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 independent variable */
85 struct variable *indep_var;
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 int compare_group_binary (const struct group_statistics *a,
244 const struct group_statistics *b,
245 const struct group_properties *p);
248 static unsigned hash_group_binary (const struct group_statistics *g,
249 const struct group_properties *p);
254 cmd_t_test (struct lexer *lexer, struct dataset *ds)
256 struct cmd_t_test cmd;
257 struct casegrouper *grouper;
258 struct casereader *group;
261 if ( !parse_t_test (lexer, ds, &cmd, NULL) )
264 if (! cmd.sbc_criteria)
269 if (cmd.sbc_testval) ++m;
270 if (cmd.sbc_groups) ++m;
271 if (cmd.sbc_pairs) ++m;
276 _ ("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
285 else if (cmd.sbc_groups)
290 if ( mode == T_PAIRED)
292 if (cmd.sbc_variables)
294 msg (SE, _ ("VARIABLES subcommand is not appropriate with PAIRS"));
300 /* Iterate through the pairs and put each variable that is a
301 member of a pair into cmd.v_variables */
304 struct hsh_iterator hi;
305 struct const_hsh_table *hash;
306 const struct variable *v;
308 hash = const_hsh_create (n_pairs, compare_vars_by_name, hash_var_by_name,
311 for (i=0; i < n_pairs; ++i)
313 const_hsh_insert (hash, pairs[i].v[0]);
314 const_hsh_insert (hash, pairs[i].v[1]);
317 assert (cmd.n_variables == 0);
318 cmd.n_variables = const_hsh_count (hash);
320 cmd.v_variables = xnrealloc (cmd.v_variables, cmd.n_variables,
321 sizeof *cmd.v_variables);
322 /* Iterate through the hash */
323 for (i=0,v = const_hsh_first (hash, &hi);
325 v = const_hsh_next (hash, &hi) )
326 cmd.v_variables[i++]=v;
327 const_hsh_destroy (hash);
330 else if ( !cmd.sbc_variables)
332 msg (SE, _ ("One or more VARIABLES must be specified."));
338 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
339 while (casegrouper_get_next_group (grouper, &group))
340 calculate (&cmd, group, ds);
341 ok = casegrouper_destroy (grouper);
342 ok = proc_commit (ds) && ok;
348 if ( mode == T_IND_SAMPLES)
351 /* Destroy any group statistics we created */
352 for (v = 0 ; v < cmd.n_variables ; ++v )
354 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
355 hsh_destroy (grpp->group_hash);
360 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
364 tts_custom_groups (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd 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_width (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_width (indep_var)))
429 if (!lex_force_match (lexer, ')'))
432 if ( n_group_values == 2 )
433 gp.criterion = CMP_EQ ;
435 gp.criterion = CMP_LE ;
438 if ( var_is_alpha (indep_var))
440 buf_copy_rpad (gp.v.g_value [0].s, var_get_width (indep_var),
441 gp.v.g_value [0].s, strlen (gp.v.g_value[0].s));
443 buf_copy_rpad (gp.v.g_value [1].s, var_get_width (indep_var),
444 gp.v.g_value [1].s, strlen (gp.v.g_value[1].s));
452 tts_custom_pairs (struct lexer *lexer, struct dataset *ds, struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
454 const struct variable **vars;
456 size_t n_pairs_local;
458 size_t n_before_WITH;
459 size_t n_after_WITH = SIZE_MAX;
460 int paired ; /* Was the PAIRED keyword given ? */
462 lex_match (lexer, '=');
465 if (!parse_variables_const (lexer, dataset_dict (ds), &vars, &n_vars,
466 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
474 if (lex_match (lexer, T_WITH))
476 n_before_WITH = n_vars;
477 if (!parse_variables_const (lexer, dataset_dict (ds), &vars, &n_vars,
478 PV_DUPLICATE | PV_APPEND
479 | PV_NUMERIC | PV_NO_SCRATCH))
484 n_after_WITH = n_vars - n_before_WITH;
487 paired = (lex_match (lexer, '(') && lex_match_id (lexer, "PAIRED") && lex_match (lexer, ')'));
489 /* Determine the number of pairs needed */
492 if (n_before_WITH != n_after_WITH)
495 msg (SE, _ ("PAIRED was specified but the number of variables "
496 "preceding WITH (%zu) did not match the number "
498 n_before_WITH, n_after_WITH);
501 n_pairs_local = n_before_WITH;
503 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
505 n_pairs_local = n_before_WITH * n_after_WITH ;
507 else /* Neither WITH nor PAIRED keyword given */
512 msg (SE, _ ("At least two variables must be specified "
517 /* how many ways can you pick 2 from n_vars ? */
518 n_pairs_local = n_vars * (n_vars - 1) / 2;
522 /* Allocate storage for the pairs */
523 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
525 /* Populate the pairs with the appropriate variables */
530 assert (n_pairs_local == n_vars / 2);
531 for (i = 0; i < n_pairs_local; ++i)
533 pairs[i].v[n_pairs] = vars[i];
534 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
537 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
