1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2005, 2009, 2010, 2011 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 <gsl/gsl_cdf.h>
20 #include <gsl/gsl_matrix.h>
21 #include <gsl/gsl_vector.h>
24 #include <data/case.h>
25 #include <data/casegrouper.h>
26 #include <data/casereader.h>
27 #include <data/dictionary.h>
28 #include <data/missing-values.h>
29 #include <data/procedure.h>
30 #include <data/transformations.h>
31 #include <data/value-labels.h>
32 #include <data/variable.h>
33 #include <language/command.h>
34 #include <language/dictionary/split-file.h>
35 #include <language/data-io/file-handle.h>
36 #include <language/lexer/lexer.h>
37 #include <libpspp/compiler.h>
38 #include <libpspp/message.h>
39 #include <libpspp/taint.h>
40 #include <math/covariance.h>
41 #include <math/linreg.h>
42 #include <math/moments.h>
43 #include <output/tab.h>
45 #include "gl/intprops.h"
46 #include "gl/xalloc.h"
49 #define _(msgid) gettext (msgid)
51 #define REG_LARGE_DATA 1000
56 "REGRESSION" (regression_):
76 +save[sv_]=resid,pred;
81 static struct cmd_regression cmd;
84 Moments for each of the variables used.
89 const struct variable *v;
93 Transformations for saving predicted values
98 int n_trns; /* Number of transformations. */
99 int trns_id; /* Which trns is this one? */
100 linreg *c; /* Linear model for this trns. */
103 Variables used (both explanatory and response).
105 static const struct variable **v_variables;
110 static size_t n_variables;
112 static bool run_regression (struct casereader *, struct cmd_regression *,
113 struct dataset *, linreg **);
116 STATISTICS subcommand output functions.
118 static void reg_stats_r (linreg *, void *);
119 static void reg_stats_coeff (linreg *, void *);
120 static void reg_stats_anova (linreg *, void *);
121 static void reg_stats_outs (linreg *, void *);
122 static void reg_stats_zpp (linreg *, void *);
123 static void reg_stats_label (linreg *, void *);
124 static void reg_stats_sha (linreg *, void *);
125 static void reg_stats_ci (linreg *, void *);
126 static void reg_stats_f (linreg *, void *);
127 static void reg_stats_bcov (linreg *, void *);
128 static void reg_stats_ses (linreg *, void *);
129 static void reg_stats_xtx (linreg *, void *);
130 static void reg_stats_collin (linreg *, void *);
131 static void reg_stats_tol (linreg *, void *);
132 static void reg_stats_selection (linreg *, void *);
133 static void statistics_keyword_output (void (*)(linreg *, void *),
134 int, linreg *, void *);
137 reg_stats_r (linreg *c, void *aux UNUSED)
147 rsq = linreg_ssreg (c) / linreg_sst (c);
148 adjrsq = 1.0 - (1.0 - rsq) * (linreg_n_obs (c) - 1.0) / (linreg_n_obs (c) - linreg_n_coeffs (c));
149 std_error = sqrt (linreg_mse (c));
150 t = tab_create (n_cols, n_rows);
151 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
152 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
153 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
154 tab_vline (t, TAL_0, 1, 0, 0);
156 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("R"));
157 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("R Square"));
158 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Adjusted R Square"));
159 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error of the Estimate"));
160 tab_double (t, 1, 1, TAB_RIGHT, sqrt (rsq), NULL);
161 tab_double (t, 2, 1, TAB_RIGHT, rsq, NULL);
162 tab_double (t, 3, 1, TAB_RIGHT, adjrsq, NULL);
163 tab_double (t, 4, 1, TAB_RIGHT, std_error, NULL);
164 tab_title (t, _("Model Summary"));
