Abstract
The aim of this study was to examine the differences in visual search behaviour between a group of expert-level and one of novice table tennis players, to determine the temporal and spatial aspects of gaze orientation associated with correct responses. Expert players were classified as successful or unsuccessful depending on their performance in a video-based test of anticipation skill involving two kinds of stroke techniques: forehand top spin and backhand drive. Eye movements were recorded binocularly with a video-based eye tracking system. Successful experts were more effective than novices and unsuccessful experts in accurately anticipating both type and direction of stroke, showing fewer fixations of longer duration. Participants fixated mainly on arm area during forehand top spin, and on hand–racket and trunk areas during backhand drive. This study can help to develop interventions that facilitate the acquisition of anticipatory skills by improving visual search strategies.
Keywords
Introduction
The anticipation of actions in fast ball sports provides a useful natural setting within which to examine the perception of human behaviour 1 and the perception-action coupling.2,3 Differences between experts and novices have been assessed in several studies showing better patterns of recognition 4 and superior anticipatory abilities 5 in expert performers.
In sports, the ability to “read” the opponent’s movements and to respond to them quickly and accurately is a characteristic of expert performers, and this is particularly true in fast ball sports such as racquet sports.6,7 Table tennis is a typical sport in which players have to decide what movement to perform in a very short time, making predictions on the opponent's stroke largely based on information collected before the opponent’s racquet–ball contact. 8 In this kind of sport, expert players are more proficient than less skilled players in “reading” the relevant information about an action in progress 9 and they are typically faster and more accurate in predicting the outcome of an opponent’s action (e.g. the direction of a serve).10–12
To investigate the importance of specific action phases and/or the scrutiny of specific body parts for anticipation, researchers have used temporal and/or spatial occlusion paradigms, as well as eye movement recording techniques (for reviews, see Williams et al. 1 and Vickers 2 ).
Raab et al. 13 documented that to perform a correct stroke (“how” decision, e.g. techniques) in table tennis, participants require a response window of at least 399 ms prior to execution. However, decisions on “what” to do (e.g. tactics) require 556 ms, based on movement durations of about 370 ms (from start of swing to racket–ball contact). These time windows indicate that selection and execution of sequential movements are performed, to some degree, in parallel. The defending player has to predict, as early as possible the information arising from the opponent pre-contact movement pattern, including direction and velocity of the opponent's approaching stroke. 6
Usually, expert tennis players are more able than novices to use information from early stages of the opponent’s action, and anticipation performance deteriorates when either the arm or the arm and the racket are occluded. 14 Moreover, compared to novices, who rely primarily on arm and racket movements, experts also perceive information from proximal body regions, such as hips, shoulders, and trunk.12,15
At the racket–ball contact, additional information can be obtained from the movement biomechanics of the ball-hitting player, in a way that the aim of his/her movements is clearly specified by the kinematic properties. 6 Abernethy 16 reported that experts are not only able to make better predictions of the direction and depth of their opponent's stroke from information available prior to racket–ball contact, but they also extract information from earlier stages in the opponent's movement patterns from which novices are unable to extract any information. Abernethy and Russell 17 studied badminton players with an occlusion technique and found that experts detect information from two particular interest areas (IAs), the racket and the arm that holds it during the game. In contrast, novices detected information from the racket only.
A considerable amount of research has focused on gaze control during visual search performance of racket sports.7,3,15,18–20 Most of those studies have investigated the athletes’ gaze behaviour in response to a service, mainly analysing the trajectory of the ball directed towards the experimental subject, in order to see if experts kept their gaze on the ball.7,20 Results showed that experts did not track the ball throughout its entire flight path, but rather at the very beginning of its trajectory. 20 According to the flight of the ball, expert cricket players 21 made a saccade to the anticipated bounce point of the ball 100–200 ms before the ball bounces. It is not reported if experts would be able to intercept the ball or make a saccade to the anticipated ball bounce point if the initial part of its trajectory is not shown, so that the only visual information given to athletes are derived from the opponent’s movements.
Furthermore, many papers investigating visual search strategies in sports have tried to identify differences in gaze behaviour comparing different levels of expertise to identify which visual search strategy obtained the best results. Few researchers have examined whether successful performers employed different visual search patterns compared to unsuccessful performers within a group of presumably similar levels of expertise. 22 The need to identify intra-group differences is important as not all experts are at the same level in a particular sport even if they play in the same team. To be able to make such a comparison, one has to create groups using a within-task criterion such as the number of penalties saved in a soccer match. 22 A within-group comparison, therefore, may disclose subtle differences in visual search behaviour that may help to reveal the determinants of successful performance which cannot be identified relying exclusively on expert – novice comparisons.
