[R-sig-ME] Should I include Items as a random effect?

[Antoine Tremblay]

Dear all,

I am unsure whether I should include Items as a random effect in my model.

I have an experiment in which all of the items were created
and I'm analyzing a substantial subset of the full set of actual
items (48 out of 64, with 16 excluded because they have
somewhat strange properties). Is it warranted (or necessary or ill-advised)
to include by-item random intercepts?

I believe one school of thought would say something like "test it
(i.e., log-likelihood ratio test --> model without the term and model
with the term)
if the test is significant, leave it in".

But another school of thought would probably say "No, don't include it because
you're not interested in generalizing and you have almost the whole population".

What should I do?

Thank you in advance.
:-)

--
Antoine Tremblay
Department of Neuroscience
Georgetown University
Washington DC