It was little more than a footnote in Tuesday’s NBA news blotter, but the retirement of veteran forward Andrei Kirilenko caused longtime members of the basketball analytics community1 to cast our thoughts back over some fond memories. Kirilenko was one of the movement’s first underground stars, a player who, because of his statistical versatility and measurable on-court impact, deserved far more credit than he got from mainstream analysts.
To most observers, Kirilenko enjoyed a decent but ultimately forgettable NBA career, mostly with the Utah Jazz. He averaged 11.8 points, 5.5 rebounds and 2.7 assists per game — practically the textbook definition of “pedestrian numbers” — and although he was voted into one All-Star game and earned a handful of All-Defensive team nods, Kirilenko was never considered an elite player. (At one point he was the league’s sixth-highest-paid player, but the contract was also widely viewed as an albatross.) Tellingly, Kirilenko has less than a 1 percent probability of making the Hall of Fame in the estimation of Basketball-Reference.com’s algorithm, which is built to mimic human voting patterns. By conventional standards, Kirilenko was nothing special.
For statheads, though, he was a superstar, especially in the early stages of his career.
When Dan Rosenbaum,2 then an economics professor at University of North Carolina at Greensboro, first successfully reverse-engineered Wayne Winston and Jeff Sagarin’s proprietary Adjusted Plus/Minus (APM) metric in 2004, a few of the usual suspects topped his player rankings: reigning MVP Kevin Garnett and statistical legend Tracy McGrady. But third on the list was a surprise: Kirilenko, then a third-year player with modest traditional numbers. It was no fluke, either: Kirilenko showed up third overall in the combined 2005 and 2006 APM ratings computed by Rosenbaum disciple Dave Lewin, and Kirilenko peaked at second overall in Jeremias Engelmann’s Real Plus/Minus (RPM)3 ratings for 2006. During his best years, you could count on one hand the number of NBA players who more positively influenced their team’s performance than Kirilenko.
The mid-2000s would end up being the finest stretch of Kirilenko’s career, after which injuries and age4 eroded his ability. But even after his decline, Kirilenko maintained a solid standard of performance. In Engelmann’s giant 14-year RPM dataset (which encompasses the 2001 through 2014 seasons), Kirilenko ranked 17th overall across all seasons, and he ranked 13th overall during his career (2002-2015) according to Daniel Myers’ value over replacement player (VORP).5 Both placements far exceed what would be suggested by Kirilenko’s modest conventional numbers.
When Rosenbaum initially released his APM ratings 11 years ago, he regressed them against basic per-minute box score statistics in an attempt to figure out which types of players tend to help their teams the most while on the floor. Rosenbaum found that some of the most productive traits a player could possess were high rates of steals (no surprise to us here at FiveThirtyEight), blocks, assists, offensive rebounds and scoring efficiency, plus a hefty dose of all-around versatility.6 Not coincidentally, Kirilenko ranked in the 65th percentile or better in each of those categories for his career, including the 96th percentile in blocks, 90th percentile in steals and 88th percentile in true shooting percentage.
In other words, unlike Shane Battier, Michael Lewis’s famous “No-Stats All-Star,” Kirilenko was an Every-Stat All-Star. Stuffing the box score in many important categories, Kirilenko (along with fellow long, rangy small forward/power forward hybrids such as Shawn Marion and Gerald Wallace) helped usher in a new era of multidimensional forwards who could do everything — and, increasingly, had to, as new rules shifted the game to make teams more reliant on smaller, quicker players.
Maybe that contribution is enough to justify Hall of Fame consideration for Kirilenko, maybe not. But it certainly justifies giving him more respect and consideration than a player with his conventional numbers would normally seem to warrant.