Quiet eye in archers: Gonzalez 2017 study

Thirty people in three groups: ten archers from the British team with at least two years of international competition (34.45 ± 7.52 hours of training a week), ten archers from a university club (4.36 ± 2.83 hours a week) and ten students who had never shot. All completed a computer task; the archers alone also completed a field task, 24 arrows at 30 metres on an 80-centimetre face. Gaze was measured with electro-oculography, and the quiet eye defined as the fixation held within one degree of the target centre for at least 100 thousandths of a second before release.

In the field, the elite archers begin the quiet eye earlier than the novices (t = 3.66, p = 0.002, d = 0.86) while its end does not differ (p = 0.057): the quiet eye therefore lasts 1.9 ± 0.3 seconds against 0.7 ± 0.1 (p = 0.012, partial eta squared 0.35), and is 29.4% of the shot's duration against 18.1%. The elite are also more accurate, with radial error 0.10 against 0.30 degrees and a score of 9.1 against 6.8. The total time from full draw to release was 6.3 ± 0.7 seconds in the elite and 4.4 ± 0.3 in the novices: the experts took longer, not less.

In the computer task participants shot with a joystick at a target on screen, with random disturbance added to the crosshair's movement at two intensities. Here the archers, experts and novices together, have a longer quiet eye than the non-archers and are more accurate, while between experts and novices radial error does not differ. The new finding is the adaptation to the disturbance: only the experts lengthen the quiet eye as the disturbance grows, from 2.6 ± 0.2 to 3.3 ± 0.4 seconds (p = 0.001), bringing its onset forward; the novices go the other way, from 2.3 to 2.0, and the non-archers stay at 1.2 in both conditions.

The correlations between quiet eye duration and error are significant in the field (R squared 0.38, p = 0.006) and in the computer task in both disturbance conditions (R squared 0.41 and 0.45), but they are computed by pooling the groups. The authors write explicitly that their results do not imply a causal relationship between quiet eye duration and performance.

All participants were right-handed, with normal eyesight and no known neurological or developmental conditions. No archer had experience of computer aiming games, and only two non-archers used video games regularly, for less than four hours a week. The study has the university ethics committee's approval and informed consent from everyone. The shooting discipline is not stated.

In the field task, after a warm-up of 12 arrows at 10 metres and 12 practice shots at 30 metres, each archer shot 24 arrows at 30 metres in four blocks of six, with breaks between blocks, seeing where the arrows landed. The 80-centimetre face at that distance corresponds to 1.5 degrees of visual angle. The session lasted about 90 minutes.

The computer task presented a ten-ring target on a laptop screen, viewed from 60 centimetres with the chin on a rest to block head movement, also 1.5 degrees of visual angle. With a joystick held in the right hand, participants pressed a button to make the crosshair and target appear, raised the crosshair by pulling the joystick to simulate the draw, and kept aiming until a green light flashed above the target 6 seconds after movement onset, for 250 thousandths; shooting was allowed once the light had gone, that is from 6.25 seconds, within a 30-second limit. The light existed to prevent a snap shot and was to be seen peripherally, without looking at it. After the button was released, a simulated arrow followed a parabolic trajectory reflecting the positional error at release. The visual disturbance consisted of random movement added to the crosshair's path, generated from thirty random tracks with two different filters, to produce either more predictable or jerkier motion. After 20 practice trials, 60 experimental trials with the two conditions alternating every 15 shots. The session lasted about 60 minutes.

Gaze was measured with an electro-oculogram, with two electrodes at the temples and two above and below the right eye, sampled at 1000 Hz and filtered between 0.2 and 30 Hz, with blinks removed and the gaps stitched by linear interpolation. The authors explain that they chose electro-oculography because the electrodes do not interfere with aiming and because some archers close or squint an eye while aiming, which caused data loss with a worn video tracker. Calibration, repeated after each block, asked participants to fixate the centre and then four lateral points, and the slopes of linear regressions on the voltages gave the equivalent of one degree for each participant.

The moment of release in the field was derived from the EMG of the flexor digitorum superficialis: trial onset at 50% of maximum activation during the draw, release at the peak of the derivative of the signal's envelope. In the computer task, pressing and releasing the button gave the same instants. In both cases the quiet eye was identified by working backwards from release to the point where the signal left the one-degree threshold, and trials with no quiet eye or a duration under 100 thousandths were excluded. Besides the absolute duration, the duration as a percentage of total response time was computed. Performance is the radial error from the centre in degrees of visual angle, plus the 10 to 0 score from the ring hit; in the field the holes were measured by hand on graph paper.

The analysis uses between-participants analysis of variance for the field and repeated measures for the computer task, t tests for onset and offset timings, Pearson correlations and Bonferroni-corrected post hoc tests, with the threshold at 0.05. The sample size calculation, done a priori on pilot data, indicated six participants per group assuming an effect size of 1.0.

