Thirty York University undergraduate students (17 to 48-year-olds, 9 males) participated in the experiment for course credit. Participants had normal or corrected-to-normal visual acuity and had normal color vision as assessed by Ishihara color plates (Ishihara, 2006). Written informed consent was obtained prior to participation in accordance with the Declaration of Helsinki. York University's Human Participants Review Committee approved all experimental procedures.

The saccade target was a red (x = 0.63, y = 0.33, L = 12.07 cd/m²), green (x = 0.29, y = 0.59, L = 12.00 cd/m²), yellow (x = 0.40, y = 0.50, L = 12.09 cd/m²), or blue (x = 0.15, y = 0.07, L = 12.04 cd/m²) square that subtended 2.0° × 2.0° (see Figure 1). Placeholders were gray (x = 0.28, y = 0.30, L = 11.20 cd/m²) squares that subtended 2.0° × 2.0°. Stimuli were embedded in a black (x = 0.26, y = 0.24, L = 0.22 cd/m²) background displayed on 21-inch CRT monitor (60 Hz, 1024 × 768). Color and luminance were calibrated using a spectrophotometer (PR-655, Photo Research, Syracuse, NY). Participants observed stimuli in a dimly lit room from a distance of 57 cm with a headrest used to stabilizing their head position.

Stimulus presentation was controlled on a computer running Presentation software (Neurobehavioral Systems, Berkeley, CA) and a serial response box (Cedrus, San Pandro, CA). Eye position was sampled using an infrared eye tracker (500 Hz, Eyelink II, SR Research, Ontario, Canada). The camera was calibrated using a nine-point grid. Calibration was conducted at the beginning of each experimental run. Drift-corrections were conducted prior to each block and as needed.

Trials were initiated by a button press on the serial response box and then maintaining fixation (1.89° square window) on a white central fixation cross (0.4° × 0.4°) for 200 ms (see Figure 2). Eight placeholders centered to the circumference of an imagery circle (radius = 7.5°) at positions along the cardinal and oblique axes were displayed for 200 ms to indicate all potential target positions (spatial phase). A randomly sampled subset of four placeholders was then replaced with red, green, yellow, and blue squares and displayed for 200 ms (color display). The colored squares then reverted back to gray placeholders and were displayed for 200 ms (mask phase). The placeholders and fixation cross were removed from the display creating a blank, gray background displayed for 1000 ms (delay phase). A target color was randomly selected from the four colors and a target cue was displayed at central fixation for 200 ms. The offset of the central target cue served as the go-signal to execute a saccade to the remembered location of the target color from the previous color display. Participants were instructed to maintain fixation throughout the trial until presented with the go-signal. The trial was complete when a saccade was made to the location of the target colored square (correct) or a non-target colored square (incorrect) from the color display, or until 500 ms had elapsed since the go-signal (time out). An error tone signified a time-out or failure to maintain fixation. Trials with a timeout or fixation break were reinserted randomly back into the block. There were eight trials for each of the four target colors on every block. Participants completed 10 blocks of 32 trials for a total of 320 trials.

Customized MATLAB (MathWorks, Natick, MA) algorithms were used to detect, visualize, filter, and analyze saccades. Trials that contained blinks, corrective saccades, saccade amplitudes < 1°, endpoint deviations > 6° from the center of the selected stimulus, or fixation drifts > 1° during the presaccadic latency period were excluded from further analysis.

To examine unresolved competition between stimuli during saccade planning, we analyzed two saccade curvature metrics: (1) sum curvature, the sum of all orthogonal deviations from a straight line between the start and endpoint of saccade trajectories sampled by the eye tracker; and (2) max curvature, the maximum orthogonal deviation from a straight line between the start and endpoint of saccade trajectories. To compute these metrics, we first translated the start point of each saccade to the Cartesian origin and then trigonometrically rotated each saccade so that the endpoint was aligned to the positive y-axis. Therefore, the saccade curvature metrics were signed so that a negative sign indicates deviation (i.e., curvature) in the counter-clockwise (CCW) direction, and conversely, a positive sign indicates deviation in the clockwise (CW) direction. This method was ideal for examining saccade curvature on circular displays with randomized target and distractor locations. Saccades with a sum curvature greater than three standard deviations from the mean were not analyzed for any curvature analyses.

