fcn

[mburton]

I am exploring the fcn notation in the design below. I need a set of attributes to add to 10, but the syntax below never finds a valid design.
I know that potentially I will generate a value for alt2.none which is negative, but I am taking the line in the manual as guidence not a rule:
"Care must be taken to ensure that no combination of explicitly defined probabilities can exceed one." (or 10 in my case). It works if I replace the 10 by 20, so it seems to choke on generating -ve values.

My intention is to limit designs which have -ve values for alt2.none using a reject condition later, but need to get beyond this first stage.

Design
;alts = alt1, alt2
;rows = 12


;eff = (mnl,d)

;alg=swap

;model:

U(alt1) = b1*none1[2]+b2*med1[4]+b3*high1[2]+b4*ext1[2]/
U(alt2) = b1*none[fcn(10-alt2.med-alt2.high-alt2.ext)]+b2*med[0,4,6,8]+b3*high[0,2,4]+b4*ext[0,2]$

[mburton]

so, to make this an explicit question:
is the failure to generate a valid design because my syntax falls foul of a restriction that the fcn command should not generate -ve values, or because of some other error on my part?

[Andrew Collins]

There is no problem with negative numbers. There appears to be some strange behaviour around the number 10. I am looking into this further, but in the meantime substituting a close number like 10.01 for 10 appears to work. This could be used for now, with the calculated attribute level updated once the design has been completed. The impact on the efficiency, if priors are used, would be minimal.