2013-14 Rookies

[Mike G]

Negative wins are not intentional, they are inevitable, they are a property of your ratings, everything I have posted is a property of you ratings (other than the 30 multiplier for margin production, which doesn't enter in to win production), and I hasten to add that these are really your wp48 values, not mine. There is no other equation you could use to properly show the win production of players given your eWins transformation for calculating team wins - you'd have to change your team wins transformation if you wanted to change the player wins values.

To reiterate, all I have done is take what your ratings do at a team level and bring it to a player level.

Well, the whole point of 'equivalent wins' by players is to separate them from the team effect. So you seem to be asking me to short circuit that process.

Indeed, the affine transformation you describe creates "player wins" which sum to a team's expected (pythagorean) win total.

But then, the ' margin per48 ' formula you describe gives me a zero value at 1.00 eW/484, as I'd have expected: You have to be above avg to outperform the avg opponent.
Somehow when you did it, your values are much different. The magnitudes are very small, and zero is around .50 e484

[v-zero]

My values are as separated from wins as your eWins values. The eWins values do not in fact represent what you think they do, the values I have posted are the only ones (given your equation) which represent players in terms of wins. I am no more adding in a team effect than you are.

The margin calculation is for point margin, not wins - I posted the values in wins. An average payer would have a wp48 of about 0.1, which translates to a margin value of 0, which is what you have observed.

Be sure to note that if the eWins value doesn't change when a player moves team then nor will the affine transformed eWins value.

[Mike G]

I've found a bit of time to investigate this proposed 'transformation' of eWins to project total player wins toward team total wins. Last year's best and worst teams, Oklahoma and Charlotte, in eWins, "transformed" eWins, and Win Shares (ala b-r.com) --

Code: Select all

. Thunders        eW    e484    TeW   Te484   WS   WS/480
Durant,Kevin     15.7   2.44   25.0   3.88   18.9   2.91
Westbrook,Russel 13.1   2.21   20.2   3.41   11.6   1.95
Ibaka,Serge       6.8   1.32    8.4   1.64    9.4   1.81
Sefolosha,Thabo   3.5    .75    2.3    .51    6.8   1.46
Martin,Kevin      4.2    .95    4.0    .90    7.0   1.57
Perkins,Kendrick  2.1    .52     .2    .05    2.5    .62
Collison,Nick     2.8    .87    2.4    .75    5.2   1.58
Jackson,Reggie    2.1   1.01    2.1   1.02    2.5   1.21
Thabeet,Hasheem   1.0    .61     .4    .22    2.2   1.34
Maynor,Eric        .2    .26    -.4   -.48     .0   -.03
Fisher,Derek       .0    .06    -.6   -.88     .3    .46
. totals         51.5          64.0          66.4   
                  

. Bobcats         eW    e484    TeW   Te484   WS   WS/480
Walker,Kemba      6.8   1.16    7.8   1.31    4.8    .80
Biyombo,Bismack   2.0    .44    -.5   -.12    1.4    .31
Henderson,Gerald  4.1    .93    3.8    .86    3.3    .73
Kidd-Gilchrist,M  3.4    .81    2.6    .61    2.1    .51
Sessions,Ramon    3.5   1.02    3.5   1.03    2.9    .84
Gordon,Ben        1.9    .58     .5    .16    -.6   -.18
Taylor,Jeffery     .7    .23   -1.7   -.53    1.2    .39
Mullens,Byron     2.4    .83    1.9    .65     .1    .05
Haywood,Brendan    .6    .26   -1.2   -.48     .1    .06
Mcroberts,Josh    1.5    .91    1.4    .82    1.5    .90
Adrien,Jeff       1.1    .74     .7    .47    1.0    .64
Warrick,Hakim      .6    .56     .1    .13    -.1   -.07
Williams,Reggie    .4    .53     .0    .05     .3    .35
Thomas,Tyrus       .3    .38    -.2   -.24    -.2   -.25
. totals         29.3          18.8          17.8   

I abbreviated the rosters to those with at least 300 minutes, so that may account for irregularities in the totals.
eWins suggests Kevin Durant is about 2 1/2 times as good as an average player. Win Shares suggests he's almost 3 times as good. eWins after 'transformation' suggests 4 times.

I think it's a stretch that anyone could be 3X as good as an avg player for a sustained interval. With one of the all-time great player-seasons, Durant approached a natural limit of sorts.
Win Shares gives Durant more credit and Westbrook less credit than what eWins gives. Westy is just the best of the rest.
eWins says he's a 2nd superstar, and Ibaka is the only other who's above average on the team.
WS says they're almost all above avg.

