1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2005, 2009, 2010, 2011, 2012, 2013, 2014 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/>. */
21 #include <gsl/gsl_cdf.h>
22 #include <gsl/gsl_matrix.h>
24 #include <data/dataset.h>
25 #include <data/casewriter.h>
27 #include "language/command.h"
28 #include "language/lexer/lexer.h"
29 #include "language/lexer/value-parser.h"
30 #include "language/lexer/variable-parser.h"
33 #include "data/casegrouper.h"
34 #include "data/casereader.h"
35 #include "data/dictionary.h"
37 #include "math/covariance.h"
38 #include "math/linreg.h"
39 #include "math/moments.h"
41 #include "libpspp/message.h"
42 #include "libpspp/taint.h"
44 #include "output/tab.h"
47 #define _(msgid) gettext (msgid)
48 #define N_(msgid) msgid
51 #include <gl/intprops.h>
53 #define REG_LARGE_DATA 1000
62 #define STATS_DEFAULT (STATS_R | STATS_COEFF | STATS_ANOVA | STATS_OUTS)
70 const struct variable **vars;
73 const struct variable **dep_vars;
83 struct regression_workspace
85 /* The new variables which will be introduced by /SAVE */
86 const struct variable **predvars;
87 const struct variable **residvars;
89 /* A reader/writer pair to temporarily hold the
90 values of the new variables */
91 struct casewriter *writer;
92 struct casereader *reader;
94 /* Indeces of the new values in the reader/writer (-1 if not applicable) */
98 /* 0, 1 or 2 depending on what new variables are to be created */
102 static void run_regression (const struct regression *cmd,
103 struct regression_workspace *ws,
104 struct casereader *input);
107 /* Return a string based on PREFIX which may be used as the name
108 of a new variable in DICT */
110 reg_get_name (const struct dictionary *dict, const char *prefix)
115 /* XXX handle too-long prefixes */
116 name = xmalloc (strlen (prefix) + INT_BUFSIZE_BOUND (i) + 1);
119 sprintf (name, "%s%d", prefix, i);
120 if (dict_lookup_var (dict, name) == NULL)
126 static const struct variable *
127 create_aux_var (struct dataset *ds, const char *prefix)
129 struct variable *var;
130 struct dictionary *dict = dataset_dict (ds);
131 char *name = reg_get_name (dict, prefix);
132 var = dict_create_var_assert (dict, name, 0);
137 /* Auxilliary data for transformation when /SAVE is entered */
138 struct save_trans_data
141 struct regression_workspace *ws;
145 save_trans_free (void *aux)
147 struct save_trans_data *save_trans_data = aux;
148 free (save_trans_data->ws->predvars);
149 free (save_trans_data->ws->residvars);
151 casereader_destroy (save_trans_data->ws->reader);
152 free (save_trans_data->ws);
153 free (save_trans_data);
158 save_trans_func (void *aux, struct ccase **c, casenumber x UNUSED)
160 struct save_trans_data *save_trans_data = aux;
161 struct regression_workspace *ws = save_trans_data->ws;
162 struct ccase *in = casereader_read (ws->reader);
167 *c = case_unshare (*c);
169 for (k = 0; k < save_trans_data->n_dep_vars; ++k)
171 if (ws->pred_idx != -1)
173 double pred = case_data_idx (in, ws->extras * k + ws->pred_idx)->f;