For all saccade curvature analyses, we only analyzed a subset of correct trials with the following characterizations: (1) a single distractor occupied the stimulus positions 45° or 90° CCW to the target and there were no distractors occupying either of the stimulus positions 45° or 90° CW to the target (subsequently referred to as the CCW condition); (2) a single distractor occupied the stimulus positions 45° or 90° CW to the target and there were no distractors occupying either of the stimulus positions 45° or 90° CCW to the target (subsequently referred to as the CW condition); or (3) no distractors occupied any of the stimulus positions within 90° of the target in either the CCW or CW direction (subsequently referred to as baseline). We refer to trials in the CCW and CW conditions as isolated distractor trials. Unpredictable distractors located 135° from a target position reportedly elicit only a marginal effect on saccade curvature (McSorley et al., 2009) and were therefore not considered in the current analyses. First, we examined whether baseline curvature was significantly biased in a particular direction by comparing baseline curvature to zero. Next, to ensure that the isolated distractor 45° or 90° away from the target position elicited a systematic effect on saccade curvature, we compared saccade curvatures in the CCW and CW conditions relative to baseline (i.e., using unbiased curvatures) to determine if saccades curved away from the isolated distractors as is commonly observed in the remote distractor paradigm (Doyle and Walker, 2001; Tipper et al., 2001). We used unbiased curvature since a significant bias may uniformly shift curvatures in either the CCW or CW direction such that they do not actually curve in opposite directions (where the sign of the metrics indicates the direction) even though the distractors are located in opposite directions.

Next, to investigate whether potential saccade goals were encoded in CIE (x,y) perceptual color space, we analyzed whether unresolved competition between potential saccade goals (as indexed by saccade curvature) was modulated by the color space distances between the target and isolated distractors, referred to as target-distractor color space distance. This analysis was performed in several steps: First, we computed an exhaustive list of pairwise Euclidian distances between each color stimulus in CIE (x,y) color space (see Table 1) using the Pythagorean equation and categorized the color space distance between the target and the isolated distractor on every trial. Second, we computed mean saccade curvature as a function of target-distractor color space distance separately for the CCW and CW conditions. We then averaged together the absolute, unsigned saccade curvature from the CCW and CW conditions for each target-distractor color space distance, as we were ultimately interested in examining the average magnitude of saccade curvature (i.e., inherent competitiveness of distractors) as a function of target-distractor color space distance regardless of the relative position of the isolated distractor to the target and the direction of the curvature. Furthermore, averaging the CCW and CW conditions would compensate for any curvature bias and thus generalize the current results such that they are not limited to any particular spatial arrangement of a color distractor relative to the target. Third, we examined absolute mean saccade curvature as a function of target-distractor color distance with regression analyses. To determine whether the inherent competitiveness of isolated distractors was modulated by surround suppression in color space, we fit the data with a Mexican hat wavelet model (a close approximation to the DoG function):

where α is the function ceiling, σ is the width of the function, κ was included to scale the depth of the inhibitory annulus, y_n is the baseline of the function and was set to the mean saccade curvature (y_i) observed at the furthest target-distractor color space distance (n), and minima are located at x= ±κ+2σ. Next, we fit the data with a generic quadratic polynomial model:

Reducing the number of fitted free parameters in the Mexican hat model ensured an equal number of fitted parameters between models. Functions were fit using a custom implementation of the maximum likelihood estimation method by maximizing the following Gaussian log-likelihood function:

where θ is the vector of fitted parameters in either the Mexican hat wavelet [θ = (α, σ, κ)′] or quadratic [θ = (a, b, c)′] models, n is the number of target-distractor color space distances being fitted to the model, ϕ is the Gaussian normal probability density function, ŷ_i is the saccade curvature value predicted by the model with parameters θ for the ith target-distractor color space distance, y_i is the average saccade curvature value observed for the ith target-distractor color space distance, and σ_i is the standard deviation for the ith target-distractor color space distance. The goodness-of-fits were evaluated using an F-test. If both models provided a significant fit to the data, the best fitting model was determined by performing an F-test on the ratio of the sum of squared residuals from each fitted model with n−k degrees of freedom, where k is the number of fitted parameters (see Kehoe and Fallah, 2017). Furthermore, to provide further evidence of a functional relationship between saccade curvature and target-distractor color space distance, we analyzed the data with a quadratic planned contrast.