The correlations between quiet eye duration and error are computed by pooling the groups: in the field experts and novices, in the computer task all three. Two groups differing in both quiet eye and accuracy produce a correlation even when there is no relationship inside either group, and this study reports no correlations computed within groups. So the R squared of 0.38 in the field largely measures the distance between elite and novices, not a relationship between how long an archer fixates and how well they shoot. The authors are careful and write that the results do not imply a causal relationship; it is worth adding that they do not show a within-person association either.

Between the two tasks there is a gap that matters. In the field the total time from full draw to release was 6.3 seconds in the elite; in the computer task 16 to 20 seconds, because the protocol required at least 6 seconds of aiming and then waiting for the green light. The quiet eye, proportionally, goes from 29.4% of the shot in the field to 17 or 18% on the computer. The computer task, with the chin on a rest, a joystick, an artificial disturbance and an aiming time three times longer, measures something that resembles aiming but is not archery; the adaptation to disturbance, which is the study's new result, appears only there.

Two results from the computer task do not hold together. The text says participants were more accurate in low disturbance, measured as radial error, and a few lines later that scores were higher in high disturbance, with F(1,27) = 111.08 and eta squared 0.81. But the score is the ring hit, that is a function of the radial error: the two measures cannot run in opposite directions on the same shots. In the table, moreover, the novices' scores rise from 6.4 to 7.7 on moving to high disturbance while their radial error worsens from 0.20 to 0.30 degrees. The table's extraction from the PDF is disordered and some value may be out of place, but the contradiction is already in the two sentences of the text.

The degrees of freedom do not add up in several places. In the field there are ten experts and ten novices, so a comparison between the two groups has 18 degrees of freedom, while the paper reports F(1,20) and t(20). In the computer task the group factor has three levels and goes with 2 degrees of freedom in the numerator, as in the interaction effects correctly reported as F(2,27), but the main effects of group are written F(1,27). Both figure captions report «p < 0.5» where 0.05 is meant.

The gaze measurement has two features to keep in mind. Electro-oculography records the eye's movement relative to the head, not the point looked at in space, and the choice is well argued by the authors, but the signal was filtered with a 0.2 Hz high-pass to remove drift: a filter of that kind attenuates components slower than about five seconds, and the quiet eye durations measured here reach 3.3 seconds. It does not follow that the estimate is distorted, but it is a question that would not arise with a video tracker. The quiet eye threshold was also set at one degree, instead of the three of the classic definition, and on a target that itself subtends 1.5 degrees: this study's durations are not directly comparable with those in the literature using the wider threshold.

The sample size remains. Ten people per group, with an a priori calculation assuming an effect size of 1.0, that is a very large one. The three groups also differ in age (26 against 31.1 against 25.5 years) and, among archers, in a training load running from 4.4 to 34.5 hours a week: the difference is not only in skill but in how much time those people spend shooting. In the field there was no non-archer group, so the three-way comparison exists only on the computer. No data are longitudinal, so nothing says that lengthening the quiet eye makes an archer improve: that needs training studies, which this work cites but does not replicate.

The figure to keep is where the difference lies. Between the British team's archers and those of a university club, the moment the gaze leaves the target does not change, which is practically the release in both groups: what changes is when the gaze settles. The experts settle earlier, and that is why the quiet eye lasts 1.9 seconds against 0.7. Anyone wanting to work on this does not need to «hold the eye still for longer afterwards», they need to still it sooner.

The second point is that the experts took longer overall: 6.3 seconds from full draw to release against the novices' 4.4. It is not a speed contest, and a long fixation is not obtained by compressing the rest. The proportion matters too: in the experts the quiet eye took up 29.4% of the shot's duration, in the novices 18.1%, so the difference is not only that the shot is longer.

The most interesting result concerns difficult conditions, and it comes from the computer task. When the crosshair was disturbed by random movement, only the experts lengthened the fixation, from 2.6 to 3.3 seconds, bringing its onset forward; the novices shortened it and the non-archers changed nothing. If the mechanism is the one the authors propose, a long fixation serves to predict where the crosshair will be, and serves more when the crosshair is less predictable. Translated to the field: it is in unstable conditions that the ability to settle the gaze early is worth most, and precisely then that the less experienced do the opposite. It is a coherent hypothesis, on a computer task with a joystick, not a measurement made shooting in wind.

One caution about how this study is cited. That a long quiet eye goes with a small error, here, comes from a correlation computed by pooling different groups: it says that experts have both longer fixations and smaller errors, not that an archer improves by lengthening the fixation. The authors themselves write that the results do not imply a causal relationship. For that question the studies in which the quiet eye is trained are what is needed.

Finally a useful reference point: in the field, at 30 metres on an 80-centimetre face, the international archers scored 9.1 points per arrow and the club archers 6.8.