542 for (i=0 ; i < n_before_WITH ; ++i )
544 for (j=0 ; j < n_after_WITH ; ++j)
546 pairs[p].v[0] = vars[i];
547 pairs[p].v[1] = vars[j+n_before_WITH];
552 else /* Neither WITH nor PAIRED given */
557 for (i=0 ; i < n_vars ; ++i )
559 for (j=i+1 ; j < n_vars ; ++j)
561 pairs[p].v[0] = vars[i];
562 pairs[p].v[1] = vars[j];
568 n_pairs+=n_pairs_local;
574 /* Parses the current token (numeric or string, depending on type)
575 value v and returns success. */
577 parse_value (struct lexer *lexer, union value * v, enum val_type type)
579 if (type == VAL_NUMERIC)
581 if (!lex_force_num (lexer))
583 v->f = lex_tokval (lexer);
587 if (!lex_force_string (lexer))
589 memset (v->s, ' ', MAX_SHORT_STRING);
590 strncpy (v->s, ds_cstr (lex_tokstr (lexer)), ds_length (lex_tokstr (lexer)));
599 /* Implementation of the SSBOX object */
601 void ssbox_base_init (struct ssbox *this, int cols,int rows);
603 void ssbox_base_finalize (struct ssbox *ssb);
605 void ssbox_one_sample_init (struct ssbox *this,
606 struct cmd_t_test *cmd );
608 void ssbox_independent_samples_init (struct ssbox *this,
609 struct cmd_t_test *cmd);
611 void ssbox_paired_init (struct ssbox *this,
612 struct cmd_t_test *cmd);
615 /* Factory to create an ssbox */
617 ssbox_create (struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
622 ssbox_one_sample_init (ssb,cmd);
625 ssbox_independent_samples_init (ssb,cmd);
628 ssbox_paired_init (ssb,cmd);
637 /* Despatcher for the populate method */
639 ssbox_populate (struct ssbox *ssb, const struct dictionary *dict,
640 struct cmd_t_test *cmd)
642 ssb->populate (ssb, dict, cmd);
646 /* Despatcher for finalize */
648 ssbox_finalize (struct ssbox *ssb)
654 /* Submit the box and clear up */
656 ssbox_base_finalize (struct ssbox *ssb)
663 /* Initialize a ssbox struct */
665 ssbox_base_init (struct ssbox *this, int cols,int rows)
667 this->finalize = ssbox_base_finalize;
668 this->t = tab_create (cols, rows, 0);
670 tab_columns (this->t, SOM_COL_DOWN, 1);
671 tab_headers (this->t,0,0,1,0);
672 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
673 tab_hline (this->t, TAL_2,0,cols-1,1);
674 tab_dim (this->t, tab_natural_dimensions);
677 void ssbox_one_sample_populate (struct ssbox *ssb,
678 const struct dictionary *,
679 struct cmd_t_test *cmd);
681 /* Initialize the one_sample ssbox */
683 ssbox_one_sample_init (struct ssbox *this,
684 struct cmd_t_test *cmd )
687 const int vsize=cmd->n_variables+1;
689 this->populate = ssbox_one_sample_populate;
691 ssbox_base_init (this, hsize,vsize);
692 tab_title (this->t, _ ("One-Sample Statistics"));
693 tab_vline (this->t, TAL_2, 1,0,vsize - 1);
694 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
695 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
696 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
697 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
700 static void ssbox_independent_samples_populate (struct ssbox *ssb,
701 const struct dictionary *,
702 struct cmd_t_test *cmd);
704 /* Initialize the independent samples ssbox */
706 ssbox_independent_samples_init (struct ssbox *this,
707 struct cmd_t_test *cmd)
710 int vsize = cmd->n_variables*2 +1;
712 this->populate = ssbox_independent_samples_populate;
714 ssbox_base_init (this, hsize,vsize);
715 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
716 tab_title (this->t, _ ("Group Statistics"));
717 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, var_get_name (indep_var));
718 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
719 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
720 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
721 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
725 /* Populate the ssbox for independent samples */
727 ssbox_independent_samples_populate (struct ssbox *ssb,
728 const struct dictionary *dict,
729 struct cmd_t_test *cmd)
733 const struct variable *wv = dict_get_weight (dict);
734 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
736 char *val_lab[2] = {NULL, NULL};
737 double indep_value[2];
739 char prefix[2][3]={"",""};
741 if ( var_is_numeric (indep_var) )
745 s = var_lookup_value_label (indep_var, &gp.v.g_value[0]);
746 val_lab[0] = s ? xstrdup (s) : NULL;
748 s = var_lookup_value_label (indep_var, &gp.v.g_value[1]);
749 val_lab[1] = s ? xstrdup (s) : NULL;
753 val_lab[0] = xcalloc (sizeof (char), MAX_SHORT_STRING + 1);
754 val_lab[1] = xcalloc (sizeof (char), MAX_SHORT_STRING + 1);
755 memcpy (val_lab[0], gp.v.g_value[0].s, MAX_SHORT_STRING);
756 memcpy (val_lab[1], gp.v.g_value[1].s, MAX_SHORT_STRING);
759 if (gp.criterion == CMP_LE )
761 strcpy (prefix[0],">=");
762 strcpy (prefix[1],"<");
763 indep_value[0] = gp.v.critical_value;
