8:38 p.m.
Allin,
Playing with the fatality panel data set from the Stock and Watson
sample files, I have notices that a problem crops up occasionally with
scatterplots.
Restricting the data set to a certain year (e.g. year=1984), I have
noticed when restoring back the full range and then applying some other
restriction (e.g. year=1982) when I plot a new scatterplot I DONT have
the OLS line appearing in the scatterplot NEITHER is there an option to
add one!
I know that I am not being specific enough as to when this happens; if
you think this merits special attention I would be happy to send a more
detailed message with specific things I did to cause this problem.
John Paravantis
University of Piraeus
2:38 p.m.
On Fri, 10 Mar 2006, John Paravantis wrote:
Playing with the fatality panel data set from the Stock and Watson
sample files, I have notices that a problem crops up occasionally
with scatterplots.
Restricting the data set to a certain year (e.g. year=1984), I
have noticed when restoring back the full range and then applying
some other restriction (e.g. year=1982) when I plot a new
scatterplot I DONT have the OLS line appearing in the scatterplot
NEITHER is there an option to add one!
The OLS fitted line is displayed in a scatter plot only if the slope
coefficient is significant at the 10 percent level. (If you want an
actual versus fitted graph regardless of the significance of the
fit, you can estimate an OLS model and explicitly request such a
plot via the menus).
If you think the fitted line is not being shown even when the slope
is significant, please give a recipe for creating the bad graph.
Thanks.
A feature that might be nice to have is the option of adding various
sorts of fitted line to any X-Y scatter: linear, quadratic,
exponential... but I don't want to hold up version 1.5.1 waiting
for that right now.
Allin Cottrell
2:49 p.m.
Allin Cottrell schrieb:
The OLS fitted line is displayed in a scatter plot only if the slope
coefficient is significant at the 10 percent level. (If you want an
actual versus fitted graph regardless of the significance of the fit,
you can estimate an OLS model and explicitly request such a plot via the
menus).
Good to know, because during lectures I couldn't explain to my students
why the line was sometimes there, sometimes not. To be honest, I think
it's better to not make it dependent on significance. At the stage of
plotting without estimating a model, I would never have guessed that
already some testing was going on.
A feature that might be nice to have is the option of adding various
sorts of fitted line to any X-Y scatter: linear, quadratic,
exponential... but I don't want to hold up version 1.5.1 waiting for
that right now.
Yes on both accounts.
-sven
3:05 p.m.
On Mon, March 13, 2006 15:49, Sven Schreiber wrote:
I agree with Sven. I'd rather have it turned off by default in any case, with
the option of plotting it if wanted. After all, if the data points align on a
stright line by themselves, the least squares interpolation gives little more
information, no?
Allin Cottrell schrieb:
> The OLS fitted line is displayed in a scatter plot only if the slope
coefficient is significant at the 10 percent level. (If you want an actual
versus fitted graph regardless of the significance of the fit, you can
estimate an OLS model and explicitly request such a plot via the menus).
Good to know, because during lectures I couldn't explain to my students why
the
line was sometimes there, sometimes not. To be honest, I think it's
better to not make it dependent on significance. At the stage of plotting
without estimating a model, I would never have guessed that already some
testing was going on.
> A feature that might be nice to have is the option of adding
various sorts
of fitted line to any X-Y scatter: linear, quadratic,
> exponential... but I don't want to hold up version 1.5.1
waiting for that
right now.
Yes on both accounts.
I agree. What I'd like even better would be different symbols for different
groups, like John suggested earlier. But this is not vital IMO, let's put it
off after release.
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona
4:15 p.m.
New subject: approaching gretl 1.5.1
I've just put up a new Windows snapshot at the usual place:
http://ricardo.ecn.wfu.edu/pub/gretl/gretl_install.exe
Please consider this a "release candidate" for 1.5.1 and let me know
if you see any show-stopping problems! Thanks.
Allin Cottrell
6882
days inactive
6885
days old
4 comments
4 participants
participants (4)
-
Allin Cottrell -
John Paravantis -
Riccardo Jack Lucchetti -
Sven Schreiber