choice-metrics.com

Dear Michiel Bliemer,
We want to create choice sets for DCE. We only knew the prior cost from theory, so we expected a negative sign with a small value. We do not know the prior value for the other attributes.Therefore, my syntax is as follows:,

Code: Select all

Design
;alts = alt1, alt2, alt3
;rows = 12
;eff = (mnl, d)
;model:
U(alt1)  = b1.dummy[0|0|0|0]*loc[0,1,2,3,4]+b2.dummy[0|0]*sup[0,1,2]+b3[-0.00001]*cost[0,500,4000]+b4.dummy[0]*lang[0,1]+b5.dummy[0|0]*time[0,1,2]/

U(alt2) = b1*loc+b2*sup+b3*cost+b4*lang+b5*time$


Please let me know if you have any suggestions for improving the syntax.
Sincerely yours,
Sukunta

See below. Yes zero priors is fine, although it is recommended to conduct a pilot study to get informative priors.

I assume that alt1 and alt2 are generic and that alt3 is an opt-out alternative.
When using zero priors, I found that it is usually best to dummy code all attributes, including numerical attribute cost.
Dominance checks cannot be performed if only one parameter has an indicated sign, so in your case you should use a zero prior for the cost coefficient.
Note that the last level is the base level when dummy coding.
Given your number of parameters, I would suggest increasing the number of rows while blocking the design.
Consider using more cost levels since they are far apart and the cost attribute may become dominant otherwise. For example, perhaps use 0, 500, 1000, 2000, 4000.

Code: Select all

Design
;alts = alt1*, alt2*, optout
;rows = 24
;block = 2
;eff = (mnl, d)
;model:
U(alt1)   = b1.dummy[0|0|0|0] * loc[1,2,3,4,0]   ? 0=...[base], 1=..., 2=..., 3=..., 4=...
          + b2.dummy[0|0]     * sup[1,2,0]       ? 0=...[base], 1=..., 2=...
          + b3.dummy[0|0]     * cost[500,4000,0]
          + b4.dummy[0]       * lang[1,0]        ? 0=...[base], 1=...
          + b5.dummy[0|0]     * time[1,2,0]      ? 0=...[base], 1=..., 2=...
          /
U(alt2)   = b1                 * loc
          + b2                 * sup
          + b3                 * cost
          + b4                 * lang
          + b5                 * time
          /
U(optout) = b6
$

Michiel

Dear Michiel Bliemer,
Thank you so much for your recommendation. I try to test d-error between two syntaxes:
The first syntax is as follows,

Code: Select all

Design
;alts = alt1*, alt2*, optout
;rows = 24
;block = 2
;eff = (mnl, d)
;model:
U(alt1)   = b1.dummy[0|0|0|0] * loc[1,2,3,4,0]   
          + b2.dummy[0|0]     * sup[1,2,0]       
          + b3.dummy[0|0]     * cost[500,4000,0]
          + b4.dummy[0]       * lang[1,0]       
          + b5.dummy[0|0]     * time[1,2,0]     
          /
U(alt2)   = b1                 * loc
          + b2                 * sup
          + b3                 * cost
          + b4                 * lang
          + b5                 * time
          /
U(optout) = b6
$

The d-error =0.448159 and the warning shows one or more attributes will not have level balance with the number of rows specified: alt1.loc, alt2.loc.

The second syntax is as follows,

Code: Select all

Design
;alts = alt1, alt2, optout
;rows = 12
;eff = (mnl, d)
;model:
U(alt1)  = b1.dummy[0|0|0|0]*loc[0,1,2,3,4]+b2.dummy[0|0]*sup[0,1,2]+b3*cost[0,500,4000]+b4.dummy[0]*lang[0,1]+b5.dummy[0|0]*time[0,1,2]/

U(alt2) = b1.*loc+b2*sup+b3*cost+b4.*lang+b5.*time/
U(optout) = b6
$

D-error= 0.178318 and The warming are "defaulting to prior values of zero for the following priors: 'b3, b6' "and "one or more attributes will not have level balance with the number of rows specified: alt1.loc, alt2.loc"
Let me know what you think of this comparison.
Sincerely yours,
Sukunta

You should not compare D-errors for different models. Using dummy coding or using a linear specification are two different models with a different number of parameters and it makes no sense comparing their D-errors.

Note that more rows will always result in a lower D-error because the D-error computation assumes that a respondent is given all choice tasks.

Michiel

Dear Michiel Bliemer,
My sincere thanks to you.
Sincerely yours,
Sukunta