choice-metrics.com

Hi all,

This question comes from the design discussed in http://choice-metrics.com/forum/viewtopic.php?f=2&t=766. Since the topic is different, I open a new post.

I modified my design following Michiel recomendations, and decided not to use priors since I'm not positive that they are totally reliable. The design is:

Code: Select all

Design

;alts = Car, Bike, Walk, None
;alg = swap(stop=total(5000 iterations))

;rows = 24
;block = 4

;eff=(mnl, d)

;cond:
if(Bike.lts_bike = 1, Walk.lts_walk = [1,2]) ,
if(Bike.lts_bike = 2, Walk.lts_walk = [1,2,3]) ,
if(Bike.lts_bike = 3, Walk.lts_walk = [2,3,4]) ,
if(Bike.lts_bike = 4, Walk.lts_walk = [3,4]),
if(Car.tt_car = 4, Bike.tt_bike = [4,5,6]), ?[(0%,25%,50%)]
if(Car.tt_car = 6, Bike.tt_bike = [6,8,9]), ?[(0%,25%,50%)]
if(Car.tt_car = 8, Bike.tt_bike = [8,10,12]), ?[(0%,25%,50%)]
if(Car.tt_car = 4, Walk.tt_walk = [8,9,10]), ?[(100%,125%,150%)]
if(Car.tt_car = 6, Walk.tt_walk = [12,14,15]), ?[(100%,125%,150%)]
if(Car.tt_car = 8, Walk.tt_walk = [16,20,22]) ?[(100%,125%,150%)]

;model:

U(Car)  = b1               * tt_car[4,6,8] + ? I was using these priors [(u,-0.025, -0.0080)]
          b2                   * tc_car[0.5] +
          b3                   * pc_car[0,1,3] /

U(Bike)  = b0 + b1      * tt_bike[4,5,6,8,9,10,12] +
           b5                  * lts_bike[1,2,3,4] /

U(Walk)  = b00 + b1    * tt_walk[8,9,10,12,14,15,16,20,22]+
           b7                  * lts_walk[1,2,3,4]

$



The problem is the following:
- If I run this code, not using the priors indicated in the comment, all utilities are exactly 0 (and therefore the probabilities of the 4 alternatives = 25%).
- If I use the priors AND tt_car, tt_bike and tt_walk have the same generic coefficient, the utilities are not zero, but always utility car > utility bike > utility walk
- If I use the priors AND tt_car, tt_bike and tt_walk have specific coefficients, only the utility of car has a value, and the others are 0.

That also happens for the other two designs, which have the same structure but larger levels of attributes.

I wonder what can be wrong and would appreciate your help.

It is correct that utilities are all zero, that is indeed what you want if you set priors to zero. These are called utility-neutral efficient designs where no prior information is used (just like orthogonal designs), where choice probabilities are equal across all alternatives. This is a design strategy that is often used across many disciplines if priors are unknown or unreliable. Note that your design will still be efficient, but it is efficient under the null hypothesis. Only if you use reliable non-zero priors you will get non-equal probabilities, allowing you to further optimise your design around those priors.

Michiel

Thanks,that helped a lot.

BR.