logbin provides methods for performing relative risk regression by fitting log-link GLMs and GAMs to binomial data. As well as providing a consistent interface to use the usual Fisher scoring algorithm (via glm or glm2) and an adaptive barrier approach (via constrOptim), it implements EM-type algorithms that have more stable convergence properties than other methods.

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

require(glm2, quietly = TRUE)

start.p <- sum(heart$Deaths) / sum(heart$Patients)
t.glm <- system.time(
  fit.glm <- logbin(cbind(Deaths, Patients-Deaths) ~ factor(AgeGroup) + factor(Severity) + 
                      factor(Delay) + factor(Region), data = heart,
                    start = c(log(start.p), -rep(1e-4, 8)), method = "glm", maxit = 10000)

The combinatorial EM method (Marschner and Gillett, 2012) provides stable convergence:

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

...but it can take a while. Using an overparameterised EM approach removes the need to run 34 = 81 separate EM algorithms:

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

...while generic EM acceleration algorithms (from the turboEM package) 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:

#>         converged    logLik iterations   time
#> glm         FALSE -186.7366      10000   2.28
#> cem          TRUE -179.9016     445161 101.01
#> em           TRUE -179.9016       7403   1.80
#> cem.acc      TRUE -179.9016       7974   8.58
#> em.acc       TRUE -179.9016         92   0.11

An adaptive barrier algorithm can also be applied using method = "ab", with user-specified options via control.method: see help(logbin) for more details.

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


Get the released version from CRAN:


Or the development version from github:

# install.packages("devtools")


mdonoghoe/logbin documentation built on May 22, 2017, 2:16 a.m.