[Limdep Nlogit List] Conditional Logit Help required!

[Griffiths, Tom (Ecad)]

Dear Limdep Users

I am a Master's student in the UK estimating a Conditional Logit model
using Limdep 8.0 (the application is a Random Utility Travel Cost site
choice model). I have generated a set of highly intuitive results, with
statistically significant explanatory variables in places where I expect
them. However, when I try to include alternative-specific constants
(i.e. site dummies) to enhance fit the model does not converge in a
reasonable number of iterations. The result is the same whether I use
the 'ONE' command to specify the ASCs or whether I do this manually from
the datasheet (of course being careful to avoid the dummy variable
trap). Below is a simplified model (including only ASCs) to illustrate
my problems.

My initial thoughts were that perhaps
i)	my sample size is too modest (although the survey covered over
200 individuals and there are a minimum of 45 observations per site), or
alternatively;
ii)	co-linearity exists between attributes across sites (although
the priors do not seem to support this).

I would be grateful for any suggestions as to where I may be going
wrong!


--> NLOGIT;LHS=TRIP1
    ; RHS=ONE,GC
    ;
CHOICES=CWM,CYB,NYA,AFAN,GWY,MACH,BRECH/.279,.206,.05,.312,.104,.023,.02
6
    ; FREQUENCIES
    ; DESCRIBE$
Hessian is not definite at current values.
Switching to BFGS (gradient based) method.
(Not a failure. Just looking for a better algorithm.)
Maximum iterations reached. Exit iterations with status=1.

+---------------------------------------------+
| Discrete choice (multinomial logit) model   |
| Maximum Likelihood Estimates                |
| Model estimated: Nov 28, 2007 at 00:25:28PM.|
| Dependent variable               Choice     |
| Weighting variable                 None     |
| Number of observations              209     |
| Iterations completed                101     |
| Log likelihood function       -9641.896     |
| Log-L for Choice   model =  -9641.89580     |
| R2=1-LogL/LogL*  Log-L fncn  R-sqrd  RsqAdj |
| No coefficients  -7536.5100 -.27936 -.28654 |
| Constants only.  Must be computed directly. |
|                  Use NLOGIT ;...; RHS=ONE $ |
| Vars. corrected for choice based sampling   |
| Response data are given as frequencies.     |
| Number of obs.=   209, skipped   0 bad obs. |
+---------------------------------------------+
+---------+--------------+----------------+--------+---------+
|Variable | Coefficient  | Standard Error |b/St.Er.|P[|Z|>z] |
+---------+--------------+----------------+--------+---------+
 GC          -2500471127.       1675.0454 ********   .0000
 A_CWM        19685662.48 ........(Fixed Parameter)........
 A_CYB        11451003.69       1678.1731 6823.494   .0000
 A_NYA       -12631396.05 ........(Fixed Parameter)........
 A_AFAN       20813002.15 ........(Fixed Parameter)........
 A_GWY       -4191204.638 ........(Fixed Parameter)........
 A_MACH      -16772019.33 ........(Fixed Parameter)........

Kind regards

Tom Griffiths

E-mail; trg22 at bath.ac.uk




The original of this email was scanned for viruses by the Government Secure Intranet virus scanning service supplied by Cable&Wireless in partnership with MessageLabs. (CCTM Certificate Number 2006/04/0007.) On leaving the GSi this email was certified virus free.
Communications via the GSi may be automatically logged, monitored and/or recorded for legal purposes.

Cafodd fersiwn gwreiddiol y neges e-bost hon ei sganio gan wasanaeth gwrthfeirysau Mewnrwyd Ddiogel y Llywodraeth (GSI) a ddarparwyd gan Cable&Wireless mewn partneriaeth â MessageLabs. (Rhif Tystysgrif CCTM 2006/04/0007). Wrth adael y GSI nid oedd unrhyw feirws yn gysylltiedig â’r neges hon.
Mae’n ddigon posibl y bydd unrhyw ohebiaeth drwy’r GSI yn cael ei logio, ei monitro a/neu ei chofnodi yn awtomatig am resymau cyfreithiol.