Social Perception of Facial Color Appearance for Human Trichromatic Versus Dichromatic Color Vision

The human face is a wellspring of social information; we can make rapid judgments about a wide range of social characteristics (e.g., sex, race, age, attractiveness, health, emotion) based on only a brief exposure to facial stimuli (Weisbuch & Ambady, 2011). Although a majority of traditional social psychological research has focused on assessing how these judgments influence downstream phenomena (e.g., stereotyping and behavior), only recently have researchers focused on elucidating the lower level perceptual mechanisms that produce these judgments in the first place (Freeman & Ambady, 2011; Oosterhof & Todorov, 2008). An emerging line of research has demonstrated that one such perceptual mechanism that can independently influence a range of social judgments is facial color appearance (for reviews, see (Rowland & Burriss, 2017; Stephen & Perrett, 2015; Thorstenson, 2018). This work suggests that human color vision plays a key role in social perception processes. In the current article, we briefly review theoretical accounts of human color vision evolution, including a recent account driven by social evolutionary considerations. We then overview the existing literature on the role of facial color appearance in the expression and perception of social characteristics (i.e., attractiveness, health, and emotion), grounded in an ecological theory of social perception (Zebrowitz-McArthur & Baron, 1983). Finally, we report two experiments that provide an empirical test of the social evolutionary account of human color vision.

Typical human (and most primate species) color vision is considered trichromatic, meaning we possess three distinct classes of wavelength-sensitive photoreceptive cells called cone photorecepters, or cones, that allow us to discriminate chromatic colors across the visible spectrum. These cones are distinguished as short wavelength (S; with peak sensitivity at ~430 nm), medium wavelength (M; with peak sensitivity at ~535 nm), and long wavelength (L; with peak sensitivity at ~562 nm; Jacobs & Deegan, 1999). Whereas most mammalian species possess dichromacy (having one S cone and one M/L cone), trichromacy in humans and most nonhuman primate species evolved when the older M/L cone opsin gene duplicated and diverged into two separate M and L cones (Dulai, von Dornum, Mollon, & Hunt, 1999).

The functional significance of trichromacy evolution is currently under debate. A popular account suggests that trichromacy was selected to aid primates in foraging by facilitating the detection of ripe fruit against green leaves (Allen, 1879; Lucas et al., 2003; Mollon, 1989; Osorio & Vorobyev, 1996; Surridge & Mundy, 2002). While there is support demonstrated for this hypothesis (Bunce, Isbell, Grote, & Jacobs, 2011; Caine & Mundy, 2000; Melin et al., 2009; Osorio, Smith, Vorobyev, & Buchanan-Smith, 2004; Regan et al., 2001; Smith, Buchanan-Smith, Surridge, & Mundy, 2003; Smith, Buchanan-Smith, Surridge, Osorio, & Mundy, 2003; Sumner & Mollon, 2000; Vorobyev, 2004), some ecological studies do not observe a benefit of trichromacy on foraging behavior (Hiramatsu et al., 2008; Vogel, Neitz, & Dominy, 2007). Other theoretical accounts of color vision variation in primates that have received less empirical attention include camouflage detection, predation detection, and nocturnal versus diurnal activity (see Kawamura & Melin, 2017, for a comprehensive review).

Alternatively, a more recent functional account posits that trichromacy may have been selected by social evolutionary pressures (Changizi, 2010; Changizi & Shimojo, 2011; Changizi, Zhang, & Shimojo, 2006). Specifically, this account suggests that the divergence of M and L cones allows for the optimal detection of socially relevant skin color appearance fluctuations in others. Changizi and colleagues (2006) note that changes in dermal hemoglobin oxygenation and concentration lead to predictable changes in the spectral reflectance (and consequently visible color) of skin. Increases in hemoglobin oxygenation heighten relative L-cone to M-cone activation, resulting in redder skin appearance (while decreases in hemoglobin oxygenation produce the opposite, resulting in greener skin appearance). Increases in hemoglobin concentration tend to increase M- and L-cone activation relative to the S cone, resulting in blue skin appearance, whereas decreases in hemoglobin concentration produce the opposite, resulting in yellow skin appearance (see also Thorstenson, 2018 for a review). Importantly, Changizi and colleagues (2006) demonstrated that M- and L-cone sensitivities in trichromats are situated in the visible spectrum such that they are able to optimally detect these skin color appearance fluctuations. This correspondence, between the spectral sensitivity of trichromatic photoreceptors and spectral fluctuations of skin enduring transient hemoglobin changes, presumably enables particular sensitivity to discriminate socially relevant physiological conditions (e.g., emotional states, sexual signals, and threat displays). It has also been noted that trichromatic primates tend to be bare-faced, allowing rapid access to visual skin color appearance modulations. See Figure 1 for an illustration of the spectral characteristics of human skin and trichromatic photoreceptors. While there is some recent work supporting this hypothesis for viewing nonhuman primate targets (Hiramatsu, Melin, Allen, Dubuc, & Higham, 2017) research in this area is still sparse and has yet to be conducted using human targets.