174 case_data_rw (*c, ws->predvars[k])->f = pred;
177 if (ws->res_idx != -1)
179 double resid = case_data_idx (in, ws->extras * k + ws->res_idx)->f;
180 case_data_rw (*c, ws->residvars[k])->f = resid;
186 return TRNS_CONTINUE;
191 cmd_regression (struct lexer *lexer, struct dataset *ds)
193 struct regression_workspace workspace;
194 struct regression regression;
195 const struct dictionary *dict = dataset_dict (ds);
198 memset (®ression, 0, sizeof (struct regression));
200 regression.ci = 0.95;
201 regression.stats = STATS_DEFAULT;
202 regression.pred = false;
203 regression.resid = false;
207 /* Accept an optional, completely pointless "/VARIABLES=" */
208 lex_match (lexer, T_SLASH);
209 if (lex_match_id (lexer, "VARIABLES"))
211 if (!lex_force_match (lexer, T_EQUALS))
215 if (!parse_variables_const (lexer, dict,
216 ®ression.vars, ®ression.n_vars,
217 PV_NO_DUPLICATE | PV_NUMERIC))
221 while (lex_token (lexer) != T_ENDCMD)
223 lex_match (lexer, T_SLASH);
225 if (lex_match_id (lexer, "DEPENDENT"))
227 if (!lex_force_match (lexer, T_EQUALS))
230 free (regression.dep_vars);
231 regression.n_dep_vars = 0;
233 if (!parse_variables_const (lexer, dict,
234 ®ression.dep_vars,
235 ®ression.n_dep_vars,
236 PV_NO_DUPLICATE | PV_NUMERIC))
239 else if (lex_match_id (lexer, "METHOD"))
241 lex_match (lexer, T_EQUALS);
243 if (!lex_force_match_id (lexer, "ENTER"))
248 else if (lex_match_id (lexer, "STATISTICS"))
250 lex_match (lexer, T_EQUALS);
252 while (lex_token (lexer) != T_ENDCMD
253 && lex_token (lexer) != T_SLASH)
255 if (lex_match (lexer, T_ALL))
257 regression.stats = ~0;
259 else if (lex_match_id (lexer, "DEFAULTS"))
261 regression.stats |= STATS_DEFAULT;
263 else if (lex_match_id (lexer, "R"))
265 regression.stats |= STATS_R;
267 else if (lex_match_id (lexer, "COEFF"))
269 regression.stats |= STATS_COEFF;
271 else if (lex_match_id (lexer, "ANOVA"))
273 regression.stats |= STATS_ANOVA;
275 else if (lex_match_id (lexer, "BCOV"))
277 regression.stats |= STATS_BCOV;
279 else if (lex_match_id (lexer, "CI"))
281 regression.stats |= STATS_CI;
283 if (lex_match (lexer, T_LPAREN))
285 regression.ci = lex_number (lexer) / 100.0;
287 lex_force_match (lexer, T_RPAREN);
292 lex_error (lexer, NULL);
297 else if (lex_match_id (lexer, "SAVE"))
299 lex_match (lexer, T_EQUALS);
301 while (lex_token (lexer) != T_ENDCMD
302 && lex_token (lexer) != T_SLASH)
304 if (lex_match_id (lexer, "PRED"))
306 regression.pred = true;
308 else if (lex_match_id (lexer, "RESID"))
310 regression.resid = true;
314 lex_error (lexer, NULL);
321 lex_error (lexer, NULL);
326 if (!regression.vars)
328 dict_get_vars (dict, ®ression.vars, ®ression.n_vars, 0);
331 save = regression.pred || regression.resid;
332 workspace.extras = 0;
333 workspace.res_idx = -1;
334 workspace.pred_idx = -1;
335 workspace.writer = NULL;
336 workspace.reader = NULL;
340 struct caseproto *proto = caseproto_create ();
342 if (regression.resid)
345 workspace.res_idx = 0;
346 workspace.residvars = xcalloc (regression.n_dep_vars, sizeof (*workspace.residvars));
