1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2005, 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 <gsl/gsl_cdf.h>
20 #include <gsl/gsl_matrix.h>
21 #include <gsl/gsl_vector.h>
25 #include <data/case.h>
26 #include <data/casegrouper.h>
27 #include <data/casereader.h>
28 #include <data/category.h>
29 #include <data/dictionary.h>
30 #include <data/missing-values.h>
31 #include <data/procedure.h>
32 #include <data/transformations.h>
33 #include <data/value-labels.h>
34 #include <data/variable.h>
35 #include <language/command.h>
36 #include <language/dictionary/split-file.h>
37 #include <language/data-io/file-handle.h>
38 #include <language/lexer/lexer.h>
39 #include <libpspp/compiler.h>
40 #include <libpspp/message.h>
41 #include <libpspp/taint.h>
42 #include <math/design-matrix.h>
43 #include <math/coefficient.h>
44 #include <math/linreg.h>
45 #include <math/moments.h>
46 #include <output/table.h>
51 #define _(msgid) gettext (msgid)
53 #define REG_LARGE_DATA 1000
58 "REGRESSION" (regression_):
78 +save[sv_]=resid,pred;
83 static struct cmd_regression cmd;
86 Moments for each of the variables used.
91 const struct variable *v;
95 Transformations for saving predicted values
100 int n_trns; /* Number of transformations. */
101 int trns_id; /* Which trns is this one? */
102 pspp_linreg_cache *c; /* Linear model for this trns. */
105 Variables used (both explanatory and response).
107 static const struct variable **v_variables;
112 static size_t n_variables;
114 static bool run_regression (struct casereader *, struct cmd_regression *,
115 struct dataset *, pspp_linreg_cache **);
118 STATISTICS subcommand output functions.
120 static void reg_stats_r (pspp_linreg_cache *);
121 static void reg_stats_coeff (pspp_linreg_cache *);
122 static void reg_stats_anova (pspp_linreg_cache *);
123 static void reg_stats_outs (pspp_linreg_cache *);
124 static void reg_stats_zpp (pspp_linreg_cache *);
125 static void reg_stats_label (pspp_linreg_cache *);
126 static void reg_stats_sha (pspp_linreg_cache *);
127 static void reg_stats_ci (pspp_linreg_cache *);
128 static void reg_stats_f (pspp_linreg_cache *);
129 static void reg_stats_bcov (pspp_linreg_cache *);
130 static void reg_stats_ses (pspp_linreg_cache *);
131 static void reg_stats_xtx (pspp_linreg_cache *);
132 static void reg_stats_collin (pspp_linreg_cache *);
133 static void reg_stats_tol (pspp_linreg_cache *);
134 static void reg_stats_selection (pspp_linreg_cache *);
135 static void statistics_keyword_output (void (*)(pspp_linreg_cache *),
136 int, pspp_linreg_cache *);
139 reg_stats_r (pspp_linreg_cache * c)
149 rsq = c->ssm / c->sst;
150 adjrsq = 1.0 - (1.0 - rsq) * (c->n_obs - 1.0) / (c->n_obs - c->n_indeps);
151 std_error = sqrt (pspp_linreg_mse (c));
152 t = tab_create (n_cols, n_rows, 0);
153 tab_dim (t, tab_natural_dimensions, NULL);
154 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
155 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
156 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
157 tab_vline (t, TAL_0, 1, 0, 0);
159 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("R"));
160 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("R Square"));
161 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Adjusted R Square"));
162 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error of the Estimate"));
163 tab_double (t, 1, 1, TAB_RIGHT, sqrt (rsq), NULL);
164 tab_double (t, 2, 1, TAB_RIGHT, rsq, NULL);
165 tab_double (t, 3, 1, TAB_RIGHT, adjrsq, NULL);
166 tab_double (t, 4, 1, TAB_RIGHT, std_error, NULL);
167 tab_title (t, _("Model Summary"));