An ecological theory of social perception holds that social perception processes serve an adaptive function (Zebrowitz, Bronstad, & Montepare, 2011; Zebrowitz-McArthur & Baron, 1983). Perception facilitates goal attainment and species propagation by informing behavior (Gibson, 1979). This ecological approach to social perception assumes that (a) the external environment provides information to guide biologically and socially functional behaviors, and (b) the successful transfer of this information relies on a compatibility between a signal (stimulus information) and the perceiver (perceptual system). In the current investigation, we focus on the stimulus information (facial color expression) and the perceptual system (facial color perception) involved in social processes related to evaluation of attractiveness, health, and emotion. We chose to focus on these evaluations because they represent interpersonally important perceptions that shape social interaction and decision making. For instance, perceptions of attractiveness and health guide mating decisions because they provide an indicator of overall mate quality and reproductive potential (Etcoff, 1999; Perrett, 2010; Rhodes et al., 2007; Weeden & Sabini, 2005). Similarly, perceptions of emotion (e.g., anger) inform situational context (e.g., dominance, hostility) and guide behavior (e.g., avoidance; Marsh, Adams, & Kleck, 2005).

Indeed there is evidence that skin color appearance subtly undergoes change as an inevitable consequence of socially relevant states. In nonhuman primates, female skin color appearance becomes redder throughout the ovulatory cycle, when sexual cues afford the most significant reproductive consequences (Bielert, Girolami, & Jowell, 1989; Deschner, Heistermann, Hodges, & Boesch, 2004; Dixson, 1983; Gerald, 2003; Setchell & Wickings, 2004; Setchell, Wickings, & Knapp, 2006; Waitt, Gerald, Little, & Kraiselburd, 2006). Skin color appearance in male nonhuman primates also influences female preferences by signaling elevated testosterone (Rhodes et al., 1997; Waitt et al., 2003). In addition, there is evidence that the same skin color appearance modulations occur in human females (Burriss et al., 2015; Edwards & Duntley, 1949; B. C. Jones et al., 2015; McGuiness, 1961; Oberzaucher et al., 2012; Snell & Turner, 1966; van den Berghe & Frost, 1986) and possibly males (due to elevated testosterone; Miller & Maner, 2010). Furthermore, skin color appearance undergoes change as a consequence of physiological states related to health, including skin vascularization (Changizi & Shimojo, 2011; Charkoudian, 2003; Henderson et al., 2017; Panza, Quyyumi, Brush, & Epstein, 1990; Ponsonby, Dwyer, & Couper, 1997; Sibenge & Gawkrodger, 1992; Wilkin, 1994), bilirubin (Knudsen & Brodersen, 1989), melanin (Stamatas, Zmudzka, Kollias, & Beer, 2004; Zonios, Bykowski, & Kollias, 2001) and carotenoids (Alaluf, Heinrich, Stahl, Tronnier, & Wiseman, 2002; Coetzee & Perrett, 2014; Tan, Graf, Mitra, & Stephen, 2015, 2017; Whitehead, Re, Xiao, Ozakinci, & Perrett, 2012). Finally, there is a sizable literature examining the physiological correlates of experiencing various emotion states, which can then be used to predict how skin color appearance likely changes with respect to these emotions (see Thorstenson, 2018; Thorstenson, Elliot, Pazda, Perrett, & Xiao, 2018, for reviews).

In accord with the above literature showing the relation between socially relevant states and skin color appearance fluctuations, recent research has demonstrated that facial color appearance can independently influence an array of social evaluations related to these states. For instance, facial color appearance can influence perceptions of attractiveness (Carrito et al., 2016; Lefevre, Ewbank, Calder a, Hagen, & Von Perrett, 2013; Pazda, Thorstenson, Elliot, & Perrett, 2016; Re, Whitehead, Xiao, & Perrett, 2011; Said & Todorov, 2011; Stephen, Oldham, Perrett, & Barton, 2012; Thorstenson, Pazda, Elliot, & Perrett, 2016; Young, 2015), health (A. L. Jones, Porcheron, Sweda, Morizot, & Russell, 2016; Re et al., 2011; Stephen, Coetzee, Law Smith, & Perrett, 2009; Stephen, Law Smith, Stirrat, & Perrett, 2009; Young, Thorstenson, & Pazda, 2016), and emotion (Benitez-Quiroz, Srinivasan, & Martinez, 2018; Thorstenson, Elliot, et al., 2018; Thorstenson, Pazda, Young, & Elliot, 2018). Notably, and most relevant to the current work, slightly redder and yellower faces are consistently positively associated with perceived attractiveness, health, and anger (for moderators of this pattern, see also Fisher, Hahn, Debruine, & Jones, 2014; Han et al., 2018; Kandrik et al., 2017).