In this research, players carried out an anticipation skill task consisting in predicting the outcome of the most common types of stroke in table tennis: forehand top spin (FT) and backhand drive (BD). In these two techniques, the IAs of the performer (trunk, head, arm and racket) are differently displayed in the visual field, more close together during BD, peripherally spread in the FT (see Figure 1). Since we expect that participants will fix mainly the hand–racket area without overlooking the opponent player, we hypothesise that during BD participants will perform fewer fixations of longer duration with respect to FT technique. This could be due to the greater distance between the opponent's body centre and his hand–racket position in FT than during BD. Successful performance in interceptive tasks depends on the acquisition of visual information about the approaching object. We tried to establish which areas contribute more, and in which time period, to the accurate prediction of the ball direction, even when the initial part of the ball trajectory is not shown. Furthermore, with this study we may be able to determine if differences are not only expertise-related but also success-related, according to the fundamental kinematics of the opponent's movement pattern. Players were classified as successful or unsuccessful based on their performance on a film-based test of anticipation skill. This within-group comparison would help to resolve the present debate regarding the key predictive source(s) of information used by players when attempting to anticipate the direction of the ball.
Participants’ point of view. (a) Backhand drive and (b) forehand top spin technique.
Thus, the aims of the present study were: (i) to examine whether there are differences in visual search behaviour between a group of expert-level and one of novices table tennis players; and (ii) to compare the gaze strategies of successful experts, unsuccessful experts and novices.
Method
Participants
Twenty-five participants, 15 students (4 female and 11 male, mean age, years and SD: 26.8 ± 3.73) and 10 elite athletes were recruited for the study. The expert players were male (M age = 23.10 years, SD = ±7.58) and had played table tennis at a professional level for an average of 10 years (±2.60). All participants were instructed about experimental procedures and informed consent was obtained before testing began. All subjects had normal or corrected-to-normal vision and none of them reported any uncompensated visual deficit. The experimental protocol was approved by the Institutional Ethic Committee of the University of Bologna.
Stimuli and procedure
A professional table tennis coach was filmed, from participants’ perspective, with a digital video camera (Casio® 300 frames/s, with a max resolution 1280 ×960 pixels), while responding to a ball emerging from a throwing machine (Speed 120 km/h). We filmed 20 strokes, equally subdivided in 10 BD (Figure1(a)) and 10 FT (Figure 1(b)), each of which were further subdivided in five strokes directed to the right and five to the left quadrant of the opponent’s game field.
The experiments were performed in a darkened room. Participants sat on a chair with their head placed on a chin rest. Stimuli were presented with a retro video projector (Epson EB-W12, 720 × 486 resolution; frame rate 60 Hz) positioned 300 cm away from a translucent screen. The screen covered 38 × 29° of visual field and was placed 180 cm from the subjects’ eyes.
The first 2000 ms of the video showed the coach in a preparation phase. Then, the ball appeared in the lower portion of the screen (subject’s point of view), bouncing on a portion of the coach’s table. The coach responded returning the ball to the left or the right opposed hemi-field. The video was stopped at the contact of the ball with the coach’s racket and the experimental subject had to predict, as soon as possible, in which hemi-field the ball would have been returned, signalling the response by a two-button gamepad.
Twenty clips were randomly selected and shown 10 times to every participant, who responded to 200 clips. Overall, a total number of 3000 videos were presented to novice group, and a total number of 2000 clips to expert group. Randomisation was kept in the same order for each participant.
Apparatus
Eye movements were recorded binocularly with a video-based eye tracking system (EyeLink® II, SR Research Ltd, Mississauga, Canada). The equipment consists of two miniature cameras mounted on a leather-padded headband. Pupil tracking was performed at 500 samples/s, with high spatial resolution <0.005° and noise limited to <0.01°. Data were encoded using a software (Eyelink Data Viewer) that allows displaying, filtering, and presenting the results. Only data regarding the right eye were analysed for this work. Gaze behaviour consisted of gaze fixations, defined as the time the eyes remained within 3° of visual angle around the target for a minimum duration of 99.99 ms. All events corresponding to eyelid occlusion (blinks), to very small pupil size and to missing or severely distorted images, were discarded. Events that occurred 100 ms before or after a blink were also discarded, because such periods are probably due to partial blinks, where the pupil is partially occluded.
