predict with Multinomial Logistic Regression

Question

If I execute the following code I have no problem:

require(foreign)
require(nnet)
require(ggplot2)
require(reshape2)

ml <- read.dta("https://stats.idre.ucla.edu/stat/data/hsbdemo.dta")
ml$prog2 <- relevel(ml$prog, ref = "academic")
test <- multinom(prog2 ~ ses + write, data = ml)
predict(test, newdata = dses, "probs")

but if I try:

require(caret)
ml <- read.dta("http://www.ats.ucla.edu/stat/data/hsbdemo.dta")
ml$prog2 <- relevel(ml$prog, ref = "academic")
test <- train(prog2 ~ ses + write,method="multinom" ,data = ml)
predict(test$finalModel, newdata = dses, "probs")

it returns Error in eval(expr, envir, enclos) : object 'sesmiddle' not found, why?

Answer 1

That's not the error I get and I'm thinking that you left out some code. I get:

predict(test$finalModel, newdata = dses, "probs")
Error in as.data.frame(newdata) : object 'dses' not found

I don't know why I see this so much, but you should avoid using the finalModel object for prediction. train is doing some things that the model from multinom sin't aware of (such as expanding dummy variables). Try using

predict(test, newdata = dses, "probs")

and never use predict(test$finalModel) again...

Max