1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2007, 2009, 2010 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 <data/case.h>
20 #include <data/casegrouper.h>
21 #include <data/casereader.h>
23 #include <math/covariance.h>
24 #include <math/categoricals.h>
25 #include <math/moments.h>
26 #include <gsl/gsl_matrix.h>
27 #include <linreg/sweep.h>
29 #include <libpspp/ll.h>
31 #include <language/lexer/lexer.h>
32 #include <language/lexer/variable-parser.h>
33 #include <language/lexer/value-parser.h>
34 #include <language/command.h>
36 #include <data/procedure.h>
37 #include <data/value.h>
38 #include <data/dictionary.h>
40 #include <language/dictionary/split-file.h>
41 #include <libpspp/hash.h>
42 #include <libpspp/taint.h>
43 #include <math/group-proc.h>
44 #include <math/levene.h>
45 #include <libpspp/misc.h>
47 #include <output/tab.h>
49 #include <gsl/gsl_cdf.h>
51 #include <data/format.h>
53 #include <libpspp/message.h>
56 #define _(msgid) gettext (msgid)
66 STATS_DESCRIPTIVES = 0x0001,
67 STATS_HOMOGENEITY = 0x0002
80 struct ll_list coefficient_list;
82 bool bad_count; /* True if the number of coefficients does not equal the number of groups */
88 const struct variable **vars;
90 const struct variable *indep_var;
92 enum statistics stats;
94 enum missing_type missing_type;
95 enum mv_class exclude;
97 /* List of contrasts */
98 struct ll_list contrast_list;
100 /* The weight variable */
101 const struct variable *wv;
105 /* Per category data */
106 struct descriptive_data
108 const struct variable *var;
109 struct moments1 *mom;
115 /* Workspace variable for each dependent variable */
118 struct covariance *cov;
129 struct oneway_workspace
131 /* The number of distinct values of the independent variable, when all
132 missing values are disregarded */
133 int actual_number_of_groups;
135 /* A hash table containing all the distinct values of the independent
137 struct hsh_table *group_hash;
139 struct per_var_ws *vws;
141 /* An array of descriptive data. One for each dependent variable */
142 struct descriptive_data **dd_total;
145 /* Routines to show the output tables */
146 static void show_anova_table (const struct oneway_spec *, const struct oneway_workspace *);
147 static void show_descriptives (const struct oneway_spec *, const struct oneway_workspace *);
148 static void show_homogeneity (const struct oneway_spec *, const struct oneway_workspace *);
150 static void output_oneway (const struct oneway_spec *, struct oneway_workspace *ws);
151 static void run_oneway (const struct oneway_spec *cmd, struct casereader *input, const struct dataset *ds);
154 cmd_oneway (struct lexer *lexer, struct dataset *ds)
156 const struct dictionary *dict = dataset_dict (ds);
157 struct oneway_spec oneway ;
160 oneway.indep_var = NULL;
162 oneway.missing_type = MISS_ANALYSIS;
163 oneway.exclude = MV_ANY;
164 oneway.wv = dict_get_weight (dict);
166 ll_init (&oneway.contrast_list);
169 if ( lex_match (lexer, '/'))
171 if (!lex_force_match_id (lexer, "VARIABLES"))
175 lex_match (lexer, '=');
178 if (!parse_variables_const (lexer, dict,
179 &oneway.vars, &oneway.n_vars,
180 PV_NO_DUPLICATE | PV_NUMERIC))
183 lex_force_match (lexer, T_BY);
185 oneway.indep_var = parse_variable_const (lexer, dict);
187 while (lex_token (lexer) != '.')
