Effects of grouping on preference for color triplets
Schloss and Palmer (2011) found that people prefer color pairs that are harmonious, or similar in hue. In the present study, we investigated the relation between spatial composition and color harmony on preference for color triplets. Specifically, we tested whether people prefer similar colors to be closer together than contrasting colors. Participants indicated which of two simultaneously presented displays of color combinations they preferred. Each display contained three squares in a row, spaced such that the central square was closer to one of the flanking squares than the other. In one combination, the central square hue was similar to the closer square and contrasted with the farther square (e.g., a cyan flanking square being closer to a central green square and a red flanking square being farther). The other display had the same spacing, but the opposite color assignment of the flanking squares (e.g., a red flanking square being closer to a central green one and a cyan flanking square being farther). Participants chose the displays in which the similar colors were closer (i.e., congruent color-spatial grouping) more often than when the similar pairs were farther (i.e., incongruent color-spatial grouping). This effect was modulated by hue, however, in that the effect was stronger for differences in redness-greenness than in blueness-yellowness. Explanations will be considered in terms of (a) Berkeley school spirit (Schloss et. al, 2011), (b) visual discomfort, and (c) natural scene statistics (Juricevic et. al, 2011).
Object Color Preferences
We investigated how object context influences color preferences by comparing preferences for "contextless" colored squares with preferences for colors of a variety of objects (e.g., walls, trim, couches, throw pillows, dress shirts/blouses, ties/scarves, and T-shirts). In Experiment 1 we found that hue preferences for contextless squares generalize relatively well to hue preferences for imagined objects, with substantial differences occurring in the saturation and lightness dimensions. In Experiments 2 and 3 we found that object color preferences are relatively invariant when the objects are (a) imagined to be the color that is presented as a small square, (b) depicted as colored images of objects, and (c) viewed as actual physical objects. In Experiment 4 we found that object color preferences are related to the degree to which colors help objects fulfill particular functions or outcomes. We also considered relations between our results and theories of color preference.
Preference Asymmetries in Color Pairs: Retinal vs. Perceived Size
Does spatial organization influence preference for color pairs? We report evidence for systematic preference asymmetries, in which participants preferred two-color figure-ground displays (a small square centered on a large square) when the color of the larger region was bluer, darker, and more preferred than that of the smaller region. Results were similar for color pairs in mosaic configurations (side-by-side, separated by a gap), showing that surroundedness is not essential for preference asymmetries to arise. We clarified the nature of the relative size effect by testing the predictions of two competing hypotheses. The retinal size hypothesis posits the dominant factor to be the relative retinal area of the figure and ground regions, whereas the perceived size hypothesis posits it to be the relative area after the ground has been amodally completed behind the figure. We measured preference asymmetries for displays in which the figure's area was smaller than (40%), equal to (50%) or larger than (60%) the retinal area of the ground to find out whether preference asymmetries would reverse when the figure's area was larger than that of the ground, as predicted by the retinal-size hypothesis. Instead, participants preferred pairs with yellower figures on bluer grounds in all three conditions, with positive correlations between the difference in blueness between the ground and figure colors for the 40% (r=+.63), 50% (r=+.58), and 60% (r=+.33) configurations. Consistent with the perceived size hypothesis, the blue-yellow effect decreased as the area of the figure approached the area of the amodally-completed ground, without reversing. The same pattern was present for the difference in lightness (darker ground preferred) and in single-color preference between the ground and the figure colors. Accordingly, aesthetic judgments of color pairs depend on the relative areas of amodally-completed regions rather than on the relative areas that are visible in the configuration.
