Our ability to form predictions about the behavior of objects outside our focus of attention and to recognize when those expectations have been violated is critical to our survival. One principle that greatly influences our beliefs about unattended stimuli is that of constancy, or the tendency to assume objects outside our attention have remained constant, and the next time we attend to them they will be unchanged. Although this phenomenon is familiar from research on inattentional blindness, it is currently unclear when constancy is assumed and what conditions are adequate to convince us that unattended stimuli have likely undergone a change while outside of our attentional spotlight. Using a simple change-detection task, we sought to show that unattended stimuli are strongly predisposed to be perceived as unchanging when presented on constant, unchanging backgrounds; however, when stimuli were presented with significant incidental visual activity, participants were no longer biased towards change rejection. We found that participants were far more likely to report that a change had occurred if target presentation was accompanied by salient, incidental visual activity. We take these results to indicate that when an object is not represented in working memory, we use environmental conditions to judge whether or not these items are likely to have undergone a change or remained constant.
Most of what we see we ignore. When we gaze upon a scene, our brain immediately goes about arranging the flood of visual information into coherent objects in space (Treisman & Gelade,
Although at any moment we are ignoring a large portion of our visual environment, attention is not a prerequisite for perception (Koch & Tsuchiya,
One known principle of inattentional vision is that of constancy (i.e., the tendency to perceive objects in a stable environment as constant and unchanging over time). Constancy is most familiar from the striking results of inattentional blindness studies, wherein changes to objects outside of our attention are extremely difficult to detect (Irwin,
Several questions remain open in regard to constancy, such as: What happens when objects seem likely to have undergone a change while outside of our attention? Or: are we biased to assume that unmemorized items remain constant in the same way that we assume unattended items do? One research paradigm that has approached this second question comes in the form of the flicker paradigm of inattentional blindness research (Rensink et al.,
Although the flicker paradigm offers a striking demonstration of change blindness, the paradigm’s lack of flexibility makes it less than ideal for further explorations of the phenomenon of change blindness. For one, because the flicker paradigm relies on ambiguous targets of change in a scene, complex naturalistic scenes with many potential targets of change are regularly used, thus presenting a set of possible change targets well above the capacity of working memory. Furthermore, because only one item (the target) can change in a flicker task, it is not possible to compare trials in which the majority of the scene remains constant with trials in which some part of the scene underwent some irrelevant changes. While it is conceivable to design a version of the flicker paradigm that addressed these issues, there already exist other tasks that could easily be adapted to answer these questions. One appealing alternative is the change-detection working-memory test designed by Luck and Vogel (
If constancy is strongly influenced by environmental factors, by embedding a simple change-detection task in complex visual environments that do not change, participants should become worse at detecting when a change has occurred to one of the memory items (i.e., what is observed in change-blindness experiments). Furthermore, by instructing participants explicitly that only the memory targets are relevant to the change-detection task, the backgrounds become available for manipulation. This opens the door to compare change-detection accuracy between trials in which the environment remained entirely stable to trials in which significant visual changes have occurred, a comparison that the flicker paradigm is incapable of making due to the ambiguous nature of the change target. It may be the case that constancy bias persists despite this added environmental activity, indicating that regardless of global events we default to assuming that objects outside our attention have remained constant. Alternatively, this salient environmental activity may represent compelling evidence that unattended items have undergone a change, resulting in the disappearance of constancy bias or even a reversal of the bias effect as background changes may be misattributed as stimuli changes, resulting in false memories and illusory changes to the memory stimuli.
To test these conditions, we chose to model our task on that used by Wheeler and Treisman in their 2002 paper. In their experiments, Wheeler and Treisman tested a variety of stimuli and change types to investigate working-memory capacity for feature conjunctions. Wheeler and Treisman’s paradigm closely copied that of Luck and Vogel except that in some experiments, the stimuli were a selection of shapes instead of colored boxes, and the types of changes that occurred in a trial were more numerous than in Luck and Vogel’s paradigm. This allowed them to observe differences in change detection for features versus binding changes. We hoped that by adopting their paradigm we would get a similarly rich view on the influences of environment of various types of change detection. Furthermore, by using shapes instead of color stimuli as memory targets, we were free to use colored backgrounds as our environmental manipulation. For our task participants needed to remember the identies and locations of four shape stimuli, which were presented on complex, multicolored backgrounds that could change at some point between learning and recall. By using four memory items, we presented participants with a challenging memory task, but one that was significantly easier than change detection in a complex, natural scene. In essence, our paradigm allows us to observe whether working memory is influenced by environmental factors, or whether working memory was entirely insensitive to irrelevant, incidental environmental activity.
