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;
75 struct group_properties
77 /* The comparison criterion */
78 enum comparison criterion;
80 /* The width of the independent variable */
84 /* The value of the independent variable at which groups are determined to
85 belong to one group or the other */
86 double critical_value;
89 /* The values of the independent variable for each group */
90 union value g_value[2];
96 static struct group_properties gp ;
100 /* PAIRS: Number of pairs to be compared ; each pair. */
101 static int n_pairs = 0 ;
104 /* The variables comprising the pair */
105 struct variable *v[2];
107 /* The number of valid variable pairs */
110 /* The sum of the members */
113 /* sum of squares of the members */
116 /* Std deviation of the members */
120 /* Sample Std deviation of the members */
123 /* The means of the members */
126 /* The correlation coefficient between the variables */
129 /* The sum of the differences */
132 /* The sum of the products */
135 /* The mean of the differences */
138 /* The sum of the squares of the differences */
141 /* The std deviation of the differences */
145 static struct pair *pairs=0;
147 static int parse_value (union value * v, int type) ;
149 /* Structures and Functions for the Statistics Summary Box */
151 typedef void populate_ssbox_func(struct ssbox *ssb,
152 struct cmd_t_test *cmd);
153 typedef void finalize_ssbox_func(struct ssbox *ssb);
159 populate_ssbox_func *populate;
160 finalize_ssbox_func *finalize;
165 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
167 /* Populate a ssbox according to cmd */
168 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
170 /* Submit and destroy a ssbox */
171 void ssbox_finalize(struct ssbox *ssb);
173 /* A function to create, populate and submit the Paired Samples Correlation
178 /* Structures and Functions for the Test Results Box */
181 typedef void populate_trbox_func(struct trbox *trb,
182 struct cmd_t_test *cmd);
183 typedef void finalize_trbox_func(struct trbox *trb);
187 populate_trbox_func *populate;
188 finalize_trbox_func *finalize;
192 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
194 /* Populate a ssbox according to cmd */
195 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
197 /* Submit and destroy a ssbox */
198 void trbox_finalize(struct trbox *trb);
200 /* Which mode was T-TEST invoked */
208 static int common_calc (const struct ccase *, void *);
209 static void common_precalc (struct cmd_t_test *);
210 static void common_postcalc (struct cmd_t_test *);
212 static int one_sample_calc (const struct ccase *, void *);
213 static void one_sample_precalc (struct cmd_t_test *);
214 static void one_sample_postcalc (struct cmd_t_test *);
216 static int paired_calc (const struct ccase *, void *);
217 static void paired_precalc (struct cmd_t_test *);
218 static void paired_postcalc (struct cmd_t_test *);
220 static void group_precalc (struct cmd_t_test *);
221 static int group_calc (const struct ccase *, struct cmd_t_test *);
222 static void group_postcalc (struct cmd_t_test *);
225 static int compare_var_name (const void *a_, const void *b_, void *v_ UNUSED);
226 static unsigned hash_var_name (const void *a_, void *v_ UNUSED);
228 static void calculate(const struct casefile *cf, void *_mode);
232 static struct cmd_t_test cmd;
234 static int bad_weight_warn;
237 static int compare_group_binary(const struct group_statistics *a,
238 const struct group_statistics *b,
239 struct group_properties *p);
242 static unsigned hash_group_binary(const struct group_statistics *g,
243 struct group_properties *p);
251 if ( !parse_t_test(&cmd) )
254 if (! cmd.sbc_criteria)
259 if (cmd.sbc_testval) ++m;
260 if (cmd.sbc_groups) ++m;
261 if (cmd.sbc_pairs) ++m;
266 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
274 else if (cmd.sbc_groups)
279 if ( mode == T_PAIRED)
281 if (cmd.sbc_variables)
283 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
288 /* Iterate through the pairs and put each variable that is a
289 member of a pair into cmd.v_variables */
292 struct hsh_iterator hi;
293 struct hsh_table *hash;
