AIC for logistic/ordinal regression

Question

Can you use Akaike Information Criterion (AIC) for model selection with either logistic or ordinal regression? I want to create multiple different logistic and ordinal models to find the best fitting model for my data, and want to know if (AIC) will work with this type of regression modeling.

Answer 1

My understanding is that in practice, the AIC calculation depends on:

A) the log-likelihood of the fitting results B) the number of model parameters C) the number of data points being modeled

Given that you know the number of data points and number of model parameters (usually true!) if whatever modeling technique is being used can produce a log-likelihood then the AIC is be valid. Since the AIC is a relative measure, any likelihood function can be used but is usually the log-likelihood.