172 Table showing estimated regression coefficients.
175 reg_stats_coeff (pspp_linreg_cache * c)
187 const struct variable *v;
188 const union value *val;
192 n_rows = c->n_coeffs + 3;
194 t = tab_create (n_cols, n_rows, 0);
195 tab_headers (t, 2, 0, 1, 0);
196 tab_dim (t, tab_natural_dimensions, NULL);
197 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
198 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
199 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
200 tab_vline (t, TAL_0, 1, 0, 0);
202 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("B"));
203 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Error"));
204 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Beta"));
205 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("t"));
206 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
207 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("(Constant)"));
208 tab_double (t, 2, 1, 0, c->intercept, NULL);
209 std_err = sqrt (gsl_matrix_get (c->cov, 0, 0));
210 tab_double (t, 3, 1, 0, std_err, NULL);
211 tab_double (t, 4, 1, 0, 0.0, NULL);
212 t_stat = c->intercept / std_err;
213 tab_double (t, 5, 1, 0, t_stat, NULL);
214 pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), 1.0);
215 tab_double (t, 6, 1, 0, pval, NULL);
216 for (j = 0; j < c->n_coeffs; j++)
219 ds_init_empty (&tstr);
222 v = pspp_coeff_get_var (c->coeff[j], 0);
223 label = var_to_string (v);
224 /* Do not overwrite the variable's name. */
225 ds_put_cstr (&tstr, label);
226 if (var_is_alpha (v))
229 Append the value associated with this coefficient.
230 This makes sense only if we us the usual binary encoding
234 val = pspp_coeff_get_value (c->coeff[j], v);
236 var_append_value_name (v, val, &tstr);
239 tab_text (t, 1, this_row, TAB_CENTER, ds_cstr (&tstr));
241 Regression coefficients.
243 tab_double (t, 2, this_row, 0, c->coeff[j]->estimate, NULL);
245 Standard error of the coefficients.
247 std_err = sqrt (gsl_matrix_get (c->cov, j + 1, j + 1));
248 tab_double (t, 3, this_row, 0, std_err, NULL);
250 Standardized coefficient, i.e., regression coefficient
251 if all variables had unit variance.
253 beta = pspp_coeff_get_sd (c->coeff[j]);
254 beta *= c->coeff[j]->estimate / c->depvar_std;
255 tab_double (t, 4, this_row, 0, beta, NULL);
258 Test statistic for H0: coefficient is 0.
260 t_stat = c->coeff[j]->estimate / std_err;
261 tab_double (t, 5, this_row, 0, t_stat, NULL);
263 P values for the test statistic above.
266 2 * gsl_cdf_tdist_Q (fabs (t_stat),
267 (double) (c->n_obs - c->n_coeffs));
268 tab_double (t, 6, this_row, 0, pval, NULL);
271 tab_title (t, _("Coefficients"));
