Everything Gary Trent Jr.

[TrueTorontoFan]

We play a different kind of defense.
If those metrics are anchored on the way that the rest of NBA plays defense, how can we effectively measure our players then?

that isn't the problem though. and if anything that would make trent look better.


you could argue oh that is why his deflections go up so much because he benefits from others disruption.. but then you look and fred was top 10 in deflections as well throughout a huge chunk of the season

[golden]

where do you find EPM?

Google is your friend.

https://dunksandthrees.com/epm

[The Great One]

Trent in college. Exactly the same player. Numbers are the same too.

[The Claw Reborn]

Trent Jr.

He is the 5th option who is a decent defender, reliable end of quarter clutch player and has range. Nick Nurse dont have to hide him on defense for his position unlike what this franchise has done for almost a decade with Derozan.

He is 23 years old with a bright future ahead as a Raptor

[chris]

Trent in college. Exactly the same player. Numbers are the same too.

i'd wager the vast majority of NBA players wish they could maintain their college production in the NBA.

[Kagemusha]

Trent Jr.

He is the 5th option who is a decent defender, reliable end of quarter clutch player and has range. Nick Nurse dont have to hide him on defense for his position unlike what this franchise has done for almost a decade with Derozan.

He is 23 years old with a bright future ahead as a Raptor

Hey, how dare you praise that player?

[Kagemusha]

that isn't the problem though. and if anything that would make trent look better.


you could argue oh that is why his deflections go up so much because he benefits from others disruption.. but then you look and fred was top 10 in deflections as well throughout a huge chunk of the season

Well then, isn't that a good thing?
That they work GREAT with our system?
You guys are picking Gary based on how the OTHER teams play D.
How stupid is that?
Nurse trusts Gary, his teammates trust Gary.
Except you guys.

[DanH]

There is not much validity there Dan but carry on what you believe is noise.

Belief has nothing to do with it, it's math.

[DanH]

As far as I can tell GTJ graded out last season as a positive on defense in almost every defensive advanced stat, except 538 RAPTOR (which is a dubious stat, tbh).

Yeah Gary grades out as roughly average in impact. His energy and effort in-season helped a lot with that. Unfortunately he's still a target once the playoffs hit, and with a defensive scheme like ours which wants to switch freely, a very easy target to exploit.

Not an issue if he's a supersub sixth man, but if people envision him as a long term starter, he needs to make more significant defensive strides so that he's not just disruptive but also solid. Which is quite possible, maybe he does exactly that.

[DanH]

I also am little curious Dan….you seem to overextend and get out more often with others, lately with TTF but never seen you dispute T-Rex with his flawed arguments about Trent Jr.
…or you are just personally encouraged when he likes your posts and tell you to preach your so called noise?

Maybe I’m wrong and you just have the same thought process, nothing wrong with that I guess

I don't care who likes my posts. I tend to have long discussions with people who a) use evidence, and b) have opinions worth debating. Evidence is the interesting bit - because stats used incorrectly can obscure the truth more than expose it, and those sorts of discussions (adding context to numbers, value of statistics, etc) are the ones that tend to stretch long as both sides present their opinions, research and reasoning.

I can assure you there is a mountain of topics I disagree with TRex on.

[The Claw Reborn]

Yeah Gary grades out as roughly average in impact. His energy and effort in-season helped a lot with that. Unfortunately he's still a target once the playoffs hit, and with a defensive scheme like ours which wants to switch freely, a very easy target to exploit.

Not an issue if he's a supersub sixth man, but if people envision him as a long term starter, he needs to make more significant defensive strides so that he's not just disruptive but also solid. Which is quite possible, maybe he does exactly that.

23 years of age, ceiling is still unknown, but fundamentals are good so hopefully he gets there.

[TrueTorontoFan]

Well then, isn't that a good thing?
That they work GREAT with our system?
You guys are picking Gary based on how the OTHER teams play D.
How stupid is that?
Nurse trusts Gary, his teammates trust Gary.
Except you guys.

I think you misunderstood what I said... and you aren't fully .... understanding my last point...


the point you made isn't the problem


also it isn't about trust ... gary can be the worst defender and yet we have an overall good defensive team. again our best defensive line up was with him out of it.

[DanH]

TLDR - without knowing all of the iterations of you looking into the stat... I am assuming you look at it only predictively because of what you stated. I could be wrong in assuming that. For that reason I hesitate to say it is outright useless. However I can understand if you looked at it only in a predictive manner that yes it would be useless entirely.

I'm not using the statistic predictively.

Predictiveness is a test for a statistic. Is it simply descriptive, showing what happened? Or is it a demonstration of a skill or an actual impact the player is having?

For various stats, it takes a certain sample size to stabilize and become predictive. Such as three point shooting. Once that happens, we can use that success rate as a descriptor for that player's ability to shoot. Until then, it is simply describing what happened, not why it happened. So the shots went in, but we don't ascribe that to the player being a good shooter until it has happened a lot. This is a straightforward concept most can grasp easily in extreme small sample cases but it holds up in larger samples too, just less obviously. If I took the court next year in training camp and hit one of one three point shots, I'm not expected to become a 100% three point shooter for my career. You would want to wait for a long while before you were sure what my talent actually was (hint, it's not good!). How long do you wait? Well the answer in statistics is you wait until the measured success rate is a better predictor of future success than a random guess or a different predictor.

