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>
38 #include "value-labels.h"
42 #include "group_proc.h"
50 variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
52 +missing=miss:!analysis/listwise,
53 incl:include/!exclude;
54 format=fmt:!labels/nolabels;
55 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
63 /* Function to use for testing for missing values */
64 static is_missing_func value_is_missing;
66 /* Variable for the GROUPS subcommand, if given. */
67 static struct variable *indep_var;
69 /* GROUPS: Number of values specified by the user; the values
73 struct group_properties
75 /* The comparison criterion */
76 enum comparison criterion;
78 /* The width of the independent variable */
82 /* The value of the independent variable at which groups are determined to
83 belong to one group or the other */
84 double critical_value;
87 /* The values of the independent variable for each group */
88 union value g_value[2];
94 static struct group_properties gp ;
98 /* PAIRS: Number of pairs to be compared ; each pair. */
99 static int n_pairs = 0 ;
102 /* The variables comprising the pair */
103 struct variable *v[2];
105 /* The number of valid variable pairs */
108 /* The sum of the members */
111 /* sum of squares of the members */
114 /* Std deviation of the members */
118 /* Sample Std deviation of the members */
121 /* The means of the members */
124 /* The correlation coefficient between the variables */
127 /* The sum of the differences */
130 /* The sum of the products */
133 /* The mean of the differences */
136 /* The sum of the squares of the differences */
139 /* The std deviation of the differences */
143 static struct pair *pairs=0;
145 static int parse_value (union value * v, int type) ;
147 /* Structures and Functions for the Statistics Summary Box */
149 typedef void populate_ssbox_func(struct ssbox *ssb,
150 struct cmd_t_test *cmd);
151 typedef void finalize_ssbox_func(struct ssbox *ssb);
157 populate_ssbox_func *populate;
158 finalize_ssbox_func *finalize;
163 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
165 /* Populate a ssbox according to cmd */
166 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
168 /* Submit and destroy a ssbox */
169 void ssbox_finalize(struct ssbox *ssb);
171 /* A function to create, populate and submit the Paired Samples Correlation
176 /* Structures and Functions for the Test Results Box */
179 typedef void populate_trbox_func(struct trbox *trb,
180 struct cmd_t_test *cmd);
181 typedef void finalize_trbox_func(struct trbox *trb);
185 populate_trbox_func *populate;
186 finalize_trbox_func *finalize;
190 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
192 /* Populate a ssbox according to cmd */
193 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
195 /* Submit and destroy a ssbox */
196 void trbox_finalize(struct trbox *trb);
198 /* Which mode was T-TEST invoked */
206 static int common_calc (const struct ccase *, void *);
207 static void common_precalc (struct cmd_t_test *);
208 static void common_postcalc (struct cmd_t_test *);
210 static int one_sample_calc (const struct ccase *, void *);
211 static void one_sample_precalc (struct cmd_t_test *);
212 static void one_sample_postcalc (struct cmd_t_test *);
214 static int paired_calc (const struct ccase *, void *);
215 static void paired_precalc (struct cmd_t_test *);
216 static void paired_postcalc (struct cmd_t_test *);
218 static void group_precalc (struct cmd_t_test *);
219 static int group_calc (const struct ccase *, struct cmd_t_test *);
220 static void group_postcalc (struct cmd_t_test *);
223 static int compare_var_name (const void *a_, const void *b_, void *v_ UNUSED);
224 static unsigned hash_var_name (const void *a_, void *v_ UNUSED);
226 static void calculate(const struct casefile *cf, void *_mode);
230 static struct cmd_t_test cmd;
232 static int bad_weight_warn;
235 static int compare_group_binary(const struct group_statistics *a,
236 const struct group_statistics *b,
237 struct group_properties *p);
240 static unsigned hash_group_binary(const struct group_statistics *g,
241 struct group_properties *p);
249 if ( !parse_t_test(&cmd) )
252 if (! cmd.sbc_criteria)
257 if (cmd.sbc_testval) ++m;
258 if (cmd.sbc_groups) ++m;
259 if (cmd.sbc_pairs) ++m;
264 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
272 else if (cmd.sbc_groups)
277 if ( mode == T_PAIRED)
279 if (cmd.sbc_variables)
281 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
286 /* Iterate through the pairs and put each variable that is a
287 member of a pair into cmd.v_variables */
290 struct hsh_iterator hi;
291 struct hsh_table *hash;
294 hash=hsh_create(n_pairs,compare_var_name,hash_var_name,0,0);
296 for (i=0; i < n_pairs; ++i)
298 hsh_insert(hash,pairs[i].v[0]);
299 hsh_insert(hash,pairs[i].v[1]);
