Reading the First Weeks of an NBA Season

Two games into a season, someone is always shooting sixty percent from three and someone else has forgotten how to make a layup. Both facts will be discussed at length. Both are almost entirely noise. The regular season opens on Oct. 20, and for the next few weeks the loudest numbers on any stat page will also be the least reliable ones.

That does not make early games useless. Some of what they show is real and important, and it shows up quickly. The research skill is sorting early information into the kind that deserves weight now and the kind that needs a month or more before it says anything. This piece is about that sort.

Start from what you knew before the first game

A prior is the distribution you hold for something before new evidence arrives. For an NBA player entering a season, the prior is built mostly from last season and the seasons before it: his minutes, his share of the offense, his shooting on a large number of attempts, and how those numbers moved as his role changed. Hundreds of minutes and many hundreds of shots sit behind that estimate. Two new games do not outweigh it.

Base rates in prop research makes the general argument for anchoring on the reference class before adjusting toward the individual. The early season is where that habit pays most, because the individual evidence is at its thinnest and the temptation to override the anchor is at its strongest.

A prior is not a promise that nothing changed, though. It is a starting position, and the question for each input is how far the new season's evidence should move it. That depends on two things: whether the player's situation actually changed over the summer, and how quickly the specific stat becomes informative.

Check what changed before trusting what carried over

Last season's numbers describe a player in last season's context. Before leaning on them, list what is different now.

• A new team. A traded player or free agent signing brings his skills, not his role. His minutes and his share of the offense are decided by a new coach and a new set of teammates.
• New teammates on an old team. A high usage arrival pushes existing players' shares down. A departure pushes them up. Either way the player did not change and his prior still needs to move.
• A new coach or a new system. Rotation length, pace, and shot selection are coaching choices, and they can shift the environment for everyone on the roster at once.
• A return from injury. A player coming back from a long absence may start on a minutes limit that is rarely stated precisely, and his early workload may describe that limit rather than his role.
• A player whose role grew late last season. His full season average understates where he ended, and the end of last season may be the better starting point.

Each of these changes the opportunity side of the prior, the minutes and the share, much more than the skill side. A shooter who changes teams is still roughly the same shooter. What changes is how many shots he gets and how open they are. That is why researching a debut recommends anchoring on role rather than talent when the history does not transfer cleanly, and the same logic covers a veteran in a new uniform.

Some early numbers mean something quickly, others do not

The core idea is simple. A stat becomes informative quickly when it varies little from game to game for the same player but differs a lot between players. It becomes informative slowly when single games swing widely and the true differences between players are small. Early season research should lean on the first kind and nearly ignore the second.

Faster: opportunity

Minutes, starting assignments, and closing lineups are decisions a coach makes on purpose, and they carry information from the first game. So do volume measures: field goal attempts, free throw attempts, and three point attempts per minute. The NBA.com stat glossary defines usage percentage as the share of team plays a player uses while he is on the floor, and that share, like the minutes behind it, reflects the role he has been given. A role can still change after a week, but a week of consistent minutes and attempts is real evidence about it.

Slower: efficiency

Shooting percentages are the opposite. Each shot is close to a coin flip weighted by skill, and the differences in true skill between most rotation players are small compared with the randomness in a few dozen attempts. Three point percentage is the extreme case: a shooter who takes five threes a game has taken maybe fifteen after three games, and fifteen attempts cannot distinguish a good shooter from an average one. Free throw percentage is steadier per attempt, but most players take so few that it too needs a long time.

The simple attempt weighted blend in the example is a rough tool. Proper shrinkage methods treat the prior as worth a certain number of observations that differs by stat, and they give the season's evidence more pull on stats that stabilize fast. The direction of the lesson is the same either way: volume moves the estimate quickly, percentages move it slowly.

Traps particular to the first weeks

Early extreme results tend to be followed by less extreme ones. That is regression toward the mean, and in October it is everywhere. The player at the top of the early scoring leaders and the one at the bottom of the shooting table are both, by selection, the most likely to drift back toward their prior. Streaks and slumps explains why a hot stretch is mostly evidence about variance unless something about the opportunity changed alongside it.

Schedules are also uneven early. One team may have played two home games against weak defenses while another opened with a road trip against strong ones. Early per game numbers mix the player with his opponents far more than full season numbers do. Before reading anything into a player's first week, look at who it came against.

Rotations are unsettled too. Coaches try things in the first weeks, and a bench player who gets 25 minutes in game two may be back to 12 in game five without any change in his play. Look for patterns that hold across several games rather than a single night that fits a story. The minutes and blowouts piece adds a further filter: an early season minutes jump that happened in a lopsided game may be garbage time rather than a new role.

Finally, lines are set with the same thin information you have. Early season prices are drawn from last season's numbers, offseason changes, and a few games, and they can lag a genuine role change or overreact to a hot week. Neither of those is a reason to trust your own early read over the line. It is a reason to be specific about which piece of evidence you think the line has weighted wrongly.

A working routine for October and November

1. Write down the prior for each player you research, built from last season and adjusted for any change in team, teammates, coach, or health.
2. After each game, update opportunity first: minutes, starts, closing role, and attempts per minute. These deserve weight early.
3. Leave shooting percentages close to the prior for weeks. Move them only a little, and only in proportion to the number of attempts behind the new evidence.
4. Tag every early outlier with its context: opponent, rest, margin, and who else was out. An outlier with an explanation is not a trend.
5. Revisit the prior itself when opportunity evidence has piled up across several games. A durable role change is the one early season signal that justifies moving a lot.

Where this approach can be wrong

Anchoring on the prior is the right default, and defaults are sometimes wrong. A player can genuinely improve over a summer, and for a few weeks the evidence of that will look exactly like noise. A careful early season process will be slow to believe him. That cost is real and worth accepting, because the alternative is chasing every hot start, and most hot starts fade.

There is also no clean point at which early numbers become meaningful. Evidence accumulates gradually, and different stats cross any useful threshold at very different times. Anyone offering a single game count after which a stat can be trusted is simplifying. The honest version is a sliding weight that moves faster for volume and slower for efficiency.

The short version for the next month: trust the role, doubt the percentages, check the schedule behind every early number, and let the prior go only as fast as the evidence earns.