We based our calculation of sample size on the publication that we used as a reference to build the task and that we are replicating here (Ferri et al., 2013). They used n = 41 participants (Study 1). There, we focused on the size of effect that the attentional manipulation had on participants’ emotional ratings. Ferri et al. reported a rating of M = 3.22, SD = 0.77 for unpleasant images with arousing focus, and a rating of M = 2.11, SD = 0.75 for unpleasant images with non-arousing focus. This results in an effect size of g = 1.4. Using this effect size, we used the software G*Power 3.1 to estimate sample size considering a matched pairs t-test, one tail, α = 0.05, and a power of 0.95. This calculation resulted in a sample size of n = 8. In a second report that used the task, sample size was n = 51 and there was no report of rating data to calculate effect size (Ferri et al., 2016). Given that it is not a recommended practice to use estimates of effect size from isolated reports, and that published effect size estimates tend to be large and misleading (Button et al., 2013; Gelman & Carlin, 2014), we simply set out to have a sample size larger than the original studies (i.e., n = 41 and n = 51). Thus, we ended data collection at n = 55.

A total of 55 participants completed the task (32 female; Age: M = 21.9, SD = 4.1 years). Participants were recruited through printed posts and social media. The data collection process occurred in two different periods. The first occurred during 2019 (n = 39 participants). The second occurred in 2022 (n = 16 participants). Our inclusion criteria were to be older than 18 years old and to have completed secondary education. The exclusion criteria were a refusal to sign the written consent or a diagnosis of a neurological condition. The institutional Ethics Committee reviewed and approved the study. All participants signed a written consent form for participation.

All data was collected at the same location, in our laboratory. Participants performed the task individually in a sound-proof, dimly lit experimental room. We used a computer screen View Sonic XG2402, with a spatial resolution of 1920 x 1080 pixels and dimensions of 53.4 cm (width) and 30.1 cm (height) to present the instructions and images throughout the task. After reviewing and signing the informed consent, the entire procedure was explained to the participants. After this explanation, and before commencing the task properly, they were familiarized with the task by executing five exercise trials that used images that were not part of the task.

The AD task adopted the procedures described by Ferri and collaborators (Ferri et al., 2013). Our main changes to the task were two: first, participants provided emotional ratings using the original Manikin 1 to 9 scale (instead of the 1 to 4 scale used by Ferri et al.) and second, participants were asked to provide two ratings, one of emotional intensity and one of emotional valence (instead of intensity only as in Ferri et al.). See Figure 1A.

Thus, depending on focus conditions and stimuli (i.e., image) types, each trial belonged to one out of five possible categories (see Figure 1B). Stimuli could be neutral or unpleasant, and they were combined with three focus conditions. Unpleasant stimuli were paired with one of the following conditions: focus-free (no circle on the image), arousing focus (circle on an arousing part of the image), and non-arousing focus (circle on a non-arousing part of the image). Neutral stimuli, by definition, did not have arousing parts, and therefore, they were paired only with focus-free or non-arousing focus. See Figure 1B for an example. The 100 IAPS images used in this task were identical to the ones employed in previous studies (Ferri et al., 2013, 2016; Hajcak et al., 2009). Stimuli were either unpleasant (low valence, high intensity) or neutral (higher valence, lower intensity). Of the 100 stimuli, 60 were unpleasant, with low valence on the 1 to 9 scale (M = 2.28, SD = 1.47), and 40 were neutral, with higher valence (M = 5.14, SD = 1.28). Conversely, unpleasant stimuli had a significantly higher level of intensity on the 1 to 9 scale (M = 6.05, SD = 2.25) compared to neutral stimuli (M = 2.91, SD = 1.92). See Figure 1C for the distributions of valence and intensity values reported for the set of stimuli.

Each trial began by presenting the instructions written on the screen for 4 seconds. Instructions were “focus your attention on the circle” or “observe the image freely”. After instructions, five images of the same type (neutral or unpleasant) were presented for 4 seconds each. Then, a screen containing an image with Numbers 1 to 9 appeared, with the instruction “Rate the valence, from 1 to 9”. Below the instruction, an image depicting the Manikin figures (Lang et al., 2008) alongside a horizontal scale showing Numbers 1 to 9. Below Number 1 was a text: “Negative”, and below Number 9 was a text: “Positive”. The participant had 5 seconds to complete the rating. They did not have the instruction to answer quickly. Participants delivered their responses through a regular computer keyboard. After delivering the response, the number corresponding to the assigned rating was shown on the screen for 0.35 seconds. If the participant did not complete it, the next screen appeared, containing instructions for Intensity rating. The procedure to record the participant’s intensity rating was similar to the one for valence. After participant ratings, a small fixation circle was presented at the center of the screen. This fixation circle was kept through five homogeneous gray backgrounds: 40%, 10%, 50%, 30%, and 20% of light saturation, which lasted 4 seconds each. After the sequence was completed, a new trial started. The task ended after 20 trials. One out of three possible trial sequences was randomly assigned to each participant to prevent possible confounding order effects (see Figure 1D). See Table 1 for the details of these sequences. Due to the random assignment procedure, we ended up with different numbers of participants in each sequence.

