1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2009 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 #include <gsl/gsl_cdf.h>
25 #include <data/case.h>
26 #include <data/casegrouper.h>
27 #include <data/casereader.h>
28 #include <data/dictionary.h>
29 #include <data/procedure.h>
30 #include <data/value-labels.h>
31 #include <data/variable.h>
32 #include <language/command.h>
33 #include <language/dictionary/split-file.h>
34 #include <language/lexer/lexer.h>
35 #include <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>
46 #include <data/format.h>
51 #define _(msgid) gettext (msgid)
59 +variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
61 missing=miss:!analysis/listwise,
62 incl:include/!exclude;
63 +format=fmt:!labels/nolabels;
64 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
70 /* Variable for the GROUPS subcommand, if given. */
71 static struct variable *indep_var;
79 struct group_properties
81 /* The comparison criterion */
82 enum comparison criterion;
84 /* The width of the independent variable */
88 /* The value of the independent variable at which groups are determined to
89 belong to one group or the other */
90 double critical_value;
93 /* The values of the independent variable for each group */
94 union value g_value[2];
100 static struct group_properties gp ;
104 /* PAIRS: Number of pairs to be compared ; each pair. */
105 static int n_pairs = 0 ;
108 /* The variables comprising the pair */
109 const struct variable *v[2];
111 /* The number of valid variable pairs */
114 /* The sum of the members */
117 /* sum of squares of the members */
120 /* Std deviation of the members */
124 /* Sample Std deviation of the members */
127 /* The means of the members */
130 /* The correlation coefficient between the variables */
133 /* The sum of the differences */
136 /* The sum of the products */
139 /* The mean of the differences */
142 /* The sum of the squares of the differences */
145 /* The std deviation of the differences */
149 static struct pair *pairs=0;
151 static int parse_value (struct lexer *lexer, union value * v, enum val_type);
153 /* Structures and Functions for the Statistics Summary Box */
155 typedef void populate_ssbox_func (struct ssbox *ssb,
156 const struct dictionary *,
157 struct cmd_t_test *cmd);
158 typedef void finalize_ssbox_func (struct ssbox *ssb);
164 populate_ssbox_func *populate;
165 finalize_ssbox_func *finalize;
170 void ssbox_create (struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
172 /* Populate a ssbox according to cmd */
173 void ssbox_populate (struct ssbox *ssb, const struct dictionary *dict,
174 struct cmd_t_test *cmd);
176 /* Submit and destroy a ssbox */
177 void ssbox_finalize (struct ssbox *ssb);
179 /* A function to create, populate and submit the Paired Samples Correlation
181 static void pscbox (const struct dictionary *);
184 /* Structures and Functions for the Test Results Box */
187 typedef void populate_trbox_func (struct trbox *trb,
188 const struct dictionary *dict,
189 struct cmd_t_test *cmd);
190 typedef void finalize_trbox_func (struct trbox *trb);
194 populate_trbox_func *populate;
195 finalize_trbox_func *finalize;
199 void trbox_create (struct trbox *trb, struct cmd_t_test *cmd, int mode);
201 /* Populate a ssbox according to cmd */
202 static void trbox_populate (struct trbox *trb, const struct dictionary *dict,
203 struct cmd_t_test *cmd);
205 /* Submit and destroy a ssbox */
206 void trbox_finalize (struct trbox *trb);
208 /* Which mode was T-TEST invoked */
216 static int common_calc (const struct dictionary *dict,
217 const struct ccase *, void *,
219 static void common_precalc (struct cmd_t_test *);
220 static void common_postcalc (struct cmd_t_test *);
222 static int one_sample_calc (const struct dictionary *dict, const struct ccase *, void *, enum mv_class);
223 static void one_sample_precalc (struct cmd_t_test *);
224 static void one_sample_postcalc (struct cmd_t_test *);
226 static int paired_calc (const struct dictionary *dict, const struct ccase *,
227 struct cmd_t_test*, enum mv_class);
228 static void paired_precalc (struct cmd_t_test *);
229 static void paired_postcalc (struct cmd_t_test *);
231 static void group_precalc (struct cmd_t_test *);
232 static int group_calc (const struct dictionary *dict, const struct ccase *,
233 struct cmd_t_test *, enum mv_class);
234 static void group_postcalc (struct cmd_t_test *);
237 static void calculate (struct cmd_t_test *,
239 const struct dataset *);
243 static struct cmd_t_test cmd;
245 static bool bad_weight_warn = false;
248 static int compare_group_binary (const struct group_statistics *a,
249 const struct group_statistics *b,
250 const struct group_properties *p);
253 static unsigned hash_group_binary (const struct group_statistics *g,
254 const struct group_properties *p);
259 cmd_t_test (struct lexer *lexer, struct dataset *ds)
261 struct casegrouper *grouper;
262 struct casereader *group;
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 const_hsh_table *hash;
310 const struct variable *v;
312 hash = const_hsh_create (n_pairs, compare_vars_by_name, hash_var_by_name,
315 for (i=0; i < n_pairs; ++i)
317 const_hsh_insert (hash, pairs[i].v[0]);
318 const_hsh_insert (hash, pairs[i].v[1]);
321 assert (cmd.n_variables == 0);
322 cmd.n_variables = const_hsh_count (hash);