296 hash=hsh_create(n_pairs,compare_var_name,hash_var_name,0,0);
298 for (i=0; i < n_pairs; ++i)
300 hsh_insert(hash,pairs[i].v[0]);
301 hsh_insert(hash,pairs[i].v[1]);
304 assert(cmd.n_variables == 0);
305 cmd.n_variables = hsh_count(hash);
307 cmd.v_variables = xrealloc(cmd.v_variables,
308 sizeof(struct variable) * cmd.n_variables);
309 /* Iterate through the hash */
310 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
312 v=hsh_next(hash,&hi) )
313 cmd.v_variables[i++]=v;
318 else if ( !cmd.sbc_variables)
320 msg(SE, _("One or more VARIABLES must be specified."));
325 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
326 if (cmd.incl == TTS_INCLUDE )
327 value_is_missing = is_system_missing;
329 value_is_missing = is_missing;
333 multipass_procedure_with_splits (calculate, &cmd);
339 if ( mode == T_IND_SAMPLES)
342 /* Destroy any group statistics we created */
343 for (v = 0 ; v < cmd.n_variables ; ++v )
345 struct group_proc *grpp = &cmd.v_variables[v]->p.grp_data;
346 free(grpp->group_hash);
354 tts_custom_groups (struct cmd_t_test *cmd UNUSED)
356 int n_group_values=0;
360 if (token != T_ALL &&
361 (token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
364 msg(SE,_("`%s' is not a variable name"),tokid);
368 indep_var = parse_variable ();
371 lex_error ("expecting variable name in GROUPS subcommand");
375 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
377 msg (SE, _("Long string variable %s is not valid here."),
382 if (!lex_match ('('))
384 if (indep_var->type == NUMERIC)
386 gp.v.g_value[0].f = 1;
387 gp.v.g_value[1].f = 2;
389 gp.criterion = CMP_EQ;
397 msg (SE, _("When applying GROUPS to a string variable, at "
398 "least one value must be specified."));
403 if (!parse_value (&gp.v.g_value[0],indep_var->type))
409 gp.criterion = CMP_LE;
410 gp.v.critical_value = gp.v.g_value[0].f;
416 if (!parse_value (&gp.v.g_value[1],indep_var->type))
420 if (!lex_force_match (')'))
423 if ( n_group_values == 2 )
424 gp.criterion = CMP_EQ ;
426 gp.criterion = CMP_LE ;
434 tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
436 struct variable **vars;
441 int n_after_WITH = -1;
442 int paired ; /* Was the PAIRED keyword given ? */
446 if ((token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
449 msg(SE,_("`%s' is not a variable name"),tokid);
454 if (!parse_variables (default_dict, &vars, &n_vars,
455 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
463 if (lex_match (T_WITH))
465 n_before_WITH = n_vars;
466 if (!parse_variables (default_dict, &vars, &n_vars,
467 PV_DUPLICATE | PV_APPEND
468 | PV_NUMERIC | PV_NO_SCRATCH))
473 n_after_WITH = n_vars - n_before_WITH;
476 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
478 /* Determine the number of pairs needed */
481 if (n_before_WITH != n_after_WITH)
484 msg (SE, _("PAIRED was specified but the number of variables "
485 "preceding WITH (%d) did not match the number "
487 n_before_WITH, n_after_WITH );
490 n_pairs_local=n_before_WITH;
492 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
494 n_pairs_local=n_before_WITH * n_after_WITH ;
496 else /* Neither WITH nor PAIRED keyword given */
501 msg (SE, _("At least two variables must be specified "
506 /* how many ways can you pick 2 from n_vars ? */
507 n_pairs_local = n_vars * (n_vars -1 ) /2 ;
511 /* Allocate storage for the pairs */
512 pairs = xrealloc(pairs, sizeof(struct pair) * (n_pairs + n_pairs_local) );
514 /* Populate the pairs with the appropriate variables */
519 assert(n_pairs_local == n_vars/2);
520 for (i = 0; i < n_pairs_local ; ++i)
522 pairs[i].v[n_pairs+0] = vars[i];
523 pairs[i].v[n_pairs+1] = vars[i+n_pairs_local];
526 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
531 for(i=0 ; i < n_before_WITH ; ++i )
533 for(j=0 ; j < n_after_WITH ; ++j)
535 pairs[p].v[0] = vars[i];
536 pairs[p].v[1] = vars[j+n_before_WITH];
541 else /* Neither WITH nor PAIRED given */
546 for(i=0 ; i < n_vars ; ++i )
548 for(j=i+1 ; j < n_vars ; ++j)
550 pairs[p].v[0] = vars[i];
551 pairs[p].v[1] = vars[j];
557 n_pairs+=n_pairs_local;
563 /* Parses the current token (numeric or string, depending on type)
564 value v and returns success. */
