reality based dominance in efficient design

[alexsydney]

thank you for your post.

the following syntax crashes Ngene ?


Design
;alts = A*, B*
;rows =6
;eff = (mnl,d)
;alg=mfederov
;require:
B.cost - A.cost <=1 and A.cost - B.cost <=1,
A.cost>B.cost and A.find<B.find,
A.cost>B.cost and A.walk<B.walk
;model:
U(A) = b2[(n,-0.2015755,0.1115085)] * walk[1,6,11] + b3[(n,-.5,0)] * cost[2,3,4] + b4[(n,-.2722374,0.0989455)] * find[0, 5] /
U(B) = b2 * walk + b3 * cost + b4 * find $

[Michiel Bliemer]

You have put too many constraints on the design. There only exist 6 choice tasks, so Ngene is not able to make any switches to improve the design.
I am not sure why you impose these constraints, as they will result in an inefficient design. The syntax below runs.

Code: Select all

Design
;alts = A*, B*
;rows = all
;fact
;require:
B.cost - A.cost <=1 and A.cost - B.cost <=1,
A.cost>B.cost and A.find<B.find,
A.cost>B.cost and A.walk<B.walk
;model:
U(A) = b2[(n,-0.2015755,0.1115085)] * walk[1,6,11] + b3[(n,-.5,0)] * cost[2,3,4] + b4[(n,-.2722374,0.0989455)] * find[0, 5] /
U(B) = b2 * walk + b3 * cost + b4 * find $


But note that this design is problematic, as you can see from the "find" levels, so you cannot estimate this model.

[alexsydney]

thank you for your reply.

Attribute with only two levels gives collinearity error during the estimation process for that attribute.

Difference in utility remains the same the so collinearity is evident.

Does this mean we should avoid 2 levels for an attribute ?

[Michiel Bliemer]

Two-level attributes are fine, if you remove your constraints the model can be estimated. You have just imposed to many restrictions.