1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 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>
25 #include <data/case.h>
26 #include <data/casegrouper.h>
27 #include <data/casereader.h>
28 #include <data/dictionary.h>
29 #include <data/procedure.h>
30 #include <data/value-labels.h>
31 #include <data/variable.h>
32 #include <language/command.h>
33 #include <language/dictionary/split-file.h>
34 #include <language/lexer/lexer.h>
35 #include <language/lexer/value-parser.h>
36 #include <libpspp/array.h>
37 #include <libpspp/assertion.h>
38 #include <libpspp/compiler.h>
39 #include <libpspp/hash.h>
40 #include <libpspp/message.h>
41 #include <libpspp/misc.h>
42 #include <libpspp/str.h>
43 #include <libpspp/taint.h>
44 #include <math/group-proc.h>
45 #include <math/levene.h>
46 #include <math/correlation.h>
47 #include <output/tab.h>
48 #include <data/format.h>
55 #define _(msgid) gettext (msgid)
63 +variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
65 missing=miss:!analysis/listwise,
66 incl:include/!exclude;
67 +format=fmt:!labels/nolabels;
68 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
79 /* A pair of variables to be compared. */
82 const struct variable *v[2]; /* The paired variables. */
83 double n; /* The number of valid variable pairs */
84 double sum[2]; /* The sum of the members */
85 double ssq[2]; /* sum of squares of the members */
86 double std_dev[2]; /* Std deviation of the members */
87 double s_std_dev[2]; /* Sample Std deviation of the members */
88 double mean[2]; /* The means of the members */
89 double correlation; /* Correlation coefficient between the variables. */
90 double sum_of_diffs; /* The sum of the differences */
91 double sum_of_prod; /* The sum of the products */
92 double mean_diff; /* The mean of the differences */
93 double ssq_diffs; /* The sum of the squares of the differences */
94 double std_dev_diff; /* The std deviation of the differences */
97 /* Which mode was T-TEST invoked */
99 T_1_SAMPLE, /* One-sample tests. */
100 T_IND_SAMPLES, /* Independent-sample tests. */
101 T_PAIRED /* Paired-sample tests. */
104 /* Total state of a T-TEST procedure. */
107 enum t_test_mode mode; /* Mode that T-TEST was invoked in. */
108 double criteria; /* Confidence interval in (0, 1). */
109 enum mv_class exclude; /* Classes of missing values to exclude. */
110 bool listwise_missing; /* Drop whole case if one missing var? */
111 struct fmt_spec weight_format; /* Format of weight variable. */
113 /* Dependent variables. */
114 const struct variable **vars;
117 /* For mode == T_1_SAMPLE. */
120 /* For mode == T_PAIRED only. */
124 /* For mode == T_IND_SAMPLES only. */
125 struct variable *indep_var; /* Independent variable. */
126 enum comparison criterion; /* Type of comparison. */
127 double critical_value; /* CMP_LE only: Grouping threshold value. */
128 union value g_value[2]; /* CMP_EQ only: Per-group indep var values. */
131 /* Statistics Summary Box */
135 void (*populate) (struct ssbox *, struct t_test_proc *);
136 void (*finalize) (struct ssbox *);
139 static void ssbox_create (struct ssbox *, struct t_test_proc *);
140 static void ssbox_populate (struct ssbox *, struct t_test_proc *);
141 static void ssbox_finalize (struct ssbox *);
143 /* Paired Samples Correlation box */
144 static void pscbox (struct t_test_proc *);
147 /* Test Results Box. */
150 void (*populate) (struct trbox *, struct t_test_proc *);
151 void (*finalize) (struct trbox *);
154 static void trbox_create (struct trbox *, struct t_test_proc *);
155 static void trbox_populate (struct trbox *, struct t_test_proc *);
156 static void trbox_finalize (struct trbox *);
158 static void calculate (struct t_test_proc *, struct casereader *,
159 const struct dataset *);
161 static int compare_group_binary (const struct group_statistics *a,
162 const struct group_statistics *b,
163 const struct t_test_proc *);
164 static unsigned hash_group_binary (const struct group_statistics *g,
165 const struct t_test_proc *p);
168 cmd_t_test (struct lexer *lexer, struct dataset *ds)
170 struct cmd_t_test cmd;
171 struct t_test_proc proc;
172 struct casegrouper *grouper;
173 struct casereader *group;
180 proc.indep_var = NULL;
181 if (!parse_t_test (lexer, ds, &cmd, &proc))
184 wv = dict_get_weight (dataset_dict (ds));
185 proc.weight_format = wv ? *var_get_print_format (wv) : F_8_0;
187 if ((cmd.sbc_testval != 0) + (cmd.sbc_groups != 0) + (cmd.sbc_pairs != 0)
190 msg (SE, _("Exactly one of TESTVAL, GROUPS and PAIRS subcommands "
191 "must be specified."));
195 proc.mode = (cmd.sbc_testval ? T_1_SAMPLE
196 : cmd.sbc_groups ? T_IND_SAMPLES
198 proc.criteria = cmd.sbc_criteria ? cmd.criteria : 0.95;
