Modelling life-history (co)variation
We used univariate linear mixed-effect models to quantify the sources of variation in the individual estimates of generation time, age of first successful reproduction, lifespan, reproductive rate, lifetime reproductive success and individual growth rate (see Appendix S1B for model equations). All these models had sex as fixed effect and random intercepts for population identity (n = eight populations) and cohort (i.e. birth year; n = 16 cohorts). We assumed normally distributed errors in these analyses. As a second step, we fitted a multivariate model to study the phenotypic covariance among individuals in the different life-history traits (see Appendix S1C for equations). This model did not contain any random or fixed effects as we were interested in estimating the overall phenotypic covariance among individual life-history traits regardless of the cause. We thus estimated the 6x6 “residual” variance-covariance matrix.
We also fitted an animal model (Kruuk 2004) to decompose individual differences in generation time into additive genetic variation versus environmental variation (see Appendix S1C for more details). This model only had sex as fixed effect and cohort year as random effects and we assumed normal error distribution. The pedigree consisted of 3116 individuals with a mean of 4.6 ancestral generations.