Player performance and Pythagorean expectations related to game outcome in NBA

Player performance rating vs game outcome indicator

Tthe influence of talent depth or an
individual superstar is keenly felt in determining the fortunes of a NBA season.
A variety of player performance ratings have developed in the NBA that can be
projected into future years along with estimated player usage to attempt to more
accurately predict a side’s expected fortunes.

The most commonly quoted
Player Efficiency Rating (PER) was developed for ESPN by John Hollinger. The
stat uses on court actions to derive a player rating per minute that accounts
for team pace. Although far from perfect, it does present a credible way to
assess both the teams with outstanding contributors, as well as depth of roster.

A broad assessment of
individual players is possible by referring to Hollinger’s reference guide to a
player’s score. 27.5 represents a potential league MVP, 22.5 an “All Star”
candidate and 13.0 a bench rotation player. Golden State’s Stephen Curry scored
a PER of 28 in 2015 and the team had eight roster spots that scored above the
league average.

Overall PER figures can
be used to estimate if a team is substantially improved or not from previous
seasons by looking at the performance figures for the roster as a whole or the
most widely used line-ups, which are available historically or on numerous
fantasy sites for upcoming games.

PER is strongly correlated to
seasonal success because the very events that determined game outcomes are
included in calculating the rating. If we look at a side’s most regularly used
players using regression analysis from previous seasons, the best average squad
in terms of PER in the NBA has in excess of a 90% chance of reaching the post
season, winning an average of 58 games.  

Combining Player ratings with
Pythagorean expectations

In attempting to project
future team performance we also have to project player ratings, adjusted for
ageing along with likely usage. A player ratings approach is not a project to
embark upon lightly and proprietary ratings are available, including Nate
Silver’s recent addition.

But they do allow us to easily
identify why simple team Pythagorean extrapolations, such as those above, may
differ greatly from win totals quoted by the books.

Foe example, Portland Trail Blazers
have lost five of their seven most frequently used players from last season.
From this, we can expect that inexperience, lack of familiarity with team mates
and uncertainty surrounding the expected output from so many new faces may
account for the large discrepancy in Pythagorean projections (50) and actual win
total quotes (27) for 2015/16.

Similarly, Oklahoma City Thunder
only got Kevin Durant onto the court for 27 games last season because of injury,
compared to an average of 77 in his previous six seasons. 2014/15’s scoring
performance, largely without Durant, depressed their 2015/16 Pythagorean based
projection.