324 cmd.v_variables = xnrealloc (cmd.v_variables, cmd.n_variables,
325 sizeof *cmd.v_variables);
326 /* Iterate through the hash */
327 for (i=0,v = const_hsh_first (hash, &hi);
329 v = const_hsh_next (hash, &hi) )
330 cmd.v_variables[i++]=v;
331 const_hsh_destroy (hash);
334 else if ( !cmd.sbc_variables)
336 msg (SE, _ ("One or more VARIABLES must be specified."));
341 bad_weight_warn = true;
344 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
345 while (casegrouper_get_next_group (grouper, &group))
346 calculate (&cmd, group, ds);
347 ok = casegrouper_destroy (grouper);
348 ok = proc_commit (ds) && ok;
354 if ( mode == T_IND_SAMPLES)
357 /* Destroy any group statistics we created */
358 for (v = 0 ; v < cmd.n_variables ; ++v )
360 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
361 hsh_destroy (grpp->group_hash);
366 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
370 tts_custom_groups (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd UNUSED,
373 int n_group_values=0;
375 lex_match (lexer, '=');
377 indep_var = parse_variable (lexer, dataset_dict (ds));
380 lex_error (lexer, "expecting variable name in GROUPS subcommand");
384 if (var_is_long_string (indep_var))
386 msg (SE, _ ("Long string variable %s is not valid here."),
387 var_get_name (indep_var));
391 if (!lex_match (lexer, '('))
393 if (var_is_numeric (indep_var))
395 gp.v.g_value[0].f = 1;
396 gp.v.g_value[1].f = 2;
398 gp.criterion = CMP_EQ;
406 msg (SE, _ ("When applying GROUPS to a string variable, two "
407 "values must be specified."));
412 if (!parse_value (lexer, &gp.v.g_value[0], var_get_type (indep_var)))
415 lex_match (lexer, ',');
416 if (lex_match (lexer, ')'))
418 if (var_is_alpha (indep_var))
420 msg (SE, _ ("When applying GROUPS to a string variable, two "
421 "values must be specified."));
424 gp.criterion = CMP_LE;
425 gp.v.critical_value = gp.v.g_value[0].f;
431 if (!parse_value (lexer, &gp.v.g_value[1], var_get_type (indep_var)))
435 if (!lex_force_match (lexer, ')'))
438 if ( n_group_values == 2 )
439 gp.criterion = CMP_EQ ;
441 gp.criterion = CMP_LE ;
444 if ( var_is_alpha (indep_var))
446 buf_copy_rpad (gp.v.g_value [0].s, var_get_width (indep_var),
447 gp.v.g_value [0].s, strlen (gp.v.g_value[0].s));
449 buf_copy_rpad (gp.v.g_value [1].s, var_get_width (indep_var),
450 gp.v.g_value [1].s, strlen (gp.v.g_value[1].s));
458 tts_custom_pairs (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
460 const struct variable **vars;
462 size_t n_pairs_local;
464 size_t n_before_WITH;
465 size_t n_after_WITH = SIZE_MAX;
466 int paired ; /* Was the PAIRED keyword given ? */
468 lex_match (lexer, '=');
471 if (!parse_variables_const (lexer, dataset_dict (ds), &vars, &n_vars,
472 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
480 if (lex_match (lexer, T_WITH))
482 n_before_WITH = n_vars;
483 if (!parse_variables_const (lexer, dataset_dict (ds), &vars, &n_vars,
484 PV_DUPLICATE | PV_APPEND
485 | PV_NUMERIC | PV_NO_SCRATCH))
490 n_after_WITH = n_vars - n_before_WITH;
493 paired = (lex_match (lexer, '(') && lex_match_id (lexer, "PAIRED") && lex_match (lexer, ')'));
495 /* Determine the number of pairs needed */
498 if (n_before_WITH != n_after_WITH)
501 msg (SE, _ ("PAIRED was specified but the number of variables "
502 "preceding WITH (%zu) did not match the number "
504 n_before_WITH, n_after_WITH);
507 n_pairs_local = n_before_WITH;
509 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
511 n_pairs_local = n_before_WITH * n_after_WITH ;
513 else /* Neither WITH nor PAIRED keyword given */
518 msg (SE, _ ("At least two variables must be specified "
523 /* how many ways can you pick 2 from n_vars ? */
524 n_pairs_local = n_vars * (n_vars - 1) / 2;
528 /* Allocate storage for the pairs */
529 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
531 /* Populate the pairs with the appropriate variables */
536 assert (n_pairs_local == n_vars / 2);
537 for (i = 0; i < n_pairs_local; ++i)
539 pairs[i].v[n_pairs] = vars[i];
540 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
543 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
548 for (i=0 ; i < n_before_WITH ; ++i )
550 for (j=0 ; j < n_after_WITH ; ++j)
552 pairs[p].v[0] = vars[i];
553 pairs[p].v[1] = vars[j+n_before_WITH];
558 else /* Neither WITH nor PAIRED given */
563 for (i=0 ; i < n_vars ; ++i )
565 for (j=i+1 ; j < n_vars ; ++j)
567 pairs[p].v[0] = vars[i];
568 pairs[p].v[1] = vars[j];
574 n_pairs+=n_pairs_local;
580 /* Parses the current token (numeric or string, depending on type)
581 value v and returns success. */
583 parse_value (struct lexer *lexer, union value * v, enum val_type type)
585 if (type == VAL_NUMERIC)
587 if (!lex_force_num (lexer))
589 v->f = lex_tokval (lexer);
593 if (!lex_force_string (lexer))
595 memset (v->s, ' ', MAX_SHORT_STRING);
596 strncpy (v->s, ds_cstr (lex_tokstr (lexer)), ds_length (lex_tokstr (lexer)));
605 /* Implementation of the SSBOX object */
607 void ssbox_base_init (struct ssbox *this, int cols,int rows);
609 void ssbox_base_finalize (struct ssbox *ssb);
611 void ssbox_one_sample_init (struct ssbox *this,
612 struct cmd_t_test *cmd );
614 void ssbox_independent_samples_init (struct ssbox *this,
615 struct cmd_t_test *cmd);
617 void ssbox_paired_init (struct ssbox *this,
618 struct cmd_t_test *cmd);
621 /* Factory to create an ssbox */
623 ssbox_create (struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
