Attributes, What's the Deal With, But for Real This Time? (1,296 views)

[eric]

So I played around with the software for a bit, and there's a lot going on. I took Raef LaFrentz, reduced all his attributes to 5 (the minimum), then added to them in a systematic way and saw what happened. He is a center so these findings may be center-specific, and I will denote the ones I am confident are.

1. As many other GMs have stated, not all grade increases are equal.
2. Not all ratings behave the same way.
3. There are effective caps on attributes separate and distinct from the hard caps based on position.
4. Attributes are otherwise linear inputs to ratings.


.

First, let's look at the rating to grade relationship. Here are the minimum ratings needed for each grade:

rating grade
1 f-
6 f
11 f+
16 d-
21 d
26 d+
31 c-
39 c
53 c+
61 b-
66 b
71 b+
76 a-
86 a
96 a+
...and to elaborate on this, here's the change in rating to obtain a given grade from the floor of the previous grade...

rating grade
5 f
5 f+
5 d-
5 d
5 d+
5 c-
8 c
14 c+
8 b-
5 b
5 b+
5 a-
10 a
10 a+
...so if your player goes C- to C in a given camp, you can just forget about getting to C+. The most efficient attribute to rating value is 0.75, and to get 14 rating points would require 14 / .75 = 19 attribute points = 3 skill camps AND 2 reward camps in the same attribute. On average the distance [C+ to C-] is slightly less than the distance [A- C+].

.

The Inside Scoring, Outside Scoring, Handling, and Rebounding ratings all behave in a straightforward way...

i o h r
0.50 inside scoring
0.50 jump shot
0.50 3 point shot
0.08
0.17 0.05 quickness
0.75 passing
0.35 offensive rebounding
0.45 defensive rebounding
0.20 0.15 strength
0.30 jumping
They all add up to 1, they have no soft attribute caps, everything makes sense. Handling is a little weird, but that might just be because I used a center. Defense is where things get super weird:

d
0.35 quickness
0.10 stealing
0.35 shot blocking
0.45 post defense
0.15 perimeter defense
0.30 drive defense
0.30 strength
This adds up to 2 instead of 1 and there are soft caps on literally every attribute...

45 quickness
65 stealing
85 shot blocking
85 post defense
55 perimeter defense
80 drive defense
85 strength
...now, I checked to see how this worked. The cap is NOT on the rating, and the relationship is always linear if not capped. Whether you're at 1 defense or 50 defense, going from 45 quickness to 100 quickness doesn't change the rating (and therefore letter grade) at all. Whether you're at 1 defense or 50 defense, going from 5 quickness to 45 quickness changes rating by .35 * 40 = 14 rating points. Looking at the caps I am reasonably sure they are position-specific (although I personally would value quickness a lot higher for centers) and this explains how position changes can change letter grades without changing the underlying attributes, which is a reasonably elegant system. These are rough (nearest 5) values, but the coincidence between 85 repeatedly cropping up and our being forbidden from raising an attribute above 85 is pretty tantalizing.

.

.

Here's the full chart:

t i o h d r
0.50 0.50 inside scoring
0.50 0.50 jump shot
0.50 0.50 3 point shot
0.08 0.08 handling
0.57 0.17 0.35 0.05 quickness
0.75 0.75 passing
0.10 0.10 stealing
0.35 0.35 shot blocking
0.45 0.45 post defense
0.15 0.15 perimeter defense
0.30 0.30 drive defense
0.35 0.35 offensive rebounding
0.45 0.45 defensive rebounding
0.65 0.20 0.30 0.15 strength
0.30 0.30 jumping
I nailed inside scoring and outside scoring in the other thread. I underestimated offensive rebounding big time but with so few camps it was a wild guess anyway. I vastly underestimated passing and defense was a shot in the dark. I anticipate defense values changing with position, I do not anticipate that with handling.

Here is the follow-up plan for whenever I get around to it (not soon):

-Run a few seasons at 45 quickness and at 100 quickness. Is performance entirely governed by ratings, or do attributes matter too?
-When that's settled, run a few seasons with widely varying heights and weights and see what that does.
-Repeat the phase 1 analysis with the other positions to see if attribute -> rating relationships hold and to see what their soft caps are.
-Quantify potential.
-See how shot generation works.

[Odin]

each position would have it's own values for defense

[Druce]

i enjoyed this.

[ANK1990]

Defense is fucked up. Especially for small forwards where there are a bunch of effective caps at like 65.

