I'm trying to understand cumulative link models and how they differ from multinom models in R. Here's a simple example of a multinom model and plot output using the nnet package:


hous1.mu <- nnet::multinom(Sat ~ Freq,data = housing)
plot_model(hous1.mu, type="pred", ci.lvl = 0.95)

multinom output

The 'ordinal' package in R states on page 17:

It is possible to fit multinomial models (i.e. with nominal effects) as the following example shows:

But, when I use the example code, the plot fits look wildly different than the multinom output:

hous1.clm <- clm(Sat ~ Freq, data = housing)
plot_model(hous1.clm, type="pred", ci.lvl = 0.95)

enter image description here

Even though the likelihood between the two models is the same:

all.equal(logLik(hous1.mu), logLik(hous1.clm))

Can anyone point me in the right direction as to what clm models are doing in this instance?

My reading of the ordinal package is that I can use clm with a nominal variable, but this gives a different plot fit.

Ultimately I would like to use clmm to model a nominal variable with random effects as: i) there's no straightforward way of doing this other than MCMCglmm, and ii) ordinal plays nicely with sjPlot, but... I'd first like to understand why clmm differs from multinom.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.