Color-Shape Associations Revealed with Implicit Association Tests

Na Chen; Kanji Tanaka; Katsumi Watanabe

doi:10.1371/journal.pone.0116954

Abstract

Kandinsky proposed a correspondence theory that suggests associations between specific colors and shapes (i.e., circle-blue, square-red, triangle-yellow). Makin and Wuerger tested the theory using the Implicit Association Test (IAT) and did not find clear evidence for Kandinsky’s color-shape associations among British participants. In the present study, we first replicated the previous study among Japanese participants and found similar results to those of Makin and Wuerger, showing little support for Kandinsky’s theory. In the subsequent experiment, we tested another set of color-shape associations that had been revealed by using an explicit matching method (circle-red, square-blue, triangle-yellow) in Japanese participants. The IAT tests showed that response times were significantly faster when circle-red, square-blue, and triangle-yellow combinations were mapped onto the same response key, rather than different key combinations, indicating that these color-shape combinations were encoded. These results provide the first empirical evidence that color-shape associations can be measured by indirect behavioral methods, and in particular, Japanese people’s color-shape associations (circle-red, square-blue, triangle-yellow) can be observed by both direct and indirect experimental methods.

Citation: Chen N, Tanaka K, Watanabe K (2015) Color-Shape Associations Revealed with Implicit Association Tests. PLoS ONE 10(1): e0116954. https://doi.org/10.1371/journal.pone.0116954

Academic Editor: Michael J. Proulx, University of Bath, UNITED KINGDOM

Received: August 27, 2014; Accepted: December 16, 2014; Published: January 27, 2015

Copyright: © 2015 Chen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited

Data Availability: All relevant data are within the paper and Supporting Information files.

Funding: This study was supported by Grants-in-Aid for Scientific Research (24300279) from the Ministry of Education, Culture, Sports, Science and Technology. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: KW is a member of PLOS ONE editorial board. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials.

Introduction

Kandinsky, a renowned abstract painter, proposed a correspondence theory suggesting an association between primary colors and shapes, according to which a circle is matched with the color blue, a square is matched with the color red, and a triangle is matched with the color yellow. He argued that the correspondence between colors and shapes was mediated by an inherent relationship between the colors and angles of the shapes [1–2]. However, the correspondence theory has evoked opposition since its proposal. Indeed, several studies have failed to replicate Kandinsky’s correspondence theory [3–8]. For example, Jacobsen used a modified Kandinsky questionnaire and found color-shape combinations of circle-yellow, square-blue, and triangle-red in German samples, suggesting that color-shape associations were based on everyday knowledge and experience [3–4]. Albertazzi et al. used a matching task to investigate color-shape associations in Italian participants, and found that both circle and square were matched with red, while a triangle was matched with yellow. They argued that color-shape associations might be determined by semantic information such as the “warmth” and “lightness” [5]. Chen et al. reported that Japanese participants showed color-shape associations similar to those of Italian participants (Japanese color-shape association: circle-red, square-blue, and triangle-yellow) [6], suggesting that color-shape associations might stem from shared semantic information between colors and shapes (e.g., warm/cold) as Albertazzi et al. These studies indicate that there exist non-random associations between colors and shapes, and these color-shape associations might be influenced by the semantic information associated with visual features and people’s learning experience.

In addition to the direct matching tasks in the aforementioned studies, some studies have adopted indirect behavioral measures to investigate color-shape associations. Moreover, studies using indirect behavioral methods have also failed to find clear evidence for Kandinsky’s color-shape associations [9–11]. Kharkhurin revisited Kandinsky’s color-shape associations in two experiments: one used with a priming technique and the other used a recognition test. He first tested whether an earlier-presented color stimulus influenced the discrimination of a later-presented shape stimulus and hypothesized that the response time (RT) would be faster in congruent conditions in which the priming color was associated with the shape. Then, in the second experiment, he measured RTs in a discrimination task using a shape filled with various colors and hypothesized that discrimination times would be faster for shapes filled with the associated color. However, both experiments failed to reveal any acceleration of RTs in the congruent conditions in Kandinsky’s correspondence theory [9]. Makin and Wuerger used implicit association tests (IAT) to examine Kandinsky’s correspondence theory. They did not find clear evidence of correspondence, but found marginal effects in only one combination (square-red and triangle-yellow associations) among the three combinations they tested (the other two combinations: circle-blue and square-red, circle-blue and triangle-yellow associations) [11].

