1 /* PSPP - computes sample statistics. -*-c-*-
3 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
4 Written by John Williams <johnr.williams@stonebow.otago.ac.nz>.
5 Almost completly re-written by John Darrington 2004
7 This program is free software; you can redistribute it and/or
8 modify it under the terms of the GNU General Public License as
9 published by the Free Software Foundation; either version 2 of the
10 License, or (at your option) any later version.
12 This program is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
24 #include <gsl/gsl_cdf.h>
29 #include <data/case.h>
30 #include <data/casefile.h>
31 #include <data/dictionary.h>
32 #include <data/procedure.h>
33 #include <data/value-labels.h>
34 #include <data/variable.h>
35 #include <language/command.h>
36 #include <language/dictionary/split-file.h>
37 #include <language/lexer/lexer.h>
38 #include <libpspp/alloc.h>
39 #include <libpspp/compiler.h>
40 #include <libpspp/hash.h>
41 #include <libpspp/magic.h>
42 #include <libpspp/message.h>
43 #include <libpspp/message.h>
44 #include <libpspp/misc.h>
45 #include <libpspp/str.h>
46 #include <math/group-proc.h>
47 #include <math/levene.h>
48 #include <output/manager.h>
49 #include <output/table.h>
54 #define _(msgid) gettext (msgid)
62 +variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
64 missing=miss:!analysis/listwise,
65 incl:include/!exclude;
66 +format=fmt:!labels/nolabels;
67 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
75 /* Function to use for testing for missing values */
76 static is_missing_func *value_is_missing;
78 /* Variable for the GROUPS subcommand, if given. */
79 static struct variable *indep_var;
87 struct group_properties
89 /* The comparison criterion */
90 enum comparison criterion;
92 /* The width of the independent variable */
96 /* The value of the independent variable at which groups are determined to
97 belong to one group or the other */
98 double critical_value;
101 /* The values of the independent variable for each group */
102 union value g_value[2];
108 static struct group_properties gp ;
112 /* PAIRS: Number of pairs to be compared ; each pair. */
113 static int n_pairs = 0 ;
116 /* The variables comprising the pair */
117 struct variable *v[2];
119 /* The number of valid variable pairs */
122 /* The sum of the members */
125 /* sum of squares of the members */
128 /* Std deviation of the members */
132 /* Sample Std deviation of the members */
135 /* The means of the members */
138 /* The correlation coefficient between the variables */
141 /* The sum of the differences */
144 /* The sum of the products */
147 /* The mean of the differences */
150 /* The sum of the squares of the differences */
153 /* The std deviation of the differences */
157 static struct pair *pairs=0;
159 static int parse_value (union value * v, int type) ;
161 /* Structures and Functions for the Statistics Summary Box */
163 typedef void populate_ssbox_func(struct ssbox *ssb,
164 struct cmd_t_test *cmd);
165 typedef void finalize_ssbox_func(struct ssbox *ssb);
171 populate_ssbox_func *populate;
172 finalize_ssbox_func *finalize;
177 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
179 /* Populate a ssbox according to cmd */
180 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
182 /* Submit and destroy a ssbox */
183 void ssbox_finalize(struct ssbox *ssb);
185 /* A function to create, populate and submit the Paired Samples Correlation
190 /* Structures and Functions for the Test Results Box */
193 typedef void populate_trbox_func(struct trbox *trb,
194 struct cmd_t_test *cmd);
195 typedef void finalize_trbox_func(struct trbox *trb);
199 populate_trbox_func *populate;
200 finalize_trbox_func *finalize;
204 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
206 /* Populate a ssbox according to cmd */
207 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
209 /* Submit and destroy a ssbox */
210 void trbox_finalize(struct trbox *trb);
212 /* Which mode was T-TEST invoked */
220 static int common_calc (const struct ccase *, void *);
221 static void common_precalc (struct cmd_t_test *);
222 static void common_postcalc (struct cmd_t_test *);
224 static int one_sample_calc (const struct ccase *, void *);
225 static void one_sample_precalc (struct cmd_t_test *);
226 static void one_sample_postcalc (struct cmd_t_test *);
228 static int paired_calc (const struct ccase *, void *);
229 static void paired_precalc (struct cmd_t_test *);
230 static void paired_postcalc (struct cmd_t_test *);
232 static void group_precalc (struct cmd_t_test *);
233 static int group_calc (const struct ccase *, struct cmd_t_test *);
234 static void group_postcalc (struct cmd_t_test *);
237 static bool calculate(const struct ccase *first,
238 const struct casefile *cf, void *_mode);
242 static struct cmd_t_test cmd;
244 static int bad_weight_warn;
247 static int compare_group_binary(const struct group_statistics *a,
248 const struct group_statistics *b,
