Color as an Indicator of Implicit Associations, and as Visual Rhetoric

One way to measure implicit attitudes is through the colors used to represent groups. A growing body of research indicates that colors can carry implicit meaning (Baldwin & Meunier, 1999; Elliot & Maier, 2007; Meier, Robinson, & Clore, 2004; Moller, Elliot, & Maier, 2010). Color-specific associations appear to happen very early, as young children have up to nine color-emotion pairs for which the emotion meaning can be linked to a specific color (Birren, 1978; Boyatzis & Varghese, 1994; Sharpe, 1974). Once formed, color-meaning associations can operate without conscious intention or awareness (Bargh, 1990; Elliot, Maier, Moller, Friedman, & Meinhardt, 2007; Maier, Elliot, & Lichtenfeld, 2008). For example, college students asked to complete a simple word categorization task (a modified Stroop task) responded more quickly when categorizing color-emotion matches (Meier, Robinson, & Clore, 2004). Negatively valenced words were categorized more quickly when they appeared in black, relative to white, indicating an implicit association (i.e., black=bad). In another experiment, referees were more likely to call penalties on a team with black (vs. white) uniforms in the same ambiguous game situation without conscious knowledge of the bias (Frank & Gilovich, 1988).

For the present investigation, we focused on the implicit meaning of the color red. At a physiological level, the perception of red initiates a high galvanic skin response, and is thought to elicit anxiety (Jacobs & Seuss, 1975; Weller & Livingston, 1988; Wilson, 1966). Red is often used as a warning of danger, as in stop signs, traffic lights, or most recently the U.S. Department of Homeland Security's recently abandoned color-coded system for conveying terrorist threat level. Further negative connotations to red can be observed in diverse phrases, such as in the red, seeing red, red tape, red herring, red faced, code red, red flag, and red-handed (e.g., Elliot & Maier, 2007; Elliot et al., 2007; Pryke, Andersson, Lawes, & Piper, 2001; Setchell & Wickings, 2005).

Elliot and colleagues (Elliot & Maier, 2007; Elliot, Maier, Binser, Friedman, & Pekrun, 2009; Elliot et al., 2007; Moller, Elliot, & Maier, 2010) have focused on the implicit associations that exist between red and failure. Elliot and colleagues (2007) demonstrated that brief exposure to the color red caused participants to solve fewer anagrams. When given an anagram test with a red cover, participants knocked fewer times to indicate they were ready for the test, and exhibited a stronger tendency to lean away from the test when it was administered (Elliot & Maier, 2007). The authors interpret these findings as an indication that perceiving the color red primes thoughts related to failure, thereby setting off an avoidant orientation. Recently, Moller and colleagues (2010) demonstrated even more directly the existence of an implicit association between red and failure using a modified Stroop task. In two studies, undergraduate participants were asked to categorize words as "success-related" or "failure-related," and to do so as quickly and accurately as possible by pressing buttons on a computer keyboard. Across both studies, when words related to failure appeared in red, the words were categorized by participants more quickly relative to when those same words appeared in green or the achromatic color white. In other words, when failure-related words were printed in red, categorization was facilitated. By contrast, when words related to success appeared in green, the words were categorized by participants more quickly relative to when those same words appeared in red or white.

Given these findings, it is not surprising that color is a powerful form of visual rhetoric (Garber & Hyatt, 2003). We are able to easily recognize chromatic arguments. Red, as a rhetorical device can be positive or negative based on the context, but is commonly used to indicate danger or something to which we should be alerted, possibly because of metonymical associations (e.g., red with blood; Caviano & López, 2010). Regardless, it is thought that the culture of the audience is reflected in the messages addressed to it. In particular, at places where data (charts, graphs) have to be processed rapidly (e.g., academic conferences), data presentations must facilitate an audience's comprehension of data, and must follow conventions that are recognizable and carry meaning to the audience (Kostelnik, 2007), whether implicit or explicit. Therefore, we believe that an academic conference would be an appropriate arena to examine the use of color to represent labels associated with autism.

