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
23 #include <gsl/gsl_cdf.h>
24 #include <libpspp/message.h>
28 #include <libpspp/alloc.h>
29 #include <data/case.h>
30 #include <data/casefile.h>
31 #include <language/command.h>
32 #include <libpspp/compiler.h>
33 #include <data/dictionary.h>
34 #include <libpspp/message.h>
35 #include <math/group-proc.h>
36 #include <libpspp/hash.h>
37 #include <math/levene.h>
38 #include <language/lexer/lexer.h>
39 #include <libpspp/magic.h>
40 #include <libpspp/misc.h>
42 #include <output/manager.h>
43 #include <libpspp/str.h>
44 #include <output/table.h>
45 #include <data/value-labels.h>
46 #include <data/variable.h>
47 #include <procedure.h>
50 #define _(msgid) gettext (msgid)
58 variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
60 +missing=miss:!analysis/listwise,
61 incl:include/!exclude;
62 format=fmt:!labels/nolabels;
63 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
71 /* Function to use for testing for missing values */
72 static is_missing_func *value_is_missing;
74 /* Variable for the GROUPS subcommand, if given. */
75 static struct variable *indep_var;
83 struct group_properties
85 /* The comparison criterion */
86 enum comparison criterion;
88 /* The width of the independent variable */
92 /* The value of the independent variable at which groups are determined to
93 belong to one group or the other */
94 double critical_value;
97 /* The values of the independent variable for each group */
98 union value g_value[2];
104 static struct group_properties gp ;
108 /* PAIRS: Number of pairs to be compared ; each pair. */
109 static int n_pairs = 0 ;
112 /* The variables comprising the pair */
113 struct variable *v[2];
115 /* The number of valid variable pairs */
118 /* The sum of the members */
121 /* sum of squares of the members */
124 /* Std deviation of the members */
128 /* Sample Std deviation of the members */
131 /* The means of the members */
134 /* The correlation coefficient between the variables */
137 /* The sum of the differences */
140 /* The sum of the products */
143 /* The mean of the differences */
146 /* The sum of the squares of the differences */
149 /* The std deviation of the differences */
153 static struct pair *pairs=0;
155 static int parse_value (union value * v, int type) ;
157 /* Structures and Functions for the Statistics Summary Box */
159 typedef void populate_ssbox_func(struct ssbox *ssb,
160 struct cmd_t_test *cmd);
161 typedef void finalize_ssbox_func(struct ssbox *ssb);
167 populate_ssbox_func *populate;
168 finalize_ssbox_func *finalize;
173 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
175 /* Populate a ssbox according to cmd */
176 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
178 /* Submit and destroy a ssbox */
179 void ssbox_finalize(struct ssbox *ssb);
181 /* A function to create, populate and submit the Paired Samples Correlation
186 /* Structures and Functions for the Test Results Box */
189 typedef void populate_trbox_func(struct trbox *trb,
190 struct cmd_t_test *cmd);
191 typedef void finalize_trbox_func(struct trbox *trb);
195 populate_trbox_func *populate;
196 finalize_trbox_func *finalize;
200 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
202 /* Populate a ssbox according to cmd */
203 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
205 /* Submit and destroy a ssbox */
206 void trbox_finalize(struct trbox *trb);
208 /* Which mode was T-TEST invoked */
216 static int common_calc (const struct ccase *, void *);
217 static void common_precalc (struct cmd_t_test *);
218 static void common_postcalc (struct cmd_t_test *);
220 static int one_sample_calc (const struct ccase *, void *);
221 static void one_sample_precalc (struct cmd_t_test *);
222 static void one_sample_postcalc (struct cmd_t_test *);
224 static int paired_calc (const struct ccase *, void *);
225 static void paired_precalc (struct cmd_t_test *);
226 static void paired_postcalc (struct cmd_t_test *);
228 static void group_precalc (struct cmd_t_test *);
229 static int group_calc (const struct ccase *, struct cmd_t_test *);
230 static void group_postcalc (struct cmd_t_test *);
233 static bool calculate(const struct casefile *cf, void *_mode);
237 static struct cmd_t_test cmd;
239 static int bad_weight_warn;
242 static int compare_group_binary(const struct group_statistics *a,
243 const struct group_statistics *b,
244 const struct group_properties *p);
247 static unsigned hash_group_binary(const struct group_statistics *g,
248 const struct group_properties *p);
257 if ( !parse_t_test(&cmd) )
260 if (! cmd.sbc_criteria)
265 if (cmd.sbc_testval) ++m;
266 if (cmd.sbc_groups) ++m;
267 if (cmd.sbc_pairs) ++m;
272 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
281 else if (cmd.sbc_groups)
286 if ( mode == T_PAIRED)
288 if (cmd.sbc_variables)
290 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
