1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000 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 <libpspp/assertion.h>
36 #include <libpspp/compiler.h>
37 #include <libpspp/hash.h>
38 #include <libpspp/message.h>
39 #include <libpspp/misc.h>
40 #include <libpspp/str.h>
41 #include <libpspp/taint.h>
42 #include <math/group-proc.h>
43 #include <math/levene.h>
44 #include <output/manager.h>
45 #include <output/table.h>
50 #define _(msgid) gettext (msgid)
58 +variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
60 missing=miss:!analysis/listwise,
61 incl:include/!exclude;
62 +format=fmt:!labels/nolabels;
63 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
69 /* Variable for the GROUPS subcommand, if given. */
70 static struct variable *indep_var;
78 struct group_properties
80 /* The comparison criterion */
81 enum comparison criterion;
83 /* The width of the independent variable */
87 /* The value of the independent variable at which groups are determined to
88 belong to one group or the other */
89 double critical_value;
92 /* The values of the independent variable for each group */
93 union value g_value[2];
99 static struct group_properties gp ;
103 /* PAIRS: Number of pairs to be compared ; each pair. */
104 static int n_pairs = 0 ;
107 /* The variables comprising the pair */
108 const struct variable *v[2];
110 /* The number of valid variable pairs */
113 /* The sum of the members */
116 /* sum of squares of the members */
119 /* Std deviation of the members */
123 /* Sample Std deviation of the members */
126 /* The means of the members */
129 /* The correlation coefficient between the variables */
132 /* The sum of the differences */
135 /* The sum of the products */
138 /* The mean of the differences */
141 /* The sum of the squares of the differences */
144 /* The std deviation of the differences */
148 static struct pair *pairs=0;
150 static int parse_value (struct lexer *lexer, union value * v, enum val_type);
152 /* Structures and Functions for the Statistics Summary Box */
154 typedef void populate_ssbox_func (struct ssbox *ssb,
155 struct cmd_t_test *cmd);
156 typedef void finalize_ssbox_func (struct ssbox *ssb);
162 populate_ssbox_func *populate;
163 finalize_ssbox_func *finalize;
168 void ssbox_create (struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
170 /* Populate a ssbox according to cmd */
171 void ssbox_populate (struct ssbox *ssb, struct cmd_t_test *cmd);
173 /* Submit and destroy a ssbox */
174 void ssbox_finalize (struct ssbox *ssb);
176 /* A function to create, populate and submit the Paired Samples Correlation
181 /* Structures and Functions for the Test Results Box */
184 typedef void populate_trbox_func (struct trbox *trb,
185 struct cmd_t_test *cmd);
186 typedef void finalize_trbox_func (struct trbox *trb);
190 populate_trbox_func *populate;
191 finalize_trbox_func *finalize;
195 void trbox_create (struct trbox *trb, struct cmd_t_test *cmd, int mode);
197 /* Populate a ssbox according to cmd */
198 void trbox_populate (struct trbox *trb, struct cmd_t_test *cmd);
200 /* Submit and destroy a ssbox */
201 void trbox_finalize (struct trbox *trb);
203 /* Which mode was T-TEST invoked */
211 static int common_calc (const struct dictionary *dict,
212 const struct ccase *, void *,
214 static void common_precalc (struct cmd_t_test *);
215 static void common_postcalc (struct cmd_t_test *);
217 static int one_sample_calc (const struct dictionary *dict, const struct ccase *, void *, enum mv_class);
218 static void one_sample_precalc (struct cmd_t_test *);
219 static void one_sample_postcalc (struct cmd_t_test *);
221 static int paired_calc (const struct dictionary *dict, const struct ccase *,
222 struct cmd_t_test*, enum mv_class);
223 static void paired_precalc (struct cmd_t_test *);
224 static void paired_postcalc (struct cmd_t_test *);
226 static void group_precalc (struct cmd_t_test *);
227 static int group_calc (const struct dictionary *dict, const struct ccase *,
228 struct cmd_t_test *, enum mv_class);
229 static void group_postcalc (struct cmd_t_test *);
232 static void calculate (struct cmd_t_test *,
234 const struct dataset *);
238 static struct cmd_t_test cmd;
240 static bool bad_weight_warn = false;
243 static int compare_group_binary (const struct group_statistics *a,
244 const struct group_statistics *b,
245 const struct group_properties *p);
248 static unsigned hash_group_binary (const struct group_statistics *g,
249 const struct group_properties *p);
254 cmd_t_test (struct lexer *lexer, struct dataset *ds)
256 struct casegrouper *grouper;
257 struct casereader *group;
260 if ( !parse_t_test (lexer, ds, &cmd, NULL) )
263 if (! cmd.sbc_criteria)
268 if (cmd.sbc_testval) ++m;
269 if (cmd.sbc_groups) ++m;
270 if (cmd.sbc_pairs) ++m;
275 _ ("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
284 else if (cmd.sbc_groups)
289 if ( mode == T_PAIRED)
291 if (cmd.sbc_variables)
293 msg (SE, _ ("VARIABLES subcommand is not appropriate with PAIRS"));
299 /* Iterate through the pairs and put each variable that is a
300 member of a pair into cmd.v_variables */
303 struct hsh_iterator hi;
304 struct const_hsh_table *hash;
