Constraining utilities to conform to a priori reference orders

Question

Dear all,

I got a set of categorial attributes (without price) in my ACBC study and most of them have an a priori reference order (e.g. the levels are basically equivalent to a low-medium-high scale, just in domain-specific wording).

In the HB estimation, should I constrain those attributes with rational order or rather not?

Answer 1

There isn't a clear answer for this.  I'd say in most situations you probably don't want to impose constraints on the a priori order attributes.  However, under the following two situations I probably would:

1) Your sample size is unusually small (say, 60 or fewer respondents)
2) Individual-level classification accuracy or individual-level prediction was the main goal of your study (usually market-level /segment-level prediction is the goal).