To examine whether there was an overall selection bias on color categories, we analyzed potential categorical color differences between overall selection proportion, which was the proportion of total trials (i.e., correct and incorrect) in which a particular color category was selected as a saccade target. This required us to include the error trials back into the data and then remove them for subsequent analyses.

We then analyzed whether the inherent competitiveness of distractors was hierarchically modulated by the color categories of the isolated distractors. As with target-distractor color space distance, absolute, unsigned saccade curvature was averaged between the CCW and CW conditions for each isolated distractor color. All subsequent analyses were performed on all correct trials of all trial types (i.e., isolated distractors and no isolated distractors).

Next, to examine whether the strength and speed of saccade target color encoding varied as a function of the target color category, we examined three metrics related to the strength and speed of target encoding: (1) Proportion of errors, which was the number of trials for a particular target color in which the saccade was executed to an incorrect location divided by the total number of trials with that target color. (2) Saccadic reaction time (SRT), which was the time between the go-signal and saccade initiation for correct saccades. (3) Saccadic precision, which was the area of a 95% confidence data ellipse fit to the 2-dimensional displacement between the target center and the saccade endpoints computed for each color and each participant for correct saccades (see Chen et al., 2011). Saccade precision was computed before and after translating the start-point of every saccade back to fixation. Our results demonstrated that this translation did not have a systematic effect on the results and thus we report the results for translated saccades.

Mean differences were analyzed with paired-samples t-tests or repeated-measures ANOVAs and Bonferroni post-hocs. If a Mauchley's test provided insufficient evidence for sphericity, the degrees of freedom of the ANOVA were adjusted using the Greenhouse-Geisser (ε ≤ 0.7) or Huynh-Feldt (ε > 0.7) adjustment. To examine whether there were hierarchical differences between color categories (red > green > yellow > blue) consistent with the color hierarchy (Tchernikov and Fallah, 2010), we examined planned polynomial contrasts of the color categories. As we performed parametric analyses, a Shapiro-Wilk test was used to determine whether the data were normally distributed. If this analysis provided insufficient evidence of normally distributed data (p < 0.05), the data was transformed to achieve normality using a log(x +1) transformation. If this did not successfully normalize the data, we alternatively utilized a x3 transformation. These transformations were chosen because they are ideal for normalizing positively skewed data such as ours.

Lastly, for each categorical analysis discussed above, we performed a complimentary functional analysis to rule out perceptual discriminability as a possible explanation for any potential differences between color categories. First, the discriminability of each color was calculated by averaging the distance between one color category and all remaining color categories in CIE (x,y) color space (see Average Distance column in Table 1). Next, a linear regression analysis using ordinary least squares was performed to assess whether there was a functional relationship between the discriminability of each color category and the means utilized in the aforementioned categorical analyses.

Sum curvature was significantly biased in the CCW direction overall as demonstrated by comparing baseline (M = −0.80°, SE = 0.34°) to zero, t₍₂₉₎ = 2.32, p = 0.027, d = 0.42. Critically, however, unbiased sum curvature was significantly different between the CCW (M = 0.74°, SE = 0.38°) and CW (M = −0.75°, SE = 0.43°) isolated distractor conditions, t₍₂₉₎ = 2.87, p = 0.008, d = 0.52 (see Figure 3). Similarly, baseline max curvature (M = −0.05°, SE = 0.03°) was also biased in the CCW direction, t₍₂₉₎ = 2.12, p = 0.042, d = 0.39; and unbiased max curvature was also significantly different between the CCW (M = 0.03°, SE = 0.03°) and CW (M = −0.07°, SE = 0.03°) isolated distractor conditions, t₍₂₉₎ = 2.65, p = 0.013, d = 0.48. Saccade curvature differences between the CCW and CW isolated distractor conditions suggested that isolated distractors systematically modulated saccade curvature and thus validated examining saccade curvature on isolated distractors trials to investigate target selection competition between color stimuli.

