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/lexer/lexer.h>
37 #include <libpspp/alloc.h>
38 #include <libpspp/compiler.h>
39 #include <libpspp/hash.h>
40 #include <libpspp/magic.h>
41 #include <libpspp/message.h>
42 #include <libpspp/message.h>
43 #include <libpspp/misc.h>
44 #include <libpspp/str.h>
45 #include <math/group-proc.h>
46 #include <math/levene.h>
47 #include <output/manager.h>
48 #include <output/table.h>
53 #define _(msgid) gettext (msgid)
61 variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
63 +missing=miss:!analysis/listwise,
64 incl:include/!exclude;
65 format=fmt:!labels/nolabels;
66 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
74 /* Function to use for testing for missing values */
75 static is_missing_func *value_is_missing;
77 /* Variable for the GROUPS subcommand, if given. */
78 static struct variable *indep_var;
86 struct group_properties
88 /* The comparison criterion */
89 enum comparison criterion;
91 /* The width of the independent variable */
95 /* The value of the independent variable at which groups are determined to
96 belong to one group or the other */
97 double critical_value;
100 /* The values of the independent variable for each group */
101 union value g_value[2];
107 static struct group_properties gp ;
111 /* PAIRS: Number of pairs to be compared ; each pair. */
112 static int n_pairs = 0 ;
115 /* The variables comprising the pair */
116 struct variable *v[2];
118 /* The number of valid variable pairs */
121 /* The sum of the members */
124 /* sum of squares of the members */
127 /* Std deviation of the members */
131 /* Sample Std deviation of the members */
134 /* The means of the members */
137 /* The correlation coefficient between the variables */
140 /* The sum of the differences */
143 /* The sum of the products */
146 /* The mean of the differences */
149 /* The sum of the squares of the differences */
152 /* The std deviation of the differences */
156 static struct pair *pairs=0;
158 static int parse_value (union value * v, int type) ;
160 /* Structures and Functions for the Statistics Summary Box */
162 typedef void populate_ssbox_func(struct ssbox *ssb,
163 struct cmd_t_test *cmd);
164 typedef void finalize_ssbox_func(struct ssbox *ssb);
170 populate_ssbox_func *populate;
171 finalize_ssbox_func *finalize;
176 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
178 /* Populate a ssbox according to cmd */
179 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
181 /* Submit and destroy a ssbox */
182 void ssbox_finalize(struct ssbox *ssb);
184 /* A function to create, populate and submit the Paired Samples Correlation
189 /* Structures and Functions for the Test Results Box */
192 typedef void populate_trbox_func(struct trbox *trb,
193 struct cmd_t_test *cmd);
194 typedef void finalize_trbox_func(struct trbox *trb);
198 populate_trbox_func *populate;
199 finalize_trbox_func *finalize;
203 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
205 /* Populate a ssbox according to cmd */
206 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
208 /* Submit and destroy a ssbox */
209 void trbox_finalize(struct trbox *trb);
211 /* Which mode was T-TEST invoked */
219 static int common_calc (const struct ccase *, void *);
220 static void common_precalc (struct cmd_t_test *);
221 static void common_postcalc (struct cmd_t_test *);
223 static int one_sample_calc (const struct ccase *, void *);
224 static void one_sample_precalc (struct cmd_t_test *);
225 static void one_sample_postcalc (struct cmd_t_test *);
227 static int paired_calc (const struct ccase *, void *);
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 ccase *, struct cmd_t_test *);
233 static void group_postcalc (struct cmd_t_test *);
236 static bool calculate(const struct casefile *cf, void *_mode);
240 static struct cmd_t_test cmd;
242 static int bad_weight_warn;
245 static int compare_group_binary(const struct group_statistics *a,
246 const struct group_statistics *b,
247 const struct group_properties *p);
250 static unsigned hash_group_binary(const struct group_statistics *g,
251 const struct group_properties *p);
260 if ( !parse_t_test(&cmd) )
263 if (! cmd.sbc_criteria)
268 if (cmd.sbc_testval) ++m;
269 if (cmd.sbc_groups) ++m;
270 if (cmd.sbc_pairs) ++m;
275 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
284 else if (cmd.sbc_groups)
289 if ( mode == T_PAIRED)
291 if (cmd.sbc_variables)
293 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
299 /* Iterate through the pairs and put each variable that is a
300 member of a pair into cmd.v_variables */
303 struct hsh_iterator hi;
304 struct hsh_table *hash;
307 hash = hsh_create (n_pairs, compare_var_names, hash_var_name, 0, 0);
309 for (i=0; i < n_pairs; ++i)
311 hsh_insert(hash,pairs[i].v[0]);
312 hsh_insert(hash,pairs[i].v[1]);
315 assert(cmd.n_variables == 0);
