Fight, Not Flight! Avoidant Goals Strengthen Attentional Biases During Increased Anxiety in Healthy Adults

Abstract

Heightened anxiety can impair perceptual-motor performance, with detrimental effects potentially arising, in part, from threat-related attentional biases and interpretations. Shifting from a flight (i.e., threat) mindset to a fight (i.e., challenge) mindset may be an effective strategy for coping with anxiety and improving performance on perceptual-motor tasks. In this context, the present study examined how differences in fight-or-flight behavioral goals, represented by hit or avoidance actions, influence attentional control in response to threatening stimuli during heightened anxiety. Healthy adult participants performed a visual probe task, with half responding to a probe target in hit mode and the other half responding in avoidance mode. Anxiety levels were manipulated using the threat-of-shock (ToS) method, which significantly increased the participants’ anxiety. Participants with avoidance goals exhibited significantly delayed responses when avoiding a target in the presence of threat-related stimulus cues under the ToS condition. Conversely, no changes in response times were observed between the ToS and no ToS conditions in those with hit goals. These results suggest that when anxiety is heightened, avoidance goals induce attentional biases toward threat-related stimuli. In conclusion, encouraging avoidance of potential threat-related sources as an action mode may be counterproductive for coping with heightened anxiety, at least in healthy adults. The study’s principal implication for clinical practice is that adopting fight-like behavioral goals in response to threats may be an effective strategy for managing anxiety in everyday life.

1. Introduction

Individuals with anxiety tend to pay increased attention to threats and are more likely to interpret emotionally ambiguous stimuli as threats [1,2]. Threat-related attentional shifts and interpretations may also occur in healthy individuals when their anxiety levels are heightened [3]. Consider, for example, a football player taking a penalty kick in the FIFA World Cup Final—a situation involving intense anxiety and psychological pressure. The kicker’s attention may be involuntarily drawn to the goalkeeper, or they may interpret the goalkeeper as more threatening than they intend to [4]. Conversely, task goals that involve intentionally avoiding the goalkeeper may ironically intensify automatic attentional shifts toward the goalkeeper, possibly resulting in worse motor performance [5,6].

In general, our attention is intentionally and consciously directed toward task-relevant properties based on current behavioral goals; this is the case in top-down or goal-directed processes [7,8]. When anxiety is heightened, attention can be automatically and unconsciously drawn away from task-relevant information and shifted toward threat-related information; this is the case in bottom-up or stimulus-driven processes [9,10]. In addition to shifting our attention toward threats, anxiety enhances the tendency to interpret task-irrelevant properties of our environment as threatening [11]. Thus, it is reasonable to hypothesize that actions taken to intentionally avoid threat-related stimuli can influence not only the degrees of attentional shifts toward potential sources of threats but also the extent to which task-irrelevant stimuli are perceived as threatening.

The phenomenon of attentional shifts toward threat-related stimuli, referred to as “attentional bias toward threat”, has been extensively studied, particularly in the context of anxiety disorder treatment. Emotional cues that threaten or induce fear in an individual, such as angry faces [12,13], aggressive words [1,3], and aversive sounds [14,15], are frequently used in these studies. Previous research has demonstrated that individuals with high trait anxiety or anxiety disorders tend to exhibit either facilitated attentional engagement (i.e., their attention is more quickly drawn to threats) or disengagement difficulty (i.e., difficulty in shifting attention away from a threat to another stimulus) [3,16,17,18]. Conversely, individuals with low trait anxiety tend to exhibit the opposite bias—avoiding threats—although this tendency is weaker and less consistent [1].

In addition to altering attentional control, heightened anxiety can impair behavioral control, which, in turn, may diminish motor performance. For instance, high levels of state anxiety affect eye gaze control during motor tasks, leading to less efficient visual search behaviors and shorter visual fixation durations [9,10,19], which likely impede the extraction of task-relevant information. High levels of state anxiety also interfere with automated movement patterns, such as walking and running, converting them into more conscious, risk-avoidant, and conservative patterns [20]. Such behavioral changes are likely associated with neural mechanisms activated in response to perceived threats—i.e., actions taken to protect one’s own life—and subconsciously cause freezing and fight-or-flight behaviors [21].

