2. Materials and Methods
2.1. Participants
One-hundred participants volunteered to take part in this study (age: 25.47 ± 2.68 years, 58% females, all of them speaking Italian as their native language). Half of them were randomly allocated to a version of the IAT with the three compatible blocks followed by the three incompatible ones. The other participants were assigned to a version with the opposite sequence. All subjects gave their informed consent for inclusion before they participated in the study. The study was conducted in accordance with the Declaration of Helsinki (1991), and the protocol was approved by the Ethics Committee of Sapienza University (Project Identification code: 1272, approved on the 5 July 2019).
Participants had normal or corrected-to-normal vision and were naïve to the purpose of the study.
Using G*Power [
27], we computed a sensitivity analysis to detect a significant effect of compatibility. Results showed that with our sample of 100 participants, considering α = 0.05 and 1 − β = 0.80, the minimum effect size that could be detected is η
2p = 0.18 (Cohen’s f = 0) corresponding to a medium effect size.
2.2. Procedure
Stimulus presentation and data collection were conducted online using the platform Testable (
www.testable.org; access date 9 December 2022) [
28]. All participants were first recruited in Rome (mostly at Sapienza University). Once volunteered to participate, each participant received by email the link to the experiment, with some instructions (e.g., the study could not run on a tablet or mobile phone; the experiment should be started only when sure to have enough time to complete it, in a quiet room, with no distractions). For each participant, start time and end time was recorded. All participants completed the experiment within 30 min. As the experiment started, each participant was first required to activate the full-screen mode to reduce all distractions (e.g., phone turned off), and to calibrate the screen to ensure that stimuli size was accordingly adjusted. At the end of the experiment, each participant received a completion code and was asked to send it back to the researchers.
The experiment consisted of 4 tasks administered in the following order: Ultimatum Game, IAT, explicit preferences ratings for stimuli used in the IAT, line bisection task. We presented the Ultimatum Game task at the beginning and the line bisection task at the end of the IAT task to avoid biases due to repetition of similar explicit choices.
2.3. Ultimatum Game
The Ultimatum Game task was performed as in Suleiman [
16]: participants were asked to choose how much of a total hypothetical amount of money (100€) they were willing to share with another imaginary participant that could refuse the offer if judged unfair, leading to a loss of money for both. Hence, they should opt for a fair division of money to reduce the risk of the other participant’s refusal.
2.4. IAT
The IAT consisted of 10 blocks of 20 trials each, followed by one block of 20 trials for the explicit preference task, in which participants rated stimuli pleasantness. Stimuli presentation was randomized across blocks, with the only constraint that, in the compatible and incompatible blocks, a word (either positive or negative) always followed an image (either golden ratio or random), and vice versa. In addition, half of participants completed the IAT with the block order reversed (incompatible block presented first, followed by the compatible block), and the practice blocks were adjusted accordingly to match the block order. Therefore, in one version of the IAT, participants first completed two practice blocks: in the first practice block, 10 golden-ratio-based (GR) and 10 random dot patterns were randomly presented, and participants were instructed to press the ‘a’ key for golden ratio dot patterns and the ‘l’ key for random dot patterns; in the second practice block, all positive and negative words were presented, and participants responded with the ‘a’ key for positive words and the ‘l’ key for negative words. Practice blocks were followed by three experimental blocks (e.g., compatible blocks), in which participants saw either dot patterns or words (interleaved between trials) and responded to golden ratio dot patterns and positive words using the practiced key associations. After the compatible blocks, participants completed two additional practice blocks to learn new key associations for the remaining 3 experimental blocks (e.g., incompatible blocks). Participants learned to press the ‘a’ key for random patterns and the ‘l’ key for golden ratio. Key assignment for positive and negative words remained the same. Following the practice trials, participants completed three incompatible blocks using the new practiced keys associations (see
Table 1).
Each trial of the IAT started with a central fixation cross for 500 ms, followed by either dot-pattern or word stimuli, which remained on screen until participant’s response. Cue words were displayed above each stimulus, indicating the response mapping for each trial. For the compatible blocks, the cues “regular” and “positive” were presented on top left of the screen, whereas the cues “irregular” and “negative” were displayed on top right (see
Figure 2 for an example of trials’ sequence).
