Craps Roll Speeds Sag 12% Before Dice Rotation, Not After
The data from 14 major US casino floors, compiled from 4.1 million resolved pass-line bets over a 14-month period, shows that shooter roll speed declines by an average of 11.8% in the five throws immediately preceding a seven-out, with the most pronounced deceleration occurring on throws three and four of a hand. The slowdown is not a post-hoc artifact of dice rotation or table congestion; it is a pre-rotational signal that becomes statistically detectable an average of 2.7 throws before the dice leave the table. This finding, which has been replicated across three independent pit audits, suggests that the physical rhythm of a craps hand is not random, and that the game’s most basic bet—the pass line—may be operating on a timing substrate that players and casinos have only partially understood.
The Methodology Behind the 11.8% Figure
The study, conducted by a consortium of table-game analytics firms and two university statistics departments, pulled data from RFID-equipped craps tables in Nevada, Pennsylvania, and New Jersey. The sensors tracked not just the outcome of each roll, but the time interval between the shooter’s release of the dice and the dice coming to rest, along with the shooter’s pre-roll preparation time—the period from when the dice are returned to the shooter to when they leave the hand.
The 11.8% decline is measured specifically on release-to-rest time, not on the shooter’s overall cadence. In the first three rolls of a hand, the average release-to-rest interval was 4.3 seconds. By rolls six through nine, that interval had compressed to 3.9 seconds. But then, on the final three rolls before a seven-out, the interval expanded to 4.8 seconds. The variance between rolls was not gradual; it was a cliff. The data shows a 9.2% acceleration in the middle of a hand, followed by a sharp 11.8% deceleration in the terminal phase.
This is not a small sample. The 4.1 million resolved bets represent roughly 220,000 individual hands, with an average hand length of 8.4 rolls. The researchers controlled for table minimums, time of day, and dealer shift changes. They also controlled for the obvious confounder: dice that leave the table or hit the chip rack. Those rolls were excluded, and the deceleration signal remained.
The key detail is that the slowdown precedes the seven-out by an average of 2.7 rolls. In a game where the house edge on the pass line is a fixed 1.41%, a pre-rotational timing signal of this magnitude is not a strategy edge—it is a behavioral artifact. But it raises a question that the casino industry has historically avoided: if the dice are physically thrown, and the thrower’s rhythm is measurable, is the game truly random in the way the regulatory framework assumes?
The Physics of a Hand: What the Sensors Actually Catch
To understand why roll speed sags before the dice rotate, you have to look at what happens to a shooter’s body over the course of a hand. The RFID data was paired with overhead camera analysis that tracked shoulder angle, wrist flexion, and grip pressure on the dice. The cameras captured 60 frames per second per table, and the researchers coded each throw for three variables: backswing height, release point, and follow-through distance.
The pattern is consistent across all 14 properties. On rolls one through three, shooters exhibit a mean backswing height of 22.4 inches and a release point that is an average of 18.3 inches from the table’s back wall. By rolls five through seven, backswing height drops to 19.1 inches, and the release point moves forward to 15.7 inches. But on the final two rolls before a seven-out, backswing height collapses to 16.8 inches, and the release point drops to 13.2 inches.
This is a physical fatigue signature, but it is not the same as muscle exhaustion. The average hand lasts only 8.4 rolls, which is under two minutes of active throwing. The researchers measured grip pressure using pressure-sensitive inserts in the dice (a method that required special regulatory approval), and the data shows that grip pressure increases by 14.3% on the final rolls. The shooter is not losing strength; they are tightening up.
The tightening manifests as a shorter backswing and a lower release point, which produces a slower dice velocity. The average velocity of the dice at release on rolls one through three was 11.2 feet per second. On the final three rolls, it drops to 9.8 feet per second. The slower velocity changes the number of bounces the dice take before coming to rest—from an average of 4.1 bounces in the early hand to 3.2 bounces in the terminal phase—which changes the distribution of outcomes.
Here is the numerical anchor that matters: across the 4.1 million resolved bets, the seven-out rate on rolls where the shooter’s release-to-rest time exceeded 4.6 seconds was 17.2%, versus 14.9% on rolls where the interval was under 4.0 seconds. That is a 2.3 percentage point swing, which is nearly double the house edge on the pass line. The effect is not large enough to beat the game, but it is large enough to be real, and it is large enough that the casino’s own surveillance systems have started flagging it.
Why the Casino Floors Haven’t Acted on This
The industry response to this data has been muted, and for good reason. The first reason is regulatory. Every major gambling jurisdiction in the US operates under a statutory assumption that craps is a game of pure chance. The dice are manufactured to tolerances of 0.0005 inches, the tables are leveled daily, and the drop boxes are sealed. The idea that the shooter’s behavior could systematically alter the outcome, even by a fraction of a percent, is a liability that no regulator wants to acknowledge.
The second reason is operational. If a casino were to publicly acknowledge that roll speed is a measurable predictor of a seven-out, it would have to decide what to do with that information. Banning slow shooters is impossible—they are not breaking any rule. Adjusting the pass line odds is a regulatory change that would require months of approval. And training dealers to subtly rush shooters who are showing the deceleration signature would be a visible change in table behavior that players would immediately notice.
The third reason is that the data, while statistically robust, is not actionable in a live setting. The 11.8% deceleration is only identifiable in hindsight, after a hand has ended. The researchers tried to build a real-time model that would predict a seven-out on the next roll, using the current roll’s timing data. The model achieved a 51.3% accuracy rate—barely above chance, and far below the threshold needed to justify any kind of intervention.
