If you’ve followed along with our NBA projection system in recent years, you probably noticed a recurring theme. During the regular season, the Golden State Warriors would either hit dry spells while key players were injured, or coach Steve Kerr, guarding against injuries, would frequently sit key players. Because of that, the Warriors would lose games that ultimately didn’t matter that much, and our Elo model — designed to track the ebbs and flows of the season’s wins and losses — would punish Golden State each time. But like clockwork, the Warriors would emerge in the spring as strong as ever — despite their blemished Elo rating.
Even after updating our model in 2015 to try to account better for the talent on each team (using a blend of our basic Elo ratings and our preseason CARMELO player projections,1 which we called “CARM-Elo”), the model would give us some weird projections because it still reacted solely to wins and losses — it didn’t understand injuries, trades or rest.
Last year, that flaw left CARM-Elo insisting that the baby-faced Sixers had a better probability of winning it all as the playoffs approached than the battle-tested Warriors did.2 The Elo-based system simply proved to be a bit too gullible, with too much of a recency bias to recognize that just because a team is playing extremely well doesn’t mean it’s going to be hoisting the Larry O’Brien Trophy at season’s end. So we’ve tried to fix these problems by launching a new set of projections specific to each team’s baseline talent.
Our NBA forecast now offers two sets of projections: one that will track teams’ basic Elo ratings just like our original forecasts and one based solely on our player projections, without focusing on daily wins or losses.3 The latter projection will account for real-time movements and talent adjustments on the depth charts for every team — which are available to check out on the new individual team pages. That marks a somewhat seismic shift for the system, since it didn’t previously have a way of factoring in long-term injuries or the idea of players resting. But now those things are considered, and it shows in the new ratings. (Read more about the methodology here.)
How CARMELO compares with previous forecasts
Side-by-side feature comparison between FiveThirtyEight’s Elo, CARM-Elo and CARMELO NBA forecasts
|Preseason team ratings use …||Last year’s final rating, regressed to the mean||Up-to-date depth charts||Up-to-date depth charts|
|In-season ratings updates use …||Game results||Game results||Player ratings and rosters|
|Accounts for postseason experience?||✓||✓|
|Accounts for trades and injuries?||✓|
|Used in 2018-19 predictions?||✓
(if you select it on the Standings page)
The Warriors are a great example of the differences between the forecasts. With our basic Elo rating projections, Golden State currently has just a 27 percent probability of reaching the NBA Finals and a puzzling 14 percent chance of winning a title. That’s lower than the Raptors’ 21 percent championship odds — and it’s a relative head-scratcher for a back-to-back champion that’s still relatively young and returned the vast majority of its core from the past two years.
Yet using the new CARMELO forecast, which takes into account a team’s talent and fully healthy potential, the Warriors’ probability of reaching the finals is a whopping 67 percent, and their chances of winning a third-consecutive championship sit at 51 percent.
Those numbers, far more in line with conventional wisdom, are based on two things: First, Golden State has an insane amount of talent when healthy — far more than any other club. (This is reflected in the full-strength CARMELO.) Second, the system believes that the Warriors are nowhere near their peak yet because of all the injuries they’ve dealt with to this point. They just recently got Stephen Curry and Draymond Green back in the lineup, and the team should see All-Star DeMarcus Cousins make his Golden State debut soon. And rotation players Andre Iguodala and Shaun Livingston rejoined the team Monday.
Looking beyond the Warriors, there are a handful of other key takeaways that stand out with the new system:
The Nuggets haven’t hit their peak, but the system doesn’t fully buy their talent
Aside from a poor stretch in mid-November, Denver has arguably been the most solid team in the NBA to this point. The Nuggets are tied for first place out West and have pieced together an unprecedented turnaround on the defensive end.
Nothing about this situation comes across as a fluke so far in our forecast. In fact, if anything, the model thinks Denver is capable of more. Earlier this week, Denver has the biggest gap in the league between its full-strength CARMELO rating and its current rating — a 63-point shortfall.
That, of course, stems from the fact that the Nuggets are awaiting the returns of Will Barton, Gary Harris, Paul Millsap and Isaiah Thomas.4 Denver hasn’t been fully healthy all year. But when the team gets everyone back, it will be interesting to see just how far it can push things.
However, while the forecast realizes that the Nuggets haven’t been healthy yet, it only recognizes them as the West’s fourth-most talented club at full strength, after Golden State, Oklahoma City and … Houston. The Thunder and Rockets currently own championship probabilities of 8 and 4 percent, respectively, while the Nuggets claim just a 2 percent chance. The numbers may be a nod to the fact that depth — which Denver has far more of than those other contenders — generally means less during the playoffs, when stars often steal the show (a fact our model now accounts for).
If there’s a saving grace for the team’s fans, it’s that the basic Elo projections are still high on Denver and have the club tied for the fourth-highest title probability, with a 9 percent chance.
The updated forecast believes in the Rockets but hates the Kings and Mavs
The CARMELO forecast model relies on real-time depth-chart information to make its projections. So the CARM-Elo system might have penalized the Rockets for technically having Carmelo Anthony on the roster, even though he’s away from the team and hasn’t suited up for more than a month. But just like Mike D’Antoni, the new mode simply assigns Anthony no minutes on the depth chart and moves forward.
And speaking of Houston, the club is an interesting one to analyze here because of how fervently the model still believes in the Rockets, who are given a 93 percent chance of reaching the playoffs despite having a record just barely over .500 more than a third of the way through the campaign. By contrast, the basic Elo forecast has them at just 72 percent to make the postseason.
Again, this is almost certainly the new model showing faith in the top of Houston’s depth chart: It likely believes that guards Chris Paul and Eric Gordon won’t continue to have career-worst years from an efficiency standpoint.5 (Those deficiencies alone don’t explain the Rockets’ slide — which many of us saw coming — on their own. The team also gambled by letting two of its better, more switchy perimeter defenders walk without having reliable replacements for them.)
Yet while the CARMELO team projections are high on Houston, they don’t make much of a pair of a clubs that have been pleasant surprises this season. The Kings and Mavericks, each with records above .500, are given just 6 and 4 percent probabilities, respectively, to make the playoffs. That’s a night-and-day difference from the basic Elo model, which gives Sacramento a 34 percent chance of getting in and Dallas a 21 percent chance.
What that spells out, aside from CARMELO’s limited belief in each roster, is the model’s view that neither club has enough battle-tested veterans in their rotation to get a boost for postseason experience. The Mavs are tied with the Nets for lowest playoff adjustment in the new system, while the Kings rank seventh-lowest. Houston, by contrast, has the fifth-highest adjustment for its postseason experience.
Overall, we think the changes to our model will help us do a better job of predicting the season. And now you can check out the inner workings of what goes into the predictions on each team’s page. You can see the depth charts we’re working with — which, by definition, are estimates, but we try to keep them updated daily by tracking trades, injuries and other changes to each player’s status — as well as an updating chart of a team’s chance to make the playoffs. Also included is a handy graph of changes to a team’s talent rating, which highlights the difference between how a club looks at full strength compared with the present moment.
So when you’re watching the NBA’s Christmas Day games with friends and family, check out our predictions to follow along with how each team has been playing and sound like an expert.
Check out our latest NBA predictions.