296 /* Iterate through the pairs and put each variable that is a
297 member of a pair into cmd.v_variables */
300 struct hsh_iterator hi;
301 struct hsh_table *hash;
304 hash = hsh_create (n_pairs, compare_var_names, hash_var_name, 0, 0);
306 for (i=0; i < n_pairs; ++i)
308 hsh_insert(hash,pairs[i].v[0]);
309 hsh_insert(hash,pairs[i].v[1]);
312 assert(cmd.n_variables == 0);
313 cmd.n_variables = hsh_count(hash);
315 cmd.v_variables = xnrealloc (cmd.v_variables, cmd.n_variables,
316 sizeof *cmd.v_variables);
317 /* Iterate through the hash */
318 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
320 v=hsh_next(hash,&hi) )
321 cmd.v_variables[i++]=v;
326 else if ( !cmd.sbc_variables)
328 msg(SE, _("One or more VARIABLES must be specified."));
334 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
335 if (cmd.incl == TTS_INCLUDE )
336 value_is_missing = mv_is_value_system_missing;
338 value_is_missing = mv_is_value_missing;
342 ok = multipass_procedure_with_splits (calculate, &cmd);
348 if ( mode == T_IND_SAMPLES)
351 /* Destroy any group statistics we created */
352 for (v = 0 ; v < cmd.n_variables ; ++v )
354 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
355 hsh_destroy (grpp->group_hash);
360 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
364 tts_custom_groups (struct cmd_t_test *cmd UNUSED)
366 int n_group_values=0;
370 indep_var = parse_variable ();
373 lex_error ("expecting variable name in GROUPS subcommand");
377 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
379 msg (SE, _("Long string variable %s is not valid here."),
384 if (!lex_match ('('))
386 if (indep_var->type == NUMERIC)
388 gp.v.g_value[0].f = 1;
389 gp.v.g_value[1].f = 2;
391 gp.criterion = CMP_EQ;
399 msg (SE, _("When applying GROUPS to a string variable, two "
400 "values must be specified."));
405 if (!parse_value (&gp.v.g_value[0], indep_var->type))
411 if (indep_var->type != NUMERIC)
414 msg (SE, _("When applying GROUPS to a string variable, two "
415 "values must be specified."));
418 gp.criterion = CMP_LE;
419 gp.v.critical_value = gp.v.g_value[0].f;
425 if (!parse_value (&gp.v.g_value[1], indep_var->type))
429 if (!lex_force_match (')'))
432 if ( n_group_values == 2 )
433 gp.criterion = CMP_EQ ;
435 gp.criterion = CMP_LE ;
443 tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
445 struct variable **vars;
447 size_t n_pairs_local;
449 size_t n_before_WITH;
450 size_t n_after_WITH = SIZE_MAX;
451 int paired ; /* Was the PAIRED keyword given ? */
456 if (!parse_variables (default_dict, &vars, &n_vars,
457 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
465 if (lex_match (T_WITH))
467 n_before_WITH = n_vars;
468 if (!parse_variables (default_dict, &vars, &n_vars,
469 PV_DUPLICATE | PV_APPEND
470 | PV_NUMERIC | PV_NO_SCRATCH))
475 n_after_WITH = n_vars - n_before_WITH;
478 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
480 /* Determine the number of pairs needed */
483 if (n_before_WITH != n_after_WITH)
486 msg (SE, _("PAIRED was specified but the number of variables "
487 "preceding WITH (%d) did not match the number "
489 n_before_WITH, n_after_WITH );
492 n_pairs_local = n_before_WITH;
494 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
496 n_pairs_local = n_before_WITH * n_after_WITH ;
498 else /* Neither WITH nor PAIRED keyword given */
503 msg (SE, _("At least two variables must be specified "
508 /* how many ways can you pick 2 from n_vars ? */
509 n_pairs_local = n_vars * (n_vars - 1) / 2;
513 /* Allocate storage for the pairs */
514 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
516 /* Populate the pairs with the appropriate variables */
521 assert(n_pairs_local == n_vars / 2);
522 for (i = 0; i < n_pairs_local; ++i)
524 pairs[i].v[n_pairs] = vars[i];
525 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
528 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
533 for(i=0 ; i < n_before_WITH ; ++i )
535 for(j=0 ; j < n_after_WITH ; ++j)
537 pairs[p].v[0] = vars[i];
538 pairs[p].v[1] = vars[j+n_before_WITH];
543 else /* Neither WITH nor PAIRED given */
548 for(i=0 ; i < n_vars ; ++i )
550 for(j=i+1 ; j < n_vars ; ++j)
552 pairs[p].v[0] = vars[i];
553 pairs[p].v[1] = vars[j];
559 n_pairs+=n_pairs_local;
565 /* Parses the current token (numeric or string, depending on type)
566 value v and returns success. */
568 parse_value (union value * v, int type )
572 if (!lex_force_num ())
578 if (!lex_force_string ())
580 strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
589 /* Implementation of the SSBOX object */
591 void ssbox_base_init(struct ssbox *this, int cols,int rows);
593 void ssbox_base_finalize(struct ssbox *ssb);
595 void ssbox_one_sample_init(struct ssbox *this,
596 struct cmd_t_test *cmd );
598 void ssbox_independent_samples_init(struct ssbox *this,
599 struct cmd_t_test *cmd);
601 void ssbox_paired_init(struct ssbox *this,
