Ever so often I’ll make these kind of posts – this actually isn’t a post per se, I’m copying/pasting a couple responses I made on the APBRmetrics board – since it may answer questions I sometimes get on other boards, email, or Twitter. I have a habit of posting explanations of my work on boards – without ever doing the same here. It’s maybe ass-backwards, but I guess I’m sort of an ass-backwards kind of guy.
First is my response to a comment made by Dick Mays, the creator of Magic Metric – a linear weights metric that’s been around for quite a while. Here’s what he said, and how I responded:
Paul George is a great player. But I wouldn’t take him ahead of Love, Carmelo, Blake Griffin, or Dirk.
You could make a case for George ahead of Cousins, Westbrook, Hardin, or Alridge, without a debate from me.
And how I responded:
“ I love ya Mr. Mays, you are kinda one of the unknown pseudo pioneers of linear weights (at least where you actually published results online – while we ALL here just dabbled in it with our TI & HP calculators and notebooks in junior high and high school) – especially since you were the first I knew online who actually rated & ranked college players. I REALLY appreciated someone at the time TRYING to tackle such a tricky endeavor.
But, as you and all of us know, players are more than the sum of their stats. I believe I’ve gotten it (my ratings) to about as good as I can get based solely on box score stats of players and teams. I am so focused for now on these “basic” box score stats (despite a very recent onslaught of new measurables) because of my love of the history of the game and my desire to better quantify historical player impact in college and the pros, AND my desire to best learn what this history can tell us (projecting NBA careers based on previous dissected NBA and college ratings). My base ratings (HnR & HnI) do adjust for playing time in relation team quality and such – and obviously SoS in college – HnR penalizes guys for missing games (don’t help team when you don’t play), HnI doesn’t (how impactful are you WHEN you’re on the court). My base ratings will have results a little closer to yours and, say, PER. My HnI rankings actually have Love (well ahead), and Blake & Carmelo (barely) ahead of George, with Dirk 2 spots behind George . I think those are good rankings, especially in a linear weights sense, and they seem to be the type of rankings your opinions of players jives with for the most part. They are also how I do the college ratings – since I really have zero idea how to best incorporate WAR at the college level (what is “replacement” with such an EXTREME difference in talent from top to bottom).
BUT, I felt my NBA WAR took that final step closer to boosting (or bumping down) the guys who are probably doing a bit more (or less) than even the “normal” rating results (HnR & HnI) tell us. I personally subjectively think Paul George is probably somewhere like the 5th best player this season (behind Durant/James/Paul/Love) or 6th best (adding Curry, who I’m inclined not to because of his D). WAR has him 3rd (playing big minutes as easily best player on a team that wins a bunch) – WAR/48 5th (Paul & Curry w/ their lesser minutes jump him – if Minnesota would win anywhere near their projected W% based on point differentials – Love would jump him also). That seems right to me. I’m guessing also, in general, it would be a little closer to a blend of statistical +/- and RAPM – without having/using the +/- info.
Plus, WAR, in a historical sense, will rank guys like Bill Russell and Bob Cousey MUCH higher than any linear weights ratings. Russell may not be Kareem/Wilt status, but a bit closer than any other metrics I know of would probably have him (unless there’s a metric out there that ONLY considers minutes played in relation to team wins and championships – then you’d have a bunch of Celtics and a couple Bulls at the top). Cousey might not be close to Magic/Stockton/etc, but he won’t be immensely far off as he would by about any metric. That seems closer to “right” I think.
But, hey, NONE of us will ever agree on some true rankings of players, especially after about two or three players in (after LeBron/Durant/Paul). We just hope our results more often than not seem to make sense with what we (or, maybe the general public/coaches/gms) “see”.
