A ‘breaking’ continuous flash suppression (b-CFS) task investigated whether differences in skin tone predicted perceptual processing fluency of faces prior to the latter entering conscious processing (Nakamura & Kawabata, 2018). During b-CFS, low-contrast faces appeared as monocular targets to the sensory non-dominant eye (Ding et al., 2018). Targets were temporarily ‘suppressed’ from conscious detection for several seconds with dynamic (10 Hz) high contrast Mondrian patterns presented to the dominant eye. Participants were tasked with emitting a localization response as soon as they could perceive target position. Accurate localization responses can indicate faces having entered subjective (modulation of sensory experience) and objective (knowledge of spatial location) awareness (Pournaghdali & Schwartz, 2020). Assuming time required for faces to enter conscious awareness (break suppression) is inversely related to their perceptual load, shorter breaking times can imply faster “transition from unconscious (to) conscious” awareness of the perceived target (Stein, 2019, p. 26).
It was presently tested whether lighter and darker variants broke suppression at different rates once target attractiveness, race, sex and identifiability (upright/inverted orientations) were controlled for. Attribute controls were motivated by earlier works demonstrating attractive/upright faces break suppression faster relative to unattractive/inverted faces, implying the former presents biologically significant and familiar configurations privileged by innate processes (Nakamura & Kawabata, 2018; Nakamura & Watanabe, 2020; Pournaghdali & Schwartz, 2020; Sui & Liu, 2009). Half of the current sample viewed inverted faces during CFS to assess influence of face identifiablity on breaking times (Experiment 2, Nakamura & Kawabata, 2018). If effects overlap across upright and inverted faces, breaking time effects could be attributed to low-level structural differences between images (e.g. median luminance levels, average nose widths). Alternatively, breaking time effects augmented across upright (relative to inverted) faces would imply skin tone influences perceptual processing following identification of precise facial configurations (Nakamura & Kawabata, 2018). Planned contrasts consisted of two-sample tests comparing breaking times for lighter and darker variants for upright and inverted face conditions independently, paralleling Study 1’s analysis strategy. Prior to CFS, all participants completed a manual sorting task designed to identify (any) explicitly held colorist biases (see Procedure). Since sorting tests reliably discriminate between ordinally valenced stimulus categories (Amd et al., 2018; Amd & Passarelli, 2020), we tested whether Study 1’s outcomes would replicate across visually processed (achromatic and ovally masked) faces which had to be ordinally sorted. Similar to Study 1, sorting tests afforded unconstrained deliberation opportunities prior to response confirmation. Additionally, exposing participants to processed faces was expected to mitigate novelty-related confounds potentially influencing breaking times.
Method
Participants
We aimed to recruit 60 participants (30 per orientation condition) in accordance with prior b-CFS sampling conventions (Ding et al., 2018; Nakamura & Watanabe, 2020; Stein et al., 2020). CFS data from 2 participants (one per orientation condition) were incomplete (~ 12% of trials not attempted) due to power outages during data collection. CFS data for one participant could not be collected due to a cyclone warning. The final pool consisted of 30 (25 female) and 30 (26 female) participants allocated to upright and inverted face conditions respectively. Sensitivity analyses for a two-sided paired t-test indicated a sample of 30 could detect moderate-to-large effects (d > .52) with 80% power and a 5% \(\alpha\) error rate. Because comparable effects and sample sizes are common to b-CFS research, increasing sample size to detect smaller effects was not deemed necessary (Stein et al., 2020). Nevertheless, frequentist analyses were complimented with Bayes factors to inform the likelihood of the data given the alternative exploratory hypothesis (of colorist bias). All participants who took part were between 20 and 34 years of age, right-handed, reported normal vision and no averse history with high-contrast visual stimuli. All reported procedures were completed in under 40 minutes.