In the incompatible blocks, the cues “irregular” and “positive” were presented on top left, and the cues “regular” and “negative” were displayed on top right. After each response, feedback was given informing participants if response was correct or wrong. Each block was preceded by written instructions, and between blocks, participants had the opportunity to take a short break. Finally, participants completed a block of 20 trials, in which they judged the aesthetic pleasantness of the dot-patterns. Twenty dot pattern stimuli were presented (10 golden ratio and 10 random stimuli, randomly selected from the full set of 40 stimuli), and participants were asked to rate them in terms of pleasantness, using a Likert scale from 1 to 7 (i.e., how much do you rate this image as pleasant? Choose a value from 1 to 7, where 1 = Not at all, and 7 = Very much).
2.5. IAT Stimuli
A set of 40 black dot-patterns consisting of 16 dots on a white background (10 cm × 10 cm) were created: half of the stimuli had configurations based on the golden proportion, whereas the other half had random configurations. The golden ratio stimuli were created according to the examples shown on
Figure 1c,d (see
Appendix A for the full set of stimuli), using two main types of golden ratio criteria: in one group (GR1), the criterion was that groups of 3 or more dots were aligned with a distance progressively increased by a factor equal to the golden ratio (
Figure 1c, upper panel), and in the other group (GR2), the criterion was that dots formed the vertices of a golden rectangle (
Figure 1c, middle panel). The different stimuli in these two groups were created according to the above criteria, modifying the distance among dots and their possible orientation. Random stimuli were extracted as a set of dots previously used in the study by Makin et al. [
19] on the preference for symmetrical dot stimuli over random ones.
Twenty words were selected from the Italian validation [
29] of the Affective Norms for English Words (ANEW) database [
30]. There were 10 positive words (Italian version of triumph, gift, holiday, music, trust, friend, laughter, health, affection, kiss), and 10 negative words (Italian version of war, deceit, slave, misery, tomb, devil, poison, nightmare, ache, crisis). The valence of the positive words was greater than the valence of the negative words (8.06 vs. 1.98,
p < 0.001, scale = 1 to 9), where the most positive word was ‘vacanza’ (Italian for ‘holiday’) and the most negative word was ‘guerra’ (Italian for ‘war’). However, the two sets of words were balanced for length (6.2 vs. 6.2,
p > 0.99) and frequency (27,278 vs. 30,614,
p = 0.87).
2.6. Line Bisection Task
Finally, participants performed a line bisection task on the computer [
31], in which they were asked to divide in two unequal parts the 100 mm segment shown on the screen by moving a virtual cutter.
2.7. Statistical Analysis
Reaction times (RT) from trials with errors (6.51%) and RT above 10 s (0.19%) were excluded from analysis (6.7%). After computing mean RT and response accuracy for each experimental block, we computed the D-score as the difference between mean RT in the incompatible and compatible blocks, divided by the standard deviations in those blocks [
19]. The D-score, which is not dependent on the order of block presentation, is close to zero if there is no difference between RT in incompatible and compatible blocks, it is positive if RT in incompatible blocks are longer than those in compatible blocks (confirming the hypothesis of a preference for the golden ratio), and it is negative if RTs are faster in incompatible than in compatible blocks (with a result opposite to the above hypothesis).
For the explicit measures, we computed the Rating Index for the judgement task as the mean score given on the Likert scale by all participants. For the line bisection task, we computed the mean position of the cutter with respect to the extreme left of the segment, and the mean length of the shorter part obtained after the cut (regardless of whether it was on the left or on the right of the segment).
Data are reported in terms of mean ± standard deviation. One sample t-test with respect to the null value was performed on the D-score. Paired t-tests were performed on the offers made at Ultimatum Game and on the bisections made at the line bisection test with respect to the golden ratio. Paired t-tests were also used to compare accuracy (computed as the sum of the correct answers) between compatible and non-compatible conditions of the IAT and to compare the ratings provided at the explicit judgement task for the two types of stimuli. Two Repeated Measure Analysis of Variance (RM-ANOVA) were used for RTs and for accuracy with compatibility, block, and their interaction. The effect size (ES) of these RM-ANOVAs was computed as the partial eta squared. If the sphericity assumption was violated, the Greenhouse–Geisser correction was adopted. Correlations between implicit and explicit measures were evaluated using Pearson coefficient (R). For all the tests, the alpha level of statistical significance was set at 0.05. Post-hoc were performed using Bonferroni correction.
