A comparison of NDM and PAE using real data.

J. SALVADOR ARIAS; M. DOLORES CASAGRANDA; J.M. DÍAZ GÓMEZ

CLADISTICS (PRINT)

WILEY-BLACKWELL PUBLISHING, INC

Lugar: Oxford; Año: 2010 vol. 26 p. 4 - 4

0748-3007

Although identification of areas of endemism is an important step in a biogeographical analysis, explicit methods for searching this patterns as PAE and NDM, follow very different criteria for its definition, and its application might result in identification of dissimilar areas of endemism from the same data set. Whereas a few a comparative analyses between methods for delimitation of areas of endemism has been done, up to now there are not numerical studies that formalize differences between those methods when real data are used. In order to do these comparison we tested three methods for delimitation of areas of endemism: PAE (Morrone, 1994); PAE-PCE (Garcia-Barros et al., 2002), and NDM as proposed by Szumik and Goloboff (2004), on 20 published distributional data, and compared the stability of results. Stability is measured by the number of areas of endemism found in a jackknife resampled data sets, which are also found in the complete matrix. Two kind of jackknife were performed, a ´taxon´ jackknife, which removes a whole taxon´s distribution from the data, and a ´cell´ jackknife which removes cells scored as presence in the data matrix. The deletion probability is 1/e (the logn base, roughly 36%). We found that (1) NDM finds more areas than PAE, or PAE-PCE; (2) PAE-PCE does not generally find more areas than those obtained with PAE; (3) NDM results were the most stable to taxa deletion, while both PAE and PAE-PCE stability is poor; (4) Similarly, cell jackknife does not affect in a relevant way the number of stable areas obtained with NDM, but it greatly affects PAE and PAE-PCE performance. Our results highlight the limitations of parsimony, or its modifications, for the identification of areas of endemism, especially when distributional data sampling is "incomplete", as occurs in real distributional data.