[Limdep Nlogit List] Marginal Effects in Truncation/Tobit Model

Kalyan Chakraborty kchakrab at emporia.edu
Tue Sep 7 09:58:00 EST 2010


Prof. Greene,

Thank you so much for your time explaining me the nature of the problem and its solution. The DEA efficiency estimates (the dep var in my model) in the data are in fact a panel estimate and now that I have arranged the data accordingly, I fould Truncated Random rather than Fixed effect model fits the data far better and the results are as expected. The scale parameter is now changed from 1 to 0.9916 for the marginal effects and the cofficients for marginal effects are close to the parameters from the REM but are not identical.

Thank you again.

Kal 

Kalyan (Kal) Chakraborty, Ph.D
Associate Professor (Economics)
http://www.emporia.edu/~chakrabk/
Department of Accounting Information Systems
School of Business
Emporia State University
1200 Commercial Street, Box 4057
Emporia, KS 66801
(Tel) 620-341-5913
(Fax) 620-341-6346
e-mail: kchakrab at emporia.edu
>>> William Greene <wgreene at stern.nyu.edu> 09/06/10 4:54 AM >>>
Professor Chakraborty.
(1) Your data are not arranged correctly for the panel data estimator.
The first four rows should be for firm 1, the second four for firm 2,
etc.
(2) The problem you note about the scale occurs when the truncation model
is not the correct model for your data. For the model you suggested, the
data should be piled up near .3 and 1. Evidently yours are not, so the MLE
is not different from OLS. When this is the case, the partial effects
will differ trivially or not at all from the OLS regression coefficients.
(3) The FEM only supports truncation at one tail of the distribution.
(4) As small as your sample is, you should expect the MLE to be a little
erratic.
(5) Since the original DEA computations did not relate to any panel data
structure in the data, chances are this second step you are computing 
will not reveal such effects either.  This is a methodological issue, not
obviously visible in the data, but it is likely to be a partial explanation
if results don't come out the way you are expecting. Since your original 
computations were computed as if the data were a cross section, you 
should probably treat the second step likewise.
(6) YOu can solve problems 3 and 5 at the same time by creating 14 firm
dummy variables (drop one of them, however) and include them in a simple
cross section version of the truncated regression. That way you can have
both limit values. (This probably will not change your ultimate results
very much however.)  Be sure to include ONE in the model with the 14 firm
dummies.
/B. Greene

----- Original Message -----
From: "Kalyan Chakraborty" <kchakrab at emporia.edu>
To: limdep at limdep.itls.usyd.edu.au
Sent: Monday, September 6, 2010 12:09:28 AM GMT -05:00 Colombia
Subject: [Limdep Nlogit List] Marginal Effects in Truncation/Tobit Model

Hi Limdep Users,

I have one quick question about the output from marginal effects from Truncate Regression. I have a panel data for 14 firms over 4 year period and the data are arranged by year. For example, first 14 rows denote data on Y and X variables for 14 firms for Year-1, the second 14 rows are for Year-2 and so on. I have a total of 56 rows. Most of the X variables are in percent. The dependent variable (Y) takes any random value between 0.30 and 1.00 (efficiency scores from a DEA model). I prefer to use Truncation command specified on page-R17-73 (Version-9).

TRUNCATION;LHS=Y
	; RHS= X1,X2,X3,X4,X5,DUMM; FEM; MARGINAL EFFECTS; PDS=4 ; LIMITS=.3,1$ 
  
I tried both FEM and REM models but the problem is the beta coefficients are identical to marginal effects. I also found that 1 is being used as a scale factor for calculating partial derivatives in marginal effects. According to the manual marginal effects could be close to estimated parameters for a best fit model but there is no mention about exactly identical numbers. Similar results are obtained in a Tobit model using the same data.

Any help would be appreciated.

Thank you.

Kal


Kalyan (Kal) Chakraborty, Ph.D
Associate Professor (Economics)
http://www.emporia.edu/~chakrabk/
Department of Accounting Information Systems
School of Business
Emporia State University
1200 Commercial Street, Box 4057
Emporia, KS 66801
(Tel) 620-341-5913
(Fax) 620-341-6346
e-mail: kchakrab at emporia.edu
_______________________________________________
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