628 ssbox_one_sample_init (ssb,cmd);
631 ssbox_independent_samples_init (ssb,cmd);
634 ssbox_paired_init (ssb,cmd);
643 /* Despatcher for the populate method */
645 ssbox_populate (struct ssbox *ssb, const struct dictionary *dict,
646 struct cmd_t_test *cmd)
648 ssb->populate (ssb, dict, cmd);
652 /* Despatcher for finalize */
654 ssbox_finalize (struct ssbox *ssb)
660 /* Submit the box and clear up */
662 ssbox_base_finalize (struct ssbox *ssb)
669 /* Initialize a ssbox struct */
671 ssbox_base_init (struct ssbox *this, int cols,int rows)
673 this->finalize = ssbox_base_finalize;
674 this->t = tab_create (cols, rows, 0);
676 tab_columns (this->t, SOM_COL_DOWN, 1);
677 tab_headers (this->t,0,0,1,0);
678 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
679 tab_hline (this->t, TAL_2,0,cols-1,1);
680 tab_dim (this->t, tab_natural_dimensions);
683 void ssbox_one_sample_populate (struct ssbox *ssb,
684 const struct dictionary *,
685 struct cmd_t_test *cmd);
687 /* Initialize the one_sample ssbox */
689 ssbox_one_sample_init (struct ssbox *this,
690 struct cmd_t_test *cmd )
693 const int vsize=cmd->n_variables+1;
695 this->populate = ssbox_one_sample_populate;
697 ssbox_base_init (this, hsize,vsize);
698 tab_title (this->t, _ ("One-Sample Statistics"));
699 tab_vline (this->t, TAL_2, 1,0,vsize - 1);
700 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
701 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
702 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
703 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
706 static void ssbox_independent_samples_populate (struct ssbox *ssb,
707 const struct dictionary *,
708 struct cmd_t_test *cmd);
710 /* Initialize the independent samples ssbox */
712 ssbox_independent_samples_init (struct ssbox *this,
713 struct cmd_t_test *cmd)
716 int vsize = cmd->n_variables*2 +1;
718 this->populate = ssbox_independent_samples_populate;
720 ssbox_base_init (this, hsize,vsize);
721 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
722 tab_title (this->t, _ ("Group Statistics"));
723 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, var_get_name (indep_var));
724 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
725 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
726 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
727 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
731 /* Populate the ssbox for independent samples */
733 ssbox_independent_samples_populate (struct ssbox *ssb,
734 const struct dictionary *dict,
735 struct cmd_t_test *cmd)
739 const struct variable *wv = dict_get_weight (dict);
740 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
742 char *val_lab[2] = {NULL, NULL};
743 double indep_value[2];
745 char prefix[2][3]={"",""};
747 if ( var_is_numeric (indep_var) )
751 s = var_lookup_value_label (indep_var, &gp.v.g_value[0]);
752 val_lab[0] = s ? strdup (s) : NULL;
754 s = var_lookup_value_label (indep_var, &gp.v.g_value[1]);
755 val_lab[1] = s ? strdup (s) : NULL;
759 val_lab[0] = calloc (sizeof (char), MAX_SHORT_STRING + 1);
760 val_lab[1] = calloc (sizeof (char), MAX_SHORT_STRING + 1);
761 memcpy (val_lab[0], gp.v.g_value[0].s, MAX_SHORT_STRING);
762 memcpy (val_lab[1], gp.v.g_value[1].s, MAX_SHORT_STRING);
765 if (gp.criterion == CMP_LE )
767 strcpy (prefix[0],">=");
768 strcpy (prefix[1],"<");
769 indep_value[0] = gp.v.critical_value;
770 indep_value[1] = gp.v.critical_value;
774 indep_value[0] = gp.v.g_value[0].f;
775 indep_value[1] = gp.v.g_value[1].f;
780 for (i=0; i < cmd->n_variables; ++i)
782 const struct variable *var = cmd->v_variables[i];
783 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
786 tab_text (ssb->t, 0, i*2+1, TAB_LEFT,
787 var_get_name (cmd->v_variables[i]));
790 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
791 "%s%s", prefix[0], val_lab[0]);
793 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
794 "%s%g", prefix[0], indep_value[0]);
798 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
799 "%s%s", prefix[1], val_lab[1]);
801 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
802 "%s%g", prefix[1], indep_value[1]);
805 /* Fill in the group statistics */
806 for ( count = 0 ; count < 2 ; ++count )
808 union value search_val;
810 struct group_statistics *gs;
812 if ( gp.criterion == CMP_LE )
817 search_val.f = gp.v.critical_value + 1.0;
821 /* less than ( < ) case */
822 search_val.f = gp.v.critical_value - 1.0;
827 search_val = gp.v.g_value[count];
830 gs = hsh_find (grp_hash, (void *) &search_val);
833 tab_double (ssb->t, 2, i*2+count+1, TAB_RIGHT, gs->n, wfmt);
834 tab_double (ssb->t, 3, i*2+count+1, TAB_RIGHT, gs->mean, NULL);
835 tab_double (ssb->t, 4, i*2+count+1, TAB_RIGHT, gs->std_dev, NULL);
836 tab_double (ssb->t, 5, i*2+count+1, TAB_RIGHT, gs->se_mean, NULL);
844 static void ssbox_paired_populate (struct ssbox *ssb,
845 const struct dictionary *dict,
846 struct cmd_t_test *cmd);
848 /* Initialize the paired values ssbox */
850 ssbox_paired_init (struct ssbox *this, struct cmd_t_test *cmd UNUSED)
854 int vsize = n_pairs*2+1;
856 this->populate = ssbox_paired_populate;
858 ssbox_base_init (this, hsize,vsize);
859 tab_title (this->t, _ ("Paired Sample Statistics"));
860 tab_vline (this->t,TAL_GAP,1,0,vsize-1);
861 tab_vline (this->t,TAL_2,2,0,vsize-1);
862 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
