Page 2 of 2

Re: Prior values (mode choice)

PostPosted: Mon Oct 05, 2020 7:39 pm
by peyman_07
Thanks Michiel for your advice. I was wondering why you suggested -0.3 as the mean? This looks to be an extreme situation (-0.0866-0.210=-0.2966), right?
I also have another question. I am going to replace some attributes with some new ones, so the utility function is going to change. Can I still use the same priors for the existing attributes?
Thanks

Re: Prior values (mode choice)

PostPosted: Tue Oct 06, 2020 11:05 am
by Michiel Bliemer
I did not suggest -0.3 as a mean, I suggested a uniform distribution with lower bound -0.3 (or any other value that you think is reasonable) and 0 as upper bound. In Ngene, we define (u,lower,upper) for a uniform distribution, while (n,mean,stdev) for a normal distribution.

Your priors would also change, but since priors are never exact and always only an approximation, you can still treat them as an approximation. It depends on how drastic your attributes and utility function change.

Michiel