[Limdep Nlogit List] problem solved: thanks!

[John C. Whitehead]

Hi all,

Last week I sent the message below to this list. I received a number of
suggestions, each of which solved the problem:

1. scale up X2, X3, and X4 (I multiplied by 1000)
2. use the SAS results as starting values
3. add ;ALG=BFGS

I'm now using #3 and cranking through LCLOGIT and RPLOGIT models with no
problems.

Many thanks!

John Whitehead

----- Original Message -----
From: "John C. Whitehead" <john.c.whitehead at gmail.com>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Saturday, March 21, 2009 11:36:59 AM GMT -05:00 US/Canada Eastern
Subject: [Limdep Nlogit List] nlogit with many alternatives

Hi All,

I'm trying to estimate conditional and nested logit models in NLOGIT. I have
1086 cases and 71 alternatives (77106 lines of data). The model is fairly
simple with only 5 independent variables:

Var  Mean     Min     Max
X1    234      .60      777
X2   .023    .00      4.500
X3    .055    .00      2.85
X4   .018     .00      1.46
X5    2.85      1.38     4.98

When I estimate the CL and NL models with only X1, X2 and X5 I'm able to
replicate my results from SAS PROC MDC. However, when I add X3 and X4 my
estimates blow up:


|Variable| Coefficient  | Standard Error |b/St.Er.|P[|Z|>z]|
+--------+--------------+----------------+--------+--------+
 X1      |    -.52294252       .00037816  ********   .0000
 X2      |    .162233D+14      .10083492  ********   .0000
 X3      |    .159318D+07      .11121567  ********   .0000
 X4      |    1231.81419       .14232936  8654.674   .0000
 X5      |    20.6940287       .05183529   399.227   .0000

I've tried scaling down X1 by dividing by 100 (no luck). The full model runs
just fine as OLS and Probit models in LIMDEP.

Any ideas about what I might be doing wrong?

Thanks for any help,

John Whitehead