The effects of imagined experiences of objects on preferences for colors
The Ecological Valence Theory (EVT) posits that color preferences are caused by individuals' emotional experiences with color-associated objects (Palmer & Schloss, 2010). In support of this causal claim, Strauss, Schloss, and Palmer (VSS-2011) showed that color preferences can be changed by exposing people to affectively biased samples of colored objects that are strongly positive or negative. Exposure to positive objects of a given color (red or green) increased people's preference for that color. Here, we attempted to change individuals' preferences for red and green by having them imagine positive or negative color-associated red and green objects from their verbal descriptions. First, participants rated their preference for 37 Berkeley Color Project (BCP) colors. Second, they generated mental images from the verbal descriptions of objects and (a) selected the BCP color that best matched the imagined object's color, (b) rated the vividness of their mental image, and (c) rated their preference for the imagined object. Half of the participants imagined positive red objects (e.g., "raspberries") and negative green objects (e.g., "slime") whereas the other half imagined negative red objects (e.g., "wounds") and positive green objects (e.g., "trees"). Both groups imagined neutral objects of other colors. When participants rated their color preferences again, there was a significant interaction between the change in color preference (before vs. after mental imagery) and the valences of the imagined objects: Participants who imagined positive red objects and negative green objects increased their preference for red and decreased their preference for green relative to those who imagined negative red objects and positive green objects. These results demonstrate that even imagined color-specific emotional experiences can influence color preferences. We also report the duration of the effects of perceptual (rather than imagined) experiences by examining the magnitude of the changes after a delay of 24 hours.
Color-Grapheme Associations in Non-Synesthetes: Evidence of Emotional Mediation
Previous research has shown that non-synesthetes have systematic associations between colors and graphemes (e.g., Simner, et al., 2005; Rich et al., 2005; Spector & Maurer, 2008). In the present study we investigated whether color-grapheme associations might be mediated by the emotional associations of the graphemes and the colors. First, participants were presented with the 26 uppercase letters of the Latin-based alphabet and 10 Arabic numerals (Helvetica font), one at a time, beside an array of the 37 Berkeley Color Project (BCP) colors (Palmer & Schloss, 2010). Their task was to select the five colors that were most consistent with the grapheme presented, followed by the five colors that were least consistent (as in Schloss, et al. (VSS-08) for color-music associations and Xu, et al. (VSS-10) for color-face associations). Next, participants rated each of the colors and graphemes along the following emotional dimensions: happy/sad, strong/weak, active/passive, and good/evil. The emotional content of the graphemes was significantly correlated with the emotional content of the colors that were associated with the graphemes for the happy/sad, (r=.64), strong/weak (r=.52), active/passive (r=.46), and good/evil (r=.51) dimensions. Participants also rated the colors and graphemes along several visual dimensions that were more weakly associated: curvy/angular (r=.23), open/closed(r=.35), balanced/unbalanced (r=.20), spacious/cramped (r=.24), and simple/complex (r=.30). These results suggest that the associations between colors and graphemes in non-synesthetes might be mediated, in part, by underlying emotional associations.
An Ecological Account of Individual Differences in Color Preferences
Schloss and Palmer (VSS-2009) reported that 80% of the variance in average color preferences for 32 chromatic colors by American participants was explained by an ecological measure of how much people like the objects that are characteristically those colors. The weighted affective valence estimate (WAVE), computed from the results of a multi-task procedure, outperformed three other models containing more free parameters. One group of participants described all objects that came to mind for each color, which were compiled into 222 categories of object descriptions. A second group rated how similar each presented color was to the color of each object described for that color. A third group rated their affective valences (positive-to-negative) for each object from its verbal description. The WAVE for each color is the average valence for each object, weighted by the rated similarity of the given color to the described object. The WAVEs were strongly correlated with average color preference ratings (r=.89). We now show that when WAVEs are calculated at the level of individual participants, they account for significantly more variance in the same individual's color preference ratings than do average WAVEs computed from the entire group. A new group of participants rated their preferences for all 32 colors, after which they provided their own idiosyncratic object descriptions for each color and rated their affective responses to them. They also rated their affective response to each of the 222 object descriptions provided by the original group. The correlation between the individual's color preferences and his/her individual WAVEs, computed from their personal ratings of the 222 object valences, proved to be reliably better than the fit of the group WAVEs, computed from the average affective ratings. Together, the cone contrast model (Hurlbert & Ling, 2007) and the WAVE predictor explain 58% of the variance in individual participants.
