Statistical analysis
Thirty-three stickleback (size range 4.0-4.1) were used for this analysis. Data from day three of the experiment was missing for four fish resulting in a total number of observations, n = 95. We used the software R v. 3.1.3 (http://www.r-project.org) (R Core Team 2012) for all statistical analysis and visualization. Preliminary data exploration showed that latency to explore and latency to forage were highly correlated and that foraging at a sheltered or open food patch did not appear to differ between individuals or day or trial. Therefore, statistical analysis was focused on latency to explore and activity. The statistical analysis and data visualization were done in R 4.0.3 (R Core Team 2021), tidyverse 1.3.0. (Wickham, 2019) and the packages cited below. All gathered data is presented in Supplementary Table 1.
First, we used lme4 1.1-27.1. (Bates et al., 2015) to fit a generalized linear mixed model using a Poisson distribution to examine change in latency to explore and activity across the three daily behavioral trials and to get estimates of intercepts and slopes for each fish for both behaviors. This allowed us to reduce the measurements of the three trials too two values. The intercept represents the rank level of that individual and in the current context highly reflects the initial response of the fish to the trial, for example, if they started exploring/foraging early in the first trial. The slope represents the plasticity of the individual across trials and in the current context is likely to reflect habituation to the trial, for example, a high negative slope for latency to explore will show a fish that was initially hesitant (first trial) but subsequently bolder. Note that fish that were either bold or shy for all trials would both have estimated slopes close to zero.
Second, we examined the correlation of individual behavior and stable isotope niche. As estimates of the individual personality we used intercept and slope for both latency to explore and activity. Slope was included to also capture plasticity in behavior across trials. We used two approaches to do this. First, we used a generalized linear model to estimate the significance of the correlation of stable isotope values (both δ13C and δ15N) to either intercept or slope of the behavioral traits. A Poisson distribution was used in these models. Second, we wanted to see if behavior correlated to within individual variation in stable isotope niche. For that purpose, we calculated the Euclidian distance from individual stable isotope values of muscle samples and fin samples (reflecting different turnover times). We then used that distance as the response variable in two linear model with the intercept and slope of each of activity and latency to explore as factors with an interaction effect. A Gaussian distribution was used in this model. The residual distributions of all models were examined using the R package DHARMa 0.4.4. (Hartig 2021).