Comparison of statistical tests for habitat associations in tropical forests: A case study of sympatric dipterocarp trees in a Bornean forest

Habitat difference is an important mechanism for maintenance of tree diversity in tropical forests. The first step in studies of habitat difference is to statistically analyze whether the spatial distributions of tree populations are skewed to species-specific habitats; this is called a habitat association test. We propose a novel habitat association test on the basis of the probability of tree occurrence along a continuous habitat variable. The test uses torus shift simulations to obtain a statistical significance level. We applied this test to 55 common dipterocarp species in a 52-ha plot of a Bornean forest to assess habitat associations along an elevation gradient. The results were compared to those of three existing habitat association tests using the same torus shift simulations. The results were considerably different from one another. In particular, the results of two existing tests using discrete habitat variables varied with differences in habitat definitions, specifically, differences in elevation break points, and the number of habitat classes. Thus, definitions of habitats must be taken into account when habitat association tests with discrete habitat variables are used. Analyses of artificial populations independent of habitat showed that all of the tests used were robust with respect to spatial autocorrelation in tree distributions, although one existing test had a higher risk of Type I errors, probably due to the use of multiple tests of significance. Power analysis of artificial populations in which distributions were skewed to certain elevations showed that the novel test had comparable statistical power to the most powerful existing test. Statistical power was affected not by the total number of a given tree but by the number of clumps in a plot, suggesting that >5 clumps were required for a reliable result.