316 cmd.n_variables = hsh_count(hash);
318 cmd.v_variables = xnrealloc (cmd.v_variables, cmd.n_variables,
319 sizeof *cmd.v_variables);
320 /* Iterate through the hash */
321 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
323 v=hsh_next(hash,&hi) )
324 cmd.v_variables[i++]=v;
329 else if ( !cmd.sbc_variables)
331 msg(SE, _("One or more VARIABLES must be specified."));
337 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
338 if (cmd.incl == TTS_INCLUDE )
339 value_is_missing = mv_is_value_system_missing;
341 value_is_missing = mv_is_value_missing;
345 ok = multipass_procedure_with_splits (calculate, &cmd);
351 if ( mode == T_IND_SAMPLES)
354 /* Destroy any group statistics we created */
355 for (v = 0 ; v < cmd.n_variables ; ++v )
357 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
358 hsh_destroy (grpp->group_hash);
363 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
367 tts_custom_groups (struct cmd_t_test *cmd UNUSED)
369 int n_group_values=0;
373 indep_var = parse_variable ();
376 lex_error ("expecting variable name in GROUPS subcommand");
380 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
382 msg (SE, _("Long string variable %s is not valid here."),
387 if (!lex_match ('('))
389 if (indep_var->type == NUMERIC)
391 gp.v.g_value[0].f = 1;
392 gp.v.g_value[1].f = 2;
394 gp.criterion = CMP_EQ;
402 msg (SE, _("When applying GROUPS to a string variable, two "
403 "values must be specified."));
408 if (!parse_value (&gp.v.g_value[0], indep_var->type))
414 if (indep_var->type != NUMERIC)
417 msg (SE, _("When applying GROUPS to a string variable, two "
418 "values must be specified."));
421 gp.criterion = CMP_LE;
422 gp.v.critical_value = gp.v.g_value[0].f;
428 if (!parse_value (&gp.v.g_value[1], indep_var->type))
432 if (!lex_force_match (')'))
435 if ( n_group_values == 2 )
436 gp.criterion = CMP_EQ ;
438 gp.criterion = CMP_LE ;
446 tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
448 struct variable **vars;
450 size_t n_pairs_local;
452 size_t n_before_WITH;
453 size_t n_after_WITH = SIZE_MAX;
454 int paired ; /* Was the PAIRED keyword given ? */
459 if (!parse_variables (default_dict, &vars, &n_vars,
460 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
468 if (lex_match (T_WITH))
470 n_before_WITH = n_vars;
471 if (!parse_variables (default_dict, &vars, &n_vars,
472 PV_DUPLICATE | PV_APPEND
473 | PV_NUMERIC | PV_NO_SCRATCH))
478 n_after_WITH = n_vars - n_before_WITH;
481 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
483 /* Determine the number of pairs needed */
486 if (n_before_WITH != n_after_WITH)
489 msg (SE, _("PAIRED was specified but the number of variables "
490 "preceding WITH (%d) did not match the number "
492 n_before_WITH, n_after_WITH );
495 n_pairs_local = n_before_WITH;
497 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
499 n_pairs_local = n_before_WITH * n_after_WITH ;
501 else /* Neither WITH nor PAIRED keyword given */
506 msg (SE, _("At least two variables must be specified "
511 /* how many ways can you pick 2 from n_vars ? */
512 n_pairs_local = n_vars * (n_vars - 1) / 2;
516 /* Allocate storage for the pairs */
517 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
519 /* Populate the pairs with the appropriate variables */
524 assert(n_pairs_local == n_vars / 2);
525 for (i = 0; i < n_pairs_local; ++i)
527 pairs[i].v[n_pairs] = vars[i];
528 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
531 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
536 for(i=0 ; i < n_before_WITH ; ++i )
538 for(j=0 ; j < n_after_WITH ; ++j)
540 pairs[p].v[0] = vars[i];
541 pairs[p].v[1] = vars[j+n_before_WITH];
546 else /* Neither WITH nor PAIRED given */
551 for(i=0 ; i < n_vars ; ++i )
553 for(j=i+1 ; j < n_vars ; ++j)
555 pairs[p].v[0] = vars[i];
556 pairs[p].v[1] = vars[j];
562 n_pairs+=n_pairs_local;
568 /* Parses the current token (numeric or string, depending on type)
569 value v and returns success. */
571 parse_value (union value * v, int type )
575 if (!lex_force_num ())
581 if (!lex_force_string ())
583 strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
592 /* Implementation of the SSBOX object */
594 void ssbox_base_init(struct ssbox *this, int cols,int rows);
596 void ssbox_base_finalize(struct ssbox *ssb);
598 void ssbox_one_sample_init(struct ssbox *this,
599 struct cmd_t_test *cmd );
601 void ssbox_independent_samples_init(struct ssbox *this,
602 struct cmd_t_test *cmd);
604 void ssbox_paired_init(struct ssbox *this,
605 struct cmd_t_test *cmd);
608 /* Factory to create an ssbox */
610 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
615 ssbox_one_sample_init(ssb,cmd);
618 ssbox_independent_samples_init(ssb,cmd);
621 ssbox_paired_init(ssb,cmd);
630 /* Despatcher for the populate method */
632 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
