Age-adapted painting descriptions change the viewing behavior of young visitors to the Rijksmuseum

Abstract

Children learn about art by actively engaging with their surroundings. This makes museums potentially rich environments for learning and development. Yet, the descriptions of paintings on show are usually written for adults rather than younger visitors. This study uses mobile eye tracking to examine how painting descriptions tailored for children influence their eye movements when viewing paintings at the Rijksmuseum - the national museum of The Netherlands. Our findings underscore the importance of adapting information specifically for children, rather than simply providing them with adult-oriented museum materials. Children who received information tailored to their developmental level showed increased glance durations to areas highlighted in the descriptions. Strikingly, the behavior of children provided with painting descriptions intended for adults was no different from their behavior when they received no information at all.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-73963-y.

Introduction

Museums in the Netherlands attract over 23 million visitors annually¹, offering inspiring environments that encompass art, history, culture and entertainment^2,3. Beyond mere recreation, museums are recognized as valuable educational tools, providing informal avenues for new experiences, knowledge acquisition, and the development of perceptual abilities³. This educational potential is particularly significant for children, given the rapid evolution of their knowledge, cognitive skills, and personal preferences during their formative years^4,5.

Theories on cognitive development posit that children learn by actively engaging with their surroundings⁴, making museums potentially rich environments for fostering learning and development. In response to an evolving societal landscape, museums are increasingly adapting their educational services and collections to create interactive and tailored experiences. Collections are becoming more accessible to children, who constitute approximately 12% of museum visitors in The Netherlands. The majority of these children are under 12 years old¹.

Despite the increased attention towards child-centric experiences, the painting descriptions museums provide their visitors are usually written for adults rather than for children. The effectiveness of such descriptions in supporting children’s understanding and perception of artworks remains unknown. In fact, there is no agreement in the current literature even about the impact of painting descriptions on aesthetic appreciation in adults^{[e.g., 6–8]}. Although studies have shown that presenting information in a narrative format enhances children’s retention of key elements^9–11, museums do not commonly use storytelling as a way to engage their younger visitors. Against this background, Castellotti et al. (2023)⁷ have suggested that museums should pay more attention to the manner in which the artworks are presented and explained, allowing them to reach a large audience of non-expert visitors, including children.

Here we investigate how painting descriptions impact children’s engagement with art, using mobile eye tracking in the Rijksmuseum – the national museum of The Netherlands, located in Amsterdam. Previous work has shown that eye movements can unveil some of the intricate cognitive and perceptual components underlying aesthetic experiences^12,13. By examining eye movements while looking at a painting, we can map the location of fixations made by the viewers, and understand which pictorial features capture their interest and prompt exploration¹². The interplay between bottom-up attention (driven by external stimuli) and top-down attention (guided by internal goals and intentions) is thought to shape visitors’ gaze patterns, influencing what they look at, for how long, and the subsequent cognitive and emotional responses^14,15. In short, eye-tracking gives us a means to probe various cognitive processes directly, which is a considerable advancement over more indirect, post-hoc measures such as questionnaires.

Essentially, the eyes can be captured by salient stimuli in the environment, resulting in bottom-up attentional allocation. Salient stimuli stand out from their surroundings due to differences in basic features such as color, luminance, shape, size, orientation or movement^15–17 and are processed automatically by the viewer¹⁴. In contrast, top-down attention is an active and volitional process, influenced by internal goals, intentions, and expectations^14,16.

In museums, the dynamics of bottom-up and top-down attention likely play a pivotal role in shaping how visitors engage with the art on display – but they may do so in different ways. Notably, while bottom-up saliency is thought to be relatively stable, the malleability of top-down attentional processes may allow for intentional guidance of viewers’ focus towards specific areas of an artwork. It is the latter type of attention where painting descriptions emerge as influential tools capable of shaping visitors’ eye movements and, consequently, leave a profound impact on their learning and aesthetic experiences.