249 const struct group_properties *p);
252 static unsigned hash_group_binary(const struct group_statistics *g,
253 const struct group_properties *p);
262 if ( !parse_t_test(&cmd, NULL) )
265 if (! cmd.sbc_criteria)
270 if (cmd.sbc_testval) ++m;
271 if (cmd.sbc_groups) ++m;
272 if (cmd.sbc_pairs) ++m;
277 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
286 else if (cmd.sbc_groups)
291 if ( mode == T_PAIRED)
293 if (cmd.sbc_variables)
295 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
301 /* Iterate through the pairs and put each variable that is a
302 member of a pair into cmd.v_variables */
305 struct hsh_iterator hi;
306 struct hsh_table *hash;
309 hash = hsh_create (n_pairs, compare_var_names, hash_var_name, 0, 0);
311 for (i=0; i < n_pairs; ++i)
313 hsh_insert(hash,pairs[i].v[0]);
314 hsh_insert(hash,pairs[i].v[1]);
317 assert(cmd.n_variables == 0);
318 cmd.n_variables = hsh_count(hash);
320 cmd.v_variables = xnrealloc (cmd.v_variables, cmd.n_variables,
321 sizeof *cmd.v_variables);
322 /* Iterate through the hash */
323 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
325 v=hsh_next(hash,&hi) )
326 cmd.v_variables[i++]=v;
331 else if ( !cmd.sbc_variables)
333 msg(SE, _("One or more VARIABLES must be specified."));
339 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
340 if (cmd.incl == TTS_INCLUDE )
341 value_is_missing = mv_is_value_system_missing;
343 value_is_missing = mv_is_value_missing;
347 ok = multipass_procedure_with_splits (calculate, &cmd);
353 if ( mode == T_IND_SAMPLES)
356 /* Destroy any group statistics we created */
357 for (v = 0 ; v < cmd.n_variables ; ++v )
359 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
360 hsh_destroy (grpp->group_hash);
365 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
369 tts_custom_groups (struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
371 int n_group_values=0;
375 indep_var = parse_variable ();
378 lex_error ("expecting variable name in GROUPS subcommand");
382 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
384 msg (SE, _("Long string variable %s is not valid here."),
389 if (!lex_match ('('))
391 if (indep_var->type == NUMERIC)
393 gp.v.g_value[0].f = 1;
394 gp.v.g_value[1].f = 2;
396 gp.criterion = CMP_EQ;
404 msg (SE, _("When applying GROUPS to a string variable, two "
405 "values must be specified."));
410 if (!parse_value (&gp.v.g_value[0], indep_var->type))
416 if (indep_var->type != NUMERIC)
419 msg (SE, _("When applying GROUPS to a string variable, two "
420 "values must be specified."));
423 gp.criterion = CMP_LE;
424 gp.v.critical_value = gp.v.g_value[0].f;
430 if (!parse_value (&gp.v.g_value[1], indep_var->type))
434 if (!lex_force_match (')'))
437 if ( n_group_values == 2 )
438 gp.criterion = CMP_EQ ;
440 gp.criterion = CMP_LE ;
448 tts_custom_pairs (struct cmd_t_test *cmd UNUSED, void *aux UNUSED)
450 struct variable **vars;
452 size_t n_pairs_local;
454 size_t n_before_WITH;
455 size_t n_after_WITH = SIZE_MAX;
456 int paired ; /* Was the PAIRED keyword given ? */
461 if (!parse_variables (default_dict, &vars, &n_vars,
462 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
470 if (lex_match (T_WITH))
472 n_before_WITH = n_vars;
473 if (!parse_variables (default_dict, &vars, &n_vars,
474 PV_DUPLICATE | PV_APPEND
475 | PV_NUMERIC | PV_NO_SCRATCH))
480 n_after_WITH = n_vars - n_before_WITH;
483 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
485 /* Determine the number of pairs needed */
488 if (n_before_WITH != n_after_WITH)
491 msg (SE, _("PAIRED was specified but the number of variables "
492 "preceding WITH (%d) did not match the number "
494 n_before_WITH, n_after_WITH );
497 n_pairs_local = n_before_WITH;
499 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
501 n_pairs_local = n_before_WITH * n_after_WITH ;
503 else /* Neither WITH nor PAIRED keyword given */
508 msg (SE, _("At least two variables must be specified "
513 /* how many ways can you pick 2 from n_vars ? */
514 n_pairs_local = n_vars * (n_vars - 1) / 2;
518 /* Allocate storage for the pairs */
519 pairs = xnrealloc (pairs, n_pairs + n_pairs_local, sizeof *pairs);
521 /* Populate the pairs with the appropriate variables */
526 assert(n_pairs_local == n_vars / 2);
527 for (i = 0; i < n_pairs_local; ++i)
529 pairs[i].v[n_pairs] = vars[i];
530 pairs[i].v[n_pairs + 1] = vars[i + n_pairs_local];
533 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
538 for(i=0 ; i < n_before_WITH ; ++i )
540 for(j=0 ; j < n_after_WITH ; ++j)
542 pairs[p].v[0] = vars[i];
543 pairs[p].v[1] = vars[j+n_before_WITH];
548 else /* Neither WITH nor PAIRED given */
553 for(i=0 ; i < n_vars ; ++i )
555 for(j=i+1 ; j < n_vars ; ++j)
557 pairs[p].v[0] = vars[i];
558 pairs[p].v[1] = vars[j];
564 n_pairs+=n_pairs_local;
570 /* Parses the current token (numeric or string, depending on type)
571 value v and returns success. */
573 parse_value (union value * v, int type )
577 if (!lex_force_num ())
583 if (!lex_force_string ())