602 struct cmd_t_test *cmd);
605 /* Factory to create an ssbox */
607 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
612 ssbox_one_sample_init(ssb,cmd);
615 ssbox_independent_samples_init(ssb,cmd);
618 ssbox_paired_init(ssb,cmd);
627 /* Despatcher for the populate method */
629 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
631 ssb->populate(ssb,cmd);
635 /* Despatcher for finalize */
637 ssbox_finalize(struct ssbox *ssb)
643 /* Submit the box and clear up */
645 ssbox_base_finalize(struct ssbox *ssb)
652 /* Initialize a ssbox struct */
654 ssbox_base_init(struct ssbox *this, int cols,int rows)
656 this->finalize = ssbox_base_finalize;
657 this->t = tab_create (cols, rows, 0);
659 tab_columns (this->t, SOM_COL_DOWN, 1);
660 tab_headers (this->t,0,0,1,0);
661 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
662 tab_hline(this->t, TAL_2,0,cols-1,1);
663 tab_dim (this->t, tab_natural_dimensions);
666 void ssbox_one_sample_populate(struct ssbox *ssb,
667 struct cmd_t_test *cmd);
669 /* Initialize the one_sample ssbox */
671 ssbox_one_sample_init(struct ssbox *this,
672 struct cmd_t_test *cmd )
675 const int vsize=cmd->n_variables+1;
677 this->populate = ssbox_one_sample_populate;
679 ssbox_base_init(this, hsize,vsize);
680 tab_title (this->t, 0, _("One-Sample Statistics"));
681 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
682 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
683 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
684 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
685 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
688 void ssbox_independent_samples_populate(struct ssbox *ssb,
689 struct cmd_t_test *cmd);
691 /* Initialize the independent samples ssbox */
693 ssbox_independent_samples_init(struct ssbox *this,
694 struct cmd_t_test *cmd)
697 int vsize = cmd->n_variables*2 +1;
699 this->populate = ssbox_independent_samples_populate;
701 ssbox_base_init(this, hsize,vsize);
702 tab_title (this->t, 0, _("Group Statistics"));
703 tab_vline(this->t,0,1,0,vsize - 1);
704 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
705 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
706 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
707 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
708 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
712 /* Populate the ssbox for independent samples */
714 ssbox_independent_samples_populate(struct ssbox *ssb,
715 struct cmd_t_test *cmd)
721 double indep_value[2];
723 char prefix[2][3]={"",""};
725 if ( indep_var->type == NUMERIC )
727 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
728 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
732 val_lab0 = gp.v.g_value[0].s;
733 val_lab1 = gp.v.g_value[1].s;
736 if (gp.criterion == CMP_LE )
738 strcpy(prefix[0],"< ");
739 strcpy(prefix[1],">=");
740 indep_value[0] = gp.v.critical_value;
741 indep_value[1] = gp.v.critical_value;
745 indep_value[0] = gp.v.g_value[0].f;
746 indep_value[1] = gp.v.g_value[1].f;
751 for (i=0; i < cmd->n_variables; ++i)
753 struct variable *var = cmd->v_variables[i];
754 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
757 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
760 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
761 "%s%s", prefix[0], val_lab0);
763 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
764 "%s%g", prefix[0], indep_value[0]);
768 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
769 "%s%s", prefix[1], val_lab1);
771 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
772 "%s%g", prefix[1], indep_value[1]);
775 /* Fill in the group statistics */
776 for ( count = 0 ; count < 2 ; ++count )
778 union value search_val;
780 struct group_statistics *gs;
782 if ( gp.criterion == CMP_LE )
786 /* less than ( < ) case */
787 search_val.f = gp.v.critical_value - 1.0;
792 search_val.f = gp.v.critical_value + 1.0;
797 search_val = gp.v.g_value[count];
800 gs = hsh_find(grp_hash, (void *) &search_val);
803 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
804 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
805 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
806 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
812 void ssbox_paired_populate(struct ssbox *ssb,
813 struct cmd_t_test *cmd);
815 /* Initialize the paired values ssbox */
817 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
821 int vsize = n_pairs*2+1;
823 this->populate = ssbox_paired_populate;
825 ssbox_base_init(this, hsize,vsize);
826 tab_title (this->t, 0, _("Paired Sample Statistics"));
827 tab_vline(this->t,TAL_0,1,0,vsize-1);
828 tab_vline(this->t,TAL_2,2,0,vsize-1);
829 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
830 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
831 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
832 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
