Nerdnumbers avatar

The Boxscore Geeks Show: The Analytical Ari Caroline

The amazing Ari Caroline(@aricaroline), Chief Analytics Officer at Memorial Sloan-Kettering Cancer Center hops on the show to talk analytics, how healthcare and sports differ in data, and, of course, the draft!

As always, thanks to our amazing producer Brian Foster (@boxscorebrian)!

This Week's Poll

In discussing the stats, Ari brings up a player the eye test says should be good, but the numbers say is average. We delve into the stats and come away with a little optimism. Players have a lot of room for growth in the NBA, so what do you think?

Analytics: Healthcare vs. Sports

I've written about healthcare before. Of course, Ari's first question was why I didn't bug him first. He's said a thing or two about this himself.

Ari's work at Memorial Sloan-Kettering is unique. Having an analytics squad with access to as much big data on cancer research makes for attacking fascinating problems.

One key difference between sports and healthcare is how structured the data is. In sports, we have clearly defined tabular data, even for the new stats. In healthcare, you may have to infer a simple yes or no question from doctors' notes.

What NBA Player Qualities to Look For?

Here's a glimpse into Ari's mind. Here's his "draft board" for the top 100 Draft Express prospects.

A little background. Ari, like Dave, has developed a regression for analyzing the value of player boxscore stats. These line up very closely with the Wins Produced coefficients. A slight change Ari makes is to use a "continuous position curve", that is based on a player's skills. For the ultimate head scratcher: Joakim Noah is considered a small forward by Ari's measure based on the skills he brings to the table. This is not as surprising as you might think.

Using Ari's coefficients, we see their per 40 minute college numbers, and as a result, who teams should look at. As we discuss, this chart is by no means the final word. It's just a good starting point for teams to look.

The problem we notice is teams don't start here. They look at other factors that are less predictive of future NBA success. As Ari has pointed out, teams have a blindside to many stats. Oddly, they do seem to get how point guards should play.

"Can players learn to play better?" is a question we discuss. While this can happen, it's not as common as people think. Of course, this is referring to after players have been in the league awhile (see Westbrook and Melo). Incoming talent may have hope. For example, Michael Carter-Williams did not look that great in Ari's numbers last year. But, he showed some flashes of greatness this year. Who knows where he'll be by his contract extension.

In terms of "learning" to play in the NBA, Ari brings up Nik Stauskas. On Michigan he just could not get any steals. Is that a quality he might learn in the NBA? Certainly things like his shot selection and shooting efficiency look promising.

We drilled down the rabbit hole on advanced stats. For instance, should players be graded against their position for shooting percentage if they take a lot of shots? For instance, Russell Westbrook shoots above average for a point guard. However, he takes many more shots than the average point guard. Should those shots be "penalized" in evaluating Westbrook. On this subject, Ari has done work looking at the value of "stretching the floor" based on different position types.

Also, we discuss the position issue. Wins Produced will look at what the player is assigned. However, should we penalize players if a coach plays them incorrectly? Dirk Nowitzki is often used as a power forward. Of course, his skill set is such that it should be easy to play him next to a power forward more regularly.

Adreian Payne seems like a Kevin Love type player. Except, he just can't grab offensive boards. Ari's position adjustment still has him as a power forward. However, he's close to average because of his offensive boards. That definitely makes you wonder how Love can shoot so many threes and still crash the glass so well.

One note on all the players and analytics we discuss is how thin the margin for greatness is. A 60% True Shooting player compared to a 55% True Shooting Player is a big difference. However, that's only five baskets (give or take depending on their three point shooting) every hundred shots. That's really hard to notice. That's why it's important to be able to explain why a metric say a player is good or bad.

Ari notes that despite his size, D.J. Stephens could be a great talent in the NBA. He played well in limited minutes for the Bucks last season. Despite being listed as a guard-forward, he plays like a power forward. Turns out that can work! (See Charles Barkley and Dennis Rodman)

Khem Birch looks to be the best draft gem, and James Young looks like this year's draft mine. Of course, we don't know yet!

Shout Outs!

Every week we will like to thank people in the sports stats realms. This week's shout outs go to

  • Kenny Pickett (@KennyPickett), who has the coolest job description ever! Thanks Kenny for for catching a trivia flub I made. It turns out the 1984 and 1986 Celtics had multiple Finals MVPs. They had just played for other teams. Thanks for the catch!
  • R.C. Buford. After a great finals win and a well deserved Executive of the Year, we still think Buford's been getting ignored for too long. We love ya Buford!
Fun stuff. I'd like to point out that wins produced handles positional shooting the way you want it to be handled, not the way you think it handles it; the regression coefficients on points and shot attempts are calculated before the position adjustment, which just moves the intercept around.

Think of it this way - the average small, shooting the positional average, is penalized under WP48 for shooting (before the position adjustment), since each shot is below the league average. When you adjust for position, that penalty gets washed out, since the average is a penalty. But if a given small shoots more at merely the positional average, they'll have a bigger penalty pre-adjustment, and a portion of that penalty will remain after the adjustment.

