Broad attention does not buffer the impact of emotionally salient stimuli on performance

ABSTRACT

It has been claimed that a broad attentional breadth buffers the impact of negative stimuli on human perception and cognition. Here we identify issues with the research on which this claim is based, and then rigorously test the claim. To induce narrow versus broad attentional breadth participants attended to the local versus global elements of Navon stimuli, and to investigate the impact of emotionally salient stimuli on performance we measured the effect of task-irrelevant stimuli of varying emotional salience (negative, neutral, or positive) on task performance. Across a series of experiments, we found that the Navon stimuli were effective in inducing different attentional breadths, and that both negative and positive task-irrelevant stimuli slowed responses relative to neutral stimuli, but that the magnitude of this emotion-induced slowing was invariant to whether attentional breadth was broad or narrow. This indicates that a broad attentional breadth did not buffer against the effect of either negative or positive emotionally salient stimuli. These results challenge the claim the broadening attentional breadth protects against the impact of emotionally salient stimuli.

KEYWORDS:

• emotion
• attention
• attentional breadth
• broad attention
• global attention

For most people, vision is the primary sensory modality. Vision allows us to navigate through the world and interact with it, such as driving safely through traffic, reading this text, and recognising the face of a friend. But at any given moment, there is far more visual information available to process than our brain is capable of processing to the level of awareness. Visual attention has a fundamental triaging role to play in shaping our perception of the world, by selecting certain relevant information for privileged processing, while filtering out other information (Fiebelkorn & Kastner, 2020).

Humans can regulate their spatial attention in a variety of ways, including via shifting the central focus of attention (i.e. shifts of attention), but also by changing the spatial extent of their attentional focus at a given location (i.e. their attentional breadth). That is, attentional breadth can be conceptualised as the spatial extent of the area over which attentional resources are applied (Goodhew, 2020). Attentional breadth can be narrow, such that attention is focussed over a small area, or it can be broad, spread over a larger area.

Attentional breadth can influence fundamental perceptual processes. The zoom-lens model espouses a trade-off between the size of the attended region and the magnitude of enhancement within it, such that going broad diminishes the perceptual enhancement relative to a narrower attentional breadth. That is, according to this model, perceptual processing is enhanced under narrower attentional breadths (Eriksen & St. James, 1986; Müller et al., 2003). Consistent with this, many studies have found that a narrower attentional breadth enhances fine-grained spatial resolution (Balz & Hock, 1997; Goodhew et al., 2017; Lawrence et al., 2020; Mounts & Edwards, 2016). In contrast, a broader attentional breadth enhances the processing of more complex stimuli such as faces (Gao et al., 2011; Gerlach & Starrfelt, 2018; Macrae & Lewis, 2002).

It has also been claimed that a broad attentional breadth mitigates the impact of negative emotionally salient stimuli on attention and performance (e.g. Gable & Harmon-Jones, 2012). If this is true, then it would be an exciting outcome because increased attentional engagement with emotionally salient stimuli is often associated with increased negative affect and anxiety in particular (Bar-Haim et al., 2007; Onie & Most, 2017), and it is claimed that training individuals to not attentionally engage with such stimuli can mitigate these symptoms (MacLeod & Clarke, 2015; but see Cristea et al., 2015). If a broader attentional breadth does make individuals more resilient against the effect of emotionally salient stimuli on attention and performance, then this could have important clinical and therapeutic implications. Consistent with this, it has been reported that regular broadening of attention can mitigate negative mood (Gu et al., 2017). However, there are reasons to question the evidence on which this claim is based. Therefore, the goal of the present work was to rigorously test this claim. First, however, we articulate the distinction between experienced emotion and its consequences and the effect of emotional salience of stimuli with respect to attention and performance.

Emotionally salient stimuli can have two dissociable effects: (1) they can induce an emotional state in the person, and (2) they can influence attention and perceptual performance, without there necessarily being a change in the viewer’s emotional state. We could consider (1) the experiential aspect of emotional-salience, because it is about the emotional experience, while a short-hand for (2) is the attentional aspect of emotional-salience, because it is about the impact that these stimuli have on attention and performance. For example, when a picture of a spider makes a person feel fear, then this is an example of emotionally salient stimuli impacting emotional experience, whereas when a picture of spider influences visual search performance, then this can be an example of emotional salience influencing the attentional aspect. Of course, the experiential and attentional components may be associated and interact. Indeed, there is a large literature investigating how the experience of emotion influences attentional allocation (e.g. Vanlessen et al., 2016). But it is important to appreciate that emotional salience can have these two dissociable effects. The focus of the present study was on whether broad attentional breadth can buffer the impact of negative emotionally salient stimuli on attention and performance, rather than on the experience of emotion. Below we review two key studies that speak to this issue.

Srinivasan and Hanif (2010) found that inducing a broad attentional breadth facilitated participants’ response speed to identify happy facial expressions relative to a narrow attentional breadth, whereas there was no reliable difference in response speed to sad faces for the different attentional breadths. While this is suggestive of attentional breadth influencing the processing of emotionally salient stimuli, there are multiple issues that preclude drawing this conclusion definitively. First, this pattern is consistent with promoting the processing of positive stimuli, rather than making individuals less susceptible to the effects of negative stimuli. Second, response classification was confounded with stimulus type. That is, participants’ task was to judge whether the face presented was happy or sad, and so an increased proportion of correct happy responses could simply be the product of a lowered response threshold to respond happy, rather than a true change in sensitivity to detecting these stimuli. Third, it is possible that happy faces are more difficult to classify than sad faces, and therefore benefit to a greater extent from the generic face-processing benefits of a broad attentional breadth described above.

Gable and Harmon-Jones (2012) found that when the task compelled participants to broaden their attention, they showed a reduced N1 amplitude following negative images. However, there is compelling evidence that perceptual processing of stimuli can be reduced under broad relative to narrow attentional breadth, consistent with the influential zoom-lens model (Eriksen & St. James, 1986; Lawrence et al., 2020; Müller et al., 2003). The N1 is a relatively early component, that is thought to gauge target-identification processes (Vogel & Luck, 2000), and is modulated by attention (Saupe et al., 2013). Therefore, according to the zoom-lens model, it is possible that the reduced N1 amplitude in the broad-attention condition was entirely a product of reduced target-related perceptual processing in the condition where attention was less focussed on the target, rather than anything selective regarding emotion processing (see also Goodhew, 2020).

