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 <math/correlation.h>
22 #include <gsl/gsl_matrix.h>
23 #include <data/casegrouper.h>
24 #include <data/casereader.h>
25 #include <data/dictionary.h>
26 #include <data/procedure.h>
27 #include <data/variable.h>
28 #include <language/command.h>
29 #include <language/dictionary/split-file.h>
30 #include <language/lexer/lexer.h>
31 #include <language/lexer/variable-parser.h>
32 #include <output/tab.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
53 const struct variable **vars;
57 /* Handling of missing values. */
58 enum corr_missing_type
60 CORR_PAIRWISE, /* Handle missing values on a per-variable-pair basis. */
61 CORR_LISTWISE /* Discard entire case if any variable is missing. */
66 STATS_DESCRIPTIVES = 0x01,
68 STATS_ALL = STATS_XPROD | STATS_DESCRIPTIVES
73 enum corr_missing_type missing_type;
74 enum mv_class exclude; /* Classes of missing values to exclude. */
76 bool sig; /* Flag significant values or not */
77 int tails; /* Report significance with how many tails ? */
78 enum stats_opts statistics;
80 const struct variable *wv; /* The weight variable (if any) */
85 output_descriptives (const struct corr *corr, const gsl_matrix *means,
86 const gsl_matrix *vars, const gsl_matrix *ns)
88 const int nr = corr->n_vars_total + 1;
92 const int heading_columns = 1;
93 const int heading_rows = 1;
95 struct tab_table *t = tab_create (nc, nr);
96 tab_title (t, _("Descriptive Statistics"));
98 tab_headers (t, heading_columns, 0, heading_rows, 0);
100 /* Outline the box */
114 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
115 tab_hline (t, TAL_1, 0, nc - 1, heading_rows);
117 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
118 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
119 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
121 for (r = 0 ; r < corr->n_vars_total ; ++r)
123 const struct variable *v = corr->vars[r];
124 tab_text (t, 0, r + heading_rows, TAB_LEFT | TAT_TITLE, var_to_string (v));
126 for (c = 1 ; c < nc ; ++c)
133 x = gsl_matrix_get (means, r, 0);
136 x = gsl_matrix_get (vars, r, 0);
138 /* Here we want to display the non-biased estimator */
139 n = gsl_matrix_get (ns, r, 0);
145 x = gsl_matrix_get (ns, r, 0);
151 tab_double (t, c, r + heading_rows, 0, x, NULL);
159 output_correlation (const struct corr *corr, const struct corr_opts *opts,
160 const gsl_matrix *cm, const gsl_matrix *samples,
161 const gsl_matrix *cv)
166 int nr = corr->n_vars1;
167 int nc = matrix_cols = corr->n_vars_total > corr->n_vars1 ?
