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 <language/command.h>
36 #include <language/dictionary/split-file.h>
37 #include <language/lexer/lexer.h>
38 #include <libpspp/alloc.h>
39 #include <libpspp/assertion.h>
40 #include <libpspp/compiler.h>
41 #include <libpspp/hash.h>
42 #include <libpspp/magic.h>
43 #include <libpspp/message.h>
44 #include <libpspp/message.h>
45 #include <libpspp/misc.h>
46 #include <libpspp/str.h>
47 #include <math/group-proc.h>
48 #include <math/levene.h>
49 #include <output/manager.h>
50 #include <output/table.h>
55 #define _(msgid) gettext (msgid)
63 +variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
65 missing=miss:!analysis/listwise,
66 incl:include/!exclude;
67 +format=fmt:!labels/nolabels;
68 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
76 /* Function to use for testing for missing values */
77 static is_missing_func *value_is_missing;
79 /* Variable for the GROUPS subcommand, if given. */
80 static struct variable *indep_var;
88 struct group_properties
90 /* The comparison criterion */
91 enum comparison criterion;
93 /* The width of the independent variable */
97 /* The value of the independent variable at which groups are determined to
98 belong to one group or the other */
99 double critical_value;
102 /* The values of the independent variable for each group */
103 union value g_value[2];
109 static struct group_properties gp ;
113 /* PAIRS: Number of pairs to be compared ; each pair. */
114 static int n_pairs = 0 ;
117 /* The variables comprising the pair */
118 struct variable *v[2];
120 /* The number of valid variable pairs */
123 /* The sum of the members */
126 /* sum of squares of the members */
129 /* Std deviation of the members */
133 /* Sample Std deviation of the members */
136 /* The means of the members */
139 /* The correlation coefficient between the variables */
142 /* The sum of the differences */
145 /* The sum of the products */
148 /* The mean of the differences */
151 /* The sum of the squares of the differences */
154 /* The std deviation of the differences */
158 static struct pair *pairs=0;
160 static int parse_value (union value * v, int type) ;
162 /* Structures and Functions for the Statistics Summary Box */
164 typedef void populate_ssbox_func(struct ssbox *ssb,
165 struct cmd_t_test *cmd);
166 typedef void finalize_ssbox_func(struct ssbox *ssb);
172 populate_ssbox_func *populate;
173 finalize_ssbox_func *finalize;
178 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
180 /* Populate a ssbox according to cmd */
181 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
183 /* Submit and destroy a ssbox */
184 void ssbox_finalize(struct ssbox *ssb);
186 /* A function to create, populate and submit the Paired Samples Correlation
191 /* Structures and Functions for the Test Results Box */
194 typedef void populate_trbox_func(struct trbox *trb,
195 struct cmd_t_test *cmd);
196 typedef void finalize_trbox_func(struct trbox *trb);
200 populate_trbox_func *populate;
201 finalize_trbox_func *finalize;
205 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
207 /* Populate a ssbox according to cmd */
208 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
210 /* Submit and destroy a ssbox */
211 void trbox_finalize(struct trbox *trb);
213 /* Which mode was T-TEST invoked */
221 static int common_calc (const struct dictionary *dict,
222 const struct ccase *, void *);
223 static void common_precalc (struct cmd_t_test *);
224 static void common_postcalc (struct cmd_t_test *);
226 static int one_sample_calc (const struct dictionary *dict, const struct ccase *, void *);
227 static void one_sample_precalc (struct cmd_t_test *);
228 static void one_sample_postcalc (struct cmd_t_test *);
230 static int paired_calc (const struct dictionary *dict, const struct ccase *, void *);
231 static void paired_precalc (struct cmd_t_test *);
232 static void paired_postcalc (struct cmd_t_test *);
234 static void group_precalc (struct cmd_t_test *);
235 static int group_calc (const struct dictionary *dict, const struct ccase *, struct cmd_t_test *);
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 dataset *ds)
265 if ( !parse_t_test (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."));
342 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
343 if (cmd.incl == TTS_INCLUDE )
344 value_is_missing = mv_is_value_system_missing;
346 value_is_missing = mv_is_value_missing;
348 bad_weight_warn = true;
350 ok = multipass_procedure_with_splits (ds, calculate, &cmd);
356 if ( mode == T_IND_SAMPLES)
359 /* Destroy any group statistics we created */
360 for (v = 0 ; v < cmd.n_variables ; ++v )
362 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
363 hsh_destroy (grpp->group_hash);
368 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