348 for (i = 0; i < regression.n_dep_vars; ++i)
350 workspace.residvars[i] = create_aux_var (ds, "RES");
351 proto = caseproto_add_width (proto, 0);
358 workspace.pred_idx = 1;
359 workspace.predvars = xcalloc (regression.n_dep_vars, sizeof (*workspace.predvars));
361 for (i = 0; i < regression.n_dep_vars; ++i)
363 workspace.predvars[i] = create_aux_var (ds, "PRED");
364 proto = caseproto_add_width (proto, 0);
368 if (proc_make_temporary_transformations_permanent (ds))
369 msg (SW, _("REGRESSION with SAVE ignores TEMPORARY. "
370 "Temporary transformations will be made permanent."));
372 workspace.writer = autopaging_writer_create (proto);
373 caseproto_unref (proto);
378 struct casegrouper *grouper;
379 struct casereader *group;
382 grouper = casegrouper_create_splits (proc_open_filtering (ds, !save), dict);
385 while (casegrouper_get_next_group (grouper, &group))
387 run_regression (®ression,
392 ok = casegrouper_destroy (grouper);
393 ok = proc_commit (ds) && ok;
396 if (workspace.writer)
398 struct save_trans_data *save_trans_data = xmalloc (sizeof *save_trans_data);
399 struct casereader *r = casewriter_make_reader (workspace.writer);
400 workspace.writer = NULL;
401 workspace.reader = r;
402 save_trans_data->ws = xmalloc (sizeof (workspace));
403 memcpy (save_trans_data->ws, &workspace, sizeof (workspace));
404 save_trans_data->n_dep_vars = regression.n_dep_vars;
406 add_transformation (ds, save_trans_func, save_trans_free, save_trans_data);
410 free (regression.vars);
411 free (regression.dep_vars);
416 free (regression.vars);
417 free (regression.dep_vars);
421 /* Return the size of the union of dependent and independent variables */
423 get_n_all_vars (const struct regression *cmd)
425 size_t result = cmd->n_vars;
429 result += cmd->n_dep_vars;
430 for (i = 0; i < cmd->n_dep_vars; i++)
432 for (j = 0; j < cmd->n_vars; j++)
434 if (cmd->vars[j] == cmd->dep_vars[i])
443 /* Fill VARS with the union of dependent and independent variables */
445 fill_all_vars (const struct variable **vars, const struct regression *cmd)
449 for (i = 0; i < cmd->n_vars; i++)
451 vars[i] = cmd->vars[i];
454 for (i = 0; i < cmd->n_dep_vars; i++)
458 for (j = 0; j < cmd->n_vars; j++)
460 if (cmd->dep_vars[i] == cmd->vars[j])
468 vars[cmd->n_vars + x++] = cmd->dep_vars[i];
474 Is variable k the dependent variable?
477 is_depvar (const struct regression *cmd, size_t k, const struct variable *v)
479 return v == cmd->vars[k];
483 /* Identify the explanatory variables in v_variables. Returns
484 the number of independent variables. */
486 identify_indep_vars (const struct regression *cmd,
487 const struct variable **indep_vars,
488 const struct variable *depvar)
490 int n_indep_vars = 0;
493 for (i = 0; i < cmd->n_vars; i++)
494 if (!is_depvar (cmd, i, depvar))
495 indep_vars[n_indep_vars++] = cmd->vars[i];
496 if ((n_indep_vars < 1) && is_depvar (cmd, 0, depvar))
499 There is only one independent variable, and it is the same
500 as the dependent variable. Print a warning and continue.
