Using Single Colors and Color Pairs to Communicate Basic Tastes II: Foreground–Background Color Combinations

Andy T Woods; Fernando Marmolejo-Ramos; Carlos Velasco; Charles Spence

doi:10.1177/2041669516663750

. 2016 Sep 20;7(5):2041669516663750. doi: 10.1177/2041669516663750

Using Single Colors and Color Pairs to Communicate Basic Tastes II: Foreground–Background Color Combinations

Andy T Woods ^1,^✉, Fernando Marmolejo-Ramos ², Carlos Velasco ³, Charles Spence ³

PMCID: PMC5034334 PMID: 27708752

Abstract

People associate basic tastes (e.g., sweet, sour, bitter, and salty) with specific colors (e.g., pink or red, green or yellow, black or purple, and white or blue). In the present study, we investigated whether a color bordered by another color (either the same or different) would give rise to stronger taste associations relative to a single patch of color. We replicate previous findings, highlighting the existence of a robust crossmodal correspondence between individual colors and basic tastes. On occasion, color pairs were found to communicate taste expectations more consistently than were single color patches. Furthermore, and in contrast to a recent study in which the color pairs were shown side-by-side, participants took no longer to match the color pairs with tastes than the single colors (they had taken twice as long to respond to the color pairs in the previous study). Possible reasons for these results are discussed, and potential applications for the results, and for the testing methodology developed, are outlined.

Keywords: color, color pairs, basic tastes, crossmodal correspondences, design

Introduction

Research shows that people tend to associate gustatory and nongustatory information in ways that are often surprising (e.g., O’Mahony, 1983). So, for example, people not only match a flavor with a color when they belong to the same object, for example, the aroma or flavor of cucumber with the color green (Garber, Hyatt, & Starr, 2000; Stillman, 1993; Velasco et al., 2015) but also match some more abstract dimensions of sensory information such as basic tastes with color cues (Koch & Koch, 2003; Spence et al., 2015). These matchings, what are often referred to as crossmodal correspondences, refer to the surprising associations that people share between sensory features, attributes, or dimensions of experience in different senses (see Marks, 1978; Parise, 2016; Spence, 2011, 2012, for reviews).. In a recent review of the literature on crossmodal correspondences between colors (or color words) and basic tastes (or taste words), Spence et al. (2015) concluded that, across a number of studies that have been conducted over the past three decades, people consistently associate certain colors with specific basic tastes. Sweet, for example, tends to be associated with colors such as red or pink, sour with yellow or green, salty with white or blue, and bitter with black or other dark colors such as brown or purple (see also Favre & November, 1979; Velasco et al., 2016). One possible reason why such associations arise is that it is actually some specific or general food memory prompted by the color, or indeed even hue, that gives rise to a particular assignment to taste (as suggested in the discussion of Woods, Poliakoff, Lloyd, Dijksterhuis, & Thomas, 2010), rather than the color itself. For example, the color pink may prompt the concept of candyfloss, and it may be this association rather than the color itself that is thought of as sweet (not all of the authors though believe that correspondences are necessarily grounded in specific experiences of environmental objects that possess both features).

Crossmodal matching studies, as well as those studies involving the matching of taste words with colors, suggest that people make reliable associations between tastes and single colors (see Koch & Koch, 2003; O’Mahony, 1983; Spence et al., 2015; Velasco et al., 2016; see also Razumiejczyk, Macbeth, Marmolejo-Ramos, & Noguchi 2015, for recent evidence that taste words can simulate actual gustatory taste experiences). Nonetheless, further research is still needed in order to clarify the extent to which pairs (or triples, etc.) of colors would lead to more consistent taste matchings (e.g., Schifferstein & Howell, 2015), be they either congruent, such as red and pink for sweet, or incongruent, such as red and green for sweet (Woods & Spence, 2016). This empirical question is particularly relevant given that the visual appearance of most foods and drinks does not include only a single color. Instead, they involve various visual characteristics with different ranges of hue, brightness, and saturation (one can even consider how people view a colored food against the background color of the plate; Piqueras-Fiszman, Alcaide, Roura, & Spence, 2012). Therefore, it can be hypothesized that multiple color cues, if chosen appropriately, may provide more information when inferring a specific taste quality.