[eric]

Alright, these results are very much preliminary but very much astonishing. I took Raef LaFrentz and set him up as follows...
25 stealing, shot blocking, perimeter/drive/post defense, strength
variable quickness
5 everything else

I made him the only center on the roster, all three levels of the depth chart, turned off injuries. I ran the same season 20 times, the first ten with 45 quickness and the second ten with 100 quickness. In each case the player in question had a Defense rating of 17 and grade of D-, while Handling went from 10/F to 19/D- and Rebounding went from 10/F to 13/F+. I haven't had time to calculate out Win Shares, but here are the raw values for Raef.

20843 21086 Minutes
672 641 FG
1765 1700 FGA
168 198 FT
469 593 FTA
1759 1710 Rebounds
400 428 Assists
414 590 Steals
155 143 Blocks
718 696 Turnovers
2102 1940 Fouls
1512 1480 Points
Now, I can calculate standard deviations of a 10 sample set and do a little magic, and we can get odds ratios to see which differences are statistically significant.

ratio 95% stat
1.01 0.01 Minutes
0.95 0.06 FG
0.96 0.03 FGA
1.18 0.11 FT
1.26 0.07 FTA
0.97 0.03 Rebounds
1.07 0.06 Assists
1.43 0.06 Steals
0.92 0.08 Blocks
0.97 0.06 Turnovers
0.92 0.03 Fouls
0.98 0.06 Points
Note that even though it's not statistically significant, there IS a 1% difference in minutes, and there will be fluctuations in Pace (league-wide and for this team) as well. For anything that's borderline (rebounds, assists, blocks) we should therefore lean towards insignificant difference. Note also that none of that means s*** for steals, which are so much higher it's ridiculous...

...with the same Defense grade. And this isn't a "oh well some D-s are higher than others" thing, they have the same numeric rating underlying the grade. This is a big deal! The entire rating system is a lie.

.

So now the mission is to abandon the rating system entirely and to see what changes in each attribute actually do. For instance, it appears quickness increases the ability to turn existing offensive uses into free throw attempts (though it does not change the overall amount of uses per minute) and decreases the rate of fouling. I haven't specifically looked for it but I am pretty sure the software does not break down fouls into types, so it's possible that these reduced fouls are less offensive fouls (and thus less turnovers) or less defensive fouls (and thus more steals). More research will reveal any numerical answer, I just don't think the specific research I'm doing will be right for that specific question. The trouble with that is that without the rating system we have to find a new system to quantify what it means to get 50 more steals (or whatever). Luckily we already have that system: Win Shares.

.

.

Bottom line: Defense has effective caps such that increasing an attribute will not increase the rating; however, increasing the attribute will still have an effect on play. (This begs the question of whether a player being +1 or whatever at a new position means anything at all.) I am going to do Win Shares analysis next, then I am going to do the height/weight analysis, then I will look at other attributes.

[eric]

Okay, so with Win Shares I can't do ratio magic because Win Shares aren't on an absolute scale the way rebounds etc. are: there's no such thing as a nega-rebound. Debound? Anyway, instead we add error in quadrature but it still works out pretty well for us, check it out:

dif 95% stat
13.6 4.0 ows
13.0 4.1 dws
26.6 6.8 ws
0.2 0.7 ows
3.0 0.5 dws
3.2 0.7 ws
The top three values are for the team, the next three are for Raef personally, and again these are comparing 10 years of 45 Quick Raef to 10 years of 100 Quick Raef. Interestingly, the effect on the team is much more pronounced than Raef's stats alone would indicate. This is plausible from a real world perspective, of course, as not everything shows up in the box score. Running the floor, boxing out, setting screens... all these things matter, and all these things would plausibly benefit from being a quicker player.

One problem we're going to run into is also illustrated here. Drawing more fouls should make a player better on the offensive end, but it so happens that ol' Raef shot 35% from the line and 38% from the floor, so he was better off devoting those touches to shooting from the floor. What I'm going to do as I proceed is keep doing both versions of the analysis, granular and overall. We're not going to be able to say something like "1 point of Quickness is always worth 0.04 team wins per year", because it depends on other factors. After I test 25 and 100 Strength against 45 and 100 Quickness (to make sure there aren't first-order interrelated effects) I'm going to build a base Raef that's more representative of the center we spend points on. This is tricky because we don't know the attributes of the players we add to, but I think we can all agree there's a more representative case than a solid F across the board guy.