764 indep_value[1] = gp.v.critical_value;
768 indep_value[0] = gp.v.g_value[0].f;
769 indep_value[1] = gp.v.g_value[1].f;
774 for (i=0; i < cmd->n_variables; ++i)
776 const struct variable *var = cmd->v_variables[i];
777 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
780 tab_text (ssb->t, 0, i*2+1, TAB_LEFT,
781 var_get_name (cmd->v_variables[i]));
784 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
785 "%s%s", prefix[0], val_lab[0]);
787 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
788 "%s%g", prefix[0], indep_value[0]);
792 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
793 "%s%s", prefix[1], val_lab[1]);
795 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
796 "%s%g", prefix[1], indep_value[1]);
799 /* Fill in the group statistics */
800 for ( count = 0 ; count < 2 ; ++count )
802 union value search_val;
804 struct group_statistics *gs;
806 if ( gp.criterion == CMP_LE )
811 search_val.f = gp.v.critical_value + 1.0;
815 /* less than ( < ) case */
816 search_val.f = gp.v.critical_value - 1.0;
821 search_val = gp.v.g_value[count];
824 gs = hsh_find (grp_hash, (void *) &search_val);
827 tab_double (ssb->t, 2, i*2+count+1, TAB_RIGHT, gs->n, wfmt);
828 tab_double (ssb->t, 3, i*2+count+1, TAB_RIGHT, gs->mean, NULL);
829 tab_double (ssb->t, 4, i*2+count+1, TAB_RIGHT, gs->std_dev, NULL);
830 tab_double (ssb->t, 5, i*2+count+1, TAB_RIGHT, gs->se_mean, NULL);
838 static void ssbox_paired_populate (struct ssbox *ssb,
839 const struct dictionary *dict,
840 struct cmd_t_test *cmd);
842 /* Initialize the paired values ssbox */
844 ssbox_paired_init (struct ssbox *this, struct cmd_t_test *cmd UNUSED)
848 int vsize = n_pairs*2+1;
850 this->populate = ssbox_paired_populate;
852 ssbox_base_init (this, hsize,vsize);
853 tab_title (this->t, _ ("Paired Sample Statistics"));
854 tab_vline (this->t,TAL_GAP,1,0,vsize-1);
855 tab_vline (this->t,TAL_2,2,0,vsize-1);
856 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _ ("Mean"));
857 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _ ("N"));
858 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
859 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _ ("SE. Mean"));
863 /* Populate the ssbox for paired values */
865 ssbox_paired_populate (struct ssbox *ssb, const struct dictionary *dict,
866 struct cmd_t_test *cmd UNUSED)
870 const struct variable *wv = dict_get_weight (dict);
871 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
875 for (i=0; i < n_pairs; ++i)
879 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _ ("Pair %d"),i);
881 for (j=0 ; j < 2 ; ++j)
883 struct group_statistics *gs;
885 gs = &group_proc_get (pairs[i].v[j])->ugs;
889 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT,
890 var_get_name (pairs[i].v[j]));
893 tab_double (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], NULL);
894 tab_double (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, wfmt);
895 tab_double (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], NULL);
896 tab_double (ssb->t,5, i*2+j+1, TAB_RIGHT,
897 pairs[i].std_dev[j]/sqrt (pairs[i].n), NULL);
903 /* Populate the one sample ssbox */
905 ssbox_one_sample_populate (struct ssbox *ssb, const struct dictionary *dict,
906 struct cmd_t_test *cmd)
910 const struct variable *wv = dict_get_weight (dict);
911 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
915 for (i=0; i < cmd->n_variables; ++i)
917 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
919 tab_text (ssb->t, 0, i+1, TAB_LEFT, var_get_name (cmd->v_variables[i]));
920 tab_double (ssb->t,1, i+1, TAB_RIGHT, gs->n, wfmt);
921 tab_double (ssb->t,2, i+1, TAB_RIGHT, gs->mean, NULL);
922 tab_double (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, NULL);
923 tab_double (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, NULL);
929 /* Implementation of the Test Results box struct */
931 void trbox_base_init (struct trbox *self,size_t n_vars, int cols);
932 void trbox_base_finalize (struct trbox *trb);
934 void trbox_independent_samples_init (struct trbox *trb,
935 struct cmd_t_test *cmd );
937 static void trbox_independent_samples_populate (struct trbox *trb,
938 const struct dictionary *dict,
939 struct cmd_t_test *cmd);
941 void trbox_one_sample_init (struct trbox *self,
942 struct cmd_t_test *cmd );
944 static void trbox_one_sample_populate (struct trbox *trb,
945 const struct dictionary *,
946 struct cmd_t_test *cmd);
948 void trbox_paired_init (struct trbox *self,
949 struct cmd_t_test *cmd );
951 static void trbox_paired_populate (struct trbox *trb,
952 const struct dictionary *,
953 struct cmd_t_test *cmd);
957 /* Create a trbox according to mode*/
959 trbox_create (struct trbox *trb,
960 struct cmd_t_test *cmd, int mode)