The Mexican hat model provided a marginal fit of sum curvature as a function of target-distractor color space distance, F_{(2, 3)} = 6.25, p = 0.085, R² = 0.75 (see Figure 4A); as did the quadratic model, F_{(2, 3)} = 7.44, p = 0.069, R² = 0.77 (see Figure 4B). Critically, the minima of the fitted Mexican hat (x_min = 0.42) and quadratic (x_min = 0.41) models were nearly identical. To provide additional evidence for a functional relationship between sum curvature and target-distractor color space distance, sum curvature was transformed to achieve normality [log(x + 1)] and was analyzed with a quadratic planned contrast, which demonstrated a significant quadratic contrast, F_{(1, 29)} = 5.41, p = 0.027, ηp2 = 0.16. In contrast, the Mexican hat model did not provide a significant fit of max curvature as a function of target-distractor color space distance, F_{(2, 3)} = 5.31, p = 0.103, R² = 0.76 (see Figure 4C). However, the quadratic model provided a marginal fit to the data, F_{(2, 3)} = 7.44, p = 0.069, R² = 0.81 (see Figure 4D). As with sum curvature, the minima of the fitted Mexican hat (x_min = 0.38) and quadratic (x_min = 0.39) models were nearly identical for the max curvature data. Again, to provide additional evidence for a functional relationship between max curvature and target-distractor color space distance, max curvature was transformed to achieve normality (x3) and was analyzed with a quadratic planned contrast, which demonstrated a significant quadratic contrast, F_{(1, 29)} = 4.76, p = 0.037, ηp2 = 0.14.

In our first analysis, we examined whether the color representations utilized by the oculomotor system for saccade planning were encoded in perceptual color space by analyzing trials on which a correct saccade was made to the target and an isolated, peripheral distractor was within 90° of the target. On these isolated distractor trials, we examined saccade curvature elicited by the distractor as saccade curvature is indicative of the competition elicited by a competing saccade goal during target selection processing. This allowed us to determine whether distractor-related competition covaried as a function of the perceptual color space relations between potential saccade goals on the task. We utilized a preliminary analysis to determine whether isolated distractors had a modulatory influence on saccade curvatures. Therefore, we compared saccade curvature between trials on which the distractor appeared counterclockwise (CCW) to the target to trials on which the distractor appeared clockwise (CW) to the target, as previous experiments suggest that saccades should curve in opposite directions in these two contexts (Doyle and Walker, 2001; Tipper et al., 2001). Furthermore, we also calculated the curvature observed on trials with no isolated distractor to establish a baseline level of inherent saccade curvature, which is expected as saccades as idiosyncratically curved (Bahill and Stark, 1975) and previous studies of saccade curvature report directional biases (Doyle and Walker, 2001; McSorley et al., 2004). As such, we observed a significant curvature bias in the CCW direction. Therefore, to determine if saccades curved in opposite directions in the CCW and CW conditions, we examined the difference in unbiased saccade curvature (i.e., baseline subtracted) between these conditions. This analysis was ideal as it merely introduces a scalar offset between the means, one that makes any potential mean difference more interpretable, but does not change the outcome of any statistical analysis since the variances are unchanged. This analysis demonstrated that saccades curved in opposite directions in the CCW and CW conditions relative to the baseline, thus validating our use of saccade curvatures to examine differences in color encoding by the oculomotor system. As such, we then measured saccade curvature as a function of the CIE (x,y) color space distance between the target and distractor. Our data demonstrated that saccade curvature elicited by these isolated distractors was functionally related to the color space distance between the target and distractor in CIE (x,y) color space. This suggested that the color representations that are utilized by the oculomotor system to plan saccadic eye movements are encoded in perceptual color space as with pursuit eye movements (Tchernikov and Fallah, 2010) and visual search (D'Zmura, 1991; Bauer et al., 1996a,b, 1998).

