Who should be this season’s NBA MVP? Many names have been mentioned, and the apparent prime candidates include Nowitzki, Nash, and Duncan, with McGrady and Bryant as possibilities. Can we predict the winner? We could build a statistical (logit) model, using a dummy for MVP win as our dependent variable, and various individual and team stats as our independent variables, but the N is very small, and I am not confident that such a model would be very robust.
We could also try to determine the criteria we believe the voters will use in making their decision, attempt to develop measures to capture these criteria, and then choose more or less objectively from there. I like this idea better — I like to measure things — but what criteria should we use?
It is often noted that the MVP is a regular season award, so we should use regular-season statistics. What should be used, though, as criteria for the decision? Bill Simmons, a good writer, but not much of an arbitrarian, suggests three questions to ask, and it is also often mentioned that the MVP should go to the best player on the best team (which would probably be Nowitzki, which means that we have four different answers to our four different questions).
I would like to bring your attention to the word valuable — how does one characterize a player’s value in the NBA? An economist would suggest we measure the player’s marginal revenue product — that is, the amount of additional money his franchise made by virtue of his production that year. This is probably a function of multiple factors: star power, marketability, performance, size of market, contribution to wins, etc. Since I again don’t want to try to regress all of these variables on team revenue (I don’t want to spend time finding the data, plus it’s hard to convert team-level coefficients to apply to players), I am going to pick wins: how many wins did the player bring to his team?
There exist several ways of calculating contribution to wins (see basketball-reference.com for a start, I think some guy named Hollinger does something like this, and probably 82games.com as well, maybe?), but I’ve got my own (of course). To see how I chose the measure, try the following thought experiment:
Imagine two teams in the same league composed entirely of clones of a single mediocre player, except each also has a different “superstar”-type player. The clones’ performance every night is always within a certain range, though it can vary randomly and according to various irregular factors. If each of these teams plays an identical schedule, and each of the clones play an identical number of minutes and have an identical number of touches, and the superstars have equal numbers of touches as well, I submit that the team with the better record is the team whose superstar is superior. That is, ceteris paribus, the more successful team has the more productive star.
Alternatively, consider the following: a team, with five players, has 1000 “possessions” (defined someway) over the course of a season, and a 15 and 5 record in 20 games. Four of the players each are credited with 150 possessions (15% of the total, each) and the fifth had 400 possessions. Which player deserves the most credit for the 15 wins (and the 5 losses?). I suggest that the player with 40% of the possessions deserves (on average) 40% of the credit for the wins. One might say that the star player here had 6.0 “win shares”, and the other four players each contributed 2.25 win shares. In a league in which each team plays the same number of games (thus having the same opportunity to garner wins), comparisons of win shares can be made across teams.
From here, it is a simple question of measurement. But, since this post is already incredibly long, I will continue next time with my proposed operationalization of player value, and perhaps my MVP prediction.