634 ssb->populate(ssb,cmd);
638 /* Despatcher for finalize */
640 ssbox_finalize(struct ssbox *ssb)
646 /* Submit the box and clear up */
648 ssbox_base_finalize(struct ssbox *ssb)
655 /* Initialize a ssbox struct */
657 ssbox_base_init(struct ssbox *this, int cols,int rows)
659 this->finalize = ssbox_base_finalize;
660 this->t = tab_create (cols, rows, 0);
662 tab_columns (this->t, SOM_COL_DOWN, 1);
663 tab_headers (this->t,0,0,1,0);
664 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
665 tab_hline(this->t, TAL_2,0,cols-1,1);
666 tab_dim (this->t, tab_natural_dimensions);
669 void ssbox_one_sample_populate(struct ssbox *ssb,
670 struct cmd_t_test *cmd);
672 /* Initialize the one_sample ssbox */
674 ssbox_one_sample_init(struct ssbox *this,
675 struct cmd_t_test *cmd )
678 const int vsize=cmd->n_variables+1;
680 this->populate = ssbox_one_sample_populate;
682 ssbox_base_init(this, hsize,vsize);
683 tab_title (this->t, _("One-Sample Statistics"));
684 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
685 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
686 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
687 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
688 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
691 void ssbox_independent_samples_populate(struct ssbox *ssb,
692 struct cmd_t_test *cmd);
694 /* Initialize the independent samples ssbox */
696 ssbox_independent_samples_init(struct ssbox *this,
697 struct cmd_t_test *cmd)
700 int vsize = cmd->n_variables*2 +1;
702 this->populate = ssbox_independent_samples_populate;
704 ssbox_base_init(this, hsize,vsize);
705 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
706 tab_title (this->t, _("Group Statistics"));
707 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
708 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
709 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
710 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
711 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
715 /* Populate the ssbox for independent samples */
717 ssbox_independent_samples_populate(struct ssbox *ssb,
718 struct cmd_t_test *cmd)
724 double indep_value[2];
726 char prefix[2][3]={"",""};
728 if ( indep_var->type == NUMERIC )
730 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
731 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
735 val_lab0 = gp.v.g_value[0].s;
736 val_lab1 = gp.v.g_value[1].s;
739 if (gp.criterion == CMP_LE )
741 strcpy(prefix[0],"< ");
742 strcpy(prefix[1],">=");
743 indep_value[0] = gp.v.critical_value;
744 indep_value[1] = gp.v.critical_value;
748 indep_value[0] = gp.v.g_value[0].f;
749 indep_value[1] = gp.v.g_value[1].f;
754 for (i=0; i < cmd->n_variables; ++i)
756 struct variable *var = cmd->v_variables[i];
757 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
760 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
763 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
764 "%s%s", prefix[0], val_lab0);
766 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
767 "%s%g", prefix[0], indep_value[0]);
771 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
772 "%s%s", prefix[1], val_lab1);
774 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
775 "%s%g", prefix[1], indep_value[1]);
778 /* Fill in the group statistics */
779 for ( count = 0 ; count < 2 ; ++count )
781 union value search_val;
783 struct group_statistics *gs;
785 if ( gp.criterion == CMP_LE )
789 /* less than ( < ) case */
790 search_val.f = gp.v.critical_value - 1.0;
795 search_val.f = gp.v.critical_value + 1.0;
800 search_val = gp.v.g_value[count];
803 gs = hsh_find(grp_hash, (void *) &search_val);
806 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
807 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
808 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
809 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
815 void ssbox_paired_populate(struct ssbox *ssb,
816 struct cmd_t_test *cmd);
818 /* Initialize the paired values ssbox */
820 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
824 int vsize = n_pairs*2+1;
826 this->populate = ssbox_paired_populate;
828 ssbox_base_init(this, hsize,vsize);
829 tab_title (this->t, _("Paired Sample Statistics"));
830 tab_vline(this->t,TAL_GAP,1,0,vsize-1);
831 tab_vline(this->t,TAL_2,2,0,vsize-1);
832 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
833 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
834 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
835 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
839 /* Populate the ssbox for paired values */
841 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
847 for (i=0; i < n_pairs; ++i)
851 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
853 for (j=0 ; j < 2 ; ++j)
855 struct group_statistics *gs;