In a recent study, Woods and Spence (2016) presented pairs of colors side-by-side (see Figure 1(a) for an example). The participants had to decide as rapidly as possible which of the four basic tastes (sweet, bitter, salty, or sour) the color pair was best associated with. One of the results to emerge from this study was that some of the 16 color pairs tested better associated with one of the basic tastes than when either of the respective colors was shown individually (e.g., for sourness, the green-yellow color pair). However, on the flip side, the participants took more than twice as long to allocate color pairs, which could be taken to imply some internal reasoning as to which color the participant decided to pair up with a given taste. According to Woods and Spence, it could also mean that the colors are individually assessed in a serial manner and then assimilated, which might well take some time to achieve. A third explanation is that the colors clashed,¹ which may have interfered with the participants’ rating. The first two explanations imply the involvement, in the study in the Woods and Spence’s article, of a quite different set of (possibly higher order) processes (e.g., operating at a decisional level) than would typically occur in everyday perceptual experience.

Both of the studies reported here are based on the methodology developed in Woods and Spence (2016). In the first experiment though, rather than presenting side-by-side color patches, as in Woods and Spence’s (2016) recent study, we decided to present a foreground–background pair of colors instead (see Figure 1(b) for an example). Our expectation was that such an arrrangement would be somewhat more likely to be treated as a unitary gestalt rather than as two separate elements to be evaluated individually, as may have been the case in Woods and Spence’s study. We tested whether combined color cues versus single colors would result in significant differences in response times when matching the color stimuli with taste words. The prediction was also that the foreground color (in Figure 1(b), blue) would dominate over the background color (pink; henceforth, stimuli are referred to according to their foreground than background color, which here is bluePink) in terms of taste expectations, be it simply because by far the majority of the patch is of the foreground color (more salient), or because, for example, foreground elements tend to be processed sooner than those in the background (Lester, Hecht, & Vecera, 2009). Given this, we predicted that simply reversing the foreground and background colors could well lead to quite different taste expectations. In our second study, we used a different paradigm that we thought would be more sensitive to association differences compared with that used in Experiment 1.

Experiment 1

Methods

In both of the studies reported here, we use robust statistics (see Wilcox, 2012 and the R package “WRS”). Specifically, whenever we report means (i.e., averages), we refer to 20% trimmed means (see Wilcox, 2005) and 95% CIs were estimated around means, via the “trimci” function. Percent-bend correlations, here r_pb, and 95% CIs around correlation values were used, which were generated via the “corb” function. One-way robust repeated measures analyses of variance were estimated for comparisons between dependent groups (via the “rmanova” function). The p values of post hoc comparisons and multiple pairwise comparisons were adjusted via the Holm-Bonferroni method (Holm, 1979). Categorical data were analyzed via the chi-square test.

Participants

One hundred participants (62 women; M_age = 31.3, SD_age = 11.3, range_age = 18–64 years) were recruited from Prolific Academic to take part in the study in return for a payment of 1.00 UK pound. By means of Prolific Academic’s “filter” feature, only those participants who reported having been born in the UK were allowed to take part in the study. The experiment was conducted on November 16, 2015 from 10:50 GMT onwards, over a period of 90 min. The participants completed the study at a rate of 1.13 participant completions per minute (see Woods, Velasco, Levitan, Wan, & Spence, 2015, for a recent methodological overview of Internet-based perception research). The participants took an average of 366 s (95% CI [343, 390]) in order to complete the study. All of the participants in both studies provided their informed consent prior to taking part. The experiments were reviewed and approved by the Central University Research Ethics Committee at Oxford University and were carried out in accordance with the World Medical Association Helsinki Declaration (World Medical Association, 2013).

