Having a random factor in model when not neededWe have a generic model we use for greenhouse studies that uses reps and subsamples as random, nested factors. Some of your factors have subsamples and some of them don't. Is it detrimental to include the subsample random, nested factor in the model when there is no subsample for that particular variable?
Re: Having a random factor in model when not neededNo, not detrimental. It will just be estimated to have a variance component of 0 and then not affect the modeling of the fixed parameters. It is more correct to remove it from the model of course, but in my experience there is very little difference.
When working with random factors we KNOW have "significant" positive variance components (when they are estimable), I normally recommend removing the tickmark in Unbounded Variance Components (in the Model Dialog), as they will then be estimated to 0 and not (by chance) to something negative. The tests on the fixed parameters will be more correct with Unbounded Variance Components, but again in my experience, the difference is very small.