199 proc.exclude = cmd.incl != TTS_INCLUDE ? MV_ANY : MV_SYSTEM;
200 proc.listwise_missing = cmd.miss == TTS_LISTWISE;
202 if (proc.mode == T_1_SAMPLE)
203 proc.testval = cmd.n_testval[0];
205 if (proc.mode == T_PAIRED)
209 if (cmd.sbc_variables)
211 msg (SE, _("VARIABLES subcommand may not be used with PAIRS."));
215 /* Fill proc.vars with the unique variables from pairs. */
216 proc.n_vars = proc.n_pairs * 2;
217 proc.vars = xmalloc (sizeof *proc.vars * proc.n_vars);
218 for (i = j = 0; i < proc.n_pairs; i++)
220 proc.vars[j++] = proc.pairs[i].v[0];
221 proc.vars[j++] = proc.pairs[i].v[1];
223 proc.n_vars = sort_unique (proc.vars, proc.n_vars, sizeof *proc.vars,
224 compare_var_ptrs_by_name, NULL);
228 if (!cmd.n_variables)
230 msg (SE, _("One or more VARIABLES must be specified."));
233 proc.n_vars = cmd.n_variables;
234 proc.vars = cmd.v_variables;
235 cmd.v_variables = NULL;
239 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
240 while (casegrouper_get_next_group (grouper, &group))
241 calculate (&proc, group, ds);
242 ok = casegrouper_destroy (grouper);
243 ok = proc_commit (ds) && ok;
245 if (proc.mode == T_IND_SAMPLES)
248 /* Destroy any group statistics we created */
249 for (v = 0; v < proc.n_vars; v++)
251 struct group_proc *grpp = group_proc_get (proc.vars[v]);
252 hsh_destroy (grpp->group_hash);
259 if (proc.indep_var != NULL)
261 int width = var_get_width (proc.indep_var);
262 value_destroy (&proc.g_value[0], width);
263 value_destroy (&proc.g_value[1], width);
267 return ok ? CMD_SUCCESS : CMD_FAILURE;
271 tts_custom_groups (struct lexer *lexer, struct dataset *ds,
272 struct cmd_t_test *cmd UNUSED, void *proc_)
274 struct t_test_proc *proc = proc_;
278 lex_match (lexer, '=');
280 proc->indep_var = parse_variable (lexer, dataset_dict (ds));
281 if (proc->indep_var == NULL)
283 lex_error (lexer, "expecting variable name in GROUPS subcommand");
286 width = var_get_width (proc->indep_var);
287 value_init (&proc->g_value[0], width);
288 value_init (&proc->g_value[1], width);
290 if (!lex_match (lexer, '('))
294 if (!parse_value (lexer, &proc->g_value[0], width))
296 lex_match (lexer, ',');
297 if (lex_match (lexer, ')'))
301 if (!parse_value (lexer, &proc->g_value[1], width)
302 || !lex_force_match (lexer, ')'))
308 if (var_is_numeric (proc->indep_var))
310 proc->criterion = n_values == 1 ? CMP_LE : CMP_EQ;
312 proc->critical_value = proc->g_value[0].f;
313 else if (n_values == 0)
315 proc->g_value[0].f = 1;
316 proc->g_value[1].f = 2;
321 proc->criterion = CMP_EQ;
324 msg (SE, _("When applying GROUPS to a string variable, two "
325 "values must be specified."));
333 add_pair (struct t_test_proc *proc,
334 const struct variable *v0, const struct variable *v1)
336 struct pair *p = &proc->pairs[proc->n_pairs++];
342 tts_custom_pairs (struct lexer *lexer, struct dataset *ds,
343 struct cmd_t_test *cmd UNUSED, void *proc_)
345 struct t_test_proc *proc = proc_;
347 const struct variable **vars1 = NULL;
350 const struct variable **vars2 = NULL;
355 size_t n_total_pairs;
358 lex_match (lexer, '=');
360 if (!parse_variables_const (lexer, dataset_dict (ds), &vars1, &n_vars1,
361 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
364 if (lex_match (lexer, T_WITH))
366 if (!parse_variables_const (lexer, dataset_dict (ds), &vars2, &n_vars2,
367 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
373 if (lex_match (lexer, '(')
374 && lex_match_id (lexer, "PAIRED")
375 && lex_match (lexer, ')'))
378 if (n_vars1 != n_vars2)
380 msg (SE, _("PAIRED was specified but the number of variables "
381 "preceding WITH (%zu) did not match the number "
395 msg (SE, _("At least two variables must be specified on PAIRS."));
400 /* Allocate storage for the new pairs. */
401 n_total_pairs = proc->n_pairs + (paired ? n_vars1
402 : n_vars2 > 0 ? n_vars1 * n_vars2
403 : n_vars1 * (n_vars1 - 1) / 2);
404 proc->pairs = xnrealloc (proc->pairs, n_total_pairs, sizeof *proc->pairs);
406 /* Populate the pairs with the appropriate variables. */
408 for (i = 0; i < n_vars1; i++)
409 add_pair (proc, vars1[i], vars2[i]);
410 else if (n_vars2 > 0)
411 for (i = 0; i < n_vars1; i++)
412 for (j = 0; j < n_vars2; j++)
413 add_pair (proc, vars1[i], vars2[j]);
415 for (i = 0; i < n_vars1; i++)
416 for (j = i + 1; j < n_vars1; j++)
417 add_pair (proc, vars1[i], vars1[j]);
418 assert (proc->n_pairs == n_total_pairs);
425 /* Implementation of the SSBOX object. */
427 static void ssbox_base_init (struct ssbox *, int cols, int rows);
428 static void ssbox_base_finalize (struct ssbox *);
429 static void ssbox_one_sample_init (struct ssbox *, struct t_test_proc *);
430 static void ssbox_independent_samples_init (struct ssbox *, struct t_test_proc *);
