Lineup performance model?

Home for all your discussion of basketball statistical analysis.
Post Reply
Posts: 6250
Joined: Thu Apr 14, 2011 11:10 pm

Lineup performance model?

Post by Crow » Sun Jun 03, 2018 2:29 pm

Has anyone ever published a lineup performance model? My first thought is that I that I haven't seen one that isn't linear and based on average season (non-lineup based) performance and that includes simulations with the possible exception of a simple adjustment to fit usage to the 100% available in any particular lineup.

What combination of weights of player efg% for season, career, that lineup or similar lineups best predict individual and summed lineup efg%? How much does the model vary by team, team type, lineup, lineup type and size? Is it better to predict team efg% by itself or in the context of all of the 4 factors and shot distance?

Is there anything like this in the academic literature or conference annuals? If not, why not? Apply machine learning, simulation, game theory, what else?

Are "large" minute lineups subject to less variation over small, medium and "large" new game use sample sizes than dink lineups? Does performance behavior by lineup size suggest a rational pattern for when to use one or other in general or for specific scoreboard contexts? How much have teams done or not done on this and how much do coaches use it?

Defense ideally would be built into the model. The usage component would ideally be more sophisticated than brute fitting. Looking at creation and pass rates and using a simulator for directing usage rates might be the way to go (and some do it) but what happens in reality by lineup (and personalities and coaching direction) versus the simple dynamic (avg. based) model?

Anyone have feedback on league practice or model construction?

Does any other sport analytic community have a lead on lineup or matchup based modeling over basketball? How far is anybody? Is this still lightly or poorly explored territory? Would "true skill" estimates help or hurt model performance? Has anyone done anything with that?

If this is being done, how is it that no one talked about it before going inside? Some people are more secretive to protect proprietary leads but lots of other types of analysis saw giveaways or least discussion. Has this topic been seen as too speculative to be done accurately? That could explain a lot of casual coaching behavior. But is that attitude "right" or too "lazy"? If this isn't being done or done with great rigor, what is raising the issue worth?

Thinking further, there are gamblers and gambling syndicates with performance models. Some are probably lineup based and rigorous. Do they have the best models? Maybe. Probably? Has any of them offered to sell access to their model? Maybe though probably not financially rationale. Any team try to encourage irrational access? Maybe. Probably should try harder unless their employees are on it and as capable. Has the league tried to identify talent, get low cost insight on this with their coding / analytic challenge weekend contest (forget the exact name)? Anybody want to start such a project, solo or collaboration?

Post Reply