This is a big part of the reason why volume shooters tend to look bad under wins produced; unless you are really, really good at hitting your shots as a wing, you're better off passing to a big, and the model penalizes guards who don't.
Interesting note on high school players picking schools is that they might be considered rational by going to the top tournament teams but if they treat college for what its is a minor league showcase, they are choosing poorly. Take Kentucky for instance, if the top 5 players are attending there then as a top 10 player choosing that school is not ideal. So high school players out there, go to a school where you have an opportunity to showcase yourself( i.e. get tons of minutes). It would be interesting to see one of the best high school prospects go to Harvard
A draft live stream with Caroline, Berri, Galletti would be amazing. Love when you guys get Ari on.
You've actually got it a little backwards. Wins Produced will actually make high volume guards/wings look better. Because their efficiency is compared with their peers (other guards) and not bigs, they are not penalized for being worse than bigs. They are only penalized for being worse than their position.
We have the classic "pizza topping" problem with more than 2-3 of us. But we've been bugging Dave to try and get on with Arturo or Ari for draft stuff.
The one thing I will say is interesting about WP is that the negative penalties are quite more substantial than other metrics I've seen (not saying that its a bad thing) and high usage guards/wings are more susceptible to these penalty type stats (mFG, TO) than high usage bigs. There is no penalty for missed RB and centers generally shoot a higher percentage. I don't really have an issue with this for most people but it does create some oddities for guards/wings that play more like big men. Basically playing like a big gives you a big advantage in efficiency. So some wings that can't/won't shoot, and are general space cloggers when paired with a standard big, look really good, I'm thinking Jeremy Evans, but can't really be utilized due to spacing issues, etc.
Not quite. In fact, apologies, but you're off.

In terms of "missed rebounds" there's two factors at play. First, a player is compared to their position. So if a center is not rebounding at a normal rate, they are penalized, see Bargnani.

Second, yes bigs shoot more efficiently, but are again compared to average bigs. That means they need to maintain an efficiency AND rate of scoring similar to these players to be considered good.

As for can't be utilized due to spacing issues. The NBA has a history of players with those limitations (Dennis Rodman, Ben Wallace), who excelled. It's about team construction. Do you hire a cheap guard that can shoot threes and also get the 3X average high rebounding wing? Or do you say "We can't use that player on our team!"
> Wins Produced will actually make high volume
> guards/wings look better. Because their efficiency
> is compared with their peers (other guards) and not
> bigs, they are not penalized for being worse than
> bigs. They are only penalized for being worse than
> their position.

More field goal attempts only helps a player's raw win score if his points-per-attempt is better than the league average points-per-possession. Isn't position adjusted win score simply the position adjustment subtracted from the raw win score? If so, I don't see where the position's average field goal percentage matters.
Nope, it's straight position to position comparison. The combination of an average player's FGA and efficiency will factor into what that is.
> .. Nope, it's straight position to position comparison. ..

Wins Produced penalizes teams for missing shots and credits them making shots. If a team makes fewer shots (as might happen if the guards shoot more, and the centers less) then that team's Wins Produced is lower. Because the team's total wins produced is the total wins produced of the players, that means that shift has to be reflected in the individuals' wins produced somewhere.

Now the guard's WP could go up at the center's expense if the marginal shift is from a super-high percentage shot such as a center lay-up to a very high percentage shot like an open corner three because wins produced doesn't take contextual opportunity cost into account. However, if the guard is taking shots that have a poor chance of going in (as compared to league average for the sample used to calculate WP) , then the extra shots will hurt the player's WP. (There may also be a negative impact on the center's WP if the lost opportunity is a high percentage shot.)

GnoiXiaK - that's the advantage of using something akin to Ari's position curve to calculate the position adjustment. The intuition is that there are only so many 'big man' stats available on the floor, and players with games focusing on those stats should be compared against their peers, regardless of what their position is listed as. Space clogging guards who don't shoot much get identified as small forwards in his model, IIRC, which is a pretty substantial penalty.

Just as a matter of reference, the "points contributed" algorithm that I use here does not have any team adjustment.

The team adjustment in WP, for that matter, is small enough that my numbers match Dave's pretty closely with only a few exceptions.

I should probably have been clearer about the context. I had hoped to contribute to the discussion about Dre's claim that "...Wins Produced will actually make high volume guards/wings look better..."

Charges seem to be reasonably consistently recorded, so, here's some data - from 2010 through 2013. Enjoy:

Thanks Nate! That's awesome!

BTW, for anyone still paying attention to this thread, I should've also mentioned that Nik Stauskus doesn't rebound either. I was just trying to give steals as an example of a potentially teachable skill at the NBA level.
Where did you have had Delon Wright if he declared? I have him as the best guy in college last year.

Sign in to write a comment.