• "Vector Generalized Linear and Additive Models: With an Implementation in R" by T. W. Yee (2015), Springer, New York, USA. This book has a webpage here.
• "The VGAM package for capture--recapture data using the conditional likelihood" by T. W. Yee and J. Stoklosa and R. H. Huggins (2015), Journal of Statistical Software,, 65, 1--33. The paper can be downloaded here and serves also as a vignette of my VGAM package. Here is the R code: v65i05.R, GAM_th_sims_prog.R. Nb. it requires VGAM 0.9-4 or higher, and VGAMdata 0.9-4 or higher; try my prerelease version and VGAMdata for the latest versions.
• "Row-column interaction models, with an R implementation" by T. W. Yee and A. F. Hadi (2014), Computational Statistics, 29, 1427--1445. Here is the code which can be cut-and-pasted into R. Nb. it requires VGAM 0.9-4 or higher, and VGAMdata 0.9-4 or higher; try my prerelease version and VGAMdata for the latest versions.
• Here is RCIMeg.R.
• Note that the function cqo() uses argument I.tolerances instead of ITolerances, and argument eq.tolerances instead of EqualTolerances. The defaults are eq.tolerances = TRUE and I.tolerances = FALSE.
• "Reduced-rank vector generalized linear models with two linear predictors" by T. W. Yee (2014), Computational Statistics & Data Analysis, 71, 899--902.
• A file with the R code is script_azpro.R.
• A file with the R code is script_nmes.R.
• A file with the R code is script_hormone.R. Note: the function normal1() has been renamed to uninormal(). Note: the function identity() has been renamed to identitylink(). Note: this example requires VGAM 0.9-4 or higher, and VGAMdata 0.9-4 or higher.
• Correction: the power-of-the-mean model cannot be fitted with the VGAM package.
• Most of the paper uses VGAM 0.9-2 or higher, and VGAMdata 0.9-2 or higher. The latest version is here (prerelease form).
• "Scoring rules, and the role of chance: Analysis of the 2008 World Fly Fishing Championships" by T. W. Yee (2014), Journal of Quantitative Analysis of Sports, 10, 397--409. Here is the code which can be cut-and-pasted into R. Nb. it requires VGAM 0.9-3 or higher, and VGAMdata 0.9-3 or higher; see my prerelease version.
• "The VGAM package for categorical data analysis" by T. W. Yee (2009), Journal of Statistical Software, 32, http://www.jstatsoft.org/v32/i10/. This paper is extremely similar to a vignette by the same name now in the VGAM package.
• The paper was run using VGAM 0.7-10.
• A file with the code is jsscda.R.
• The data set nzmarital can be source()ed: nzmarital.R.
• Documentation on RR-VGLMs (including the stereotype model and Goodman's RC model): rrvglm.pdf (816K).
• Further documentation on categorical data in general: categorical.pdf (248K).
• "VGLMs and VGAMs: an overview for applications in fisheries research" by T. W. Yee (2010), Fisheries Research 101(1--2): 116--126. Here is the code which can be cut-and-pasted into R. Nb. it requires VGAM 0.7-9 or higher; see my prerelease version.
• "Models for analysing species' presence/absence data at two time points" by T. W. Yee and T. Dirnboeck (2009), Journal of Theoretical Biology 259(4): 684--694. Here is some R code which shows how an exchangeable bivariate (cloglog) odds ratio model may be fitted: ebcom.R. It needs the VGAM package to be installed.
• "Vector generalized linear and additive extreme value models" by T. W. Yee and A. G. Stephenson (2007), Extremes 10(1): 1--19. Here is the code which can be cut-and-pasted into R.
• "Constrained additive ordination" by T. W. Yee (2006). Ecology, 87(1): 203--213. The figures are printed in black and white, but here are the colour versions:
• Fig. 1. Fig. 2. Fig. 3.
• A correction: p.204,LHS column: "write \nu = c^T = x_2 when R=1." should be "write \nu = c^T x_2 when R=1."
• "Local regression for vector responses" by A. H. Welsh and T. W. Yee (2006). Journal of Statistical Planning and Inference, 136(9): 3007--3031.
• "A new technique for maximum-likelihood canonical Gaussian ordination" by Thomas W. Yee (2004). Ecological Monographs, 74(4): 685--701.
• May 2014: the function cqo() uses argument I.tolerances instead of ITolerances, and argument eq.tolerances instead of EqualTolerances. The defaults are eq.tolerances = TRUE and I.tolerances = FALSE.
• November 2005: the function cqo() uses a new undocumented algorithm for the case when the argument ITolerances=TRUE. This is for the equal-tolerances asssumption.
• March 2005: the function cgo() has been renamed cqo() in my VGAM package. Additionally, there are now related functions uqo() and cao(). The function uqo() is not functioning well yet.
• Jan 2005: the function cgo() will soon be renamed cqo(), for constrained quadratic ordination. This better reflects an improved nomenclature (see the "Constrained additive ordination" paper) for this and other related ordination methods. Hence all calls to cgo() in the paper needs to be to cqo() instead.
• May 2004: the function cgo() now implements an undocumented fast algorithm which is a large improvement on the algorithm presented in the paper.
• Here are the colour versions of the figures.
• Here are some corrections:
• pp.687,688: the word "sigmoid" is true only for binary responses. For Poisson data, it is exponential (of the form K*exp(lv) where lv is the latent variable axis.)
• p.694: the canonical coefficients of r2 are obtained by typing
print(t(Coef(r2)@C), digits = 3)

Better is
print(t(concoef(r2)), digits = 3)

• p.696: the canonical coefficients of r2.75 are obtained by typing
t(round(Coef(r2.75)@C, digits = 3))

Better is
t(round(concoef(r2.75), digits = 3))

• p.697: the canonical coefficients of b1 are obtained by typing
round(t(Coef(b1, ITolerances = FALSE)@C), digits = 3)

• "Reduced-rank vector generalized linear models" by Thomas W. Yee and Trevor J. Hastie (2003). Statistical Modelling, 3(1): 15--41.
• Corrections:
in the middle of p.21.

• "Vector generalized additive models in plant ecology" by Thomas W. Yee and Monique Mackenzie (2002). Ecological Modelling, 157(2--3): 141--156.
• The VGAM package now fits quadratic reduced-rank vector generalized linear models (QRR-VGLMs), which is a new technique for maximum likelihood estimated canonical Gaussian ordination (CGO). This is an alternative to canonical correspondence analysis (CCA). The function to use is cgo(). Some relevant documentation in the VGAM website is rrvglm.pdf.
• Here are some changes:
1. The VGAM packages runs under Version 4 of the S language.
Hence,
vglm() and vgam() objects have slots, so fit@x and fit@y etc.
should be used instead of fit$x and fit$y.
2. qtplot() now replaces ceplot() for plotting regression quantiles.
3. Reduced-rank VGLMs are treated in detail in a paper by the
same name, by Yee and Hastie (2003).
Canonical Gaussian ordination (CGO) can be performed by