Efficient Design for Pilot Study

[Michiel Bliemer]

I generally prefer a Bayesian efficient design. When choosing the parameters to give distributions, it is important to select at least the attributes that contribute most to utility and perhaps make others fixed (to reduce the number of draws required). Level of autonomy and the ASC are for example good to give Bayesian priors.

I do not see an issue with balance. In your design, level 3 of 'org' appears more than 20 times across both alternatives (you should not look only at a single alternative but should consider both alt1 and alt2 because these alternatives are generic). If there was too much imbalance, one of the dummy coded coefficients could not be estimated and lead to a large D-error, so by minimising the D-error balance is automatically sufficiently satisfied in a model with dummy coded attributes.

Michiel