302 assert(cmd.n_variables == 0);
303 cmd.n_variables = hsh_count(hash);
305 cmd.v_variables = xrealloc(cmd.v_variables,
306 sizeof(struct variable) * cmd.n_variables);
307 /* Iterate through the hash */
308 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
310 v=hsh_next(hash,&hi) )
311 cmd.v_variables[i++]=v;
316 else if ( !cmd.sbc_variables)
318 msg(SE, _("One or more VARIABLES must be specified."));
323 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
324 if (cmd.incl == TTS_INCLUDE )
325 value_is_missing = is_system_missing;
327 value_is_missing = is_missing;
331 multipass_procedure_with_splits (calculate, &cmd);
337 if ( mode == T_IND_SAMPLES)
340 /* Destroy any group statistics we created */
341 for (v = 0 ; v < cmd.n_variables ; ++v )
343 struct group_proc *grpp = &cmd.v_variables[v]->p.grp_data;
344 free(grpp->group_hash);
352 tts_custom_groups (struct cmd_t_test *cmd UNUSED)
354 int n_group_values=0;
358 if (token != T_ALL &&
359 (token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
362 msg(SE,_("`%s' is not a variable name"),tokid);
366 indep_var = parse_variable ();
369 lex_error ("expecting variable name in GROUPS subcommand");
373 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
375 msg (SE, _("Long string variable %s is not valid here."),
380 if (!lex_match ('('))
382 if (indep_var->type == NUMERIC)
384 gp.v.g_value[0].f = 1;
385 gp.v.g_value[1].f = 2;
387 gp.criterion = CMP_EQ;
395 msg (SE, _("When applying GROUPS to a string variable, at "
396 "least one value must be specified."));
401 if (!parse_value (&gp.v.g_value[0],indep_var->type))
407 gp.criterion = CMP_LE;
408 gp.v.critical_value = gp.v.g_value[0].f;
414 if (!parse_value (&gp.v.g_value[1],indep_var->type))
418 if (!lex_force_match (')'))
421 if ( n_group_values == 2 )
422 gp.criterion = CMP_EQ ;
424 gp.criterion = CMP_LE ;
432 tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
434 struct variable **vars;
439 int n_after_WITH = -1;
440 int paired ; /* Was the PAIRED keyword given ? */
444 if ((token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
447 msg(SE,_("`%s' is not a variable name"),tokid);
452 if (!parse_variables (default_dict, &vars, &n_vars,
453 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
461 if (lex_match (T_WITH))
463 n_before_WITH = n_vars;
464 if (!parse_variables (default_dict, &vars, &n_vars,
465 PV_DUPLICATE | PV_APPEND
466 | PV_NUMERIC | PV_NO_SCRATCH))
471 n_after_WITH = n_vars - n_before_WITH;
474 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
476 /* Determine the number of pairs needed */
479 if (n_before_WITH != n_after_WITH)
482 msg (SE, _("PAIRED was specified but the number of variables "
483 "preceding WITH (%d) did not match the number "
485 n_before_WITH, n_after_WITH );
488 n_pairs_local=n_before_WITH;
490 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
492 n_pairs_local=n_before_WITH * n_after_WITH ;
494 else /* Neither WITH nor PAIRED keyword given */
499 msg (SE, _("At least two variables must be specified "
504 /* how many ways can you pick 2 from n_vars ? */
505 n_pairs_local = n_vars * (n_vars -1 ) /2 ;
509 /* Allocate storage for the pairs */
510 pairs = xrealloc(pairs, sizeof(struct pair) * (n_pairs + n_pairs_local) );
512 /* Populate the pairs with the appropriate variables */
517 assert(n_pairs_local == n_vars/2);
518 for (i = 0; i < n_pairs_local ; ++i)
520 pairs[i].v[n_pairs+0] = vars[i];
521 pairs[i].v[n_pairs+1] = vars[i+n_pairs_local];
524 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
529 for(i=0 ; i < n_before_WITH ; ++i )
531 for(j=0 ; j < n_after_WITH ; ++j)
533 pairs[p].v[0] = vars[i];
534 pairs[p].v[1] = vars[j+n_before_WITH];
539 else /* Neither WITH nor PAIRED given */
544 for(i=0 ; i < n_vars ; ++i )
546 for(j=i+1 ; j < n_vars ; ++j)
548 pairs[p].v[0] = vars[i];
549 pairs[p].v[1] = vars[j];
555 n_pairs+=n_pairs_local;
561 /* Parses the current token (numeric or string, depending on type)
562 value v and returns success. */
564 parse_value (union value * v, int type )
568 if (!lex_force_num ())
574 if (!lex_force_string ())
576 strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
585 /* Implementation of the SSBOX object */
587 void ssbox_base_init(struct ssbox *this, int cols,int rows);
589 void ssbox_base_finalize(struct ssbox *ssb);
591 void ssbox_one_sample_init(struct ssbox *this,
592 struct cmd_t_test *cmd );
594 void ssbox_independent_samples_init(struct ssbox *this,
595 struct cmd_t_test *cmd);
597 void ssbox_paired_init(struct ssbox *this,
598 struct cmd_t_test *cmd);
601 /* Factory to create an ssbox */
603 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
608 ssbox_one_sample_init(ssb,cmd);
611 ssbox_independent_samples_init(ssb,cmd);
614 ssbox_paired_init(ssb,cmd);
623 /* Despatcher for the populate method */