504 ("The dependent variable is equal to the independent variable. "
505 "The least squares line is therefore Y=X. "
506 "Standard errors and related statistics may be meaningless."));
508 indep_vars[0] = cmd->vars[0];
515 fill_covariance (gsl_matrix * cov, struct covariance *all_cov,
516 const struct variable **vars,
517 size_t n_vars, const struct variable *dep_var,
518 const struct variable **all_vars, size_t n_all_vars,
523 size_t dep_subscript;
525 const gsl_matrix *ssizes;
526 const gsl_matrix *mean_matrix;
527 const gsl_matrix *ssize_matrix;
530 const gsl_matrix *cm = covariance_calculate_unnormalized (all_cov);
535 rows = xnmalloc (cov->size1 - 1, sizeof (*rows));
537 for (i = 0; i < n_all_vars; i++)
539 for (j = 0; j < n_vars; j++)
541 if (vars[j] == all_vars[i])
546 if (all_vars[i] == dep_var)
551 mean_matrix = covariance_moments (all_cov, MOMENT_MEAN);
552 ssize_matrix = covariance_moments (all_cov, MOMENT_NONE);
553 for (i = 0; i < cov->size1 - 1; i++)
555 means[i] = gsl_matrix_get (mean_matrix, rows[i], 0)
556 / gsl_matrix_get (ssize_matrix, rows[i], 0);
557 for (j = 0; j < cov->size2 - 1; j++)
559 gsl_matrix_set (cov, i, j, gsl_matrix_get (cm, rows[i], rows[j]));
560 gsl_matrix_set (cov, j, i, gsl_matrix_get (cm, rows[j], rows[i]));
563 means[cov->size1 - 1] = gsl_matrix_get (mean_matrix, dep_subscript, 0)
564 / gsl_matrix_get (ssize_matrix, dep_subscript, 0);
565 ssizes = covariance_moments (all_cov, MOMENT_NONE);
566 result = gsl_matrix_get (ssizes, dep_subscript, rows[0]);
567 for (i = 0; i < cov->size1 - 1; i++)
569 gsl_matrix_set (cov, i, cov->size1 - 1,
570 gsl_matrix_get (cm, rows[i], dep_subscript));
571 gsl_matrix_set (cov, cov->size1 - 1, i,
572 gsl_matrix_get (cm, rows[i], dep_subscript));
573 if (result > gsl_matrix_get (ssizes, rows[i], dep_subscript))
575 result = gsl_matrix_get (ssizes, rows[i], dep_subscript);
578 gsl_matrix_set (cov, cov->size1 - 1, cov->size1 - 1,
579 gsl_matrix_get (cm, dep_subscript, dep_subscript));
587 STATISTICS subcommand output functions.
589 static void reg_stats_r (const linreg *, const struct variable *);
590 static void reg_stats_coeff (const linreg *, const gsl_matrix *, const struct variable *, const struct regression *);
591 static void reg_stats_anova (const linreg *, const struct variable *);
592 static void reg_stats_bcov (const linreg *, const struct variable *);
596 subcommand_statistics (const struct regression *cmd, const linreg * c, const gsl_matrix * cm,
597 const struct variable *var)
599 if (cmd->stats & STATS_R)
600 reg_stats_r (c, var);
602 if (cmd->stats & STATS_ANOVA)
603 reg_stats_anova (c, var);
605 if (cmd->stats & STATS_COEFF)
606 reg_stats_coeff (c, cm, var, cmd);
608 if (cmd->stats & STATS_BCOV)
609 reg_stats_bcov (c, var);
614 run_regression (const struct regression *cmd,
615 struct regression_workspace *ws,
616 struct casereader *input)
623 struct covariance *cov;
624 struct casereader *reader;
625 size_t n_all_vars = get_n_all_vars (cmd);
626 const struct variable **all_vars = xnmalloc (n_all_vars, sizeof (*all_vars));
628 double *means = xnmalloc (n_all_vars, sizeof (*means));
630 fill_all_vars (all_vars, cmd);
631 cov = covariance_1pass_create (n_all_vars, all_vars,
632 dict_get_weight (dataset_dict (cmd->ds)),
635 reader = casereader_clone (input);
636 reader = casereader_create_filter_missing (reader, all_vars, n_all_vars,
641 struct casereader *r = casereader_clone (reader);
643 for (; (c = casereader_read (r)) != NULL; case_unref (c))
645 covariance_accumulate (cov, c);
647 casereader_destroy (r);
650 models = xcalloc (cmd->n_dep_vars, sizeof (*models));
651 for (k = 0; k < cmd->n_dep_vars; k++)
653 const struct variable **vars = xnmalloc (cmd->n_vars, sizeof (*vars));
654 const struct variable *dep_var = cmd->dep_vars[k];
655 int n_indep = identify_indep_vars (cmd, vars, dep_var);
656 gsl_matrix *this_cm = gsl_matrix_alloc (n_indep + 1, n_indep + 1);
657 double n_data = fill_covariance (this_cm, cov, vars, n_indep,
658 dep_var, all_vars, n_all_vars, means);
659 models[k] = linreg_alloc (dep_var, vars, n_data, n_indep);
660 models[k]->depvar = dep_var;
661 for (i = 0; i < n_indep; i++)
663 linreg_set_indep_variable_mean (models[k], i, means[i]);
665 linreg_set_depvar_mean (models[k], means[i]);
