# Nylon Calculus: Updating DRE with a few tweaks

Two years ago, I put together a metric for measuring player performance on a game by game basis. It is a linear weights metric called Daily RAPM Estimate or “DRE,” for short.

DRE is a metric like John Hollinger’s Game Score that can be used to get a rough estimate of a player’s bulk productivity in a given game. Unlike Game Score, DRE’s weights are derived empirically through regression analysis.

At the time I put DRE together, I knew very little about running regressions in R. I knew just enough to be dangerous. In fact, I knew so little, that I was unaware that you could weight a regression to emphasize some observations as against others. Well, I knew you could, I just didn’t know how. A while back, I learned how to do just that. So I re-ran the regression of the per 100 possession version of most simple box score metrics against 15-year Regularized Adjusted Plus-Minus weighing players who had more minutes played more heavily in the regression. I also opted to use 2-point and 3-point shot attempts broken out separately, rather than rolled up into the overall field goal attempt rubric so as to give a slight bump for floor-spacers.

The results of the updated regression look like so:

You can see that offensive rebounds grade out as not being statistically significant, but I included them in the metric because I wanted to give defensive rebounds more credit and splitting them out this way does that, unlike the prior version of DRE which included both rebound types under the umbrella of total rebounds, with each being given the same weight.

Additionally, personal fouls don’t grade out as being significant either but I included them also because in a linear weights metric it felt inappropriate to not have some sort of debit for fouls, given that we know that they are a negative thing for a player to tally.

Rounding the coefficients for simplicity, you get these weights for the new version of DRE:

If you’re using DRE in a way similar to Game Score, i.e. as a bulk production metric, I would recommend not using the -8.4 per 100 possession intercept, but if you want to use DRE as a statistical plus-minus fully on the +/- scale, setting your inputs to a per-possession basis and using the intercept is more appropriate.

I don’t envision making any additional changes to DRE, so this should be the last iteration. If you use DRE in the future, this is the version I would recommend.

For fun, here’s the top 10 in DRE per 100 possessions among players qualified for the MPG leaderboard (via basketball-reference).

This largely jives with the players at the top of the MVP leaderboard for most people, if not exactly in the same order as most would have it. It certainly passes the smell test, for me.

Next: Nylon Calculus -- Al Horford and the Celtics' defensive rebounding struggles

Hopefully this updated and improved formula will prove useful for everyone. Thanks!

Ed. Note: Kevin hosts our analytics podcast, Nothing But Nylon. You can subscribe to his newsletter at Patreon.