Here's Part Three of our discussion of advanced stats: If you've missed all or any part of Parts One and Two, you'll find them here and here. In part one, we covered Plus/Minus, On/Off, and Adjusted Plus Minus (APM). Steve Perrin, our stat guru, got very excited over that last one, we're still cleaning up the mess. In part two, we moved onto True Shooting Percentage, Effective Shooting Percentage (another of Steve's favorites) and Points Per Shot, where Steve threw Corey Maggette under the bus (only figuratively, Steve's not a particularly big guy and he would not have been foolish enough to try and throw the magnificent Corey under anything. He might have, perhaps, tried to coax Maggs under a bus, but I digress).
For the third part of our series, we'll be dipping our collective toes into some "estimated" stats. I am the Ignoramus, Steve Perrin is the smart guy. Here we go:
The Ignoramus: So, Steve, let's just skip the pleasantries, take off the gloves, crack our knuckles, sharpen our pencils, turn up the lights, damn the torpedoes, and get started. There are all these "estimated stats" I see everywhere: Rebound Percentages (Offensive, Defensive, Total), Assist percentages, Steal, Block, Turnover percentages, and my favorite, Usage percentage. First of all, these are all actually, percentage stats, but I've recently discovered that these percentages are based on estimates. But stat is short for statistic. Isn't the term "estimated stat" oxymoronic?
Steve Perrin: Is "estimated statistic" an oxymoron?
TI: Yeah, I think I just said that. Let's go over this again: I ask the question, you answer the question.
SP: It's such a good question, and, coming from you, so unexpected, that now I'm having trouble thinking of anything else. But this is important so I'm going to focus. If I ask you, who led the NBA in scoring a decade ago, during the 2001-2002 season, what would your answer be? You'd probably look it up and say Allen Iverson, who averaged 31.4 points per game. But Paul Pierce actually scored the most total points that year (2144) and in fact, Iverson was only fifth in total points scored - he missed 22 games that season. Still, we call Iverson the scoring champ because it's points per game that matters - Pierce and some others just had more games in which to score points. But wait. In his 60 games, Iverson averaged almost 44 minutes per game. Shaquille O'Neal played just over 36 minutes per game, and actually scored more than Iverson on a per minute basis (27.1 points per 36 minutes versus 25.9). So why wasn't Shaq the scoring champ that year? It's exactly the same situation isn't it? Pierce played more games in order to score more points than Iverson, but Iverson played more minutes in order to outscore Shaq.
In baseball, the home run champ for the season is the one who hits the most home runs. Doesn't matter how many games or how many at bats. In basketball, the focus is almost always on per game production. But just as per game is a more accurate measure of productivity than raw totals, per minute is a more accurate measure of productivity than per game. And guess what? You can take this even further. And, yes, eventually we're going to get to those percentage stats.
TI: That wasn't a yawn, I was just relaxing my jaw.
SP: Remember Paul Westhead's first season in Denver, 1990-1991? He installed his Loyola Marymount 'system' and the Nuggets led the NBA in scoring at almost 120 points per game. But they were a terrible team, winning only 20 games. How could they have been so unsuccessful if they led the league in scoring? Shouldn't the best offensive team in the league win more than 20 games? The answer of course is (a) they were terrible on defense and (b) they weren't the best offensive team in the league - not by a long shot. Just as a player can score more points in more games or in more minutes, a team can score more points in more possessions - a crucial concept in advanced stats. The 90-91 Nuggets played at a record pace, creating almost 114 possessions per game. But they actually weren't much good at scoring on those possessions: their offensive efficiency, which is their points divided by possessions times 100 (in other words, points per 100 possessions) was 21st in the league (and there were only 27 teams at the time).
So per game is better than total, per minute is better than per game, and per possession is better than per minute (and until someone thinks of something new, that's as good as it gets).
TI: La la la la...
SP: All of those percentage stats you see on basketball-reference use this per possession (or you might call it per opportunity) idea. Let's take rebounds. When does a player have an opportunity to get a rebound? Well, when he's in the game, right? But more to the point, you can only get a rebound when there is a missed shot, and the higher the pace of the game, the more shots (and probably misses) there will be. Rebound percentage is the percentage of available rebounds collected by the player - since there are 10 players on the floor at any one time, in principal, each person should get about 10% of them. Of course, power forwards and centers play closer to the basket, and so have an advantage and will naturally get more - but no one's going to get a majority of the available rebounds. Kevin Love was a complete monster rebounding last season, but his rebound% of 23.6% was only fourth best in the league; Reggie Evans, Hamed Haddadi and Marcus Camby all had slightly better rebound %s.