169 Table showing estimated regression coefficients.
172 reg_stats_coeff (linreg * c, void *aux_)
184 const struct variable *v;
186 gsl_matrix *cov = aux_;
189 n_rows = linreg_n_coeffs (c) + 3;
191 t = tab_create (n_cols, n_rows);
192 tab_headers (t, 2, 0, 1, 0);
193 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
194 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
195 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
196 tab_vline (t, TAL_0, 1, 0, 0);
198 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("B"));
199 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Error"));
200 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Beta"));
201 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("t"));
202 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
203 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("(Constant)"));
204 tab_double (t, 2, 1, 0, linreg_intercept (c), NULL);
205 std_err = sqrt (gsl_matrix_get (linreg_cov (c), 0, 0));
206 tab_double (t, 3, 1, 0, std_err, NULL);
207 tab_double (t, 4, 1, 0, 0.0, NULL);
208 t_stat = linreg_intercept (c) / std_err;
209 tab_double (t, 5, 1, 0, t_stat, NULL);
210 pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), (double) (linreg_n_obs (c) - linreg_n_coeffs (c)));
211 tab_double (t, 6, 1, 0, pval, NULL);
212 for (j = 0; j < linreg_n_coeffs (c); j++)
215 ds_init_empty (&tstr);
218 v = linreg_indep_var (c, j);
219 label = var_to_string (v);
220 /* Do not overwrite the variable's name. */
221 ds_put_cstr (&tstr, label);
222 tab_text (t, 1, this_row, TAB_CENTER, ds_cstr (&tstr));
224 Regression coefficients.
226 tab_double (t, 2, this_row, 0, linreg_coeff (c, j), NULL);
228 Standard error of the coefficients.
230 std_err = sqrt (gsl_matrix_get (linreg_cov (c), j + 1, j + 1));
231 tab_double (t, 3, this_row, 0, std_err, NULL);
233 Standardized coefficient, i.e., regression coefficient
234 if all variables had unit variance.
236 beta = sqrt (gsl_matrix_get (cov, j, j));
237 beta *= linreg_coeff (c, j) /
238 sqrt (gsl_matrix_get (cov, cov->size1 - 1, cov->size2 - 1));
239 tab_double (t, 4, this_row, 0, beta, NULL);
242 Test statistic for H0: coefficient is 0.
244 t_stat = linreg_coeff (c, j) / std_err;
245 tab_double (t, 5, this_row, 0, t_stat, NULL);
247 P values for the test statistic above.
250 2 * gsl_cdf_tdist_Q (fabs (t_stat),
251 (double) (linreg_n_obs (c) - linreg_n_coeffs (c)));
252 tab_double (t, 6, this_row, 0, pval, NULL);
255 tab_title (t, _("Coefficients"));
260 Display the ANOVA table.
263 reg_stats_anova (linreg * c, void *aux UNUSED)
267 const double msm = linreg_ssreg (c) / linreg_dfmodel (c);
268 const double mse = linreg_mse (c);
269 const double F = msm / mse;
270 const double pval = gsl_cdf_fdist_Q (F, c->dfm, c->dfe);
275 t = tab_create (n_cols, n_rows);
276 tab_headers (t, 2, 0, 1, 0);
278 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
280 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
281 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
282 tab_vline (t, TAL_0, 1, 0, 0);
284 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
285 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
286 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
287 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
288 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
290 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Regression"));
291 tab_text (t, 1, 2, TAB_LEFT | TAT_TITLE, _("Residual"));
292 tab_text (t, 1, 3, TAB_LEFT | TAT_TITLE, _("Total"));
294 /* Sums of Squares */
295 tab_double (t, 2, 1, 0, linreg_ssreg (c), NULL);
296 tab_double (t, 2, 3, 0, linreg_sst (c), NULL);
297 tab_double (t, 2, 2, 0, linreg_sse (c), NULL);