585 strncpy (v->s, ds_cstr (&tokstr), ds_length (&tokstr));
594 /* Implementation of the SSBOX object */
596 void ssbox_base_init(struct ssbox *this, int cols,int rows);
598 void ssbox_base_finalize(struct ssbox *ssb);
600 void ssbox_one_sample_init(struct ssbox *this,
601 struct cmd_t_test *cmd );
603 void ssbox_independent_samples_init(struct ssbox *this,
604 struct cmd_t_test *cmd);
606 void ssbox_paired_init(struct ssbox *this,
607 struct cmd_t_test *cmd);
610 /* Factory to create an ssbox */
612 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
617 ssbox_one_sample_init(ssb,cmd);
620 ssbox_independent_samples_init(ssb,cmd);
623 ssbox_paired_init(ssb,cmd);
632 /* Despatcher for the populate method */
634 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
636 ssb->populate(ssb,cmd);
640 /* Despatcher for finalize */
642 ssbox_finalize(struct ssbox *ssb)
648 /* Submit the box and clear up */
650 ssbox_base_finalize(struct ssbox *ssb)
657 /* Initialize a ssbox struct */
659 ssbox_base_init(struct ssbox *this, int cols,int rows)
661 this->finalize = ssbox_base_finalize;
662 this->t = tab_create (cols, rows, 0);
664 tab_columns (this->t, SOM_COL_DOWN, 1);
665 tab_headers (this->t,0,0,1,0);
666 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
667 tab_hline(this->t, TAL_2,0,cols-1,1);
668 tab_dim (this->t, tab_natural_dimensions);
671 void ssbox_one_sample_populate(struct ssbox *ssb,
672 struct cmd_t_test *cmd);
674 /* Initialize the one_sample ssbox */
676 ssbox_one_sample_init(struct ssbox *this,
677 struct cmd_t_test *cmd )
680 const int vsize=cmd->n_variables+1;
682 this->populate = ssbox_one_sample_populate;
684 ssbox_base_init(this, hsize,vsize);
685 tab_title (this->t, _("One-Sample Statistics"));
686 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
687 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
688 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
689 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
690 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
693 void ssbox_independent_samples_populate(struct ssbox *ssb,
694 struct cmd_t_test *cmd);
696 /* Initialize the independent samples ssbox */
698 ssbox_independent_samples_init(struct ssbox *this,
699 struct cmd_t_test *cmd)
702 int vsize = cmd->n_variables*2 +1;
704 this->populate = ssbox_independent_samples_populate;
706 ssbox_base_init(this, hsize,vsize);
707 tab_vline (this->t, TAL_GAP, 1, 0,vsize - 1);
708 tab_title (this->t, _("Group Statistics"));
709 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
710 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
711 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
712 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
713 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
717 /* Populate the ssbox for independent samples */
719 ssbox_independent_samples_populate(struct ssbox *ssb,
720 struct cmd_t_test *cmd)
726 double indep_value[2];
728 char prefix[2][3]={"",""};
730 if ( indep_var->type == NUMERIC )
732 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
733 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
737 val_lab0 = gp.v.g_value[0].s;
738 val_lab1 = gp.v.g_value[1].s;
741 if (gp.criterion == CMP_LE )
743 strcpy(prefix[0],"< ");
744 strcpy(prefix[1],">=");
745 indep_value[0] = gp.v.critical_value;
746 indep_value[1] = gp.v.critical_value;
750 indep_value[0] = gp.v.g_value[0].f;
751 indep_value[1] = gp.v.g_value[1].f;
756 for (i=0; i < cmd->n_variables; ++i)
758 struct variable *var = cmd->v_variables[i];
759 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
762 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
765 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
766 "%s%s", prefix[0], val_lab0);
768 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
769 "%s%g", prefix[0], indep_value[0]);
773 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
774 "%s%s", prefix[1], val_lab1);
776 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
777 "%s%g", prefix[1], indep_value[1]);
780 /* Fill in the group statistics */
781 for ( count = 0 ; count < 2 ; ++count )
783 union value search_val;
785 struct group_statistics *gs;
787 if ( gp.criterion == CMP_LE )
791 /* less than ( < ) case */
792 search_val.f = gp.v.critical_value - 1.0;
797 search_val.f = gp.v.critical_value + 1.0;
802 search_val = gp.v.g_value[count];
805 gs = hsh_find(grp_hash, (void *) &search_val);
808 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
809 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
810 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
811 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
817 void ssbox_paired_populate(struct ssbox *ssb,
818 struct cmd_t_test *cmd);
820 /* Initialize the paired values ssbox */
822 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
826 int vsize = n_pairs*2+1;
828 this->populate = ssbox_paired_populate;
830 ssbox_base_init(this, hsize,vsize);
831 tab_title (this->t, _("Paired Sample Statistics"));
