Avoiding duplicated choice tasks

Experiments
,
1

Dear all

I am trying to generate an efficient design of a fairly simple unlabelled experiment with 2 alternatives, 3 attributes and 2, 2 and 4 levels, respectively. I define strictly signed priors to avoid dominant alternatives. However, the designs generated by ngene usually contains at least one duplicate choice task with alternatives in reverse order, and often several duplicates. For instance, one choice task details profile A vs. B and another choice task shows B vs. A. Since this is an unlabelled experiment, respondents are effectively asked the same question twice.

Is there any way to avoid this and restrict ngene to only generate unique choice tasks?

Thanks a lot in advance for any suggestions.

Below is a concrete example with 12 rows, in which rows 1 and 3 are the same (swapping alt 1 and alt 2). So are rows 2<->6, 4<->12, 5<->8, 7<->10 and 9<->11. Effectively, the 12 rows are reduced to 6 unique choice tasks! The number of duplicates vary whenever I “push the button”, but at least the first duplicate consistently shows up.

image
image519×515 11.2 KB

2

Ktb,
We’re not really Ngene experts. But we can weigh in on the practicalities.

Reversed duplicate profiles are valid and likely required for models to identify alternative effects. Which may be either lexographic (left vs right) or because of the label. If the design doesn’t account for these, any left right bias effects will erroneously end up in one of the utilities estimates. These duplicates correctly control for these effects so they can be checked and discarded.

Nevertheless, I suggest is you get the Ngene script that was generated and post to their forum.

They are pretty active.