625 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
627 ssb->populate(ssb,cmd);
631 /* Despatcher for finalize */
633 ssbox_finalize(struct ssbox *ssb)
639 /* Submit the box and clear up */
641 ssbox_base_finalize(struct ssbox *ssb)
648 /* Initialize a ssbox struct */
650 ssbox_base_init(struct ssbox *this, int cols,int rows)
652 this->finalize = ssbox_base_finalize;
653 this->t = tab_create (cols, rows, 0);
655 tab_columns (this->t, SOM_COL_DOWN, 1);
656 tab_headers (this->t,0,0,1,0);
657 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
658 tab_hline(this->t, TAL_2,0,cols-1,1);
659 tab_dim (this->t, tab_natural_dimensions);
662 void ssbox_one_sample_populate(struct ssbox *ssb,
663 struct cmd_t_test *cmd);
665 /* Initialize the one_sample ssbox */
667 ssbox_one_sample_init(struct ssbox *this,
668 struct cmd_t_test *cmd )
671 const int vsize=cmd->n_variables+1;
673 this->populate = ssbox_one_sample_populate;
675 ssbox_base_init(this, hsize,vsize);
676 tab_title (this->t, 0, _("One-Sample Statistics"));
677 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
678 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
679 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
680 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
681 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
684 void ssbox_independent_samples_populate(struct ssbox *ssb,
685 struct cmd_t_test *cmd);
687 /* Initialize the independent samples ssbox */
689 ssbox_independent_samples_init(struct ssbox *this,
690 struct cmd_t_test *cmd)
693 int vsize = cmd->n_variables*2 +1;
695 this->populate = ssbox_independent_samples_populate;
697 ssbox_base_init(this, hsize,vsize);
698 tab_title (this->t, 0, _("Group Statistics"));
699 tab_vline(this->t,0,1,0,vsize - 1);
700 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
701 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
702 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
703 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
704 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
708 /* Populate the ssbox for independent samples */
710 ssbox_independent_samples_populate(struct ssbox *ssb,
711 struct cmd_t_test *cmd)
717 double indep_value[2];
719 char prefix[2][3]={"",""};
721 if ( indep_var->type == NUMERIC )
723 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
724 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
728 val_lab0 = gp.v.g_value[0].s;
729 val_lab1 = gp.v.g_value[1].s;
732 if (gp.criterion == CMP_LE )
734 strcpy(prefix[0],"< ");
735 strcpy(prefix[1],">=");
736 indep_value[0] = gp.v.critical_value;
737 indep_value[1] = gp.v.critical_value;
741 indep_value[0] = gp.v.g_value[0].f;
742 indep_value[1] = gp.v.g_value[1].f;
747 for (i=0; i < cmd->n_variables; ++i)
749 struct variable *var = cmd->v_variables[i];
750 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
753 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
756 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
757 "%s%s", prefix[0], val_lab0);
759 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
760 "%s%g", prefix[0], indep_value[0]);
764 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
765 "%s%s", prefix[1], val_lab1);
767 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
768 "%s%g", prefix[1], indep_value[1]);
771 /* Fill in the group statistics */
772 for ( count = 0 ; count < 2 ; ++count )
774 union value search_val;
776 struct group_statistics *gs;
778 if ( gp.criterion == CMP_LE )
782 /* less than ( < ) case */
783 search_val.f = gp.v.critical_value - 1.0;
788 search_val.f = gp.v.critical_value + 1.0;
793 search_val = gp.v.g_value[count];
796 gs = hsh_find(grp_hash, (void *) &search_val);
799 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
800 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
801 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
802 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
808 void ssbox_paired_populate(struct ssbox *ssb,
809 struct cmd_t_test *cmd);
811 /* Initialize the paired values ssbox */
813 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
817 int vsize = n_pairs*2+1;
819 this->populate = ssbox_paired_populate;
821 ssbox_base_init(this, hsize,vsize);
822 tab_title (this->t, 0, _("Paired Sample Statistics"));
823 tab_vline(this->t,TAL_0,1,0,vsize-1);
824 tab_vline(this->t,TAL_2,2,0,vsize-1);
825 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
826 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
827 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
828 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
832 /* Populate the ssbox for paired values */
834 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
840 for (i=0; i < n_pairs; ++i)