832 tab_vline(this->t,TAL_GAP,1,0,vsize-1);
833 tab_vline(this->t,TAL_2,2,0,vsize-1);
834 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
835 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
836 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
837 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
841 /* Populate the ssbox for paired values */
843 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
849 for (i=0; i < n_pairs; ++i)
853 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
855 for (j=0 ; j < 2 ; ++j)
857 struct group_statistics *gs;
859 gs = &group_proc_get (pairs[i].v[j])->ugs;
863 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
866 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
867 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
868 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
869 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
875 /* Populate the one sample ssbox */
877 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
883 for (i=0; i < cmd->n_variables; ++i)
885 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
887 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
888 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
889 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
890 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
891 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
898 /* Implementation of the Test Results box struct */
900 void trbox_base_init(struct trbox *self,size_t n_vars, int cols);
901 void trbox_base_finalize(struct trbox *trb);
903 void trbox_independent_samples_init(struct trbox *trb,
904 struct cmd_t_test *cmd );
906 void trbox_independent_samples_populate(struct trbox *trb,
907 struct cmd_t_test *cmd);
909 void trbox_one_sample_init(struct trbox *self,
910 struct cmd_t_test *cmd );
912 void trbox_one_sample_populate(struct trbox *trb,
913 struct cmd_t_test *cmd);
915 void trbox_paired_init(struct trbox *self,
916 struct cmd_t_test *cmd );
918 void trbox_paired_populate(struct trbox *trb,
919 struct cmd_t_test *cmd);
923 /* Create a trbox according to mode*/
925 trbox_create(struct trbox *trb,
926 struct cmd_t_test *cmd, int mode)
931 trbox_one_sample_init(trb,cmd);
934 trbox_independent_samples_init(trb,cmd);
937 trbox_paired_init(trb,cmd);
944 /* Populate a trbox according to cmd */
946 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
948 trb->populate(trb,cmd);
951 /* Submit and destroy a trbox */
953 trbox_finalize(struct trbox *trb)
958 /* Initialize the independent samples trbox */
960 trbox_independent_samples_init(struct trbox *self,
961 struct cmd_t_test *cmd UNUSED)
964 const int vsize=cmd->n_variables*2+3;
967 self->populate = trbox_independent_samples_populate;
969 trbox_base_init(self,cmd->n_variables*2,hsize);
970 tab_title(self->t,_("Independent Samples Test"));
971 tab_hline(self->t,TAL_1,2,hsize-1,1);
972 tab_vline(self->t,TAL_2,2,0,vsize-1);
973 tab_vline(self->t,TAL_1,4,0,vsize-1);
974 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
975 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
976 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
977 tab_joint_text(self->t, 2, 0, 3, 0,
978 TAB_CENTER,_("Levene's Test for Equality of Variances"));
979 tab_joint_text(self->t, 4,0,hsize-1,0,
980 TAB_CENTER,_("t-test for Equality of Means"));
982 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
983 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
984 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
985 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
986 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
987 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
988 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
989 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
990 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
992 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
993 _("%g%% Confidence Interval of the Difference"),
994 cmd->criteria*100.0);
998 /* Populate the independent samples trbox */
1000 trbox_independent_samples_populate(struct trbox *self,
1001 struct cmd_t_test *cmd )
1006 for (i=0; i < cmd->n_variables; ++i)
1015 double pooled_variance;
1016 double std_err_diff;
1019 struct variable *var = cmd->v_variables[i];
1020 struct group_proc *grp_data = group_proc_get (var);
1022 struct hsh_table *grp_hash = grp_data->group_hash;
1024 struct group_statistics *gs0 ;
1025 struct group_statistics *gs1 ;
1027 union value search_val;
1029 if ( gp.criterion == CMP_LE )
1030 search_val.f = gp.v.critical_value - 1.0;
1032 search_val = gp.v.g_value[0];
1034 gs0 = hsh_find(grp_hash, (void *) &search_val);
1037 if ( gp.criterion == CMP_LE )
1038 search_val.f = gp.v.critical_value + 1.0;
1040 search_val = gp.v.g_value[1];
1042 gs1 = hsh_find(grp_hash, (void *) &search_val);