372 tts_custom_groups (struct dataset *ds, struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
374 int n_group_values=0;
378 indep_var = parse_variable (dataset_dict (ds));
381 lex_error ("expecting variable name in GROUPS subcommand");
385 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
387 msg (SE, _("Long string variable %s is not valid here."),
392 if (!lex_match ('('))
394 if (indep_var->type == NUMERIC)
396 gp.v.g_value[0].f = 1;
397 gp.v.g_value[1].f = 2;
399 gp.criterion = CMP_EQ;
407 msg (SE, _("When applying GROUPS to a string variable, two "
408 "values must be specified."));
413 if (!parse_value (&gp.v.g_value[0], indep_var->type))
419 if (indep_var->type != NUMERIC)
422 msg (SE, _("When applying GROUPS to a string variable, two "
423 "values must be specified."));
426 gp.criterion = CMP_LE;
427 gp.v.critical_value = gp.v.g_value[0].f;
433 if (!parse_value (&gp.v.g_value[1], indep_var->type))
437 if (!lex_force_match (')'))
440 if ( n_group_values == 2 )
441 gp.criterion = CMP_EQ ;
443 gp.criterion = CMP_LE ;
451 tts_custom_pairs (struct dataset *ds, struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
453 struct variable **vars;
455 size_t n_pairs_local;
457 size_t n_before_WITH;
458 size_t n_after_WITH = SIZE_MAX;
459 int paired ; /* Was the PAIRED keyword given ? */
464 if (!parse_variables (dataset_dict (ds), &vars, &n_vars,
465 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
473 if (lex_match (T_WITH))
475 n_before_WITH = n_vars;
476 if (!parse_variables (dataset_dict (ds), &vars, &n_vars,
477 PV_DUPLICATE | PV_APPEND
478 | PV_NUMERIC | PV_NO_SCRATCH))
483 n_after_WITH = n_vars - n_before_WITH;
486 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
488 /* Determine the number of pairs needed */
491 if (n_before_WITH != n_after_WITH)
494 msg (SE, _("PAIRED was specified but the number of variables "
495 "preceding WITH (%d) did not match the number "
497 n_before_WITH, n_after_WITH );
500 n_pairs_local = n_before_WITH;
502 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
504 n_pairs_local = n_before_WITH * n_after_WITH ;
506 else /* Neither WITH nor PAIRED keyword given */
511 msg (SE, _("At least two variables must be specified "
516 /* how many ways can you pick 2 from n_vars ? */
517 n_pairs_local = n_vars * (n_vars - 1) / 2;
521 /* Allocate storage for the pairs */
522 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
524 /* Populate the pairs with the appropriate variables */
529 assert(n_pairs_local == n_vars / 2);
530 for (i = 0; i < n_pairs_local; ++i)
532 pairs[i].v[n_pairs] = vars[i];
533 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
536 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
541 for(i=0 ; i < n_before_WITH ; ++i )
543 for(j=0 ; j < n_after_WITH ; ++j)
545 pairs[p].v[0] = vars[i];
546 pairs[p].v[1] = vars[j+n_before_WITH];
551 else /* Neither WITH nor PAIRED given */
556 for(i=0 ; i < n_vars ; ++i )
558 for(j=i+1 ; j < n_vars ; ++j)
560 pairs[p].v[0] = vars[i];
561 pairs[p].v[1] = vars[j];
567 n_pairs+=n_pairs_local;
573 /* Parses the current token (numeric or string, depending on type)
574 value v and returns success. */
576 parse_value (union value * v, int type )
580 if (!lex_force_num ())
586 if (!lex_force_string ())
588 strncpy (v->s, ds_cstr (&tokstr), ds_length (&tokstr));
597 /* Implementation of the SSBOX object */
599 void ssbox_base_init(struct ssbox *this, int cols,int rows);
601 void ssbox_base_finalize(struct ssbox *ssb);
603 void ssbox_one_sample_init(struct ssbox *this,
604 struct cmd_t_test *cmd );
606 void ssbox_independent_samples_init(struct ssbox *this,
607 struct cmd_t_test *cmd);
609 void ssbox_paired_init(struct ssbox *this,
610 struct cmd_t_test *cmd);
613 /* Factory to create an ssbox */
615 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
620 ssbox_one_sample_init(ssb,cmd);
623 ssbox_independent_samples_init(ssb,cmd);
626 ssbox_paired_init(ssb,cmd);
635 /* Despatcher for the populate method */
637 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
639 ssb->populate(ssb,cmd);
643 /* Despatcher for finalize */
645 ssbox_finalize(struct ssbox *ssb)
651 /* Submit the box and clear up */
653 ssbox_base_finalize(struct ssbox *ssb)
660 /* Initialize a ssbox struct */
662 ssbox_base_init(struct ssbox *this, int cols,int rows)
664 this->finalize = ssbox_base_finalize;
665 this->t = tab_create (cols, rows, 0);
667 tab_columns (this->t, SOM_COL_DOWN, 1);
668 tab_headers (this->t,0,0,1,0);
669 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
670 tab_hline(this->t, TAL_2,0,cols-1,1);
671 tab_dim (this->t, tab_natural_dimensions);
674 void ssbox_one_sample_populate(struct ssbox *ssb,
675 struct cmd_t_test *cmd);
677 /* Initialize the one_sample ssbox */
679 ssbox_one_sample_init(struct ssbox *this,
