When an NBA Player Is Listed “Questionable,” Do They End Up Playing?
We count what share of players listed as “Questionable” actually end up playing. Whether the market gets this uncertainty right is the question for the next step.
- Result status
- Preliminary results
- Last updated
- Start date
- Duration
- Day 22 of 21
- Research team
- Mysports.AI Research Team
- Plan registered
NBA “Questionable” Players: Key Findings
Injury statuses come from the NBA’s official injury report (NBA.com); whether a player played comes from the NBA’s official game stats. 2024-25 and 2025-26 regular seasons: the player played in 1,423 of 2,559 cases (range 53%–58%). Key player = averaging 20+ minutes over the last 10 games. Other websites have published similar overall figures, so this part is not a blind test.
- Almost no “Doubtful” players played (1.3%); almost all “Probable” players did (97.3%).
- Players first listed as “Questionable” less than 12 hours before tip-off played only 47% of the time; those listed earlier played 60% of the time.
- Among players still listed as questionable two hours out, 98.6% of those finally cleared played; none of those finally ruled out played in this sample.
- Odds: when the home team has one more “Questionable” key player than the away team, we can’t see which way the odds will move before tip-off (on average only +0.01 percentage points of implied win chance, range −0.23 to +0.25). This doesn’t mean the market got it right — only that this method can’t see a pattern.
How to use this yourself
Open the NBA’s official injury report page (official.nba.com) and click the latest report of the day (a PDF). Times on the report are US Eastern; since the mid-2025-26 rule change, a new report comes out roughly every 15 minutes. Open source ↗
Find the game (Matchup column) and the player (Player Name column), then look at the “Current Status” column: Questionable means the player may or may not play. Other labels: Probable = likely to play, Doubtful = unlikely to play, Out = will not play, Available = cleared to play; a player not on the report is normally available.
Check the latest report again about 2 hours before tip-off. If the player is still Questionable, treat it as “a bit more than half the time he plays” (we found about 56%), not as a sure yes or a sure no.
Go back through earlier reports to see when he was first listed: players first listed less than 12 hours before tip-off played less often (about 47%) than those listed earlier (about 60%). Exception: for the second game of a back-to-back, teams often list players later because of the reporting rules, so don’t automatically lower his chances in that case.
If the last report before tip-off changes him to Available, he almost always plays (98.6%); if it changes him to Out, he doesn’t play. Also, Doubtful players almost never played (1.3%) and Probable players almost always did (97.3%).
These rates only cover key players averaging 20+ minutes over their last 10 games in the 2024-25 and 2025-26 regular seasons. They are past averages and guarantee nothing for any single game; use them as a starting point, then consider the player, the injury and when he was listed, rather than applying one fixed percentage to every player. We also could not see that “Questionable” predicts which way the odds move before tip-off, so don’t treat it as a betting signal.
In plain words
Why study this?
“Questionable” is the most uncertain and also the most common status on the injury report. Turning it into a playing rate helps the model estimate each game’s lineup more accurately.
What do we want to know?
The NBA’s official injury report lists players with statuses such as “Out,” “Doubtful,” “Questionable” and “Probable.” First, we want to know: what share of players listed as “Questionable” actually end up playing? Does it differ by how important the player is, or how long before the game it is? The next step is to check whether the market gets this uncertainty right.
How do we study it?
We used the official injury reports already parsed in R1 with their release times (the reporting frequency changed; after the mid-2025-26 rule change, roughly every 15 minutes) and player participation records to compute playing rates. Polymarket prices are for the team winning, not for whether a player plays, so this version only asks whether the pre-game status predicts later odds moves; it does not infer playing chances from the odds.
From questionable to actual appearance
- 01Record pre-game status
At a pre-specified time, record whether the player is listed as questionable.
- 02Check the game record
Count players who appeared and those who did not; do not select only appearances.
- 03Calculate appearance rate
Appearances divided by all eligible player-game observations; separate games are counted separately.
- 04Compare conditions
For example, compare time to tip-off and player importance.
