How to set defaut of draws for Bayesien design ?

[sukunta]

Dear All,
I have got the MNL model of the pilot study. I plan to use the Bayesian that is

Code: Select all

design
;alts=alt1*,alt2*,alt3
;rows=6
;eff = (mnl,d,mean)
;con
;bdraws = .....?
;model:
U(alt1)=b0 [(n,5.51892, 2.400485)]+b1.dummy[(n,-.1969752, 0.6256593) |(n, -1.777203, 1.348568)]*mode[1,2,0]+b2.dummy[(n, -0.0287256, 0.10309630)]*format[1,0]+b3.dummy[(n,1.368347, 1.700064)]*con_ser[1,0]+b4[(n,-.0268496, .0316639)]*time[10,20,30]+ b5 [(n,-.0297015, .0246112)]*cost[0,50,100]/
U(alt2)= b0 [(n,5.51892, 2.400485)]+b1*mode+b2*format+b3*con_ser+b4*time+b5*cost
$

My questions are
1. How I choose the type of draws?
2. How many the number of draws?
3. The direction of b2 is not consistent with my qualitative result. Can I add the value of SE equal b3?

Sincerely yours,
Sukunta

[Michiel Bliemer]

I refer to the following publication:

Bliemer, M.C.J., J.M. Rose, and S. Hess (2008) Approximation of Bayesian efficiency in experimental choice designs. Journal of Choice Modelling, Vol. 1, pp. 98-127.

Gaussian draws are generally my preference, so I suggest using ;bdraws = gauss(3), which will do 3^7 = 2187 draws per design evaluation. The number of Gaussian draws increases expontentially with the number of Bayesian priors, so in case of a large number of Bayesian priors (e.g. 10-20) I would recommend using Sobol draws. Choosing the number of draws is always tricky, but the more the better, e.g. ;bdraws = sobol(5000), although this will mean a slowdown in generating designs so you may want to run the syntax for quite some time. My rule of thumb is to choose the number of draws larger or equal to 3^(number of Bayesian priors), where I would use 2^(number of Bayesian priors) as some lower limit on the number of draws that cover all dimensions.

I am not sure what you mean with question 3. If the sign of the prior is unexpected then I would change b2 to something like (n,0.001,0.10309630) to indicate that on average it is positive (0.001, which is used for dominance checks) but has a large unreliability. I am not sure what it has to do with b3.

Michiel

[sukunta]

Dear Prof. Michiel,

Thank you so much. I will apply;bdraws =gausss (3) in my syntax.
In question 3, you understand me. If the sign of the prior is unexpected. Can I still use the negative sign but add the value of SE such as (n, -.0287256 ,.2030963 ) or remain the sign and the value of SE from the pilot study in the main survey syntax.

Sincerely yours,
Sukunta

[Michiel Bliemer]

Bayesian priors are merely best guesses of the coefficients, therefore you are free to make manual modifications that you deem necessary to the value of the coefficient as well as the unreliability of the coefficient. As long as the distribution of coefficients is reasonable and not extreme, it is fine.

Michiel

[sukunta]

Dear Prof Michiel,
Thank you so much.

Sincerely yours,
Sukunta