189 lex_match (lexer, '/');
191 if (lex_match_id (lexer, "STATISTICS"))
193 lex_match (lexer, '=');
194 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
196 if (lex_match_id (lexer, "DESCRIPTIVES"))
198 oneway.stats |= STATS_DESCRIPTIVES;
200 else if (lex_match_id (lexer, "HOMOGENEITY"))
202 oneway.stats |= STATS_HOMOGENEITY;
206 lex_error (lexer, NULL);
211 else if (lex_match_id (lexer, "CONTRAST"))
213 struct contrasts_node *cl = xzalloc (sizeof *cl);
215 struct ll_list *coefficient_list = &cl->coefficient_list;
216 lex_match (lexer, '=');
218 ll_init (coefficient_list);
220 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
222 if ( lex_is_number (lexer))
224 struct coeff_node *cc = xmalloc (sizeof *cc);
225 cc->coeff = lex_number (lexer);
227 ll_push_tail (coefficient_list, &cc->ll);
232 lex_error (lexer, NULL);
237 ll_push_tail (&oneway.contrast_list, &cl->ll);
239 else if (lex_match_id (lexer, "MISSING"))
241 lex_match (lexer, '=');
242 while (lex_token (lexer) != '.' && lex_token (lexer) != '/')
244 if (lex_match_id (lexer, "INCLUDE"))
246 oneway.exclude = MV_SYSTEM;
248 else if (lex_match_id (lexer, "EXCLUDE"))
250 oneway.exclude = MV_ANY;
252 else if (lex_match_id (lexer, "LISTWISE"))
254 oneway.missing_type = MISS_LISTWISE;
256 else if (lex_match_id (lexer, "ANALYSIS"))
258 oneway.missing_type = MISS_ANALYSIS;
262 lex_error (lexer, NULL);
269 lex_error (lexer, NULL);
276 struct casegrouper *grouper;
277 struct casereader *group;
280 grouper = casegrouper_create_splits (proc_open (ds), dict);
281 while (casegrouper_get_next_group (grouper, &group))
282 run_oneway (&oneway, group, ds);
283 ok = casegrouper_destroy (grouper);
284 ok = proc_commit (ds) && ok;
299 compare_double_3way (const void *a_, const void *b_, const void *aux UNUSED)
301 const double *a = a_;
302 const double *b = b_;
303 return *a < *b ? -1 : *a > *b;
307 do_hash_double (const void *value_, const void *aux UNUSED)
309 const double *value = value_;
310 return hash_double (*value, 0);
314 free_double (void *value_, const void *aux UNUSED)
316 double *value = value_;
322 static void postcalc (const struct oneway_spec *cmd);
323 static void precalc (const struct oneway_spec *cmd);
325 static struct descriptive_data *
326 dd_create (const struct variable *var)
328 struct descriptive_data *dd = xmalloc (sizeof *dd);
330 dd->mom = moments1_create (MOMENT_VARIANCE);
331 dd->minimum = DBL_MAX;
332 dd->maximum = -DBL_MAX;
340 makeit (void *aux1, void *aux2 UNUSED)
342 const struct variable *var = aux1;
344 struct descriptive_data *dd = dd_create (var);
350 updateit (void *user_data,
351 enum mv_class exclude,
352 const struct variable *wv,
353 const struct variable *catvar UNUSED,
354 const struct ccase *c,
355 void *aux1, void *aux2)
357 struct descriptive_data *dd = user_data;
359 const struct variable *varp = aux1;
361 const union value *valx = case_data (c, varp);
363 if ( var_is_value_missing (varp, valx, exclude))
366 struct descriptive_data *dd_total = aux2;
370 weight = case_data (c, wv)->f;
372 moments1_add (dd->mom, valx->f, weight);
373 if (valx->f * weight < dd->minimum)
374 dd->minimum = valx->f * weight;
376 if (valx->f * weight > dd->maximum)
377 dd->maximum = valx->f * weight;
380 const struct variable *var = dd_total->var;
381 const union value *val = case_data (c, var);
383 moments1_add (dd_total->mom,
387 if (val->f * weight < dd_total->minimum)
388 dd_total->minimum = val->f * weight;
390 if (val->f * weight > dd_total->maximum)
391 dd_total->maximum = val->f * weight;
396 run_oneway (const struct oneway_spec *cmd,
397 struct casereader *input,
398 const struct dataset *ds)
402 struct dictionary *dict = dataset_dict (ds);
403 struct casereader *reader;
406 struct oneway_workspace ws;
408 ws.actual_number_of_groups = 0;
409 ws.vws = xmalloc (cmd->n_vars * sizeof (*ws.vws));
410 ws.dd_total = xmalloc (sizeof (struct descriptive_data) * cmd->n_vars);
412 for (v = 0 ; v < cmd->n_vars; ++v)
413 ws.dd_total[v] = dd_create (cmd->vars[v]);
415 for (v = 0; v < cmd->n_vars; ++v)
417 struct categoricals *cats = categoricals_create (&cmd->indep_var, 1,
418 cmd->wv, cmd->exclude,
