Questions regarding "e-Wins"

[BasketDork]

Stop me if I've asked before, but Mike G., how exactly do you arrive at your e-Win totals ? I'm not looking to copy anything or attempting to improve upon it, I was just curious as to the steps taken to arrive at a total. I'm interested in the nuts and bolts, I could have sworn I've seen it on here explained rather thoroughly, but I can't locate said post or comment right now.

[Mike G]

eWins or 'equivalent wins' sprang from the notion that 'wins added' rates tend to be scaled to the arbitrary 48 minute NBA game, some 96 possessions per team. The per-96 win% is quite different from the per-possession number: A team that 'wins' 55% of its possessions might win 80% of its 40-minute games, or 90% of its 48-minute games.

Scaling to a 100-possession (or 48 minute) game has the effect of exaggerating the per-possession effectiveness of a player. So the objective was to neutralize that effect as much as practicable.

I had this statistical standardization measure for players that dated back to the 1980's in some form. Per 36 minutes seemed right for visualization purposes; points are scaled to 100 per team, and to player, team, and opponent shooting%. Rebounds are scaled to 1980's (and current) average of 44 per team per game. In general, there was decent correlation between a player's minutes (situational to team) and his weighted sum of counted stats, standardized. I put a T on that column.

In 2004, I decided to make my player T rates interactive with team point-differential. This made for a working relationship between team-adjusted player stats and that team's success on the scoreboard.
Feedback drives the stat weights, and they change from year to year, playoff series to playoff series. Most years, some additional factors are added, while others are tweaked.

This constant tinkering makes the stat unavailable to most of the people most of the time.

[Mike G]

I don't bother to show T Rate in the tables, as it's just an intermediate composite leading directly to eWins. As of Sunday,
eWins = (T-13.32)*Min/5347
T is a weighted sum of Sco, Reb, Ast, PF, Stl, TO, and Blk rates. Right now, the weights are:
Sco 1.00
Reb 1.00
Ast 1.33
PF -.24
Stl 1.37
TO -1.49
Blk .90
These are all very close to normal. Subtracting 13.32 from T gives the least error between player stats and team success. It's what I thing of 'replacement level T rate', creating on avg zero wins vs. avg NBA competition.
5347 is the Excel-chosen value that scales eWins to total games played.

[Mike G]

eWins de-exaggerates the possession-to-game extrapolation, by exactly 50%.
This is just how my old T rate converted to pythagorean wins; without any substantial change to the original stat weights.
There's been substantial fine tuning to some rates. Home and away assist and block rates are factored in. Estimated % of minutes vs opponent starters. A versatility index.

Player eWins don't sum to team expected wins. That relationship is:
xW = eW*2 - G/2

If after 82 games, a team's MOV suggests 61 wins, their eWins will total (in theory) 51.
A team of players totaling 31 eWins has xW = 21.
This is consistent with a good team being more than the sum of its parts, while a weak team is less than the sum of its talents.

Since an average player is 1.0 times as good as avg, and one NBA win requires on avg 484 player minutes, 1.0 eWins per 484 minutes is avg. This essentially moves the decimal point one spot from comparisons to WS/48, and it also saves a decimal place later.

[Crow]

I am not trying to stop you but you have asked before viewtopic.php?f=2&t=9021&hilit=eWins

I don't really understand the difficulty with search. Using the search tool I immediately found plenty / probably all the threads with eWins in the title. Here are a few more.

viewtopic.php?f=2&t=70&hilit=eWins

viewtopic.php?f=2&t=209&hilit=eWins

viewtopic.php?f=2&t=162&hilit=eWins

viewtopic.php?f=2&t=101&hilit=eWins

There are plenty more year to year that may have explanation. These were the ones focused on explanation in title.

[BasketDork]

Appreciate the thorough explanation, Mike ! As far as posting the question, I was in a discussion with a casual fan about this year's MVP race and where Westbrook, Harden, LeBron, KD fit in, they asked me if there was a was a way to estimate how many wins a player added to a team, to which I pointed to a few, with e-Wins being one, which requires some explaining, and doesn't have the quick Google link others do. Crow caught my laziness. It wasn't an inability to search the page effectively, it was just being lazy at the moment, and too wrapped up in the conversation I was in to take the time. I think this explanation was an even better one than what Crow pointed to me inquiring about last time. Thanks guys !