Red for Danger, Warmth for Everything: How Children Assign Colours to Communicative Light Signals
Sandhya Rao, Sreemoyee Chakraborty, and Raunak Sharma
- DOI
- 10.5281/zenodo.22125690
- Type
- Published Record

Abstract
This study examines how children map colour onto communicative function using the openly available Firefly Design Task dataset (Rao, Sharma, & Reddy Boddu, 2026). In the task, 89 children (aged 3–15 years; 50 boys, 39 girls) each assigned a colour to three imagined light signals on a firefly: a Glow light, a Hello light, and a Danger light, named an animal they wished to communicate with, and wrote a short message to that animal. Free-text colour entries were normalised to a dominant hue and classified by warmth; animals and messages were coded into categories. Since every child provided three color responses to signal meanings, signal-colour associations were evaluated using Cochran's Q and pair-wise McNemar tests. Warm colours were abundant overall and their presence did not differ across the three signals (Cochran's Q = 1.58, df=2, p=.45). Red, in particular, was assigned to the Danger signal disproportionately more than to the Glow or Hello signals (Cochran's Q = 24.27, df=2, p<.001); 38% of the children coded Red for Danger whereas only 14% and 8% coded Red for Glow and Hello respectively (both pair-wise contrast tests to Danger were significant after adjusting for multiple testing in line with the Bonferroni adjustment procedure). The percentage of engaged participants in the experiment was high (90% (95% confidence interval, 82% – 95%) and messages were more likely to be longer as age increased (Spearman's rho = .26, p=.021). There was a marked effect of gender on message content; affectionate messages were nearly exclusively written by females (90% of these messages; Fisher's exact p=.004).The results suggest that by early childhood children reserve red specifically for danger while treating warm colours as a general-purpose palette, and they illustrate the value of open datasets for studying children's symbolic use of colour.
Keywords: colour cognition; child development; early childhood; danger signalling; human–animal communication; open data; Firefly Design Task
Introduction
Colour carries meaning long before children can articulate why. Adults readily associate red with danger, prohibition and warnings and use this knowledge for stop-signs, alarm sounds or labels warning of potential risks which holds true transculturally (Elliot & Maier, 2014; Pravossoudovitch, Cury, Young, & Elliot, 2014). Thus, a major research topic in developmental colour research is the development and timing of this symbolic mapping of specific colours for specific meanings. Do children rely on a colourful palate where warm colours signal salience and cool colours signal something else, or do they already allot colour tokens for specific functions? Play-based design tasks offer a natural window onto these mappings, because they ask children to make colour choices in the service of a communicative goal rather than to report preferences in the abstract (Zentner, 2001). The Firefly Design Task is one such activity: children imagine a firefly that communicates through coloured light and decide what colour each of its signals should be. Crucially, the task elicits three distinct signals from every child: a neutral Glow, a friendly Hello, and a Danger warning, allowing within-child comparison of colour choice across communicative functions. The activity is an embodied biomimicry design for teaching bioluminescent communication to children (Rao & Sharma, 2026).
This paper analyses the openly published Firefly Design Task dataset (Rao, Sharma, & Reddy Boddu, 2026) to ask three questions. First, do children assign colours differentially across the three signals, and in particular is red reserved for danger? Second, how do children engage with the communicative element of the task whom do they choose to address, and what do they say? Third, do colour and message choices vary with age or gender? We treat the first question as the primary confirmatory analysis and the remainder as exploratory.
Methods
Data source
All analyses use the Firefly Design Task dataset, version 2.0, deposited on Zenodo under a persistent DOI (Rao, Sharma, & Reddy Boddu, 2026; https://doi.org/10.5281/zenodo.22124612). No new data were collected for the present study. The dataset provides one record per child
comprising the child's verbatim colour choices for the three signals, the animal named, and the message written, together with derived variables supplied by the data creators. Five first names that appeared in message text had been redacted to a placeholder in the source, and this study introduces no further identifiers. The dataset was produced within a practitioner-led, longitudinal programme of embedded educational research (Blue Blocks Micro Research Institute, Goyal, Chakraborty, & Ediga, 2026), following a standardised embedded-observation protocol (Blue Blocks Micro Research Institute, 2026b) and the programme's ethics and child-data-governance standards (Blue Blocks Micro Research Institute, 2026a, 2026c).
Participants
The dataset contains 89 children. Fifty were recorded as boys and 39 as girls. Age was recorded for 86 children (mean = 6.73 years, SD = 2.18, range 3–15); three ages were missing and were excluded from age-based analyses. The age distribution was concentrated in the 6–8 year band, with a single 15-year-old.
Variables and coding
Each free-text colour entry was normalised by the data creators to a dominant basic colour and to a warmth class: warm (red, orange, yellow, pink), cool (blue, green, purple), neutral (black, grey, brown, white), or mixed (multi-colour, rainbow, gradient). A binary multi-colour flag marked entries naming more than one hue. Animals were assigned to categories (pet/domestic, wild/large, insect, bird, fantasy, firefly self-reference, all animals, unclear/other, not specified).