All hardware controls and data acquisition (behavioral and physiological) routines were written in Matlab (Version: 9.6.0.1472908 (R2019a) Update 9), using the Psychophysics Toolbox extension (Brainard, 1997; Kleiner et al., 2007).

Eye movement data, used to monitor instruction-following behavior, was acquired using an Eyelink 1000 (SR Research Ltd., Mississauga, Ontario, Canada) with a 500 Hz sampling frequency. Throughout the task, participants sat in front of the computer screen and the eye tracker device and kept their heads in a forehead/chin rest (SR Research Ltd.) placed 56 cm from the screen. Gaze data files were converted to the ASCII format using Eyelink’s EDFConverter, and then analyzed using custom routines written in Matlab (Version: 9.6.0.1472908 (R2019a) Update 9). We discarded gaze data from trials in which gaze detection was lost due to excessive blinks or other software/recording errors (for example). This resulted in completely discarding 12 participants from the statistical descriptions and comparisons.

Gaze behavior was quantified as dwelling time: the percentage of total time spent by the gaze within the focus circle during the image presentation (4 seconds). The raw data contained the (X, Y) screen coordinates where the gaze was located on each sampled data point. We classified each sample by marking whether it was inside or outside the focus circle. This allowed us to count the number of samples inside the circle, divide it by the total number (4 s x 500 samples/s = 2000 samples), and express it as a percentage to obtain the dwelling time. We calculated this dwelling time for each image and the mean of the five images of a trial. This gave us a single value of dwelling time per trial. The overall dwelling time was obtained by averaging dwell times across trials.

To estimate the participants’ attentional abilities, we used a computerized version of the ANT, which measures the capacity for alertness, orientation, and attentional executive control (Fan et al., 2002). The efficiency of the alerting network was examined through differences in reaction times in response to a warning signal before the presentation of stimuli. The efficiency of the orienting network was examined through differences in reaction times in response to a cue indicating where a stimulus will appear. The efficiency of the executive control network was examined through differences in reaction times in response to the presentation of a central arrow surrounded by congruent or incongruent indicators (arrows in the same or different directions to the central arrow). The differences in the reaction times associated with using each attentional system were used as a score to evaluate the performance of each attentional network (Alertness, Orientation, and Executive Control Score). For all these measurements, we followed the standard previously published procedures (Fan et al., 2002).

Given that participants completed the ANT after the experimental task, we had two periods of data collection (See Method section, Participants subsection). However, due to technical problems in retrieving the ANT data from the software, we were able to use only the data from the first period. And in this period (n = 39), data was lost for 4 participants. Therefore, we have ANT data only for 35 participants.

In general, data are summarized by means (M), medians (Mdn), standard deviations (SD), and inter-quantile range (IQR). For two-group comparisons, we implemented a non-parametrical permutation test based on the t-statistic. This allowed us to deal with heterogeneity in our data, which were not always normally distributed. We based our tests on previously published protocols (Ernst, 2004; Maris & Oostenveld, 2007). First, we computed the regular two-sample t-test and saved the corresponding t-value (i.e., the t-statistic value). We then randomly permuted (reorganized) the values from both groups, forming two new groups, and implemented the t-test, calculating the new t-value and saving it. We repeated this permutation procedure 1500 times and constructed a distribution of the permutation-based t-values. We then computed the p-value as the proportion of t-values in the distribution equal to or larger than the non-permuted t-value. Therefore, we report the non-permuted t-value, the corresponding degrees of freedom, and the p-value. For each two-group comparison, we also present the Hedges’ g statistic as a measure of effect size (Hedges, 1981), which calculates effect size as a difference in means standardized by the pooled standard deviation. Hedges’ g value and associated confidence intervals were calculated using a previously available Matlab toolbox (Hentschke & Stüttgen, 2011).

