APBRmetrics

I’ve come up with a way to estimate each player’s impact on every one of his teammates BoxScore statistics

Essentially, this is an extension of this, but instead of estimating each player’s impact - generally - on teammates, I can now estimate the impact for each existing pair

To come up with the estimates, each player pair turns into two variables in the regression
A_impacting_B, and
B_impacting_A
As you can imagine, that creates lots of columns in the matrix

At the same time, I create a predictive estimate of how much of a certain BoxScore statistic a player will accumulate, per possession

As the signal/noise is different for the player / pair_impact columns, different penalization parameters are used depending on column type. The two optimal lambdas are then found through crossvalidation

So far I’ve only computed estimates for cumulative DFS points (steals are worth 3 “points”, assists are worth 1.2 etc),
So I get an estimate for how many DFS points a player will accumulate for a possession, and how each teammate of that player would impact that number if they’re sharing time on the court

Results can be found here

Not surprisingly, DeMarcus Cousins "leads" the league in negative impact on a teammate, in this case Davis

Thanks, will look further later.

Have you checked how well the sum of players by box score minus interaction effects does to predict lineup results? Which does better on average, that or sum of player RPM or a blend or something else?

Any insuperable obstacle to take this down to the four factors? That would be very interesting.

Could you do something on the impacts of defensive players? Probably not by player but maybe by counterpart starter or deserve or by player type (of a modest number of types, say 3-10)?

The DFS weights I saw in past seemed a bit biased / weird to me. Are you comfortable with them? Any desire to modify?

Do you have commentary / analysis on the recent "luck-adjusted" trend for opponent shooting? Do you luck adjust in any way or have you considered doing so?

What does DFS mean?

Daily Fantasy Sports. You pick a roster and compete against others for cumulative best next game performance as opposed to originally all season.