1 /* PSPP - computes sample statistics. -*-c-*-
3 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
4 Written by John Williams <johnr.williams@stonebow.otago.ac.nz>.
5 Almost completly re-written by John Darrington 2004
7 This program is free software; you can redistribute it and/or
8 modify it under the terms of the GNU General Public License as
9 published by the Free Software Foundation; either version 2 of the
10 License, or (at your option) any later version.
12 This program is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
24 #include <gsl/gsl_cdf.h>
29 #include <data/case.h>
30 #include <data/casefile.h>
31 #include <data/dictionary.h>
32 #include <data/procedure.h>
33 #include <data/value-labels.h>
34 #include <data/variable.h>
35 #include <data/casefilter.h>
37 #include <language/command.h>
38 #include <language/dictionary/split-file.h>
39 #include <language/lexer/lexer.h>
40 #include <libpspp/alloc.h>
41 #include <libpspp/assertion.h>
42 #include <libpspp/compiler.h>
43 #include <libpspp/hash.h>
44 #include <libpspp/magic.h>
45 #include <libpspp/message.h>
46 #include <libpspp/message.h>
47 #include <libpspp/misc.h>
48 #include <libpspp/str.h>
49 #include <math/group-proc.h>
50 #include <math/levene.h>
51 #include <output/manager.h>
52 #include <output/table.h>
57 #define _(msgid) gettext (msgid)
65 +variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
67 missing=miss:!analysis/listwise,
68 incl:include/!exclude;
69 +format=fmt:!labels/nolabels;
70 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
76 /* Variable for the GROUPS subcommand, if given. */
77 static struct variable *indep_var;
85 struct group_properties
87 /* The comparison criterion */
88 enum comparison criterion;
90 /* The width of the independent variable */
94 /* The value of the independent variable at which groups are determined to
95 belong to one group or the other */
96 double critical_value;
99 /* The values of the independent variable for each group */
100 union value g_value[2];
106 static struct group_properties gp ;
110 /* PAIRS: Number of pairs to be compared ; each pair. */
111 static int n_pairs = 0 ;
114 /* The variables comprising the pair */
115 struct variable *v[2];
117 /* The number of valid variable pairs */
120 /* The sum of the members */
123 /* sum of squares of the members */
126 /* Std deviation of the members */
130 /* Sample Std deviation of the members */
133 /* The means of the members */
136 /* The correlation coefficient between the variables */
139 /* The sum of the differences */
142 /* The sum of the products */
145 /* The mean of the differences */
148 /* The sum of the squares of the differences */
151 /* The std deviation of the differences */
155 static struct pair *pairs=0;
157 static int parse_value (struct lexer *lexer, union value * v, int type) ;
159 /* Structures and Functions for the Statistics Summary Box */
161 typedef void populate_ssbox_func(struct ssbox *ssb,
162 struct cmd_t_test *cmd);
163 typedef void finalize_ssbox_func(struct ssbox *ssb);
169 populate_ssbox_func *populate;
170 finalize_ssbox_func *finalize;
175 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
177 /* Populate a ssbox according to cmd */
178 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
180 /* Submit and destroy a ssbox */
181 void ssbox_finalize(struct ssbox *ssb);
183 /* A function to create, populate and submit the Paired Samples Correlation
188 /* Structures and Functions for the Test Results Box */
191 typedef void populate_trbox_func(struct trbox *trb,
192 struct cmd_t_test *cmd);
193 typedef void finalize_trbox_func(struct trbox *trb);
197 populate_trbox_func *populate;
198 finalize_trbox_func *finalize;
202 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
204 /* Populate a ssbox according to cmd */
205 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
207 /* Submit and destroy a ssbox */
208 void trbox_finalize(struct trbox *trb);
210 /* Which mode was T-TEST invoked */
218 static int common_calc (const struct dictionary *dict,
219 const struct ccase *, void *,
220 const struct casefilter *filter);
221 static void common_precalc (struct cmd_t_test *);
222 static void common_postcalc (struct cmd_t_test *);
224 static int one_sample_calc (const struct dictionary *dict, const struct ccase *, void *, const struct casefilter *);
225 static void one_sample_precalc (struct cmd_t_test *);
226 static void one_sample_postcalc (struct cmd_t_test *);
228 static int paired_calc (const struct dictionary *dict, const struct ccase *,
229 struct cmd_t_test*, const struct casefilter *);
230 static void paired_precalc (struct cmd_t_test *);
231 static void paired_postcalc (struct cmd_t_test *);
233 static void group_precalc (struct cmd_t_test *);
234 static int group_calc (const struct dictionary *dict, const struct ccase *,
235 struct cmd_t_test *, const struct casefilter *);
