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/manager.h>
34 #include <output/table.h>
35 #include <libpspp/message.h>
36 #include <data/format.h>
37 #include <math/moments.h>
42 #include <libpspp/misc.h>
43 #include <gsl/gsl_cdf.h>
46 #define _(msgid) gettext (msgid)
47 #define N_(msgid) msgid
55 const struct variable **vars;
59 /* Handling of missing values. */
60 enum corr_missing_type
62 CORR_PAIRWISE, /* Handle missing values on a per-variable-pair basis. */
63 CORR_LISTWISE /* Discard entire case if any variable is missing. */
68 STATS_DESCRIPTIVES = 0x01,
70 STATS_ALL = STATS_XPROD | STATS_DESCRIPTIVES
75 enum corr_missing_type missing_type;
76 enum mv_class exclude; /* Classes of missing values to exclude. */
78 bool sig; /* Flag significant values or not */
79 int tails; /* Report significance with how many tails ? */
80 enum stats_opts statistics;
82 const struct variable *wv; /* The weight variable (if any) */
87 output_descriptives (const struct corr *corr, const gsl_matrix *means,
88 const gsl_matrix *vars, const gsl_matrix *ns)
90 const int nr = corr->n_vars_total + 1;
94 const int heading_columns = 1;
95 const int heading_rows = 1;
97 struct tab_table *t = tab_create (nc, nr);
98 tab_title (t, _("Descriptive Statistics"));
99 tab_dim (t, tab_natural_dimensions, NULL, NULL);
101 tab_headers (t, heading_columns, 0, heading_rows, 0);
103 /* Outline the box */
117 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
118 tab_hline (t, TAL_1, 0, nc - 1, heading_rows);
120 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
121 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
122 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
124 for (r = 0 ; r < corr->n_vars_total ; ++r)
126 const struct variable *v = corr->vars[r];
127 tab_text (t, 0, r + heading_rows, TAB_LEFT | TAT_TITLE, var_to_string (v));
129 for (c = 1 ; c < nc ; ++c)
136 x = gsl_matrix_get (means, r, 0);
139 x = gsl_matrix_get (vars, r, 0);
141 /* Here we want to display the non-biased estimator */
142 n = gsl_matrix_get (ns, r, 0);
148 x = gsl_matrix_get (ns, r, 0);
154 tab_double (t, c, r + heading_rows, 0, x, NULL);
162 output_correlation (const struct corr *corr, const struct corr_opts *opts,
163 const gsl_matrix *cm, const gsl_matrix *samples,
164 const gsl_matrix *cv)
169 int nr = corr->n_vars1;
170 int nc = matrix_cols = corr->n_vars_total > corr->n_vars1 ?
171 corr->n_vars_total - corr->n_vars1 : corr->n_vars1;
173 const struct fmt_spec *wfmt = opts->wv ? var_get_print_format (opts->wv) : & F_8_0;
175 const int heading_columns = 2;
176 const int heading_rows = 1;
178 int rows_per_variable = opts->missing_type == CORR_LISTWISE ? 2 : 3;
180 if (opts->statistics & STATS_XPROD)
181 rows_per_variable += 2;
183 /* Two header columns */
184 nc += heading_columns;
186 /* Three data per variable */
187 nr *= rows_per_variable;
192 t = tab_create (nc, nr);
193 tab_title (t, _("Correlations"));
194 tab_dim (t, tab_natural_dimensions, NULL, NULL);
196 tab_headers (t, heading_columns, 0, heading_rows, 0);
198 /* Outline the box */
212 tab_vline (t, TAL_2, heading_columns, 0, nr - 1);
213 tab_vline (t, TAL_1, 1, heading_rows, nr - 1);
215 for (r = 0 ; r < corr->n_vars1 ; ++r)
217 tab_text (t, 0, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
218 var_to_string (corr->vars[r]));