A zoom-lens based explanation for their N1 results could have been directly tested if Gable and Harmon-Jones (2012) had included a positive condition. For example, if it was found that processing of positive stimuli was enhanced under a broad relative to a narrow attentional breadth (i.e. a crossover interaction), this would refute a zoom-lens based explanation. Gable and Harmon-Jones (2012) did not include such a condition. However, in one of these authors’ earlier studies, they found that the N1 amplitude to appetitive (positive) images was also reduced under the global (broad) relative to the local (narrow) condition (Gable & Harmon-Jones, 2011). This means that across these two studies, the N1 in response to both positive and negative images was reduced when attention was broad, which is consistent with the zoom-lens-model explanation. In other words, it is entirely possible that the reduction in N1 amplitude observed in the global condition in both studies (i.e. for both positive and negative stimuli) is a product of a generic zoom-lens effect whereby all early perceptual processing is attenuated under a broad attentional breadth, with no need to refer to emotion-specific mechanisms. If so, then this undermines any potential therapeutic benefit of going broad, because it means that any buffering against the effect of emotionally salient stimuli by going broad comes at the cost of perceptual resolution. It is akin to a person who needs glasses taking off their glasses so that they do not perceive threatening stimuli, but therefore also cannot drive safely, read this text, or recognise the face of a friend. However, to test this more rigorously, both positive and negative emotionally salient stimuli, and neutral control stimuli, should all be included within the same experiment, where the effect of attentional breadth is assessed. This is what we did here across a series of studies.

Experiment 1: emotion-induced slowing intermixed valence

The purpose of Experiment 1 was to assess if broadened attentional breadth resulted in reduced impact of emotionally salient stimuli on performance relative to a narrower attentional breadth. Here, therefore, attentional breadth was manipulated to be broad versus narrow in different blocks, and the effect of emotionally salient stimuli was assessed within these different blocks. That is, in one block of trials participants’ task always required them to attend to the global level of the Navon stimuli and would therefore induce a broad attentional breadth throughout this block. In the other block of trials, participants’ task always required them to attend to the local level of the Navon stimuli and would therefore induce a narrow attentional breadth throughout that block.

We then assessed the influence of task-irrelevant negative, neutral, and positive stimuli presented prior to the Navon stimulus on each trial on task performance in these different blocks. Previous work has shown that negative and positive stimuli result in slowed responses to subsequent stimuli relative to neutral stimuli, an effect called emotion-induced slowing (Goodhew & Edwards, 2022). We used this as the metric of emotionally salient stimuli impacting perception and cognition, and then examined how broad versus narrow attentional breadth affected the magnitude of this metric.

If a broadened attentional breadth selectively reduces the impact of negative emotionally salient stimuli on performance, then emotion-induced slowing following negative stimuli should be reduced when participants attend to the global level of the Navon stimuli. If a broadened attentional breadth reduces the impact of both negative and positive stimuli, then emotion-induced slowing following negative and positive stimuli should be reduced when participants are attending to the global level. If a broadened attentional breadth has no impact, then equivalent-magnitude emotion-induced slowing should be present in both the global and local conditions. These patterns are shown in Figure 1.

Figure 1. Predicted pattern of results according to different hypotheses.

Note. Emotion-induced slowing is anticipated when participants have a narrow breadth of attention (i.e. when attending to the local level of Navon). This is depicted in the solid line producing a V-shape in all three graphs. The dashed lines indicate the differential predictions for what will happen when participants have a broad attentional breadth (i.e. when attending to the global level of Navon) according to different hypotheses, as described below. Note. Selective Negative: If a broadened attentional breadth selectively reduces the impact of negative emotionally salient stimuli on performance, emotion-induced slowing should be selectively reduced following negative stimuli in the broad attentional breadth condition relative to the narrow attentional breadth condition. This is depicted by the dashed line showing responses quickening for the negative condition under broad attentional breadth, such that the difference in RTs between the negative and neutral conditions is reduced. In other words, the steepness of the V-shape should be selectively reduced for negative stimuli under broad attentional breadth. Note. All Emotional: If a broadened attentional breadth reduces the impact of all emotionally salient stimuli on performance, emotion-induced slowing should be reduced for both negative and positive stimuli in the broad attentional breadth condition relative to the narrow attentional breadth condition. This is depicted by the dashed line showing responses quickening for the negative and positive condition under broad attentional breadth, such that the difference in RTs between the neutral condition and both the negative and positive conditions is reduced. In other words, the steepness of the V-shape should be reduced for negative and positive stimuli under broad attentional breadth. Note. No Effect: If a broadened attentional breadth does not reduce the impact of any emotionally salient stimuli on performance, then emotion-induced slowing should be observed for both negative and positive stimuli and be of equivalent magnitude in both the narrow and broad attentional breadth conditions. In other words, an equivalent V-shape should be observed in both the broad and narrow attentional breadth conditions. Note. Diagrams are not necessarily to scale and are used for illustrative purposes only.

Method

Data

We report how we determined our sample size, data exclusions, manipulations, and measures in the study. Raw data from Testable output files was collated into a spreadsheet for analysis via custom written code in MATLAB, and then statistical analysis was performed in JASP version 0.14.1 (Team JASP, 2020). All ds reported are Cohen’s d. Note that in JASP, Cohen’s d in paired samples t-tests is calculated via the mean of the difference divided by the standard deviation of the difference. Data are available at Open Science Framework (OSF): https://osf.io/n9kq5/.

Participants

A power analysis conducted in G*Power (t-tests > Means: difference between two dependent means) assuming a medium effect size¹ and an alpha of .05 indicates that N = 54 is required for 95% power. We assumed that up to 10% of participants may be excluded for failure to comply with task instructions, and therefore we sought to recruit N = 60. This also satisfies the power requirements for the ANOVA, see Supplementary Material.

