Illusory Awareness in Professional Table Tennis Players

We tested 10 professional table tennis players in a simple hit–reproduce paradigm designed to probe their awareness of racket pose at ball contact. In each trial, a confederate served a ball and the player hit it to a cued table location (the last third of the table) using either forehand or backhand strokes. Immediately after a hit, the player was asked to recreate, as precisely as possible, the posture in which they believed they had contacted the ball and to hold that position for 2–3 seconds. They then rated how confident they were in this reproduction on a three-point scale (1 = low, 2 = medium, 3 = high). Across the session, each player completed roughly 60–75 valid trials, with both stroke types represented. Throughout, a motion sensor mounted on the racket recorded its 3D orientation. From these data, we derived three interpretable angles—inclination (tilt), heading (where the face points in the horizontal plane), and twist (rotation about the face). For each trial, we paired the reproduction with its original hit, computed absolute angular error on each dimension, and then used mixed-effects models and Bayesian analyses to relate these errors to confidence. A complementary analysis searched within a ±100 ms window around impact to identify which moment in the stroke best matched the reproduced pose. Across trials, reproductions were systematically inaccurate. Linear mixed models on absolute angular error showed that, collapsing across confidence and stroke type, mean inclination error was about 13.73° (95% CI [10.42, 17.05]), heading error about 31.62° (95% CI [26.29, 36.95]), and twist error about 37.58° (95% CI [29.41, 45.74]), all ps < .001. Confidence (low, medium, high) did not reliably predict error for any angle (all ps ≥ .05), and Bayesian follow-ups provided evidence for the null: Bayes factors for the confidence contrasts ranged roughly from 0.03 to 0.08 for medium vs. low and from 0.01 to 0.03 for high vs. low, providing strong evidence for the null hypothesis that conscious representations do not play a role in performance. To identify what moment in the stroke players were actually reproducing, we searched within ±100 ms of each hit for the racket pose that minimised the average discrepancy across inclination, heading, and twist, and recorded its timing relative to ball–racket contact. The resulting time offsets were robustly positive: the mean lag was 27.9 ms (95% CI [21.85, 33.96]), with an overwhelming Bayes factor of about 541021254068.79 in favour of a post-impact match.