431 static void ssbox_paired_init (struct ssbox *, struct t_test_proc *);
433 /* Factory to create an ssbox. */
435 ssbox_create (struct ssbox *ssb, struct t_test_proc *proc)
440 ssbox_one_sample_init (ssb, proc);
443 ssbox_independent_samples_init (ssb, proc);
446 ssbox_paired_init (ssb, proc);
453 /* Despatcher for the populate method */
455 ssbox_populate (struct ssbox *ssb, struct t_test_proc *proc)
457 ssb->populate (ssb, proc);
460 /* Despatcher for finalize */
462 ssbox_finalize (struct ssbox *ssb)
467 /* Submit the box and clear up */
469 ssbox_base_finalize (struct ssbox *ssb)
474 /* Initialize a ssbox struct */
476 ssbox_base_init (struct ssbox *this, int cols, int rows)
478 this->finalize = ssbox_base_finalize;
479 this->t = tab_create (cols, rows);
481 tab_headers (this->t, 0, 0, 1, 0);
482 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols - 1, rows - 1);
483 tab_hline (this->t, TAL_2, 0, cols- 1, 1);
486 /* ssbox implementations. */
488 static void ssbox_one_sample_populate (struct ssbox *, struct t_test_proc *);
489 static void ssbox_independent_samples_populate (struct ssbox *,
490 struct t_test_proc *);
491 static void ssbox_paired_populate (struct ssbox *, struct t_test_proc *);
493 /* Initialize the one_sample ssbox */
495 ssbox_one_sample_init (struct ssbox *this, struct t_test_proc *proc)
498 const int vsize = proc->n_vars + 1;
500 this->populate = ssbox_one_sample_populate;
502 ssbox_base_init (this, hsize, vsize);
503 tab_title (this->t, _("One-Sample Statistics"));
504 tab_vline (this->t, TAL_2, 1, 0, vsize - 1);
505 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
506 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
507 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
508 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("S.E. Mean"));
511 /* Initialize the independent samples ssbox */
513 ssbox_independent_samples_init (struct ssbox *this, struct t_test_proc *proc)
516 int vsize = proc->n_vars * 2 + 1;
518 this->populate = ssbox_independent_samples_populate;
520 ssbox_base_init (this, hsize, vsize);
521 tab_vline (this->t, TAL_GAP, 1, 0, vsize - 1);
522 tab_title (this->t, _("Group Statistics"));
523 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE,
524 var_get_name (proc->indep_var));
525 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
526 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
527 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
528 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("S.E. Mean"));
531 /* Populate the ssbox for independent samples */
533 ssbox_independent_samples_populate (struct ssbox *ssb,
534 struct t_test_proc *proc)
539 double indep_value[2];
543 for (i = 0; i < 2; i++)
545 union value *value = &proc->g_value[i];
546 int width = var_get_width (proc->indep_var);
548 indep_value[i] = (proc->criterion == CMP_LE ? proc->critical_value
551 if (val_type_from_width (width) == VAL_NUMERIC)
553 const char *s = var_lookup_value_label (proc->indep_var, value);
554 val_lab[i] = s ? xstrdup (s) : xasprintf ("%g", indep_value[i]);
557 val_lab[i] = xmemdup0 (value_str (value, width), width);
560 if (proc->criterion == CMP_LE)
562 strcpy (prefix[0], ">=");
563 strcpy (prefix[1], "<");
567 strcpy (prefix[0], "");
568 strcpy (prefix[1], "");
571 for (i = 0; i < proc->n_vars; i++)
573 const struct variable *var = proc->vars[i];
574 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
577 tab_text (ssb->t, 0, i * 2 + 1, TAB_LEFT,
578 var_get_name (proc->vars[i]));
579 tab_text_format (ssb->t, 1, i * 2 + 1, TAB_LEFT,
580 "%s%s", prefix[0], val_lab[0]);
581 tab_text_format (ssb->t, 1, i * 2 + 1+ 1, TAB_LEFT,
582 "%s%s", prefix[1], val_lab[1]);
584 /* Fill in the group statistics */
585 for (count = 0; count < 2; count++)
587 union value search_val;
588 struct group_statistics *gs;
590 if (proc->criterion == CMP_LE)
591 search_val.f = proc->critical_value + (count == 0 ? 1.0 : -1.0);
593 search_val = proc->g_value[count];
595 gs = hsh_find (grp_hash, &search_val);
598 tab_double (ssb->t, 2, i * 2 + count+ 1, TAB_RIGHT, gs->n,
599 &proc->weight_format);
600 tab_double (ssb->t, 3, i * 2 + count+ 1, TAB_RIGHT, gs->mean, NULL);
601 tab_double (ssb->t, 4, i * 2 + count+ 1, TAB_RIGHT, gs->std_dev,
603 tab_double (ssb->t, 5, i * 2 + count+ 1, TAB_RIGHT, gs->se_mean,
611 /* Initialize the paired values ssbox */
613 ssbox_paired_init (struct ssbox *this, struct t_test_proc *proc)
616 int vsize = proc->n_pairs * 2 + 1;
618 this->populate = ssbox_paired_populate;
620 ssbox_base_init (this, hsize, vsize);
621 tab_title (this->t, _("Paired Sample Statistics"));
622 tab_vline (this->t, TAL_GAP, 1, 0, vsize - 1);