[Druce]

this article must have been written in a mcdonalds...because i'm lovin it.

[eric]

Alright, so I did the four sets of ten seasons. Here's how to read this chart. All stats are explicitly pace-adjusted except for Win Shares which is pace-adjusted already. All counting stats except minutes are then adjusted to be per-time: per-36 for base stats and per-48 for personal Win Shares. All ratio stats and team Win Shares are left alone. Finally, the values are calculated per 1000 points of attribute.

quick strong stat
522 76 MP
-2 9 FG
-3 21 FGA
0 4 FT
1 13 FTA
0 0 3P
0 0 3PA
-2 19 REB
0 3 AST
4 1 STL
0 0 BLK
-1 5 TOV
-5 2 PF
-4 23 PT
-3 32 uses

0.07 0.19 dws
0.05 0.03 ws

-0.15 0.07 fga/use
0.19 0.39 fta/use
-0.05 -0.46 tov/use

-0.40 0.42 pts/tsa
0.53 -0.33 ast/tov
1.42 0.01 def/pf

8.90 41.50 dws
9.67 68.28 ws
So 5 points of Quickness (to this player) would be expected to give another 2.6 minutes per season, -0.025 personal fouls per 36 minutes, 0.05 team wins, and so on. These values were mostly similar whether I was going from 45 to 100 quickness with 25 strength or with 100 strength, but not similar enough to me, so I'm going to collect 20 seasons per sample from now on. Here are the conclusions I'm still comfortable drawing:

Quickness
-reduces fouling and therefore increases minutes
-increases passing efficiency but not passing volume
-increases the percentage of uses devoted to free throw attempts at the expense of FG:TOV 3:1 but not total uses
-increases steals and team defense

Strength
-increases all usage: scoring, passing, and defensive
-does not increase passing or defensive efficiency
-increases scoring efficiency by increasing FG% (not FT%) and reducing turnovers per use (charging fouls? illegal screens?)
-increases team defense and team offense

The last point with strength is worth elaborating on. It's possible that Raef's defensive efficiency is above average for this team, therefore him taking on more defensive plays (and thus leaving less to other people) increases the team's overall DRtg, but I doubt it. It's definitely not the case his taking more offensive plays can help the team's overall ORtg, and because the team's overall pace stayed the same he definitely wasn't creating looks ex nihilo.

When I redo this analysis with a better player it will be interesting to see how similar the values are.

[Lazy Pete]

how this only has 10 votes blows my mind. We should bring back +10s for this

[KwYawnza]

This is wild

[eric]

Really appreciate all the feedback gents

I rebuilt Raef to have 50 in everything but Handling, Passing, and Perimeter Defense (all 25) so I could add 50 points to everything. Three quick hits here.

1. Weight
I tried him at 140, 240, and 340 pounds. I found one thing that changed: FG%, and at about the rate of 0.1% (.001) per 5 pounds, so each 100 pounds was worth 2%. Literally every other stat was identical, which I found astonishing.

2. Current Rating
Tucked away in one of the .mdb sections is a field called "current rating", next to a mysterious field called "future rating". I started looking for it because when I added 100 pounds (of solid muscle) to Raef he jumped a spot on the roster, indicating that the software has some composite metric for players, and I'm pretty sure this is it. I regressed it against everyone's attributes and ratings and it always had a positive relationship, but it would take a lot more work to determine how it's calculated, and it's probably different per position, and it's probably got weird wrinkles to it. For instance, 340 pound Raef is +5 Current Rating over 240 pounds, but 240 and 140 have identical Current Ratings. I'm going to collect data on it as I go but I'm not going to put any real effort into cracking it, because it doesn't really have anything to do with anything we do here.

3. More Mystery
In the same .mdb section are fields labeled "Scoring", "Outside Scoring" (in the attribute rather than rating section), and "Stamina". Outside Scoring is always a 0 for everyone in the league, Scoring ranges from 0 to 100, Stamina from 0 to 1000. My guess is that Scoring is the baseline for how much a player wants to score, Outside Scoring for how much they want to take 3s, and Stamina I have no idea. Outside Scoring always being 0 explains why there's no Korver-esque spot-up guys in this league, the highest 3PA/FGA ratio last year was JR Smith's 40%, IRL Korver's this year was 75%. Again I'll keep an eye on these, but my main focus is height and then going attribute by attribute to see what happens, mainly because I can't edit any of these three fields directly so it would be really intricate to try and test them systematically.

