I am currently trying to generate a design for a project in which we need to value watershed restoration. Specifically we are looking to find the best policy mix with which to manage stormwater problems in the area.
The problem at hand is very complex in that many of my attributes are nested and so I need to specify a lot of conditions in Ngene. I ran the following syntax:
- Code: Select all
Design
;alts = X, Y, sq
;rows = 24
;block = 3
;eff= (mnl, d)
;cond:
If (X.A=0, X.B=0),
If (Y.A=0, Y.B=0),
If (X.A=0, X.C=0),
If (Y.A=0, Y.C=0),
If(X.D=0, X.C=0),
If(Y.D=0, Y.C=0),
If (X.D=0, X.E=0),
If (Y.D=0, Y.E=0),
If (X.D>0, X.E>0),
If (Y.D>0, Y.E>0),
If (X.D=0, X.B=0),
If (Y.D=0, Y.B=0)
;model:
U(X) = b1[-0.1]* A[1,2,3,0](5-8,5-8,5-8,5-8) + b2[0.3]* B[1,2,3,0] + b3[0.3] * C[1,2,3,0] + b4[-0.1]* D[1,2,3,0] (5-8,5-8,5-8,5-8) + b5[-0.15] * E[1,2,3,0] + b6[0.5] *F [1,2,0] /
U(Y) = b1 * A + b2 * B + b3 * C + b4*D + b5*E +b6*F
$
Although in the Ngene Manual it mentions that only the nested attributes will not display attribute level balance (p.154 Ngene Manual), this does not appear to be the case in my problem. For example, while attribute A and D are not nested per se, these attributes don’t display attribute level balance in the design generated by Ngene. I guess this is because some of the nesting structures overlap?
Because I wanted the attributes to be balanced as far as possible for attribute A & D (attribute level balance would require each level occurring 6 times in the design – I wanted to have each level occur at least 5 times), I specified that the levels of attributes A & D should occur between 5 and 8 times (following the explanation on p.151 of the manual):
- Code: Select all
;model:
U(X) = b1[-0.1]* A[1,2,3,0](5-8,5-8,5-8,5-8) + b2[0.3]* B[1,2,3,0] + b3[0.3] * C[1,2,3,0] + b4[-0.1]* D[1,2,3,0](5-8,5-8,5-8,5-8) + b5[-0.15] * E[1,2,3,0] + b6[0.5] *F [1,2,0] /
U(Y) = b1 * A + b2 * B + b3 * C + b4*D + b5*E +b6*F
$
When I run this however, Ngene doesn’t seem to follow the specifications I stipulated with levels in attribute A & D still occurring 4 or even 9 times.
Am I doing something wrong?
Then, what is the consequence of using so many conditions? Is there anything else that I need to keep in mind besides specifying an adequate number of rows?
On this point, are 24 rows enough? As all my variables will be nominal, I have (4-1) x 6 + (3-1) x 1 + 1 = 3x6 +2 + 1= 21 degrees of freedom. So given this, 24 rows should be sufficient right?
What rule do I follow to find the minimum number of rows required? In the Ngene manual on p.72 I find the following formula: S= K/(J-1) which in my case = 20/(3-1) = 10 rows. But I suspect this is not for efficient designs?
Any assistance would be much appreciated!