3. Results
Mauchly’s Test revealed a significant violation of sphericity for Block W = 0.84, Chi-square (2) = 17.08, p < 0.001, and Compatibility by Block interaction, W = 0.91, Chi-square (2) = 8.76, p = 0.013, therefore, the Greenhouse–Geisser correction is reported.
As shown in
Figure 3 and
Table 2, the mean response times for compatible trials was shorter than for incompatible trials in the three blocks. Indeed, the RM-ANOVA on response time showed a significant main effect of compatibility (
F(1,99) = 129.8,
p < 0.001, η
2p = 0.567), with faster responses in the compatible blocks (M = 971, SD = ± 324 ms), compared to incompatible blocks (M = 1456, SD = ± 633 ms). The main effect of block was also significant (
F(1.7,171) = 66.1,
p < 0.001, η
2p = 0.40) due to faster responses as the experiment proceeded (B1: 1375 ± 675; B2: 1170 ± 509; B3: 1095 ± 427 ms). The interaction was also significant (
F(1.8,182) = 26.1,
p < 0.001, η
2p = 0.21). Post-hoc comparisons showed statistically significant differences between compatible and incompatible trials on all blocks (
p < 0.001). No significant differences were found based on block order. The D-score was 0.64 ± 0.43, significantly higher than the null value (
t = 14.92, df = 99,
p < 0.001).
ANOVA results for accuracy showed a main effect of compatibility, F(1,99) = 33.79, p < 0.001, η2p = 0.25), with greater accuracy in the compatible blocks (M = 0.95) compared to incompatible blocks (M = 0.92). The main effect of block was also significant, F(2,198) = 8.59, p < 0.001, η2p = 0.08 due to greater accuracy in the third block (M = 0.95) compared to the first block (M = 0.92, p = 0.001), but no difference was found between block 2 and 3, or block 1 and 2. The interaction was not significant.
To test if there was any effect of the two types of GR configurations, we performed a one-sample t-test on the D-index separately for GR1 (M = 1.45, SD = 1.45) and GR2 (M = 0.80, SD = 0.68) stimuli. Findings showed a statistically significant difference from the null value for both types of GR (t = 9.9, df = 99, p < 0.001 and t = 11.7, df = 99, p < 0.001, respectively), therefore, GR was implicitly preferred regardless of how it was implemented. However, the two D-scores were statistically different from each other (t = −6.10, df = 99, p < 0.001), but this difference was only due to a higher variability (i.e., standard deviation) for GR1. In fact, when comparing the differences in RT between incompatible and compatible trials (without dividing for the standard deviation), this difference was no longer statistically significant (t = −0.70, df = 99, p = 0.49).
Explicit Tasks
For the Ultimatum Game, the mean response was 54.7 ± 13.8, which was significantly different from the golden ratio (t = −5.15, df = 99, p < 0.001). Seventy-two participants chose the fifty–fifty division, the second more common division was that in golden ratio: twelve subjects used a division close (between 60:40 and 65:35) to the golden ratio (61.8:38:2). A post-hoc exploratory analysis was conducted on these 12 subjects who chose a golden ratio division for the Ultimatum Game. They also showed a statistically significant preference for the golden ratio in the explicit pleasantness rating index compared to the other participants (2.1 ± 1.5 vs. 1.2 ± 1.3, t = −2.16, df = 98, p = 0.033). However, there were no statistically significant differences in the D-score of the 12 subjects who chose a golden ratio as the solution for the Ultimatum Game compared to the other subjects (0.73 ± 0.36 vs. 0.62 ± 0.43, t = 0.83, df = 98, p = 0.407) or compared to performance on the line bisection task (69.6 ± 14.5 vs. 72.6 ± 13.8, t = 0.71, df = 98, p = 0.476).
Findings for the preference ratings showed that the golden ratio stimuli were rated as more pleasant, with a mean rating score of 4.0 ± 1.2, compared to random patterns that were 2.7 ± 1.1 (rating index: 1.3 ± 1.4, p < 0.001). The correlation between D-score and explicit pleasantness rating was R = 0.158, p = 0.116.
