[Gretl-users] Re: gretl, regression
On 11/06/2024 17:30, gene shackman wrote:
Hi all
I'm working on regression, in this case, just OLS. I used Birth Rate,
Agriculture, Services, GDP Per Capita and Population to predict Infant
Mortality.
So far, there is good news. The values that gretl generates are the same
values, or close enough, to the values SAS generates.
Well done, SAS guys!
There are some
differences with naming outcomes, but I'm okay with that.
I have some questions. The default output (what first appears when doing
the model / OLS) includes these
Mean dependent var 20.75258
Sum squared resid 19737.38
S.E. of regression 9.581323
R-squared 0.759917
Adjusted R-squared 0.754334
F(5, 215) 136.1049
P-value(F) 1.39E-64
S.D. dependent var 19.33093
Any particular reason the Sum squared resid, S.E. of regression and the
S.D. dependent var are included?
Because these are descriptive statistics that an econometrician might
want to see, especially the SSR. It's no coincidence that the SSR is
reported by all econometrics-oriented software packages.
The S.E. of regression is the Root MSE on SAS's default output,
so that
makes sense.
And the output has the mean of the dependent var, so okay, the S.D.
dependent var could make sense too.
But why is the Sum squared resid on the default output? That's on the
ANOVA (as the error sum of squares), but error mean squares is also on
ANOVA but not on the default output. And the model sum of squares and
mean square are also on the ANOVA but not on the default output.
Again, this is something you may need: for example, you may want to use
the SSR to compare two different models via a Wald test, or an F test
"by hand".
That is, the ANOVA has regression sum of squares and regression mean
square, and the residual sum of squares and the residual mean square.
But the the only thing from the ANOVA that's on the default output is
the residual sum of squares. Why that one?
Because hardly anybody uses ANOVA tables in econometrics. In case you
need it, it's just a couple of clicks away.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------