[2poor]

Holy shit, I've wasted a lot of bucks on weight for apparently no reason

[Deleted]

so weight does not matter?

[eric]

Weight matters a little, FG% is always useful, so you should always make use of the free weight. When it comes to spending money on it, let's compare Strength to Weight: 3500 bucks will get you +3 strength or +5 pounds, so we do...

1000 strength = 0.42 pts/tsa
100 pounds = 0.035 pts/tsa

3 strength = .42 * 3 / 1000 = .00126
5 pounds = .035 * 5 / 100 = .00175

So just from the perspective of shooting efficiency, weight is more efficient. The problem is that Strength does so much more stuff that it has a measurable effect on team Win Shares... weight does not. It must help your team wins on some level, but if I can measure 20 seasons a piece at 200 pounds difference and not see it, the difference for 5 pounds is going to be very very very tiny.

It's important to keep in mind that Strength is famously good, so comparing weight to it might be unfair. For all we know Strength could be the only attribute that produces results in this methodology. I doubt it's going to be that extreme, Quickness has already produced some, but we'll see.

.

I've now checked height as well, at 5'11", 6'11", and 7'11", and these results are pretty shocking too:

1. Height makes for about 0.3 more rebounds per 36 minutes per foot, or about half what Strength gives for rebounding. This isn't shocking.
2. Height makes for about 0.2 less blocks per 36 minutes per foot. That's f***ing baffling.
3. Height makes for about 1.5 more team wins per season, most of which come on the offensive end.
4. But height does help team defense.

Here's my best guess. One big gap in our knowledge is offensive vs. defensive rebounds, because the software doesn't record them. The coder decided that height would generate way more offensive rebounds, maybe a few more defensive rebounds, less blocks BUT more shots altered or deterred. Why they made it that way instead of the more obvious "better defense AND more blocks" is beyond me.

Bottom line: height is good. A foot of it is worth about 500 Strength. This isn't super useful for player development because we can't add height as freely or purchase it at all, but it is (potentially) helpful for player comparison.

[eric]

I forgot to mention above, each foot of height was worth 3 Current Rating. Clearly Current Rating is garbage, the foot of height was worth way more wins than the 240->340 pounds transition that was worth 5 Current Rating.

Since I have all these bins of 20 seasons of absolutely identical players and teams and schedules and health, I thought I would go ahead and see how much team wins varied. I didn't keep track of actual team wins and wouldn't want to anyway, because that introduces fluke wins in close games. The standard deviation for each of the 5 bins of 20 I've done (1 base, 2 height, 2 weight) were...

4.1
3.1
3.2
3.8
4.9

I think a reasonable estimate of the true standard deviation is 4 wins. This means that if your team has a true talent level of .500 (41 wins), you could reasonably expect to see anywhere from 49 to 33 wins. Put another way, there's no way for us to tell that an actual team that gets 49 wins is any different from an actual team that gets 33.

I think this is a useful reality check (so to speak) for any of us trying to evaluate our teams season-to-season. 16 games seems like a huge swing, but every bit of it could be just luck... in either direction!

[eric]

Scoring Options
So I held Raef constant and ran 20 seasons with Paul Pierce #1 option and Eric Williams #3 option, then 20 seasons vice versa. Things changed dramatically, so the main takeaway is that the options are ordered (most then medium then less) rather than listed (three equals). How they change in detail is pretty interesting.

Paul Pierce going from #1 to #3 option lost 4300 uses over the 20 years, but Eric Williams gained 5800 even though they moved the same amount. This isn't because the team pace went up; in fact, team pace went down slightly from 91.6 to 91.2. I suspect this is because Paul Pierce has 93 scoring (little low imo) compared to Eric William's 23, and there's a limit to how many touches a player can use. The way to test this would be to repeat the analysis with players who had identical Scoring scores, but nah. What matters is that if you have three very efficient players and a bunch of bums (e.g. the 2011 Heat), and one of the players happens to be a naturally low volume guy vs. two naturally high volume guys (e.g. Chris Bosh vs. LeBron and Wade), you would be best off making Bosh your #1 option because he'll take the most uses away from the bums. If he was just taking touches from LeBron, obviously you'd stick with LeBron #1, but if he's taking a bunch from Mike Bibby that LeBron wouldn't, that massive gain makes up for the minor loss on the touches he does take from LeBron.