965 trbox_one_sample_init (trb,cmd);
968 trbox_independent_samples_init (trb,cmd);
971 trbox_paired_init (trb,cmd);
978 /* Populate a trbox according to cmd */
980 trbox_populate (struct trbox *trb, const struct dictionary *dict,
981 struct cmd_t_test *cmd)
983 trb->populate (trb, dict, cmd);
986 /* Submit and destroy a trbox */
988 trbox_finalize (struct trbox *trb)
993 /* Initialize the independent samples trbox */
995 trbox_independent_samples_init (struct trbox *self,
996 struct cmd_t_test *cmd UNUSED)
999 const int vsize=cmd->n_variables*2+3;
1002 self->populate = trbox_independent_samples_populate;
1004 trbox_base_init (self,cmd->n_variables*2,hsize);
1005 tab_title (self->t,_ ("Independent Samples Test"));
1006 tab_hline (self->t,TAL_1,2,hsize-1,1);
1007 tab_vline (self->t,TAL_2,2,0,vsize-1);
1008 tab_vline (self->t,TAL_1,4,0,vsize-1);
1009 tab_box (self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
1010 tab_hline (self->t,TAL_1, hsize-2,hsize-1,2);
1011 tab_box (self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
1012 tab_joint_text (self->t, 2, 0, 3, 0,
1013 TAB_CENTER,_ ("Levene's Test for Equality of Variances"));
1014 tab_joint_text (self->t, 4,0,hsize-1,0,
1015 TAB_CENTER,_ ("t-test for Equality of Means"));
1017 tab_text (self->t,2,2, TAB_CENTER | TAT_TITLE,_ ("F"));
1018 tab_text (self->t,3,2, TAB_CENTER | TAT_TITLE,_ ("Sig."));
1019 tab_text (self->t,4,2, TAB_CENTER | TAT_TITLE,_ ("t"));
1020 tab_text (self->t,5,2, TAB_CENTER | TAT_TITLE,_ ("df"));
1021 tab_text (self->t,6,2, TAB_CENTER | TAT_TITLE,_ ("Sig. (2-tailed)"));
1022 tab_text (self->t,7,2, TAB_CENTER | TAT_TITLE,_ ("Mean Difference"));
1023 tab_text (self->t,8,2, TAB_CENTER | TAT_TITLE,_ ("Std. Error Difference"));
1024 tab_text (self->t,9,2, TAB_CENTER | TAT_TITLE,_ ("Lower"));
1025 tab_text (self->t,10,2, TAB_CENTER | TAT_TITLE,_ ("Upper"));
1027 tab_joint_text (self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
1028 _ ("%g%% Confidence Interval of the Difference"),
1029 cmd->criteria*100.0);
1033 /* Populate the independent samples trbox */
1035 trbox_independent_samples_populate (struct trbox *self,
1036 const struct dictionary *dict UNUSED,
1037 struct cmd_t_test *cmd)
1042 for (i=0; i < cmd->n_variables; ++i)
1051 double pooled_variance;
1052 double std_err_diff;
1055 const struct variable *var = cmd->v_variables[i];
1056 struct group_proc *grp_data = group_proc_get (var);
1058 struct hsh_table *grp_hash = grp_data->group_hash;
1060 struct group_statistics *gs0 ;
1061 struct group_statistics *gs1 ;
1063 union value search_val;
1065 if ( gp.criterion == CMP_LE )
1066 search_val.f = gp.v.critical_value - 1.0;
1068 search_val = gp.v.g_value[0];
1070 gs0 = hsh_find (grp_hash, (void *) &search_val);
1073 if ( gp.criterion == CMP_LE )
1074 search_val.f = gp.v.critical_value + 1.0;
1076 search_val = gp.v.g_value[1];
1078 gs1 = hsh_find (grp_hash, (void *) &search_val);
1082 tab_text (self->t, 0, i*2+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1084 tab_text (self->t, 1, i*2+3, TAB_LEFT, _ ("Equal variances assumed"));
1087 tab_double (self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, NULL);
1089 /* Now work out the significance of the Levene test */
1090 df1 = 1; df2 = grp_data->ugs.n - 2;
1091 q = gsl_cdf_fdist_Q (grp_data->levene, df1, df2);
1093 tab_double (self->t, 3, i*2+3, TAB_CENTER, q, NULL);
1095 df = gs0->n + gs1->n - 2.0 ;
1096 tab_double (self->t, 5, i*2+3, TAB_RIGHT, df, NULL);
1098 pooled_variance = ( (gs0->n )*pow2 (gs0->s_std_dev)
1100 (gs1->n )*pow2 (gs1->s_std_dev)
1103 t = (gs0->mean - gs1->mean) / sqrt (pooled_variance) ;
1104 t /= sqrt ((gs0->n + gs1->n)/ (gs0->n*gs1->n));
1106 tab_double (self->t, 4, i*2+3, TAB_RIGHT, t, NULL);
1108 p = gsl_cdf_tdist_P (t, df);
1109 q = gsl_cdf_tdist_Q (t, df);
1111 tab_double (self->t, 6, i*2+3, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1113 mean_diff = gs0->mean - gs1->mean;
1114 tab_double (self->t, 7, i*2+3, TAB_RIGHT, mean_diff, NULL);
1117 std_err_diff = sqrt ( pow2 (gs0->se_mean) + pow2 (gs1->se_mean));
1118 tab_double (self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, NULL);
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);
1125 tab_double (self->t, 9, i*2+3, TAB_RIGHT,
1126 mean_diff - t * std_err_diff, NULL);
1128 tab_double (self->t, 10, i*2+3, TAB_RIGHT,
1129 mean_diff + t * std_err_diff, NULL);
1134 /* Now for the \sigma_1 != \sigma_2 case */
1135 tab_text (self->t, 1, i*2+3+1,
1136 TAB_LEFT, _ ("Equal variances not assumed"));
1139 se2 = (pow2 (gs0->s_std_dev)/ (gs0->n -1) ) +
1140 (pow2 (gs1->s_std_dev)/ (gs1->n -1) );
1142 t = mean_diff / sqrt (se2) ;
1143 tab_double (self->t, 4, i*2+3+1, TAB_RIGHT, t, NULL);