In summary, heightened anxiety can impair attentional control and perceptual-motor performance, potentially because of threat-related attentional biases and interpretations. Effective strategies for coping with anxiety are crucial to prevent the decline of perceptual-motor performance in healthy individuals under conditions of increased anxiety. To the best of our knowledge, few studies have explored the relationship between anxiety and attentional bias in individuals with low trait anxiety [22], and none have examined these relationships under the condition of temporarily elevated anxiety among healthy adults.

The biopsychosocial model of challenge and threat (BMCT) [23,24,25] offers valuable guidance on developing methods for coping with anxiety and improving perceptual-motor performance. Generally, individuals tend to approach stimuli with positive valence and avoid stimuli with negative valence [26,27]. Emotions also influence whether individuals are more likely to freeze or move. These two kinds of motivated behavior are referred to as approach/avoidance behavior and behavioral freezing/activation [28].

According to the BMCT, adopting a “challenge” mindset when facing negative emotional stimuli may increase one’s tolerance for anxiety, while adopting a “threat” mindset may increase one’s vulnerability to anxiety [29]. If the degree and/or direction of attentional bias are influenced by whether one confronts or avoids threat-related stimuli through voluntary (or goal-directed) attentional control [30], then shifting from a flight (avoidant) to a fight (challenge) action goal may be an effective strategy for coping with anxiety [29].

With these general issues in mind, the present study examined how avoidant and hit action goals affect automatic attentional control toward threatening stimuli during experimentally induced anxiety. A threat-of-shock (ToS) method was used to manipulate state anxiety in healthy participants as they performed a visual probe task. Previous studies have validated the effectiveness of the ToS method in inducing anxiety in healthy individuals [31,32,33]. In a typical visual probe task trial, emotional cues (e.g., angry or happy faces) are briefly presented, followed by a probe target that replaces the cues. Attentional bias toward an emotional cue is inferred from how quickly the participant responds to the probe. As Sharpe et al. [34] argued, conventional visual probe tasks may not fully distinguish between the two components of attentional bias involved in anxiety: facilitated engagement and disengagement difficulty. This issue may be resolved by requiring participants to either hit or avoid the probe target as their response mode.

Given that heightened anxiety causes attention to be drawn more quickly to threat-related stimuli, we predicted that, under ToS conditions, participants would more quickly hit a probe target appearing in the same location as an angry face (a threat-congruent condition) than a probe appearing in the opposite location (a threat-incongruent condition); this was Hypothesis I regarding facilitated engagement [12,13]. Additionally, since heightened anxiety makes it more difficult to disengage one’s attention from threat-related stimuli, we predicted that, under ToS conditions, participants would respond more slowly when avoiding a probe target in the threat-congruent condition than in the threat-incongruent condition; this was Hypothesis II regarding disengagement difficulty [14,15,17]. Finally, assuming that heightened anxiety leads to emotionally ambiguous stimuli being perceived as more threatening, we predicted that, in a ToS setting, attentional bias would more likely be directed toward neutral faces than toward happy faces, which carry less threatening valences; this was Hypothesis III regarding threat-related interpretation [2,35].

As we will show, the results of this study support Hypotheses II and III pertaining to disengagement difficulty and threat-related interpretation in the context of the avoidant action mode. This study highlights that interventions that encourage the avoidance of potential threat-related stimuli as an action mode may be counterproductive for coping with heightened state anxiety, at least in healthy individuals.

2. Materials and Methods

2.1. Participants

Thirty-six university students (19 males and 17 females aged 20–28 years, $\overline{X}$ = 21.6 years) participated in this experiment after providing written informed consent. All participants reported normal or corrected-to-normal vision and no anxiety disorders. The experimental protocol was approved by the research ethics committee of the authors’ institution. The requisite sample size was determined by using the G*Power program [36] to conduct a one-tailed paired t-test to determine the difference between safe and threatening conditions. We supposed that α = 0.05, that the power (1 − β) = 0.80, and that the effect size as represented by Cohen’s d = 0.65.