680 struct cmd_t_test *cmd )
683 const int vsize=cmd->n_variables+1;
685 this->populate = ssbox_one_sample_populate;
687 ssbox_base_init(this, hsize,vsize);
688 tab_title (this->t, _("One-Sample Statistics"));
689 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
690 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
691 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
692 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
693 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
696 void ssbox_independent_samples_populate(struct ssbox *ssb,
697 struct cmd_t_test *cmd);
699 /* Initialize the independent samples ssbox */
701 ssbox_independent_samples_init(struct ssbox *this,
702 struct cmd_t_test *cmd)
705 int vsize = cmd->n_variables*2 +1;
707 this->populate = ssbox_independent_samples_populate;
709 ssbox_base_init(this, hsize,vsize);
710 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
711 tab_title (this->t, _("Group Statistics"));
712 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
713 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
714 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
715 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
716 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
720 /* Populate the ssbox for independent samples */
722 ssbox_independent_samples_populate(struct ssbox *ssb,
723 struct cmd_t_test *cmd)
729 double indep_value[2];
731 char prefix[2][3]={"",""};
733 if ( indep_var->type == NUMERIC )
735 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
736 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
740 val_lab0 = gp.v.g_value[0].s;
741 val_lab1 = gp.v.g_value[1].s;
744 if (gp.criterion == CMP_LE )
746 strcpy(prefix[0],"< ");
747 strcpy(prefix[1],">=");
748 indep_value[0] = gp.v.critical_value;
749 indep_value[1] = gp.v.critical_value;
753 indep_value[0] = gp.v.g_value[0].f;
754 indep_value[1] = gp.v.g_value[1].f;
759 for (i=0; i < cmd->n_variables; ++i)
761 struct variable *var = cmd->v_variables[i];
762 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
765 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
768 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
769 "%s%s", prefix[0], val_lab0);
771 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
772 "%s%g", prefix[0], indep_value[0]);
776 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
777 "%s%s", prefix[1], val_lab1);
779 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
780 "%s%g", prefix[1], indep_value[1]);
783 /* Fill in the group statistics */
784 for ( count = 0 ; count < 2 ; ++count )
786 union value search_val;
788 struct group_statistics *gs;
790 if ( gp.criterion == CMP_LE )
794 /* less than ( < ) case */
795 search_val.f = gp.v.critical_value - 1.0;
800 search_val.f = gp.v.critical_value + 1.0;
805 search_val = gp.v.g_value[count];
808 gs = hsh_find(grp_hash, (void *) &search_val);
811 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
812 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
813 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
814 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
820 void ssbox_paired_populate(struct ssbox *ssb,
821 struct cmd_t_test *cmd);
823 /* Initialize the paired values ssbox */
825 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
829 int vsize = n_pairs*2+1;
831 this->populate = ssbox_paired_populate;
833 ssbox_base_init(this, hsize,vsize);
834 tab_title (this->t, _("Paired Sample Statistics"));
835 tab_vline(this->t,TAL_GAP,1,0,vsize-1);
836 tab_vline(this->t,TAL_2,2,0,vsize-1);
837 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
838 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
839 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
840 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
844 /* Populate the ssbox for paired values */
846 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
852 for (i=0; i < n_pairs; ++i)
856 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
858 for (j=0 ; j < 2 ; ++j)
860 struct group_statistics *gs;
862 gs = &group_proc_get (pairs[i].v[j])->ugs;
866 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
869 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
870 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
871 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
872 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
878 /* Populate the one sample ssbox */
880 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
886 for (i=0; i < cmd->n_variables; ++i)
888 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
890 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
891 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
892 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