566 parse_value (union value * v, int type )
570 if (!lex_force_num ())
576 if (!lex_force_string ())
578 strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
587 /* Implementation of the SSBOX object */
589 void ssbox_base_init(struct ssbox *this, int cols,int rows);
591 void ssbox_base_finalize(struct ssbox *ssb);
593 void ssbox_one_sample_init(struct ssbox *this,
594 struct cmd_t_test *cmd );
596 void ssbox_independent_samples_init(struct ssbox *this,
597 struct cmd_t_test *cmd);
599 void ssbox_paired_init(struct ssbox *this,
600 struct cmd_t_test *cmd);
603 /* Factory to create an ssbox */
605 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
610 ssbox_one_sample_init(ssb,cmd);
613 ssbox_independent_samples_init(ssb,cmd);
616 ssbox_paired_init(ssb,cmd);
625 /* Despatcher for the populate method */
627 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
629 ssb->populate(ssb,cmd);
633 /* Despatcher for finalize */
635 ssbox_finalize(struct ssbox *ssb)
641 /* Submit the box and clear up */
643 ssbox_base_finalize(struct ssbox *ssb)
650 /* Initialize a ssbox struct */
652 ssbox_base_init(struct ssbox *this, int cols,int rows)
654 this->finalize = ssbox_base_finalize;
655 this->t = tab_create (cols, rows, 0);
657 tab_columns (this->t, SOM_COL_DOWN, 1);
658 tab_headers (this->t,0,0,1,0);
659 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
660 tab_hline(this->t, TAL_2,0,cols-1,1);
661 tab_dim (this->t, tab_natural_dimensions);
664 void ssbox_one_sample_populate(struct ssbox *ssb,
665 struct cmd_t_test *cmd);
667 /* Initialize the one_sample ssbox */
669 ssbox_one_sample_init(struct ssbox *this,
670 struct cmd_t_test *cmd )
673 const int vsize=cmd->n_variables+1;
675 this->populate = ssbox_one_sample_populate;
677 ssbox_base_init(this, hsize,vsize);
678 tab_title (this->t, 0, _("One-Sample Statistics"));
679 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
680 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
681 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
682 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
683 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
686 void ssbox_independent_samples_populate(struct ssbox *ssb,
687 struct cmd_t_test *cmd);
689 /* Initialize the independent samples ssbox */
691 ssbox_independent_samples_init(struct ssbox *this,
692 struct cmd_t_test *cmd)
695 int vsize = cmd->n_variables*2 +1;
697 this->populate = ssbox_independent_samples_populate;
699 ssbox_base_init(this, hsize,vsize);
700 tab_title (this->t, 0, _("Group Statistics"));
701 tab_vline(this->t,0,1,0,vsize - 1);
702 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
703 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
704 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
705 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
706 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
710 /* Populate the ssbox for independent samples */
712 ssbox_independent_samples_populate(struct ssbox *ssb,
713 struct cmd_t_test *cmd)
719 double indep_value[2];
721 char prefix[2][3]={"",""};
723 if ( indep_var->type == NUMERIC )
725 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
726 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
730 val_lab0 = gp.v.g_value[0].s;
731 val_lab1 = gp.v.g_value[1].s;
734 if (gp.criterion == CMP_LE )
736 strcpy(prefix[0],"< ");
737 strcpy(prefix[1],">=");
738 indep_value[0] = gp.v.critical_value;
739 indep_value[1] = gp.v.critical_value;
743 indep_value[0] = gp.v.g_value[0].f;
744 indep_value[1] = gp.v.g_value[1].f;
749 for (i=0; i < cmd->n_variables; ++i)
751 struct variable *var = cmd->v_variables[i];
752 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
755 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
758 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
759 "%s%s", prefix[0], val_lab0);
761 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
762 "%s%g", prefix[0], indep_value[0]);
766 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
767 "%s%s", prefix[1], val_lab1);
769 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