I'm comfortable with what eWins tells me. Not at all comfortable with the suggested transformation. It just doubles a player's difference from average: Durant isn't 1.44 times better than avg, he's 2.88 times better? Nah. This defeats the purpose.

eWins says Perkins was a bit better than Biyombo. WS says he was twice as good. 'Transformed' says they were both crap.
Earlier I noticed the Te484 were negative for players < .50 e484, and of course it's because difference from avg (1.00) is doubled.
Personally, I cringe to see a system which says players are getting lots of minutes and creating negative wins. Why are they getting minutes to create negative wins?

[v-zero]

That's just what your numbers say, don't ask me why. You can't use multiplicative comparisons for a linear rating, it doesn't make sense, the players are just linearly different, you should use additive comparisons.


This is what your numbers do when you transform them to wins using your transformation, it is nothing to do with me.

[Bobbofitos]

This is what your numbers do when you transform them to wins using your transformation, it is nothing to do with me.

No no, you're personally distorting them for comedic effect.

[Mike G]

It's OK; I think I'm done with 'my' transformation now.
It doesn't serve any purpose (that I can see) to either (a) exaggerate how good are players for good teams and how bad they are for bad teams; or (b) exaggerate how much better or worse than average a player is.

Also, if anyone wondered, I hate leading zeroes.

[Mike G]

Meanwhile, at 8.6% of the season, rookies have hardly shown any improvement.

Code: Select all

e82   per36 rates   tm   Min   Eff%   Sco   Reb   Ast   PF   Stl   TO   Blk  e484
8.9 Carter-Williams Phl   37  .481   16.4   5.3   6.3   2.3  2.6   2.1   .7  1.43
4.2  Wolters,Nate   Mil   33  .458    9.7   3.8   6.6   2.6   .7   1.0   .4   .75
4.1  Oladipo,Victor Orl   26  .469   16.0   5.1   4.0   3.7  2.2   4.0   .7   .94
3.5  Adams,Steven   Okl   20  .554   10.8  10.4   1.5   5.0   .8    .5  2.2  1.03
2.0  Mclemore,Ben   Sac   20  .541   15.6   4.2   1.0   4.0  1.1   1.6   .3   .59

1.7  Mekel,Gal      Dal   15  .485    9.8   3.5   7.6   2.2   .3   2.6   .3   .67
.8   Zeller,Cody    Cha   17  .449    9.4   6.6   1.0   4.9  1.4   1.9   .5   .29
.7   Gobert,Rudy    Uta   13  .318    3.3  12.9    .3   6.4   .9   2.7  2.1   .30
.6 Caldwell-Pope,Ke Det   14  .519   15.9    .6   1.7   4.1  1.2   1.1   .0   .38
.5 Christmas,Dionte Phx   10  .531   11.3   6.3    .7   3.1  1.2   1.1   .0   .43

.4  Franklin,Jamaal Mem    8  .500    7.8   9.9   1.2   3.8   .0    .0  1.2   .60
.3   Davies,Brandon Phl    9  .452    8.5   3.9   1.1   3.4  2.7    .9   .7   .28
.2   Goodwin,Archie Phx   11  .365    7.8   3.9   1.0   2.4   .9   2.4  2.3   .11
.2   Calathes,Nick  Mem   14  .308    4.4   3.7   6.9   4.2   .9   3.9   .5   .09
.1   Hardaway,Tim   NYK   19  .432   11.5   2.2   1.3   3.6   .0    .9   .0   .04

The rookie crop is now looking at 28 eWins on the year. In the East, 19 eW; just 9 in the whole wide West.

By position:
14.7 - PG - MCW, Wolters, Mekel
8.0 - SG - Oladipo, McLemore
4.1 - C - Adams, Gobert
1.2 - PF - Zeller, Davies
0.1 - SF - Datome ...

[v-zero]

It's OK; I think I'm done with 'my' transformation now.
It doesn't serve any purpose (that I can see) to either (a) exaggerate how good are players for good teams and how bad they are for bad teams; or (b) exaggerate how much better or worse than average a player is.

Also, if anyone wondered, I hate leading zeroes.

The whole purpose was to put your numbers on a meaningful scale (or two) which the rest of the internet/community has agreed are most useful and meaningful. Direct comparison to SPM and RAPM, and WS, rather than meaningless abstraction like PER.

Obviously I'm incredibly frustrated that I cannot turn you, but I guess messing with a man's metric is like messing with his Bible, even if all you're trying to do is make it legible to a wider audience.

[Mike G]

You do realize there's a serious transformation from +/- to alleged Player Wins.
The reason I call these kinds of PW (I hope that is generic) "exaggerated" and team-strength-dominated is that the rather arbitrary length of an NBA game determines the probability of a win.
In a 40 minute game, with fewer possessions, the average value of a possession (relative to the opponent) is less certain to determine the outcome of a game.

When a guy is going in for, say, 10 possessions, how relevant is it that the game will go 90-some possessions? If he's a .95 eWins/484 guy, he'll get the better of his opponent often enough, in theory. He's not a 33% winner vs NBA avg; he's more like a 45% winner.