893 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
894 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
901 /* Implementation of the Test Results box struct */
903 void trbox_base_init(struct trbox *self,size_t n_vars, int cols);
904 void trbox_base_finalize(struct trbox *trb);
906 void trbox_independent_samples_init(struct trbox *trb,
907 struct cmd_t_test *cmd );
909 void trbox_independent_samples_populate(struct trbox *trb,
910 struct cmd_t_test *cmd);
912 void trbox_one_sample_init(struct trbox *self,
913 struct cmd_t_test *cmd );
915 void trbox_one_sample_populate(struct trbox *trb,
916 struct cmd_t_test *cmd);
918 void trbox_paired_init(struct trbox *self,
919 struct cmd_t_test *cmd );
921 void trbox_paired_populate(struct trbox *trb,
922 struct cmd_t_test *cmd);
926 /* Create a trbox according to mode*/
928 trbox_create(struct trbox *trb,
929 struct cmd_t_test *cmd, int mode)
934 trbox_one_sample_init(trb,cmd);
937 trbox_independent_samples_init(trb,cmd);
940 trbox_paired_init(trb,cmd);
947 /* Populate a trbox according to cmd */
949 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
951 trb->populate(trb,cmd);
954 /* Submit and destroy a trbox */
956 trbox_finalize(struct trbox *trb)
961 /* Initialize the independent samples trbox */
963 trbox_independent_samples_init(struct trbox *self,
964 struct cmd_t_test *cmd UNUSED)
967 const int vsize=cmd->n_variables*2+3;
970 self->populate = trbox_independent_samples_populate;
972 trbox_base_init(self,cmd->n_variables*2,hsize);
973 tab_title(self->t,_("Independent Samples Test"));
974 tab_hline(self->t,TAL_1,2,hsize-1,1);
975 tab_vline(self->t,TAL_2,2,0,vsize-1);
976 tab_vline(self->t,TAL_1,4,0,vsize-1);
977 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
978 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
979 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
980 tab_joint_text(self->t, 2, 0, 3, 0,
981 TAB_CENTER,_("Levene's Test for Equality of Variances"));
982 tab_joint_text(self->t, 4,0,hsize-1,0,
983 TAB_CENTER,_("t-test for Equality of Means"));
985 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
986 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
987 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
988 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
989 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
990 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
991 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
992 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
993 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
995 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
996 _("%g%% Confidence Interval of the Difference"),
997 cmd->criteria*100.0);
1001 /* Populate the independent samples trbox */
1003 trbox_independent_samples_populate(struct trbox *self,
1004 struct cmd_t_test *cmd )
1009 for (i=0; i < cmd->n_variables; ++i)
1018 double pooled_variance;
1019 double std_err_diff;
1022 struct variable *var = cmd->v_variables[i];
1023 struct group_proc *grp_data = group_proc_get (var);
1025 struct hsh_table *grp_hash = grp_data->group_hash;
1027 struct group_statistics *gs0 ;
1028 struct group_statistics *gs1 ;
1030 union value search_val;
1032 if ( gp.criterion == CMP_LE )
1033 search_val.f = gp.v.critical_value - 1.0;
1035 search_val = gp.v.g_value[0];
1037 gs0 = hsh_find(grp_hash, (void *) &search_val);
1040 if ( gp.criterion == CMP_LE )
1041 search_val.f = gp.v.critical_value + 1.0;
1043 search_val = gp.v.g_value[1];
1045 gs1 = hsh_find(grp_hash, (void *) &search_val);
1049 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1051 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1054 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1056 /* Now work out the significance of the Levene test */
1057 df1 = 1; df2 = grp_data->ugs.n - 2;
1058 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1060 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1062 df = gs0->n + gs1->n - 2.0 ;
1063 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1065 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1067 (gs1->n )*pow2(gs1->s_std_dev)
1070 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1071 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1073 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1075 p = gsl_cdf_tdist_P(t, df);
1076 q = gsl_cdf_tdist_Q(t, df);
1078 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1080 mean_diff = gs0->mean - gs1->mean;
1081 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1084 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1085 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1088 /* Now work out the confidence interval */
1089 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1091 t = gsl_cdf_tdist_Qinv(q,df);
