Today’s Fastbreak colleague Charley Rosen is an esteemed writer with a plethora of books to his credit. He coached under Phil Jackson. He was a head coach in the now-defunct CBA and for both men and women’s teams in college basketball. He has been in and around the game his entire life. He also hates stats.
He doesn’t hold back in his assessment, declaring unequivocally, “One of the most basic of the several cultural factors that are destroying the beauty of basketball-as-we-should-know-it is America’s general obsession with numbers…”
“Stats are destroying the beauty of basketball,” is a pretty strong statement. Rosen’s issue with them seems to boil down to this statement:
While statistics can certainly be valuable and even necessary in many applications, the only stats that count — and are trustworthy — in basketball can be read on the scoreboards. Indeed, the new fad of using metrics — and even the traditional numerologies — to evaluate players is bogus.
I would agree, in principle, that using stats solely to evaluate players is bogus. But Rosen is arguing:
To truly evaluate a player, simply watch him play over an extended number of games. And since there are 10 players and only one ball, the trick is to watch that player during the 80-to-90 percent of the game when he doesn’t have the ball.
And that, I have a problem with. I’ll grant that Rosen has the expertise to do that. I would also venture to say that most basketball fans don’t. I’ve seen too many people use the “eye test” to prove mutually exclusive positions. Beauty is in the eye of the beholder. I also don’t know how many analysts have the time to extensively watch every single player in the league.
Therefore, it might be fine to use the eye test to asses the strengths and weaknesses of one player (provided we have Rosen’s capacity for that), but how does that help when you’re analytically trying to compare him to everyone else? Who has the time to extensively scout 450 or more players?
Statistics allow us to get a ballpark determination in a couple of moments. The eye test doesn’t.
There’s also a danger that we can fall into the opposite end of the spectrum where we dismiss stats, metrics and their corresponding value out of hand. Stats are a valuable aid in player evaluation; they’re just not the only lens we should be looking through.
To understand the proper way to view them, we first have to distinguish between what stats do and don’t do.
What they don’t do is tell the whole picture. In very broad terms, stats tell us (to a point) what happened, but they don’t tell us why they happened. With some of the more advanced stats, that line gets a little more blurry, but none of them completely dismisses the need for “the eye test.”
The first of these is that the dangers of depending completely on subjective observation are just as ubiquitous, though. The greatest danger here is something called “confirmation bias” that Science Daily defines accordingly:
In psychology and cognitive science, confirmation bias (or confirmatory bias) is a tendency to search for or interpret information in a way that confirms one’s preconceptions, leading to statistical errors.
Confirmation bias is a phenomenon wherein decision makers have been shown to actively seek out and assign more weight to evidence that confirms their hypothesis, and ignore or underweigh evidence that could disconfirm their hypothesis.
As such, it can be thought of as a form of selection bias in collecting evidence.
Kobe Bryant’s “clutch” shooting is a classic example of confirmation bias at play. According to Basketball-Reference.com, since the 2000-01 season, in both the regular and postseason combined, he’s hit a whopping 53 shots to tie or take the lead with 24 seconds or less in the fourth quarter and overtime. That’s 53 instances of success.
His fans just remember the makes and would like to ignore that he’s missed 128 in the same situation. Bryant’s detractors would like to just focus on the clanks, saying they disavow the makes. They would like to ignore Bryant’s career 34.2 effective field goal percentage in those spots, which is almost exactly what the league average was last year.
Bottom line, Bryant is neither great nor horrid when it comes to game-winning shots, but the middle of the road is where you find almost no one taking the correct position. Why? Confirmation bias.
In essence, we see the thing that we expect to see. This bias is blind to expertise and is rarely deliberate. It’s human nature to seek out that which confirms what we already believe. That’s the first big danger of relying solely on the eye test. Bryant’s fans see what they want. His detractors focus on that which proves their position.
And the solution is stats because they show us what actually happened. Like it or lump it, there’s no getting around the fact that since 2000-01, Bryant is 53-181 with 18 treys when the game is on the line. Stats present a safeguard from observational biases because they show us both what we do and don’t “agree with.”
For better or worse, to support our position or even change it, we first need to what the facts are. The stats are the fundamental facts of the discussion. There’s a danger of misusing them, yes. But it’s just as fallacious to fail to use them because of that as it is to cherry pick them and misuse them.
Understanding of why things happen that stems from the eye test can never conflict with the stats because it’s still assessing what happened. But if a position starts contradicting what actually happened, it can’t be true, because the premise is flawed.
The other problem inherent in Mr. Rosen’s arguments is that of conflating the notion of traditional box score stats with advanced stats. Ironically, his observations about the flaws in the traditional stats are exactly what precipitated the rise of advanced ones.
Rosen criticizes the use of rebounds by saying, “Since poor shooting teams miss more shots, their big men have more opportunities to snatch offensive rebounds. Likewise do opposing bigs have more opportunities to rack up defensive rebounds.” But that’s why we track things like offensive rebound percentage and can tell you that Andre Drummond still comes out great.
Rosen anticipates that, though, asking rhetorically, “And what about the uncontested defensive rebounds corralled when the other team is hustling back to prevent a fast break? The corollary here is that rebound percentages are also meaningless.”
