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 (var_is_long_string (indep_var))
380 msg (SE, _("Long string variable %s is not valid here."),
381 var_get_name (indep_var));
385 if (!lex_match (lexer, '('))
387 if (var_is_numeric (indep_var))
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], var_get_type (indep_var)))
409 lex_match (lexer, ',');
410 if (lex_match (lexer, ')'))
412 if (var_is_alpha (indep_var))
414 msg (SE, _("When applying GROUPS to a string variable, two "
415 "values must be specified."));
418 gp.criterion = CMP_LE;
419 gp.v.critical_value = gp.v.g_value[0].f;
425 if (!parse_value (lexer, &gp.v.g_value[1], var_get_type (indep_var)))
429 if (!lex_force_match (lexer, ')'))
432 if ( n_group_values == 2 )
433 gp.criterion = CMP_EQ ;
435 gp.criterion = CMP_LE ;
443 tts_custom_pairs (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
445 struct variable **vars;
447 size_t n_pairs_local;
449 size_t n_before_WITH;
450 size_t n_after_WITH = SIZE_MAX;
451 int paired ; /* Was the PAIRED keyword given ? */
453 lex_match (lexer, '=');
456 if (!parse_variables (lexer, dataset_dict (ds), &vars, &n_vars,
457 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
465 if (lex_match (lexer, T_WITH))
467 n_before_WITH = n_vars;
468 if (!parse_variables (lexer, dataset_dict (ds), &vars, &n_vars,
469 PV_DUPLICATE | PV_APPEND
470 | PV_NUMERIC | PV_NO_SCRATCH))
475 n_after_WITH = n_vars - n_before_WITH;
478 paired = (lex_match (lexer, '(') && lex_match_id (lexer, "PAIRED") && lex_match (lexer, ')'));
480 /* Determine the number of pairs needed */
483 if (n_before_WITH != n_after_WITH)
486 msg (SE, _("PAIRED was specified but the number of variables "
487 "preceding WITH (%d) did not match the number "
489 n_before_WITH, n_after_WITH );
492 n_pairs_local = n_before_WITH;
494 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
496 n_pairs_local = n_before_WITH * n_after_WITH ;
498 else /* Neither WITH nor PAIRED keyword given */
503 msg (SE, _("At least two variables must be specified "
508 /* how many ways can you pick 2 from n_vars ? */
509 n_pairs_local = n_vars * (n_vars - 1) / 2;
513 /* Allocate storage for the pairs */
514 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
516 /* Populate the pairs with the appropriate variables */
521 assert(n_pairs_local == n_vars / 2);
522 for (i = 0; i < n_pairs_local; ++i)
524 pairs[i].v[n_pairs] = vars[i];
525 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
528 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
533 for(i=0 ; i < n_before_WITH ; ++i )
535 for(j=0 ; j < n_after_WITH ; ++j)
537 pairs[p].v[0] = vars[i];
538 pairs[p].v[1] = vars[j+n_before_WITH];
543 else /* Neither WITH nor PAIRED given */
548 for(i=0 ; i < n_vars ; ++i )
550 for(j=i+1 ; j < n_vars ; ++j)
552 pairs[p].v[0] = vars[i];
553 pairs[p].v[1] = vars[j];
559 n_pairs+=n_pairs_local;
565 /* Parses the current token (numeric or string, depending on type)
566 value v and returns success. */
568 parse_value (struct lexer *lexer, union value * v, int type )
572 if (!lex_force_num (lexer))
574 v->f = lex_tokval (lexer);
578 if (!lex_force_string (lexer))
580 strncpy (v->s, ds_cstr (lex_tokstr (lexer)), ds_length (lex_tokstr (lexer)));
589 /* Implementation of the SSBOX object */
591 void ssbox_base_init(struct ssbox *this, int cols,int rows);
593 void ssbox_base_finalize(struct ssbox *ssb);
595 void ssbox_one_sample_init(struct ssbox *this,
596 struct cmd_t_test *cmd );
598 void ssbox_independent_samples_init(struct ssbox *this,
599 struct cmd_t_test *cmd);
601 void ssbox_paired_init(struct ssbox *this,
602 struct cmd_t_test *cmd);
605 /* Factory to create an ssbox */
607 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
612 ssbox_one_sample_init(ssb,cmd);
615 ssbox_independent_samples_init(ssb,cmd);
618 ssbox_paired_init(ssb,cmd);
627 /* Despatcher for the populate method */
629 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
631 ssb->populate(ssb,cmd);
635 /* Despatcher for finalize */
637 ssbox_finalize(struct ssbox *ssb)
643 /* Submit the box and clear up */
645 ssbox_base_finalize(struct ssbox *ssb)
652 /* Initialize a ssbox struct */
654 ssbox_base_init(struct ssbox *this, int cols,int rows)
656 this->finalize = ssbox_base_finalize;
657 this->t = tab_create (cols, rows, 0);
659 tab_columns (this->t, SOM_COL_DOWN, 1);
660 tab_headers (this->t,0,0,1,0);
