1 /* PSPP - linear regression.
2 Copyright (C) 2005 Free Software Foundation, Inc.
3 Written by Jason H Stover <jason@sakla.net>.
5 This program is free software; you can redistribute it and/or
6 modify it under the terms of the GNU General Public License as
7 published by the Free Software Foundation; either version 2 of the
8 License, or (at your option) any later version.
10 This program is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 #include <gsl/gsl_cdf.h>
23 #include <gsl/gsl_vector.h>
24 #include <gsl/gsl_matrix.h>
27 #include "dictionary.h"
28 #include "file-handle.h"
35 #include <linreg/pspp_linreg.h>
41 "REGRESSION" (regression_):
65 static struct cmd_regression cmd;
68 Array holding the subscripts of the independent variables.
72 static void run_regression (const struct casefile *, void *);
74 STATISTICS subcommand output functions.
76 static void reg_stats_r (pspp_linreg_cache *);
77 static void reg_stats_coeff (pspp_linreg_cache *);
78 static void reg_stats_anova (pspp_linreg_cache *);
79 static void reg_stats_outs (pspp_linreg_cache *);
80 static void reg_stats_zpp (pspp_linreg_cache *);
81 static void reg_stats_label (pspp_linreg_cache *);
82 static void reg_stats_sha (pspp_linreg_cache *);
83 static void reg_stats_ci (pspp_linreg_cache *);
84 static void reg_stats_f (pspp_linreg_cache *);
85 static void reg_stats_bcov (pspp_linreg_cache *);
86 static void reg_stats_ses (pspp_linreg_cache *);
87 static void reg_stats_xtx (pspp_linreg_cache *);
88 static void reg_stats_collin (pspp_linreg_cache *);
89 static void reg_stats_tol (pspp_linreg_cache *);
90 static void reg_stats_selection (pspp_linreg_cache *);
91 static void statistics_keyword_output (void (*)(pspp_linreg_cache *),
92 int, pspp_linreg_cache *);
95 reg_stats_r (pspp_linreg_cache * c)
105 rsq = c->ssm / c->sst;
106 adjrsq = 1.0 - (1.0 - rsq) * (c->n_obs - 1.0) / (c->n_obs - c->n_indeps);
107 std_error = sqrt ((c->n_indeps - 1.0) / (c->n_obs - 1.0));
108 t = tab_create (n_cols, n_rows, 0);
109 tab_dim (t, tab_natural_dimensions);
110 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
111 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
112 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
113 tab_vline (t, TAL_0, 1, 0, 0);
115 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("R"));
116 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("R Square"));
117 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Adjusted R Square"));
118 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error of the Estimate"));
119 tab_float (t, 1, 1, TAB_RIGHT, sqrt (rsq), 10, 2);
120 tab_float (t, 2, 1, TAB_RIGHT, rsq, 10, 2);
121 tab_float (t, 3, 1, TAB_RIGHT, adjrsq, 10, 2);
122 tab_float (t, 4, 1, TAB_RIGHT, std_error, 10, 2);
123 tab_title (t, 0, _("Model Summary"));
128 Table showing estimated regression coefficients.
131 reg_stats_coeff (pspp_linreg_cache * c)
146 n_rows = 2 + c->param_estimates->size;
147 t = tab_create (n_cols, n_rows, 0);
148 tab_headers (t, 2, 0, 1, 0);
149 tab_dim (t, tab_natural_dimensions);
150 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
151 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
152 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
153 tab_vline (t, TAL_0, 1, 0, 0);
155 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("B"));
156 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Error"));
157 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Beta"));
158 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("t"));
159 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
160 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("(Constant)"));
161 coeff = gsl_vector_get (c->param_estimates, 0);
162 tab_float (t, 2, 1, 0, coeff, 10, 2);
163 std_err = sqrt (gsl_matrix_get (c->cov, 0, 0));
164 tab_float (t, 3, 1, 0, std_err, 10, 2);
165 beta = coeff / c->depvar_std;
166 tab_float (t, 4, 1, 0, beta, 10, 2);
167 t_stat = coeff / std_err;
168 tab_float (t, 5, 1, 0, t_stat, 10, 2);
169 pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), 1.0);
170 tab_float (t, 6, 1, 0, pval, 10, 2);
171 for (j = 0; j < c->n_indeps; j++)
174 struct variable *v = cmd.v_variables[i];
175 label = var_to_string (v);
176 tab_text (t, 1, j + 2, TAB_CENTER, label);
178 Regression coefficients.
180 coeff = gsl_vector_get (c->param_estimates, j + 1);
181 tab_float (t, 2, j + 2, 0, coeff, 10, 2);
183 Standard error of the coefficients.
185 std_err = sqrt (gsl_matrix_get (c->cov, j + 1, j + 1));
186 tab_float (t, 3, j + 2, 0, std_err, 10, 2);
188 'Standardized' coefficient, i.e., regression coefficient
189 if all variables had unit variance.