844 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
846 for (j=0 ; j < 2 ; ++j)
848 struct group_statistics *gs;
850 gs=&pairs[i].v[j]->p.grp_data.ugs;
854 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
857 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
858 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
859 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
860 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
866 /* Populate the one sample ssbox */
868 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
874 for (i=0; i < cmd->n_variables; ++i)
876 struct group_statistics *gs;
877 gs= &cmd->v_variables[i]->p.grp_data.ugs;
879 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
880 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
881 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
882 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
883 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
890 /* Implementation of the Test Results box struct */
892 void trbox_base_init(struct trbox *self,int n_vars, int cols);
893 void trbox_base_finalize(struct trbox *trb);
895 void trbox_independent_samples_init(struct trbox *trb,
896 struct cmd_t_test *cmd );
898 void trbox_independent_samples_populate(struct trbox *trb,
899 struct cmd_t_test *cmd);
901 void trbox_one_sample_init(struct trbox *self,
902 struct cmd_t_test *cmd );
904 void trbox_one_sample_populate(struct trbox *trb,
905 struct cmd_t_test *cmd);
907 void trbox_paired_init(struct trbox *self,
908 struct cmd_t_test *cmd );
910 void trbox_paired_populate(struct trbox *trb,
911 struct cmd_t_test *cmd);
915 /* Create a trbox according to mode*/
917 trbox_create(struct trbox *trb,
918 struct cmd_t_test *cmd, int mode)
923 trbox_one_sample_init(trb,cmd);
926 trbox_independent_samples_init(trb,cmd);
929 trbox_paired_init(trb,cmd);
936 /* Populate a trbox according to cmd */
938 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
940 trb->populate(trb,cmd);
943 /* Submit and destroy a trbox */
945 trbox_finalize(struct trbox *trb)
950 /* Initialize the independent samples trbox */
952 trbox_independent_samples_init(struct trbox *self,
953 struct cmd_t_test *cmd UNUSED)
956 const int vsize=cmd->n_variables*2+3;
959 self->populate = trbox_independent_samples_populate;
961 trbox_base_init(self,cmd->n_variables*2,hsize);
962 tab_title(self->t,0,_("Independent Samples Test"));
963 tab_hline(self->t,TAL_1,2,hsize-1,1);
964 tab_vline(self->t,TAL_2,2,0,vsize-1);
965 tab_vline(self->t,TAL_1,4,0,vsize-1);
966 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
967 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
968 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
969 tab_joint_text(self->t, 2, 0, 3, 0,
970 TAB_CENTER,_("Levene's Test for Equality of Variances"));
971 tab_joint_text(self->t, 4,0,hsize-1,0,
972 TAB_CENTER,_("t-test for Equality of Means"));
974 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
975 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
976 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
977 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
978 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
979 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
980 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
981 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
982 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
984 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
985 _("%g%% Confidence Interval of the Difference"),
986 cmd->criteria*100.0);
990 /* Populate the independent samples trbox */
992 trbox_independent_samples_populate(struct trbox *self,
993 struct cmd_t_test *cmd )
998 for (i=0; i < cmd->n_variables; ++i)
1007 double pooled_variance;
1008 double std_err_diff;
1011 struct variable *var = cmd->v_variables[i];
1013 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
1015 struct group_statistics *gs0 ;
1016 struct group_statistics *gs1 ;
1018 union value search_val;
1020 if ( gp.criterion == CMP_LE )
1021 search_val.f = gp.v.critical_value - 1.0;
1023 search_val = gp.v.g_value[0];
1025 gs0 = hsh_find(grp_hash, (void *) &search_val);
1028 if ( gp.criterion == CMP_LE )
1029 search_val.f = gp.v.critical_value + 1.0;
1031 search_val = gp.v.g_value[1];
1033 gs1 = hsh_find(grp_hash, (void *) &search_val);
1037 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1039 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1042 tab_float(self->t, 2, i*2+3, TAB_CENTER,
1043 cmd->v_variables[i]->p.grp_data.levene, 8,3);
1045 /* Now work out the significance of the Levene test */
1046 df1 = 1; df2 = cmd->v_variables[i]->p.grp_data.ugs.n - 2;
1047 q = gsl_cdf_fdist_Q(cmd->v_variables[i]->p.grp_data.levene, df1, df2);
1049 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1051 df = gs0->n + gs1->n - 2.0 ;
