Page 2 of 2

Re: Beginner: Optimise MNL, evaluate RPL efficient design

PostPosted: Tue Jan 07, 2020 11:27 am
by Michiel Bliemer
Those are entirely different distributions. The mean of the normally distributed prior is -0.000093, while the mean of the uniformly distributed parameter is +0.0017035 (yes, positive). Using the negative prior, the alt3 alternative is mostly chosen because it has a higher (zero) utility, whereas with the positive prior mostly alt1 and alt2 are chosen, which provides more information.

You cannot simply change distributions and set priors manually, this needs to be done with great care and informed through a pilot study. Setting b5 to zero is problematic as I indicated earlier. If you are not sure of the priors, then it is best to set them all to zero.