Posted by dahsdebater on 1/22/2020 12:48:00 AM (view original):
Here are all the guys who have averaged at least 3 3 point attempts per game for my team over the past 20 seasons, in their senior seasons. Obviously it's a fairly small sample, but to expand it meaningfully I'd have to expand it A LOT to compensate for differences in eras, coaching and team building strategies, systems, etc.
| SPD |
PER |
BH |
Three% |
PPG |
| 71 |
100 |
60 |
0.497 |
24.9 |
| 84 |
100 |
100 |
0.537 |
22.6 |
| 99 |
99 |
82 |
0.482 |
22.5 |
| 87 |
99 |
100 |
0.513 |
18.3 |
| 77 |
98 |
56 |
0.497 |
13.9 |
| 83 |
98 |
85 |
0.486 |
17.7 |
| 75 |
97 |
97 |
0.467 |
13.6 |
| 57 |
96 |
53 |
0.5 |
15.1 |
| 81 |
96 |
78 |
0.472 |
13.9 |
| 72 |
95 |
88 |
0.488 |
16.8 |
| 72 |
94 |
72 |
0.463 |
17.9 |
| 72 |
94 |
72 |
0.463 |
17.9 |
| 64 |
92 |
72 |
0.452 |
13.6 |
| 74 |
92 |
42 |
0.464 |
16.6 |
| 70 |
92 |
73 |
0.449 |
17 |
| 60 |
91 |
81 |
0.463 |
18.2 |
| 84 |
91 |
56 |
0.459 |
11.6 |
| 82 |
91 |
60 |
0.433 |
17.5 |
| 63 |
87 |
75 |
0.437 |
14.7 |
| 87 |
71 |
74 |
0.394 |
10 |
| 64 |
70 |
75 |
0.37 |
11.2 |
| 79 |
62 |
75 |
0.381 |
8.3 |
If I only look at the guys with elite PER - basically dropping the bottom 3 guys from the list - the remaining 3% vs PER can be fit to a line with 3%=.0061PER-.0994 with an R^2 of .722. So basically, even ignoring all the variation on the list in SPD and BH, PER differentiation between 87 and 100 is still strongly correlated to 3 point shooting efficiency at better than half a percent per point of PER. FWIW, BH and SPD each contribute about .05%/point with R^2 of .04 and .02, respectively. So like I said, you might not notice a huge difference between 85 and 100 BH or SPD, but you sure as heck notice the difference from 15 points of PER.
interesting stuff, dahs - sorry for the greatly delayed response, i had to chew on this one for a bit. one issue is that i have been having an internal debate about how much stock can go into any of this. one of the things i picked up on in my rookie days & have carried with me ever since is that you have to play top competition, and use those lessons, if you want to calibrate your scheme for top competition. the lessons simply differ too much from those one might glean from lower competition. outside of the standard disclaimers (like yours about small sample size), this is my biggest objection about drawing conclusions from these stats. you are playing an extremely poor schedule - mostly sims - so i feel like in general, this data tells you a lot more about how to beat terrible sims, than how to win championships.
anyway, i don't think that statement, along with the usual disclaimers, can just brush away all of the data - as you say, its the same sim engine. i would have liked to see some really good spd/bh ratings with more moderate per mixed in, things of that nature, to get a bit more variety in the players (or maybe even just including the 3 you excluded - but i guess would that go against the premise, of trying to consider near maxed-out guys?) - i guess things like strength of opponents, pg passing and such, these are relatively significant compared to a guy having 4 per more than another. i'm generally surprised by that R^2 you get on per alone, which correct me if im wrong, suggests the per differences alone explain the large majority of the 3pt% differences? in general, that is just not realistic - playing a top 5 opponent vs a bottom 100 should dwarf a few per - which again tells me we have to be pretty careful about drawing conclusions. i don't know if you just run such a consistent scheme, with such consistent caliber of opponents, that you are minimizing all those differences - or if small sample size (or something else) is skewing this data in ways that are unreasonable, making all conclusions highly suspect.