770 "%s%g", prefix[1], indep_value[1]);
773 /* Fill in the group statistics */
774 for ( count = 0 ; count < 2 ; ++count )
776 union value search_val;
778 struct group_statistics *gs;
780 if ( gp.criterion == CMP_LE )
784 /* less than ( < ) case */
785 search_val.f = gp.v.critical_value - 1.0;
790 search_val.f = gp.v.critical_value + 1.0;
795 search_val = gp.v.g_value[count];
798 gs = hsh_find(grp_hash, (void *) &search_val);
801 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
802 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
803 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
804 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
810 void ssbox_paired_populate(struct ssbox *ssb,
811 struct cmd_t_test *cmd);
813 /* Initialize the paired values ssbox */
815 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
819 int vsize = n_pairs*2+1;
821 this->populate = ssbox_paired_populate;
823 ssbox_base_init(this, hsize,vsize);
824 tab_title (this->t, 0, _("Paired Sample Statistics"));
825 tab_vline(this->t,TAL_0,1,0,vsize-1);
826 tab_vline(this->t,TAL_2,2,0,vsize-1);
827 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
828 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
829 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
830 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
834 /* Populate the ssbox for paired values */
836 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
842 for (i=0; i < n_pairs; ++i)
846 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
848 for (j=0 ; j < 2 ; ++j)
850 struct group_statistics *gs;
852 gs=&pairs[i].v[j]->p.grp_data.ugs;
856 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
859 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
860 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
861 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
862 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
868 /* Populate the one sample ssbox */
870 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
876 for (i=0; i < cmd->n_variables; ++i)
878 struct group_statistics *gs;
879 gs= &cmd->v_variables[i]->p.grp_data.ugs;
881 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
882 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
883 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
884 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
885 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
892 /* Implementation of the Test Results box struct */
894 void trbox_base_init(struct trbox *self,int n_vars, int cols);
895 void trbox_base_finalize(struct trbox *trb);
897 void trbox_independent_samples_init(struct trbox *trb,
898 struct cmd_t_test *cmd );
900 void trbox_independent_samples_populate(struct trbox *trb,
901 struct cmd_t_test *cmd);
903 void trbox_one_sample_init(struct trbox *self,
904 struct cmd_t_test *cmd );
906 void trbox_one_sample_populate(struct trbox *trb,
907 struct cmd_t_test *cmd);
909 void trbox_paired_init(struct trbox *self,
910 struct cmd_t_test *cmd );
912 void trbox_paired_populate(struct trbox *trb,
913 struct cmd_t_test *cmd);
917 /* Create a trbox according to mode*/
919 trbox_create(struct trbox *trb,
920 struct cmd_t_test *cmd, int mode)
925 trbox_one_sample_init(trb,cmd);
928 trbox_independent_samples_init(trb,cmd);
931 trbox_paired_init(trb,cmd);
938 /* Populate a trbox according to cmd */
940 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
942 trb->populate(trb,cmd);
945 /* Submit and destroy a trbox */
947 trbox_finalize(struct trbox *trb)
952 /* Initialize the independent samples trbox */
954 trbox_independent_samples_init(struct trbox *self,
955 struct cmd_t_test *cmd UNUSED)
958 const int vsize=cmd->n_variables*2+3;
961 self->populate = trbox_independent_samples_populate;
963 trbox_base_init(self,cmd->n_variables*2,hsize);
964 tab_title(self->t,0,_("Independent Samples Test"));
965 tab_hline(self->t,TAL_1,2,hsize-1,1);
966 tab_vline(self->t,TAL_2,2,0,vsize-1);
967 tab_vline(self->t,TAL_1,4,0,vsize-1);
968 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
969 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
