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 <math/design-matrix.h>
23 #include <gsl/gsl_matrix.h>
24 #include <data/casegrouper.h>
25 #include <data/casereader.h>
26 #include <data/dictionary.h>
27 #include <data/procedure.h>
28 #include <data/variable.h>
29 #include <language/command.h>
30 #include <language/dictionary/split-file.h>
31 #include <language/lexer/lexer.h>
32 #include <language/lexer/variable-parser.h>
33 #include <output/tab.h>
34 #include <libpspp/message.h>
35 #include <data/format.h>
36 #include <math/moments.h>
41 #include <libpspp/misc.h>
42 #include <gsl/gsl_cdf.h>
45 #define _(msgid) gettext (msgid)
46 #define N_(msgid) msgid
54 const struct variable **vars;
58 /* Handling of missing values. */
59 enum corr_missing_type
61 CORR_PAIRWISE, /* Handle missing values on a per-variable-pair basis. */
62 CORR_LISTWISE /* Discard entire case if any variable is missing. */
67 STATS_DESCRIPTIVES = 0x01,
69 STATS_ALL = STATS_XPROD | STATS_DESCRIPTIVES
74 enum corr_missing_type missing_type;
75 enum mv_class exclude; /* Classes of missing values to exclude. */
77 bool sig; /* Flag significant values or not */
78 int tails; /* Report significance with how many tails ? */
79 enum stats_opts statistics;
81 const struct variable *wv; /* The weight variable (if any) */
86 output_descriptives (const struct corr *corr, const gsl_matrix *means,
87 const gsl_matrix *vars, const gsl_matrix *ns)
89 const int nr = corr->n_vars_total + 1;
93 const int heading_columns = 1;
94 const int heading_rows = 1;
96 struct tab_table *t = tab_create (nc, nr);
97 tab_title (t, _("Descriptive Statistics"));
99 tab_headers (t, heading_columns, 0, heading_rows, 0);
101 /* Outline the box */
115 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
116 tab_hline (t, TAL_1, 0, nc - 1, heading_rows);
118 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
119 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
120 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
122 for (r = 0 ; r < corr->n_vars_total ; ++r)
124 const struct variable *v = corr->vars[r];
125 tab_text (t, 0, r + heading_rows, TAB_LEFT | TAT_TITLE, var_to_string (v));
127 for (c = 1 ; c < nc ; ++c)
134 x = gsl_matrix_get (means, r, 0);
137 x = gsl_matrix_get (vars, r, 0);
139 /* Here we want to display the non-biased estimator */
140 n = gsl_matrix_get (ns, r, 0);
146 x = gsl_matrix_get (ns, r, 0);
152 tab_double (t, c, r + heading_rows, 0, x, NULL);
160 output_correlation (const struct corr *corr, const struct corr_opts *opts,
161 const gsl_matrix *cm, const gsl_matrix *samples,
162 const gsl_matrix *cv)
167 int nr = corr->n_vars1;
168 int nc = matrix_cols = corr->n_vars_total > corr->n_vars1 ?