836 /* Populate the ssbox for paired values */
838 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
844 for (i=0; i < n_pairs; ++i)
848 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
850 for (j=0 ; j < 2 ; ++j)
852 struct group_statistics *gs;
854 gs = &group_proc_get (pairs[i].v[j])->ugs;
858 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
861 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
862 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
863 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
864 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
870 /* Populate the one sample ssbox */
872 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
878 for (i=0; i < cmd->n_variables; ++i)
880 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
882 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
883 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
884 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
885 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
886 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
893 /* Implementation of the Test Results box struct */
895 void trbox_base_init(struct trbox *self,size_t n_vars, int cols);
896 void trbox_base_finalize(struct trbox *trb);
898 void trbox_independent_samples_init(struct trbox *trb,
899 struct cmd_t_test *cmd );
901 void trbox_independent_samples_populate(struct trbox *trb,
902 struct cmd_t_test *cmd);
904 void trbox_one_sample_init(struct trbox *self,
905 struct cmd_t_test *cmd );
907 void trbox_one_sample_populate(struct trbox *trb,
908 struct cmd_t_test *cmd);
910 void trbox_paired_init(struct trbox *self,
911 struct cmd_t_test *cmd );
913 void trbox_paired_populate(struct trbox *trb,
914 struct cmd_t_test *cmd);
918 /* Create a trbox according to mode*/
920 trbox_create(struct trbox *trb,
921 struct cmd_t_test *cmd, int mode)
926 trbox_one_sample_init(trb,cmd);
929 trbox_independent_samples_init(trb,cmd);
932 trbox_paired_init(trb,cmd);
939 /* Populate a trbox according to cmd */
941 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
943 trb->populate(trb,cmd);
946 /* Submit and destroy a trbox */
948 trbox_finalize(struct trbox *trb)
953 /* Initialize the independent samples trbox */
955 trbox_independent_samples_init(struct trbox *self,
956 struct cmd_t_test *cmd UNUSED)
959 const int vsize=cmd->n_variables*2+3;
962 self->populate = trbox_independent_samples_populate;
964 trbox_base_init(self,cmd->n_variables*2,hsize);
965 tab_title(self->t,0,_("Independent Samples Test"));
966 tab_hline(self->t,TAL_1,2,hsize-1,1);
967 tab_vline(self->t,TAL_2,2,0,vsize-1);
968 tab_vline(self->t,TAL_1,4,0,vsize-1);
969 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
970 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
971 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
972 tab_joint_text(self->t, 2, 0, 3, 0,
973 TAB_CENTER,_("Levene's Test for Equality of Variances"));
974 tab_joint_text(self->t, 4,0,hsize-1,0,
975 TAB_CENTER,_("t-test for Equality of Means"));
977 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
978 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
979 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
980 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
981 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
982 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
983 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
984 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
985 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
987 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
988 _("%g%% Confidence Interval of the Difference"),
989 cmd->criteria*100.0);
993 /* Populate the independent samples trbox */
995 trbox_independent_samples_populate(struct trbox *self,
996 struct cmd_t_test *cmd )
1001 for (i=0; i < cmd->n_variables; ++i)
1010 double pooled_variance;
1011 double std_err_diff;
1014 struct variable *var = cmd->v_variables[i];
1015 struct group_proc *grp_data = group_proc_get (var);
1017 struct hsh_table *grp_hash = grp_data->group_hash;
1019 struct group_statistics *gs0 ;
1020 struct group_statistics *gs1 ;
1022 union value search_val;
1024 if ( gp.criterion == CMP_LE )
1025 search_val.f = gp.v.critical_value - 1.0;
1027 search_val = gp.v.g_value[0];
1029 gs0 = hsh_find(grp_hash, (void *) &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[1];
1037 gs1 = hsh_find(grp_hash, (void *) &search_val);
1041 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1043 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1046 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1048 /* Now work out the significance of the Levene test */
1049 df1 = 1; df2 = grp_data->ugs.n - 2;
1050 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1052 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1054 df = gs0->n + gs1->n - 2.0 ;
