[Limdep Nlogit List] Question regarding generalized mixed logit (GMXLOGIT)
miq at wne.uw.edu.pl
Fri Jul 15 05:19:53 AEST 2016
The GMXL model is tough to converge. Try using more draws and different
The only explanation for why the restricted model would result in a
better LL than the unrestricted model is because optimizer did not find
global optimum for at least the unrestricted one.
On 2016-07-14 18:51, Yau-Huo Shr wrote:
> I am running generalized mixed logit models with different specification
> using the GMXLOGIT command in NLOGIT 5. Because I am polling two samples
> together and interested to know if the scale and its variance are different
> between two samples, I have the option HFT and HFR included.
> After getting the results, I then wanted to do a likelihood ratio test to
> see if the theta and lambda are jointly zero and so remove the option HFT
> and HFR.
> Then I found out the restricted model have higher (less negative)
> likelihood ratio compared and pseudo R-square to the model with the option
> HFT and HFR, which I have no idea why it is the case. Could someone
> provide me some advice?
> Here is the essential part of my code:
> CREATE; IF(VERSION=1) V1=1 $
> CREATE; IF(VERSION=2) V2=1 $
> GMXLOGIT; lhs=CHOICE ; choices=1,2
> ; rhs= DSG2, DSG3, SPC2, SPC3, WTR2, WTR3, Cost,
> ; pds=12; maxit=1000; halton; draws=250; Wts=Weight;
> ; fcn= DSG2(n), DSG3(n), SPC2(n), SPC3(n), WTR2(n), WTR3(n)
> ; hft = V2; hfr=V2; cor =
> Jimmy Shr
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
More information about the Limdep