863 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
864 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
865 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
869 /* Populate the ssbox for paired values */
871 ssbox_paired_populate (struct ssbox *ssb, const struct dictionary *dict,
872 struct cmd_t_test *cmd UNUSED)
876 const struct variable *wv = dict_get_weight (dict);
877 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
881 for (i=0; i < n_pairs; ++i)
885 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _ ("Pair %d"),i);
887 for (j=0 ; j < 2 ; ++j)
889 struct group_statistics *gs;
891 gs = &group_proc_get (pairs[i].v[j])->ugs;
895 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT,
896 var_get_name (pairs[i].v[j]));
899 tab_double (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], NULL);
900 tab_double (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, wfmt);
901 tab_double (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], NULL);
902 tab_double (ssb->t,5, i*2+j+1, TAB_RIGHT,
903 pairs[i].std_dev[j]/sqrt (pairs[i].n), NULL);
909 /* Populate the one sample ssbox */
911 ssbox_one_sample_populate (struct ssbox *ssb, const struct dictionary *dict,
912 struct cmd_t_test *cmd)
916 const struct variable *wv = dict_get_weight (dict);
917 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
921 for (i=0; i < cmd->n_variables; ++i)
923 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
925 tab_text (ssb->t, 0, i+1, TAB_LEFT, var_get_name (cmd->v_variables[i]));
926 tab_double (ssb->t,1, i+1, TAB_RIGHT, gs->n, wfmt);
927 tab_double (ssb->t,2, i+1, TAB_RIGHT, gs->mean, NULL);
928 tab_double (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, NULL);
929 tab_double (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, NULL);
935 /* Implementation of the Test Results box struct */
937 void trbox_base_init (struct trbox *self,size_t n_vars, int cols);
938 void trbox_base_finalize (struct trbox *trb);
940 void trbox_independent_samples_init (struct trbox *trb,
941 struct cmd_t_test *cmd );
943 static void trbox_independent_samples_populate (struct trbox *trb,
944 const struct dictionary *dict,
945 struct cmd_t_test *cmd);
947 void trbox_one_sample_init (struct trbox *self,
948 struct cmd_t_test *cmd );
950 static void trbox_one_sample_populate (struct trbox *trb,
951 const struct dictionary *,
952 struct cmd_t_test *cmd);
954 void trbox_paired_init (struct trbox *self,
955 struct cmd_t_test *cmd );
957 static void trbox_paired_populate (struct trbox *trb,
958 const struct dictionary *,
959 struct cmd_t_test *cmd);
963 /* Create a trbox according to mode*/
965 trbox_create (struct trbox *trb,
966 struct cmd_t_test *cmd, int mode)
971 trbox_one_sample_init (trb,cmd);
974 trbox_independent_samples_init (trb,cmd);
977 trbox_paired_init (trb,cmd);
984 /* Populate a trbox according to cmd */
986 trbox_populate (struct trbox *trb, const struct dictionary *dict,
987 struct cmd_t_test *cmd)
989 trb->populate (trb, dict, cmd);
992 /* Submit and destroy a trbox */
994 trbox_finalize (struct trbox *trb)
999 /* Initialize the independent samples trbox */
1001 trbox_independent_samples_init (struct trbox *self,
1002 struct cmd_t_test *cmd UNUSED)
1005 const int vsize=cmd->n_variables*2+3;
1008 self->populate = trbox_independent_samples_populate;
1010 trbox_base_init (self,cmd->n_variables*2,hsize);
1011 tab_title (self->t,_ ("Independent Samples Test"));
1012 tab_hline (self->t,TAL_1,2,hsize-1,1);
1013 tab_vline (self->t,TAL_2,2,0,vsize-1);
1014 tab_vline (self->t,TAL_1,4,0,vsize-1);
1015 tab_box (self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
1016 tab_hline (self->t,TAL_1, hsize-2,hsize-1,2);
1017 tab_box (self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
1018 tab_joint_text (self->t, 2, 0, 3, 0,
1019 TAB_CENTER,_ ("Levene's Test for Equality of Variances"));
1020 tab_joint_text (self->t, 4,0,hsize-1,0,
1021 TAB_CENTER,_ ("t-test for Equality of Means"));
1023 tab_text (self->t,2,2, TAB_CENTER | TAT_TITLE,_ ("F"));
1024 tab_text (self->t,3,2, TAB_CENTER | TAT_TITLE,_ ("Sig."));
1025 tab_text (self->t,4,2, TAB_CENTER | TAT_TITLE,_ ("t"));
1026 tab_text (self->t,5,2, TAB_CENTER | TAT_TITLE,_ ("df"));
1027 tab_text (self->t,6,2, TAB_CENTER | TAT_TITLE,_ ("Sig. (2-tailed)"));
1028 tab_text (self->t,7,2, TAB_CENTER | TAT_TITLE,_ ("Mean Difference"));
1029 tab_text (self->t,8,2, TAB_CENTER | TAT_TITLE,_ ("Std. Error Difference"));
1030 tab_text (self->t,9,2, TAB_CENTER | TAT_TITLE,_ ("Lower"));
1031 tab_text (self->t,10,2, TAB_CENTER | TAT_TITLE,_ ("Upper"));
1033 tab_joint_text (self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
1034 _ ("%g%% Confidence Interval of the Difference"),
1035 cmd->criteria*100.0);
1039 /* Populate the independent samples trbox */
1041 trbox_independent_samples_populate (struct trbox *self,
1042 const struct dictionary *dict UNUSED,
1043 struct cmd_t_test *cmd)
1048 for (i=0; i < cmd->n_variables; ++i)
1057 double pooled_variance;
1058 double std_err_diff;
1061 const struct variable *var = cmd->v_variables[i];
1062 struct group_proc *grp_data = group_proc_get (var);
1064 struct hsh_table *grp_hash = grp_data->group_hash;
1066 struct group_statistics *gs0 ;
1067 struct group_statistics *gs1 ;
1069 union value search_val;
1071 if ( gp.criterion == CMP_LE )