236 static void group_postcalc (struct cmd_t_test *);
239 static bool calculate(const struct ccase *first,
240 const struct casefile *cf, void *_mode,
241 const struct dataset *ds);
245 static struct cmd_t_test cmd;
247 static bool bad_weight_warn = false;
250 static int compare_group_binary(const struct group_statistics *a,
251 const struct group_statistics *b,
252 const struct group_properties *p);
255 static unsigned hash_group_binary(const struct group_statistics *g,
256 const struct group_properties *p);
261 cmd_t_test (struct lexer *lexer, struct dataset *ds)
265 if ( !parse_t_test (lexer, ds, &cmd, NULL) )
268 if (! cmd.sbc_criteria)
273 if (cmd.sbc_testval) ++m;
274 if (cmd.sbc_groups) ++m;
275 if (cmd.sbc_pairs) ++m;
280 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
289 else if (cmd.sbc_groups)
294 if ( mode == T_PAIRED)
296 if (cmd.sbc_variables)
298 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
304 /* Iterate through the pairs and put each variable that is a
305 member of a pair into cmd.v_variables */
308 struct hsh_iterator hi;
309 struct hsh_table *hash;
312 hash = hsh_create (n_pairs, compare_var_names, hash_var_name, 0, 0);
314 for (i=0; i < n_pairs; ++i)
316 hsh_insert(hash,pairs[i].v[0]);
317 hsh_insert(hash,pairs[i].v[1]);
320 assert(cmd.n_variables == 0);
321 cmd.n_variables = hsh_count(hash);
323 cmd.v_variables = xnrealloc (cmd.v_variables, cmd.n_variables,
324 sizeof *cmd.v_variables);
325 /* Iterate through the hash */
326 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
328 v=hsh_next(hash,&hi) )
329 cmd.v_variables[i++]=v;
334 else if ( !cmd.sbc_variables)
336 msg(SE, _("One or more VARIABLES must be specified."));
341 bad_weight_warn = true;
343 ok = multipass_procedure_with_splits (ds, calculate, &cmd);
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, void *aux UNUSED)
367 int n_group_values=0;
369 lex_match (lexer, '=');
371 indep_var = parse_variable (lexer, dataset_dict (ds));
374 lex_error (lexer, "expecting variable name in GROUPS subcommand");
378 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
380 msg (SE, _("Long string variable %s is not valid here."),
385 if (!lex_match (lexer, '('))
387 if (indep_var->type == NUMERIC)
389 gp.v.g_value[0].f = 1;
390 gp.v.g_value[1].f = 2;
392 gp.criterion = CMP_EQ;
400 msg (SE, _("When applying GROUPS to a string variable, two "
401 "values must be specified."));
406 if (!parse_value (lexer, &gp.v.g_value[0], indep_var->type))
409 lex_match (lexer, ',');
410 if (lex_match (lexer, ')'))
412 if (indep_var->type != NUMERIC)
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], indep_var->type))
430 if (!lex_force_match (lexer, ')'))
433 if ( n_group_values == 2 )
434 gp.criterion = CMP_EQ ;
436 gp.criterion = CMP_LE ;
444 tts_custom_pairs (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
446 struct variable **vars;
448 size_t n_pairs_local;
450 size_t n_before_WITH;
451 size_t n_after_WITH = SIZE_MAX;
452 int paired ; /* Was the PAIRED keyword given ? */
454 lex_match (lexer, '=');
457 if (!parse_variables (lexer, dataset_dict (ds), &vars, &n_vars,
458 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
466 if (lex_match (lexer, T_WITH))
468 n_before_WITH = n_vars;
469 if (!parse_variables (lexer, dataset_dict (ds), &vars, &n_vars,
470 PV_DUPLICATE | PV_APPEND
471 | PV_NUMERIC | PV_NO_SCRATCH))
476 n_after_WITH = n_vars - n_before_WITH;
479 paired = (lex_match (lexer, '(') && lex_match_id (lexer, "PAIRED") && lex_match (lexer, ')'));
481 /* Determine the number of pairs needed */
484 if (n_before_WITH != n_after_WITH)
487 msg (SE, _("PAIRED was specified but the number of variables "
488 "preceding WITH (%d) did not match the number "
490 n_before_WITH, n_after_WITH );
493 n_pairs_local = n_before_WITH;
495 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
497 n_pairs_local = n_before_WITH * n_after_WITH ;
499 else /* Neither WITH nor PAIRED keyword given */
504 msg (SE, _("At least two variables must be specified "
509 /* how many ways can you pick 2 from n_vars ? */
510 n_pairs_local = n_vars * (n_vars - 1) / 2;
514 /* Allocate storage for the pairs */
515 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
517 /* Populate the pairs with the appropriate variables */
522 assert(n_pairs_local == n_vars / 2);
523 for (i = 0; i < n_pairs_local; ++i)
525 pairs[i].v[n_pairs] = vars[i];
526 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
529 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
534 for(i=0 ; i < n_before_WITH ; ++i )
536 for(j=0 ; j < n_after_WITH ; ++j)
538 pairs[p].v[0] = vars[i];
539 pairs[p].v[1] = vars[j+n_before_WITH];
544 else /* Neither WITH nor PAIRED given */
549 for(i=0 ; i < n_vars ; ++i )
551 for(j=i+1 ; j < n_vars ; ++j)
553 pairs[p].v[0] = vars[i];
554 pairs[p].v[1] = vars[j];
560 n_pairs+=n_pairs_local;
566 /* Parses the current token (numeric or string, depending on type)
567 value v and returns success. */
