Everything Gary Trent Jr.

[TrueTorontoFan]

[TrueTorontoFan]

Here, a miniature demonstration of the sort of comparison you can do:

I took the top 100 shot defenders (in terms of DFGA) last season for 3 point attempts. I took the top 100 from the season before as well. As it turned out, there were ~50 guys who were on both lists. (I did this to make the data gathering quicker for this demonstration, I have previously done this for the whole league). But these highest volume defenders are where you'd expect the most stable results.

The average deviation (good or bad) from expected 3PT% for the opponent last year was ~2%. So the good defenders on average reduced their opponents' 3PT% by 2%, and the bad ones increased it by 2%.

Then I looked at how much each player's defensive FG% impact changed from last year to the year prior. The average change from one year to the next was 2.6%. That means, on average, the top 50 guys in terms of defended shots had their defensive FG% impact change by more than it's starting point value - meaning they were, on average, likely to see a performance change big enough to swing them from positive to negative or vice versa. These are the highest volume, theoretically most stable sample of players in the league, and the year to year instability is larger than the value we are trying to measure on average!

Now, 3PT% is an extreme case, but I've done the legwork before and spoiler alert, this pattern holds up for all non-paint field goal ranges, over the entire league or high volume samples like the above. It's only in the paint (rim protection) that the highest volume group starts to look like the numbers could have anything remotely close to stability, and even that goes away if looking at lower volume rim protectors.

So correct me if I am wrong you did not run any sort of regression on this data set in the past? Did you run any sort of thing that spit out actual p-values?

[TrueTorontoFan]

Here, a miniature demonstration of the sort of comparison you can do:

I took the top 100 shot defenders (in terms of DFGA) last season for 3 point attempts. I took the top 100 from the season before as well. As it turned out, there were ~50 guys who were on both lists. (I did this to make the data gathering quicker for this demonstration, I have previously done this for the whole league). But these highest volume defenders are where you'd expect the most stable results.

The average deviation (good or bad) from expected 3PT% for the opponent last year was ~2%. So the good defenders on average reduced their opponents' 3PT% by 2%, and the bad ones increased it by 2%.

Then I looked at how much each player's defensive FG% impact changed from last year to the year prior. The average change from one year to the next was 2.6%. That means, on average, the top 50 guys in terms of defended shots had their defensive FG% impact change by more than it's starting point value - meaning they were, on average, likely to see a performance change big enough to swing them from positive to negative or vice versa. These are the highest volume, theoretically most stable sample of players in the league, and the year to year instability is larger than the value we are trying to measure on average!

Now, 3PT% is an extreme case, but I've done the legwork before and spoiler alert, this pattern holds up for all non-paint field goal ranges, over the entire league or high volume samples like the above. It's only in the paint (rim protection) that the highest volume group starts to look like the numbers could have anything remotely close to stability, and even that goes away if looking at lower volume rim protectors.

I think this is where the key difference lies. I was not looking for consistency across years but within a year does player x looking backward have a positive or negative differential and how was that players overall defensive impact during said year.
I could be wrong which is why I am leaving it open... but it seems like you are outlining that you are trying to find some sort of pattern between years for players. this goes back to my point where I said all stats are past-orientated but not necessarily useful for predictions.



Now I will say this again. I am lightly pushing back. I am assuming you know the difference between me being harsher or lighter iwth my pushbacks. I and lightly pushing back .... but do not mistake my push backs for me arguing in good or bad faith because like I said above I am not going there and I am not assuming that so don't. have the tiniest bit of respect for me

[Kagemusha]

Scottie checks all the boxes in his definition of a PG.

[DanH]

you made a point earlier about there being al ot of noise with 3 point shot but I asked what about beyond 15 feet but within 3 point shot and between 6-15 feet...

and again don't ever suggest I am doing this in bad faith again.

OK. Dig up the same numbers for those ranges then, show me how wrong I am. I'm telling you I've done it all and it's all noise. Midrange jump shots are just as noisy and high variance as three pointers, they are still long jumpers.

[DanH]

I think this is where the key difference lies. I was not looking for consistency across years but within a year does player x looking backward have a positive or negative differential and how was that players overall defensive impact during said year.
I could be wrong which is why I am leaving it open... but it seems like you are outlining that you are trying to find some sort of pattern between years for players. this goes back to my point where I said all stats are past-orientated but not necessarily useful for predictions.



Now I will say this again. I am lightly pushing back. I am assuming you know the difference between me being harsher or lighter iwth my pushbacks. I and lightly pushing back .... but do not mistake my push backs for me arguing in good or bad faith because like I said above I am not going there and I am not assuming that so don't. have the tiniest bit of respect for me

I feel like my responding to every request you've made with more information has been a pretty good sign of respect, no?