1046 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1048 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1051 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1053 /* Now work out the significance of the Levene test */
1054 df1 = 1; df2 = grp_data->ugs.n - 2;
1055 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1057 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1059 df = gs0->n + gs1->n - 2.0 ;
1060 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1062 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1064 (gs1->n )*pow2(gs1->s_std_dev)
1067 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1068 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1070 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1072 p = gsl_cdf_tdist_P(t, df);
1073 q = gsl_cdf_tdist_Q(t, df);
1075 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1077 mean_diff = gs0->mean - gs1->mean;
1078 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1081 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1082 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1085 /* Now work out the confidence interval */
1086 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1088 t = gsl_cdf_tdist_Qinv(q,df);
1089 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1090 mean_diff - t * std_err_diff, 8, 3);
1092 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1093 mean_diff + t * std_err_diff, 8, 3);
1098 /* Now for the \sigma_1 != \sigma_2 case */
1099 tab_text (self->t, 1, i*2+3+1,
1100 TAB_LEFT, _("Equal variances not assumed"));
1103 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1104 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1106 t = mean_diff / sqrt(se2) ;
1107 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1110 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1114 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1118 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1120 p = gsl_cdf_tdist_P(t, df);
1121 q = gsl_cdf_tdist_Q(t, df);
1123 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1125 /* Now work out the confidence interval */
1126 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1128 t = gsl_cdf_tdist_Qinv(q, df);
1130 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1133 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1136 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1137 mean_diff - t * std_err_diff, 8, 3);
1139 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1140 mean_diff + t * std_err_diff, 8, 3);
1146 /* Initialize the paired samples trbox */
1148 trbox_paired_init(struct trbox *self,
1149 struct cmd_t_test *cmd UNUSED)
1153 const int vsize=n_pairs+3;
1155 self->populate = trbox_paired_populate;
1157 trbox_base_init(self,n_pairs,hsize);
1158 tab_title (self->t, _("Paired Samples Test"));
1159 tab_hline(self->t,TAL_1,2,6,1);
1160 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1161 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1162 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1163 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1164 tab_hline(self->t,TAL_1,5,6, 2);
1165 tab_vline(self->t,TAL_GAP,6,0,1);
1167 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1168 _("%g%% Confidence Interval of the Difference"),
1169 cmd->criteria*100.0);
1171 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1172 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1173 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1174 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1175 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1176 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1177 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1178 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1181 /* Populate the paired samples trbox */
1183 trbox_paired_populate(struct trbox *trb,
1184 struct cmd_t_test *cmd UNUSED)
1188 for (i=0; i < n_pairs; ++i)
1193 double n = pairs[i].n;
1197 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1199 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1200 pairs[i].v[0]->name, pairs[i].v[1]->name);
1202 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1204 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1207 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1208 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1210 /* Now work out the confidence interval */
1211 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1213 t = gsl_cdf_tdist_Qinv(q, df);
1215 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1216 pairs[i].mean_diff - t * se_mean , 8, 4);
1218 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1219 pairs[i].mean_diff + t * se_mean , 8, 4);
1221 t = (pairs[i].mean[0] - pairs[i].mean[1])
1223 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1224 2 * pairs[i].correlation *
1225 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1229 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1231 /* Degrees of freedom */
1232 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1234 p = gsl_cdf_tdist_P(t,df);