Materials
Target faces were despeckled, passed through a Gaussian spatial noise filter, rendered achromatic and standardized along contrast levels using the imager R package and commercial image processing software (BatchPhoto). Peripheral distinguishing features (hair, ears, chin) were concealed with a uniform oval mask, retaining facial configuration and skin tone as distinguishing features. Two-sample tests indicated image luminance and contrast levels between lighter and darker variants within HAF/HAM/LAF/LAM categories were non-significantly different (all p’s > 0.07). Processed faces appeared as monocular targets during suppression. The same faces were also printed and laminated on 3 by 4 inch cards for the pre-CFS sorting task. For masks, 256 unique Mondrian-like patterns containing anywhere between 50 and 200 randomly sized achromatic rectangles with black, grey and white hues were generated. Faces and Mondrians were achromatic to reduce spectral leakage between displays (Pournaghdali & Schwartz, 2020, p. 5). The CFS task was developed on E-Prime 3 (Psychology Software Tools, Inc. [E-Prime Go], 2020) and administered in a quiet dark room. Stimuli during CFS appeared on a 24 inch LCD monitor with a 144 Hz refresh rate. The monitor was located 20 inches from the principal axis of a fixed chin-and-head rest with a black vertical divider splitting the enclosed display vertically along the middle. All participants were fitted with customized \(+{2.5}_{D}\) prism glasses, with each prism lens oriented to one half of the display (bases facing inward). Stimulus legibility was confirmed by having participants read instructions presented to alternating eyes prior to CFS onset. Samples of stimuli and a recorded demo of the CFS task are available in the online file.
Procedure
Sorting tests.
Upon arrival, participants were seated in front of a desk with four decks of cards located equidistantly from each other. Each deck contained eight achromatic faces printed on individual laminated cards against a grey background. Decks contained faces from HAF, HAM, LAM and LAF categories respectively. Deck and card placement was randomized between participants. Participants were instructed to randomly select a deck, then ‘spread out the cards’ (faces) on the desk and carefully examine them. Near the wide edge of the desk, eight ‘boxes’ were visually demarcated along a row. Participants were instructed to ‘move the face (they) liked the most’ into the left anchor box, then place relatively less preferred faces across adjacent boxes. So, eight faces were to be ordinally ranked from ‘most liked’ to ‘least liked’. Participants could freely reallocate faces between boxes any number of times and were under no time constraints. Once the participant signaled completion, the experimenter quietly collected the cards and placed them out of sight. The participant was prompted to select from one of the three remaining decks and repeat the task (ordinally rank eight faces by subjective preference). After faces from all four decks had been ranked and collected, participants were moved to a separate desk with the CFS setup.
Continuous flash suppression.
After setting up the prism glasses, participant heads were stabilized on the chin-and-head rest and their hands situated on the response keys. This included the left index and middle fingers on the ‘z’ and ‘a’ letters respectively, and the right hand on the spacebar. All responses were collected on a QWERTY low-latency (< 2 ms) mechanical keyboard. Letters were taped with up- and down-facing arrowheads to facilitate tactile discrimination. Each eye independently viewed 600 by 800 pixel grey ‘boxes’ against black backgrounds enclosed by black and white (15 pixels wide) fusion contours throughout the task, with a white fixation circle appearing in the center (of each box). Participants were instructed to center their eyes near fixation until contours appeared to binocularly converge, at which point they were to press the spacebar to begin. Binocular fusion was reported by most participants within 10 seconds of the starting trial, and by all participants by the second trial. Pressing a spacebar produced a 10 Hz mask (10 Mondrian patterns per second) to one eye. To the other eye, only the grey box remained visible for 1800–2000 milliseconds (ms). Then, a low-contrast face emerged 120 pixels above/below fixation to the non-suppressed eye. The contrast of the face relative to the grey background was linearly increased for 2000 ms. The maximum-contrast target remained on screen for an additional 7000 ms or until a localization response was detected. Participants were isntructed to press ‘z’ (‘a’) as soon as they could detect a face near the bottom (top) of the display. Pressing either key replaced both displays with blank grey boxes. During the non-timed inter-trial interval, participants could move their eyes and relax. Before continuing, participants had been instructed to achieve binocular fusion before progressing with a spacebar press. All target faces appeared an equal number of times above/below fixation for each eye condition, culminating in 128 trials per participant and face orientation condition. Completion of all trials terminated both displays and ended the experiment.
