NBA MVP Metric

By Christopher Vela and Daniel Schwartz

Introduction

Link to PDF Right Click to Save: adafinalproject%20(1).pdf

At the end of each season, the National Basketball Association designates the most valuable player (MVP) based on a voting system whereby broadcasters and journalists rank the ﬁve best performing players. Points are then assigned inversely with rank for each ballot cast. The rank-point equivalencies are as follows:

rank for a given ballot points

1st: 10

2nd: 7

3rd: 5

4th: 3

5th: 1

The award is given to the player receiving the most points. In recent years, the MVP has been the subject of much criticism from fans and sports commenta- tors. Wrote Fansided writer Phil Daniels: “The NBA fails to concretely deﬁne the criteria for the Maurice Podoloﬀ Trophy winner. The only criterion is that the MVP is awarded annually to the ’most valuable player.’ Value is not deﬁned. Scope is not limited.” Many have also claimed that the award has become a function of player popularity and media attention rather than performance. To investigate voting behavior, we examine historical data from 1986-2012, looking at metrics in several areas: player age, position, individual performance, contribution in the context of a team and salary. We assume here that player salary functions as a proxy variable for understanding popularity and overall market demand for a particular player. We narrow our focus to the sub- set of players receiving positive point counts (typically a handful of players each season). Our question of interest can then be distilled to: “Among the top-performing players, which character- istics best predict total points?” Data pertaining to basketball metrics, player age, position and MVP voting was drawn from www.basketball-reference.com. Player salary data was taken from www.eskimo.com/~pbender.

Two seasons of data (1998-1999 and 2011-2012) were excluded from our analysis as player negotia- tions prevented completion of these seasons’ games. The 1987-1988 and 1990-1991 seasons were also excluded as player salary data could not be found. Total point counts per player were normalized by the total number of points available since the total number of ballots cast diﬀers from year to year, giving a response variable that can be interpreted as the proportion of the vote captured by a particular player. Player salaries were normalized by the cumulative league salary for the particular year to account for inﬂation.

After aggregating all of the positive point-receiving players over 23 seasons we are left with a sample size of n = 374. This method makes two important assumptions. The ﬁrst is that the tastes and preferences of the sports commentariat are roughly stable over time. To check the validity of this assumption, we will examine models for both the full 23 seasons as well as for the most recent 10 seasons (n = 166). The second assumption is that the joint distribution of the covariates for players in the running for MVP are independent from year to year. This assumption becomes important as there are many recurring players in our sample.

Implementation

Our prediction for MVP for 2013-14 Season: 2nd column is number of games played and third column is percentage of votes attained by each player.