It is reasonable to decide after the data have been collected whether the levels of a factor can be considered fixed or random.

[rlyhim]

This is a true/false question, and I can’t figure out whether or not this is true. On the one hand, it should be declared fixed or random based on the collection methodology, so it should be done beforehand, but alternatively, as long as you’re right about whether it is fixed or random it doesn’t really matter when you decide.