Question about coding some attributes and levels

[susiezhao]

So if I want to study uncertainty as this author does should I only consider orthogonal design? Efficient design might not be appropriate?

[Michiel Bliemer]

That paper uses expected utility theory (see my earlier comment) and uses random regret theory and hybrid theories. None of these are models based on random utility maximisation, which is what Ngene assumes. Ngene can optimise for random regret models, see paper below, but it cannot optimise for expected utility models and hybrid models. The authors in that paper used an orthogonal design, not an efficient design because they were also not able to optimise for their specific model. You may also want to consider using an orthogonal design.

https://www.sciencedirect.com/science/article/pii/S1755534518300940

[susiezhao]

Thank you Professor. I have one last question.
In my scenario there is an attribute "Check-in luggage" with levels "yes" and "no". When the level is "no", the "Luggage check-in time" at the terminal is 0. The "Luggage check-in time" at the airport is also 0. "I don't know how to code this condition. I am attaching my code. Please help me to check.

Design
;alts = mode1, mode2
;rows = 64
;eff = (mnl,d)
;cond:
if(mode1.luggage=0, mode1.luggage_airport=0 AND mode2.luggage_terminal=0)
;model:
U(mode1) = b1[-0.05] * luggage_airport[0,10,20,30] +
b2[-0.05] * security_airport[5,12,19,26] +
luggage.dummy[0.1] * luggage[1,0] /

U(mode2) = b3[-0.1] * luggage_terminal[0,3,6,9] +
b4[-0.1] * security_terminal[3,5,7,9]
luggage * luggage[luggage]

$

[susiezhao]

Dear professor,

I also have an SP experiment with a more complex design. The d error generated using efficient design coding is 1.1. Can this generated experiment be used for a trial study?

I appreciate you continuing to answer my questions. Thank you very much.

[Michiel Bliemer]

The model in your syntax is not identifiable, the value for luggage is constant across the two alternatives and therefore the luggage dummy cannot be estimated. You need to interact the luggage attribute with another attribute to make it identifiable, or you need to add an opt-out alternative. Both these options have different behavioural meaning, so you need to define your utility function such that it makes most behavioural sense.

Conditional constraints with the default algorithm are somewhat restrictive. For full flexibility of constraints, please use ;alg = mfederov in combination with ;reject to apply your constraints.

See my previous comment on D-errors. A value of 1.1 can be very bad, or can be very good, this is case specific. It is quite high.

[susiezhao]

Dear professor,

Thank you very much. Your answers are really helpful.