300 /* Degrees of freedom */
301 tab_text_format (t, 3, 1, TAB_RIGHT, "%g", c->dfm);
302 tab_text_format (t, 3, 2, TAB_RIGHT, "%g", c->dfe);
303 tab_text_format (t, 3, 3, TAB_RIGHT, "%g", c->dft);
306 tab_double (t, 4, 1, TAB_RIGHT, msm, NULL);
307 tab_double (t, 4, 2, TAB_RIGHT, mse, NULL);
309 tab_double (t, 5, 1, 0, F, NULL);
311 tab_double (t, 6, 1, 0, pval, NULL);
313 tab_title (t, _("ANOVA"));
318 reg_stats_outs (linreg * c, void *aux UNUSED)
324 reg_stats_zpp (linreg * c, void *aux UNUSED)
330 reg_stats_label (linreg * c, void *aux UNUSED)
336 reg_stats_sha (linreg * c, void *aux UNUSED)
341 reg_stats_ci (linreg * c, void *aux UNUSED)
346 reg_stats_f (linreg * c, void *aux UNUSED)
351 reg_stats_bcov (linreg * c, void *aux UNUSED)
363 n_cols = c->n_indeps + 1 + 2;
364 n_rows = 2 * (c->n_indeps + 1);
365 t = tab_create (n_cols, n_rows);
366 tab_headers (t, 2, 0, 1, 0);
367 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
368 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
369 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
370 tab_vline (t, TAL_0, 1, 0, 0);
371 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Model"));
372 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Covariances"));
373 for (i = 0; i < linreg_n_coeffs (c); i++)
375 const struct variable *v = linreg_indep_var (c, i);
376 label = var_to_string (v);
377 tab_text (t, 2, i, TAB_CENTER, label);
378 tab_text (t, i + 2, 0, TAB_CENTER, label);
379 for (k = 1; k < linreg_n_coeffs (c); k++)
381 col = (i <= k) ? k : i;
382 row = (i <= k) ? i : k;
383 tab_double (t, k + 2, i, TAB_CENTER,
384 gsl_matrix_get (c->cov, row, col), NULL);
387 tab_title (t, _("Coefficient Correlations"));
391 reg_stats_ses (linreg * c, void *aux UNUSED)
396 reg_stats_xtx (linreg * c, void *aux UNUSED)
401 reg_stats_collin (linreg * c, void *aux UNUSED)
406 reg_stats_tol (linreg * c, void *aux UNUSED)
411 reg_stats_selection (linreg * c, void *aux UNUSED)
417 statistics_keyword_output (void (*function) (linreg *, void *),
418 int keyword, linreg * c, void *aux)
422 (*function) (c, aux);
427 subcommand_statistics (int *keywords, linreg * c, void *aux)
430 The order here must match the order in which the STATISTICS
431 keywords appear in the specification section above.
458 Set everything but F.
460 for (i = 0; i < f; i++)
467 for (i = 0; i < all; i++)
475 Default output: ANOVA table, parameter estimates,
476 and statistics for variables not entered into model,
479 if (keywords[defaults] | d)
487 statistics_keyword_output (reg_stats_r, keywords[r], c, aux);
488 statistics_keyword_output (reg_stats_anova, keywords[anova], c, aux);
489 statistics_keyword_output (reg_stats_coeff, keywords[coeff], c, aux);
490 statistics_keyword_output (reg_stats_outs, keywords[outs], c, aux);
491 statistics_keyword_output (reg_stats_zpp, keywords[zpp], c, aux);
492 statistics_keyword_output (reg_stats_label, keywords[label], c, aux);
493 statistics_keyword_output (reg_stats_sha, keywords[sha], c, aux);
494 statistics_keyword_output (reg_stats_ci, keywords[ci], c, aux);
495 statistics_keyword_output (reg_stats_f, keywords[f], c, aux);
496 statistics_keyword_output (reg_stats_bcov, keywords[bcov], c, aux);
497 statistics_keyword_output (reg_stats_ses, keywords[ses], c, aux);
498 statistics_keyword_output (reg_stats_xtx, keywords[xtx], c, aux);
499 statistics_keyword_output (reg_stats_collin, keywords[collin], c, aux);
500 statistics_keyword_output (reg_stats_tol, keywords[tol], c, aux);
501 statistics_keyword_output (reg_stats_selection, keywords[selection], c, aux);
505 Free the transformation. Free its linear model if this
506 transformation is the last one.
