Restricting attributes levels for selected alternative.

[izakro]

Hi all,

The following design ought to fix the att.levels highlighted in pink, and reject the option that led.quality equals to 3.
I'm not sure what am I doing wrong.
Best regards,
Rotem

design
;alts = led*, noled*, bau*
;rows = 18
;block = 3
;eff = (mnl,d)
;alg = mfederov
;reject:
led.quality=3,
led.quality=1 and led.cost_l>5,
noled.esaving_nl=20 and noled.cost_nl<250
;model:
U(led) = led[0]
+ intensity.dummy[0.001|0.002]*intensity[2,3,1] ? 1 = 20% reduction, 2 = no change, 3 =20% increase
+ quality.dummy[0.001|0.002] * quality[2,3,1] ? 1 = high glair and discomfort , 2 = moderate glair and discomfort, 3 = no glair and discomfort
+ esaving [0.00001]* esaving_l[40,60]
+ cost[-0.00001] * cost_l[5,40,80,120]/

U(noled) =noled[0]
+ intensity * intensity_nl[2,1]
+ quality * quality_nl[3]
+ esaving* esaving_nl[20,40]
+ cost * cost_nl[150,200,250]/

U(bau)= intensity *intensity_b[2]
+quality * quality_b[3]
+esaving * esaving_b[0]
+cost*cost_b[350] $

[Michiel Bliemer]

There are a few issues:

1. Your design is extremely constrained, the candidate set is small and requires using ;alg = mfederov(candidates = 200)
2. Fixing levels for dummy coded variables needs to be done using the ;require constraint
3. led never has quality level 3, while noloed and bau always have quality level 3. This means that quality level 3 is perfectly confounded with the constants and cannot be estimated (and hence Ngene will produce Undefined, infinite, D-errors). The only way to avoid this issue is to remove this level from your model, which means that this information will be merged with the alternative-specific constants, or to allow variation of quality within noled or bau, or to allow level 3 to appear in led.

I fixed up the syntax, see below.

Code: Select all

design
;alts = led*, noled*, bau*
;rows = 18
;block = 3
;eff = (mnl,d)
;alg = mfederov(candidates=200)
;require:
bau.intensity=2
;reject:
led.quality=3,
led.quality=1 and led.cost_l>5,
noled.esaving_nl=20 and noled.cost_nl<250,
noled.intensity=3
;model:
U(led)   = led[0]
         + intensity.dummy[0.001|0.002] * intensity[2,3,1] ? 1 = 20% reduction, 2 = no change, 3 =20% increase
         + quality.dummy[0.001]         * quality[2,1]     ? 1 = high glair and discomfort , 2 = moderate glair and discomfort
         + esaving [0.00001]            * esaving_l[40,60]
         + cost[-0.00001]               * cost_l[5,40,80,120]
         /
U(noled) = noled[0]
         + intensity                    * intensity
         + esaving                      * esaving_nl[20,40]
         + cost                         * cost_nl[150,200,250]
         /
U(bau)  = intensity                     * intensity
        + esaving                       * esaving_b[0]
        + cost                          * cost_b[350] 
$

Michiel

[izakro]

Thank you Michiel. All three comments really helped me understand!
Last query.. Since the design it meant for Hebrew speaker participants, I'm trying to format the atts levels in Hebrew.
This worked selectively should I 'fight' with ngene? or leave it in English and then translate the formatted scenarios to Hebrew?
BTW, It doesn't let me send the message with the design here (I received General Error, SQL ERROR [ mysql4 ]

Best,
Rotem.

[Michiel Bliemer]

The formatted scenarios in Ngene are merely for the analyst to check whether the choice tasks make sense. For the actual data collection you need to transfer the design that Ngene creates into a survey instrument such as SurveyEngine (which can read Ngene designs), Qualtrics (using their CBC add-on module), or other survey software that has the capability of presenting choice experiments. You would do the translation to Hebrew in this survey instrument, not in Ngene.

Michiel

[izakro]

Thanks!