We recruited 16 participants
Our experiment took place in a room with black walls and with the lights turned off. Visual stimuli were presented on a 54.61 cm, 1,920 × 1,080 resolution, 60 Hz LG Flatron W2261 monitor connected to a Macintosh MacPro version 10.10.5 running the Python library program PsychoPy (v1.85.6) (Peirce,
Our stimuli were based on those used by Wheeler and Treisman in their 2002 paper (Experiment 4). Memory targets were a set of nine possible shape stimuli. The shapes were chosen due to their familiarity and were: a circle, triangle, diamond, trapezoid, pentagon, arrow, hexagon, star, and cross. All shape stimuli were approximately 2 visual degrees in size and dark gray in color. The locations that the shapes could appear were pseudo random in that the test screen was divided into an invisible 3 × 4 grid of evenly spaced locations approximately 5 cm apart from one another and the screen edge. To make the shapes seem like they were not being placed on a grid, a random (x, y) value was added to each position when it was selected to hold a shape, making it seem that shapes appeared at random locations in space.
While many change-detection working-memory tasks use color stimuli, by using shape stimuli we were able to vary background colors without overlapping with the relevant memory features. The background colors used were all light in shade and highly luminous to ensure that the dark shapes were suitably visible regardless of what color was used. Colors were selected for their discriminability and familiarity. The colors used were red (10 candelas), blue (11.4 cd), green (16.1 cd), yellow (18.5 cd), pink (10.8 cd), purple (10.1 cd), orange (12.4 cd), and gray (16 cd). The background design we used was a two-colored vanishing point design where the screen was divided into four colored triangles converging at the centre of the screen (see Fig. The experimental flow for the two onset conditions
The experiment lasted approximately one and a half hours including five breaks, with a break occurring approximately every 10 min. The length of the break was left up to the participant, though they were made aware that longer breaks would lead to later finish times and as a result many participants took short breaks or even chose to periodically skip breaks.
The experiment began with both a verbal description of the experiment as well as a comprehensive written description on the computer that included visual demonstrations of the types of stimuli and changes to expect. The experiment consisted of 576 trials broken up into six blocks of 96 trials each, with a break between each block. (A breakdown of the number of trials for each condition is included in Table Accuracy (% correct) and trial count (per-participant) across no-change trials and the three change-trial types separated between the two onset conditions Trials (Per-Participant) Notice that position changes were consistently easy to detect and were uninfluenced by the various experimental conditionsLate-Onset Background Change Early-Onset Background Change Trials (Per-Participant) Same Background Different Background Same Background Different Background No-Change 78.47 71.81 87.15 86.28 288 Position 92.71 93.75 90.63 89.06 96 Binding 58.85 64.84 52.86 53.65 96 Feature 50.26 56.24 40.36 45.83 96 144 144 144 144 576 (Total)
The progression of a standard trial is given in Fig.
The test screen would be presented for 2 s, after which the test stimuli would disappear, and the subsequent answer screen would prompt participants to indicate whether they believed it was a change or no-change trial. Participants indicated their answers using the left and right arrow keys corresponding to “no-change” and “change,” respectively. As is normal in change-detection working-memory tasks (e.g., Luck & Vogel,
Additionally, on half of the trials, the background would remain constant throughout the trial, and in the other half it would change sometime between learning and test screens. Introducing background changes presented us with the interesting question of when in a trial should the background change. We treated this question as non-trivial as the onset timing of the background change can have a positive or negative effect on the change-detection tasks (Baker & Levin,
We therefore chose to use two onset timings for the background changes: either the background could change early in the trial (early-onset) or late in the trial (late-onset). In early-onset trials, the change would occur immediately after the learning stimuli disappeared, and thus participants would see the new background during the 2-s retention interval. In late-onset trials the change would occur immediately before the test stimuli were presented, so participants would be confronted with the test stimuli and a new background in the same moment. If participants were only sensitive to the identity of visual environments as the same or different, we did not expect the different onsets to exhibit different results. However, if participants were sensitive to when a change occurred in temporal proximity to stimuli learning or recall, then we expected this manipulation to illicit different patterns of results.
A last element of our experimental design was the inclusion of a salient white flash on all trials. We chose to include a white flash so that there would be some level of environmental activity in background constant trials. If we had not included this activity, there would have been an additional dimension of difference between our background conditions as in background-change trials there is necessarily always background activity in the form of the background change, but on background-same trials there need not be any environmental activity at all, creating a potential confound for any observed results. In order to ensure that in all trials there would be some level of background activity, we chose to add salient visual activity to all trials in the form of a 100-ms white flash. This flash would occur immediately before the background change, corresponding to the trial’s onset-timing condition. On trials in which the background remained the same, the only difference between early- and late-onset trials was when this salient flash occurred.
All results were analyzed using JASP, an open-source data analysis software similar to SPSS (JASP team, Experiment
Main effects were found for onset timing (
Several interactions were observed between onset and stimuli conditions (
To simplify our illustration of these interactions, we have broken down our 2 × 2 × 2 repeated-measures ANOVA into two separate 2 × 2 repeated-measures ANOVAs by separating the onset-timings conditions while keeping background and stimuli change conditions as factors. We chose to separate our analysis based on onset timing as they nicely illustrate very different participant behaviors when stimuli were or were not presented with incidental visual activity. These differences are shown in Fig.
