[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Frontier estimation using truncated normal option

Subject   st: Frontier estimation using truncated normal option
Date   Fri, 8 Feb 2008 23:50:56 -0800 (PST)

Dear Statalisters,

My question is fairly simple but I have struggled to
find a good answer. I am estimating a stochastic
production frontier and I want to properly test
whether technical inefficiency is present (and thus
whether frontier estimation is appropriate). The
stata routine does this for you when the default
half-normal model is used, but I am using the
truncated normal model with explanatory variables for
inefficiency (u = d*Z + w) through the cm(variables)

If I want to do a likelihood ratio test of
H_o: sigma_u = 0
With a test statistic of LR = -2*(L(H_o)-L(H_a),
what is the appropriate value for L(H_o)?

Is it the e(ll_c) value saved in the frontier results
(in which case e(chi2_c) is my test statistic) or is
it the implied log-likelihood from an OLS version of
the production function (assuming normal errors),
including Z directly as a set of controls on the
right hand side?

I thought they would be numerically equivalent (if
sigma_u = 0, the frontier collapses to a simple
linear regression - and the likelihoods for normal
errors should be the same). In my data, e(ll) from
the OLS and e(ll_c) from the frontier are only the
same when Z is empty (no cm(variables) option
specified). Does anyone know why they are not
generally equivalent and which one is correct to use
when Z is included? What is this "ll_c" value stata
is calculating?

Thank you!
Ben Gilbert

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index