421 cmd->vars[v], ws.dd_total[v]);
423 ws.vws[v].cov = covariance_2pass_create (1, &cmd->vars[v],
425 cmd->wv, cmd->exclude);
428 c = casereader_peek (input, 0);
431 casereader_destroy (input);
434 output_split_file_values (ds, c);
437 taint = taint_clone (casereader_get_taint (input));
439 ws.group_hash = hsh_create (4,
447 input = casereader_create_filter_missing (input, &cmd->indep_var, 1,
448 cmd->exclude, NULL, NULL);
449 if (cmd->missing_type == MISS_LISTWISE)
450 input = casereader_create_filter_missing (input, cmd->vars, cmd->n_vars,
451 cmd->exclude, NULL, NULL);
452 input = casereader_create_filter_weight (input, dict, NULL, NULL);
454 reader = casereader_clone (input);
456 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
460 const double weight = dict_get_case_weight (dict, c, NULL);
462 const union value *indep_val = case_data (c, cmd->indep_var);
463 void **p = hsh_probe (ws.group_hash, &indep_val->f);
466 double *value = *p = xmalloc (sizeof *value);
467 *value = indep_val->f;
470 for (i = 0; i < cmd->n_vars; ++i)
472 const struct variable *v = cmd->vars[i];
473 const union value *val = case_data (c, v);
475 if ( MISS_ANALYSIS == cmd->missing_type)
477 if ( var_is_value_missing (v, val, cmd->exclude))
482 struct per_var_ws *pvw = &ws.vws[i];
484 covariance_accumulate_pass1 (pvw->cov, c);
487 struct group_proc *gp = group_proc_get (cmd->vars[i]);
488 struct hsh_table *group_hash = gp->group_hash;
490 struct group_statistics *gs;
492 gs = hsh_find (group_hash, indep_val );
496 gs = xmalloc (sizeof *gs);
502 gs->minimum = DBL_MAX;
503 gs->maximum = -DBL_MAX;
505 hsh_insert ( group_hash, gs );
508 if (!var_is_value_missing (v, val, cmd->exclude))
510 struct group_statistics *totals = &gp->ugs;
513 totals->sum += weight * val->f;
514 totals->ssq += weight * pow2 (val->f);
516 if ( val->f * weight < totals->minimum )
517 totals->minimum = val->f * weight;
519 if ( val->f * weight > totals->maximum )
520 totals->maximum = val->f * weight;
523 gs->sum += weight * val->f;
524 gs->ssq += weight * pow2 (val->f);
526 if ( val->f * weight < gs->minimum )
527 gs->minimum = val->f * weight;
529 if ( val->f * weight > gs->maximum )
530 gs->maximum = val->f * weight;
533 gp->n_groups = hsh_count (group_hash );
537 casereader_destroy (reader);
538 reader = casereader_clone (input);
539 for ( ; (c = casereader_read (reader) ); case_unref (c))
542 for (i = 0; i < cmd->n_vars; ++i)
544 struct per_var_ws *pvw = &ws.vws[i];
545 const struct variable *v = cmd->vars[i];
546 const union value *val = case_data (c, v);
548 if ( MISS_ANALYSIS == cmd->missing_type)
550 if ( var_is_value_missing (v, val, cmd->exclude))
554 covariance_accumulate_pass2 (pvw->cov, c);
557 casereader_destroy (reader);
559 for (v = 0; v < cmd->n_vars; ++v)
561 struct per_var_ws *pvw = &ws.vws[v];
562 gsl_matrix *cm = covariance_calculate_unnormalized (pvw->cov);
563 const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
566 moments1_calculate (ws.dd_total[v]->mom, &n, NULL, NULL, NULL, NULL);
568 pvw->sst = gsl_matrix_get (cm, 0, 0);
570 // gsl_matrix_fprintf (stdout, cm, "%g ");
574 pvw->sse = gsl_matrix_get (cm, 0, 0);
576 pvw->ssa = pvw->sst - pvw->sse;
578 pvw->n_groups = categoricals_total (cats);
580 pvw->mse = (pvw->sst - pvw->ssa) / (n - pvw->n_groups);
586 for (v = 0; v < cmd->n_vars; ++v)
588 struct categoricals *cats = covariance_get_categoricals (ws.vws[v].cov);
590 categoricals_done (cats);
592 if (categoricals_total (cats) > ws.actual_number_of_groups)
593 ws.actual_number_of_groups = categoricals_total (cats);
596 if ( cmd->stats & STATS_HOMOGENEITY )
597 levene (dict, casereader_clone (input), cmd->indep_var,
598 cmd->n_vars, cmd->vars, cmd->exclude);
600 casereader_destroy (input);
602 if (!taint_has_tainted_successor (taint))
603 output_oneway (cmd, &ws);
605 taint_destroy (taint);
608 /* Pre calculations */
610 precalc (const struct oneway_spec *cmd)
614 for (i = 0; i < cmd->n_vars; ++i)
616 struct group_proc *gp = group_proc_get (cmd->vars[i]);
617 struct group_statistics *totals = &gp->ugs;
619 /* Create a hash for each of the dependent variables.
