NBA Player Prop Hit Rates, and How to Predict Them

What our points, assists, rebounds and double-double props hit this season, graded in public, and the method behind the points projection.

The short answer. An NBA player prop is a bet on one player's stat line (points, rebounds, assists, or a double-double) against a number the sportsbook sets. To predict one, start with the minutes he will play, then his scoring or passing rate per minute, then adjust for pace, rest and who is out, and ask how often his recent games clear the line. Our graded alt lines, set below the sportsbook's main line, hit 90.3% for points (647 graded, Nov 1 to Nov 20, 2026) and 91.1% for assists (731 graded, a replay: the rule was set on Nov 16 and run back over every night from Nov 1, and tracked forward from Nov 17). At the sportsbook's main line, our points plays hit 51.8% (677 graded) from Oct 20 to Nov 21, 2026.

Every number on this page is a hit rate: how often the pick happened, graded the next morning against the box score. It is not a return, and none of it says a bet is better than its price. The numbers update each morning from the same record as the Alt lines tab and the NBA performance page.

From our data: what our NBA props hit this season

Alt lines (Nov 1 to Nov 20, 2026)

An alt line is a lower line than the main one: “over 18.5 points” for a player whose main line is 22.5. Lower lines clear more often, so the hit rates below are high by design. The useful question is whether the model's stated % matches what happened.

StatHow the line is setRecordHit rateThe model said (average)
PointsThe highest over line the model puts at 85% or more, at least 2 points under the sportsbook's main line584-6390.3% (647)86.4%
Assistsreplay3 under the sportsbook's main line (DraftKings), with the model's % at that line. Replay: rule set Nov 16, run back from Nov 1666-6591.1% (731)82.0%
Rebounds †4 under our own rebounds line (no sportsbook line)827-23977.6% (1066)no % given

† Rebounds are graded against a line our model sets itself, from the player's own recent games, because our odds feed carries no rebounds line. It is a real hit rate, but not against a sportsbook's number, so it does not compare with the rows above. On these rebounds rows our average projection was 7.38 rebounds and the average actual was 5.88: the projection ran high. Window: Nov 1 to Nov 20, 2026.

About the assists row. It is a replay, not a forward record: the rule (3 under the main line) was set on Nov 16 and run back over every night from Nov 1; nights from Nov 17 on are tracked forward. The rule is defined from the sportsbook's main line, but the model never takes that line as an input; the line only picks which rung is graded. The rule it replaced, where the model chose the line, went 310-35 (89.9%, 345 graded, model said 89.1%) over Nov 1 to Nov 16, 2026. The points rule was fixed before Nov 1 and has been tracked forward since.

Stated vs actual, assists by the model's % (replay), Nov 1 to Nov 20, 2026

Model's % at the lineRecordHit rateModel said (average)
Under 70%85-1188.5% (96)61.7%
70-80%169-2288.5% (191)75.8%
80-85%121-695.3% (127)82.6%
85-90%109-1488.6% (123)87.5%
90% and up182-1293.8% (194)94.4%

Read it plainly: on these two graded rules, the stated % has not run high. Below 85% the model undersold the line, by 12 points or more; from 85% up it was close (within 2 points). That is not true of every alt line. When we checked the full alt-line ladder in late October, lines the model put at 85-95% hit noticeably less often than it said, and calls right on the sportsbook's main line ran well under the model's %. That is why only these rules are graded and shown in the Alt lines tab, and why the board never prints a hit figure on the book's own line.

These lines still miss. 20 nights is a short window, and a few misses in one night are expected.

Main-line play tier (Oct 20 to Nov 21, 2026)

MarketWhat counts as a playRecordHit rate
Rebounds †model 65%+ against our own line141-5571.9% (196)
Double-doublemodel 60%+66-3068.8% (96)
Assistsmodel 65%+ at the main line466-30160.8% (767)
Pointsmodel 65%+ at the main line351-32651.8% (677)
Triple-doublemodel 30%+4-7 so farsmall sample

† Same disclosure as above: our own line, not a sportsbook's.