The Matlab code used for controlling hardware and implementing the task, and all the code used to analyze the data, produce plots, and compute statistics are available at Salas et al. (2025a) and Salas et al. (2025b). All the data collected and used to compute the results presented in this article are available at Rojas Libano (2025). All the image files used as stimuli for the task trials are available at Salas et al. (2025a). We report in the methods section how we determined our sample size, all data exclusions, manipulations, and all study measures. We also report effect sizes and confidence intervals. This study was not preregistered.

We monitored participants’ gaze through eye tracking to observe their instruction-following behavior while viewing the task images. Participants generally followed the instructions given (for some examples see Figure 3A and 3B). The gaze behavior was quantified as the dwelling time of the gaze within the blue circle presented in the images and expressed as a percentage of the total time (4 s) of image presentation. The overall dwelling time was high (Mdn = 81.97%, M = 69.76%, SD = 30.6%, n = 43), meaning participants tended to keep their gaze within the circle when required (Figure 3C, left-side plot). When we computed the difference in dwelling time between image types (neutral vs. unpleasant), we found that participants spent slightly less time within the circle for unpleasant than neutral images (Neutral: Mdn = 86.3%, M = 71.94%, SD = 30.95%; Unpleasant: Mdn = 81.87%, M = 70.31%, SD = 28.96%), with t(41)= 4.39, p < .01 (see Figure 3C, right-side plot). When we considered different experimental conditions (arousing focus vs. non-arousing focus), differences in dwelling time were smaller (Arousing: Mdn = 85.52%, M = 76.97%, SD = 23.43%; Non-arousing: Mdn = 82.43%, M = 76.92%, SD = 21.19%). In summary, these results show that participants engaged in the task and followed the instructions related to the visual attentional focus, irrespective of image or focus type.

Participants had 5 seconds to provide their responses of emotional ratings. Within this temporal frame, they were faster to produce emotional ratings in the arousing than in the non-arousing focus conditions. This difference was presented for intensity (non-arousing Mdn = 2.43 s; arousing Mdn = 2.2 s), t(54) = -1.8, p = .03, g = -0.22 [-0.6, 0.16], with 60% of the participants showing faster responses for the arousing focus. The same was observed for valence ratings (non-arousing: Mdn = 2.93 s; arousing: Mdn = 2.66 s), t(54) = -3.8, p < .01, g = -0.44, [-0.84, -0.09], where 67% of the participants displayed faster responses for the arousing focus (see Figure 4A).

To further characterize the behavioral responses, we then assessed for associations, computing Spearman’s rank correlation coefficient between the ratings’ response times and the value of the ratings, parsed by attentional focus condition. For the relation between intensity response times and intensity ratings, we obtained rho = -0.24, p = .07 (Non-Arousing focus), and rho = -0.51, p < .01 (Arousing focus). For the relation between valence response times and valence ratings, we obtained rho = 0.19, p = .16 (Non-Arousing focus), and rho = 0.32, p = .02 (Arousing focus). Thus, specifically for the arousing focus condition, we report a small albeit detectable relationship. Under this condition, the larger the intensity rating, the shorter the response time, and conversely, the larger the valence, the larger the response time.

Emotional ratings were expressed on a 1 to 9 scale. For emotional intensity, we found that the non-arousing focus condition elicited lower ratings than the arousing focus, evidencing a regulation of the emotional experience (non-arousing: M = 5.5, SD = 1.67; arousing: M = 6.5; SD = 1.99), with t(54) = 4.37, p < .01, g = 0.43 [0.06, 0.86]. Conversely, we found that for emotional valence, the non-arousing focus elicited higher ratings than the arousing focus, again evidencing regulation (non-arousing: M = 2.82, SD = 1.16; arousing: M = 2.34, SD = 1.23), with t(54) = -3.89, p < .01, g = -0.4 [-0.85, -0.04]. These results are shown in Figure 4B.

We then computed our AD estimates, constituted by the differences in rating between the non-arousing and arousing focus conditions. The rating difference was calculated as the participant’s mean rating for the non-arousing focus minus the mean rating for the arousing focus. Given that we had ratings for emotional intensity and valence, we computed two estimates, one for each emotional dimension. For emotional intensity, we found a distribution that was centered on negative values, Mdn (IQR) = -1(1.5), reflecting a decrease in emotional intensity in the presence of a non-arousing focus. The opposite was true for emotional valence, with a distribution of differences centered in positive values, Mdn (IQR) = 0.5(1), reflecting an increase in valence for the non-arousing focus.

