Page 1 of 1

Problem with efficient design

PostPosted: Fri Feb 21, 2020 9:55 pm
by andreb84
I am generating an efficient design to conduct a pilot study. As the signs are partly known, I've used an efficient design with small positive/negative values.
The problem is that Ngene is running since Monday, and after 4 days it's still running. However, if I click on the last iteration, I get my choice task but I don't know if I can use them.
Is it common or am I doing some mistakes? Are there any tips to reduce the running time? Sorry for the question but it is the first time I'm using Ngene
Here the script:

Code: Select all
Design
;alts=SQ, alt2, alt3
;rows=10
;eff=(mnl,d)
;model:
U(SQ) = b1 + b3[-0.0001] * TAX[109] + b4[0.0001] * PROD[60] + b5[0.0001] * PER[100] + b6[0.0001] * CHIL[100]/
U(alt2) = b2 + b3 * TAX1[25,50,109,200,300] + b4 * PROD1[30,60,90,120] + b5 * PERF1[50,100,150,200] + b6 * CHIL1[50,100,200,300] /
U(alt3) =  b3 * TAX1 + b4 * PROD1 + b5 * PERF1 + b6 * CHIL1
$


Thank you for your attention

Re: Problem with efficient design

PostPosted: Sat Feb 22, 2020 11:52 am
by Michiel Bliemer
There is no mistake, most algorithms in Ngene use randomised search to explore the solution space because typically there are billions/trillions of possible designs and Ngene cannot evaluate all of them. This means that Ngene will keep randomising parts of the design and try to find a more efficient design until the user stops the algorithm. For a simple design like this, you can probably stop the algorithm after about 10 minutes or so as Ngene will no longer be able to find more efficient designs. Note that you can use ANY design with a finite D-error reported by Ngene, but clearly designs with a lower D-error are more efficient. At some point, there is little more efficiency to be squeezed out of the design, at which point you can stop the search and take the last design.

Michiel