A total of 60 participants completed the experiment. They were recruited from the Australian National University’s (ANU) research participation website (SONA), and they were offered research participation course credit in compensation for their time if they were enrolled in eligible courses and were informed that they were was no compensation offered if they were not. The Study ad warned about the graphic images. Participants were required to be at least 18 years of age. For all the present experiments, prior to participation, participants were presented with an onscreen Information Sheet, which included highlighted information about the nature of the pictures and that participation was entirely voluntary. Only participants who selected “Agree” to indicate their explicit consent continued to the study. The ethical components of all the experiments were approved by the ANU Science and Medical Delegated Human Ethics Committee (Protocol number 2018.633). Participant demographics for this and all subsequent experiments is available in the Supplementary Material.

Stimuli

Navon stimuli are commonly used to induce different attentional breadths (Flevaris et al., 2014; Gable & Harmon-Jones, 2012; Gao et al., 2011; Goodhew & Edwards, 2022; Hanif et al., 2012; Macrae & Lewis, 2002). Navon stimuli are hierarchical stimuli where a larger letter is made up of multiple instances of a smaller letter (Navon, 1977; for a review see Goodhew, 2020). There is compelling evidence that when participants attend to the local elements, the region of enhanced activation in early visual areas is narrower in scope compared with when participants attend to the global letter (Sasaki et al., 2001). Further, the magnitude of activation within that region is more strongly enhanced when attending to the local letters relative to the global letter, consistent with the zoom-lens model (Sasaki et al., 2001). Here, therefore, we manipulated participants’ attentional breadth to be narrow for one block of trials by having them attend to the local elements and broad for another block of trials by having them attend to the global elements.

Navon stimuli come in different varieties. Here, we selected those that have produced demonstrably different behavioural results when participants are attending to their global versus local elements (Goodhew & Edwards, 2022; Goodhew & Plummer, 2019). This set consists of eight hierarchical letters, where the letter at the global level always differs from the letter at the local level. The target letters are T and H, while the non-target letters are E and F. There is always one and only one target in each stimulus. For half of the stimuli, the target appears at the global level, for the other half, the target appears at the local level. Specifically, the stimuli are: Eh, Et, Fh, Ft, He, Hf, Te, and Tf (where the uppercase letter denotes the global letter, and the lowercase letter denotes the local letter, although in the Navon stimuli themselves the letters at both levels were always uppercase). For reference, following calibration, these subtended 8 × 8 cm on an iMac monitor.

To manipulate emotional salience, we selected images from the International Affective Picture System (IAPS). IAPS images have been rated on the dimensions of valence (pleasantness: unpleasant to pleasant, with neutral in between) and arousal (low arousal to high arousal) (Lang et al., 2008). Physiological evidence provides converging evidence for these subjective ratings (e.g. Cuthbert et al., 2000; Lang et al., 1998). We selected 20 negative images (e.g. mutilated bodies and violence), 20 neutral images (e.g. person sitting on a bench with neutral expression, mushrooms), and 20 positive images (e.g. couples engaged in romantic and sexual acts, people participating in extreme sports, and cupcakes). Information on their ratings and specific IAPS numbers are available in the Supplementary Material. Following calibration, the images subtended 8 cm high by 10 cm wide on an iMac screen.

Apparatus

The experiment was run online remotely via Testable (www.testable.org). Testable to has been found to have inter-trial timing variability in the range of 3.2–8.4 ms (Bridges et al., 2020). We have found studies run online via Testable sensitive to effects of experimental manipulations on RT (Goodhew & Edwards, 2022). We used Testable’s inbuilt calibration function (for more information, see Supplementary Material). The screen background was white. Text and Navon stimuli were shown in black, while the IAPS images were shown in colour. All images were presented centred on the centre of the screen.

Procedure

Each trial began with a fixation cross shown for 800 ms, followed by the task-irrelevant image (negative, neutral, or positive) for 1000 ms, then the screen was blank for 100 ms, and then the Navon stimulus was presented until response (see Figure 2). Participants’ task was to identify whether the letter “T” or “H” was at the prescribed level for that block as quickly and accurately as possible. RT was measured from the onset of the Navon stimulus. Participants were randomly assigned to complete either the Global or Local block first. Each block consisted of 60 trials randomly intermixed, 20 of each image valence (i.e. negative, neutral, positive), where each of the four Navon stimuli with the target letter at the prescribed level was used an equal number of times for each valence.

Figure 2. An illustration of an example trial in Experiment 1.

Note. This is an example of a positive trial from the global target block. The cupcakes image is a photo by Brian Chan on Unsplash (a database of images freely available for use without permission), designed to be indicative of an IAPS image without showing one here. Stimuli are not necessarily to scale.

The experiment began with a CAPTCHA (designed to detect bots), then calibration, followed by the Information Sheet, with a Disagree and an Agree option. For those who selected Agree, they then entered their demographic information, before receiving a reminder about how to withdraw and contact details of helplines if in distress that were provided in the Information Sheet. They then completed either the Global or Local condition first (randomly assigned), and received onscreen task instructions, including that their task was to identify whether the T or H was present at the prescribed level in the stimulus as quickly and accurately as possible using the T and H keys on the keyboard. They then completed eight practice trials with onscreen feedback on each trial (i.e. “correct” or “incorrect” presented in the centre of the screen for 1000 ms). The practice block used eight neutral images that were different from the specific neutral images used in the experimenter proper, and each of the four Navon stimuli for that block was presented twice (randomly intermixed). Following the practice, participants were reminded of the task instructions before progressing to the main block. Once they had completed their first block, they were shown the instructions for their second block, completed an eight-trial practice for that block, before being reminded of the instructions and then completing that block. After they had completed both blocks, they presented with 20 pleasant images (e.g. gardens, cute puppies, smiling people) designed to mitigate any effect of exposure to the negative images during the experiment. Finally, they were presented with a debrief text about the study and reminded again of helpline contact details if in distress.

