Statistical analysis:
All statistical analyses were conducted in RStudio (R Core Team 2021, version 4.1.2) and all models were built using the lme4 package (Bates et al. 2015). Model assumptions of residual normality and homoscedasticity were checked visually via residual vs fitted and Q-Q plots, with log transformations made as necessary to best meet assumptions. To compare initial egg mass across populations, we used a linear mixed-effects model (LMM) including a fixed-effect of site, while controlling for clutch as a random effect. To determine whether hatch probability or survival to 10-DPH differed across temperatures or sites, we used a generalized linear-mixed model with a binomial distribution including temperature, site, and their interaction as fixed effects with clutch nested within site as a random effect. Post-hoc pairwise comparisons were conducted using the emmeans package (Lenth et al. 2023) with Kenward-Roger degrees of freedom and correcting for multiple testing using Tukey’s method.
To test for differences in the relative contributions of egg mass and incubation temperature to phenotypic traits across populations, we constructed separate LMMs for every phenotype at each site. In every model, we included fixed effects of egg mass and incubation temperature, with the exception of incubation duration, in which we only included a fixed effect of incubation temperature, while controlling for clutch as a random effect. We then compared model estimates across populations by extracting beta values and 95% confidence intervals (CIs) using the confint function in R. Model beta estimates in which CIs did not overlap zero or another population were considered statistically significant.
To further examine how embryos respond to temperature and maternal provisioning across populations, we used the ggeffects package (Lüdecke 2018) to predict temperature-specific mean values of each phenotype at a common egg mass, corresponding to the average egg mass across the dataset (83g), from each population-specific model. By comparing egg mass-corrected mean phenotypes across temperatures and populations, we were able to determine whether populations differed in mean trait values irrespective of egg mass at either or both temperatures and whether variation in the influence of incubation temperature was driven by phenotypic differences at 29.5°C, 33.5°C, or both. Mean values in which 95% CIs did not overlap were considered statistically significant.
Given that populations can also vary in how maternal resources are allocated towards particular phenotypes, we compared ratios of SVL, TG, HL, HW, residual yolk mass and fat body mass to hatchling mass across populations within and across temperatures using LMMs. For this analysis, we included temperature, site, and their interaction as predictors, along with egg mass as a covariate, while controlling for clutch nested within site as a random effect. We then compared predicted mean values from the model within and among temperatures across populations using the emmeans package. Values in which CIs did not overlap were considered statistically significant. We used ratios of traits to hatchling mass instead of egg mass for this analysis because there were significant differences in temperature-specific mass across populations (see below), and as a result, differences in the ratio of traits to egg mass would be confounded by population-specific effects of temperature on mass and may not represent differences in the allocation of maternal resources towards specific phenotypes. All figures were created using the R package ggplot2 (Wickham 2016).