如何遍历glm测试不同的模型

Question

I have a vector a which contains some combinations of my variables. In order to get the best possible model (smallest BIC), I hoped to be able to loop over the glm function - well, that didn't work.

The data looks something like this:

de.dat <- data.frame(death=c(0,0,0,1,0), ac=c(1, 2, 2, 3, 2), fv=c(1,0,0,0,1), vs=c(0,0,0,0,0), v2=c(0,0,1,1,1), ms=c(0,0,0,0,0), yv=c(0,0,1,0,1))

vars <- c("ac", "fv", "vs", "v2", "ms", "yv")
a <- apply(data.frame(t(combn(vars, 5))), 1, paste, collapse="+")

The model should look like this, with the dependent variable death and the independent variables of each element of a :

glm(death ~ a, data=de.dat, family="binomial")

In order to get the best model, I then need to extract the BIC of every model and then to be able to identify which BIC belongs to which model. Thank in advance.

Answer 1

This one should help you. Here is your data:

de.dat <- data.frame(death=c(0,0,0,1,0), ac=c(1, 2, 2, 3, 2), fv=c(1,0,0,0,1), vs=c(0,0,0,0,0), v2=c(0,0,1,1,1), ms=c(0,0,0,0,0), yv=c(0,0,1,0,1))
vars <- c("ac", "fv", "vs", "v2", "ms", "yv")

You should create character vector with whole formula inside (not only the right part):

a <- apply(cbind(paste(names(de.dat)[1],"~"),data.frame(t(combn(vars, 5)))), 1, paste, collapse="+")

Then, using lapply function you can create a list, which consists of all your 6 models:

models <- lapply(a,FUN = function(X) glm(X, data=de.dat, family="binomial"))