My apologies for the silly question: I've been running a mixed effects model and since I have a small number of clusters (11) I've been using REML. Having plotted the residuals for one of my models, and seeing they were nowhere near being normal (variable is heavily negatively skewed), I ran the model again using robust standard errors. It gave me completely different results, and different variables being significant. I've been reading around the subject to understand it, and was wondering if anyone had any insight into why they're so different, and what would be the best approach here: REML or robust SE's? TIA.