James Harden: An advanced stats case study

I learn best by examining real world examples. A lot of teaching theory indicates that most people learn that way. Concerning a relatively complex subject like advanced basketball statistics, we could talk about formulas and algorithms and perceived strengths and weaknesses all day long, but for me at least, it takes a real world example to bring the point home.

Two of the more widely quoted statistics that attempt to distill a players contribution on the basketball court to a single number are John Hollinger's Player Efficiency Rating (PER) and David Berri's Wins Produced (WP). There's long been disagreement between devotees on both sides of the PER/WP debate, with Berri and Hollinger leveling pretty harsh criticism at each other from time to time. Long time readers of this blog know that I can find fault with both calculations, but that by and large I prefer PER and find WP to have some serious flaws.

But we'll leave that discussion for another time, and just look at the Case Study of James Harden for the time being. I think it does a terrific job of illustrating some significant differences between these two rating systems.

There is still over a third of the NBA season remaining, but because last season was shortened by the lockout and because Harden is receiving more minutes as a starter in Houston this season, he's actually played more total minutes already this season (2026) than he played all of last season (1946). The point here being that we have a very comparable sample size of data to use between the two seasons (about a 4% difference in total minutes played).

The tables above show what I believe to be the stats most germane to this discussion. All stats are taken from basketball-reference with the exception of the WP data, which come from The NBA Geek. It goes without saying that other statistics are taken into consideration in these calculations, but I hope you'll trust me when I say that they won't have a major impact on the calculations so I've omitted them for simplicity's sake. For instance Harden's steals per minute are up this season, but so are his turnovers per minute, so I'm calling that a wash.

Of course all of the individual stats are per 36 minutes. It's not news that Harden's scoring more points per game (almost 10 points more, from 16.8 to 26.1) because he's playing more and shooting more. Both PER and WP are rate adjusted stats that take minutes played into consideration, as any worthwhile stat does, so the numbers of interest to us here are the per minute numbers. I'm using per 36 because they're easiest to relate to.

I'm not however spending a lot of extra effort to take pace into consideration. The Rockets play at the fastest pace in the NBA this season, so Harden's per 36 numbers will be inflated versus those of a player on a team that plays at a slower pace. But the Thunder played at the sixth highest pace in the league last season, so it's not as if Harden has gone from a very slow team to a very fast team. His per 36 numbers are a bit inflated this season compared to last, but it won't make a big differences in these comparisons.

Looking at the numbers, Harden's rebounding is just a bit lower, his assists are up some, but the numbers are pretty small there. Those differences will move the needle a bit (the better assist numbers probably compensating for the slight dropoff in rebounds for a net gain), but not a lot.

The real story is in the shooting numbers. Before we get into them, a reminder of what true shooting percentage represents. The true shooting percentage (TS%) takes into consideration both three pointers made (since they are more valuable than two pointers) and free throws to come up with an infinitely more useful number than simple shooting percentage. Since TS% accounts both for the points scored from the line and the possessions used on trips to the line, it is a thorough accounting of how much a player scores on the possessions he uses. Harden remains a super efficient scorer this season even as his volume of shots has increased (from 11.6 per 36 to 16.3 per 36). But his efficiency was simply off the charts last season. A TS% of .660 is crazy good, and in the past ten seasons only a handful of non-centers have achieved that level of scoring efficiency.

It makes sense that Harden's efficiency has decreased some as his shot volume has increased. He's taking about 40% more shots per game in Houston than he took in Oklahoma City. Presumably those extra shots are of lower quality than the ones he took with the Thunder, so it stands to reason his efficiency would go down. A TS% of .600 is still amazingly good -- but it's about 9% worse than he was last season. He's shooting worse overall from the field, and he's shooting worse from three point range, both of which hurt his TS% of course. It helps his TS% that he's getting to the line more while making a high percentage, but it doesn't make up the difference for the decrease in his other shooting numbers.

So how do these statistical differences manifest themselves in PER and WP? Well his PER has improved from 21.1 to 23.3 while his WP/48 (that's wins produced per 48 minutes) has decreased from .263 to .214.

Interesting, right? Same player, two seasons, one metric says he's better, the other says he's worse. Note that these are not insignificant changes in these metrics. Among players with at least 1000 minutes played, Harden was the 26th best player in the NBA by PER last season, and he's ninth this season. Meanwhile, he has dropped from ninth in WP/48 last season to 16th this season. In other words, the two seasons have had an very similar though inverse effect on the two metrics.

Let's look at why this might be. Starting with PER, I'll begin by saying that the metric is probably a bit misnamed, or that the name is at least misleading in this case. I think it helps to think of it as a Player Effectiveness Rating as opposed to a Player Efficiency Rating. Because let's face it, there's not much of a case that James Harden has been more efficient this season than he was last season. He produced 1.32 points for every scoring chance he took last season, and he's scoring 1.2 points on those chances this season. He was more efficient with his opportunities last season, period.

BUT, it's not at all difficult to argue that he's been more effective for his team this season. A TS% of .600 is far above the league average. There is certainly a point of diminishing returns with extra shots, but Harden hasn't come close to reaching it. He's taking almost five more shots per 36 minutes this year than last year, but if he weren't taking those shots, someone else would. By taking more shots, he's clearly helping his team more, even if it means his personal scoring efficiency is decreasing. So by shooting more, he's become a more effective player and his PER has gone up. The goal of the game is to outscore your opponent after all, so while it's important to look at other contributions as well, putting points on the board is still paramount.

Meanwhile, WP/48 comes out looking a bit too ideological and purist in this particular case study. Yes, Harden was more efficient last season. But the highest TS% in the league only helps a team so much if the player rarely shoots. It's partly for this reason that players like Kenneth Faried (the second best player in the league by WP/48 last season) and Tyson Chandler (the fourth best) are so highly ranked by WP -- they shoot an incredibly high percentage, while also contributing in many other ways, especially rebounding. Faried and Chandler are great -- but top four in the league? PER (and conventional wisdom for that matter) would argue that their overall impact is limited by the fact that they just don't score a lot.

If you want to know whether you generally favor PER or WP as an accurate measurement of a player's overall impact in the NBA, look no further than this James Harden case study. If you prefer Houston James Harden, you're a PER man. If you prefer the version from last season in Oklahoma City, then WP is for you.