Additive Regression for Discrete Data

Methods for fitting identity-link GLMs and GAMs to discrete data, using EM-type algorithms with more stable convergence properties than standard methods.


addreg

CRAN_Status_Badge

addreg provides methods for fitting identity-link GLMs and GAMs to discrete data, using EM-type algorithms with more stable convergence properties than standard methods.

An example of periodic non-convergence using glm (run with trace = TRUE to see deviance at each iteration):

require(glm2, quietly = TRUE)
data(crabs)
 
crabs.boot <- crabs[crabs$Rep1, -c(5:6)]
 
t.glm <- system.time(
  fit.glm <- glm(Satellites ~ Width + Dark + GoodSpine, data = crabs.boot, family = poisson(identity),
                    start = rep(1, 4), maxit = 500)
)

The combinatorial EM method (Marschner, 2010) provides stable convergence:

require(addreg, quietly = TRUE)
t.cem <- system.time(
  fit.cem <- addreg(Satellites ~ Width + Dark + GoodSpine, data = crabs.boot, family = poisson,
                    start = rep(1, 4))
)

...but it can take a while. Using an overparameterised EM approach removes the need to run (2^3 = 8) separate EM algorithms:

t.em <- system.time(fit.em <- update(fit.cem, method = "em"))

while generic EM acceleration algorithms from the turboEM package --- implemented in version (\geq) 3.0 --- can speed this up further still:

t.cem.acc <- system.time(fit.cem.acc <- update(fit.cem, accelerate = "squarem"))
t.em.acc <- system.time(fit.em.acc <- update(fit.em, accelerate = "squarem"))

Comparison of results: #> glm FALSE -518.2579 500 0.06 #> cem TRUE -500.8886 6101 0.69 #> em TRUE -500.8886 1680 0.13 #> cem.acc TRUE -500.8886 128 0.11 #> em.acc TRUE -500.8886 38 0.05

The combinatorial EM algorithms for identity-link binomial (Donoghoe and Marschner, 2014) and negative binomial (Donoghoe and Marschner, 2016) models are also available, using family = binomial and family = negbin1, respectively.

Semi-parametric regression using B-splines (Donoghoe and Marschner, 2015) can be incorporated by using the addreg.smooth function. See example(addreg.smooth) for a simple example.

Installation

Get the released version from CRAN:

install.packages("addreg")

Or the development version from github:

# install.packages("devtools")
devtools::install_github("mdonoghoe/addreg")

References

  • Donoghoe, M. W. and I. C. Marschner (2014). Stable computational methods for additive binomial models with application to adjusted risk differences. Computational Statistics and Data Analysis 80: 184-196.
  • Donoghoe, M. W. and I. C. Marschner (2015). Flexible regression models for rate differences, risk differences and relative risks. International Journal of Biostatistics 11(1): 91-108.
  • Donoghoe, M. W. and I. C. Marschner (2016). Estimation of adjusted rate differences using additive negative binomial regression. Statistics in Medicine 35(18): 3166-3178.
  • Marschner, I. C. (2010). Stable computation of maximum likelihood estimates in identity link Poisson regression. Journal of Computational and Graphical Statistics 19(3): 666-683.

News

addreg 3.0

  • method option added: fit using combinatorial EM or single parameter expanded EM algorithm
  • accelerate option added: use turboEM to allow acceleration of EM algorithm
  • Corrected an error in the calculation of deviance residuals for family = negbin1
  • Corrected an error in the calculation of null deviance for family = negbin1
  • (Minor change) Change iter to return a vector of two values

addreg 2.0

  • Added additive negative binomial regression (family = negbin1)
  • (Minor change) Removed bin.identity link function; avoided CRAN error in a simpler way
  • (Minor change) Documentation changes, including updated references

addreg 1.2

  • Introduced bin.identity link function: identical to the standard identity link, but does not cause an error when used with the binomial family.
  • (Minor change) Corrections to avoid codetools errors.

addreg 1.1

  • The object returned by nnpois() now includes component standard.
  • (Minor change) Various very minor corrections.

addreg 1.0

  • Added na.action argument to addreg() and addreg.smooth()

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("addreg")

3.0 by Mark W. Donoghoe, a year ago


https://github.com/mdonoghoe/addreg


Browse source code at https://github.com/cran/addreg


Authors: Mark W. Donoghoe [aut, cre] , Ian C. Marschner [ths]


Documentation:   PDF Manual  


GPL (>= 2) license


Imports splines, combinat, glm2, turboEM


See at CRAN