I have a study that was conducted on 2 farms. This wasn't part of the initial experimental design, and wasn't a main focus of the study. I have therefore included it as a random effect in most of my analyses. However, I am also interested in whether the treatment effects were consistent across the two farms, so I have conducted another analysis with farm as a fixed effect, and included the interaction with treatment.
My advisor is arguing that the farm effect should be either fixed or random for all analyses, and not different across analyses. My question is, am I correct in thinking that it should be random if we're not interested in the effect itself, and then fixed when we are, or should it be fixed at all times? And does anyone have references I could use to support either decision?
Aucun commentaire:
Enregistrer un commentaire