857 gs = &group_proc_get (pairs[i].v[j])->ugs;
861 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
864 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
865 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
866 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
867 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
873 /* Populate the one sample ssbox */
875 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
881 for (i=0; i < cmd->n_variables; ++i)
883 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
885 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
886 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
887 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
888 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
889 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
896 /* Implementation of the Test Results box struct */
898 void trbox_base_init(struct trbox *self,size_t n_vars, int cols);
899 void trbox_base_finalize(struct trbox *trb);
901 void trbox_independent_samples_init(struct trbox *trb,
902 struct cmd_t_test *cmd );
904 void trbox_independent_samples_populate(struct trbox *trb,
905 struct cmd_t_test *cmd);
907 void trbox_one_sample_init(struct trbox *self,
908 struct cmd_t_test *cmd );
910 void trbox_one_sample_populate(struct trbox *trb,
911 struct cmd_t_test *cmd);
913 void trbox_paired_init(struct trbox *self,
914 struct cmd_t_test *cmd );
916 void trbox_paired_populate(struct trbox *trb,
917 struct cmd_t_test *cmd);
921 /* Create a trbox according to mode*/
923 trbox_create(struct trbox *trb,
924 struct cmd_t_test *cmd, int mode)
929 trbox_one_sample_init(trb,cmd);
932 trbox_independent_samples_init(trb,cmd);
935 trbox_paired_init(trb,cmd);
942 /* Populate a trbox according to cmd */
944 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
946 trb->populate(trb,cmd);
949 /* Submit and destroy a trbox */
951 trbox_finalize(struct trbox *trb)
956 /* Initialize the independent samples trbox */
958 trbox_independent_samples_init(struct trbox *self,
959 struct cmd_t_test *cmd UNUSED)
962 const int vsize=cmd->n_variables*2+3;
965 self->populate = trbox_independent_samples_populate;
967 trbox_base_init(self,cmd->n_variables*2,hsize);
968 tab_title(self->t,_("Independent Samples Test"));
969 tab_hline(self->t,TAL_1,2,hsize-1,1);
970 tab_vline(self->t,TAL_2,2,0,vsize-1);
971 tab_vline(self->t,TAL_1,4,0,vsize-1);
972 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
973 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
974 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
975 tab_joint_text(self->t, 2, 0, 3, 0,
976 TAB_CENTER,_("Levene's Test for Equality of Variances"));
977 tab_joint_text(self->t, 4,0,hsize-1,0,
978 TAB_CENTER,_("t-test for Equality of Means"));
980 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
981 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
982 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
983 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
984 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
985 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
986 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
987 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
988 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
990 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
991 _("%g%% Confidence Interval of the Difference"),
992 cmd->criteria*100.0);
996 /* Populate the independent samples trbox */
998 trbox_independent_samples_populate(struct trbox *self,
999 struct cmd_t_test *cmd )
1004 for (i=0; i < cmd->n_variables; ++i)
1013 double pooled_variance;
1014 double std_err_diff;
1017 struct variable *var = cmd->v_variables[i];
1018 struct group_proc *grp_data = group_proc_get (var);
1020 struct hsh_table *grp_hash = grp_data->group_hash;
1022 struct group_statistics *gs0 ;
1023 struct group_statistics *gs1 ;
1025 union value search_val;
1027 if ( gp.criterion == CMP_LE )
1028 search_val.f = gp.v.critical_value - 1.0;
1030 search_val = gp.v.g_value[0];
1032 gs0 = hsh_find(grp_hash, (void *) &search_val);
1035 if ( gp.criterion == CMP_LE )
1036 search_val.f = gp.v.critical_value + 1.0;
1038 search_val = gp.v.g_value[1];
1040 gs1 = hsh_find(grp_hash, (void *) &search_val);
1044 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1046 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1049 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1051 /* Now work out the significance of the Levene test */
1052 df1 = 1; df2 = grp_data->ugs.n - 2;
1053 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1055 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1057 df = gs0->n + gs1->n - 2.0 ;
1058 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1060 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1062 (gs1->n )*pow2(gs1->s_std_dev)
