Best way to evaluate players? How should we use stats?

[akira]

I've watched basketball for many years, but ever since I was young I was curious if there were statistics that made more sense for evaluating players than ppg, rpg, assists etc.

I'll admit I'm not really technically proficient in statistics (although I've taken basic stats, a couple years of Calculus) but learning about so called "advanced" stats such as BPM, RPM, WS, WP, RAPM, PT-PM, intrigues me because it tickles my desire to find truth.

So my question is really several-fold.

Is there a single stat that can explain how much a player has added or subtracted on average? Maybe a 50-50 blend like 538's CARMELO projections. If so, can that be used to explain who the "best" players are in the NBA or are stats like RPM better used to create a tier system? If not, what is the point of these all-in-one stats? How much of the equation should be "eye test" if you will? What should the role of analytics in decision making?

[Mike G]

You're close to suggesting a proposal that's often been discussed, yet it's never quite been done here, as far as I know: Make a blend of these stats, that best predicts team point differential, from one season to the next.

Adjust the weighting on each stat to arrive at the optimal recipe. You likely discard some of them, as their weight approaches zero.
Good luck! and welcome to the conversation.

[Nate]

...

Is there a single stat that can explain how much a player has added or subtracted on average? Maybe a 50-50 blend like 538's CARMELO projections...

There are many stats that "can explain" the size of a player's contribution. The problem is that they don't always agree with each other, so odds are some of them are incorrect. Once you've worked out a practical test for comparing the various stats, you can work backward to produce a stat to match that test.

People who claim to have the one true single number stat are lying. (Often they're lying to themselves.)

...What should the role of analytics in decision making?

Rather that starting with some preconceived 'magic number' stat that's supposed to be all things for all purposes, you're going to be much better off starting with the choice, and working the analytics in the context of that choice.