169 corr->n_vars_total - corr->n_vars1 : corr->n_vars1;
171 const struct fmt_spec *wfmt = opts->wv ? var_get_print_format (opts->wv) : & F_8_0;
173 const int heading_columns = 2;
174 const int heading_rows = 1;
176 int rows_per_variable = opts->missing_type == CORR_LISTWISE ? 2 : 3;
178 if (opts->statistics & STATS_XPROD)
179 rows_per_variable += 2;
181 /* Two header columns */
182 nc += heading_columns;
184 /* Three data per variable */
185 nr *= rows_per_variable;
190 t = tab_create (nc, nr);
191 tab_title (t, _("Correlations"));
193 tab_headers (t, heading_columns, 0, heading_rows, 0);
195 /* Outline the box */
209 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
210 tab_vline (t, TAL_1, 1, heading_rows, nr - 1);
212 for (r = 0 ; r < corr->n_vars1 ; ++r)
214 tab_text (t, 0, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
215 var_to_string (corr->vars[r]));
217 tab_text (t, 1, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Pearson Correlation"));
218 tab_text (t, 1, 2 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
219 (opts->tails == 2) ? _("Sig. (2-tailed)") : _("Sig. (1-tailed)"));
221 if (opts->statistics & STATS_XPROD)
223 tab_text (t, 1, 3 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Cross-products"));
224 tab_text (t, 1, 4 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Covariance"));
227 if ( opts->missing_type != CORR_LISTWISE )
228 tab_text (t, 1, rows_per_variable + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("N"));
230 tab_hline (t, TAL_1, 0, nc - 1, r * rows_per_variable + 1);
233 for (c = 0 ; c < matrix_cols ; ++c)
235 const struct variable *v = corr->n_vars_total > corr->n_vars1 ? corr->vars[corr->n_vars_total - corr->n_vars1 + c] : corr->vars[c];
236 tab_text (t, heading_columns + c, 0, TAB_LEFT | TAT_TITLE, var_to_string (v));
239 for (r = 0 ; r < corr->n_vars1 ; ++r)
241 const int row = r * rows_per_variable + heading_rows;
242 for (c = 0 ; c < matrix_cols ; ++c)
244 unsigned char flags = 0;
245 const int col_index = corr->n_vars_total - corr->n_vars1 + c;
246 double pearson = gsl_matrix_get (cm, r, col_index);
247 double w = gsl_matrix_get (samples, r, col_index);
248 double sig = opts->tails * significance_of_correlation (pearson, w);
250 if ( opts->missing_type != CORR_LISTWISE )
251 tab_double (t, c + heading_columns, row + rows_per_variable - 1, 0, w, wfmt);
254 tab_double (t, c + heading_columns, row + 1, 0, sig, NULL);
256 if ( opts->sig && c != r && sig < 0.05)
259 tab_double (t, c + heading_columns, row, flags, pearson, NULL);
261 if (opts->statistics & STATS_XPROD)
263 double cov = gsl_matrix_get (cv, r, col_index);
264 const double xprod_dev = cov * w;
265 cov *= w / (w - 1.0);
267 tab_double (t, c + heading_columns, row + 2, 0, xprod_dev, NULL);
268 tab_double (t, c + heading_columns, row + 3, 0, cov, NULL);
278 run_corr (struct casereader *r, const struct corr_opts *opts, const struct corr *corr)
281 const gsl_matrix *var_matrix, *samples_matrix, *mean_matrix;
282 const gsl_matrix *cov_matrix;
283 gsl_matrix *corr_matrix;
284 struct covariance *cov = covariance_create (corr->n_vars_total, corr->vars,
285 opts->wv, opts->exclude);
287 for ( ; (c = casereader_read (r) ); case_unref (c))
289 covariance_accumulate (cov, c);
292 cov_matrix = covariance_calculate (cov);
294 samples_matrix = covariance_moments (cov, MOMENT_NONE);
295 var_matrix = covariance_moments (cov, MOMENT_VARIANCE);
296 mean_matrix = covariance_moments (cov, MOMENT_MEAN);
298 corr_matrix = correlation_from_covariance (cov_matrix, var_matrix);
300 if ( opts->statistics & STATS_DESCRIPTIVES)
301 output_descriptives (corr, mean_matrix, var_matrix, samples_matrix);
303 output_correlation (corr, opts,
308 covariance_destroy (cov);
309 gsl_matrix_free (corr_matrix);
313 cmd_correlation (struct lexer *lexer, struct dataset *ds)
316 int n_all_vars = 0; /* Total number of variables involved in this command */
317 const struct variable **all_vars ;
318 const struct dictionary *dict = dataset_dict (ds);
321 struct casegrouper *grouper;
322 struct casereader *group;
324 struct corr *corr = NULL;
327 struct corr_opts opts;
328 opts.missing_type = CORR_PAIRWISE;
329 opts.wv = dict_get_weight (dict);
332 opts.exclude = MV_ANY;
335 /* Parse CORRELATIONS. */
336 while (lex_token (lexer) != '.')