276 Display the ANOVA table.
279 reg_stats_anova (pspp_linreg_cache * c)
283 const double msm = c->ssm / c->dfm;
284 const double mse = pspp_linreg_mse (c);
285 const double F = msm / mse;
286 const double pval = gsl_cdf_fdist_Q (F, c->dfm, c->dfe);
291 t = tab_create (n_cols, n_rows, 0);
292 tab_headers (t, 2, 0, 1, 0);
293 tab_dim (t, tab_natural_dimensions, NULL);
295 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
297 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
298 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
299 tab_vline (t, TAL_0, 1, 0, 0);
301 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
302 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
303 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
304 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
305 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
307 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Regression"));
308 tab_text (t, 1, 2, TAB_LEFT | TAT_TITLE, _("Residual"));
309 tab_text (t, 1, 3, TAB_LEFT | TAT_TITLE, _("Total"));
311 /* Sums of Squares */
312 tab_double (t, 2, 1, 0, c->ssm, NULL);
313 tab_double (t, 2, 3, 0, c->sst, NULL);
314 tab_double (t, 2, 2, 0, c->sse, NULL);
317 /* Degrees of freedom */
318 tab_text (t, 3, 1, TAB_RIGHT | TAT_PRINTF, "%g", c->dfm);
319 tab_text (t, 3, 2, TAB_RIGHT | TAT_PRINTF, "%g", c->dfe);
320 tab_text (t, 3, 3, TAB_RIGHT | TAT_PRINTF, "%g", c->dft);
323 tab_double (t, 4, 1, TAB_RIGHT, msm, NULL);
324 tab_double (t, 4, 2, TAB_RIGHT, mse, NULL);
326 tab_double (t, 5, 1, 0, F, NULL);
328 tab_double (t, 6, 1, 0, pval, NULL);
330 tab_title (t, _("ANOVA"));
335 reg_stats_outs (pspp_linreg_cache * c)
341 reg_stats_zpp (pspp_linreg_cache * c)
347 reg_stats_label (pspp_linreg_cache * c)
353 reg_stats_sha (pspp_linreg_cache * c)
358 reg_stats_ci (pspp_linreg_cache * c)
363 reg_stats_f (pspp_linreg_cache * c)
368 reg_stats_bcov (pspp_linreg_cache * c)
380 n_cols = c->n_indeps + 1 + 2;
381 n_rows = 2 * (c->n_indeps + 1);
382 t = tab_create (n_cols, n_rows, 0);
383 tab_headers (t, 2, 0, 1, 0);
384 tab_dim (t, tab_natural_dimensions, NULL);
385 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
386 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
387 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
388 tab_vline (t, TAL_0, 1, 0, 0);
389 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Model"));
390 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Covariances"));
391 for (i = 0; i < c->n_coeffs; i++)
393 const struct variable *v = pspp_coeff_get_var (c->coeff[i], 0);
394 label = var_to_string (v);
395 tab_text (t, 2, i, TAB_CENTER, label);
396 tab_text (t, i + 2, 0, TAB_CENTER, label);
397 for (k = 1; k < c->n_coeffs; k++)
399 col = (i <= k) ? k : i;
400 row = (i <= k) ? i : k;
401 tab_double (t, k + 2, i, TAB_CENTER,
402 gsl_matrix_get (c->cov, row, col), NULL);
405 tab_title (t, _("Coefficient Correlations"));
409 reg_stats_ses (pspp_linreg_cache * c)
414 reg_stats_xtx (pspp_linreg_cache * c)
419 reg_stats_collin (pspp_linreg_cache * c)
424 reg_stats_tol (pspp_linreg_cache * c)
429 reg_stats_selection (pspp_linreg_cache * c)
435 statistics_keyword_output (void (*function) (pspp_linreg_cache *),
436 int keyword, pspp_linreg_cache * c)
445 subcommand_statistics (int *keywords, pspp_linreg_cache * c)
448 The order here must match the order in which the STATISTICS
449 keywords appear in the specification section above.
476 Set everything but F.
478 for (i = 0; i < f; i++)
485 for (i = 0; i < all; i++)
493 Default output: ANOVA table, parameter estimates,
494 and statistics for variables not entered into model,
497 if (keywords[defaults] | d)
505 statistics_keyword_output (reg_stats_r, keywords[r], c);
506 statistics_keyword_output (reg_stats_anova, keywords[anova], c);
507 statistics_keyword_output (reg_stats_coeff, keywords[coeff], c);
508 statistics_keyword_output (reg_stats_outs, keywords[outs], c);
509 statistics_keyword_output (reg_stats_zpp, keywords[zpp], c);
510 statistics_keyword_output (reg_stats_label, keywords[label], c);
511 statistics_keyword_output (reg_stats_sha, keywords[sha], c);
512 statistics_keyword_output (reg_stats_ci, keywords[ci], c);
513 statistics_keyword_output (reg_stats_f, keywords[f], c);
514 statistics_keyword_output (reg_stats_bcov, keywords[bcov], c);
515 statistics_keyword_output (reg_stats_ses, keywords[ses], c);
516 statistics_keyword_output (reg_stats_xtx, keywords[xtx], c);
517 statistics_keyword_output (reg_stats_collin, keywords[collin], c);
518 statistics_keyword_output (reg_stats_tol, keywords[tol], c);
519 statistics_keyword_output (reg_stats_selection, keywords[selection], c);
523 Free the transformation. Free its linear model if this
524 transformation is the last one.