This explains the method; see Key findings above for the actual rates. A team’s chance of winning is not a player’s chance of playing.
What have we found?
About 56% of key players listed as questionable two hours before tip-off played; doubtful players almost never did and probable players almost always did. 47.3% of players first listed less than 12 hours out played, versus 59.8% of those listed earlier. On odds, we could not see that the status predicts later moves.
How does this help with predictions?
Treat “questionable” as “a bit more than half will play” rather than certain either way. In this sample, players first listed less than 12 hours before tip-off played less often overall—but the second game of a back-to-back is often listed late because of the reporting rules, so do not lower the chance across the board. Players finally cleared almost always play.
What can’t we say yet?
Polymarket only has odds on “will the team win”; there is no market on “will the player play.” So this version does not judge whether the market gets playing chances right. Seeing no change in the odds only means this method couldn’t see one (a change needs to be about 0.34 percentage points or more to show up). Other websites have published some playing rates, so part of our numbers is not a blind test.
Terms used on this page
- Questionable
- A status on the NBA injury report meaning it is still uncertain whether the player can play.
- Team win odds
- The odds on Polymarket that “a certain team will win,” which roughly show the chance people think that team has of winning. They are not odds on whether a player will play. A price of $0.60 ≈ a 60% implied win chance ≈ odds of about 1.67; lower odds = more favored.
Detailed results
Share who actually played, by injury status 2 hours before tip-off
Key players (averaging 20+ minutes over their last 10 games), 2024-25 and 2025-26 regular seasons.
Status source: the NBA’s official injury report (NBA.com)｜Calculated following the plan written in advance｜Status taken at the moment 2 hours before tip-off, and already posted for at least 15 minutes｜“Listed early” = first listed 12 hours or more before tip-off; “listed late” = less than 12 hours｜Played = has an appearance in the official game stats (including under 1 minute).
HypothesesFor deeper reading
- H1Being listed as “Questionable” 2 hours before tip-off predicts how the odds change until tip-off → not seen (+0.01 percentage points of implied win chance, range −0.23 to +0.25)Insufficient evidence
- H2The “Doubtful” status predicts changes in the odds → not seenInsufficient evidence
- H3The “Probable” status predicts changes in the odds → a little when looked at alone, but it doesn’t hold after correcting for testing several things at once; the two seasons also point in different directionsInsufficient evidence
- H4The playing rate for “Questionable” differs by how important the player is → slightly higher for stars (57% vs. 54%), but it doesn’t hold after the correctionInsufficient evidence
- H5Second night of a back-to-back differs → supported in the model, but both raw rates are about 56%. A split added after seeing the results shows 536 of 540 back-to-back cases were listed late; this helps explain the model gap but does not show back-to-backs themselves change playing ratesSupported in this sample
- H8Players first listed as “Questionable” less than 12 hours before the game have a different playing rate → holds (47% vs. 60%)Supported in this sample
Further analysis
Further analysis done after the main results. Items marked “post-hoc” were added after seeing the data and are for reference only; items marked “pre-registered” or “Extension 1” had their rules written and timestamped before anything was calculated.
- Whether the player played last game matters a lot (post-hoc exploration): among key players listed "Questionable", 65.7% of those who missed their team's previous game (possibly returning) ended up playing, versus only 47.4% of those who had played the previous game.
- The group most likely to sit (post-hoc exploration): players who played the previous game and were newly listed "Questionable" less than 12 hours before tip-off (not the second night of a back-to-back) played only 31.8% of the time — about one in three.
- A more accurate play-probability model (post-hoc exploration): adding "played last game" and "games missed in a row", choosing the model on one season and testing on the next, lowered the prediction error (Brier score, lower is better) from 0.1557 to 0.1512. The model is now frozen and will be re-tested on the 2026-27 season. However, on two seasons R7 never used (2022-23 and 2023-24), the improved model was barely more accurate than the original (differences of +0.00002 and −0.0004, no clear difference); most of the gain comes from using the pre-game injury report at all.