1065 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1066 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1068 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1070 p = gsl_cdf_tdist_P(t, df);
1071 q = gsl_cdf_tdist_Q(t, df);
1073 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1075 mean_diff = gs0->mean - gs1->mean;
1076 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1079 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1080 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1083 /* Now work out the confidence interval */
1084 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1086 t = gsl_cdf_tdist_Qinv(q,df);
1087 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1088 mean_diff - t * std_err_diff, 8, 3);
1090 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1091 mean_diff + t * std_err_diff, 8, 3);
1096 /* Now for the \sigma_1 != \sigma_2 case */
1097 tab_text (self->t, 1, i*2+3+1,
1098 TAB_LEFT, _("Equal variances not assumed"));
1101 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1102 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1104 t = mean_diff / sqrt(se2) ;
1105 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1108 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1112 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1116 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1118 p = gsl_cdf_tdist_P(t, df);
1119 q = gsl_cdf_tdist_Q(t, df);
1121 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1123 /* Now work out the confidence interval */
1124 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1126 t = gsl_cdf_tdist_Qinv(q, df);
1128 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1131 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1134 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1135 mean_diff - t * std_err_diff, 8, 3);
1137 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1138 mean_diff + t * std_err_diff, 8, 3);
1144 /* Initialize the paired samples trbox */
1146 trbox_paired_init(struct trbox *self,
1147 struct cmd_t_test *cmd UNUSED)
1151 const int vsize=n_pairs+3;
1153 self->populate = trbox_paired_populate;
1155 trbox_base_init(self,n_pairs,hsize);
1156 tab_title (self->t, _("Paired Samples Test"));
1157 tab_hline(self->t,TAL_1,2,6,1);
1158 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1159 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1160 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1161 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1162 tab_hline(self->t,TAL_1,5,6, 2);
1163 tab_vline(self->t,TAL_GAP,6,0,1);
1165 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1166 _("%g%% Confidence Interval of the Difference"),
1167 cmd->criteria*100.0);
1169 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1170 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1171 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1172 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1173 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1174 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1175 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1176 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1179 /* Populate the paired samples trbox */
1181 trbox_paired_populate(struct trbox *trb,
1182 struct cmd_t_test *cmd UNUSED)
1186 for (i=0; i < n_pairs; ++i)
1191 double n = pairs[i].n;
1195 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1197 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1198 pairs[i].v[0]->name, pairs[i].v[1]->name);
1200 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1202 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1205 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1206 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1208 /* Now work out the confidence interval */
1209 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1211 t = gsl_cdf_tdist_Qinv(q, df);
1213 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1214 pairs[i].mean_diff - t * se_mean , 8, 4);
1216 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1217 pairs[i].mean_diff + t * se_mean , 8, 4);
1219 t = (pairs[i].mean[0] - pairs[i].mean[1])
1221 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1222 2 * pairs[i].correlation *
1223 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1227 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1229 /* Degrees of freedom */
1230 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1232 p = gsl_cdf_tdist_P(t,df);
1233 q = gsl_cdf_tdist_P(t,df);
1235 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1240 /* Initialize the one sample trbox */
1242 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1245 const int vsize=cmd->n_variables+3;
1247 self->populate = trbox_one_sample_populate;
1249 trbox_base_init(self, cmd->n_variables,hsize);