623 tab_vline (this->t, TAL_2, 2, 0, vsize - 1);
624 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
625 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
626 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
627 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("S.E. Mean"));
630 /* Populate the ssbox for paired values */
632 ssbox_paired_populate (struct ssbox *ssb, struct t_test_proc *proc)
636 for (i = 0; i < proc->n_pairs; i++)
638 struct pair *p = &proc->pairs[i];
641 tab_text_format (ssb->t, 0, i * 2 + 1, TAB_LEFT, _("Pair %d"), i);
642 for (j=0; j < 2; j++)
645 tab_text (ssb->t, 1, i * 2 + j + 1, TAB_LEFT,
646 var_get_name (p->v[j]));
649 tab_double (ssb->t, 2, i * 2 + j + 1, TAB_RIGHT, p->mean[j], NULL);
650 tab_double (ssb->t, 3, i * 2 + j + 1, TAB_RIGHT, p->n,
651 &proc->weight_format);
652 tab_double (ssb->t, 4, i * 2 + j + 1, TAB_RIGHT, p->std_dev[j],
654 tab_double (ssb->t, 5, i * 2 + j + 1, TAB_RIGHT,
655 p->std_dev[j] /sqrt (p->n), NULL);
660 /* Populate the one sample ssbox */
662 ssbox_one_sample_populate (struct ssbox *ssb, struct t_test_proc *proc)
666 for (i = 0; i < proc->n_vars; i++)
668 struct group_statistics *gs = &group_proc_get (proc->vars[i])->ugs;
670 tab_text (ssb->t, 0, i + 1, TAB_LEFT, var_get_name (proc->vars[i]));
671 tab_double (ssb->t, 1, i + 1, TAB_RIGHT, gs->n, &proc->weight_format);
672 tab_double (ssb->t, 2, i + 1, TAB_RIGHT, gs->mean, NULL);
673 tab_double (ssb->t, 3, i + 1, TAB_RIGHT, gs->std_dev, NULL);
674 tab_double (ssb->t, 4, i + 1, TAB_RIGHT, gs->se_mean, NULL);
678 /* Implementation of the Test Results box struct */
680 static void trbox_base_init (struct trbox *, size_t n_vars, int cols);
681 static void trbox_base_finalize (struct trbox *);
682 static void trbox_independent_samples_init (struct trbox *,
683 struct t_test_proc *);
684 static void trbox_independent_samples_populate (struct trbox *,
685 struct t_test_proc *);
686 static void trbox_one_sample_init (struct trbox *, struct t_test_proc *);
687 static void trbox_one_sample_populate (struct trbox *, struct t_test_proc *);
688 static void trbox_paired_init (struct trbox *, struct t_test_proc *);
689 static void trbox_paired_populate (struct trbox *, struct t_test_proc *);
691 /* Create a trbox according to mode*/
693 trbox_create (struct trbox *trb, struct t_test_proc *proc)
698 trbox_one_sample_init (trb, proc);
701 trbox_independent_samples_init (trb, proc);
704 trbox_paired_init (trb, proc);
711 /* Populate a trbox according to proc */
713 trbox_populate (struct trbox *trb, struct t_test_proc *proc)
715 trb->populate (trb, proc);
718 /* Submit and destroy a trbox */
720 trbox_finalize (struct trbox *trb)
725 /* Initialize the independent samples trbox */
727 trbox_independent_samples_init (struct trbox *self,
728 struct t_test_proc *proc)
730 const int hsize = 11;
731 const int vsize = proc->n_vars * 2 + 3;
734 self->populate = trbox_independent_samples_populate;
736 trbox_base_init (self, proc->n_vars * 2, hsize);
737 tab_title (self->t, _("Independent Samples Test"));
738 tab_hline (self->t, TAL_1, 2, hsize - 1, 1);
739 tab_vline (self->t, TAL_2, 2, 0, vsize - 1);
740 tab_vline (self->t, TAL_1, 4, 0, vsize - 1);
741 tab_box (self->t, -1, -1, -1, TAL_1, 2, 1, hsize - 2, vsize - 1);
742 tab_hline (self->t, TAL_1, hsize - 2, hsize - 1, 2);
743 tab_box (self->t, -1, -1, -1, TAL_1, hsize - 2, 2, hsize - 1, vsize - 1);
744 tab_joint_text (self->t, 2, 0, 3, 0,
745 TAB_CENTER, _("Levene's Test for Equality of Variances"));
746 tab_joint_text (self->t, 4, 0, hsize- 1, 0,
747 TAB_CENTER, _("t-test for Equality of Means"));
749 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("F"));
750 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig."));
751 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("t"));
752 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("df"));
753 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
754 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
755 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Difference"));
756 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
757 tab_text (self->t, 10, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
759 tab_joint_text_format (self->t, 9, 1, 10, 1, TAB_CENTER,
760 _("%g%% Confidence Interval of the Difference"),
761 proc->criteria * 100.0);
764 /* Populate the independent samples trbox */
766 trbox_independent_samples_populate (struct trbox *self,
767 struct t_test_proc *proc)
771 for (i = 0; i < proc->n_vars; i++)
780 double pooled_variance;
786 const struct variable *var = proc->vars[i];
787 struct group_proc *grp_data = group_proc_get (var);
