In November 2015, the journal Communications of the ACM published a study titled “Hidden In-Game Intelligence in NBA Players’ Tweets.” Instead of arguing about a player’s mental state based on a highlight reel, three researchers tried to measure it from what players wrote and check it against the box score.
| Item | Detail |
|---|---|
| Published | Communications of the ACM, November 2015 |
| Researchers | Chenyan Xu (Stockton University), Yang Yu and Chun-Keung Hoi (Rochester Institute of Technology) |
| Season | 2012-13 NBA season |
| Players | 353 active players on Twitter, about 78% of the league |
| Tweets | 91,659 |
| Mood score | AFINN word list, which rates words from -5 to +5, plus emoticons |
| Window | 7 p.m. the day before to 7 p.m. on game day. The league bars players from social media starting 45 minutes before tipoff. |
| Performance | Box-score stats, including shooting percentages, plus/minus and minutes |
What They Found
The central finding was specific enough to be useful. Players whose pre-game tweets read as more positive tended to perform better that night, and the paper concludes that players’ tweets are a real source of game-performance information.
| Test | Result |
|---|---|
| Pre-game mood vs. that night’s performance | Positive link, coefficient 0.17, statistically significant |
| Same test with controls such as salary and home or road | Positive link, coefficient 0.16, statistically significant |
| Away games | Mood mattered more |
| Higher-paid players | Mood mattered more |
The study did not claim to have solved player psychology. It claimed, with a published method anyone can check, that there is a measurable, non-random relationship between a player’s expressed mood and his performance.

How Studio Coverage Handles the Same Question
Compare that to how sports media often handles the same underlying question: is this player in a good headspace right now, and should that change how we evaluate what we just watched?
A common approach is a reporter reading a facial expression in a postgame interview, or a studio panel speculating about a player’s mentality off a single tense exchange with a teammate, and then running that speculation as if it were established fact for the next 48 hours of coverage. That approach has no published method and no way of being wrong that anyone has to account for.
The ACM study, whatever its limitations, comes with a stated sample size, a stated time frame and a claim that other researchers can try to replicate or challenge.
The Limits
Those limitations are real. The study covers one season, and an association is not proof that mood causes performance. Its sample is built entirely from public tweets, which are a narrower and more curated signal than a player’s actual internal state. The data are from 2012-13, so the finding may not describe today’s players, and this piece does not claim it has been replicated.
The Gap That Matters
This is not an argument that sentiment analysis should replace journalism. It is an argument that the gap between “researchers measured a season of tweets and published a specific, checkable finding” and “a panel guessed at a player’s mood from one camera angle and ran with it for a week” is the same gap that separates evidence-based coverage from the five-minute outrage cycle.
A player’s mood is hard to know, but it can at least be approximated with a disclosed method. That takes someone willing to do the measuring instead of performing a reaction on camera.

