what parameters stepAIC accepts?

Question

I am using stepAIC() function from library (MASS).

And per the documentation it says that I can pass an 'object':

a model of an appropriate class. This is used as the initial model in the stepwise search.

What is an appropriate class?

I created a model using ksvm() but, stepAIC seems not to accept it.

I am using the code:

stepAIC(svmFit, direction="both")

And I get the error

Error: $ operator not defined for this S4 class

The object svmFit I am passing is of class:

[1] "ksvm"
attr(,"package")
[1] "kernlab"

Answer 1

It's not exactly clear to me what models are accepted, but we can take some guesses based on the source.

It needs a model that can be accessed using $ (i.e. typically an S3 class). Your model is an S4 class, and therefore by definition will not work.

In addition, there need to be at least a working method for the model class for the the following functions: formula, terms, update.formula, nobs, extractAIC, deviance, update (judging from a scan of the source). This can be a default method, such as stats:::formula.default, which finds either x$formula, $terms, $call$formula or a formula attribute.

If one or more of these methods have not been implemented, then it will also fail.

Basically, it is making a lot of assumptions that work for many of the modelling functions, such as lm and glm. One can write a new modelling function that will create models that are compatible with stepAIC, and a comprehensive list may be difficult to create!