569 parse_value (struct lexer *lexer, union value * v, int type )
573 if (!lex_force_num (lexer))
575 v->f = lex_tokval (lexer);
579 if (!lex_force_string (lexer))
581 strncpy (v->s, ds_cstr (lex_tokstr (lexer)), ds_length (lex_tokstr (lexer)));
590 /* Implementation of the SSBOX object */
592 void ssbox_base_init(struct ssbox *this, int cols,int rows);
594 void ssbox_base_finalize(struct ssbox *ssb);
596 void ssbox_one_sample_init(struct ssbox *this,
597 struct cmd_t_test *cmd );
599 void ssbox_independent_samples_init(struct ssbox *this,
600 struct cmd_t_test *cmd);
602 void ssbox_paired_init(struct ssbox *this,
603 struct cmd_t_test *cmd);
606 /* Factory to create an ssbox */
608 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
613 ssbox_one_sample_init(ssb,cmd);
616 ssbox_independent_samples_init(ssb,cmd);
619 ssbox_paired_init(ssb,cmd);
628 /* Despatcher for the populate method */
630 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
632 ssb->populate(ssb,cmd);
636 /* Despatcher for finalize */
638 ssbox_finalize(struct ssbox *ssb)
644 /* Submit the box and clear up */
646 ssbox_base_finalize(struct ssbox *ssb)
653 /* Initialize a ssbox struct */
655 ssbox_base_init(struct ssbox *this, int cols,int rows)
657 this->finalize = ssbox_base_finalize;
658 this->t = tab_create (cols, rows, 0);
660 tab_columns (this->t, SOM_COL_DOWN, 1);
661 tab_headers (this->t,0,0,1,0);
662 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
663 tab_hline(this->t, TAL_2,0,cols-1,1);
664 tab_dim (this->t, tab_natural_dimensions);
667 void ssbox_one_sample_populate(struct ssbox *ssb,
668 struct cmd_t_test *cmd);
670 /* Initialize the one_sample ssbox */
672 ssbox_one_sample_init(struct ssbox *this,
673 struct cmd_t_test *cmd )
676 const int vsize=cmd->n_variables+1;
678 this->populate = ssbox_one_sample_populate;
680 ssbox_base_init(this, hsize,vsize);
681 tab_title (this->t, _("One-Sample Statistics"));
682 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
683 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
684 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
685 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
686 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
689 void ssbox_independent_samples_populate(struct ssbox *ssb,
690 struct cmd_t_test *cmd);
692 /* Initialize the independent samples ssbox */
694 ssbox_independent_samples_init(struct ssbox *this,
695 struct cmd_t_test *cmd)
698 int vsize = cmd->n_variables*2 +1;
700 this->populate = ssbox_independent_samples_populate;
702 ssbox_base_init(this, hsize,vsize);
703 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
704 tab_title (this->t, _("Group Statistics"));
705 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
706 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
707 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
708 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
709 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
713 /* Populate the ssbox for independent samples */
715 ssbox_independent_samples_populate(struct ssbox *ssb,
716 struct cmd_t_test *cmd)
722 double indep_value[2];
724 char prefix[2][3]={"",""};
726 if ( indep_var->type == NUMERIC )
728 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
729 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
733 val_lab0 = gp.v.g_value[0].s;
734 val_lab1 = gp.v.g_value[1].s;
737 if (gp.criterion == CMP_LE )
739 strcpy(prefix[0],"< ");
740 strcpy(prefix[1],">=");
741 indep_value[0] = gp.v.critical_value;
742 indep_value[1] = gp.v.critical_value;
746 indep_value[0] = gp.v.g_value[0].f;
747 indep_value[1] = gp.v.g_value[1].f;
752 for (i=0; i < cmd->n_variables; ++i)
754 struct variable *var = cmd->v_variables[i];
755 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
758 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
761 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
762 "%s%s", prefix[0], val_lab0);
764 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
765 "%s%g", prefix[0], indep_value[0]);
769 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
770 "%s%s", prefix[1], val_lab1);
772 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
773 "%s%g", prefix[1], indep_value[1]);
776 /* Fill in the group statistics */
777 for ( count = 0 ; count < 2 ; ++count )
779 union value search_val;
781 struct group_statistics *gs;
783 if ( gp.criterion == CMP_LE )
787 /* less than ( < ) case */
788 search_val.f = gp.v.critical_value - 1.0;
793 search_val.f = gp.v.critical_value + 1.0;
798 search_val = gp.v.g_value[count];
801 gs = hsh_find(grp_hash, (void *) &search_val);
804 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
805 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
806 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
807 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
813 void ssbox_paired_populate(struct ssbox *ssb,