305 const struct variable *v;
307 hash = const_hsh_create (n_pairs, compare_vars_by_name, hash_var_by_name,
310 for (i=0; i < n_pairs; ++i)
312 const_hsh_insert (hash, pairs[i].v[0]);
313 const_hsh_insert (hash, pairs[i].v[1]);
316 assert (cmd.n_variables == 0);
317 cmd.n_variables = const_hsh_count (hash);
319 cmd.v_variables = xnrealloc (cmd.v_variables, cmd.n_variables,
320 sizeof *cmd.v_variables);
321 /* Iterate through the hash */
322 for (i=0,v = const_hsh_first (hash, &hi);
324 v = const_hsh_next (hash, &hi) )
325 cmd.v_variables[i++]=v;
326 const_hsh_destroy (hash);
329 else if ( !cmd.sbc_variables)
331 msg (SE, _ ("One or more VARIABLES must be specified."));
336 bad_weight_warn = true;
339 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
340 while (casegrouper_get_next_group (grouper, &group))
341 calculate (&cmd, group, ds);
342 ok = casegrouper_destroy (grouper);
343 ok = proc_commit (ds) && ok;
349 if ( mode == T_IND_SAMPLES)
352 /* Destroy any group statistics we created */
353 for (v = 0 ; v < cmd.n_variables ; ++v )
355 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
356 hsh_destroy (grpp->group_hash);
361 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
365 tts_custom_groups (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd UNUSED,
368 int n_group_values=0;
370 lex_match (lexer, '=');
372 indep_var = parse_variable (lexer, dataset_dict (ds));
375 lex_error (lexer, "expecting variable name in GROUPS subcommand");
379 if (var_is_long_string (indep_var))
381 msg (SE, _ ("Long string variable %s is not valid here."),
382 var_get_name (indep_var));
386 if (!lex_match (lexer, '('))
388 if (var_is_numeric (indep_var))
390 gp.v.g_value[0].f = 1;
391 gp.v.g_value[1].f = 2;
393 gp.criterion = CMP_EQ;
401 msg (SE, _ ("When applying GROUPS to a string variable, two "
402 "values must be specified."));
407 if (!parse_value (lexer, &gp.v.g_value[0], var_get_type (indep_var)))
410 lex_match (lexer, ',');
411 if (lex_match (lexer, ')'))
413 if (var_is_alpha (indep_var))
415 msg (SE, _ ("When applying GROUPS to a string variable, two "
416 "values must be specified."));
419 gp.criterion = CMP_LE;
420 gp.v.critical_value = gp.v.g_value[0].f;
426 if (!parse_value (lexer, &gp.v.g_value[1], var_get_type (indep_var)))
430 if (!lex_force_match (lexer, ')'))
433 if ( n_group_values == 2 )
434 gp.criterion = CMP_EQ ;
436 gp.criterion = CMP_LE ;
439 if ( var_is_alpha (indep_var))
441 buf_copy_rpad (gp.v.g_value [0].s, var_get_width (indep_var),
442 gp.v.g_value [0].s, strlen (gp.v.g_value[0].s));
444 buf_copy_rpad (gp.v.g_value [1].s, var_get_width (indep_var),
445 gp.v.g_value [1].s, strlen (gp.v.g_value[1].s));
453 tts_custom_pairs (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
455 const struct variable **vars;
457 size_t n_pairs_local;
459 size_t n_before_WITH;
460 size_t n_after_WITH = SIZE_MAX;
461 int paired ; /* Was the PAIRED keyword given ? */
463 lex_match (lexer, '=');
466 if (!parse_variables_const (lexer, dataset_dict (ds), &vars, &n_vars,
467 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
475 if (lex_match (lexer, T_WITH))
477 n_before_WITH = n_vars;
478 if (!parse_variables_const (lexer, dataset_dict (ds), &vars, &n_vars,
479 PV_DUPLICATE | PV_APPEND
480 | PV_NUMERIC | PV_NO_SCRATCH))
485 n_after_WITH = n_vars - n_before_WITH;
488 paired = (lex_match (lexer, '(') && lex_match_id (lexer, "PAIRED") && lex_match (lexer, ')'));
490 /* Determine the number of pairs needed */
493 if (n_before_WITH != n_after_WITH)
496 msg (SE, _ ("PAIRED was specified but the number of variables "
497 "preceding WITH (%zu) did not match the number "
499 n_before_WITH, n_after_WITH);
502 n_pairs_local = n_before_WITH;
504 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
506 n_pairs_local = n_before_WITH * n_after_WITH ;
508 else /* Neither WITH nor PAIRED keyword given */
513 msg (SE, _ ("At least two variables must be specified "
518 /* how many ways can you pick 2 from n_vars ? */
519 n_pairs_local = n_vars * (n_vars - 1) / 2;
523 /* Allocate storage for the pairs */
524 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
526 /* Populate the pairs with the appropriate variables */
531 assert (n_pairs_local == n_vars / 2);
532 for (i = 0; i < n_pairs_local; ++i)
534 pairs[i].v[n_pairs] = vars[i];
535 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
538 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
543 for (i=0 ; i < n_before_WITH ; ++i )
545 for (j=0 ; j < n_after_WITH ; ++j)
547 pairs[p].v[0] = vars[i];
548 pairs[p].v[1] = vars[j+n_before_WITH];
553 else /* Neither WITH nor PAIRED given */
558 for (i=0 ; i < n_vars ; ++i )
560 for (j=i+1 ; j < n_vars ; ++j)
562 pairs[p].v[0] = vars[i];
563 pairs[p].v[1] = vars[j];
569 n_pairs+=n_pairs_local;
575 /* Parses the current token (numeric or string, depending on type)
576 value v and returns success. */
578 parse_value (struct lexer *lexer, union value * v, enum val_type type)
580 if (type == VAL_NUMERIC)
582 if (!lex_force_num (lexer))
584 v->f = lex_tokval (lexer);
588 if (!lex_force_string (lexer))
590 memset (v->s, ' ', MAX_SHORT_STRING);
591 strncpy (v->s, ds_cstr (lex_tokstr (lexer)), ds_length (lex_tokstr (lexer)));