1235 q = gsl_cdf_tdist_P(t,df);
1237 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1242 /* Initialize the one sample trbox */
1244 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1247 const int vsize=cmd->n_variables+3;
1249 self->populate = trbox_one_sample_populate;
1251 trbox_base_init(self, cmd->n_variables,hsize);
1252 tab_title (self->t, _("One-Sample Test"));
1253 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1254 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1256 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1257 _("Test Value = %f"), cmd->n_testval[0]);
1259 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1262 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1263 _("%g%% Confidence Interval of the Difference"),
1264 cmd->criteria*100.0);
1266 tab_vline(self->t,TAL_GAP,6,1,1);
1267 tab_hline(self->t,TAL_1,5,6,2);
1268 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1269 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1270 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1271 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1272 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1273 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1278 /* Populate the one sample trbox */
1280 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1286 for (i=0; i < cmd->n_variables; ++i)
1291 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1294 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1296 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1298 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1300 /* degrees of freedom */
1303 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1305 p = gsl_cdf_tdist_P(t, df);
1306 q = gsl_cdf_tdist_Q(t, df);
1308 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1310 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1312 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1315 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1316 t = gsl_cdf_tdist_Qinv(q, df);
1318 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1319 gs->mean_diff - t * gs->se_mean, 8,4);
1321 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1322 gs->mean_diff + t * gs->se_mean, 8,4);
1326 /* Base initializer for the generalized trbox */
1328 trbox_base_init(struct trbox *self, size_t data_rows, int cols)
1330 const size_t rows = 3 + data_rows;
1332 self->finalize = trbox_base_finalize;
1333 self->t = tab_create (cols, rows, 0);
1334 tab_headers (self->t,0,0,3,0);
1335 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1336 tab_hline(self->t, TAL_2,0,cols-1,3);
1337 tab_dim (self->t, tab_natural_dimensions);
1341 /* Base finalizer for the trbox */
1343 trbox_base_finalize(struct trbox *trb)
1349 /* Create , populate and submit the Paired Samples Correlation box */
1353 const int rows=1+n_pairs;
1357 struct tab_table *table;
1359 table = tab_create (cols,rows,0);
1361 tab_columns (table, SOM_COL_DOWN, 1);
1362 tab_headers (table,0,0,1,0);
1363 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1364 tab_hline(table, TAL_2, 0, cols - 1, 1);
1365 tab_vline(table, TAL_2, 2, 0, rows - 1);
1366 tab_dim(table, tab_natural_dimensions);
1367 tab_title(table, _("Paired Samples Correlations"));
1369 /* column headings */
1370 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1371 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1372 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1374 for (i=0; i < n_pairs; ++i)
1378 double df = pairs[i].n -2;
1380 double correlation_t =
1381 pairs[i].correlation * sqrt(df) /
1382 sqrt(1 - pow2(pairs[i].correlation));
1386 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1389 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1390 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1394 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1395 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1397 p = gsl_cdf_tdist_P(correlation_t, df);
1398 q = gsl_cdf_tdist_Q(correlation_t, df);
1400 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1409 /* Calculation Implementation */
1411 /* Per case calculations common to all variants of the T test */
1413 common_calc (const struct ccase *c, void *_cmd)
1416 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1418 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1421 /* Skip the entire case if /MISSING=LISTWISE is set */
1422 if ( cmd->miss == TTS_LISTWISE )
1424 for(i=0; i< cmd->n_variables ; ++i)
1426 struct variable *v = cmd->v_variables[i];
1427 const union value *val = case_data (c, v->fv);
1429 if (value_is_missing(&v->miss, val) )
1436 /* Listwise has to be implicit if the independent variable is missing ?? */
1437 if ( cmd->sbc_groups )
1439 const union value *gv = case_data (c, indep_var->fv);
1440 if ( value_is_missing(&indep_var->miss, gv) )
1447 for(i=0; i< cmd->n_variables ; ++i)
1449 struct group_statistics *gs;
1450 struct variable *v = cmd->v_variables[i];
