Hi all, I am trying to model (continuous) spatial variation in a response variable as a function of one or more of several explanatory variables. I am principally interested in obtaining some measure of the relative "importance" of the explanatory variables. I have found several R libraries that are tailored to this sort of problem (geoR, geoRglm, gstat, etc.); however, as near as I can tell, the appropriate functions (e.g., krige.bayes in geoR), only allow for linear modelling of the large-scale variation. In my case, covariate effects are probably nonlinear. Moreover, although there is theory to suggest the general form of some of these relationships (e.g., "monotonically increasing"), there is no theory that specifies the specific functional form. I have two questions:
(1) Are there R functions available that allow one to model nonlinear covariate effects on the large scale variation within a "universal kriging" type framework? (2) Are there any (spatial or otherwise) regression models in R that could be used to fit models where only general restrictions are placed on the functional form? I am thinking of something like a GAM, where the fit is constrained to be monotonically increasing, to be monotonically increasing with a negative second derivative, etc. Thanks in advance. Regards, Sean ******************************************** Sean R. Connolly, PhD Senior Lecturer Department of Marine Biology James Cook University Townsville, QLD 4811 AUSTRALIA Ph: 61 7 4781 4242 Fax: 61 7 4725 1570 http://www.jcu.edu.au/school/mbiolaq/mbiol/staff/sconnolly.html VISIT THE NEW CENTRE FOR CORAL REEF BIODIVERSITY at http://www.jcu.edu.au/school/mbiolaq/ccrbio/ ********************************************* "We are raised to honor all the wrong explorers and discoverers-- thieves planting flags, murderers carrying crosses. Let us at last praise the colonizers of dreams." -- Peter Beagle [[alternative HTML version deleted]] ______________________________________________ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help
