Dr Spaceman wrote:It's the best stat currently in existence.
Some people will scoff at this, but just think on this: what are we actually trying to evaluate when we look at a basketball player? Answer: How well does he help his team win? (Can be restated as: what effect does he have on the scoring margin?) The reason I hold RAPM in such high regard is because it is literally the only stat that actually attempts to answer this question. Any box score stat you can think of doesn't even try. Quite literally, RAPM is the only stat that has any validity for what people are actually looking for in a stat, even if they don't quite realize it.
There's nothing inherently wrong with using the box score, as long as you realize that it is, at best, a proxy for what you actually want to know. There can be great players who score 20+ ppg, and terrible ones who do the same. But there will never ever, by definition, be a terrible player who makes a hugely positive impact on his team.
Once I came to realize this, I became a big RAPM convert, and I live with the flaws because it's the only thing that can actually tell me what I want to know. It might miss the mark by more than the box score will, but at least the mark in this case is clear, and it's exactly what I want it to be.
So really this is the whole reliability vs. validity issue. People dislike RAPM generally because it challenges what they thought they knew. The fallacy is in thinking that you ever knew anything by looking at the box score anyway. So yeah, in small samples we have some results that are absolutely nuts. Granted. But with big samples, and enough noise correction, we zero in on exactly what we really want to know. That's beautiful, and it's something no other stat in existence can accomplish in the slightest.
Now I'll add the caveat that RAPM never thinks for me. I would never use it to rank players, or the crux of an argument, or anything of the sort. But as you guys know, I watch a ton of film, and generally find that what I see lines up with what plus/minus data shows. Obviously Manu Ginobili isn't the best player in the league, but when he comes in the Spurs play really well, and that's what RAPM tells us. Now Kevin Durant has a lower RAPM, but a much huger role and more minutes, so quite obviously he's the better player. I think sometimes people who use RAPM get pisgeonholed into using it as the be-all-end-all, and that's not what I do at all.
Final thing: RAPM is at a crossroads right now, and it's either going to head into RPM (wrong direction) or PTPM (right direction). This has everything to do with how I feel about the box score.
It must be comforting to believe that critics of the strong RAPM hypothesis oppose it because it “challenges what they thought they knew” or because people are unable to “realize it” contains the truth that they are looking for. It must be bothersome to believe that NBA teams are run by idiots who are leaving titles on the table with their box score fetish. That is shown by the fact box score stats still impact player salaries.
I am going to quote and then summarize the critical portions of this post:
what are we actually trying to evaluate when we look at a basketball player? Answer: How well does he help his team win? . . . with big samples, and enough noise correction, we zero in on exactly what we really want to know. . . RAPM never thinks for me. I would never use it to rank players . . . as you guys know, I watch a ton of film . . . RAPM is at a crossroads right now, and it's either going to head into RPM (wrong direction) or PTPM (right direction). This has everything to do with how I feel about the box score.
The argument being made here with regards to player evaluation is as follows:
I) on/off stats with a sufficient sample size can “zero in” on exactly how much a player helps his team win
ii) Incorporating a box score stat into on/off stats is the addition of arsenic into a cake.
iii) One should still watch games to determine player value while using on/off stats.
The distinction between the box score and usage of game footage collapses upon careful examination.
The box score is best understood as the first attempt to record events that occur on the court. The original box score was limited to such matters as points, rebounds, assists, etc. It did not contain until the seventies many pieces of information that we take for granted today such as:
Offensive Rebounds
Defensive Rebounds
Turnovers
Steals
Blocks
Box score stats are counting stats.
The video tracking stats are an attempt to record additional information that was not included in the current box score. The video tracking stats record such information as player shooting percentages on various spots on the court or their effectiveness at contesting shots. Video tracking stats are counting stats.
While it is not currently recognized they belong to the same category of stats as the traditional box score eventually it will occur.
The only important difference than between video tracking stats and traditional box score stats is that the new video tracking stats are generally automated while the traditional box score stats are recorded by human beings.
To incorporate the video tracking stats in your analysis but to discard the traditional box score means either (i) the original box score categories have no value but the video tracking stats categories do or (ii) the human error is so substantial that it cannot be trusted. The first argument hits me as bizarre. If you actually hold that view I would welcome an explanation. The second argument is more tenable but still weak.
The NBA during its only days was a minor league. The professionalism of the traditional box score keepers could be questioned. That really isn’t the case anymore due to greater scrutiny from media and millions of fans around the globe. While occasionally there will be mistakes in the box score there are no regular, extreme errors in recent years that justify not trusting the information in the box score.
When you utilize game footage in the evaluation of players it is likely that you are watching the game to make records of actions taken by that player on the court to illuminate what it is they do and whether it has value. If you are more ambitious and have the time you may begin keeping detailed records of what occurs on the court. ElGee did as part of his Opportunities Created stat. I have attached a webarchive link to an old Opportunities Created stat.
http://web.archive.org/web/201111262117 ... ted-value/
If you notice while there is a difference between what is being recorded it is still the counting of events that occur on the court. As an image it looks the traditional box score which makes sense because it belongs to the same category.
The scouting of players, whether in person or through footage, is no different at its core than the traditional box score. When you utilize this information you are not doing something different than the box score but rather you are attempting to count different pieces of information. Thus your hostility to RPM cannot be square with your view that you would never let RAPM do your thinking for you.
In conclusion, you should not use game footage when determining player values with a sufficient on/off sample size if you believe on/off stats can zero in on player value. If you don’t believe on/off stats can zero in on player value you should have no problem with the incorporation of box score stats or other counting stats into on/off stats provided that they can be shown to have value. Hybrid stats actually out performed pure on/off stats in predicting team performance. Until that is no longer the case, which I suspect will be never, the box score should be utilized by intelligent followers of the game. The best method of determining player value is using the best performing hybrid stats, along with whatever information you can discern from watching games.
