#98 2SLS summary: use chisq test for model

[KTTK]

I noticed a slight difference between the summary statistics for 2SLS between Stata and Gretl:
Gretl uses a F test while Stata uses a chisq test
(see e.g., https://www.stata.com/manuals13/rivregress.pdf , example 1, p. 6)
coefficient tests seems to be carried out by both programmes using the normal distribution (z values).

Note: This might be related to ticket #93 (random effects model: F statistic or wald chisq statistic ))

Any reference about the "correct" test and the distribution? Same argument as in #93?

[Sven S.]

First, I would really recommend to re-join the gretl (users') mailing list, where such discussions are very welcome.
So, since no feature is requested, I will close this.
However, the question as such is valid; my answer would be in principle, yes, similar to that in the other ticket. (Asymptotically it's all the same anyway.) I'm currently looking at eq. (8.55) in Davidson&MacKinnon for example, where an F statistic for the IV case is constructed via artificial regressions. So the only question is, which d.o.f. does gretl use here? (I haven't looked at the actual output right now, probably it's given in the output.)