661 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
662 tab_hline(this->t, TAL_2,0,cols-1,1);
663 tab_dim (this->t, tab_natural_dimensions);
666 void ssbox_one_sample_populate(struct ssbox *ssb,
667 struct cmd_t_test *cmd);
669 /* Initialize the one_sample ssbox */
671 ssbox_one_sample_init(struct ssbox *this,
672 struct cmd_t_test *cmd )
675 const int vsize=cmd->n_variables+1;
677 this->populate = ssbox_one_sample_populate;
679 ssbox_base_init(this, hsize,vsize);
680 tab_title (this->t, _("One-Sample Statistics"));
681 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
682 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
683 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
684 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
685 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
688 void ssbox_independent_samples_populate(struct ssbox *ssb,
689 struct cmd_t_test *cmd);
691 /* Initialize the independent samples ssbox */
693 ssbox_independent_samples_init(struct ssbox *this,
694 struct cmd_t_test *cmd)
697 int vsize = cmd->n_variables*2 +1;
699 this->populate = ssbox_independent_samples_populate;
701 ssbox_base_init(this, hsize,vsize);
702 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
703 tab_title (this->t, _("Group Statistics"));
704 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, var_get_name (indep_var));
705 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
706 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
707 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
708 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
712 /* Populate the ssbox for independent samples */
714 ssbox_independent_samples_populate(struct ssbox *ssb,
715 struct cmd_t_test *cmd)
721 double indep_value[2];
723 char prefix[2][3]={"",""};
725 if ( var_is_numeric (indep_var) )
727 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
728 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
732 val_lab0 = gp.v.g_value[0].s;
733 val_lab1 = gp.v.g_value[1].s;
736 if (gp.criterion == CMP_LE )
738 strcpy(prefix[0],"< ");
739 strcpy(prefix[1],">=");
740 indep_value[0] = gp.v.critical_value;
741 indep_value[1] = gp.v.critical_value;
745 indep_value[0] = gp.v.g_value[0].f;
746 indep_value[1] = gp.v.g_value[1].f;
751 for (i=0; i < cmd->n_variables; ++i)
753 struct variable *var = cmd->v_variables[i];
754 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
757 tab_text (ssb->t, 0, i*2+1, TAB_LEFT,
758 var_get_name (cmd->v_variables[i]));
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,
860 var_get_name (pairs[i].v[j]));
863 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
864 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
865 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
866 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
872 /* Populate the one sample ssbox */
874 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
880 for (i=0; i < cmd->n_variables; ++i)
882 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
884 tab_text (ssb->t, 0, i+1, TAB_LEFT, var_get_name (cmd->v_variables[i]));
885 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
886 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
887 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
888 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
895 /* Implementation of the Test Results box struct */
897 void trbox_base_init(struct trbox *self,size_t n_vars, int cols);
898 void trbox_base_finalize(struct trbox *trb);
900 void trbox_independent_samples_init(struct trbox *trb,
901 struct cmd_t_test *cmd );
903 void trbox_independent_samples_populate(struct trbox *trb,
904 struct cmd_t_test *cmd);
906 void trbox_one_sample_init(struct trbox *self,
907 struct cmd_t_test *cmd );
909 void trbox_one_sample_populate(struct trbox *trb,
910 struct cmd_t_test *cmd);
912 void trbox_paired_init(struct trbox *self,
913 struct cmd_t_test *cmd );
915 void trbox_paired_populate(struct trbox *trb,
916 struct cmd_t_test *cmd);
920 /* Create a trbox according to mode*/
922 trbox_create(struct trbox *trb,
923 struct cmd_t_test *cmd, int mode)
928 trbox_one_sample_init(trb,cmd);
931 trbox_independent_samples_init(trb,cmd);
934 trbox_paired_init(trb,cmd);
941 /* Populate a trbox according to cmd */
943 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
945 trb->populate(trb,cmd);
948 /* Submit and destroy a trbox */