220 tab_text (t, 1, 1 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Pearson Correlation"));
221 tab_text (t, 1, 2 + r * rows_per_variable, TAB_LEFT | TAT_TITLE,
222 (opts->tails == 2) ? _("Sig. (2-tailed)") : _("Sig. (1-tailed)"));
224 if (opts->statistics & STATS_XPROD)
226 tab_text (t, 1, 3 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Cross-products"));
227 tab_text (t, 1, 4 + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("Covariance"));
230 if ( opts->missing_type != CORR_LISTWISE )
231 tab_text (t, 1, rows_per_variable + r * rows_per_variable, TAB_LEFT | TAT_TITLE, _("N"));
233 tab_hline (t, TAL_1, 0, nc - 1, r * rows_per_variable + 1);
236 for (c = 0 ; c < matrix_cols ; ++c)
238 const struct variable *v = corr->n_vars_total > corr->n_vars1 ? corr->vars[corr->n_vars_total - corr->n_vars1 + c] : corr->vars[c];
239 tab_text (t, heading_columns + c, 0, TAB_LEFT | TAT_TITLE, var_to_string (v));
242 for (r = 0 ; r < corr->n_vars1 ; ++r)
244 const int row = r * rows_per_variable + heading_rows;
245 for (c = 0 ; c < matrix_cols ; ++c)
247 unsigned char flags = 0;
248 const int col_index = corr->n_vars_total - corr->n_vars1 + c;
249 double pearson = gsl_matrix_get (cm, r, col_index);
250 double w = gsl_matrix_get (samples, r, col_index);
251 double sig = opts->tails * significance_of_correlation (pearson, w);
253 if ( opts->missing_type != CORR_LISTWISE )
254 tab_double (t, c + heading_columns, row + rows_per_variable - 1, 0, w, wfmt);
257 tab_double (t, c + heading_columns, row + 1, 0, sig, NULL);
259 if ( opts->sig && c != r && sig < 0.05)
262 tab_double (t, c + heading_columns, row, flags, pearson, NULL);
264 if (opts->statistics & STATS_XPROD)
266 double cov = gsl_matrix_get (cv, r, col_index);
267 const double xprod_dev = cov * w;
268 cov *= w / (w - 1.0);
270 tab_double (t, c + heading_columns, row + 2, 0, xprod_dev, NULL);
271 tab_double (t, c + heading_columns, row + 3, 0, cov, NULL);
281 run_corr (struct casereader *r, const struct corr_opts *opts, const struct corr *corr)
284 const gsl_matrix *var_matrix, *samples_matrix, *mean_matrix;
285 const gsl_matrix *cov_matrix;
286 gsl_matrix *corr_matrix;
287 struct covariance *cov = covariance_create (corr->n_vars_total, corr->vars,
288 opts->wv, opts->exclude);
290 for ( ; (c = casereader_read (r) ); case_unref (c))
292 covariance_accumulate (cov, c);
295 cov_matrix = covariance_calculate (cov);
297 samples_matrix = covariance_moments (cov, MOMENT_NONE);
298 var_matrix = covariance_moments (cov, MOMENT_VARIANCE);
299 mean_matrix = covariance_moments (cov, MOMENT_MEAN);
301 corr_matrix = correlation_from_covariance (cov_matrix, var_matrix);
303 if ( opts->statistics & STATS_DESCRIPTIVES)
304 output_descriptives (corr, mean_matrix, var_matrix, samples_matrix);
306 output_correlation (corr, opts,
311 covariance_destroy (cov);
312 gsl_matrix_free (corr_matrix);
316 cmd_correlation (struct lexer *lexer, struct dataset *ds)
319 int n_all_vars = 0; /* Total number of variables involved in this command */
320 const struct variable **all_vars ;
321 const struct dictionary *dict = dataset_dict (ds);
324 struct casegrouper *grouper;
325 struct casereader *group;
327 struct corr *corr = NULL;
330 struct corr_opts opts;
331 opts.missing_type = CORR_PAIRWISE;
332 opts.wv = dict_get_weight (dict);
335 opts.exclude = MV_ANY;
338 /* Parse CORRELATIONS. */
339 while (lex_token (lexer) != '.')
