Dataset DOI: 10.5061/dryad.msbcc2g9d

Description of the data and file structure

The repository contains the data and scripts necessary to reproduce the results and figures related to the article and supplementary materials by Mottier, H., Pascalis, O., Quinn, P. C., Méary, D. (2026). Do Infants Have a Sense of Beauty? A Study Using Kinetic Dot Displays. Accepted for publication in Proceedings of the Royal Society B: Biological Sciences.

The study involved two tasks.

In the preference judgment task, an independent group of adults judged the relative beauty of the dot displays, which were presented in pairs.

In the visual preference task, the same pairs of dot displays were shown to adults and infants, without instructions, while their eye movements were measured with an Eyelink® 1000 system (SR Research Ltd., Mississauga, Ontario, Canada) with a 500 Hz sampling frequency.

All measures were performed using MatLab (R2023b) and the Eyelink® toolbox from the PsychToolbox. Statistical analyses were performed with Matlab’s Statistical toolbox.

The repository contains one folder for each task and a third folder with example raw files from the EyeLink.

Files and variables

File: PreferenceJudgmentTask.7z

Description: The folder PreferenceJudgmentTask.7z contains a subfolder (Rating_Results), containing the experimental data, and a script (Step1_Judgement.m) detailing how to generate Figure S1.tif and the WinRatePair.mat file used for analysis of the results of the second visual preference task. It also includes a function (ICC.m, © 2008 Arash Salarian) used to calculate the intraclass correlation coefficients and a file (MovParamNew20.mat) containing the descriptors of the motion pattern as in Zeki et al 2012.

Figure S1. Preference Ratios as a Function of Kinetic Pattern Pairing. Note. Frequency of choices for kinetic patterns P_x, depending on the opponent pattern (dots with pattern number). Dashed lines give the boundaries of the 95% confidence interval under the binomial hypothesis of null preference in judgment for a pair. Patterns P₂, P₄, P₅, and P₇ had above chance ratings over opponents like P₁ and P₆. Pattern P₁ was judged less beautiful than P₂, P₄, P₅, P_7, and P₈, but was equivalent with P₃ and P₆. The grey plain lines give the average preference ratio over the seven possible pairings of a pattern.

The Rating_Results subfolder contain 35 matlab data files each containing 2 matrices

resp is a 56 x 4 matrix. The columns are 1) kinetic pattern played on the left, 2) kinetic pattern played on the right, 3) decision time from movement onset to response, 4) response (1 for left pattern and -1 for right pattern). This atypical coding comes from the Psychtoolbox script used to conduct the experiment. It was recoded (-1 for left, 1 for right) line 30 of the script Step1_Judgement.m. The 56 lines correspond to the 56 pairs of patterns used in the preference judgment task.

Subject is a structure containing 6 fields with basic participant information. In addition to figure S1, the generated file for further processing is WinRatePair.mat:

WinRatePair.mat is a file containing 2 matrices (WinRatePair and NumWinS)

WinRatePair is a 56 x 19 matrix. The columns are: 1) kinetic pattern played on the left 2) kinetic pattern played on the right, 3) frequency of choice for the left pattern, 4) frequency of choice for the right pattern), 5) number of choices (2 * 35 judges), 6) an index identifying the mirror pairs (eg., {1,2} and {2,1}), 7) number of choices for the left pattern, 8) number of choices for the right pattern, 9) the mean decision time, 10-11) Zeki’s rating for left and right patterns, 12-13) Disuniformity measure for left and right patterns, 14-15) Incoherence measure for left and right patterns, 16-17) Roughness measure for left and right patterns, 18-19) mean preference score across pattern for the left and right patterns. The 56 lines correspond to the 56 pairs of patterns used in the preference judgment task. This matrix is used to predict the looking time results in the visual preference task.

NumWinS is a vector giving the mean preference ratio (beauty judgment) for each of the eight patterns (over all comparisons patterns). It is used to check the correspondence of judgments with those from Zeki and Stutters (2012) and calculate the correlation between their rating task and our preference judgment task reported in the article.

File: VisualPreferenceTask.7z

Description: The folder VisualPreferenceTask.7z contains a subfolder Results and 3 scripts reproducing the results reported in the article (Step1_GetData; Step2_LinearModel; Step3_TimeCourse). It also contains 2 scripts used to do the violin plot (Violin.m and violinplot.m, © 2016, Bastian Bechtold), the WinRatePair.mat file from the preference judgment analysis and 2 files (DataForStats.mat, LinearModelOverTime.mat) that are generated when running the Step1_GetData and Step3_TimeCourse scripts, respectively.

Results: Contains data extracted from the raw .edf file returned by the EyeLink. Examples of the raw files (.edf, .asc, .mat) are given in the third folder RawFiles (see below section 3). Each of the 323 Data_*.mat files in Results contains 7 elements resuming the raw .edf files.

