The effect of image category and incidental arousal on boundary restriction

People ’ s memory for scenes has consequences, including for eyewitness testimony. Negative scenes may lead to a particular memory error, where narrowed scene boundaries lead people to recall being closer to a scene than they were. But boundary restriction — including attenuation of the opposite phenomenon boundary extension — has been difficult to replicate, perhaps because heightened arousal accompanying negative scenes, rather than negative valence itself, drives the effect. Indeed, in Green et al. (2019) arousal alone, conditioned to a particular neutral image category, increased boundary restriction for images in that category. But systematic differences between image categories may have driven these results, irrespective of arousal. Here, we clarify whether boundary restriction stems from the external arousal stimulus or image category differences. Presenting one image category (everyday-objects), half accompanied by arousal (Experiment 1), and presenting both neutral image categories (everyday-objects, nature), without arousal (Experiment 2), resulted in no difference in boundary judgement errors. These findings suggest that image features — including inherent valence, arousal, and complexity — are not sufficient to induce boundary restriction or reduce boundary extension for neutral images, perhaps explaining why boundary restriction is inconsistently demonstrated in the lab.


Introduction
Where someone's attention is directed when viewing a scene can lead visual details from that scene to be misremembered, missed, or distorted.These memory errors can affect important decisions, such as criminal conviction based on an eyewitness's memory report.For emotional scenes, people sometimes recall less of the periphery than they saw (e.g., Christianson & Loftus, 1987).This error is known as boundary restriction (e.g., Mathews & Mackintosh, 2004), an opposite phenomenon to the robust boundary extension error, whereby people tend to remember the periphery as being more expansive than it actually was (e.g., Intraub, & Richardson, 1989).One explanation for boundary restriction is that people's attention is captured under conditions of heightened arousal, which benefits memory for the arousing stimuli (e.g., a weapon; Mather, 2007).But boundary restriction has been inconsistently replicated (e.g., Candel et al., 2003).Some studies find a reduction in boundary extension-though not pure boundary restriction-under some conditions (e.g., Ménétrier et al., 2013).But most studies find that boundary extension is the more common memory error (see Hubbard et al., 2010 for review).These mixed findings may be due to past researchers relying on the incidental arousal of negative images to capture and hold attention (Anderson, 2005), rather than manipulating arousal itself.Consistent with this idea, we previously demonstrated reduced boundary extension for neutral images accompanied by an arousing stimulus (white noise; Green et al., 2019).But an alternative explanation for our finding is that systematic differences between the image categories we used-nature and everyday objects-drove the differences in boundary judgement errors, irrespective of the arousal stimulus.Therefore here we aimed to clarify whether the tendency toward boundary restriction rather than boundary extension occurs because of the arousal stimulus alone (Experiment 1) or differences between the two image categories we used (Experiment 2).These possibilities must be tested before concluding that arousal drives boundary restriction.
Boundary restriction errors, including reductions in the typical boundary extension pattern, have sometimes been shown for negatively valenced images (e.g., Mathews & Mackintosh, 2004;Ménétrier et al., 2013;Safer et al., 1998), using various memory tests (drawing and forced choice) and stimuli types (e.g., dynamic vs. static images, facial expressions vs. scenes).These findings fit with related research showing that people have poorer memory for other event details when their attention is focused on a weapon (Loftus, Loftus, & Messo, 1987), poorer memory for the association between an unrelated object presented in the periphery of a negative scene and the scene itself (Touryan et al., 2007), and enhanced memory for the specific visual details of negative object images, but not the general theme of those images (Kensinger et al., 2007).But other research on boundary errors specifically, has not demonstrated a reversal or reduction in the boundary extension pattern (e.g., Candel et al., 2003;Candel et al., 2004;Mathews, 1996), or has found increased boundary extension for emotional images (e.g., Patel et al., 2023).
One possibility for these mixed findings is that incidental high arousal from viewing negative imagesrather than the negative valence of those images per sedrives boundary restriction.We investigated this possibility in a previous study (Green et al., 2019).Specifically, we disentangled arousal from the incidental arousal associated with negative valence by presenting neutral images to participants either with or without an arousing noise.Across three experiments-in which we varied the type of view memory test (2AFC, distance ratings)-participants made more boundary restriction errors and fewer boundary extension errors when the noise was present vs. absent.This pattern of results may be evidence that arousal affects boundary judgements.
However, an alternative explanation is that the results are an artefact of the image categories we used and their associated complexity.Specifically, in our experiments, we counterbalanced the arousal stimulus (noise) and image category-creating two versions of the encoding phase (one where noise was paired with nature images, and one where it was paired with object images)-to produce a conditioned response where participants associated the noise with one category of images and not the other.But our two image categories-nature and object images-differed in complexity, which has recently been shown to affect boundary judgments.Bainbridge and Baker (2020) found that simple prosaic images (e.g., objects on patterned backgrounds) induced boundary extension, and complex, naturalistic images (e.g., nature scenes) induced boundary restriction.Thus, perhaps it was differences between image categories (i.e., nature vs. everyday objects), rather than the arousal stimulus, that influenced boundary judgement errors.Additionally, the relative complexity of nature and object images may lead these two image categories to differ on arousal, independent of any accompanying arousal stimulus, which in turn might also explain the boundary error differences we previously found.Indeed, participants in our previous work rated the nature images as more arousing than the object images, despite pilot testing to match the images on arousal (among other dimensions; Green et al., 2019).In fact, the combination-of arousal from the aversive noise in addition to the arousal arising from the nature images, and/or complexity of the nature images-may have had a greater influence on boundary judgements than either alone.
Taken together then, we do not know whether the arousal stimulus itself-and its conditioning to a particular category-caused the tendency toward boundary restriction we found in our previous work, or whether these findings were simply a product of differences in image complexity or naturally arising arousal between the image categories we used.Importantly though, if image category did have a (systematic) influence on boundary restriction or extension, we should have found a significant three-way interaction between noise, type of error, and noise-category-pairing, and we did not.Nevertheless, in the current research we aimed to directly test whether our previous results (in Green et al., 2019) were due to manipulated arousal (via aversive noise), or to the image categories we used and their associated visual complexity and inherent arousal.To investigate the role of the arousal stimulus in the absence of association to an independent category, we compared boundary restriction and extension errors for images from the same image category presented with and without noise (Experiment 1).To investigate the role of the image categories (Experiment 2), we compared boundary restriction and extension errors for nature images vs object images presented without noise (Experiment 2).This study was not preregistered, but the data are publicly available (https://osf.io/vdyu9/).Below, we report how we determined sample size, data exclusions, all manipulations, and all measures in the study.