It is important to note here that this is not the first study to use physiological measures to investigate art experience. Previous research has demonstrated the influence of instructions and information on adults’ attention while viewing art, showing that information can strongly and positively impact viewing time, scan paths, visual elements viewed, as well as skin conductance and pupil size^{[e.g., 7,18–20]}. However, those studies have been conducted with small samples and mostly in laboratory settings, and as such may lack generalizability to larger audiences in natural museum contexts – and especially to children, whose viewing behaviors change during cognitive development between the ages of 10 and 12^4,5. Several studies have shown that viewing times differ between lab and museum settings^{[e.g., 21,22]}  and that museum visitors show a more active and exploratory viewing behavior²³. As emphasized in the current literature^{[e.g., 23,24]}, while laboratory studies have made significant contributions to our understanding of art perception, there is a growing need for ecologically valid examinations to capture more naturalistic viewing behaviors. A study by Walker et al. (2017)²⁵ on 11-12-year-old children and adults’ visual attention in the Van Gogh Museum Amsterdam highlighted the role of bottom-up attention in children’s initial viewing, transitioning to top-down attention with the aid of painting descriptions. However, the same descriptions were used for both children and adults. The potential differential impact of tailored descriptions for children remains an unexplored area. Furthermore, in Walker et al. (2017)²⁵ painting descriptions were predominantly factual, exclusively focusing on works by Vincent Van Gogh and tested on a relatively small sample size comprising twelve adults and nine children. In contrast, the present study uses narrative painting descriptions depicting artworks from the 17th century and is conducted on a substantially larger sample of sixty-two children.

The current study employs mobile eye tracking technology to examine how painting descriptions tailored for children influence their eye movements when viewing paintings at the Rijksmuseum. We hypothesize that children exposed to tailored descriptions will spend more time in Areas of Interest (AOIs), defined as the key visual elements highlighted in the descriptions. We expect this effect to be more pronounced relative to when children receive descriptions tailored for adults or no descriptions at all. Additionally, we expect that these tailored descriptions will impact children’s aesthetic appreciation of the artworks.

Overall, we aim to provide insights that can inform educational practices and contribute to a deeper understanding of children’s engagement with art in cultural institutions. We and the other stakeholders in this study share the goal of improving the visitor’s experience, and firmly believe that incorporating scientific evidence alongside qualitative data can help museums in planning their future displays.

Results

Effect of description type on gaze behavior

We used an analysis of variance (ANOVA) to test whether child descriptions are more effective in guiding participant’s fixations compared to adult descriptions or free viewing. The ANOVA on normalized glance duration in AOIs with painting (Schuttersmaaltijd, Pronkstilleven, Winterlandschap), AOI (child, adult) and phase (phase 1, phase 2) as within-subjects factors and description condition (CDC, ADC, FVC) as between-subjects factor, showed a significant main effect of painting, F(2, 106) = 57.126, p < .001, η_p² = 0.519, a significant main effect of AOI, F(1, 53) = 8.203, p = .006, η_p² = 0.134, a significant main effect of phase, with participants spending more time in AOIs in phase 2 than in phase 1, F(1, 53) = 22.758, p < .001, η_p² = 0.300, and a significant main effect of condition, F(2, 53) = 8.939, p < .001, η_p² = 0.252.

Furthermore, there was a significant two-way interaction between painting and condition, F(4, 106) = 8.100, p < .001, η_p² = 0.234, between AOI and condition, F(2, 53) = 23.309, p < .001, η_p² = 0.468, between phase and condition, F(2, 53) = 13.311, p < .001, η_p² = 0.334, between painting and AOI, F(2, 106) = 7.480, p = .001, η_p² = 0.124, between painting and phase, F(2, 106) = 20.585, p < .001, η_p² = 0.280, and between AOI and phase, F(1, 53) = 14.239, p < .001, η_p² = 0.212.

Finally, significant three-way interactions were observed between painting, AOI and condition, F(4, 106) = 21.620, p < .001, η_p² = 0.449, between painting, phase and condition, F(4, 106) = 12.198, p < .001, η_p² = 0.315, between AOI, phase and condition, F(2, 53) = 20.334, p < .001, η_p² = 0.434, and between painting, AOI and phase, F(2, 106) = 13.724, p < .001, η_p² = 0.206.

Given that the paintings were intrinsically different, between-painting differences were not analyzed any further. Nonetheless, post hoc pairwise comparisons were conducted to delve deeper into the two and three-way interactions between AOI, phase and condition. Bonferroni corrections were applied to avoid Type 1 errors. As expected, no significant differences between the conditions were observed in phase 1 (all p’s > 0.05), meaning that any differences that would be observed subsequently in phase 2 could not be attributed to unforeseen differences among participant groups.