509 regression_trns_free (void *t_)
512 struct reg_trns *t = t_;
514 if (t->trns_id == t->n_trns)
516 result = linreg_free (t->c);
524 Gets the predicted values.
527 regression_trns_pred_proc (void *t_, struct ccase **c,
528 casenumber case_idx UNUSED)
532 struct reg_trns *trns = t_;
534 union value *output = NULL;
535 const union value *tmp;
537 const struct variable **vars = NULL;
539 assert (trns != NULL);
541 assert (model != NULL);
542 assert (model->depvar != NULL);
543 assert (model->pred != NULL);
545 vars = linreg_get_vars (model);
546 n_vals = linreg_n_coeffs (model);
547 vals = xnmalloc (n_vals, sizeof (*vals));
548 *c = case_unshare (*c);
550 output = case_data_rw (*c, model->pred);
552 for (i = 0; i < n_vals; i++)
554 tmp = case_data (*c, vars[i]);
557 output->f = linreg_predict (model, vals, n_vals);
559 return TRNS_CONTINUE;
566 regression_trns_resid_proc (void *t_, struct ccase **c,
567 casenumber case_idx UNUSED)
571 struct reg_trns *trns = t_;
573 union value *output = NULL;
574 const union value *tmp;
577 const struct variable **vars = NULL;
579 assert (trns != NULL);
581 assert (model != NULL);
582 assert (model->depvar != NULL);
583 assert (model->resid != NULL);
585 vars = linreg_get_vars (model);
586 n_vals = linreg_n_coeffs (model);
588 vals = xnmalloc (n_vals, sizeof (*vals));
589 *c = case_unshare (*c);
590 output = case_data_rw (*c, model->resid);
591 assert (output != NULL);
593 for (i = 0; i < n_vals; i++)
595 tmp = case_data (*c, vars[i]);
598 tmp = case_data (*c, model->depvar);
600 output->f = linreg_residual (model, obs, vals, n_vals);
603 return TRNS_CONTINUE;
607 reg_get_name (const struct dictionary *dict, const char *prefix)
612 /* XXX handle too-long prefixes */
613 name = xmalloc (strlen (prefix) + INT_BUFSIZE_BOUND (i) + 1);
616 sprintf (name, "%s%d", prefix, i);
617 if (dict_lookup_var (dict, name) == NULL)
623 reg_save_var (struct dataset *ds, const char *prefix, trns_proc_func * f,
624 linreg * c, struct variable **v, int n_trns)
626 struct dictionary *dict = dataset_dict (ds);
627 static int trns_index = 1;
629 struct variable *new_var;
630 struct reg_trns *t = NULL;
632 t = xmalloc (sizeof (*t));
633 t->trns_id = trns_index;
637 name = reg_get_name (dict, prefix);
638 new_var = dict_create_var_assert (dict, name, 0);
642 add_transformation (ds, f, regression_trns_free, t);
646 subcommand_save (struct dataset *ds, int save, linreg ** models)
654 /* Count the number of transformations we will need. */
655 for (i = 0; i < REGRESSION_SV_count; i++)
662 n_trns *= cmd.n_dependent;
664 for (lc = models; lc < models + cmd.n_dependent; lc++)
668 if ((*lc)->depvar != NULL)
670 if (cmd.a_save[REGRESSION_SV_RESID])
672 reg_save_var (ds, "RES", regression_trns_resid_proc, *lc,
673 &(*lc)->resid, n_trns);
675 if (cmd.a_save[REGRESSION_SV_PRED])
677 reg_save_var (ds, "PRED", regression_trns_pred_proc, *lc,
678 &(*lc)->pred, n_trns);
686 for (lc = models; lc < models + cmd.n_dependent; lc++)
697 cmd_regression (struct lexer *lexer, struct dataset *ds)
699 struct casegrouper *grouper;
700 struct casereader *group;
705 if (!parse_regression (lexer, ds, &cmd, NULL))
710 models = xnmalloc (cmd.n_dependent, sizeof *models);
711 for (i = 0; i < cmd.n_dependent; i++)
717 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
718 while (casegrouper_get_next_group (grouper, &group))
719 run_regression (group, &cmd, ds, models);
720 ok = casegrouper_destroy (grouper);
721 ok = proc_commit (ds) && ok;
723 subcommand_save (ds, cmd.sbc_save, models);
726 free_regression (&cmd);
728 return ok ? CMD_SUCCESS : CMD_FAILURE;