789 struct hsh_table *grp_hash = grp_data->group_hash;
791 struct group_statistics *gs0;
792 struct group_statistics *gs1;
794 union value search_val;
796 if (proc->criterion == CMP_LE)
797 search_val.f = proc->critical_value - 1.0;
799 search_val = proc->g_value[0];
801 gs0 = hsh_find (grp_hash, &search_val);
804 if (proc->criterion == CMP_LE)
805 search_val.f = proc->critical_value + 1.0;
807 search_val = proc->g_value[1];
809 gs1 = hsh_find (grp_hash, &search_val);
813 tab_text (self->t, 0, i * 2 + 3, TAB_LEFT, var_get_name (proc->vars[i]));
814 tab_text (self->t, 1, i * 2 + 3, TAB_LEFT, _("Equal variances assumed"));
815 tab_double (self->t, 2, i * 2 + 3, TAB_CENTER, grp_data->levene, NULL);
817 /* Now work out the significance of the Levene test */
819 df2 = grp_data->ugs.n - 2;
820 q = gsl_cdf_fdist_Q (grp_data->levene, df1, df2);
821 tab_double (self->t, 3, i * 2 + 3, TAB_CENTER, q, NULL);
823 df = gs0->n + gs1->n - 2.0;
824 tab_double (self->t, 5, i * 2 + 3, TAB_RIGHT, df, NULL);
826 pooled_variance = (gs0->n * pow2 (gs0->s_std_dev)
827 + gs1->n *pow2 (gs1->s_std_dev)) / df ;
829 t = (gs0->mean - gs1->mean) / sqrt (pooled_variance);
830 t /= sqrt ((gs0->n + gs1->n) / (gs0->n * gs1->n));
832 tab_double (self->t, 4, i * 2 + 3, TAB_RIGHT, t, NULL);
834 p = gsl_cdf_tdist_P (t, df);
835 q = gsl_cdf_tdist_Q (t, df);
837 tab_double (self->t, 6, i * 2 + 3, TAB_RIGHT, 2.0 * (t > 0 ? q : p),
840 mean_diff = gs0->mean - gs1->mean;
841 tab_double (self->t, 7, i * 2 + 3, TAB_RIGHT, mean_diff, NULL);
844 std_err_diff = sqrt (pow2 (gs0->se_mean) + pow2 (gs1->se_mean));
845 tab_double (self->t, 8, i * 2 + 3, TAB_RIGHT, std_err_diff, NULL);
847 /* Now work out the confidence interval */
848 q = (1 - proc->criteria)/2.0; /* 2-tailed test */
850 t = gsl_cdf_tdist_Qinv (q, df);
851 tab_double (self->t, 9, i * 2 + 3, TAB_RIGHT,
852 mean_diff - t * std_err_diff, NULL);
854 tab_double (self->t, 10, i * 2 + 3, TAB_RIGHT,
855 mean_diff + t * std_err_diff, NULL);
858 /* Now for the \sigma_1 != \sigma_2 case */
859 tab_text (self->t, 1, i * 2 + 3 + 1,
860 TAB_LEFT, _("Equal variances not assumed"));
862 se2 = ((pow2 (gs0->s_std_dev) / (gs0->n - 1)) +
863 (pow2 (gs1->s_std_dev) / (gs1->n - 1)));
865 t = mean_diff / sqrt (se2);
866 tab_double (self->t, 4, i * 2 + 3 + 1, TAB_RIGHT, t, NULL);
868 df = pow2 (se2) / ((pow2 (pow2 (gs0->s_std_dev) / (gs0->n - 1))
870 + (pow2 (pow2 (gs1->s_std_dev) / (gs1->n - 1))
872 tab_double (self->t, 5, i * 2 + 3 + 1, TAB_RIGHT, df, NULL);
874 p = gsl_cdf_tdist_P (t, df);
875 q = gsl_cdf_tdist_Q (t, df);
877 tab_double (self->t, 6, i * 2 + 3 + 1, TAB_RIGHT, 2.0 * (t > 0 ? q : p),
880 /* Now work out the confidence interval */
881 q = (1 - proc->criteria) / 2.0; /* 2-tailed test */
883 t = gsl_cdf_tdist_Qinv (q, df);
885 tab_double (self->t, 7, i * 2 + 3 + 1, TAB_RIGHT, mean_diff, NULL);
886 tab_double (self->t, 8, i * 2 + 3 + 1, TAB_RIGHT, std_err_diff, NULL);
887 tab_double (self->t, 9, i * 2 + 3 + 1, TAB_RIGHT,
888 mean_diff - t * std_err_diff, NULL);
889 tab_double (self->t, 10, i * 2 + 3 + 1, TAB_RIGHT,
890 mean_diff + t * std_err_diff, NULL);
894 /* Initialize the paired samples trbox */
896 trbox_paired_init (struct trbox *self, struct t_test_proc *proc)
899 const int vsize=proc->n_pairs+ 3;
901 self->populate = trbox_paired_populate;
903 trbox_base_init (self, proc->n_pairs, hsize);
904 tab_title (self->t, _("Paired Samples Test"));
905 tab_hline (self->t, TAL_1, 2, 6, 1);
906 tab_vline (self->t, TAL_2, 2, 0, vsize - 1);
907 tab_joint_text (self->t, 2, 0, 6, 0, TAB_CENTER, _("Paired Differences"));
908 tab_box (self->t, -1, -1, -1, TAL_1, 2, 1, 6, vsize - 1);
909 tab_box (self->t, -1, -1, -1, TAL_1, 6, 0, hsize - 1, vsize - 1);
910 tab_hline (self->t, TAL_1, 5, 6, 2);
911 tab_vline (self->t, TAL_GAP, 6, 0, 1);
913 tab_joint_text_format (self->t, 5, 1, 6, 1, TAB_CENTER,
914 _("%g%% Confidence Interval of the Difference"),
915 proc->criteria*100.0);
917 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
918 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
919 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
920 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
921 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
922 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
923 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
924 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
927 /* Populate the paired samples trbox */
929 trbox_paired_populate (struct trbox *trb,
930 struct t_test_proc *proc)
934 for (i = 0; i < proc->n_pairs; i++)
936 struct pair *pair = &proc->pairs[i];
944 tab_text_format (trb->t, 0, i + 3, TAB_LEFT, _("Pair %d"), i);
