Advanced NBA Stats vs. the Eye Test

Advanced stats have taken over NBA analysis. While PER and even Win Shares have been around for a while, new metrics such as RPM, adjusted +/-, and more have become increasingly prominent in recent seasons. As these numbers have spread in influence, their proponents have come into conflict with more traditional fans who use basic stats (points, rebounds, etc.) to analyze the game, if they use numbers at all. This war between “eye test” supporters and “numbers” guys shows no signs of stopping. However, the real problem is that nobody wants to find a middle ground.

Look, ignoring all advanced metrics in favor of eyeballing it when discussing basketball is just not a wise strategy in 2018. While watching games is incredibly important (more on that in a bit), the whole point of advanced numbers is that they are at work every minute of action. They therefore are broader and more knowing than one person’s lone observations. Nobody can watch all 82 games of all 30 teams in real time, but stats can. This ability makes them valuable… when used correctly.

The big issue with advanced metrics is improper usage, or incorrect interpretation. Reciting numbers by rote without any context or other kind of analysis is even more futile than ignoring them entirely. These stats are useful, and are far better than what was available 10 or 15 years ago, but none of them so are infallible as to start or end an argument on their own. Arguing that Otto Porter Jr. is better than LeBron James or Kevin Durant because he had a higher RPM than them last season is foolish. Not only is it a bad basketball take (no offense to Porter, who is in fact very good), but it’s also based on a poor reading of what RPM is and what it means (it’s not really meant to rank players).

My favorite way to use advanced metrics is as a jumping-off point for further inquiry/research. For example, CJ McCollum had an RPM of -0.33 last season, ranking him 26^th among NBA shooting guards. Since I know McCollum wasn’t a negative player for the Blazers, I’d instead look for reasons as to why that might be the case. A quick look on NBA.com’s lineup data for the Blazers shows that several lineups with CJ weren’t all that great, while a couple without him performed quite well. So maybe it wasn’t necessarily that CJ was bad, but that his replacements were good, or that the Blazers just happened to have some success without him last season through no fault of his own. Hours and hours of research later, involving both study of stats and analysis, and a compelling argument might be drawn up one way or another. Basically, while numbers can’t just be shrugged off (unless there’s a very strong reason why), they can’t be taken as law without backup either.

An argument took place on Twitter for several hours today about whether current Rudy Gobert is better than prime Dwight Howard. Shockingly, many advanced numbers indicate this is, in fact, the case. Personally, having extensively watched both Dwight in 2009 and Gobert in 2018, there’s no question as to who I’d rather have: Dwight Howard will be a 1^st ballot Hall of Fame inductee for a reason, and it’s because he was a top 3 player in his prime. Gobert is a phenomenal defensive player who brings some things t o the table offensively, but I just can’t get behind his being more impactful than Dwight.

Rather than regarding all of the stats as suspect merely for suggesting such things, however, it’s again wise to take into consideration why they might indicate Gobert is better. Gobert is a more efficient scorer than Howard was, true. But that’s because he uses a lot less possessions than Howard did, and takes much easier looks on average. His defense, too, might rate better by some measures, but he also had far superior teammates on that end than Howard ever had during his prime in Orlando. These are just two easy explanations why Gobert might be favored over Howard by some advanced stats.

Basketball is contextual. Five players on each team take the court at once, all of whom are constantly in the middle of some action or other. It’s nearly impossible to break down statistically how much value each player contributes to their team’s success. Similarly, with so much action taking place, it’s equally difficult for one person watching a game to separate one player’s impact separated from all that his teammates are doing to help or hinder him. Blending both styles of analysis together is, for right now, the best that we can do.