Altogether, 22 football players (defenders and/or defensive midfielders) were separated into two groups using a median split approach based on their tactical performance measured under low-workload conditions: high-performing group (N = 11, age: m = 24.75, sd = 6.43 years; performance: m = 61.08, sd = 4.70%), and low-performing (N = 11, age: m = 24.75, sd = 6.43 years; performance: m = 53.29, sd = 2.43%). The player's defensive tactical accuracy (DTA) was assessed based on defensive tactical principles of football defined from the analysis and identification of each player's efficiency in performing the task (Teoldo et al., 2011). This assessment enables the accuracy of the player's position and movement according to a set of spatial references, as well as the analysis and categorisation of the tactical actions. We only evaluated actions related to the defensive tactical principles of play. The scores were calculated considering both negative (five-point scale) and positive actions (ten-point scale) observed during their performance in the 2 vs. 1 + GK game subphase. Previously, researchers have reported reliability values over 0.79 in the analysis of actions (Padilha et al., 2017). The G^*Power version 3.1.9.7 was used to calculate sample size and a minimum of 16 players were required (f ² = 0.34; α = 0.05, and β = 0.95) (Verma and Verma, 2020). The group inclusion criteria adopted were: (i) participants played at professional (Liga Portugal SABSEG) or semi-professional level (Liga 3); (ii) players reported normal levels of visual function.

The procedures were presented to the participants, who then provided informed consent. The study was approved by the Ethics Committee of the host university (protocol number CEFADE.02.2019) and all procedures were conducted in accordance with the Declaration of Helsinki.

The experimental task consisted of 12 trials involving a goalkeeper and three outfield players (i.e., 2 vs. 1 + GK). The 2 vs. 1 + GK game subphase took place on a football field with 27 m length x 20 m width (see Figure 1). These dimensions were defined according to on-field/player ratio provided by the International Football Association Board (Hugues, 1994). This game subphase is considered one of the most prevalent during match-play (see Ramos et al., 2019; Ribeiro et al., 2020).

The experimental task consisted of a free-play situation. The participants were evaluated only on their defensive actions (e.g., disarm, interception, tackle), whereas two attacking players (Liga Portugal SABSEG) performed the offensive actions. The trials started and finished on the command of the researcher. The trials finished either when: (i) the defender recovered ball possession; (ii) the attackers shot at goal; or (iii) when an offside or a foul was observed. To begin a new trial, the attacking players had to return to the starting point, pick up the ball, and wait for the signal to proceed. The attackers were instructed to perform an attack quickly and accurately as in a match situation.

The Yo-Yo Intermittent Recovery Test (level 2) (Bangsbo et al., 2008) was used to identify the workload thresholds. There were two workload test conditions: (i) low-workload–players were required to achieve 60% of their Maximal Heart Rate (HRmax); and ii) high-workload–players were required to achieve at least 90%; both according to the age-predicted HRmax equation (Tanaka et al., 2001; Robergs and Landwehr, 2002).

Players were asked to warm-up before the test (Karsten et al., 2016). The eye-movement system was fitted onto each participant's head. To calibrate the Tobii Pro eye-movement system, the participant had to focus on the centre of the calibration card, held in front of him, for 5 s. To ensure familiarity with the test procedure, players were given three practise trials (Williams and Davids, 1998). An Suunto Ambit3 chest band was worn by each player to measure heart frequency.

Initially, players were assigned to perform Yo-Yo Intermittent Recovery Test (level 2) to achieve at least 60% of HRmax (low-workload condition), immediately following a total of 12 game trials was prescribed to each player. After each trial, they completed the rating scale for mental effort (RSME) and the Borg scale for perceived exertion (RPE). The RSME was used to evaluate the mental effort during the task, this scale required participants to indicate the level of effort associated with task performance. The RSME ranges from 0 to 150, with three verbal anchors corresponding to 0 (not at all effortful), 75 (moderately effortful), and 150 (very effortful) (Zijlstra, 1993). The rating of perceived physical exertion was used with verbal anchors, which comprehended a 15-grade scale ranging from 6 (minimum effort) to 20 (maximum effort) (Borg, 1982).