667 For large data sets, use QR decomposition.
669 if (n_data > sqrt (n_indep) && n_data > REG_LARGE_DATA)
671 models[k]->method = LINREG_QR;
677 Find the least-squares estimates and other statistics.
679 linreg_fit (this_cm, models[k]);
681 if (!taint_has_tainted_successor (casereader_get_taint (input)))
683 subcommand_statistics (cmd, models[k], this_cm, dep_var);
688 msg (SE, _("No valid data found. This command was skipped."));
690 gsl_matrix_free (this_cm);
697 struct casereader *r = casereader_clone (reader);
699 for (; (c = casereader_read (r)) != NULL; case_unref (c))
701 struct ccase *outc = case_clone (c);
702 for (k = 0; k < cmd->n_dep_vars; k++)
704 const struct variable **vars = xnmalloc (cmd->n_vars, sizeof (*vars));
705 const struct variable *dep_var = cmd->dep_vars[k];
706 int n_indep = identify_indep_vars (cmd, vars, dep_var);
707 double *vals = xnmalloc (n_indep, sizeof (*vals));
708 for (i = 0; i < n_indep; i++)
710 const union value *tmp = case_data (c, vars[i]);
716 double pred = linreg_predict (models[k], vals, n_indep);
717 case_data_rw_idx (outc, k * ws->extras + ws->pred_idx)->f = pred;
722 double obs = case_data (c, models[k]->depvar)->f;
723 double res = linreg_residual (models[k], obs, vals, n_indep);
724 case_data_rw_idx (outc, k * ws->extras + ws->res_idx)->f = res;
729 casewriter_write (ws->writer, outc);
731 casereader_destroy (r);
734 casereader_destroy (reader);
736 for (k = 0; k < cmd->n_dep_vars; k++)
738 linreg_unref (models[k]);
744 casereader_destroy (input);
745 covariance_destroy (cov);
752 reg_stats_r (const linreg * c, const struct variable *var)
762 rsq = linreg_ssreg (c) / linreg_sst (c);
764 (1.0 - rsq) * linreg_n_coeffs (c) / (linreg_n_obs (c) -
765 linreg_n_coeffs (c) - 1);
766 std_error = sqrt (linreg_mse (c));
767 t = tab_create (n_cols, n_rows);
768 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
769 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
770 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
771 tab_vline (t, TAL_0, 1, 0, 0);
773 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("R"));
774 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("R Square"));
775 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Adjusted R Square"));
776 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error of the Estimate"));
777 tab_double (t, 1, 1, TAB_RIGHT, sqrt (rsq), NULL);
778 tab_double (t, 2, 1, TAB_RIGHT, rsq, NULL);
779 tab_double (t, 3, 1, TAB_RIGHT, adjrsq, NULL);
780 tab_double (t, 4, 1, TAB_RIGHT, std_error, NULL);
781 tab_title (t, _("Model Summary (%s)"), var_to_string (var));