Finally, we estimated the relationship between these two estimates to assess their congruence using the Pearson product-moment correlation coefficient. We found a correlation between intensity and valence rating differences, r² = 0.35; p < .01. This result implied that, on average, the effects of changing attentional focus (i.e., attentional deployment) were consistent across the two emotional dimensions: the larger the effect in intensity, the larger in valence. Therefore, our task produced valence-based and intensity-based estimates of AD.

We also studied the ratings and rating differences for the images in the focus-free condition. In these cases, participants did not have an instructed attentional focus on the image and were free to explore it visually. Regarding response times, we found no differences between non-arousing focus and focus-free conditions, both for intensity (focus-free Mdn = 2.4 s; arousing Mdn = 2.2 s), with t(54) = 1.42, p = .93, g = 0.13 [-0.24, 0.53], and valence (focus-free Mdn = 2.68 s; arousing Mdn = 2.66 s), with t(54) = 0.44, p = .67, g = 0.04 [-0.34, 0.42]. When examining ratings, we found very small differences between focus-free and arousing focus, for intensity (focus-free: M = 6.5, SD = 1.76; arousing: M = 6.5, SD = 1.99), t(54) = 1.69, p = .05, g = 0.12 [-0.26, 0.46], and also for valence (focus-free: M = 2.1, SD = 1.08; arousing: M = 2.34, SD = 1.23), t(54) = -2.72, p = .003, g = -0.2 [-0.57, 0.15]. These slim differences meant the focus-free attentional condition was similar to the arousing focus one regarding the emotional response it elicited.

The ANT allowed us to measure performance in three types of attention or attentional systems (Fan et al., 2002). Data from this sample showed a linear increase in the amount of time used by participants, reflecting an increase in the cognitive load of each attentional sub-task; from alerting to orientation to executive: Alerting Network (Mdn = 33 ms, M = 35.46 ms, SD = 22.88 ms), Orientation Network (Mdn = 46, M = 49.43 ms, SD = 20.1 ms), Executive Network (Mdn = 111 ms, M = 116.16 ms, SD = 30.82 ms). When error rates were analyzed, a sharp increase in the percentage of errors in the Executive Network task compared to the Alerting Network and Orienting Network was found: Alerting Network (Mdn = 0%, M = 0.66%, SD = 1.21%), Orientation Network (Mdn = 0%, M = 0.26%, SD = 0.78%), Executive Network (Mdn = 4%, M = 4.86%, SD = 5.27%). A key question of this study was the relationship between attentional ability -measured by performance on the ANT- and AD ability (AD estimate). Contrary to our hypotheses, no relationship was found between AD ability and performance on attentional tasks (See Figure 5B). We examined the Pearson product-moment correlation coefficient between the intensity-based AD estimate and the Alert Network, r = 0.31, p = .06, Orientation Network, r = 0.11, p = .53, and Executive Network, r = 0.14, p = .43, subdomains of the test. The same was true for the valence-based AD estimate and the Alert Network, r = -0.33, p = 0.05, Orientation Network, r = -0.16, p = .34, and Executive Network, r = -0.27, p = .11, although in this case all correlation values were negative. These results suggest that attentional capacities, as assessed by ANT, do not seem to explain the observed AD ability in our sample of participants.

In summary, we replicated a previously published AD task (Ferri et al., 2013, 2016), extending the sample to a different country, and studied in more detail some of its behavioral and neuropsychological correlates. This is a simple and brief experimental task that requires no more than 20 minutes to complete. In terms of emotional elicitation, our data showed that the task is effective, as emotional intensity ratings increase for unpleasant images compared to neutral ones, and emotional valence ratings are lower for unpleasant images (see Figure 2). In terms of behavioral correlates of the task, we observed for unpleasant images that rating response times were shorter (quicker) for ratings involving images with an arousing, as compared to non-arousing, attentional focus. We also found that, specifically for the arousing focus conditions, intensity ratings were inversely related to response times, and conversely, valence ratings were positively associated with response times. In addition, data from eye tracking showed that participants generally followed instructions, with no significant difference in gaze dwelling time between image types or focus conditions. Regarding estimates of AD, we operationalized it as a rating difference between attentional conditions (Non-Arousing and Arousing focus). We observed that the manipulation of attentional focus generated a decrease in emotional intensity and an increase in emotional valence, consistent with previous reports. As for the neuropsychological correlates of AD, no significant relationships were found between AD ability and performance on any of the ANT subtasks.

