[Limdep Nlogit List] Question regarding generalized mixed logit (GMXLOGIT)

[Mikołaj Czajkowski]

Dear Jimmy

The GMXL model is tough to converge. Try using more draws and different 
starting points.
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.

Best regards,
Mik



On 2016-07-14 18:51, Yau-Huo Shr wrote:
> Hi
>
> 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 =
> 1,
> 1,1,
> 0,0,1,
> 0,0,1,1,
> 0,0,0,0,1,
> 0,0,0,0,1,1
> $
>
> Thanks!
> Jimmy Shr
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au