Experiment 1
Recall that in Green et al. (2019), we conditioned the arousal stimulus to an image category such that one group of participants saw objects paired with noise (and nature scenes paired with no noise), and the other group saw nature images paired with noise (and objects paired with no noise).Our rationale for this decision came from the conditioning literature.Visual stimuli (such as images) can be easily conditioned to other sensory stimuli (such as sounds; see Jacobs & Shams, 2010 for a review).However, once a visual stimulus (e.g., an image) has been associated with a particular secondary non-visual stimulus (e.g., a noise), that association can transfer to other perceptually similar visual stimuli (i.e., perceptually similar images then become associated with the same noise; Davies et al., 1982;Stussi et al., 2018).Further, people are very good at categorizing visual images into semantic categories (e.g., identifying animals from objects); it happens seemingly instantaneously (Grill-Spector & Kanwisher, 2005).Thus, we expected that participants would automatically associate the arousal stimulus with the compositionally similar (i.e., outdoor scenes: nature; rather than indoor D.M. Green et al. scenes: objects) and perceptually similar (i.e., far distances: nature; rather than close distances: objects) images in the same category, such that seeing any image from that category, with or without noise, would elicit the same response.We wanted to condition the arousal stimulus to one specific image category and avoid the association of that stimulus with images in the other category-or "carry over" from one category to another.Thus, the image categories in Green et al. (2019) were compositionally and perceptually different (Dunsmoor et al., 2015; see also Stussi et al., 2018).We created two versions of the encoding phase, one where the noise accompanied nature images, and one where the noise accompanied everyday object images.
In the current experiment we replicated Green et al., (2019;Experiment 1a) without conditioning the arousal stimulus to an image category.We used images from only one category (everyday objects) but presented half of them with noise.If the arousal stimulus alone influences boundary judgements, then we would expect images presented with noise to lead to reduced boundary extension/ boundary restriction, compared to images presented without noise.However, we expected that with no conditioning, we would see more "carry over" of the arousal stimulus to the images presented without noise.In other words, we predicted that participants would associate the arousal stimulus with all of the images from that category rather than just the images that were actually paired with the arousal stimulus, and thus would make the same type of boundary errors for all the images regardless of the arousal stimulus.This finding would suggest that the differences in boundary restriction errors we found in Green et al. were either due to using two different image categories-the possibility we go on to test in Experiment 2 -or due specifically to the fact that the arousal stimulus was conditioned to one category of images.We also predicted that there would be no difference in image arousal ratings, and no difference in image pleasantness ratings.

