Posted by Neil Paine on January 11, 2010
Last week, I attempted to replicate an old Bill James study on high-peak (since people apparently don't like the word "peaky") vs. consistent pitching aces, adapting it to basketball. The goal was to see whether a team would expect to win more championships over a 20-year span with a guy whose peak was fast and meteoric -- but whose decline was just as abrupt and total -- or a guy whose career slowly built to become a solid star, never contending for MVP but performing at a high level for a long time. To accomplish this, I took real-world data and built a composite player for each type of All-Star, each with the same number of career Win Shares, and then ran a Monte Carlo simulation of 10,000 careers, tallying how many times each player won a title when surrounded by average teammates. The results were that the consistent player won a ring slightly more often in his career, which contradicted James' findings in baseball, but the # of rings/season for each type was so negligibly different that there was an advantage of just 0.011 championships per 20-year career for Mr. Consistency, not enough to make any difference for real players.
However, there were a number of flaws with the first study, chief among which was the fact that I lumped all "peaky" players into one group and all "consistent" ones into another. In effect, I took a bunch of players whose careers peaked at different times, threw them together, and the peaks for one subgroup were cancelled out by the valleys of another -- I basically averaged the individuals' peaks into a consistent career arc for each type of player, which defeats the entire purpose of the exercise.
Luckily, I get a second chance at these things, and a number of users suggested that instead of taking real-world data like I did on Friday, I should create an exaggerated version of each player type and re-run the simulation. So I put together these fictional career arcs for each player type:
Again, the career totals were equal, 112 for each player, but this time the distributions are radically different and it is very easy to tell at a glance which player type is which. Notice as well that the High-Peak player doesn't even play out the final 4 years of the 20-season block; in those years, he was replaced by a bench-level player to punish the otherwise average team for having to find a replacement for their one-time star player.
Now, it was time to run the 10,000-season simulation again. With our new and improved model, would Mr. Consistency still cop the most NBA titles, or would the brief brillance of the High-Peak star be a better formula for postseason immortality?
|# of Careers||10000||10000|
|% of Years||1.3%||0.9%|
|Careers w/ Rings >= 1||2337||1613|
Well, that's quite a different result. Exaggerating the "peakiness" of the High-Peak player and the stability of the Consistent one, we see that having a very short stretch of essentially being the NBA's best player is more conducive to winning championships with a typical team than being a consistent lower-end All-Star. This supports what James found in baseball, which was basically that you needed to be better than just good for a long period of time to win a pennant -- you needed to be dominant in your best years, without much regard to how bad you were over the rest of your career.
Now, maybe I made the apex of Mr. Peaky's career a bit too amazing. I mean, very few players will ever reach 16 Win Shares in a season; in fact, it's only been done 62 times in the history of the NBA. Perhaps we should reduce the High-Peak star's best years to something more reasonable. Would that change the outcome?
As it turns out, no. Even if you reduce the peak to 10-11-12-11-10 WS instead of 11-14-16-14-11, the High-Peak star wins you a significantly larger number of championships over 20 years than his Consistent counterpart -- and that's all while being replaced with a bench scrub for the final four seasons of his career.
Given that the first study was both fatally flawed and showed very little in the way of conclusive results either way, I'm inclined to trust these results more, especially since they seem to dovetail with what you see in a cursory analysis of real-life championship results. Dominant players with ridiculous peaks own the Larry O'Brien Trophy, with all but 5 of the past 58 NBA champions being led by a player with double-digit Win Shares (one was Tim Duncan in '99, a season shortened to 50 games; had he played all 82, he was on pace for 14.3), more than half being led by 13+ win players, and a quarter being carried by stars with 15.8 or more WS. Given that the same six franchises (Boston, Chicago, Detroit, Houston, LA Lakers, and San Antonio) have controlled 28 of the past 30 NBA crowns, it also makes sense that it takes a truly epic peak performance to break through and capture a championship. Lower-tier All-Stars who can be counted on for solid production every year are nice, but it seems that your best chance for a ring lies with a superstar capable of a handful of monster seasons, even if the rest of his career is mediocre.