1055 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1057 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1059 (gs1->n )*pow2(gs1->s_std_dev)
1062 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1063 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1065 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1067 p = gsl_cdf_tdist_P(t, df);
1068 q = gsl_cdf_tdist_Q(t, df);
1070 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1072 mean_diff = gs0->mean - gs1->mean;
1073 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1076 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1077 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1080 /* Now work out the confidence interval */
1081 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1083 t = gsl_cdf_tdist_Qinv(q,df);
1084 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1085 mean_diff - t * std_err_diff, 8, 3);
1087 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1088 mean_diff + t * std_err_diff, 8, 3);
1093 /* Now for the \sigma_1 != \sigma_2 case */
1094 tab_text (self->t, 1, i*2+3+1,
1095 TAB_LEFT, _("Equal variances not assumed"));
1098 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1099 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1101 t = mean_diff / sqrt(se2) ;
1102 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1105 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1109 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1113 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1115 p = gsl_cdf_tdist_P(t, df);
1116 q = gsl_cdf_tdist_Q(t, df);
1118 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1120 /* Now work out the confidence interval */
1121 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1123 t = gsl_cdf_tdist_Qinv(q, df);
1125 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1128 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1131 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1132 mean_diff - t * std_err_diff, 8, 3);
1134 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1135 mean_diff + t * std_err_diff, 8, 3);
1141 /* Initialize the paired samples trbox */
1143 trbox_paired_init(struct trbox *self,
1144 struct cmd_t_test *cmd UNUSED)
1148 const int vsize=n_pairs+3;
1150 self->populate = trbox_paired_populate;
1152 trbox_base_init(self,n_pairs,hsize);
1153 tab_title (self->t, 0, _("Paired Samples Test"));
1154 tab_hline(self->t,TAL_1,2,6,1);
1155 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1156 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1157 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1158 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1159 tab_hline(self->t,TAL_1,5,6, 2);
1160 tab_vline(self->t,TAL_0,6,0,1);
1162 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1163 _("%g%% Confidence Interval of the Difference"),
1164 cmd->criteria*100.0);
1166 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1167 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1168 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1169 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1170 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1171 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1172 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1173 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1176 /* Populate the paired samples trbox */
1178 trbox_paired_populate(struct trbox *trb,
1179 struct cmd_t_test *cmd UNUSED)
1183 for (i=0; i < n_pairs; ++i)
1188 double n = pairs[i].n;
1192 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1194 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1195 pairs[i].v[0]->name, pairs[i].v[1]->name);
1197 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1199 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1202 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1203 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1205 /* Now work out the confidence interval */
1206 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1208 t = gsl_cdf_tdist_Qinv(q, df);
1210 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1211 pairs[i].mean_diff - t * se_mean , 8, 4);
1213 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1214 pairs[i].mean_diff + t * se_mean , 8, 4);
1216 t = (pairs[i].mean[0] - pairs[i].mean[1])
1218 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1219 2 * pairs[i].correlation *
1220 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1224 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1226 /* Degrees of freedom */
1227 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1229 p = gsl_cdf_tdist_P(t,df);
1230 q = gsl_cdf_tdist_P(t,df);
1232 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1237 /* Initialize the one sample trbox */
1239 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1242 const int vsize=cmd->n_variables+3;