Results

Individual trials were excluded from further analysis if the participant had responded quicker than 100 ms, or slower than 3 standard deviations above their mean RT for all trials (see Goodhew et al., 2020 for recommendations of relative RT screening). This led to the exclusion of only a small proportion of trials (M = 1.42%, SD = 0.82%, range = 0–3.33%). Next, each participants’ average accuracy and average RT on correct response trials was calculated. Participants whose average accuracy in any condition (i.e. any of the six combinations of Navon level and image valence) fell below 75% was excluded from further analysis, as we reasoned that this is indicative of being either unwilling or unable to comply with the task instructions. This is because the target stimuli were displayed until response, and therefore near-ceiling accuracy is expected: 75% was chosen as the particular minimum performance criterion for inclusion because it is halfway between chance-level performance (50%) and ceiling (100%) for a two-alternative forced-choice task, and therefore scores below this are closer to chance than ceiling. Three participants’ data were excluded on this basis, leaving a final sample for analysis of N = 57.

Participants’ average RTs were submitted to a 2 (Navon Level: Global versus Local) × 3 (Image Valence: Negative, Neutral, Positive) repeated-measures ANOVA.² For all effects reported here involving variables with more than two levels, if Mauchly’s test indicated sphericity was violated, then the Greenhouse Geisser correction was applied. This revealed a significant main effect of Valence, F(1.70, 95.08) = 6.93, p = .003,  = .110. In contrast, neither the main effect of Navon Level, F(1, 56) = 0.10, p = .756, ${\eta }_p^2$ = .002, nor the interaction between Valence and Navon Level, F(1.40, 95.08) = 0.56, p = .513, ${\eta }_p^2$ = .010, were significant.³,⁴

Post hoc tests indicated that RTs were significantly slower in the Negative condition compared with the Neutral condition (p_holm =.021, d = .35), and RTs were significantly slower in the Positive condition than the Neutral Condition (p_holm = .001, d = .48), whereas they did not significantly differ between the Positive and Negative conditions (p_holm = .320, d = .13). This demonstrates that both the Positive and Negative conditions produced emotion-induced slowing relative to Neutral, and the absence of the interaction indicates that the magnitude of this emotion-induced slowing was invariant to Navon level (see Figure 3). Bayesian analyses are reported in the Supplementary Material and converge with the conclusions from the frequentist approach.

Figure 3. Mean RTs to identify targets at Global versus Local level following each image valence in Experiment 1.

Note. Error bars represent standard errors.

Average accuracy for each condition was high (Means > 95%). Participants’ average accuracies were submitted to a 2 (Navon Level: Global versus Local) x 3 (Image Valence: Negative, Neutral, Positive) repeated-measures ANOVA. This revealed no significant main effect of Navon Level (p = .061, ${\eta }_p^2$ = .061), no significant main effect of Valence (p = .500, ${\eta }_p^2$ = .012), and no significant interaction between Navon Level and Valence, (p = .223, ${\eta }_p^2$ = .026).

Discussion

The purpose of this study was to determine whether broadening attentional breadth buffers against the detrimental impact of emotionally salient stimuli on performance. To test this, we had participants attend to the global versus local elements of Navon stimuli to induce broad versus narrow attentional breadths in different blocks of trials. On each trial in these blocks, prior to the Navon stimulus, a task-irrelevant image was presented that was either negative, neutral, or positive in valence. We observed emotion-induced slowing, whereby responses to the Navon stimuli were slower following negative and positive images relative to neutral images. Crucially, the magnitude of this emotion-induced slowing effect was unchanged in the global relative to the local Navon level condition. In other words, emotionally salient stimuli demonstrably impacted performance, and broadening attentional breadth did not protect against this impact.

Experiment 2: zoom-lens perceptual outcome

The results of Experiment 1 indicated that broadening attentional breadth by virtue of having participants attend to the global (vs local) elements of Navon stimuli had no discernible mitigating influence on the impact of emotionally salient stimuli. While we selected the Navon stimuli to induce different attentional breadths because previous research demonstrates their effectiveness, here we sought to confirm their effectiveness via a control experiment. Here, we used the identical stimuli to Experiment 1 to induce attentional breadth and measured their impact on the perceptual processing of small spatially-detailed stimulus.

In doing so, a key methodological decision was the proportion of Navon trials versus perceptual outcome trials. There need to be sufficient Navon trials to successfully induce attentional breadth, and sufficient single letter trials to reliably measure the perceptual outcome of this induced attentional breadth. However, the proportion of perceptual outcome trials cannot be too large, because prolonged series of single small-stimulus trials will narrow attentional breadth away from the intended broad attentional breadth in the Global condition, thereby undermining the differential manipulation of attentional breadth. Previous research has shown that 80/20 induction/outcome trials can successfully achieve this balance (e.g. Goodhew & Plummer, 2019). Therefore, this split was used here. Specifically, in one block, the majority (80%) of trials required participants to attend to the global element of the Navon to broaden their attentional breadth for that block, and in another block, the majority (80%) of trials required participants to attend to the local elements of the Navon to narrow their attentional breadth for that block. We then gauged the influence of these attentional breadths on the processing of a small letter stimulus which appeared on the minority (20%) of trials in each block. If attending to the different levels of these Navon stimuli does indeed induce different attentional breadths, then according to the zoom lens model, participants’ responses to identify the small letter stimulus should be quicker and/or more accurate when they have a narrower (relative to a broader) attentional breadth.

Method

Participants

Based on the same power analysis as Experiment 1, 60 participants were recruited from Testable Minds and were offered $4 (USD) for their time.

Stimuli and apparatus

The identical Navon stimuli as Experiment 1 were used here to induce narrow versus broad attentional breadth. Small black letters “T” and “H” (approximately 0.5 cm × 0.5 cm) presented in isolation in the centre of the screen were used to assess the perceptual outcome of these different attentional breadths. The IAPS images were not used in Experiment 2. The identical apparatus as Experiment 1 was used.