620 The hash contains a group_statistics structure,
621 and is keyed by value of the independent variable */
623 gp->group_hash = hsh_create (4, compare_group, hash_group,
624 (hsh_free_func *) free_group,
630 totals->sum_diff = 0;
631 totals->maximum = -DBL_MAX;
632 totals->minimum = DBL_MAX;
636 /* Post calculations for the ONEWAY command */
638 postcalc (const struct oneway_spec *cmd)
642 for (i = 0; i < cmd->n_vars; ++i)
644 struct group_proc *gp = group_proc_get (cmd->vars[i]);
645 struct hsh_table *group_hash = gp->group_hash;
646 struct group_statistics *totals = &gp->ugs;
648 struct hsh_iterator g;
649 struct group_statistics *gs;
651 for (gs = hsh_first (group_hash, &g);
653 gs = hsh_next (group_hash, &g))
655 gs->mean = gs->sum / gs->n;
656 gs->s_std_dev = sqrt (gs->ssq / gs->n - pow2 (gs->mean));
659 gs->n / (gs->n - 1) *
660 ( gs->ssq / gs->n - pow2 (gs->mean))
663 gs->se_mean = gs->std_dev / sqrt (gs->n);
664 gs->mean_diff = gs->sum_diff / gs->n;
667 totals->mean = totals->sum / totals->n;
668 totals->std_dev = sqrt (
669 totals->n / (totals->n - 1) *
670 (totals->ssq / totals->n - pow2 (totals->mean))
673 totals->se_mean = totals->std_dev / sqrt (totals->n);
677 static void show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspace *ws);
678 static void show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspace *ws);
681 output_oneway (const struct oneway_spec *cmd, struct oneway_workspace *ws)
685 /* Check the sanity of the given contrast values */
686 struct contrasts_node *coeff_list = NULL;
687 ll_for_each (coeff_list, struct contrasts_node, ll, &cmd->contrast_list)
689 struct coeff_node *cn = NULL;
691 struct ll_list *cl = &coeff_list->coefficient_list;
694 if (ll_count (cl) != ws->actual_number_of_groups)
697 _("Number of contrast coefficients must equal the number of groups"));
698 coeff_list->bad_count = true;
702 ll_for_each (cn, struct coeff_node, ll, cl)
706 msg (SW, _("Coefficients for contrast %zu do not total zero"), i);
709 if (cmd->stats & STATS_DESCRIPTIVES)
710 show_descriptives (cmd, ws);
712 if (cmd->stats & STATS_HOMOGENEITY)
713 show_homogeneity (cmd, ws);
715 show_anova_table (cmd, ws);
717 if (ll_count (&cmd->contrast_list) > 0)
719 show_contrast_coeffs (cmd, ws);
720 show_contrast_tests (cmd, ws);
724 for (i = 0; i < cmd->n_vars; ++i )
726 struct hsh_table *group_hash = group_proc_get (cmd->vars[i])->group_hash;
728 hsh_destroy (group_hash);
731 hsh_destroy (ws->group_hash);
735 /* Show the ANOVA table */
737 show_anova_table (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
741 size_t n_rows = cmd->n_vars * 3 + 1;
743 struct tab_table *t = tab_create (n_cols, n_rows);
745 tab_headers (t, 2, 0, 1, 0);
751 n_cols - 1, n_rows - 1);
753 tab_hline (t, TAL_2, 0, n_cols - 1, 1 );
754 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
755 tab_vline (t, TAL_0, 1, 0, 0);
757 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
758 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
759 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
760 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
761 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
764 for (i = 0; i < cmd->n_vars; ++i)
767 moments1_calculate (ws->dd_total[i]->mom, &n, NULL, NULL, NULL, NULL);
769 const struct per_var_ws *pvw = &ws->vws[i];
770 const double df1 = pvw->n_groups - 1;