1052 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1054 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1056 (gs1->n )*pow2(gs1->s_std_dev)
1059 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1060 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1062 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1064 p = gsl_cdf_tdist_P(t, df);
1065 q = gsl_cdf_tdist_Q(t, df);
1067 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1069 mean_diff = gs0->mean - gs1->mean;
1070 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1073 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1074 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1077 /* Now work out the confidence interval */
1078 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1080 t = gsl_cdf_tdist_Qinv(q,df);
1081 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1082 mean_diff - t * std_err_diff, 8, 3);
1084 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1085 mean_diff + t * std_err_diff, 8, 3);
1090 /* Now for the \sigma_1 != \sigma_2 case */
1091 tab_text (self->t, 1, i*2+3+1,
1092 TAB_LEFT, _("Equal variances not assumed"));
1095 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1096 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1098 t = mean_diff / sqrt(se2) ;
1099 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1102 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1106 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1110 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1112 p = gsl_cdf_tdist_P(t, df);
1113 q = gsl_cdf_tdist_Q(t, df);
1115 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1117 /* Now work out the confidence interval */
1118 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1120 t = gsl_cdf_tdist_Qinv(q, df);
1122 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1125 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1128 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1129 mean_diff - t * std_err_diff, 8, 3);
1131 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1132 mean_diff + t * std_err_diff, 8, 3);
1138 /* Initialize the paired samples trbox */
1140 trbox_paired_init(struct trbox *self,
1141 struct cmd_t_test *cmd UNUSED)
1145 const int vsize=n_pairs+3;
1147 self->populate = trbox_paired_populate;
1149 trbox_base_init(self,n_pairs,hsize);
1150 tab_title (self->t, 0, _("Paired Samples Test"));
1151 tab_hline(self->t,TAL_1,2,6,1);
1152 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1153 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1154 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1155 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1156 tab_hline(self->t,TAL_1,5,6, 2);
1157 tab_vline(self->t,TAL_0,6,0,1);
1159 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1160 _("%g%% Confidence Interval of the Difference"),
1161 cmd->criteria*100.0);
1163 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1164 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1165 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1166 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1167 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1168 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1169 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1170 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1173 /* Populate the paired samples trbox */
1175 trbox_paired_populate(struct trbox *trb,
1176 struct cmd_t_test *cmd UNUSED)
1180 for (i=0; i < n_pairs; ++i)
1185 double n = pairs[i].n;
1189 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1191 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1192 pairs[i].v[0]->name, pairs[i].v[1]->name);
1194 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1196 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1199 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1200 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1202 /* Now work out the confidence interval */
1203 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1205 t = gsl_cdf_tdist_Qinv(q, df);
1207 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1208 pairs[i].mean_diff - t * se_mean , 8, 4);
1210 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1211 pairs[i].mean_diff + t * se_mean , 8, 4);
1213 t = (pairs[i].mean[0] - pairs[i].mean[1])
1215 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1216 2 * pairs[i].correlation *
1217 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1221 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1223 /* Degrees of freedom */
1224 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1226 p = gsl_cdf_tdist_P(t,df);
1227 q = gsl_cdf_tdist_P(t,df);
1229 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1234 /* Initialize the one sample trbox */
