We have done a multiple regression (glm) with non-orthogonal design and two correlated explanatory variables, and poisson errors, because the response variable is a count (number of species). The parameter estimates calculated were opposite to the tendency observed in the plot. As we excluded the most influencial point, the results changed radically, but as we took the second influential point the results changed again, to a third form. Thefore I tried to use robust regression, with rlm. My questions are: 1 - how do I get the test of bias of the robust regression? In the S-Plus guide (p. 338), it is said that this test should be used to choose between the robust regression and the other one. Is this correct? 2 - when I run an anova, comparing the complete rlm model to a more simple model, I get no values for F nor P. How do I get these? 3 - may I use robust regression for count response variables, taking the log of the data? Thank you all! Carlos Sperber, Carla Galbiati & Carla Ribas Carlos Frankl Sperber Ecology - Orthoptera: Grylloidea Departamento de Biologia Geral Universidade Federal de Vi?osa 3.570-000 - Vi?osa - MG Brazil tel: +55 (31) 3899 2556 fax: +55 (31) 3899 2549 E-mail: sperber at ufv.br -------------- next part -------------- ---