dcreate (efficient designs / choice experiments)

[Juan Price Elton]

Dear friends: I am working with the dcreate command to build efficient designs for choice experiments. Does any of you know if there is an option to avoid dominant strategies? I´d really appreciate if any of you can help me. Best regards, Juan

[Bryan Parthum]

For what it's worth, you can wrap your design sequence inside of a loop and manually identify any alternatives that you would not like to include in your final design (such as strictly dominated strategies). What I've written also allows you to store the number of iterations before your conditions are met, and also how many choice cards failed your conditions in each iteration.

For example:

******************************************************
************** UNCOMMENT AFTER FIRST ITERATION (START)
******************************************************
/*

use N_dominated, clear   /* This file is created after the first iteration of the design */

while N_dominated > 0 {

qui{

clear all
set more off

** RANDOMLY GENERATE RANDOM SEED
local rand = runiformint(0,200000)
set seed `rand'

*/
******************************************************
**************** UNCOMMENT AFTER FIRST ITERATION (END)
******************************************************

** GENERATE EXPERIMENT DESIGN HERE

*****************************************
*************************  TEST DOMINATED
*****************************************

save CEdesign_temp, replace

reshape wide attribute1 attribute2 cost, i(card_id) j(alt)


gen dominated = (attribute11  > attribute12 &     ///
                 cost1 < cost2) |  ///
                (attribute11  < attribute12 &      ///
                 cost1 > cost2)  ///

egen N_dominated = sum(dominated)
keep N_dominated
duplicates drop

tempfile dom
save `dom'
save N_dominated, replace    /*Comment this line after first iteration*/
*append using N_dominated    /*Uncomment this line after first iteration*/
save N_dominated, replace
}
sum N_dominated              /*This allows you to see how many dominated strategies occured in the current iteration*/
qui{
use `dom', clear
}
}


*****************************************
*********************************  EXPORT
*****************************************

use CEdesign_temp, clear
save CEdesign, replace

[Juan Price Elton]

Many thanks for your prompt response Arne
Best regards

[Arne Risa Hole]

Juan

There is no option, as such, but using non-zero priors (from e.g. a pilot study) should help avoiding dominant alternatives.

Arne

[Juan Price Elton]

sorry I meant to write "dominant alternatives"