Now, Eric Williams is no Chris Bosh, so the team lost about a win and a half per year with him as the #1 guy. It was also interesting to see that Williams became slightly more inefficient (shooting and passing), while Pierce was about as efficient shooting but even less efficient passing. My guess is that all of these fell within statistical uncertainty, and that the program doesn't have a usage/efficiency curve for a given player.

Buckets
I did Raef analysis on Inside Scoring, Jump Shot, and Three Point Shot. None of these changed the Scoring or Outside Scoring attributes, confirming that they are like height or college and are simply declared at creation. Because this analysis repeats the same year over and over, I don't have any way of knowing if Scoring changes over time. Outside Scoring is always 0.

50 Inside Scoring was worth 73 team wins...
+10 uses per 36 minutes
more FGA and FTA per use at the cost of TOV per use, or all the new uses are scoring uses rather than passing
+.15 points per true shot attempt
+250 minutes per season

50 Jump Shot was worth 34 team wins...
+.11 points per true shot attempt, albeit entirely from FT% and 3P%
+.02 3PA/use, which might not even be significant, and definitely no significant change in FTA/use

50 Three Point Shot was worth 22 team wins...
+.06 points per true shot attempt, entirely from 3P%, no effect on FT%
+.05 3PA/use

Now, it's worth reiterating that these values are all for a center, and one in a balanced offense. It stands to reason that a perimeter player in an outside offense would get a lot more use out of Three Point Shot's effect even if it had the same +14% 3P%. The software is from the early aughts, it's not going to factor in the spacing concept so prevalent in today's NBA because that was only invented as a response to Thibodeau's defense made famous in 2008. This is an unavoidable limitation of the analysis. I am eventually going to repeat it with a point guard to give us a better picture, though.

[eric]

Dimes

Raef analysis.

50 Handling was worth 57 team wins...
-17 turnovers per 36 minutes per 20 seasons
-15 uses per 36 minutes per 20 seasons

Pretty straightforward effect here. Raef just stopped turning the ball over so much. He didn't try to touch the ball more, or pass it more once he did get it, he just made the catch and shot. Turnovers are very harshly penalized by Win Shares, so it makes sense that cutting down on that alone would have such an outsized impact. It's also possible that team pace dropped slightly, indicating that sometimes Raef made the catch and kept the offense running for good offensive players (i.e. not him) to get a look, but the effect was so tiny I'd want to see more data before ruling on it.

50 Quickness was worth 0 team wins...
+7 steals per 36 minutes per 20 seasons
+37 minutes per season

It was also worth -6 personal fouls per 36 minutes, but to me that's just on the wrong edge of statistical significance, a claim buttressed by the fact it was worth -6 rebounds per 36 minutes for an attribute that allegedly helps rebounding. Quickness for centers is just garbage, bottom line.

50 passing was worth 0 team wins...
+2 assists per 36 minutes
+1 turnovers per 36 minutes
+2 uses per 36 minutes
+.5 ast/tov ratio
no change in team pace

This one would probably be better if the player wasn't such a dreadful passer to begin with, because in addition to making the player more efficient at passing it makes him more voluminous at it, and Win Shares is notoriously harsh on passing. You need to be well above 3:1 ast:tov for WS to think you should pass, and even the marginal increase for Raef is well below that. The no change in team pace also contributes to this, because it means Raef took uses away from better (i.e. dang near any other) player. It is useful to notice, however, that Passing is the only attribute besides Inside Scoring so far that increases a player's usage, even though the effect is about one sixth as big.

[Lazy Pete]

Next time you and raef hit the hardwood, can you take a look at the improvement (or lack thereof) from an increase in jumping alone?

[eric]

Jun 14, 2015 20:43:18 GMT -6 Lazy Pete said:

Next time you and raef hit the hardwood, can you take a look at the improvement (or lack thereof) from an increase in jumping alone?

I'm making my way down the list of attributes as ordered in the software, so Jumping is going to be dead last, but it will be done. Halfway through Post Defense right now.

[eric]

Raef analysis

D Fense

50 Stealing was worth 50 team wins...
17 steals per 36 minutes per 20 seasons
Pretty straightforward, but part of the increase was in Offensive Win Shares, which does make sense: live ball turnovers generate fast breaks, fast breaks are good for offense.

50 Shot Blocking was worth 125(!!!) team wins...
57 blocks per 36 minutes per 20 seasons
No change in rebounds or steals, which is not real life accurate.
Massive increase in team defense, which is also not real life accurate.
A slightly smaller increase in team offense than 50 points in Jump Shot. Think about that one for a second.
It's a joke how much better shot blocking is than everything else.