Considering the social significance of facial color perception, and the hypothesis that trichromatic color vision was selected by social evolutionary pressures to detect socially relevant states in others, it follows that trichromatic color vision should provide an advantage relative to dichromatic color vision in social perceptions arising from differences in facial color appearance. Hiramatsu and colleagues (2017) provided initial evidence of this hypothesis. The researchers presented human observers with pairs of rhesus macaque photographs—proceptive versus non-proceptive faces—and asked them to select the proceptive face. In addition, the researchers presented images that simulated what these face pairs would look like to trichromats (equivalent to the original images) and dichromats (protanopia, deuteranopia, tritanopia), by computationally modeling the visual response from the corresponding cone sensitivities. The researchers found support that the trichromatic image representations conferred an advantage relative to the dichromatic image representations in detecting the proceptive faces.

Although Hiramatsu and colleagues’ (2017) data provide initial support for a social evolutionary account of trichromatic color vision, some methodological issues prohibit drawing conclusions with regard to human interaction. Although proceptivity is an important characteristic for male rhesus macaques to identify in females, humans experience and exhibit a much more complex range of social states and emotions, and the degree to which trichromatic (vs. dichromatic) color vision influences judgments of these states and emotions remains unknown. Second, facial coloration differences within and outside of the proceptive behavioral period for rhesus macaques are likely to be much greater than facial coloration appearance changes that occur within humans. In other words, the differences in facial coloration that participants viewed in Hiramatsu et al.’s (2017) study were much more conspicuous than would be expected in human interaction. Examining how judgments of socially relevant variables differ as a function of trichromatic/dichromatic vision across subtle, real-life skin color appearance variations therefore seems essential. Third, one of the photographs in each trial of the Hiramatsu et al. (2017) study was objectively proceptive, and participants were provided with accuracy feedback in a series of practice trials as to which was the correct selection, essentially training participants to associate facial redness appearance with proceptivity. This type of feedback is unlikely to occur in human interactions. For example, people make snap judgments of socially relevant factors (e.g., health) without necessarily obtaining subsequent information to confirm or disconfirm their initial judgment.

The current research aims to extend our understanding of a social functional account of trichromatic color vision in humans. Our central focus is to manipulate human facial coloration in a manner consistent with a typical blushing response, simulate how this would be perceived by trichromatic versus dichromatic observers, and assess whether these low-level perceptual differences influence subsequent judgments of attractiveness, health, and emotion, each of which has important socially relevant consequences.

Previous work has demonstrated that computational modeling can be used to simulate what images would look like to dichromatic (and animal) observers, by adjusting the modeled cone sensitivities according to the missing photoreceptor (Brettel, Viénot, & Mollon, 1997; Hiramatsu et al., 2017; Melin, Kline, Hickey, & Fedigan, 2013; Viénot, Brettel, & Mollon, 1999; Viénot, Brettel, Ott, Ben M’Barek, & Mollon, 1995). We use this approach herein to generate image representations that simulate what the images would look like to routine trichromats (color-normal vision; sensitivity along three distinct L, M and S cones) and three dichromatic conditions: protanopia (missing sensitivity of L cone; red-green color deficiency), deuteranopia (missing sensitivity of M cone; red-green color deficiency), and tritanopia (missing sensitivity of S cone; blue-yellow color deficiency). We used two sets of faces: baseline (unmanipulated) faces, and the same faces manipulated to have increased skin lightness (CIELAB L* + 0.23), redness (CIELAB a* + 5.43), and yellowness (CIELAB b* + 1.88). These color values were obtained from previous data (Thorstenson, 2017) by measuring a hyperemic skin response (increased blood perfusion beneath the skin), simulating a flushed face. In Experiment 1, we simultaneously presented the baseline and flushed faces in pairs for the four color vision conditions and asked participants to select the faces that looked more attractive, healthy, and angry. In Experiment 2, we used a similar experimental design, except that participants were asked to rate on a continuous scale how attractive, healthy, and angry the faces looked. For both experiments, we expected flushed faces (relative to baseline faces) to be perceived as more attractive, healthy, and angry in the trichromatic condition. In addition, we expected that the difference between these evaluations in baseline and flushed faces would be stronger in the trichromatic condition relative to the dichromatic conditions. Particularly, we expected to observe the greatest difference between the trichromatic and the two red-green color-deficient conditions (protanopia and deuteranopia).

For both experiments, all data exclusions, manipulations, and variables analyzed are reported, and data collection was completed prior to any analysis. All participants were unique; each individual only participated in one experiment. All analyses included only participants with color-normal vision, assessed by self-report at the end of each experiment; participants who reported a color vision deficiency were excluded from all analyses a priori. Sample size was determined by a simultaneous a priori consideration of power and expected participant availability. In both experiments, sample size (target n = 67) was determined a priori via power analysis (targeting .80 power to detect a d = 0.35 effect at p < .05 in a within-subjects design). We fell slightly short of this target for Experiment 1 (n = 66) but were able to slightly exceed it for Experiment 2 (n = 81), due to participant availability in both cases.