Environmental variables
To test whether there are differences in niche width and range size between species with different nest types we used geographical distribution information from BirdLife (2019). We used a published dataset (Cally et al. 2021) of 19 bioclimatic variables from Worldclim (Fick & Hijmans 2017) (details in Cally et al. 2021). These variables were sampled in 1000 random points across the distribution of each species, and provide information on temperature and rainfall across the range. We also extracted information on range size from Cally et al. 2021 (n=3174). For a smaller set of species for which breeding range information was available (n=3049), we extracted information on the same 19 bioclimatic variables following the protocol in Cally et al. 2021, but this time we used a more recently developed dataset (CHELSA), which has an algorithm that predicts precipitation patterns more precisely than Worldclim (Karger & Zimmermann 2019). For each climatic variable, and each species, we calculated the standard deviation across the 1000 points sampled, to estimate climatic variability across the species range (and breeding range). Since species restricted to islands are limited both in the extent of their range and the niche width, we performed analyses using both the whole dataset and only continental species.
To summarise information on variation in temperature and variation in precipitation across a species range, we performed two principal component analyses, one for temperature and one for precipitation variables. We split precipitation from temperature because we expected that nest types would be more linked to temperature than precipitation variables, given the proposed thermoregulatory capabilities of domed nests (Martin et al. 2017). The same was done for the breeding range, leading to four principal component analyses that summarise information on how variable temperature and precipitation are both within total and only breeding ranges. We report the PC loadings and percentage of variance explained for PCs. We refer to the first principal component from temperature variables as PCTEMPand to the first principal component based on precipitation variables as PCPRE. These PCs (four in total) were used each as a different response variable in statistical analyses, but many of them were very highly correlated (e.g. r2 = 0.99, supp. material Figure 1). Given that results were very similar for both breeding and whole range, we present in the main text the results from the whole range, because it holds a larger sample.