Population matrices
We obtained tropical tree population matrices for 58 species from the COMPADRE Plant Matrix Database (Salguero-Gómez et al. 2015). Population matrices were included in the analysis if they (1) were classified as a ‘tree’ or ‘palm’, (2) were from tropical or subtropical regions, (3) did not include clonal reproduction, (4) included life stages that could be categorized as ‘seedling’, ‘juvenile’, and ‘adult’ stage classes, (5) included transitions between major stages, and (6) included populations in ‘unmanipulated’ or ‘unmanaged’ habitat treatments. For species with multiple population matrix studies that fit these criteria, we selected the study that included the most populations. For studies with multiple matrices of a single species (N = 25), we averaged matrix elements across populations to create a composite representative matrix for that species. Averaging matrices across populations did not affect the ultimate results of our study (Table S1 in supporting information).
We reduced the sizes of each matrix to 4 × 4 with the following stages: seed, seedling, juvenile, and adult. If a matrix had multiple sub-classes within a single stage (e.g. multiple ‘adult’ size classes), we combined stage classes such that the new composite matrix had the same stable stage distribution and the same asymptotic growth rate as the original matrix (Yearsley & Fletcher 2002). Some studies considered ‘seedlings’ to be plants with a diameter at breast height <1 cm, while in most other studies ‘seedlings’ were individuals under a certain height (though the exact threshold varied across species). For our ‘seedling’ stage, we followed the definitions used by each study. ‘Juvenile’ encompassed non-reproductive stages between seedling and adult. If matrices lacked a seed stage (i.e. the fecundity element in the matrix was the number of seedlings produced), we used seed survival values from the original studies to add a seed stage. For species that lacked seed survival data, we created a seed stage using the average seed survival value among non-cycads or cycads. We divided our data in this way because cycads had significantly higher seed survival than other major lineages (i.e. conifer, magnoliid, eudicot, monocot), but seed survival was not significantly different among other clades (Figure S1). Using averaged seed survival values did not affect the ultimate results of our study (Table S1). No regression transitions, such as juveniles becoming seedlings, were allowed.