1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2009 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 #include <libpspp/assertion.h>
20 #include <math/covariance.h>
21 #include <gsl/gsl_matrix.h>
22 #include <data/casegrouper.h>
23 #include <data/casereader.h>
24 #include <data/dictionary.h>
25 #include <data/procedure.h>
26 #include <data/variable.h>
27 #include <language/command.h>
28 #include <language/dictionary/split-file.h>
29 #include <language/lexer/lexer.h>
30 #include <language/lexer/variable-parser.h>
31 #include <output/manager.h>
32 #include <output/table.h>
33 #include <libpspp/message.h>
34 #include <data/format.h>
35 #include <math/moments.h>
40 #include <libpspp/misc.h>
41 #include <gsl/gsl_cdf.h>
44 #define _(msgid) gettext (msgid)
45 #define N_(msgid) msgid
49 significance_of_correlation (double rho, double w)
52 t /= 1 - MIN (1, pow2 (rho));
57 return gsl_cdf_tdist_Q (t, w - 2);
59 return gsl_cdf_tdist_P (t, w - 2);
68 const struct variable **vars;
72 /* Handling of missing values. */
73 enum corr_missing_type
75 CORR_PAIRWISE, /* Handle missing values on a per-variable-pair basis. */
76 CORR_LISTWISE /* Discard entire case if any variable is missing. */
81 STATS_DESCRIPTIVES = 0x01,
83 STATS_ALL = STATS_XPROD | STATS_DESCRIPTIVES
88 enum corr_missing_type missing_type;
89 enum mv_class exclude; /* Classes of missing values to exclude. */
91 bool sig; /* Flag significant values or not */
92 int tails; /* Report significance with how many tails ? */
93 enum stats_opts statistics;
95 const struct variable *wv; /* The weight variable (if any) */
100 output_descriptives (const struct corr *corr, const gsl_matrix *means,
101 const gsl_matrix *vars, const gsl_matrix *ns)
103 const int nr = corr->n_vars_total + 1;
107 const int heading_columns = 1;
108 const int heading_rows = 1;
110 struct tab_table *t = tab_create (nc, nr, 0);
111 tab_title (t, _("Descriptive Statistics"));
112 tab_dim (t, tab_natural_dimensions, NULL);
114 tab_headers (t, heading_columns, 0, heading_rows, 0);
116 /* Outline the box */
130 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
131 tab_hline (t, TAL_1, 0, nc - 1, heading_rows);
133 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
134 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
135 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
137 for (r = 0 ; r < corr->n_vars_total ; ++r)
139 const struct variable *v = corr->vars[r];
140 tab_text (t, 0, r + heading_rows, TAB_LEFT | TAT_TITLE, var_to_string (v));
142 for (c = 1 ; c < nc ; ++c)
149 x = gsl_matrix_get (means, r, 0);
152 x = gsl_matrix_get (vars, r, 0);
154 /* Here we want to display the non-biased estimator */
155 n = gsl_matrix_get (ns, r, 0);
161 x = gsl_matrix_get (ns, r, 0);
167 tab_double (t, c, r + heading_rows, 0, x, NULL);
175 output_correlation (const struct corr *corr, const struct corr_opts *opts,
176 const gsl_matrix *cm, const gsl_matrix *samples,
177 const gsl_matrix *cv)
182 int nr = corr->n_vars1;
183 int nc = matrix_cols = corr->n_vars_total > corr->n_vars1 ?