786 Table showing estimated regression coefficients.
789 reg_stats_coeff (const linreg * c, const gsl_matrix *cov, const struct variable *var, const struct regression *cmd)
793 const int heading_rows = 2;
802 const struct variable *v;
805 const double df = linreg_n_obs (c) - linreg_n_coeffs (c) - 1;
806 double q = (1 - cmd->ci) / 2.0; /* 2-tailed test */
807 double tval = gsl_cdf_tdist_Qinv (q, df);
810 n_rows = linreg_n_coeffs (c) + heading_rows + 1;
812 if (cmd->stats & STATS_CI)
815 t = tab_create (n_cols, n_rows);
816 tab_headers (t, 2, 0, 1, 0);
817 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
818 tab_hline (t, TAL_2, 0, n_cols - 1, heading_rows);
819 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
820 tab_vline (t, TAL_0, 1, 0, 0);
823 tab_hline (t, TAL_1, 2, 4, 1);
824 tab_joint_text (t, 2, 0, 3, 0, TAB_CENTER | TAT_TITLE, _("Unstandardized Coefficients"));
825 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("B"));
826 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Std. Error"));
827 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Standardized Coefficients"));
828 tab_text (t, 4, 1, TAB_CENTER | TAT_TITLE, _("Beta"));
829 tab_text (t, 5, 1, TAB_CENTER | TAT_TITLE, _("t"));
830 tab_text (t, 6, 1, TAB_CENTER | TAT_TITLE, _("Sig."));
831 tab_text (t, 1, heading_rows, TAB_LEFT | TAT_TITLE, _("(Constant)"));
832 tab_double (t, 2, heading_rows, 0, linreg_intercept (c), NULL);
833 std_err = sqrt (gsl_matrix_get (linreg_cov (c), 0, 0));
835 if (cmd->stats & STATS_CI)
837 double lower = linreg_intercept (c) - tval * std_err ;
838 double upper = linreg_intercept (c) + tval * std_err ;
839 tab_double (t, 7, heading_rows, 0, lower, NULL);
840 tab_double (t, 8, heading_rows, 0, upper, NULL);
842 tab_joint_text_format (t, 7, 0, 8, 0, TAB_CENTER | TAT_TITLE, _("%g%% Confidence Interval for B"), cmd->ci * 100);
843 tab_hline (t, TAL_1, 7, 8, 1);
844 tab_text (t, 7, 1, TAB_CENTER | TAT_TITLE, _("Lower Bound"));
845 tab_text (t, 8, 1, TAB_CENTER | TAT_TITLE, _("Upper Bound"));
847 tab_double (t, 3, heading_rows, 0, std_err, NULL);
848 tab_double (t, 4, heading_rows, 0, 0.0, NULL);
849 t_stat = linreg_intercept (c) / std_err;
850 tab_double (t, 5, heading_rows, 0, t_stat, NULL);
852 2 * gsl_cdf_tdist_Q (fabs (t_stat),
853 (double) (linreg_n_obs (c) - linreg_n_coeffs (c)));
854 tab_double (t, 6, heading_rows, 0, pval, NULL);
856 for (j = 0; j < linreg_n_coeffs (c); j++)
859 ds_init_empty (&tstr);
860 this_row = j + heading_rows + 1;
862 v = linreg_indep_var (c, j);
863 label = var_to_string (v);
864 /* Do not overwrite the variable's name. */
865 ds_put_cstr (&tstr, label);
866 tab_text (t, 1, this_row, TAB_LEFT, ds_cstr (&tstr));
868 Regression coefficients.
870 tab_double (t, 2, this_row, 0, linreg_coeff (c, j), NULL);
872 Standard error of the coefficients.
874 std_err = sqrt (gsl_matrix_get (linreg_cov (c), j + 1, j + 1));
875 tab_double (t, 3, this_row, 0, std_err, NULL);
877 Standardized coefficient, i.e., regression coefficient
878 if all variables had unit variance.