771 const double df2 = n - pvw->n_groups;
772 const double msa = pvw->ssa / df1;
774 const char *s = var_to_string (cmd->vars[i]);
776 tab_text (t, 0, i * 3 + 1, TAB_LEFT | TAT_TITLE, s);
777 tab_text (t, 1, i * 3 + 1, TAB_LEFT | TAT_TITLE, _("Between Groups"));
778 tab_text (t, 1, i * 3 + 2, TAB_LEFT | TAT_TITLE, _("Within Groups"));
779 tab_text (t, 1, i * 3 + 3, TAB_LEFT | TAT_TITLE, _("Total"));
782 tab_hline (t, TAL_1, 0, n_cols - 1, i * 3 + 1);
785 /* Sums of Squares */
786 tab_double (t, 2, i * 3 + 1, 0, pvw->ssa, NULL);
787 tab_double (t, 2, i * 3 + 3, 0, pvw->sst, NULL);
788 tab_double (t, 2, i * 3 + 2, 0, pvw->sse, NULL);
791 /* Degrees of freedom */
792 tab_fixed (t, 3, i * 3 + 1, 0, df1, 4, 0);
793 tab_fixed (t, 3, i * 3 + 2, 0, df2, 4, 0);
794 tab_fixed (t, 3, i * 3 + 3, 0, n - 1, 4, 0);
797 tab_double (t, 4, i * 3 + 1, TAB_RIGHT, msa, NULL);
798 tab_double (t, 4, i * 3 + 2, TAB_RIGHT, pvw->mse, NULL);
801 const double F = msa / pvw->mse ;
804 tab_double (t, 5, i * 3 + 1, 0, F, NULL);
806 /* The significance */
807 tab_double (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q (F, df1, df2), NULL);
811 tab_title (t, _("ANOVA"));
816 /* Show the descriptives table */
818 show_descriptives (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
825 const double confidence = 0.95;
826 const double q = (1.0 - confidence) / 2.0;
828 const struct fmt_spec *wfmt = cmd->wv ? var_get_print_format (cmd->wv) : &F_8_0;
832 for (v = 0; v < cmd->n_vars; ++v)
833 n_rows += ws->actual_number_of_groups + 1;
835 t = tab_create (n_cols, n_rows);
836 tab_headers (t, 2, 0, 2, 0);
838 /* Put a frame around the entire box, and vertical lines inside */
843 n_cols - 1, n_rows - 1);
845 /* Underline headers */
846 tab_hline (t, TAL_2, 0, n_cols - 1, 2);
847 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
849 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("N"));
850 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Mean"));
851 tab_text (t, 4, 1, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
852 tab_text (t, 5, 1, TAB_CENTER | TAT_TITLE, _("Std. Error"));
855 tab_vline (t, TAL_0, 7, 0, 0);
856 tab_hline (t, TAL_1, 6, 7, 1);
857 tab_joint_text_format (t, 6, 0, 7, 0, TAB_CENTER | TAT_TITLE,
858 _("%g%% Confidence Interval for Mean"),
861 tab_text (t, 6, 1, TAB_CENTER | TAT_TITLE, _("Lower Bound"));
862 tab_text (t, 7, 1, TAB_CENTER | TAT_TITLE, _("Upper Bound"));
864 tab_text (t, 8, 1, TAB_CENTER | TAT_TITLE, _("Minimum"));
865 tab_text (t, 9, 1, TAB_CENTER | TAT_TITLE, _("Maximum"));
867 tab_title (t, _("Descriptives"));
870 for (v = 0; v < cmd->n_vars; ++v)
872 const char *s = var_to_string (cmd->vars[v]);
873 const struct fmt_spec *fmt = var_get_print_format (cmd->vars[v]);
877 struct per_var_ws *pvw = &ws->vws[v];
878 const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
880 tab_text (t, 0, row, TAB_LEFT | TAT_TITLE, s);
882 tab_hline (t, TAL_1, 0, n_cols - 1, row);
884 for (count = 0; count < categoricals_total (cats); ++count)
887 double n, mean, variance;
888 double std_dev, std_error ;
892 const union value *gval = categoricals_get_value_by_subscript (cats, count);
893 const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, count);
895 moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL);
897 std_dev = sqrt (variance);
898 std_error = std_dev / sqrt (n) ;
900 ds_init_empty (&vstr);