341 lex_match (lexer, '/');
342 if (lex_match_id (lexer, "MISSING"))
344 lex_match (lexer, '=');
345 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
347 if (lex_match_id (lexer, "PAIRWISE"))
348 opts.missing_type = CORR_PAIRWISE;
349 else if (lex_match_id (lexer, "LISTWISE"))
350 opts.missing_type = CORR_LISTWISE;
352 else if (lex_match_id (lexer, "INCLUDE"))
353 opts.exclude = MV_SYSTEM;
354 else if (lex_match_id (lexer, "EXCLUDE"))
355 opts.exclude = MV_ANY;
358 lex_error (lexer, NULL);
361 lex_match (lexer, ',');
364 else if (lex_match_id (lexer, "PRINT"))
366 lex_match (lexer, '=');
367 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
369 if ( lex_match_id (lexer, "TWOTAIL"))
371 else if (lex_match_id (lexer, "ONETAIL"))
373 else if (lex_match_id (lexer, "SIG"))
375 else if (lex_match_id (lexer, "NOSIG"))
379 lex_error (lexer, NULL);
383 lex_match (lexer, ',');
386 else if (lex_match_id (lexer, "STATISTICS"))
388 lex_match (lexer, '=');
389 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
391 if ( lex_match_id (lexer, "DESCRIPTIVES"))
392 opts.statistics = STATS_DESCRIPTIVES;
393 else if (lex_match_id (lexer, "XPROD"))
394 opts.statistics = STATS_XPROD;
395 else if (lex_token (lexer) == T_ALL)
397 opts.statistics = STATS_ALL;
402 lex_error (lexer, NULL);
406 lex_match (lexer, ',');
411 if (lex_match_id (lexer, "VARIABLES"))
413 lex_match (lexer, '=');
416 corr = xrealloc (corr, sizeof (*corr) * (n_corrs + 1));
417 corr[n_corrs].n_vars_total = corr[n_corrs].n_vars1 = 0;
419 if ( ! parse_variables_const (lexer, dict, &corr[n_corrs].vars,
420 &corr[n_corrs].n_vars_total,
428 corr[n_corrs].n_vars1 = corr[n_corrs].n_vars_total;
430 if ( lex_match (lexer, T_WITH))
432 if ( ! parse_variables_const (lexer, dict,
433 &corr[n_corrs].vars, &corr[n_corrs].n_vars_total,
434 PV_NUMERIC | PV_APPEND))
441 n_all_vars += corr[n_corrs].n_vars_total;
449 msg (SE, _("No variables specified."));
454 all_vars = xmalloc (sizeof (*all_vars) * n_all_vars);
457 /* FIXME: Using a hash here would make more sense */
458 const struct variable **vv = all_vars;
460 for (i = 0 ; i < n_corrs; ++i)
463 const struct corr *c = &corr[i];
464 for (v = 0 ; v < c->n_vars_total; ++v)
469 grouper = casegrouper_create_splits (proc_open (ds), dict);
471 while (casegrouper_get_next_group (grouper, &group))
473 for (i = 0 ; i < n_corrs; ++i)
475 /* FIXME: No need to iterate the data multiple times */
476 struct casereader *r = casereader_clone (group);
478 if ( opts.missing_type == CORR_LISTWISE)
479 r = casereader_create_filter_missing (r, all_vars, n_all_vars,
480 opts.exclude, NULL, NULL);
483 run_corr (r, &opts, &corr[i]);
484 casereader_destroy (r);
486 casereader_destroy (group);
489 ok = casegrouper_destroy (grouper);
490 ok = proc_commit (ds) && ok;
497 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;