My point is looking backward is only of value if the statistic actually tells you anything. Remember the coin flip example? Looking backward at those coin tosses didn't tell me anything about your defence. We could pretend it did, but we know it didn't. Same thing here. The statistic is only of value if it is describing an effect the defender had, rather than just a pure outcome. And every test we can run on DFG% shows it does not really describe any effect the defender is having.

You want to run a full regression you go right ahead. Anything that fails a simple standard deviation check is going to be garbage in a regression too.

[Dr Hannibal Lecter]

I've said from day one..that Scottie should be our starting pg.

[TrueTorontoFan]

I feel like my responding to every request you've made with more information has been a pretty good sign of respect, no?

My point is looking backward is only of value if the statistic actually tells you anything. Remember the coin flip example? Looking backward at those coin tosses didn't tell me anything about your defence. We could pretend it did, but we know it didn't. Same thing here. The statistic is only of value if it is describing an effect the defender had, rather than just a pure outcome. And every test we can run on DFG% shows it does not really describe any effect the defender is having.

You want to run a full regression you go right ahead. Anything that fails a simple standard deviation check is going to be garbage in a regression too.

OK. Dig up the same numbers for those ranges then, show me how wrong I am. I'm telling you I've done it all and it's all noise. Midrange jump shots are just as noisy and high variance as three pointers, they are still long jumpers.

I told you that I want to dig into it just not at this exact second and I will. In the mean time I am asking you questions. For a few reasons. If I run the same analysis and get the same results ok checks out fair enough but if I run a different anaysis then what... so that is why I am asking you what you specifically did... not to be an asshole.

In terms of the notion of indepedence yes I fully understand that but I am just saying I haven't looked into it.




That is what I have said for three responses in a row and yet you still keep trying to paint me as bad faith ....

"My point is looking backward is only of value if the statistic actually tells you anything" this specifically with stats in general... I mean its debateable. So yes if a stat tells you nothing at all and is just a random phenomenon fair enough I think its fair to ask if it is just coincidence or not. Take for example you are trying to compare to species amino acid sequence because you think the protein structure is responsible for... shifts in lifespan of an organism. You initially look at it one way and find no specific connection but then you run it another way and you find indeed there is a connection ... but perhaps it isn't the one you thought ... instead of being related to cysteine residues it has to do with tyrosine residues ... which ironically enough have a tendency to interact with cysteine residues... do you just ignore the analysis. (Real thing that has happened).

[DanH]

I told you that I want to dig into it just not at this exact second and I will. In the mean time I am asking you questions. For a few reasons. If I run the same analysis and get the same results ok checks out fair enough but if I run a different anaysis then what... so that is why I am asking you what you specifically did... not to be an asshole.

In terms of the notion of indepedence yes I fully understand that but I am just saying I haven't looked into it.




That is what I have said for three responses in a row and yet you still keep trying to paint me as bad faith ....

"My point is looking backward is only of value if the statistic actually tells you anything" this specifically with stats in general... I mean its debateable. So yes if a stat tells you nothing at all and is just a random phenomenon fair enough I think its fair to ask if it is just coincidence or not. Take for example you are trying to compare to species amino acid sequence because you think the protein structure is responsible for... shifts in lifespan of an organism. You initially look at it one way and find no specific connection but then you run it another way and you find indeed there is a connection ... but perhaps it isn't the one you thought ... instead of being related to cysteine residues it has to do with tyrosine residues ... which ironically enough have a tendency to interact with cysteine residues... do you just ignore the analysis. (Real thing that has happened).

OK man, whatever. I've given you everything I know on DFG%, have fun with it or ignore it as you see fit.

[Kagemusha]

I've said from day one..that Scottie should be our starting pg.

And Fred should be our SG,
Gary his backup.

[TrueTorontoFan]

And Fred should be our SG,
Gary his backup.

that is what the video hinted at and this is a thought that is becoming more prevalent... so why am I pegged a hater when I basically outline the reasons why this should be the case?

[Kagemusha]

that is what the video hinted at and this is a thought that is becoming more prevalent... so why am I pegged a hater when I basically outline the reasons why this should be the case?

Fred , or Gary hater?

[TrueTorontoFan]

Fred , or Gary hater?

Gary

[TrueTorontoFan]

OK man, whatever. I've given you everything I know on DFG%, have fun with it or ignore it as you see fit.

I will say one thing. It is hard because of how schemes shift even throughout a season and given the different defensive tasks that a player is asked and how that isn't standardized but I suppose that gets back to the noise. Can it not be used along side impact stats ?

[DanH]

I will say one thing. It is hard because of how schemes shift even throughout a season and given the different defensive tasks that a player is asked and how that isn't standardized but I suppose that gets back to the noise. Can it not be used along side impact stats ?

You can use anything you like alongside impact stats. It doesn't provide any additional value though.