Taken together, studies have shown little support for Kandinsky’s color-shape associations using either direct matching or indirect behavioral experimental methods [3–11]. One possible reason might be that Kandinsky’s correspondence theory is not universal or simply incorrect. Previous studies using explicit matching task have shown that there exist other color-shape associations [5–6]. Then, the question is: Is it possible to demonstrate color-shape associations in the indirect behavioral measures? One possibility is that these associations are constructed only through subjective introspections, and may not be measured through indirect behavioral performance reflecting quick judgments. But, the other possibility that the indirect behavioral measures have not been tested for color-shape associations with empirical supports remains [6, 11].

The present study aimed to examine whether two forms of color-shape associations reported in direct matching tasks in the previous studies could be observed by an indirect behavioral method: one referred Kandinsky’s correspondence theory [11] and the other referred to Japanese color-shape association [6]. For this, we adopted the IAT, which is known to be a useful tool to reveal people’s unconscious associations between dimensions by response time [12–13]. In the IAT experiment, participants have to classify four stimulus categories by pressing two different keys. Participants are instructed to press one key rapidly and accurately for one pair of visual stimuli and to press the other key for the other pair of visual stimuli. If the two stimuli assigned to the same key are associated (congruent condition), the RT should theoretically be shorter than in the condition in which two stimuli assigned to the same key are unassociated (incongruent conditions). Hence, the IAT reveals associations at the level of response selection. It has been widely used to assess the strength of associations between different dimensions [14–16]. In the present study, we used multiple IAT tasks to explore the associations between colors and shapes [11, 16]. We first replicated the study of Makin and Wuerger [11] to examine whether Kandinsky’s correspondence theory could be observed in Japanese participants. Having negative results in the first experiment, in the second experiment, we explored a different pattern of color-shape associations, reflecting our previous findings of color-shape associations among Japanese participants [6].

Experiment 1

Method

Participants.

Thirty-eight Japanese college students took part (24 males, mean age = 21.5, SD = 2.1). All participants had normal or corrected-to-normal visual acuity and normal motor functions, and were naïve to the purpose of this study. This experiment, as well as the subsequent experiments, were approved by the institutional review board (IRB) of The University of Tokyo, and conducted in accordance with the ethical standards in the 1964 Declaration of Helsinki. Written informed consent was obtained from all participants in advance.

Apparatus and stimuli.

Stimuli were presented at 60cm on a 1280 × 1024 pixel CRT monitor, with a refresh rate of 60 Hz. E-Prime 2.0 software (Psychology Software Tools, Inc.) was used to present the stimuli and collect the data. The three shape stimuli were all white line drawings (with a width of 26mm, 0.03° visual angle) on a black background (8 cd/m²). The circle was 4.8° in diameter, the square was 4.8° (in height) × 4.8° (in width), and the triangle was 4.8° (in height) × 5.8° (in width). The three shape stimuli were always presented in the upward orientation. Color patches were ~9.5° wide (Gaussian-modulated). The colored Gaussian patches were measured using PR-655 (Photo Research, Chatsworth, CA, USA) and each color was consecutively measured 10 times and averaged. The color information was as follows: yellow: L* = 46.44, a* = -2.57, b* = 53.73; red: L* = 20.99, a* = 37.82, b* = 27.66; blue: L* = 27.79, a* = -14.58, b* = -15.12. Example stimuli are shown in Fig. 1.

Download:

Figure 1. Example stimuli used in the experiment.

The words of shapes or colors appeared on the left and right sides of the screen. For example, “赤” (red), “青” (blue), “円” (circle), and “四角” (square) were used as cue words to remind participants of the key press.

https://doi.org/10.1371/journal.pone.0116954.g001

Procedure.

The experiment was carried out in a laboratory with dimmed lighting conditions (1 lux on the wall). The structure of the IAT was based on the study conducted by Makin and Wuerger [11]. The experiment was composed of three IAT tasks, each consisting of 8 blocks (Tables 1 and 2). During the task, participants were asked to discriminate visual stimulus of color or shape by pressing two keys (i.e., left key “z” and right key “m”) in the keyboard using the two index fingers. In the first block (Block 1), participants were trained to discriminate a shape stimulus by pressing a key (for example, the left key for circle and the right key for square in IAT 1; Table 1). In the second block (Block 2), participants were trained to discriminate a color stimulus (e.g., the left key for blue, the right key for red) by pressing a key. In the third and fourth blocks (Blocks 3 and 4), color and shape discriminations were combined, and responses were mapped according to Kandinsky’s color-shape associations (e.g., the left key for circle or blue, the right key for square or red; congruent blocks). In the following two training blocks (Blocks 5 and 6), participants were instructed to learn a shape-key mapping opposite to that of the previous blocks (e.g., left key for square, right key for circle). In the last two blocks (Blocks 7 and 8), the response mapping was the opposite of Kandinsky’s color-shape associations (e.g., the left key for square or blue, the right key for circle or red; incongruent blocks).

