I generally simulate choice responses manually in Excel, which is quite easy, Ngene does not have a separate functionality to do this, but it can be tricked to do this. Random draws are used for drawing parameter values from random coefficients to compute the D-error for mixed logit models, so that is not what you need.
Suppose that you have created a design with the following syntax:
- Code: Select all
design
;alts = route1, route2
;rows = 12
;eff = (mnl,d)
;model:
U(route1) = tt-0.2] * traveltime[5,10,15,20] + tc[-1.2] * travelcost[1,2,3,4] /
U(route2) = tt[-0.2] * traveltime[5,10,15,20] + tc[-1.2] * travelcost[1,2,3,4]
$
You can then save the design in Ngene, say as design.ngd. You can then evaluate the design, and put in extra syntax that mimics a panel mixed logit model with random coefficients, such as below, where a normally distributed pameters with a mean of -0.2 and a standard deviation of 0 is essentially a fixed parameter with a value of -0.2. This syntax will create a sample of 100 respondents with simulated choices, ;rep = 100 indicates creating a sample of 100 respondents. This command only works with panel mixed logit models as sample generation is not needed to compute the D-error for other model types.
- Code: Select all
design
;alts = route1, route2
;rows = 12
;eff = (mnl,d)
;eval = design.ngd
;rdraws = gauss(1)
;rep = 100
;model:
U(route1) = tt[n,-0.2,0] * traveltime[5,10,15,20] + tc[n,-1.2,0] * travelcost[1,2,3,4] /
U(route2) = tt[n,-0.2,0] * traveltime[5,10,15,20] + tc[n,-1.2,0] * travelcost[1,2,3,4]
$
You can find the sample by opening the design, clcking on "Design properties, RP Panel" and then click on "Sample", where you will see simulated choice observations in the column "Choice". You can use this data to estimate a model in model estimation software like Nlogit, Biogeme, or Apollo to verify that you can estimate your model parameters using the design.
Michiel