1244 self->populate = trbox_one_sample_populate;
1246 trbox_base_init(self, cmd->n_variables,hsize);
1247 tab_title (self->t, 0, _("One-Sample Test"));
1248 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1249 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1251 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1252 _("Test Value = %f"), cmd->n_testval[0]);
1254 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1257 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1258 _("%g%% Confidence Interval of the Difference"),
1259 cmd->criteria*100.0);
1261 tab_vline(self->t,TAL_0,6,1,1);
1262 tab_hline(self->t,TAL_1,5,6,2);
1263 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1264 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1265 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1266 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1267 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1268 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1273 /* Populate the one sample trbox */
1275 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1281 for (i=0; i < cmd->n_variables; ++i)
1286 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1289 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1291 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1293 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1295 /* degrees of freedom */
1298 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1300 p = gsl_cdf_tdist_P(t, df);
1301 q = gsl_cdf_tdist_Q(t, df);
1303 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1305 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1307 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1310 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1311 t = gsl_cdf_tdist_Qinv(q, df);
1313 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1314 gs->mean_diff - t * gs->se_mean, 8,4);
1316 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1317 gs->mean_diff + t * gs->se_mean, 8,4);
1321 /* Base initializer for the generalized trbox */
1323 trbox_base_init(struct trbox *self, size_t data_rows, int cols)
1325 const size_t rows = 3 + data_rows;
1327 self->finalize = trbox_base_finalize;
1328 self->t = tab_create (cols, rows, 0);
1329 tab_headers (self->t,0,0,3,0);
1330 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1331 tab_hline(self->t, TAL_2,0,cols-1,3);
1332 tab_dim (self->t, tab_natural_dimensions);
1336 /* Base finalizer for the trbox */
1338 trbox_base_finalize(struct trbox *trb)
1344 /* Create , populate and submit the Paired Samples Correlation box */
1348 const int rows=1+n_pairs;
1352 struct tab_table *table;
1354 table = tab_create (cols,rows,0);
1356 tab_columns (table, SOM_COL_DOWN, 1);
1357 tab_headers (table,0,0,1,0);
1358 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1359 tab_hline(table, TAL_2, 0, cols - 1, 1);
1360 tab_vline(table, TAL_2, 2, 0, rows - 1);
1361 tab_dim(table, tab_natural_dimensions);
1362 tab_title(table, 0, _("Paired Samples Correlations"));
1364 /* column headings */
1365 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1366 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1367 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1369 for (i=0; i < n_pairs; ++i)
1373 double df = pairs[i].n -2;
1375 double correlation_t =
1376 pairs[i].correlation * sqrt(df) /
1377 sqrt(1 - pow2(pairs[i].correlation));
1381 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1384 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1385 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1389 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1390 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1392 p = gsl_cdf_tdist_P(correlation_t, df);
1393 q = gsl_cdf_tdist_Q(correlation_t, df);
1395 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1404 /* Calculation Implementation */
1406 /* Per case calculations common to all variants of the T test */
1408 common_calc (const struct ccase *c, void *_cmd)
1411 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1413 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1416 /* Skip the entire case if /MISSING=LISTWISE is set */
1417 if ( cmd->miss == TTS_LISTWISE )
1419 for(i=0; i< cmd->n_variables ; ++i)
1421 struct variable *v = cmd->v_variables[i];
1422 const union value *val = case_data (c, v->fv);
1424 if (value_is_missing(&v->miss, val) )
1431 /* Listwise has to be implicit if the independent variable is missing ?? */
1432 if ( cmd->sbc_groups )
1434 const union value *gv = case_data (c, indep_var->fv);
1435 if ( value_is_missing(&indep_var->miss, gv) )
1442 for(i=0; i< cmd->n_variables ; ++i)
1444 struct group_statistics *gs;
1445 struct variable *v = cmd->v_variables[i];
1446 const union value *val = case_data (c, v->fv);
1448 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1450 if (! value_is_missing(&v->miss, val) )