1451 const union value *val = case_data (c, v->fv);
1453 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1455 if (! value_is_missing(&v->miss, val) )
1458 gs->sum+=weight * val->f;
1459 gs->ssq+=weight * val->f * val->f;
1465 /* Pre calculations common to all variants of the T test */
1467 common_precalc ( struct cmd_t_test *cmd )
1471 for(i=0; i< cmd->n_variables ; ++i)
1473 struct group_statistics *gs;
1474 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1483 /* Post calculations common to all variants of the T test */
1485 common_postcalc ( struct cmd_t_test *cmd )
1490 for(i=0; i< cmd->n_variables ; ++i)
1492 struct group_statistics *gs;
1493 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1495 gs->mean=gs->sum / gs->n;
1496 gs->s_std_dev= sqrt(
1497 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1502 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1505 gs->se_mean = gs->std_dev / sqrt(gs->n);
1506 gs->mean_diff= gs->sum_diff / gs->n;
1510 /* Per case calculations for one sample t test */
1512 one_sample_calc (const struct ccase *c, void *cmd_)
1515 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1518 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1520 /* Skip the entire case if /MISSING=LISTWISE is set */
1521 if ( cmd->miss == TTS_LISTWISE )
1523 for(i=0; i< cmd->n_variables ; ++i)
1525 struct variable *v = cmd->v_variables[i];
1526 const union value *val = case_data (c, v->fv);
1528 if (value_is_missing(&v->miss, val) )
1535 for(i=0; i< cmd->n_variables ; ++i)
1537 struct group_statistics *gs;
1538 struct variable *v = cmd->v_variables[i];
1539 const union value *val = case_data (c, v->fv);
1541 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1543 if ( ! value_is_missing(&v->miss, val))
1544 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1550 /* Pre calculations for one sample t test */
1552 one_sample_precalc ( struct cmd_t_test *cmd )
1556 for(i=0; i< cmd->n_variables ; ++i)
1558 struct group_statistics *gs;
1559 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1565 /* Post calculations for one sample t test */
1567 one_sample_postcalc (struct cmd_t_test *cmd)
1571 for(i=0; i< cmd->n_variables ; ++i)
1573 struct group_statistics *gs;
1574 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1576 gs->mean_diff = gs->sum_diff / gs->n ;
1583 paired_precalc (struct cmd_t_test *cmd UNUSED)
1587 for(i=0; i < n_pairs ; ++i )
1590 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1591 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1592 pairs[i].sum_of_prod = 0;
1593 pairs[i].correlation = 0;
1594 pairs[i].sum_of_diffs = 0;
1595 pairs[i].ssq_diffs = 0;
1602 paired_calc (const struct ccase *c, void *cmd_)
1606 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1608 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1610 /* Skip the entire case if /MISSING=LISTWISE is set ,
1611 AND one member of a pair is missing */
1612 if ( cmd->miss == TTS_LISTWISE )
1614 for(i=0; i < n_pairs ; ++i )
1616 struct variable *v0 = pairs[i].v[0];
1617 struct variable *v1 = pairs[i].v[1];
1619 const union value *val0 = case_data (c, v0->fv);
1620 const union value *val1 = case_data (c, v1->fv);
1622 if ( value_is_missing(&v0->miss, val0) ||
1623 value_is_missing(&v1->miss, val1) )
1630 for(i=0; i < n_pairs ; ++i )
1632 struct variable *v0 = pairs[i].v[0];
1633 struct variable *v1 = pairs[i].v[1];
1635 const union value *val0 = case_data (c, v0->fv);
1636 const union value *val1 = case_data (c, v1->fv);
1638 if ( ( !value_is_missing(&v0->miss, val0)
1639 && !value_is_missing(&v1->miss, val1) ) )
1641 pairs[i].n += weight;
1642 pairs[i].sum[0] += weight * val0->f;
1643 pairs[i].sum[1] += weight * val1->f;
1645 pairs[i].ssq[0] += weight * pow2(val0->f);
1646 pairs[i].ssq[1] += weight * pow2(val1->f);
1648 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1650 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1651 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1659 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1663 for(i=0; i < n_pairs ; ++i )
1666 const double n = pairs[i].n;
1668 for (j=0; j < 2 ; ++j)
1670 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1671 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1672 pow2(pairs[i].mean[j]))
1675 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1676 pow2(pairs[i].mean[j]))
1680 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1681 pairs[i].mean[0] * pairs[i].mean[1] ;
1682 /* correlation now actually contains the covariance */
1684 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1685 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1687 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1689 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1690 ( pairs[i].ssq_diffs / n )
1692 pow2(pairs[i].mean_diff )
1698 group_precalc (struct cmd_t_test *cmd )
1703 for(i=0; i< cmd->n_variables ; ++i)
1705 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1707 /* There's always 2 groups for a T - TEST */