168 corr->n_vars_total - corr->n_vars1 : corr->n_vars1;
170 const struct fmt_spec *wfmt = opts->wv ? var_get_print_format (opts->wv) : & F_8_0;
172 const int heading_columns = 2;
173 const int heading_rows = 1;
175 int rows_per_variable = opts->missing_type == CORR_LISTWISE ? 2 : 3;
177 if (opts->statistics & STATS_XPROD)
178 rows_per_variable += 2;
180 /* Two header columns */
181 nc += heading_columns;
183 /* Three data per variable */
184 nr *= rows_per_variable;
189 t = tab_create (nc, nr);
190 tab_title (t, _("Correlations"));
192 tab_headers (t, heading_columns, 0, heading_rows, 0);
194 /* Outline the box */
208 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
209 tab_vline (t, TAL_1, 1, heading_rows, nr - 1);
211 for (r = 0 ; r < corr->n_vars1 ; ++r)
213 tab_text (t, 0, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
214 var_to_string (corr->vars[r]));
216 tab_text (t, 1, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Pearson Correlation"));
217 tab_text (t, 1, 2 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
218 (opts->tails == 2) ? _("Sig. (2-tailed)") : _("Sig. (1-tailed)"));
220 if (opts->statistics & STATS_XPROD)
222 tab_text (t, 1, 3 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Cross-products"));
223 tab_text (t, 1, 4 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Covariance"));
226 if ( opts->missing_type != CORR_LISTWISE )
227 tab_text (t, 1, rows_per_variable + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("N"));
229 tab_hline (t, TAL_1, 0, nc - 1, r * rows_per_variable + 1);
232 for (c = 0 ; c < matrix_cols ; ++c)
234 const struct variable *v = corr->n_vars_total > corr->n_vars1 ? corr->vars[corr->n_vars_total - corr->n_vars1 + c] : corr->vars[c];
235 tab_text (t, heading_columns + c, 0, TAB_LEFT | TAT_TITLE, var_to_string (v));
238 for (r = 0 ; r < corr->n_vars1 ; ++r)
240 const int row = r * rows_per_variable + heading_rows;
241 for (c = 0 ; c < matrix_cols ; ++c)
243 unsigned char flags = 0;
244 const int col_index = corr->n_vars_total - corr->n_vars1 + c;
245 double pearson = gsl_matrix_get (cm, r, col_index);
246 double w = gsl_matrix_get (samples, r, col_index);
247 double sig = opts->tails * significance_of_correlation (pearson, w);
249 if ( opts->missing_type != CORR_LISTWISE )
250 tab_double (t, c + heading_columns, row + rows_per_variable - 1, 0, w, wfmt);
253 tab_double (t, c + heading_columns, row + 1, 0, sig, NULL);
255 if ( opts->sig && c != r && sig < 0.05)
258 tab_double (t, c + heading_columns, row, flags, pearson, NULL);
260 if (opts->statistics & STATS_XPROD)
262 double cov = gsl_matrix_get (cv, r, col_index);
263 const double xprod_dev = cov * w;
264 cov *= w / (w - 1.0);
266 tab_double (t, c + heading_columns, row + 2, 0, xprod_dev, NULL);
267 tab_double (t, c + heading_columns, row + 3, 0, cov, NULL);
277 run_corr (struct casereader *r, const struct corr_opts *opts, const struct corr *corr)
280 const gsl_matrix *var_matrix, *samples_matrix, *mean_matrix;
281 const gsl_matrix *cov_matrix;
282 gsl_matrix *corr_matrix;
283 struct covariance *cov = covariance_2pass_create (corr->n_vars_total, corr->vars,
285 opts->wv, opts->exclude);
287 struct casereader *rc = casereader_clone (r);
288 for ( ; (c = casereader_read (r) ); case_unref (c))
290 covariance_accumulate_pass1 (cov, c);
293 for ( ; (c = casereader_read (rc) ); case_unref (c))
295 covariance_accumulate_pass2 (cov, c);
298 cov_matrix = covariance_calculate (cov);
300 casereader_destroy (rc);
302 samples_matrix = covariance_moments (cov, MOMENT_NONE);
303 var_matrix = covariance_moments (cov, MOMENT_VARIANCE);
304 mean_matrix = covariance_moments (cov, MOMENT_MEAN);
306 corr_matrix = correlation_from_covariance (cov_matrix, var_matrix);
308 if ( opts->statistics & STATS_DESCRIPTIVES)
309 output_descriptives (corr, mean_matrix, var_matrix, samples_matrix);
311 output_correlation (corr, opts,
316 covariance_destroy (cov);
317 gsl_matrix_free (corr_matrix);
321 cmd_correlation (struct lexer *lexer, struct dataset *ds)
324 int n_all_vars = 0; /* Total number of variables involved in this command */
325 const struct variable **all_vars ;
326 const struct dictionary *dict = dataset_dict (ds);
329 struct casegrouper *grouper;
330 struct casereader *group;
332 struct corr *corr = NULL;
335 struct corr_opts opts;
336 opts.missing_type = CORR_PAIRWISE;
337 opts.wv = dict_get_weight (dict);
340 opts.exclude = MV_ANY;
343 /* Parse CORRELATIONS. */
344 while (lex_token (lexer) != '.')
