Hi,
I've just studied the results from our MaxDiff survey and found that correlation between raw and anchored utilities is very low for the last Item (r = 0.25 compared to average of 0.8 for other items). I've also found the largest discrepancy (between anchored and standard estimation) on aggregate ranking of utils for this last item. It can be accidental, but I wonder if it has sth to do with different estimation for anchored and unanchored model. In the first one the last item is treated just like all others but in the second the last item is a reference level. Has anybody found that behaviour of utilities for the last item? Are there any procedures to avoid it (if it's an estimation problem)?
I should add that I've checked for both rescaled and raw (Zero-centered) utilities.