Participants
According to Brysbaert (2019), d = 0.40 is the smallest effect size we should consider in psychology, because this is the effect size that starts to have meaningful, replicable effects (e.g., in a repeated measures study, an effect size of d = 0.40 corresponds to a 61 % chance of finding the expected difference for a random participant from each sample).For paired t-tests, Brysbaert (2019) recommends a minimum of 52 participants to detect d = 0.40.Thus for our paired samples design, we recruited 52 participants with normal or corrected-to-normal color vision from the Flinders University undergraduate research participation pool; 67 % were female, 33 % were male.Most participants were Caucasian (including "White"; 64.60 %).Participants ranged from 18 to 46 years (M = 23.19,SD = 6.12) and received course credit, or payment (AUD$10) for their time.

Design
We used a 2 (noise: noise, no noise) x 2 (encoded image: cropped, uncropped) within-subjects design.Cropped trials are close-up views of the image and uncropped trials are wide angle views of the image.We created two matched sets of object images and counterbalanced the set that was presented with noise so that all images were presented with noise an equal number of times across the sample.Our key dependent variable was the proportion of images on which participants made errors, either boundary restriction errors (selecting the cropped image) or boundary extension errors (selecting the uncropped image).

Images
For Experiment 1, we required 24 images from one semantic category.We used the pilot data discussed in Green et al. (2019) to identify everyday objects as the category with the most unique images.We excluded images that did not contain a central object on a wider background, or that we could not crop without cropping the central object in the image.Thus, we chose 24 everyday-object images with the best category fit ratings, and neutral valence.These images originally came from various sources including the International Affective Picture System (IAPS; Lang et al., 2008), the Nencki Affective Picture System (NAPS; Marchewka et al., 2014) the stock photography database "Shutterstock", or were created by the research team (see Supplementary Tables S3 and S4 for details).In each image, an object (e.g., lamp/clock/fan) appeared on a naturalistic background (e.g., table/wall/floor, respectively)-see Fig. 1 for examples.We created two sets of 12 object images that were matched1 on valence and arousal (see Table 1 for descriptive statistics).We cropped the background of each image to 75 % of its original size and then resized it, which resulted in two versions of each image (i.e., 75 % cropped and 100 % uncropped).

Anxiety measures
We measured participants' anxiety using the 20-item trait and state subscales of the State-Trait Anxiety Inventory (STAI-T, STAI-S; Spielberger, et al., 1970).For STAI-T, participants read a series of statements about how they feel generally (e.g., I am "calm, cool and collected"), and rated how often they felt this way on a 4-point Likert scale (see online supplementary material).Participants read a series of statements about how they felt at the present time, (e.g., "I feel calm") and rated how much they agreed with each statement on a 4-point Likert scale (from 1 = almost never to 4 = almost always).We measured state, rather than trait, anxiety because state anxiety is malleable over time (unlike trait anxiety).Overall, the STAI has good reliability and validity (α = 0.88; Grös et al., 2007).
given that we treated images as participants.
D.M. Green et al.

Manipulation of noise
White noise above 90 dB (db), in short bursts, briefly increases stress, fear and surprise similar to that of emotional arousal (Rhudy & Meagher, 2001).Half the participants saw image set 1 paired with noise and image set 2 paired with no noise.The other half saw image set 2 paired with noise, and image set 1 paired with no noise.We administered a 95-98 db range of white noise with headphones for the full two seconds that each image was presented.Images appeared in a random order such that noise and no noise trials were interleaved.

Procedure
The Flinders University's Social and Behavioural Research Ethics Committee approved this research.Participants were positioned approximately 40 cm from an Apple iMac desktop computer screen and the experiment was presented using Superlab software.