1072 search_val.f = gp.v.critical_value - 1.0;
1074 search_val = gp.v.g_value[0];
1076 gs0 = hsh_find (grp_hash, (void *) &search_val);
1079 if ( gp.criterion == CMP_LE )
1080 search_val.f = gp.v.critical_value + 1.0;
1082 search_val = gp.v.g_value[1];
1084 gs1 = hsh_find (grp_hash, (void *) &search_val);
1088 tab_text (self->t, 0, i*2+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1090 tab_text (self->t, 1, i*2+3, TAB_LEFT, _ ("Equal variances assumed"));
1093 tab_double (self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, NULL);
1095 /* Now work out the significance of the Levene test */
1096 df1 = 1; df2 = grp_data->ugs.n - 2;
1097 q = gsl_cdf_fdist_Q (grp_data->levene, df1, df2);
1099 tab_double (self->t, 3, i*2+3, TAB_CENTER, q, NULL);
1101 df = gs0->n + gs1->n - 2.0 ;
1102 tab_double (self->t, 5, i*2+3, TAB_RIGHT, df, NULL);
1104 pooled_variance = ( (gs0->n )*pow2 (gs0->s_std_dev)
1106 (gs1->n )*pow2 (gs1->s_std_dev)
1109 t = (gs0->mean - gs1->mean) / sqrt (pooled_variance) ;
1110 t /= sqrt ((gs0->n + gs1->n)/ (gs0->n*gs1->n));
1112 tab_double (self->t, 4, i*2+3, TAB_RIGHT, t, NULL);
1114 p = gsl_cdf_tdist_P (t, df);
1115 q = gsl_cdf_tdist_Q (t, df);
1117 tab_double (self->t, 6, i*2+3, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1119 mean_diff = gs0->mean - gs1->mean;
1120 tab_double (self->t, 7, i*2+3, TAB_RIGHT, mean_diff, NULL);
1123 std_err_diff = sqrt ( pow2 (gs0->se_mean) + pow2 (gs1->se_mean));
1124 tab_double (self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, NULL);
1127 /* Now work out the confidence interval */
1128 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1130 t = gsl_cdf_tdist_Qinv (q,df);
1131 tab_double (self->t, 9, i*2+3, TAB_RIGHT,
1132 mean_diff - t * std_err_diff, NULL);
1134 tab_double (self->t, 10, i*2+3, TAB_RIGHT,
1135 mean_diff + t * std_err_diff, NULL);
1140 /* Now for the \sigma_1 != \sigma_2 case */
1141 tab_text (self->t, 1, i*2+3+1,
1142 TAB_LEFT, _ ("Equal variances not assumed"));
1145 se2 = (pow2 (gs0->s_std_dev)/ (gs0->n -1) ) +
1146 (pow2 (gs1->s_std_dev)/ (gs1->n -1) );
1148 t = mean_diff / sqrt (se2) ;
1149 tab_double (self->t, 4, i*2+3+1, TAB_RIGHT, t, NULL);
1152 (pow2 (pow2 (gs0->s_std_dev)/ (gs0->n - 1 ))
1156 (pow2 (pow2 (gs1->s_std_dev)/ (gs1->n - 1 ))
1161 tab_double (self->t, 5, i*2+3+1, TAB_RIGHT, df, NULL);
1163 p = gsl_cdf_tdist_P (t, df);
1164 q = gsl_cdf_tdist_Q (t, df);
1166 tab_double (self->t, 6, i*2+3+1, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1168 /* Now work out the confidence interval */
1169 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1171 t = gsl_cdf_tdist_Qinv (q, df);
1173 tab_double (self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, NULL);
1176 tab_double (self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, NULL);
1179 tab_double (self->t, 9, i*2+3+1, TAB_RIGHT,
1180 mean_diff - t * std_err_diff, NULL);
1182 tab_double (self->t, 10, i*2+3+1, TAB_RIGHT,
1183 mean_diff + t * std_err_diff, NULL);
1188 /* Initialize the paired samples trbox */
1190 trbox_paired_init (struct trbox *self,
1191 struct cmd_t_test *cmd UNUSED)
1195 const int vsize=n_pairs+3;
1197 self->populate = trbox_paired_populate;
1199 trbox_base_init (self,n_pairs,hsize);
1200 tab_title (self->t, _ ("Paired Samples Test"));
1201 tab_hline (self->t,TAL_1,2,6,1);
1202 tab_vline (self->t,TAL_2,2,0,vsize - 1);
1203 tab_joint_text (self->t,2,0,6,0,TAB_CENTER,_ ("Paired Differences"));
1204 tab_box (self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1205 tab_box (self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1206 tab_hline (self->t,TAL_1,5,6, 2);
1207 tab_vline (self->t,TAL_GAP,6,0,1);
1209 tab_joint_text (self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1210 _ ("%g%% Confidence Interval of the Difference"),
1211 cmd->criteria*100.0);
1213 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _ ("Mean"));
1214 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
1215 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _ ("Std. Error Mean"));
1216 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _ ("Lower"));
1217 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _ ("Upper"));
1218 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _ ("t"));
1219 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _ ("df"));
1220 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _ ("Sig. (2-tailed)"));
1223 /* Populate the paired samples trbox */
1225 trbox_paired_populate (struct trbox *trb,
1226 const struct dictionary *dict,
1227 struct cmd_t_test *cmd UNUSED)
1231 const struct variable *wv = dict_get_weight (dict);
1232 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1234 for (i=0; i < n_pairs; ++i)
1239 double n = pairs[i].n;
1243 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _ ("Pair %d"),i);
1245 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1246 var_get_name (pairs[i].v[0]),
1247 var_get_name (pairs[i].v[1]));
1249 tab_double (trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, NULL);
1251 tab_double (trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, NULL);
1254 se_mean = pairs[i].std_dev_diff / sqrt (n) ;
1255 tab_double (trb->t, 4, i+3, TAB_RIGHT, se_mean, NULL);
1257 /* Now work out the confidence interval */
1258 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1260 t = gsl_cdf_tdist_Qinv (q, df);
