Three weeks into the NBA season, the standings don’t look quite as expected. With most teams having played at least 10 games in the first 82-game campaign—fingers crossed—in three years, favorites like the Bucks and Lakers are sitting in play-in position, while the conferences’ top tiers are populated by more surprising candidates.
The greatest early overachiever, relative to expectations, is Golden State, which entered the season with an over/under projection of 48.5 wins, according to Basketball-Reference’s archive of Vegas odds. That translates to 5.9 expected wins out of every 10 games—but the Warriors are a league-best 9-1, meaning they’re already 3.1 wins, or 31 percent, above expectation.
Two teams that were projected to finish with losing records have also started the season hot and made similarly sized leaps. The Wizards are 7-3, in contrast with an over/under of 33.5 wins, or 4.1 wins in 10 games. And the Cavaliers—who entered the season with the fourth-lowest over/under total in the league (26.5, or 3.2 wins in 10 games)—have started 6-4 (now 7-4), basically double the expectation.
Typically, over a full season, Vegas odds are pretty good indicators of team quality. Looking at all 82-game seasons this century (a sample of 565 teams), the 10 teams that entered the season with the best over/unders finished with an average record of 59-23; the 10 teams with the worst over/unders finished, conveniently, with the reverse average record of 23-59. Overall, Vegas projections alone could predict 56 percent of teams’ final records over this span (the R-squared between the two variables).
But once a season begins, what happens in the cases of teams like the Warriors or Wizards or Cavaliers? Should their hot starts affect our impressions of their future chances, versus preseason expectations? Or, conversely, should we lose some faith in early underachievers like the Bucks, who entered the season with a 54.5-win over/under but have started 4-6?
The answer, at least for now, is not to change much. Consider the biggest overachiever through 10 games in the 2018-19 season (the last with 82 games): Denver, which started 9-1 despite an over/under projection of 47.5 wins. Over the rest of the season, the Nuggets regressed and won at a 51-win pace—much closer to their initial projection than their play over the first 10 games.
In many cases, the most surprising teams early on end up performing worse than expected over the rest of the season, even if by a small margin. The 2017-18 campaign included two major early overachievers: The Pistons started 7-3 despite an over/under of just 38.5, then won at a 36.5-win pace the rest of the season; the Knicks started 6-4 despite an over/under of 30.5, then won at a 26-win pace the rest of the way.
In the season before that, the biggest overachiever was the Lakers, who started 6-4 despite an over/under of 24.5. They won at a 23-win pace the rest of the season. The second-biggest overachiever was the Hawks, who started 8-2 despite an over/under of 43.5, then won at a 40-win pace the rest of the way.
And so on and so on. Overall, we find that the 10 percent of teams in this sample that performed the best in their first 10 games, relative to preseason expectations, won 77 percent of their first 10 games, versus a Vegas-projected win rate of just 49 percent. Across the rest of the season, they won 55 percent of their games—much closer to preseason estimates. The same relationship appears with early underachievers, in the other direction.
So if 10 games isn’t enough to significantly change perceptions of a team’s quality, then how much data do we need in order to come to firmer conclusions? The answer is 25 games. After 10 games, a team’s record offers 37 percent predictiveness for its performance over the rest of the regular season; at the same time, a team’s preseason Vegas projection offers a more robust 54 percent predictiveness. But by Game 25, a team’s actual early-season record becomes more predictive than its Vegas projection going forward.
Another way to predict teams’ future success is using point differential, which is useful more quickly than record, as it offers a much wider range of outcomes than the simple win-loss binary. Point differential catches up to Vegas projections by Game 17, and overtakes it in terms of rest-of-season predictiveness by Game 18. This timeline generally aligns with previous research from The Athletic’s Seth Partnow, who used a different method of analysis and found that a team’s net rating tends to settle after the first 15-20 games.
It is readily apparent why early-season deviations from expectations often look flukish in retrospect: Numerous contextual factors can influence a team’s initial record, from injuries (and, this season, COVID-19-related absences) to random luck to schedule disparities. The Wizards, for instance, are 5-0 in “clutch” games that have been within five points in the last five minutes—a surely unsustainable pace.
And the Warriors’ nine wins have come against the league’s easiest schedule so far, according to Basketball-Reference, as they’ve already feasted on the Rockets, Pelicans, and Thunder twice and haven’t faced any other top team. (No, the Lakers have not looked like a top team so far this season.) That disparity will even out somewhat by Game 25, as the Warriors play Brooklyn, Philadelphia, and Phoenix twice, among other higher-quality teams, in that span—just another reason to wait on delivering verdicts about their outlandish performance above expectation so far.
Those early-season wins count, of course—even if a 9-1 record doesn’t mean much for predicting the Warriors’ record over the rest of the season, that’s nine banked victories and a 3.5-game cushion on the play-in spots, which could prove vital as the Western titans battle for seeding later in the season. And there’s reason for legitimate encouragement about these three teams’ outlooks, from the more diversified offense in Golden State to the deeper rotation in Washington to Evan Mobley’s two-way emergence in Cleveland.
But year after year, teams off to surprisingly scorching starts fall back to their expected level as the season continues. It’s important not to overreact to the vagaries and randomness inherent in a small sliver of a season. There’s not much reason to forecast the playoffs or title races until Thanksgiving—roughly the 18-game mark, when point differential becomes a more meaningful indicator—at the earliest. Everyone will be more than eager to spin forward by that point.