814 struct cmd_t_test *cmd);
816 /* Initialize the paired values ssbox */
818 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
822 int vsize = n_pairs*2+1;
824 this->populate = ssbox_paired_populate;
826 ssbox_base_init(this, hsize,vsize);
827 tab_title (this->t, _("Paired Sample Statistics"));
828 tab_vline(this->t,TAL_GAP,1,0,vsize-1);
829 tab_vline(this->t,TAL_2,2,0,vsize-1);
830 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
831 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
832 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
833 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
837 /* Populate the ssbox for paired values */
839 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
845 for (i=0; i < n_pairs; ++i)
849 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
851 for (j=0 ; j < 2 ; ++j)
853 struct group_statistics *gs;
855 gs = &group_proc_get (pairs[i].v[j])->ugs;
859 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
862 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
863 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
864 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
865 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
871 /* Populate the one sample ssbox */
873 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
879 for (i=0; i < cmd->n_variables; ++i)
881 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
883 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
884 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
885 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
886 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
887 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
894 /* Implementation of the Test Results box struct */
896 void trbox_base_init(struct trbox *self,size_t n_vars, int cols);
897 void trbox_base_finalize(struct trbox *trb);
899 void trbox_independent_samples_init(struct trbox *trb,
900 struct cmd_t_test *cmd );
902 void trbox_independent_samples_populate(struct trbox *trb,
903 struct cmd_t_test *cmd);
905 void trbox_one_sample_init(struct trbox *self,
906 struct cmd_t_test *cmd );
908 void trbox_one_sample_populate(struct trbox *trb,
909 struct cmd_t_test *cmd);
911 void trbox_paired_init(struct trbox *self,
912 struct cmd_t_test *cmd );
914 void trbox_paired_populate(struct trbox *trb,
915 struct cmd_t_test *cmd);
919 /* Create a trbox according to mode*/
921 trbox_create(struct trbox *trb,
922 struct cmd_t_test *cmd, int mode)
927 trbox_one_sample_init(trb,cmd);
930 trbox_independent_samples_init(trb,cmd);
933 trbox_paired_init(trb,cmd);
940 /* Populate a trbox according to cmd */
942 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
944 trb->populate(trb,cmd);
947 /* Submit and destroy a trbox */
949 trbox_finalize(struct trbox *trb)
954 /* Initialize the independent samples trbox */
956 trbox_independent_samples_init(struct trbox *self,
957 struct cmd_t_test *cmd UNUSED)
960 const int vsize=cmd->n_variables*2+3;
963 self->populate = trbox_independent_samples_populate;
965 trbox_base_init(self,cmd->n_variables*2,hsize);
966 tab_title(self->t,_("Independent Samples Test"));
967 tab_hline(self->t,TAL_1,2,hsize-1,1);
968 tab_vline(self->t,TAL_2,2,0,vsize-1);
969 tab_vline(self->t,TAL_1,4,0,vsize-1);
970 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
971 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
972 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
973 tab_joint_text(self->t, 2, 0, 3, 0,
974 TAB_CENTER,_("Levene's Test for Equality of Variances"));
975 tab_joint_text(self->t, 4,0,hsize-1,0,
976 TAB_CENTER,_("t-test for Equality of Means"));
978 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
979 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
980 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
981 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
982 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
983 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
984 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
985 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
986 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
988 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
989 _("%g%% Confidence Interval of the Difference"),
990 cmd->criteria*100.0);
994 /* Populate the independent samples trbox */
996 trbox_independent_samples_populate(struct trbox *self,
997 struct cmd_t_test *cmd )
1002 for (i=0; i < cmd->n_variables; ++i)
1011 double pooled_variance;
1012 double std_err_diff;
1015 struct variable *var = cmd->v_variables[i];
1016 struct group_proc *grp_data = group_proc_get (var);
1018 struct hsh_table *grp_hash = grp_data->group_hash;
1020 struct group_statistics *gs0 ;
1021 struct group_statistics *gs1 ;
1023 union value search_val;
1025 if ( gp.criterion == CMP_LE )
1026 search_val.f = gp.v.critical_value - 1.0;
1028 search_val = gp.v.g_value[0];