The participants were randomly divided into two groups: one performed a task involving hitting, and the other performed an avoidant task. The participants completed the trait scale included in the Spielberger State-Trait Anxiety Inventory—Japanese Version [37] prior to beginning the experiment. There was no significant difference in trait anxiety scores between the two groups (t[34] = 1.11, p = 0.273, Cohen’s d = 0.35), as $\overline{X}$ = 50.4 and the standard deviation (SD) = 8.3 in the hit group, while $\overline{X}$ = 46.8 and SD = 10.7 in the avoidant group. The group means did not significantly differ from those of the general Japanese population, as µ = 48.2 and σ = 10.0 [37] in the hit group (Z = 0.92, p = 0.358, 95% confidence interval [CI] [45.8, 55.0]) and the avoidant group (Z = −0.59, p = 0.556, 95% CI [42.2, 51.5]).

2.2. Apparatus and Materials

This study followed the ToS procedure delineated in previous studies [31,32,33]. A portable constant current stimulator (USE-100, Unique Medical, Komae, Japan) was calibrated to deliver bipolar electric pulses with a duration of 5.0 ms and a frequency of 5.0 Hz. The stimulation intensity varied between 4 mA and 10 mA. In the experimental blocks, exposure to the stimulus was limited to one bipolar pulse per block to minimize any physiological risks to the participants. We used a watch-type heart rate (HR) monitor (Polar Vantage M, Polar, Finland) and three visual analog scales (VASs) (representing the participants’ levels of fear, anxiety, and happiness) to assess the degrees of threat manipulation.

To execute the visual probe task, the participants pressed three buttons on a hand-held response box. The three buttons, each of which had a diameter of 20 mm, were arranged in a horizontal row with gaps of 25 mm. During the task trials, each participant pressed the buttons with their index finger while resting their palm on the device. The participants’ responses to manipulating the buttons were acquired at a sampling rate of 1 kHz.

A fixation point, face images, and a probe image were presented in a light-gray target frame against a black background. Images of faces were collected from the FACES database developed by the Center for Lifespan Psychology, Max Planck Institute for Human Development [38]. Three face images representing emotions with angry, happy, and neutral valences were selected, which represented two male models (IDs 008 and 114) and two female models (IDs 048 and 140). In total, 12 images were prepared. Irrelevant backgrounds surrounding the faces in the photographs were removed, and all pictures were transformed into grayscale images. A silhouette of a human figure was used as a probe target to indicate the start signal and the direction of the response actions. All visual stimuli were displayed on a 27-inch LCD system with a refresh rate of 60 Hz. During the experiment, each participant faced the screen at a distance of 1.2 m. Thus, the visual angle of each face image was approximately 5.8° × 7.5°, with a 13.5° disparity between the centers of the two face images.

2.3. Experimental Design and Procedure

The experimental task employed a block design with two ToS conditions (i.e., no ToS/safe and ToS/threat), each of which consisted of three blocks. No shocks were delivered in the three safe blocks. Electric shocks were delivered after the tenth trial in the first threat block and after the thirtieth trial in the second threat block. No shocks were administered in the third threat block. Since the two trials after the electric shocks had to be excluded from the data, two additional trials were added at the end of those blocks. Each of the six blocks contained forty trials, resulting in 240 trials in total. The safe and threat blocks were alternated, and the order of the sequence was counterbalanced across the participants within each task group.

Figure 1 illustrates the sequence of stimulus presentations and motor responses in each trial. The illustration exemplifies a threat-congruent trial in which the location of a threat valence cue (i.e., an angry face) is spatially congruent with the location of the probe. Each trial began with a beeping sound that directed each participant to keep pressing the center button (“Standby”, the home button) on the response box. The fixation point was then displayed at the center of the target frame. It remained visible until each participant pressed either the right or left button (i.e., the response buttons). Five hundred ms after the presentation of the fixation point, a pair of emotional cues (i.e., a pair of angry and neutral faces or a pair of happy and neutral faces made by the same model) were simultaneously presented on the left and right sides of the fixation point. They remained visible for 200 ms.

In the next step, the probe target (a silhouette of a human figure) appeared on either the left or right side of the fixation point and remained visible for 150 ms. The interstimulus interval was fixed at 350 ms. The hit task group was instructed to hit the human silhouette as quickly as possible by pressing the response button that spatially corresponded to the side on which the silhouette was presented (represented by a yellow arrow in Figure 1). The avoidant task group was instructed to avoid the human silhouette as quickly as possible by pressing the response button on the opposite side of the silhouette (represented by a red arrow in Figure 1). The side on which the emotional cues and probe target appeared, along with their spatial combinations (i.e., congruent or incongruent), were randomized within each block of the ToS and no ToS conditions.

Before beginning the “hit or avoid” task, each participant was asked to decide which hand they would use to manipulate the response device. The HR monitor was attached to their contralateral wrist. Two disposable electrodes were placed on the surface of their dorsal skin above the third proximal and third intermediate phalanx bones of their contralateral hand. For each participant, the intensity of the sock was calibrated to a level that was “quite unpleasant/uncomfortable but not painful”, following the established protocol [33]. The participants were informed that electric shocks might arrive during some of the trials within the threat blocks, but never within the safe blocks.

While executing the task, each participant was restricted to using their index finger alone to press the buttons. They were strictly asked not to release their finger from the home button before the probe target appeared. The participants performed 40 practice trials without electric shocks, and then the test trials began. At the beginning of each block, a warning appeared on the display: “You are safe from shocks in this next set of trials”, written in black letters against a blue background for the safe condition blocks, and “You may receive electric shocks at some points during the next set of trials”, written in black letters against a red background for the threat condition blocks. Three-minute breaks were taken between the blocks. Immediately after completing each block, each participant answered a VAS-based questionnaire about their mood during the block. The experiment took 60 min.

2.4. Measurements and Statistical Analysis

2.4.1. Manipulation Checks

The raw time sequence data of the HR records during each participant’s test block were averaged. The average HR and raw VAS scores of the three test blocks were then averaged for each of the ToS conditions (no ToS/safe and ToS/threat). The participants’ mean measurements were used as dependent variables to statistically test the effect of the anxiety manipulation protocol—that is, the difference in anxiety levels between the “safe” and “threat” condition blocks. Some of these variables deviated significantly from the normal distribution (p < 0.05). Thus, Wilcoxon signed-rank tests were used to evaluate the effect of the anxiety manipulations on the dependent variables. To measure effect sizes, r was calculated as the Z-statistics divided by the square root of the sample size.

2.4.2. Task Performance

The participants’ task performance was evaluated by measuring their reaction time (RT), which was defined as the time between the presentation of the probe target and the pressing of the response button. Because four participants executed an anticipatory response strategy—that is, releasing the home button before the probe target appeared—in multiple trials, all data on their performance were excluded from the statistical analysis. The remaining 32 participants’ data (16 participants for each group) were subjected to statistical analysis, and a post hoc power analysis was conducted.

Errors, such as pressing the wrong response button, were negligible (there were 8 errors in the hit trials, representing 0.21% of all trials; in the avoidant trials, there were 16 errors, representing 0.42%). Thus, these erroneous responses were excluded from the full dataset, and the error rates were not statistically analyzed. Trials with RTs smaller than 100 ms or RTs larger than 1000 ms were eliminated from the analysis as outliers (2.6% of all hit trials and 6.7% of all avoidant trials in the full dataset).

For each participant’s dataset, the RT measurement was averaged within the relevant emotional cue conditions (e.g., an angry and neutral face pair, or a happy and neutral face pair), ToS conditions (i.e., no ToS/safe, or a ToS/threat), and threat-congruent condition (i.e., spatial congruency between the location of a threat-related stimulus and the location of the probe target). The degree of attentional bias was assessed as the difference between RT in the threat-congruent condition (RTtc) and RT in the threat-incongruent condition (RTti):

Attentional Bias Index (ABI) [ms] = RTtc − RTti,

(1)

where the more positive or more negative the ABI value, the larger the degree of attentional bias toward the threat-related cue compared to the other cue. Negative ABIs represent faster overall responses, including target detection, motor selection, and motor execution, whereas positive ABIs represent delayed overall responses. Figure 2 illustrates the ABI formulae used to test Hypotheses I, II, and III.

The mean ABIs of the participants’ RTs were used as dependent variables to statistically test the effects of ToS on the degree of attentional bias. Wilcoxon signed-rank tests were used for variables that significantly deviated from the normal distribution (p < 0.05). Paired t-tests with one tail were used for other variables that were normally distributed. We calculated r and Cohen’s d as measures of effect size for the Wilcoxon signed-rank test and paired t-test, respectively. Statistical calculations were performed using the Statistics and Machine Learning Toolbox (MATLAB; R2024a, MathWorks, Natick, MA, USA). In the post hoc analysis, the power (1 − β) was calculated based on the means and SDs of the differences between the ToS and no ToS conditions using the G*Power program [36].

3. Results

3.1. Manipulation Checks

Table 1. Comparison of VAS scores and HRs between the safe (no ToS) and threat (ToS) conditions for each action goal group.

Action Group	Measurement	No ToS		ToS			No ToS < ToS
		Mdn	IQR	Mdn	IQR	N	Z	p	r
Hit	Anxiety [%]	18.2	33.3	52.2	37.2	16	2.74	0.006 *	0.69
	Fear [%]	12.2	33.0	38.5	24.7	16	3.28	0.001 *	0.82
	Happiness [%]	41.2	29.8	41.3	30.3	16	–0.67	0.501	–0.17
	HR [bpm]	74.8	21.5	77.5	21.2	16	0.43	0.670	0.11
Avoidant	Anxiety [%]	27.0	20.7	34.7	29.2	16	3.41	0.001 *	0.85
	Fear [%]	12.8	23.3	38.2	20.7	16	3.14	0.002 *	0.78
	Happiness [%]	50.0	22.8	50.0	26.8	16	–1.60	0.109	–0.40
	HR [bpm]	75.0	6.0	75.7	5.8	16	–0.66	0.509	–0.17

bpm—beats per minute; ToS—threat of shock; Mdn—median; IQR—interquartile range; r—effect size, calculated as Z divided by the square root of the sample size; N; * indicates a significant difference between the ToS and no ToS conditions (p < 0.05).

summarizes the results of the statistical tests, indicating the effects of the anxiety manipulations on the participants’ HRs and VAS scores. In HRs and happiness VAS scores, there were no significant differences between the ToS and no ToS condition blocks for either task group (p > 0.05). Compared to the no ToS block, the anxiety VAS scores were higher in the ToS block for the hit group (Z[15] = 2.74, p = 0.006, r = 0.69) and the avoidant group (Z[15] = 3.41, p = 0.001, r = 0.85). The fear VAS scores were also higher under the ToS block for the hit group (Z[15] = 3.28, p = 0.001, r = 0.82) and the avoidant group (Z[15] = 3.14, p = 0.002, r = 0.78). These results confirm that, by providing only a few electric shocks during the task trials, the threat manipulation protocol increased anxiety in the participants.

3.2. Task Performance

To test Hypotheses I and II, a threat-congruent condition was set up, meaning that the probe target appeared in the same location as angry face cues and the opposite location to neutral face cues (Figure 2a). Figure 3a illustrates the group means and standard errors (SEs) of the ABIs for the responses to a pair of angry and neutral faces as a function of the ToS condition. In the hit mode, the ABI mean values were negative under the ToS and no ToS conditions, indicating relatively faster responses in the threat-congruent condition (see also Table S1). However, there was no significant difference between the ToS and no ToS conditions (t[(15]) = −0.36, p = 0.639, d = 0.13, $\overline{X}$ = 1.7, SD = 19.3, 95% CI [−8.5, 12.0], (1 − β) = 0.095). These results suggest that, when responding in the hit mode, the tendency to adopt an attentional bias toward threat stimuli (i.e., facilitated engagement) may remain stable and is not substantially modulated by temporarily elevated anxiety. Therefore, the facilitated engagement hypothesis was not supported in the context of the hit mode (Hypothesis I).

In the avoidant mode, the ABI mean was slightly negative under the no ToS condition, whereas it became positive under the ToS condition (Figure 3a). A significant difference was discovered between the ToS and no ToS conditions (Z[15] = 2.30, p = 0.011, r = 0.58, $\overline{X}$ = 11.0, SD = 15.9, 95% CI [2.5, 19.5], (1 − β) = 0.823). The ToS significantly increased ABIs, leading to more delayed avoidant responses when cued by angry faces (see Table S1). This result supports the disengagement difficulty hypothesis (Hypothesis II).

