When the periphery dominates perception: Confusion errors and spatial attention explain the inner-outer asymmetry of visual crowding

Crowding, the failure to identify a peripheral item in clutter, is an essential bottleneck in visual information processing. A hallmark characteristic of crowding is the inner-outer asymmetry in which the outer flanker (more eccentric) produces stronger interference than the inner one (closer to the fovea). We tested the contribution of the inner-outer asymmetry to the pattern of crowding errors in a typical radial crowding display in which both flankers are presented simultaneously on the horizontal meridian. In two experiments observers were asked to estimate the orientation of a Gabor target. Instead of the target, observers reported the outer flanker much more frequently than the inner one. When the target was the outer Gabor, crowding was reduced. These findings suggest that orientation crowding reflects source confusion (substitution) between the target and the outer flanker, but not vice versa. Furthermore, when there were four flankers, two on each side of the target, observers misreported the outer flanker adjacent to the target, not the outermost flanker. This finding postulates the role of spatial selection in crowding. Our findings reveal a counterintuitive phenomenon: in a radial arrangement of orientation crowding, within a region of selection, the outer item dominates appearance more than the inner one.


Introduction
Crowding refers to our inability to identify an object (primarily in the peripheral visual field) because it is presented with nearby objects [1][2][3][4][5][6] . Crowding is considered to be an impediment to reading 7 , face recognition 8 , eye saccade and hand movement, visual search 2 as well as deficits like macular degeneration 9 , amblyopia 10 and dyslexia 11 . Researches have characterized crowding and distinguished it from other spatial interferences, such as masking 5,12 , lateral interaction 13 and surround suppression 14 . Yet, the underlying processes of crowding are still unclear 2,4,15,16 . Researchers have associated crowding with a variety of factors: the reduction in spatial resolution in the periphery due to increased receptive field size of cells at early visual areas [17][18][19] , decreased cortical magnification 20 and reduced attentional resolution [21][22][23][24] .
A hallmark characteristic of crowding is the inner-outer asymmetry: an outer flanker (more peripheral) produces stronger interference than an inner one (closer to the fovea) 6,14,21,22 . The asymmetry effect has been demonstrated with different stimulus types and tasks, such as letter recognition 25 , face recognition 26 and Gabor orientation discrimination 14 . Nevertheless, the mechanisms underlying this effect remain a matter of debate.
Researchers have proposed various causes of inner-outer asymmetry. According to the cortical magnification account, the smaller cortical distance between the outer flanker and the target leads to integration errors between these two items 20,27 . Other accounts postulate that increased receptive field size in the periphery biases sampling rate toward the outer flanker 28 or leads to sparse selection in the visual periphery 29 . Finally, the attentional account attributes display layout and experimental task in biasing spatial attention outwards, leading to the selection of the outer flanker 21,22 .
All investigations of the inner-outer asymmetry, with the exception of one study, presented only one flanker at a time, either the inner or the outer one, and showed lower performance with the outer flanker than the inner one 14,[29][30][31][32][33] . The inner-outer asymmetry effect in this single flanker display is mediated by an outward bias of attention 21,22 ; however, it is unclear whether such attentional bias exists when both flankers are presented simultaneously. Thus, based on studies using a single flanker, it is unclear whether and how the inner-outer asymmetry contributes to a typical radial crowding display -i.e. when the target is flanked on both sides.
Contrary to studies using a single flanker display, the one study that investigated the inner-outer asymmetry using a presentation of both flankers simultaneously showed a reversed inner-outer asymmetry effect. Namely, in a letter identification task, in which three letters were presented on the horizontal meridian instead of the target (the middle one), observers often reported the inner letter more than the outer letter 34 . To explain this conflicting result, Strasburger 35 assumed a pooling process-under crowding, due to inappropriate integration field size in the periphery, observers simultaneously detect and pool excessive information of low-level features, including those that belong to the flankers 5,18,[36][37][38][39][40] . These pooling models suggest that crowding reflects some type of integration or averaging of target and flanker features. According to Strasburger, observers reported the inner flanker because it was less affected by the integration field and, therefore, more intact compared to the target and the outer flanker 35 .
In the crowding literature, pooling models compete with substitution models 16,[41][42][43] . According to substitution models, increased location uncertainty in the periphery or failure of attentional selection 44 , renders observers unable to spatially differentiate between the target and the flankers. These models predict a source confusion between the flankers and the target, which leads to reporting the flanker instead of the target 16,[41][42][43] .
Recent investigations of the two models used estimation reports in which observers estimated the target feature in a continuous space; for example, reporting the orientation of the target by adjusting the orientation of a probe 41,45,46 . By fitting probabilistic models to the distribution of the estimation errors, researchers could distinguish between averaging errors (reporting a combination of the target and the flanker values) and misreport errors (reporting the flanker value instead of the target). Using this method, studies have consistently shown that, under crowded conditions, observers often reported the orientation of a flanker instead of the target 16,41,46 . Interestingly, Yashar et al showed that when observers had to estimate an entire object -i.e. simultaneously estimate two feature-dimensions such as orientation and colourthey misreported orientations or colours in an independent manner within a trial, i.e., reporting the orientation of one item and the colour of another 46 . Thus, whereas the findings at the object level (integration of feature-dimensions) are consistent with pooling models, the findings at the feature level are consistent with substitution models.
However, if we consider the inner-outer asymmetry, pooling models may also explain misreport errors at the level of a feature. Following Strasburger's 35 explanation, observers reported the inner-flanker feature, which was less affected by crowding, and therefore, not integrated with the target. Thus, in a basic feature (e.g. orientation) estimation task, pooling models predict misreports of the flanker less involved in crowding, namely the inner flanker. Conversely, substitution models predict misreports of the flanker most involved in crowding-the outer flanker.
Here, we tested these predictions in a typical Gabor orientation crowding display with an orientation estimation task. We separately assessed the contribution of the inner and outer flankers to the pattern of crowding errors by fitting probabilistic models with a separate misreport component for each flanker. As in previous studies, under crowded conditions, observers often misreported a flanker as the target. Interestingly, we revealed that the misreport rate was much higher for the outer flanker than the inner one. Moreover, when the target was the outer flanker, crowding was substantially reduced. Thus, in a typical crowding display of a basic feature such as orientation, the inner-outer asymmetry reflects the dominance of the outer flanker over perception. By manipulating the region of selection between Experiment 1 and Experiment 2, our findings also point to the role of attentional selection in crowding.