902 var_append_value_name (cmd->indep_var, gval, &vstr);
904 tab_text (t, 1, row + count,
905 TAB_LEFT | TAT_TITLE,
910 /* Now fill in the numbers ... */
912 tab_fixed (t, 2, row + count, 0, n, 8, 0);
914 tab_double (t, 3, row + count, 0, mean, NULL);
916 tab_double (t, 4, row + count, 0, std_dev, NULL);
919 tab_double (t, 5, row + count, 0, std_error, NULL);
921 /* Now the confidence interval */
923 T = gsl_cdf_tdist_Qinv (q, n - 1);
925 tab_double (t, 6, row + count, 0,
926 mean - T * std_error, NULL);
928 tab_double (t, 7, row + count, 0,
929 mean + T * std_error, NULL);
933 tab_double (t, 8, row + count, 0, dd->minimum, fmt);
934 tab_double (t, 9, row + count, 0, dd->maximum, fmt);
939 double n, mean, variance;
943 moments1_calculate (ws->dd_total[v]->mom, &n, &mean, &variance, NULL, NULL);
945 std_dev = sqrt (variance);
946 std_error = std_dev / sqrt (n) ;
948 tab_text (t, 1, row + count,
949 TAB_LEFT | TAT_TITLE, _("Total"));
951 tab_double (t, 2, row + count, 0, n, wfmt);
953 tab_double (t, 3, row + count, 0, mean, NULL);
955 tab_double (t, 4, row + count, 0, std_dev, NULL);
957 tab_double (t, 5, row + count, 0, std_error, NULL);
959 /* Now the confidence interval */
960 T = gsl_cdf_tdist_Qinv (q, n - 1);
962 tab_double (t, 6, row + count, 0,
963 mean - T * std_error, NULL);
965 tab_double (t, 7, row + count, 0,
966 mean + T * std_error, NULL);
969 tab_double (t, 8, row + count, 0, ws->dd_total[v]->minimum, fmt);
970 tab_double (t, 9, row + count, 0, ws->dd_total[v]->maximum, fmt);
973 row += categoricals_total (cats) + 1;
979 /* Show the homogeneity table */
981 show_homogeneity (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
985 size_t n_rows = cmd->n_vars + 1;
987 struct tab_table *t = tab_create (n_cols, n_rows);
988 tab_headers (t, 1, 0, 1, 0);
990 /* Put a frame around the entire box, and vertical lines inside */
995 n_cols - 1, n_rows - 1);
998 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
999 tab_vline (t, TAL_2, 1, 0, n_rows - 1);
1001 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Levene Statistic"));
1002 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("df1"));
1003 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df2"));
1004 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
1006 tab_title (t, _("Test of Homogeneity of Variances"));
1008 for (v = 0; v < cmd->n_vars; ++v)
1011 moments1_calculate (ws->dd_total[v]->mom, &n, NULL, NULL, NULL, NULL);
1013 const struct per_var_ws *pvw = &ws->vws[v];
1014 const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
1016 const struct variable *var = cmd->vars[v];
1017 const struct group_proc *gp = group_proc_get (cmd->vars[v]);
1018 const char *s = var_to_string (var);
1020 const double df1 = pvw->n_groups - 1;
1021 const double df2 = n - pvw->n_groups;
1022 double F = gp->levene;
1024 tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s);
1027 tab_double (t, 1, v + 1, TAB_RIGHT, F, NULL);
1028 tab_fixed (t, 2, v + 1, TAB_RIGHT, df1, 8, 0);
1029 tab_fixed (t, 3, v + 1, TAB_RIGHT, df2, 8, 0);
1031 /* Now the significance */
1032 tab_double (t, 4, v + 1, TAB_RIGHT, gsl_cdf_fdist_Q (F, df1, df2), NULL);
1039 /* Show the contrast coefficients table */
1041 show_contrast_coeffs (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
1046 int n_contrasts = ll_count (&cmd->contrast_list);
1047 int n_cols = 2 + ws->actual_number_of_groups;