But the fact that the timing signal exists at all has already changed how some casinos think about table rotation. Three properties in the study have quietly changed their stickman rotation schedules, moving the stickman to the other end of the table more frequently during long hands. The stated reason is "dealer fatigue management," but the internal memos that were obtained through a public records request in Pennsylvania reference the roll-speed data explicitly.
The Player Side: Do Regulars Know?
The pit data is one thing, but the player behavior is another. The study also included interviews with 47 "regular" craps players who each logged at least 200 hours at the tables during the study period. Of those 47, 31 said they could "feel" when a hand was about to end. They described it as a shift in the shooter’s energy, a change in the sound of the dice hitting the table, or a sense that the shooter was "forcing it."
The researchers tested this perception. They showed the 47 players video footage of 40 hands—20 that ended in a seven-out on the next roll and 20 that continued for at least three more rolls—without telling them which was which. The players correctly identified the upcoming seven-out 58.7% of the time. That is not a huge edge, but it is statistically significant at the p<0.01 level, and it is far above the 50% baseline.
The players were not using any conscious strategy. When asked to explain their choice, they gave vague answers: "he looked tired," "the dice felt flat," "she was rushing." But their subconscious was picking up on the same timing signals that the RFID sensors were recording. This raises an uncomfortable question for the casino industry: if a subset of players can feel the deceleration, even without knowing the numbers, is the house edge on the pass line actually as fixed as the regulators claim?
The practical answer is no, but only in a narrow sense. The 2.3 percentage point swing in seven-out rates on slow rolls does not translate into a player edge because the player cannot know, in the moment, whether a given roll is a "slow" roll. The timing data is only meaningful in aggregate. A player who tries to bet against the shooter when they sense a slowdown will lose money over time because the base rate of seven-outs is still 14.9%, and the timing signal is too noisy to act on in real time.
But the fact that the perception exists is a marketing problem. The casinos have spent decades selling craps as a pure game of chance, where every roll is independent and the dice have no memory. If a meaningful minority of players believe they can detect the end of a hand, that belief undermines the game’s core appeal. The casinos’ response has been to double down on the randomness narrative, which is why the roll-speed data has not been published in any industry trade journal. It exists in internal audits and in the academic paper that is currently under peer review at a statistics journal.
What the Deceleration Means for the House Edge
The house edge on the pass line is 1.41%, and that number is computed under the assumption that every roll has an identical probability distribution. The roll-speed data does not change that assumption in a way that matters for the average player. But it does change the variance profile of the game. A hand where the shooter is decelerating is more likely to end, which means the distribution of hand lengths is not purely geometric. It is a mixture of two distributions: a "fresh" distribution for the first few rolls and a "fatigued" distribution for the later rolls.
The researchers modeled this as a two-state Markov process, where the shooter transitions from a "fast" state to a "slow" state with a probability that increases with each roll. The model fits the observed data better than a pure geometric distribution, with a log-likelihood improvement of 4.7% across the 220,000 hands. Under this model, the effective house edge on the pass line remains 1.41% over the long run, but the conditional house edge on the seventh roll of a hand is slightly higher than the house edge on the second roll.
This has implications for how casinos structure their table limits. Several properties in the study have already moved to a "roll count" pricing model for their highest-limit tables, where the maximum bet on the pass line is reduced after the fifth roll of a hand. The official explanation is that "long hands increase the theoretical hold per table," but the internal data shows that the reduction is a direct response to the deceleration finding. The casinos are not trying to stop the slowdown; they are trying to cap their exposure on the rolls where the seven-out is most likely.
The long-term question is whether this data will lead to a change in how craps is regulated. The Nevada Gaming Control Board has not commented on the study, but the board’s technical staff has requested the raw data for review. A source within the board, speaking on condition of anonymity, said that the request is "routine" and that "no regulatory action is contemplated." But the request itself is notable, because the board does not typically ask for raw data from academic studies unless it is considering a change to the game’s certification standards.
The Open Question: Is the Dice Rotation the Cause or the Symptom?
The title of this article makes a specific claim: roll speeds sag before the dice rotation, not after. The data supports that claim. The deceleration is measurable in the throws preceding the seven-out, and the seven-out itself is just the terminal event. But the deeper question is whether the deceleration is a cause of the seven-out or a symptom of something else.
The researchers lean toward a psychomotor explanation. As a shooter becomes more invested in a hand—as the pass line accumulates more odds, as the table gets louder, as the pressure to repeat a number builds—their motor control degrades. The grip pressure increases, the backswing shortens, and the release becomes less fluid. The dice are not being thrown with less force because the shooter is tired; they are being thrown with less force because the shooter is thinking too much.
If that explanation is correct, then the seven-out is not a random event. It is the physical manifestation of a cognitive state. And that would mean that craps, at its core, is a game that punishes confidence. The more a shooter believes they are on a hot streak, the more likely they are to tighten up and seven out. The data supports this: the deceleration is most pronounced on hands where the shooter has already made their point twice, which are the hands where the table is most likely to be cheering.
The implication is not that players should try to stay relaxed—that is the kind of folk wisdom that has existed at craps tables for decades. The implication is that the casino’s own house edge may be partially a function of the players’ psychology, not just the dice’s physics. If that is true, then the 1.41% pass line edge is not a fixed constant. It is a number that fluctuates based on the emotional state of the person holding the dice.
That is a difficult thought for the industry to sit with, because it suggests that the game is not as purely random as the regulatory framework assumes. It also suggests that the casinos, by creating an environment that heightens tension—loud music, bright lights, aggressive dealers—may be inadvertently increasing the deceleration signal. The question is whether they will ever admit that, or whether the roll-speed data will remain buried in internal audits, waiting for a journalist to file a public records request. The 11.8% number is out there now, and it is not going back in the box.