1092 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1093 mean_diff - t * std_err_diff, 8, 3);
1095 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1096 mean_diff + t * std_err_diff, 8, 3);
1101 /* Now for the \sigma_1 != \sigma_2 case */
1102 tab_text (self->t, 1, i*2+3+1,
1103 TAB_LEFT, _("Equal variances not assumed"));
1106 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1107 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1109 t = mean_diff / sqrt(se2) ;
1110 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1113 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1117 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1121 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1123 p = gsl_cdf_tdist_P(t, df);
1124 q = gsl_cdf_tdist_Q(t, df);
1126 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1128 /* Now work out the confidence interval */
1129 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1131 t = gsl_cdf_tdist_Qinv(q, df);
1133 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1136 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1139 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1140 mean_diff - t * std_err_diff, 8, 3);
1142 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1143 mean_diff + t * std_err_diff, 8, 3);
1149 /* Initialize the paired samples trbox */
1151 trbox_paired_init(struct trbox *self,
1152 struct cmd_t_test *cmd UNUSED)
1156 const int vsize=n_pairs+3;
1158 self->populate = trbox_paired_populate;
1160 trbox_base_init(self,n_pairs,hsize);
1161 tab_title (self->t, _("Paired Samples Test"));
1162 tab_hline(self->t,TAL_1,2,6,1);
1163 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1164 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1165 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1166 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1167 tab_hline(self->t,TAL_1,5,6, 2);
1168 tab_vline(self->t,TAL_GAP,6,0,1);
1170 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1171 _("%g%% Confidence Interval of the Difference"),
1172 cmd->criteria*100.0);
1174 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1175 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1176 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1177 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1178 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1179 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1180 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1181 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1184 /* Populate the paired samples trbox */
1186 trbox_paired_populate(struct trbox *trb,
1187 struct cmd_t_test *cmd UNUSED)
1191 for (i=0; i < n_pairs; ++i)
1196 double n = pairs[i].n;
1200 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1202 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1203 pairs[i].v[0]->name, pairs[i].v[1]->name);
1205 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1207 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1210 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1211 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1213 /* Now work out the confidence interval */
1214 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1216 t = gsl_cdf_tdist_Qinv(q, df);
1218 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1219 pairs[i].mean_diff - t * se_mean , 8, 4);
1221 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1222 pairs[i].mean_diff + t * se_mean , 8, 4);
1224 t = (pairs[i].mean[0] - pairs[i].mean[1])
1226 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1227 2 * pairs[i].correlation *
1228 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1232 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1234 /* Degrees of freedom */
1235 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1237 p = gsl_cdf_tdist_P(t,df);
1238 q = gsl_cdf_tdist_P(t,df);
1240 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1245 /* Initialize the one sample trbox */
1247 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1250 const int vsize=cmd->n_variables+3;
1252 self->populate = trbox_one_sample_populate;
1254 trbox_base_init(self, cmd->n_variables,hsize);
1255 tab_title (self->t, _("One-Sample Test"));
1256 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1257 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1259 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1260 _("Test Value = %f"), cmd->n_testval[0]);
1262 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1265 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1266 _("%g%% Confidence Interval of the Difference"),
1267 cmd->criteria*100.0);
1269 tab_vline(self->t,TAL_GAP,6,1,1);
1270 tab_hline(self->t,TAL_1,5,6,2);