1262 tab_double (trb->t, 5, i+3, TAB_RIGHT,
1263 pairs[i].mean_diff - t * se_mean , NULL);
1265 tab_double (trb->t, 6, i+3, TAB_RIGHT,
1266 pairs[i].mean_diff + t * se_mean , NULL);
1268 t = (pairs[i].mean[0] - pairs[i].mean[1])
1270 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1271 2 * pairs[i].correlation *
1272 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1276 tab_double (trb->t, 7, i+3, TAB_RIGHT, t, NULL);
1278 /* Degrees of freedom */
1279 tab_double (trb->t, 8, i+3, TAB_RIGHT, df, wfmt);
1281 p = gsl_cdf_tdist_P (t,df);
1282 q = gsl_cdf_tdist_Q (t,df);
1284 tab_double (trb->t, 9, i+3, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1289 /* Initialize the one sample trbox */
1291 trbox_one_sample_init (struct trbox *self, struct cmd_t_test *cmd )
1294 const int vsize=cmd->n_variables+3;
1296 self->populate = trbox_one_sample_populate;
1298 trbox_base_init (self, cmd->n_variables,hsize);
1299 tab_title (self->t, _ ("One-Sample Test"));
1300 tab_hline (self->t, TAL_1, 1, hsize - 1, 1);
1301 tab_vline (self->t, TAL_2, 1, 0, vsize - 1);
1303 tab_joint_text (self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1304 _ ("Test Value = %f"), cmd->n_testval[0]);
1306 tab_box (self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1309 tab_joint_text (self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1310 _ ("%g%% Confidence Interval of the Difference"),
1311 cmd->criteria*100.0);
1313 tab_vline (self->t,TAL_GAP,6,1,1);
1314 tab_hline (self->t,TAL_1,5,6,2);
1315 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _ ("t"));
1316 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _ ("df"));
1317 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _ ("Sig. (2-tailed)"));
1318 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _ ("Mean Difference"));
1319 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _ ("Lower"));
1320 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _ ("Upper"));
1325 /* Populate the one sample trbox */
1327 trbox_one_sample_populate (struct trbox *trb,
1328 const struct dictionary *dict,
1329 struct cmd_t_test *cmd)
1333 const struct variable *wv = dict_get_weight (dict);
1334 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1338 for (i=0; i < cmd->n_variables; ++i)
1343 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1346 tab_text (trb->t, 0, i+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1348 t = (gs->mean - cmd->n_testval[0] ) * sqrt (gs->n) / gs->std_dev ;
1350 tab_double (trb->t, 1, i+3, TAB_RIGHT, t, NULL);
1352 /* degrees of freedom */
1355 tab_double (trb->t, 2, i+3, TAB_RIGHT, df, wfmt);
1357 p = gsl_cdf_tdist_P (t, df);
1358 q = gsl_cdf_tdist_Q (t, df);
1360 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1362 tab_double (trb->t, 3, i+3, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1364 tab_double (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, NULL);
1367 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1368 t = gsl_cdf_tdist_Qinv (q, df);
1370 tab_double (trb->t, 5, i+3, TAB_RIGHT,
1371 gs->mean_diff - t * gs->se_mean, NULL);
1373 tab_double (trb->t, 6, i+3, TAB_RIGHT,
1374 gs->mean_diff + t * gs->se_mean, NULL);
1378 /* Base initializer for the generalized trbox */
1380 trbox_base_init (struct trbox *self, size_t data_rows, int cols)
1382 const size_t rows = 3 + data_rows;
1384 self->finalize = trbox_base_finalize;
1385 self->t = tab_create (cols, rows, 0);
1386 tab_headers (self->t,0,0,3,0);
1387 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1388 tab_hline (self->t, TAL_2,0,cols-1,3);
1389 tab_dim (self->t, tab_natural_dimensions);
1393 /* Base finalizer for the trbox */
1395 trbox_base_finalize (struct trbox *trb)
1397 tab_submit (trb->t);
1401 /* Create , populate and submit the Paired Samples Correlation box */
1403 pscbox (const struct dictionary *dict)
1405 const struct variable *wv = dict_get_weight (dict);
1406 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1408 const int rows = 1 + n_pairs;
1412 struct tab_table *table;
1414 table = tab_create (cols,rows,0);
1416 tab_columns (table, SOM_COL_DOWN, 1);
1417 tab_headers (table,0,0,1,0);
1418 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1419 tab_hline (table, TAL_2, 0, cols - 1, 1);
1420 tab_vline (table, TAL_2, 2, 0, rows - 1);
1421 tab_dim (table, tab_natural_dimensions);
1422 tab_title (table, _ ("Paired Samples Correlations"));
1424 /* column headings */
1425 tab_text (table, 2,0, TAB_CENTER | TAT_TITLE, _ ("N"));
1426 tab_text (table, 3,0, TAB_CENTER | TAT_TITLE, _ ("Correlation"));
1427 tab_text (table, 4,0, TAB_CENTER | TAT_TITLE, _ ("Sig."));
1429 for (i=0; i < n_pairs; ++i)
1433 double df = pairs[i].n -2;
1435 /* corr2 will mathematically always be in the range [0, 1.0]. Inaccurate
1436 calculations sometimes cause it to be slightly greater than 1.0, so
1437 force it into the correct range to avoid NaN from sqrt(). */
1438 double corr2 = MIN (1.0, pow2 (pairs[i].correlation));
1439 double correlation_t =
1440 pairs[i].correlation * sqrt (df) /
1445 tab_text (table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1448 tab_text (table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1450 var_get_name (pairs[i].v[0]),
1451 var_get_name (pairs[i].v[1]));
1455 tab_double (table, 2, i+1, TAB_RIGHT, pairs[i].n, wfmt);
1456 tab_double (table, 3, i+1, TAB_RIGHT, pairs[i].correlation, NULL);
