Mike G wrote:talkingpractice wrote:Our published values are not "player values", and never claimed to be...
"Individual Player Values" are not "player values", and never claimed to be? OK ...
LOL, touche. I meant to say "player rankings", and not player values. Sadly, as it was the source of some fun, having added data from the past few nights, Reggie has slipped to 17th and Lebron is now 12th, in the sense of 'rankings' (which again they are not).
We've gotten a bunch of q/a. I'm going to answer here quickly, and then we'll just update the site every few days going forward. We can often answer specific q/a on specific players not on the site (or on the methodology we're using) via email/Twitter, depending on who is asking.
- Added a brief description of the method to the site, and introduced our FORPM prior (a random forest based SPM). Its also below.
Individual Player Value (“IPV”) is a stabilized in-season RAPM model which uses a robust machine learning based SPM metric (“FORPM”) as a prior for RAPM. There is no previous season information used, to put it on par with other in-season metrics such as NPI RAPM, ASPM, or PER. The choice of a FORPM metric as prior (using an ensemble consisting of random forest regressions and gradient boosting), rather than a traditional SPM metric, was made in part to eliminate discretion in variable selection, with the goal of making IPV a “pure” metric. In addition, a properly specified FORPM model (fit to SOS-adjusted PD) performs much better out of sample than more plain vanilla regression-based models (especially with regard to ‘defense’). Due to not using any previous year info or an aging/experience curve, these values should be considered as descriptive more so than as predictive. The model is based on ‘basketball’, and not on ‘offense’ nor on ‘defense’, and as such there is only one coefficient for each player. This is again both for purity of the metric, and due to this approach being more predictive out of sample. Individual Player Values here are not meant to imply ‘player rankings’, nor are they meant to imply that they are the players value if he were to be traded, or have his role changed on his team.
- How it differs from what we posted last year: Last year's published results also used data from previous seasons. The goal for this year was for IPV to be 'pure' (like pure APM). So, IPV is simply a robust SPM used as a prior for RAPM, for purposes of stabilization. As such, it should be much more useful than plain vanilla NPI RAPM. We're calculating it and reporting it, but its not our 'absolute opinion on player value' anymore than in-season FG% is the 'absolute opinion on shooting efficiency' of the guy who first decided to divide FG by FGA.
- Expanded from top 10 to top 20 today because i like the guy at #20 today. Again, these are not rankings, perse.
- Someone that we do not know asked whether or not we like Mike G (I swear this was a true question). Simple answer is that I/we have learned a ton from his posts over many years, and continue to. And he seems like a nice guy. He appears to not be a very large proponent of practicing (and much prefers the game itself). If there are olive branches in basketball, consider one extended. And again, we've learned a ton from his posts here over the years, and I respect his opinions on the game quite a lot.
- How large of a sample do we need: Presumably less of a sample than is needed for NPI, and ofc way more of a sample than is needed for a traditional SPM or boxscore metric.
- To answer Crow's question from yesterday, the notable people not doing so well in RAPM thus far (ie net negative), but doing very well in the FORPM portion of the model (and I assume in any other sort of vanilla SPM or boxscore model): The King, Howard, Bledsoe, Drummond, Kawhi.
- As I said in another thread a few weeks ago, if there's a team out there without the internal capability for this (currently) that would like a copy of our IPV values (and the underlying components), then the first team to write us can have it all every morning via email (for all players, obviously), totally gratis. At the end of the season, we'd provide the supporting documentation to verify/calculate the results internally (also gratis). I doubt any team will take us up on that, but the offer is out there now. If a team is concerned re our other basketball businesses, then this can be done via a 3rd party (someone very known in the analytics community who we talk to every day but is not affiliated with us), and with no direct contact between us and the team. We have no interest in any gains/profit/benefit of any kind here other than that 'it would be cool' to send values to an NBA team each morning, to setup a currently non-analytics team with FORPM/RAPM/IPV, etc