It's also important to note that even though Shot Blocking's impact on Defense grade is capped at 85, the increase from 75 to 100 shot blocking produced 31 blocks while the increase from 50 to 75 produced 26. The evidence that the soft attribute caps are meaningless is overwhelming at this point, which should radically alter the way we think about evaluating defensive prospects.

50 Post Defense was worth 46 team wins...
No change in any counting stat, as one would expect.

50 Perimeter Defense was somewhat inconclusive.
No change in any counting stat.
The first increase of 25 points saw -7 team WS and +21 team DWS, the second saw +34 and -15. None of that makes sense, but I'm comfortable concluding that this attribute (the only one to behave this way) was especially prone to noise (there's one expected in any group of twenty) and its effect is small, probably 27 team wins.

50 Drive Defense was worth 13 team wins.
No change in any counting stat.

13+46+27+50 = 136
The four other stats put together are barely more than Shot Blocking alone. This is heavy.

.

Rebounds

50 Offensive Rebounding was worth 67 team wins...
31 rebounds per 36 per 20
12 uses per 36 per 20, and the use distribution of shots/free throws/turnovers remained the same
Helped offense and defense at about 50/50

50 Defensive Rebounding was worth 53 team wins...
54 rebounds per 36 per 20
Helped offense and defense at about 20/80

Rebounding is the special teams of basketball, and so it makes sense that each type of rebounding would help both ordinary teams. It's much easier to fast break off a miss than a make, though some teams (e.g. 2013 Miami) did the latter effectively too, so if I can keep the trip alive I've removed the possibility of a team fast breaking off that miss, and vice versa if I get a defensive rebound, but because defensive rebounds are already so common the marginal offensive rebound is worth more than the marginal defensive rebound.

I also hoped we could try to back out an offensive/defensive rebound split, and that looks pretty possible. The relative increases in rebounding suggest that Raef got 36% of his rebounds on the offensive end, which is a little high for a contemporary center but the coder wasn't operating today. It will be interesting to see if the ratio works out to be smaller for the point guard, and if so by how much. We'll see!

[eric]

Atha Leticism

50 points of Strength was worth 65 team wins...
14 rebounds per 36 minutes per 20 years
14 assists
15 turnovers
219 uses
more fga/use and fta/use, less tov/use
250 minutes per year
slight increase in pts/tsa, def/pf
slight increase in team pace

Strength increases really everything. This isn't a shock to anyone, it's a really really really good attribute even when the player is a lousy scorer and passer to start with. More on that in a minute.

50 points of Jumping was worth 7 team wins...
13 assists per 36 minutes per 20 years
10 turnovers
103 uses
more fga/use, less tov/use
increase in ast/tov

Jumping doesn't increase efficiency, so adding on uses to a bad player doesn't help the team very much. People are skeptical of this attribute, but it has its place. Inside Scoring and Strength increase usage the most, Jumping about half as much, Passing and Offensive Rebounding about an eighth of the top. If you really want your player to shoot more and can't or won't increase Inside Scoring and Strength, Jumping is by far and away your best option, and if you want a player to shoot more he's probably an efficient scorer and you'll get more than 7 team wins over 20 years.

.

Here's the full table.

tws characteristic
0.0 pound
2.5 inch
1.0 experience
1.5 inside scoring
0.7 jump shot
0.4 three shot
1.1 handling
0.0 quickness
0.2 passing
1.0 stealing
2.5 shot blocking
0.9 post defense
0.5 perimeter defense
0.3 drive defense
1.3 offensive rebounding
1.1 defensive rebounding
1.3 strength
0.1 jumping

2.5 shot blocking
1.5 inside scoring
1.3 offensive rebounding
1.3 strength
1.1 handling
1.1 defensive rebounding
1.0 stealing
0.9 post defense
0.7 jump shot
0.5 perimeter defense
0.4 three shot
0.3 drive defense
0.2 passing
0.1 jumping
0.0 quickness
By and large the attributes people like did well on this metric, and like Jumping you would expect Inside Scoring and Strength to do better for better players. Shot Blocking and Offensive Rebounding are underrated, Jump Shot and Post Defense overrated.

Edited to add, this table is team Win Shares per point of attribute per 20 years, or equivalently team Win Shares per SC of attribute per 4 years.

Next up... training camps!