Results
Sorting performances
61 participants ordinally sorted eight faces from most (level 1) to least (level 8) preferred across HAF, HAM, LAF or LAM categories separately. Before analysis, rankings were classified under preferred (levels 1–4) and non-preferred (levels 5–8) categories in part to control for racially motivated rankings. Since each deck contained equal numbers of lighter and darker variants from each race, participants who sorted faces based on racial characteristics should select comparable frequencies of lighter/darker variants of the preferred race over variants of the non-preferred race, which would produce a distribution indistinguishable from the null. Alternatively, if participants sorted faces by skin tone, frequencies of lighter variant selections should collectively differ between preferred and non-preferred categories. Two-sample tests confirmed mean frequencies of lighter variant selections were statistically greater for HAM, t(120) = 4.18; p = .001; g = 0.76, LAM, t(120) = 5; p = .001; g = 0.91, and LAF t(119.4) = 2.47; p = 0.015; g = 0.45 faces across preferred relative to non-preferred categories.
Continous Flash Suppression
Before analysis, the first six CFS trials were dropped to control for learning effects. Next, 1.20% of trials with responses detected within 2000 ms of trial onset (during which no targets were present) were dropped, followed by the removal of all incorrect localization responses. This left 95.2% and 95.6% of trials across upright and inverted face conditions respectively. Ocular non-dominance was estimated by comparing across breaking times between eye conditions for individual participants. The eye associated with longer breaking times was targeted for suppression (Ding et al., 2018). Breaking times for non-dominant eyes were significantly longer then for dominant eyes across upright and inverted face conditions (all p’s < .001).
Planned contrasts confirmed significantly faster breaking times across lighter HAM, (M = 5912.5, SE = 88.6 ms) relative to darker HAM (M = 6343.4, SE = 102.1 ms) when upright faces appeared as targets, t(423.7) = 3.19; p = 0.002; g = 0.31 (Panel C, Fig. 2). Contrasts across remaining categories were non-significant (all p’s > 0.09). Bayes factors estimated with .707 Cauchy priors indicated the data for 14 times more likely to occur if the alternative hypothesis of HAM colorism was true, \(\varDelta M=418.49\), 95% HDI \(\left[159.49,687.10\right]\), \({\text{B}\text{F}}_{\text{10}}=14.01\). Bayes factors for remaining categories revealed the data was extremely unlikely (all \(B{F}_{10}{\prime }s<.8\)) if alternative hypotheses (of LAM/LAF/HAF colorism) were true.
Discussion
Breaking time differences indicate lighter HAM had been processed more fluently compared to darker HAM when targets were upright. Corresponding Bayes factors produced credible evidence for the alternative hypothesis (of HAM colorism) being true. Breaking times did not vary between lighter and darker variants for any of the remaining upright face categories, nor across any inverted face category. Associated Bayes factors indicated extremely weak evidence of the alternative hypotheses across non-HAM faces. Sorting performances produced reliable colorist effects across HAM, LAM and LAF categories, largely replicating Study 1’s outcomes.
Standardized effects with 1000-bootstrapped 95% confidence intervals (CIs) were estimated for HAM/HAF/LAM/LAF categories across studies as they tested conceptually comparable hypotheses. Effects and CIs across studies are summarized in Table 1 (also see Figure S2 in the Supplementary file). CIs that did not overlap with null intercepts represent statistically non-equivalent effects (Lakens, 2017). Across unconstrained measures (attractiveness ratings, sorting tests), non-negligible (d’s > .2) statistically non-equivalent colorism was detected across HAM, LAM and (during sorting tests) LAF categories. Across constrained measures (implicit tests, flash suppression), non-negligible colorist biases were detected in the presence of upright HAM only.