A survival analysis to discover which skills determine a higher scoring in basketball

Abstract

Over the years data analytics for sports has developed consistently. Survival
analysis is a method that allows to study the occurrence of a particular event during a
period of follow-up. This work aims studying the main achievements associated to the
probability of reaching a certain amount of points during a NBA season segment. A
Stepwise Cox regression model and LASSO Cox regression were used to select the most
important variables. Two settings were examined: in the first one only a 20% of censoring
was admitted, while in the second one 50% of the subjects was censored, i.e. did not
experience the event. It emerged that attempting more shots, gaining more achievements
(double doubles) and having been selected for the All-Star game increase the probability
of success, i.e. exceeding the given threshold of points. Moreover, a higher number of
steals seems to decrease the probability of gaining a certain amount of points. Thus,
players more involved in this fundamental are penalized in terms of gained points.