Skip to main content
Menu
NBA Nerds, Rejoice: Our Projections Now Reflect Injuries And Lineup Changes

Eagle-eyed FiveThirtyEight readers might have noticed something different about our NBA predictions interactive on Wednesday afternoon: We rolled out a slightly different look to the table based on a pretty significant upgrade to the way we handle NBA forecasts.

We’ll have a full methodology page ready soon, but for now we wanted to highlight a few of the key features in our new, improved system — and what they mean for predicting the NBA season.

We explained a few of these concepts at the bottom of our 2018-19 NBA preview post, but our new predictions hinge on a couple of important elements:

Updating talent ratings. In the past, our system relied on preseason player ratings derived from our CARMELO projection system, which informed initial ratings for every team. Those would then go up or down during the season based on a team’s game results, without regard to how the individual players on a given roster performed.

Now, that’s changed. We still use CARMELO as the starting point for every player’s talent projection, but those preseason expectations for minutes per game and offensive/defensive plus-minus impact will evolve throughout the season depending on how a player performs.1 So if a player outperforms his preseason projection and/or gets more playing time than expected, his talent ratings will change to reflect that, which will then be summed up at the team level for each roster.

Up-to-date rosters and depth charts. Another shortcoming of our old system is that it had no way of knowing when a team made a change to its roster composition, beyond simply inferring it from changes in the team’s record. This wasn’t ideal — particularly during the NBA’s regular season, when resting stars and coasting superteams are the norm.

Now, we have created a method that algorithmically generates depth charts for every team in every game, based on our updating talent ratings, the team’s roster and a list of which players are or are not available on a given night. With a little bit of manual curation, we’ll be able to keep track of how a team’s talent level fluctuates over time, accounting for trades, injuries, suspensions, rest days and various other factors. (In the future, we plan to release these depth charts in interactive form, so stay tuned.)


As a consequence of this approach, our CARMELO ratings no longer necessarily care about wins and losses, except indirectly (as they are reflected in changes to player performance or roster composition). Also, for each team in our interactive, we will now track a pair of CARMELO ratings: “Current” and “Full-Strength.” Current represents the team’s estimated strength rating for its next game, including injuries and other absences. Full-strength is a team’s rating if it had all of its players available to play right now. Generally speaking, teams will be closer to full strength as they begin the playoffs, barring long-term injuries, making that rating a better indicator of a team’s potential ability in the big picture. (For the purposes of our playoff simulations, we also tweak rotations to give more playing time to top players and award a bonus to teams with more postseason experience.)

We hope all of these modifications will make for better predictions within the season, both on a day-to-day basis and over the long haul, with an eye toward the playoffs in particular. You can already see the changes taking hold in the interactive: The Golden State Warriors’ current rating dropped dramatically to reflect all of their key injuries at the moment, though their championship odds actually improved (thanks to their talent at full-strength, plus our ability to regulate expected playing time during the postseason, when top stars log more minutes). The Toronto Raptors saw the opposite effect — their current rating improved to reflect the excellent play of so many of their rotation members this season, but Toronto’s title chances dipped slightly because its outstanding depth comes in less handy during the playoffs.

Let us know what you think about the new changes, and keep an eye out for more additions to the interactive as the season goes on.

Check out our latest NBA predictions.

Footnotes

  1. As usual, we are judging player performance using a mix of two-thirds Real Plus-Minus and one-third Box Plus/Minus.

Neil Paine is a senior sportswriter for FiveThirtyEight.

Comments