Posted by Neil Paine on February 18, 2010
Two summers ago, Doug made a great post over at PFR that asked a simple question: Who is the current Dave Duerson? If you don't know who Duerson was, he was a good (4-time Pro Bowl) but not great (he won't make the Hall of Fame) safety for the Bears in the mid-1980s; seeing as I was born in December '85, I only know who he was because I've watched tapes of the 1985 Patriots and he was one of the Chicago defenders in that regrettable game that ended the Patties' season. Anyway, the point of the question was to find a bridge that related Doug's frame of reference (he was a huge sports fan by the time '85 rolled around) to that of someone my age who didn't compulsively watch old Patriots tapes and therefore wouldn't have any way of knowing who a semi-notable SS from 1985 was.
To that end, Doug built a simple similarity system using Approximate Value that would find players with careers of similar shape and quality -- this wouldn't find players who played a similar style to Duerson, of course, but it could give you a broad idea of how good Duerson was and how he fit in the world of football in 1985 by finding players who were basically as good as Duerson that you have actually heard of. So for someone my age, Duerson is Merton Hanks, he's Mike Brown, he's Blaine Bishop -- guys who made a couple pro Bowls, pretty good players but not all-timers. And as a result, everyone now has a common frame of reference.
It's a terrific concept, and it's easy to pull off in any sport that has an all-in-one, single-number value type of metric. Luckily for us at BBR, we happen to have just such a metric: Win Shares. Now, in Doug's method he started with 100 "similarity points" and subtracted one similarity point for each point of difference between 2 players' best-season AVs, .95 of a similarity point for each point of difference in their second-best-season AVs, .90 of a similarity point for each point of difference in their third-best-season AVs, and so on, which has the effect of really emphasizing that 2 players have peaks of similar quality. But I want to find players whose entire careers look the same year-by-year -- their first seasons are similar, their 5th seasons are similar, their 12th seasons are similar, etc. To accomplish this, I used least squares to measure how similar a player's WS were to another at every season of their careers, and minimized the sum of squared errors. Here's an example, Kobe Bryant through 14 NBA seasons (2010 WS were pro-rated to 82 games):
So among guards (or G-F), whose careers end up looking similar to Kobe's through 14 pro seasons?
Here are Bryant and Payton side-by-side:
Yes, the numbers came at different ages, and Bryant is clearly better, but it does give a nice idea of how the two fit together in the history of the NBA -- you could explain to someone who didn't know GP in his prime that Payton was to PGs what a slightly lesser version of Kobe Bryant would be to SGs.
It's even more useful for past players you may not know... for instance, who was Guy Rodgers?
|Win Shares by Season#|
Here are some other older players and their more modern equivalents:
Calvin Murphy = Andre Miller, Stephon Marbury
Darrall Imhoff = Will Perdue, Aaron Williams
Willie Naulls = Antoine Walker, Ken Norman
Phil Jackson = Malik Rose, Luc Longley
Dan Roundfield = Shareef Abdur-Rahim, Corey Maggette
Don Buse = Vinnie Johnson, David Wesley
Marques Johnson = Andrei Kirilenko, Anfernee Hardaway
Bob McAdoo = Grant Hill, Yao Ming
Sam Jones = Glen Rice, Jason Terry
Joe Caldwell = Travis Best, Bobby Jackson
I can do this for any player in NBA history, so let me know in the comments who you want to see the equivalents for, and I'll post them below...