527 regression_trns_free (void *t_)
530 struct reg_trns *t = t_;
532 if (t->trns_id == t->n_trns)
534 result = pspp_linreg_cache_free (t->c);
542 Gets the predicted values.
545 regression_trns_pred_proc (void *t_, struct ccase **c,
546 casenumber case_idx UNUSED)
550 struct reg_trns *trns = t_;
551 pspp_linreg_cache *model;
552 union value *output = NULL;
553 const union value **vals = NULL;
554 const struct variable **vars = NULL;
556 assert (trns != NULL);
558 assert (model != NULL);
559 assert (model->depvar != NULL);
560 assert (model->pred != NULL);
562 vars = xnmalloc (model->n_coeffs, sizeof (*vars));
563 n_vals = (*model->get_vars) (model, vars);
565 vals = xnmalloc (n_vals, sizeof (*vals));
566 *c = case_unshare (*c);
567 output = case_data_rw (*c, model->pred);
569 for (i = 0; i < n_vals; i++)
571 vals[i] = case_data (*c, vars[i]);
573 output->f = (*model->predict) ((const struct variable **) vars,
574 vals, model, n_vals);
577 return TRNS_CONTINUE;
584 regression_trns_resid_proc (void *t_, struct ccase **c,
585 casenumber case_idx UNUSED)
589 struct reg_trns *trns = t_;
590 pspp_linreg_cache *model;
591 union value *output = NULL;
592 const union value **vals = NULL;
593 const union value *obs = NULL;
594 const struct variable **vars = NULL;
596 assert (trns != NULL);
598 assert (model != NULL);
599 assert (model->depvar != NULL);
600 assert (model->resid != NULL);
602 vars = xnmalloc (model->n_coeffs, sizeof (*vars));
603 n_vals = (*model->get_vars) (model, vars);
605 vals = xnmalloc (n_vals, sizeof (*vals));
606 *c = case_unshare (*c);
607 output = case_data_rw (*c, model->resid);
608 assert (output != NULL);
610 for (i = 0; i < n_vals; i++)
612 vals[i] = case_data (*c, vars[i]);
614 obs = case_data (*c, model->depvar);
615 output->f = (*model->residual) ((const struct variable **) vars,
616 vals, obs, model, n_vals);
619 return TRNS_CONTINUE;