Points at the main line is close to a coin flip, and it was last season too. That is the honest headline: the main points line is the most heavily bet prop in the sport, and our projection mostly lands where the book already is. The alt lines above ask a different question of the same projection: how low is his floor, rather than over or under the book's own number.

What is a player performance double?

A double-double is 10 or more in two of the five counting stats (points, rebounds, assists, steals, blocks) in one game. Sportsbooks list it with the player performance props as a yes/no bet. The most common route is points and rebounds for a big man, or points and assists for a lead guard. A triple-double is 10 or more in three.

It is a minutes-and-role bet more than a shooting bet: a player averaging 18 points and 9.5 rebounds is one rebound away most nights. Our double-double plays (model 60%+) are 66-30 (68.8%, 96 graded) from Oct 20 to Nov 21, 2026. Last season they went 93-53 (63.7%, 146 graded).

How to predict NBA player points, in four steps

This is the method that runs on our points board every game day.

  1. Measure recent form, not the season average. Keep the player's last eight games and split scoring into two rates: minutes per game and points per minute. A player at 34 minutes and 0.65 points a minute is a different prop from one at 28 minutes and 0.8, even though both average about 22. Keep the spread of his points too; it becomes the width of the forecast. Games under 20 minutes are dropped, and players averaging under 15 minutes are not projected.
  2. Project a mean with a regression. An XGBoost regressor, trained on this season's box scores, takes seven inputs: the two rates, rest days, back-to-back, home or away, and the opponent's defensive rating and pace. It outputs a projected points total. Seven inputs is deliberately few; a bigger model on player-game data memorises last month.
  3. Put a range around it. A projection of 24.1 is not a pick. The model wraps it in a normal distribution as wide as the player's recent spread, with two corrections our graded record forced: the width never drops below 60% of his full-season spread, and it is multiplied by 1.5 (added May 14, 2026, after 90%+ calls were found to hit far less).
  4. Score the line. The chance of the over is the share of that range above the line. Half a point is added first, so “over 24.5” means 25 or more. The same sum is repeated at every line from 5.5 to 40.5, which is where the alt-line %s come from.

The model never takes the sportsbook's line as an input. The line only comes in afterwards, as the number the projection is graded against. (The assists alt-line rule is defined from the sportsbook's main line, 3 under it, but that line only picks which rung is graded; it never goes into the projection.)

What moves a prop (all markets)

  • Minutes first. A starter at 36 minutes has a quarter more chances than one at 28. Track the last five games' minutes, not the season figure; a coach quietly extending a role shows up in the box score before the line.
  • Who is out. When a team's second scorer sits, the main option's shots go up. When the starting point guard sits, the backup runs the offence for 34 minutes. Check the NBA injury report in the 90 minutes before tip-off; our board is built before the day's games and is not updated for late scratches.
  • Pace. A 230-point total means more possessions for everyone than a 210-point grind.
  • Blowouts. A 12-point favourite's starters may sit the whole fourth quarter, and every over on that team dies together. It is the most common reason overs miss.
  • Rest. The second night of a back-to-back trims minutes, most for older players and big men.

The markets, briefly

  • Points: the most popular prop and the hardest to read at the main line (see the table above).
  • Rebounds: position, minutes and the opponent's misses. Our board sets its own line because no sportsbook line reaches our feed; hold it against your book's number.
  • Assists: pace, teammates' shooting and who else handles the ball.
  • Double-doubles and triple-doubles: role props, covered above.
  • PRA (points + rebounds + assists): three stats in one line, so one quiet category can be made up by another. Suits all-round players more than pure scorers. We do not publish PRA picks; add up the individual projections.