600 /* Implementation of the SSBOX object */
602 void ssbox_base_init (struct ssbox *this, int cols,int rows);
604 void ssbox_base_finalize (struct ssbox *ssb);
606 void ssbox_one_sample_init (struct ssbox *this,
607 struct cmd_t_test *cmd );
609 void ssbox_independent_samples_init (struct ssbox *this,
610 struct cmd_t_test *cmd);
612 void ssbox_paired_init (struct ssbox *this,
613 struct cmd_t_test *cmd);
616 /* Factory to create an ssbox */
618 ssbox_create (struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
623 ssbox_one_sample_init (ssb,cmd);
626 ssbox_independent_samples_init (ssb,cmd);
629 ssbox_paired_init (ssb,cmd);
638 /* Despatcher for the populate method */
640 ssbox_populate (struct ssbox *ssb,struct cmd_t_test *cmd)
642 ssb->populate (ssb,cmd);
646 /* Despatcher for finalize */
648 ssbox_finalize (struct ssbox *ssb)
654 /* Submit the box and clear up */
656 ssbox_base_finalize (struct ssbox *ssb)
663 /* Initialize a ssbox struct */
665 ssbox_base_init (struct ssbox *this, int cols,int rows)
667 this->finalize = ssbox_base_finalize;
668 this->t = tab_create (cols, rows, 0);
670 tab_columns (this->t, SOM_COL_DOWN, 1);
671 tab_headers (this->t,0,0,1,0);
672 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
673 tab_hline (this->t, TAL_2,0,cols-1,1);
674 tab_dim (this->t, tab_natural_dimensions);
677 void ssbox_one_sample_populate (struct ssbox *ssb,
678 struct cmd_t_test *cmd);
680 /* Initialize the one_sample ssbox */
682 ssbox_one_sample_init (struct ssbox *this,
683 struct cmd_t_test *cmd )
686 const int vsize=cmd->n_variables+1;
688 this->populate = ssbox_one_sample_populate;
690 ssbox_base_init (this, hsize,vsize);
691 tab_title (this->t, _ ("One-Sample Statistics"));
692 tab_vline (this->t, TAL_2, 1,0,vsize - 1);
693 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
694 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
695 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
696 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
699 void ssbox_independent_samples_populate (struct ssbox *ssb,
700 struct cmd_t_test *cmd);
702 /* Initialize the independent samples ssbox */
704 ssbox_independent_samples_init (struct ssbox *this,
705 struct cmd_t_test *cmd)
708 int vsize = cmd->n_variables*2 +1;
710 this->populate = ssbox_independent_samples_populate;
712 ssbox_base_init (this, hsize,vsize);
713 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
714 tab_title (this->t, _ ("Group Statistics"));
715 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, var_get_name (indep_var));
716 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
717 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
718 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
719 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
723 /* Populate the ssbox for independent samples */
725 ssbox_independent_samples_populate (struct ssbox *ssb,
726 struct cmd_t_test *cmd)
730 char *val_lab[2] = {NULL, NULL};
731 double indep_value[2];
733 char prefix[2][3]={"",""};
735 if ( var_is_numeric (indep_var) )
739 s = var_lookup_value_label (indep_var, &gp.v.g_value[0]);
740 val_lab[0] = s ? strdup (s) : NULL;
742 s = var_lookup_value_label (indep_var, &gp.v.g_value[1]);
743 val_lab[1] = s ? strdup (s) : NULL;
747 val_lab[0] = calloc (sizeof (char), MAX_SHORT_STRING + 1);
748 val_lab[1] = calloc (sizeof (char), MAX_SHORT_STRING + 1);
749 memcpy (val_lab[0], gp.v.g_value[0].s, MAX_SHORT_STRING);
750 memcpy (val_lab[1], gp.v.g_value[1].s, MAX_SHORT_STRING);
753 if (gp.criterion == CMP_LE )
755 strcpy (prefix[0],">=");
756 strcpy (prefix[1],"<");
757 indep_value[0] = gp.v.critical_value;
758 indep_value[1] = gp.v.critical_value;
762 indep_value[0] = gp.v.g_value[0].f;
763 indep_value[1] = gp.v.g_value[1].f;
768 for (i=0; i < cmd->n_variables; ++i)
770 const struct variable *var = cmd->v_variables[i];
771 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
774 tab_text (ssb->t, 0, i*2+1, TAB_LEFT,
775 var_get_name (cmd->v_variables[i]));
778 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
779 "%s%s", prefix[0], val_lab[0]);
781 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
782 "%s%g", prefix[0], indep_value[0]);
786 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
787 "%s%s", prefix[1], val_lab[1]);
789 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
790 "%s%g", prefix[1], indep_value[1]);
793 /* Fill in the group statistics */
794 for ( count = 0 ; count < 2 ; ++count )
796 union value search_val;
798 struct group_statistics *gs;
800 if ( gp.criterion == CMP_LE )
805 search_val.f = gp.v.critical_value + 1.0;
809 /* less than ( < ) case */
810 search_val.f = gp.v.critical_value - 1.0;
815 search_val = gp.v.g_value[count];
818 gs = hsh_find (grp_hash, (void *) &search_val);
821 tab_float (ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 10, 0);
822 tab_float (ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
823 tab_float (ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
824 tab_float (ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
832 void ssbox_paired_populate (struct ssbox *ssb,
833 struct cmd_t_test *cmd);