623 Returns false if NAME is a duplicate of any existing variable name.
626 try_name (const struct dictionary *dict, const char *name)
628 if (dict_lookup_var (dict, name) != NULL)
635 reg_get_name (const struct dictionary *dict, char name[VAR_NAME_LEN],
636 const char prefix[VAR_NAME_LEN])
640 snprintf (name, VAR_NAME_LEN, "%s%d", prefix, i);
641 while (!try_name (dict, name))
644 snprintf (name, VAR_NAME_LEN, "%s%d", prefix, i);
649 reg_save_var (struct dataset *ds, const char *prefix, trns_proc_func * f,
650 pspp_linreg_cache * c, struct variable **v, int n_trns)
652 struct dictionary *dict = dataset_dict (ds);
653 static int trns_index = 1;
654 char name[VAR_NAME_LEN];
655 struct variable *new_var;
656 struct reg_trns *t = NULL;
658 t = xmalloc (sizeof (*t));
659 t->trns_id = trns_index;
662 reg_get_name (dict, name, prefix);
663 new_var = dict_create_var (dict, name, 0);
664 assert (new_var != NULL);
666 add_transformation (ds, f, regression_trns_free, t);
670 subcommand_save (struct dataset *ds, int save, pspp_linreg_cache ** models)
672 pspp_linreg_cache **lc;
676 assert (models != NULL);
680 /* Count the number of transformations we will need. */
681 for (i = 0; i < REGRESSION_SV_count; i++)
688 n_trns *= cmd.n_dependent;
690 for (lc = models; lc < models + cmd.n_dependent; lc++)
694 if ((*lc)->depvar != NULL)
696 if (cmd.a_save[REGRESSION_SV_RESID])
698 reg_save_var (ds, "RES", regression_trns_resid_proc, *lc,
699 &(*lc)->resid, n_trns);
701 if (cmd.a_save[REGRESSION_SV_PRED])
703 reg_save_var (ds, "PRED", regression_trns_pred_proc, *lc,
704 &(*lc)->pred, n_trns);
712 for (lc = models; lc < models + cmd.n_dependent; lc++)
716 pspp_linreg_cache_free (*lc);
723 cmd_regression (struct lexer *lexer, struct dataset *ds)
725 struct casegrouper *grouper;
726 struct casereader *group;
727 pspp_linreg_cache **models;
731 if (!parse_regression (lexer, ds, &cmd, NULL))
736 models = xnmalloc (cmd.n_dependent, sizeof *models);
737 for (i = 0; i < cmd.n_dependent; i++)
743 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
744 while (casegrouper_get_next_group (grouper, &group))
745 run_regression (group, &cmd, ds, models);
746 ok = casegrouper_destroy (grouper);
747 ok = proc_commit (ds) && ok;
749 subcommand_save (ds, cmd.sbc_save, models);
752 free_regression (&cmd);
754 return ok ? CMD_SUCCESS : CMD_FAILURE;
758 Is variable k the dependent variable?
761 is_depvar (size_t k, const struct variable *v)
763 return v == v_variables[k];
766 /* Parser for the variables sub command */
768 regression_custom_variables (struct lexer *lexer, struct dataset *ds,
769 struct cmd_regression *cmd UNUSED,
772 const struct dictionary *dict = dataset_dict (ds);
774 lex_match (lexer, '=');
776 if ((lex_token (lexer) != T_ID
777 || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
778 && lex_token (lexer) != T_ALL)
782 if (!parse_variables_const
783 (lexer, dict, &v_variables, &n_variables, PV_NONE))
788 assert (n_variables);
793 /* Identify the explanatory variables in v_variables. Returns
794 the number of independent variables. */
796 identify_indep_vars (const struct variable **indep_vars,
797 const struct variable *depvar)
799 int n_indep_vars = 0;
802 for (i = 0; i < n_variables; i++)
803 if (!is_depvar (i, depvar))
804 indep_vars[n_indep_vars++] = v_variables[i];
805 if ((n_indep_vars < 1) && is_depvar (0, depvar))
808 There is only one independent variable, and it is the same
809 as the dependent variable. Print a warning and continue.
812 gettext ("The dependent variable is equal to the independent variable."
813 "The least squares line is therefore Y=X."
814 "Standard errors and related statistics may be meaningless."));
816 indep_vars[0] = v_variables[0];
821 /* Encode categorical variables.
