[R] multinom() and parameters

Vladimir Bezlyak vladimir at stats.gla.ac.uk
Mon May 18 18:07:15 CEST 2009

Dear All,

I am doing multinomial logistic regression with multinom().

I have to process many variables and I created a function which biulds the models and format the results (see below).
It works prefect when I set the variables and the dataset locally
        Outcome <- "Group"
        covars <- "age"
        dta <- my.data
        mod<-multinom(as.formula(paste(Outcome,"~",covars)), data = dta,na.action =na.omit,Hess=T)

But it says that it cannot find the variables and dataset when I start using the function
        my.multinom("Group", "age", dta = my.data)

There should be some problems with passing the parameters from multinom() to nnet().

Help me please!

With the best regards,
Vladimir Bezlyak

Robertson Centre for Biostatistics
University of Glasgow

# the function
 mod<-multinom(as.formula(paste(Outcome,"~",covars)), data = dta,na.action =na.omit,Hess=T)
 mod.sum<- summary(mod,Wald=T)
 p <- 1-pchisq((mod.sum$Wald[,-1])^2,1)
        coef <-c(mod.sum$coefficients[,-1])
        se <-c(mod.sum$standard.errors[,-1])
        ci <- paste(round(exp(coef),3), "(",round(exp(coef-1.96*se),3), ":",round(exp(coef+1.96*se),3),"), p=",p.format(p),sep="")

                mm<- match(zz,names(HV.all))
                        if(!is.na(mm)) attributes(HV.all[,zz])$label
                                else zz

        varnms<- unlist(lapply(Variable,function(zz){paste (dimnames(mod.sum$coefficients)[[1]],":",zz)}))
        out<-   cbind(varnms, ci)
        dimnames(out)[[2]]<-c("Parameter","HR (95%CI),p" )

The University of Glasgow, charity number SC004401

More information about the R-help mailing list