1250 tab_title (self->t, _("One-Sample Test"));
1251 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1252 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1254 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1255 _("Test Value = %f"), cmd->n_testval[0]);
1257 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1260 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1261 _("%g%% Confidence Interval of the Difference"),
1262 cmd->criteria*100.0);
1264 tab_vline(self->t,TAL_GAP,6,1,1);
1265 tab_hline(self->t,TAL_1,5,6,2);
1266 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1267 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1268 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1269 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1270 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1271 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1276 /* Populate the one sample trbox */
1278 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1284 for (i=0; i < cmd->n_variables; ++i)
1289 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1292 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1294 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1296 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1298 /* degrees of freedom */
1301 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1303 p = gsl_cdf_tdist_P(t, df);
1304 q = gsl_cdf_tdist_Q(t, df);
1306 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1308 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1310 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1313 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1314 t = gsl_cdf_tdist_Qinv(q, df);
1316 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1317 gs->mean_diff - t * gs->se_mean, 8,4);
1319 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1320 gs->mean_diff + t * gs->se_mean, 8,4);
1324 /* Base initializer for the generalized trbox */
1326 trbox_base_init(struct trbox *self, size_t data_rows, int cols)
1328 const size_t rows = 3 + data_rows;
1330 self->finalize = trbox_base_finalize;
1331 self->t = tab_create (cols, rows, 0);
1332 tab_headers (self->t,0,0,3,0);
1333 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1334 tab_hline(self->t, TAL_2,0,cols-1,3);
1335 tab_dim (self->t, tab_natural_dimensions);
1339 /* Base finalizer for the trbox */
1341 trbox_base_finalize(struct trbox *trb)
1347 /* Create , populate and submit the Paired Samples Correlation box */
1351 const int rows=1+n_pairs;
1355 struct tab_table *table;
1357 table = tab_create (cols,rows,0);
1359 tab_columns (table, SOM_COL_DOWN, 1);
1360 tab_headers (table,0,0,1,0);
1361 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1362 tab_hline(table, TAL_2, 0, cols - 1, 1);
1363 tab_vline(table, TAL_2, 2, 0, rows - 1);
1364 tab_dim(table, tab_natural_dimensions);
1365 tab_title(table, _("Paired Samples Correlations"));
1367 /* column headings */
1368 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1369 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1370 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1372 for (i=0; i < n_pairs; ++i)
1376 double df = pairs[i].n -2;
1378 double correlation_t =
1379 pairs[i].correlation * sqrt(df) /
1380 sqrt(1 - pow2(pairs[i].correlation));
1384 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1387 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1388 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1392 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1393 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1395 p = gsl_cdf_tdist_P(correlation_t, df);
1396 q = gsl_cdf_tdist_Q(correlation_t, df);
1398 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1407 /* Calculation Implementation */
1409 /* Per case calculations common to all variants of the T test */
1411 common_calc (const struct ccase *c, void *_cmd)
1414 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1416 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1419 /* Skip the entire case if /MISSING=LISTWISE is set */
1420 if ( cmd->miss == TTS_LISTWISE )
1422 for(i=0; i< cmd->n_variables ; ++i)
1424 struct variable *v = cmd->v_variables[i];
1425 const union value *val = case_data (c, v->fv);
1427 if (value_is_missing(&v->miss, val) )
1434 /* Listwise has to be implicit if the independent variable is missing ?? */
1435 if ( cmd->sbc_groups )
1437 const union value *gv = case_data (c, indep_var->fv);
1438 if ( value_is_missing(&indep_var->miss, gv) )
1445 for(i=0; i< cmd->n_variables ; ++i)
1447 struct group_statistics *gs;
1448 struct variable *v = cmd->v_variables[i];
1449 const union value *val = case_data (c, v->fv);
1451 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1453 if (! value_is_missing(&v->miss, val) )
1456 gs->sum+=weight * val->f;
1457 gs->ssq+=weight * val->f * val->f;
