ambchang
🧑🔧🔛🕕
- Joined
- Sep 5, 2025
- Messages
- 264
- Reaction score
- 226
So after watching Luka shoot 7 free throws in the 1st quarter yesterday, I thought to myself, "Is it because he's Luka, or is it because he's a Laker?", and so I spent this noon hacking out a regression model of a player's FTr, taking into account their shot profile (0-3/3-10/10-16/16-3pt/3pt) and their team.
I did a similar analysis back a few years during that Kobe-Nash-Dwight season where they were close to misisng the playoffs, only to come storming back. I noticed something and did an analysis, then confirmed that not only did the Lakers FTA increased dramatically during that stretch where they went from like 10th seed to 7th seed (don't remember exactly the seeding), but the team directly above them in the standings see their FTA DECREASE dramatically, until they are surpassed by the Lakers, then their FTA went back to normal. it was one of the most hilarious things I have seen, but we already knew what was going on.
Anways, some background:
Luka last season was traded from the Mavs to the Lakers. His FTr with the Mavs was .313, but increased to .445 once he became a Laker. So happened to Unibrow? Well, the opposite (or is that the same?) happened, his FTr was .423 with the Lakers, then dropped to .316 with the Mavs (only 9 games though). Speaking of Unibrow, his prime and most dominant was with the Pels, he scored close to the paint much more with the Pels/Hornets than he did with the Lakers, and yet the two seasons he had the highest FTr was with the Lakers. Go figure.
Rui Hachimura? .179 with the Wiz, then .235 with the Lakers. (the season he was traded). Ayton FTr has been cratering since he was with Portland (went from a regular low to mid .2s to low .1s, but mostly because he is playing like wuss away from the basket. This year so far, his FGA% is LOWER than he had last year with the Blazers, and yet his FTr went up from .128 to .172. So I am like ... does playing for the Lakers really give you more FTs?
So this is what I did. I looked up all player-team combinations from last year, and did a regression model of a player's FTr as a dependent variable, then the following as the independent variables: FGA% from 0-3/3-10/10-16/16-3P/3-P, and finally the team. (1 for belong to a certain team, 0 for not), so say Wemby will be 1 for Spurs but 0 for all other teams. This isn't the best as there are just too many boolean variables, but alas. The other problem is that my analysis pack only accepts up to 16 independent variables at a time, so with the 5 mandatory fields filled up by the shot profile, I have to divide up the model into 3 groups (ATL to GSW, HOU to NYK, and finally OKC to WAS), and there are issues with this as the coefficients to each batch is different even for the shot profile. I ran a fourth regression model without any team details as a reference.
Comparing the numbers, the intercepts are fairly reliable for all four batches, at around 0.00045, which basically means that you get 0 FTs with 0 shots, which is quite on the ball, batch 3 has 0.048, which is a bit high for my taste but not something I will lose sleep over. All four models have the shot profile having high absolute t-Stat numbers and low P-value numbers for 0-3 (0.00005 to 0.0003) and 3-10 (0.01 to 0.04), which basically shows the shot profile from close range has a large impact on the FTr, which isn't a surprise at all. Note that I am using the FTr and shot profile (% of attempts) and not the raw numbers, which means that an end of the bench guy who played 4 minutes all season will count as much as an all-star or even a superstar who get superstar calls. I did that on purpose as I wanted to look at impact of team, not players, so this will remove the bias of superstars skewing the numbers for a team.
I cannot, for the life of me, figure out why all shot profile, other than 16-3P, have +ve coefficients with the FTr, I would have thought the further you go, the lower the number gets and will be negative for anything 10+, but for 16-3P, only the 4th batch has a -ve number, but at the same time, the P-value is quite high for 3P (0.08 to 0.18) and 16-3P (0.20 to 0.31), which I am not surprised by as those should have lesser impact than say 0-3, which consistently showed numbers approaching 0. In other words, shots from 16+ feet doesn't have any significant impact on the FTr. So I am relatively comfortable with the model.
One final thing, 10-16 has by far the lowest P-value, whcih means it has the most signficant impact on the FTr AND it has the highest Coefficient, which means that it is closely tied. The Coefficient I can understand as these are just percentages. Most players have low FTr and low 10-16 shots, so they have a high Co-efficient. It doesn't mean that they are closely tied by any means. What I cannot get is why they have such a low P-value. (9.2 x 10^-6 to 5.2x10^-5), which are far lower than the 0-3 numbers. But overall, I am comfortable with the model itself.
