# [Limdep Nlogit List] Marginal effects in a random parameter model

William Greene wgreene at stern.nyu.edu
Fri Dec 5 03:25:30 EST 2008

```Jaap. I would not compute those in that fashion. The computation
of the "firm specific" estimates is automated in limdep. Add
;PARAMETERS to the command, and a new matrix, in your case with
two columns, named BETA_I will be computed.  But, these are the
E[coefficient | all data for i] described in the manual. It is
much more involved than your expression. These are essentially
the same as a Bayesian posterior mean for beta(i).  Note that the
MEs for x1 and x2 are just the raw coefficients.
Regards,
Bill

----- Original Message -----
From: "Jaap Bos" <jwb_bos at yahoo.com>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Thursday, December 4, 2008 10:23:16 AM GMT -05:00 US/Canada Eastern
Subject: [Limdep Nlogit List] Marginal effects in a random parameter model

Dear all,

I have estimated the following specification:

frontier;lhs=tc;rhs=one,x1,x2,y1,y2;pds;fcn=y1(N),y2(N);rpm=y1_,y2_;cost\$

Where y1_ and y2_ are the firm specific means for y1 and y2.

Now i want to calculate the marginal effects for x1, x2 and y1 and y2 for all firms (>5000) in my data set.

My general question is: how to go about to do this?

More specifically:

(a) Is the average marginal effect for y1, the coefficient on y1 + the scale parameter for y1 * y1_ + scale parameter for y2 * y2_ ?

(b) Do i need the "Heterogeneity in the means of random parameters" to adjust for correlations among my z's?

gr + thanks, Jaap Bos
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au

```