Procedure

Each trial began with a fixation cross shown for 800 ms, followed by either the Navon stimulus until response, or the isolated small letter stimulus until response. If the Navon stimulus was presented, then participants’ task was to identify whether the letter “T” or “H” was at the prescribed level for that block as quickly and accurately as possible. If the small letter was presented, then participants’ task was to identify whether it was the letter “T” or “H” as quickly and accurately as possible. RT was measured from the onset of the Navon or letter stimulus. Participants were randomly assigned to complete either the Global or Local block first. Each block consisted of 120 trials, where 96 trials were Navon trials, and 24 trials were letter trials. These trial types were randomly intermixed within each block, with the constraint that the first four trials of the block were always Navon trials.

The experiment began with a CAPTCHA, then calibration, then the Information Sheet. For those who selected Agree, they then entered their demographic information, before progressing to complete either the Global or Local condition first (randomly assigned), where they received onscreen task instructions. Next, they completed six practice trials (four Navon, and two letter stimuli trials randomly intermixed) with onscreen feedback, before being reminded of the task instructions for that block and then completing the block. Then they received instructions for their second block, completed six practice trials with onscreen feedback, before being reminded of the task instructions and progressing to complete the block.

Results

Individual trials were excluded from further analysis if the participant had responded quicker than 100 ms, or slower than 3 standard deviations above their mean RT for all trials. This led to the exclusion of only a small number of trials (M = 1.56%, SD = 0.75, range = 0.42–3.33%). Next, each participants’ average accuracy and average RT on correct response trials was calculated. Participants whose average accuracy in responding to the Navon stimulus in any condition (i.e. any of the four averages of global or local Navon accuracy) fell below 75% was excluded from further analysis. One participant’s data was excluded on this basis, leaving a final sample for analysis of N = 59. Given that the perceptual outcome task is one that participants perform only infrequently and therefore may be more difficult, we did not impose an accuracy criterion for it, but minimum mean accuracy from an individual participant was 83.3% and 82.6% for the Global and Local outcome conditions.

For the small single letter perceptual outcome stimulus, a paired samples t-test showed that responses were significantly faster in the Local condition compared with the Global condition, t(58) = −2.39, p = .020, d = −.31⁵ (see Figure 4). This is consistent with the zoom-lens model. Crucially, it confirms that the Navon stimuli and task were effective in manipulating different attentional breadths that demonstrably impacted performance.

Figure 4. Mean RTs to identify the small letter perceptual outcome as a function of Navon block level in Experiment 2.

Note. Error bars represent standard errors.

A paired-samples t-test showed that accuracy for the small letter identification was greater in the Local block (M = 97.99%, SD = 3.06) than the Global block (M = 95.74%, SD = 5.05), t(58) = 3.44, p = .001, d = .45.⁶ Given that the Global condition resulted in both slower RTs and reduced accuracy, this analysis demonstrates that a speed-accuracy trade-off does not contaminate the interpretation of the RT data above.

Discussion

Experiment 2 demonstrated that the Navon stimuli used in Experiment 1 could produce demonstrably different perceptual outcomes. That is, responses to simple stimuli interleaved on a minority of trials were faster and more accurate when the main task required attending to the local rather than the global level of the Navon stimuli. This is consistent with the Navon stimuli inducing narrow versus broad attentional breadths.

Experiment 3: emotion-induced slowing valence blocked

Experiment 1 demonstrated emotion-induced slowing that was impervious to attentional breadth, and Experiment 2 validated that the Navon stimuli could induce different attentional breadths. However, in Experiment 1, while the task-relevant Navon level was fixed within a block, image valence type was intermixed within blocks. It is possible that the impact of emotional salience on performance will be larger, and/or that the effect modulatory effect of attentional breadth will become apparent when image valence is also blocked. This was assessed in Experiment 3.

Method

Based on the same power analysis as Experiment 1, 60 participants were recruited from the same ANU SONA system as Experiment 1. The same stimuli, apparatus, and procedure was used as Experiment 1, except that here, both Navon Level and Image Valence was blocked. Participants were randomly assigned to complete either the Global or Local condition first and the other one second, and within each of those conditions, the order of image valence (Negative, Neutral, and Positive) was independently randomised for each participant. For example, a participant assigned to the Global condition first would complete the Global Negative, Global Neutral, and Global Positive conditions in a randomised order, before progressing to complete the Local Negative, Local Neutral, and Local Positive conditions in a randomised order. Participants completed a practice block (eight practice trials with neutral images and Navon target letters at the prescribed level for that block) prior to commencing the Local and Global conditions (i.e., two practice blocks total).

Results

Individual trials were excluded from further analysis if they were quicker than 100 ms or slower than 3 standard deviations above that participant’s mean RT for all trials (M excluded = 1.5%, SD = 1.0%, range = 0–4.2%). One participant’s data was excluded for falling below 75% accuracy in a condition, leaving a final sample for analysis of N = 59.

Participants’ average RTs were submitted to a 2 (Navon Level: Global versus Local) × 3 (Image Valence: Negative, Neutral, or Positive) repeated-measures ANOVA. This revealed a significant main effect of Valence, F(1.82, 105.28) = 6.49, p = .003, ${\eta }_p^2$= .101. Neither the main effect of Navon Level, F(1, 58) = .43, p = .513, ${\eta }_p^2$= .007, nor the interaction between Valence and Navon Level, F(2, 116) = .36, p = .702, ${\eta }_p^2$= .006, were significant. These results are shown in Figure 5.

Figure 5. Mean RTs to identify targets at Global versus Local level following each image valence in Experiment 3.

Note. Error bars represent standard error.

Post hoc tests indicated that RTs were significantly slower in the Negative condition compared with the Neutral condition (p_holm = .007, d = .39), and significantly slower in the Positive condition compared with the Neutral condition (p_holm = .005, d = .42), whereas the Positive and Negative conditions did not reliably differ (p_holm = .781, d = .04).

Accuracy was high (Means > =  97%). Participants’ average accuracy was submitted to the same repeated-measures ANOVA, in which no effects were significant (ps > =  .108, ${\eta }_p^2$s < =  .040).

Discussion

Experiment 3 replicated the same results as Experiment 1 when image valence type was blocked. That is, once again, emotion-induced slowing was observed, which was impervious to whether attentional breadth was manipulated to be broad or narrow.