945 tab_text_format (trb->t, 1, i + 3, TAB_LEFT, "%s - %s",
946 var_get_name (pair->v[0]),
947 var_get_name (pair->v[1]));
948 tab_double (trb->t, 2, i + 3, TAB_RIGHT, pair->mean_diff, NULL);
949 tab_double (trb->t, 3, i + 3, TAB_RIGHT, pair->std_dev_diff, NULL);
952 se_mean = pair->std_dev_diff / sqrt (n);
953 tab_double (trb->t, 4, i + 3, TAB_RIGHT, se_mean, NULL);
955 /* Now work out the confidence interval */
956 q = (1 - proc->criteria) / 2.0; /* 2-tailed test */
958 t = gsl_cdf_tdist_Qinv (q, df);
960 tab_double (trb->t, 5, i + 3, TAB_RIGHT,
961 pair->mean_diff - t * se_mean, NULL);
962 tab_double (trb->t, 6, i + 3, TAB_RIGHT,
963 pair->mean_diff + t * se_mean, NULL);
965 t = ((pair->mean[0] - pair->mean[1])
966 / sqrt ((pow2 (pair->s_std_dev[0]) + pow2 (pair->s_std_dev[1])
967 - (2 * pair->correlation
968 * pair->s_std_dev[0] * pair->s_std_dev[1]))
971 tab_double (trb->t, 7, i + 3, TAB_RIGHT, t, NULL);
973 /* Degrees of freedom */
974 tab_double (trb->t, 8, i + 3, TAB_RIGHT, df, &proc->weight_format);
976 p = gsl_cdf_tdist_P (t,df);
977 q = gsl_cdf_tdist_Q (t,df);
979 tab_double (trb->t, 9, i + 3, TAB_RIGHT, 2.0 * (t > 0 ? q : p), NULL);
983 /* Initialize the one sample trbox */
985 trbox_one_sample_init (struct trbox *self, struct t_test_proc *proc)
988 const int vsize = proc->n_vars + 3;
990 self->populate = trbox_one_sample_populate;
992 trbox_base_init (self, proc->n_vars, hsize);
993 tab_title (self->t, _("One-Sample Test"));
994 tab_hline (self->t, TAL_1, 1, hsize - 1, 1);
995 tab_vline (self->t, TAL_2, 1, 0, vsize - 1);
997 tab_joint_text_format (self->t, 1, 0, hsize - 1, 0, TAB_CENTER,
998 _("Test Value = %f"), proc->testval);
1000 tab_box (self->t, -1, -1, -1, TAL_1, 1, 1, hsize - 1, vsize - 1);
1003 tab_joint_text_format (self->t, 5, 1, 6, 1, TAB_CENTER,
1004 _("%g%% Confidence Interval of the Difference"),
1005 proc->criteria * 100.0);
1007 tab_vline (self->t, TAL_GAP, 6, 1, 1);
1008 tab_hline (self->t, TAL_1, 5, 6, 2);
1009 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1010 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1011 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1012 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1013 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1014 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1017 /* Populate the one sample trbox */
1019 trbox_one_sample_populate (struct trbox *trb, struct t_test_proc *proc)
1025 for (i = 0; i < proc->n_vars; i++)
1030 struct group_statistics *gs = &group_proc_get (proc->vars[i])->ugs;
1032 tab_text (trb->t, 0, i + 3, TAB_LEFT, var_get_name (proc->vars[i]));
1034 t = (gs->mean - proc->testval) * sqrt (gs->n) / gs->std_dev;
1036 tab_double (trb->t, 1, i + 3, TAB_RIGHT, t, NULL);
1038 /* degrees of freedom */
1041 tab_double (trb->t, 2, i + 3, TAB_RIGHT, df, &proc->weight_format);
1043 p = gsl_cdf_tdist_P (t, df);
1044 q = gsl_cdf_tdist_Q (t, df);
1046 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1048 tab_double (trb->t, 3, i + 3, TAB_RIGHT, 2.0 * (t > 0 ? q : p), NULL);
1049 tab_double (trb->t, 4, i + 3, TAB_RIGHT, gs->mean_diff, NULL);
1052 q = (1 - proc->criteria) / 2.0; /* 2-tailed test */
1053 t = gsl_cdf_tdist_Qinv (q, df);
1055 tab_double (trb->t, 5, i + 3, TAB_RIGHT,
1056 gs->mean_diff - t * gs->se_mean, NULL);
1057 tab_double (trb->t, 6, i + 3, TAB_RIGHT,
1058 gs->mean_diff + t * gs->se_mean, NULL);
1062 /* Base initializer for the generalized trbox */
1064 trbox_base_init (struct trbox *self, size_t data_rows, int cols)
1066 const size_t rows = 3 + data_rows;
1068 self->finalize = trbox_base_finalize;
1069 self->t = tab_create (cols, rows);
1070 tab_headers (self->t, 0, 0, 3, 0);
1071 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols - 1, rows - 1);
1072 tab_hline (self->t, TAL_2, 0, cols- 1, 3);
1075 /* Base finalizer for the trbox */
1077 trbox_base_finalize (struct trbox *trb)
1079 tab_submit (trb->t);
1082 /* Create, populate and submit the Paired Samples Correlation box */
1084 pscbox (struct t_test_proc *proc)
1086 const int rows=1+proc->n_pairs;
1090 struct tab_table *table;
1092 table = tab_create (cols, rows);
1094 tab_headers (table, 0, 0, 1, 0);
1095 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols - 1, rows - 1);
1096 tab_hline (table, TAL_2, 0, cols - 1, 1);
1097 tab_vline (table, TAL_2, 2, 0, rows - 1);
1098 tab_title (table, _("Paired Samples Correlations"));
1100 /* column headings */
1101 tab_text (table, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
1102 tab_text (table, 3, 0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1103 tab_text (table, 4, 0, TAB_CENTER | TAT_TITLE, _("Sig."));
1105 for (i = 0; i < proc->n_pairs; i++)
