I would like test AICc as a criteria for model selection for a glm using
stepAIC() from MASS package.
Based on various information available in WEB, stepAIC() use
extractAIC() to get the criteria used for model selection.
I have created a new extractAIC() function (and extractAIC.glm() and
extractAIC.lm() ones) that use a new parameter criteria that can be AIC,
BIC or AICc.
It works as expected using extractAIC() but when I run stepAIC(), the
first AIC shown in the result is correct, but after it still shows the
original AIC:
For example (the full code is below) the "Start: AIC=70.06" is indeed
the AICc but after, "<none> 47.548 67.874" is the AIC.
> stepAIC(G1, criteria="AICc")
Start: AIC=70.06
x ~ A + B
Df Deviance AIC
- A 1 48.506 66.173
<none> 47.548 67.874
- B 1 57.350 68.685
Thanks if you can help me that stepAIC() use always the new extractAIC()
function.
Marc
library("MASS")
set.seed(1)
df <- data.frame(x=rnorm(15, 15, 2))
for (i in 1:10) {
df <- cbind(df, matrix(data = rnorm(15, 15, 2), ncol=1,
dimnames=list(NULL, LETTERS[i])))
}
G1 <- glm(formula = x ~ A + B,
data=df, family = gaussian(link = "identity"))
extractAIC(G1)
stepAIC(G1)
extractAIC.glm <- function(fit, scale, k = 2, criteria = c("AIC",
"AICc", "BIC"), ...) {
criteria <- match.arg(arg=criteria, choice=c("AIC", "AICc", "BIC"))
AIC <- fit$aic
edf <- length(fit$coefficients)
n <- nobs(fit, use.fallback = TRUE)
if (criteria == "AICc") return(c(edf, AIC + (2*edf*(edf+1))/(n - edf
-1)))
if (criteria == "AIC") return(c(edf, AIC-2*edf + k*edf))
if (criteria == "BIC") return(c(edf, AIC -2*edf + log(n)*edf))
}
extractAIC <- extractAIC.lm <- extractAIC.glm
extractAIC(G1, criteria="AIC")
extractAIC(G1, k=log(15))
extractAIC(G1, criteria="BIC")
stats:::extractAIC.glm(G1, k=log(15))
extractAIC(G1, criteria="AICc")
stepAIC(G1)
stepAIC(G1, criteria="AICc")
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