1463 /* Pre calculations common to all variants of the T test */
1465 common_precalc ( struct cmd_t_test *cmd )
1469 for(i=0; i< cmd->n_variables ; ++i)
1471 struct group_statistics *gs;
1472 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1481 /* Post calculations common to all variants of the T test */
1483 common_postcalc ( struct cmd_t_test *cmd )
1488 for(i=0; i< cmd->n_variables ; ++i)
1490 struct group_statistics *gs;
1491 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1493 gs->mean=gs->sum / gs->n;
1494 gs->s_std_dev= sqrt(
1495 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1500 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1503 gs->se_mean = gs->std_dev / sqrt(gs->n);
1504 gs->mean_diff= gs->sum_diff / gs->n;
1508 /* Per case calculations for one sample t test */
1510 one_sample_calc (const struct ccase *c, void *cmd_)
1513 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1516 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1518 /* Skip the entire case if /MISSING=LISTWISE is set */
1519 if ( cmd->miss == TTS_LISTWISE )
1521 for(i=0; i< cmd->n_variables ; ++i)
1523 struct variable *v = cmd->v_variables[i];
1524 const union value *val = case_data (c, v->fv);
1526 if (value_is_missing(&v->miss, val) )
1533 for(i=0; i< cmd->n_variables ; ++i)
1535 struct group_statistics *gs;
1536 struct variable *v = cmd->v_variables[i];
1537 const union value *val = case_data (c, v->fv);
1539 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1541 if ( ! value_is_missing(&v->miss, val))
1542 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1548 /* Pre calculations for one sample t test */
1550 one_sample_precalc ( struct cmd_t_test *cmd )
1554 for(i=0; i< cmd->n_variables ; ++i)
1556 struct group_statistics *gs;
1557 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1563 /* Post calculations for one sample t test */
1565 one_sample_postcalc (struct cmd_t_test *cmd)
1569 for(i=0; i< cmd->n_variables ; ++i)
1571 struct group_statistics *gs;
1572 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1574 gs->mean_diff = gs->sum_diff / gs->n ;
1581 paired_precalc (struct cmd_t_test *cmd UNUSED)
1585 for(i=0; i < n_pairs ; ++i )
1588 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1589 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1590 pairs[i].sum_of_prod = 0;
1591 pairs[i].correlation = 0;
1592 pairs[i].sum_of_diffs = 0;
1593 pairs[i].ssq_diffs = 0;
1600 paired_calc (const struct ccase *c, void *cmd_)
1604 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1606 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1608 /* Skip the entire case if /MISSING=LISTWISE is set ,
1609 AND one member of a pair is missing */
1610 if ( cmd->miss == TTS_LISTWISE )
1612 for(i=0; i < n_pairs ; ++i )
1614 struct variable *v0 = pairs[i].v[0];
1615 struct variable *v1 = pairs[i].v[1];
1617 const union value *val0 = case_data (c, v0->fv);
1618 const union value *val1 = case_data (c, v1->fv);
1620 if ( value_is_missing(&v0->miss, val0) ||
1621 value_is_missing(&v1->miss, val1) )
1628 for(i=0; i < n_pairs ; ++i )
1630 struct variable *v0 = pairs[i].v[0];
1631 struct variable *v1 = pairs[i].v[1];
1633 const union value *val0 = case_data (c, v0->fv);
1634 const union value *val1 = case_data (c, v1->fv);
1636 if ( ( !value_is_missing(&v0->miss, val0)
1637 && !value_is_missing(&v1->miss, val1) ) )
1639 pairs[i].n += weight;
1640 pairs[i].sum[0] += weight * val0->f;
1641 pairs[i].sum[1] += weight * val1->f;
1643 pairs[i].ssq[0] += weight * pow2(val0->f);
1644 pairs[i].ssq[1] += weight * pow2(val1->f);
1646 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1648 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1649 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1657 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1661 for(i=0; i < n_pairs ; ++i )
1664 const double n = pairs[i].n;
1666 for (j=0; j < 2 ; ++j)
1668 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1669 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1670 pow2(pairs[i].mean[j]))
1673 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1674 pow2(pairs[i].mean[j]))
1678 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1679 pairs[i].mean[0] * pairs[i].mean[1] ;
1680 /* correlation now actually contains the covariance */
1682 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1683 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1685 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1687 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1688 ( pairs[i].ssq_diffs / n )
1690 pow2(pairs[i].mean_diff )
1696 group_precalc (struct cmd_t_test *cmd )
1701 for(i=0; i< cmd->n_variables ; ++i)
1703 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1705 /* There's always 2 groups for a T - TEST */
1708 gp.indep_width = indep_var->width;