835 /* Initialize the paired values ssbox */
837 ssbox_paired_init (struct ssbox *this, struct cmd_t_test *cmd UNUSED)
841 int vsize = n_pairs*2+1;
843 this->populate = ssbox_paired_populate;
845 ssbox_base_init (this, hsize,vsize);
846 tab_title (this->t, _ ("Paired Sample Statistics"));
847 tab_vline (this->t,TAL_GAP,1,0,vsize-1);
848 tab_vline (this->t,TAL_2,2,0,vsize-1);
849 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
850 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
851 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
852 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
856 /* Populate the ssbox for paired values */
858 ssbox_paired_populate (struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
864 for (i=0; i < n_pairs; ++i)
868 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _ ("Pair %d"),i);
870 for (j=0 ; j < 2 ; ++j)
872 struct group_statistics *gs;
874 gs = &group_proc_get (pairs[i].v[j])->ugs;
878 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT,
879 var_get_name (pairs[i].v[j]));
882 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
883 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 10, 0);
884 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
885 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt (pairs[i].n), 8, 3);
891 /* Populate the one sample ssbox */
893 ssbox_one_sample_populate (struct ssbox *ssb, struct cmd_t_test *cmd)
899 for (i=0; i < cmd->n_variables; ++i)
901 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
903 tab_text (ssb->t, 0, i+1, TAB_LEFT, var_get_name (cmd->v_variables[i]));
904 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 10, 0);
905 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
906 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
907 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
914 /* Implementation of the Test Results box struct */
916 void trbox_base_init (struct trbox *self,size_t n_vars, int cols);
917 void trbox_base_finalize (struct trbox *trb);
919 void trbox_independent_samples_init (struct trbox *trb,
920 struct cmd_t_test *cmd );
922 void trbox_independent_samples_populate (struct trbox *trb,
923 struct cmd_t_test *cmd);
925 void trbox_one_sample_init (struct trbox *self,
926 struct cmd_t_test *cmd );
928 void trbox_one_sample_populate (struct trbox *trb,
929 struct cmd_t_test *cmd);
931 void trbox_paired_init (struct trbox *self,
932 struct cmd_t_test *cmd );
934 void trbox_paired_populate (struct trbox *trb,
935 struct cmd_t_test *cmd);
939 /* Create a trbox according to mode*/
941 trbox_create (struct trbox *trb,
942 struct cmd_t_test *cmd, int mode)
947 trbox_one_sample_init (trb,cmd);
950 trbox_independent_samples_init (trb,cmd);
953 trbox_paired_init (trb,cmd);
960 /* Populate a trbox according to cmd */
962 trbox_populate (struct trbox *trb, struct cmd_t_test *cmd)
964 trb->populate (trb,cmd);
967 /* Submit and destroy a trbox */
969 trbox_finalize (struct trbox *trb)
974 /* Initialize the independent samples trbox */
976 trbox_independent_samples_init (struct trbox *self,
977 struct cmd_t_test *cmd UNUSED)
980 const int vsize=cmd->n_variables*2+3;
983 self->populate = trbox_independent_samples_populate;
985 trbox_base_init (self,cmd->n_variables*2,hsize);
986 tab_title (self->t,_ ("Independent Samples Test"));
987 tab_hline (self->t,TAL_1,2,hsize-1,1);
988 tab_vline (self->t,TAL_2,2,0,vsize-1);
989 tab_vline (self->t,TAL_1,4,0,vsize-1);
990 tab_box (self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
991 tab_hline (self->t,TAL_1, hsize-2,hsize-1,2);
992 tab_box (self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
993 tab_joint_text (self->t, 2, 0, 3, 0,
994 TAB_CENTER,_ ("Levene's Test for Equality of Variances"));
995 tab_joint_text (self->t, 4,0,hsize-1,0,
996 TAB_CENTER,_ ("t-test for Equality of Means"));
998 tab_text (self->t,2,2, TAB_CENTER | TAT_TITLE,_ ("F"));
999 tab_text (self->t,3,2, TAB_CENTER | TAT_TITLE,_ ("Sig."));
1000 tab_text (self->t,4,2, TAB_CENTER | TAT_TITLE,_ ("t"));
1001 tab_text (self->t,5,2, TAB_CENTER | TAT_TITLE,_ ("df"));
1002 tab_text (self->t,6,2, TAB_CENTER | TAT_TITLE,_ ("Sig. (2-tailed)"));
1003 tab_text (self->t,7,2, TAB_CENTER | TAT_TITLE,_ ("Mean Difference"));
1004 tab_text (self->t,8,2, TAB_CENTER | TAT_TITLE,_ ("Std. Error Difference"));
1005 tab_text (self->t,9,2, TAB_CENTER | TAT_TITLE,_ ("Lower"));
1006 tab_text (self->t,10,2, TAB_CENTER | TAT_TITLE,_ ("Upper"));
1008 tab_joint_text (self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
1009 _ ("%g%% Confidence Interval of the Difference"),
1010 cmd->criteria*100.0);
1014 /* Populate the independent samples trbox */
1016 trbox_independent_samples_populate (struct trbox *self,
1017 struct cmd_t_test *cmd )
1022 for (i=0; i < cmd->n_variables; ++i)
1031 double pooled_variance;
1032 double std_err_diff;
1035 const struct variable *var = cmd->v_variables[i];
1036 struct group_proc *grp_data = group_proc_get (var);
1038 struct hsh_table *grp_hash = grp_data->group_hash;
1040 struct group_statistics *gs0 ;
1041 struct group_statistics *gs1 ;
1043 union value search_val;
1045 if ( gp.criterion == CMP_LE )
