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., 59 Temple Place - Suite 330, Boston, MA
23 #include <gsl/gsl_cdf.h>
32 #include "dictionary.h"
39 #include "value-labels.h"
43 #include "group_proc.h"
52 variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
54 +missing=miss:!analysis/listwise,
55 incl:include/!exclude;
56 format=fmt:!labels/nolabels;
57 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
65 /* Function to use for testing for missing values */
66 static is_missing_func value_is_missing;
68 /* Variable for the GROUPS subcommand, if given. */
69 static struct variable *indep_var;
77 struct group_properties
79 /* The comparison criterion */
80 enum comparison criterion;
82 /* The width of the independent variable */
86 /* The value of the independent variable at which groups are determined to
87 belong to one group or the other */
88 double critical_value;
91 /* The values of the independent variable for each group */
92 union value g_value[2];
98 static struct group_properties gp ;
102 /* PAIRS: Number of pairs to be compared ; each pair. */
103 static int n_pairs = 0 ;
106 /* The variables comprising the pair */
107 struct variable *v[2];
109 /* The number of valid variable pairs */
112 /* The sum of the members */
115 /* sum of squares of the members */
118 /* Std deviation of the members */
122 /* Sample Std deviation of the members */
125 /* The means of the members */
128 /* The correlation coefficient between the variables */
131 /* The sum of the differences */
134 /* The sum of the products */
137 /* The mean of the differences */
140 /* The sum of the squares of the differences */
143 /* The std deviation of the differences */
147 static struct pair *pairs=0;
149 static int parse_value (union value * v, int type) ;
151 /* Structures and Functions for the Statistics Summary Box */
153 typedef void populate_ssbox_func(struct ssbox *ssb,
154 struct cmd_t_test *cmd);
155 typedef void finalize_ssbox_func(struct ssbox *ssb);
161 populate_ssbox_func *populate;
162 finalize_ssbox_func *finalize;
167 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
169 /* Populate a ssbox according to cmd */
170 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
172 /* Submit and destroy a ssbox */
173 void ssbox_finalize(struct ssbox *ssb);
175 /* A function to create, populate and submit the Paired Samples Correlation
180 /* Structures and Functions for the Test Results Box */
183 typedef void populate_trbox_func(struct trbox *trb,
184 struct cmd_t_test *cmd);
185 typedef void finalize_trbox_func(struct trbox *trb);
189 populate_trbox_func *populate;
190 finalize_trbox_func *finalize;
194 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
196 /* Populate a ssbox according to cmd */
197 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
199 /* Submit and destroy a ssbox */
200 void trbox_finalize(struct trbox *trb);
202 /* Which mode was T-TEST invoked */
210 static int common_calc (const struct ccase *, void *);
211 static void common_precalc (struct cmd_t_test *);
212 static void common_postcalc (struct cmd_t_test *);
214 static int one_sample_calc (const struct ccase *, void *);
215 static void one_sample_precalc (struct cmd_t_test *);
216 static void one_sample_postcalc (struct cmd_t_test *);
218 static int paired_calc (const struct ccase *, void *);
219 static void paired_precalc (struct cmd_t_test *);
220 static void paired_postcalc (struct cmd_t_test *);
222 static void group_precalc (struct cmd_t_test *);
223 static int group_calc (const struct ccase *, struct cmd_t_test *);
224 static void group_postcalc (struct cmd_t_test *);
227 static int compare_var_name (const void *a_, const void *b_, void *v_ UNUSED);
228 static unsigned hash_var_name (const void *a_, void *v_ UNUSED);
230 static void calculate(const struct casefile *cf, void *_mode);
234 static struct cmd_t_test cmd;
236 static int bad_weight_warn;
239 static int compare_group_binary(const struct group_statistics *a,
240 const struct group_statistics *b,
241 const struct group_properties *p);
244 static unsigned hash_group_binary(const struct group_statistics *g,
245 const struct group_properties *p);
253 if ( !parse_t_test(&cmd) )
256 if (! cmd.sbc_criteria)
261 if (cmd.sbc_testval) ++m;
262 if (cmd.sbc_groups) ++m;
263 if (cmd.sbc_pairs) ++m;
268 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
277 else if (cmd.sbc_groups)
282 if ( mode == T_PAIRED)
284 if (cmd.sbc_variables)
286 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
292 /* Iterate through the pairs and put each variable that is a
293 member of a pair into cmd.v_variables */
296 struct hsh_iterator hi;
297 struct hsh_table *hash;
300 hash=hsh_create(n_pairs,compare_var_name,hash_var_name,0,0);
302 for (i=0; i < n_pairs; ++i)