TI: Well, I knew Reggie Evans goes after a lot of balls. But Hamed Haddadi is better than Kevin Love?
SP: All of those other percent stats are similar - block percentage is the percent of shots a player blocked out of the shots he could have blocked, turnover percentage is the ratio of turnovers to total touches for the player, etc.
TI: Okay. But let's get back to the oxymoron thing. One of our Clips Nation contributors, Citizen Buddahfan (link), suggested that even something like number of possessions is actually a formula... and not based on a real calculation. I assumed number of possessions was a fact. Certainly one of those guys with the cool courtside seats behind the table, you know the guys watching TV (Do you think they get HBO on those things?) has a little clicker and is counting possessions. I mean why not?) And while I'm at it, what about the security guys in blazers with wires in their ears? Those guys who sit in the second row and stare up into the crowd? They're backs are turned! They're sitting in five-hundred dollar seats specifically so they can not watch the game! Actually, maybe they're the guys who are counting possessions, because they might as well be if just gonna use estimates instead of real stats. But I digress.)
SP: Ya think? Breathe Swami, breathe. Now, you have to bear something in mind - there's a difference between what 'someone' is counting and what is an official 'stat'. So no doubt some intern from every team in the league is counting possessions with a clicker (not to mention deflections and rotations and blown rotations and lots and lots of other things), but only official stats are available across the board for all games. Since possessions aren't official, they're not available - if you want to get League Pass and watch every game with a clicker, you can be the keeper of the true possession stat - but everyone else seems to be comfortable with just doing some math and coming up with something that is pretty accurate.
TI: You're right of course, I would be terrible at the clicker... anyway, that hand is holding a beer bottle or a wine glass, depending on my mood.
SP: Besides - what does a possession even mean? When you have the ball and shoot, and then get an offensive rebound, is that one long possession for your team, or two short ones? What about a flagrant 2 foul? Two free throws and the ball... one possession or 2? The simple fact is that the formula just removes the guess work and derives possessions from all the other stats that everyone agrees with. (By the way, offensive rebound is one long possession, flagrant 2 foul is 1.88 possessions - it was a trick question.)
TI: Okay. I'm actually beginning to understand; even if I put down my Martini, how can I click the clicker 1.88 times? I can't. So the estimate is actually more accurate than the machine. Hey, I get that! Finally, I understand something! Now, what hell is "usage" and why should I care?
SP: OK, now we're getting into the heart of the advanced stats debate.
TI: (And it only took us 5,000 words.)
SP: Usage percentage is simply the percentage of a team's possessions that an individual player uses... a possession is considered used by a field goal attempt, or a free throw (once again multiplied by a factor or .44) or a turnover. A player doesn't have to play big minutes to have a high usage rate - it only looks at what happens while the player is on the floor. The Clippers had four players last season - Blake Griffin, Eric Gordon, Chris Kaman and Baron Davis - with usage rates greater than 24%, which means that if they were on the floor at the same time, there'd be no touches for anyone else. Of course, they weren't all on the floor at the same time, or very rarely, since Kaman was hurt most of the time before Baron was traded.
TI: You just had to go there didn't you? Sheesh. Tell me, Sensai, is usage good or bad?
SP: Good question.
TI: That's two.
SP: Most of the ultra high usage players in the NBA are All Stars and All NBA, and most of the All Stars and All NBA players have very high usage rates. So does the usage rate make them All Stars, or do they have high usage rates because they are All Stars? And what does it mean when we see a player like Michael Beasley in the top 10 in usage? It makes sense that top players are going to have high usage rates - teams want their best players handling the ball, taking the shots. But a high usage rate more or less requires a lot of productivity also - and some high usage players don't happen to be highly efficient players - but I think that seems like a subject for next time.
TI: Oh, that's just great. I feel like you tied me to the railroad tracks and broke for lunch. But, you're coming back, right? Right? Steve? This blindfold's a little tight. Hey Steve? You're coming back, right?
Well, stay tuned for Part Four where we either will or will not be wrapping things up and talking about those infamous single number stats.
Here's the links from last time and some new ones: