Visual features drive attentional bias for threat

ABSTRACT

Past studies argued that the attentional capture by threats is hard to inhibit. When using threats as task-irrelevant stimuli, this effect can deteriorate performance on the primary task. Whether attentional capture is driven by affective information (threat) or visual features (shape) is still debated. Here we aimed to investigate the role of threat value and shape in modulating attentional resources by conducting two experiments (total N = 87). Participants engaged in a semantic vigilance task responding to masked words appearing at the centre of the screen while ignoring threat-relevant (threatening or visually similar but nonthreatening) and neutral control distractor images placed at different distances from the target word. We found no performance difference between participants exposed to threat-related stimuli via affective or shape features. Moreover, while performance decreased when a neutral distractor appeared close (compared to further away) to the target word, stimulus eccentricity had no effect when the distractor (irrespective of the conveying feature) was threat relevant. Our findings are in line with previous studies showing an initial capture of attention by threat-relevant information but that this negative effect is compensated by an increase in arousal. We conclude that even the visual features of a stimulus can modulate attention toward threats.

KEYWORDS:

• Signal suppression
• inhibition
• attentional bias towards threat
• general feature detection
• semantic vigilance

Introduction

Threatening stimuli seem to be prioritized in visual perception, resulting in faster detection of threatening objects in the environment (Brown et al., 2010; Fox et al., 2001; Hedger et al., 2016; Öhman & Mineka, 2001; Subra et al., 2017; Zsido, Stecina, et al., 2022). Past studies found that reaction times are significantly faster when it comes to locating threatening target stimuli over neutral ones (Becker et al., 2011; Blanchette, 2006; Brosch & Sharma, 2005; LoBue, 2010; Subra et al., 2017; Williams et al., 2006). Furthermore, threats tend to hold attentional focus delaying attentional shifts (Burra et al., 2019; Fox et al., 2007; Holmes et al., 2014). Threats seem to produce these attentional biases stronger and more reliably than other emotional valences (e.g., positive or negative non-threatening) and visually salient neutral stimuli (Csathó et al., 2008; March et al., 2017; Williams et al., 2006; Zsido, Bali, et al., 2022). Attentional biases and attentional prioritization allow our nervous system to initiate a quick and adaptive behavioural response in dangerous situations (LeDoux, 2022; Reinecke et al., 2009; Trujillo et al., 2021). However, attentional biases toward threat-related information also serve as the basis of the acquisition and maintenance of anxiety disorders, such as phobias; a bias that can be targeted in interventions to reduce fear and symptoms of anxiety (Cisler & Koster, 2010; McNally, 2018).

Yet, to date, it is still debated whether the advantage that threatening stimuli receive over neutral stimuli in visual processing is driven by visual or affective features. According to the general feature detection theory, threatening stimuli are salient because of their specific visual features (e.g., shape, skin texture, movement) (Coelho & Purkis, 2009; Davey, 1995). Visual search studies found that curvilinear shapes (like the body of a snake) are detected faster than straight or zigzag lines (LoBue et al., 2014; Van Strien et al., 2016; Wolfe et al., 1992). An attentional advantage to downward-pointing V shape (that is geometrically similar to the head of a snake) has also been observed (Larson et al., 2007). In contrast, the fear module theory (Mineka & Öhman, 2002; Öhman & Mineka, 2001) suggests that threatening objects are salient due to their affective features (e.g., threat-relevance, valence, arousal). More specifically, according to the arousal-biased competition theory (Mather & Sutherland, 2011), emotional arousal cannot only drive cognitive processes and mental representations, but it can also improve memory and modulate selective attention. Better understanding the roles of visual and affective features in the processing of threat-relevant information could have both theoretical (i.e., a unified theory) and practical (e.g., refining attention retraining methodologies) implications.

It was suggested that task-irrelevant threatening distractors compared to other valences and neutral ones are more likely to capture attention and decrease performance (Burra et al., 2017, 2019; Fox et al., 2005; Mancini et al., 2020; Zsido, Bali, et al., 2022), yet prior work concerning the inhibition or suppression of threatening stimuli is still scarce. Consequently, the underlying mechanisms are still not well understood although it could prove crucial because attentional inhibitory biases to threat-related information are core to the development and maintenance of anxiety and phobias (Koster et al., 2006; Koster et al., 2006). In contrast, for visually salient (but emotionally neutral) stimuli the signal suppression hypothesis (Sawaki & Luck, 2010a, 2011) proposes that the attentional inhibition of objects is possible. According to the hypothesis salient stimulus in the visual field creates a signal that grabs attention even though it is irrelevant to the observer’s goals. This signal, however, can be actively inhibited with top-down control before attentional capture happens. This means that people are capable of attentional control through top-down, goal-directed mechanisms when they want to perform well. Inhibition also plays an important role in emotion regulation and helps us downregulate the automatic evaluation of salient objects, such as threatening stimuli (Mogg & Bradley, 2018). Studying how threatening stimuli and visually similar but neutral distractors can be effectively inhibited compared to control neutral stimuli offers an opportunity to gain deeper insights into the interplay between arousal, shape, and the advantageous role of threat in attentional processes.

Past studies suggest that the prioritization of threats over neutral stimuli is even greater in the peripheral vision (Almeida et al., 2015; Öhman et al., 2012). The brainstem-amygdala-cortex pathway (Gu et al., 2020; Liddell et al., 2005) and the dorsolateral prefrontal cortex (Cinq-Mars et al., 2022) ensures effortless evaluation and rapid orienting towards threats, both in and out of attentional focus. Stimuli that appear outside of this centre lose details, which makes it harder to identify their content. To make up for the loss of this information, spatial attention works with different eye movements to bring important objects in the environment to the centre of the visual field. However, some stimuli that are characterized by higher arousal and valence levels, seem to have the ability to grab attention without corresponding eye movements (Bayle et al., 2009; Gao et al., 2017; Rigoulot et al., 2012; Zsido et al., 2019). According to a previous study (Calvo et al., 2008), the specific contents of emotional priming scenes presented in peripheral vision were not precisely processed, however, an impression was extracted that later oriented selective attention or caused false alarms for related probes in a recognition task. Consequently, presenting a threatening stimulus in the periphery should be harder to ignore than a neutral stimulus even if it has shared visual features with the threat. That is, by manipulating the eccentricity of threat-related stimuli, we may investigate the difference between the effect of visual and emotional features on visual processing.

In the present study, we aimed to test whether the visual or affective features of threat-related distractor stimuli are more important in determining attentional biases to threats. Additionally, we aimed to test whether the spatial distance (stimulus eccentricity) between the task and the distractors would have any effect on inhibition. We used a semantic vigilance task where participants had to respond to a centrally presented word and ignore task-irrelevant distractor images surrounding that word. We employed this paradigm over a more common visual search task because attention was fixated on the central task allowing for a more convenient and reliable manipulation of stimulus eccentricity. This addresses a critique of past studies (Pakai-Stecina et al., 2023). Further, using this paradigm we can extract more behavioural variables besides the usually presented accuracy and reaction time potentially resulting in a more robust interpretation. The semantic vigilance task has been proven successful in a past attempt to demonstrate the effect of auditory negative emotions on visual attentional performance (Zsido et al., 2023).

We expected higher vigilance decrement (decline in the rate of the correct detection of signals) for threat-related (i.e., threatening and shape-matched nonthreatening) compared to (visually dissimilar) neutral stimuli. Further, we hypothesized that threatening distractors would have a more pronounced effect than nonthreatening but visually similar ones. Our second hypothesis was that the eccentricity of the distractor would have a greater effect on the performance of the threat condition compared to the shape-matched nonthreatening condition. We expected that threat compared to shape-matched nonthreatening distractors would be harder to inhibit regardless of stimulus eccentricity.

Experiment 1

We adopted a semantic vigilance task similar to previous studies (Epling et al., 2016; Zsido et al., 2023). Participants were instructed to concentrate on masked words appearing on the centre of the screen one at a time and respond to living words with the spacebar and ignore the non-living words. Irrelevant distractive stimuli (pictures of two living and two non-living things) also appeared on the screen at three different distances to the target word: close (visual angle of 5°), middle (30°), and far (45°). The distractor pictures were of neutral valence in general, however, there we also used two special distractors. Participants were divided into two groups, with one group (threatening distractor) seeing a threatening picture among the distractors (snake) and the other group (shape-matched distractor) seeing a neutral but shape-similar picture to the threatening one (caterpillar) among the distractors. We introduced this manipulation as a between-subject factor to avoid carry-over effects between seeing an actual threat and an object that visually resembles it. In both groups, a neutral, visually dissimilar control distractor (fish) was used alongside the threatening or the shape-matched one. Figure 1 shows the trial structure of the paradigm used along with sample trials from both visual feature and affective feature groups in all three distance conditions.

Figure 1. The top panel (A) shows the trial structure of the paradigm used. First, a fixation cross of 0.5s was shown, then the semantic decision task followed. Each trial was shown for 3.25s regardless of being a target or non-target word trial and the reaction of the participants. The task was presented in three blocks (distractors in the close, middle, far positions) and the blocks were randomized. The bottom panel (B) shows sample trials from both threatening and shape-matched groups in all three (top row: close, middle row: middle, bottom row: far) distance conditions. Threat distractors are marked with red circles, shape-matched nonthreatening ones are marked with green circles and visually dissimilar neutral control distractors are marked with blue rectangles.

Top panel shows two grey squares. There is a black fixation cross in the middle of the top grey square. There is a word in the middle of the bottom grey square. The word is masked by black dotes to make it harder to read. There are four objects, one in each corner of the word.Bottom panel shows six grey squares that are similar to the bottom grey square in the top panel. The difference is that the objects appear at different distances from the word.

Materials and method

Participants

The required sample size for this experiment was determined by computing estimated statistical power based on previous studies of singletons and threat suppression (Sawaki & Luck, 2010b; Zsido et al., 2021, 2023). The analysis (f = .40, 1-β > .8, r = .5) indicated that the minimum required total sample size was 12 (or 28 with a more conservative approach of f = .25). We recruited a total of 29 students (21 females, mean age = 22.6 SD = 3.56) who participated in exchange for course credit. The threatening distractor group comprised 16 participants (mean age = 22.1 SD = 2.45). The shape-matched distractor group comprised 13 participants (mean age = 23.2 SD = 4.63).

All participants reported normal or corrected to normal vision and normal colour vision. Two participants were excluded because they failed to follow instructions. The research was approved by the national ethics committee and was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). All participants provided written informed consent.

Stimuli

We created a semantic vigilance task based on the methodology of past studies (Epling et al., 2016). A list of words, taken from a previous study (Zsido et al., 2023), consisted of 384 nontarget (non-living) and 96 target (living) words, with a signal-to-noise ratio of 1:4 throughout the experiment (target word probability was .2 and non-target word probability was .8). At the start of the experiment, the words were sorted into six lists (counterbalanced across participants) to create a unique set of words for each condition. Table 1 shows the number of words and trials broken down into six conditions (2 types of distractors and 3 stimulus eccentricities). The words consisted of three to seven letters (counterbalanced across word categories and lists). The words were positioned on the centre of a 1920 × 1080 pixel-sized grey background with black ink colour of Arial size 9, set to a transparency level of 35% (see Figure 1). A mask (dark grey dots on grey background sized 135 × 62 pixels) was placed under the text to make it more difficult to read.

Table 1. Number of words and trials broken down to six experimental blocks. Special distractors were snakes for those in the affective feature group and caterpillars for those in the visual feature group.

Condition	Distractor type	Number of words
Close condition	With special distractor (snake or caterpillar)	16 target, 64 non-target
	With neutral control distractor (fish)	16 target, 64 non-target
Middle condition	With special distractor (snake or caterpillar)	16 target, 64 non-target
	With neutral control distractor (fish)	16 target, 64 non-target
Far condition	With special distractor (snake or caterpillar)	16 target, 64 non-target
	With neutral control distractor (fish)	16 target, 64 non-target
Total		96 target, 384 non-target

The special distractor image categories (i.e., snake for the threatening and caterpillar for the shape-matched group) were determined based on the results of an online survey (see Supplementary material 1) we conducted on an independent sample (N = 77) where we asked participants to write objects that they find threatening and pair them with an object that is visually similar but is non-threatening. We used the stimuli pair that was mentioned most frequently. Our goal was to match threatening objects to ones that people find visually similar but non-threatening in line with past studies investigating similar questions (Almeida et al., 2015; LoBue, 2014; Van Strien et al., 2016; Zsido et al., 2018; Zsido, Stecina, et al., 2022)

Distractor images were colourful photographs of real-world objects. Most of the images were collected from the Massive Memory Database (Hout et al., 2014) and some images of the snakes, caterpillars, and fish were sourced from the Internet. None of these stimuli had a background. The images were resized to approximately the same size (i.e., no larger than 100 × 100 pixels) maintaining the original proportions. We used a large number of special distractors (20 exemplars per category) and other distractors (i.e., 240 categories with 15–16 exemplars per category) that were randomly sampled across trials (and participants) to ensure that distractors and targets were comparable and to reduce the possible nuisance effects of low- and mid-level visual features of the individual objects.

Distractors were placed at one of three relative distances to the four corners of the mask on every picture: visual angle of 5° (close), 30° (middle), and 45° (far). In all trials, there was either a threat-relevant special distractor (snake or caterpillar) or the neutral control distractor (fish) presented among three other random objects (e.g., butterfly, leaf, rock, clock, etc.). Special and control distractors appeared with equal probability across all word types (living and non-living), eccentricities (close, middle, and far), and experimental blocks.

Procedure

Data was collected in small groups on up to 10 computers simultaneously (with non-identical hardware and software profiles) in a computer room. Participants were seated in separate work-station booths, approximately 60cm in front of 17-inch CRT monitors (resolution 1024 × 768, 4:3 aspect ratio, refresh rate of 60 Hz, colour depth of 65.536k). Stimuli were presented using the PsychoPy v3.0 software (Peirce, 2007). Data collection sessions were monitored by a research assistant. After both verbal and written instructions, participants completed a test run of 10 trials with 5 target present and 5 target-absent trials. There were no distractor pictures present during the practice trials and participants got feedback on their reactions (correct/incorrect). Practice trials were excluded from the analysis. Participants also had their chance to ask questions if they had any before starting the real experiment. Then, all participants present at the data collection site started the task at the same time, having to press the spacebar when a living word appeared on the screen. One stimulus picture was presented for 3.25 s preceded by a fixation cross of 0.5 s (see Figure 1). Stimuli were presented in three blocks according to the three eccentricity conditions (close, middle, far). The presentation of the blocks was randomized across participants. The task took approximately 30 min to complete.

Statistical analysis

There were no outliers, defined as those more than 2 standard deviations below or above the group mean, for accuracy; while we identified and removed the outlier for reaction time (less than 3% of trials). Unlike past studies that used a visual search paradigm, we also calculated the signal detection theory metrics of d prime (d’, sensitivity) and response bias (c) in line with semantic vigilance studies (Epling et al., 2016; Zsido et al., 2023). We used the formulas d’ = z(H)-z(FA) and c = −1/2*[z(H)+z(FA)] for this purpose. Here, z(H) is the z-transformed value of the proportion of corrected detection (Hits) and z(FA) is the z-transformed value of the proportion of False Alarms.

Statistical analyses were completed with the help of the JAMOVI Statistics Program v2.0 (Jamovi Project, 2022). We performed a 2 × 3 × 2 repeated measures analysis of variance (rANOVA) with Distractor Type (threat-related and control) and Distance (close, middle, far) as within-subject factors and the Groups (threatening and shape-matched distractor) as a between-subject factor to test the effect of threat and shape-matched threat-relevant but nonthreatening distractors on vigilance performance. The accuracy, d’, c, and RT values were analyzed separately. Main effects and interactions are reported separately, paired with relevant follow-up analyses to further investigate the significant interactions. Effect sizes are also presented: partial eta squared (${\eta }_p^2$) for the rANOVAs. Tukey corrections were used to account for multiple comparisons. Both the normality and homogeneity of variances assumptions for the ANOVA analysis were met. Please note that in the interest of brevity and clarity, the results of the statistical analyses are presented in tables. See Supplementary material 2 for the detailed descriptive statistics including accuracy, d’, c, and RT across all conditions.

Results

Accuracy

We began by examining accuracy to test our prediction that the performance of the threatening distractor group compared to the shape-matched distractor group will be worse for threat-related distractors compared to neutral distractors. Figure 2 presents the descriptive statistics; see Table 2 for significant statistical results and Supplementary material 3 for the full report on all statistical results. The interaction between the Distractor Type and Group was nonsignificant; thus, our hypothesis was not supported, and neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli. We also expected that the threatening distractor group would be more affected by the distance of distractors than the shape-matched distractor group, with the distractors only interfering with the close distractor stimuli in the latter and interfering regardless of distance in the former group. Contrary to our hypothesis, we did not find a significant interaction between Distance, Distractor Type, and Group either.

Figure 2. Accuracy in Experiment 1 for the threatening distractor and shape-matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

This is a box plot visualising the descriptive statistics for accuracy. There are two panels, the left is for the control and the right is for the threat-related condition. The results of the groups are shown in different colour. Shape-matched group is in green, threatening group is in orange.

Table 2. Significant main effect and interactions for all behavioural measures in Experiment 1. Main effects are broken down by pairwise comparisons, while interactions by follow-up ANOVAs.

Measure	Variable		df	F/t	p	${\eta }_p^2$
Accuracy	Distance		2, 54	8.660	< .001	0.243
		Close – Middle	27	−3.747	0.002
		Close – Far	27	−3.564	0.004
		Middle – Far	27	−0.520	0.862
	Type * Distance		2, 54	3.970	0.025	0.128
	Threat relevant		2, 54	0.403	0.670	0.014
	Non-threatening control		2, 54	14.2	< .001	0.337
		Close – Middle	27	−5.012	< .001
		Close – Far	27	−4.749	< .001
		Middle – Far	27	−0.780	0.718
d'	Distance		2, 54	8.496	< .001	0.239
		Close – Middle	27	−3.379	0.006
		Close – Far	27	−3.844	0.002
		Middle – Far	27	−0.551	0.847
	Type * Distance		2, 54	4.684	0.013	0.148
	Threat relevant		2, 54	0.309	0.735	0.011
	Non-threatening control		2, 54	14.7	< .001	0.345
		Close – Middle	27	−4.651	< .001
		Close – Far	27	−4.819	< .001
		Middle – Far	27	−0.864	0.667
c	Distance		2, 54	7.437	0.001	0.216
		Close – Middle	27	2.859	0.021
		Close – Far	27	3.844	0.002
		Middle – Far	27	0.816	0.697
	Type * Distance		2, 54	4.424	0.017	0.141
	Threat relevant		2, 54	0.278	0.758	0.010
	Non-threatening control		2, 54	12.8	< .001	0.314
		Close – Middle	27	4.01	0.001
		Close – Far	27	4.82	< .001
		Middle – Far	27	1.15	0.492
Reaction time	Type * Distance		2, 54	3.3520	0.042	0.110
	Threat relevant		2, 54	0.608	0.548	0.021
	Non-threatening control		2, 54	2.81	0.069	0.091
		Close – Middle	27	1.26	0.428
		Close – Far	27	1.95	0.144
		Middle – Far	27	1.57	0.275

The main effect of Distance and the interaction between Distance and Distractor Type was significant. Teasing apart the interaction revealed that the main effect of Distance was only significant in the neutral control condition. Participants were less accurate when the neutral target was close to the semantic task compared to when it was further away. The effect of Distance was nonsignificant in the Threat-relevant condition. All other effects were nonsignificant.

d’ scores

We then examined d’ scores to check for our predictions. Figure 3 presents the descriptive statistics; see Table 2 and Supplementary material 3 for the statistical results. Contrary to our expectations both the Distractor Type x Group and the Distance x Distractor Type x Group interactions were nonsignificant meaning that neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli.

Figure 3. Sensitivity in Experiment 1 for the threatening distractor and shape- matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

This is a box plot visualising the descriptive statistics for sensitivity. There are two panels, the left is for the control and the right is for the threat-related condition. The results of the groups are shown in different colour. Shape-matched group is in green, threatening group is in orange.

Again, both the Distance main effect and Distance x Distractor Type interaction were significant. The Distance effect was only significant for the control but not the threat-relevant conditions. That is, participants’ performance was worse in the neutral control condition when distractors were presented close to the target word than when they were presented in middle and far eccentricities. All other effects were nonsignificant.

c scores

We next examined the c scores, again to check for our predictions regarding the threat-relevant distractors. Figure 4 presents the descriptive statistics; see Table 2 and Supplementary material 3 for the statistical results. Contrary to our expectations, again, both the Distractor Type x Group and the Distance x Distractor Type x Group interactions were nonsignificant meaning that neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli.

Figure 4. Response bias in Experiment 1 for the threatening distractor and shape- matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

This is a box plot visualising the descriptive statistics for response bias. There are two panels, the left is for the control and the right is for the threat-related condition. The results of the groups are shown in different colour. Shape-matched group is in green, threatening group is in orange.

The main effect of Distance and the interaction between Distance and Distractor Type were significant. The Distance effect was only significant for the control but not the threat-relevant conditions. That is, participants’ performance was worse in the control condition when distractors were presented close to the target word than when they were presented in middle and far eccentricities. All other effects were nonsignificant.

Reaction time

We next examined the RTs to check for our predictions regarding the threat-relevant distractors. Figure 5 presents the descriptive statistics; see Table 2 and Supplementary material 3 for the statistical results. Contrary to our expectations, again, both the Distractor Type x Group and the Distance x Distractor Type x Group interactions were nonsignificant meaning that neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli.

Figure 5. Reaction time in Experiment 1 for the threatening distractor and shape- matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

This is a box plot visualising the descriptive statistics for reaction time. There are two panels, the left is for the control and the right is for the threat-related condition. The results of the groups are shown in different colour. Shape-matched group is in green, threatening group is in orange.

While the interaction between Distance and Distractor Type was significant, the follow-up analyses revealed nonsignificant main effects of Distance in both the control (p = .548) and the threat-relevant conditions (p = .069). All other effects were nonsignificant.

Discussion

The main goal of Experiment 1 was to determine whether visual or affective features of a threat-relevant distractor stimulus are more defining in attentional capture and how the effect changes with the spatial distance between the task and the distractors. Overall, we found evidence of a distance effect; that is, the performance of participants was lower when a distractor appeared close to the task compared to when it appeared in the periphery. Surprisingly, however, we found no evidence of differences in performance for threatening (snake) versus shape-matched (caterpillar) distractors. In addition to this, the distance effect was only observable for the neutral control distractor (fish) but not for the threat-relevant distractors. The results may provide evidence for an attentional prioritization of threat-related information based on visual features.

Before we can dive into the discussion of the possible theoretical explanations behind these results, we need to check whether the results from a unique class of images (i.e., snakes and caterpillars) can be generalized to other types of threatening information. Further, the lack of significant results for the threat-related distractors might mean that the threat manipulation failed. Consequently, we next sought to rule out stimulus idiosyncrasies or flukish results as an explanation for what we observed. For the dual purposes of replication and to rule out stimulus idiosyncrasies, (and also to test whether the visual or affective features of the threat caused this pattern of results), we conducted a second experiment.

Experiment 2

In Experiment 2, participants performed the same semantic vigilance task as in Experiment 1. Here, in addition to the snake, the threatening special distractor category also included spider, syringe, and gun; consequently, in addition to caterpillars, the shape-matched special distractor category also included stinkbug, knitting-pin, and hairdryer; finally, in addition to fish, the neutral control distractor category included cat, kitchen utensil, and perfume bottle. This was necessary to address the concern left by Experiment 1 that our results were not generalizable to threats. Further, the sample size in Experiment 1 was rather low (although the minimum sample size requirement was met), which also precluded making generalized claims about the results. Therefore, we aimed to collect a significantly larger number of responses in Experiment 2. Our modified design thus allowed us to explore the effects of affective and visual features of a threat-related stimulus more broadly.

Materials and method

Participants

We sought to double the sample size of Experiment 1. We collected data from 58 students (mean age = 20.7, SD = 1.63) for partial course credit. Five participants were identified as outliers (defined as those more than 2 standard deviations below or above the group mean) and removed, resulting in a total sample size of 53 participants. The threatening distractor group comprised 26. The shape-matched distractor group comprised 27 participants.

All participants reported normal or corrected to normal vision and normal colour vision. The research was approved by the national ethics committee and was carried out in accordance with the Code of Ethics of the World Medical Association (Declaration of Helsinki). All participants provided written informed consent.

Stimuli and procedure

The semantic vigilance task and the distractor images were identical to Experiment 1. In all trials, there was either a special distractor (threatening: snake, spider, syringe, gun or shape-matched nonthreatening: caterpillar, stinkbug, knitting-pin, hairdryer) or a neutral control distractor (fish, cat, kitchen utensil, perfume bottle) presented among three other random objects.

Data was collected in small groups on up to 10 computers simultaneously (with non-identical hardware and software profiles) in a computer room. Participants were seated in separate work-station booths, approximately 60cm in front of the computer screens. In contrast to Experiment 1, stimuli were presented on 21.5-inch LCD screens (resolution 1920 × 1080, 16:9 aspect ratio, refresh rate of 60 Hz, colour depth of 16.7M) because we did not have access to the computer room with CRT monitors.

Statistical analysis

We removed participants with outlier values, defined as those more than 2 standard deviations below or above the group mean (less than 5% of trials). We calculated the signal detection theory metrics of d’ and c.

Statistical analyses were completed with the help of the JAMOVI Statistics Program v2.0 (Jamovi Project, 2022). We performed a 2 × 3 × 2 repeated measures analysis of variance (rANOVA) to test the effect of Distractor Type (threat-related and control) and Distance (close, middle, far) as within-subject factors and the Groups (threatening and shape-matched) as a between-subject factor on performance (indicated by accuracy, d’, c, and RT). Only correct trial RTs were analyzed. Both the normality and homogeneity of variances assumptions for the ANOVA analysis were met. Statistical results are presented in tables instead of in text to make the description of the results easier to follow. See Supplementary material 2 for the detailed descriptive statistics including accuracy, d, c, and RT across all conditions.

Results

Accuracy

We began by examining accuracy to test our prediction that the performance of the threatening distractor group compared to the shape-matched distractor group will be worse for threat-related distractors compared to neutral control distractors. Figure 6 presents the descriptive statistics; see Table 3 and Supplementary material 3 for the statistical results. Replicating the results of Experiment 1, the interaction between Distractor Type and Group was nonsignificant; thus, our hypothesis was not supported, and neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli. We also expected that the threatening distractor group would be more affected by the distance of distractors than the shape-matched distractor group, with the distractors only interfering with the close distractor stimuli in the latter and interfering regardless of distance in the former group. Again, in line with the results of Experiment 1 but contrary to our hypothesis, we did not find a significant interaction between Distance, Distractor Type, and Group either.

Figure 6. Accuracy in Experiment 2 for the threatening distractor and shape- matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

Table 3. Significant main effect and interactions for all behavioural measures in Experiment 2. Main effects are broken down by pairwise comparisons, while interactions by follow-up ANOVAs.

Measure	Variable		df	F/t	p	${\eta }_p^2$
Accuracy	Distance		2, 102	3.165	0.046	0.058
		Close – Middle	51.0	1.524	0.288
		Close – Far	51.0	−2.430	0.048
		Middle – Far	51.0	−1.000	0.580
	Type * Distance		2, 102	4.454	0.014	0.080
	Threat relevant		2, 102	2.040	0.134	0.030
	Non-threatening control		2, 102	3.35	0.038	0.048
		Close – Middle	51.0	−2.0198	0.042
		Close – Far	51.0	−2.3464	0.022
		Middle – Far	51.0	−0.0449	0.964
d'	Type * Distance		2, 102	4.191	0.018	0.076
	Threat relevant		2, 102	0.859	0.426	0.013
	Non-threatening control		2, 102	3.67	0.028	0.052
		Close – Middle	51.0	−2.238	0.029
		Close – Far	51.0	−2.383	0.020
		Middle – Far	51.0	0.122	0.903
c	Distance		2, 102	3.2757	0.042	0.060
		Close – Middle	51.0	−1.01	0.576
		Close – Far	51.0	1.57	0.035
		Middle – Far	51.0	2.56	0.269
	Type * Distance		2, 102	4.4317	0.014	0.080
	Threat relevant		2, 102	3.73	0.027	0.054
		Close – Middle	51.0	−1.13	0.263
		Close – Far	51.0	1.48	0.142
		Middle – Far	51.0	2.99	0.004
	Non-threatening control		2, 102	1.85	0.161	0.027
Reaction time	Type		1, 51	4.296	0.043	0.072
	Distance * Groups		2, 102	4.313	0.016	0.073
	Threat relevant		2, 102	2.267	0.114	0.080
	Non-threatening control		2, 102	3.06	0.054	0.096
		Close – Middle	51.0	0.664	0.512
		Close – Far	51.0	2.290	0.029
		Middle – Far	51.0	1.671	0.105

Similarly to Experiment 1, the main effect of Distance and the interaction between Distance and Distractor Type was significant. Teasing apart the interaction revealed that the main effect of Distance was only significant in the neutral control condition. Participants were less accurate when the neutral target was close to the semantic task compared to when it was further away. The effect of Distance was nonsignificant in the Threat-relevant condition. All other effects were nonsignificant.

d’ scores

We then examined d’ scores to check for our predictions. Figure 7 presents the descriptive statistics; see Table 3 for significant statistical results and Supplementary material 3 for the full report on all statistical results. Replicating the results of Experiment 1 both the Distractor Type x Group and the Distance x Distractor Type x Group interactions were nonsignificant. Neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli.

Figure 7. Sensitivity in Experiment 2 for the threatening distractor and shape- matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

Similarly to Experiment 1, the Distance x Distractor Type interaction was significant with the Distance effect only being significant for the neutral control but not the threat-relevant conditions. That is, participants’ performance was worse in the neutral control condition when distractors were presented close to the target word than when they were presented in middle and far eccentricities. All other effects were nonsignificant.

c scores

We next examined the c scores, again to check for our predictions regarding the threat-relevant distractors. Figure 8 presents the descriptive statistics; see Table 3 and Supplementary material 3 for the statistical results. Contrary to our expectations, but in line with Experiment 1, both the Distractor Type x Group and the Distance x Distractor Type x Group interactions were nonsignificant meaning that neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli.

Figure 8. Response bias in Experiment 2 for the threatening distractor and shape- matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

Again, similarly to Experiment 1, the main effect of Distance and the interaction between Distance and Distractor Type were significant. The Distance effect was only significant for the neutral control but not the threat-relevant conditions. That is, participants’ performance was worse in the control condition when distractors were presented close to the target word than when they were presented in middle and far eccentricities. All other effects were nonsignificant.

Reaction time

We next examined the RTs to check for our predictions regarding the threat-relevant distractors. Figure 9 presents the descriptive statistics; see Table 3 and Supplementary material 3 for the statistical results. Contrary to our expectations, replicating the results of Experiment 1, the Distractor Type x Group and the Distance x Distractor Type x Group interactions were nonsignificant meaning that neither the effect of threatening nor that of visually similar stimuli was different when compared to neutral stimuli.

Figure 9. Reaction time in Experiment 2 for the threatening distractor and shape- matched distractor groups across the three distractor eccentricities visualized as boxplots (separately for the two types of distractors).

In contrast to previous results, the interaction between Distance and Group was significant; however, the follow-up analyses revealed that the main effect of Distance was only significant in the neutral control but not in the threat-relevant group. In the control group, participants were faster to respond when the distractors were presented far from the target word than when they were presented in close and middle eccentricities. All other effects were nonsignificant.

Discussion

In Experiment 2, we replicated the results of Experiment 1; that is, performance was lower when a neutral distractor appeared close to the task compared to when it was on the periphery. We found no such effect for threat-relevant distractors; further, we did not find any differences between the threatening distractor and shape-matched distractor groups. This was true even though, compared to Experiment 1, we used other types of threatening stimuli, not just snakes and caterpillars. Thus, the effects we found in Experiment 1 are likely not due to the specific shape or a possible difference in the visibility of the targets but rather may be generalized to a wider range of threat-relevant information. Similarly, it seems unlikely that the results are due to a failed threat manipulation because the effect of threat-related distractors was different compared to the baseline effect seen for the neutral control distractors in both the threatening distractor and the shape-matched distractor groups. In contrast to Experiment 1, here we also found a distance effect for threat-relevant targets for response bias; that is, the conservative response bias was lower when the distractor was presented farther from the task compared to when it appeared closer. In sum, again, the results of Experiment 2 suggest an attentional prioritization of threat-related information based on visual features.

General discussion

The goal of our two experiments was to test if salient but task-irrelevant stimuli, i.e., threat-related distractors, capture attention during a semantic vigilance task at different stimulus eccentricities. Further, we sought to investigate whether the attentional capture is driven by the affective (threat value) or visual (shape) features of a threatening stimulus. We did find an effect of stimulus eccentricity for neutral control distractors – performance was lower when they were presented closer (compared to farther away) to the task. Contrary to our expectations, we did not find such an effect for threat-related distractors. Also contradicting our hypothesis, we did not find a difference in task performance between the threatening and shape-matched distractor groups.

Our results suggest that shape information alone if threat-related, can affect task performance similarly to actual threats. This can be seen as further proof of the general feature detection theory (Coelho & Purkis, 2009; Davey, 1995) the smoke detector principle – i.e., better to err on the side of caution – (Nesse, 2006). Further, this result is in line with previous experimental results indicating that shapes that are strongly associated with threats (e.g., curvy shapes – snakes, downward pointing V shapes – snakeheads) can elicit the same responses as threatening images (of real snakes) (Larson et al., 2007; LoBue, 2014; Van Strien et al., 2016; Wolfe et al., 1992).

A previous study (Zsido, Stecina, et al., 2022) using the Rapid Serial Visual Presentation paradigm with task-relevant threat-related objects showed that when visual features are sufficient to discriminate the target from the other items in the stream, there was no effect of affective feature (i.e., threat level) on reaction time or accuracy. While that study was more focused on working memory resources and used task-relevant objects, our findings here are similar insofar as only the visual feature was sufficient to elicit the same response as the affective feature of a threat. Somewhat contrary to our findings a recent study (Pakai-Stecina et al., 2023) using threat-related objects as distractors found that visual features of threats are easier to suppress than affective features. However, in that study, the presentation of the two features was mixed; i.e., trials with threatening distractors and nonthreatening but visually similar distractors were randomly presented to the participants. This might have caused a generalization of the threat effect from the real threatening objects to those with the same visual features. Participants who once saw the snake might think that all curvy distractors were snakes. Thus, a strength of our study is that the presentation of the two features was not mixed, i.e., participants only saw threatening or shape-matched neutral distractors but not both. This means that the effect we found could not be caused by a generalization of the threat value to the visual feature.

Further, Pakai-Stecina et al. (2023) used a visual search paradigm allowing the participants to freely explore the visual scene which might have caused a confound in interpreting the distractor eccentricity (the distance between the target and distractor). In the present study, participants fixated on the target appearing in the centre of the screen while distractors appeared at the same time in different eccentricities ensuring that distractors were presented in the fovea, parafovea, or periphery. Consequently, we propose that just the shape of the threatening object can cause the same effect – and, therefore, lack of distraction – relative to neutral stimuli.

In the present study, we focused solely on threatening and non-threatening images without manipulating arousal using other valences such as positive images. This restricts the generalizability of our findings to other types of emotional stimuli. While the assumption of higher arousal and more negative valence for the threat compared to neutral categories is reasonable, we did not measure this directly. There are other differences between threats and their visually similar counterparts that could be relevant including, e.g., knowledge of the threat.¹ The dual implicit process model of evaluation (March et al., 2018) proposes two interconnected automatic mechanisms, where threat perception influences valence processing, which in turn affects explicit processes like evaluation. While the emotion (“fear”) elicited by an object precedes its evaluation (“this is dangerous”), participants saw several representations so ongoing attentional processes shall also be considered. Again, we did not find a difference between threatening and shape-matched nonthreatening distractors suggesting that conscious processes (such as evaluation) play a lesser role in attentional biases towards threats. Given that future studies confirm these findings, this could be an important step towards an understanding of the cognitive mechanisms involved in threat processing and perception as well as the maintenance of specific phobias.

Stimulus eccentricity mostly affected performance in trials with neutral control distractors while seemingly it had no effect in trials with a threat-related distractor. Concerning neutral control stimuli the results are in line with expectations based on, e.g., the guided search theory (Wolfe, 2021). Stimuli that are closer to fixation are given priority in attentional processing (thus it is harder to inhibit them by top-down control), while distractors appearing further from the task are easier to inhibit (and have considerably smaller effects on task performance). Interestingly, for both threatening and shape-matched nonthreatening distractors the performance remained unchanged across stimulus eccentricities suggesting that the presence of threat-related information overrides the distance-related variations. Considering the arousal stimulation effect theory (Zsido et al., 2018; Zsido et al., 2020, 2021) increased levels of arousal elicited by threat-relevant information may compensate for a negative effect of distractors, resulting in overall better performance in the close condition compared to neutral control stimuli. While past research has only demonstrated this effect with threatening stimuli, our results suggest that it is triggered by visual features strongly associated with threats.

Some limitations of the study shall be noted. First, we solely examined behavioural outcomes and did not investigate neural correlates, which could have provided a more comprehensive understanding of the underlying mechanisms. In future directions, our study can be expanded by investigating the neural pathways underlying attentional biases towards threatening stimuli, utilizing techniques such as EEG and ERP to explore the temporal dynamics and neural signatures associated with these biases. Second, individual differences such as anxiety and (both objective and subjective) fatigue levels may interact with the arousal and distracting effects of threats. Incorporating measures of anxiety and fatigue would enable direct monitoring of participants’ vigilance and anxiety levels during the task and may provide deeper insights into the results.

In conclusion, our findings support the notion that threat information affects attentional processing based on visual features. The effect of threat-related distractors seems to be independent of their distance from the fovea they seem to enhance performance on the primary task when presented near it. Understanding this attentional bias towards threat-related information is crucial, as it forms the foundation for the development and persistence of anxiety disorders, including phobias. Targeting the attentional bias associated with threat-relevant features through interventions aimed at reducing fear and anxiety symptoms can have significant implications for improving treatment outcomes (Cisler & Koster, 2010; McNally, 2018).

Availability of data and material

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Author contribution

DTPS: Conceptualization, Methodology, Software, Data curation, Visualization, Formal analysis, Project administration, Funding acquisition, Writing – Original draft, Reviewing and Editing; JB: Data curation, Visualization, Formal analysis, Writing – Reviewing and Editing; BK: Data curation, Visualization, Formal analysis, Writing – Reviewing and Editing; ANZS: Conceptualization, Methodology, Supervision, Funding acquisition, Writing – Original draft, Reviewing and Editing.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

DTPS was supported by the German Academic Exchange Service (DAAD) (Funding ID: 57588369) and the New National Excellence Program (ÚNKP-21-3) of the Ministry of Innovation and Technology. ANZS was supported by the ÚNKP-23-5 New National Excellence Program of the Ministry for Innovation and Technology from the source of the National Research, Development and Innovation Fund, the OTKA PD 137588 and OTKA FK 146604 research grants, and the János Bolyai Research Scholarship provided by the Hungarian Academy of Sciences. BLK and JB was supported by the OTKA K 143254 research grants of the National Research, Development and Innovation Office.

Notes

1 We thank Reviewer 2 for pointing this out and helping us to make this explicit.