1146 (pow2 (pow2 (gs0->s_std_dev)/ (gs0->n - 1 ))
1150 (pow2 (pow2 (gs1->s_std_dev)/ (gs1->n - 1 ))
1155 tab_double (self->t, 5, i*2+3+1, TAB_RIGHT, df, NULL);
1157 p = gsl_cdf_tdist_P (t, df);
1158 q = gsl_cdf_tdist_Q (t, df);
1160 tab_double (self->t, 6, i*2+3+1, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1162 /* Now work out the confidence interval */
1163 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1165 t = gsl_cdf_tdist_Qinv (q, df);
1167 tab_double (self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, NULL);
1170 tab_double (self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, NULL);
1173 tab_double (self->t, 9, i*2+3+1, TAB_RIGHT,
1174 mean_diff - t * std_err_diff, NULL);
1176 tab_double (self->t, 10, i*2+3+1, TAB_RIGHT,
1177 mean_diff + t * std_err_diff, NULL);
1182 /* Initialize the paired samples trbox */
1184 trbox_paired_init (struct trbox *self,
1185 struct cmd_t_test *cmd UNUSED)
1189 const int vsize=n_pairs+3;
1191 self->populate = trbox_paired_populate;
1193 trbox_base_init (self,n_pairs,hsize);
1194 tab_title (self->t, _ ("Paired Samples Test"));
1195 tab_hline (self->t,TAL_1,2,6,1);
1196 tab_vline (self->t,TAL_2,2,0,vsize - 1);
1197 tab_joint_text (self->t,2,0,6,0,TAB_CENTER,_ ("Paired Differences"));
1198 tab_box (self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1199 tab_box (self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1200 tab_hline (self->t,TAL_1,5,6, 2);
1201 tab_vline (self->t,TAL_GAP,6,0,1);
1203 tab_joint_text (self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1204 _ ("%g%% Confidence Interval of the Difference"),
1205 cmd->criteria*100.0);
1207 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _ ("Mean"));
1208 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _ ("Std. Deviation"));
1209 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _ ("Std. Error Mean"));
1210 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _ ("Lower"));
1211 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _ ("Upper"));
1212 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _ ("t"));
1213 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _ ("df"));
1214 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _ ("Sig. (2-tailed)"));
1217 /* Populate the paired samples trbox */
1219 trbox_paired_populate (struct trbox *trb,
1220 const struct dictionary *dict,
1221 struct cmd_t_test *cmd UNUSED)
1225 const struct variable *wv = dict_get_weight (dict);
1226 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1228 for (i=0; i < n_pairs; ++i)
1233 double n = pairs[i].n;
1237 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _ ("Pair %d"),i);
1239 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1240 var_get_name (pairs[i].v[0]),
1241 var_get_name (pairs[i].v[1]));
1243 tab_double (trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, NULL);
1245 tab_double (trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, NULL);
1248 se_mean = pairs[i].std_dev_diff / sqrt (n) ;
1249 tab_double (trb->t, 4, i+3, TAB_RIGHT, se_mean, NULL);
1251 /* Now work out the confidence interval */
1252 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1254 t = gsl_cdf_tdist_Qinv (q, df);
1256 tab_double (trb->t, 5, i+3, TAB_RIGHT,
1257 pairs[i].mean_diff - t * se_mean , NULL);
1259 tab_double (trb->t, 6, i+3, TAB_RIGHT,
1260 pairs[i].mean_diff + t * se_mean , NULL);
1262 t = (pairs[i].mean[0] - pairs[i].mean[1])
1264 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1265 2 * pairs[i].correlation *
1266 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1270 tab_double (trb->t, 7, i+3, TAB_RIGHT, t, NULL);
1272 /* Degrees of freedom */
1273 tab_double (trb->t, 8, i+3, TAB_RIGHT, df, wfmt);
1275 p = gsl_cdf_tdist_P (t,df);
1276 q = gsl_cdf_tdist_P (t,df);
1278 tab_double (trb->t, 9, i+3, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1283 /* Initialize the one sample trbox */
1285 trbox_one_sample_init (struct trbox *self, struct cmd_t_test *cmd )
1288 const int vsize=cmd->n_variables+3;
1290 self->populate = trbox_one_sample_populate;
1292 trbox_base_init (self, cmd->n_variables,hsize);
1293 tab_title (self->t, _ ("One-Sample Test"));
1294 tab_hline (self->t, TAL_1, 1, hsize - 1, 1);
1295 tab_vline (self->t, TAL_2, 1, 0, vsize - 1);
1297 tab_joint_text (self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1298 _ ("Test Value = %f"), cmd->n_testval[0]);
1300 tab_box (self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1303 tab_joint_text (self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1304 _ ("%g%% Confidence Interval of the Difference"),
1305 cmd->criteria*100.0);
1307 tab_vline (self->t,TAL_GAP,6,1,1);