1030 gs0 = hsh_find(grp_hash, (void *) &search_val);
1033 if ( gp.criterion == CMP_LE )
1034 search_val.f = gp.v.critical_value + 1.0;
1036 search_val = gp.v.g_value[1];
1038 gs1 = hsh_find(grp_hash, (void *) &search_val);
1042 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1044 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1047 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1049 /* Now work out the significance of the Levene test */
1050 df1 = 1; df2 = grp_data->ugs.n - 2;
1051 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1053 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1055 df = gs0->n + gs1->n - 2.0 ;
1056 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1058 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1060 (gs1->n )*pow2(gs1->s_std_dev)
1063 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1064 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1066 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1068 p = gsl_cdf_tdist_P(t, df);
1069 q = gsl_cdf_tdist_Q(t, df);
1071 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1073 mean_diff = gs0->mean - gs1->mean;
1074 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1077 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1078 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1081 /* Now work out the confidence interval */
1082 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1084 t = gsl_cdf_tdist_Qinv(q,df);
1085 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1086 mean_diff - t * std_err_diff, 8, 3);
1088 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1089 mean_diff + t * std_err_diff, 8, 3);
1094 /* Now for the \sigma_1 != \sigma_2 case */
1095 tab_text (self->t, 1, i*2+3+1,
1096 TAB_LEFT, _("Equal variances not assumed"));
1099 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1100 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1102 t = mean_diff / sqrt(se2) ;
1103 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1106 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1110 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1114 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1116 p = gsl_cdf_tdist_P(t, df);
1117 q = gsl_cdf_tdist_Q(t, df);
1119 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1121 /* Now work out the confidence interval */
1122 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1124 t = gsl_cdf_tdist_Qinv(q, df);
1126 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1129 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1132 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1133 mean_diff - t * std_err_diff, 8, 3);
1135 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1136 mean_diff + t * std_err_diff, 8, 3);
1142 /* Initialize the paired samples trbox */
1144 trbox_paired_init(struct trbox *self,
1145 struct cmd_t_test *cmd UNUSED)
1149 const int vsize=n_pairs+3;
1151 self->populate = trbox_paired_populate;
1153 trbox_base_init(self,n_pairs,hsize);
1154 tab_title (self->t, _("Paired Samples Test"));
1155 tab_hline(self->t,TAL_1,2,6,1);
1156 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1157 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1158 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1159 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1160 tab_hline(self->t,TAL_1,5,6, 2);
1161 tab_vline(self->t,TAL_GAP,6,0,1);
1163 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1164 _("%g%% Confidence Interval of the Difference"),
1165 cmd->criteria*100.0);
1167 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1168 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1169 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1170 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1171 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1172 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1173 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1174 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1177 /* Populate the paired samples trbox */
1179 trbox_paired_populate(struct trbox *trb,
1180 struct cmd_t_test *cmd UNUSED)
1184 for (i=0; i < n_pairs; ++i)
1189 double n = pairs[i].n;
1193 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1195 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1196 pairs[i].v[0]->name, pairs[i].v[1]->name);
1198 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1200 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1203 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1204 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1206 /* Now work out the confidence interval */
1207 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1209 t = gsl_cdf_tdist_Qinv(q, df);
1211 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1212 pairs[i].mean_diff - t * se_mean , 8, 4);
1214 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1215 pairs[i].mean_diff + t * se_mean , 8, 4);
