Life-History and Ecological Traits of Bintan Bird Species To investigate the correlates of local extinction of bird species in Bintan, we measured 10 life-history and ecological traits for each species. The data we used came from a global bird-ecology database (Sekercioglu et al.2004), which we updated with data from recent publications (del Hoyo et al. 2004–2008). The 10 traits were Indomalayan realm endemism (endemic or not endemic);forest specialist (forest most preferred habitat among 14 different habitats; specialist or nonspecialist); mass(average body mass); disturbed habitat use (occurs in disturbed habitats, e.g., secondary growth; yes or no);
habitat breadth (observed number of habitat a species occurs in; range 1–10); diet type of a species (primary diet type: invertebrates; vertebrate and fish; fruits, nectar and seeds); diet breadth (observed number of major diet type a species has; range 1–7); nest type (exposed [no nest,platform, saucer, scrape] or not); nest substrate (ground or aboveground [e.g., shrub, tree]); and minimum clutch size (minimum number of eggs laid; range 1–6).
We used a classification-tree procedure to examine the key correlates of local extinction. Classification trees repeatedly partition the data set into twomutually exclusive groups (i.e., binary splitting) to generate a tree-like classification and each split partitions deviance in successively smaller subsets of values of the dependent variable on the basis of a threshold value of an independent predictor (McCune & Grace 2002).We preferred tree modeling
to logistic regression because tree modeling is free from assumptions of parametric methods and provides analysis of complex ecological data, including missing values,nonlinear relations between variables and high-order interactions (Breiman et al. 1984). The response (i.e., extinct
or extant) and predictors (i.e., the 10 traits of life history or ecology; Supporting Information) were used to grow an overlarge tree, which was subsequently pruned to the optimum tree size (i.e., a tree size that minimizes the cost-complexity measure by snipping off the least important splits and hence reducing data overfitting and is within 1 SE of the minimum-error tree) through 10-fold cross-validations. We used the library rpart in Program R
(R Development Core Team 2008) with a minimum splitting group of size one and cost complexity measure of 0.0001.