Download:

Table 1. Sequence of blocks and response mappings (example from IAT 1).

https://doi.org/10.1371/journal.pone.0116954.t001

Download:

Table 2. The response mappings in the IATs of Experiments 1 and 2.

https://doi.org/10.1371/journal.pone.0116954.t002

In each block, the stimuli were presented in a novel random order for each participant. Cue words shown on the top two sides of the screen were used to remind participants which side of key should be pressed to report their answers. For example, in the first training block of IAT 1, the words “円” (circle) and “四角” (square) appeared on the top left and top right sides of the screen according to the response mapping. When participants pressed the wrong button, the word “Error” appeared centrally in red, and a beep sounded at the same time. Before each block, participants were appropriately informed of the task, stimuli, response mapping, and cue words. For instance, the instruction for block 1 in IAT 1 was as follows. “You are going to take part in a test measures your response time in discriminating visual stimulus for colors and shapes. In this block, when you see a circle (○), please press key “z” with your left index finger. when you see a square (□), please press key “m” with your right index finger. The cue words of “円” (circle) and “四角” (square) will be appeared at the top left or right sides of the screen to remind you which side of key to press. Please press the key as accurately and fast as possible. When you press the wrong key, a red word “ERROR” will appear and a beep will sound.” For another example as in the block 3, the main instruction would be changed into “when you see a circle (○) or color blue, please press key “z” with your left index finger; when you see a square (□) or color red, please press key “m” with your right index finger. The cue words of “円/青” (circle/blue) and “四角/赤” (square/red) will be appeared at the top left or right side of the screen to remind you which side of key to press.”

The order of the congruent and incongruent blocks was counterbalanced between participants; half of the participants were presented the congruent blocks first, and the other half of participants were presented the incongruent blocks first [11]. Each participant completed three types of IAT tasks (response mappings are shown in Table 2). The order of the three IAT tasks was also counterbalanced across participants. Each IAT task took about 7 minutes, and participants rested 3 minutes between tasks. The RTs and the number of error trials were collected for data analysis. The whole experiment took about 30 minutes.

At the end of the experiment, participants were instructed to complete a paper questionnaire. On the questionnaire, three shapes (i.e., circle, square, and triangle) and three colors (i.e., red, blue, and yellow) were presented in two rows, with the locations counterbalanced across participants. Then, participants were asked to intuitively connect the matched colors and shapes by drawing three lines, resulting in no overlapped colors or shapes and six types of color-shape associations (e.g., circle-red, square-blue, triangle-yellow; Table 3).

Download:

Table 3. Explicit questionnaire results in Experiment 1 and Experiment 2.

https://doi.org/10.1371/journal.pone.0116954.t003

Analysis.

We adopted the D score measure, which was widely used in the IAT studies [11, 12, 13, 16, 17]. We computed the D score of every IAT task for each participant, with reference to the study of Makin and Wuerger [11]. The step-by-step procedure for computing D scores was as follows:

1). Include trials from block 3, 4, 7, 8 in analysis.
2). Compute mean RT and standard deviation for each individual in each IAT and each congruent and incongruent condition separately.
3). Compute boundary values by adding or subtracting 2 SDs to the mean RT.
4). Include the trials [mean − 2 SD, mean + 2 SD].
5). Compute the mean RTs from block 3, 4, 7, 8.
6). Compute one pooled SD for trials in blocks 3 and 7 as SD_(3,7), another one for blocks 4 and 8 as SD_(4,8).
7). Compute two difference RTs in block 3 and 7 (RT 7 − RT 3) and block 4 and 8 (RT 8 − RT 4).
8). Divide each difference of RTs by its associated pooled SD from step 6, producing two D scores (e.g., D score_(3,7) = (RT 7 − RT 3) / SD_(3,7)).
9). Compute the average of two D scores (i.e., D score_(3,7) and D score_(4,8) in step 8), generate a final D score for each participant in each IAT.

This procedure was repeated for the three IAT tasks; therefore, each participant had three D scores for the three IATs. The D score represents the difference between the means of congruent and incongruent blocks, as a function of the standard deviation of the distribution. For participants who did congruent first, a positive D score value indicates support for the proposed hypothesis, and a negative value indicates that the participant showed the opposite associations to the hypothesis.

Results and Discussion

We excluded error trials (3.6%) and trials with RTs longer than 2000 ms (0.8%) from data analysis. We calculated D-scores and response times in congruent and incongruent blocks from each of the three IAT tasks. All individual-level data including participant information, D score, RT and error rate are provided in S1 Table.