To test Hypothesis III, a threat-congruent condition was set up, meaning that the probe target appeared in the same location as neutral face cues and opposite happy face cues (Figure 2b). Figure 3b illustrates the group ABIs (means and SE) for responses to a pair of happy and neutral face cues as a function of the ToS condition. In the hit mode, no significant difference was observed between the ToS and no ToS conditions (t[15] = −0.12, p = 0.454, d = 0.04, $\overline{X}$ = 0.4, SD = 14.4, 95% CI [−7.3, 8.1], (1 − β) = 0.062). In the avoidant mode, a significant difference was found between the ToS and no ToS conditions (t[15] = 2.63, p = 0.009, d = 0.80, $\overline{X}$ = 12.3, SD = 18.7, 95% CI [2.3, 22.3], (1 − β) = 0.806). The ToS significantly increased ABIs, leading to more delayed avoidant responses when cued by neutral faces than by happy faces. These results suggest that attentional bias was drawn toward emotionally neutral face cues when participants responded in the avoidant action mode. Therefore, the threat-related interpretation hypothesis (Hypothesis III) was supported in the context of the avoidant mode.

4. Discussion

Our experiment reveals two key associations between increased anxiety and attentional control in response to task-irrelevant emotional stimuli in healthy individuals. When state anxiety was heightened, (1) the avoidant action goal led to delayed responses to a target if it was in the same location as angry faces (compared to neutral faces), supporting the disengagement difficulty hypothesis, and (2) the avoidant action goal also led to delayed responses to a target if it was in the same location as neutral faces (compared to happy faces), supporting the threat-related interpretation hypothesis.

Overall, the current study lends further support to the idea that, in individuals with high trait anxiety and anxiety disorders, attentional biases are characterized by facilitated attentional engagement or disengagement difficulty [3,16,17,18]. While inconsistent tendencies have been reported regarding the attentional biases of individuals with low trait anxiety [1], this study confirms the presence of disengagement difficulty in attention in healthy individuals using the avoidant action mode and the ToS method. Our discovery of the presence of threat-related interpretations during heightened state anxiety in healthy individuals is also new.

It is intriguing that healthy individuals exhibited disengagement difficulties when experiencing heightened state anxiety in the avoidant action mode, but not when in the hit action mode. This distinction between the action modes may stem from the interplay between action goals and corresponding mindsets. In this study, the avoidant action, while itself only a task goal, corresponded to flight behavior in response to threat-related stimulus cues in the threat-congruent trials. As anxiety was heightened via ToS, this flight-like behavior may have unconsciously reinforced a “threat” mindset rather than a “challenge” mindset, as suggested by the embodied cognition framework [39]. Adopting a “threat” mindset may increase one’s vulnerability to anxiety [29]. This heightened vulnerability may impair attentional flexibility by leading one to prioritize threat-related stimuli over other task-relevant targets. Indeed, studies of attentional biases in anxiety have shown that prolonged engagement with threat cues and delayed disengagement are hallmark features of anxious cognition [16]. These findings thus suggest the importance of understanding the interplay between action goals, mindsets, and attentional control mechanisms in developing clinical interventions for anxiety disorders, such as embodiment techniques [40].

To date, visual probe tasks, all of which have employed hit action modes, have been used to examine attentional biases toward emotional stimuli. Previous research has hypothesized that faster key presses in response to probe targets in threat-congruent trials reflect an attentional bias toward threat-related cues [12,13]. In experiments based on this methodology, healthy individuals with low trait anxiety have exhibited weaker and less consistent tendencies toward attentional bias [41]. Recently, however, visual probe tasks have been criticized for failing to consistently measure attentional bias [34]. These inconsistent results may arise because visual probe tasks typically compare individuals with and without anxiety using a between-group design and rely solely on the hit mode. Taking these criticisms into account, the present study adopted the avoidant response mode and manipulated anxiety via ToS to demonstrate the presence of significantly delayed avoidant responses to targets associated with threat-related emotional cues during heightened state anxiety in healthy individuals.

Even so, one may argue that the results reached using the avoidant action mode are inconsistent with what is predicted by the innate approach–avoidance control mechanism: that avoidance responses to angry faces should be faster. In fact, organisms are endowed with mechanisms that automatically trigger approach behavior toward positive objects, such as happy faces, and avoidance of negative objects, such as angry faces [42,43]. One possible explanation for this inconsistent result is that the participants were not required to respond directly to the emotional stimuli, whereas approach–avoidance research requires such responses. Thus, it is safe to assume that the difference in reaction times between the angry/happy face cues and neutral face cues reflects a function of attentional control subconsciously evoked by the presence of a temporally prior cue [3,16,17,18].