Previous studies have found that remote distractor-related saccade curvature is greater when the featural similarity of the distractor to the target is categorically greater (Ludwig and Gilchrist, 2003; Mulckhuyse et al., 2009). However, the current results demonstrate that oculomotor suppression of a distractor can vary continuously along a particular feature dimension that determines the behavioral relevance of the distractor. This continuous relationship between saccade curvature and distractor behavioral relevance may result from a gradient of oculomotor activation/inhibition in critical oculomotor neural substrates such as SCi where the spatiotemporal interactions of multiple saccade vectors can elicit saccade curvature toward a distractor when the target and distractor vectors are co-activated (McPeek et al., 2003) or away from a distractor when the distractor vector is inhibited (Aizawa and Wurtz, 1998). This reasoning is supported by the observation that saccade curvature toward (McPeek et al., 2003) and away (White et al., 2012) from a distractor is correlated with the magnitude of neural activation encoding the distractor vector. Interestingly, in the critical oculomotor substrates that encode movement vectors and have been most strongly associated with saccade curvature, namely SCi (McPeek et al., 2003; Port and Wurtz, 2003; White et al., 2012) and FEF (McPeek, 2006), there are many cells with visuomotor properties and some that are strictly visual (SCi: McPeek and Keller, 2002; FEF: Mohler et al., 1973; Bruce and Goldberg, 1985). Therefore, what is not clear from the current experiment is whether the featural computations necessary to determine the behavioral relevance of the stimuli is performed locally in these oculomotor substrates or whether alternatively these areas integrate visual and cognitive signals to related saccade vectors and select a winner for the movement. The latter mechanism is supported by research demonstrating behavioral relevance or priority encoding in SCi (reviewed by Fecteau and Munoz, 2006) and with a winner-take-all saccade triggering mechanism in SCi (see White and Munoz, 2011 for a discussion). Critically, this latter mechanism is also consistent with the current observation of surround suppression.

The relationship between saccade curvature and target-distractor color space distance was marginally fit by both Mexican hat and quadratic polynomial functions, and when treating color space distances categorically, there was a quadratic contrast between the means. This is consistent with surround suppression of the cued saccade target color in color space and provides experimental support for selective tuning (Tsotsos et al., 1995; Tsotsos, 2011). Critically, if surround suppression modulates the strength of visuomotor representations in the oculomotor system, this implies that the featural computation necessary to determine the behavioral relevance of potential saccade goals was not performed locally in oculomotor substrates. This is because surround suppression in feature space results from task-based priming of a relevant visual feature at the top level of representation in the visual hierarchy and the subsequent recursive pruning of connections to unrelated features in a recurrent feedback sweep through the visual hierarchy (Tsotsos et al., 1995; Tsotsos, 2011). This theoretical prediction of ST has been well-supported by recent behavioral (Boehler et al., 2009; Hopf et al., 2010) and neurophysiological (Mehta et al., 2000; Roelfsema et al., 2007; Buffalo et al., 2010) experiments. Since the oculomotor system does not have intrinsic featural representations, this feedback sweep would not propagate through the oculomotor system. Therefore, surround suppression modulating oculomotor representation implies that sensory signals projected into the oculomotor system have already been attenuated in the respective representational networks from which they originate. The behavioral relevance of a potential oculomotor movement goal is likely then determined by the cumulative strength of the attenuated sensory signals across feature domains. Furthermore, we would not expect similar results had we pre-cued the memory-guided saccade target, as this would have introduced the influence of spatial attention, which over the considerably lengthy delay period of 1000 ms, likely would have eliminated the featural competition elicited by distractors that we observed.

We examined whether hierarchical differences between color categories were apparent in various aspects of saccadic target selection processing and saccadic vector encoding as they are in the task irrelevant selection of certain color categories for pursuit eye movements (Tchernikov and Fallah, 2010) or the task relevant selection of visual search targets (Lindsey et al., 2010; Pomerleau et al., 2014) and response inhibition signals (Blizzard et al., 2017). We began by examining whether there was an overall selection bias for certain color categories regardless of task instructions by adding error trials back into the dataset and examining the proportion of total trials on which any particular color category was selected. Consistent with the automatic (i.e., task irrelevant) color selection bias observed by Tchernikov and Fallah (2010) in which participants hierarchically selected color RDKs (red > green > yellow > blue) for pursuit despite never being instructed to do so, we observed a strong overall bias for selecting the color red over green and yellow, and marginally over blue, which was accompanied by a significant linear contrast between the aforementioned color categories. Next, we examined if the inherent competitiveness of an isolated distractor was also modulated by the color category of the distractor. Here we observed only a marginal hierarchical effect of color category, but it was qualitatively the same as the overall selection bias above. Neither the overall selection bias nor the marginal modulation of inherent competitiveness by color category could be explained by differences in the discriminability of color categories in CIE (x,y) color space.