If everyone else is trucking down the wrong road, or barking up the wrong tree, is that what you need to do also? How about an alternative to the way everyone else is doing it? What if that alternative looks and feels more right?

A playoff series is a sort of laboratory for eWins; I don't know if anyone else does it at this level. I just noticed basketball-reference now has 'Advanced Stats' for individual playoff series. But they don't have WS nor PER.

Unless I can see a good reason to redo a good metric, I'm disinclined to do so. The reason eWins came into being was a perceived need for a PW metric independent of team strength. You add up player eWins on a team and double the difference from .500, to reach team Wins. It's pretty cool, and very simple. What's not to love?

[v-zero]

Serious as in difficult? If so I don't agree. Moving between wins and margin (plus minus scale) is trivial.

I do understand the lack of love for metrics that rely on the length of the game, which is why I favour the plus minus scale, it does what I think you want to do with eWins in terms of being ignorant of game length.

My basic issue is that the eWins values cannot be interpreted well without transforming them, so why not remove that step?

Also I feel I should point out that if you can transform eWins to get back team wins then they are in no way independent of team wins.

[Mike G]

..if you can transform eWins to get back team wins then they are in no way independent of team wins.

A player's eWins tells us what the he did for a team. The team is just the sum of its players. A team full of pretty good players is a very good team.

The player's rate of producing eWins does not jump around when his team gets better or worse. The player didn't suddenly start playing better just because the team got better. He was in more winning games, of course. But his contributions were constant, and his eWins are constant.

Win Shares (as an example) attributes these fluctuations to Defensive improvements. There are no individual boxscore stats for team defense. Player performance appears to change radically and inexplicably from year to year, whenever team success changes. This is all to accommodate team wins as the necessary target for (sum of) player wins.

Without that allegedly necessary condition, player rates may be left alone; and still a team's success is determined by those player rates, through a fairly simple transformation.
Team win% is an exaggeration of player win%, which is an exaggeration of possession win%.

...the eWins values cannot be interpreted well without transforming them...

I see it exactly the opposite. Independent of team strength, players are directly comparable to one another.

[v-zero]

I feel like I'm being pranked here. I don't know what you think I'm doing to your numbers, but I'm not doing anything to them that changes the information within. Can you not see that if the eWins value doesn't change then neither does the transformed eWins value? In which case where is your issue? Seriously, one of us is building enormous straw men here, and I'm pretty sure it's not me, unless anybody can point out some error I have made?

[Mike G]

You haven't made an error in calculation. As Bobbo suggested, it's just a grotesque kind of transformation. eWins are fitted to player performance and team wins as described. Your adjustment makes players twice as good or twice as bad (relative to avg).

I don't know how one could adjust a stat like eWins, which interactively adjusts parameters to best fit team eWins to team Pythagorean Wins. I don't think stats could be weighted equally for all teams, to have player wins = team wins. It would necessarily involve a very heavy team adjustment for the best and worst teams. That's what WS does, and the results are unsatisfactory for many reasons.

...if the eWins value doesn't change then neither does the transformed eWins value?

That's correct. But eWins is the proper value, and transformed it's not.
I don't have any issue with eWins. It behaves as expected. You seem to have an issue, but I'm not sure what it is. You can't use a number like 1.50 eW/484 because you have to transform it to 2.00 ?

I don't believe players for a 60-win team are really 3 times as good as players for a 20-win team. They don't win 3/4 of their possessions, nor quarters, nor matchups. Just games. The game ends at an arbitrary time. We don't have to be beholden to that fact alone.

[v-zero]

He'll correct me if I'm wrong, but I believe Bobbo was being sarcastic.

Ok. I'm out. I have tackled every point you keep making and you have ignored that and reiterated things to which I have already responded, and I will not succumb to the signs of madness.

[Mike G]

It's too bad we aren't face to face, and we would say normal stuff like, "yes, I get that, but ..." Since I haven't really figured out your point, I can't really tell what needs more clarification.

If your Ast/TO ratio is unchanged, your (Ast/TO)^2 will also be unchanged. That doesn't make the latter more meaningful.
If I multiply TO/36 by 1.5 as an element in arriving at eWins, that doesn't imply when I intend to display TO/36, that I should multiply it by 1.5

...point out some error I have made?

Here's a logical error:

if you can transform eWins to get back team wins then they are in no way independent of team wins.

Not all players are with average teams. eWins describes equivalent performance with an average team.
Teams are above average or below average because their players are one or the other. Not vise versa.

Look again at the tables of last year's Thunder and Bobcats. Two sets of columns are terribly skewed to team W%. The other looks quite reasonable.

I really do hate to repeat myself unnecessarily -- but:

Team win% is an exaggeration of player win%, which is an exaggeration of possession win%.