732 Is variable k the dependent variable?
735 is_depvar (size_t k, const struct variable *v)
737 return v == v_variables[k];
740 /* Parser for the variables sub command */
742 regression_custom_variables (struct lexer *lexer, struct dataset *ds,
743 struct cmd_regression *cmd UNUSED,
746 const struct dictionary *dict = dataset_dict (ds);
748 lex_match (lexer, T_EQUALS);
750 if ((lex_token (lexer) != T_ID
751 || dict_lookup_var (dict, lex_tokcstr (lexer)) == NULL)
752 && lex_token (lexer) != T_ALL)
756 if (!parse_variables_const
757 (lexer, dict, &v_variables, &n_variables, PV_NONE))
762 assert (n_variables);
767 /* Identify the explanatory variables in v_variables. Returns
768 the number of independent variables. */
770 identify_indep_vars (const struct variable **indep_vars,
771 const struct variable *depvar)
773 int n_indep_vars = 0;
776 for (i = 0; i < n_variables; i++)
777 if (!is_depvar (i, depvar))
778 indep_vars[n_indep_vars++] = v_variables[i];
779 if ((n_indep_vars < 1) && is_depvar (0, depvar))
782 There is only one independent variable, and it is the same
783 as the dependent variable. Print a warning and continue.
786 gettext ("The dependent variable is equal to the independent variable."
787 "The least squares line is therefore Y=X."
788 "Standard errors and related statistics may be meaningless."));
790 indep_vars[0] = v_variables[0];
795 fill_covariance (gsl_matrix *cov, struct covariance *all_cov,
796 const struct variable **vars,
797 size_t n_vars, const struct variable *dep_var,
798 const struct variable **all_vars, size_t n_all_vars,
803 size_t dep_subscript;
805 const gsl_matrix *ssizes;
807 const gsl_matrix *mean_matrix;
808 const gsl_matrix *ssize_matrix;
811 cm = covariance_calculate_unnormalized (all_cov);
812 rows = xnmalloc (cov->size1 - 1, sizeof (*rows));
814 for (i = 0; i < n_all_vars; i++)
816 for (j = 0; j < n_vars; j++)
818 if (vars[j] == all_vars[i])
823 if (all_vars[i] == dep_var)
828 mean_matrix = covariance_moments (all_cov, MOMENT_MEAN);
829 ssize_matrix = covariance_moments (all_cov, MOMENT_NONE);
830 for (i = 0; i < cov->size1 - 1; i++)
832 means[i] = gsl_matrix_get (mean_matrix, rows[i], 0)
833 / gsl_matrix_get (ssize_matrix, rows[i], 0);
834 for (j = 0; j < cov->size2 - 1; j++)
836 gsl_matrix_set (cov, i, j, gsl_matrix_get (cm, rows[i], rows[j]));
837 gsl_matrix_set (cov, j, i, gsl_matrix_get (cm, rows[j], rows[i]));
840 means[cov->size1 - 1] = gsl_matrix_get (mean_matrix, dep_subscript, 0)
841 / gsl_matrix_get (ssize_matrix, dep_subscript, 0);
842 ssizes = covariance_moments (all_cov, MOMENT_NONE);
843 result = gsl_matrix_get (ssizes, dep_subscript, rows[0]);
844 for (i = 0; i < cov->size1 - 1; i++)
846 gsl_matrix_set (cov, i, cov->size1 - 1,
847 gsl_matrix_get (cm, rows[i], dep_subscript));
848 gsl_matrix_set (cov, cov->size1 - 1, i,
849 gsl_matrix_get (cm, rows[i], dep_subscript));
850 if (result > gsl_matrix_get (ssizes, rows[i], dep_subscript))
852 result = gsl_matrix_get (ssizes, rows[i], dep_subscript);
855 gsl_matrix_set (cov, cov->size1 - 1, cov->size1 - 1,