The results from examining overall selection proportion and inherent competitiveness of isolated distractors were qualitatively distinct from those obtained by examining the speed and strength of saccadic vector encoding of the target color (i.e., SRT and the proportion of error trials respectively). Our data demonstrated that when saccades were cued to red targets, they were faster, more precise, and were executed with fewer errors than saccades directed to green and yellow targets. However, blue demonstrated a similar advantage for the proportion of errors and speed, but not for precision. These results were accompanied by significant quadratic contrasts between the aforementioned color categories for the proportion of errors, saccadic latency, and saccadic precision and with insufficient evidence for a linear contrast across these metrics. Clearly, blue was processed more efficiently or elicited more selection bias than would otherwise be predicted by the original color hierarchy (red > green > yellow > blue) of Tchernikov and Fallah (2010). Recall that overall selection proportion and distractor inherent competitiveness showed no such equal advantage for the blue color category. There are several possible explanations for this discrepancy.

Our data may suggest a fundamental variation on the color hierarchy specifically for the speed and strength of saccade vector encoding (red = blue > green = yellow). For example, faster saccades executed to both red and blue targets is consistent with Pomerleau et al. (2014) who observed faster N2pc ERPs to both red and blue targets than green and yellow targets in visual search. Such results could reflect a fundamental difference in the speed of transmission between red/blue and green/yellow color signals. However, this is unlikely given that red/blue and green/yellow color signals are projected through separate anatomical color channels (De Valois and De Valois, 1993). More plausible is that the proportion of errors and response latency metrics are reflective of the perceptual decision component of the task, which is highly related to the discriminability of the stimuli: a perceptual decision threshold is surpassed faster as the target becomes more easily perceptually discriminable (Hanes and Schall, 1996; Gold and Shadlen, 2000; Ditterich et al., 2003). We find this explanation more plausible given that subsequent analyses suggested that lower error rates and faster latencies for red and blue saccades might be explained by the perceptual discriminability of these colors in color space, as SRTs showed a marginal linear relationship with the discriminability of the current color categories in color space, and although there was an insignificant relationship for error rates, a subjective inspection of these means suggests that errors might vary as a sigmoidal or step-like function of discriminability in color space. These results were also consistent with previous studies demonstrating that the discriminability of color stimuli in color space facilitates the response latency in visual search (D'Zmura, 1991; Bauer et al., 1996a,b, 1998). Furthermore, higher perceptual discriminability between stimuli also improves the signal-to-noise ratio in the perceptual accumulation process (Gold et al., 1999; Romo et al., 2003), which could account for our observation of fewer errors for red and blue than green and yellow color categories.

The differences may also be explained by the fact that color was strictly task irrelevant in Tchernikov and Fallah and was crucial for the current task. As such, the perceptual discriminability of the stimuli may impede the featural processing of the saccade targets therefore degrading task performance as is observed in visual search (Verghese and Nakayama, 1994; Verghese, 2001). Another possibility is that the current effects were influenced by differences in the strength of visual working memory representations between color categories. Previous studies that have reported preferential processing for red over other color categories have utilized tasks in which color processing occurs during sustained color stimulation (Lindsey et al., 2010; Tchernikov and Fallah, 2010; Pomerleau et al., 2014; Blizzard et al., 2017). Given that the signal-to-noise ratio of prefrontal neural activity is proportional to visual working task performance (Sawaguchi and Goldman-Rakic, 1991, 1994; Williams and Goldman-Rakic, 1995), utilizing a task that relies on visual working memory color representations may have introduced the influence of perceptual discriminability on localization performance as decreased perceptual discriminability may have degraded working memory representations. In any case, our data provided evidence for privileged processing of red stimuli, consistent with a growing number of studies (Lindsey et al., 2010; Tchernikov and Fallah, 2010; Pomerleau et al., 2014; Blizzard et al., 2017). Therefore, the current results cannot be simply accounted for just by differences in perceptual discriminability and likely arise from an interaction of inherent selection biases for red and the perceptual discriminability of the color stimuli.

We were interested in determining where in the visual processing hierarchy does the biased selection and processing of color categories occur. We utilized a memory-guided saccade task and observed clear hierarchical effects in the selection and inherent competitiveness of color categories independent of perceptual discriminability. Given the delay period of 1000 ms, previous research suggests that the color representations that guided this saccade task were those maintained through recurrent projections in the cortical visual system (Lee et al., 2005) perhaps as early as V1 (Supèr et al., 2001), as supposed to feedforward sensory signals from the retina, which would have decayed after this considerable delay. Therefore, the current categorical color effects likely do not arise from differing proportions of chromatic photoreceptors as has been proposed for speeded visual search (Lindsey et al., 2010).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.