Encoding phase
Following information and consent, participants completed a demographic questionnaire, the STAI-T,2 and the STAI-S (pre-image exposure).Participants then viewed 24 images one at a time for two seconds each.Half of the images were cropped (75 %), the other half uncropped (100 %), participants viewed one version of each image.Presentation of cropped and uncropped versions of the images was counterbalanced and the order of image presentation was randomized.
Participants rated the pleasantness (1 = very unpleasant, 7 = very pleasant) and emotional arousal (1 = not at all emotionally arousing, 7 = highly emotionally arousing; from Porter et al., 2014) of each image, following its presentation, on a subsequent screen. 3ollowing the presentation of the images, participants completed the STAI-S-measuring their change in state arousal after viewing the images (post-image exposure)-and completed an unrelated filler task (a multiple-choice general knowledge quiz) for a total of 10 min.In line with Mathews and Mackintosh (2004), and other studies investigating boundary restriction (Candel et al., 2003, Green et al., 2019;Safer et al., 1998, Takarangi et al., 2016), we did not inform participants that we would test their memory of the images.

Test phase
Following the filler task, participants completed a memory test identical to the one used in previous studies (Green et al., 2019;

Note.
Valence (1 = very unpleasant to 7 = very pleasant) and arousal (1 = least arousing to 7 = most arousing).These ratings combine IAPS normed ratings (using a 1-9 scale; see Lang et al., 2008) for the IAPS-sourced images with our own ratings (where pilot participants rated from 1 = very pleasant to 5 = very unpleasant and 1 = relaxed/calm, 5 = aroused/excited; see Green et al., 2019) for images we created or sourced elsewhere.We have converted all ratings to a 1-7 scale-as used by participants in the current experiments-to allow easy comparison to those data.
Experiment 1a/1b: Takarangi et al., 2016).Participants saw two versions of each image they originally viewed.One version was identical to the one presented at encoding, the other was in the opposite format (i.e., 75 % cropped vs. 100 % uncropped).Participants selected which of the two images they originally viewed.Note that for each trial, participants could only either make one type of error (boundary restriction or extension) or be correct (see Fig. 2).When participants viewed an uncropped image at encoding (half the time), incorrectly selecting the cropped version at test was a boundary restriction error.When participants viewed a cropped image at encoding, incorrectly selecting the uncropped version at test was a boundary extension error.Participants were then debriefed.

Image rating analysis
First, we examined whether exposure to noise led participants to rate images as less pleasant and/or more emotionally arousing.Recall we expected thatwithout conditioning the arousal stimulus to an image typeparticipants would rate images presented with and without noise as similarly pleasant and arousing.However, contrary to our prediction (but consistent with Green et al., 2019), a paired samples t-test revealed that participants rated images as less pleasant when presented with noise (M = 3.01, SD = 0.87) compared to without noise (M = 3.53, SD = 0.99), t( 51 This result suggests that participants associated the unpleasantness of the noise with only those images accompanied by noise, rather than all images in the same category.However, it is important to note that we did not use a conditioning paradigm where participants first spend time associating the noise with a particular type of image category, prior to rating the images.Instead, participants associated the noise with an image category during encoding.Thus, perhaps the noise only affected pleasantness ratings with their paired image for the first few presentations of the noise, with a diminishing effect occurring for later images once the noise had been associated with the category.To examine this possibility, we split the data into three phases (the first, middle, and last 8 trials) and used a 3 x (time: time 1, time 2, time 3) x 2 (noise: noise, no noise) repeated measures ANOVA to investigate potential changes in pleasantness ratings over the three time points.There was no interaction, F(2, 102) = 3.19, p = .05,η p 2 = .06,and no main effect of time, F(2, 102) = 1.35, p = .26,η p 2 = .03,and there remained a main effect of noise, with images without noise being rated as more pleasant than images with noise, F(1, 51) = 22.07, p < .001,η p 2 = .30.Thus, we can conclude that there was no diminishing of the arousal stimulus's effect on pleasantness ratings.
We next compared the arousal ratings for images presented with noise versus without noise.As we predicted, participants rated images as similarly arousing when presented with noise (M = 2.33, SD = 1.16) compared to without noise (M = 2.16, SD = 0.96), t(51) = 3.50, p = .18,d = 0.19, 95 % CI [− 0.09, 0.46].Note that this finding is consistent with Green et al. (2019;Experiments 1a and 1b) and is most likely the result of asking participants-in both studies-to rate the perceived image arousal ("How emotionally arousing was that picture" 1 = not at all emotionally arousing, 7 = highly emotionally arousing), rather than reflect on their own state of emotional arousal (though see Kurdi et al., 2017).Indeed, when participants were asked-in Green et al. (Experiment 2) to reflect on their state ("How did you feel while viewing that image?"), they reported higher arousal overall and higher arousal after viewing either nature images and objects presented with noise compared to without noise.Thus, we feel confident that we induced the type of arousal necessary to potentially drive boundary restriction, even if the arousal measure we used here did not capture it.
We also wanted to find out whether-consistent with participants' pleasantness ratings-there was a similar change in arousal ratings during the encoding phase.We hypothesized that in the absence of conditioning, we would see a "carry over" of the effect of noise onto the no noise images, and therefore images presented without noise would also become more arousing over time.To investigate this possibility, we again split the data into three phases (the first, middle, and last 8 trials) and used a repeated measures ANOVA to investigate changes in arousal ratings over time.Our hypothesis would be supported by a significant main effect of time and no interaction with noise condition.We found a main effect of noise condition, F(1, 51) = 3.46, p = .045,η p 2 = .07.We also found a main effect of time, F(2, 102) = 5.43, p = .006,η p 2 = .10,and no interaction between time and noise condition on arousal ratings, F < 1.
These findings suggest in the absence of conditioning, the arousal effect of the noise "carried over" to the images without noise.An alternative explanation is that this pattern represents a general sensitization-and progressively amplified response-to the noise stimulus.
Our data suggest that participants perceived images with noise as more unpleasant, but not as more emotionally arousing, than images without noise.However, despite no change in image arousal, participants' state anxiety (on the STAI-S) increased from pre-(M = 36.30,SD = 9.71), to post-image exposure (M = 39.40,SD = 8.97, t(51) = 3.12, p = .003,d = 0.43, 95 % CI [0.15, 0.72].This finding supports the assertion that the noise exposure increased participants' own arousal, though the increase was small with participants' anxiety post-image exposure still close to the "somewhat" scale anchor.This raw change in state anxiety is slightly lower to that induced by a fearful video-clip (Tovilovic et al., 2009), but similar4 to that induced by exposure to a series of sensitive-content screens-that showed a blurred negative image and warned of graphic or violent content-embedded within positive and neutral images (Takarangi et al., 2023).Thus, the change in state anxiety from the noise stimulus is akin to other studies using negative images.

