Single profile DCE with ordered choices

[jb_ox]

Hi all,

I'm trying to work up the experimental design for a DCE but I'm not sure how to proceed in Ngene. The format of the choice question is as follows:

You are prescribing treatment A. Given the scenario described below, what length prescription will you give to the patient.

<Single profile presented that includes attributes such as patient signs and symptoms, drug name, patient comorbidities, diagnostic test result>

Would you give this patient a prescription for 7, 10, 14 or 18 days.

So respondents will be presented with a single profile, but need to make 1 of 4 ordered choices. I initially assumed I would need an ordered probit/logit model to analyse the results, but can I code such a design in Ngene, and is this even the correct approach?

Apologies if I've missed an answer to this question in the manual, I did have a look.

[Michiel Bliemer]

What you are describing is a scenario, not a profile. The scenario is constant across all choice alternatives.
The utility functions of the four alternatives (prescription duration) have the same attributes and levels, but the coefficients are alternative-specific. Further, one of the alternatives needs to be normalised to zero, which can be done by simply not defining the utility function for this alternative.

Your Ngene script will look something like this (although Ngene cannot do ordered logit so it would consider regular logit here):

Code: Select all

design
;alts = 7days, 10days, 14days, 18days
;rows = ...
;eff = (mnl,d)
;alg = mfederov
;require:
7days.symptoms = 10days.symptoms, 10days.symptoms = 14days.symptoms, 14days.symptoms = 18days.symptoms,
7days.drug = 10days.drug, 10days.drug = 14days.drug, 14days.drug = 18days.drug

;model:
U(7days)  = asc1 + b2.dummy[...] * symptoms[...] + b3.dummy[...] * drug[...] + ... /
U(10days) = asc2 + b4.dummy[...] * symptoms[...] + b5.dummy[...] * drug[...] + ... /
U(14days) = asc3 + b6.dummy[...] * symptoms[...] + b7.dummy[...] * drug[...] + ...

$

Have a look at this paper, which also considers choice tasks with scenarios only: https://journals.plos.org/plosone/artic ... ne.0142507

Michiel