Players were asked to perform the same task under the high-workload condition during which they performed Yo-Yo Intermittent Recovery Test (level 2) to achieve at least 90% of HRmax and completed the RSME and RPE scales. The individual HR values were monitored during the whole Yo-Yo test (see Figure 1). All trials were performed according to the official laws of the game.

To control for possible learning biases, four attacking players were recruited in the task. They played in alternate pairs for each trial. No feedback was provided during performance.

The distribution of data sets was analysed using Shapiro-Wilk tests. Factorial ANOVAs were performed using Group (high- and low-performing) as the between-participant factor and Workload (high- and low-workload) as the within-participant factor to examine how these factors impacted on performance. Moreover, separate factorial ANOVAs were used to analyse the RSME and RPE data respectively, with Group (high- and low-performing) as the between-participant factor and Workload (high- and low-workload) as the within-participant factor.

The different gaze data measures (NF, NFL, and FD) were analysed using separate factorial ANOVAs with Group as the between-participant factor and Workload as the within-participant factor. The %VT data across the different AOIs were analysed using a three-way ANOVA with Group as the between-participants factor and Workload and AOIs as within-participants factors, respectively. Greenhouse Geisser procedures were used to correct for violations of the sphericity assumption. Any significant main and interaction effects were followed up using Bonferroni-corrected pairwise comparisons and Bonferroni post hoc tests, respectively. Partial eta squared (ηp2) values were provided as a measure of effect size for all main effects and interactions, considering that ηp2 = 0.02 represents a small effect, ηp2 = 0.13 represents a medium effect, and ηp2 = 0.26 represents a large effect (Cohen, 1988). The alpha level for significance was set at p < 0.05.

There were no main effects for Group on both the RSME (F_1,20 = 1.949; p = 0.170; ηp2 = 0.046) and Borg scales (F_1,20 = 0.163; p = 0.689; ηp2 = 0.004). However, there were significant main effects for Workload on the RSME (F_1,20 = 6.360, p = 0.016; ηp2 = 0.137) and Borg scales (F_1,20 = 18.312; p < 0.0001; ηp2 = 0.314). There was no significant Group^*workload interaction on the Borg scale (F_1,20 = 1.884; p = 0.178; ηp2 = 0.045) or on the RSME scale (F_1,20 = 0.397, p = 0.532; ηp2 = 0.010) (see Table 1). The high-performing group achieved higher values on both RSME (p = 0.031) and Borg scales (p < 0.0001) for the high- compared with the low-workload condition. In contrast, the low-performing group presented higher values only on the Borg scale during the high-workload condition when compared with the low-workload condition (p = 0.046). No other significant main effects were reported for scores on the RSME and Borg scales when comparing groups under low- (RSME: p = 0.591; Borg: p = 0.497) and high-workload conditions (RSME: p = 0.160; Borg: p = 0.216).

There were significant effects for Group on NF (F_1,20 = 8.629; p = 0.003; ηp2 = 0.016), NFL (F_1,20 = 5.518; p = 0.019; ηp2 = 0.010), and FD (F_1,20 = 16.862; p < 0.0001; ηp2 = 0.031). Also, there were main effects for the workload conditions on NF (F_1,20 = 9.340; p = 0.002; ηp2 = 0.018) and NFL (F_1,20 = 9.340; p = 0.003; ηp2 = 0.018). There was no significant effect for FD (F_1,20 = 2.04; p = 0.15; ηp2 = 0.004). There were significant Group^*Workload interactions for NF (F_1,20 = 2.482; p = 0.048; ηp2 = 0.015) and FD (F_1,20 = 14.86; p = 0.049; ηp2 = 0.021). There were no main effects for NFL (F_1,20 = 9.171; p = 0.056; ηp2 = 0.007) (see Table 2). Participants in the low-performing group showed differences in the NF and NFL (NF: p < 0.0001; NFL: p < 0.0001) and increased their average FD (p = 0.029) between the low- and high-workload conditions. The high-performing players employed less fixations to fewer locations compared with the low-performing group on the low-workload condition (NF: p < 0.0001; NFL: p < 0.0001; FD: p < 0.0001). On the high-workload condition there were significant differences in FD (p = 0.012) but not in NF (p = 0.585) and NFL (p = 0.631) across groups.