Analysis of boundary errors
Recall that our main aim was to clarify whether the reduced boundary extension and boundary restriction we observed in Green et al. (2019) was due to heightened arousal (induced by a noise; Experiment 1).Examining the mean proportion of participants' errors overall-such that larger values indicate worse performance-we found that participants were more likely to identify the correct image than make boundary judgement errors (M = 0.44, SD = 0.11, range = 0.25-0.75).Next, we addressed whether the arousal stimulus increased boundary restriction and attenuated boundary extension, in the absence of image-noise conditioning.Recall we hypothesized that-since we did not condition the noise to one semantic category of images-we would not see an effect of noise on boundary judgement errors.Thus, we investigated differences in error types.A paired samples t-test5 showed that-consistent with boundary extension research (e.g., Takarangi et al., 2016) Our results indicated that there was no significant difference in boundary extension or boundary restriction errors for images presented with or without noise.To confirm this null difference, we next ran a paired-samples Bayesian t-test with default Cauchy prior (Rouder et al., 2009) on boundary restriction errors on (uncropped) images presented with noise vs. no noise and found BF 10 = 0.15.According to the statistical interpretation that Wetzels et al. (2011) suggest, this Bayes factor indicates substantial evidence for no difference in boundary restriction errors.A second paired samples Bayesian t-test for boundary extension errors on (cropped) images presented with and without noise revealed BF 10 = 0.18, demonstrating substantial evidence that noise made no difference to boundary extension errors.Consistent with our hypothesis, participants made a similar number of boundary restriction and boundary extension errors regardless of whether the image was presented with or without noise.
Experiment 1 revealed three important findings.First, there was a difference in valence (pleasantness ratings) for images presented with and without the arousal stimulus, indicating that conditioning the arousal stimulus to a specific category was not necessary to manipulate the valence of a particular subset of images.This valence difference was consistent with Green et al. (2019), where the noise was paired with one image category only.Despite this difference, valence did not affect boundary judgement errors.Second, there was no difference in arousal ratings for images presented with and without the arousal stimulus, indicating that conditioning may have been necessary to manipulate image arousal.However, participants rated images presented later in the experiment as more arousing, compared to the images presented earlier in the experiment.Thus, it appears that the negative arousal effect of the noise may have carried over to the images presented without noise, or, perhaps participants became more sensitized to the noise.Thirdunlike Green Taken together, the results from Experiment 1 demonstrated that arousalin the absence of conditioning to an image categorywas not enough to induce boundary restriction, or even an attenuation of boundary extension.It is possible that the effect of the arousal stimulus carried over to all of the images, and/or that participants felt heightened arousal for all of the images at retrieval rather than only those images presented with noise.Thus conditioning an arousal stimulus to a particular category may therefore be necessary to observe the impact of the arousal stimulus.However, it remains possible that arousal does not influence boundary restriction and that the effects we observed in Green et al. (2019) were tied to image category, which we did not vary in Experiment 1. Teasing these two possibilities apart requires a second experiment that examines the role of image category independent of arousal.