Stimuli

Sixty-four color squares (75 × 75 pixels) or circles (diameter 75 pixels), each with a 5-pixel color border, were used as stimuli. The central color and border color consisted of all possible combinations of green, yellow, red, pink, blue, white, black, and purple, the so-called internet safe colors (hex codes were 00FF00, FFFF00, FF0000, FFC0CB, 0000FF, FFFFFF, 000000, 800080, respectively). Whether a participant saw square or circular stimuli was determined randomly at the start of each participant’s experimental session. The background against which the stimuli were presented was gray (hex code 7A7A79).

Apparatus

Given that the experiment was conducted online, the apparatus varied by participant. Nevertheless, the experiment utilized “full screen” mode (i.e., utilizing the entirety of the participant’s monitor), and took place within a 1024 × 768 pixel box in the center of the screen, irrespective of the size of the participant’s monitor. The experiment was conducted on the Internet using the Adobe Flash-based version of Xperiment (http://www.xperiment.mobi).

Design

A within-participants experimental design was adopted with all of the participants undertaking all 64 of the experimental trials in a random order. There were two dependent variables: The taste word chosen for each color stimulus and the time taken by the participant to decide on that taste word.

Procedure

The participants had to select the taste word that they felt best represented the color stimulus; the position of the response words, see Figure 2, was randomized across participants. The selection was made by pressing one of four response keys that labeled the responses (o, k, z, or x). The next trial commenced after a 100-ms pause. There were five practice trials before the experiment commenced; the stimuli shown during the practice trials were selected at random from those shown during the actual study. After completing all the trials, the participants were debriefed as to the nature of the study.

Results and Discussion

The number of participants selecting each taste for each given color stimulus is shown in Figure 3. Descriptively, certain of the stimuli would appear to be linked by participants with a given taste much more often than the others. So, for example, the single black color patch (or, in other words, black with a black border) would seem to be the stimulus that is most strongly linked with bitterness. Meanwhile, the green stimulus with a red border seems most strongly linked with sourness (68 individuals associated it such); note that the inverse of this patch, the red stimulus with a green border, did not so strongly link with sour (50 such individuals), which relates to the topic of discussion next.

As predicted, there was evidence for our hypothesis that patches consisting of the same two colors (e.g., blue and yellow), but, with one patch having a particular color as the foreground (blueYellow), while the other patch had that color as the border (yellowBlue) would not necessarily have the same pattern of taste associations. As shown in Figure 4, which shows the degree of difference between pairs of such patches, 39.3% of such patches statistically differed in terms of their pattern of association with the basic tastes. For example, the pattern of association strength for blueYellow for bitter, salty, sour, sweet (respective scores were 21, 36, 32, 12) differed significantly from that of the yellowBlue patch (16, 16, 46, 22), χ²(3) = 13.21, p = .0043, p_crit = .0045, w = .26.

Figure 4. — A color matrix plot depicting the extent to which color patches of X foreground color and Y border color differ from those of Y foreground and X border colors, differ in terms of patch to taste association differences (2 × 2 χ² test value) to the four basic tastes. The horizontal distance between each XY pair reflects this difference. For example, at the top of the figure are the whiteBlack and blackWhite patches which are the most horizontally separated, and thus, most differing in terms of how participants assigned each patch to each of the basic tastes. Space permitting, the actual percentage tastes-word associations to each word have been plotted between each XY pair, by means of accumulative bar plots (the boxes of ascending lightness in each bar reflect, respectively, the tastes of bitter, salty, sour, sweet); the chart closest to a given patch reflects that patch’s % taste associations. XY pairs consisting of more than one color that statistically differ in terms of their degree of difference have a white star inserted directly between them. Note that the eight single color patches at the bottom of the figure have been added for completeness, and also to make available to the reader, how each was assigned to each of the individual colors.

By means of a simulation (see Appendix A for the code; to run the simulation online visit http://try.haxe.org/#5F5d1), as shown in Figure 5, certain of the stimuli were selected more frequently for a given taste than expected by chance (stimuli falling in the darker gray box in each segment fit this criteria). As noted earlier, for bitter, the black stimulus with a black border is indeed shown to be the most strongly linked stimulus. Furthermore, as noted, the green stimulus with a red border most often links with sour. Indeed, this greenRed stimulus links more strongly with sour than either of its constituent colors (green and red)—a star has been suffixed to the right of each stimulus that has this property. Given the small degree of difference, however, we are not so sure that we would observe the same small differences if we replicated the study, given the sample size of Experiment 1.