1710 gp.indep_width = indep_var->width;
1712 ttpr->group_hash = hsh_create(2,
1713 (hsh_compare_func *) compare_group_binary,
1714 (hsh_hash_func *) hash_group_binary,
1715 (hsh_free_func *) free_group,
1718 for (j=0 ; j < 2 ; ++j)
1721 struct group_statistics *gs = xmalloc (sizeof *gs);
1727 if ( gp.criterion == CMP_EQ )
1729 gs->id = gp.v.g_value[j];
1734 gs->id.f = gp.v.critical_value - 1.0 ;
1736 gs->id.f = gp.v.critical_value + 1.0 ;
1739 hsh_insert ( ttpr->group_hash, (void *) gs );
1747 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1751 const union value *gv = case_data (c, indep_var->fv);
1753 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1755 if ( value_is_missing(&indep_var->miss, gv) )
1760 if ( cmd->miss == TTS_LISTWISE )
1762 for(i=0; i< cmd->n_variables ; ++i)
1764 struct variable *v = cmd->v_variables[i];
1765 const union value *val = case_data (c, v->fv);
1767 if (value_is_missing(&v->miss, val) )
1774 gv = case_data (c, indep_var->fv);
1776 for(i=0; i< cmd->n_variables ; ++i)
1778 struct variable *var = cmd->v_variables[i];
1779 const union value *val = case_data (c, var->fv);
1780 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1781 struct group_statistics *gs;
1783 gs = hsh_find(grp_hash, (void *) gv);
1785 /* If the independent variable doesn't match either of the values
1786 for this case then move on to the next case */
1790 if ( !value_is_missing(&var->miss, val) )
1793 gs->sum+=weight * val->f;
1794 gs->ssq+=weight * pow2(val->f);
1803 group_postcalc ( struct cmd_t_test *cmd )
1807 for(i=0; i< cmd->n_variables ; ++i)
1809 struct variable *var = cmd->v_variables[i];
1810 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1811 struct hsh_iterator g;
1812 struct group_statistics *gs;
1815 for (gs = hsh_first (grp_hash,&g);
1817 gs = hsh_next(grp_hash,&g))
1819 gs->mean = gs->sum / gs->n;
1821 gs->s_std_dev= sqrt(
1822 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1827 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1830 gs->se_mean = gs->std_dev / sqrt(gs->n);
1840 calculate(const struct ccase *first, const struct casefile *cf, void *cmd_)
1842 struct ssbox stat_summary_box;
1843 struct trbox test_results_box;
1845 struct casereader *r;
1848 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1850 output_split_file_values (first);
1851 common_precalc(cmd);
1852 for(r = casefile_get_reader (cf);
1853 casereader_read (r, &c) ;
1856 common_calc(&c,cmd);
1858 casereader_destroy (r);
1859 common_postcalc(cmd);
1864 one_sample_precalc(cmd);
1865 for(r = casefile_get_reader (cf);
1866 casereader_read (r, &c) ;
1869 one_sample_calc(&c,cmd);
1871 casereader_destroy (r);
1872 one_sample_postcalc(cmd);
1876 paired_precalc(cmd);
1877 for(r = casefile_get_reader (cf);
1878 casereader_read (r, &c) ;
1881 paired_calc(&c,cmd);
1883 casereader_destroy (r);
1884 paired_postcalc(cmd);
1890 for(r = casefile_get_reader (cf);
1891 casereader_read (r, &c) ;
1896 casereader_destroy (r);
1897 group_postcalc(cmd);
1899 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1900 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1905 ssbox_create(&stat_summary_box,cmd,mode);
1906 ssbox_populate(&stat_summary_box,cmd);
1907 ssbox_finalize(&stat_summary_box);
1909 if ( mode == T_PAIRED)
1912 trbox_create(&test_results_box,cmd,mode);
1913 trbox_populate(&test_results_box,cmd);
1914 trbox_finalize(&test_results_box);
1919 short which_group(const struct group_statistics *g,
1920 const struct group_properties *p);
1922 /* Return -1 if the id of a is less than b; +1 if greater than and
1925 compare_group_binary(const struct group_statistics *a,
1926 const struct group_statistics *b,
1927 const struct group_properties *p)
1932 if ( p->criterion == CMP_LE )
1934 /* less-than-or-equal comparision is not meaningfull for
1935 alpha variables, so we shouldn't ever arrive here */
1936 assert(p->indep_width == 0 ) ;
1938 flag_a = ( a->id.f < p->v.critical_value ) ;
1939 flag_b = ( b->id.f < p->v.critical_value ) ;
1943 flag_a = which_group(a, p);
1944 flag_b = which_group(b, p);
1947 if (flag_a < flag_b )
1950 return (flag_a > flag_b);
1953 /* This is a degenerate case of a hash, since it can only return three possible
1954 values. It's really a comparison, being used as a hash function */
1957 hash_group_binary(const struct group_statistics *g,
1958 const struct group_properties *p)
1962 if ( p->criterion == CMP_LE )
1964 /* Not meaningfull to do a less than compare for alpha values ? */
1965 assert(p->indep_width == 0 ) ;
1966 flag = ( g->id.f < p->v.critical_value ) ;
1968 else if ( p->criterion == CMP_EQ)
1970 flag = which_group(g,p);
1978 /* return 0 if G belongs to group 0,
1979 1 if it belongs to group 1,
1980 2 if it belongs to neither group */
1982 which_group(const struct group_statistics *g,
1983 const struct group_properties *p)
1986 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1989 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))