1236 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1239 const int vsize=cmd->n_variables+3;
1241 self->populate = trbox_one_sample_populate;
1243 trbox_base_init(self, cmd->n_variables,hsize);
1244 tab_title (self->t, 0, _("One-Sample Test"));
1245 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1246 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1248 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1249 _("Test Value = %f"),cmd->n_testval);
1251 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1254 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1255 _("%g%% Confidence Interval of the Difference"),
1256 cmd->criteria*100.0);
1258 tab_vline(self->t,TAL_0,6,1,1);
1259 tab_hline(self->t,TAL_1,5,6,2);
1260 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1261 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1262 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1263 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1264 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1265 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1270 /* Populate the one sample trbox */
1272 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1278 for (i=0; i < cmd->n_variables; ++i)
1283 struct group_statistics *gs;
1284 gs= &cmd->v_variables[i]->p.grp_data.ugs;
1287 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1289 t = (gs->mean - cmd->n_testval ) * sqrt(gs->n) / gs->std_dev ;
1291 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1293 /* degrees of freedom */
1296 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1298 p = gsl_cdf_tdist_P(t, df);
1299 q = gsl_cdf_tdist_Q(t, df);
1301 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1303 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1305 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1308 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1309 t = gsl_cdf_tdist_Qinv(q, df);
1311 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1312 gs->mean_diff - t * gs->se_mean, 8,4);
1314 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1315 gs->mean_diff + t * gs->se_mean, 8,4);
1319 /* Base initializer for the generalized trbox */
1321 trbox_base_init(struct trbox *self, int data_rows, int cols)
1323 const int rows = 3 + data_rows;
1325 self->finalize = trbox_base_finalize;
1326 self->t = tab_create (cols, rows, 0);
1327 tab_headers (self->t,0,0,3,0);
1328 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1329 tab_hline(self->t, TAL_2,0,cols-1,3);
1330 tab_dim (self->t, tab_natural_dimensions);
1334 /* Base finalizer for the trbox */
1336 trbox_base_finalize(struct trbox *trb)
1342 /* Create , populate and submit the Paired Samples Correlation box */
1346 const int rows=1+n_pairs;
1350 struct tab_table *table;
1352 table = tab_create (cols,rows,0);
1354 tab_columns (table, SOM_COL_DOWN, 1);
1355 tab_headers (table,0,0,1,0);
1356 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1357 tab_hline(table, TAL_2, 0, cols - 1, 1);
1358 tab_vline(table, TAL_2, 2, 0, rows - 1);
1359 tab_dim(table, tab_natural_dimensions);
1360 tab_title(table, 0, _("Paired Samples Correlations"));
1362 /* column headings */
1363 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1364 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1365 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1367 for (i=0; i < n_pairs; ++i)
1371 double df = pairs[i].n -2;
1373 double correlation_t =
1374 pairs[i].correlation * sqrt(df) /
1375 sqrt(1 - pow2(pairs[i].correlation));
1379 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1382 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1383 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1387 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1388 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1390 p = gsl_cdf_tdist_P(correlation_t, df);
1391 q = gsl_cdf_tdist_Q(correlation_t, df);
1393 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1402 /* Calculation Implementation */
1404 /* Per case calculations common to all variants of the T test */
1406 common_calc (const struct ccase *c, void *_cmd)
1409 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1411 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1414 /* Skip the entire case if /MISSING=LISTWISE is set */
1415 if ( cmd->miss == TTS_LISTWISE )
1417 for(i=0; i< cmd->n_variables ; ++i)
1419 struct variable *v = cmd->v_variables[i];
1420 const union value *val = case_data (c, v->fv);
1422 if (value_is_missing(val,v) )
1429 /* Listwise has to be implicit if the independent variable is missing ?? */
1430 if ( cmd->sbc_groups )
1432 const union value *gv = case_data (c, indep_var->fv);
1433 if ( value_is_missing(gv,indep_var) )
1440 for(i=0; i< cmd->n_variables ; ++i)
