David Airey wrote:
I asked if anyone knew why the command "test" did not work for me
after the command "boxcox". Stata support said this occurs because
the required standard errors are not available for "test" to take
advantage of following a boxcox model. Clearly, however, boxcox
postestimation lists test as an available command. Maybe Stata Corp
should update their help files to reflect what postestimation
commands are actually available after boxcox?
--------------------------------------------------------------------------------
You can use -test- after -boxcox-. It will only be able to work for those
coefficients that have a nonzero diagonal cell in the e(V) matrix, however.
For the others, you'll need to use the built-in likelihood-ratio test as
your example illustrated in your original post, or, perhaps, a resampling
test--the help file states that both -bootstrap- and -jackknife- are
allowed.
You can still use the likelihood-ratio test if you need to simultaneously
test two or more coefficients, such as with categorical interaction terms in
your example, but you'll need to do it manually, using -estimates save-
and -lrtest-.
