|
Note: This FAQ is for users of Stata 6, an older version of Stata.
It is not relevant for more recent versions.
Stata 6: Where are the Wald tests for zinb that appear in the manual?
|
Title
|
|
Stata 6: Wald tests associated with the alpha parameter
|
|
Author
|
James Hardin, StataCorp
|
|
Date
|
January 1999
|
The alpha parameter is a positive value that is parameterized using the
natural log. The equivalent test of alpha=0 is thus log(alpha) = -infinity.
In the maximization, the value of log(alpha) is limited to the range [-18,
18] such that we limit alpha to the range [1.523e-08, 65659969].
Note: the actual range that can be achieved is somewhat wide
as the optimizer will take one final step after any correction is made in
the calculation.
The limiting case is alpha=0 where the zero-inflated negative binomial
estimator does not differ from the zero-inflated Poisson estimator or, when
combined with testing that all of the inflation equation parameters are
zero, where the zero-inflated negative binomial model does not differ from
the Poisson estimator.
At the initial time of preparing the documentation of the manuals, we
considered testing whether log(alpha)=-18 (since this was the limiting
value). After the manuals were printed, we revisited this subject and
discovered that we did not, in fact, want to include this test. The
likelihood-ratio test is the only test that one should consider in these
limiting cases.
So, a test involving the alpha parameter is only included if you specify the
nbreg and/or zip options.
|