Messages were coded for presence, word count, and theme (greeting, question/curious, affection/friendly, warning/threat, invitation, other/unclear, none). Category coding reflects a single rule-based pass; we treat these as exploratory and discuss the implications in Section 5.
Statistical analysis
Because each child contributed a colour choice to all three signals, the three responses are statistically dependent, and pooling them would constitute pseudoreplication. We therefore tested differences in the proportion of a given colour class across signals with Cochran's Q test for paired binary data, following significant omnibus results with pairwise McNemar exact tests and Bonferroni correction for the three comparisons. Proportions are reported with Wilson 95%
confidence intervals. The non-uniformity of the Danger colour distribution was tested with a chi-square goodness-of-fit test. Associations with age used Spearman's rank correlation (message length is right-skewed) and the Kruskal–Wallis test across age bands. Gender differences in message length used the Mann–Whitney U test; gender differences in categorical message themes used Fisher's exact test, with Cramér's V as an effect-size measure for contingency tables. Analyses were conducted in Python 3 with SciPy 1.17. Given the exploratory nature of the secondary analyses, p-values there are interpreted descriptively.
Results
Sample characteristics
Table 1 summarises the sample. Boys outnumbered girls, and most children were of early-primary-school age.
|
|
|
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Table 1. Sample characteristics (Firefly Design Task dataset, v2.0).
Colour choice across the three signals
Warm colours predominated for every signal: 61% of children chose a warm colour for the Glow light, 55% for the Hello light, and 64% for the Danger light. These proportions did not differ across signals (Cochran's Q = 1.58, df = 2, p = .45), indicating that children did not treat any one signal as warmer than the others in general terms. The distribution of primary colours by signal is shown in Table 2.
The picture changed sharply when red was considered on its own. Red was the modal choice for the Danger signal (34 of 89, 38%) but was uncommon for Glow (12, 14%) and Hello (7, 8%). A Cochran's Q test on the paired red/not-red responses was highly significant (Q = 24.27, df = 2, p
< .001).
|
|
|
|
|
|---|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Table 2. Frequency of normalised primary colour by signal (N = 89 children × 3 signals).
|
|
|
|
|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Table 3. Warm-colour and red proportions across signals, with paired tests. p_adj = Bonferroni-adjusted.
Pairwise McNemar exact tests showed that red was chosen for Danger significantly more often than for Hello (Bonferroni-corrected p < .001) and for Glow (corrected p = .004), whereas Glow and Hello did not differ (corrected p > .99). The Danger colour distribution as a whole departed strongly from uniformity (goodness-of-fit χ² = 112.6, df = 10, p < .001). Table 3 reports the key proportions with confidence intervals. In short, children did not reserve warmth for danger; they used warm colours freely everywhere, but they did reserve red for danger specifically. Multi-colour and gradient responses were rare and evenly distributed across signals (three or fewer per signal).
Chosen animals
The most popular type of imagined conversant was a known animal, with pets and domestic animals (24) and wild, non-small animals (24) equally the most cited forms. Next came insects
(10). In 18 cases (20% of participants), participants identified not an animal but the firefly frequently in a stylised or technological format rather than as a clearly distinct entity to which the participant can communicate, indicating participant investment in the central role of the subject. Four children provided no animal. See Table 4 for frequency breakdown per category.
|
|
|
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Table 4. Researcher-coded animal categories (N = 89).
Messages: engagement, length, and theme
Engagement with the writing element was high: 80 of 89 children (90%; 95% CI [82, 95]) left a message. Messages were short (median = 2 words) but ranged up to 18 words. Message length increased modestly with age (Spearman's ρ = .26, p = .021, n = 77); the same trend across age bands did not reach significance in the grouped Kruskal–Wallis test (H = 4.13, df = 3, p = .25), although median length rose from 2 words in the youngest bands to 8 in the oldest. Greetings were the most common theme, followed by warnings/threats, affectionate messages, and questions.
Message length did not differ by gender (Mann–Whitney U = 825.0, p = .71; median 2 words for both). Message content, however, did. Affectionate messages were written almost exclusively by girls: 9 of the 10 affectionate messages came from girls and only 1 from a boy (Fisher's exact p =
.004, odds ratio ≈ 14.7). Warning or threatening messages, by contrast, were split evenly (6 boys, 6 girls; p = .76), as was the choice of a warm colour for the Danger signal (χ² = 1.26, df = 1, p =
.26, Cramér's V = .12). Table 5 shows theme by gender.
|
|
|
|
|---|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Table 5. Message theme by gender. * Fisher's exact test, significant after considering the exploratory context.