Our analysis of the AD task proposed an estimate of the AD capacity based on the difference in participants’ ratings between non-arousing and arousing focus conditions. We reasoned that the change in emotional rating, depending on the focus (i.e., attentional) condition, would constitute a behavioral readout of such an estimate. It is interesting to note that while our AD estimates clearly show that participants indeed regulate their emotions during the task and that they are consistent, the valence-based and intensity-based metrics show an important degree of inter-individual variability. We think that explaining this variability is an important task to better understand AD as a process. One way forward in this regard is to study the physiological correlates of behavioral readouts. For instance, several reports have shown a relationship between emotion regulation capacities and physiological markers such as heart rate variability (Min et al., 2023). Another interesting way to unpack this variability is considering personality traits or ER traits to explain how individuals respond to negative visual stimuli and down-regulate their emotions. Neuroticism, for example, has been described as the tendency presented by some individuals to experience negative emotions and distress (Ormel et al., 2013). Several studies have offered evidence suggesting that neuroticism is associated with specific patterns of gaze behavior toward negative emotional stimuli (e.g., Armstrong et al., 2010) and may influence selective attention (Richards et al., 2014). Unfortunately, no instruments are available to measure individuals' disposition to use AD as a strategy. This contrasts with other ER strategies, such as reappraisal and suppression, which have received more attention from the research community and have more established measurement tools (McRae & Gross, 2020). Future studies should develop tools that can contribute to measuring this ability.

We aimed to fill the gap in the current literature regarding a deeper characterization of AD. Concerning behavioral correlates, interesting data were found in terms of response time. Specifically, participants were faster to produce emotional ratings in the arousing focus condition compared to the non-arousing focus condition. This difference was evident for both intensity and valence ratings, with 60% and 67% of participants, respectively, showing faster responses for the arousing focus. The difference was larger for the valence rating, which is the first rating the participants had to report after watching IAPS images. Interestingly, we found associations between the ratings’ response times and the rating values, specifically for the arousing focus condition in the unpleasant images; higher intensity ratings were associated with shorter time responses, and conversely, lower valence ratings (more negative) were associated with faster response time. Furthermore, we observed that focusing attention on arousing portions of pictures, within unpleasant emotional stimuli, tends to prompt faster responses. This aligns well with previous studies, where faster recognition of emotional stimuli was associated with higher intensity and lower valence ratings, in the case of emotional faces (Sato & Yoshikawa, 2010), abstract emotional stimuli (Bartoszek & Cervone, 2022) and also of emotional events in experience sampling methodologies (Arndt et al., 2018). Overall, these data offer supporting evidence to the view that attentional focus is key to emotional experience and particularly to the process by which emotions are generated (Storbeck & Clore, 2007).

No robust associations were found when exploring the relationship between attentional capacities, evaluated through the ANT, and AD performance. Only performance on the Alerting Network exhibited a small and negative association with valence-based AD (r = -.33, p = .05) and a positive association with intensity-based AD (r = .31, p = 0.06). The Alerting Network has been closely linked to arousal (Petersen & Posner, 2012) and is described as responsible for achieving and maintaining an alert state. According to the review by Compton (2003), this network is crucial for processing emotional stimuli, as emotional states can enhance alertness and thus influence how attentional resources are allocated. Our data does not support the involvement of other attentional networks related to the selection of information, shifting attention (Orienting Network), or resolving conflicts amongst competing responses (Executive Network). One plausible explanation is that AD, as measured by our task, does not recruit these networks, since participants are simply instructed to fix their gaze on arousing and non-arousing areas of emotionally negative pictures. To explore further the relationship between AD and attentional abilities, future studies could replicate this task with other well-known neuropsychological tools that measure attention, such as the Paced Auditory Serial Addition Test (PASAT, Gronwall, 1977) or the Continuous Performance Test (CPT, Conners & Sitarenios, 2011). Another relevant area to explore is the assessment of AD in individuals with attentional disorders due to neurological damage, where AD impairment has been described (Salas et al., 2019).

In conclusion, our study contributes to the growing body of literature on AD specifically, and emotion-cognition interaction more generally, by providing a detailed behavioral characterization of a task that measures this emotion regulation strategy. It also offers novel data regarding the attentional correlates of AD, suggesting a potential role for the Alerting Network. While some of our findings challenge existing assumptions about arousing focus conditions in AD paradigms, they also open new avenues for methodological refinement in this field. Future research should continue to investigate the cognitive and neural bases of AD, employing diverse measures and paradigms to unravel the complex interplay between attention, emotional reactivity, and emotion regulation.