1 /* PSPP - computes sample statistics. -*-c-*-
3 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
4 Written by John Williams <johnr.williams@stonebow.otago.ac.nz>.
5 Almost completly re-written by John Darrington 2004
7 This program is free software; you can redistribute it and/or
8 modify it under the terms of the GNU General Public License as
9 published by the Free Software Foundation; either version 2 of the
10 License, or (at your option) any later version.
12 This program is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
23 #include <gsl/gsl_cdf.h>
32 #include "dictionary.h"
39 #include "value-labels.h"
43 #include "group_proc.h"
52 variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
54 +missing=miss:!analysis/listwise,
55 incl:include/!exclude;
56 format=fmt:!labels/nolabels;
57 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
65 /* Function to use for testing for missing values */
66 static is_missing_func value_is_missing;
68 /* Variable for the GROUPS subcommand, if given. */
69 static struct variable *indep_var;
77 struct group_properties
79 /* The comparison criterion */
80 enum comparison criterion;
82 /* The width of the independent variable */
86 /* The value of the independent variable at which groups are determined to
87 belong to one group or the other */
88 double critical_value;
91 /* The values of the independent variable for each group */
92 union value g_value[2];
98 static struct group_properties gp ;
102 /* PAIRS: Number of pairs to be compared ; each pair. */
103 static int n_pairs = 0 ;
106 /* The variables comprising the pair */
107 struct variable *v[2];
109 /* The number of valid variable pairs */
112 /* The sum of the members */
115 /* sum of squares of the members */
118 /* Std deviation of the members */
122 /* Sample Std deviation of the members */
125 /* The means of the members */
128 /* The correlation coefficient between the variables */
131 /* The sum of the differences */
134 /* The sum of the products */
137 /* The mean of the differences */
140 /* The sum of the squares of the differences */
143 /* The std deviation of the differences */
147 static struct pair *pairs=0;
149 static int parse_value (union value * v, int type) ;
151 /* Structures and Functions for the Statistics Summary Box */
153 typedef void populate_ssbox_func(struct ssbox *ssb,
154 struct cmd_t_test *cmd);
155 typedef void finalize_ssbox_func(struct ssbox *ssb);
161 populate_ssbox_func *populate;
162 finalize_ssbox_func *finalize;
167 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
169 /* Populate a ssbox according to cmd */
170 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
172 /* Submit and destroy a ssbox */
173 void ssbox_finalize(struct ssbox *ssb);
175 /* A function to create, populate and submit the Paired Samples Correlation
180 /* Structures and Functions for the Test Results Box */
183 typedef void populate_trbox_func(struct trbox *trb,
184 struct cmd_t_test *cmd);
185 typedef void finalize_trbox_func(struct trbox *trb);
189 populate_trbox_func *populate;
190 finalize_trbox_func *finalize;
194 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
196 /* Populate a ssbox according to cmd */
197 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
199 /* Submit and destroy a ssbox */
200 void trbox_finalize(struct trbox *trb);
202 /* Which mode was T-TEST invoked */
210 static int common_calc (const struct ccase *, void *);
211 static void common_precalc (struct cmd_t_test *);
212 static void common_postcalc (struct cmd_t_test *);
214 static int one_sample_calc (const struct ccase *, void *);
215 static void one_sample_precalc (struct cmd_t_test *);
216 static void one_sample_postcalc (struct cmd_t_test *);
218 static int paired_calc (const struct ccase *, void *);
219 static void paired_precalc (struct cmd_t_test *);
220 static void paired_postcalc (struct cmd_t_test *);
222 static void group_precalc (struct cmd_t_test *);
223 static int group_calc (const struct ccase *, struct cmd_t_test *);
224 static void group_postcalc (struct cmd_t_test *);