1271 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1272 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1273 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1274 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1275 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1276 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1281 /* Populate the one sample trbox */
1283 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1289 for (i=0; i < cmd->n_variables; ++i)
1294 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1297 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1299 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1301 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1303 /* degrees of freedom */
1306 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1308 p = gsl_cdf_tdist_P(t, df);
1309 q = gsl_cdf_tdist_Q(t, df);
1311 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1313 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1315 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1318 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1319 t = gsl_cdf_tdist_Qinv(q, df);
1321 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1322 gs->mean_diff - t * gs->se_mean, 8,4);
1324 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1325 gs->mean_diff + t * gs->se_mean, 8,4);
1329 /* Base initializer for the generalized trbox */
1331 trbox_base_init(struct trbox *self, size_t data_rows, int cols)
1333 const size_t rows = 3 + data_rows;
1335 self->finalize = trbox_base_finalize;
1336 self->t = tab_create (cols, rows, 0);
1337 tab_headers (self->t,0,0,3,0);
1338 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1339 tab_hline(self->t, TAL_2,0,cols-1,3);
1340 tab_dim (self->t, tab_natural_dimensions);
1344 /* Base finalizer for the trbox */
1346 trbox_base_finalize(struct trbox *trb)
1352 /* Create , populate and submit the Paired Samples Correlation box */
1356 const int rows=1+n_pairs;
1360 struct tab_table *table;
1362 table = tab_create (cols,rows,0);
1364 tab_columns (table, SOM_COL_DOWN, 1);
1365 tab_headers (table,0,0,1,0);
1366 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1367 tab_hline(table, TAL_2, 0, cols - 1, 1);
1368 tab_vline(table, TAL_2, 2, 0, rows - 1);
1369 tab_dim(table, tab_natural_dimensions);
1370 tab_title(table, _("Paired Samples Correlations"));
1372 /* column headings */
1373 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1374 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1375 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1377 for (i=0; i < n_pairs; ++i)
1381 double df = pairs[i].n -2;
1383 double correlation_t =
1384 pairs[i].correlation * sqrt(df) /
1385 sqrt(1 - pow2(pairs[i].correlation));
1389 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1392 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1393 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1397 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1398 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1400 p = gsl_cdf_tdist_P(correlation_t, df);
1401 q = gsl_cdf_tdist_Q(correlation_t, df);
1403 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1412 /* Calculation Implementation */
1414 /* Per case calculations common to all variants of the T test */
1416 common_calc (const struct dictionary *dict, const struct ccase *c, void *_cmd)
1419 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1421 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1424 /* Skip the entire case if /MISSING=LISTWISE is set */
1425 if ( cmd->miss == TTS_LISTWISE )
1427 for(i=0; i< cmd->n_variables ; ++i)
1429 struct variable *v = cmd->v_variables[i];
1430 const union value *val = case_data (c, v->fv);
1432 if (value_is_missing(&v->miss, val) )
1439 /* Listwise has to be implicit if the independent variable is missing ?? */
1440 if ( cmd->sbc_groups )
1442 const union value *gv = case_data (c, indep_var->fv);
1443 if ( value_is_missing(&indep_var->miss, gv) )
1450 for(i=0; i< cmd->n_variables ; ++i)
1452 struct group_statistics *gs;
1453 struct variable *v = cmd->v_variables[i];
1454 const union value *val = case_data (c, v->fv);
1456 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1458 if (! value_is_missing(&v->miss, val) )
1461 gs->sum+=weight * val->f;
1462 gs->ssq+=weight * val->f * val->f;
1468 /* Pre calculations common to all variants of the T test */
1470 common_precalc ( struct cmd_t_test *cmd )
1474 for(i=0; i< cmd->n_variables ; ++i)
1476 struct group_statistics *gs;
1477 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1486 /* Post calculations common to all variants of the T test */
1488 common_postcalc ( struct cmd_t_test *cmd )
1493 for(i=0; i< cmd->n_variables ; ++i)
1495 struct group_statistics *gs;
1496 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1498 gs->mean=gs->sum / gs->n;