Purpose of the Study

The purpose of this study was to examine the colors used to represent autism and related group labels in poster graphs at a major international autism research conference. Based on the premises that red represents negative concepts (such as failure or danger) in situations where evaluation of competence is salient and the notion that there is an implicit negative societal association with the label autism, we predicted that red would be the most commonly used color to represent group labels that are typically defined as being an "Autism Spectrum Disorder" (ASD), including Autism, Asperger syndrome, and Pervasive Developmental Disorder - Not Otherwise Specified (PDD-NOS). We also predicted that the use of red to represent the groups with an autism spectrum label would be significantly more frequent than the use of red to represent groups that are not considered to be part of the autism spectrum. Third, we predicted that the use of red would be greater for groups labeled specifically as having autism than for groups labeled with qualifiers such as "high-functioning autism" or "autism spectrum disorders." Finally, we predicted that the use of red would be less frequent in groups labeled as Asperger syndrome or PDD-NOS relative to groups whose label explicitly included the word autism.

Graph Selection

The 108 posters, selected at random from a total of 657 posters at the conference, contained 115 graphs. Posters came from 13 different countries, although the United States, United Kingdom, and France accounted for 75% of the posters selected. Graphs were coded if they depicted more than one group (e.g., autism vs. comparison group) and/or more than one condition (e.g., time 1 vs. time 2) where comparisons could be made. If a poster contained more than one graph, the first graph was coded, and subsequent graphs were coded only if they differed from the original graph in terms of color scheme or groups. For example, if a poster that had two groups (e.g., autism vs. comparison group) and several graphs that all represented the autism group as blue and the comparison group as orange, only the first graph was coded. If, however, the second graph used green for autism and brown for the comparison group, or introduced groups other than autism versus the original comparison group, then this second graph was also coded.

From the randomly selected posters, there were several types of graphs that the coders encountered (see Table 1). First, there were graphs where colors or shading specifically differentiated between a group with an ASD label and comparison groups (Type 1 — ASD vs. comparison group, n=36 graphs). In other words, the bar or line representing the ASD group was one color, and the bar or line representing the comparison group was a different color. For these graphs, an ASD group was represented by a color 30 times and comparison groups were represented by a color 45 times (i.e., sometimes there was more than one comparison group in a graph). Type 1 graphs allowed us to make direct comparisons on color choices for ASD and comparison groups when used in the same graph.

Note: Studies using an animal model, such as knockout mice, were considered Type 1 if the graph explicitly had an ASD or related label (e.g., "autism-like") in the graph. Otherwise, these graphs were considered to be Type 3. ASD = Autism Spectrum Disorder.

There were two types of graphs that did not use colors to differentiate between an ASD and a comparison group (Type 2 and Type 3, see Table 1). These graphs were coded in order to have representative samples of color distribution in graphs where a color did not explicitly differentiate an ASD group from another group. One of these graph types used color to signify different within-subject conditions rather than a particular group (Type 2- Between Conditions, n=67 graphs). In other words, a condition (e.g. Time 1) was represented by one color regardless of group, and a second condition (e.g. Time 2) was highlighted using a different color. In Type 2 graphs, there were 244 conditions that were represented by a color. In another type of graph, no groups contained an ASD label at all (Type 3 — No Autism, n=12). A common example of this would be a graph that represents the provision of services in different areas of a city, county, or state. Whereas Type 2 graphs included ASD groups but color did not differentiate them from other conditions, Type 3 graphs had no mention of an ASD group in the graph. Overall, there were 30 groups or conditions that were represented by a color in the Type 3 graphs.

Coding

Each rater was given a notebook that contained the list of posters they were assigned to code. They were unaware of which posters would be used for testing reliability. For each poster, the raters collected the primary affiliation of the first author, the colors used in the graph, and group labels associated with each color. Raters put each of the colors from the graph into one of the following categories: red, orange, yellow, green, blue, purple, gray, brown, black, and white. Although additional colors could have been included, pilot testing of the coding system on prints of posters presented at previous conferences revealed that as color categories became more fine-tuned, inter-rater reliability decreased (e.g., what one rater might code as fuchsia, another might code as magenta). Therefore, this specific subset of color categories was used to improve reliability.

