as an everyday event. i know CBS uses such information to assess the effectiveness of certain combinations of players.

again, good job! it is fun to read the analyses. thanks.

[The Oracle]

  The single biggest inhibitor to K.O.'s game is putting him on the floor with a lane clogging big man.  He can beat his man into the lane but when he is on the floor with Lee/Zeller he is then met with resistance from the man defending them.  This does not occur near so much when on the floor with Jerebko for example as he will have that other big man pulled far enough away from the lane to not be able to affect attacking the rim.  It should show in some advanced analysis of where he gets his shots and his shooting percentages.  But it clearly shows in his plus/minus data already.  His +/- has been incredible next to Sully, Amir, Jerebko.  His +/- has been very average when paired with Bass, Humphries.  And his +/- has been terrible next to Zeller, Faverani, and Lee.  The more the lane is clogged with him on the floor the less successful the team has been and most likely when dug into deep enough would show in his personal numbers also.