For trials in which the background change and salient flash occurred early in the trial and were separated from the test stimuli by several seconds, a main effect of stimuli condition was observed (
A very different pattern of results was observed when either backgrounds changed or the salient white flash occurred immediately before the test stimuli were presented. While no main effects were found for stimuli or background conditions (
Additionally, participants were consistently less biased towards change rejection in late-onset trials than early-onset trials, even when the backgrounds remained constant. This can be observed by comparing accuracy on the two stimuli conditions between onset conditions on trials in which the backgrounds remained the same. Late-onset background same trials were significantly less accurate on no-change trials (
Additionally, the three different change types were compared to see how change detection accuracy varied depending on what type of change occurred in a trial. A repeat-measures ANOVA taking the three types of changes as factors showed a significant difference in accuracy depending on which type of change had occurred in a trial (
It is worth observing that 50% accuracy on a four-shape-change detection task suggests that participants were consistently able to remember about two of the four items. One useful measurement for working memory capacity is Pashler’s K measurement (Pashler,
Change-detection tasks have been a standard in the measurement of working memory, both with colored stimuli (Luck & Vogel) and with shape stimuli (Allen et al.
Fifteen new participants (eight females, age 19–28 years) were recruited in the exact same way as in Experiment
The stimuli and experimental design were exactly the same as in Experiment
Our observed results closely matched those observed in the background same condition of Experiment
The aim of this study was to test whether constancy bias could be observed and measured in an adaptation of a simple working-memory change-detection task (Luck & Vogel,
The assertion that participants were only sensitive to the amount of visual activity immediately before stimuli presentation is further supported by the observation that the constancy bias was reduced on background-same trials when the salient flash had a late onset. For trials that did not feature a background change, the only difference between early- and late-onset conditions was when during the trial the salient flash occurred. The fact that the presence of a flash immediately before stimuli presentation led to a weaker constancy bias indicates that participants were in tune to the amount of visual activity at stimuli presentation, even when it was only a short flash, and adjusted their bias to account for this activity. Furthermore, the only conditions that did not show any constancy bias were late-onset background-change trials, where the most visual activity accompanied the presentation of the test stimuli.
Clear differences in behavior were also observed between the three change types, where position changes were consistently very easy to detect and entirely insensitive to experimental conditions. This was in contrast to feature and binding change trials, which were both much harder to detect and varied in accuracy along the different experimental conditions. While feature-location binding was used to test binding behavior in Wheeler and Treisman’s
Interestingly, binding changes were consistently easier to detect than feature changes, a pattern of results that was not predicted based on Wheeler and Treisman’s findings in their 2002 paper from which our experimental paradigm borrowed heavily. Wheeler and Treisman suggested that object binding should be the most difficult type of change to detect (Wheeler & Treisman,
Following the completion of Experiment
Because our task was a two-alternatives forced-choice task, it is worth considering how much of our observed effect was due to guess-work caused by the stimulus set being larger than the natural working-memory capacity of our participants. With an average Pashler’s K score of 2.3 items from our experiments, participants consistently would have had to guess whether one or two of the items had changed on a trial. The resulting bias can thus be accounted for as differences in guess behavior, where environments influence perception in situations of uncertainty. We take this to be a plausible explanation of our results, though the answers provided still represent the perceptual beliefs of participants, and the biasing effect of background was invisible to the participants.
Working memory is a critical tool in change detection; if we have a clear memory of an item then we will not struggle to detect a change to that item (Luck & Vogel,
Despite the fact that inattentional visual processes govern the majority of the visual information that enters our eyes, researchers have tended to ignore inattentional visual processes, instead preferring to study the behavior of objects in our immediate attention. Unattended visual information is not presented in a raw, unstructured way, but instead is represented as coherent objects obeying rules and principles in the same way that attended visual objects do. We are able to comfortably ignore such a large part of our visual environment exactly because inattentional principles are so efficient at governing this ignored information, forming predictions about their behavior while outside of our focus of attention and informing us when this information may have become relevant. A critical principle of inattentional vision is that of constancy, or the tendency of objects to remain constant without accompanying visual stimulus. This principle can be easily flipped to say that when something happens to an object in space, it is often accompanied with salient visual activity, an easy flag for attentional relevance. Without this flag, though, the simple principle of constancy is the guiding heuristic, and we will remember and react to these unattended items as if they were constant and unchanging.
The data and experimental code are available upon request. None of the experiments were preregistered.
We chose to use a relatively small sample size as this experiment was heavily piloted as part of the first author’s Master’s Thesis. The effect size was thus approximately known, and the sample size of 16 was deemed appropriate. This study was not pre-registered.
During piloting, some participants reported lingering afterimages caused by viewing a strong computer monitor in a dark room. We wanted to be sure these afterimages were no longer present as they could be used to verify whether shapes had changed positions.
We chose not to analyze scores using d` or A` detection theory measures as bias was shown to vary greatly between conditions, a variation that sensitivity scores fails to capture (Allen et al.,
It may be observed that a K-score of 2.3 is quite low; normally K-scores on a four-item task should be in the neighborhood of 3–3.5. The lower sores observed in our experiment may have been due to the fact that we used shape stimuli rather than the more commonly used color stimuli (the reasons for this choice are given in the methods section) and shape stimuli have regularly been shown to be harder to remember than color stimuli (Allen et al.,
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.