184 corr->n_vars_total - corr->n_vars1 : corr->n_vars1;
186 const struct fmt_spec *wfmt = opts->wv ? var_get_print_format (opts->wv) : & F_8_0;
188 const int heading_columns = 2;
189 const int heading_rows = 1;
191 int rows_per_variable = opts->missing_type == CORR_LISTWISE ? 2 : 3;
193 if (opts->statistics & STATS_XPROD)
194 rows_per_variable += 2;
196 /* Two header columns */
197 nc += heading_columns;
199 /* Three data per variable */
200 nr *= rows_per_variable;
205 t = tab_create (nc, nr, 0);
206 tab_title (t, _("Correlations"));
207 tab_dim (t, tab_natural_dimensions, NULL);
209 tab_headers (t, heading_columns, 0, heading_rows, 0);
211 /* Outline the box */
225 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
226 tab_vline (t, TAL_1, 1, heading_rows, nr - 1);
228 for (r = 0 ; r < corr->n_vars1 ; ++r)
230 tab_text (t, 0, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
231 var_to_string (corr->vars[r]));
233 tab_text (t, 1, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Pearson Correlation"));
234 tab_text (t, 1, 2 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
235 (opts->tails == 2) ? _("Sig. (2-tailed)") : _("Sig. (1-tailed)"));
237 if (opts->statistics & STATS_XPROD)
239 tab_text (t, 1, 3 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Cross-products"));
240 tab_text (t, 1, 4 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Covariance"));
243 if ( opts->missing_type != CORR_LISTWISE )
244 tab_text (t, 1, rows_per_variable + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("N"));
246 tab_hline (t, TAL_1, 0, nc - 1, r * rows_per_variable + 1);
249 for (c = 0 ; c < matrix_cols ; ++c)
251 const struct variable *v = corr->n_vars_total > corr->n_vars1 ? corr->vars[corr->n_vars_total - corr->n_vars1 + c] : corr->vars[c];
252 tab_text (t, heading_columns + c, 0, TAB_LEFT | TAT_TITLE, var_to_string (v));
255 for (r = 0 ; r < corr->n_vars1 ; ++r)
257 const int row = r * rows_per_variable + heading_rows;
258 for (c = 0 ; c < matrix_cols ; ++c)
260 unsigned char flags = 0;
261 const int col_index = corr->n_vars_total - corr->n_vars1 + c;
262 double pearson = gsl_matrix_get (cm, r, col_index);
263 double w = gsl_matrix_get (samples, r, col_index);
264 double sig = opts->tails * significance_of_correlation (pearson, w);
266 if ( opts->missing_type != CORR_LISTWISE )
267 tab_double (t, c + heading_columns, row + rows_per_variable - 1, 0, w, wfmt);
270 tab_double (t, c + heading_columns, row + 1, 0, sig, NULL);
272 if ( opts->sig && c != r && sig < 0.05)
275 tab_double (t, c + heading_columns, row, flags, pearson, NULL);
277 if (opts->statistics & STATS_XPROD)
279 double cov = gsl_matrix_get (cv, r, col_index);
280 const double xprod_dev = cov * w;
281 cov *= w / (w - 1.0);
283 tab_double (t, c + heading_columns, row + 2, 0, xprod_dev, NULL);
284 tab_double (t, c + heading_columns, row + 3, 0, cov, NULL);
294 correlation_from_covariance (const gsl_matrix *cv, const gsl_matrix *v)
297 gsl_matrix *corr = gsl_matrix_calloc (cv->size1, cv->size2);
299 for (i = 0 ; i < cv->size1; ++i)
301 for (j = 0 ; j < cv->size2; ++j)
303 double rho = gsl_matrix_get (cv, i, j);
305 rho /= sqrt (gsl_matrix_get (v, i, j))
307 sqrt (gsl_matrix_get (v, j, i));
309 gsl_matrix_set (corr, i, j, rho);
320 run_corr (struct casereader *r, const struct corr_opts *opts, const struct corr *corr)
323 const gsl_matrix *var_matrix, *samples_matrix, *mean_matrix;
324 const gsl_matrix *cov_matrix;
325 gsl_matrix *corr_matrix;
326 struct covariance *cov = covariance_2pass_create (corr->n_vars_total, corr->vars,
328 opts->wv, opts->exclude);
330 struct casereader *rc = casereader_clone (r);
331 for ( ; (c = casereader_read (r) ); case_unref (c))
333 covariance_accumulate_pass1 (cov, c);
336 for ( ; (c = casereader_read (rc) ); case_unref (c))
338 covariance_accumulate_pass2 (cov, c);
341 cov_matrix = covariance_calculate (cov);
343 casereader_destroy (rc);
345 samples_matrix = covariance_moments (cov, MOMENT_NONE);
346 var_matrix = covariance_moments (cov, MOMENT_VARIANCE);
347 mean_matrix = covariance_moments (cov, MOMENT_MEAN);
349 corr_matrix = correlation_from_covariance (cov_matrix, var_matrix);
351 if ( opts->statistics & STATS_DESCRIPTIVES)
352 output_descriptives (corr, mean_matrix, var_matrix, samples_matrix);
354 output_correlation (corr, opts,
359 covariance_destroy (cov);
360 gsl_matrix_free (corr_matrix);
364 cmd_correlation (struct lexer *lexer, struct dataset *ds)
367 int n_all_vars = 0; /* Total number of variables involved in this command */
368 const struct variable **all_vars ;
369 const struct dictionary *dict = dataset_dict (ds);
372 struct casegrouper *grouper;
373 struct casereader *group;
375 struct corr *corr = NULL;
378 struct corr_opts opts;
379 opts.missing_type = CORR_PAIRWISE;
380 opts.wv = dict_get_weight (dict);
383 opts.exclude = MV_ANY;
386 /* Parse CORRELATIONS. */
387 while (lex_token (lexer) != '.')