950 trbox_finalize(struct trbox *trb)
955 /* Initialize the independent samples trbox */
957 trbox_independent_samples_init(struct trbox *self,
958 struct cmd_t_test *cmd UNUSED)
961 const int vsize=cmd->n_variables*2+3;
964 self->populate = trbox_independent_samples_populate;
966 trbox_base_init(self,cmd->n_variables*2,hsize);
967 tab_title(self->t,_("Independent Samples Test"));
968 tab_hline(self->t,TAL_1,2,hsize-1,1);
969 tab_vline(self->t,TAL_2,2,0,vsize-1);
970 tab_vline(self->t,TAL_1,4,0,vsize-1);
971 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
972 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
973 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
974 tab_joint_text(self->t, 2, 0, 3, 0,
975 TAB_CENTER,_("Levene's Test for Equality of Variances"));
976 tab_joint_text(self->t, 4,0,hsize-1,0,
977 TAB_CENTER,_("t-test for Equality of Means"));
979 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
980 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
981 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
982 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
983 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
984 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
985 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
986 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
987 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
989 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
990 _("%g%% Confidence Interval of the Difference"),
991 cmd->criteria*100.0);
995 /* Populate the independent samples trbox */
997 trbox_independent_samples_populate(struct trbox *self,
998 struct cmd_t_test *cmd )
1003 for (i=0; i < cmd->n_variables; ++i)
1012 double pooled_variance;
1013 double std_err_diff;
1016 struct variable *var = cmd->v_variables[i];
1017 struct group_proc *grp_data = group_proc_get (var);
1019 struct hsh_table *grp_hash = grp_data->group_hash;
1021 struct group_statistics *gs0 ;
1022 struct group_statistics *gs1 ;
1024 union value search_val;
1026 if ( gp.criterion == CMP_LE )
1027 search_val.f = gp.v.critical_value - 1.0;
1029 search_val = gp.v.g_value[0];
1031 gs0 = hsh_find(grp_hash, (void *) &search_val);
1034 if ( gp.criterion == CMP_LE )
1035 search_val.f = gp.v.critical_value + 1.0;
1037 search_val = gp.v.g_value[1];
1039 gs1 = hsh_find(grp_hash, (void *) &search_val);
1043 tab_text (self->t, 0, i*2+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1045 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1048 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1050 /* Now work out the significance of the Levene test */
1051 df1 = 1; df2 = grp_data->ugs.n - 2;
1052 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1054 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1056 df = gs0->n + gs1->n - 2.0 ;
1057 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1059 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1061 (gs1->n )*pow2(gs1->s_std_dev)
1064 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1065 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1067 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1069 p = gsl_cdf_tdist_P(t, df);
1070 q = gsl_cdf_tdist_Q(t, df);
1072 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1074 mean_diff = gs0->mean - gs1->mean;
1075 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1078 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1079 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1082 /* Now work out the confidence interval */
1083 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1085 t = gsl_cdf_tdist_Qinv(q,df);
1086 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1087 mean_diff - t * std_err_diff, 8, 3);
1089 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1090 mean_diff + t * std_err_diff, 8, 3);
1095 /* Now for the \sigma_1 != \sigma_2 case */
1096 tab_text (self->t, 1, i*2+3+1,
1097 TAB_LEFT, _("Equal variances not assumed"));
1100 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1101 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1103 t = mean_diff / sqrt(se2) ;
1104 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1107 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1111 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1115 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1117 p = gsl_cdf_tdist_P(t, df);