We analysed the gaze positions in relation to the positions of the moving objects using dynamic interest areas (IAs; see paragraph below), that represent the position and dimensions of important visual scene content over time.
Data analysis
Anticipation test
The ability to make accurate predictions from advanced sources of information was measured as follows.
Response accuracy
The percentage of trials in which the subject’s response was correct.
Key-press response time
The time (ms) from the coach’s racket–ball contact to the button pressed by the participant.
Gaze behaviour was measured as:
Search rate. Mean number of fixations per trial, that is the total number of fixations in a trial divided by the duration of the same trial; Mean fixation duration per trial. Average duration of all fixations that occur in a trial; Mean viewing time. Mean time participants spent fixating their gaze on each displayed interest area when trying to anticipate the ball’s direction. For this purpose, the screen was divided into seven IAs: (i) IA-1, called head (2.0° × 2.0°), included the coach’s head; (ii) IA-2, called trunk (3.6° × 4.0°), included the coach’s trunk; (iii) IA-3 called hand–racket (1.4° × 0.4°), included the coach’s hand and the racket (iv) IA-4 called arm (1.37° × 1.94°), included the coach’s arm from the side of the racket; (v) IA-5 called ball (1.0° × 1.0°); (vi) IA-6 called right quarter (9.5° × 1.0°), included the right portion of the table on the side of the coach (subjects’ point of view), (vii) IA-7 called left quarter (9.5° × 1.0°), included the left portion of the table on the side of the coach (subjects’ point of view). All fixations outside these IAs were referred to as “out” fixations. We have chosen these IAs to determine whether the important cues for success are the opponent’s body parts, the racket–ball, or both as suggested by other authors12,15 With the inclusion of the table’s right/left portion on the coach side, we intended to verify if the participants watch the bounce point of the ball.
21
Response accuracy, as percent of correct responses, was analysed with 2 × 2 × 3 mixed design analysis of variance (ANOVA), in which stroke technique (FT, BD) or stroke direction (left, right) were the within-subjects factors, and expertise (S_Experts, U_Experts and novices) the between-subjects factor.
Key-press response time, search rate, and mean fixation duration were analysed separately by 2 × 2 ×2 × 3 mixed design ANOVA, in which stroke technique (FT, BD), stroke direction (left, right) and response accuracy (correct, incorrect) were the within-subjects factors, expertise (S_Experts, U_Experts and novices) the between-subjects factor.
Mean viewing time-dependent measures (mean fixation durations and numbers on each IAs) were analysed by 8 × 2 × 2 × 2 × 3 mixed design ANOVA, in which IAs (head, trunk, hand–racket, arm, ball, right and left quarter of the table and out), stroke technique (FT, BD), stroke direction (right, left), and response accuracy (correct, incorrect) were the within-subject factors, and expertise (S_Experts, U_Experts and novices) the between-subjects factor.
Mauchly's test was considered for each variable to assess assumptions of sphericity. If assumptions of sphericity were violated, the Greenhouse–Geisser epsilon corrections of degrees of freedom were used. 23 Post hoc unpaired t-test with Bonferroni’s correction was used to examine in which experimental conditions groups were different. Paired sample t-test was used for interaction effects.
A cumulative method was used to determine the probability that a stroke could be intercepted by randomly selecting from one of two possible directions. It indicates the number of correct responses verified in a group in relation to the total number of given responses. The cumulative method is a proportion, and it is calculated dividing the number of correct responses with the number of total responses (correct/incorrect), taking into account the two kinds of strokes. The cumulative method has shown that, only if a participant correctly predicts at least 140 out of 200 strokes, the probability that a gambling strategy was adopted is 5% or less. 22
Data were analysed with SPSS v13.0 (SPSS, Chicago, IL, USA). Means were considered significantly different at p < 0.05.
Results
Pre-processing data
Responses with RTs shorter than 150 ms and longer than 1000 ms were discarded (early or delayed responses). After pre-processing data, we used 1710 videos for experts (out of 2000 videos) and 2671 for novices (out of 3000 videos).
On the basis of these performance scores, two groups of experts were created for further analyses. Figure 2 shows the four participants who succeeded in more than 140 correct response (>70%) so that this group was labelled as successful experts (S_Experts), the other participants formed the group of unsuccessful experts (U_Experts).
The expert table tennis players ranked according to the percentage of correct (black bars) and incorrect (light bars) responses. The horizontal axis indicates the subject’s code number.
