std error of estimate?
When I do a linear regression analysis, the model summary always shows what appear to be incorrect values for (overall) Std. Error of the Estimate, regardless of the data set. It does show individual std. errors for the constant and coefficient on X in "Coefficients" section of output and these are correct, checked against R and SPSS. Am I missing something? (probably... ;-) This is pspp 0.6.1 for 32bit i586, run on SuSE 10.3, installed from Dr. Steuer's repository. example: REGRESSION /VARIABLES= x /DEPENDENT= y /STATISTICS=COEFF R ANOVA. (choose any x or y variable from any data I've tried) Is this part of the regression procedure not complete yet? -- Erik Frebold Vancouver, BC ___ Pspp-users mailing list Pspp-users@gnu.org http://lists.gnu.org/mailman/listinfo/pspp-users
Re: std error of estimate?
On Mon, Feb 23, 2009 at 02:30:20AM -0800, Erik Frebold wrote: > When I do a linear regression analysis, the model summary always shows what > appear to be incorrect values for (overall) Std. Error of the Estimate, > regardless of the data set. It does show individual std. errors for the > constant and coefficient on X in "Coefficients" section of output and these > are correct, checked against R and SPSS. Am I missing something? > (probably... ;-) Oops. I coded this to place the standard error of R-squared in the box, rather than root MSE. I'll file a bug and fix it soon. -Jason ___ Pspp-users mailing list Pspp-users@gnu.org http://lists.gnu.org/mailman/listinfo/pspp-users
Re: std error of estimate?
I just committed a fix to the source repository. To fix your own local copy,
try this patch. Apply by cd'ing to the directory with the source
code, and run
patch -p1 < regression.patch
Then rebuild pspp.
--- pspp.0/src/math/linreg.c2009-01-19 14:38:06.0 -0500
+++ pspp/src/math/linreg.c 2009-02-23 11:09:16.0 -0500
@@ -737,3 +737,9 @@
design_matrix_destroy (cov);
}
+double pspp_linreg_mse (const pspp_linreg_cache *c)
+{
+ assert (c != NULL);
+ return (c->sse / c->dfe);
+}
+
--- pspp.0/src/math/linreg.h2009-01-19 14:38:06.0 -0500
+++ pspp/src/math/linreg.h 2009-02-23 11:09:20.0 -0500
@@ -222,4 +222,5 @@
Regression using only the covariance matrix.
*/
void pspp_linreg_with_cov (const struct covariance_matrix *, pspp_linreg_cache
*);
+double pspp_linreg_mse (const pspp_linreg_cache *);
#endif
--- pspp.0/tests/command/regression.sh 2009-01-19 14:38:06.0 -0500
+++ pspp/tests/command/regression.sh2009-02-23 11:16:21.0 -0500
@@ -95,7 +95,7 @@
###=#==#
# R #R Square|Adjusted R Square|Std. Error of the Estimate#
##===##=#==#
-#|.97# .94| .93| .33#
+#|.97# .94| .93| 1.34#
##===##=#==#
2.2 REGRESSION. ANOVA
#===#==#==#===#==##
--- pspp.0/tests/command/regression-qr.sh 2009-01-19 14:38:06.0
-0500
+++ pspp/tests/command/regression-qr.sh 2009-02-23 11:11:15.0 -0500
@@ -1582,7 +1582,7 @@
###=#==#
# R #R Square|Adjusted R Square|Std. Error of the Estimate#
##===##=#==#
-#|.05# .00| .00| .00#
+#|.05# .00| .00| 8.11#
##===##=#==#
2.2 REGRESSION. ANOVA
#===#==##===#=##
--- pspp.0/src/language/stats/regression.q 2009-01-19 14:38:06.0
-0500
+++ pspp/src/language/stats/regression.q2009-02-23 11:10:07.0
-0500
@@ -148,7 +148,7 @@
assert (c != NULL);
rsq = c->ssm / c->sst;
adjrsq = 1.0 - (1.0 - rsq) * (c->n_obs - 1.0) / (c->n_obs - c->n_indeps);
- std_error = sqrt ((c->n_indeps - 1.0) / (c->n_obs - 1.0));
+ std_error = sqrt (pspp_linreg_mse (c));
t = tab_create (n_cols, n_rows, 0);
tab_dim (t, tab_natural_dimensions);
tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
@@ -281,7 +281,7 @@
int n_cols = 7;
int n_rows = 4;
const double msm = c->ssm / c->dfm;
- const double mse = c->sse / c->dfe;
+ const double mse = pspp_linreg_mse (c);
const double F = msm / mse;
const double pval = gsl_cdf_fdist_Q (F, c->dfm, c->dfe);
On Mon, Feb 23, 2009 at 02:30:20AM -0800, Erik Frebold wrote:
> When I do a linear regression analysis, the model summary always shows what
> appear to be incorrect values for (overall) Std. Error of the Estimate,
> regardless of the data set. It does show individual std. errors for the
> constant and coefficient on X in "Coefficients" section of output and these
> are correct, checked against R and SPSS. Am I missing something?
> (probably... ;-)
>
> This is pspp 0.6.1 for 32bit i586, run on SuSE 10.3, installed from Dr.
> Steuer's repository.
>
> example:
> REGRESSION
> /VARIABLES= x
> /DEPENDENT= y
> /STATISTICS=COEFF R ANOVA.
>
> (choose any x or y variable from any data I've tried)
>
> Is this part of the regression procedure not complete yet?
> --
> Erik Frebold
> Vancouver, BC
>
>
>
> ___
> Pspp-users mailing list
> Pspp-users@gnu.org
> http://lists.gnu.org/mailman/listinfo/pspp-users
___
Pspp-users mailing list
Pspp-users@gnu.org
http://lists.gnu.org/mailman/listinfo/pspp-users
Re: std error of estimate?
Wow, that was quick! :-) Forgive my ignorance as I'm new to this-- re: your instructions as follows: "To fix your own local copy, try this patch. Apply by cd'ing to the directory with the source code, and run patch -p1 < regression.patch Then rebuild pspp." Wasn't sure where the patch file goes physically (and I did read man patch...) so I created a .txt file called regression.patch containing the patch text and put it in the source directory. Here's what happened: --- patch -p1 < regression.patch patching file src/math/linreg.c patch: malformed patch at line 4: design_matrix_destroy (cov); --- Any suggestions? -- Erik Frebold Vancouver, BC ___ Pspp-users mailing list Pspp-users@gnu.org http://lists.gnu.org/mailman/listinfo/pspp-users
Re: std error of estimate?
On Mon, Feb 23, 2009 at 01:54:33PM -0800, Erik Frebold wrote: > patch -p1 < regression.patch > Then rebuild pspp." > > Wasn't sure where the patch file goes physically (and I did read man > patch...) so I created a .txt file called regression.patch containing the > patch text and put it in the source directory. Here's what happened: > > --- > patch -p1 < regression.patch > > patching file src/math/linreg.c > patch: malformed patch at line 4: design_matrix_destroy (cov); > --- > > Any suggestions? The line breaks in the email confused the patch program. Here is another copy, without the line breaks: http://math.gcsu.edu/~jhs/regression.patch But there may be a more important problem here: You mentioned getting PSPP from someone else. Did you compile it from source? If not, then the patch won't help unless you either get the source code for PSPP and build it on your own machine, or get a new, precompiled copy. ___ Pspp-users mailing list Pspp-users@gnu.org http://lists.gnu.org/mailman/listinfo/pspp-users