For the line bisection task, results showed that the cutting mean value was at 39.7 ± 24.2 mm from the left border, a value not significantly different from the golden ratio division 38.2:61.8 (t = 0.611, df = 99, p = 0.543). However, when considering only the mean length of the shorter bisected parts, the mean value was 27.7 ± 13.9 (t = −7.5, df = 99, p < 0.001).
4. Discussion
In the present study, we investigated individuals’ implicit and explicit preferences for golden ratio patterns. Previous research has suggested that the golden ratio can evoke aesthetic preferences, but evidence on individual implicit preferences for the golden ratio stimuli is scarce. To address this, one-hundred participants completed the explicit preference tasks and an implicit association task, which involved presenting dot-patterns based on the golden ratio and random dot-patterns, as well as positive and negative words.
The results of the study revealed the predicted IAT effect, with faster and more accurate responses in the compatible compared to incompatible blocks. This resulted in a positive D-score, which represents a measure of the strength of the hypothesized association between the categories being investigated. In this case, the positive D-score indicates an implicit preference for the golden ratio stimuli. This implicit preference was relatively independent of the criterion used to design the golden ratio stimuli. In fact, the implicit preference effect as expressed by the D-score was present for both types of golden ratio stimuli. Although it should be mentioned that the D-score for the two types of golden ratio differed from each other mainly for the higher variance present for GR1. In fact, when RT between incompatible and compatible trials were compared, the difference in implicit preference scores between the two types of golden ratio stimuli was not present anymore.
The aim of our study was to assess whether there is an implicit preference for golden ratio over random patterns, and we did so by using the same methodology as Makin et al. [
19] where symmetric dot patterns were compared to random dot patterns to assess whether there was an implicit preference for symmetry. Hence, in the present study, instead of symmetrical dot patterns, we used golden ratio patterns. We did not include other stimuli as we focused on the potential differences between implicit and explicit preferences for the golden ratio, and using other stimuli not only could act as confounders, but it would also have entailed increasing the number of condition and pattern combinations. Although we did not include other non-random stimuli and we could not compare the golden ratio patterns with different types of regularities, it is interesting that the values of the D-score were in line with those reported for symmetry in Makin et al. [
19] as they found a D-score >0.5 when reflective and rotative symmetric patterns were compared to random patterns. In our study, we found a D-score of 0.64, higher than the value found for reflective symmetry but lower than that found for rotational symmetry. Whether this suggests that the implicit preferences for different regularities decrease linearly from reflective symmetry to golden ratio and to rotational symmetry is for future studies to address.
We also asked participants to explicitly rate dot-pattern stimuli for pleasantness, but these findings are less clear. This is because although the golden ratio stimuli were rated more positively than random stimuli, the implicit preference D-score did not correlate with the explicit preference rating index. This finding suggests that implicit and explicit measures may tap on different processes. The lack of a correlation between the D-score and the explicit ratings is not surprising, due to correlations between implicit and explicit measures being typically weak [
32], which is attributed to these measures being able to elicit different, but still related constructs. It is possible that the lack of explicit preferences for golden ratio stimuli reported in previous studies is due to these stimuli being preferred only when considered in association to a positive-valence category (e.g., IAT). Gawronski et al. [
32] suggest that the IAT can elicit implicit attitudes, which participants are not always able to express in explicit self-report measures, due to being unaware of such inclinations. However, there is also evidence that performance at the IAT can be predicted by some participants [
33,
34], challenging the assumption that the implicit preferences are not accessible by consciousness.
In addition, when participants were asked to ‘produce’ ratios by dividing quantities (regardless of whether instructed to do so asymmetrically as for bisecting a line or simply instructed to divide an amount of money) there was no evidence of the golden ratio for sharing an amount of money [
15,
16]. It is interesting that we found a lack of evidence for the golden ratio also for the line bisection task (i.e., mean length of the shorter bisected parts), which participants completed after the IAT (as it could have primed golden ratio) and with the explicit instruction to bisect asymmetrically. Therefore, our findings indicate that people do not spontaneously use golden ratio rules in these tasks. It is possible that individuals may not have explicit knowledge of the golden ratio, yet the implicit association with beauty can still emerge. Although the present findings are at odds with those reported by Stieger and Swami [
24], who also used the IAT and found an implicit preference for symmetry but not for the golden ratio, this may be because they compared golden ratio stimuli to symmetric ones. In fact, they also found that the golden ratio stimuli were preferred only when compared to the ¾ proportion.