880 beta = sqrt (gsl_matrix_get (cov, j, j));
881 beta *= linreg_coeff (c, j) /
882 sqrt (gsl_matrix_get (cov, cov->size1 - 1, cov->size2 - 1));
883 tab_double (t, 4, this_row, 0, beta, NULL);
886 Test statistic for H0: coefficient is 0.
888 t_stat = linreg_coeff (c, j) / std_err;
889 tab_double (t, 5, this_row, 0, t_stat, NULL);
891 P values for the test statistic above.
893 pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), df);
894 tab_double (t, 6, this_row, 0, pval, NULL);
897 if (cmd->stats & STATS_CI)
899 double lower = linreg_coeff (c, j) - tval * std_err ;
900 double upper = linreg_coeff (c, j) + tval * std_err ;
902 tab_double (t, 7, this_row, 0, lower, NULL);
903 tab_double (t, 8, this_row, 0, upper, NULL);
906 tab_title (t, _("Coefficients (%s)"), var_to_string (var));
911 Display the ANOVA table.
914 reg_stats_anova (const linreg * c, const struct variable *var)
918 const double msm = linreg_ssreg (c) / linreg_dfmodel (c);
919 const double mse = linreg_mse (c);
920 const double F = msm / mse;
921 const double pval = gsl_cdf_fdist_Q (F, c->dfm, c->dfe);
926 t = tab_create (n_cols, n_rows);
927 tab_headers (t, 2, 0, 1, 0);
929 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
931 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
932 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
933 tab_vline (t, TAL_0, 1, 0, 0);
935 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
936 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
937 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
938 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
939 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Sig."));
941 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Regression"));
942 tab_text (t, 1, 2, TAB_LEFT | TAT_TITLE, _("Residual"));
943 tab_text (t, 1, 3, TAB_LEFT | TAT_TITLE, _("Total"));
945 /* Sums of Squares */
946 tab_double (t, 2, 1, 0, linreg_ssreg (c), NULL);
947 tab_double (t, 2, 3, 0, linreg_sst (c), NULL);
948 tab_double (t, 2, 2, 0, linreg_sse (c), NULL);
951 /* Degrees of freedom */
952 tab_text_format (t, 3, 1, TAB_RIGHT, "%g", c->dfm);
953 tab_text_format (t, 3, 2, TAB_RIGHT, "%g", c->dfe);
954 tab_text_format (t, 3, 3, TAB_RIGHT, "%g", c->dft);
957 tab_double (t, 4, 1, TAB_RIGHT, msm, NULL);
958 tab_double (t, 4, 2, TAB_RIGHT, mse, NULL);
960 tab_double (t, 5, 1, 0, F, NULL);
962 tab_double (t, 6, 1, 0, pval, NULL);
964 tab_title (t, _("ANOVA (%s)"), var_to_string (var));
970 reg_stats_bcov (const linreg * c, const struct variable *var)
982 n_cols = c->n_indeps + 1 + 2;
983 n_rows = 2 * (c->n_indeps + 1);
984 t = tab_create (n_cols, n_rows);
985 tab_headers (t, 2, 0, 1, 0);
986 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
987 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
988 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
989 tab_vline (t, TAL_0, 1, 0, 0);
990 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Model"));
991 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Covariances"));
992 for (i = 0; i < linreg_n_coeffs (c); i++)
994 const struct variable *v = linreg_indep_var (c, i);
995 label = var_to_string (v);
996 tab_text (t, 2, i, TAB_CENTER, label);
997 tab_text (t, i + 2, 0, TAB_CENTER, label);
998 for (k = 1; k < linreg_n_coeffs (c); k++)
1000 col = (i <= k) ? k : i;
1001 row = (i <= k) ? i : k;
1002 tab_double (t, k + 2, i, TAB_CENTER,
1003 gsl_matrix_get (c->cov, row, col), NULL);
1006 tab_title (t, _("Coefficient Correlations (%s)"), var_to_string (var));