In line with the hypotheses, strong significant differences were observed in phase 2. Here, as expected, participants in CDC (M = 0.11, SD = 0.06) spent more time in CDC AOIs than participants in ADC (M = 0.03, SD = 0.01, p < .001, d = 1.63) and FVC (M = 0.03, SD = 0.01, p < .001, d = 1.86). Not surprisingly, participants in ADC (M = 0.06, SD = 0.03) spent more time in ADC AOIs than participants in CDC (M = 0.03, SD = 0.01, p = .001, d = 1.34). Yet, they did not spend significantly more time viewing ADC AOIs compared to participants in FVC (M = 0.04, SD = 0.02, p = .055).

The uncorrected data showed that in phase 2 participants in CDC spent 31% of their time viewing CDC AOIs, while children in ADC spent only 22% viewing ADC AOIs. To ensure that this difference was not due to variations in the size of CDC and ADC AOIs, we analyzed the three-way-interaction between AOI, phase and condition using normalized glance duration data. An independent samples t-test indicated that, in phase 2, participants in CDC spent more time viewing CDC AOIs (M = 0.11, SD = 0.06), than participants in ADC spent viewing ADC AOIs (M = 0.06, SD = 0.03), t(37) = 3.486, p = .001, d = 1.117 (Fig. 1).

Fig. 1.

Normalized glance duration towards CDC and ADC AOIs. Note. In phase 2, participants in CDC spent more time viewing CDC AOIs, than participants in ADC spent viewing ADC AOIs. Error bars are based on the standard error of the means. For calculation of normalized glance duration, see “Raw data processing” section.

Overall, these results indicate that children directed their eyes more to relevant AOIs when receiving a description tailored to them compared to receiving a description written for adults or receiving no description at all. Crucially, the viewing behavior of children who received the adult description mirrored the behavior of participants who received no description at all. The absence of such differences in phase 1 strongly suggests that the findings in phase 2 can be attributed to the variations in painting descriptions rather than to other potential confounding factors.

Many of the findings described above are already apparent in heatmaps, showing patterns of participants fixations for the different paintings. In phase 1 (Fig. 2), the heatmaps for all three conditions are similar and concentrations of fixations in particular areas are rare. No large inter-group differences are apparent. Conversely, in phase 2 (Fig. 3), participants in CDC paid more attention to AOIs mentioned in the descriptions, compared to participants in the ADC and FVC.

Fig. 2.

Heatmaps of glance duration during Phase 1, averaged across participants. Note. a: Schuttersmaaltijd; b: Pronkstilleven; c: Winterlandschap. CDC AOIs are presented in orange; ADC AOIs in white. In b, “Lobster” was in both conditions and is, therefore, presented in light blue. Parts that are fixated for longer are presented in red, scaling down to green for areas that are fixated for shorter times. Painting images were downloaded from Rijksmuseum.nl and are used in accordance with the Rijksmuseum’s policy.

Fig. 3.

Heatmaps of glance duration during Phase 2, averaged across participants. Note. a: Schuttersmaaltijd; b: Pronkstilleven; c: Winterlandschap. CDC AOIs are presented in orange; ADC AOIs in white. In b, “Lobster” was in both conditions and is, therefore, presented in light blue. Color coded as in Fig. 2. Painting images were downloaded from Rijksmuseum.nl and are used in accordance with the Rijksmuseum’s policy.

Given that paintings Schuttersmaaltijd and Pronkstilleven contain a relatively large number of AOIs, the difference may not be immediately apparent. In painting Winterlandschap, on the other hand, we observe a clear difference between the CDC and ADC conditions. It can be observed that participants in CDC (Fig. 4, panel a) tend to focus on the AOI mentioned in the description, while viewing patterns in ADC (Fig. 4, panel b) and FVC (Fig. 4, panel c) are more diffuse. In fact, there are no apparent differences between the observed behaviors in these two conditions.

Fig. 4.

Heatmaps showing glance duration in painting Winterlandschap during Phase 2, averaged across participants. Note. a: Heatmap of glance duration in CDC; b: Heatmap of glance duration in ADC; c: Heatmap of glance duration in FVC. Painting images were downloaded from Rijksmuseum.nl and are used in accordance with the Rijksmuseum’s policy.

