Diversification analyses and evolutionary models
We estimated speciation rates across a range of phylogenetic trees using the
cladogenetic diversification rate shift (ClaDS) Bayesian model (Malietet al. 2019) that infers branch-specific diversification rates. ClaDS allows for gradual variation in diversification rates and was shown to perform well in identifying both small and large rate shifts (Maliet et al. 2019). We used the model that accounts for a scenario with a constant turnover rate (ClaDS2 model) and the faster implementation in Julia (v.1.1.0, (Bezanson et al. 2017) that uses data augmentation (Maliet & Morlon 2020), setting the sampling probability to 0.63. To test whether nest type is associated with variation in speciation rates, we used the tip-rate estimates (estimates for each terminal branch at present) extracted from ClaDS as response variable in a PGLS model, with nest type and log(range size) as predictors, since range size has been shown to be associated with speciation (Cally et al. 2021). The nest type category used for this model was open vs. domed, given that we are interested in the actual nest building behaviour, not where the nest is placed. A main model was run using the MCC tree and we also performed the same analysis in a sample of 20 phylogenetic trees to account for phylogenetic uncertainty.
To gain a more mechanistic understanding on the feedback between species traits and their influence on diversification, we used the Multiple State Speciation Extinction (MuSSE) model to jointly estimate state dependent speciation and extinction rates (FitzJohn 2012). Modelling trait evolution while accounting for the diversification process results in more appropriate estimates of trait transition rates and ancestral estimates, while accounting for uncertainty in the species for which we do not have state information (Maddison et al. 2007; FitzJohn 2012). We modelled the correlated evolution of niche width andnest type data by aggregating them into four different states: narrow niche and domed nest, narrow niche and open nest, wide niche and domed nest and wide niche and open nest. We assigned a species as having a wide or narrow niche based on whether the PCTEMP was below or above the median value of the whole dataset (PCTEMP = -0.639). Because we are interested in the interaction among these states, we disallowed transition rates between states that effectively would represent simultaneous changes in nest type and niche width (e.g. going from wide niche and domed nest directly to narrow niche and open nest is not allowed).
Finally, to allow for unobserved taxonomic variance in the diversification process, thereby not forcing the model to explain all the heterogeneity in diversification, we ran HiSSE (Hidden State Speciation Extinction) with two concealed states (Beaulieu & O’Meara 2016). Both analyses were performed using the MCC tree within a Bayesian Framework using RevBayes, including ancestral state estimation (details in supplementary material). To evaluate the effect of phylogenetic uncertainty, we ran the Maximum Likelihood approach to MuSSE implemented in the R package ‘castor’ (Louca & Doebeli 2018) across 20 phylogenetic trees. Because phylogenetic uncertainty did not influence the results, and results were consistent across methods, we present the HiSSE analyses in the main text and castor and MuSSE results in the supplementary material.