Ratings were forced choice; in other words, if a color appeared to be in between two of the listed categories, raters were required to choose just one color label. To guide coders in choosing a label for ambiguous colors, raters were provided with a color "key," which depicted the labels that were most commonly chosen to correspond to each color during pilot testing. For forced choice responses, the raters achieved good reliability (unweighted ĸ=.70).

Descriptive Characteristics

Across all graphs, regardless of group or condition, blue was the most common color used, followed by red, green, purple, and yellow (see Figure 1 for overall distribution). Red and green were the second and third most commonly used colors, despite the suggestion not to use these colors given in the pre-conference instructions to authors. It is important to note that, in the instructions to presenters that were emailed to the poster presenters several weeks before the conference, presenters were discouraged from using the colors red and green in poster graphs in order to be accommodating for attendees who were colorblind. Thus, it is possible that the use of these colors was less frequent overall because of this stipulation. It should also be noted that in our coding system, both Excel default colors (i.e., the two colors that would appear on a graph if no colors were manually selected by the author) would be coded as a shade of purple. Because purple was only used to represent 12.1% of the groups, the use of Excel defaults was not a common occurrence in these graphs.

Figure 1. Distribution of colors used in a random sampling of poster graphs at an international autism conference. Numbers represent the percentage of the total number of graphs for which a color was used to highlight a group or condition in a line, bar, or section of the graph. Red represented 18.4% of poster graphs, orange 4.8%, yellow 7.3%, green 16.9%, blue 30.5%, purple 12.1%, black 6.5%, white 1.7%, grey 1.5%, and brown 0.4%.

Main Analyses

First, we examined the Type 1 graphs to compare the distribution of colors used to represent ASD vs. comparison groups (see Figure 2). The overall distribution of colors used for ASD (n=30) and comparison groups (n=45) were significantly different, Χ²=19.28, p<.05. We then used a 2x2 Chi-square to directly compare the use of red to that of all other colors combined in each group. Red was used to represent an ASD group significantly more often than comparison groups, Χ²=10.30, p<.01. As predicted, red was the most commonly used color to represent groups with an ASD label1 (50%). For comparison groups, however, green was the most commonly used color, followed by blue and purple. This is striking, given that across all graphs, blue is the most commonly used color (Figure 2). Thus, poster authors used a significantly different color distribution for posters comparing an ASD group to a comparison group than for posters in general, and red was used more frequently to represent an ASD group than comparison groups.

Figure 2. Color distribution used in graphs when the purpose or designation of the color is considered. In the sample graph color could specifically designate a group labeled with an autism spectrum disorder (Type 1), a comparison group that does not have an autism spectrum disorder diagnosis (Type 1), comparisons between experimental conditions rather than groups (Type 2), or when graphs are not about autism (Type 3). ASD=Autism Spectrum Disorder.

Next, we compared the colors used to represent ASD groups in the Type 1 graphs to the colors used to depict contrast conditions in the Type 2 graphs (n=244, see Figure 2). The Type 2 graphs were thought to be more representative of the normal distribution of colors that would be used in graphs in general, because they do not differentiate between groups. Overall, the distributions were significantly different, Χ²=20.34, p<.05. Moreover, a group with an ASD label was significantly more likely to be represented by red than were conditions in Type 2 graphs, Χ²=16.16, p<.001. When colors signified conditions (as opposed to groups), blue was almost two times more likely to be used than any other color.

The color distribution was then explored within Type 3 graphs, which made no mention of an ASD group (n=30, see Figure 2). The color distribution of Type 3 (no ASD) groups was significantly different from the color distribution within groups with an ASD label in Type 1 graphs, Χ²=16.60, p<.05. An ASD group was significantly more likely to be represented by red than were groups in Type 3 graphs, Χ²=11.43, p<.01. In Type 3 graphs, blue was the most commonly used color, followed by green and purple. Red was used relatively infrequently in these graphs (10%).