Experiment 1
Observers performed an orientation estimation task of peripheral sinusoidal gratings (Gabor patches) in which the target (7º eccentricity) appeared alone (uncrowded conditions) or flanked (crowded conditions) by either two (two-flanker condition) or four (four-flanker condition) Gabors. Target and flankers were arranged on the horizontal meridian, either to the left or to the right of fixation. The centre-to-centre distance between two adjacent items was 1.5º. In crowded conditions, the target was always in the middle of the string of Gabors, such that in the two-flanker condition there was one inner flanker and one outer flanker, whereas in the four-flanker condition there were two inner flankers and two outer flankers with respect to the target (Fig. 1A & B).

Method
Observers. Thirteen undergraduate and graduate students (11 females: age range= 20 -31 years, M=25.23, SD=3.32) from The University of Haifa participated in this experiment for course credit or a 40 ILS per hour monetary payment (around $12). On the basis of a priori power analysis, using effect sizes from previous studies 46 , we estimated that a sample size of 12 observers was required to detect a crowding effect with 95% power, given a .05 significance criterion. We collected data from one more observer in anticipation of possible dropouts or equipment failure. All observers were naive to the purposes of the experiment. All observers reported having normal or corrected-to-normal visual acuity and normal colour vision, and none reported attention deficits or epilepsy. The observers signed a written informed consent form before the experiment. The University Committee on Activities Involving Human Subjects at The University of Haifa approved the experimental procedures (No. 373/18).
Apparatus. Stimuli were programmed in Matlab (The MathWorks, Inc., Natick, MA) with the Psychophysics Toolbox extensions 47 and presented on a gamma-corrected 21-in CRT monitor (SGI, with 1280 × 960 resolution and 85-Hz refresh rate) connected with an IMac. Eye movements were monitored with an Eyelink 1000 Plus (SR Research, Ottawa, ON, Canada), by using a chin rest at a 57-cm viewing distance. Observers reported their responses by using the mouse.
Stimuli and procedure. Figure 1A and B illustrate trial sequence and stimulus conditions in Experiment 1. All stimuli were presented on a grey background (56 cd/m 2 ). Each trial began with the presentation of a fixation display. The fixation display consisted of a fixation mark (a centred dot subtending 0.1°) along with two peripheral pairs of place holders, one pair in the right hemifield and one in the left hemifield (7º eccentricity and ±2º vertical offset from the horizontal meridian), which indicated the eccentricity of the upcoming target. Both fixation mark and place holders were white (112 cd/m 2 ).
Following observers' fixation for a random duration lasting between 500 and 800 ms, the fixation mark appeared for an interstimulus interval (ISI) of 100ms. The stimulus display appeared for 400ms. The stimulus display consisted of the fixation mark along with the target; a sinusoidal grating (Gabors) with a 2D Gaussian spatial envelope (standard deviation 0.55° and 85% contrast); and a spatial frequency of 1.5 cycles per degree (cpd). The target appeared on the horizontal meridian with 7º eccentricity, either in the right or left hemifield. The target could appear alone (uncrowded condition) or along with two flankers (two-flanker crowded condition) or four flankers (four-flanker crowded condition), which were located on the horizontal meridian on either side of the target. The centre-to-centre distance between two adjacent items was 1.5º. In the twoflanker condition, flankers' eccentricities were 5.5º and 8.5º for the first inner (1I) and first outer (1O) flankers respectively. In the four-flanker condition, flankers' eccentricities were 4º, 5.5º, 8.5º and 10º for the second inner (2I), first inner (1I), first outer (1O) and second outer (2O) flankers respectively. Thus, all flankers appeared within the window of crowding (< 0.5 of eccentricity). Target and flanker orientation were selected at random from a circular parameter space of 180 values evenly distributed between 1° and 180°, with the restriction that each of the flanker's orientation differed from that of the target by at least 15º.