191 beta = gsl_vector_get (c->indep_std, j + 1);
192 beta *= coeff / c->depvar_std;
193 tab_float (t, 4, j + 2, 0, beta, 10, 2);
196 Test statistic for H0: coefficient is 0.
198 t_stat = coeff / std_err;
199 tab_float (t, 5, j + 2, 0, t_stat, 10, 2);
201 P values for the test statistic above.
203 pval = 2 * gsl_cdf_tdist_Q (fabs (t_stat), 1.0);
204 tab_float (t, 6, j + 2, 0, pval, 10, 2);
206 tab_title (t, 0, _("Coefficients"));
211 Display the ANOVA table.
214 reg_stats_anova (pspp_linreg_cache * c)
218 const double msm = c->ssm / c->dfm;
219 const double mse = c->sse / c->dfe;
220 const double F = msm / mse;
221 const double pval = gsl_cdf_fdist_Q (F, c->dfm, c->dfe);
226 t = tab_create (n_cols, n_rows, 0);
227 tab_headers (t, 2, 0, 1, 0);
228 tab_dim (t, tab_natural_dimensions);
230 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
232 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
233 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
234 tab_vline (t, TAL_0, 1, 0, 0);
236 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
237 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
238 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
239 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
240 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
242 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Regression"));
243 tab_text (t, 1, 2, TAB_LEFT | TAT_TITLE, _("Residual"));
244 tab_text (t, 1, 3, TAB_LEFT | TAT_TITLE, _("Total"));
246 /* Sums of Squares */
247 tab_float (t, 2, 1, 0, c->ssm, 10, 2);
248 tab_float (t, 2, 3, 0, c->sst, 10, 2);
249 tab_float (t, 2, 2, 0, c->sse, 10, 2);
252 /* Degrees of freedom */
253 tab_float (t, 3, 1, 0, c->dfm, 4, 0);
254 tab_float (t, 3, 2, 0, c->dfe, 4, 0);
255 tab_float (t, 3, 3, 0, c->dft, 4, 0);
259 tab_float (t, 4, 1, TAB_RIGHT, msm, 8, 3);
260 tab_float (t, 4, 2, TAB_RIGHT, mse, 8, 3);
262 tab_float (t, 5, 1, 0, F, 8, 3);
264 tab_float (t, 6, 1, 0, pval, 8, 3);
266 tab_title (t, 0, _("ANOVA"));
270 reg_stats_outs (pspp_linreg_cache * c)
275 reg_stats_zpp (pspp_linreg_cache * c)
280 reg_stats_label (pspp_linreg_cache * c)
285 reg_stats_sha (pspp_linreg_cache * c)
290 reg_stats_ci (pspp_linreg_cache * c)
295 reg_stats_f (pspp_linreg_cache * c)
300 reg_stats_bcov (pspp_linreg_cache * c)
313 n_cols = c->n_indeps + 1 + 2;
314 n_rows = 2 * (c->n_indeps + 1);
315 t = tab_create (n_cols, n_rows, 0);
316 tab_headers (t, 2, 0, 1, 0);
317 tab_dim (t, tab_natural_dimensions);
318 tab_box (t, TAL_2, TAL_2, -1, TAL_1, 0, 0, n_cols - 1, n_rows - 1);
319 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
320 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
321 tab_vline (t, TAL_0, 1, 0, 0);
322 tab_text (t, 0, 0, TAB_CENTER | TAT_TITLE, _("Model"));
323 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Covariances"));
324 for (i = 1; i < c->n_indeps + 1; i++)
326 j = indep_vars[(i - 1)];
327 struct variable *v = cmd.v_variables[j];
328 label = var_to_string (v);
329 tab_text (t, 2, i, TAB_CENTER, label);
330 tab_text (t, i + 2, 0, TAB_CENTER, label);
331 for (k = 1; k < c->n_indeps + 1; k++)
333 col = (i <= k) ? k : i;
334 row = (i <= k) ? i : k;
335 tab_float (t, k + 2, i, TAB_CENTER,
336 gsl_matrix_get (c->cov, row, col), 8, 3);
339 tab_title (t, 0, _("Coefficient Correlations"));
343 reg_stats_ses (pspp_linreg_cache * c)
348 reg_stats_xtx (pspp_linreg_cache * c)
353 reg_stats_collin (pspp_linreg_cache * c)
358 reg_stats_tol (pspp_linreg_cache * c)
363 reg_stats_selection (pspp_linreg_cache * c)
369 statistics_keyword_output (void (*function) (pspp_linreg_cache *),
370 int keyword, pspp_linreg_cache * c)
379 subcommand_statistics (int *keywords, pspp_linreg_cache * c)
382 The order here must match the order in which the STATISTICS
383 keywords appear in the specification section above.