1442 struct group_statistics *gs;
1443 struct variable *v = cmd->v_variables[i];
1444 const union value *val = case_data (c, v->fv);
1446 gs= &cmd->v_variables[i]->p.grp_data.ugs;
1448 if (! value_is_missing(val,v) )
1451 gs->sum+=weight * val->f;
1452 gs->ssq+=weight * val->f * val->f;
1458 /* Pre calculations common to all variants of the T test */
1460 common_precalc ( struct cmd_t_test *cmd )
1464 for(i=0; i< cmd->n_variables ; ++i)
1466 struct group_statistics *gs;
1467 gs= &cmd->v_variables[i]->p.grp_data.ugs;
1476 /* Post calculations common to all variants of the T test */
1478 common_postcalc ( struct cmd_t_test *cmd )
1483 for(i=0; i< cmd->n_variables ; ++i)
1485 struct group_statistics *gs;
1486 gs= &cmd->v_variables[i]->p.grp_data.ugs;
1488 gs->mean=gs->sum / gs->n;
1489 gs->s_std_dev= sqrt(
1490 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1495 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1498 gs->se_mean = gs->std_dev / sqrt(gs->n);
1499 gs->mean_diff= gs->sum_diff / gs->n;
1503 /* Per case calculations for one sample t test */
1505 one_sample_calc (const struct ccase *c, void *cmd_)
1508 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1511 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1513 /* Skip the entire case if /MISSING=LISTWISE is set */
1514 if ( cmd->miss == TTS_LISTWISE )
1516 for(i=0; i< cmd->n_variables ; ++i)
1518 struct variable *v = cmd->v_variables[i];
1519 const union value *val = case_data (c, v->fv);
1521 if (value_is_missing(val,v) )
1528 for(i=0; i< cmd->n_variables ; ++i)
1530 struct group_statistics *gs;
1531 struct variable *v = cmd->v_variables[i];
1532 const union value *val = case_data (c, v->fv);
1534 gs= &cmd->v_variables[i]->p.grp_data.ugs;
1536 if ( ! value_is_missing(val,v))
1537 gs->sum_diff += weight * (val->f - cmd->n_testval);
1543 /* Pre calculations for one sample t test */
1545 one_sample_precalc ( struct cmd_t_test *cmd )
1549 for(i=0; i< cmd->n_variables ; ++i)
1551 struct group_statistics *gs;
1552 gs= &cmd->v_variables[i]->p.grp_data.ugs;
1558 /* Post calculations for one sample t test */
1560 one_sample_postcalc (struct cmd_t_test *cmd)
1564 for(i=0; i< cmd->n_variables ; ++i)
1566 struct group_statistics *gs;
1567 gs= &cmd->v_variables[i]->p.grp_data.ugs;
1569 gs->mean_diff = gs->sum_diff / gs->n ;
1576 compare_var_name (const void *a_, const void *b_, void *v_ UNUSED)
1578 const struct variable *a = a_;
1579 const struct variable *b = b_;
1581 return strcmp(a->name,b->name);
1585 hash_var_name (const void *a_, void *v_ UNUSED)
1587 const struct variable *a = a_;
1589 return hsh_hash_bytes (a->name, strlen(a->name));
1595 paired_precalc (struct cmd_t_test *cmd UNUSED)
1599 for(i=0; i < n_pairs ; ++i )
1602 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1603 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1604 pairs[i].sum_of_prod = 0;
1605 pairs[i].correlation = 0;
1606 pairs[i].sum_of_diffs = 0;
1607 pairs[i].ssq_diffs = 0;
1614 paired_calc (const struct ccase *c, void *cmd_)
1618 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1620 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1622 /* Skip the entire case if /MISSING=LISTWISE is set ,
1623 AND one member of a pair is missing */
1624 if ( cmd->miss == TTS_LISTWISE )
1626 for(i=0; i < n_pairs ; ++i )
1628 struct variable *v0 = pairs[i].v[0];
1629 struct variable *v1 = pairs[i].v[1];
1631 const union value *val0 = case_data (c, v0->fv);
1632 const union value *val1 = case_data (c, v1->fv);
1634 if ( value_is_missing(val0,v0) ||
1635 value_is_missing(val1,v1) )
1642 for(i=0; i < n_pairs ; ++i )
1644 struct variable *v0 = pairs[i].v[0];
1645 struct variable *v1 = pairs[i].v[1];
1647 const union value *val0 = case_data (c, v0->fv);
1648 const union value *val1 = case_data (c, v1->fv);
1650 if ( ( !value_is_missing(val0,v0) && !value_is_missing(val1,v1) ) )
1652 pairs[i].n += weight;
1653 pairs[i].sum[0] += weight * val0->f;
1654 pairs[i].sum[1] += weight * val1->f;
1656 pairs[i].ssq[0] += weight * pow2(val0->f);
1657 pairs[i].ssq[1] += weight * pow2(val1->f);
1659 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1661 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1662 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1670 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1674 for(i=0; i < n_pairs ; ++i )
1677 const double n = pairs[i].n;
1679 for (j=0; j < 2 ; ++j)
1681 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1682 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1683 pow2(pairs[i].mean[j]))
1686 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1687 pow2(pairs[i].mean[j]))
1691 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1692 pairs[i].mean[0] * pairs[i].mean[1] ;
1693 /* correlation now actually contains the covariance */
