#include <config.h>
+#include <libpspp/assertion.h>
#include <math/covariance.h>
+#include <math/correlation.h>
#include <math/design-matrix.h>
#include <gsl/gsl_matrix.h>
#include <data/casegrouper.h>
#include <language/dictionary/split-file.h>
#include <language/lexer/lexer.h>
#include <language/lexer/variable-parser.h>
-#include <output/manager.h>
-#include <output/table.h>
+#include <output/tab.h>
#include <libpspp/message.h>
#include <data/format.h>
#include <math/moments.h>
#define N_(msgid) msgid
-static double
-significance_of_correlation (double rho, double w)
-{
- double t = w - 2;
- t /= 1 - MIN (1, pow2 (rho));
- t = sqrt (t);
- t *= rho;
-
- if (t > 0)
- return gsl_cdf_tdist_Q (t, w - 2);
- else
- return gsl_cdf_tdist_P (t, w - 2);
-}
-
-
struct corr
{
size_t n_vars_total;
CORR_LISTWISE /* Discard entire case if any variable is missing. */
};
+enum stats_opts
+ {
+ STATS_DESCRIPTIVES = 0x01,
+ STATS_XPROD = 0x02,
+ STATS_ALL = STATS_XPROD | STATS_DESCRIPTIVES
+ };
+
struct corr_opts
{
enum corr_missing_type missing_type;
bool sig; /* Flag significant values or not */
int tails; /* Report significance with how many tails ? */
+ enum stats_opts statistics;
const struct variable *wv; /* The weight variable (if any) */
};
+static void
+output_descriptives (const struct corr *corr, const gsl_matrix *means,
+ const gsl_matrix *vars, const gsl_matrix *ns)
+{
+ const int nr = corr->n_vars_total + 1;
+ const int nc = 4;
+ int c, r;
+
+ const int heading_columns = 1;
+ const int heading_rows = 1;
+
+ struct tab_table *t = tab_create (nc, nr);
+ tab_title (t, _("Descriptive Statistics"));
+
+ tab_headers (t, heading_columns, 0, heading_rows, 0);
+
+ /* Outline the box */
+ tab_box (t,
+ TAL_2, TAL_2,
+ -1, -1,
+ 0, 0,
+ nc - 1, nr - 1);
+
+ /* Vertical lines */
+ tab_box (t,
+ -1, -1,
+ -1, TAL_1,
+ heading_columns, 0,
+ nc - 1, nr - 1);
+
+ tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
+ tab_hline (t, TAL_1, 0, nc - 1, heading_rows);
+
+ tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
+ tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
+ tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
+
+ for (r = 0 ; r < corr->n_vars_total ; ++r)
+ {
+ const struct variable *v = corr->vars[r];
+ tab_text (t, 0, r + heading_rows, TAB_LEFT | TAT_TITLE, var_to_string (v));
+
+ for (c = 1 ; c < nc ; ++c)
+ {
+ double x ;
+ double n;
+ switch (c)
+ {
+ case 1:
+ x = gsl_matrix_get (means, r, 0);
+ break;
+ case 2:
+ x = gsl_matrix_get (vars, r, 0);
+
+ /* Here we want to display the non-biased estimator */
+ n = gsl_matrix_get (ns, r, 0);
+ x *= n / (n -1);
+
+ x = sqrt (x);
+ break;
+ case 3:
+ x = gsl_matrix_get (ns, r, 0);
+ break;
+ default:
+ NOT_REACHED ();
+ };
+
+ tab_double (t, c, r + heading_rows, 0, x, NULL);
+ }
+ }
+
+ tab_submit (t);
+}
+
static void
output_correlation (const struct corr *corr, const struct corr_opts *opts,
- const gsl_matrix *cm, const gsl_matrix *samples)
+ const gsl_matrix *cm, const gsl_matrix *samples,
+ const gsl_matrix *cv)
{
int r, c;
struct tab_table *t;
const int heading_columns = 2;
const int heading_rows = 1;
- const int rows_per_variable = opts->missing_type == CORR_LISTWISE ? 2 : 3;
+ int rows_per_variable = opts->missing_type == CORR_LISTWISE ? 2 : 3;
+
+ if (opts->statistics & STATS_XPROD)
+ rows_per_variable += 2;
/* Two header columns */
nc += heading_columns;
/* One header row */
nr += heading_rows;
- t = tab_create (nc, nr, 0);
+ t = tab_create (nc, nr);
tab_title (t, _("Correlations"));
- tab_dim (t, tab_natural_dimensions, NULL);
tab_headers (t, heading_columns, 0, heading_rows, 0);
tab_text (t, 1, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Pearson Correlation"));