304 hsh_insert(hash,pairs[i].v[0]);
305 hsh_insert(hash,pairs[i].v[1]);
308 assert(cmd.n_variables == 0);
309 cmd.n_variables = hsh_count(hash);
311 cmd.v_variables = xrealloc(cmd.v_variables,
312 sizeof(struct variable) * cmd.n_variables);
313 /* Iterate through the hash */
314 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
316 v=hsh_next(hash,&hi) )
317 cmd.v_variables[i++]=v;
322 else if ( !cmd.sbc_variables)
324 msg(SE, _("One or more VARIABLES must be specified."));
330 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
331 if (cmd.incl == TTS_INCLUDE )
332 value_is_missing = is_system_missing;
334 value_is_missing = is_missing;
338 multipass_procedure_with_splits (calculate, &cmd);
344 if ( mode == T_IND_SAMPLES)
347 /* Destroy any group statistics we created */
348 for (v = 0 ; v < cmd.n_variables ; ++v )
350 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
351 hsh_destroy (grpp->group_hash);
360 tts_custom_groups (struct cmd_t_test *cmd UNUSED)
362 int n_group_values=0;
366 if (token != T_ALL &&
367 (token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
370 msg(SE,_("`%s' is not a variable name"),tokid);
374 indep_var = parse_variable ();
377 lex_error ("expecting variable name in GROUPS subcommand");
381 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
383 msg (SE, _("Long string variable %s is not valid here."),
388 if (!lex_match ('('))
390 if (indep_var->type == NUMERIC)
392 gp.v.g_value[0].f = 1;
393 gp.v.g_value[1].f = 2;
395 gp.criterion = CMP_EQ;
403 msg (SE, _("When applying GROUPS to a string variable, two "
404 "values must be specified."));
409 if (!parse_value (&gp.v.g_value[0], indep_var->type))
415 if (indep_var->type != NUMERIC)
418 msg (SE, _("When applying GROUPS to a string variable, two "
419 "values must be specified."));
422 gp.criterion = CMP_LE;
423 gp.v.critical_value = gp.v.g_value[0].f;
429 if (!parse_value (&gp.v.g_value[1], indep_var->type))
433 if (!lex_force_match (')'))
436 if ( n_group_values == 2 )
437 gp.criterion = CMP_EQ ;
439 gp.criterion = CMP_LE ;
447 tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
449 struct variable **vars;
454 int n_after_WITH = -1;
455 int paired ; /* Was the PAIRED keyword given ? */
459 if ((token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
462 msg(SE,_("`%s' is not a variable name"),tokid);
467 if (!parse_variables (default_dict, &vars, &n_vars,
468 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
476 if (lex_match (T_WITH))
478 n_before_WITH = n_vars;
479 if (!parse_variables (default_dict, &vars, &n_vars,
480 PV_DUPLICATE | PV_APPEND
481 | PV_NUMERIC | PV_NO_SCRATCH))
486 n_after_WITH = n_vars - n_before_WITH;
489 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
491 /* Determine the number of pairs needed */
494 if (n_before_WITH != n_after_WITH)
497 msg (SE, _("PAIRED was specified but the number of variables "
498 "preceding WITH (%d) did not match the number "
500 n_before_WITH, n_after_WITH );
503 n_pairs_local=n_before_WITH;
505 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
507 n_pairs_local=n_before_WITH * n_after_WITH ;
509 else /* Neither WITH nor PAIRED keyword given */
514 msg (SE, _("At least two variables must be specified "
519 /* how many ways can you pick 2 from n_vars ? */
520 n_pairs_local = n_vars * (n_vars -1 ) /2 ;
524 /* Allocate storage for the pairs */
525 pairs = xrealloc(pairs, sizeof(struct pair) * (n_pairs + n_pairs_local) );
527 /* Populate the pairs with the appropriate variables */
532 assert(n_pairs_local == n_vars/2);
533 for (i = 0; i < n_pairs_local ; ++i)
535 pairs[i].v[n_pairs+0] = vars[i];
536 pairs[i].v[n_pairs+1] = vars[i+n_pairs_local];
539 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
544 for(i=0 ; i < n_before_WITH ; ++i )
546 for(j=0 ; j < n_after_WITH ; ++j)
548 pairs[p].v[0] = vars[i];
549 pairs[p].v[1] = vars[j+n_before_WITH];
554 else /* Neither WITH nor PAIRED given */
559 for(i=0 ; i < n_vars ; ++i )
561 for(j=i+1 ; j < n_vars ; ++j)
563 pairs[p].v[0] = vars[i];
564 pairs[p].v[1] = vars[j];
570 n_pairs+=n_pairs_local;
576 /* Parses the current token (numeric or string, depending on type)
577 value v and returns success. */
579 parse_value (union value * v, int type )
583 if (!lex_force_num ())
589 if (!lex_force_string ())
591 strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
600 /* Implementation of the SSBOX object */
602 void ssbox_base_init(struct ssbox *this, int cols,int rows);
604 void ssbox_base_finalize(struct ssbox *ssb);
606 void ssbox_one_sample_init(struct ssbox *this,
607 struct cmd_t_test *cmd );
609 void ssbox_independent_samples_init(struct ssbox *this,
610 struct cmd_t_test *cmd);
612 void ssbox_paired_init(struct ssbox *this,
613 struct cmd_t_test *cmd);
616 /* Factory to create an ssbox */
618 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
623 ssbox_one_sample_init(ssb,cmd);
626 ssbox_independent_samples_init(ssb,cmd);
629 ssbox_paired_init(ssb,cmd);
638 /* Despatcher for the populate method */
640 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
642 ssb->populate(ssb,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 struct cmd_t_test *cmd);
680 /* Initialize the one_sample ssbox */
682 ssbox_one_sample_init(struct ssbox *this,
683 struct cmd_t_test *cmd )
686 const int vsize=cmd->n_variables+1;
688 this->populate = ssbox_one_sample_populate;
690 ssbox_base_init(this, hsize,vsize);
691 tab_title (this->t, 0, _("One-Sample Statistics"));
692 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
693 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
694 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
695 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
696 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
699 void ssbox_independent_samples_populate(struct ssbox *ssb,
700 struct cmd_t_test *cmd);
702 /* Initialize the independent samples ssbox */
704 ssbox_independent_samples_init(struct ssbox *this,
705 struct cmd_t_test *cmd)
708 int vsize = cmd->n_variables*2 +1;
710 this->populate = ssbox_independent_samples_populate;
712 ssbox_base_init(this, hsize,vsize);
713 tab_title (this->t, 0, _("Group Statistics"));
714 tab_vline(this->t,0,1,0,vsize - 1);
715 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
716 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
717 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
718 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
719 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
723 /* Populate the ssbox for independent samples */
725 ssbox_independent_samples_populate(struct ssbox *ssb,
726 struct cmd_t_test *cmd)
732 double indep_value[2];
734 char prefix[2][3]={"",""};
736 if ( indep_var->type == NUMERIC )
738 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
739 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
743 val_lab0 = gp.v.g_value[0].s;
744 val_lab1 = gp.v.g_value[1].s;
747 if (gp.criterion == CMP_LE )
749 strcpy(prefix[0],"< ");
750 strcpy(prefix[1],">=");
751 indep_value[0] = gp.v.critical_value;
752 indep_value[1] = gp.v.critical_value;
756 indep_value[0] = gp.v.g_value[0].f;
757 indep_value[1] = gp.v.g_value[1].f;
762 for (i=0; i < cmd->n_variables; ++i)
764 struct variable *var = cmd->v_variables[i];
765 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
768 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
771 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
772 "%s%s", prefix[0], val_lab0);
774 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
775 "%s%g", prefix[0], indep_value[0]);
779 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
780 "%s%s", prefix[1], val_lab1);
782 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
783 "%s%g", prefix[1], indep_value[1]);
786 /* Fill in the group statistics */
787 for ( count = 0 ; count < 2 ; ++count )
789 union value search_val;
791 struct group_statistics *gs;
793 if ( gp.criterion == CMP_LE )
797 /* less than ( < ) case */
798 search_val.f = gp.v.critical_value - 1.0;
803 search_val.f = gp.v.critical_value + 1.0;
808 search_val = gp.v.g_value[count];
811 gs = hsh_find(grp_hash, (void *) &search_val);
814 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
815 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
816 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
817 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
823 void ssbox_paired_populate(struct ssbox *ssb,
824 struct cmd_t_test *cmd);
826 /* Initialize the paired values ssbox */
828 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
832 int vsize = n_pairs*2+1;
834 this->populate = ssbox_paired_populate;
836 ssbox_base_init(this, hsize,vsize);
837 tab_title (this->t, 0, _("Paired Sample Statistics"));
838 tab_vline(this->t,TAL_0,1,0,vsize-1);
839 tab_vline(this->t,TAL_2,2,0,vsize-1);
840 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
841 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
842 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
843 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
847 /* Populate the ssbox for paired values */
849 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
855 for (i=0; i < n_pairs; ++i)
859 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
861 for (j=0 ; j < 2 ; ++j)
863 struct group_statistics *gs;
865 gs = &group_proc_get (pairs[i].v[j])->ugs;
869 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
872 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
873 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
874 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
875 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
881 /* Populate the one sample ssbox */
883 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
889 for (i=0; i < cmd->n_variables; ++i)
891 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
893 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
894 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
895 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
896 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
897 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
904 /* Implementation of the Test Results box struct */
906 void trbox_base_init(struct trbox *self,int n_vars, int cols);
907 void trbox_base_finalize(struct trbox *trb);
909 void trbox_independent_samples_init(struct trbox *trb,
910 struct cmd_t_test *cmd );
912 void trbox_independent_samples_populate(struct trbox *trb,
913 struct cmd_t_test *cmd);
915 void trbox_one_sample_init(struct trbox *self,
916 struct cmd_t_test *cmd );
918 void trbox_one_sample_populate(struct trbox *trb,
919 struct cmd_t_test *cmd);
921 void trbox_paired_init(struct trbox *self,
922 struct cmd_t_test *cmd );
924 void trbox_paired_populate(struct trbox *trb,
925 struct cmd_t_test *cmd);
929 /* Create a trbox according to mode*/
931 trbox_create(struct trbox *trb,
932 struct cmd_t_test *cmd, int mode)
937 trbox_one_sample_init(trb,cmd);
940 trbox_independent_samples_init(trb,cmd);
943 trbox_paired_init(trb,cmd);
950 /* Populate a trbox according to cmd */
952 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
954 trb->populate(trb,cmd);
957 /* Submit and destroy a trbox */
959 trbox_finalize(struct trbox *trb)
964 /* Initialize the independent samples trbox */
966 trbox_independent_samples_init(struct trbox *self,
967 struct cmd_t_test *cmd UNUSED)
970 const int vsize=cmd->n_variables*2+3;
973 self->populate = trbox_independent_samples_populate;
975 trbox_base_init(self,cmd->n_variables*2,hsize);
976 tab_title(self->t,0,_("Independent Samples Test"));
977 tab_hline(self->t,TAL_1,2,hsize-1,1);
978 tab_vline(self->t,TAL_2,2,0,vsize-1);
979 tab_vline(self->t,TAL_1,4,0,vsize-1);
980 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
981 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
982 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
983 tab_joint_text(self->t, 2, 0, 3, 0,
984 TAB_CENTER,_("Levene's Test for Equality of Variances"));
985 tab_joint_text(self->t, 4,0,hsize-1,0,
986 TAB_CENTER,_("t-test for Equality of Means"));
988 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
989 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
990 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
991 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
992 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
993 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
994 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
995 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
996 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
998 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
999 _("%g%% Confidence Interval of the Difference"),
1000 cmd->criteria*100.0);
1004 /* Populate the independent samples trbox */
1006 trbox_independent_samples_populate(struct trbox *self,
1007 struct cmd_t_test *cmd )
1012 for (i=0; i < cmd->n_variables; ++i)
1021 double pooled_variance;
1022 double std_err_diff;
1025 struct variable *var = cmd->v_variables[i];
1026 struct group_proc *grp_data = group_proc_get (var);
1028 struct hsh_table *grp_hash = grp_data->group_hash;
1030 struct group_statistics *gs0 ;
1031 struct group_statistics *gs1 ;
1033 union value 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[0];
1040 gs0 = hsh_find(grp_hash, (void *) &search_val);
1043 if ( gp.criterion == CMP_LE )
1044 search_val.f = gp.v.critical_value + 1.0;
1046 search_val = gp.v.g_value[1];
1048 gs1 = hsh_find(grp_hash, (void *) &search_val);
1052 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1054 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1057 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1059 /* Now work out the significance of the Levene test */
1060 df1 = 1; df2 = grp_data->ugs.n - 2;
1061 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1063 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1065 df = gs0->n + gs1->n - 2.0 ;
1066 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1068 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1070 (gs1->n )*pow2(gs1->s_std_dev)
1073 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1074 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1076 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1078 p = gsl_cdf_tdist_P(t, df);
1079 q = gsl_cdf_tdist_Q(t, df);
1081 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1083 mean_diff = gs0->mean - gs1->mean;
1084 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1087 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1088 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1091 /* Now work out the confidence interval */
1092 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1094 t = gsl_cdf_tdist_Qinv(q,df);
1095 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1096 mean_diff - t * std_err_diff, 8, 3);
1098 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1099 mean_diff + t * std_err_diff, 8, 3);
1104 /* Now for the \sigma_1 != \sigma_2 case */
1105 tab_text (self->t, 1, i*2+3+1,
1106 TAB_LEFT, _("Equal variances not assumed"));
1109 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1110 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1112 t = mean_diff / sqrt(se2) ;
1113 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1116 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1120 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1124 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1126 p = gsl_cdf_tdist_P(t, df);
1127 q = gsl_cdf_tdist_Q(t, df);
1129 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1131 /* Now work out the confidence interval */
1132 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1134 t = gsl_cdf_tdist_Qinv(q, df);
1136 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1139 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1142 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1143 mean_diff - t * std_err_diff, 8, 3);
1145 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1146 mean_diff + t * std_err_diff, 8, 3);
1152 /* Initialize the paired samples trbox */
1154 trbox_paired_init(struct trbox *self,
1155 struct cmd_t_test *cmd UNUSED)
1159 const int vsize=n_pairs+3;
1161 self->populate = trbox_paired_populate;
1163 trbox_base_init(self,n_pairs,hsize);
1164 tab_title (self->t, 0, _("Paired Samples Test"));
1165 tab_hline(self->t,TAL_1,2,6,1);
1166 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1167 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1168 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1169 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1170 tab_hline(self->t,TAL_1,5,6, 2);
1171 tab_vline(self->t,TAL_0,6,0,1);
1173 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1174 _("%g%% Confidence Interval of the Difference"),
1175 cmd->criteria*100.0);
1177 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1178 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1179 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1180 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1181 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1182 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1183 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1184 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1187 /* Populate the paired samples trbox */
1189 trbox_paired_populate(struct trbox *trb,
1190 struct cmd_t_test *cmd UNUSED)
1194 for (i=0; i < n_pairs; ++i)
1199 double n = pairs[i].n;
1203 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1205 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1206 pairs[i].v[0]->name, pairs[i].v[1]->name);
1208 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1210 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1213 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1214 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1216 /* Now work out the confidence interval */
1217 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1219 t = gsl_cdf_tdist_Qinv(q, df);
1221 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1222 pairs[i].mean_diff - t * se_mean , 8, 4);
1224 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1225 pairs[i].mean_diff + t * se_mean , 8, 4);
1227 t = (pairs[i].mean[0] - pairs[i].mean[1])
1229 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1230 2 * pairs[i].correlation *
1231 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1235 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1237 /* Degrees of freedom */
1238 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1240 p = gsl_cdf_tdist_P(t,df);
1241 q = gsl_cdf_tdist_P(t,df);
1243 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1248 /* Initialize the one sample trbox */
1250 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1253 const int vsize=cmd->n_variables+3;
1255 self->populate = trbox_one_sample_populate;
1257 trbox_base_init(self, cmd->n_variables,hsize);
1258 tab_title (self->t, 0, _("One-Sample Test"));
1259 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1260 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1262 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1263 _("Test Value = %f"), cmd->n_testval[0]);
1265 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1268 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1269 _("%g%% Confidence Interval of the Difference"),
1270 cmd->criteria*100.0);
1272 tab_vline(self->t,TAL_0,6,1,1);
1273 tab_hline(self->t,TAL_1,5,6,2);
1274 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1275 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1276 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1277 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1278 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1279 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1284 /* Populate the one sample trbox */
1286 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1292 for (i=0; i < cmd->n_variables; ++i)
1297 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1300 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1302 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1304 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1306 /* degrees of freedom */
1309 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1311 p = gsl_cdf_tdist_P(t, df);
1312 q = gsl_cdf_tdist_Q(t, df);
1314 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1316 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1318 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1321 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1322 t = gsl_cdf_tdist_Qinv(q, df);
1324 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1325 gs->mean_diff - t * gs->se_mean, 8,4);
1327 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1328 gs->mean_diff + t * gs->se_mean, 8,4);
1332 /* Base initializer for the generalized trbox */
1334 trbox_base_init(struct trbox *self, int data_rows, int cols)
1336 const int rows = 3 + data_rows;
1338 self->finalize = trbox_base_finalize;
1339 self->t = tab_create (cols, rows, 0);
1340 tab_headers (self->t,0,0,3,0);
1341 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1342 tab_hline(self->t, TAL_2,0,cols-1,3);
1343 tab_dim (self->t, tab_natural_dimensions);
1347 /* Base finalizer for the trbox */
1349 trbox_base_finalize(struct trbox *trb)
1355 /* Create , populate and submit the Paired Samples Correlation box */
1359 const int rows=1+n_pairs;
1363 struct tab_table *table;
1365 table = tab_create (cols,rows,0);
1367 tab_columns (table, SOM_COL_DOWN, 1);
1368 tab_headers (table,0,0,1,0);
1369 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1370 tab_hline(table, TAL_2, 0, cols - 1, 1);
1371 tab_vline(table, TAL_2, 2, 0, rows - 1);
1372 tab_dim(table, tab_natural_dimensions);
1373 tab_title(table, 0, _("Paired Samples Correlations"));
1375 /* column headings */
1376 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1377 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1378 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1380 for (i=0; i < n_pairs; ++i)
1384 double df = pairs[i].n -2;
1386 double correlation_t =
1387 pairs[i].correlation * sqrt(df) /
1388 sqrt(1 - pow2(pairs[i].correlation));
1392 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1395 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1396 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1400 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1401 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1403 p = gsl_cdf_tdist_P(correlation_t, df);
1404 q = gsl_cdf_tdist_Q(correlation_t, df);
1406 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1415 /* Calculation Implementation */
1417 /* Per case calculations common to all variants of the T test */
1419 common_calc (const struct ccase *c, void *_cmd)
1422 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1424 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1427 /* Skip the entire case if /MISSING=LISTWISE is set */
1428 if ( cmd->miss == TTS_LISTWISE )
1430 for(i=0; i< cmd->n_variables ; ++i)
1432 struct variable *v = cmd->v_variables[i];
1433 const union value *val = case_data (c, v->fv);
1435 if (value_is_missing(val,v) )
1442 /* Listwise has to be implicit if the independent variable is missing ?? */
1443 if ( cmd->sbc_groups )
1445 const union value *gv = case_data (c, indep_var->fv);
1446 if ( value_is_missing(gv,indep_var) )
1453 for(i=0; i< cmd->n_variables ; ++i)
1455 struct group_statistics *gs;
1456 struct variable *v = cmd->v_variables[i];
1457 const union value *val = case_data (c, v->fv);
1459 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1461 if (! value_is_missing(val,v) )
1464 gs->sum+=weight * val->f;
1465 gs->ssq+=weight * val->f * val->f;
1471 /* Pre calculations common to all variants of the T test */
