The presence of bystanders can influence the behaviour of a forager, which has mainly been studied in primates and birds. We tested the effect of the absence and presence of an unfamiliar audience (females, males, and their combination) near or far from a food patch, on the foraging behaviour of guppies (Poecilia reticulata). Our investigation includes both males and females, recognising that different social dynamics and reproductive strategies between the sexes could lead to varied responses to audience effects. For each focal fish, we measured the latency to start feeding, bite frequency, time spent near the audience, and overall mobility. Both males and females started feeding faster when food was close to any audience type. Specifically, females exhibited a faster feeding response in the presence of a female audience. Males reduced their feeding rate in the presence of male and mixed audiences, while both sexes increased their consumption when food was close to the audience. Focal fish, irrespective of their sex, spent more time in the vicinity of the audience zone when the audience was present but, surprisingly, females spent less time with a female audience compared to others. Only females increased their mobility when the food patch was far from any audience. Here we show that guppies adjust their foraging behaviour in the presence of an audience. The specific responses observed varied between the sexes, reflecting the distinct social trade-offs faced by each sex.

This dataset was collected through a series of behavioural observation trials focusing on the foraging behaviour of guppies, differentiated by sex, in the presence and absence of an audience and when the food was close or far from the audience. The collection occurred over two periods: for female guppies in November 2018, January, and February 2019, and for male guppies in November 2019, January, and February 2020. The methodology was consistent across these periods, involving the selection of focal guppies from a stock population, which were then isolated in small tanks and deprived of food to standardize hunger levels before trials. For the collection process, guppies were selected haphazardly 24 hours before their trial, photographed for identification, and isolated in individual tanks. The trials were held in experimental tanks divided using a transparent barrier that allowed visual and chemical communication but segregated focal guppies from an "audience" of other four fish selected based on size criteria.

The processing of this dataset involved several steps:

Image Analysis: Photographs of the isolated guppies were analyzed using Image J software to measure their standard lengths, providing a quantitative basis for comparisons and ensuring that audience fish were appropriately sized in relation to focal fish.

Behavioural Recording: During the trials, behaviours were meticulously recorded by two observers (NTV, GLN), including latency to start foraging, frequency of bites, time spent near the audience, mobility, instances of aggression, frequency of courtship displays, and gonopodial swings. This data was captured using BORIS software, ensuring systematic and accurate recording.

Treatments: We tested 4 audiences: 1) No audience, 2) female audience, 3) male audience, and 4) mixed audience. We also tested the proximity of food (close or far) from the audience.

Data processing:

1. Initial Data Handling: Upon collecting behavioural observations from our experimental trials, we meticulously cleaned the raw data to ensure accuracy. This involved verifying the correctness of all entries and addressing any missing or anomalous data points through established protocols.
2. Data Analysis Software: For the analysis of our data, we used R (version 4.0.0, R Core Team 2020), which facilitated statistical modelling and graphical representation of our results. The 'survival' package (Thernau 2020) was used in our analysis of latency to start foraging using accelerated failure time regression models.
3. Statistical Modelling: Our analysis incorporated a variety of statistical methods tailored to the specific nature of each behavioural parameter observed. This included:
   • Accelerated Failure Time Regression Models: To analyse latency to start foraging, allowing us to compare survival curves over time and identify the best fit among four common survival distributions (exponential, Weibull, log-normal, log-logistic) based on the Akaike information criterion (AIC).
   • Generalized Linear Models (GLM): Employed to assess the influence of the audience and food patch position on foraging behaviour, including frequency of bites and mobility. We addressed overdispersion in the frequency of bites and time near the audience using the quasi-likelihood method, which is adept at handling variance-mean relationships.
4. Model Simplification: Following initial model fitting, we engaged in a process of backward selection to simplify our models, focusing on the removal of non-significant interactions and factors based on probability values and AIC for model comparison. This approach enabled us to refine our models to those most statistically significant and relevant to our research questions.
5. Independent Analysis of Male and Female Data: Given the temporal separation of observations for male and female guppies, we analysed the data for each sex independently to avoid confounding effects and ensure the precision of our findings.
6. Exclusion of Certain Behaviours: Due to limited occurrences, aggressive and sexual behaviours recorded in focal males were not included in the final statistical analysis. This decision was based on the premise that the low frequency of these behaviours would not yield statistically significant insights.
7. Correlation Analysis: We also conducted correlation analyses for key variables (latency to start foraging, frequency of bites, time near and far from the audience, and mobility) for each sex, with detailed results presented in the supplementary data.
8. No Multiple Testing Correction: Our approach did not apply a multiple testing correction due to the independent analysis of separate datasets for each model. This strategy allowed us to thoroughly assess the effects within each specific model without the need for cross-model comparison or adjustment for multiple comparisons.
9. Compliance with Ethical Standards: It's noteworthy to mention that our study adhered strictly to the ASAB/ABS Guidelines for the Use of Animals in Research, with all procedures approved by BUAP's ethical guidelines.