410 Set everything but F.
412 for (i = 0; i < f; i++)
419 for (i = 0; i < all; i++)
427 Default output: ANOVA table, parameter estimates,
428 and statistics for variables not entered into model,
431 if (keywords[defaults] | d)
439 statistics_keyword_output (reg_stats_r, keywords[r], c);
440 statistics_keyword_output (reg_stats_anova, keywords[anova], c);
441 statistics_keyword_output (reg_stats_coeff, keywords[coeff], c);
442 statistics_keyword_output (reg_stats_outs, keywords[outs], c);
443 statistics_keyword_output (reg_stats_zpp, keywords[zpp], c);
444 statistics_keyword_output (reg_stats_label, keywords[label], c);
445 statistics_keyword_output (reg_stats_sha, keywords[sha], c);
446 statistics_keyword_output (reg_stats_ci, keywords[ci], c);
447 statistics_keyword_output (reg_stats_f, keywords[f], c);
448 statistics_keyword_output (reg_stats_bcov, keywords[bcov], c);
449 statistics_keyword_output (reg_stats_ses, keywords[ses], c);
450 statistics_keyword_output (reg_stats_xtx, keywords[xtx], c);
451 statistics_keyword_output (reg_stats_collin, keywords[collin], c);
452 statistics_keyword_output (reg_stats_tol, keywords[tol], c);
453 statistics_keyword_output (reg_stats_selection, keywords[selection], c);
457 cmd_regression (void)
459 if (!parse_regression (&cmd))
463 multipass_procedure_with_splits (run_regression, &cmd);
469 Is variable k one of the dependent variables?
475 for (j = 0; j < cmd.n_dependent; j++)
478 compare_var_names returns 0 if the variable
481 if (!compare_var_names (cmd.v_dependent[j], cmd.v_variables[k], NULL))
488 run_regression (const struct casefile *cf, void *cmd_ UNUSED)
495 const union value *val;
496 struct casereader *r;
497 struct casereader *r2;
499 const struct variable *v;
500 struct recoded_categorical_array *ca;
501 struct recoded_categorical *rc;
502 struct design_matrix *X;
504 pspp_linreg_cache *lcache;
505 pspp_linreg_opts lopts;
507 n_data = casefile_get_case_cnt (cf);
508 n_indep = cmd.n_variables - cmd.n_dependent;
509 indep_vars = xnmalloc (n_indep, sizeof *indep_vars);
511 Y = gsl_vector_alloc (n_data);
512 lopts.get_depvar_mean_std = 1;
513 lopts.get_indep_mean_std = xnmalloc (n_indep, sizeof (int));
515 lcache = pspp_linreg_cache_alloc (n_data, n_indep);
516 lcache->indep_means = gsl_vector_alloc (n_indep);
517 lcache->indep_std = gsl_vector_alloc (n_indep);
520 Read from the active file. The first pass encodes categorical
523 ca = cr_recoded_cat_ar_create (cmd.n_variables, cmd.v_variables);
524 for (r = casefile_get_reader (cf);
525 casereader_read (r, &c); case_destroy (&c))
527 for (i = 0; i < ca->n_vars; i++)
529 v = (*(ca->a + i))->v;
530 val = case_data (&c, v->fv);
531 cr_value_update (*(ca->a + i), val);
534 cr_create_value_matrices (ca);
536 design_matrix_create (n_indep, (const struct variable **) cmd.v_variables,
540 The second pass creates the design matrix.
542 for (r2 = casefile_get_reader (cf); casereader_read (r2, &c);
544 /* Iterate over the cases. */
547 row = casereader_cnum (r2) - 1;
548 for (i = 0; i < cmd.n_variables; ++i) /* Iterate over the variables
549 for the current case.
552 v = cmd.v_variables[i];
553 val = case_data (&c, v->fv);
555 Independent/dependent variable separation. The
556 'variables' subcommand specifies a varlist which contains
557 both dependent and independent variables. The dependent
558 variables are specified with the 'dependent'
559 subcommand. We need to separate the two.
563 assert (v->type == NUMERIC);
564 gsl_vector_set (Y, row, val->f);
568 if (v->type == ALPHA)
570 rc = cr_var_to_recoded_categorical (v, ca);
571 design_matrix_set_categorical (X, row, v, val, rc);
573 else if (v->type == NUMERIC)
575 design_matrix_set_numeric (X, row, v, val);
580 lopts.get_indep_mean_std[i] = 1;
585 Find the least-squares estimates and other statistics.
587 pspp_linreg ((const gsl_vector *) Y, X->m, &lopts, lcache);
588 subcommand_statistics (cmd.a_statistics, lcache);
590 design_matrix_destroy (X);
591 pspp_linreg_cache_free (lcache);
592 free (lopts.get_indep_mean_std);
594 casereader_destroy (r);