1499 gs->s_std_dev= sqrt(
1500 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1505 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1508 gs->se_mean = gs->std_dev / sqrt(gs->n);
1509 gs->mean_diff= gs->sum_diff / gs->n;
1513 /* Per case calculations for one sample t test */
1515 one_sample_calc (const struct dictionary *dict,
1516 const struct ccase *c, void *cmd_)
1519 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1522 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1524 /* Skip the entire case if /MISSING=LISTWISE is set */
1525 if ( cmd->miss == TTS_LISTWISE )
1527 for(i=0; i< cmd->n_variables ; ++i)
1529 struct variable *v = cmd->v_variables[i];
1530 const union value *val = case_data (c, v->fv);
1532 if (value_is_missing(&v->miss, val) )
1539 for(i=0; i< cmd->n_variables ; ++i)
1541 struct group_statistics *gs;
1542 struct variable *v = cmd->v_variables[i];
1543 const union value *val = case_data (c, v->fv);
1545 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1547 if ( ! value_is_missing(&v->miss, val))
1548 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1554 /* Pre calculations for one sample t test */
1556 one_sample_precalc ( struct cmd_t_test *cmd )
1560 for(i=0; i< cmd->n_variables ; ++i)
1562 struct group_statistics *gs;
1563 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1569 /* Post calculations for one sample t test */
1571 one_sample_postcalc (struct cmd_t_test *cmd)
1575 for(i=0; i< cmd->n_variables ; ++i)
1577 struct group_statistics *gs;
1578 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1580 gs->mean_diff = gs->sum_diff / gs->n ;
1587 paired_precalc (struct cmd_t_test *cmd UNUSED)
1591 for(i=0; i < n_pairs ; ++i )
1594 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1595 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1596 pairs[i].sum_of_prod = 0;
1597 pairs[i].correlation = 0;
1598 pairs[i].sum_of_diffs = 0;
1599 pairs[i].ssq_diffs = 0;
1606 paired_calc (const struct dictionary *dict, const struct ccase *c, void *cmd_)
1610 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1612 double weight = dict_get_case_weight (dict, c, &bad_weight_warn);
1614 /* Skip the entire case if /MISSING=LISTWISE is set ,
1615 AND one member of a pair is missing */
1616 if ( cmd->miss == TTS_LISTWISE )
1618 for(i=0; i < n_pairs ; ++i )
1620 struct variable *v0 = pairs[i].v[0];
1621 struct variable *v1 = pairs[i].v[1];
1623 const union value *val0 = case_data (c, v0->fv);
1624 const union value *val1 = case_data (c, v1->fv);
1626 if ( value_is_missing(&v0->miss, val0) ||
1627 value_is_missing(&v1->miss, val1) )
1634 for(i=0; i < n_pairs ; ++i )
1636 struct variable *v0 = pairs[i].v[0];
1637 struct variable *v1 = pairs[i].v[1];
1639 const union value *val0 = case_data (c, v0->fv);
1640 const union value *val1 = case_data (c, v1->fv);
1642 if ( ( !value_is_missing(&v0->miss, val0)
1643 && !value_is_missing(&v1->miss, val1) ) )
1645 pairs[i].n += weight;
1646 pairs[i].sum[0] += weight * val0->f;
1647 pairs[i].sum[1] += weight * val1->f;
1649 pairs[i].ssq[0] += weight * pow2(val0->f);
1650 pairs[i].ssq[1] += weight * pow2(val1->f);
1652 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1654 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1655 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1663 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1667 for(i=0; i < n_pairs ; ++i )
1670 const double n = pairs[i].n;
1672 for (j=0; j < 2 ; ++j)
1674 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1675 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1676 pow2(pairs[i].mean[j]))
1679 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1680 pow2(pairs[i].mean[j]))
1684 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1685 pairs[i].mean[0] * pairs[i].mean[1] ;
1686 /* correlation now actually contains the covariance */
1688 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1689 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1691 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1693 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1694 ( pairs[i].ssq_diffs / n )
1696 pow2(pairs[i].mean_diff )
1702 group_precalc (struct cmd_t_test *cmd )
1707 for(i=0; i< cmd->n_variables ; ++i)
1709 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1711 /* There's always 2 groups for a T - TEST */
1714 gp.indep_width = indep_var->width;
1716 ttpr->group_hash = hsh_create(2,
1717 (hsh_compare_func *) compare_group_binary,
1718 (hsh_hash_func *) hash_group_binary,
1719 (hsh_free_func *) free_group,
1722 for (j=0 ; j < 2 ; ++j)
1725 struct group_statistics *gs = xmalloc (sizeof *gs);
1731 if ( gp.criterion == CMP_EQ )
1733 gs->id = gp.v.g_value[j];
1738 gs->id.f = gp.v.critical_value - 1.0 ;