Following the stimulus display, an additional fixation screen was presented for 200ms, which was followed by a response display. During the response display, observers were required to use a mouse cursor to adjust the probe Gabor's orientation to match the orientation of the target by selecting a position on the orientation wheel (marked by a black circle 0.08° thick with an inner radius of 3.8° at the centre of the screen). Following an observer's response, a blank inter trial interval (ITI) appeared for 400ms. In each trial, we monitored eye fixation using an eye tracker (see Apparatus). Trials in which fixation was broken (>1.5° from fixation mark) were terminated and rerun at the end of the block. Design. Each observer completed 15 blocks of 40 trials (600 trials in total) over a 60min session. There were 200 trials in each of the three crowding conditions (uncrowded condition, two-flanker condition and four-flanker condition). Crowding conditions were randomly mixed within each block. To reduce location uncertainty, observers were told that the pairs of dots have the same eccentricity as the upcoming target. The experiment began with two practice blocks of 10 trials. Observers were encouraged to take a short rest between blocks.

Models and analyses.
For each trial, we calculated the estimation error for orientation by subtracting the true value of the target from the estimation value. We analysed the error distributions by fitting probabilistic mixture models, which were developed from the standard model as well as the standard with misreport model 48 . We compared among four models: The standard mixture model (Equation 1) uses a von Mises (circular) distribution to describe the probability density of the pooling estimation of the target's orientation and a uniform component to reflect the guessing in estimation. The model has two free parameters ( , ): (1) Where is the value of the estimation error, is the proportion of trials in which participants are randomly guessing (guessing rate), and is the von Mises distribution with a standard deviation (variability) (the mean is set to zero), and is the total number of possible values for the target's feature. ( 2) where is the probability of reporting a flanker as the target, is the total number of flankers-target items (two or four in the present study), and is the error to the feature of the flanker. Notice that the von Mises distribution of the estimation errors here describes the distribution when the observer correctly estimated the target's feature, thus its mean is zero; whereas, for the distribution of estimating one flanker, the mean would be the feature distance of the corresponding flanker to that of the target. The variability of the distributions for each stimulus was assumed to be the same.
The two-misreport model (Equation 3) has a separate misreport component for each of the flankers in the two-flanker condition, one for the inner flanker and one for the outer flanker. The model has four free parameters ( , , , where is the probability of misreporting the inner flanker as the target, and is the probability of misreporting the outer flanker as the target. Where is the probability of misreporting the second inner flanker (innermost) as the target and is the probability of misreporting the second outer (outermost) flanker as the target.
We fit the models to the individual data using the MemToolBox 49 . We compared models' performance by comparing means Akaike information criterion with correction (AICc) to the individual data.
For each fitted model, we calculated target reporting rate (PT) by subtracting the accumulative guessing rate and misreport rate from 1, such that , , . , , for the standard mixture, misreport, two-misreport and four-misreport, respectively.

