# David Firth

#### 11 packages on CRAN

Functions to facilitate inference on the relative importance of predictors in a linear or generalized linear model, and a couple of useful Tcl/Tk widgets.

Functions to Accompany J. Fox and S. Weisberg, An R Companion to Applied Regression, Third Edition, Sage, in press.

Functions and datasets to support Venables and Ripley, "Modern Applied Statistics with S" (4th edition, 2002).

Graphical and tabular effect displays, e.g., of interactions, for various statistical models with linear predictors.

Specify and fit the Bradley-Terry model, including structured versions in which the parameters are related to explanatory variables through a linear predictor and versions with contest-specific effects, such as a home advantage.

Functions to specify and fit generalized nonlinear models, including models with multiplicative interaction terms such as the UNIDIFF model from sociology and the AMMI model from crop science, and many others. Over-parameterized representations of models are used throughout; functions are provided for inference on estimable parameter combinations, as well as standard methods for diagnostics etc.

Provides additional data sets, methods and documentation to complement the 'vcd' package for Visualizing Categorical Data and the 'gnm' package for Generalized Nonlinear Models. In particular, 'vcdExtra' extends mosaic, assoc and sieve plots from 'vcd' to handle 'glm()' and 'gnm()' models and adds a 3D version in 'mosaic3d'. Additionally, methods are provided for comparing and visualizing lists of 'glm' and 'loglm' objects. This package is now a support package for the book, "Discrete Data Analysis with R" by Michael Friendly and David Meyer.

Functions to prepare rankings data and fit the Plackett-Luce model jointly attributed to Plackett (1975) <doi:10.2307/2346567> and Luce (1959, ISBN:0486441369). The standard Plackett-Luce model is generalized to accommodate ties of any order in the ranking. Partial rankings, in which only a subset of items are ranked in each ranking, are also accommodated in the implementation. Disconnected/weakly connected networks implied by the rankings may be handled by adding pseudo-rankings with a hypothetical item. Optionally, a multivariate normal prior may be set on the log-worth parameters and ranker reliabilities may be incorporated as proposed by Raman and Joachims (2014) <doi:10.1145/2623330.2623654>. Maximum a posteriori estimation is used when priors are set. Methods are provided to estimate standard errors or quasi-standard errors for inference as well as to fit Plackett-Luce trees. See the package website or vignette for further details.

Facilities are provided for fitting the simple, unstructured Bradley-Terry model to networks of binary comparisons. The implemented methods are designed to scale well to large, potentially sparse, networks. A fairly high degree of scalability is achieved through the use of EM and MM algorithms, which are relatively undemanding in terms of memory usage (relative to some other commonly used methods such as iterative weighted least squares, for example). Both maximum likelihood and Bayesian MAP estimation methods are implemented. The package provides various standard methods for a newly defined 'btfit' model class, such as the extraction and summarisation of model parameters and the simulation of new datasets from a fitted model. Tools are also provided for reshaping data into the newly defined "btdata" class, and for analysing the comparison network, prior to fitting the Bradley-Terry model. This package complements, rather than replaces, the existing 'BradleyTerry2' package. (BradleyTerry2 has rather different aims, which are mainly the specification and fitting of "structured" Bradley-Terry models in which the strength parameters depend on covariates.)