choice-metrics.com

hi NGENE team,

I created a D-efficient MNL design for my DCE. We are planning to launch an online and offline version of this for 500 participants. The rationale behind having an offline version is to capture non-tech savvy individuals.
How do I determine what proportion of the sample population should be online vs. offline to be able to estimate if preferences may be different between online vs. offline participants?
Is it reasonable to say having 20% of the sample use the offline survey has enough "power" to detect differences in preferences (if it exists)?
thanks for your help

Kind regards,

Jason Ong

This is not an Ngene question (so you may want to post in the stated choice experimental design forum), but there is not a simple answer to your question. Whether you can detect differences in preferences depends completely on your study and how important the attributes are. It is easier to detect differences for attributes that are deemed very important, and it is easier to detect differences for attributes with a wide attribute level range. Maybe you need 10 respondents, maybe you need 100 or 1000. The only way to provide a guesstimate is to consider parameter priors, calculate standard errors, assume sample sizes, and do a compare means t test.

Michiel