1453 gs->sum+=weight * val->f;
1454 gs->ssq+=weight * val->f * val->f;
1460 /* Pre calculations common to all variants of the T test */
1462 common_precalc ( struct cmd_t_test *cmd )
1466 for(i=0; i< cmd->n_variables ; ++i)
1468 struct group_statistics *gs;
1469 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1478 /* Post calculations common to all variants of the T test */
1480 common_postcalc ( struct cmd_t_test *cmd )
1485 for(i=0; i< cmd->n_variables ; ++i)
1487 struct group_statistics *gs;
1488 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1490 gs->mean=gs->sum / gs->n;
1491 gs->s_std_dev= sqrt(
1492 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1497 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1500 gs->se_mean = gs->std_dev / sqrt(gs->n);
1501 gs->mean_diff= gs->sum_diff / gs->n;
1505 /* Per case calculations for one sample t test */
1507 one_sample_calc (const struct ccase *c, void *cmd_)
1510 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1513 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1515 /* Skip the entire case if /MISSING=LISTWISE is set */
1516 if ( cmd->miss == TTS_LISTWISE )
1518 for(i=0; i< cmd->n_variables ; ++i)
1520 struct variable *v = cmd->v_variables[i];
1521 const union value *val = case_data (c, v->fv);
1523 if (value_is_missing(&v->miss, val) )
1530 for(i=0; i< cmd->n_variables ; ++i)
1532 struct group_statistics *gs;
1533 struct variable *v = cmd->v_variables[i];
1534 const union value *val = case_data (c, v->fv);
1536 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1538 if ( ! value_is_missing(&v->miss, val))
1539 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1545 /* Pre calculations for one sample t test */
1547 one_sample_precalc ( struct cmd_t_test *cmd )
1551 for(i=0; i< cmd->n_variables ; ++i)
1553 struct group_statistics *gs;
1554 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1560 /* Post calculations for one sample t test */
1562 one_sample_postcalc (struct cmd_t_test *cmd)
1566 for(i=0; i< cmd->n_variables ; ++i)
1568 struct group_statistics *gs;
1569 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1571 gs->mean_diff = gs->sum_diff / gs->n ;
1578 paired_precalc (struct cmd_t_test *cmd UNUSED)
1582 for(i=0; i < n_pairs ; ++i )
1585 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1586 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1587 pairs[i].sum_of_prod = 0;
1588 pairs[i].correlation = 0;
1589 pairs[i].sum_of_diffs = 0;
1590 pairs[i].ssq_diffs = 0;
1597 paired_calc (const struct ccase *c, void *cmd_)
1601 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1603 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1605 /* Skip the entire case if /MISSING=LISTWISE is set ,
1606 AND one member of a pair is missing */
1607 if ( cmd->miss == TTS_LISTWISE )
1609 for(i=0; i < n_pairs ; ++i )
1611 struct variable *v0 = pairs[i].v[0];
1612 struct variable *v1 = pairs[i].v[1];
1614 const union value *val0 = case_data (c, v0->fv);
1615 const union value *val1 = case_data (c, v1->fv);
1617 if ( value_is_missing(&v0->miss, val0) ||
1618 value_is_missing(&v1->miss, val1) )
1625 for(i=0; i < n_pairs ; ++i )
1627 struct variable *v0 = pairs[i].v[0];
1628 struct variable *v1 = pairs[i].v[1];
1630 const union value *val0 = case_data (c, v0->fv);
1631 const union value *val1 = case_data (c, v1->fv);
1633 if ( ( !value_is_missing(&v0->miss, val0)
1634 && !value_is_missing(&v1->miss, val1) ) )
1636 pairs[i].n += weight;
1637 pairs[i].sum[0] += weight * val0->f;
1638 pairs[i].sum[1] += weight * val1->f;
1640 pairs[i].ssq[0] += weight * pow2(val0->f);
1641 pairs[i].ssq[1] += weight * pow2(val1->f);
1643 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1645 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1646 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1654 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1658 for(i=0; i < n_pairs ; ++i )
1661 const double n = pairs[i].n;
1663 for (j=0; j < 2 ; ++j)
1665 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1666 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1667 pow2(pairs[i].mean[j]))
1670 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1671 pow2(pairs[i].mean[j]))
1675 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1676 pairs[i].mean[0] * pairs[i].mean[1] ;
1677 /* correlation now actually contains the covariance */
1679 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1680 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1682 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1684 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1685 ( pairs[i].ssq_diffs / n )
1687 pow2(pairs[i].mean_diff )
1693 group_precalc (struct cmd_t_test *cmd )
1698 for(i=0; i< cmd->n_variables ; ++i)
1700 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1702 /* There's always 2 groups for a T - TEST */