Response accuracy
ANOVA showed a significant main effect for expertise (F2,22 = 16.07, p < .001, ηp2 = .59), stroke technique (F1,22 = 6.96, p = .015, ηp2 = .24), and stroke direction (F1,22 = 4.91, p = .037, ηp2 = .18). A Bonferroni post hoc test confirmed that S_Experts showed more correct responses in comparison to U_Experts (p < .001) and novices (p < .001). All participants during FT and strokes directed to the right revealed more correct responses than during BD and strokes directed to the left (Figure 3). No interaction effects were observed.
Response accuracy. Histograms represent percentage values (±SE) of correct responses, between groups (S_Experts, U_Experts and novices), stroke techniques (forehand top spin -FT-; backhand drive -BD-), and between-stroke directions (left; right).
Key-press response time
ANOVA showed a significant interaction effect of stroke technique × response accuracy (F1,22 = 4.50, p = .045, ηp2 = .16). Paired samples t-test showed that participants responded faster to FT than BD in correct trials [t (24) = 2.50; p = .020] (M = 473 ± 39 vs. 482 ±24 ms). However, in incorrect trials they responded faster to BD than FT [t (24) = 3.40; p < .001] (M = 451 ± 25 vs. 496 ± 25 ms). We did not find significant differences among groups (S_Experts =470 ± 53; U_Experts = 438 ± 43; Novice = 518 ± 27 ms, mean ± SE).
Search rate and mean fixation duration
Mean fixation duration (ms) and search rate between groups, and between forehand top spin (FT) and backhand drive (BD) techniques (mean ± se).
Mean fixation duration on each interest area
For fixation duration, ANOVA indicated a significant main effect of interest area (F7,154 = 31.30, p < .001, ηp2 = .59) and stroke technique (F1,22 = 7.25, p = .013, ηp2 = .25). Analysis also demonstrated an interaction effect for interest areas × stroke technique (F7,154 =38.36, p < .001, ηp2 = .64), interest area × response accuracy (F7,154 = 4.00, p < .001, ηp2 = .15), and interest area ×stroke technique × response accuracy × expertise interaction effects (F14,154 = 1.95, p = .025, ηp2 = .15). The between-subjects analysis showed no significant difference.
For both techniques, participants fixated mainly on the hand–racket area. Paired sample t-test for interest areas ×stroke technique interaction effects showed that participants looked at the arm [t (24) = 8.14; p = .000] during FT and at the trunk [t (24) = 9.61; p = .001] and hand–racket [t (24) = 4.74; p < .001] during BD. For interest areas × response accuracy interaction effects, analysis demonstrated that participants looked at the hand–racket [t (24) = 2.99; p = .006] when the responses were correct.
Post hoc tests with Bonferroni correction revealed that, during BD, when the responses were correct, S_Experts had longer fixations on the hand–racket area compared to novices and U_Experts (p < .001) (Figure 4). Right and left quarters of the table tennis table were not inserted in the histogram due to the lack of fixations on them.
Mean fixation duration on each interest areas in correct/incorrect responses. Histograms representing mean duration values (±SE) for each interest area during forehand top spin (upper panel) and backhand drive (lower panel), in correct (C) and incorrect (IN) responses, across groups (S_Experts, U_Experts and Novices).
Discussion
The aims of the present study were to examine whether there are differences in visual search behaviour between a group of expert-level and one of novice table tennis players, and to compare the gaze strategy of successful experts with that of unsuccessful experts and novice players.
As expected and in line with a previous study, 22 successful experts had superior anticipatory performance compared to unsuccessful experts and novice players. All participants responded more accurately during FT than during BD, and when the strokes were directed to the right part of their visual field. This could be explained by the fact that FT is the main hitting stroke in the modern game of table tennis, as showed by statistical results of the most important world competitions. BD counts for about 45% of all strokes made during one table tennis match (excluding the serve), since it is a difficult shot to make. It could be that, being less used to this kind of stroke, participants have more difficulty in intercepting it compared to FT.
In our experiments, there were clear differences in visual search across groups. The S_Expert players used a search strategy involving fewer fixations of longer duration. It means that they are able to reduce the amount of information to be processed, or require fewer fixations to create a coherent perceptual representation of the display. 24 As we have hypothesised, all participants showed fewer fixations of longer duration when they responded to BD in comparison to FT technique. One possible explanation could be that in FT the position of the racket is farther from the coach’s body centre, so that a higher search rate is necessary to explore the areas of interest. On the contrary, it is more likely that the subjects use parafoveal vision throughout BD, since IAs are more gathered together.