1118 q = gsl_cdf_tdist_Q(t, df);
1120 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1122 /* Now work out the confidence interval */
1123 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1125 t = gsl_cdf_tdist_Qinv(q, df);
1127 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1130 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1133 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1134 mean_diff - t * std_err_diff, 8, 3);
1136 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1137 mean_diff + t * std_err_diff, 8, 3);
1143 /* Initialize the paired samples trbox */
1145 trbox_paired_init(struct trbox *self,
1146 struct cmd_t_test *cmd UNUSED)
1150 const int vsize=n_pairs+3;
1152 self->populate = trbox_paired_populate;
1154 trbox_base_init(self,n_pairs,hsize);
1155 tab_title (self->t, _("Paired Samples Test"));
1156 tab_hline(self->t,TAL_1,2,6,1);
1157 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1158 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1159 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1160 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1161 tab_hline(self->t,TAL_1,5,6, 2);
1162 tab_vline(self->t,TAL_GAP,6,0,1);
1164 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1165 _("%g%% Confidence Interval of the Difference"),
1166 cmd->criteria*100.0);
1168 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1169 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1170 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1171 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1172 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1173 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1174 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1175 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1178 /* Populate the paired samples trbox */
1180 trbox_paired_populate(struct trbox *trb,
1181 struct cmd_t_test *cmd UNUSED)
1185 for (i=0; i < n_pairs; ++i)
1190 double n = pairs[i].n;
1194 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1196 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1197 var_get_name (pairs[i].v[0]),
1198 var_get_name (pairs[i].v[1]));
1200 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1202 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1205 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1206 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1208 /* Now work out the confidence interval */
1209 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1211 t = gsl_cdf_tdist_Qinv(q, df);
1213 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1214 pairs[i].mean_diff - t * se_mean , 8, 4);
1216 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1217 pairs[i].mean_diff + t * se_mean , 8, 4);
1219 t = (pairs[i].mean[0] - pairs[i].mean[1])
1221 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1222 2 * pairs[i].correlation *
1223 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1227 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1229 /* Degrees of freedom */
1230 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1232 p = gsl_cdf_tdist_P(t,df);
1233 q = gsl_cdf_tdist_P(t,df);
1235 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1240 /* Initialize the one sample trbox */
1242 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1245 const int vsize=cmd->n_variables+3;
1247 self->populate = trbox_one_sample_populate;
1249 trbox_base_init(self, cmd->n_variables,hsize);
1250 tab_title (self->t, _("One-Sample Test"));
1251 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1252 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1254 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1255 _("Test Value = %f"), cmd->n_testval[0]);
1257 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1260 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1261 _("%g%% Confidence Interval of the Difference"),
1262 cmd->criteria*100.0);
1264 tab_vline(self->t,TAL_GAP,6,1,1);
1265 tab_hline(self->t,TAL_1,5,6,2);
1266 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1267 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1268 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1269 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1270 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1271 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1276 /* Populate the one sample trbox */