338 lex_match (lexer, '/');
339 if (lex_match_id (lexer, "MISSING"))
341 lex_match (lexer, '=');
342 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
344 if (lex_match_id (lexer, "PAIRWISE"))
345 opts.missing_type = CORR_PAIRWISE;
346 else if (lex_match_id (lexer, "LISTWISE"))
347 opts.missing_type = CORR_LISTWISE;
349 else if (lex_match_id (lexer, "INCLUDE"))
350 opts.exclude = MV_SYSTEM;
351 else if (lex_match_id (lexer, "EXCLUDE"))
352 opts.exclude = MV_ANY;
355 lex_error (lexer, NULL);
358 lex_match (lexer, ',');
361 else if (lex_match_id (lexer, "PRINT"))
363 lex_match (lexer, '=');
364 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
366 if ( lex_match_id (lexer, "TWOTAIL"))
368 else if (lex_match_id (lexer, "ONETAIL"))
370 else if (lex_match_id (lexer, "SIG"))
372 else if (lex_match_id (lexer, "NOSIG"))
376 lex_error (lexer, NULL);
380 lex_match (lexer, ',');
383 else if (lex_match_id (lexer, "STATISTICS"))
385 lex_match (lexer, '=');
386 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
388 if ( lex_match_id (lexer, "DESCRIPTIVES"))
389 opts.statistics = STATS_DESCRIPTIVES;
390 else if (lex_match_id (lexer, "XPROD"))
391 opts.statistics = STATS_XPROD;
392 else if (lex_token (lexer) == T_ALL)
394 opts.statistics = STATS_ALL;
399 lex_error (lexer, NULL);
403 lex_match (lexer, ',');
408 if (lex_match_id (lexer, "VARIABLES"))
410 lex_match (lexer, '=');
413 corr = xrealloc (corr, sizeof (*corr) * (n_corrs + 1));
414 corr[n_corrs].n_vars_total = corr[n_corrs].n_vars1 = 0;
416 if ( ! parse_variables_const (lexer, dict, &corr[n_corrs].vars,
417 &corr[n_corrs].n_vars_total,
425 corr[n_corrs].n_vars1 = corr[n_corrs].n_vars_total;
427 if ( lex_match (lexer, T_WITH))
429 if ( ! parse_variables_const (lexer, dict,
430 &corr[n_corrs].vars, &corr[n_corrs].n_vars_total,
431 PV_NUMERIC | PV_APPEND))
438 n_all_vars += corr[n_corrs].n_vars_total;
446 msg (SE, _("No variables specified."));
451 all_vars = xmalloc (sizeof (*all_vars) * n_all_vars);
454 /* FIXME: Using a hash here would make more sense */
455 const struct variable **vv = all_vars;
457 for (i = 0 ; i < n_corrs; ++i)
460 const struct corr *c = &corr[i];
461 for (v = 0 ; v < c->n_vars_total; ++v)
466 grouper = casegrouper_create_splits (proc_open (ds), dict);
468 while (casegrouper_get_next_group (grouper, &group))
470 for (i = 0 ; i < n_corrs; ++i)
472 /* FIXME: No need to iterate the data multiple times */
473 struct casereader *r = casereader_clone (group);
475 if ( opts.missing_type == CORR_LISTWISE)
476 r = casereader_create_filter_missing (r, all_vars, n_all_vars,
477 opts.exclude, NULL, NULL);
480 run_corr (r, &opts, &corr[i]);
481 casereader_destroy (r);
483 casereader_destroy (group);
486 ok = casegrouper_destroy (grouper);
487 ok = proc_commit (ds) && ok;
494 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;