Re: [Gretl-devel] chow test underflow (?) bug
On Tue, 8 Dec 2009, Sven Schreiber wrote:
running a regression that includes a shift dummy for 1989q3 and then
doing a Chow test for that exact date gretl omits its own split dummy
due to collinearity (ok, of course) and yields:
a test statistic F(0,148) = 1.#INF,
which is still not so problematic, but then:
a p-value of P(F(0,148)> 1.#INF) = 1.79769e+308
with a positive exponent and that is clearly wrong. I guess NA would be
more appropriate.
That is in fact NA, but we're showing its numerical value, which
is a bug. Fixed in CVS.
Just to let people know, I'll be mostly offline for a while,
resuming "normal service" later this month.
Allin.