1278 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1284 for (i=0; i < cmd->n_variables; ++i)
1289 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1292 tab_text (trb->t, 0, i+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1294 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1296 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1298 /* degrees of freedom */
1301 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1303 p = gsl_cdf_tdist_P(t, df);
1304 q = gsl_cdf_tdist_Q(t, df);
1306 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1308 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1310 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1313 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1314 t = gsl_cdf_tdist_Qinv(q, df);
1316 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1317 gs->mean_diff - t * gs->se_mean, 8,4);
1319 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1320 gs->mean_diff + t * gs->se_mean, 8,4);
1324 /* Base initializer for the generalized trbox */
1326 trbox_base_init(struct trbox *self, size_t data_rows, int cols)
1328 const size_t rows = 3 + data_rows;
1330 self->finalize = trbox_base_finalize;
1331 self->t = tab_create (cols, rows, 0);
1332 tab_headers (self->t,0,0,3,0);
1333 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1334 tab_hline(self->t, TAL_2,0,cols-1,3);
1335 tab_dim (self->t, tab_natural_dimensions);
1339 /* Base finalizer for the trbox */
1341 trbox_base_finalize(struct trbox *trb)
1347 /* Create , populate and submit the Paired Samples Correlation box */
1351 const int rows=1+n_pairs;
1355 struct tab_table *table;
1357 table = tab_create (cols,rows,0);
1359 tab_columns (table, SOM_COL_DOWN, 1);
1360 tab_headers (table,0,0,1,0);
1361 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1362 tab_hline(table, TAL_2, 0, cols - 1, 1);
1363 tab_vline(table, TAL_2, 2, 0, rows - 1);
1364 tab_dim(table, tab_natural_dimensions);
1365 tab_title(table, _("Paired Samples Correlations"));
1367 /* column headings */
1368 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1369 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1370 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1372 for (i=0; i < n_pairs; ++i)
1376 double df = pairs[i].n -2;
1378 double correlation_t =
1379 pairs[i].correlation * sqrt(df) /
1380 sqrt(1 - pow2(pairs[i].correlation));
1384 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1387 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1389 var_get_name (pairs[i].v[0]),
1390 var_get_name (pairs[i].v[1]));
1394 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1395 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1397 p = gsl_cdf_tdist_P(correlation_t, df);
1398 q = gsl_cdf_tdist_Q(correlation_t, df);
1400 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1409 /* Calculation Implementation */
1411 /* Per case calculations common to all variants of the T test */
1413 common_calc (const struct dictionary *dict,
1414 const struct ccase *c,
1416 const struct casefilter *filter)
1419 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1421 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1424 /* Listwise has to be implicit if the independent variable is missing ?? */
1425 if ( cmd->sbc_groups )
1427 if ( casefilter_variable_missing (filter, c, indep_var) )
1431 for(i = 0; i < cmd->n_variables ; ++i)
1433 struct variable *v = cmd->v_variables[i];
1435 if (! casefilter_variable_missing (filter, c, v) )
1437 struct group_statistics *gs;
1438 const union value *val = case_data (c, v->fv);
1439 gs = &group_proc_get (cmd->v_variables[i])->ugs;
1442 gs->sum += weight * val->f;
1443 gs->ssq += weight * val->f * val->f;
1449 /* Pre calculations common to all variants of the T test */
1451 common_precalc ( struct cmd_t_test *cmd )
1455 for(i=0; i< cmd->n_variables ; ++i)
1457 struct group_statistics *gs;
1458 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1467 /* Post calculations common to all variants of the T test */
1469 common_postcalc (struct cmd_t_test *cmd)
1473 for(i=0; i< cmd->n_variables ; ++i)
1475 struct group_statistics *gs;
1476 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1478 gs->mean=gs->sum / gs->n;
1479 gs->s_std_dev= sqrt(
1480 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1485 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1488 gs->se_mean = gs->std_dev / sqrt(gs->n);