D scores.

One-sample t-tests with Bonferroni corrections (as there were six tests, the adjusted p value was derived from p value of each t-test multiplied by six, and then compared with the significant level of 0.05) were performed to compare D scores to zero (Fig. 2a). In IAT 1, D scores were distributed in the left (i.e., unexpected) direction (IAT 1: t[37] = 3.42, adjusted p < .01; mean D score = -0.26). This result indicates an association pattern opposite to that of Kandinsky’s color-shape associations; our participants associated circle with red and square with blue. In IAT 2, D scores were distributed around zero (IAT 2: t[37] = 0.88, adjusted p > .99; mean D score = 0.06). Thus, participants did not show any specific associations in this type of color-shape combination (i.e., circle/triangle and blue/yellow). In IAT 3, D scores were distributed in the right (i.e., expected) direction (IAT 3: t[37] = 4.65, adjusted p < .01; mean D score = 0.38), which suggested a congruency effect of triangle-yellow and square-red associations (i.e., consistent with Kandinsky’s color-shape associations).

Download:

Figure 2. Results of D scores and mean RTs in Experiment 1.

(a) Distribution of D scores for the three IAT tasks with Kandinsky’s color-shape associations. Positive D scores indicate that trials in the left color-shape combinations were faster than those in the right. (b) Mean RTs in congruent and incongruent blocks for the three IAT tasks. Error bars represent the standard errors of the mean.

https://doi.org/10.1371/journal.pone.0116954.g002

Response time.

We calculated and compared mean RTs in congruent and incongruent conditions (Fig. 2b). Paired t-tests with Bonferroni corrections showed similar patterns in the results of D scores. In IAT 1, participants responded faster in the incongruent (512 ms) than in the congruent (563 ms) conditions (t[37] = 3.04, adjusted p <.05). In IAT 2, there was no significant difference between the congruent (513 ms) and incongruent (529 ms) conditions (t[37] = 1.46, adjusted p = .92). In IAT 3, participants responded faster in the congruent (491 ms) than in the incongruent (568 ms) conditions (t[37] = 4.02, adjusted p <.01).

Comparison with Makin and Wuerger [11].

Here, we discuss differences of the present results and Makin and Wuerger who run the same experiments with a different set of participants. In the present study, we recruited thirty-eight Japanese college students (24 males, mean age = 21.5, SD = 1.5) and in Makin and Wuerger, they recruited thirty-six British participants (age 18–45, 11 males) who were mostly undergraduate or postgraduate students at the university of Liverpool [11].

In the present IAT 1, the D scores were significantly shifted to the left direction (mean D score = -0.26), and the RTs in the incongruent blocks (mean RT: 512 ms) were significantly faster than those in the congruent blocks (563 ms). This indicates that most of participants associated circle with red and square with blue (i.e., opposite of Kandinsky’s color-shape associations). Makin and Wuerger [11] used the same set of color and shape stimuli (circle/square and red/blue in their IAT 2), and D scores in their study were distributed around zero (mean D score = -0.04), and there was no significant difference in RTs (t[35] = -1.502, p = 0.142, congruent = 630ms vs. incongruent = 601 ms). This discrepancy may need further investigations. Nevertheless, these results indicate that most people do not have circle-blue and square-red associations proposed by Kandinsky. In addition, the previous studies using direct matching task reported circle-red and square-blue associations, which was opposite to Kandinsky’s association [6, 7, 9].

In IAT 2, the D scores were distributed around zero (mean D score = 0.06) and RTs did not show any differences between the congruent (513 ms) and incongruent (529 ms) conditions. It is likely similar to the null result in Makin and Wuerger [11] (mean D score = -0.08; RT: 609 ms vs. 588 ms in the congruent and incongruent conditions, respectively, in their IAT 1). These results indicate that there was little congruency effect consistent with Kandinsky’s circle-blue and triangle-yellow associations.

In IAT 3, we observed that D scores were significantly shifted to the positive direction (mean D score = 0.38), and participants responded faster in the congruent (491 ms) than in the incongruent (568 ms) condition, suggesting a congruency effect of square-red and triangle-yellow associations. This result supports Kandinsky’s correspondence theory. Makin and Wuerger also observed a marginally significant difference in their IAT 3 for this color-shape combination; D scores were marginally shifted to the right (p = 0.051, mean D score = 0.13) and RTs in the congruent conditions (575 ms) were faster than in the incongruent conditions (621 ms) [11].

Explicit questionnaire.