Probabilistic models
In uncrowded conditions, the standard mixture model described well the distribution of the errors (Fig. 2A). For each crowding condition we chose the best model among two relevant models by calculating the Akaike information criterion with correction (AICc) for each observer. Figure 2C shows   In order to test the contribution of each flanker to crowding errors, we compared the misreport rates of the different flankers. Figure 4 depicts the distributions of reports around the value of each flanker and the probability of reporting each presented item. Importantly, in the two-flanker condition, observers significantly misreported the outer flanker more than the inner one, t(12) = -2.59, p = 0.024, Cohen's d = 1.24. In the fourflanker condition, we conducted a one-way ANOVA on misreport rate with flanker positions (second inner, first inner, first outer and second outer) as within subject factor. Flanker position had a significant effect on misreport rate, F(3,36) = 5.74, p = 0.003, partial η 2 = 0.32, suggesting that observers confused the target with some flankers more than others. Planned comparisons revealed that the misreport rate of the first (  The results of Experiment 1 show that a reduction in overall precision of orientation estimation tasks under crowding is due to reporting a flanker instead of the target rather than an increase in variability ( ) over the representation of the target. This finding replicates that of Yashar et al (2019). Importantly, the results show that when the target is flanked radially from both sides, observers often misreport the orientation of the outer flanker that is adjacent to the target instead of the target. Observers misreported a flanker that was inner or not adjacent to the target in a much smaller proportion of the trials.

Experiment 2
Following the results of Experiment 1, in Experiment 2 we examined whether the dominance over perception of the adjacent flanker depends on the region of spatial selection. We used the same display as in Experiment 1, except that now the target was shifted one Gabor outward. That is, in the two-flanker condition, the target was the outermost Gabor, and, in the four-flanker condition, the target was flanked by the outermost Gabor (see Fig. 1C).
We predicted that if the outer item dominates perception, then a substantial reduction in crowding would occur in the two-flanker condition, in which the outer item is the target. However, in the four-flanker condition, in which the target is flanked by an outer flanker, we predicted that many trials would result in observers reporting the outer flanker instead of the target.

Method
Observers. Fourteen undergraduate and graduate students (10 females: age range 19 to 36 years old, M=25.71, SD=5.03), from The University of Haifa participated in this experiment for a course credit or a 40 ILS per hour monetary payment (around $12). All observers were naive to the purposes of the experiment. All observers reported having normal or corrected-to-normal visual acuity and normal colour vision, and none reported attention deficits or epilepsy. The observers signed a written informed consent form before the experiment. The experimental procedures were approved by The University Committee on Activities Involving Human Subjects at The University of Haifa (No. 373/18).

Apparatus.
The apparatus was the same as described in Experiment 1. Figure 1A and C illustrate a trial sequence and stimulus condition in Experiment 2. Stimuli and procedure were the same as in Experiment 1, expect for the location of the target. In this experiment, the target appeared on the horizontal meridian with 8.5º eccentricity, either in the right or left hemifield. In the twoflanker condition, the target was the outer item, whereas in the four-flanker condition, the target was located near the outer flanker. Here too, the centre-to-centre distance between two adjacent items was 1.5º. In the two-flanker condition, flanker eccentricities were 5.5º and 7º for the second inner (2I) and first inner (1I) flankers respectively. In the four-flanker condition, flanker eccentricities were 4º, 5.5º, 7º and 10º for the third inner (3I), second inner (2I), first inner (1I) and first outer (1O) flankers respectively.

Stimuli and procedure.
Design. The design was the same as described in Experiment 1.

Models and analyses.
The statistical analyses were the same as described in Experiment 1. Figure 5A illustrates the error distribution for each condition. Here again, we first analysed the bias of the errors by calculating error means for each subject in each condition. Error mean was close to zero in the uncrowded condition (M= -1.01, SD = 2.29), the two-flanker condition (M = -0.41, SD = 2.75) and the four-flanker condition (M= 0.34, SD = 4.03). As in Experiment 1, we calculated precision as the inverse of the variance of the errors for each observer in each condition (Fig. 5B). One-way ANOVA analyses on precision and crowding condition as within subjects factor revealed a significant main effect of crowding conditions on precision, F(2,26) = 44.30, p = 0.000, partial η 2 = 0.77, indicating higher precision in the uncrowded condition compared to the two-flanker condition, t(13) = 6.13, p = 0.000, Cohen's d = 1.45, and in the two-flanker condition compared to the four-flanker condition, t(13) = 4.19, p = 0.001, Cohen's d = 1.60.

