[Limdep Nlogit List] dummy coding vs effects coding

Kandice L. Kleiber kleiberk at onid.orst.edu
Thu Jul 23 08:49:17 EST 2009

Hi -

So I have estimated a nested logit model and basically done everything  
that I want to do ... except find the marginal effects of my  
categorical variables.

One variable in question is education, and I have broken it down into  
3 levels, eda, edb, edc, where eda is the lowest level of education.   
In running my model I used eda as the base (ie dropped it from the  
equations).  I coded each new variable as a 0-1, thus if edb=1 then  
eda=edc=0 and if eda=1 then edb=edc=0.

However, the code that I found to run the marginal effects uses  
"effects code" ie making eda=-1 instead of =0.  If you use the effects  
coding and want to compare edb vs eda you use:

scenario: edb=1/edc=0 & edb=-1/edc=-1$

But since I have already run my model and have all my other results,  
I'm wondering if I can still use this code, given that I have used  
"dummy coding" (0-1 variables).

If so, can I use:

scenario: edb1/edc=0 & edb=0/edc=0$ ?

I guess I'm not sure if LimDep is doing the same thing that it would  
with effects coding?
Any input would be appreciated!

Kandice L. Kleiber
Graduate Teaching Assistant
Department of Economics
Oregon State University
e: kleiberk at onid.orst.edu
w: 541.737.6628

More information about the Limdep mailing list