1048 int n_rows = 2 + n_contrasts;
1050 struct tab_table *t;
1052 const struct covariance *cov = ws->vws[0].cov ;
1054 t = tab_create (n_cols, n_rows);
1055 tab_headers (t, 2, 0, 2, 0);
1057 /* Put a frame around the entire box, and vertical lines inside */
1062 n_cols - 1, n_rows - 1);
1076 tab_hline (t, TAL_1, 2, n_cols - 1, 1);
1077 tab_hline (t, TAL_2, 0, n_cols - 1, 2);
1079 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
1081 tab_title (t, _("Contrast Coefficients"));
1083 tab_text (t, 0, 2, TAB_LEFT | TAT_TITLE, _("Contrast"));
1086 tab_joint_text (t, 2, 0, n_cols - 1, 0, TAB_CENTER | TAT_TITLE,
1087 var_to_string (cmd->indep_var));
1089 for ( cli = ll_head (&cmd->contrast_list);
1090 cli != ll_null (&cmd->contrast_list);
1091 cli = ll_next (cli))
1094 struct contrasts_node *cn = ll_data (cli, struct contrasts_node, ll);
1097 tab_text_format (t, 1, c_num + 2, TAB_CENTER, "%d", c_num + 1);
1099 for (coeffi = ll_head (&cn->coefficient_list);
1100 coeffi != ll_null (&cn->coefficient_list);
1101 ++count, coeffi = ll_next (coeffi))
1103 const struct categoricals *cats = covariance_get_categoricals (cov);
1104 const union value *val = categoricals_get_value_by_subscript (cats, count);
1107 ds_init_empty (&vstr);
1109 var_append_value_name (cmd->indep_var, val, &vstr);
1111 tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE, ds_cstr (&vstr));
1116 tab_text (t, count + 2, c_num + 2, TAB_RIGHT, "?" );
1119 struct coeff_node *coeffn = ll_data (coeffi, struct coeff_node, ll);
1121 tab_text_format (t, count + 2, c_num + 2, TAB_RIGHT, "%g", coeffn->coeff);
1131 /* Show the results of the contrast tests */
1133 show_contrast_tests (const struct oneway_spec *cmd, const struct oneway_workspace *ws)
1135 int n_contrasts = ll_count (&cmd->contrast_list);
1138 size_t n_rows = 1 + cmd->n_vars * 2 * n_contrasts;
1140 struct tab_table *t;
1142 t = tab_create (n_cols, n_rows);
1143 tab_headers (t, 3, 0, 1, 0);
1145 /* Put a frame around the entire box, and vertical lines inside */
1150 n_cols - 1, n_rows - 1);
1158 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
1159 tab_vline (t, TAL_2, 3, 0, n_rows - 1);
1161 tab_title (t, _("Contrast Tests"));
1163 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Contrast"));
1164 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Value of Contrast"));
1165 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error"));
1166 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("t"));
1167 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("df"));
1168 tab_text (t, 7, 0, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1170 for (v = 0; v < cmd->n_vars; ++v)
1172 const struct per_var_ws *pvw = &ws->vws[v];
1173 const struct categoricals *cats = covariance_get_categoricals (pvw->cov);
1176 int lines_per_variable = 2 * n_contrasts;
1178 tab_text (t, 0, (v * lines_per_variable) + 1, TAB_LEFT | TAT_TITLE,
1179 var_to_string (cmd->vars[v]));
1181 for ( cli = ll_head (&cmd->contrast_list);
1182 cli != ll_null (&cmd->contrast_list);
1183 ++i, cli = ll_next (cli))
1185 struct contrasts_node *cn = ll_data (cli, struct contrasts_node, ll);
1188 double contrast_value = 0.0;
1189 double coef_msq = 0.0;
1192 double std_error_contrast;
1194 double sec_vneq = 0.0;
1196 /* Note: The calculation of the degrees of freedom in the
1197 "variances not equal" case is painfull!!