1710 ttpr->group_hash = hsh_create(2,
1711 (hsh_compare_func *) compare_group_binary,
1712 (hsh_hash_func *) hash_group_binary,
1713 (hsh_free_func *) free_group,
1716 for (j=0 ; j < 2 ; ++j)
1719 struct group_statistics *gs = xmalloc (sizeof *gs);
1725 if ( gp.criterion == CMP_EQ )
1727 gs->id = gp.v.g_value[j];
1732 gs->id.f = gp.v.critical_value - 1.0 ;
1734 gs->id.f = gp.v.critical_value + 1.0 ;
1737 hsh_insert ( ttpr->group_hash, (void *) gs );
1745 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1749 const union value *gv = case_data (c, indep_var->fv);
1751 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1753 if ( value_is_missing(&indep_var->miss, gv) )
1758 if ( cmd->miss == TTS_LISTWISE )
1760 for(i=0; i< cmd->n_variables ; ++i)
1762 struct variable *v = cmd->v_variables[i];
1763 const union value *val = case_data (c, v->fv);
1765 if (value_is_missing(&v->miss, val) )
1772 gv = case_data (c, indep_var->fv);
1774 for(i=0; i< cmd->n_variables ; ++i)
1776 struct variable *var = cmd->v_variables[i];
1777 const union value *val = case_data (c, var->fv);
1778 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1779 struct group_statistics *gs;
1781 gs = hsh_find(grp_hash, (void *) gv);
1783 /* If the independent variable doesn't match either of the values
1784 for this case then move on to the next case */
1788 if ( !value_is_missing(&var->miss, val) )
1791 gs->sum+=weight * val->f;
1792 gs->ssq+=weight * pow2(val->f);
1801 group_postcalc ( struct cmd_t_test *cmd )
1805 for(i=0; i< cmd->n_variables ; ++i)
1807 struct variable *var = cmd->v_variables[i];
1808 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1809 struct hsh_iterator g;
1810 struct group_statistics *gs;
1813 for (gs = hsh_first (grp_hash,&g);
1815 gs = hsh_next(grp_hash,&g))
1817 gs->mean = gs->sum / gs->n;
1819 gs->s_std_dev= sqrt(
1820 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1825 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1828 gs->se_mean = gs->std_dev / sqrt(gs->n);
1838 calculate(const struct casefile *cf, void *cmd_)
1840 struct ssbox stat_summary_box;
1841 struct trbox test_results_box;
1843 struct casereader *r;
1846 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1848 common_precalc(cmd);
1849 for(r = casefile_get_reader (cf);
1850 casereader_read (r, &c) ;
1853 common_calc(&c,cmd);
1855 casereader_destroy (r);
1856 common_postcalc(cmd);
1861 one_sample_precalc(cmd);
1862 for(r = casefile_get_reader (cf);
1863 casereader_read (r, &c) ;
1866 one_sample_calc(&c,cmd);
1868 casereader_destroy (r);
1869 one_sample_postcalc(cmd);
1873 paired_precalc(cmd);
1874 for(r = casefile_get_reader (cf);
1875 casereader_read (r, &c) ;
1878 paired_calc(&c,cmd);
1880 casereader_destroy (r);
1881 paired_postcalc(cmd);
1887 for(r = casefile_get_reader (cf);
1888 casereader_read (r, &c) ;
1893 casereader_destroy (r);
1894 group_postcalc(cmd);
1896 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1897 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1902 ssbox_create(&stat_summary_box,cmd,mode);
1903 ssbox_populate(&stat_summary_box,cmd);
1904 ssbox_finalize(&stat_summary_box);
1906 if ( mode == T_PAIRED)
1909 trbox_create(&test_results_box,cmd,mode);
1910 trbox_populate(&test_results_box,cmd);
1911 trbox_finalize(&test_results_box);
1916 short which_group(const struct group_statistics *g,
1917 const struct group_properties *p);
1919 /* Return -1 if the id of a is less than b; +1 if greater than and
1922 compare_group_binary(const struct group_statistics *a,
1923 const struct group_statistics *b,
1924 const struct group_properties *p)
1929 if ( p->criterion == CMP_LE )
1931 /* less-than-or-equal comparision is not meaningfull for
1932 alpha variables, so we shouldn't ever arrive here */
1933 assert(p->indep_width == 0 ) ;
1935 flag_a = ( a->id.f < p->v.critical_value ) ;
1936 flag_b = ( b->id.f < p->v.critical_value ) ;
1940 flag_a = which_group(a, p);
1941 flag_b = which_group(b, p);
1944 if (flag_a < flag_b )
1947 return (flag_a > flag_b);
1950 /* This is a degenerate case of a hash, since it can only return three possible
1951 values. It's really a comparison, being used as a hash function */
1954 hash_group_binary(const struct group_statistics *g,
1955 const struct group_properties *p)
1959 if ( p->criterion == CMP_LE )
1961 /* Not meaningfull to do a less than compare for alpha values ? */
1962 assert(p->indep_width == 0 ) ;
1963 flag = ( g->id.f < p->v.critical_value ) ;
1965 else if ( p->criterion == CMP_EQ)
1967 flag = which_group(g,p);
1975 /* return 0 if G belongs to group 0,
1976 1 if it belongs to group 1,
1977 2 if it belongs to neither group */
1979 which_group(const struct group_statistics *g,
1980 const struct group_properties *p)
1983 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1986 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))