5 Comments
User's avatar
Charles Hardin's avatar

Learning a bit of bayesian modeling in school, and I was initially a bit confused on the reasoning for applying priors & bayesian thinking to player ratings but after reading it makes a lot of sense. This is brilliant.

Expand full comment
Evan Miyakawa's avatar

Thank you so much!

Expand full comment
Jeremy's avatar

Zach Edey appears in the all time list TWICE. I guess a lot of the top guys go straight to the NBA, but not a single one of the other guys turned in two Top 30 BPR performances?

So is it best all-time SEASONS/PEFORMANCES, then, and not best PLAYERS? Zach Edey is only one player.

I'm pretty sure.

Expand full comment
Ron's avatar

What makes this model Bayesian?

Expand full comment
Evan Miyakawa's avatar

This means two things:

1. The output of every model is not just a point estimate (single value), it's actually a probability distribution. So you can know not just the best guess at a player's value, but all the possible "true" values. All bayesian models also require prior distributions for each coefficient as well, which I mention in the article

2. The computation itself has to be done through some bayesian computation method. That's a bit more of a complicated subject, but if you do some googling you can learn more about that.

Expand full comment