1107 struct pair *pair = &proc->pairs[i];
1110 tab_text_format (table, 0, i + 1, TAB_LEFT | TAT_TITLE,
1112 tab_text_format (table, 1, i + 1, TAB_LEFT | TAT_TITLE,
1114 var_get_name (pair->v[0]),
1115 var_get_name (pair->v[1]));
1118 tab_double (table, 2, i + 1, TAB_RIGHT, pair->n, &proc->weight_format);
1119 tab_double (table, 3, i + 1, TAB_RIGHT, pair->correlation, NULL);
1121 tab_double (table, 4, i + 1, TAB_RIGHT,
1122 2.0 * significance_of_correlation (pair->correlation, pair->n), NULL);
1128 /* Calculation Implementation */
1130 /* Calculations common to all variants of the T test. */
1132 common_calc (const struct dictionary *dict,
1133 struct t_test_proc *proc,
1134 struct casereader *reader)
1139 for (i = 0; i < proc->n_vars; i++)
1141 struct group_statistics *gs = &group_proc_get (proc->vars[i])->ugs;
1148 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
1150 double weight = dict_get_case_weight (dict, c, NULL);
1152 /* Listwise has to be implicit if the independent variable
1154 if (proc->mode == T_IND_SAMPLES)
1156 if (var_is_value_missing (proc->indep_var,
1157 case_data (c, proc->indep_var),
1162 for (i = 0; i < proc->n_vars; i++)
1164 const struct variable *v = proc->vars[i];
1165 const union value *val = case_data (c, v);
1167 if (!var_is_value_missing (v, val, proc->exclude))
1169 struct group_statistics *gs;
1170 gs = &group_proc_get (v)->ugs;
1173 gs->sum += weight * val->f;
1174 gs->ssq += weight * pow2 (val->f);
1178 casereader_destroy (reader);
1180 for (i = 0; i < proc->n_vars; i++)
1182 struct group_statistics *gs = &group_proc_get (proc->vars[i])->ugs;
1184 gs->mean = gs->sum / gs->n;
1185 gs->s_std_dev = sqrt (((gs->ssq / gs->n) - pow2 (gs->mean)));
1186 gs->std_dev = sqrt (gs->n / (gs->n- 1)
1187 * ((gs->ssq / gs->n) - pow2 (gs->mean)));
1188 gs->se_mean = gs->std_dev / sqrt (gs->n);
1189 gs->mean_diff = gs->sum_diff / gs->n;
1193 /* Calculations for one sample T test. */
1195 one_sample_calc (const struct dictionary *dict, struct t_test_proc *proc,
1196 struct casereader *reader)
1201 for (i = 0; i < proc->n_vars; i++)
1203 struct group_statistics *gs = &group_proc_get (proc->vars[i])->ugs;
1207 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
1209 double weight = dict_get_case_weight (dict, c, NULL);
1210 for (i = 0; i < proc->n_vars; i++)
1212 const struct variable *v = proc->vars[i];
1213 struct group_statistics *gs = &group_proc_get (v)->ugs;
1214 const union value *val = case_data (c, v);
1215 if (!var_is_value_missing (v, val, proc->exclude))
1216 gs->sum_diff += weight * (val->f - proc->testval);
1220 for (i = 0; i < proc->n_vars; i++)
1222 struct group_statistics *gs = &group_proc_get (proc->vars[i])->ugs;
1223 gs->mean_diff = gs->sum_diff / gs->n;
1226 casereader_destroy (reader);
1232 paired_calc (const struct dictionary *dict, struct t_test_proc *proc,
1233 struct casereader *reader)
1238 for (i = 0; i < proc->n_pairs; i++)
1240 struct pair *pair = &proc->pairs[i];
1242 pair->sum[0] = pair->sum[1] = 0;
1243 pair->ssq[0] = pair->ssq[1] = 0;
1244 pair->sum_of_prod = 0;
1245 pair->correlation = 0;
1246 pair->sum_of_diffs = 0;
1247 pair->ssq_diffs = 0;
1250 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
1252 double weight = dict_get_case_weight (dict, c, NULL);
1253 for (i = 0; i < proc->n_pairs; i++)
1255 struct pair *pair = &proc->pairs[i];
1256 const struct variable *v0 = pair->v[0];
1257 const struct variable *v1 = pair->v[1];
1259 const union value *val0 = case_data (c, v0);
1260 const union value *val1 = case_data (c, v1);
1262 if (!var_is_value_missing (v0, val0, proc->exclude)
1263 && !var_is_value_missing (v1, val1, proc->exclude))
1266 pair->sum[0] += weight * val0->f;
1267 pair->sum[1] += weight * val1->f;
1268 pair->ssq[0] += weight * pow2 (val0->f);
1269 pair->ssq[1] += weight * pow2 (val1->f);
1270 pair->sum_of_prod += weight * val0->f * val1->f;
1271 pair->sum_of_diffs += weight * (val0->f - val1->f);
1272 pair->ssq_diffs += weight * pow2 (val0->f - val1->f);
1277 for (i = 0; i < proc->n_pairs; i++)
1279 struct pair *pair = &proc->pairs[i];
1280 const double n = pair->n;
1283 for (j=0; j < 2; j++)
1285 pair->mean[j] = pair->sum[j] / n;
1286 pair->s_std_dev[j] = sqrt ((pair->ssq[j] / n
1287 - pow2 (pair->mean[j])));
1288 pair->std_dev[j] = sqrt (n / (n- 1) * (pair->ssq[j] / n
1289 - pow2 (pair->mean[j])));
1292 pair->correlation = (pair->sum_of_prod / pair->n
1293 - pair->mean[0] * pair->mean[1]);
1294 /* correlation now actually contains the covariance */
1295 pair->correlation /= pair->std_dev[0] * pair->std_dev[1];
1296 pair->correlation *= pair->n / (pair->n - 1);
1298 pair->mean_diff = pair->sum_of_diffs / n;
