non statistically significant attributes as priors

[Rajwane]

Hello,

I have conducted a pilot DCE on 100 respondants in order to calculate priors (using mlogit estimation technique in Stata) to later use them in the generation of the design that will be used in the final DCE on 1500 respondents.

I have run the regression on Stata and have found some attributes to be statistically non-significant. Can I still use the betta coefficient of these attributes as priors ?

Sincerely,
Rajwane

[Michiel Bliemer]

Yes you can still use the parameter estimates as priors, even if they are not statistically significant, since they are still the best guesses you have. Only if the sign of the parameter seems incorrect or if the parameters of an ordinal dummy coded variable do not have the expected preference order, then I would manually adjust the priors to make sure that they are sensible.

You can use Bayesian priors in Ngene using (n,beta,se), where beta is the parameter estimate and se is the standard error, which indicates its unreliability.

Michiel