1458 p = gsl_cdf_tdist_P (correlation_t, df);
1459 q = gsl_cdf_tdist_Q (correlation_t, df);
1461 tab_double (table, 4, i+1, TAB_RIGHT, 2.0* (correlation_t>0?q:p), NULL);
1470 /* Calculation Implementation */
1472 /* Per case calculations common to all variants of the T test */
1474 common_calc (const struct dictionary *dict,
1475 const struct ccase *c,
1477 enum mv_class exclude)
1480 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1482 double weight = dict_get_case_weight (dict, c, NULL);
1485 /* Listwise has to be implicit if the independent variable is missing ?? */
1486 if ( cmd->sbc_groups )
1488 if (var_is_value_missing (indep_var, case_data (c, indep_var), exclude))
1492 for (i = 0; i < cmd->n_variables ; ++i)
1494 const struct variable *v = cmd->v_variables[i];
1495 const union value *val = case_data (c, v);
1497 if (!var_is_value_missing (v, val, exclude))
1499 struct group_statistics *gs;
1500 gs = &group_proc_get (v)->ugs;
1503 gs->sum += weight * val->f;
1504 gs->ssq += weight * val->f * val->f;
1510 /* Pre calculations common to all variants of the T test */
1512 common_precalc ( struct cmd_t_test *cmd )
1516 for (i=0; i< cmd->n_variables ; ++i)
1518 struct group_statistics *gs;
1519 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1528 /* Post calculations common to all variants of the T test */
1530 common_postcalc (struct cmd_t_test *cmd)
1534 for (i=0; i< cmd->n_variables ; ++i)
1536 struct group_statistics *gs;
1537 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1539 gs->mean=gs->sum / gs->n;
1540 gs->s_std_dev= sqrt (
1541 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1546 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1549 gs->se_mean = gs->std_dev / sqrt (gs->n);
1550 gs->mean_diff= gs->sum_diff / gs->n;
1554 /* Per case calculations for one sample t test */
1556 one_sample_calc (const struct dictionary *dict,
1557 const struct ccase *c, void *cmd_,
1558 enum mv_class exclude)
1562 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1564 double weight = dict_get_case_weight (dict, c, NULL);
1567 for (i=0; i< cmd->n_variables ; ++i)
1569 struct group_statistics *gs;
1570 const struct variable *v = cmd->v_variables[i];
1571 const union value *val = case_data (c, v);
1573 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1575 if (!var_is_value_missing (v, val, exclude))
1576 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1582 /* Pre calculations for one sample t test */
1584 one_sample_precalc ( struct cmd_t_test *cmd )
1588 for (i=0; i< cmd->n_variables ; ++i)
1590 struct group_statistics *gs;
1591 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1597 /* Post calculations for one sample t test */
1599 one_sample_postcalc (struct cmd_t_test *cmd)
1603 for (i=0; i< cmd->n_variables ; ++i)
1605 struct group_statistics *gs;
1606 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1608 gs->mean_diff = gs->sum_diff / gs->n ;
1615 paired_precalc (struct cmd_t_test *cmd UNUSED)
1619 for (i=0; i < n_pairs ; ++i )
1622 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1623 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1624 pairs[i].sum_of_prod = 0;
1625 pairs[i].correlation = 0;
1626 pairs[i].sum_of_diffs = 0;
1627 pairs[i].ssq_diffs = 0;
1634 paired_calc (const struct dictionary *dict, const struct ccase *c,
1635 struct cmd_t_test *cmd UNUSED, enum mv_class exclude)
1639 double weight = dict_get_case_weight (dict, c, NULL);
1641 for (i=0; i < n_pairs ; ++i )
1643 const struct variable *v0 = pairs[i].v[0];
1644 const struct variable *v1 = pairs[i].v[1];
1646 const union value *val0 = case_data (c, v0);
1647 const union value *val1 = case_data (c, v1);
1649 if (!var_is_value_missing (v0, val0, exclude) &&
1650 !var_is_value_missing (v1, val1, exclude))
1652 pairs[i].n += weight;
1653 pairs[i].sum[0] += weight * val0->f;
1654 pairs[i].sum[1] += weight * val1->f;
1656 pairs[i].ssq[0] += weight * pow2 (val0->f);
1657 pairs[i].ssq[1] += weight * pow2 (val1->f);
1659 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1661 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1662 pairs[i].ssq_diffs += weight * pow2 (val0->f - val1->f);
1670 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1674 for (i=0; i < n_pairs ; ++i )
1677 const double n = pairs[i].n;
1679 for (j=0; j < 2 ; ++j)
1681 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1682 pairs[i].s_std_dev[j] = sqrt ((pairs[i].ssq[j] / n -
1683 pow2 (pairs[i].mean[j]))
1686 pairs[i].std_dev[j] = sqrt (n/ (n-1)* (pairs[i].ssq[j] / n -
1687 pow2 (pairs[i].mean[j]))
1691 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1692 pairs[i].mean[0] * pairs[i].mean[1] ;
1693 /* correlation now actually contains the covariance */
1695 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1696 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1698 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1700 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1701 ( pairs[i].ssq_diffs / n )
1703 pow2 (pairs[i].mean_diff )
1709 group_precalc (struct cmd_t_test *cmd )
1714 for (i=0; i< cmd->n_variables ; ++i)
1716 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1718 /* There's always 2 groups for a T - TEST */
1721 gp.indep_width = var_get_width (indep_var);
1723 ttpr->group_hash = hsh_create (2,
1724 (hsh_compare_func *) compare_group_binary,