Cross-Cultural Studies of Color Preferences: US, Japan, and Mexico
Consistent with Schloss and Palmer's (VSS-2009) Ecological Valence Theory (EVT) of color preference, 80% of the variance in average American preferences for 32 chromatic colors was explained by the Weighted Affective Valence Estimate (WAVE) of American preferences for the objects that are characteristically those colors. To test predictions of the EVT cross-culturally, corresponding color preferences and ecological WAVE measures were collected in Japan and Mexico for the same 32 colors. American participants showed a broad preference for cool over warm hues, an aversion to dark orange (brown) and dark yellow (olive), and greater preference for more saturated than less saturated colors. Japanese participants showed similar preferences for cool over warm colors, dislike for brown and olive, and high preference for saturated colors, but a greater preference for light colors (pastels) and a lesser preference for dark colors relative to Americans. Mexican participants showed the same aversion to brown and olive, but liked warm and cool colors about equally and tended to like both light and saturated colors less than American and Japanese participants. The WAVEs in each culture were computed from the results of the same three-part procedure: eliciting object descriptions for each of the 32 colors, rating the similarity of the presented color to the colors of the described objects, and rating the affective valence (degree of liking) of each described object. The WAVE for each color is the average valence over objects weighted by the average similarity of the given color to the described object. American WAVEs predict American preferences (r=.89) better than Japanese (r=.77) or Mexican preferences (r=.54). Similarly, Japanese WAVEs predict Japanese color preferences (r=.66) better than American preferences (r=.55) or Mexican preferences (r=.29). These findings are consistent with the EVT, which predicts that culturally specific WAVEs should predict within-culture preferences better than between-culture preferences.
The Good the Bad and the Ugly: Effects of Object Exposure on Color Preferences
Palmer and Schloss (submitted) proposed an Ecological Valence Theory (EVT) of color preferences, which states that color preferences are determined by individuals' emotional experiences with objects characteristically associated with those colors. An implication of the EVT is that an individual's color preferences change as he/she has new emotional experiences with colored objects. The present experiment tests whether exposing subjects to emotional objects of particular colors produces a reliable change in preferences for those colors. Participants first rated their color preferences for the 37 BCP colors. Half then completed four "spatial aesthetics" tasks in which they were exposed to positive green images (e.g., trees and grass) and negative red images (e.g., wounds and lesions), and the other half did the same with negative green images (e.g., slime and mold) and positive red images (e.g., berries and roses). Both groups also saw neutral objects of other colors. The "spatial aesthetics" tasks were designed to insure that participants had processed the content of the images: judging whether a verbal label was appropriate, clicking on the center of the focal objects, rating the complexity of the image, and rating their preference for the depicted object. Following these four tasks, participants rated their color preferences again, and difference scores were computed for the corresponding red and green colors. There was an interaction between the change in color preference and the images viewed: Those who saw positive images of a given color (either red or green) showed an increase in preference relative to those who saw negative images of the same color. These results provide causal evidence in support of the EVT by showing that exposure to (or priming of) emotional objects of a particular color can increase or decrease preference for that color, depending on the emotional valence of the objects.