In terms of reaction times (RTs), by comparing confidence intervals, there were a range of two-color patches for each taste that did not take any longer to assign relative to the equivalent color patches of their constituent colors (see Figure 6). For example, referring to the green stimulus with a red border, it is worth noting that our participants did not take any longer to assign it as “sour” as compared with the purely green (or purely red) stimulus (the RT data are plotted in Figure 6). This might suggest that the green-red stimulus better “portrays” sourness than either of its constituents (and indeed, any other) colors. Note that for sweet, the pink with purple background stimulus, and for salty the white with blue background stimulus, are also both better able to portray the tastes they were most strongly associated with.

Figure 6. — Time taken to assign a taste to a given color patch (represented via the bars) and frequency with which each patch was assigned as that taste (represented via the squares). Error bars represent the 95% CIs (note that the size of the error bars increases from left to right as the number of data points contributing to each bar decreases; this reflects the fact that the bars have been rank ordered according to frequency). Darker gray bars were those that were chosen at a level greater than that expected by chance (n ≥ 45; as defined in Figure 7). We have also included in the figure color patches that consisted only of a single color, when that given color was present in patches that *were* chosen by participants at a level shown greater than that expected by chance.

Thus, as hypothesized, some color patches were found to be more consistently linked with sweet (pinkPurple, purpleWhite), salty (whiteBlue), and sour (greenRed, redYellow), than their constituent colors, and at a level that was significantly greater than that expected by chance. As there was no difference in terms of RTs between color patches of a single color as compared with those double-colored (as the confidence intervals overlapped), it can be concluded that participants found it just as easy to link single colors with tastes as to link multicolor patches with tastes. Note that this was not the case in Woods and Spence’s (2016) study, where it took participants more than twice as long to assign a taste to side-by-side color pairs, than it did to single color patches.

We were curious to see whether the differences between the top five associated patches for each color were statistically discriminable with our sample size, by means of a power analysis (via G*Power, version 3.1.9.2, Faul, Erdfelder, Buchner, & Lang, 2009). One issue here was that the data for each taste were not independent, with, for example, a high number of assignments for a patch as sweet, salty, and sour leading to a low number of assignments as bitter. Although thus only approximate, and potentially influenced by how a patch was assigned to each of the other tastes, we decided to use the ratio of a given taste assignment for two patches (e.g., whiteBlue and whiteBlack were assigned as sweet 59 and 47 times, respectively, or .56 and .44; n.b. we chose these patches as out of all tastes, they differed most in terms of magnitude) to calculate that the required sample size was 1,015 data points (w = .11, α = .5, 1−β = .95; we currently have 106 data points assigned to these patches). About 10 times as many participants were found to be needed to reliably test for this effect. It stands to reason, then, that given the current sample size, there was no significant difference between whiteBlue and whiteBlack patches for saltiness, χ²(1) = 1.36, p = .24, w = .11.²

The relationship between the time taken to assign each color to a taste and the number of times this assignment was chosen are plotted in Figure 7. Retrospectively, we split the data according to whether the number of assignments expected was above chance levels (≥45) or below this threshold (p < .001); it stands to reason that only when there were clear associations between color and tastes (i.e., above the threshold level), would one expect a relationship between RT and strength of association. Below the threshold, however, color-taste assignments would be more likely due to chance, and we would expect both RT and association strength to be rather random and unlikely correlate. When above the critical threshold, the dependent variables did indeed correlate significantly, r_pb = −.57, N = 39, p = .0033, 95% CI [−.25, −.77]. Below the threshold, there was also a significant, albeit much weaker, relationship r_pb = .14, N = 217, p = .0057, 95% CI [−.0043, .2722].

Experiment 2

In Experiment 1, an order of magnitude more participants would have been needed to reliably distinguish between the top ranking color patches for each taste. The purpose of our second study was to see whether we could replicate the results of Experiment 1, but using a paradigm that we thought should be more sensitive to differences of association.