856 gsl_matrix_get (cm, dep_subscript, dep_subscript));
858 gsl_matrix_free (cm);
862 get_n_all_vars (struct cmd_regression *cmd)
864 size_t result = n_variables;
868 result += cmd->n_dependent;
869 for (i = 0; i < cmd->n_dependent; i++)
871 for (j = 0; j < n_variables; j++)
873 if (v_variables[j] == cmd->v_dependent[i])
882 fill_all_vars (const struct variable **vars, struct cmd_regression *cmd)
888 for (i = 0; i < n_variables; i++)
890 vars[i] = v_variables[i];
892 for (i = 0; i < cmd->n_dependent; i++)
895 for (j = 0; j < n_variables; j++)
897 if (cmd->v_dependent[i] == v_variables[j])
905 vars[i + n_variables] = cmd->v_dependent[i];
910 run_regression (struct casereader *input, struct cmd_regression *cmd,
911 struct dataset *ds, linreg **models)
919 struct covariance *cov;
920 const struct variable **vars;
921 const struct variable **all_vars;
922 const struct variable *dep_var;
923 struct casereader *reader;
924 const struct dictionary *dict;
927 assert (models != NULL);
929 for (i = 0; i < n_variables; i++)
931 if (!var_is_numeric (v_variables[i]))
933 msg (SE, _("REGRESSION requires numeric variables."));
938 c = casereader_peek (input, 0);
941 casereader_destroy (input);
944 output_split_file_values (ds, c);
947 dict = dataset_dict (ds);
950 dict_get_vars (dict, &v_variables, &n_variables, 0);
952 n_all_vars = get_n_all_vars (cmd);
953 all_vars = xnmalloc (n_all_vars, sizeof (*all_vars));
954 fill_all_vars (all_vars, cmd);
955 vars = xnmalloc (n_variables, sizeof (*vars));
956 means = xnmalloc (n_all_vars, sizeof (*means));
957 cov = covariance_1pass_create (n_all_vars, all_vars,
958 dict_get_weight (dict), MV_ANY);
960 reader = casereader_clone (input);
961 reader = casereader_create_filter_missing (reader, v_variables, n_variables,
963 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
965 covariance_accumulate (cov, c);
968 for (k = 0; k < cmd->n_dependent; k++)
971 dep_var = cmd->v_dependent[k];
972 n_indep = identify_indep_vars (vars, dep_var);
974 this_cm = gsl_matrix_alloc (n_indep + 1, n_indep + 1);
975 n_data = fill_covariance (this_cm, cov, vars, n_indep,
976 dep_var, all_vars, n_all_vars, means);
977 models[k] = linreg_alloc (dep_var, (const struct variable **) vars,
979 models[k]->depvar = dep_var;
980 for (i = 0; i < n_indep; i++)
982 linreg_set_indep_variable_mean (models[k], i, means[i]);
984 linreg_set_depvar_mean (models[k], means[i]);
986 For large data sets, use QR decomposition.
988 if (n_data > sqrt (n_indep) && n_data > REG_LARGE_DATA)
990 models[k]->method = LINREG_QR;
996 Find the least-squares estimates and other statistics.
998 linreg_fit (this_cm, models[k]);
1000 if (!taint_has_tainted_successor (casereader_get_taint (input)))
1002 subcommand_statistics (cmd->a_statistics, models[k], this_cm);
1008 gettext ("No valid data found. This command was skipped."));
1009 linreg_free (models[k]);
1012 gsl_matrix_free (this_cm);
1015 casereader_destroy (reader);
1019 casereader_destroy (input);
1020 covariance_destroy (cov);