i do think there is some interesting stuff in here though. first off, the per comes back as 10:1 important on the bh/spd, which is quite extreme. i think we would pretty much all agree that as a general rule of thumb, that would be a terrible one! as in, for a guy at 35% 3s, you'd never choose to add 1 per over 10 points of bh or spd (purely from a 3pt shooting standpoint). but, i do think that creates an interesting way which this data can fit with my model - albeit with a conclusion that my model isn't exactly right. i'm not sure this is enough to draw a strong conclusion on - but like you say, it IS the same sim engine across the board, and i struggle justify just hand-waving away stuff like this entirely. anyway, here is my take on the data you presented:
for starters, because i have to refer back to it, here is roughly how i look at high d1 guards - my claim about 90/90/90 spd/per/bh is that there is near-negligible difference between that and 100/100/100. however, obviously not all guards are so even in their ratings. i convert between those ratings with a loose, poorly calibrated scheme of 2:1 per:spd/bh, so i'll consider 90/90/90 equal to 80/95/90 (-10 spd for +5 per) or 80/100/80 (-10 spd, -10 bh, +10 per). however, i won't go the other way. no way a 100spd/bh 80per guy is equivalent. so on some level, im tacitly recognizing something at least similar to your point, about giving up per at the really high end being different than giving up spd/bh. i definitely don't think they are all 100% interchangeable, that's in no way the point i'm trying to make.
on interchanging per for spd/bh - my theory on shooting has long been that, regardless of underlying mechanics, looking at shooting as a 2-step process gives you a model that feels more in line with how the game actually works, than a 1-step model (step 1 - shot). specifically, i look at shooting has having a phase 1 of 'getting open', and phase 2 of 'taking the shot'. saying 'getting open' is sort of unclear - the player is already selected by this point - but i'm suggesting basically the game first has to figure out how decent of a shot the guy has, before trying to figure out how well of a shot he takes. specifically, i site this concept regularly to explain why a high per, garbage spd/bh player can be so successful against **** opponents, but how it would be a grave mistake for coaches to assume that flies against top competition (it does not).
to be clear, im suggesting if you can't get a decent shot in phase 1, you can't have good odds on the RNG in phase 2. i have no idea if there are really 2 phases mechanically, but it sure feels like that more closely describes the behavior, than thinking of it as 1 step. with 1 step, you'd think, a guy is as good as he is, and the better the defense, the worse he gets - but you wouldn't expect inversion - like where player A is a bit better than player B against a crap team, but it flips against a better team. i definitely believe this happens, specifically, that you can tolerate crap ath/spd/bh with great lp/per (this concept holds across the spectrum IMO, not just for guards) against **** defense, but that such a player gets destroyed by a crippling defense - they just can't get open - so the ceiling on how good of a shot they can get is low - maxing it (with great lp/per) still results in a fairly low value. this is why i tend to have my mid tier scorers on my crappier teams (who play crappier opponents) come from those higher lp/per guys with crappier ath/spd, but for great teams, my mid tier of offense tends to come from 90 ath/spd types who have lesser lp/per.
now, back to the data. i could reconcile your data, with some of my observations, as follows - if we assume you are operating in a space where your better players are way better than the opposing D (which seems like a given, with the sim-laden schedule and the quality of the ratings you posted there), maybe you are maxing out that phase 1 of getting open, which is really a spd/bh thing - but not the phase 2 of shooting, where per is king. maybe per doesn't ever max out, its just that by 90 against tough defenses, the benefit of going to 100 can't be seen by the naked eye? or else, appears negligible? we know the spd/bh to per value on 3pt shooting is nowhere near 10:1, but it would be interesting if there was a space where that was true - after spd/bh max - as your data suggests. i do wonder however - even if you take the data at face value (very dangerous), and assume in your case, high d3 scorers against **** defense, there is a 10:1 value on per to bh/spd - does this space even exist (at the extreme 10:1 ratio) at competitive levels of d3 play, even? it almost certainly does not for d1, but its nearly impossible to accurately speculate where that type of behavior is possible and where it isnt. top d3 scorers against **** d3 defenses is the most extreme case there is.