1198 The following formula may help to understand it:
1199 \frac{\left (\sum_{i=1}^k{c_i^2\frac{s_i^2}{n_i}}\right)^2}
1202 \frac{\left (c_i^2\frac{s_i^2}{n_i}\right)^2} {n_i-1}
1207 double df_denominator = 0.0;
1208 double df_numerator = 0.0;
1211 moments1_calculate (ws->dd_total[v]->mom, &grand_n, NULL, NULL, NULL, NULL);
1212 df = grand_n - pvw->n_groups;
1216 tab_text (t, 1, (v * lines_per_variable) + i + 1,
1217 TAB_LEFT | TAT_TITLE,
1218 _("Assume equal variances"));
1220 tab_text (t, 1, (v * lines_per_variable) + i + 1 + n_contrasts,
1221 TAB_LEFT | TAT_TITLE,
1222 _("Does not assume equal"));
1225 tab_text_format (t, 2, (v * lines_per_variable) + i + 1,
1226 TAB_CENTER | TAT_TITLE, "%d", i + 1);
1229 tab_text_format (t, 2,
1230 (v * lines_per_variable) + i + 1 + n_contrasts,
1231 TAB_CENTER | TAT_TITLE, "%d", i + 1);
1236 for (coeffi = ll_head (&cn->coefficient_list);
1237 coeffi != ll_null (&cn->coefficient_list);
1238 ++ci, coeffi = ll_next (coeffi))
1240 double n, mean, variance;
1241 const struct descriptive_data *dd = categoricals_get_user_data_by_subscript (cats, ci);
1243 moments1_calculate (dd->mom, &n, &mean, &variance, NULL, NULL);
1245 struct coeff_node *cn = ll_data (coeffi, struct coeff_node, ll);
1246 const double coef = cn->coeff;
1248 const double winv = variance / n;
1250 contrast_value += coef * mean;
1252 coef_msq += (pow2 (coef)) / n;
1254 sec_vneq += (pow2 (coef)) * variance / n;
1256 df_numerator += (pow2 (coef)) * winv;
1257 df_denominator += pow2((pow2 (coef)) * winv) / (n - 1);
1260 sec_vneq = sqrt (sec_vneq);
1262 df_numerator = pow2 (df_numerator);
1264 tab_double (t, 3, (v * lines_per_variable) + i + 1,
1265 TAB_RIGHT, contrast_value, NULL);
1267 tab_double (t, 3, (v * lines_per_variable) + i + 1 +
1269 TAB_RIGHT, contrast_value, NULL);
1271 std_error_contrast = sqrt (pvw->mse * coef_msq);
1274 tab_double (t, 4, (v * lines_per_variable) + i + 1,
1275 TAB_RIGHT, std_error_contrast,
1278 T = fabs (contrast_value / std_error_contrast);
1282 tab_double (t, 5, (v * lines_per_variable) + i + 1,
1287 /* Degrees of Freedom */
1288 tab_fixed (t, 6, (v * lines_per_variable) + i + 1,
1293 /* Significance TWO TAILED !!*/
1294 tab_double (t, 7, (v * lines_per_variable) + i + 1,
1295 TAB_RIGHT, 2 * gsl_cdf_tdist_Q (T, df),
1298 /* Now for the Variances NOT Equal case */
1302 (v * lines_per_variable) + i + 1 + n_contrasts,
1303 TAB_RIGHT, sec_vneq,
1306 T = contrast_value / sec_vneq;
1308 (v * lines_per_variable) + i + 1 + n_contrasts,
1312 df = df_numerator / df_denominator;
1315 (v * lines_per_variable) + i + 1 + n_contrasts,
1319 /* The Significance */
1320 tab_double (t, 7, (v * lines_per_variable) + i + 1 + n_contrasts,
1321 TAB_RIGHT, 2 * gsl_cdf_tdist_Q (T,df),
1326 tab_hline (t, TAL_1, 0, n_cols - 1, (v * lines_per_variable) + 1);