389 lex_match (lexer, '/');
390 if (lex_match_id (lexer, "MISSING"))
392 lex_match (lexer, '=');
393 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
395 if (lex_match_id (lexer, "PAIRWISE"))
396 opts.missing_type = CORR_PAIRWISE;
397 else if (lex_match_id (lexer, "LISTWISE"))
398 opts.missing_type = CORR_LISTWISE;
400 else if (lex_match_id (lexer, "INCLUDE"))
401 opts.exclude = MV_SYSTEM;
402 else if (lex_match_id (lexer, "EXCLUDE"))
403 opts.exclude = MV_ANY;
406 lex_error (lexer, NULL);
409 lex_match (lexer, ',');
412 else if (lex_match_id (lexer, "PRINT"))
414 lex_match (lexer, '=');
415 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
417 if ( lex_match_id (lexer, "TWOTAIL"))
419 else if (lex_match_id (lexer, "ONETAIL"))
421 else if (lex_match_id (lexer, "SIG"))
423 else if (lex_match_id (lexer, "NOSIG"))
427 lex_error (lexer, NULL);
431 lex_match (lexer, ',');
434 else if (lex_match_id (lexer, "STATISTICS"))
436 lex_match (lexer, '=');
437 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
439 if ( lex_match_id (lexer, "DESCRIPTIVES"))
440 opts.statistics = STATS_DESCRIPTIVES;
441 else if (lex_match_id (lexer, "XPROD"))
442 opts.statistics = STATS_XPROD;
443 else if (lex_token (lexer) == T_ALL)
445 opts.statistics = STATS_ALL;
450 lex_error (lexer, NULL);
454 lex_match (lexer, ',');
459 if (lex_match_id (lexer, "VARIABLES"))
461 lex_match (lexer, '=');
464 corr = xrealloc (corr, sizeof (*corr) * (n_corrs + 1));
465 corr[n_corrs].n_vars_total = corr[n_corrs].n_vars1 = 0;
467 if ( ! parse_variables_const (lexer, dict, &corr[n_corrs].vars,
468 &corr[n_corrs].n_vars_total,
476 corr[n_corrs].n_vars1 = corr[n_corrs].n_vars_total;
478 if ( lex_match (lexer, T_WITH))
480 if ( ! parse_variables_const (lexer, dict,
481 &corr[n_corrs].vars, &corr[n_corrs].n_vars_total,
482 PV_NUMERIC | PV_APPEND))
489 n_all_vars += corr[n_corrs].n_vars_total;
497 msg (SE, _("No variables specified."));
502 all_vars = xmalloc (sizeof (*all_vars) * n_all_vars);
505 /* FIXME: Using a hash here would make more sense */
506 const struct variable **vv = all_vars;
508 for (i = 0 ; i < n_corrs; ++i)
511 const struct corr *c = &corr[i];
512 for (v = 0 ; v < c->n_vars_total; ++v)
517 grouper = casegrouper_create_splits (proc_open (ds), dict);
519 while (casegrouper_get_next_group (grouper, &group))
521 for (i = 0 ; i < n_corrs; ++i)
523 /* FIXME: No need to iterate the data multiple times */
524 struct casereader *r = casereader_clone (group);
526 if ( opts.missing_type == CORR_LISTWISE)
527 r = casereader_create_filter_missing (r, all_vars, n_all_vars,
528 opts.exclude, NULL, NULL);
531 run_corr (r, &opts, &corr[i]);
532 casereader_destroy (r);
534 casereader_destroy (group);
537 ok = casegrouper_destroy (grouper);
538 ok = proc_commit (ds) && ok;
545 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;