• DATA is a n x 8 matrix with the EyeLink measures.
• ELBLINK is a n x 5 matrix indexing Blink in the raw data.
• ELFIX is a n x 8 matrix describing and indexing fixations in the DATA matrix.
• ELSACC is a n x 11 matrix describing and indexing saccades in the DATA matrix.
• Expe is a structure giving details of experimental setting.
• Subject is a structure with anonymized information about the participant.
• TRIAL is a n x 3 matrix indexing trial event in the DATA matrix (attention-getter on, stimulus onset and stimulus end.

Extended description of the elements and of their use are given in the script Step1_GetData

The scripts:

Step1_GetData.m goes through the Results folder and output DataForStats.mat containing ‘AllRes'; 'ntrial'; 'AllLeftRes'; 'AllRightRes'; 'AgeUGr', and ‘ListSu’. Detailed explanations are given in the script. These matrices are used in the next steps to produce the results reported in the manuscript.

Step2_LinearModel uses DataForStats.mat to produce Table 2 (or Table S2 if a different setting is selected) and Figures 2, S2, S3 and S4. Details and legends of figures are given in the script and below.

Figure 2. Relation Between Preference Judgment and Looking Time in the Visual Preference Task for a Randomly Chosen Participant in Each Age Group. Note. Each dot represents the looking time to one pattern with PJ value x within a pair (2 dots per trial). Dashed lines give the 99% confidence interval for the slope and intercept parameter of the predicted values (plain line).

Figure S2. Distribution of Individual Values for the Intercept (β₀) in Each Age Group. Note. Each dot represents the participant’s mean LT at one pattern within the pair as estimated using linear regression LT_j = β₀ + β₁PJ_j + ε_j (as described in the section on Preference Judgment and Participants’ Preferential Looking Times in the main text of the manuscript). Note the correspondence of the groups’ mean values with the results from the group analysis (Table 2 in the main manuscript).

Figure S3. Distribution of Individual Values for the Slopes (β₁) in Each Age Group. Note. Each dot represents a slope value from a participant.

Figure S4. Scatterplots of Looking Time as a Function of Preference Judgment and Age Group. Note. Each point represents the preferential looking time to a pattern (within a pair) as a function of adults’ pattern rating and age group. Grey areas show the 99% confidence interval for the mean intercept and slope of the group.

Step3_TimeCourse use LinearModelOverTime.mat to produce Figure 3 and the additional time course figure for B0 shown in the supplemental materials (Figure_S5). It can also be used to recompute the LinearModelOverTime.mat matrix using DataForStats.mat. Detailed explanations are given in the script. Details and legends of figures are given in the script and below.

Figure 3. Variation of the Effect of Preference Judgment (β₁PJ) During the 5 s of Display Duration. Note. The thicker lines give the time-step where the effects of preference judgment were significant (all ps < 0.01).

Figure S5. Variation of Mean Looking Time (β₀) During the 5 s of Display Duration. Note. The thicker lines give the time-step where the effects of β₀ were significant (all ps < 0.01).

File: ExampleRawFiles.7z

Description: The third subfolder (ExampleRawFiles.7z) contains examples of the raw files (.edf, .asc, .mat) as they were before the pre-processing needed to produce the Data_*.mat files given in the Results folder and used by the script Step1_GetData.m. We arbitrarily chose the first participant of each age group.

The reason why we did not directly upload the raw files is practical. The DATA_*.mat files in Results are 10 times smaller in size than the equivalent .asc file. Pre-processing is based on a script (Step_1_CheckFiles.mat, not furnished) using the subjects’ .asc and .mat files. The script is basically 1) going through the .asc files to extract fixations, blinks, and saccades reported by the EyeLink, 2) using the subjects’ .mat files from the experiment to define age in days and other variables, and 3) identifying the trial triggers (given in the matrix TRIAL available in the DATA_*.mat files in Results) and used in Step1_GetData.m.

In the DATA_*.mat files, there are the matrix DATA that reproduce in a more compact format the data from the .asc raw file. To show what is done by Step_1_CheckFiles.m one can compare the content of the .asc files in ExampleRaw to those of the corresponding file in the Results folder.

For example, in DATA_22210601.mat the TRIAL matrix informs that the first trigger (attention getter on) is occurring in line 65 of the DATA matrix. The last column of the data matrix indicates that this sample in associated with the time value 4686985 (time value of the EyeLink sample). The 22210601.asc file at this time value (line 99) shows that the trigger occurred in between this time step and the next one 4686987.

The whole raw Data set (213 Mo) can be made available on motivated request to the corresponding author.

Code/software

All measures were performed using MatLab (R2023b) and the Eyelink® toolbox from the PsychToolbox. Statistical analyses were performed with Matlab’s Statistical toolbox. We used 2 third party functions.

• ICC.m © 2008 Arash Salarian (used in PreferenceJudgmentTask.7z)
• Violin.m and violinplot.m © 2016, Bastian Bechtold (used in VisualPreferenceTask.7z).

Both functions are available from MatLab exchange site.

Human subjects data

The study involved human infants and their parents. The participant data have been anonymized. The study is not a health-related study and was not involving extended collection of personal information. The study was conducted in accordance with the ethical principles of the declaration of Helsinki and was approved by the ethics committee from the University Grenoble-Alpes under the reference IRB00010290-2018-02-06-39.