Hello, I am having a problem figuring out how to model a continuous outcome (y) given a continuous predictor (x1) and two levels of nested categorical predictors (x3 nested in x2). The data are observational, not from a designed experiment. There are about 15 levels of x2 and between 3 and 14 levels of x3 nested within each level of x2. There are between 6 and 50 x1, y observations for each unique x3 (i.e. the data are unbalanced). I would like to get a prediction equation for y based on x1 and the levels of x2 and x3, and be able to test for significant effects of the levels of x2 and x3. The variables x2 and x3 are drawn from a fixed population. [[alternative HTML version deleted]]