1705 gp.indep_width = indep_var->width;
1707 ttpr->group_hash = hsh_create(2,
1708 (hsh_compare_func *) compare_group_binary,
1709 (hsh_hash_func *) hash_group_binary,
1710 (hsh_free_func *) free_group,
1713 for (j=0 ; j < 2 ; ++j)
1716 struct group_statistics *gs = xmalloc (sizeof *gs);
1722 if ( gp.criterion == CMP_EQ )
1724 gs->id = gp.v.g_value[j];
1729 gs->id.f = gp.v.critical_value - 1.0 ;
1731 gs->id.f = gp.v.critical_value + 1.0 ;
1734 hsh_insert ( ttpr->group_hash, (void *) gs );
1742 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1746 const union value *gv = case_data (c, indep_var->fv);
1748 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1750 if ( value_is_missing(&indep_var->miss, gv) )
1755 if ( cmd->miss == TTS_LISTWISE )
1757 for(i=0; i< cmd->n_variables ; ++i)
1759 struct variable *v = cmd->v_variables[i];
1760 const union value *val = case_data (c, v->fv);
1762 if (value_is_missing(&v->miss, val) )
1769 gv = case_data (c, indep_var->fv);
1771 for(i=0; i< cmd->n_variables ; ++i)
1773 struct variable *var = cmd->v_variables[i];
1774 const union value *val = case_data (c, var->fv);
1775 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1776 struct group_statistics *gs;
1778 gs = hsh_find(grp_hash, (void *) gv);
1780 /* If the independent variable doesn't match either of the values
1781 for this case then move on to the next case */
1785 if ( !value_is_missing(&var->miss, val) )
1788 gs->sum+=weight * val->f;
1789 gs->ssq+=weight * pow2(val->f);
1798 group_postcalc ( struct cmd_t_test *cmd )
1802 for(i=0; i< cmd->n_variables ; ++i)
1804 struct variable *var = cmd->v_variables[i];
1805 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1806 struct hsh_iterator g;
1807 struct group_statistics *gs;
1810 for (gs = hsh_first (grp_hash,&g);
1812 gs = hsh_next(grp_hash,&g))
1814 gs->mean = gs->sum / gs->n;
1816 gs->s_std_dev= sqrt(
1817 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1822 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1825 gs->se_mean = gs->std_dev / sqrt(gs->n);
1835 calculate(const struct casefile *cf, void *cmd_)
1837 struct ssbox stat_summary_box;
1838 struct trbox test_results_box;
1840 struct casereader *r;
1843 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1845 common_precalc(cmd);
1846 for(r = casefile_get_reader (cf);
1847 casereader_read (r, &c) ;
1850 common_calc(&c,cmd);
1852 casereader_destroy (r);
1853 common_postcalc(cmd);
1858 one_sample_precalc(cmd);
1859 for(r = casefile_get_reader (cf);
1860 casereader_read (r, &c) ;
1863 one_sample_calc(&c,cmd);
1865 casereader_destroy (r);
1866 one_sample_postcalc(cmd);
1870 paired_precalc(cmd);
1871 for(r = casefile_get_reader (cf);
1872 casereader_read (r, &c) ;
1875 paired_calc(&c,cmd);
1877 casereader_destroy (r);
1878 paired_postcalc(cmd);
1884 for(r = casefile_get_reader (cf);
1885 casereader_read (r, &c) ;
1890 casereader_destroy (r);
1891 group_postcalc(cmd);
1893 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1894 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1899 ssbox_create(&stat_summary_box,cmd,mode);
1900 ssbox_populate(&stat_summary_box,cmd);
1901 ssbox_finalize(&stat_summary_box);
1903 if ( mode == T_PAIRED)
1906 trbox_create(&test_results_box,cmd,mode);
1907 trbox_populate(&test_results_box,cmd);
1908 trbox_finalize(&test_results_box);
1913 short which_group(const struct group_statistics *g,
1914 const struct group_properties *p);
1916 /* Return -1 if the id of a is less than b; +1 if greater than and
1919 compare_group_binary(const struct group_statistics *a,
1920 const struct group_statistics *b,
1921 const struct group_properties *p)
1926 if ( p->criterion == CMP_LE )
1928 /* less-than-or-equal comparision is not meaningfull for
1929 alpha variables, so we shouldn't ever arrive here */
1930 assert(p->indep_width == 0 ) ;
1932 flag_a = ( a->id.f < p->v.critical_value ) ;
1933 flag_b = ( b->id.f < p->v.critical_value ) ;
1937 flag_a = which_group(a, p);
1938 flag_b = which_group(b, p);
1941 if (flag_a < flag_b )
1944 return (flag_a > flag_b);
1947 /* This is a degenerate case of a hash, since it can only return three possible
1948 values. It's really a comparison, being used as a hash function */
1951 hash_group_binary(const struct group_statistics *g,
1952 const struct group_properties *p)
1956 if ( p->criterion == CMP_LE )
1958 /* Not meaningfull to do a less than compare for alpha values ? */
1959 assert(p->indep_width == 0 ) ;
1960 flag = ( g->id.f < p->v.critical_value ) ;
1962 else if ( p->criterion == CMP_EQ)
1964 flag = which_group(g,p);
1972 /* return 0 if G belongs to group 0,
1973 1 if it belongs to group 1,
1974 2 if it belongs to neither group */
1976 which_group(const struct group_statistics *g,
1977 const struct group_properties *p)
1980 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1983 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))