Effects of school spirit on color preferences: Berkeley's Blue-and-Gold vs. Stanford's Red-and-White
According to the Ecological Valence Theory (EVT), people's color preferences are determined by their average affective response to all "things" associated with those colors (Palmer & Schloss, submitted). Accordingly, preference for a color should increase with increasingly positive feelings for a strong associate of that color (e.g., one's university) and decrease with increasingly negative feelings about that same associate. We tested this prediction by comparing color preference ratings from Berkeley and Stanford undergraduates a few weeks before the intensely rivalrous "Big Game." The EVT predicts that students should like their own school colors more than their rival's school colors, and that the degree of preference for these colors should be related to their amount of school spirit. Berkeley and Stanford undergraduates, rated their preferences for 40 single colors (37 colors of the Berkeley Color Project plus Berkeley-blue, Berkeley-gold, and Stanford-red) and 42 figure-ground color pairs (all pair-wise permutations of Berkeley-blue, Berkeley-gold, Stanford-red, white, light-blue, dark-yellow and light-red). Participants then rated their degree of agreement with five statements designed to assess school spirit. Total school spirit scores from Berkeley and Stanford were combined into a single bipolar dimension by multiplying the Stanford scores by -1. For single colors, there was a significant positive correlation (r=0.44) between school spirit and the signed difference in preference (Berkeley-blue plus Berkeley-gold minus Stanford-red), showing that Berkeley students like blue and gold more than red, whereas Stanford students like red more than blue and gold. Preference for color pairs showed analogous effects: School spirit was significantly correlated with the difference in preference for pairs containing Berkeley's blue-and gold and those containing Stanford's red-and-white (r=0.36). These results support the EVT by showing that positive feelings towards one's university promote higher preference for colors associated with that university than for colors associated with a rival university.
The Color of Emotionally Expressive Faces
Schloss, Lawler, and Palmer (VSS-2008) investigated the relation between color and classical music by having participants select the 5 colors that "went best" (and, later, the 5 colors that "went worst") with 18 musical pieces from among the 37 colors of the Berkeley Color Project (Palmer & Schloss, submitted). They found that the emotional associations of the colors that were chosen for a particular musical selection were closely related to the emotional associations of that musical selection. They proposed that when people perform this task, they have an emotional response to the music and chose the colors that are most (or least) closely associated with those same emotions. In this study we used the same paradigm to test for analogous associations between colors and emotionally expressive faces. In the color-face task, the participants were presented with the entire array of 37 colors beside a photograph of a face that appeared happy, sad, angry, or calm to varying degrees. Their task was to choose the five colors that were most consistent with the face and (later) the five colors that were least consistent with the face. In the color-emotion task, the same participants rated the strength of association for each of 37 colors with three emotional dimensions: happy-sad, angry-contented, and strong-weak. In the face-emotion task, the same participants rated the strength of association for the faces along the same three emotional dimensions. Analogous to color-music associations, the emotional associations of the colors chosen to go with the faces were highly correlated with the emotional content of the faces. The results are consistent with the general hypothesis that associations between colors and stimuli that have clear emotional content (e.g., classical music and emotionally expressive faces) are mediated by emotion: People choose the colors that have the most similar emotional content.
An Ecological Valence Theory of Human Color Preferences
We studied color preferences for 37 colors: 8 hues x 4 brightness/saturation levels plus five grays. Average color preference for saturated, desaturated, and light colors was a relatively smooth function of hue with a strong peak at blue and a trough at chartreuse. The hue function differed for dark colors: dark orange and yellow were much less preferred, and dark red and green were somewhat more preferred. Males preferred saturated colors to desaturated ones, whereas females preferred desaturated colors to saturated ones. Male-female differences were highly correlated with rated activeness/passiveness of colors (r=0.73), with males preferring active colors. People with greater color expertise preferred chromatic colors more than novices, but both preferred achromatic colors equally. Despite enormous individual differences, hue preference was relatively stable within observers across brightness/saturation levels (average r=0.35). Hue preference was also relatively stable across contextual objects: walls, trim, couches, pillows, shirts/blouses, ties/scarves (average r=0.65). A colorimetric model based on yellowness/blueness, saturation and brightness/darkness explained 60% of the group variance, whereas an emotional-association model based on calmness, courageousness, and frustratedness of colors explained 79%. We also tested an ecological-valence model, hypothesizing that people like colors that remind them of positive things (sky, trees) and dislike colors that remind them of negative things (vomit, feces). We asked participants to name all the objects that came to mind for each presented color and later obtained ratings of how positive/negative they felt about each object. We then weighted the average valence for each object by the frequency with which it was reported for each color. Color preference was strongly related to this weighted object valence measure (r=0.83), suggesting that color preference may be analogous to taste in food: a heuristic designed to help guide us toward potentially beneficial objects and situations and away from potentially harmful ones.