Experiment 2
In Experiment 2, we investigated whether participants would make more boundary restriction errors for the nature images than the object images, in the absence of an arousal stimulus.If we were to find the same pattern of results as we did in Green et al. (2019) in the absence of the arousal stimulus, then we could conclude that differences in the image categories themselves-due to image complexity and/or inherent differences in arousal-are sufficient to reduce boundary extension and/or induce boundary restriction.In other words, because nature images are more complex and arousing than object images, we would expect nature images to lead to reduced boundary extension compared to object images, and more likely to lead to boundary restriction than object images.However, finding no such pattern-i.e., participants make the same type of errors for both image types-would support the assertion that the arousal stimulus, conditioned to a particular category, was necessary to change boundary error judgements.We replicated Green et al. (Experiment 1a) without an arousal stimulus.This procedure also allowed us to investigate whether the change in STAI state anxiety seen in Green et al. was attributable to the arousal stimulus.We predicted that in the absence of noise, participants' STAI ratings would not change from pre-to post-image exposure.

Participants
We again recruited 52 participants with normal or corrected-to-normal color vision from an undergraduate research participation pool at Flinders University.Most (71.20 %) were female, and Caucasian (including "White"; 61.50 %).Participants ranged from 18 to 53 years (M = 21.87,SD = 6.04).Participants received course credit, or payment (AUD$10) for their time.

Materials
For Experiment 2, we used the same 12 nature and 12 object images used in Green et al. (2019), which were matched as closely as possible-albeit not perfectly-on valence and arousal (Table 1).The content of the nature images included a tree in a wheat field, a leaf on the ground, a duck on a pond.Fig. 4 shows an example image from each category (object, nature).These images were resized and cropped in the same way as the Experiment 1 images, and presented in a random-interleaved-order at encoding.We used the same anxiety measure used in Experiment 1, outlined above.

Procedure
We used the same procedure as Experiment 1, without the arousal stimulus.

Analysis of boundary errors
Our aim was to determine whether boundary restriction errors-or attenuation of boundary extension errors-could be induced by differences in the complexity of, and/or inherent arousal differences in image category, regardless of noise.Overall, when we examined the mean proportion of errors, we found that participants were fairly inaccurate in identifying the correct image (M = 0.40, SD = 0.11, range = .17-.63), but slightly more accurate than in Experiment 1 (M = 0.44, SD = 0.11, range = 0.25-0.75;(t(1 0 2) = 1.99, p = .049,d = 0.39, 95 % CI [0.02, 0.79]).Next, we investigated differences in error types.A paired samples t-test showed that-consistent with boundary extension research, and our previous experiments-participants made more boundary extension (M = 0.71, SD = 0.22) than boundary restriction errors (M = 0.08, SD = .09),t(51) = 17.28, p < .001, d = 2.40, [1.86, 2.93].As shown in Fig. 5, there was no significant difference between object and nature images for boundary extension errors, t( 51 Our results indicated that there was no significant difference in boundary extension or boundary restriction errors for object versus nature images.To confirm these null differences, we next ran a paired-samples Bayesian t-test with default Cauchy prior (Rouder et al., 2009) on boundary restriction errors for object vs. nature images, which revealed BF 10 = 0.17.Similarly, a paired-samples Bayesian ttest on boundary extension errors for object vs. nature images revealed BF 10 = 0.17.Both these Bayes factors demonstrate that there was substantial evidence (according to Wetzels et al., 2011) of no difference in boundary restriction errors based on image category.In other words, participants were not more likely to make boundary restriction errors for nature images than for object images.This finding suggests that boundary judgments do not vary as a function of image category, via differences in the complexity or inherent arousal of nature vs. object images.
Experiment 2 produced three key findings.First, the absence of noise reduced overall boundary restriction errors, when compared to all of our previous experiments (including Green et al., 2019).Second, despite participants rating nature images as more arousing and more pleasant than object images, this difference in perceived image arousal and valence did not affect boundary errors.Third, we found no difference in the STAI between pre-and post-image exposure (in the absence of noise), indicating that the noise was likely responsible for this change in Experiment 1, and in Green et al.Overall, it appears that the arousal stimulus, conditioned to one image category, is necessary to reduce boundary extension errors.