1217 t = (pairs[i].mean[0] - pairs[i].mean[1])
1219 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1220 2 * pairs[i].correlation *
1221 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1225 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1227 /* Degrees of freedom */
1228 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1230 p = gsl_cdf_tdist_P(t,df);
1231 q = gsl_cdf_tdist_P(t,df);
1233 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1238 /* Initialize the one sample trbox */
1240 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1243 const int vsize=cmd->n_variables+3;
1245 self->populate = trbox_one_sample_populate;
1247 trbox_base_init(self, cmd->n_variables,hsize);
1248 tab_title (self->t, _("One-Sample Test"));
1249 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1250 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1252 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1253 _("Test Value = %f"), cmd->n_testval[0]);
1255 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1258 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1259 _("%g%% Confidence Interval of the Difference"),
1260 cmd->criteria*100.0);
1262 tab_vline(self->t,TAL_GAP,6,1,1);
1263 tab_hline(self->t,TAL_1,5,6,2);
1264 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1265 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1266 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1267 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1268 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1269 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1274 /* Populate the one sample trbox */
1276 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1282 for (i=0; i < cmd->n_variables; ++i)
1287 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1290 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1292 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1294 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1296 /* degrees of freedom */
1299 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1301 p = gsl_cdf_tdist_P(t, df);
1302 q = gsl_cdf_tdist_Q(t, df);
1304 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1306 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1308 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1311 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1312 t = gsl_cdf_tdist_Qinv(q, df);
1314 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1315 gs->mean_diff - t * gs->se_mean, 8,4);
1317 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1318 gs->mean_diff + t * gs->se_mean, 8,4);
1322 /* Base initializer for the generalized trbox */
1324 trbox_base_init(struct trbox *self, size_t data_rows, int cols)
1326 const size_t rows = 3 + data_rows;
1328 self->finalize = trbox_base_finalize;
1329 self->t = tab_create (cols, rows, 0);
1330 tab_headers (self->t,0,0,3,0);
1331 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1332 tab_hline(self->t, TAL_2,0,cols-1,3);
1333 tab_dim (self->t, tab_natural_dimensions);
1337 /* Base finalizer for the trbox */
1339 trbox_base_finalize(struct trbox *trb)
1345 /* Create , populate and submit the Paired Samples Correlation box */
1349 const int rows=1+n_pairs;
1353 struct tab_table *table;
1355 table = tab_create (cols,rows,0);
1357 tab_columns (table, SOM_COL_DOWN, 1);
1358 tab_headers (table,0,0,1,0);
1359 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1360 tab_hline(table, TAL_2, 0, cols - 1, 1);
1361 tab_vline(table, TAL_2, 2, 0, rows - 1);
1362 tab_dim(table, tab_natural_dimensions);
1363 tab_title(table, _("Paired Samples Correlations"));
1365 /* column headings */
1366 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1367 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1368 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1370 for (i=0; i < n_pairs; ++i)
1374 double df = pairs[i].n -2;
1376 double correlation_t =
1377 pairs[i].correlation * sqrt(df) /
1378 sqrt(1 - pow2(pairs[i].correlation));
1382 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1385 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1386 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1390 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1391 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1393 p = gsl_cdf_tdist_P(correlation_t, df);
1394 q = gsl_cdf_tdist_Q(correlation_t, df);
1396 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1405 /* Calculation Implementation */
1407 /* Per case calculations common to all variants of the T test */
1409 common_calc (const struct dictionary *dict,
1410 const struct ccase *c,
1412 const struct casefilter *filter)
1415 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1417 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1420 /* Listwise has to be implicit if the independent variable is missing ?? */
1421 if ( cmd->sbc_groups )