Preference for Three-Color Combinations in Varying Proportions
Previous research on preference for color combinations investigated pairs of colors (Schloss & Palmer, VSS-07; Ou & Luo, 2006). The current project investigated preference for combinations of three colors in varying proportions. The full set of 37 colors included eight hues (red, orange, yellow, chartreuse, green, cyan, blue and purple) at four saturation/lightness cuts through color space (high-saturation, medium-saturation, light, and dark), as well as five grays. In Experiment 1, displays were 45-degree-rotated checkerboards whose two colors were always adjacent high-saturation hues. When a third color of any of the remaining six hues was present, it formed squares of four different sizes (large, medium, small and absent) at the intersections of the checkerboard. Observers rated their preferences for each display. Later they also rated their preferences for the individual colors in isolation. The results showed that displays containing no third color were most preferred, and preference decreased as the area of the third color increased (p < .05). When the third color was present, the display was most preferred when the third color was closest in hue to the other two colors, consistent with prior studies of color harmony (Schloss & Palmer, VSS-07). A regression model accounted for 62% of the variance with the following predictors: size of the third color square, average preference for the two individual checkerboard colors, distance in hue between the third color and the checkerboard colors, and redness/greenness of the checkerboard colors. Further experiments examined cases in which adding the third color increased the harmony of the two-color combination, and in which combinations of colors in other cuts of the color space were investigated.
The Role of Spatial Composition in Preference for Color Pairs
In this project we investigated the role of spatial composition in preference for color pairs. Our 37 colors included: 8 hues (red, orange, yellow, chartreuse, green, cyan, blue, purple) x 4 brightness/saturation levels (saturated, desaturated, light, dark) plus five grays. In the first experiment, displays contained two figure-ground pairs (small squares on larger squares), side-by-side. Both pairs contained the same two colors in opposite spatial locations (e.g., yellow figure on blue ground and blue figure on yellow ground). Participants were asked whether they preferred the left, right, or neither pair. A regression model showed that 32% of the variance was explainable by difference in preference for figure and ground colors when rated in isolation: pairs in which the ground color was more preferred than the figure color were more preferred overall. An additional 33% of the variance, however, was explainable by yellowness/blueness and lightness darkness: pairs with yellower, lighter figures and bluer, darker grounds were more preferred. In the second experiment we tested whether relative area, surroundedness or shared perimeter size influenced how spatial composition affected pair preference. In figure-ground displays we varied the area (and perimeter) of the figure; in bipartite displays we varied the relative size of the left and right regions; and in "plus sign" displays we varied the perimeter of the figure and held area constant. Results show that relative area was the most important factor influencing preference. For both figure-ground displays participants preferred color combinations in which the smaller region was yellower and the larger region was bluer. This was also true for bipartite displays, showing that surroundedness was not an important factor. Changes in perimeter when area was held constant had no effect on preference.
Preference for color-pairs within finely sampled color space
In previous research we found that preference for color pairs increased as a function of color similarity (Schloss & Palmer, VSS07). The 37 colors we used were only coarsely sampled within color space, however: 8 hues (red, orange, yellow, chartreuse, green, cyan, blue and purple) x 4 brightness/saturation levels (saturated, light, muted and dark) plus five grays. In this experiment, we tested whether preferences for color pairs remained monotonic when sampled more finely between the colors previously examined. Four equally spaced colors were interpolated between (a) adjacent saturated hues (hue change), (b) saturated and muted colors of the same hue (saturation change), (c) light and muted colors of the same hue (lightness change), and (d) muted and dark colors of the same hue (darkness change). We used checkerboard displays rather than figure-ground pairs with a central square on a larger ground square to avoid potential artifacts due to relative depth. Half of the displays contained narrow gaps between the squares of the checkerboard and half did not, but this factor did not matter. If preference were simply a function of color similarly, then there should be a monotonic increase in preference as color similarity increased. This was not the case: preference increased as similarity in saturation and lightness decreased (p < .001), and it was not significantly affected by similarity in hue (p > .05). In addition, checkerboards with differences in lightness/darkness were always more preferred than those with differences in hue or saturation (p < .001), regardless color similarity. These findings support the hypothesis that very similar colors "clash" more than less similar colors within a finely sampled color space, especially for differences in saturation and lightness. Further experiments will examine the relation between color discriminability and color preference.
