On Sun, 20 Jun 2010, artur bala wrote:
On Wed, Jun 16, 2010 at 6:22 PM, Allin Cottrell
<cottrell(a)wfu.edu> wrote:
> * Main window right-click menu with multiple series selected,
> correlation matrix option: this now presents an initial dialog
> with the option of imposing a uniform sample for all the
> pairwise correlations, in case of missing data.
Thank you Allin for your efforts. It looks nice and the changes are
really welcomed.
Thanks, glad you like it.
Now the correlation matrix has a “uniforme sample” option which
makes sense of gretl displaying the number of observations used.
But, in the case of a pairwise correlation matrix the number of
observations may very from pair to pair. In such a case, it may
be misleading to still display the number of observations.
What happens at present is this:
* The correlation matrix header always says, e.g. "using the
observations 1 - 100", as a reminder of the current sample.
* If any missing values were skipped, you also see the text
"missing values were skipped".
* You also get a note: "5% critical value (two-tailed) = <value>
for n = <nmin>", where nmin is the minimum number of observations
used in calculating any of the pairwise coefficients.
If there are no missing values, or the missing values line up
across the series, or the uniform option has been selected, then
nmin is in fact the true n for all the coefficients shown.
If there are non-aligned missing values and the uniform option has
not been selected, then some of the correlations will in fact have
been computed on a larger sample than the "n" for which the
critical value is printed, in which case the critical value is
conservative for those coefficients. I don't think that's a
problem, but perhaps it should be better documented.
Allin