1695 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1696 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1698 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1700 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1701 ( pairs[i].ssq_diffs / n )
1703 pow2(pairs[i].mean_diff )
1709 group_precalc (struct cmd_t_test *cmd )
1714 for(i=0; i< cmd->n_variables ; ++i)
1716 struct group_proc *ttpr = &cmd->v_variables[i]->p.grp_data;
1718 /* There's always 2 groups for a T - TEST */
1721 gp.indep_width = indep_var->width;
1723 ttpr->group_hash = hsh_create(2,
1724 (hsh_compare_func *) compare_group_binary,
1725 (hsh_hash_func *) hash_group_binary,
1726 (hsh_free_func *) free_group,
1729 for (j=0 ; j < 2 ; ++j)
1732 struct group_statistics *gs = (struct group_statistics *)
1733 xmalloc (sizeof(struct group_statistics));
1739 if ( gp.criterion == CMP_EQ )
1741 gs->id = gp.v.g_value[j];
1746 gs->id.f = gp.v.critical_value - 1.0 ;
1748 gs->id.f = gp.v.critical_value + 1.0 ;
1751 hsh_insert ( ttpr->group_hash, (void *) gs );
1759 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1763 const union value *gv = case_data (c, indep_var->fv);
1765 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1767 if ( value_is_missing(gv,indep_var) )
1772 if ( cmd->miss == TTS_LISTWISE )
1774 for(i=0; i< cmd->n_variables ; ++i)
1776 struct variable *v = cmd->v_variables[i];
1777 const union value *val = case_data (c, v->fv);
1779 if (value_is_missing(val,v) )
1786 gv = case_data (c, indep_var->fv);
1788 for(i=0; i< cmd->n_variables ; ++i)
1790 struct variable *var = cmd->v_variables[i];
1791 const union value *val = case_data (c, var->fv);
1792 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
1793 struct group_statistics *gs;
1795 gs = hsh_find(grp_hash, (void *) gv);
1797 /* If the independent variable doesn't match either of the values
1798 for this case then move on to the next case */
1802 if ( !value_is_missing(val,var) )
1805 gs->sum+=weight * val->f;
1806 gs->ssq+=weight * pow2(val->f);
1815 group_postcalc ( struct cmd_t_test *cmd )
1819 for(i=0; i< cmd->n_variables ; ++i)
1821 struct variable *var = cmd->v_variables[i];
1822 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
1823 struct hsh_iterator g;
1824 struct group_statistics *gs;
1827 for (gs = hsh_first (grp_hash,&g);
1829 gs = hsh_next(grp_hash,&g))
1831 gs->mean = gs->sum / gs->n;
1833 gs->s_std_dev= sqrt(
1834 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1839 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1842 gs->se_mean = gs->std_dev / sqrt(gs->n);
1852 calculate(const struct casefile *cf, void *cmd_)
1854 struct ssbox stat_summary_box;
1855 struct trbox test_results_box;
1857 struct casereader *r;
1860 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1862 common_precalc(cmd);
1863 for(r = casefile_get_reader (cf);
1864 casereader_read (r, &c) ;
1867 common_calc(&c,cmd);
1869 casereader_destroy (r);
1870 common_postcalc(cmd);
1875 one_sample_precalc(cmd);
1876 for(r = casefile_get_reader (cf);
1877 casereader_read (r, &c) ;
1880 one_sample_calc(&c,cmd);
1882 casereader_destroy (r);
1883 one_sample_postcalc(cmd);
1887 paired_precalc(cmd);
1888 for(r = casefile_get_reader (cf);
1889 casereader_read (r, &c) ;
1892 paired_calc(&c,cmd);
1894 casereader_destroy (r);
1895 paired_postcalc(cmd);
1901 for(r = casefile_get_reader (cf);
1902 casereader_read (r, &c) ;
1907 casereader_destroy (r);
1908 group_postcalc(cmd);
1910 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1911 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1916 ssbox_create(&stat_summary_box,cmd,mode);
1917 ssbox_populate(&stat_summary_box,cmd);
1918 ssbox_finalize(&stat_summary_box);
1920 if ( mode == T_PAIRED)
1923 trbox_create(&test_results_box,cmd,mode);
1924 trbox_populate(&test_results_box,cmd);
1925 trbox_finalize(&test_results_box);
1930 /* Return -1 if the id of a is less than b; +1 if greater than and
1933 compare_group_binary(const struct group_statistics *a,
1934 const struct group_statistics *b,
1935 struct group_properties *p)
1941 assert(p->indep_width == 0 ) ;
1943 if ( p->criterion == CMP_LE )
1945 flag_a = ( a->id.f < p->v.critical_value ) ;
1946 flag_b = ( b->id.f < p->v.critical_value ) ;
1950 flag_a = ( a->id.f == p->v.critical_value ) ;
1951 flag_b = ( b->id.f == p->v.critical_value ) ;
1955 if ( flag_a == flag_b)
1958 return ( flag_a < flag_b);
1962 hash_group_binary(const struct group_statistics *g, struct group_properties *p)
1966 assert(p->indep_width == 0 ) ;
1968 /* FIXME: should compare union values */
1969 if ( p->criterion == CMP_LE )
1971 flag = ( g->id.f < p->v.critical_value ) ;
1973 else if ( p->criterion == CMP_EQ)
1975 if ( g->id.f == p->v.g_value[0].f )
1977 else if ( g->id.f == p->v.g_value[1].f )