Investigating differences in multivariate phenotypes between
groups and hybrid populations
We used a Bayesian mixed-effects model to investigate if trait means
varied between Caribbean, SSI generalist-only, and SSI radiating groups
while accounting for the effects of species and population. We used the
brms function from the BRMS package to fit a model with a student’s T
distribution, and uninformative priors (Bürkner 2017). The model
included the 18 trait measurements as the multivariate response
variable, group type and species designation as fixed effects, and
population ID as a random effect.
We fit a Bayesian fixed-effects model with a student’s T distribution,
and uninformative priors to investigate differences in trait values
between Crescent Pond and Little Lake F2 hybrids. This model included
the 18 trait measurements as the multivariate response variable and
population as a fixed effect. While testing for differences in trait
values between hybrid groups is not a primary goal of this study,
detected differences could indicate population level differences in the
genetic architecture of these traits. Furthermore, the output of this
model provides estimates of uncertainty around average trait values
which were used to investigate departures from null expectations.
Specifically, we used these estimates to determine which phenotypes
deviated from additive trait value expectations (Appendix Table E4)
For both models we extracted posterior samples and estimated the median
values and 95% highest density intervals (HDI) to make inferences about
differences in trait values between the levels of our fixed effects
using functions from the brms and emmeans packages (Lenth 2023; Figures
3 & 7).