I respectfully disagree. Rebound percentages are not “meaningless”; they just aren’t all-telling. There’s a difference.
But the SportVU data, while not specifying why the rebounds are uncontested, does distinguish them from contested ones.
Since the data at NBA.com is so thorough, we can look not only at the percentage of rebounds a player grabs but also the percentage of those which are contested. And furthermore, we can figure out the percentage of the times he wins that particular battle.
The chart below shows both the percentage of missed shots a player secured while he was on the court (horizontal axis) and the percentage of battles he won (vertical axis):
As you can see, guys like DeAndre Jordan and Drummond excel in both areas. But where this helps with the type of thing that Rosen is concerned about is in distinguishing between someone like Anthony Davis (10.2 rebounds per game, 16.1 percent rebounding percentage) and Greg Monroe (10.2 rebounds and 17.9 percent).
But Davis wins 50.54 percent of his contested battles and Monroe only wins 42.0 percent of his. At the same time, we can’t equate Kawhi Leonard (43.64 winning percentage and 12.9 rebound percentage) with Rudy Gobert (44.8 percent and 20.7 percent).
Ergo, Rosen’s argument about not factoring in contested boards is valid, but the answer to that is not fewer stats; it’s a deeper dive into them. Part of the anti-stats rhetoric, in general, is that it points to a flaw in one stat, but that’s only a flaw if we look at a stat exclusive of everything else.
Rosen takes issue with points, suggesting that you can’t distinguish between those against teams in consequential games versus meaningless ones. But HoopsStats.com does just that. Rosen argues that assists don’t factor in free throws, but SportVU data does.
He argues that you can’t distinguish between an assist for a good shot and one where “the receiver scores after making a twisting, off-balance, fadeaway jumper with an astronomical degree-of-difficulty against a double-team.” Setting aside the fact that the receiver should be passing himself instead of taking that shot, is there a way that factors that in?
By looking at assist opportunities and the average number of points their teammates score on those, we can determine who’s doing the best job of setting up their teammates to score, as I demonstrated here.
Rosen points to numerous defensive factors that are impossible to measure, yet also argues against the notion of plus-minus stats, saying:
Plus and minus … Again, no consideration of what players are on the court, what the matchups are, if a game was a blowout or a nail-biter, nor how significant was the particular game where this misleading stat was recorded. Two cellar-dwelling teams with little at stake? Or a critical battle between two top contenders?”
Rosen is right. If someone is using a single game of plus-minus to make an argument (beyond that one game), they’re wrong to do so. However, this is another case of statisticians recognizing and addressing the issues Rosen points out.
There are numerous adjusted plus-minus stats that specifically attempt to address the very factors that Rosen complains are missing. ESPN’s Real Plus-Minus (RPM) is explained by Steve Ilardi:
[The] metric isolates the unique plus-minus impact of each NBA player by adjusting for the effects of each teammate, opposing player and coach. … The RPM model sifts through more than 230,000 possessions each NBA season to tease apart the “real” plus-minus effects attributable to each player, employing techniques similar to those used by scientific researchers when they need to model the effects of numerous variables at the same time.
Note that exactly what Rosen is worried about is what is taken care of here. Who the player is on the court with and against, who the opposition is, etc. So if that’s the concern, shouldn’t addressing it give it more value?
Now, it should be noted that RPM only tells us what a player does within his system and on his team. But this also doubles the irony of the complaint about the defenses. Rosen essentially argues that the problem with stats is they don’t tell you all the “hidden” things that happen — particularly on the defensive end.
But that’s what DRPM does. It might not tell you exactly what percentage of time a player trails his man on a baseline screen per se (one of the specifics Rosen mentions are omitted by stats), but it tells you the effect of making good decisions. Ostensibly, the reasoning behind Rosen’s list of things that aren’t measured is that they prevent opponents from scoring.
By extending that logic, a player who does those things will have a good DRPM, which is why a guy like Draymond Green led the NBA last year in it even though he’s not a box score hero. He does the right things, and those decisions ultimately show up on what Rosen argues is the most important stat of all: the scoreboard.
In a vacuum, Rosen is correct about many of his observations. The problem is that he seems so steeped in that opinion that he hasn’t done much to explore the newer stats, and thus is unaware that many of his issues aren’t only acknowledged by the statisticians, but even resolved.
And while there’s always a need for observational evaluation, there remains the need for the statistical. It not only safeguards us from our own innate biases, but it also helps us notice things we might not have otherwise recognized by just “watching” games.
Rosen argues that using stats is “dehumanizing” because it can’t account for the human spirit. But the important thing here is not about the human spirit; it’s human nature. And honest analysis requires we challenge our own concepts, challenging our observations with the hard edge of factual reality. Otherwise, we’ll never know when we’re wrong, and we’ll just entrench ourselves in what we already believe.
Perhaps Rosen has the experience he needs to avoid that pitfall. The average fan, and even most writers, don’t.
 By subtracting uncontested rebounds from rebound opportunities we can get an estimate of contested rebound opportunities (assuming that a player secures his uncontested rebound opportunities). Then by dividing the contested rebounds by contested rebound opportunities we can determine contested rebound winning percentage.