Visual Salience

To exclude the possibility that children’s observed gaze behavior was due to the visual salience of particular AOIs rather than to the descriptions provided them, we calculated mean salience values for the AOIs mentioned in each description. We then compared the seven AOIs mentioned in the CDC with the six present in the ADC. AOI “Lobster” (painting: Pronkstilleven) was present in both conditions and was therefore excluded from the analysis. No differences were found between the two set of descriptions, t(11) = 0.467, p = .66. We conclude that the AOIs mentioned in the child and adult descriptions did not differ in terms of visual salience.

Interestingly, although CDC and ADC AOIs did not differ in terms of visual salience, salience at fixation of children in the CDC increased significantly from phase 1 to phase 2, t(54) = 4.804, p < .001, d = 0.38 (Fig. 5), while it did not for the ADC and FVC (all p’s > 0.05). We will address the interpretation of this unexpected result in the Discussion.

Fig. 5.

Condition per phase interaction. Note. Salience at fixation of children in the CDC increased significantly from phase 1 to phase 2. Error bars are based on the standard error of the means.

Self-reported Aesthetic Emotion Intensity

One-way ANOVAs were used to compare aesthetic emotion intensity between the three conditions. The results showed no significant differences in aesthetic emotion intensity between the three conditions. From the scores, it is evident that while all paintings elicited more pleasing than negative emotions, none evoked strong emotional responses. Participants’ ratings tended to cluster slightly below the middle of the scale (Table 1).

Table 1.

Overview of the ANOVA results on aesthetic emotion intensity.

Painting	Subclass	Scores	F value	Significance p	Effect size η2
Schuttersmaaltijd	Full questionnaire	27.3	0.86	0.430	0.03
	Prototypical emotions	7.16	1.37	0.262	0.04
	Pleasing emotions	10.18	1.44	0.245	0.05
	Negative emotions	5.54	0.41	0.792	0.01
	Epistemic emotions	6.74	0.23	0.665	0.01
Pronkstilleven	Full questionnaire	30.5	0.50	0.611	0.02
	Prototypical emotions	7.31	1.54	0.223	0.05
	Pleasing emotions	11.02	1.08	0.347	0.04
	Negative emotions	5.45	0.26	0.774	< 0.01
	Epistemic emotions	6.73	0.08	0.924	0.01
Winterlandschap	Full questionnaire	31.32	0.74	0.480	0.03
	Prototypical emotions	7.9	1.18	0.315	0.04
	Pleasing emotions	11.87	0.09	0.918	< 0.01
	Negative emotions	4.94	1.03	0.127	0.07
	Epistemic emotions	6.61	2.14	0.364	0.03
All Paintings Combined	Full questionnaire	29.71	0.94	0.398	0.03
	Prototypical emotions	7.46	2.02	0.142	0.06
	Pleasing emotions	11.02	0.87	0.424	0.03
	Negative emotions	5.31	0.18	0.837	0.01
	Epistemic emotions	6.69	0.31	0.736	0.01

Children’s Speech

We conducted a one-way ANOVA to explore the language used by children across the three different conditions. Our analysis revealed no significant differences in word count (F(2, 53) = 1.672, p = .197; ADC: M = 146.381; CDC: M = 108.762; FVC: M = 112.1). We observed several differences in the frequency of interjections and nouns (data not shown), but given the number of POS tests performed, these differences cannot be considered as significant. Measures of text complexity, such as number of unique words and sentence length, did not show significant differences.

Finally, we observed that children occasionally pointed to the paintings rather than verbally describing the areas they were referring to. Analysis revealed a significant difference between conditions, F(2, 53) = 6.857, p = .002, η_p² = 0.083. Children in ADC (M = 1.38, SD = 1.72) pointed to the pictures more frequently than participants in CDC (M = 0.24, SD = 0.44, p = .004, d = 0.91) and FVC (M = 0.35, SD = 0.67, p = .011, d = 0.78).

Discussion

Many museums tailor painting descriptions primarily for adult visitors. The effectiveness of such information in supporting children’s perception of artworks remains uncertain. Equally unclear is whether modifying these descriptions to be more accessible for children can enhance their museum experience. Addressing these knowledge gaps, our paper explores the hypothesis that tailoring painting descriptions specifically for children can lead to more effective guidance of their visual attention, ultimately resulting in a more enriching museum experience.