tab_text (t, 1, 2 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
(opts->tails == 2) ? _("Sig. (2-tailed)") : _("Sig. (1-tailed)"));
+
+ if (opts->statistics & STATS_XPROD)
+ {
+ tab_text (t, 1, 3 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Cross-products"));
+ tab_text (t, 1, 4 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Covariance"));
+ }
+
if ( opts->missing_type != CORR_LISTWISE )
- tab_text (t, 1, 3 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("N"));
+ tab_text (t, 1, rows_per_variable + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("N"));
+
tab_hline (t, TAL_1, 0, nc - 1, r * rows_per_variable + 1);
}
for (c = 0 ; c < matrix_cols ; ++c)
{
unsigned char flags = 0;
- int col_index = corr->n_vars_total - corr->n_vars1 + c;
+ const int col_index = corr->n_vars_total - corr->n_vars1 + c;
double pearson = gsl_matrix_get (cm, r, col_index);
double w = gsl_matrix_get (samples, r, col_index);
double sig = opts->tails * significance_of_correlation (pearson, w);
if ( opts->missing_type != CORR_LISTWISE )
- tab_double (t, c + heading_columns, row + 2, 0, w, wfmt);
+ tab_double (t, c + heading_columns, row + rows_per_variable - 1, 0, w, wfmt);
if ( c != r)
tab_double (t, c + heading_columns, row + 1, 0, sig, NULL);
flags = TAB_EMPH;
tab_double (t, c + heading_columns, row, flags, pearson, NULL);
- }
- }
-
- tab_submit (t);
-}
+ if (opts->statistics & STATS_XPROD)
+ {
+ double cov = gsl_matrix_get (cv, r, col_index);
+ const double xprod_dev = cov * w;
+ cov *= w / (w - 1.0);
-static gsl_matrix *
-correlation_from_covariance (const gsl_matrix *cv, const gsl_matrix *v)
-{
- size_t i, j;
- gsl_matrix *corr = gsl_matrix_calloc (cv->size1, cv->size2);
-
- for (i = 0 ; i < cv->size1; ++i)
- {
- for (j = 0 ; j < cv->size2; ++j)
- {
- double rho = gsl_matrix_get (cv, i, j);
-
- rho /= sqrt (gsl_matrix_get (v, i, j))
- *
- sqrt (gsl_matrix_get (v, j, i));
-
- gsl_matrix_set (corr, i, j, rho);
+ tab_double (t, c + heading_columns, row + 2, 0, xprod_dev, NULL);
+ tab_double (t, c + heading_columns, row + 3, 0, cov, NULL);
+ }
}
}
-
- return corr;
-}
-
+ tab_submit (t);
+}
static void
run_corr (struct casereader *r, const struct corr_opts *opts, const struct corr *corr)
{
struct ccase *c;
- const gsl_matrix *var_matrix;
- const gsl_matrix *samples_matrix;
+ const gsl_matrix *var_matrix, *samples_matrix, *mean_matrix;
const gsl_matrix *cov_matrix;
gsl_matrix *corr_matrix;
struct covariance *cov = covariance_create (corr->n_vars_total, corr->vars,
samples_matrix = covariance_moments (cov, MOMENT_NONE);
var_matrix = covariance_moments (cov, MOMENT_VARIANCE);
+ mean_matrix = covariance_moments (cov, MOMENT_MEAN);
corr_matrix = correlation_from_covariance (cov_matrix, var_matrix);
+ if ( opts->statistics & STATS_DESCRIPTIVES)
+ output_descriptives (corr, mean_matrix, var_matrix, samples_matrix);
+
output_correlation (corr, opts,
corr_matrix,
- samples_matrix );
+ samples_matrix,
+ cov_matrix);
covariance_destroy (cov);
gsl_matrix_free (corr_matrix);
opts.tails = 2;
opts.sig = false;
opts.exclude = MV_ANY;
+ opts.statistics = 0;
/* Parse CORRELATIONS. */
while (lex_token (lexer) != '.')
goto error;
}
+ lex_match (lexer, ',');
+ }
+ }
+ else if (lex_match_id (lexer, "STATISTICS"))
+ {
+ lex_match (lexer, '=');
+ while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
+ {
+ if ( lex_match_id (lexer, "DESCRIPTIVES"))
+ opts.statistics = STATS_DESCRIPTIVES;
+ else if (lex_match_id (lexer, "XPROD"))
+ opts.statistics = STATS_XPROD;
+ else if (lex_token (lexer) == T_ALL)
+ {
+ opts.statistics = STATS_ALL;
+ lex_get (lexer);
+ }
+ else
+ {
+ lex_error (lexer, NULL);
+ goto error;
+ }
+
lex_match (lexer, ',');
}
}