1473 common_precalc ( struct cmd_t_test *cmd )
1477 for(i=0; i< cmd->n_variables ; ++i)
1479 struct group_statistics *gs;
1480 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1489 /* Post calculations common to all variants of the T test */
1491 common_postcalc ( struct cmd_t_test *cmd )
1496 for(i=0; i< cmd->n_variables ; ++i)
1498 struct group_statistics *gs;
1499 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1501 gs->mean=gs->sum / gs->n;
1502 gs->s_std_dev= sqrt(
1503 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1508 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1511 gs->se_mean = gs->std_dev / sqrt(gs->n);
1512 gs->mean_diff= gs->sum_diff / gs->n;
1516 /* Per case calculations for one sample t test */
1518 one_sample_calc (const struct ccase *c, void *cmd_)
1521 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1524 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1526 /* Skip the entire case if /MISSING=LISTWISE is set */
1527 if ( cmd->miss == TTS_LISTWISE )
1529 for(i=0; i< cmd->n_variables ; ++i)
1531 struct variable *v = cmd->v_variables[i];
1532 const union value *val = case_data (c, v->fv);
1534 if (value_is_missing(val,v) )
1541 for(i=0; i< cmd->n_variables ; ++i)
1543 struct group_statistics *gs;
1544 struct variable *v = cmd->v_variables[i];
1545 const union value *val = case_data (c, v->fv);
1547 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1549 if ( ! value_is_missing(val,v))
1550 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1556 /* Pre calculations for one sample t test */
1558 one_sample_precalc ( struct cmd_t_test *cmd )
1562 for(i=0; i< cmd->n_variables ; ++i)
1564 struct group_statistics *gs;
1565 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1571 /* Post calculations for one sample t test */
1573 one_sample_postcalc (struct cmd_t_test *cmd)
1577 for(i=0; i< cmd->n_variables ; ++i)
1579 struct group_statistics *gs;
1580 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1582 gs->mean_diff = gs->sum_diff / gs->n ;
1589 compare_var_name (const void *a_, const void *b_, void *v_ UNUSED)
1591 const struct variable *a = a_;
1592 const struct variable *b = b_;
1594 return strcmp(a->name,b->name);
1598 hash_var_name (const void *a_, void *v_ UNUSED)
1600 const struct variable *a = a_;
1602 return hsh_hash_bytes (a->name, strlen(a->name));
1608 paired_precalc (struct cmd_t_test *cmd UNUSED)
1612 for(i=0; i < n_pairs ; ++i )
1615 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1616 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1617 pairs[i].sum_of_prod = 0;
1618 pairs[i].correlation = 0;
1619 pairs[i].sum_of_diffs = 0;
1620 pairs[i].ssq_diffs = 0;
1627 paired_calc (const struct ccase *c, void *cmd_)
1631 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1633 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1635 /* Skip the entire case if /MISSING=LISTWISE is set ,
1636 AND one member of a pair is missing */
1637 if ( cmd->miss == TTS_LISTWISE )
1639 for(i=0; i < n_pairs ; ++i )
1641 struct variable *v0 = pairs[i].v[0];
1642 struct variable *v1 = pairs[i].v[1];
1644 const union value *val0 = case_data (c, v0->fv);
1645 const union value *val1 = case_data (c, v1->fv);
1647 if ( value_is_missing(val0,v0) ||
1648 value_is_missing(val1,v1) )
1655 for(i=0; i < n_pairs ; ++i )
1657 struct variable *v0 = pairs[i].v[0];
1658 struct variable *v1 = pairs[i].v[1];
1660 const union value *val0 = case_data (c, v0->fv);
1661 const union value *val1 = case_data (c, v1->fv);
1663 if ( ( !value_is_missing(val0,v0) && !value_is_missing(val1,v1) ) )
1665 pairs[i].n += weight;
1666 pairs[i].sum[0] += weight * val0->f;
1667 pairs[i].sum[1] += weight * val1->f;
1669 pairs[i].ssq[0] += weight * pow2(val0->f);
1670 pairs[i].ssq[1] += weight * pow2(val1->f);
1672 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1674 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1675 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1683 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1687 for(i=0; i < n_pairs ; ++i )
1690 const double n = pairs[i].n;
1692 for (j=0; j < 2 ; ++j)
1694 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1695 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1696 pow2(pairs[i].mean[j]))
1699 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1700 pow2(pairs[i].mean[j]))
1704 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1705 pairs[i].mean[0] * pairs[i].mean[1] ;
1706 /* correlation now actually contains the covariance */
1708 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1709 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1711 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1713 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1714 ( pairs[i].ssq_diffs / n )