970 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
971 tab_joint_text(self->t, 2, 0, 3, 0,
972 TAB_CENTER,_("Levene's Test for Equality of Variances"));
973 tab_joint_text(self->t, 4,0,hsize-1,0,
974 TAB_CENTER,_("t-test for Equality of Means"));
976 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
977 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
978 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
979 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
980 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
981 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
982 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
983 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
984 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
986 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
987 _("%g%% Confidence Interval of the Difference"),
988 cmd->criteria*100.0);
992 /* Populate the independent samples trbox */
994 trbox_independent_samples_populate(struct trbox *self,
995 struct cmd_t_test *cmd )
1000 for (i=0; i < cmd->n_variables; ++i)
1009 double pooled_variance;
1010 double std_err_diff;
1013 struct variable *var = cmd->v_variables[i];
1015 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
1017 struct group_statistics *gs0 ;
1018 struct group_statistics *gs1 ;
1020 union value search_val;
1022 if ( gp.criterion == CMP_LE )
1023 search_val.f = gp.v.critical_value - 1.0;
1025 search_val = gp.v.g_value[0];
1027 gs0 = hsh_find(grp_hash, (void *) &search_val);
1030 if ( gp.criterion == CMP_LE )
1031 search_val.f = gp.v.critical_value + 1.0;
1033 search_val = gp.v.g_value[1];
1035 gs1 = hsh_find(grp_hash, (void *) &search_val);
1039 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1041 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1044 tab_float(self->t, 2, i*2+3, TAB_CENTER,
1045 cmd->v_variables[i]->p.grp_data.levene, 8,3);
1047 /* Now work out the significance of the Levene test */
1048 df1 = 1; df2 = cmd->v_variables[i]->p.grp_data.ugs.n - 2;
1049 q = gsl_cdf_fdist_Q(cmd->v_variables[i]->p.grp_data.levene, df1, df2);
1051 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1053 df = gs0->n + gs1->n - 2.0 ;
1054 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1056 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1058 (gs1->n )*pow2(gs1->s_std_dev)
1061 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1062 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1064 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1066 p = gsl_cdf_tdist_P(t, df);
1067 q = gsl_cdf_tdist_Q(t, df);
1069 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1071 mean_diff = gs0->mean - gs1->mean;
1072 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1075 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1076 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1079 /* Now work out the confidence interval */
1080 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1082 t = gsl_cdf_tdist_Qinv(q,df);
1083 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1084 mean_diff - t * std_err_diff, 8, 3);
1086 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1087 mean_diff + t * std_err_diff, 8, 3);
1092 /* Now for the \sigma_1 != \sigma_2 case */
1093 tab_text (self->t, 1, i*2+3+1,
1094 TAB_LEFT, _("Equal variances not assumed"));
1097 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1098 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1100 t = mean_diff / sqrt(se2) ;
1101 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1104 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1108 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1112 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1114 p = gsl_cdf_tdist_P(t, df);
1115 q = gsl_cdf_tdist_Q(t, df);
1117 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1119 /* Now work out the confidence interval */
1120 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1122 t = gsl_cdf_tdist_Qinv(q, df);
1124 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1127 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1130 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1131 mean_diff - t * std_err_diff, 8, 3);