1489 gs->mean_diff= gs->sum_diff / gs->n;
1493 /* Per case calculations for one sample t test */
1495 one_sample_calc (const struct dictionary *dict,
1496 const struct ccase *c, void *cmd_,
1497 const struct casefilter *filter)
1501 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1503 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1506 for(i=0; i< cmd->n_variables ; ++i)
1508 struct group_statistics *gs;
1509 struct variable *v = cmd->v_variables[i];
1510 const union value *val = case_data (c, v->fv);
1512 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1514 if ( ! casefilter_variable_missing (filter, c, v))
1515 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1521 /* Pre calculations for one sample t test */
1523 one_sample_precalc ( struct cmd_t_test *cmd )
1527 for(i=0; i< cmd->n_variables ; ++i)
1529 struct group_statistics *gs;
1530 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1536 /* Post calculations for one sample t test */
1538 one_sample_postcalc (struct cmd_t_test *cmd)
1542 for(i=0; i< cmd->n_variables ; ++i)
1544 struct group_statistics *gs;
1545 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1547 gs->mean_diff = gs->sum_diff / gs->n ;
1554 paired_precalc (struct cmd_t_test *cmd UNUSED)
1558 for(i=0; i < n_pairs ; ++i )
1561 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1562 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1563 pairs[i].sum_of_prod = 0;
1564 pairs[i].correlation = 0;
1565 pairs[i].sum_of_diffs = 0;
1566 pairs[i].ssq_diffs = 0;
1573 paired_calc (const struct dictionary *dict, const struct ccase *c,
1574 struct cmd_t_test *cmd UNUSED, const struct casefilter *filter)
1578 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1580 for(i=0; i < n_pairs ; ++i )
1582 struct variable *v0 = pairs[i].v[0];
1583 struct variable *v1 = pairs[i].v[1];
1585 const union value *val0 = case_data (c, v0->fv);
1586 const union value *val1 = case_data (c, v1->fv);
1588 if ( ! casefilter_variable_missing (filter, c, v0) &&
1589 ! casefilter_variable_missing (filter, c, v1) )
1591 pairs[i].n += weight;
1592 pairs[i].sum[0] += weight * val0->f;
1593 pairs[i].sum[1] += weight * val1->f;
1595 pairs[i].ssq[0] += weight * pow2(val0->f);
1596 pairs[i].ssq[1] += weight * pow2(val1->f);
1598 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1600 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1601 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1609 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1613 for(i=0; i < n_pairs ; ++i )
1616 const double n = pairs[i].n;
1618 for (j=0; j < 2 ; ++j)
1620 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1621 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1622 pow2(pairs[i].mean[j]))
1625 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1626 pow2(pairs[i].mean[j]))
1630 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1631 pairs[i].mean[0] * pairs[i].mean[1] ;
1632 /* correlation now actually contains the covariance */
1634 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1635 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1637 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1639 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1640 ( pairs[i].ssq_diffs / n )
1642 pow2(pairs[i].mean_diff )
1648 group_precalc (struct cmd_t_test *cmd )
1653 for(i=0; i< cmd->n_variables ; ++i)
1655 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1657 /* There's always 2 groups for a T - TEST */
1660 gp.indep_width = var_get_width (indep_var);
1662 ttpr->group_hash = hsh_create(2,
1663 (hsh_compare_func *) compare_group_binary,
1664 (hsh_hash_func *) hash_group_binary,
1665 (hsh_free_func *) free_group,
1668 for (j=0 ; j < 2 ; ++j)
1671 struct group_statistics *gs = xmalloc (sizeof *gs);
1677 if ( gp.criterion == CMP_EQ )
1679 gs->id = gp.v.g_value[j];
1684 gs->id.f = gp.v.critical_value - 1.0 ;
1686 gs->id.f = gp.v.critical_value + 1.0 ;
1689 hsh_insert ( ttpr->group_hash, (void *) gs );
1697 group_calc (const struct dictionary *dict,
1698 const struct ccase *c, struct cmd_t_test *cmd,
1699 const struct casefilter *filter)
1703 const double weight =
1704 dict_get_case_weight (dict, c, &bad_weight_warn);
1706 const union value *gv;
1708 if ( casefilter_variable_missing (filter, c, indep_var))
1711 gv = case_data (c, indep_var->fv);