Cross-Cultural Differences in Color Preference: Japan vs. the USA
We studied cross-cultural color preferences for the 37 colors of the Berkeley Color Project: 8 hues (unique-red, orange, unique-yellow, chartreuse, unique-green, cyan, unique-blue, and purple) x 4 brightness/saturation levels (saturated, desaturation, light, and dark) plus five grays. Forty observers in Tokyo, Japan, and 48 observers in Berkeley, USA, rated aesthetic responses to all 37 colors using a line-mark ratings scale. The hue preference functions (averaged over brightness/saturation levels) were very similar for Japanese and American observers, with a broad peak around blue and a trough around yellow-chartreuse. Reliable differences were present in preferences for brightness/saturation levels, however. In particular, Japanese observers had a greater relative preference for light colors, rating light colors higher than Americans did and rating dark colors lower than Americans did. Japanese observers also liked desaturated (muted) colors less than American observers for warm colors (chartreuse, yellow, orange, red, and purple) but not for cool colors (green, cyan, and blue). Some gender effects were similar in the two cultures, but others were different. Males in both cultures tended to prefer saturated colors more than females, whereas females in both cultures tended to prefer desaturated colors more than males. Japanese females also preferred relatively lighter colors more than Japanese males, rating light colors more highly and dark colors less highly than their male counterparts. No similar interaction was present in the American observers, however. A colorimetric model based on yellowness/blueness, saturation, and brightness/darkness explained 79% of the Japanese group variance, versus 60% for the American data. An ecological-valence model, based on the assumption that people like colors that remind them of positive things (sky, trees) and dislike colors that remind them of negative things (vomit, feces), will also be fit to the data to determine whether the cultural differences we found can be predicted by this model.
In this portion of the Berkeley Color Project we investigate preferences for single colors and color combinations. Average preferences for single colors yielded a relatively smooth function of hue with a peak at blue and a trough at chartreuse, for which rated blueness/yellowness explained 88% of the variance. Preferences were largely invariant across saturation and lightness levels, except that (a) dark yellow and dark orange (browns) were less preferred than their lighter counterparts, (b) dark red was more preferred. Preference for color combinations could only weakly be predicted by preference for individual colors in each pair. Adding participants' ratings of color harmony, however, dramatically increased the amount of variance explained. Individuals differed in the degree to which they preferred combinations that they found harmonious (correlations raging from -0.11 to 0.74). Originally when this poster was presented we concluded that these individual differences could be accounted for by measures of personality (BFI), but this was no longer true once we had our full set of data. Artistic experience does, however relate to preference for harmony: people with no artistic experience and people with a lot of artistic experience are less influenced by harmony than people who have a moderate amount of artistic experience.
The Berkeley Color Project
The Berkeley Color Project is a large-scale study of color perception and aesthetics. We argue that this complex domain, replete with marked individual differences and emotional overtones, is best approached using a massive repeated measures (MRM) design in which the same subjects evaluate the same set of colors over a wide range of tasks, including preferences (for single colors, color pairs, colors in different contexts), colorimetric assessments (red/green, yellow/blue, light/dark, and saturation), and associations (with various emotional dimensions and classical music). Forty-eight participants balanced for both gender and artistic experience, performed 32 tasks with 37 colors sampled from color space: 8 hues (unique red, green, blue, and yellow plus their approximate angle bisectors: orange, purple, cyan, and chartreuse) at 4 brightness/saturation levels (light, dark, saturated, and medium), plus 5 matched achromatic colors.
All of the above research is funded in part by NSF Grant BCS-0745820 to Stephen E. Palmer and by a gift from Google, Inc.