- Broadly consistent with an outside site: using isheplayingtonight.com's published method as closely as our data allow, the "Questionable" play rate is 46.3% in our data versus 48.6% on that site, and the team-by-team ranking is similar.
Limits of this analysis
- These splits and model features were thought of after seeing the results, so the 2025-26 model test is not blind; firm conclusions wait for the pre-registered 2026-27 checks.
- "Played last game" only uses official box scores and cannot tell injury, coach's decision or a trade apart.
- Our count of listings is about 17% higher than the outside site's; the reason is unknown.
NBA Player Status: Research Methods
Research questions
- What share of players listed as “Questionable” actually end up playing? Does it differ by how important the player is, whether the team played the day before, or how long before the game it is?
- Can the injury status 2 hours before tip-off predict how the odds change from then until tip-off?
Method summary
- Playing rate: key players (averaging 20+ minutes over their last 10 games) in the 2024-25 and 2025-26 regular seasons. We take their status on the official injury report 2 hours before tip-off and check the official game stats for whether they played.
- Odds: 1,091 games in the 2025-26 season, comparing Polymarket odds 2 hours and 5 minutes before tip-off. For every game, we checked that the final odds reading came before the game actually started.
- There is only one main test (H1). The other five are corrected together for testing many things at once, to avoid “try enough times and something will hit by luck.”
- We hand-checked 40 player-games against the game stats on the NBA’s official website: 0 errors. All six data checkpoints passed.
Data and limitations
Data we use (all public and free)
- Official NBA injury reports (NBA.com; compiled by R1)
- NBA player game logs (NBA.com / stats.nba.com)
- Polymarket odds for single NBA games
- Actual tip-off check (first version): game records from ESPN’s public API (time of the first play; about 2,780 requests; no longer fetched)
Limitations
- Polymarket only has odds on “will the team win,” so this version neither estimates nor judges what the market thinks a player’s chance of playing is.
- We have one odds reading every 5 minutes, with no buy and sell prices; only changes of about 0.34 percentage points or more can be seen.
- The trading volume cutoff for the price sample uses total volume collected after the games, so it can’t be used directly as a rule for picking games before they start.
- Other websites have published some playing rates that overlap with our sample, so the playing rates are not a blind test.
Latest progress
Latest：10/29: a shared pre-registered add-on (timestamped) for the player-news studies—the 2026-27 re-check will also report a version using official tip-off times
Next step：Re-check in 2026-27 with the frozen rules (decision 2027-04-14); the original J1 is expected to stop at a data gate, in which case the clearly labelled “J1 alternative” is reported instead
Study timeline
- Novelty checked; plan finalized and timestamped after two rounds of AI-tool code review
- Data organized; all six checkpoints passed
- Main analysis done; corrected after AI-tool code review
- Publish after confirming data licensingIn progress
- Check again on new 2026-27 games
Update log
Version history
- v1.1-forward
Forward addendum a1 timestamped: if the original J1 stops at a data gate it is reported as “not done”, and a separately named “J1 alternative (without verifying actual tip-off)” is reported instead; J2 and J3 are unaffected
- ext1
Further analysis (post-hoc) and a 2026-27 prospective check plan, timestamped
- v1.0-results
Main results are in; after an AI-tool code review, we fixed the report’s wording (not describing “no drift seen” as “the market got it right”) and added a code check
- v1.0
Study plan finalized and timestamped (this first version does not judge whether the market gets playing chances right)
Study record
d687a2aa3c048468c1436f58d688e1a9b5a1f776689653f5dcf8dac215b8007a049743c91b6bcbe505f26cc381ea4c6ef415d4ee2bd06f53ab621e35cb9723dbcd57532497b90c6ffb27a4691c4bfaacff57cddb30f69fc02380a0d883e6c5d0d1f4c8aef4da8fe36bbb48a112749c400705a6b79d39e6377692a65033ddfa26bb14c6886d787b81196a12071d5b2aed89248777ccaba5798956827e18c6e812This page is a research log for reference only and is not betting advice or a guarantee of profit.