In one of my estimations, the order of attribute levels within one attribute (in terms of increasing part-worth utility values) changes when you switch from raw values to zero-centered differences. What does that mean for interpretation?
In general: How would you interpret the values of zero-centered differences (in contrast to raw utilities)?