1308 tab_hline (self->t,TAL_1,5,6,2);
1309 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _ ("t"));
1310 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _ ("df"));
1311 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _ ("Sig. (2-tailed)"));
1312 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _ ("Mean Difference"));
1313 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _ ("Lower"));
1314 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _ ("Upper"));
1319 /* Populate the one sample trbox */
1321 trbox_one_sample_populate (struct trbox *trb,
1322 const struct dictionary *dict,
1323 struct cmd_t_test *cmd)
1327 const struct variable *wv = dict_get_weight (dict);
1328 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1332 for (i=0; i < cmd->n_variables; ++i)
1337 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1340 tab_text (trb->t, 0, i+3, TAB_LEFT, var_get_name (cmd->v_variables[i]));
1342 t = (gs->mean - cmd->n_testval[0] ) * sqrt (gs->n) / gs->std_dev ;
1344 tab_double (trb->t, 1, i+3, TAB_RIGHT, t, NULL);
1346 /* degrees of freedom */
1349 tab_double (trb->t, 2, i+3, TAB_RIGHT, df, wfmt);
1351 p = gsl_cdf_tdist_P (t, df);
1352 q = gsl_cdf_tdist_Q (t, df);
1354 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1356 tab_double (trb->t, 3, i+3, TAB_RIGHT, 2.0* (t>0?q:p), NULL);
1358 tab_double (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, NULL);
1361 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1362 t = gsl_cdf_tdist_Qinv (q, df);
1364 tab_double (trb->t, 5, i+3, TAB_RIGHT,
1365 gs->mean_diff - t * gs->se_mean, NULL);
1367 tab_double (trb->t, 6, i+3, TAB_RIGHT,
1368 gs->mean_diff + t * gs->se_mean, NULL);
1372 /* Base initializer for the generalized trbox */
1374 trbox_base_init (struct trbox *self, size_t data_rows, int cols)
1376 const size_t rows = 3 + data_rows;
1378 self->finalize = trbox_base_finalize;
1379 self->t = tab_create (cols, rows, 0);
1380 tab_headers (self->t,0,0,3,0);
1381 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1382 tab_hline (self->t, TAL_2,0,cols-1,3);
1383 tab_dim (self->t, tab_natural_dimensions);
1387 /* Base finalizer for the trbox */
1389 trbox_base_finalize (struct trbox *trb)
1391 tab_submit (trb->t);
1395 /* Create , populate and submit the Paired Samples Correlation box */
1397 pscbox (const struct dictionary *dict)
1399 const struct variable *wv = dict_get_weight (dict);
1400 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : &F_8_0;
1402 const int rows = 1 + n_pairs;
1406 struct tab_table *table;
1408 table = tab_create (cols,rows,0);
1410 tab_columns (table, SOM_COL_DOWN, 1);
1411 tab_headers (table,0,0,1,0);
1412 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1413 tab_hline (table, TAL_2, 0, cols - 1, 1);
1414 tab_vline (table, TAL_2, 2, 0, rows - 1);
1415 tab_dim (table, tab_natural_dimensions);
1416 tab_title (table, _ ("Paired Samples Correlations"));
1418 /* column headings */
1419 tab_text (table, 2,0, TAB_CENTER | TAT_TITLE, _ ("N"));
1420 tab_text (table, 3,0, TAB_CENTER | TAT_TITLE, _ ("Correlation"));
1421 tab_text (table, 4,0, TAB_CENTER | TAT_TITLE, _ ("Sig."));
1423 for (i=0; i < n_pairs; ++i)
1427 double df = pairs[i].n -2;
1429 double correlation_t =
1430 pairs[i].correlation * sqrt (df) /
1431 sqrt (1 - pow2 (pairs[i].correlation));
1435 tab_text (table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1438 tab_text (table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1440 var_get_name (pairs[i].v[0]),
1441 var_get_name (pairs[i].v[1]));
1445 tab_double (table, 2, i+1, TAB_RIGHT, pairs[i].n, wfmt);
1446 tab_double (table, 3, i+1, TAB_RIGHT, pairs[i].correlation, NULL);
1448 p = gsl_cdf_tdist_P (correlation_t, df);
1449 q = gsl_cdf_tdist_Q (correlation_t, df);
1451 tab_double (table, 4, i+1, TAB_RIGHT, 2.0* (correlation_t>0?q:p), NULL);
1460 /* Calculation Implementation */
1462 /* Per case calculations common to all variants of the T test */
1464 common_calc (const struct dictionary *dict,
1465 const struct ccase *c,
1467 enum mv_class exclude)
1470 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1472 double weight = dict_get_case_weight (dict, c, NULL);
1475 /* Listwise has to be implicit if the independent variable is missing ?? */
1476 if ( cmd->sbc_groups )
1478 if (var_is_value_missing (indep_var, case_data (c, indep_var), exclude))
1482 for (i = 0; i < cmd->n_variables ; ++i)
1484 const struct variable *v = cmd->v_variables[i];
1485 const union value *val = case_data (c, v);
1487 if (!var_is_value_missing (v, val, exclude))
1489 struct group_statistics *gs;
1490 gs = &group_proc_get (v)->ugs;
1493 gs->sum += weight * val->f;
1494 gs->ssq += weight * pow2 (val->f);
