Ryoga Hibiki wrote:By definition, the moment you reduce thousands of data points to one number you are losing depth and context.
No matter what, you would need way more information to get to the ultimate goal, that is estimating the VALUE each player is creating.
The way it's designed, this statistic measures something very tangible, how many points each player is adding vs the average. Pay attention, I didn't normalize to pace but to points. And it's quite easy to obtain. The real question is what is your ultimate goal, what do you want to measure? How would you use this stat?
By itself I think it's nice because it measures something clear, intuitive and not ambiguous.
I'll show you why it doesn't make sense.
Two different interpretations of what you're suggesting;
(Data for examples to come;
The teams Kareem Abdul-Jabbar played on had 112.8 ppg average.
The teams Ray Allen played on had 98.3 ppg average.)
-Adjusting TS Add first, then calculating a new Gain/Loss%
Kareem's TS Add total was 4718.8. By your method it's 4183.3. Kareem lost 535.5 TS Add. Kareem's gain/loss% goes from 14.0% to 12.2%.
Allen's TS Add total was 2130.1. Now it's 2166.9. Allen gained 36.8 TS Add. Allen's gain/loss% goes from 9.5% to 9.7%.
- Adjusting Gain/Loss% directly;
Kareem's gain/loss% goes from 14.0% to 12.4%.
Allen's gain/loss% goes from 9.5% to 9.7%.
But Abdul-Jabbar scored bigger portion of his team's point totals than Allen considering Allen missed way more games than Abdul-Jabbar. (Their games played ratios are similar. Abdul-Jabbar 95.1%, Allen 96.6%.)
24.6/112.8 = 21.8%
18.9/98.3 = 19.2%
So, why would we take roughly 1.5% from Kareem for playing a higher scoring era/team while he contributed more? Especially when the efficiency was already calculated relative to the era?
It doesn't make sense to penalize high volume / high efficiency players.
Also, using league average would reward players on high scoring teams and penalize players on low scoring teams within that season.
If I used league average (108.9 ppg) instead of directly team average, Abdul-Jabbar's +14.0 gain/loss ratio would be reduced to 12.7% on interpretation 1 and 12.9% on interpretation 2. I already corrected your approach in that aspect for this particular example and it's still highly flawed with or without that correction.