To test our hypotheses, we conducted a study with children aged 10 to 12 at the renowned Rijksmuseum in Amsterdam. In the first phase, participants freely viewed three selected paintings. Subsequently, in the second phase, participants received painting descriptions tailored for children (CDC), written for adults (ADC), or no descriptions at all (FVC).

As anticipated, children who received information tailored to their developmental level showed increased glance durations to areas highlighted in the descriptions. Strikingly, children receiving adult-oriented information did not significantly differ from those with no descriptions at all, showing that typical painting descriptions for (adult) visitors may be ineffective in guiding the visual attention of young viewers. This raises concerns about the potential limited impact of such descriptions on children’s learning and cognitive development during museum visits.

Interestingly, children in the ADC group pointed significantly more to the areas they were describing than in any other condition. We conjecture that children who did not receive information tailored to their age had more difficulty naming the areas. In future work it would be interesting to test whether tailoring descriptions to children enhances their ability to articulate observations effectively.

Unexpectedly, we found that children in the CDC not only directed their attention towards described areas but also demonstrated increased spontaneous attention towards visually salient regions. Given that visual salience of CDC and ADC AOIs does not significantly differ, this is not a confound. We speculate that the provision of descriptions adapted to children has an impact not just on their specific attentional strategies, but also on their general level of engagement. We plan to investigate this effect further in future studies.

Contrary to our expectations, painting descriptions did not impact children’s self-reported aesthetic appreciation of artworks. This null result may be due to the inherent weaknesses of self-reports, emphasizing the need for more refined measures in assessing aesthetic appreciation, especially in children. However, it is plausible that the use of child-oriented descriptions did not result in stronger emotional responses. This is what the aesthetic appreciation scores suggest, with all ratings clustering slightly below the middle of the scale.

Importantly, the study was conducted in a natural museum setting during regular hours, strengthening the generalizability of our conclusions. The sample size, significantly higher than previous studies in similar settings^{[e.g., 7,18]}, further contributes to the reliability of our statistical analyses. Given the natural setting in which the study was conducted, we controlled meticulously for variations in size of paintings and AOIs.

Our study provides compelling evidence that customizing painting descriptions is a highly effective strategy for guiding children’s visual attention when viewing art. Given that the child and adult AOIs did not significantly differ in terms of visual salience, the between-group variations may be due to differences in language use or to differences in the content. We conjecture that the storytelling format employed in the child descriptions may have contributed to their effectiveness. This is a theme to be explored in future studies.

More generally, our findings underscore the importance of adapting information specifically for children rather than simply providing them with adult-oriented museum materials. In fact, our results strongly suggest that when children are provided with painting descriptions intended for adult visitors, their behavior is no different from when they receive no information at all. Interestingly, no specific guidance was provided to the educational experts who wrote the descriptions. We conclude that even in the absence of tight guidelines, museum experts can tailor descriptions for children in ways that significantly impact how they experience the artworks on display.

In conclusion, our study emphasizes the critical importance of tailoring information to match children’s cognitive and developmental needs. Recognizing and addressing these unique requirements holds the key to significantly enhancing museum education, while improving children’s learning experiences and engagement.

Methods

Study design

We conducted a study amongst children aged 10 to 12 at the Rijksmuseum in Amsterdam during opening hours. Children were asked to view three paintings while wearing a mobile eye tracker. The experiment followed a between-subjects design, and all children were randomly assigned to one of three conditions: a child description condition (CDC), an adult description condition (ADC), or a free viewing condition (FVC). The experiment consisted of two phases. In phase 1, all children were asked to freely view the three paintings. In phase 2, before viewing the paintings again, children in the CDC were provided with a painting description specifically tailored towards their level of understanding and development. Children in the ADC were given a painting description written for adults. Lastly, children in the FVC received no descriptions and were instructed to view the three paintings again. Importantly, we did not expect to observe any between-group differences in phase 1. This phase acted as a control, ensuring that variations observed in phase 2 would be attributable to our manipulation rather than to inherent differences among the groups. The experimental protocol was approved by Leiden University’s Research Ethics Committee (reference number: 2022-08-31-F. Walker-V1-4205), and was performed in accordance with all the relevant guidelines and regulations of the committee. Informed consent was obtained from all participants. All gathered data was collected, processed, and stored anonymously. The research was performed in accordance with the Declaration of Helsinki.

