Is omitting a reference category always needed for calculating MaxDiff with CBC/HB?

Question

We fielded a MaxDiff excercise outside of SSIweb and I have created a .csv file that CBC/HB can read (columns: respID; task; Concept; Alternatives 1 through x; response).
In the CBC/HB manual appendix there is "A Suggested Rescaling Procedure" in which I have to fill in 0's for the omitted variable.
Therefore, I would have to omit an alternative, say x, when creating my .csv input so that there is x-1 alternatives. The concept row would be all 0's when the concept was x.
Is that always necessary?

Thanks for a quick clarification,
Alex

Answer 1

In order to allow for degrees of freedom, one of the items is used as the reference item (with a value of zero) and so is excluded from analysis.

Comments

To avoid confusion: the reference item is not "excluded" from analysis.  It's actually an essential part of the analysis--as the reference item that is fixed at 0, so that the scores for all items are identified.  Otherwise, they'd all float around and an infinite number of sets of scores could explain the data equally well.

Thus, a column is dropped from the matrix.  But, the row representing the reference item must be included in the analysis (and the row is a vector of all zeros in the design matrix).

---

Thanks for the responses. I wasn't sure if the "reference category" clause (i.e. dropping a concept / column to tie down "floating" scores) necessarily applies to MaxDiff too. I know it from estimating with ASCs and other parameters when there is no none-option in a discrete choice setting (which is essentially also the case in MaxDiff).