[Limdep Nlogit List] Controlling for number of parameters in model selection tests

[Mikołaj Czajkowski]

Dear All,

I am using the Vuong test as a model-selection test, when choosing 
between NL, ECM, RPL, MNP etc.

In order to control for number of parameters in different models I want 
to use the Schwarz's (1978) correction based on Bayesian Information 
criteria, that is subtracting:

- ( (p/2)ln(n) - (q/2)ln(n) )
from the usual Voung test statistic. (where p and q are numbers of 
parameters in models 1 and 2 respectively).

Now this might be a simple question, I would like to consult my 
understanding of this. My question regards NL. After estimating the 
model the output would give parameter estimates, IV estimates and 
Underlying standard deviation. This way B matrix contains also the 
results for underlying standard deviation for each IV parameter.

(1) Should these standard deviations be counted as parameters of the 
model or not?

(2) In addition - if one of the IV parameters is restricted (say to 1 or 
two parameters restricted to be equal) should they be counted as 
separate parameters anyway?

Thank you very much in advance for your help.

Best regards,

-- 
     Mikołaj Czajkowski

     Warsaw Ecological Economics Center
     Warsaw University
     http://www.woee.pl/