1046 search_val.f = gp.v.critical_value - 1.0;
1048 search_val = gp.v.g_value[0];
1050 gs0 = hsh_find (grp_hash, (void *) &search_val);
1053 if ( gp.criterion == CMP_LE )
1054 search_val.f = gp.v.critical_value + 1.0;
1056 search_val = gp.v.g_value[1];
1058 gs1 = hsh_find (grp_hash, (void *) &search_val);
1062 tab_text (self->t, 0, i*2+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1064 tab_text (self->t, 1, i*2+3, TAB_LEFT, _ ("Equal variances assumed"));
1067 tab_float (self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1069 /* Now work out the significance of the Levene test */
1070 df1 = 1; df2 = grp_data->ugs.n - 2;
1071 q = gsl_cdf_fdist_Q (grp_data->levene, df1, df2);
1073 tab_float (self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1075 df = gs0->n + gs1->n - 2.0 ;
1076 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 10, 0);
1078 pooled_variance = ( (gs0->n )*pow2 (gs0->s_std_dev)
1080 (gs1->n )*pow2 (gs1->s_std_dev)
1083 t = (gs0->mean - gs1->mean) / sqrt (pooled_variance) ;
1084 t /= sqrt ((gs0->n + gs1->n)/ (gs0->n*gs1->n));
1086 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1088 p = gsl_cdf_tdist_P (t, df);
1089 q = gsl_cdf_tdist_Q (t, df);
1091 tab_float (self->t, 6, i*2+3, TAB_RIGHT, 2.0* (t>0?q:p) , 8, 3);
1093 mean_diff = gs0->mean - gs1->mean;
1094 tab_float (self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1097 std_err_diff = sqrt ( pow2 (gs0->se_mean) + pow2 (gs1->se_mean));
1098 tab_float (self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1101 /* Now work out the confidence interval */
1102 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1104 t = gsl_cdf_tdist_Qinv (q,df);
1105 tab_float (self->t, 9, i*2+3, TAB_RIGHT,
1106 mean_diff - t * std_err_diff, 8, 3);
1108 tab_float (self->t, 10, i*2+3, TAB_RIGHT,
1109 mean_diff + t * std_err_diff, 8, 3);
1114 /* Now for the \sigma_1 != \sigma_2 case */
1115 tab_text (self->t, 1, i*2+3+1,
1116 TAB_LEFT, _ ("Equal variances not assumed"));
1119 se2 = (pow2 (gs0->s_std_dev)/ (gs0->n -1) ) +
1120 (pow2 (gs1->s_std_dev)/ (gs1->n -1) );
1122 t = mean_diff / sqrt (se2) ;
1123 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1126 (pow2 (pow2 (gs0->s_std_dev)/ (gs0->n - 1 ))
1130 (pow2 (pow2 (gs1->s_std_dev)/ (gs1->n - 1 ))
1134 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1136 p = gsl_cdf_tdist_P (t, df);
1137 q = gsl_cdf_tdist_Q (t, df);
1139 tab_float (self->t, 6, i*2+3+1, TAB_RIGHT, 2.0* (t>0?q:p) , 8, 3);
1141 /* Now work out the confidence interval */
1142 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1144 t = gsl_cdf_tdist_Qinv (q, df);
1146 tab_float (self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1149 tab_float (self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1152 tab_float (self->t, 9, i*2+3+1, TAB_RIGHT,
1153 mean_diff - t * std_err_diff, 8, 3);
1155 tab_float (self->t, 10, i*2+3+1, TAB_RIGHT,
1156 mean_diff + t * std_err_diff, 8, 3);
1162 /* Initialize the paired samples trbox */
1164 trbox_paired_init (struct trbox *self,
1165 struct cmd_t_test *cmd UNUSED)
1169 const int vsize=n_pairs+3;
1171 self->populate = trbox_paired_populate;
1173 trbox_base_init (self,n_pairs,hsize);
1174 tab_title (self->t, _ ("Paired Samples Test"));
1175 tab_hline (self->t,TAL_1,2,6,1);
1176 tab_vline (self->t,TAL_2,2,0,vsize - 1);
1177 tab_joint_text (self->t,2,0,6,0,TAB_CENTER,_ ("Paired Differences"));
1178 tab_box (self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1179 tab_box (self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1180 tab_hline (self->t,TAL_1,5,6, 2);
1181 tab_vline (self->t,TAL_GAP,6,0,1);
1183 tab_joint_text (self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1184 _ ("%g%% Confidence Interval of the Difference"),
1185 cmd->criteria*100.0);
1187 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _ ("Mean"));
1188 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
1189 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _ ("Std. Error Mean"));
1190 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _ ("Lower"));
1191 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _ ("Upper"));
1192 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _ ("t"));
1193 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _ ("df"));
1194 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _ ("Sig. (2-tailed)"));
1197 /* Populate the paired samples trbox */
1199 trbox_paired_populate (struct trbox *trb,
1200 struct cmd_t_test *cmd UNUSED)
1204 for (i=0; i < n_pairs; ++i)
1209 double n = pairs[i].n;
1213 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _ ("Pair %d"),i);
1215 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1216 var_get_name (pairs[i].v[0]),
1217 var_get_name (pairs[i].v[1]));
1219 tab_float (trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1221 tab_float (trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1224 se_mean = pairs[i].std_dev_diff / sqrt (n) ;
1225 tab_float (trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1227 /* Now work out the confidence interval */
1228 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1230 t = gsl_cdf_tdist_Qinv (q, df);