Probabilistic models
In the uncrowded condition, the standard mixture model described well the distribution of the errors (Fig. 5A). For each flanker condition, we calculated the Akaike information criterion with correction (AICc) for each observer, in order to choose the best model among the two relevant models. Figure 5C presents mean AICc for the relevant models in each crowding condition. In the two-flanker condition, there was no bestfitting model, meaning that no significant differences were found between the standard misreport model and the standard mixture model, t(13) = 1.30, p = 0.216 and between the independent two-misreport model and the standard misreport model, t(13) = 1.25, p = 0.232. This result is explained by small crowding interference in this condition. However, in the four-flanker condition, the standard misreport model outperformed the standard mixture model, t(13) = 4.06, p = 0.001, Cohen's d = 0.69, and the independent four-misreport outperformed the standard misreport model, t(13) = 3.80, p = 0.002, Cohen's d = 0.91. In this condition, observers misreported specific flankers more than others.
Then, we analysed the fitted parameters of the best performing model in each condition. Figure 6 depicts the mean fitted variabilities ( ) and guess rates ( ) of the best fitted models in each condition. In order to examine the effect of crowding on performance, we conducted a one-way ANOVA with crowding condition (uncrowded vs. two-flanker vs. four-flanker) as a within subject factor and variability ( ), guess rate ( ) and target reporting rate (PT) as dependent variables. Significant differences were found between the crowding conditions for , F(2,26) = 9.86, p = 0.001, partial  We assessed the contribution of each flanker to crowding errors by comparing the misreport rates of the different flankers. Figure 7 depicts the report distribution around the value of each flanker and the probability of reporting each presented item. In the two-flanker condition, in which subjects were asked to report on the outer flanker, we found insignificant effect for flankers' positions, (t(13) = -0.35, p = 0.730). In the fourflanker condition, we conducted a one-way ANOVA on misreport rate with flanker position (third inner, second inner, first inner, and first outer) as within subject factor.

B D
The results of Experiment 2 show that the reduction of overall precision in orientation estimation task under crowding is due to both reporting a flanker instead of the target and increased variability ( ) over the representation of the target. Here again, we found that observers reported the adjacent flanker that was more eccentric rather than the target. However, when the target was the outer item, crowding was reduced and there was no significant difference in the misreport rate (PT) between the flankers.
Finally, we examined the effect of flanker position on crowding interference. To do so, we conducted an independent t-test for each crowded condition and compared the target reporting rate (PT) and standard deviation ( ) between Experiment 1 and Experiment 2. PT analyses revealed that in the uncrowded condition, the strength of crowding interference did not differ significantly, t(25) = -0.39, p = 0.701, meaning that observers were unaffected by item eccentricity. However, significant differences were found between the two experiments in the two crowded conditions. In the two-flanker condition, when observers reported the outer item (Experiment 2) crowding was smaller than when observes reported the item in the middle (Experiment 1), t(25) = -2.69, p = 0.013, Cohen's d = 1.04. In contrast, in the four-flanker condition, when the outer target was flanked by an outer flanker (Experiment 2) crowding was stronger than when the central target was flanked by two outer flankers (Experiment 1), t(25) = 2.40, p = 0.024, Cohen's d = 0.92. No differences were observed for between Experiment 1 and Experiment 2, with t-values bigger than 0.30 in each crowding condition.