1725 (hsh_hash_func *) hash_group_binary,
1726 (hsh_free_func *) free_group,
1729 for (j=0 ; j < 2 ; ++j)
1731 struct group_statistics *gs = xmalloc (sizeof *gs);
1737 if ( gp.criterion == CMP_EQ )
1739 gs->id = gp.v.g_value[j];
1744 gs->id.f = gp.v.critical_value - 1.0 ;
1746 gs->id.f = gp.v.critical_value + 1.0 ;
1749 hsh_insert ( ttpr->group_hash, (void *) gs );
1756 group_calc (const struct dictionary *dict,
1757 const struct ccase *c, struct cmd_t_test *cmd,
1758 enum mv_class exclude)
1762 const double weight = dict_get_case_weight (dict, c, NULL);
1764 const union value *gv;
1766 if (var_is_value_missing (indep_var, case_data (c, indep_var), exclude))
1769 gv = case_data (c, indep_var);
1771 for (i=0; i< cmd->n_variables ; ++i)
1773 const struct variable *var = cmd->v_variables[i];
1774 const union value *val = case_data (c, var);
1775 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1776 struct group_statistics *gs;
1778 gs = hsh_find (grp_hash, (void *) gv);
1780 /* If the independent variable doesn't match either of the values
1781 for this case then move on to the next case */
1785 if (!var_is_value_missing (var, val, exclude))
1788 gs->sum += weight * val->f;
1789 gs->ssq += weight * pow2 (val->f);
1798 group_postcalc ( struct cmd_t_test *cmd )
1802 for (i = 0; i < cmd->n_variables ; ++i)
1804 const struct variable *var = cmd->v_variables[i];
1805 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1806 struct hsh_iterator g;
1807 struct group_statistics *gs;
1810 for (gs = hsh_first (grp_hash,&g);
1812 gs = hsh_next (grp_hash,&g))
1814 gs->mean = gs->sum / gs->n;
1816 gs->s_std_dev= sqrt (
1817 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1822 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1825 gs->se_mean = gs->std_dev / sqrt (gs->n);
1828 assert (count == 2);
1835 calculate (struct cmd_t_test *cmd,
1836 struct casereader *input, const struct dataset *ds)
1838 const struct dictionary *dict = dataset_dict (ds);
1839 struct ssbox stat_summary_box;
1840 struct trbox test_results_box;
1842 struct casereader *pass1, *pass2, *pass3;
1843 struct taint *taint;
1846 enum mv_class exclude = cmd->miss != TTS_INCLUDE ? MV_ANY : MV_SYSTEM;
1848 if (!casereader_peek (input, 0, &c))
1850 casereader_destroy (input);
1853 output_split_file_values (ds, &c);
1856 if ( cmd->miss == TTS_LISTWISE )
1857 input = casereader_create_filter_missing (input,
1862 input = casereader_create_filter_weight (input, dict, NULL, NULL);
1864 taint = taint_clone (casereader_get_taint (input));
1865 casereader_split (input, &pass1, &pass2);
1867 common_precalc (cmd);
1868 for (; casereader_read (pass1, &c); case_destroy (&c))
1869 common_calc (dict, &c, cmd, exclude);
1870 casereader_destroy (pass1);
1871 common_postcalc (cmd);
1876 one_sample_precalc (cmd);
1877 for (; casereader_read (pass2, &c); case_destroy (&c))
1878 one_sample_calc (dict, &c, cmd, exclude);
1879 one_sample_postcalc (cmd);
1882 paired_precalc (cmd);
1883 for (; casereader_read (pass2, &c); case_destroy (&c))
1884 paired_calc (dict, &c, cmd, exclude);
1885 paired_postcalc (cmd);
1888 pass3 = casereader_clone (pass2);
1890 group_precalc (cmd);
1891 for (; casereader_read (pass2, &c); case_destroy (&c))
1892 group_calc (dict, &c, cmd, exclude);
1893 group_postcalc (cmd);
1895 levene (dict, pass3, indep_var, cmd->n_variables, cmd->v_variables,
1899 casereader_destroy (pass2);
1901 if (!taint_has_tainted_successor (taint))
1903 ssbox_create (&stat_summary_box,cmd,mode);
1904 ssbox_populate (&stat_summary_box, dict, cmd);
1905 ssbox_finalize (&stat_summary_box);
1907 if ( mode == T_PAIRED )
1910 trbox_create (&test_results_box, cmd, mode);
1911 trbox_populate (&test_results_box, dict, cmd);
1912 trbox_finalize (&test_results_box);
1915 taint_destroy (taint);
1918 short which_group (const struct group_statistics *g,
1919 const struct group_properties *p);
1921 /* Return -1 if the id of a is less than b; +1 if greater than and
1924 compare_group_binary (const struct group_statistics *a,
1925 const struct group_statistics *b,
1926 const struct group_properties *p)
1931 if ( p->criterion == CMP_LE )
1933 /* less-than comparision is not meaningfull for
1934 alpha variables, so we shouldn't ever arrive here */
1935 assert (p->indep_width == 0 ) ;
1937 flag_a = ( a->id.f < p->v.critical_value ) ;
1938 flag_b = ( b->id.f < p->v.critical_value ) ;
1942 flag_a = which_group (a, p);
1943 flag_b = which_group (b, p);
1946 if (flag_a < flag_b )
1949 return (flag_a > flag_b);
1952 /* This is a degenerate case of a hash, since it can only return three possible
1953 values. It's really a comparison, being used as a hash function */
1956 hash_group_binary (const struct group_statistics *g,
1957 const struct group_properties *p)
1961 if ( p->criterion == CMP_LE )
1963 /* Not meaningfull to do a less than compare for alpha values ? */
1964 assert (p->indep_width == 0 ) ;
1965 flag = ( g->id.f < p->v.critical_value ) ;
1967 else if ( p->criterion == CMP_EQ)
1969 flag = which_group (g,p);
1977 /* return 0 if G belongs to group 0,
1978 1 if it belongs to group 1,
1979 2 if it belongs to neither group */
1981 which_group (const struct group_statistics *g,
1982 const struct group_properties *p)
1984 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1987 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))