How many games until an NBA player’s averages mean something?
About ten games for points, and longer for everything else. We checked about 235 regular players in each of the last two NBA seasons. After five games, a player’s scoring average this season was a better guide to the rest of his season than last season’s average only half the time. For rebounds and threes, last season was still as good a guide after twenty games.
So in the first few weeks, the number you are most likely to anchor on, what a player has done so far this season, is the weaker of the two. Here is what we measured, how to use it, and where our own projections disagree with it.
What we measured
For every player who played at least 40 games in a season and at least 40 the season before, we took his average over his first few games and asked a plain question. Which was closer to what he averaged for the rest of that season: those first few games, or his whole previous season?
The table shows how often this season’s early average was the closer one. Fifty means a coin flip. Under fifty means last season was the better guide.
| Stat | After 3 | After 5 | After 10 | After 15 | After 20 |
|---|---|---|---|---|---|
| Points | 39% | 50% | 56% | 58% | 58% |
| Rebounds | 32% | 37% | 46% | 52% | 53% |
| Assists | 39% | 44% | 53% | 57% | 58% |
| Threes made | 33% | 34% | 44% | 48% | 51% |
| Minutes | 48% | 52% | 56% | 62% | 64% |
Share of players whose average after that many games was closer to the rest of the season than last season’s average. Based on 237 players in 2024-25 and 236 in 2025-26, each with 40 or more games that season and the season before. Regular season and NBA Cup games, games played only. Red is where last season was the better guide. Differences of a few points either side of 50 are inside the noise for a sample this size.
Three things stand out. Points settle first: last season wins through about three games, it is even at five, and this season pulls ahead by ten. Assists take about fifteen. And rebounds and threes never clearly flip inside twenty games. A month into the season, last season’s rebounding average was still as good a guide as this season’s.
Minutes behave like points. That matters, because minutes drive every counting stat, and a minutes change you can see after ten games is worth more to you than a hot shooting week.
What one big night is worth
Opening night produces the loudest numbers of the year, and the least reliable ones. After a single game, a player’s scoring number missed his rest-of-season average by five points or more 40 times in 100. Last season’s average missed by that much 13 times in 100.
Here is what that looks like with a made-up but typical case. A guard averaged 18 points last season and scores 27 in his first game. His next line goes up to 21.5. If you bet the over at -110, you risk $11 to win $10, so you need to be right 52.4% of the time to break even. The number pushing you toward that over is the one that misses by five or more four times in ten. Last season’s 18 is the number that usually holds up, and it says under.
That does not make the under a good bet. The book saw the same 27 and the same 18. It means the single game is not the reason to bet either side.
Use both, not one
The better answer was not either number. An even mix of last season and this season was closer than either alone for points and assists by five games, and for all four prop stats by ten. Last season tells you the player. The first few games tell you what changed: a new team, a bigger role, a starter traded away. You want both until this season has earned the weight.
A rough rule that fits the table: for the first two weeks, give last season at least half the say. Let this season take over points and minutes after about ten games, and assists after about fifteen. For rebounds and threes, keep last season in the mix well into November.
Where this stops being true
- It is about regulars. Everyone in the study played 40 or more games in both seasons. Rookies, and players who barely played last year, are exactly the ones with no last season to lean on, and there were 197 of them among the regulars in these two seasons. For them, early games are all you have, and they are noisy.
- Role changes break it. A player traded to a team that needs his shots, or a backup who becomes a starter, has a last season that describes a different job. The average still won overall, but not for him.
- It compares averages, not bets. Being closer to a player’s average is not the same as clearing a book’s line, and nothing here is a hit rate.
- Two seasons is a modest sample. The direction was the same in both, and the shares near 50 could move a few points with another season.
What our own numbers do, and where they disagree
Our NBA projections lean on last season for a player’s first 20 games this season. Each game played gives this season more of the say: about a sixth after one game, half after five, four fifths by game 20. After that the projection runs on this season alone. Whenever a projection is leaning on last season, the Edge board shows a Last season figure beside it, so you can see how much of the number is the old one. Rookies get no projection at all until five games. When the Lab’s Matchup, Rotation and Recon tabs fill in with last season’s games for a player who has barely played, they label them as last season.
The first version handed back to this season’s numbers at five games, which the averages above called early. So we replayed our own projector over the first 20 games of every player’s last two seasons, picked the bridge on 2024-25 and checked it on 2025-26. The longer bridge missed by less in both seasons for all four stats, points included, which the plain averages did not predict. The gain is small. On points in 2025-26, the average miss went from 4.73 to 4.67 points a game. Its chances at each line also matched what happened a little more closely. We could see the 2025-26 results while choosing, so that check was not blind.
The cost shows up in the number you see. Leaning on last season longer pulls a hot start down: a player scoring 31 a game through 12 games, who scored 24 last season, now projects about three points lower than before the change. So the highest numbers on our board each day sit lower, and they clear the line just above our number more often. In the replay’s first six weeks of 2025-26, the top 25 points projections each day cleared it about 47 times in 100, up from 43, which means they missed 53. Part of that rise is the lower number, not a sharper read. Our NBA record started on opening night, October 20, and its first nights used the five-game version. Read its early-season hit rate with that in mind.
Free, no account. Slip Check shows the books’ chance at each NBA leg’s exact line.
Questions
How many games until NBA player stats are reliable?
For points, about ten. Across about 235 regulars in each of the last two seasons, a player’s scoring average after ten games was closer to the rest of his season than last season’s average 56 times in 100. For assists it took about fifteen. For rebounds and threes, last season’s average was still as good a guide after twenty games.
Should I use last season’s stats for early-season NBA props?
Yes, alongside this season’s, not instead of them. After five games, an even mix of the two was closer to the rest of the season than either one alone for points and assists. By ten games the mix was closest for all four prop stats: points, rebounds, assists and threes.
Does a big opening night mean anything?
Less than it looks. After one game, a player’s scoring number missed his rest-of-season average by five points or more 40 times in 100. Last season’s average missed by that much 13 times in 100. One game is mostly noise, and early lines move on exactly those games.
Why do rebounds and threes take longer to settle?
We measured that they do, not why. One likely reason is that they are small counts that swing a lot from one game to the next, so a handful of games says little about a player’s real level. Points and minutes are bigger numbers and settle faster.
What does ET Parlays do early in the NBA season?
Our projections lean on last season for a player’s first 20 games this season, a little less each game, and the Edge board shows a “Last season” figure whenever they do. Rookies get no projection until five games. The first version handed back at five games. When we replayed our own projector over the last two seasons, the longer bridge missed by a little less for all four prop stats, so we changed it.
Nothing here is betting advice. Bet only what you can afford to lose. If it stops being fun, take a break.