Discussion
Using an orientation estimation task and probabilistic models designed to assess the contribution of each flanker to crowding errors, we showed that the hallmark innerouter asymmetry in crowding is due to confusion errors and contingent on spatial region of selection. First, as in previous studies 16,41,46 , our results showed that orientation crowding errors reflect reporting a flanker instead of the target (misreport errors). Second, in contrast to Strasburger and Malania 34 , we found that in a radial crowding display (Experiment 1), in which a target is flanked on both sides, observers misreported the outer flanker much more than the inner one. Third, when two flankers appeared on each side of the target (four-flanker condition), observers reported the outer flanker adjacent to the target, but not the most eccentric flanker. Finally, when the target was the outer item (Experiment 2, two-flanker condition) crowding errors were substantially reduced. These findings suggest that in a radial arrangement of orientation crowding, the outer flanker within a region of selection becomes dominant over perception.
The asymmetry effect has been typically examined under asymmetrical display of crowding in which only one flanker was presented at a time 6,14,30,32 . However, since such a display layout may bias attention outward 21 , it was not clear whether and how the inner-outer asymmetry emerges-particularly in a typical symmetrical crowding display of a basic feature, such as orientation. Here, we reduced attentional bias by using a typical radial crowding display in which both flankers are presented, one on each side of the target.
Among the various theories of crowding, the perceptual framework of pooling versus substitution processes dominates the literature. Recent investigations showed misreport errors in which observers reported the orientation of the flanker instead of the target 16,41,46 . Under the perceptual framework of the asymmetry effect, these findings, although intuitively supporting the substitution account, might result from pooling processes. Accordingly, due to pooling of the target with the outer item, the dominant flanker was assumed to be the inner one, which was not integrated with the target. In contrast, according to the substitution account, the dominant flanker was assumed to be the flanker that was mostly involved in crowding.
Supporting evidence for the pooling account was reported by Strasburger and Malania (2013) 34 . Similar to our two-flanker condition, in their study the asymmetry effect was investigated in a typical symmetrical crowding display in which the target was flanked by two items simultaneously. Surprisingly, using a letter identification task, they observed a reverse inner-outer asymmetry. That is, observers confused the target with the inner letter, which was less impaired by pooling processes compared to the outer letter 35 . In contrast to Strasburger and Malania (2013) 34 , our findings support the substitution model. We found that observers were unaffected by the flanker that was less involved in crowding, the inner flanker. Instead, they mistakenly reported the orientation of the outer adjacent flanker. This inconsistency might be explained by variations in processing levels due to the variation in stimulus type: Strasburger and Malania 34 examined asymmetry for letter crowding, which demands high-level representations of objects, whereas we assessed orientation asymmetry, which demands processing of low-level features. This pattern of results corresponds with Yashar et al 46 findings, indicating pooling processes at the object level and substitution processes at the feature level.
Our findings are consistent with crowding models, which suggest that crowding is due to sampling over a larger receptive field size in the visual periphery 28,29,46 . In particular, according to the receptive field size account 28 , the dominant flanker is assumed to be the most eccentric one since the size of receptive fields grow with eccentricity. However, rather than simply reporting on the outer flanker, which was within a typical critical spacing of crowding (<0.45 of eccentricity), observers reported the adjacent outer flanker. This finding indicates that bottom-up, low-level processes alone cannot explain crowding and postulates on the role of top-down attentional selection 21,22 .
Our results also contradict the cortical magnification account. According to this view, crowding asymmetry results from cortical mapping in which the outer flanker is closer to the target than the inner one 5,27 . This explanation predicts that within the same display (i.e. same cortical distances), crowding interference will be the same regardless of whether observers are reporting on the middle or the outer item. Thus, the reduction of crowding interference in the two-flanker condition of Experiment 2 compared to Experiment 1 suggests that cortical magnification alone is insufficient to explain the inner-outer asymmetry of crowding 28 .
In this study we investigated the inner-outer asymmetry along the horizontal meridian by using a symmetrical display. Although previous studies found the asymmetry effect along the horizontal, but not along the vertical meridian 22 , it is still unclear whether and how this effect is reflected in a typical symmetrical crowding display across the visual field. Furthermore, our results are limited to orientation errors involved in crowding asymmetry. Since recent studies found dissociation of crowding errors across dimensions, such as orientation, colour and SF 46 , and between colour and motion 50 , further work is needed to determine whether the current findings apply to other feature dimensions. Moreover, the use of Gabor patches also limited our ability to distinguish between object level and feature level. Due to previous studies that pointed to different processes involved at these levels 34,46,51 , further work should use a different type of stimulus, which would allow a clear distinction between those levels.

Conclusions
Our findings reveal that in a typical radial crowding display, observers confuse the target with the nearby outer flanker, but not vice versa. This confusion occurs regardless of the number of flankers and reported item position. Our findings suggest that crowding cannot be explained by a simple averaging process and demonstrate a counterintuitive phenomenon: within a region of selection, the more eccentric item dominates appearance.