822 Returns number of valid cases. */
824 prepare_categories (struct casereader *input,
825 const struct variable **vars, size_t n_vars,
826 struct moments_var *mom)
832 assert (vars != NULL);
833 assert (mom != NULL);
835 for (i = 0; i < n_vars; i++)
836 if (var_is_alpha (vars[i]))
837 cat_stored_values_create (vars[i]);
840 for (; (c = casereader_read (input)) != NULL; case_unref (c))
843 The second condition ensures the program will run even if
844 there is only one variable to act as both explanatory and
847 for (i = 0; i < n_vars; i++)
849 const union value *val = case_data (c, vars[i]);
850 if (var_is_alpha (vars[i]))
851 cat_value_update (vars[i], val);
853 moments1_add (mom[i].m, val->f, 1.0);
857 casereader_destroy (input);
863 coeff_init (pspp_linreg_cache * c, struct design_matrix *dm)
865 c->coeff = xnmalloc (dm->m->size2, sizeof (*c->coeff));
866 pspp_coeff_init (c->coeff, dm);
870 run_regression (struct casereader *input, struct cmd_regression *cmd,
871 struct dataset *ds, pspp_linreg_cache **models)
877 const struct variable **indep_vars;
878 struct design_matrix *X;
879 struct moments_var *mom;
882 pspp_linreg_opts lopts;
884 assert (models != NULL);
886 c = casereader_peek (input, 0);
889 casereader_destroy (input);
892 output_split_file_values (ds, c);
897 dict_get_vars (dataset_dict (ds), &v_variables, &n_variables, 0);
900 for (i = 0; i < cmd->n_dependent; i++)
902 if (!var_is_numeric (cmd->v_dependent[i]))
904 msg (SE, _("Dependent variable must be numeric."));
909 mom = xnmalloc (n_variables, sizeof (*mom));
910 for (i = 0; i < n_variables; i++)
912 (mom + i)->m = moments1_create (MOMENT_VARIANCE);
913 (mom + i)->v = v_variables[i];
915 lopts.get_depvar_mean_std = 1;
917 lopts.get_indep_mean_std = xnmalloc (n_variables, sizeof (int));
918 indep_vars = xnmalloc (n_variables, sizeof *indep_vars);
920 for (k = 0; k < cmd->n_dependent; k++)
922 const struct variable *dep_var;
923 struct casereader *reader;
926 size_t n_data; /* Number of valid cases. */
928 dep_var = cmd->v_dependent[k];
929 n_indep = identify_indep_vars (indep_vars, dep_var);
930 reader = casereader_clone (input);
931 reader = casereader_create_filter_missing (reader, indep_vars, n_indep,
933 reader = casereader_create_filter_missing (reader, &dep_var, 1,
935 n_data = prepare_categories (casereader_clone (reader),
936 indep_vars, n_indep, mom);
938 if ((n_data > 0) && (n_indep > 0))
940 Y = gsl_vector_alloc (n_data);
942 design_matrix_create (n_indep,
943 (const struct variable **) indep_vars,
945 for (i = 0; i < X->m->size2; i++)
947 lopts.get_indep_mean_std[i] = 1;
949 models[k] = pspp_linreg_cache_alloc (dep_var, (const struct variable **) indep_vars,
950 X->m->size1, X->m->size2);
951 models[k]->depvar = dep_var;
953 For large data sets, use QR decomposition.
955 if (n_data > sqrt (n_indep) && n_data > REG_LARGE_DATA)
957 models[k]->method = PSPP_LINREG_QR;
961 The second pass fills the design matrix.
963 reader = casereader_create_counter (reader, &row, -1);
964 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
966 for (i = 0; i < n_indep; ++i)
968 const struct variable *v = indep_vars[i];
969 const union value *val = case_data (c, v);
970 if (var_is_alpha (v))
971 design_matrix_set_categorical (X, row, v, val);
973 design_matrix_set_numeric (X, row, v, val);
975 gsl_vector_set (Y, row, case_num (c, dep_var));
978 Now that we know the number of coefficients, allocate space
979 and store pointers to the variables that correspond to the
982 coeff_init (models[k], X);
985 Find the least-squares estimates and other statistics.
987 pspp_linreg ((const gsl_vector *) Y, X, &lopts, models[k]);
989 if (!taint_has_tainted_successor (casereader_get_taint (input)))
991 subcommand_statistics (cmd->a_statistics, models[k]);
995 design_matrix_destroy (X);
1000 gettext ("No valid data found. This command was skipped."));
1002 casereader_destroy (reader);
1004 for (i = 0; i < n_variables; i++)
1006 moments1_destroy ((mom + i)->m);
1010 free (lopts.get_indep_mean_std);
1011 casereader_destroy (input);