1232 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1233 pairs[i].mean_diff - t * se_mean , 8, 4);
1235 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1236 pairs[i].mean_diff + t * se_mean , 8, 4);
1238 t = (pairs[i].mean[0] - pairs[i].mean[1])
1240 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1241 2 * pairs[i].correlation *
1242 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1246 tab_float (trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1248 /* Degrees of freedom */
1249 tab_float (trb->t, 8, i+3, TAB_RIGHT, df , 10, 0 );
1251 p = gsl_cdf_tdist_P (t,df);
1252 q = gsl_cdf_tdist_P (t,df);
1254 tab_float (trb->t, 9, i+3, TAB_RIGHT, 2.0* (t>0?q:p) , 8, 3);
1259 /* Initialize the one sample trbox */
1261 trbox_one_sample_init (struct trbox *self, struct cmd_t_test *cmd )
1264 const int vsize=cmd->n_variables+3;
1266 self->populate = trbox_one_sample_populate;
1268 trbox_base_init (self, cmd->n_variables,hsize);
1269 tab_title (self->t, _ ("One-Sample Test"));
1270 tab_hline (self->t, TAL_1, 1, hsize - 1, 1);
1271 tab_vline (self->t, TAL_2, 1, 0, vsize - 1);
1273 tab_joint_text (self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1274 _ ("Test Value = %f"), cmd->n_testval[0]);
1276 tab_box (self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1279 tab_joint_text (self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1280 _ ("%g%% Confidence Interval of the Difference"),
1281 cmd->criteria*100.0);
1283 tab_vline (self->t,TAL_GAP,6,1,1);
1284 tab_hline (self->t,TAL_1,5,6,2);
1285 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _ ("t"));
1286 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _ ("df"));
1287 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _ ("Sig. (2-tailed)"));
1288 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _ ("Mean Difference"));
1289 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _ ("Lower"));
1290 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _ ("Upper"));
1295 /* Populate the one sample trbox */
1297 trbox_one_sample_populate (struct trbox *trb, struct cmd_t_test *cmd)
1303 for (i=0; i < cmd->n_variables; ++i)
1308 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1311 tab_text (trb->t, 0, i+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1313 t = (gs->mean - cmd->n_testval[0] ) * sqrt (gs->n) / gs->std_dev ;
1315 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1317 /* degrees of freedom */
1320 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1322 p = gsl_cdf_tdist_P (t, df);
1323 q = gsl_cdf_tdist_Q (t, df);
1325 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1327 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0* (t>0?q:p), 8,3);
1329 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1332 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1333 t = gsl_cdf_tdist_Qinv (q, df);
1335 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1336 gs->mean_diff - t * gs->se_mean, 8,4);
1338 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1339 gs->mean_diff + t * gs->se_mean, 8,4);
1343 /* Base initializer for the generalized trbox */
1345 trbox_base_init (struct trbox *self, size_t data_rows, int cols)
1347 const size_t rows = 3 + data_rows;
1349 self->finalize = trbox_base_finalize;
1350 self->t = tab_create (cols, rows, 0);
1351 tab_headers (self->t,0,0,3,0);
1352 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1353 tab_hline (self->t, TAL_2,0,cols-1,3);
1354 tab_dim (self->t, tab_natural_dimensions);
1358 /* Base finalizer for the trbox */
1360 trbox_base_finalize (struct trbox *trb)
1362 tab_submit (trb->t);
1366 /* Create , populate and submit the Paired Samples Correlation box */
1370 const int rows=1+n_pairs;
1374 struct tab_table *table;
1376 table = tab_create (cols,rows,0);
1378 tab_columns (table, SOM_COL_DOWN, 1);
1379 tab_headers (table,0,0,1,0);
1380 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1381 tab_hline (table, TAL_2, 0, cols - 1, 1);
1382 tab_vline (table, TAL_2, 2, 0, rows - 1);
1383 tab_dim (table, tab_natural_dimensions);
1384 tab_title (table, _ ("Paired Samples Correlations"));
1386 /* column headings */
1387 tab_text (table, 2,0, TAB_CENTER | TAT_TITLE, _ ("N"));
1388 tab_text (table, 3,0, TAB_CENTER | TAT_TITLE, _ ("Correlation"));
1389 tab_text (table, 4,0, TAB_CENTER | TAT_TITLE, _ ("Sig."));
1391 for (i=0; i < n_pairs; ++i)
1395 double df = pairs[i].n -2;
1397 double correlation_t =
1398 pairs[i].correlation * sqrt (df) /
1399 sqrt (1 - pow2 (pairs[i].correlation));
1403 tab_text (table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1406 tab_text (table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1408 var_get_name (pairs[i].v[0]),
1409 var_get_name (pairs[i].v[1]));
1413 tab_float (table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1414 tab_float (table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1416 p = gsl_cdf_tdist_P (correlation_t, df);
1417 q = gsl_cdf_tdist_Q (correlation_t, df);
1419 tab_float (table, 4, i+1, TAB_RIGHT, 2.0* (correlation_t>0?q:p), 8, 3);
1428 /* Calculation Implementation */
1430 /* Per case calculations common to all variants of the T test */
1432 common_calc (const struct dictionary *dict,
