[Limdep Nlogit List] Parametric survival models
Gustavo Garcia Manzato
gusmanzato at gmail.com
Wed Jun 2 00:38:28 EST 2010
Dear colleagues,
I'm doing some analysis using parametric survival models. I have 286860
observations and 136 variables. Running the model using any type of
distribution (Weibull, Exponential, Loglogistic, etc.), I get an error
message saying:
Line search does not improve fn. Exit iterations. Status=3
Check derivatives (with ;OUTPUT=3). This may be a solution
if several iterations have been computed, not if only one.
Error 806: (The log likelihood is flat at the current estimates.)
I already checked whether the data was loaded correctly and tried to
start from a base model, increasing the variables step by step. Except
for one trial (when I included ONE and another specific variable), all
resulted in the same error message. Then, I ran the model (the same
trials, step by step) with this ";OUTPUT=3" as suggested. With this, I
get the value of the function when it starts and when it ends. What
happens is that the values in each step are very close, even compared to
that successful trial (when I get a "normal exit from iterations").
Additionally, the value of the function remains very high, at both start
and end stages.
My concern is whether the successful trial is wrong or the other trials
are correct (given that lots of iterations - more than 100 - have been
computed). Has anyone faced a similar problem? Any ideas/suggestions on
how to resolve this?
I appreciate your attention!
Best regards,
Gustavo
More information about the Limdep
mailing list