Best design if biased pilot sample

[marianne.lefebvre]

Dear Ngene community,

I have collected priors from a subsample of n=25. However, I am pretty sure this subsample is biased and may be less reluctant towards one of the level (attribute B=1) than the targetted population. What is the best way to deal with that ?

Here is the design I have in mind at the moment

Code: Select all

Design
;alts=GA*, GB*, SQ
;rows = 8
;block = 1

;eff = (mnl,d)
;model:
U(GA) =  b1.dummy[(n,-0.505,0.186)]*A[1,0] + b2.dummy[(n,0.359,0.185)]*B[1,0] + b3[(n,0.0306,0.0085)]*C[0,20,30,40,50] /
U(GB) =  b1*A + b2*B + b3*C/
U(SQ) = b0[(n,0.904,0.381)]
$

Note that I have the below error message if am modifying to rows=16 and Block=2, most likely because I have 2 attributes with only 2 levels.
Is that problematic ? especially given the uncertainty on priors ?

A valid initial random design could not be generated after approximately 10 seconds. In this time, of the 518699 attempts made, there were 0 row repetitions, 41630 alternative repetitions, and 477069 cases of dominance. There are a number of possible causes for this, including the specification of too many constraints, not having enough attributes or attribute levels for the number of rows required, and the use of too many scenario attributes. A design may yet be found, and the search will continue for 10 minutes. Alternatively, you can stop the run and alter the syntax.

Thanks for your help
Marianne

[marianne.lefebvre]

Please let me being more precise on the second question

This code, with less information on priors, is running.

Code: Select all

Design
;alts=GA*, GB*, SQ
?place an asterisk next to the names to prevent from having i) within choice task alternative repetition, ii) strict attribute level dominance and iii) choice task repetition given attribute bundle ordering

;rows = 24
?make sure rows>nb of parameters beta/(nb of alternatives-1) AND nb rows multiplies the number of levels of all attributes (2,4)

;block = 3

;eff = (mnl,d)
;model:
U(GA) =  b1.dummy[0]*A[1,0] + b2.dummy[0]*B[1,0] + i[0]*A*B + b3[0]*C[0,20,30,40,50](1-5,1-5,1-5,1-5,1-5) /
U(GB) =  b1*A + b2*B + i*A*B + b3*C/
U(SQ) = b0[(n,0.8295,0.3996)]
$

This code is not running. To make it run, I have to specify rows=8 and block=1. But I am concerned by having only one block. What is the most appropriate design ?

Code: Select all

Design
;alts=GA*, GB*, SQ
?place an asterisk next to the names to prevent from having i) within choice task alternative repetition, ii) strict attribute level dominance and iii) choice task repetition given attribute bundle ordering

;rows = 24
?make sure rows>nb of parameters beta/(nb of alternatives-1) AND nb rows multiplies the number of levels of all attributes (2,4)

;block = 3

;eff = (mnl,d)
;model:
U(GA) =  b1.dummy[(n,-0.6812,0.3421)]*A[1,0] + b2.dummy[0]*B[1,0] + i[0]*A*B + b3[(n,0.0307,0.0085)]*C[0,20,30,40,50](1-5,1-5,1-5,1-5,1-5) /
U(GB) =  b1*A + b2*B + i*A*B + b3*C/
U(SQ) = b0[(n,0.8295,0.3996)]
$

Thanks