choice-metrics.com

Dear Professor, I am conducting a DCE study.
My survey targets three types of respondents, eg staff, superintendent, and stakeholder.
One of my main research objectives includes comparing the preference differences of three types of people. The alternatives, attributes, and levels are the same for different respondents.

I want to conduct a two-stage design. First, in conducting a pilot, I would use D-efficient designs with the sign's value. Then, with the outcome of the pilot, I used the Bayesian efficient design for conducting the formal survey.

To achieve the above objectives, I think there are several options.
1. Three pilot designs, then investigate different respondents separately, and based on the results of the pilots, design three formal surveys separately.
2. One pilot design, then investigate different respondents separately, and based on the results of the pilots, design three formal surveys separately.
3. Following Ngene's manual Page 165, refer to pivot design (Although not using the pivot design), using a single design that can be applied to different respondent segments, using “ homogenous design ” for the pilot, then investigate different respondents separately, and based on the results of the pilots, design one formal survey also using “homogenous design”
4. Same as option 3, the difference is designing one formal survey using “heterogeneous design”
5. Similar to option 3, but use “heterogeneous design” for pilots, then use “homogenous design” for formal surveys
6. Similar to option 3, but use “heterogeneous design” for both pilots and formal surveys


So my question is:
Q1 Which of the above options is correct?

Q2 Is there a difference in the selection of the above methods for labeled and unlabeled experiments, and which one should be chosen respectively? I may conduct labeled and unlabeled experiments separately.

Q3 What are the key points to pay attention to after choosing the correct method above? Can you provide some examples or code syntax?

Thank you for patiently guiding me multiple times.

Q1) If you specifically want to compare preferences across different types of people, and all alternatives/attributes/levels are the same, then it is best to use the exact same design for all people since this ensures that any different you find in model estimation is due to differences in types of people and not due to differences in the experimental design. So I would generate one pilot design, then estimate a different model for each type of people, and then generate a homogeneous design based on three models that only differ in the priors.

Q2) There is no difference between labelled or unlabelled regarding the methodology above.

Q3) Once you have created your pilot design, I can assist you with the script for the homogeneous design based on three sets of priors.

Michiel

Thank you, Professor Bliemer.
I will continue to conduct research. If I encounter any difficulties, I will continue to seek your help and guidance. Thank you.