Participants

We recruited sixty-two participants. These joined the study in exchange for a children’s cookbook, provided by the Rijksmuseum. 21 participants joined the CDC, 21 the ADC, and 20 the FVC. 43.5% of participants identified as female and 56.5% as male. The mean age of the sample was 10.8 years (SD = 0.88). There were no significant differences between the average age (F(2,59) = 2.54, p = .088) or gender identity among the conditions (X²(2, N = 62) = 1.96, p = .375).

None of the children attended art school, and none had visited the Rijksmuseum before. All participants had normal or corrected to normal vision (and no color vision deficits). As the study was conducted in Dutch, proficiency in Dutch was a prerequisite.

Measures and materials

Paintings

We selected three paintings: Schuttersmaaltijd ter viering van de Vrede van Munster by Van Der Helst (1648), Pronkstilleven by Van Utrecht (1644) and Winterlandschap met schaatsers by Avercamp (c. 1608) (Fig. 6). In what follows, these will be referred to as Schuttersmaaltijd, Pronkstilleven and Winterlandschap. The three paintings date back to the 17th century, and are part of the Rijksmuseum’s renowned and extensive collection of masterpieces that capture the Dutch art and history of that period. The paintings were selected because they contained multiple people, items, and animals, making them visually complex and providing multiple possible AOIs. Each of the three paintings shows a different scene: a group portrait in Schuttersmaaltijd, a still life in Pronkstilleven, and a landscape view of a village in Winterlandschap. This variation in content allows us to generalize our findings above and beyond the specific paintings included in the study.

Fig. 6.

Paintings and AOIs mentioned in the descriptions. Note. a: Schuttersmaaltijd (1648 , 232 × 547 cm.) ; CDC: Banner ( Vaandel ), Man in suit ( Man in pak ); ADC: Drum ( Trommel ), Handshake ( Handen schudden ), Drinking hoorn ( Drinkhoorn ). b: Pronkstilleven (1644 , 186 × 243 cm.) ; CDC: Parrot ( Papegaai ), Ham ( Ham ), Monkey ( Aapje ), Dog ( Hondje ); ADC: Pie ( Pastei ), Glass and Dishware ( Glas en vaatwerk ); mentioned in both CDC and ADC: Lobster ( Kreeft ). c: Winterlandschap (c. 1608 , 77.3 × 131.9 cm.); CDC: Skaters ( Schaatsers ); ADC: Horse ( Paard ). Painting images were downloaded from Rijksmuseum.nl and are used in accordance with the Rijksmuseum’s policy.

Painting descriptions

Different painting descriptions were used for each condition (Supplementary Information). The descriptions used for participants in the ADC corresponded to the labels displayed in the museum, located next to the three paintings. For the CDC, new descriptions were created by the educational staff of the museum, who specifically tailored the descriptions to 10-12-year-olds. Compared to the adult descriptions, the child descriptions were written from the viewpoint of specific characters in the paintings, thus relying on a “storytelling” format.

From a linguistic point of view, both descriptions were similar. Both were approximately 60 words long (Child’s descriptions: M = 60.67, SD = 1.25; Adult’s descriptions: M = 57.67, SD = 5.73).

Descriptions’ average word frequency was computed by retrieving descriptions’ constituent word frequencies, whereby the value zero was attributed to words that do not occur in the database. Word frequencies (derived from the Dutch Lexicon Project database²⁶) were approximately equal for both descriptions (Child’s descriptions: M = 3.39 Zipf, SD = 0.05; Adult’s descriptions: M = 3.42 Zipf, SD = 0.16). However, the average word length was shorter in the child descriptions than in the adult descriptions, at respectively M = 4.46 (SD = 0.08) versus M = 5.11 (SD = 0.16) characters per word.

Areas of interest (AOIs)

Specific elements mentioned in the descriptions (e.g., a dog, a drum, an ice skater) were used to define AOIs. These allowed us to examine whether eye movements were influenced by the information participants received. Since children might be interested in different aspects of the painting than adults, the AOIs mentioned in the child description (7) differed from AOIs mentioned in the adult description (6). One AOI, defined in painting Pronkstilleven, was present in both descriptions. In total, we defined 14 AOIs over the three paintings (Fig. 6).

Apparatus

Participants’ eye movements were recorded using Tobii Pro Glasses 2, a wearable mobile eye tracker with an accuracy of 0.62° and a precision of 0.05°. The eye tracker was equipped with two infrared cameras for each eye, and a wide-angle full HD scene camera to record the external world. The glasses were connected to a recording unit and a laptop, running the Tobii Pro Glasses Controller software²⁷. Participants’ gaze behavior was extracted and analyzed using Tobii Pro Lab software, version 1.194²⁸.