General discussion
In the present experiments, we aimed to determine whether a reduction in boundary extension and/or boundary restriction can be induced by an arousal stimulus alone, not associated with a unique image category (Experiment 1), or due to differences resulting from using two image categories of differing complexity and/or inherent arousal (Experiment 2).
In Experiment 1 we found no difference in boundary judgements for object images presented with and without an arousing noise, even though images with noise had higher negative valence ratings than images presented without noise.In Experiment 2 we found no difference in boundary judgements for nature images compared to object images, despite the nature images being more complex and arousing.In other words, boundary restriction effects do not vary as a function of image category.
These results have several conceptual and methodological implications that are novel to the literature.First, taken together with the findings of Green et al. (2019), it seems that negative arousal alone is not sufficient to reduce boundary extension or induce boundary restriction for neutral images, and that it is necessary to condition the aversive stimulus (e.g., noise) to an image category to affect boundary judgements.This conditioning likely increases the arousal associated with the conditioned category and reduces the possibility that the arousal stimulus becomes associated with images that are not part of that category.
Second, in the absence of a separate noise stimulus, the valence, inherent arousal, complexity-and perhaps even other features  specific to the images themselves-may not affect boundary judgements.It remains possible that the combination of arousal from the aversive noise and inherent arousal or complexity from the images themselves has a greater influence on boundary judgements than either arousal source alone, especially when the images induce greater arousal levels than our neutral images here did.Nonetheless, our work highlights the importance of independently manipulating arousal, rather than, for example, relying on the incidental arousal of negatively valenced images to test the role of arousal in boundary judgements.Of course, it is also possible that any naturally arising arousal from the image categories themselves may be fundamentally different from the arousal arising from the noise stimulus we used.For example, it is likely that the noise stimulus was a higher intensity of arousal than the arousal arising from images, and perhaps this level of intensity is necessary to increase boundary restriction and/or reduce boundary extension.Indeed, despite no group differences within experiments, boundary restriction errors were generally lower-and boundary extension errors higher-in Experiment 2 compared to Experiment 1.We suspect these differences arose from the presence of noise leading to more arousal and higher state anxiety in Experiment 1 (vs.no noise in Experiment 2), resulting overall in reduced boundary extension in Experiment 1 (vs.2).To directly test this explanation, future research could manipulate arousal (absent/present) between participants instead of within participants.
Further, arousal arising from the noise is an example of aversive arousal, whereas arousal arising from image content (e.g., nature images) is likely an example of pleasant arousal.Aversive and pleasant arousal are similar in some ways: stress hormones are released for both types of arousal (e.g., Merali et al., 1998), and both types of arousal have been shown to improve memory accuracy (Bradley et al., 1992;Cahill & McGaugh, 1998).Thus, we might expect arousal to have the same impact on boundary judgments regardless of its valence.However, although both types of arousal lead to attentional capture (Sarter et al., 2001), people preferentially attend to aversive stimuli over pleasant stimuli (e.g., Eysenck, 1988; see Yiend, 2010 for a review).Given these differences, we would expect aversive arousal, through its capacity to capture attention, to result in more boundary restriction and less boundary extension while pleasant arousal would not affect boundary judgments via attentional capture.Alternatively, pleasant arousal might lead to increased boundary extension.Indeed, perhaps this is one reason why people made more extension errors in Experiment 2 (which included nature images) than in Experiment 1.According to Fredrickson's (1998Fredrickson's ( , 2001) ) broaden hypothesis-and the evidence for it (e.g., Fredrickson & Branigan, 2005)-positive states, including low to high arousal positive emotions, broaden people's attentional scope.Future research could explicitly test these possibilities by systematically manipulating the pleasantness of arousal (for example, by using music, which has previously been used to successfully alter arousal levels in both directions; e.g., Thompson, et al., 2001).Another interesting future research avenue is to investigate whether different types of aversive arousal influence boundary judgments, given discrete negative emotions can hold attention differently.For example, despite both eliciting aversive arousal, disgust typically holds attention longer than fear (e.g., Moeck et al., 2021).