There were no significant main effects for Group (F_5,240 = 0.014; p = 0.905; ηp2 = 0.379) or Workload (F_5,240 = 0.310; p = 0.578; ηp2 = 0.001). In contrast, we observed significant differences on AOIs (F_5,240 = 29.315; p < 0.0001; ηp2 = 0.379). There were significant interactions for Group^*Workload (F_5,240 = 3.661; p = 0.003; ηp2 = 0.071) and Workload^*AOIs (F_5,240 = 3.661; p = 0.003; ηp2 = 071). The high-performing group spent more time fixating on the “SpaceFrontPlayer” than “Ball” (p = 0.034) and “Another Opponent” (p < 0.0001) under both workload conditions. Particularly, players spent less time fixating on “Another Opponent” compared with other AOIs (p > 0.05) in the low-workload condition. For the high-workload, the high-performing players fixated more time on “SpaceFrontPlayer” than “Ball” (p = 0.046) and “Another Opponent” (p = 0.046).

In contrast, the low-performing group spent more time fixating on “SpacePlayer” and SpaceFrontPlayer” than “PlayerBall” (p = 0.001; p < 0.0001), “Ball” (p = 0.016; p < 0.0001) and “Another Opponent” (p = 0.0001; p < 0.0001) on low-workload conditions. In the high-workload condition they spent more time fixating on “SpacePlayer” and “SpaceFrontPlayer” than “Ball” (p = 0.015; p = 0.015) and “Another Opponent” (p = 0.018; p = 0.018) (see Figure 2).

The high-performing group reported higher values on both RSME and Borg scales under high- compared with low-workload conditions. The increase in mental effort observed for the high-performing group might be linked with efforts to maintain performance by investing more resources. Players displayed higher values for RSME on the high-workload condition (high group: 90.27–115.73; low group: 84.09–99.36) (Vickers and Williams, 2007; Vater et al., 2016). Once players invest more mental effort on the task to maintain performance, the resources available to maintain optimal gaze behaviour on the task are reduced. Consequently, the potential decline in performance efficiency, may lead to a deterioration in performance effectiveness, as evident in impaired decision-making and technical-tactical performance during defensive actions (Badin et al., 2016; Smith et al., 2016).

As hypothesised, players exhibited higher values on the Borg scale under high- compared to low-workload conditions. However, there were no significant differences between groups in either condition. Previously, different physical intensities have been reported to create higher physiological stress, leading to fatigue (Casamichana et al., 2012; Casanova et al., 2013). Fatigue is associated with a decline in performance, decreasing the ability to identify relevant information in the game environment, and suppressing irrelevant cues (Gilis et al., 2008; Smith et al., 2016). Probably, for the high-performing players, a greater perceived exertion may not hold the same interconnection with the gaze behaviour processes as depicted in the mental effort evaluation. In the future, researchers could better examine the impact on mental effort and perceived exertion across different skill levels and positional roles.

The low-performing group displayed changes in search strategy across conditions (low-workload: FD: 197.60. NF: 18.30. NFL: 2.94; high-workload: FD: 215.55. NF: 15,94. NFL: 2.50). Notably, the low-performing group decreased the number of fixations and the mean number of fixations per location. In contrast, the high-performing group did not show significant changes on most of the gaze variables across low- and high-workload conditions (low-workload: FD: 232.76. NF: 16.00. NFL: 2.55; high-workload: FD: 228.45. NF: 15.59. NFL: 2.55), albeit they showed a tendency to decrease the number of fixations employed under high-workload conditions.

In agreement with our findings, Badin et al. (2016) suggested that when players are fatigued, their mean FD tends to increase. Such changes in visual search behaviour during the 2 vs. 1 game situations may have been influenced by task complexity and intensity, as shown by changes in effort perception. Also, such differences between groups suggest that high-performing players showed greater selectivity in searching for information cues in more relevant places than their counterparts when performing the task under low-workload conditions (Casanova et al., 2013).

Casanova et al. (2013) reported that elite players employ more fixations of shorter duration than non-elite participants and fixate on more locations at the beginning of each playing half, under low-intensity workloads. However, elite players continued to fixate on a greater number of locations, for the same amount of time, during the high-intensity physiological workload protocol. Previously, researchers have shown that different physiological workloads changes gaze behaviour (Vickers and Williams, 2007), whereas, in our study, no differences between groups were observed under high-workload conditions. While differences were evident when performing under low-workloads, the differences were less evident under high-workloads, implying that the players who show superior decision-making under low-workload may not be the ones that perform better under higher workload conditions. Moreover, during low-workloads conditions, shorter fixation durations were found in our study, which are in accordance with the previous published reports (Williams et al., 1994; Cañal-Bruland et al., 2011; Aksum et al., 2020). Participants with superior tactical expertise may have been able to react quicker and to selectively extract relevant information form the performance environment.