Comparisons within the Spectrum Labels

Figure 3 shows the percentage of times red was used to represent a group or condition when all ASD diagnostic labels are considered separately. For these analyses, we split the ASD groups as represented in Type 1 graphs into three groups (Autism, n=14; ASD/HFA, n=10; Asperger syndrome, n=5) based on the explicit label used by the poster author in the graph. We predicted that adding qualifiers (such as "high-functioning" or "spectrum disorder") would reduce the frequency that red was used, and that Asperger syndrome would have the lowest frequency of red use. As predicted, red was most commonly used to represent autism groups (64%), followed by HFA/ASD groups (50%), and then by Asperger syndrome groups (20%).

Figure 3. Percentage of times red was used to represent a particular group or contrast condition when subcategories of autism spectrum disorder labels are considered. Type 1 graphs are split into four (autism, HFA/ASD, Asperger Syndrome, comparison groups). These categories were determined by the explicit terminology used by the authors to describe the group that was being represented in the graph. ASD = Autism Spectrum Disorder, HFA = High-Functioning Autism.

Limitations and Future Directions of Research

An alternate interpretation of our findings is that in the context of research presentations /posters, red could represent the group that a researcher wants to highlight, rather than an implicit perception related to that group. At an autism conference, it would follow that an ASD group would be more likely to be highlighted in graphs to draw attention to the important data. Still, this hypothesis does not account well for the differences observed between various autism spectrum disorder labels in this study. It also does not explain why the use of red was significantly less frequent in graphs that did not include groups with an ASD label but presumably still had a condition of interest to highlight. Still, future research using this methodology should attempt to replicate this pattern at a venue where a specific disorder is not the focus of the conference, such as at a general child development conference.

A second limitation of this study was the small number of graphs representing potential labels of interest, making it difficult to draw definitive conclusions about the distribution of colors that would be used for these populations. There were only five graphs that specified Asperger syndrome as a group label, and none that explicitly used PDD-NOS. Also, there were very few instances of comparison groups with a specific disability/disorder label that was not considered to be on the autism spectrum (n=13), although red was never used to represent these groups in the instances when they did occur. Future studies should utilize stratification in the random selection process in order to ensure that all ASD labels are meaningfully represented in terms of sample size.

Future research on this topic should consider both the implicit nature of negative associations, and societal factors that might be perpetuating these negative conceptions. Contemporary reaction time study designs measure differences in participants' responses to stimuli on the order of milliseconds, thereby providing a window into implicit or nonconscious attitudes that may not be explicitly articulated or measured via self-report (Gawronski & Payne, 2010). Tools like these may be used in the future to explore whether the effect demonstrated in this study is, in fact, nonconscious. For example, several paradigms have been developed that demonstrate implicit associations between words and colors can increase reaction time. On a broader level, it would be fruitful to explore the relationship of "awareness" campaigns to negative conceptions. It is possible that the success of these campaigns at raising awareness, but not knowledge of autism has contributed to the perpetuation of negative associations with autism.

Implications for Research and Practice

One message to take from this study is that there may be negative social conceptions associated with the label autism that operate at an implicit level, outside of conscious awareness. Importantly, even if we assume no implicit prejudice on the part of the researcher, it should be seriously considered that colors used to represent a group in a graph may implicitly communicate unintended meaning to a presenter's audience. There may be inadvertent consequences for using red to represent ASDs or any specific group with a disability in posters, presentations, on book covers, and in other media. That is, using red may contribute to inadvertently perpetuating negative implicit attitudes towards individuals with the autism label.

Additionally, it is possible that different representations of disorders in the autism spectrum could lead the children to be perceived differently in research, educational, or other settings. Research has clearly shown that labels affect the interpretation of ambiguous behavior (Hart et al., 1997; Hinshaw & Cicchetti, 2000; Lefley, 1992; Woods et al., 1998). If the findings from this study are interpreted as an indication of implicit or nonconscious differences in the representation and perception of specific ASD labels within the research community, then subjective measurements of behavior, especially if they are collected or coded during the course of a testing session, are potentially susceptible to such implicit biases. Moreover, implicit reactions or biases associated with a diagnosis could very likely effect how a teacher or therapist behaves around a child with an autism spectrum label. Exploring these issues warrants significant empirical attention in the future.