Experiment 4: emotion induced slowing with shorter image duration and combined perceptual outcome combined

The purpose of Experiment 4 was twofold. First, while Experiment 2 demonstrated that the Navon stimuli used to manipulate attentional breadth in Experiment 1 and Experiment 3 were effective in inducing different attentional breadths, this was demonstrated in an experiment that did not contain emotionally salient stimuli, which could themselves have an impact on attentional breadth. Therefore, one goal of Experiment 4 was to test whether attending to the local versus global elements of the Navon stimuli was effective in inducing different attentional breadths within the same experimental context in which emotion-induced slowing effect was measured. To do this, 80% of trials were induction trials (which contained both a task-irrelevant emotionally salient stimulus and a Navon stimulus, as per trials in Experiments 1 and 3), and 20% of trials were outcome trials where a single letter was presented alone (as per outcome trials in Experiment 2). The induction trials were designed to both (a) induce different attentional breadths and (b) measure emotion-induced slowing, while the outcome trials were designed to assess the perceptual effect of the induced attentional breadth. If the attentional breadth induction is successful, then as per Experiment 2, a zoom-lens effect should be observed for the outcome trials. Then the effect of these different attentional breadths on emotion-induced slowing can be assessed (i.e. whether a broad attentional breadth reduces emotion-induced slowing).

Second, in Experiments 1 and 3, the emotionally salient images were shown for 1000 ms. However, attentional biases to emotionally salient stimuli can vary over time, such as showing patterns of initial vigilance toward threat followed by avoidance of it (e.g. Derakshan et al., 2007; Mogg et al., 2004; Proud et al., 2020). Here, therefore, we reduced their duration to 500 ms, to make it more akin to durations where attentional engagement effects are typically observed (e.g. in the dot-probe) (e.g. Mogg et al., 2004). Image valence was blocked, and the order of block completion was randomised as per Experiment 3.

Method

Sixty participants were recruited from the ANU SONA system. The same stimuli, apparatus, and procedure was used as Experiment 3, with two key exceptions. First, the majority of trials were induction trials, where a task-irrelevant emotionally salient stimulus was presented for 500 ms, followed by a Navon stimulus until response (100 ms interstimulus interval between these two stimuli still retained). These were akin to all trials in Experiments 1 and 3, except for the change in stimulus duration (500 ms rather than 1000 ms). Second, the minority of trials were outcome trials, where a single letter stimulus was presented until response (as per outcome trials in Experiment 2). As per the previous experiments, on induction trials, participants’ task was to identify the target letter at the prescribed level of the Navon (T or H), while on outcome trials, it was to identify the isolated letter (T or H).

Like Experiment 2, we were aiming for approximately an 80%/20% split of induction versus outcome trials. However, here we also sought to preserve the same 20 images per valence category of Experiment 1 and 3. Therefore, in each combination of Navon Level and Valence block (e.g. Local Positive), we presented the 20 images (e.g. 20 positive images) in that category twice (i.e. 40 induction trials), and presented the 12 single letter trials (6 H, 6 T) randomly intermixed within that block (this corresponds to approximately a 77%/23% split of induction versus outcome trials). The random assignment of block order was as per Experiment 3.

Practice blocks occurred prior to participants completing the Global and Local conditions (i.e., two practice blocks total). In these practice blocks, the Navon level of eight neutral induction trials corresponded to the one for that upcoming three blocks, and two outcome trials (one H one T), were randomly intermixed.

Results

Individual trials were excluded from further analysis if quicker than 100 ms or slower than 3 standard deviations above that participant’s mean RT for all trials (M excluded = 1.5%, SD = 0.7%, range = 0–3.2%). Two participants’ data was excluded for falling below 75% accuracy in a condition, leaving a final sample of N = 58. As per Experiment 2, this accuracy screening was only applied to induction trials. The lowest condition accuracy on outcome trials for any participant was 70%.

Participants’ average RTs to Navon stimuli were submitted to a 2 (Navon Level: Global versus Local) × 3 (Image Valence: Negative, Neutral, Positive) repeated-measures ANOVA. This revealed a significant main effect of Navon Level, F(1, 57) = 18.12, p < .001, ${\eta }_p^2$= .241, such that responses were slower in the Global condition compared with the Local. There was also a significant main effect of Valence, F(1.16, 66.27) = 7.55, p = .006, ${\eta }_p^2$= .117, which did not interact with Target Level, F(1.42, 66.27) = 1.13, p = .311, ${\eta }_p^2$= .019. Post-hoc tests showed that responses in the Negative condition were significantly slower than in the Neutral condition (p_holm < .001, d = .51). None of the other comparisons were significant (ps > =  .065, ds < =  |.28|). In other words, emotion-induced slowing was reliably present only for the Negative condition (see Figure 6). When accuracy on Navon trials was submitted to the same analysis, there were no main effects or interactions (ps > =  .196 and ${\eta }_p^2$s < =  .028).

Figure 6. Mean RTs to identify targets at Global versus Local level following each image valence in Experiment 4.

Note. Error bars represent standard error.

Given that emotion-induced slowing was specific to the Negative condition (relative to Neutral), we focussed on assessing the perceptual outcome trials only for the Negative and Neutral trials (analysis with Positive included is reported in the Supplementary Material). That is, RTs on the single letter outcome trials were submitted to a 2 (Navon Level: Global versus Local) × 2 (Image Valence Condition: Negative versus Neutral) repeated-measures ANOVA. This demonstrated a significant main effect of Navon Level, F(1, 57) = 5.28, p = .025, ${\eta }_p^2$= .085, such that responses were slower on Global relative to Local trials, consistent with a zoom-lens effect. Neither the main effect of Valence, F(1, 57) = 2.94, p = .092, ${\eta }_p^2$= .049, nor the interaction between Target Level and Valence, F(1, 57) = .49, p = .486, ${\eta }_p^2$= .009, were significant. Accuracy to identify the single letter on the outcome trials were submitted to the same ANOVA, which showed no main effects or interactions (ps > =  .069, ${\eta }_p^2$_s < =  .057).