346 lex_match (lexer, '/');
347 if (lex_match_id (lexer, "MISSING"))
349 lex_match (lexer, '=');
350 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
352 if (lex_match_id (lexer, "PAIRWISE"))
353 opts.missing_type = CORR_PAIRWISE;
354 else if (lex_match_id (lexer, "LISTWISE"))
355 opts.missing_type = CORR_LISTWISE;
357 else if (lex_match_id (lexer, "INCLUDE"))
358 opts.exclude = MV_SYSTEM;
359 else if (lex_match_id (lexer, "EXCLUDE"))
360 opts.exclude = MV_ANY;
363 lex_error (lexer, NULL);
366 lex_match (lexer, ',');
369 else if (lex_match_id (lexer, "PRINT"))
371 lex_match (lexer, '=');
372 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
374 if ( lex_match_id (lexer, "TWOTAIL"))
376 else if (lex_match_id (lexer, "ONETAIL"))
378 else if (lex_match_id (lexer, "SIG"))
380 else if (lex_match_id (lexer, "NOSIG"))
384 lex_error (lexer, NULL);
388 lex_match (lexer, ',');
391 else if (lex_match_id (lexer, "STATISTICS"))
393 lex_match (lexer, '=');
394 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
396 if ( lex_match_id (lexer, "DESCRIPTIVES"))
397 opts.statistics = STATS_DESCRIPTIVES;
398 else if (lex_match_id (lexer, "XPROD"))
399 opts.statistics = STATS_XPROD;
400 else if (lex_token (lexer) == T_ALL)
402 opts.statistics = STATS_ALL;
407 lex_error (lexer, NULL);
411 lex_match (lexer, ',');
416 if (lex_match_id (lexer, "VARIABLES"))
418 lex_match (lexer, '=');
421 corr = xrealloc (corr, sizeof (*corr) * (n_corrs + 1));
422 corr[n_corrs].n_vars_total = corr[n_corrs].n_vars1 = 0;
424 if ( ! parse_variables_const (lexer, dict, &corr[n_corrs].vars,
425 &corr[n_corrs].n_vars_total,
433 corr[n_corrs].n_vars1 = corr[n_corrs].n_vars_total;
435 if ( lex_match (lexer, T_WITH))
437 if ( ! parse_variables_const (lexer, dict,
438 &corr[n_corrs].vars, &corr[n_corrs].n_vars_total,
439 PV_NUMERIC | PV_APPEND))
446 n_all_vars += corr[n_corrs].n_vars_total;
454 msg (SE, _("No variables specified."));
459 all_vars = xmalloc (sizeof (*all_vars) * n_all_vars);
462 /* FIXME: Using a hash here would make more sense */
463 const struct variable **vv = all_vars;
465 for (i = 0 ; i < n_corrs; ++i)
468 const struct corr *c = &corr[i];
469 for (v = 0 ; v < c->n_vars_total; ++v)
474 grouper = casegrouper_create_splits (proc_open (ds), dict);
476 while (casegrouper_get_next_group (grouper, &group))
478 for (i = 0 ; i < n_corrs; ++i)
480 /* FIXME: No need to iterate the data multiple times */
481 struct casereader *r = casereader_clone (group);
483 if ( opts.missing_type == CORR_LISTWISE)
484 r = casereader_create_filter_missing (r, all_vars, n_all_vars,
485 opts.exclude, NULL, NULL);
488 run_corr (r, &opts, &corr[i]);
489 casereader_destroy (r);
491 casereader_destroy (group);
494 ok = casegrouper_destroy (grouper);
495 ok = proc_commit (ds) && ok;
502 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;