Vaeyens et al. (2007) reported that visual search behaviours vary as a function of decision-making skill, as measured under laboratory conditions. Successful players spent more time fixating on the player in ball-possession of the ball and alternated their fixations to other AOIs (Vaeyens et al., 2007). In contrast, in our study, the high-performing players spent more time fixating on the “SpaceFrontPlayer” when compared with the “Ball”, in both conditions (low-workload: 25–12%; high-workload: 31–11%). However, the high-performing players spent more time fixating “AnotherOpponent” in the high-workload conditions. The fact that our experimental task was designed and developed in-situ may have contributed to these contrasting results. More research is needed to corroborate the present results, potentially involving a direct comparison between in-situ and laboratory-based tests.

As per Vaeyens et al. (2007), in our study players alternated their gaze between “SpaceFrontPlayer” and other locations (“PlayerBall,” “SpacePlayer,” and “Space”), although there were no significant differences among AOIs. Such changes may have been provoked by the fatigue induced by the physiological workload (Vickers and Williams, 2007; Eysenck and Derakshan, 2011) and expertise level (Casanova et al., 2013). In fact, experienced players spend more time fixating in AOIs that they considered more pertinent to their success (Williams and Davids, 1998). Moreover, it may have influenced players' search rate with clear implications on the adjustments of attentional focus away from goal-directed stimuli, potentially impairing anticipation, motor response, and decision-making (Williams et al., 2004).

Additionally, players spent more time fixating on the “AnotherOpponent” in the high-workload than the low-workload conditions. It may have been associated with physiological and mental fatigue, thus affecting processing effectiveness and changing the attentional focus from goal-directed to stimulus-driven attention (Boksem et al., 2005; Ackerman, 2011). An increase in the use of these bottom-up stimulus-driven processes under pressure might lead to threat-related attention and response tendencies (i.e., AnotherOpponent), resulting in reduced attentional control and impairment of the inhibition, shifting functions, and affecting visual perception (Eysenck and Derakshan, 2011).

Our findings provide important information for coaches when designing training sessions. Coaches need to consider how best to manipulate constraints during training to consider the key information cues used by defenders. The main challenge would be to manipulate constraints to replicate the conditions that exist during actual competitive match-play situations. Coaches could manipulate playing space, the number of players involved, and physiological workload to create realistic practise conditions. Such practise environments will allow players to engage in trial-and-error learning and lead to the development of flexible and adaptative gaze behaviours that underpin superior decision-making and creativity (Memmert and Roca, 2019).

Some limitations are worth acknowledging. First, the sample size in this study was relatively small, however, the players were mostly professional level performers. We prioritised the need to secure highly elite participants to generate more insightful findings. Second, we manipulated physiological workload by controlling the percentage of HRmax achieved by the players during both conditions. However, we did not consider the HR values during the entire trial as a continuous variable to verify the physiological impact of the task.

Our findings show that football players employ different gaze behaviours during low- and high-workload conditions. The high-performing group focused more time on the “SpaceFrontPlayer” during both conditions when compared to their counterparts, who elicited “SpaceFrontPlayer” and “SpacePlayers” areas, under the low-workload condition and, “SpaceFrontPlayer” and “SpacePlayers” under the high-workload condition. Furthermore, the results revealed that physiological workload and tactical expertise interact to drive visual search behaviours in in-situ game situations. From a practical perspective, coaches and practitioners should consider how best to adapt their interventions to improve visual search behaviours and performance under different workload conditions. The challenge remains how best to create practise settings that mimic the demands of competition and present desirable difficulties that enable players to search for creative solutions to task demands. Moreover, to improve the quality of instruction, particularly related to gaze behaviour, our study highlights some of the specific visual cues that may be helpful in improving defensive play. In the future, researchers need to continue to consider how gaze behaviours may change as a function of the different constraints imposed during practise and competition.