Mean viewing time analysis on correct responses revealed that, during BD, successful experts make longer fixations on the hand–racket area compared to novices and unsuccessful experts. Participants, in both correct and incorrect responses, showed longer fixation durations on the trunk and the hand–racket areas during BD and on the arm during FT. Moreover, during both techniques, and when the responses were correct, all subjects fixated for longer time on the hand–racket area.
With reference to the ball interest area, previous research4,20,21,25,26 reported that participants kept their eyes on the ball early in flight but not during the final portion of its trajectory. In the current study, participants did not fixate the ball but directed their gaze mainly to the coach’s body, since they do not see the initial portion of the ball flight after the racket–ball contact. As it was documented in other ball sports, individuals directed their gaze towards the position of the opponent’s body areas prior to direct their eyes on the ball. 27 Piras and Vickers 3 found that, during soccer penalty kick, if goalkeepers’ final fixation was too long on the ball, then goals occurred, while a longer fixation on the visual pivot (a point between ball, kicking leg and non-kicking leg) was a characteristic of saves. When a fixation was maintained on the ball, valuable information was missed from the kicking leg action; however, when a visual pivot was used, the goalkeepers were better able to detect the type of kick being delivered and anticipate the ball direction.
In summary, the visual searching scheme, in addition to determining eye movement patterns and fixation locations, can be used for assessing anticipatory capabilities. Based on gaze strategy, we demonstrate that S_Experts are more effective than novices and U_Experts in accurately anticipating the type of stroke (forehand or backhand) and its direction (left or right). Similar results have been found in tennis by Singer et al. 28 and in soccer by Helsen and Pauwels. 29 It would be expected that experts obtain contextual-specific information, and, due to previous similar experiences, they are able to anticipate the action accurately. Given the position of the racket during FT, participants focused more on distal cues (i.e. arm) to anticipate location of the ball. During BD they focused on more proximal cues (i.e. trunk and hand–racket) to guide their anticipatory responses, “anchoring” the fovea in the middle of the coach’s chest and probably using the parafovea and visual periphery to pick up relevant information. 3 This is confirmed by the fewer number of fixations of longer duration during BD.
Conclusion
The results of this study highlight several issues useful for scientists interested in further investigation about perceptual-motor skill acquisition, as well as coaches interested in designing anticipatory and gaze behavioural training programs to improve sport-specific attention capacity. This task could be organised to allow athletes to experience repetition of key situations from table tennis in a shorter space of time than they would normally experience when actually playing. The aim is to train anticipation, highlighting the links between important environmental or opponent cues and outcomes. For example, temporal and spatial occlusion methodologies have been used to train anticipation and decision making in athletes. In Jones and Miles 30 study, professional coaches were able to pick up early information emanating from opponent movements, which led to significantly more accurate predictions in conditions in which racket–ball contact was occluded. Otherwise, spatial occlusion could involve removing particular areas or information sources from the opponent, such as hand–racket. It enables researchers and coaches to infer which body region provides information that cannot be picked up elsewhere, through decrements in anticipation occurring when that body region is occluded. However, this does not necessarily mean that the body region or cue in isolation is critical. It may be the removal of the cue that distorts or removes the relative motion between regions of the body. 31
The current study demonstrates that the differences in anticipation skill between successful and unsuccessful players are associated with differences in visual search behaviour consisting in the extraction of information from arm areas during FT, and from trunk and hand–racket areas during BD. Our findings suggest that situational probabilities influenced by different actions, and the ability to extract information from an opponent’s postural orientation, may influence perceptual-motor skills necessary for anticipatory processes and, in racket sports, it may be more important than the ability to identify patterns of play. Within a specific task, situation constraints influence the relevance of each perceptual-cognitive skill, and visual behaviours employed by athletes vary significantly based on task constraints such as the participant’s specific functional role (defender vs. attacker), or the type of technique (FT vs. BD). For example, the position of the player in the field may determine the influence of each perceptual-cognitive skill (i.e. central vs. lateral position in table tennis) as well as the distance between the player and the opponent/ball. The above factors often influence the time available to formulate a response as well as the cost and benefits associated with accurate and inaccurate judgments. 32 This study can help to develop interventions that facilitate the acquisition of anticipatory skills by improving visual search strategies.
Footnotes
Acknowledgments
Authors are grateful to Dr. Andrea Giovanardi for technical assistance in the experiments.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is supported by University of Bologna and Italian Ministry for University and Scientific Research (MIUR).