Telling R to change the reference level of a categorical variable

[eloiz]

Dear all,

Does anyone know how to change the reference level of a dependent variable?

I have computed a multinomial regression on the basis of a dataset including 8 categorical variables. 

- Assuming a four-level dependent variable and its individual levels: L1, L2, L3 and L4;
- Assuming L2 as the reference level;
- Assuming that I want to plot the predicted values for L4 in decreasing order.

The output I keep getting is a graph showing predicted values for L4 but in the decreasing order of the predicted values for L2.

Does anyone know how to fix this?

Thanks!!

Eloiz

[Stefan Th. Gries]

?relevel

# from the second edition:
x <- factor(rep(c("long", "intmed", "short"), 1:3)); x # create a factor x
x <- relevel(x, "short"); x # change order of levels: choose a
different 1st reference level (you can also use the number of the new
reference level)

Also see:

?reorder

STG
--
Stefan Th. Gries
-----------------------------------------------
University of California, Santa Barbara
http://www.linguistics.ucsb.edu/faculty/stgries
-----------------------------------------------

[eloiz]

Wow! Thanks for the quick response!

So, am I right to understand that in order to be able to plot the predicted values for each of the four levels of the depend variables, I need to 'relevel' 3 times (not counting the default level) and then run the regression three times and also get predicted values (with 'fitted') 3 times?

Thanks for your help, it's greatly appreciated!

Eloiz

[Stefan Th. Gries]

> So, am I right to understand that in order to be able to plot the predicted values for each of the four levels of the depend variables

depend variableS?? Without a replicable example I am not sure I
understand what you're trying to do.

rm(list=ls(all=TRUE))
set.seed(1); library(effects); library(nnet)
A <- factor(rep(c("o", "l", "t"), each=100))
S <- factor(c(sample(c("f", "m"), 150, replace=TRUE, prob=c(0.7,
0.3)), sample(c("f", "m"), 150, replace=TRUE, prob=c(0.3, 0.7))))
model.01 <- multinom(A ~ S)
allEffects(model.01); plot(allEffects(model.01))