Re: [Gretl-users] Wald test for joint significance of time dummies
On Thu, 28 Jul 2016, klazar wrote:
Dear Allin Cottrell,
I would like to ask 2 questions:
1* I am doing panel data analysis with --time-dummies option. There is
a Wald test for joint significance of time dummies implemented in
Gretl. I just want to be sure that the null hypothesis is "none of
these regressors (years) has any effect on the explained variable". So
if Chi-square(9) = 505.436 with p-value = 3.9413e-103, it means that I
should reject H0, so as a result the time-dummies should be in the
model.
Yes, that's right.
Gretl Result (for panel regression with --time----time-dummies
option)
Wald test for joint significance of time dummies
Asymptotic test statistic: Chi-square(9) = 505.436
with p-value = 3.9413e-103
2* My humble recommendation -
I am wondering, whether the expression "Wald test for joint NON
significance of time dummies" OR "Null hypothesis - all regressors are
zero" would not be better for nonadvanced users (like me) to derive
the proper results more easily.
It's not particularly easy to come up with a suitably terse phrase
that is also precise. "All regressors are zero", if you think about
it, is not right: "The true coefficients on all regressors are zero"
would be required. Arguably the word "significance" should not be in
the string, either in the positive or the negative, since what's
we're actually _testing_ for is not significance or its absence but
"zero-ness" of the parameters, so to speak. A more defensible short
phrase (but not all that informative for beginners) might be just
"Joint Wald test on time dummies".
Have a nice day and I wish you all the best in work (gretl is the
excellent tool) and also in personal life.
Thank you!
Allin Cottrell