Discussion
Two findings stand out. The first is a dissociation between warmth and hue. Children in this sample did not treat the Danger signal as warmer than the other warm colours were the default
across all three signals, yet they singled out red for danger with striking specificity. This pattern is consistent with an early, culturally reinforced red–danger association (Elliot & Maier, 2014; Pravossoudovitch et al., 2014) operating on top of a broad preference for bright, warm colours characteristic of early childhood (Zentner, 2001). It suggests that the red–danger link is not merely a by-product of red being warm and salient; if it were, orange and yellow should have been elevated for Danger too, and they were not.
The second finding is a gendered pattern in message content but not in colour or message length. Girls were far more likely than boys to address the animal affectionately, whereas warnings and threats were produced equally by both. This dissociation a gender difference in emotional framing without a corresponding difference in engagement or in the danger-colour mapping indicates that the gender effect lies in social-emotional expression rather than in the colour-cognition process that is the study's primary focus.
The prominence of firefly self-reference (one child in five) is a reminder that children readily inhabit the premise of a design task, sometimes elaborating the protagonist ("future firefly 3.2") rather than answering literally. Such responses are a feature of imaginative engagement, not noise, and they merit attention in any study using narrative design prompts with children.
Limitations
Several limitations temper these conclusions. The sample is a modest convenience sample with an uneven age distribution and no accompanying metadata on setting, prompt wording, or administration, so generalisation beyond the task is uncertain. Colour, animal, and message variables were derived by a single rule-based coding pass; some entries were genuinely ambiguous, and independent double-coding with an inter-rater reliability statistic (e.g., Cohen's κ) would strengthen the categorical analyses. The secondary analyses are exploratory and were not corrected for the full family of comparisons, so the gender and age effects in particular should be treated as hypotheses for confirmation in a pre-registered study. Finally, participant codes in the source are not unique and some are placeholders, so the data should not be used to track individuals across records.
Conclusion
Using an open dataset of children's responses to a firefly design task, we found that children aged 3–15 apply warm colours broadly but reserve red specifically for a danger signal, a dissociation that points to an early and selective red–danger mapping rather than a general warmth heuristic. Children engaged readily with the communicative element of the task, and while message length grew with age and did not differ by gender, affectionate content was overwhelmingly produced by girls. These results demonstrate how a compact, well-documented open dataset can support quantitative developmental inference, and they offer concrete hypotheses — the hue-specificity of the red–danger link, and the social-emotional locus of gender differences — for confirmatory study.
Data Availability
The data analysed in this study are openly available from Zenodo: Rao, S., Sharma, R., & Reddy Boddu, S. (2026). The Firefly Design Task: A Dataset of Children's Colour-Signal, Animal, and Message Choices (Version 2.0) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.22124612.
Analysis code is available from the author on request.
References
Blue Blocks Micro Research Institute. (2026a). Micro Research Ethics Framework (MREF v1.0): An ethics standard for embedded educational research by Blue Blocks Micro Research Institute. Zenodo. https://doi.org/10.5281/zenodo.19047669
Blue Blocks Micro Research Institute. (2026b). Blue Blocks Embedded Observation Protocol (BEOP v1.0): A standard for practitioner-led observation in educational settings. Blue Blocks Micro Research Institute. https://doi.org/10.5281/zenodo.19087415
Blue Blocks Micro Research Institute. (2026c). Child Data Classification Standard (CDCS v1.0): A data governance standard for child development research. Blue Blocks Micro Research Institute. https://doi.org/10.5281/zenodo.19202499
Blue Blocks Micro Research Institute, Goyal, P., Chakraborty, S., & Ediga, S. (2026). Blue Blocks Micro Research Methodology: A Practitioner-Led, Longitudinal Framework for Embedded Educational Research. Zenodo. https://doi.org/10.5281/zenodo.18584816
Elliot, A. J. (2015). Color and psychological functioning: A review of theoretical and empirical work. Frontiers in Psychology, 6, 368.
Elliot, A. J., & Maier, M. A. (2014). Color psychology: Effects of perceiving color on psychological functioning. Annual Review of Psychology, 65, 95–120.
Pravossoudovitch, K., Cury, F., Young, S. G., & Elliot, A. J. (2014). Is red the colour of danger? Testing the relationship between red and perceived danger. Ergonomics, 57(4), 503–510.
Rao, S., & Sharma, R. (2026). Becoming Fireflies: An Embodied Biomimicry Activity for Teaching Bioluminescent Communication to Children Aged 9–12. Zenodo. https://doi.org/10.5281/zenodo.22094474
Rao, S., Sharma, R., & Reddy Boddu, S. (2026). The Firefly Design Task: A Dataset of Children's Colour-Signal, Animal, and Message Choices (Dataset) (Version 2.0) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.22124612
Zentner, M. R. (2001). Preferences for colours and colour–emotion combinations in early childhood. Developmental Science, 4(4), 389–398.