The adoption of an avoidant action mode in the visual probe task revealed another unique phenomenon. Heightened anxiety can promote threat-related interpretations [2,35]. This means that, even if one is paying visual attention to task-relevant information, this information (or surrounding information) may be perceived differently or misinterpreted depending on one’s current emotions or state [30]. If the ToS led the participants to perceive neutral faces as having more threatening valence than happy faces, the anxiety-induced disengagement difficulty function would work to slow down the redirection of attention away from the neutral stimuli.

Indeed, in our experiment, delayed responses occurred in the trials in which neutral faces were congruent with the target (Figure 2b and Figure 3b). Thus, the threat-related interpretation hypothesis [2,35] is plausible, and it is difficult to explain the results with the approach/avoidance hypothesis [42,43]. In any case, this appears to be an interesting specific phenomenon induced by the avoidance action mode. Although the ToS method requires caution when applied to individuals with anxiety, we propose that the present methodology, based on combining the avoidant action mode and a ToS, will shed further light on the relationship between attentional bias and anxiety, potentially informing the treatment of anxiety disorders [44].

Finally, heightened state anxiety did not affect attentional control with respect to threat-related stimuli while hitting a target. These results provide indirect experimental evidence that a “challenge” mindset is important for reducing impairment of perceptual-motor performance when state anxiety is heightened—e.g., in situations of high psychological pressure. The BMCT theory suggests that shifting from a threat mindset (i.e., flight) to a challenge mindset (i.e., fight) may be an effective way to cope with anxiety [29]. Elucidating the relationship between the challenge mindset and the extent of attentional bias toward threat-related stimuli remains a future challenge.

In conclusion, behavioral goals—such as fighting or fleeing, represented in our experiment by hitting or avoiding a target—were found to interact with state anxiety, leading to different degrees of attentional bias. Avoidance behavioral goals may function to strengthen attentional bias (in this case, disengagement difficulty) and thereby mediate anxiety. Furthermore, they may induce a tendency to interpret stimuli with less threatening valence as threatening. Therefore, it is safe to conclude that encouraging the avoidance of task-relevant stimuli as an action mode may be counterproductive for coping with heightened state anxiety, at least in healthy individuals.

Limitations and Directions for Further Research

In terms of the mechanisms underlying the present findings, three limitations arise, particularly regarding the research methodology. First, the ToS protocol in the present study did not induce a high enough level of state anxiety to significantly increase HRs, although it did increase levels of subjective state anxiety. The finding that state anxiety is higher to the extent that it manifests itself in physiological responses could help elucidate the background mechanism that explains the present findings.

Second, we did not test the effect of altering action goals on biased attention using a within-participant design. Instead, a group comparison design was adopted because frequent repetition of electric shocks may allow each participant to become accustomed to the stimulus intensity and consequently reduce state anxiety levels. Therefore, to deepen our understanding of this issue, future research should focus on developing methods that can consistently induce high state anxiety levels in healthy individuals.

The third limitation of the study was the relatively small sample size. Although significant results were obtained in the avoidant mode, and the power was somewhat greater than that resulting from a priori calculation in testing Hypotheses II and III, further validation with a larger, independent sample population is recommended to strengthen the support for the hypotheses. In addition, the study did not include athletes, whose perceptual and motor performance is often impaired in anxious situations. Some athletes have sufficient resources to cope with the demands of tasks in very anxious situations, while others do not [45]. Investigating the relationship between attentional bias and action goals in a large and diverse population, including athletes, is necessary to further understand the cognitive mechanisms underlying the deterioration of perceptual and motor performance in anxious situations.

Finally, to apply the findings of this study in clinical settings, it would be necessary to address potential limitations, such as the use of standardized visual stimuli, i.e., facial expressions, and the lack of ecological validity in real-world contexts. Given these limitations, the findings of this study could inform the development of clinical intervention methods based on psychological tasks, such as threat-avoidance training [44] and approach-avoidance training [46].

5. Conclusions

Avoidant behavior intensifies attentional bias toward emotional stimuli when anxiety is temporarily heightened, even in healthy adult individuals without anxiety disorders. This study provides the first evidence to suggest that voluntary control of avoidance behavior may intensify attentional bias toward threat-related stimuli and delay motor responses, with anxiety as a mediator. For clinical applications, this study suggests that adopting fight-like behavioral goals in response to threats, potentially in conjunction with a “challenge” mindset, may be effective for managing anxiety in everyday life.