1433 const struct ccase *c,
1435 enum mv_class exclude)
1438 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1440 double weight = dict_get_case_weight (dict, c, NULL);
1443 /* Listwise has to be implicit if the independent variable is missing ?? */
1444 if ( cmd->sbc_groups )
1446 if (var_is_value_missing (indep_var, case_data (c, indep_var), exclude))
1450 for (i = 0; i < cmd->n_variables ; ++i)
1452 const struct variable *v = cmd->v_variables[i];
1453 const union value *val = case_data (c, v);
1455 if (!var_is_value_missing (v, val, exclude))
1457 struct group_statistics *gs;
1458 gs = &group_proc_get (v)->ugs;
1461 gs->sum += weight * val->f;
1462 gs->ssq += weight * val->f * val->f;
1468 /* Pre calculations common to all variants of the T test */
1470 common_precalc ( struct cmd_t_test *cmd )
1474 for (i=0; i< cmd->n_variables ; ++i)
1476 struct group_statistics *gs;
1477 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1486 /* Post calculations common to all variants of the T test */
1488 common_postcalc (struct cmd_t_test *cmd)
1492 for (i=0; i< cmd->n_variables ; ++i)
1494 struct group_statistics *gs;
1495 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1497 gs->mean=gs->sum / gs->n;
1498 gs->s_std_dev= sqrt (
1499 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1504 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1507 gs->se_mean = gs->std_dev / sqrt (gs->n);
1508 gs->mean_diff= gs->sum_diff / gs->n;
1512 /* Per case calculations for one sample t test */
1514 one_sample_calc (const struct dictionary *dict,
1515 const struct ccase *c, void *cmd_,
1516 enum mv_class exclude)
1520 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1522 double weight = dict_get_case_weight (dict, c, NULL);
1525 for (i=0; i< cmd->n_variables ; ++i)
1527 struct group_statistics *gs;
1528 const struct variable *v = cmd->v_variables[i];
1529 const union value *val = case_data (c, v);
1531 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1533 if (!var_is_value_missing (v, val, exclude))
1534 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1540 /* Pre calculations for one sample t test */
1542 one_sample_precalc ( struct cmd_t_test *cmd )
1546 for (i=0; i< cmd->n_variables ; ++i)
1548 struct group_statistics *gs;
1549 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1555 /* Post calculations for one sample t test */
1557 one_sample_postcalc (struct cmd_t_test *cmd)
1561 for (i=0; i< cmd->n_variables ; ++i)
1563 struct group_statistics *gs;
1564 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1566 gs->mean_diff = gs->sum_diff / gs->n ;
1573 paired_precalc (struct cmd_t_test *cmd UNUSED)
1577 for (i=0; i < n_pairs ; ++i )
1580 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1581 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1582 pairs[i].sum_of_prod = 0;
1583 pairs[i].correlation = 0;
1584 pairs[i].sum_of_diffs = 0;
1585 pairs[i].ssq_diffs = 0;
1592 paired_calc (const struct dictionary *dict, const struct ccase *c,
1593 struct cmd_t_test *cmd UNUSED, enum mv_class exclude)
1597 double weight = dict_get_case_weight (dict, c, NULL);
1599 for (i=0; i < n_pairs ; ++i )
1601 const struct variable *v0 = pairs[i].v[0];
1602 const struct variable *v1 = pairs[i].v[1];
1604 const union value *val0 = case_data (c, v0);
1605 const union value *val1 = case_data (c, v1);
1607 if (!var_is_value_missing (v0, val0, exclude) &&
1608 !var_is_value_missing (v1, val1, exclude))
1610 pairs[i].n += weight;
1611 pairs[i].sum[0] += weight * val0->f;
1612 pairs[i].sum[1] += weight * val1->f;
1614 pairs[i].ssq[0] += weight * pow2 (val0->f);
1615 pairs[i].ssq[1] += weight * pow2 (val1->f);
1617 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1619 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1620 pairs[i].ssq_diffs += weight * pow2 (val0->f - val1->f);
1628 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1632 for (i=0; i < n_pairs ; ++i )
1635 const double n = pairs[i].n;
1637 for (j=0; j < 2 ; ++j)
1639 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1640 pairs[i].s_std_dev[j] = sqrt ((pairs[i].ssq[j] / n -
1641 pow2 (pairs[i].mean[j]))
1644 pairs[i].std_dev[j] = sqrt (n/ (n-1)* (pairs[i].ssq[j] / n -
1645 pow2 (pairs[i].mean[j]))
1649 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1650 pairs[i].mean[0] * pairs[i].mean[1] ;
1651 /* correlation now actually contains the covariance */
1653 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1654 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1656 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1658 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1659 ( pairs[i].ssq_diffs / n )
1661 pow2 (pairs[i].mean_diff )
1667 group_precalc (struct cmd_t_test *cmd )
1672 for (i=0; i< cmd->n_variables ; ++i)
1674 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1676 /* There's always 2 groups for a T - TEST */
1679 gp.indep_width = var_get_width (indep_var);
1681 ttpr->group_hash = hsh_create (2,
1682 (hsh_compare_func *) compare_group_binary,
1683 (hsh_hash_func *) hash_group_binary,
1684 (hsh_free_func *) free_group,
1687 for (j=0 ; j < 2 ; ++j)
1689 struct group_statistics *gs = xmalloc (sizeof *gs);