1423 if ( casefilter_variable_missing (filter, c, indep_var) )
1427 for(i = 0; i < cmd->n_variables ; ++i)
1429 struct variable *v = cmd->v_variables[i];
1431 if (! casefilter_variable_missing (filter, c, v) )
1433 struct group_statistics *gs;
1434 const union value *val = case_data (c, v->fv);
1435 gs = &group_proc_get (cmd->v_variables[i])->ugs;
1438 gs->sum += weight * val->f;
1439 gs->ssq += weight * val->f * val->f;
1445 /* Pre calculations common to all variants of the T test */
1447 common_precalc ( struct cmd_t_test *cmd )
1451 for(i=0; i< cmd->n_variables ; ++i)
1453 struct group_statistics *gs;
1454 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1463 /* Post calculations common to all variants of the T test */
1465 common_postcalc (struct cmd_t_test *cmd)
1469 for(i=0; i< cmd->n_variables ; ++i)
1471 struct group_statistics *gs;
1472 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1474 gs->mean=gs->sum / gs->n;
1475 gs->s_std_dev= sqrt(
1476 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1481 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1484 gs->se_mean = gs->std_dev / sqrt(gs->n);
1485 gs->mean_diff= gs->sum_diff / gs->n;
1489 /* Per case calculations for one sample t test */
1491 one_sample_calc (const struct dictionary *dict,
1492 const struct ccase *c, void *cmd_,
1493 const struct casefilter *filter)
1497 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1499 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1502 for(i=0; i< cmd->n_variables ; ++i)
1504 struct group_statistics *gs;
1505 struct variable *v = cmd->v_variables[i];
1506 const union value *val = case_data (c, v->fv);
1508 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1510 if ( ! casefilter_variable_missing (filter, c, v))
1511 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1517 /* Pre calculations for one sample t test */
1519 one_sample_precalc ( struct cmd_t_test *cmd )
1523 for(i=0; i< cmd->n_variables ; ++i)
1525 struct group_statistics *gs;
1526 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1532 /* Post calculations for one sample t test */
1534 one_sample_postcalc (struct cmd_t_test *cmd)
1538 for(i=0; i< cmd->n_variables ; ++i)
1540 struct group_statistics *gs;
1541 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1543 gs->mean_diff = gs->sum_diff / gs->n ;
1550 paired_precalc (struct cmd_t_test *cmd UNUSED)
1554 for(i=0; i < n_pairs ; ++i )
1557 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1558 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1559 pairs[i].sum_of_prod = 0;
1560 pairs[i].correlation = 0;
1561 pairs[i].sum_of_diffs = 0;
1562 pairs[i].ssq_diffs = 0;
1569 paired_calc (const struct dictionary *dict, const struct ccase *c,
1570 struct cmd_t_test *cmd UNUSED, const struct casefilter *filter)
1574 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1576 for(i=0; i < n_pairs ; ++i )
1578 struct variable *v0 = pairs[i].v[0];
1579 struct variable *v1 = pairs[i].v[1];
1581 const union value *val0 = case_data (c, v0->fv);
1582 const union value *val1 = case_data (c, v1->fv);
1584 if ( ! casefilter_variable_missing (filter, c, v0) &&
1585 ! casefilter_variable_missing (filter, c, v1) )
1587 pairs[i].n += weight;
1588 pairs[i].sum[0] += weight * val0->f;
1589 pairs[i].sum[1] += weight * val1->f;
1591 pairs[i].ssq[0] += weight * pow2(val0->f);
1592 pairs[i].ssq[1] += weight * pow2(val1->f);
1594 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1596 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1597 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1605 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1609 for(i=0; i < n_pairs ; ++i )
1612 const double n = pairs[i].n;
1614 for (j=0; j < 2 ; ++j)
1616 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1617 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1618 pow2(pairs[i].mean[j]))
1621 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1622 pow2(pairs[i].mean[j]))
1626 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1627 pairs[i].mean[0] * pairs[i].mean[1] ;
1628 /* correlation now actually contains the covariance */
1630 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1631 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1633 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1635 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1636 ( pairs[i].ssq_diffs / n )
1638 pow2(pairs[i].mean_diff )
1644 group_precalc (struct cmd_t_test *cmd )
1649 for(i=0; i< cmd->n_variables ; ++i)
1651 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1653 /* There's always 2 groups for a T - TEST */
1656 gp.indep_width = indep_var->width;
1658 ttpr->group_hash = hsh_create(2,
1659 (hsh_compare_func *) compare_group_binary,
1660 (hsh_hash_func *) hash_group_binary,
1661 (hsh_free_func *) free_group,
1664 for (j=0 ; j < 2 ; ++j)
1667 struct group_statistics *gs = xmalloc (sizeof *gs);
1673 if ( gp.criterion == CMP_EQ )
