By Johnathan Cruzen-Myer · Updated 28 October 2026 · How this is measured

Home › NBA › Player prop trends

NBA player prop trends, and what survives a null

Built from 113,832 appearances across 5,342 games, 585 players with enough history to read. Every split is measured against that player's own rate in the same market — never 50%, never the league.

What this cannot tell you

Tonight's games

3 games on, and the splits below are only for players expected to play in them — 89 of 9,925 found across the league. The arithmetic above is still computed over everything, because searching one night and quoting the chance figures of a full search would understate how many tests were run.

Best picks, game by game → The rows below that clear both tests (form beats the price, expected return positive), up to three per game.

Player trends

 Green  his last 20 games beat the chance the price implies by 5+ points (form says the price is generous).  Red  they fall 5+ points short (form says the price is mean). No colour: within 5 points, or no price yet. Example: at 2.80 the price implies 35.7%; a player who hit it 10 of his last 20 (50%) is green.

Each player's split for tonight's situation only: this opponent, this arena, this rest, tonight's crew, who starts beside him and who starts against him at his position. Thresholds are round numbers resembling prop lines, not prices anyone offered — we hold no history of posted NBA player lines, so a hit rate here is a frequency and says nothing about value.

Every row is an OVER at a milestone line, one row per player and milestone, with every one of tonight's situations that cleared the filters listed in it. Each situation has 15+ games behind it, and a milestone only appears when it sits near his average over his last 20 games, so lines he clears almost every night (and that price around 1.03) are left out, as are 4 priced under 1.30. All together counts only the games where all of those situations held at once; records of overlapping situations cannot be added. Expected return ranks the table: the record blended with his own rate (a prior worth 10 games, because raw splits overstated their hit rate by about nine points in the backtest), times the price. It still cannot see what the price knows (minutes, injuries, matchup), so a positive figure is a reason to look, not proof of value. Prices are the best over at each milestone across US books, taken Fri 12:49 pm Sydney time, from PropLine. The record is hits–misses. Last 20 is how often he reached the milestone in his last 20 games, the figure to hold against the price:  green  where it beats the price's implied chance by 5+ points,  red  where it falls 5+ short. The main line cell does the same at that line: "last 20: 9 over, 11 under" is how many of his last 20 games went over and under it, and its colour compares that over rate with the over price, so a row and its main line can disagree.

When a teammate is out

For each club with a rotation player (20+ minutes a game) out tonight: what his teammates have done in the games he missed, against the games he played, counting only games where both were on the club. When two or more are out, the first table counts only games where all of them sat, against games where all of them played. Each needs 8+ games without him, a lift of 12+ points, and an exact chance figure. 337 player-market pairs were tested to find these, so expect some to be chance.

The same caution as everything above, and more of it: a teammate's absence is the first thing a bookmaker prices, usually before the news reaches anyone else. A big lift here is a reason the line has already moved, not evidence that it has not moved enough.

When both agree: a teammate is out, and he beats his average against tonight's opponent

Players in both lists below for the same stat. Each side's evidence is shown as it stands on its own: the teammate split (his rate at this milestone without the absent player) and his average against tonight's opponent against his overall average. Agreement is two separate reasons pointing the same way, not a stronger probability: the figures are not multiplied together.

Same starting five as their last game

Clubs expected to start tonight with the five who started their last game. Each player's average when these five started together (bold), beside his average in the club's other games over the same stretch, and tonight's line under it. Over or Under marks an opportunity: the together average sits 3+ points (1.5+ rebounds or assists, 0.8+ threes, 4+ PRA, 3.5+ points and rebounds or points and assists, 2.5+ rebounds and assists) off tonight's line. A five that has started together often is usually a healthy, settled rotation, so read the difference as minutes and role. Replayed through two seasons, the together average predicted a player's next game about 1% better than his last 20 games did: a small edge in accuracy, not a betting edge on its own. 1 club is expected to change their starting five tonight and are not listed.

Better than their average against tonight's opponent

Players in tonight's rotations whose average against tonight's opponent beats their own average across every game on file, by a real margin (3+ points, 1.5+ rebounds or assists, 0.8+ threes, 4+ PRA) and with a chance under 5%. Every meeting counts, at any club the player was with. 235 player-stat pairs were tested to find these 9, so around 12 would clear the bar by luck alone, and opposing rosters turn over between seasons.

Leans compares his average against this opponent with tonight's main line: over when it is higher, under when lower. Replayed through 2024-25 and 2025-26, players on this list scored no more than their own average in those games, so treat a lean as a description, not a recommendation.

How a split has to earn its place

1. Its own baseline. Compared with that player's rate in that market, not the league's. A centre who grabs 8 rebounds 70% of the time has found nothing by doing it 70% of the time at home.
2. A real margin. At least 12 points above his own rate, or the list fills with restatements of an average wearing a label.
3. Not most of his games. A split covering more than 70% of a player's games is his baseline again under another name.
4. An exact chance figure. Reshuffling a player's results across his own games is hypergeometric sampling, so the tail is computed in closed form rather than simulated.
5. 108,612 cells were tested to produce this list. At a 5% level that is thousands of false positives expected, which is why the per-market null above matters more than any individual row below it.

Does any of it survive a null?

Across every selection, the method beats each player's own rate by +0.8 points. Shuffling the outcomes within each player's own games 200 times produced a mean of 0.06 and never reached 0.8. That edge is real.

And it holds for the strongest single market too, which it did not on two seasons of data. 10+ rebounds at +2.7 is measured against the best-of-18 null — what the top market reaches when eighteen are tested and none of them is real — which averages 1.5 and reaches 2.3 one time in twenty. The real figure clears it, at 1.99%.

The null shuffles each player's outcomes across his own games. His overall rate, his sample size, the market mix and every filter are untouched; the only thing destroyed is the link between a context and a result. 200 shuffles were run, so the smallest chance figure this page can resolve is 0.5%.

Highlighted rows cleared 5% against their own null: 7 of 18, where chance alone would give about 0.9. That is more than chance produces, so something is there.

What was thrown out of the data, and why

The ingest returns every game ESPN lists. 294 of them are preseason or All-Star fixtures — one against a German club side — and are excluded, because thirty minutes in October against non-NBA opposition is not evidence about a January scoring line. 48 rows arrived with no identifiable player. On 12 games ESPN's own player list does not add up to ESPN's own final score, fourteen of the original sixteen involving one club; those games are flagged. Player points were checked against the final scores recorded from a separate feed, and they agree.

Informational only, not betting advice · 18+ · Gamble responsibly. Break-even at standard −110 pricing is 52.4%.