1695 if ( gp.criterion == CMP_EQ )
1697 gs->id = gp.v.g_value[j];
1702 gs->id.f = gp.v.critical_value - 1.0 ;
1704 gs->id.f = gp.v.critical_value + 1.0 ;
1707 hsh_insert ( ttpr->group_hash, (void *) gs );
1714 group_calc (const struct dictionary *dict,
1715 const struct ccase *c, struct cmd_t_test *cmd,
1716 enum mv_class exclude)
1720 const double weight = dict_get_case_weight (dict, c, NULL);
1722 const union value *gv;
1724 if (var_is_value_missing (indep_var, case_data (c, indep_var), exclude))
1727 gv = case_data (c, indep_var);
1729 for (i=0; i< cmd->n_variables ; ++i)
1731 const struct variable *var = cmd->v_variables[i];
1732 const union value *val = case_data (c, var);
1733 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1734 struct group_statistics *gs;
1736 gs = hsh_find (grp_hash, (void *) gv);
1738 /* If the independent variable doesn't match either of the values
1739 for this case then move on to the next case */
1743 if (!var_is_value_missing (var, val, exclude))
1746 gs->sum += weight * val->f;
1747 gs->ssq += weight * pow2 (val->f);
1756 group_postcalc ( struct cmd_t_test *cmd )
1760 for (i = 0; i < cmd->n_variables ; ++i)
1762 const struct variable *var = cmd->v_variables[i];
1763 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1764 struct hsh_iterator g;
1765 struct group_statistics *gs;
1768 for (gs = hsh_first (grp_hash,&g);
1770 gs = hsh_next (grp_hash,&g))
1772 gs->mean = gs->sum / gs->n;
1774 gs->s_std_dev= sqrt (
1775 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1780 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1783 gs->se_mean = gs->std_dev / sqrt (gs->n);
1786 assert (count == 2);
1793 calculate (struct cmd_t_test *cmd,
1794 struct casereader *input, const struct dataset *ds)
1796 const struct dictionary *dict = dataset_dict (ds);
1797 struct ssbox stat_summary_box;
1798 struct trbox test_results_box;
1800 struct casereader *pass1, *pass2, *pass3;
1801 struct taint *taint;
1804 enum mv_class exclude = cmd->miss != TTS_INCLUDE ? MV_ANY : MV_SYSTEM;
1806 if (!casereader_peek (input, 0, &c))
1808 casereader_destroy (input);
1811 output_split_file_values (ds, &c);
1814 if ( cmd->miss == TTS_LISTWISE )
1815 input = casereader_create_filter_missing (input,
1820 input = casereader_create_filter_weight (input, dict, NULL, NULL);
1822 taint = taint_clone (casereader_get_taint (input));
1823 casereader_split (input, &pass1, &pass2);
1825 common_precalc (cmd);
1826 for (; casereader_read (pass1, &c); case_destroy (&c))
1827 common_calc (dict, &c, cmd, exclude);
1828 casereader_destroy (pass1);
1829 common_postcalc (cmd);
1834 one_sample_precalc (cmd);
1835 for (; casereader_read (pass2, &c); case_destroy (&c))
1836 one_sample_calc (dict, &c, cmd, exclude);
1837 one_sample_postcalc (cmd);
1840 paired_precalc (cmd);
1841 for (; casereader_read (pass2, &c); case_destroy (&c))
1842 paired_calc (dict, &c, cmd, exclude);
1843 paired_postcalc (cmd);
1846 pass3 = casereader_clone (pass2);
1848 group_precalc (cmd);
1849 for (; casereader_read (pass2, &c); case_destroy (&c))
1850 group_calc (dict, &c, cmd, exclude);
1851 group_postcalc (cmd);
1853 levene (dict, pass3, indep_var, cmd->n_variables, cmd->v_variables,
1857 casereader_destroy (pass2);
1859 if (!taint_has_tainted_successor (taint))
1861 ssbox_create (&stat_summary_box,cmd,mode);
1862 ssbox_populate (&stat_summary_box,cmd);
1863 ssbox_finalize (&stat_summary_box);
1865 if ( mode == T_PAIRED )
1868 trbox_create (&test_results_box,cmd,mode);
1869 trbox_populate (&test_results_box,cmd);
1870 trbox_finalize (&test_results_box);
1873 taint_destroy (taint);
1876 short which_group (const struct group_statistics *g,
1877 const struct group_properties *p);
1879 /* Return -1 if the id of a is less than b; +1 if greater than and
1882 compare_group_binary (const struct group_statistics *a,
1883 const struct group_statistics *b,
1884 const struct group_properties *p)
1889 if ( p->criterion == CMP_LE )
1891 /* less-than comparision is not meaningfull for
1892 alpha variables, so we shouldn't ever arrive here */
1893 assert (p->indep_width == 0 ) ;
1895 flag_a = ( a->id.f < p->v.critical_value ) ;
1896 flag_b = ( b->id.f < p->v.critical_value ) ;
1900 flag_a = which_group (a, p);
1901 flag_b = which_group (b, p);
1904 if (flag_a < flag_b )
1907 return (flag_a > flag_b);
1910 /* This is a degenerate case of a hash, since it can only return three possible
1911 values. It's really a comparison, being used as a hash function */
1914 hash_group_binary (const struct group_statistics *g,
1915 const struct group_properties *p)
1919 if ( p->criterion == CMP_LE )
1921 /* Not meaningfull to do a less than compare for alpha values ? */
1922 assert (p->indep_width == 0 ) ;
1923 flag = ( g->id.f < p->v.critical_value ) ;
1925 else if ( p->criterion == CMP_EQ)
1927 flag = which_group (g,p);
1935 /* return 0 if G belongs to group 0,
1936 1 if it belongs to group 1,
1937 2 if it belongs to neither group */
1939 which_group (const struct group_statistics *g,
1940 const struct group_properties *p)
1942 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1945 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))