The results of the explicit questionnaire were shown in Table 3. A chi-square goodness of fit test showed that our participants made some color-shape combinations more frequently than others (x² = 27.99, df = 5, p <.01). We can see that 14 out of 38 participants (36.84%) made color-shape associations consistent with Kandinsky’s theory (Label 2 of Table 3) and the same number of participants (14 out of 38 participants; 36.84%) chosen Japanese color-shape associations (Label 1 of Table 3). However, the participants who expressed color-shape associations consistent with Kandinsky’s theory may be influenced by the IAT task they had performed [3, 16]. In the present study, we run the IAT first and then the explicit questionnaire because our aim was to investigate whether the proposed color-shape associations could be observed by the IAT, resulting in that the explicit questionnaire was influenced by the priming effect of the IAT paradigm. However, if we run the explicit questionnaire first, and the IAT next, the performance in the IAT would be inevitably affected by the choices in the explicit questionnaire. As the main purpose of the present study was to examine whether color-shape associations could be measured by the IAT, and the explicit questionnaire was used to test whether Japanese color-shape associations [6] could be observed even under the priming effect of the IAT. Therefore, we run the experiments in the present order. In the present experimental paradigms, the congruent color-shape associations appeared twice while the incongruent color-shape associations appeared only once (e.g., red-square appeared in IAT 1 and 3 while blue-square and yellow-square appeared once in IAT 1 and 3, respectively); therefore, the difference of the number of responses might affect the answers of explicit questionnaire to some extent. Hence, here, we mainly discuss the results of participants who showed unmatched color-shape associations with Kandinsky’s theory in the explicit questionnaire (14 participants in the matched group and 24 participants in the unmatched group).

For participants who showed matched color-shape associations, one sample t-tests with Bonferroni correction revealed significant differences between the D scores and zero only in IAT 3 (IAT 1: t[13] = 2.07, adjusted p = .12, mean D score = -0.27; IAT 2: t[13] = 1.66, adjusted p = .32, mean D score = 0.17; IAT 3: t[13] = 3.58, adjusted p <.05, mean D score = 0.42). This indicates that participants who endorsed Kandinsky’s color-shape associations in the explicit matching task did not clearly show faster responses in the expected congruent conditions in the IAT. However, one possibility should be considered that some participants chose Kandinsky’s color-shape associations due to the imbalance of the responses while they might have other associations in fact. For this understanding, improvement of experimental paradigms is required in the future study.

For participants who showed unmatched color-shape associations, D scores revealed significant congruency effects in IAT 1 and IAT 3 (IAT 1: t[23] = 2.65, adjusted p <.05, mean D score = -0.26; IAT 2: t[23] = 0.01, adjusted p >.99, mean D score = -0.001; IAT 3: t[23] = 3.17, adjusted p <.05, mean D score = 0.34). The results of IAT 1 showed circle-red and square-blue associations and those of IAT 3 showed triangle-yellow and square-red. These faster responses might be caused by Japanese color-shape associations (circle-red, square-blue, and triangle-yellow; Label 1 in Table 3) because 14 out of 24 participants in the unmatched group (58.33%) chose the association.

Interim summary.

Overall, in Experiment 1, we replicated the study of Makin and Wuerger [11] in Japanese participants, which suggested little support for Kandinsky’s color-shape associations. In IAT 1, we observed circle-red and square-blue congruent association effects, and in IAT 3, we observed triangle-yellow and square-red congruent association effects. Interestingly, these color-shape associations have been observed in the previous studies using explicit matching tasks [5–6]. Albertazzi et al. asked Italian participants to choose the color that best matched a shape from a color wheel (40 hue colors from Natural Color System), and found that circle and square were associated with red, and a triangle was matched with yellow [5]. Chen et al. replicated the study in Japanese participants and found that circle was associated with red, square was associated with blue, and triangle was associated with yellow [6]. Indeed, in the explicit questionnaire, the circle-red, square-blue, triangle-yellow associations were prominent.

Experiment 2

In the previous study [6], Chen et al. used an explicit matching task and asked participants to choose a best matched hue color for a geometric shape from 40 color circles. The results showed that participants systematically associated shapes with colors. Chen et al. found that about 34% of participants matched circle with red colors, 33% of participants matched triangle with yellow colors, and 19% of participants matched square with blue colors. Therefore, we reason that the Japanese color-shape associations might be represented as circle-red, triangle-yellow, and square-blue. In Experiment 2, in order to examine whether there were any color-shape associations that could be consistently shown using the IAT, we examined the recently proposed color-shape associations found in Japanese participants (circle-red, square-blue, triangle-yellow; [6]), instead of Kandinsky’s color-shape associations.