Simulating multinomial logit with three outcomes

[Tinatin Mumladze]

I am currently trying to simulate multinomial logit in stata (obs 1000). I have five explanatory variables and pre-determined slope coefficients. I generate the depended variable using the following:
gen p0 = 1 / ( 1 + exp(xb1) + exp(xb2) ) gen p1 = exp(xb1) / ( 1 + exp(xb1) + exp(xb2) ) gen p2 = exp(xb2) / ( 1 + exp(xb1) + exp(xb2) ) gen u = runiform() gen y = 0 replace y=1 if u>p0 % u<p1+p2 replace y=2 if u>p1+p2 After running mlogit command with base(0), slope coefficient do not converge to the true values. Also all predicted probabilities are between 0 and 0.5.

Can someone help me figure out what can be the problem?

[FernandoRios]

Hi Tinatin
Not sure about your specific problem, but perhaps the following code may help:

webuse sysdsn1, clear
mlogit insure age male nonwhite i.site
keep if e(sample)
predict p*, pr
 gen double ru=.
 gen byte insure_sim=0
 label values insure_sim insure
** simulation

program mlogit_sim, eclass
* simulate insure
replace ru=runiform()
replace insure_sim = 1 
replace insure_sim = 2  if ru> p1
replace insure_sim = 3  if ru>(p1+p2)
mlogit insure_sim age male nonwhite i.site, base(1)
end

simulate, reps(100):mlogit_sim
sum