We processed the data using the Tobii I-VT Attention Filter with the minimum fixation duration set to 60 ms. and a velocity threshold of 100°/s. A fixation duration of 60 ms. appears appropriate for this setup, as it allows for the detection of shorter fixations, enhancing the analysis’ sensitivity to brief periods of attention.

In this context, a glance is defined as the combination of fixations and saccades directed towards a specific location. Therefore, glance duration – the total time spent viewing a specific location – was computed as the sum of the duration of all fixations on that location.

Aesthetic appreciation

We measured participants’ aesthetic appreciation through a modified version of the AESTHEMOS questionnaire²⁹. AESTHEMOS assesses the intensity of aesthetic emotions experienced towards aesthetic stimuli such as art. The original questionnaire was developed for adults and consisted of 42 questions. These questions comprised 21 subscales, providing a broad range of emotions. The reliability scores for these 21 subscales range between α = 0.55 and α = 0.85²⁹. From the 21 subscales, Schindler et al. (2017)²⁹ established four main subclasses of emotion: prototypical emotions, pleasing emotions, epistemic emotions, and negative emotions.

To make the instrument more suitable for 10-12-year-olds, the original AESTHEMOS was adapted to a shorter 12-item version, and the questions were translated from English to Dutch. The adapted version included at least two questions relating to each of the four subclasses of emotion. For each aesthetic emotion, participants were asked: “how intensely did you feel this emotion?”. Answers were given on a Likert scale from 1 (not at all) to 5 (very). The total scores ranged from 12 to 60. A score of 12 represented almost no aesthetic emotion intensity, and 60 represented extremely high aesthetic emotion intensity.

Procedure

Participants were recruited at the central lobby of the Rijksmuseum. Before participating, the children and their parents/caregivers received a general explanation of the study. The parents/caregivers were then asked to sign informed consent before participation. The children were then guided to the museum gallery for testing. Here the children received further instructions, and parents/caregivers were asked not to interfere during the experiment. Next, the experimenters helped the children to wear the eye tracker. The latter was then calibrated, using the one-point calibration of the manufacturer’s software.

The study lasted approximately 30 min and took place in two phases. In the first phase, children were instructed to view each of the three paintings freely for 15 s. To minimize unintentional viewing of each painting, participants were instructed to look at the floor when approaching the paintings and wait with their backs towards the artworks. Tape was placed on the floor to mark the position where the children should stand. Given the varying sizes of the paintings, they were viewed from different distances: Schuttersmaaltijd was observed from 430 cm. with a corresponding visual angle of 30.22° x 64.95°; Pronkstilleven from 280 cm., visual angle 36.75° x 46.91°; and Winterlandschap from 145 cm., with a visual angle of 29.85° x 48.91°. The researchers made sure that museum visitors did not block the view of the paintings. When the view was clear, participants could turn around and view the paintings for 15 s. The three paintings were located in different rooms on the same floor, with less than a minute walking time between each painting.

In the second phase, participants were asked to view the three paintings again, but this time for 30 s each. Before viewing each painting, one of the experimenters read aloud a short description of the painting. The description given depended on the participant’s condition. In the ADC, participants heard the adult descriptions (Supplementary Information), while in the CDC, they heard the child descriptions (Supplementary Information). In the FVC, participants were not provided with any painting descriptions, and simply viewed the three paintings again. After viewing each painting, participants were asked to face the researcher to answer the items in the adapted AESTHEMOS. All answers were recorded on paper by the experimenters.

After phase two, children were asked to sit on a bench in the museum gallery. The experimenters showed a laminated A4 picture of the first painting and asked the open question “what struck you about this painting?”. This procedure was repeated for all three paintings. The participants spoke their answers out loud, and the researcher would repeat and specify the children’s answers if they were unclear. Answers were registered with a voice recorder. After the experiment, participants were returned to their parents, debriefed, and any questions were answered.