1299 pair->std_dev_diff = sqrt (n / (n - 1) * ((pair->ssq_diffs / n)
1300 - pow2 (pair->mean_diff)));
1303 casereader_destroy (reader);
1308 group_calc (const struct dictionary *dict, struct t_test_proc *proc,
1309 struct casereader *reader)
1314 for (i = 0; i < proc->n_vars; i++)
1316 struct group_proc *ttpr = group_proc_get (proc->vars[i]);
1319 /* There's always 2 groups for a T - TEST */
1321 ttpr->group_hash = hsh_create (2,
1322 (hsh_compare_func *) compare_group_binary,
1323 (hsh_hash_func *) hash_group_binary,
1324 (hsh_free_func *) free_group,
1327 for (j = 0; j < 2; j++)
1329 struct group_statistics *gs = xmalloc (sizeof *gs);
1333 if (proc->criterion == CMP_EQ)
1334 gs->id = proc->g_value[j];
1338 gs->id.f = proc->critical_value - 1.0;
1340 gs->id.f = proc->critical_value + 1.0;
1343 hsh_insert (ttpr->group_hash, gs);
1347 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
1349 const double weight = dict_get_case_weight (dict, c, NULL);
1350 const union value *gv;
1352 if (var_is_value_missing (proc->indep_var,
1353 case_data (c, proc->indep_var), proc->exclude))
1356 gv = case_data (c, proc->indep_var);
1357 for (i = 0; i < proc->n_vars; i++)
1359 const struct variable *var = proc->vars[i];
1360 const union value *val = case_data (c, var);
1361 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1362 struct group_statistics *gs = hsh_find (grp_hash, gv);
1364 /* If the independent variable doesn't match either of the values
1365 for this case then move on to the next case. */
1369 if (!var_is_value_missing (var, val, proc->exclude))
1372 gs->sum += weight * val->f;
1373 gs->ssq += weight * pow2 (val->f);
1378 for (i = 0; i < proc->n_vars; i++)
1380 const struct variable *var = proc->vars[i];
1381 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1382 struct hsh_iterator g;
1383 struct group_statistics *gs;
1386 for (gs = hsh_first (grp_hash, &g); gs != NULL;
1387 gs = hsh_next (grp_hash, &g))
1389 gs->mean = gs->sum / gs->n;
1390 gs->s_std_dev = sqrt (((gs->ssq / gs->n) - pow2 (gs->mean)));
1391 gs->std_dev = sqrt (gs->n / (gs->n- 1)
1392 * ((gs->ssq / gs->n) - pow2 (gs->mean)));
1393 gs->se_mean = gs->std_dev / sqrt (gs->n);
1396 assert (count == 2);
1399 casereader_destroy (reader);
1405 calculate (struct t_test_proc *proc,
1406 struct casereader *input, const struct dataset *ds)
1408 const struct dictionary *dict = dataset_dict (ds);
1409 struct ssbox stat_summary_box;
1410 struct trbox test_results_box;
1411 struct taint *taint;
1414 c = casereader_peek (input, 0);
1417 casereader_destroy (input);
1420 output_split_file_values (ds, c);
1423 if (proc->listwise_missing)
1424 input = casereader_create_filter_missing (input,
1427 proc->exclude, NULL, NULL);
1428 input = casereader_create_filter_weight (input, dict, NULL, NULL);
1429 taint = taint_clone (casereader_get_taint (input));
1431 common_calc (dict, proc, casereader_clone (input));
1435 one_sample_calc (dict, proc, input);
1438 paired_calc (dict, proc, input);
1441 group_calc (dict, proc, casereader_clone (input));
1442 levene (dict, input, proc->indep_var, proc->n_vars, proc->vars,
1449 if (!taint_has_tainted_successor (taint))
1451 ssbox_create (&stat_summary_box, proc);
1452 ssbox_populate (&stat_summary_box, proc);
1453 ssbox_finalize (&stat_summary_box);
1455 if (proc->mode == T_PAIRED)
1458 trbox_create (&test_results_box, proc);
1459 trbox_populate (&test_results_box, proc);
1460 trbox_finalize (&test_results_box);
1463 taint_destroy (taint);
1466 /* return 0 if G belongs to group 0,
1467 1 if it belongs to group 1,
1468 2 if it belongs to neither group */
1470 which_group (const struct group_statistics *g,
1471 const struct t_test_proc *proc)
1473 int width = var_get_width (proc->indep_var);
1475 if (0 == value_compare_3way (&g->id, &proc->g_value[0], width))
1478 if (0 == value_compare_3way (&g->id, &proc->g_value[1], width))
1484 /* Return -1 if the id of a is less than b; +1 if greater than and
1487 compare_group_binary (const struct group_statistics *a,
1488 const struct group_statistics *b,
1489 const struct t_test_proc *proc)
1494 if (proc->criterion == CMP_LE)
1496 flag_a = (a->id.f < proc->critical_value);
1497 flag_b = (b->id.f < proc->critical_value);
1501 flag_a = which_group (a, proc);
1502 flag_b = which_group (b, proc);
1505 if (flag_a < flag_b)
1508 return (flag_a > flag_b);
1511 /* This is a degenerate case of a hash, since it can only return three possible
1512 values. It's really a comparison, being used as a hash function */
1515 hash_group_binary (const struct group_statistics *g,
1516 const struct t_test_proc *proc)
1518 return (proc->criterion == CMP_LE
1519 ? g->id.f < proc->critical_value
1520 : which_group (g, proc));