1716 pow2(pairs[i].mean_diff )
1722 group_precalc (struct cmd_t_test *cmd )
1727 for(i=0; i< cmd->n_variables ; ++i)
1729 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1731 /* There's always 2 groups for a T - TEST */
1734 gp.indep_width = indep_var->width;
1736 ttpr->group_hash = hsh_create(2,
1737 (hsh_compare_func *) compare_group_binary,
1738 (hsh_hash_func *) hash_group_binary,
1739 (hsh_free_func *) free_group,
1742 for (j=0 ; j < 2 ; ++j)
1745 struct group_statistics *gs = (struct group_statistics *)
1746 xmalloc (sizeof(struct group_statistics));
1752 if ( gp.criterion == CMP_EQ )
1754 gs->id = gp.v.g_value[j];
1759 gs->id.f = gp.v.critical_value - 1.0 ;
1761 gs->id.f = gp.v.critical_value + 1.0 ;
1764 hsh_insert ( ttpr->group_hash, (void *) gs );
1772 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1776 const union value *gv = case_data (c, indep_var->fv);
1778 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1780 if ( value_is_missing(gv,indep_var) )
1785 if ( cmd->miss == TTS_LISTWISE )
1787 for(i=0; i< cmd->n_variables ; ++i)
1789 struct variable *v = cmd->v_variables[i];
1790 const union value *val = case_data (c, v->fv);
1792 if (value_is_missing(val,v) )
1799 gv = case_data (c, indep_var->fv);
1801 for(i=0; i< cmd->n_variables ; ++i)
1803 struct variable *var = cmd->v_variables[i];
1804 const union value *val = case_data (c, var->fv);
1805 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1806 struct group_statistics *gs;
1808 gs = hsh_find(grp_hash, (void *) gv);
1810 /* If the independent variable doesn't match either of the values
1811 for this case then move on to the next case */
1815 if ( !value_is_missing(val,var) )
1818 gs->sum+=weight * val->f;
1819 gs->ssq+=weight * pow2(val->f);
1828 group_postcalc ( struct cmd_t_test *cmd )
1832 for(i=0; i< cmd->n_variables ; ++i)
1834 struct variable *var = cmd->v_variables[i];
1835 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1836 struct hsh_iterator g;
1837 struct group_statistics *gs;
1840 for (gs = hsh_first (grp_hash,&g);
1842 gs = hsh_next(grp_hash,&g))
1844 gs->mean = gs->sum / gs->n;
1846 gs->s_std_dev= sqrt(
1847 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1852 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1855 gs->se_mean = gs->std_dev / sqrt(gs->n);
1865 calculate(const struct casefile *cf, void *cmd_)
1867 struct ssbox stat_summary_box;
1868 struct trbox test_results_box;
1870 struct casereader *r;
1873 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1875 common_precalc(cmd);
1876 for(r = casefile_get_reader (cf);
1877 casereader_read (r, &c) ;
1880 common_calc(&c,cmd);
1882 casereader_destroy (r);
1883 common_postcalc(cmd);
1888 one_sample_precalc(cmd);
1889 for(r = casefile_get_reader (cf);
1890 casereader_read (r, &c) ;
1893 one_sample_calc(&c,cmd);
1895 casereader_destroy (r);
1896 one_sample_postcalc(cmd);
1900 paired_precalc(cmd);
1901 for(r = casefile_get_reader (cf);
1902 casereader_read (r, &c) ;
1905 paired_calc(&c,cmd);
1907 casereader_destroy (r);
1908 paired_postcalc(cmd);
1914 for(r = casefile_get_reader (cf);
1915 casereader_read (r, &c) ;
1920 casereader_destroy (r);
1921 group_postcalc(cmd);
1923 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1924 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1929 ssbox_create(&stat_summary_box,cmd,mode);
1930 ssbox_populate(&stat_summary_box,cmd);
1931 ssbox_finalize(&stat_summary_box);
1933 if ( mode == T_PAIRED)
1936 trbox_create(&test_results_box,cmd,mode);
1937 trbox_populate(&test_results_box,cmd);
1938 trbox_finalize(&test_results_box);
1943 /* Return -1 if the id of a is less than b; +1 if greater than and
1946 compare_group_binary(const struct group_statistics *a,
1947 const struct group_statistics *b,
1948 const struct group_properties *p)
1954 if ( p->criterion == CMP_LE )
1956 assert(p->indep_width == 0 ) ;
1957 flag_a = ( a->id.f < p->v.critical_value ) ;
1958 flag_b = ( b->id.f < p->v.critical_value ) ;
1962 flag_a = ( a->id.f == p->v.critical_value ) ;
1963 flag_b = ( b->id.f == p->v.critical_value ) ;
1966 if ( flag_a == flag_b)
1969 return ( flag_a < flag_b);
1972 /* This is a degenerate case of a hash, since it can only return three possible
1973 values. It's really a comparison, being used as a hash function */
1976 hash_group_binary(const struct group_statistics *g,
1977 const struct group_properties *p)
1981 if ( p->criterion == CMP_LE )
1983 /* Not meaningfull to do a less than compare for alpha values ? */
1984 assert(p->indep_width == 0 ) ;
1985 flag = ( g->id.f < p->v.critical_value ) ;
1987 else if ( p->criterion == CMP_EQ)
1989 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width ))
1991 else if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width ))