I’m really glad you post on this board Dick, I like the general box score metrics debates (everyone is so into prior informed RAPM and such, which I have never felt “comfortable” with in terms of results and using multiple seasons) – I suggest you check Mike G’s posts if you haven’t been – his stuff will have a little similar results to yours, and he’s pretty proficient about getting the stuff posted here. Plus, he tries to tie all the historical rankings the best he can – I’m looking forward to seeing how my historical player rankings compare to his when I finally actually do mine – since I love that stuff.
Hey RAPM guys – don’t take it as a diss btw what I just posted – but since there isn’t quality +/- data pre 2000s in the NBA and pre 2010s in college (and even that raw data is often very flawed, missing games, etc) – it just isn’t my thing. If I was working for some team or something like some of you have or aspire to do I’d probably be delving deep into it also – obviously it’s maybe an important part of the analytics “puzzle” on players if done well. But, for what I’m trying to do at the moment, I don’t have time for it, that’s all.”
And, next, my response to Bob Chaikin, on his response to Mr. Mayes. By the way, Bob Chaikin created possibly the most well known NBA basketball game simulation based on real life results, in which you could change players playing time, teams, etc. – and rerun and see how it changed team results. I don’t think his stuff it out for the general public anymore, since I think he does work for some NBA team (or teams)? Not sure, doesn’t really matter here I guess:
I think having Paul George in your top five, ahead of Kevin Love, should raise an eyebrow! LOL.
paul george is a key reason, perhaps the key reason, why the pacers are – by far – the league’s best defensive team, along with hibbert, and why they have the league’s 2nd best W-L record. he leads the team in – by far – scoring, is an excellent defensive rebounder for a SF, an excellent man defender, and force turnovers at a solid rate. he generates wins at the 3rd best rate among SFs behind kevin durant and lebron james, and both durant and james as PFs generate wins at rates similar to kevin love…
i can easily see someone having paul george rated among the top 5 players in the league, and above kevin love, especially when you properly account for defense…
And my response:
“I’m trying – when my WAR/48 results differ quite a bit from general linear metrics results (PER?) – I usually assume it’s maybe individual defense. Guys like George/Aldridge/Howard/Duncan/Ibaka rank higher (or way higher) in WAR/48 than PER would have them – while others like Cousins & Thomas/Love/Carmelo/Irving/Gay rank lower than PER would suggest. It seems USUALLY the results end up being a group of guys that people think of as good/great defenders when WAR/48 loves them more, and a group of guys that people think of as maybe mediocre defenders when PER loves them more.
And, since I have people ask (DeanO specifically has questioned me when it came to my college ratings) – my WAR differs from WS mainly because my stuff doesn’t put as much assumption in thinking a high blk/stl/Dreb rate equates to great defense. It does some, but not nearly as much. For example, right now Deandre Jordan is 9th in WS, my WAR has him 31st. WS/48 has Pekovic 16th, WAR/48 has him 55th. So, yes, while a similar approach, my results can differ quite a bit from WS – and in my opinion usually for the better, or I wouldn’t bother with doing the work. My results don’t tend to be quite as defensive rebounding/shot blocking big man loving as WS – ESPECIALLY if that player plays for an underperforming team – there’s more of a general position “balance” without ANY subjective position “adjustments”. However, my WAR likes Duncan this season quite a bit more than WS, and a lot more than PER – go figure.”
Now, I mention DeanO above - for those who dont know, that’s Dean Oliver, writer of Basketball on Paper, and creator of Win Shares (the WS I reference above) and a bunch of other approaches to basketball analytics that guys like myself steal bits and pieces from. He’s now the Director of Sports Analytics for ESPN - and has always been very cool with me anytime I had a question (heck, we’re actually friends on Facebook – even though I never use facebook except for when my wife reminds me to log in and “like” my kids’ pics she’s posted). I HIGHLY recommend Basketball on Paper for ANYONE who thinks they want to delve into basketball analytics.
Well, there you go. I figure this may help some better understand how my work differs from some other linear metrics out there – and how/why I’m still trying to advance that type of approach in light of all the new shiny “advanced” stats that are FINALLY being compiled.