GNM: a package for generalized nonlinear models

摘要: This talk will introduce the gnm package which provides functions for specification, estimation and evaluation of generalized nonlinear models. The class models extends that linear by allowing non-linear terms in predictor. Examples include with multiplicative interaction such as row-column association from sociology GAMMI crop science; stereotype an ordered categorical response, diagonal reference dependence on a square two-way classification. main have been patterned base function glm its methods (for models), so integrates well into R useRs should find it straight-forward to pick up. includes some may be used context (generalized) models, particular setting up structured interactions between factors. However major contribution is provide facilities specification terms. From user’s perspective, this achieved through two functions: Mult common case Nonlin any other differentiable term. generality made possible features package. First can specify term use “plug-in” functions, number are provided also userdefined. Second over-parameterized representation throughout, rules applying constraints do not need defined. A set tools estimable parameter combinations their standard errors obtained after model has fitted. Although user-level small, functionality large only snapshot capabilities. more detailed overview vignette available webpage (http://www.warwick.ac.uk/go/heatherturner/gnm) or part itself, downloaded CRAN (http://cran.r-project.org).