Post by eric on Jun 15, 2018 17:57:55 GMT
*Kittensfish, actually
Here's how many bad/average/good players there are at each position in 2002 compared to 3045.
.
Now here's a lot more detail.
Win Shares are universally acclaimed as the best stat ever by posters on tmbsl named eric. League Win Shares per 48 minutes played tends to be right around .100 although there's nothing explicit in the formula to make that happen.
In 2002 there were 292 players to play at least 500 minutes. They generated a total of 1194 WS in 559745 MP, good for .102. Nice.
In 3045 there were 283 players to play at least 500 minutes. They generated a total of 1201 WS in 560277 MP, good for .103. Real smooth.
In 2002 Athansios Cathasach happened to generate .1000 WS/48, and Basketball Joe is funny, so I put it up there.
Now, technically the average WS/48 is not .100 because better players tend to play more minutes and worse players less. This is because we (to lesser and greater extents) don't set our rotations at random and we (again to lesser and greater extents) have an idea of who is good and who is bad. Similarly, talent is not distributed evenly at each position: in 5.0 the bigs are better, in 4.0 they are worse.
We can account for all of these factors and objectively define good and bad players with a z score, which is simply the standard deviation of WS/48 at each position in each year divided into the distance from each average. A player within one standard deviation from the average is average, a player one standard deviation above but not two is good, a player above two standard deviations is very good, and the same for bad and below. (I told you it was simple!)
A normal distribution (i.e. one caused by entirely random chance) will be distributed such that 68% are average, 27.5% will be evenly split between good and bad, and the remaining 4.5% will be evenly split between very good and very bad. A priori we don't really know whether our distribution will be skewed in either direction: while we play to win the games and so play and nurture players who are good, some of us also tank or otherwise rebuild and so employ players who are not good (i.e. rookies).
Here's how the overall distros look:
Both leagues are significantly non-normal: they are heavily skewed towards both extremes. GMs in both leagues compared to the model prefer above or below average players to average players, but they still employ mostly average players.
We also see that the skews are not the same. 2002 is slightly skewed towards below average players while 3046 towards above. I don't know if this difference is statistically significant. Another complication is that GMs are now able to stash prospects in G League, and we have seen multiple times already that players in G League can contribute significantly in the majors. Pending future data I would say the distributions are pretty much the same.
.
To close, here are the players who hit 3s:
Jock Landale
Dolph Schayes
Brain Winter
Gary Bossert
Neon Boudeaux
Firsto Picko
Cameron Reddish
Yante Maten
3046
Rashard Lewis
Joel Berry (-3)
Nenê Hilário (-3)
That's it!
Here's how many bad/average/good players there are at each position in 2002 compared to 3045.
.
Now here's a lot more detail.
Win Shares are universally acclaimed as the best stat ever by posters on tmbsl named eric. League Win Shares per 48 minutes played tends to be right around .100 although there's nothing explicit in the formula to make that happen.
In 2002 there were 292 players to play at least 500 minutes. They generated a total of 1194 WS in 559745 MP, good for .102. Nice.
In 3045 there were 283 players to play at least 500 minutes. They generated a total of 1201 WS in 560277 MP, good for .103. Real smooth.
In 2002 Athansios Cathasach happened to generate .1000 WS/48, and Basketball Joe is funny, so I put it up there.
Now, technically the average WS/48 is not .100 because better players tend to play more minutes and worse players less. This is because we (to lesser and greater extents) don't set our rotations at random and we (again to lesser and greater extents) have an idea of who is good and who is bad. Similarly, talent is not distributed evenly at each position: in 5.0 the bigs are better, in 4.0 they are worse.
We can account for all of these factors and objectively define good and bad players with a z score, which is simply the standard deviation of WS/48 at each position in each year divided into the distance from each average. A player within one standard deviation from the average is average, a player one standard deviation above but not two is good, a player above two standard deviations is very good, and the same for bad and below. (I told you it was simple!)
A normal distribution (i.e. one caused by entirely random chance) will be distributed such that 68% are average, 27.5% will be evenly split between good and bad, and the remaining 4.5% will be evenly split between very good and very bad. A priori we don't really know whether our distribution will be skewed in either direction: while we play to win the games and so play and nurture players who are good, some of us also tank or otherwise rebuild and so employ players who are not good (i.e. rookies).
Here's how the overall distros look:
Both leagues are significantly non-normal: they are heavily skewed towards both extremes. GMs in both leagues compared to the model prefer above or below average players to average players, but they still employ mostly average players.
We also see that the skews are not the same. 2002 is slightly skewed towards below average players while 3046 towards above. I don't know if this difference is statistically significant. Another complication is that GMs are now able to stash prospects in G League, and we have seen multiple times already that players in G League can contribute significantly in the majors. Pending future data I would say the distributions are pretty much the same.
.
To close, here are the players who hit 3s:
Jock Landale
Dolph Schayes
Brain Winter
Gary Bossert
Neon Boudeaux
Firsto Picko
Cameron Reddish
Yante Maten
3046
Rashard Lewis
Joel Berry (-3)
Nenê Hilário (-3)
That's it!