Discussion

Experiment 4 showed that: (a) negative emotionally salient stimuli had a detrimental effect on performance relative to neutral stimuli (i.e. slowed responses following negative versus neutral stimuli), and (b) attending to the local versus global elements of the Navon stimuli successfully induced different attentional breadths for these conditions (i.e. faster responses to small outcome targets in local versus global target blocks). Crucially, however, going broad did not buffer, because the effect of broadened attentional breadth did not reduce the impact of negative stimuli on performance (i.e. no reduction of emotion-induced slowing following negative images for broad versus narrow attentional breadth).

While Experiment 4 did not yield reliable emotion-induced slowing for positive stimuli, given that the primary clinical potential for broadening attention is via mitigating the effect of negative stimuli, it is most important that this experiment was able to show that attentional breadths that were sufficiently potent to produce demonstrably different perceptual outcomes did not mitigate emotion-induced slowing following negative stimuli. The purpose of the positive condition is in allowing assessment of whether any observed mitigation of emotion-inducing slowing is selective to negative valence, and since such mitigation was not observed for negative, the absence of emotion-induced slowing following positive stimuli here does not impede drawing key conclusions from this study.

Experiment 4 was the first study to show slower RTs to the Global level targets on the main induction trials. One possible explanation for this pattern was that the presence of the small letter outcome trials biased participants’ attentional breadth to be narrower that it would otherwise be in the Global blocks to facilitate their responses to these small stimuli. This would explain why responses were slowed to the Global targets – because attentional breadth was narrower than that which was optimal for the Global level. This effect did not eliminate the differential attentional breadth across conditions – there was demonstrable evidence that attentional breadth was still broader in the Global condition than it was in the Local condition, and thus we were able to draw conclusions about the effect of the attentional breadth. But it is possible that attentional breadth in the Global condition was narrower than it was in the Global condition in the previous experiments. For this reason, in Experiment 5, we removed the perceptual outcome component of the design.

Experiment 5: milder emotionally salient stimuli

The goal of Experiment 5 was to assess whether broadened attentional breadth would mitigate the impact of emotionally salient stimuli on attention and performance when the emotionally salient stimuli were less intense. The design akin to Experiment 3, but with less intense images.

Method

Sixty-two participants were recruited from the ANU SONA system. The same stimuli, apparatus, and procedure was used as Experiment 3, with one key change: we selected 20 less intense negative and 20 less intense positive emotionally salient IAPS images. The negative stimuli included images of animals (e.g. snakes, snarling dog), weapons, injuries, skulls, and negative facial expressions, while the positive stimuli included images of attractive men and women, couples engaged in romantic acts, and sports (see Supplementary Material for details). The same neutral images were used as Experiment 4. Another change was that the IAPS images were presented for 500ms (as per Experiment 4, rather than Experiment 3).

Results

Individual trials were excluded from further analysis if quicker than 100 ms or slower than 3 standard deviations above that participant’s mean RT for all trials (M excluded = 1.3%, SD = .9%, range = 0–3.3%). One participant’s data was excluded for falling below 75% accuracy in a condition, leaving a final sample of N = 61.

Participants’ average RTs to Navon stimuli were submitted to a 2 (Navon Level: Global versus Local) × 3 (Image Valence: Negative, Neutral, Positive) repeated-measures ANOVA. This showed a significant main effect of Navon Level, F(1, 60) = 9.48, p = .003, ${\eta }_p^2$= .136, such that responses were faster in the Local relative to the Global condition. There was a significant main effect of Valence, F(1.79, 107.21) = 8.85, p < .001, ${\eta }_p^2$= .129, and no significant interaction, F(2, 120) = 1.25, p = .290, ${\eta }_p^2$= .020. Post-hoc tests showed that responses were slower in the Negative condition compared with Neutral (p_holm = .042, d = .30), and were slower in the Positive condition compared with Neutral (p_holm < .001, d = −.54), while the Negative and Positive condition did not differ (p_holm = .066, d = −.24) (see Figure 7).

Figure 7. Mean RTs to identify targets at Global versus Local level following each image valence in Experiment 5.

Note. Error bars represent standard error.

When accuracy data were submitted to the same ANOVA, there were no main effects or interactions (ps > =  .363, ${\eta }_p^2$ < =  .014).

Discussion

Experiment 5 demonstrated that even when milder (i.e. less intense) emotionally salient stimuli were used, emotion-induced slowing was observed in equivalent magnitude under a broad and a narrow attentional breadth.

General discussion

Here, the goal was to rigorously test the claim that a broadened attentional breadth can buffer against the impact of negative emotionally salient stimuli on attention and performance. Across a series of experiments, we found robust effects of both attentional breadth and emotional salience on performance, but they never interacted. That is, there was no evidence that broad attentional breadth reduced the impact of emotionally salient stimuli. This challenges previous claims that broad attention buffers against the effect of negative emotionally salient stimuli (Gable & Harmon-Jones, 2012). Of course, it is impossible to rule out the possibility that under a different set of conditions, such an effect would emerge. For example, while we found that significantly reducing the intensity of the images did not result in broad attentional breadth buffering the impact of negative stimuli, it remains possible that for even lower intensity negative stimuli this effect may emerge. However, given that emotional salience and attentional breadth were each able to exert demonstrable effects on perception across multiple experiments and across different stimulus sets (higher versus lower intensity emotional salience), we propose that the absence of their interaction here is meaningful. It is also possible that broad attentional breadth mitigates the impact of negative stimuli, but that the effect size is smaller than that which the present experiments were powered to detect. While we cannot rule out this possibility, there are two important counterpoints to this possibility. First, here the studies were powered to detect a medium effect – if there is an effect but it is considerably smaller than this, then this would curtail its clinical utility. Second, the Bayesian analyses provided support in favour of the null hypothesis, rather than indicating that there was simply insufficient power to reject the null hypothesis in the frequentist analysis.