1713 for(i=0; i< cmd->n_variables ; ++i)
1715 struct variable *var = cmd->v_variables[i];
1716 const union value *val = case_data (c, var->fv);
1717 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1718 struct group_statistics *gs;
1720 gs = hsh_find(grp_hash, (void *) gv);
1722 /* If the independent variable doesn't match either of the values
1723 for this case then move on to the next case */
1727 if ( ! casefilter_variable_missing (filter, c, var) )
1730 gs->sum += weight * val->f;
1731 gs->ssq += weight * pow2(val->f);
1740 group_postcalc ( struct cmd_t_test *cmd )
1744 for (i = 0; i < cmd->n_variables ; ++i)
1746 struct variable *var = cmd->v_variables[i];
1747 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1748 struct hsh_iterator g;
1749 struct group_statistics *gs;
1752 for (gs = hsh_first (grp_hash,&g);
1754 gs = hsh_next(grp_hash,&g))
1756 gs->mean = gs->sum / gs->n;
1758 gs->s_std_dev= sqrt(
1759 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1764 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1767 gs->se_mean = gs->std_dev / sqrt(gs->n);
1777 calculate(const struct ccase *first, const struct casefile *cf,
1778 void *cmd_, const struct dataset *ds)
1780 const struct dictionary *dict = dataset_dict (ds);
1781 struct ssbox stat_summary_box;
1782 struct trbox test_results_box;
1784 struct casereader *r;
1787 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1789 struct casefilter *filter = casefilter_create (cmd->miss != TTS_INCLUDE,
1792 if ( cmd->miss == TTS_LISTWISE )
1793 casefilter_add_variables (filter,
1794 cmd->v_variables, cmd->n_variables);
1796 output_split_file_values (ds, first);
1797 common_precalc (cmd);
1798 for(r = casefile_get_reader (cf, filter);
1799 casereader_read (r, &c) ;
1802 common_calc (dict, &c, cmd, filter);
1805 casereader_destroy (r);
1806 common_postcalc (cmd);
1811 one_sample_precalc (cmd);
1812 for(r = casefile_get_reader (cf, filter);
1813 casereader_read (r, &c) ;
1816 one_sample_calc (dict, &c, cmd, filter);
1818 casereader_destroy (r);
1819 one_sample_postcalc (cmd);
1822 paired_precalc(cmd);
1823 for(r = casefile_get_reader (cf, filter);
1824 casereader_read (r, &c) ;
1827 paired_calc (dict, &c, cmd, filter);
1829 casereader_destroy (r);
1830 paired_postcalc (cmd);
1836 for(r = casefile_get_reader (cf, filter);
1837 casereader_read (r, &c) ;
1840 group_calc (dict, &c, cmd, filter);
1842 casereader_destroy (r);
1843 group_postcalc(cmd);
1845 levene (dict, cf, indep_var, cmd->n_variables, cmd->v_variables,
1850 casefilter_destroy (filter);
1852 ssbox_create(&stat_summary_box,cmd,mode);
1853 ssbox_populate(&stat_summary_box,cmd);
1854 ssbox_finalize(&stat_summary_box);
1856 if ( mode == T_PAIRED)
1859 trbox_create(&test_results_box,cmd,mode);
1860 trbox_populate(&test_results_box,cmd);
1861 trbox_finalize(&test_results_box);
1866 short which_group(const struct group_statistics *g,
1867 const struct group_properties *p);
1869 /* Return -1 if the id of a is less than b; +1 if greater than and
1872 compare_group_binary(const struct group_statistics *a,
1873 const struct group_statistics *b,
1874 const struct group_properties *p)
1879 if ( p->criterion == CMP_LE )
1881 /* less-than-or-equal comparision is not meaningfull for
1882 alpha variables, so we shouldn't ever arrive here */
1883 assert(p->indep_width == 0 ) ;
1885 flag_a = ( a->id.f < p->v.critical_value ) ;
1886 flag_b = ( b->id.f < p->v.critical_value ) ;
1890 flag_a = which_group(a, p);
1891 flag_b = which_group(b, p);
1894 if (flag_a < flag_b )
1897 return (flag_a > flag_b);
1900 /* This is a degenerate case of a hash, since it can only return three possible
1901 values. It's really a comparison, being used as a hash function */
1904 hash_group_binary(const struct group_statistics *g,
1905 const struct group_properties *p)
1909 if ( p->criterion == CMP_LE )
1911 /* Not meaningfull to do a less than compare for alpha values ? */
1912 assert(p->indep_width == 0 ) ;
1913 flag = ( g->id.f < p->v.critical_value ) ;
1915 else if ( p->criterion == CMP_EQ)
1917 flag = which_group(g,p);
1925 /* return 0 if G belongs to group 0,
1926 1 if it belongs to group 1,
1927 2 if it belongs to neither group */
1929 which_group(const struct group_statistics *g,
1930 const struct group_properties *p)
1933 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1936 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))