Page 1 of 1

The efficient design is taking too long to run

PostPosted: Wed May 03, 2017 12:04 pm
by thames_k
Hi,
I used the following efficient design for estimating the MNL model (as explained in section 7.2.2. of the manual) but its running for the past 5 days. Can you please check if there's anything wrong in my coding or is this normal? It's a labelled choice experiment with 2 alternatives+ a status quo.

Design
;alts =alt1, alt2, alt3
;rows = 12
;eff= (mnl,d)
;model:
U(alt1) =b1[-1.4] +b2[-2.6]*A[0,1,2]+b3[-0.0]*B[0,1,2]+b4[0.03]*C[0,1,2]+b5[-0.03]*D[0,1,2]/
U(alt2) = b6[-2.0]+ b2 *A + b3 *B + b4 *C + b5 *D$

Many thanks
Thames

Re: The efficient design is taking too long to run

PostPosted: Wed May 10, 2017 8:08 pm
by Michiel Bliemer
Your syntax is fine. In case of efficient designs, the algorithm in Ngene will NEVER terminate by itself (unless you specify a maximum time limit in your syntax or in the settings), since there exist literally billions of possible designs, and it may take years for Ngene to go through all combinations. So we leave it to the user to terminate the algorithm. You can see in the output window that the D-error decreases, but at some point the D-error no longer decreases (this happens typically already after a few minutes for simple models like these), and then you can simply stop it and use the last design it has found. There may exist a more efficient design, but it will only be marginally more efficient while it may take hours or days to find it.