We suggest that the evidence for reduced N1 amplitudes in response to negative and positive stimuli under broad versus narrow attentional breadths in previous work (Gable & Harmon-Jones, 2011, p. 2012) can be explained within the zoom-lens framework, whereby N1 amplitudes in response to all stimuli is reduced under a broad attentional breadth. This means that these previous studies do not support the supposition of attentional breadth selectively attenuating the effect of negative emotionally salient stimuli. Instead, it is analogous to a person removing their glasses, thereby reducing their perceptual sensitivity to all stimuli.

What then does attentional breadth do? It does impact spatial acuity (Balz & Hock, 1997; Goodhew et al., 2016; Goodhew et al., 2017; Lawrence et al., 2020; Mounts & Edwards, 2016). This is consistent with the functional argument that spatial attention mitigates the loss of spatial acuity in peripheral vision (Carrasco, 2011). It may also impact the processing of more complex stimuli such as faces, but via already well-understood means. That is, it is well known that faces benefit from holistic rather than piecemeal featural processing (Maurer et al., 2002), and a broad attentional breadth could simply promote the former over the latter. Both effects are relatively low-level. If the primary functional purpose of attentional breadth is to allow narrowing to compensate for spatial acuity decrements, or broadening to encompass the entirety of large stimuli that need to be processed holistically, then more complex processes may simply fall outside its ambit. The present results are consistent with this notion. Given this, other claims of attentional breadth affecting diverse psychological processes such as creativity (Friedman et al., 2003) and self-regulation (Hanif et al., 2012) may need to be revisited.

In the following sections, we address an alternative explanation for the results that were observed here. That is, could the effect we called “emotion-induced slowing” here be a product of induced experienced emotion influencing attentional breadth, rather than their emotional salience influencing attention and performance? We think this is unlikely for several reasons. First, studies where pictures are used to induce different mood states typically use much longer presentation times than what we used here (e.g. 6 s rather than 1 s or less) (Gable & Harmon-Jones, 2008, 2010). Our stimulus durations are more akin to dot-probe studies, from which it is inferred that the emotional salience of stimuli influence attention, rather than inducing an experienced emotion. Second, the observed pattern of results is not consistent with any of the existing models of how emotion and motivation influence attentional breadth. That is, one influential model espouses that the valence of an emotion determines attentional breadth, such that people have a broader attentional breadth in a positive mood and narrow attentional breadth in a negative mood (Fredrickson, 2001; Fredrickson & Branigan, 2005). In the present study, such an effect would emerge as more efficient responses to global Navon targets following positive images and more efficient responses to local Navon targets following negative images, relative to neutral. This would entail an interaction between Navon target level and image valence, and this pattern did not emerge in any of our experiments.

The motivational-intensity model instead claims that affective states high in motivational intensity (e.g. desire, fear) narrow attentional breadth, whereas affective states low in motivational intensity (e.g. contentment, sadness) broaden attentional breadth (Gable & Harmon-Jones, 2008, 2010; Harmon-Jones et al., 2012). Elsewhere, the issues with common existing approaches to operationalising motivational intensity have been discussed (Campbell et al., 2021, 2023, in press). We did not measure participants’ motivational intensity in response to these specific images, but given the types of images that have been found to differ in their motivational intensity ratings (Campbell et al., 2021, 2023), it would be reasonable to hypothesise that the positive and negative conditions both entailed higher motivational intensity than the neutral, and thus according to this model, both of these conditions should narrow attentional breadth. This would emerge as more efficient responses to local Navon targets following negative and positive images relative to neutral. Again, this would entail an interaction between Navon target level and image valence, and we did not see this pattern emerge in any of our experiments. This suggests that what we were observing was a product of emotional salience influencing attention and performance, rather than experienced emotion doing so.

We cannot rule out the possibility that the repeated presentation of emotionally salient images did indeed induce an affective state in participants. That said, a similar pattern of results was observed across both experiments that used intermixed versus blocked presentations of similar-valence images. Further, whether participants completed the Local or Global condition trials had no demonstrable impact on the results. This is consistent with effects that are relatively invariant across time, whereas affective state is more likely to increase across the course of blocked presentation of an image type. However, it would be useful for future research to collect participant reports of experienced emotion throughout brief but repeated exposure to such stimuli.

Conclusion

In conclusion, here we found that the presentation of both positive and negative task-irrelevant pictures slowed responses to a subsequent target, an effect we call emotion-induced slowing. The magnitude of emotion-induced slowing was invariant to whether a broad or narrow attentional breadth was induced. These findings challenge the claim that a broadened attentional breadth buffers against the impact of emotionally salient stimuli on attention and performance.

Disclosure statement

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

Data availability statement

Raw data are available here: https://osf.io/n9kq5/.

Additional information

Funding

This work was supported by an Australian Research Council (ARC) Future Fellowship (FT170100021) awarded to SCG.

Notes

1 Note that the Size x Valence interaction in Experiment 1 of Goodhew and Edwards (2022) indicative of emotion-induced slowing following large images (the size used here) had a partial eta-squared of .070, which is slightly larger than a medium effect. This means that assuming a medium effect for this analysis is the more conservative approach to ensuring sufficient power.

2 The residuals of this ANOVA were not normally distributed. When a series of steps including transformation and outlier exclusion were used to make the residuals of this analysis not significantly different from normal, then the results were equivalent (see Supplementary Material for this analysis for Experiments 1, 3, 4, and 5, i.e. all Experiments where ANOVAs were performed).

3 Estimates of variability around the effect size for this Valence by Navon Level interaction are reported in the Supplementary Material for Experiments 1, 3, 4, and 5 (i.e. all Experiments where both these variables were included).

4 When Block Order (i.e. whether participants completed the Global or Local block first) was entered as a between-subjects factor in this ANOVA, this had no main effect or interaction with any other variables (ps > =  .105 and η_p²s < =  .042). This was true for subsequent Experiments, see Supplementary Material.

5 While normality was violated (Shapiro-Wilk p < .001), Wilcoxon signed-ranks tests showed the same result (p < .001, d = -.50)

6 While normality was violated, Wilcoxon signed-ranks tests showed the same result (p < .001, d = .64).