1500 /* Pre calculations common to all variants of the T test */
1502 common_precalc ( struct cmd_t_test *cmd )
1506 for (i=0; i< cmd->n_variables ; ++i)
1508 struct group_statistics *gs;
1509 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1518 /* Post calculations common to all variants of the T test */
1520 common_postcalc (struct cmd_t_test *cmd)
1524 for (i=0; i< cmd->n_variables ; ++i)
1526 struct group_statistics *gs;
1527 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1529 gs->mean=gs->sum / gs->n;
1530 gs->s_std_dev= sqrt (
1531 ( (gs->ssq / gs->n ) - pow2 (gs->mean))
1536 ( (gs->ssq / gs->n ) - pow2 (gs->mean))
1539 gs->se_mean = gs->std_dev / sqrt (gs->n);
1540 gs->mean_diff= gs->sum_diff / gs->n;
1544 /* Per case calculations for one sample t test */
1546 one_sample_calc (const struct dictionary *dict,
1547 const struct ccase *c, void *cmd_,
1548 enum mv_class exclude)
1552 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1554 double weight = dict_get_case_weight (dict, c, NULL);
1557 for (i=0; i< cmd->n_variables ; ++i)
1559 struct group_statistics *gs;
1560 const struct variable *v = cmd->v_variables[i];
1561 const union value *val = case_data (c, v);
1563 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1565 if (!var_is_value_missing (v, val, exclude))
1566 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1572 /* Pre calculations for one sample t test */
1574 one_sample_precalc ( struct cmd_t_test *cmd )
1578 for (i=0; i< cmd->n_variables ; ++i)
1580 struct group_statistics *gs;
1581 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1587 /* Post calculations for one sample t test */
1589 one_sample_postcalc (struct cmd_t_test *cmd)
1593 for (i=0; i< cmd->n_variables ; ++i)
1595 struct group_statistics *gs;
1596 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1598 gs->mean_diff = gs->sum_diff / gs->n ;
1605 paired_precalc (struct cmd_t_test *cmd UNUSED)
1609 for (i=0; i < n_pairs ; ++i )
1612 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1613 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1614 pairs[i].sum_of_prod = 0;
1615 pairs[i].correlation = 0;
1616 pairs[i].sum_of_diffs = 0;
1617 pairs[i].ssq_diffs = 0;
1624 paired_calc (const struct dictionary *dict, const struct ccase *c,
1625 struct cmd_t_test *cmd UNUSED, enum mv_class exclude)
1629 double weight = dict_get_case_weight (dict, c, NULL);
1631 for (i=0; i < n_pairs ; ++i )
1633 const struct variable *v0 = pairs[i].v[0];
1634 const struct variable *v1 = pairs[i].v[1];
1636 const union value *val0 = case_data (c, v0);
1637 const union value *val1 = case_data (c, v1);
1639 if (!var_is_value_missing (v0, val0, exclude) &&
1640 !var_is_value_missing (v1, val1, exclude))
1642 pairs[i].n += weight;
1643 pairs[i].sum[0] += weight * val0->f;
1644 pairs[i].sum[1] += weight * val1->f;
1646 pairs[i].ssq[0] += weight * pow2 (val0->f);
1647 pairs[i].ssq[1] += weight * pow2 (val1->f);
1649 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1651 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1652 pairs[i].ssq_diffs += weight * pow2 (val0->f - val1->f);
1660 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1664 for (i=0; i < n_pairs ; ++i )
1667 const double n = pairs[i].n;
1669 for (j=0; j < 2 ; ++j)
1671 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1672 pairs[i].s_std_dev[j] = sqrt ((pairs[i].ssq[j] / n -
1673 pow2 (pairs[i].mean[j]))
1676 pairs[i].std_dev[j] = sqrt (n/ (n-1)* (pairs[i].ssq[j] / n -
1677 pow2 (pairs[i].mean[j]))
1681 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1682 pairs[i].mean[0] * pairs[i].mean[1] ;
1683 /* correlation now actually contains the covariance */
1685 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1686 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1688 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1690 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1691 ( pairs[i].ssq_diffs / n )
1693 pow2 (pairs[i].mean_diff )
1699 group_precalc (struct cmd_t_test *cmd )
1704 for (i=0; i< cmd->n_variables ; ++i)
1706 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1708 /* There's always 2 groups for a T - TEST */
1711 gp.indep_var = indep_var;
1713 ttpr->group_hash = hsh_create (2,
1714 (hsh_compare_func *) compare_group_binary,
1715 (hsh_hash_func *) hash_group_binary,
1716 (hsh_free_func *) free_group,
1719 for (j=0 ; j < 2 ; ++j)
1721 struct group_statistics *gs = xmalloc (sizeof *gs);
1727 if ( gp.criterion == CMP_EQ )
1729 gs->id = gp.v.g_value[j];
1734 gs->id.f = gp.v.critical_value - 1.0 ;
1736 gs->id.f = gp.v.critical_value + 1.0 ;
1739 hsh_insert ( ttpr->group_hash, (void *) gs );
1746 group_calc (const struct dictionary *dict,
1747 const struct ccase *c, struct cmd_t_test *cmd,
