Dear all, I'm fitting a mixed linear model with 2 levels, a fixed effects level and random effects level.
When plotting the standardized residuals versus the predicted (or fitted) values at the random-effects level, i find the standardized residuals are correlated with the predicted values. There appears to be positive correlation between the two variables. The plots are the same in S-PLUS and SAS. Does anyone know whether the residuals should be orthogonal to the predicted values at all levels? Is there a way of checking assumptions relating to homoscedasticity and independent residuals, using the residuals versus predicted values at the random-effects level? Does you know how to overcome this problem or a reference to an article that discusses this problem further? Thanks in advance for any help. Hars . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