For other stats, such as DFG%, this threshold is simply never reached. It never becomes predictive in any way (in a single season, anyway). Which means it's not describing something repeatable, such as a player's talent. It's only describing the outcomes on the court and has no ability to describe the "why" of those things happening.

The test I ran for DFG% was whether a single season DFG% could better predict, on average across the entire league, each player's next season DFG% than, say, their team's overall DFG%. Team DFG% won out (by a lot) on average - so being on a good defensive team was a better predictor of future DFG% success than who on that team was having success. Actually, even worse, using the league average DFG% for every player yielded on average a better prediction than using each player's DFG%. That's a brutal result for any statistic that intends to describe a player's ability to impact the game (in this case, to affect their opponent's ability to score). Actually tried comparing the previous year's DFG% to a random number generator pulling from the range of DFG%'s from the prior year, and the results were not significantly different.

Will some players see consistent trends? Of course! In a sample of 500 players you'd expect a fairly large number of them to show those trends, just based on random sampling, but also of course due to players most often staying with the same team year to year. Unfortunately, it does not mean they are meaningful.

[TrueTorontoFan]

I'm not using the statistic predictively.

Predictiveness is a test for a statistic. Is it simply descriptive, showing what happened? Or is it a demonstration of a skill or an actual impact the player is having?

For various stats, it takes a certain sample size to stabilize and become predictive. Such as three point shooting. Once that happens, we can use that success rate as a descriptor for that player's ability to shoot. Until then, it is simply describing what happened, not why it happened. So the shots went in, but we don't ascribe that to the player being a good shooter until it has happened a lot. This is a straightforward concept most can grasp easily in extreme small sample cases but it holds up in larger samples too, just less obviously. If I took the court next year in training camp and hit one of one three point shots, I'm not expected to become a 100% three point shooter for my career. You would want to wait for a long while before you were sure what my talent actually was (hint, it's not good!). How long do you wait? Well the answer in statistics is you wait until the measured success rate is a better predictor of future success than a random guess or a different predictor.

For other stats, such as DFG%, this threshold is simply never reached. It never becomes predictive in any way (in a single season, anyway). Which means it's not describing something repeatable, such as a player's talent. It's only describing the outcomes on the court and has no ability to describe the "why" of those things happening.

The test I ran for DFG% was whether a single season DFG% could better predict, on average across the entire league, each player's next season DFG% than, say, their team's overall DFG%. Team DFG% won out (by a lot) on average - so being on a good defensive team was a better predictor of future DFG% success than who on that team was having success. Actually, even worse, using the league average DFG% for every player yielded on average a better prediction than using each player's DFG%. That's a brutal result for any statistic that intends to describe a player's ability to impact the game (in this case, to affect their opponent's ability to score). Actually tried comparing the previous year's DFG% to a random number generator pulling from the range of DFG%'s from the prior year, and the results were not significantly different.

Will some players see consistent trends? Of course! In a sample of 500 players you'd expect a fairly large number of them to show those trends, just based on random sampling, but also of course due to players most often staying with the same team year to year. Unfortunately, it does not mean they are meaningful.

.......
It sounds like you ran predictive modelling which is exactly what I suggested you did Dan.


Uneless I am missing something because you keep talking about using the stat to predict future trends which is not what I did.

Will some players see consistent trends? Of course! In a sample of 500 players you'd expect a fairly large number of them to show those trends, just based on random sampling

'''

of course which I accounted for.

[DanH]

It sounds like you ran predictive modelling which is exactly what I suggested you did Dan.

Uneless I am missing something because you keep talking about using the stat to predict future trends which is not what I did.

of course which I accounted for.

The thing you are missing is the thing I explicitly said. I am not using the DFG% stat to predict future trends. I am comparing the stat as a predictor of future trends to any other random number, so see which is better, as a test of the relationship between the statistic and the talent or control of the player in question. I'm not coming up with projections for how a player will defend in the future, I'm testing how well the statistic captures the player's actual defensive impacts now. Predictive testing of statistics is extremely common and is the most basic pass/fail test of whether that statistic is capturing anything other than noise, and DFG% fails that test pretty handily.

What I'm saying is this: if I was going to flip a coin, and asked you to "defend" my attempts, and I wanted heads, and I flipped 10 heads in a row, are you a bad defender? Or were you just unlucky?

One way to check is to see if we did that experiment for 500 people, are their first 10 flips more predictive of the second ten flips than just a random guess, or an average value? In that case, it would not be, the vast majority of those samples would skew back to roughly 4-6 heads no matter what the first set of flips were. So the statistic we captured about you defending my coin flips fails the test - it doesn't predict your ability to defend the next ten "shots", therefore it probably isn't actually describing your defence on the first ten shots at all either, it was just random luck.

DFG% fails the same test, with much bigger samples. It's not about using the statistic to actually predict future successes, it's about seeing whether there is any predictive value at all to determine whether the stat is actually measuring what you think it is (in this case, a player's ability to play good defence and prevent efficient scoring by the guy they are guarding). And it rather emphatically is not.