1133 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1134 mean_diff + t * std_err_diff, 8, 3);
1140 /* Initialize the paired samples trbox */
1142 trbox_paired_init(struct trbox *self,
1143 struct cmd_t_test *cmd UNUSED)
1147 const int vsize=n_pairs+3;
1149 self->populate = trbox_paired_populate;
1151 trbox_base_init(self,n_pairs,hsize);
1152 tab_title (self->t, 0, _("Paired Samples Test"));
1153 tab_hline(self->t,TAL_1,2,6,1);
1154 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1155 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1156 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1157 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1158 tab_hline(self->t,TAL_1,5,6, 2);
1159 tab_vline(self->t,TAL_0,6,0,1);
1161 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1162 _("%g%% Confidence Interval of the Difference"),
1163 cmd->criteria*100.0);
1165 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1166 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1167 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1168 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1169 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1170 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1171 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1172 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1175 /* Populate the paired samples trbox */
1177 trbox_paired_populate(struct trbox *trb,
1178 struct cmd_t_test *cmd UNUSED)
1182 for (i=0; i < n_pairs; ++i)
1187 double n = pairs[i].n;
1191 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1193 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1194 pairs[i].v[0]->name, pairs[i].v[1]->name);
1196 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1198 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1201 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1202 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1204 /* Now work out the confidence interval */
1205 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1207 t = gsl_cdf_tdist_Qinv(q, df);
1209 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1210 pairs[i].mean_diff - t * se_mean , 8, 4);
1212 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1213 pairs[i].mean_diff + t * se_mean , 8, 4);
1215 t = (pairs[i].mean[0] - pairs[i].mean[1])
1217 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1218 2 * pairs[i].correlation *
1219 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1223 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1225 /* Degrees of freedom */
1226 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1228 p = gsl_cdf_tdist_P(t,df);
1229 q = gsl_cdf_tdist_P(t,df);
1231 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1236 /* Initialize the one sample trbox */
1238 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1241 const int vsize=cmd->n_variables+3;
1243 self->populate = trbox_one_sample_populate;
1245 trbox_base_init(self, cmd->n_variables,hsize);
1246 tab_title (self->t, 0, _("One-Sample Test"));
1247 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1248 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1250 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1251 _("Test Value = %f"), cmd->n_testval[0]);
1253 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1256 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1257 _("%g%% Confidence Interval of the Difference"),
1258 cmd->criteria*100.0);
1260 tab_vline(self->t,TAL_0,6,1,1);
1261 tab_hline(self->t,TAL_1,5,6,2);
1262 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1263 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1264 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1265 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1266 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1267 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1272 /* Populate the one sample trbox */
1274 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1280 for (i=0; i < cmd->n_variables; ++i)
1285 struct group_statistics *gs;