Third, taken alongside Green et al. (2019), these experiments show that when suitably conditioned to a unique image category, inducing arousal only at encoding (not retrieval) can reduce boundary extension.In typical boundary restriction studies where participants view negative images, the source of arousal is the images themselves.Given participants view negative images at both encoding and retrieval, arousal effects at encoding and retrieval are confounded.By contrast, when pairing the arousal stimulus with neutral images (as in Experiment 1), arousal occurs only during encoding, not during retrieval.This methodological feature improves ecological validity-arousal is likely higher when participants experience a stressful event than when they recall it later-and means that any effect on boundary judgements can be attributed to arousal at encoding rather than retrieval, because arousal does not occur during retrieval.Indeed, recent research demonstrating that negative valence does affect the extent of boundary extensionbut only during the period at which a participant first feels the emotion-supports the idea that experiences at the time of encoding images affect boundary judgements (Ménétrier et al., 2020).However, it is possible that experiencing increased arousal at retrieval might also affect boundary restriction; thus, future studies should test arousal effects on boundary judgments at retrieval.Fourth, our data are consistent with the broader idea that conditions of heightened arousal (which often co-occurs with a negative stimulus) capture attention, which in turn restricts memory for the boundaries of an image.Recent research demonstrates a similar pattern of reduced boundary extension and/or evidence of boundary restriction when people's attention is focused to the central visual information-and/or their opportunity to extrapolate from the scene is constrained-by time pressure (Yildirim & Intraub, 2020), or executive abilities (e.g., ability to resist interference; Ménétrier et al., 2019).
Of course, our experiments have several important limitations.First, in a natural setting, neutral objects usually do not heighten arousal.Previous research shows that people preferentially associate aversive stimuli with congruent, naturalistic images (such as animals, nature scenes, or faces), rather than the objects on a neutral background used in our study.For example, participants have shown faster conditioning for face stimuli, compared with neutral images: specifically, conditioning occurs more rapidly, and extinction more slowly, for face stimuli than simple neutral images (a learning bias; Stussi et al., 2018).In fact, many studies have found that people show a learning bias for aversive stimuli; with angry faces ( Öhman & Dimberg, 1978), or potentially phobic objects (e.g., snakes; Öhman et al., 1976;Olsson et al., 2005;Mallan et al., 2013).Thus, the mismatch between the arousal stimulus and image type in Experiment 1 may have made conditioning more difficult, ultimately leading to no difference between object images presented with and without noise.Future research could tap into this possibility by pairing both image categories with noise vs. no-noise in a random order.Relatedly, overall the noise produced only relatively small changes in arousal.Perhaps larger changes in arousal are required to detect larger boundary judgement effects.Future research could test this possibility by increasing the intensity of the noise (e.g., duration) and/or the number of trials.
Second, using only two image categories might have systematically biased valence and arousal ratings because the images in one category were rated in the context of only one other category.Specifically, perhaps participants in the present experiment rated the nature images as more arousing than the object images (whereas pilot participants did not; see also Green et al., 2019; Experiments 1a and 1b) because they were making a direct comparison between the two image types.In addition, relative to the prosaic object images, D.M. Green et al.

Fig. 2 .
Fig. 2. Test paradigm for both Experiment 1 and Experiment 2 Note.Participants saw one version of each image (cropped or uncropped) at encoding, then viewed both versions at test.Participants chose the image they originally saw, giving them the opportunity to make one type of boundary judgement error, or get the correct answer.

Fig. 3 .
Fig. 3. Mean Proportion of Boundary Errors by Error Type and Noise for Experiment 1. Note.Data collapsed across noise-object group combination.Error bars represent 95% confidence interval around the mean (Masson & Loftus, 2003).

Fig. 5 .
Fig. 5. Mean Proportion of Boundary Errors by Error Type and Image Category for Experiment 2. Note.Error bars represent 95% confidence interval around the mean (Masson & Loftus, 2003).

Table 1
Pilot ratings of valence and arousal per image group for experiment 1 and 2.