We would like to acknowledge some limitations of the present study. Firstly, we used online data collection, but to minimize the effect of possible noise, we oversampled our subject pool to one-hundred participants, whereas typical IAT studies use smaller sample sizes [
35]. Moreover, there is evidence of similar results obtained from both online and in-lab studies [
36]. We also checked for outliers in our data (e.g., higher percentage of errors or longer time for experiment completion, compared to the total sample), and none were found, therefore, the data of all participants were included in the data analyses. A second limitation is that the sample involved only young-adults, and future research should investigate differences across different populations. Furthermore, one can argue that a third limitation of our study concerns the difference in the instructions for the line bisection task and for the Ultimatum Game as we explicitly asked participants to divide the line not symmetrically for the bisection task, whereas for the Ultimatum Game we left participants free to choose, including choosing the symmetric 50:50 ratio. Indeed, most of the participants opted for the symmetric division in the Ultimatum Game, suggesting that symmetry is explicitly preferred to golden ratio not only for visual tasks as those used by Stieger and Swami [
24], but also for tasks such as the Ultimatum Game. In contrast, for the line bisection task there was an effect of the golden ratio only when considering the average division, which, as the presentation order of the explicit tasks was not counterbalanced, could be due to previous exposure to golden ratio stimuli during the IAT.
It is also important to point out that our study was not aimed at comparing symmetry and the golden ratio, but rather to assess whether there is a preference—implicit and/or explicit—for the golden ratio with respect to chance. Finally, we would like to acknowledge that the present findings are informative of how golden ratio stimuli are perceived and cannot be used as a predictor of possible individuals’ behaviour. This is particularly important as an interesting review has reported how IAT findings on individuals’ implicit preferences are often a bad predictor of actual behaviour [
22].
To sum up, the present findings converge with Makin and colleagues [
19] in pointing out that the IAT represents a valuable tool to investigate visual preferences and, by extension, aesthetic experiences not only for symmetry but also for the golden ratio, although literature presents discussions on IAT limitations [
37,
38,
39].
Future studies may further investigate whether the golden ratio is explicitly preferred when stimuli are not geometrical, like when the golden ratio is embedded in an artistic stimulus [
25], or when participants perceive a harmony without an explicit knowledge of the geometrical rule behind it. In fact, the aesthetic preference for the golden ratio may emerge when this proportion is embedded in an ecological/artistic stimulus more than when it is present in an abstract geometrical/virtual image. Our findings cannot clarify this point as we compared the golden ratio to random geometrical patterns. The present findings are also consistent with evidence showing that explicit preferences are task dependent. Finally, our results did not show a clear preference for the golden ratio in the Ultimatum Game, which is at odds with previous studies [
15,
16]. However, it should be mentioned that McManus et al. [
40] found that individuals had preferences for squares (corresponding to a 50:50 proportion, much like that preferred by our subjects in the Ultimatum Game) and for the golden ratio, but they also observed high inter-individual differences, which were not explained by personality, need for cognition, tolerance of ambiguity, vocational types, or aesthetic activities. The issue of the role of individual differences on preferences for the golden ratio is a complex one and still unresolved. In fact, whereas Stieger and Swami [
24] reported that preferences for the golden ratio may be affected by aesthetic experiences, De Bartolo failed to find any significant correlations between preferences for the golden ratio and cultural skills, but observed a relation between self-assessed art and math knowledge. Indeed, they suggested that the golden ratio may be a sort of affordance present in the environment that could be associated with an easier visual processing, and this could explain why symmetry and the golden ratio are preferred at the implicit more than at the explicit level.
To conclude, the present findings provided evidence of an implicit preference for the golden ratio. Human beings have an implicit and explicit aesthetic preference for symmetric stimuli [
24], probably because of processing fluency [
26] and/or due to the detection of a gestalt-like pattern, or because in general, they prefer ordered versus random stimuli [
19,
20]. When considered in this context, the golden ratio could represent a second level of order, probably more difficult to be explicitly preferred, but implicitly engendering some preference both in abstract (as in our study) or artistic [
25] stimuli. Future research could investigate more in depth the association between the golden ratio and positive valence category, and under which conditions differences in explicit judgments emerge but also to what extent the golden ratio is preferred to other types of regularities.