1286 gs= &cmd->v_variables[i]->p.grp_data.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, int data_rows, int cols)
1325 const int 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(val,v) )
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(gv,indep_var) )
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= &cmd->v_variables[i]->p.grp_data.ugs;
1450 if (! value_is_missing(val,v) )
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= &cmd->v_variables[i]->p.grp_data.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= &cmd->v_variables[i]->p.grp_data.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(val,v) )
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= &cmd->v_variables[i]->p.grp_data.ugs;
1538 if ( ! value_is_missing(val,v))
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= &cmd->v_variables[i]->p.grp_data.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= &cmd->v_variables[i]->p.grp_data.ugs;
1571 gs->mean_diff = gs->sum_diff / gs->n ;
1578 compare_var_name (const void *a_, const void *b_, void *v_ UNUSED)
1580 const struct variable *a = a_;
1581 const struct variable *b = b_;
1583 return strcmp(a->name,b->name);
1587 hash_var_name (const void *a_, void *v_ UNUSED)
1589 const struct variable *a = a_;
1591 return hsh_hash_bytes (a->name, strlen(a->name));
1597 paired_precalc (struct cmd_t_test *cmd UNUSED)
1601 for(i=0; i < n_pairs ; ++i )
1604 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1605 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1606 pairs[i].sum_of_prod = 0;
1607 pairs[i].correlation = 0;
1608 pairs[i].sum_of_diffs = 0;
1609 pairs[i].ssq_diffs = 0;
1616 paired_calc (const struct ccase *c, void *cmd_)
1620 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1622 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1624 /* Skip the entire case if /MISSING=LISTWISE is set ,
1625 AND one member of a pair is missing */
1626 if ( cmd->miss == TTS_LISTWISE )
1628 for(i=0; i < n_pairs ; ++i )
1630 struct variable *v0 = pairs[i].v[0];
1631 struct variable *v1 = pairs[i].v[1];
1633 const union value *val0 = case_data (c, v0->fv);
1634 const union value *val1 = case_data (c, v1->fv);
1636 if ( value_is_missing(val0,v0) ||
1637 value_is_missing(val1,v1) )
1644 for(i=0; i < n_pairs ; ++i )
1646 struct variable *v0 = pairs[i].v[0];
1647 struct variable *v1 = pairs[i].v[1];
1649 const union value *val0 = case_data (c, v0->fv);
1650 const union value *val1 = case_data (c, v1->fv);
1652 if ( ( !value_is_missing(val0,v0) && !value_is_missing(val1,v1) ) )
1654 pairs[i].n += weight;
1655 pairs[i].sum[0] += weight * val0->f;
1656 pairs[i].sum[1] += weight * val1->f;
1658 pairs[i].ssq[0] += weight * pow2(val0->f);
1659 pairs[i].ssq[1] += weight * pow2(val1->f);
1661 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1663 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1664 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1672 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1676 for(i=0; i < n_pairs ; ++i )
1679 const double n = pairs[i].n;
1681 for (j=0; j < 2 ; ++j)
1683 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1684 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1685 pow2(pairs[i].mean[j]))
1688 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1689 pow2(pairs[i].mean[j]))
1693 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1694 pairs[i].mean[0] * pairs[i].mean[1] ;
1695 /* correlation now actually contains the covariance */
1697 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1698 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1700 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1702 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1703 ( pairs[i].ssq_diffs / n )
1705 pow2(pairs[i].mean_diff )
1711 group_precalc (struct cmd_t_test *cmd )
1716 for(i=0; i< cmd->n_variables ; ++i)
1718 struct group_proc *ttpr = &cmd->v_variables[i]->p.grp_data;
1720 /* There's always 2 groups for a T - TEST */
1723 gp.indep_width = indep_var->width;
1725 ttpr->group_hash = hsh_create(2,
1726 (hsh_compare_func *) compare_group_binary,
1727 (hsh_hash_func *) hash_group_binary,
