[Limdep Nlogit List] Sample Selection Model - Error 806

Shane Murray shane.murray at memetrics.com
Tue Aug 7 20:47:59 EST 2007


Mr Greene,

Thankyou very much for your detailed response.
I will attempt to re-specify the model.

Shane Murray


-----Original Message-----
From:	William Greene [mailto:wgreene at stern.nyu.edu]
Sent:	Tue 07/08/2007 20:11
To:	Limdep and Nlogit Mailing List
Cc:	
Subject:	Re: [Limdep Nlogit List] Sample Selection Model - Error 806

Dear Mr. Murray:  You are essentially fitting a model to a column of zeros.  In such
a case, anything (or more likely, nothing) goes.  Note also that you are using the 
same variables in the probit equation that you are using in the regression, so the
control variable is probably nearly collinear with the rest of the variables in X.  The 
806 warning means exactly that - in the dimension of RHO, the log likelihood is 
almost flat - different values of RHO produce the same logL.  In the situation you
have described, the only thing you can do to produce a better solution is to improve
the specification of the model.  As it stands, particularly since the regressors are
dummy variables, you are probably able to predict YGT0 almost perfectly.  This in 
itself is not good.  There should be some differences between the two equations, 
as well.
   The diagnostics you are getting are a result of the data set, not the program.
There is no program setting that you can use to fix the problem.  The specification
of the model and the very sparse nature of the LHS variable look like the culprits.
/B. Greene

************************************************
Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012
Ph. 212.998.0876
Fax. 212.995.4218
URL. http://www.stern.nyu.edu/~wgreene
Email. wgreene at stern.nyu.edu
************************************************

----- Original Message -----
From: Shane Murray <shane.murray at memetrics.com>
Date: Tuesday, August 7, 2007 3:47 am
Subject: [Limdep Nlogit List] Sample Selection Model - Error 806

> I'm attempting to run a sample selection model on a continuous 
> dependentvariable with a high degree of censoring (approximately 
> 88% zeros). 
> 
> Commands as follows:
> Namelist	;x=one,x12,x13,x14,x15,x16,x22,x31,x33,x41,x42,x43,x52 $
> Probit	;lhs=YGT0 ;rhs=x ;hold $
> selection	;lhs=LOGY ;rhs=x ;mle; maxit=1000 $
> 
> Note that my independent variables are nominal variables, which I have
> dummy coded.
> 
> I have taken a couple of random slices of the data (the dataset has
> approx 600,000 records), and when I run the model I am seeing very
> different results for each slice.
> 
> In particular, RHO can vary from +1 to -1.
> 
> The following message is also presented:
> "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.)"
> 
> What does it mean by "the log likelihood is flat at the current
> estimates"?
> Is there something I can do to arrive at a more stable solution?
> 
> 
> Any assistance is greatly appreciated.
> 
> Thanks,
> Shane
> 
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
> 

_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au






More information about the Limdep mailing list