i will say this - i am genuinely surprised you have so many guys up around 50%. that is a figure i consider beyond the range of possible - but like just everything i say, i am usually only concerned with high level competition. you simply cannot pull numbers anywhere close to that, against elite 8 and up level competition (which is where titles are really won and lost) - especially at d1, but frankly, anywhere. still, i would have anticipated that ceiling was a little lower in the real world (as opposed to a theoretical team with 1 in everything), even against garbage sims. very interesting.
anyway, in d1 against the highest tier of d1 defense, an elite guy by my measure (90/90/90) is barely topping 40%, on average. that's pretty different from the 50% or so your guys are pulling, but its not altogether clear to me how strong defense impacts the way per/spd/bh affect 3pt%. its pretty clear from even really simple observation that the 100/100/100 guys are not in excess of 46% (+6%), not against top competition - not even close. i could maybe buy something in the 2% range?
so, what makes the difference, what makes that marginal value of per so different in high d1 and your scenario here, i'm not sure - of course, we should expect significant outcome differences between how players play against the worst and best of defenses. but why specifically, i don't know. i do feel like the situations are too apples to oranges to draw too many conclusion - but you do have me wondering if by 90/90/90 in d1, the reason 100/100/100 feels meaningless is because the spd/bh may actually be (or may not - maybe its impossible to 'max' those, against a top tier defense - but maybe it comes close?), and because the +10 per is only worth a % or 2, which is pretty hard to discern with the massive amount of noise in the signal (what the rest of your team looks like, the opponent, your rate of 3pta and how that impacts the enemy defensive strategy, etc).
so, perhaps, i am a bit misleading in my 90/90/90 = 100/100/100 claim. the part i have no doubt about is that you can build crushingly dominant teams with those 90/90/90 guys leading the way - this was possible even when top d1 teams were quite significantly better than top d1 teams today, so it must be at least as true today as it was (i think that tracks?). your take, that per isn't capped but maybe spd/bh, kinda jives with how i'll trade per for spd/bh, but not the reverse, in assessing who is elite and who is not (how i convert real players to that 90/90/90 standard). so maybe the extra 10 per helps, but its just a small effect against those good of defenses - but where say, if you are doing EE planning, you should be a lot more willing to hold back the bh than the per? or practice planning for growth, maybe you prioritize the per over bh more than i'd think?
it could also very well be that in high d1, the difference is fairly negligible, as i suggest - but in d2/d3, things are different, because you can build more separation from the defense (because top d2/d3 scorers are a lot better relative to top defenses, than they are in d1)? maybe in d2/d3, assuming you max at 80/80/80 is pretty misleading - both because of the d2/d3 vs d1 differences, but maybe also just from the raw difference there, like 80 per to 100 is +20, where the d1 version, 90->100, is only +10?
i will concede your data definitely has me questioning how valid my claim is in the lower divisions, on a broad basis (not even just the per part) - i only had 1 ultra dominant d2 team, and it was about 10 years ago, when the game was very different. i'm not really questioning my claim for d1 in a broad sense, just because i've been too successful there too many times, and have too much experience with the subject matter in question (i get to see 90/90/90 vs near perfect guys literally all the time - in d2/d3, those near-perfect guys were insanely rare, at least in my day). however - i will say, even for d2/d3, i would definitely stick with the overall approach of diminishing returns being key - about how one must focus on getting 'enough' offense, rebounding, and guard skills, before really trying to push the pedal all the way to the floor on defense, if you want to be successful. that general approach i think is right, the diminishing returns and how they impact team abilities and all that - but i could definitely be wrong more specifically about the 80/80/80=100/100/100 thing (for d3), especially on the per. maybe all of though? (assuming good defenses - i have to imagine bh/spd max out less against top teams than sims)
finally, what i think would be really valuable for further study, is running the numbers on a situation that approximates the situation most folks care about (what should you do against top teams, when you have a championship hopeful). it is always interesting to run the numbers in the extreme cases - it sometimes shows things you just can't pick out otherwise - maybe this is one of those times? it at least throws out an idea for further inspection, but i do think the difference in scenarios is so extreme, that it is hard to draw any concrete conclusions about how things work in very competitive play. i'll definitely keep a closer eye on this going forward.
thanks for taking the time to put that data together, and to push back on my statements in a thoughtful and substantive way! its not very often i get to experience that in this setting, and i very much appreciate it!