1728 (hsh_free_func *) free_group,
1731 for (j=0 ; j < 2 ; ++j)
1734 struct group_statistics *gs = (struct group_statistics *)
1735 xmalloc (sizeof(struct group_statistics));
1741 if ( gp.criterion == CMP_EQ )
1743 gs->id = gp.v.g_value[j];
1748 gs->id.f = gp.v.critical_value - 1.0 ;
1750 gs->id.f = gp.v.critical_value + 1.0 ;
1753 hsh_insert ( ttpr->group_hash, (void *) gs );
1761 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1765 const union value *gv = case_data (c, indep_var->fv);
1767 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1769 if ( value_is_missing(gv,indep_var) )
1774 if ( cmd->miss == TTS_LISTWISE )
1776 for(i=0; i< cmd->n_variables ; ++i)
1778 struct variable *v = cmd->v_variables[i];
1779 const union value *val = case_data (c, v->fv);
1781 if (value_is_missing(val,v) )
1788 gv = case_data (c, indep_var->fv);
1790 for(i=0; i< cmd->n_variables ; ++i)
1792 struct variable *var = cmd->v_variables[i];
1793 const union value *val = case_data (c, var->fv);
1794 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
1795 struct group_statistics *gs;
1797 gs = hsh_find(grp_hash, (void *) gv);
1799 /* If the independent variable doesn't match either of the values
1800 for this case then move on to the next case */
1804 if ( !value_is_missing(val,var) )
1807 gs->sum+=weight * val->f;
1808 gs->ssq+=weight * pow2(val->f);
1817 group_postcalc ( struct cmd_t_test *cmd )
1821 for(i=0; i< cmd->n_variables ; ++i)
1823 struct variable *var = cmd->v_variables[i];
1824 struct hsh_table *grp_hash = var->p.grp_data.group_hash;
1825 struct hsh_iterator g;
1826 struct group_statistics *gs;
1829 for (gs = hsh_first (grp_hash,&g);
1831 gs = hsh_next(grp_hash,&g))
1833 gs->mean = gs->sum / gs->n;
1835 gs->s_std_dev= sqrt(
1836 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1841 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1844 gs->se_mean = gs->std_dev / sqrt(gs->n);
1854 calculate(const struct casefile *cf, void *cmd_)
1856 struct ssbox stat_summary_box;
1857 struct trbox test_results_box;
1859 struct casereader *r;
1862 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1864 common_precalc(cmd);
1865 for(r = casefile_get_reader (cf);
1866 casereader_read (r, &c) ;
1869 common_calc(&c,cmd);
1871 casereader_destroy (r);
1872 common_postcalc(cmd);
1877 one_sample_precalc(cmd);
1878 for(r = casefile_get_reader (cf);
1879 casereader_read (r, &c) ;
1882 one_sample_calc(&c,cmd);
1884 casereader_destroy (r);
1885 one_sample_postcalc(cmd);
1889 paired_precalc(cmd);
1890 for(r = casefile_get_reader (cf);
1891 casereader_read (r, &c) ;
1894 paired_calc(&c,cmd);
1896 casereader_destroy (r);
1897 paired_postcalc(cmd);
1903 for(r = casefile_get_reader (cf);
1904 casereader_read (r, &c) ;
1909 casereader_destroy (r);
1910 group_postcalc(cmd);
1912 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1913 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1918 ssbox_create(&stat_summary_box,cmd,mode);
1919 ssbox_populate(&stat_summary_box,cmd);
1920 ssbox_finalize(&stat_summary_box);
1922 if ( mode == T_PAIRED)
1925 trbox_create(&test_results_box,cmd,mode);
1926 trbox_populate(&test_results_box,cmd);
1927 trbox_finalize(&test_results_box);
1932 /* Return -1 if the id of a is less than b; +1 if greater than and
1935 compare_group_binary(const struct group_statistics *a,
1936 const struct group_statistics *b,
1937 struct group_properties *p)
1943 assert(p->indep_width == 0 ) ;
1945 if ( p->criterion == CMP_LE )
1947 flag_a = ( a->id.f < p->v.critical_value ) ;
1948 flag_b = ( b->id.f < p->v.critical_value ) ;
1952 flag_a = ( a->id.f == p->v.critical_value ) ;
1953 flag_b = ( b->id.f == p->v.critical_value ) ;
1957 if ( flag_a == flag_b)
1960 return ( flag_a < flag_b);
1964 hash_group_binary(const struct group_statistics *g, struct group_properties *p)
1968 assert(p->indep_width == 0 ) ;
1970 /* FIXME: should compare union values */
1971 if ( p->criterion == CMP_LE )
1973 flag = ( g->id.f < p->v.critical_value ) ;
1975 else if ( p->criterion == CMP_EQ)
1977 if ( g->id.f == p->v.g_value[0].f )
1979 else if ( g->id.f == p->v.g_value[1].f )