1675 gs->id = gp.v.g_value[j];
1680 gs->id.f = gp.v.critical_value - 1.0 ;
1682 gs->id.f = gp.v.critical_value + 1.0 ;
1685 hsh_insert ( ttpr->group_hash, (void *) gs );
1693 group_calc (const struct dictionary *dict,
1694 const struct ccase *c, struct cmd_t_test *cmd,
1695 const struct casefilter *filter)
1699 const double weight =
1700 dict_get_case_weight (dict, c, &bad_weight_warn);
1702 const union value *gv;
1704 if ( casefilter_variable_missing (filter, c, indep_var))
1707 gv = case_data (c, indep_var->fv);
1709 for(i=0; i< cmd->n_variables ; ++i)
1711 struct variable *var = cmd->v_variables[i];
1712 const union value *val = case_data (c, var->fv);
1713 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1714 struct group_statistics *gs;
1716 gs = hsh_find(grp_hash, (void *) gv);
1718 /* If the independent variable doesn't match either of the values
1719 for this case then move on to the next case */
1723 if ( ! casefilter_variable_missing (filter, c, var) )
1726 gs->sum += weight * val->f;
1727 gs->ssq += weight * pow2(val->f);
1736 group_postcalc ( struct cmd_t_test *cmd )
1740 for (i = 0; i < cmd->n_variables ; ++i)
1742 struct variable *var = cmd->v_variables[i];
1743 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1744 struct hsh_iterator g;
1745 struct group_statistics *gs;
1748 for (gs = hsh_first (grp_hash,&g);
1750 gs = hsh_next(grp_hash,&g))
1752 gs->mean = gs->sum / gs->n;
1754 gs->s_std_dev= sqrt(
1755 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1760 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1763 gs->se_mean = gs->std_dev / sqrt(gs->n);
1773 calculate(const struct ccase *first, const struct casefile *cf,
1774 void *cmd_, const struct dataset *ds)
1776 const struct dictionary *dict = dataset_dict (ds);
1777 struct ssbox stat_summary_box;
1778 struct trbox test_results_box;
1780 struct casereader *r;
1783 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1785 struct casefilter *filter = casefilter_create (cmd->miss != TTS_INCLUDE,
1788 if ( cmd->miss == TTS_LISTWISE )
1789 casefilter_add_variables (filter,
1790 cmd->v_variables, cmd->n_variables);
1792 output_split_file_values (ds, first);
1793 common_precalc (cmd);
1794 for(r = casefile_get_reader (cf, filter);
1795 casereader_read (r, &c) ;
1798 common_calc (dict, &c, cmd, filter);
1801 casereader_destroy (r);
1802 common_postcalc (cmd);
1807 one_sample_precalc (cmd);
1808 for(r = casefile_get_reader (cf, filter);
1809 casereader_read (r, &c) ;
1812 one_sample_calc (dict, &c, cmd, filter);
1814 casereader_destroy (r);
1815 one_sample_postcalc (cmd);
1818 paired_precalc(cmd);
1819 for(r = casefile_get_reader (cf, filter);
1820 casereader_read (r, &c) ;
1823 paired_calc (dict, &c, cmd, filter);
1825 casereader_destroy (r);
1826 paired_postcalc (cmd);
1832 for(r = casefile_get_reader (cf, filter);
1833 casereader_read (r, &c) ;
1836 group_calc (dict, &c, cmd, filter);
1838 casereader_destroy (r);
1839 group_postcalc(cmd);
1841 levene (dict, cf, indep_var, cmd->n_variables, cmd->v_variables,
1846 casefilter_destroy (filter);
1848 ssbox_create(&stat_summary_box,cmd,mode);
1849 ssbox_populate(&stat_summary_box,cmd);
1850 ssbox_finalize(&stat_summary_box);
1852 if ( mode == T_PAIRED)
1855 trbox_create(&test_results_box,cmd,mode);
1856 trbox_populate(&test_results_box,cmd);
1857 trbox_finalize(&test_results_box);
1862 short which_group(const struct group_statistics *g,
1863 const struct group_properties *p);
1865 /* Return -1 if the id of a is less than b; +1 if greater than and
1868 compare_group_binary(const struct group_statistics *a,
1869 const struct group_statistics *b,
1870 const struct group_properties *p)
1875 if ( p->criterion == CMP_LE )
1877 /* less-than-or-equal comparision is not meaningfull for
1878 alpha variables, so we shouldn't ever arrive here */
1879 assert(p->indep_width == 0 ) ;
1881 flag_a = ( a->id.f < p->v.critical_value ) ;
1882 flag_b = ( b->id.f < p->v.critical_value ) ;
1886 flag_a = which_group(a, p);
1887 flag_b = which_group(b, p);
1890 if (flag_a < flag_b )
1893 return (flag_a > flag_b);
1896 /* This is a degenerate case of a hash, since it can only return three possible
1897 values. It's really a comparison, being used as a hash function */
1900 hash_group_binary(const struct group_statistics *g,
1901 const struct group_properties *p)
1905 if ( p->criterion == CMP_LE )
1907 /* Not meaningfull to do a less than compare for alpha values ? */
1908 assert(p->indep_width == 0 ) ;
1909 flag = ( g->id.f < p->v.critical_value ) ;
1911 else if ( p->criterion == CMP_EQ)
1913 flag = which_group(g,p);
1921 /* return 0 if G belongs to group 0,
1922 1 if it belongs to group 1,
1923 2 if it belongs to neither group */
1925 which_group(const struct group_statistics *g,
1926 const struct group_properties *p)
1929 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1932 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))