Now to the teams.
I did a similar analysis back a few years during that Kobe-Nash-Dwight season where they were close to misisng the playoffs, only to come storming back. I noticed something and did an analysis, then confirmed that not only did the Lakers FTA increased dramatically during that stretch where they went from like 10th seed to 7th seed (don't remember exactly the seeding), but the team directly above them in the standings see their FTA DECREASE dramatically, until they are surpassed by the Lakers, then their FTA went back to normal. it was one of the most hilarious things I have seen, but we already knew what was going on.
Anways, some background:
Luka last season was traded from the Mavs to the Lakers. His FTr with the Mavs was .313, but increased to .445 once he became a Laker. So happened to Unibrow? Well, the opposite (or is that the same?) happened, his FTr was .423 with the Lakers, then dropped to .316 with the Mavs (only 9 games though). Speaking of Unibrow, his prime and most dominant was with the Pels, he scored close to the paint much more with the Pels/Hornets than he did with the Lakers, and yet the two seasons he had the highest FTr was with the Lakers. Go figure.
Rui Hachimura? .179 with the Wiz, then .235 with the Lakers. (the season he was traded). Ayton FTr has been cratering since he was with Portland (went from a regular low to mid .2s to low .1s, but mostly because he is playing like wuss away from the basket. This year so far, his FGA% is LOWER than he had last year with the Blazers, and yet his FTr went up from .128 to .172. So I am like ... does playing for the Lakers really give you more FTs?
So this is what I did. I looked up all player-team combinations from last year, and did a regression model of a player's FTr as a dependent variable, then the following as the independent variables: FGA% from 0-3/3-10/10-16/16-3P/3-P, and finally the team. (1 for belong to a certain team, 0 for not), so say Wemby will be 1 for Spurs but 0 for all other teams. This isn't the best as there are just too many boolean variables, but alas. The other problem is that my analysis pack only accepts up to 16 independent variables at a time, so with the 5 mandatory fields filled up by the shot profile, I have to divide up the model into 3 groups (ATL to GSW, HOU to NYK, and finally OKC to WAS), and there are issues with this as the coefficients to each batch is different even for the shot profile. I ran a fourth regression model without any team details as a reference.
Comparing the numbers, the intercepts are fairly reliable for all four batches, at around 0.00045, which basically means that you get 0 FTs with 0 shots, which is quite on the ball, batch 3 has 0.048, which is a bit high for my taste but not something I will lose sleep over. All four models have the shot profile having high absolute t-Stat numbers and low P-value numbers for 0-3 (0.00005 to 0.0003) and 3-10 (0.01 to 0.04), which basically shows the shot profile from close range has a large impact on the FTr, which isn't a surprise at all. Note that I am using the FTr and shot profile (% of attempts) and not the raw numbers, which means that an end of the bench guy who played 4 minutes all season will count as much as an all-star or even a superstar who get superstar calls. I did that on purpose as I wanted to look at impact of team, not players, so this will remove the bias of superstars skewing the numbers for a team.
I cannot, for the life of me, figure out why all shot profile, other than 16-3P, have +ve coefficients with the FTr, I would have thought the further you go, the lower the number gets and will be negative for anything 10+, but for 16-3P, only the 4th batch has a -ve number, but at the same time, the P-value is quite high for 3P (0.08 to 0.18) and 16-3P (0.20 to 0.31), which I am not surprised by as those should have lesser impact than say 0-3, which consistently showed numbers approaching 0. In other words, shots from 16+ feet doesn't have any significant impact on the FTr. So I am relatively comfortable with the model.
One final thing, 10-16 has by far the lowest P-value, whcih means it has the most signficant impact on the FTr AND it has the highest Coefficient, which means that it is closely tied. The Coefficient I can understand as these are just percentages. Most players have low FTr and low 10-16 shots, so they have a high Co-efficient. It doesn't mean that they are closely tied by any means. What I cannot get is why they have such a low P-value. (9.2 x 10^-6 to 5.2x10^-5), which are far lower than the 0-3 numbers. But overall, I am comfortable with the model itself.
Now to the teams.
Last edited:
