Decision situations with dominan alternative

[Peter_C]

Hi Michiel,

Sorry for the many questions!

In the meantime, I have successfully completed a pilot study using the recommended design:

design
;alts = Service_A*, Service_B*, Service_C*, Service_D*
;rows = 8
;block = 2
;eff = (mnl,d)
;alg = mfederov
;model:
U(Service_A) = b2[0.0001] * int[0,1] ?discrete attribute levels 0-no 1-yes
+ b3[0.0001] * iter[0,1] ?discrete attribute levels 0-no 1-yes
+ b4[0.0001] * mobil[0,1] ?discrete attribute levels 0-no 1-yes
+ b5[-0.0001] * cost[10,20,30,40,50](1-2,1-2,1-2,1-2,1-2) ?continuous attribute levels
/
U(Service_B) = b2*int + b3*iter + b4*mobil + b5*cost /
U(Service_C) = b2*int + b3*iter + b4*mobil + b5*cost /
U(Service_D) = b2*int + b3*iter + b4*mobil + b5*cost
$

I am currently facing the final questionnaire and using the results of the pilot study as a priors, I have generated a bayesian design.

Results from my pilot study:

b_int: 0.46765 s.e.:0.154934 t.value:3.0184
b_iter: 0.42613 s.e.:0.197380 t.value:2.1589
b_mobil: 1.31031 s.e.:0.187561 t.value:6.9860
b_cost: -0.02847 s.e.:0.007360 t.value:-3.8690

I have used the following bayesian syntax:

design
;alts = Service_A*, Service_B*, Service_C*, Service_D*
;rows = 8
;block = 2
;eff = (mnl,d,mean)
;bdraws = gauss(5)
;alg = mfederov
;model:
U(Service_A) = b2[(n,0.47,0.15)] * int[0,1]
+ b3[(n,0.43,0.20)] * iter[0,1]
+ b4[(n,1.31,0.19)] * mobil[0,1]
+ b5[(n,-0.03,0.01)] * cost[10,20,30,40,50] (1-2,1-2,1-2,1-2,1-2)
/
U(Service_B) = b2*int + b3*iter + b4*mobil + b5*cost /
U(Service_C) = b2*int + b3*iter + b4*mobil + b5*cost /
U(Service_D) = b2*int + b3*iter + b4*mobil + b5*cost
$

I got the following results:

Fixed Bayesian mean
D error 0.109019 0.111571
A error 0.495706 0.508849
B estimate 57.346949 0.55082
S estimate 12.529809 63.938391

Prior b2 b3 b4 b5
Fixed prior value 0.47 0.43 1.31 -0.03
Sp estimates 10.653272 12.529809 1.715609 3.328466
Sp t-ratios 0.600502 0.553712 1.496397 1.074321
Sb mean estimates 29.694358 29.434689 1.85483 19.98548
Sb mean t-ratios 0.592145 0.547779 1.469743 1.046393

What do you think? Can I start the final questionnaire with this design specification?


In addition, some decision situation contain the following alternative:

int: no
iter: no
mobil: no
cost: 10

This is similar to an opt-out alternative, but you have to pay for it. So it is not to realistic. Can I put any restrictions in order to avoid such alternatives?


Thanks a lot for the help!

Best regards,
Peter

[Michiel Bliemer]

That looks good yes.

You can add constraints such as in below syntax to avoid choice tasks with those combinations of attribute levels.
I would recommend using more than 8 rows to increase variation in your data, e.g. using 16 rows as in syntax below, blocked in 4.

Code: Select all

design
;alts = Service_A*, Service_B*, Service_C*, Service_D*
;rows = 16
;block = 4
;eff = (mnl,d,mean)
;bdraws = gauss(5)
;alg = mfederov
;reject:
Service_A.int = 0 and Service_A.iter = 0 and Service_A.mobil = 0 and Service_A.cost = 10,
Service_B.int = 0 and Service_B.iter = 0 and Service_B.mobil = 0 and Service_B.cost = 10,
Service_C.int = 0 and Service_C.iter = 0 and Service_C.mobil = 0 and Service_C.cost = 10,
Service_D.int = 0 and Service_D.iter = 0 and Service_D.mobil = 0 and Service_D.cost = 10
;model:
U(Service_A) = b2[(n,0.47,0.15)] * int[0,1]
+ b3[(n,0.43,0.20)] * iter[0,1]
+ b4[(n,1.31,0.19)] * mobil[0,1]
+ b5[(n,-0.03,0.01)] * cost[10,20,30,40,50] (2-4,2-4,2-4,2-4,2-4)
/
U(Service_B) = b2*int + b3*iter + b4*mobil + b5*cost /
U(Service_C) = b2*int + b3*iter + b4*mobil + b5*cost /
U(Service_D) = b2*int + b3*iter + b4*mobil + b5*cost
$

Michiel

[Peter_C]

It worked perfectly!

Thank you very much for your help!