Two researchers oversaw the testing. One set up the eye tracker, managed the eye tracking recording and timed the viewings. The other gave instructions, read the descriptions, and conducted the AESTHEMOS and open questions. All tasks were counterbalanced across the researchers. To reduce order effects, the order in which the paintings were viewed was either from Schuttersmaaltijd to Winterlandschap to Pronkstilleven or vice versa, counterbalanced between the participants. The order we used respected the physical layout of the museum, ensuring that participants were not inadvertently exposed to any painting outside the test phases, thus avoiding unwanted repeated-exposure effects.

Data analyses

Raw data processing

Glance duration was standardized across the three paintings. More specifically, participants’ glance duration was divided by their total viewing time, compensating for minor differences between participants.

The ratio between AOIs and painting size differed between paintings and between the adult and child descriptions. To compensate for these differences, we computed normalized glance durations for each painting and each description. These values represent the time-corrected glance duration (see above) multiplied by the painting size and divided by the sum of the areas of the relevant AOIs.

Furthermore, a total score of participants’ aesthetic emotion intensity was calculated, as well as separate scores for each of the four subclasses of the AESTHEMOS questionnaire.

Lastly, the audio recordings of answers to the open question were transcribed using the speech-to-text function in Microsoft Word, followed by a manual check to eliminate mistakes.

Salience maps and baseline-corrected salience values

To explore the impact of bottom-up factors on eye movement behavior, salience maps for each painting were generated using the Salience Toolbox³⁰. This analytical tool assesses paintings based on low-level intensity, color, and orientation channels, resulting in a map that delineates regions standing out in terms of these features. Bright areas on the map denote relatively salient regions. As in Walker et al. (2017)²⁵, we utilized an intermediate stage in the salience modeling procedure to produce maps without inhibitory processes, fostering smoother maps and heightened salience variability across the images³¹. The resulting salience maps were normalized to values between 0 and 100.

An established method to determine the predictive capacity of the salience model is to compare participants’ actual locations fixated on the paintings with the salience model values at corresponding positions. This measure is commonly known as “salience at fixation”³¹. For this study, salience at fixation was defined as the average of salience map values within a 40 × 40 pixel region surrounding each participant’s first six fixation locations. To determine whether low-level image features exerted influence beyond chance expectations on fixation location, salience at fixation was compared to a baseline level, calculated as outlined in Walker et al. (2017)²⁵. In this context, values above 0 signify higher-than-expected salience at fixation, while values below 0 indicate lower-than-expected salience at fixation³². These baseline values were computed separately for each phase. Given that visual salience only plays a role during initial fixations^[e.g.,33], our analysis focused exclusively on the first six fixations made by participants.

Speech analysis

We hypothesized that the language of participants in CDC would show signs of more active cognitive engagement with the paintings compared to participants in the other two conditions. To explore this question, we created a corpus of children’s answers to the question “what struck you about this painting?”. The analysis considered not just the content of the children’s descriptions but also the way they articulated their thoughts.

Answers were recorded in individual text files, and speech by researchers was excluded. The resulting text was then tokenized and subjected to Part-of-Speech (POS) tagging using spaCy, a well-known open-source library in Python for natural language processing (NLP). Using the resulting processed speech, we analyzed differences between the three experimental conditions in terms of word count and the relative frequency of different POS. We also counted the frequency with which participants pointed at areas of the paintings without necessarily describing them verbally.

Exclusions

The total sample (N = 62) was used for analyzing the AESTHEMOS data and participants’ answers to the open question. Due to missing data, a smaller sample of 56 participants was used to investigate children’s glance duration and visual saliency.

For recordings with less than 85% gaze samples, researchers manually inspected where data loss occurred. This was important, since the total eye tracking recording included the time participants took to move between paintings. Therefore, data loss could occur outside the relevant experimental phases. As a result of this procedure, a few participants were excluded due to missing eye tracking data (n = 4) and low gaze samples (n = 2), resulting in 19 participants in CDC, 20 in ADC and 17 in FVC.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

FW: conceptualization, methodology, validation, formal analysis, resources, data curation, writing – original draft, visualization, supervision, project administration. BB and JS: conceptualization, methodology, writing – review & editing. NA and ZP: formal analysis, data curation. KH and RVdB: investigation, data curation. PK and IdV: resources, project administration. JT: conceptualization, methodology, writing – review & editing, funding acquisition.

Data availability

All data is freely available and can be accessed at the following link: https://www.doi.org/10.17605/OSF.IO/D7YUW.

Declarations

Competing interests

The authors declare no competing interests.