1748 enum mv_class exclude)
1752 const double weight = dict_get_case_weight (dict, c, NULL);
1754 const union value *gv;
1756 if (var_is_value_missing (indep_var, case_data (c, indep_var), exclude))
1759 gv = case_data (c, indep_var);
1761 for (i=0; i< cmd->n_variables ; ++i)
1763 const struct variable *var = cmd->v_variables[i];
1764 const union value *val = case_data (c, var);
1765 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1766 struct group_statistics *gs;
1768 gs = hsh_find (grp_hash, (void *) gv);
1770 /* If the independent variable doesn't match either of the values
1771 for this case then move on to the next case */
1775 if (!var_is_value_missing (var, val, exclude))
1778 gs->sum += weight * val->f;
1779 gs->ssq += weight * pow2 (val->f);
1788 group_postcalc ( struct cmd_t_test *cmd )
1792 for (i = 0; i < cmd->n_variables ; ++i)
1794 const struct variable *var = cmd->v_variables[i];
1795 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1796 struct hsh_iterator g;
1797 struct group_statistics *gs;
1800 for (gs = hsh_first (grp_hash,&g);
1802 gs = hsh_next (grp_hash,&g))
1804 gs->mean = gs->sum / gs->n;
1806 gs->s_std_dev= sqrt (
1807 ( (gs->ssq / gs->n ) - pow2 (gs->mean))
1812 ( (gs->ssq / gs->n ) - pow2 (gs->mean))
1815 gs->se_mean = gs->std_dev / sqrt (gs->n);
1818 assert (count == 2);
1825 calculate (struct cmd_t_test *cmd,
1826 struct casereader *input, const struct dataset *ds)
1828 const struct dictionary *dict = dataset_dict (ds);
1829 struct ssbox stat_summary_box;
1830 struct trbox test_results_box;
1832 struct casereader *pass1, *pass2, *pass3;
1833 struct taint *taint;
1836 enum mv_class exclude = cmd->miss != TTS_INCLUDE ? MV_ANY : MV_SYSTEM;
1838 c = casereader_peek (input, 0);
1841 casereader_destroy (input);
1844 output_split_file_values (ds, c);
1847 if ( cmd->miss == TTS_LISTWISE )
1848 input = casereader_create_filter_missing (input,
1851 exclude, NULL, NULL);
1853 input = casereader_create_filter_weight (input, dict, NULL, NULL);
1855 taint = taint_clone (casereader_get_taint (input));
1856 casereader_split (input, &pass1, &pass2);
1858 common_precalc (cmd);
1859 for (; (c = casereader_read (pass1)) != NULL; case_unref (c))
1860 common_calc (dict, c, cmd, exclude);
1861 casereader_destroy (pass1);
1862 common_postcalc (cmd);
1867 one_sample_precalc (cmd);
1868 for (; (c = casereader_read (pass2)) != NULL; case_unref (c))
1869 one_sample_calc (dict, c, cmd, exclude);
1870 one_sample_postcalc (cmd);
1873 paired_precalc (cmd);
1874 for (; (c = casereader_read (pass2)) != NULL; case_unref (c))
1875 paired_calc (dict, c, cmd, exclude);
1876 paired_postcalc (cmd);
1879 pass3 = casereader_clone (pass2);
1881 group_precalc (cmd);
1882 for (; (c = casereader_read (pass2)) != NULL; case_unref (c))
1883 group_calc (dict, c, cmd, exclude);
1884 group_postcalc (cmd);
1886 levene (dict, pass3, indep_var, cmd->n_variables, cmd->v_variables,
1890 casereader_destroy (pass2);
1892 if (!taint_has_tainted_successor (taint))
1894 ssbox_create (&stat_summary_box,cmd,mode);
1895 ssbox_populate (&stat_summary_box, dict, cmd);
1896 ssbox_finalize (&stat_summary_box);
1898 if ( mode == T_PAIRED )
1901 trbox_create (&test_results_box, cmd, mode);
1902 trbox_populate (&test_results_box, dict, cmd);
1903 trbox_finalize (&test_results_box);
1906 taint_destroy (taint);
1909 short which_group (const struct group_statistics *g,
1910 const struct group_properties *p);
1912 /* Return -1 if the id of a is less than b; +1 if greater than and
1915 compare_group_binary (const struct group_statistics *a,
1916 const struct group_statistics *b,
1917 const struct group_properties *p)
1922 if ( p->criterion == CMP_LE )
1924 flag_a = ( a->id.f < p->v.critical_value ) ;
1925 flag_b = ( b->id.f < p->v.critical_value ) ;
1929 flag_a = which_group (a, p);
1930 flag_b = which_group (b, p);
1933 if (flag_a < flag_b )
1936 return (flag_a > flag_b);
1939 /* This is a degenerate case of a hash, since it can only return three possible
1940 values. It's really a comparison, being used as a hash function */
1943 hash_group_binary (const struct group_statistics *g,
1944 const struct group_properties *p)
1948 if ( p->criterion == CMP_LE )
1950 flag = ( g->id.f < p->v.critical_value ) ;
1952 else if ( p->criterion == CMP_EQ)
1954 flag = which_group (g,p);
1962 /* return 0 if G belongs to group 0,
1963 1 if it belongs to group 1,
1964 2 if it belongs to neither group */
1966 which_group (const struct group_statistics *g,
1967 const struct group_properties *p)
1969 if ( 0 == compare_values_short (&g->id, &p->v.g_value[0], p->indep_var))
1972 if ( 0 == compare_values_short (&g->id, &p->v.g_value[1], p->indep_var))