Is the list aware of any research that compare the pros and cons of
full and conditional maximum likelihood based inference?
I have the choice of using conditional and full maximum likelihood
methods in estimating parameters relating to cancer screening. In
practice, I shall probably use both. The full method provides
inference about three sets of parameters. The conditional method
conditions over one set of parameters and provides inference about the
remaining two.
I don't expect the two methods to provide radically different
parameter estimates (of the two sets of parameters shared between both
approaches). If the two methods do provide quite different parameter
estimates, I shall look carefully at how well the full likelihood
model estimates the parameter that is conditioned over in the
conditional model.
But I'm wondering whether full vs conditional likelihood inference is
actually a pretty general issue, which is probably well covered in the
literature. Maybe I'm missing something important.
Any help gratefully received (and summarised to list).
Thank you,
Dominic Muston
|