1740 gs->id.f = gp.v.critical_value + 1.0 ;
1743 hsh_insert ( ttpr->group_hash, (void *) gs );
1751 group_calc (const struct dictionary *dict,
1752 const struct ccase *c, struct cmd_t_test *cmd)
1756 const union value *gv = case_data (c, indep_var->fv);
1758 const double weight =
1759 dict_get_case_weight (dict, c, &bad_weight_warn);
1761 if ( value_is_missing(&indep_var->miss, gv) )
1766 if ( cmd->miss == TTS_LISTWISE )
1768 for(i=0; i< cmd->n_variables ; ++i)
1770 struct variable *v = cmd->v_variables[i];
1771 const union value *val = case_data (c, v->fv);
1773 if (value_is_missing(&v->miss, val) )
1780 gv = case_data (c, indep_var->fv);
1782 for(i=0; i< cmd->n_variables ; ++i)
1784 struct variable *var = cmd->v_variables[i];
1785 const union value *val = case_data (c, var->fv);
1786 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1787 struct group_statistics *gs;
1789 gs = hsh_find(grp_hash, (void *) gv);
1791 /* If the independent variable doesn't match either of the values
1792 for this case then move on to the next case */
1796 if ( !value_is_missing(&var->miss, val) )
1799 gs->sum+=weight * val->f;
1800 gs->ssq+=weight * pow2(val->f);
1809 group_postcalc ( struct cmd_t_test *cmd )
1813 for(i=0; i< cmd->n_variables ; ++i)
1815 struct variable *var = cmd->v_variables[i];
1816 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1817 struct hsh_iterator g;
1818 struct group_statistics *gs;
1821 for (gs = hsh_first (grp_hash,&g);
1823 gs = hsh_next(grp_hash,&g))
1825 gs->mean = gs->sum / gs->n;
1827 gs->s_std_dev= sqrt(
1828 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1833 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1836 gs->se_mean = gs->std_dev / sqrt(gs->n);
1846 calculate(const struct ccase *first, const struct casefile *cf,
1847 void *cmd_, const struct dataset *ds)
1849 const struct dictionary *dict = dataset_dict (ds);
1850 struct ssbox stat_summary_box;
1851 struct trbox test_results_box;
1853 struct casereader *r;
1856 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1858 output_split_file_values (ds, first);
1859 common_precalc(cmd);
1860 for(r = casefile_get_reader (cf);
1861 casereader_read (r, &c) ;
1864 common_calc(dict, &c,cmd);
1866 casereader_destroy (r);
1867 common_postcalc(cmd);
1872 one_sample_precalc(cmd);
1873 for(r = casefile_get_reader (cf);
1874 casereader_read (r, &c) ;
1877 one_sample_calc (dict, &c,cmd);
1879 casereader_destroy (r);
1880 one_sample_postcalc(cmd);
1884 paired_precalc(cmd);
1885 for(r = casefile_get_reader (cf);
1886 casereader_read (r, &c) ;
1889 paired_calc (dict, &c,cmd);
1891 casereader_destroy (r);
1892 paired_postcalc (cmd);
1898 for(r = casefile_get_reader (cf);
1899 casereader_read (r, &c) ;
1902 group_calc (dict, &c, cmd);
1904 casereader_destroy (r);
1905 group_postcalc(cmd);
1907 levene (dict, cf, indep_var, cmd->n_variables, cmd->v_variables,
1908 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1913 ssbox_create(&stat_summary_box,cmd,mode);
1914 ssbox_populate(&stat_summary_box,cmd);
1915 ssbox_finalize(&stat_summary_box);
1917 if ( mode == T_PAIRED)
1920 trbox_create(&test_results_box,cmd,mode);
1921 trbox_populate(&test_results_box,cmd);
1922 trbox_finalize(&test_results_box);
1927 short which_group(const struct group_statistics *g,
1928 const struct group_properties *p);
1930 /* Return -1 if the id of a is less than b; +1 if greater than and
1933 compare_group_binary(const struct group_statistics *a,
1934 const struct group_statistics *b,
1935 const struct group_properties *p)
1940 if ( p->criterion == CMP_LE )
1942 /* less-than-or-equal comparision is not meaningfull for
1943 alpha variables, so we shouldn't ever arrive here */
1944 assert(p->indep_width == 0 ) ;
1946 flag_a = ( a->id.f < p->v.critical_value ) ;
1947 flag_b = ( b->id.f < p->v.critical_value ) ;
1951 flag_a = which_group(a, p);
1952 flag_b = which_group(b, p);
1955 if (flag_a < flag_b )
1958 return (flag_a > flag_b);
1961 /* This is a degenerate case of a hash, since it can only return three possible
1962 values. It's really a comparison, being used as a hash function */
1965 hash_group_binary(const struct group_statistics *g,
1966 const struct group_properties *p)
1970 if ( p->criterion == CMP_LE )
1972 /* Not meaningfull to do a less than compare for alpha values ? */
1973 assert(p->indep_width == 0 ) ;
1974 flag = ( g->id.f < p->v.critical_value ) ;
1976 else if ( p->criterion == CMP_EQ)
1978 flag = which_group(g,p);
1986 /* return 0 if G belongs to group 0,
1987 1 if it belongs to group 1,
1988 2 if it belongs to neither group */
1990 which_group(const struct group_statistics *g,
1991 const struct group_properties *p)
1994 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1997 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))