Last season, for reference (Mar 12 to Jun 13, 2026, playoffs included)

MarketWhat counts as a playRecordHit rate
Rebounds †model 65%+ against our own line196-8769.3% (283)
Double-doublemodel 60%+93-5363.7% (146)
Assistsmodel 65%+449-34256.8% (791)
Pointsmodel 65%+463-45450.5% (917)
Triple-doublemodel 30%+11-10small sample (21)

† Rebounds: graded against our own line, not a sportsbook's.

See tonight's alt lines

Tonight's NBA props board has the Alt lines tab (points, assists and rebounds, one line per player with the model's %), plus the full points, assists, rebounds and double-double boards with each player's last ten games against the line. For members. Every line is graded the next morning and added to the record above.

See the NBA props board →

Read the methods

Frequently asked questions

What are your NBA player prop hit rates?

From Nov 1 to Nov 20, 2026, our points alt lines hit 90.3% (647 graded). Assists alt lines hit 91.1% (731 graded, Nov 1 to Nov 20, 2026), but that is a replay: the rule was set on Nov 16 and run back over every night from Nov 1, and tracked forward from Nov 17. At the main line from Oct 20 to Nov 21, 2026, assists plays hit 60.8% (767 graded), double-doubles 68.8% (96 graded), points 51.8% (677 graded). Rebounds are graded against our own line, not a sportsbook's.

What is a player performance double?

A double-double: 10 or more in two of points, rebounds, assists, steals and blocks in one game. Sportsbooks list it with the player performance props as a yes/no bet. A triple-double is 10 or more in three of them.

How do you predict NBA player points?

Project the player's minutes and points per minute from his last eight games, adjust for rest, home or away and the opponent's defence and pace with a regression, put a range around that projection using his recent spread, then measure how much of the range sits above the line.

Why is points the hardest NBA prop to read?

It is the most heavily bet player prop, so the line already reflects minutes, pace and matchup, and scoring is the noisiest thing a player does. Our points plays at the main line hit 51.8% (677 graded) this season (Oct 20 to Nov 21, 2026) and 50.5% (917 graded) last season (Mar 12 to Jun 13, 2026).

What is a z-score in a player prop?

The distance between the line and the projection, measured in the player's recent spread. Zero is a coin flip; a line one spread below the projection gives roughly an 84% over.

What does PRA mean in NBA props?

PRA is points, rebounds and assists added together into one line. A player averaging 22 points, 8 rebounds and 6 assists has a PRA around 36. One quiet category can be made up by another, so PRA suits well-rounded players more than pure scorers. We do not publish PRA picks; add up the individual projections.

Do alt lines hit as often as the model says?

On the two lines we grade, as often or more so far: points said 86.4% and hit 90.3% (647 graded, Nov 1 to Nov 20, 2026); assists said 82.0% and hit 91.1% (731 graded, a replay: the rule was set on Nov 16 and run back over every night from Nov 1, and tracked forward from Nov 17). Across every rung of the alt-line ladder, no: high %s ran above what happened, so only the graded rules are shown in the Alt lines tab.

How do you grade a prop?

Every published play is settled against the official box score the next morning, by player and date. A player who does not play is no result, not a loss. Losses stay on the record, as they do on the performance page.

How do injuries change NBA player props?

A starter ruled out moves minutes and shots to teammates, and their lines can lag the news. A player back from injury is often on a minutes limit. Our board is built before the day's games and is not updated for late scratches, so check the injury report before tip-off.

The week ahead, once a week

Last week's graded record — wins and losses — and the plays for the week ahead. One email, no daily noise, unsubscribe in a click.

Responsible Gambling Disclaimer: Sports betting involves risk and should be done responsibly. Never bet more than you can afford to lose. Prediction Engine provides data-driven analysis for educational and informational purposes only. Hit rates are historical and do not promise future results. If you or someone you know has a gambling problem, call 1-800-GAMBLER or visit ncpgambling.org. Must be 21+ and located in a jurisdiction where sports betting is legal.