227 static int compare_var_name (const void *a_, const void *b_, void *v_ UNUSED);
228 static unsigned hash_var_name (const void *a_, void *v_ UNUSED);
230 static void calculate(const struct casefile *cf, void *_mode);
234 static struct cmd_t_test cmd;
236 static int bad_weight_warn;
239 static int compare_group_binary(const struct group_statistics *a,
240 const struct group_statistics *b,
241 const struct group_properties *p);
244 static unsigned hash_group_binary(const struct group_statistics *g,
245 const struct group_properties *p);
253 if ( !parse_t_test(&cmd) )
256 if (! cmd.sbc_criteria)
261 if (cmd.sbc_testval) ++m;
262 if (cmd.sbc_groups) ++m;
263 if (cmd.sbc_pairs) ++m;
268 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
276 else if (cmd.sbc_groups)
281 if ( mode == T_PAIRED)
283 if (cmd.sbc_variables)
285 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
290 /* Iterate through the pairs and put each variable that is a
291 member of a pair into cmd.v_variables */
294 struct hsh_iterator hi;
295 struct hsh_table *hash;
298 hash=hsh_create(n_pairs,compare_var_name,hash_var_name,0,0);
300 for (i=0; i < n_pairs; ++i)
302 hsh_insert(hash,pairs[i].v[0]);
303 hsh_insert(hash,pairs[i].v[1]);
306 assert(cmd.n_variables == 0);
307 cmd.n_variables = hsh_count(hash);
309 cmd.v_variables = xrealloc(cmd.v_variables,
310 sizeof(struct variable) * cmd.n_variables);
311 /* Iterate through the hash */
312 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
314 v=hsh_next(hash,&hi) )
315 cmd.v_variables[i++]=v;
320 else if ( !cmd.sbc_variables)
322 msg(SE, _("One or more VARIABLES must be specified."));
327 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
328 if (cmd.incl == TTS_INCLUDE )
329 value_is_missing = is_system_missing;
331 value_is_missing = is_missing;
335 multipass_procedure_with_splits (calculate, &cmd);
341 if ( mode == T_IND_SAMPLES)
344 /* Destroy any group statistics we created */
345 for (v = 0 ; v < cmd.n_variables ; ++v )
347 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
348 free(grpp->group_hash);
356 tts_custom_groups (struct cmd_t_test *cmd UNUSED)
358 int n_group_values=0;
362 if (token != T_ALL &&
363 (token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
366 msg(SE,_("`%s' is not a variable name"),tokid);
370 indep_var = parse_variable ();
373 lex_error ("expecting variable name in GROUPS subcommand");
377 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
379 msg (SE, _("Long string variable %s is not valid here."),
384 if (!lex_match ('('))
386 if (indep_var->type == NUMERIC)
388 gp.v.g_value[0].f = 1;
389 gp.v.g_value[1].f = 2;
391 gp.criterion = CMP_EQ;
399 msg (SE, _("When applying GROUPS to a string variable, two "
400 "values must be specified."));
405 if (!parse_value (&gp.v.g_value[0], indep_var->type))
411 if (indep_var->type != NUMERIC)
414 msg (SE, _("When applying GROUPS to a string variable, two "
415 "values must be specified."));
418 gp.criterion = CMP_LE;
419 gp.v.critical_value = gp.v.g_value[0].f;
425 if (!parse_value (&gp.v.g_value[1], indep_var->type))
429 if (!lex_force_match (')'))
432 if ( n_group_values == 2 )
433 gp.criterion = CMP_EQ ;
435 gp.criterion = CMP_LE ;
443 tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
445 struct variable **vars;
450 int n_after_WITH = -1;
451 int paired ; /* Was the PAIRED keyword given ? */
455 if ((token != T_ID || dict_lookup_var (default_dict, tokid) == NULL)
458 msg(SE,_("`%s' is not a variable name"),tokid);
463 if (!parse_variables (default_dict, &vars, &n_vars,
464 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
472 if (lex_match (T_WITH))
474 n_before_WITH = n_vars;
475 if (!parse_variables (default_dict, &vars, &n_vars,
476 PV_DUPLICATE | PV_APPEND
477 | PV_NUMERIC | PV_NO_SCRATCH))
482 n_after_WITH = n_vars - n_before_WITH;
485 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
487 /* Determine the number of pairs needed */
490 if (n_before_WITH != n_after_WITH)
493 msg (SE, _("PAIRED was specified but the number of variables "
494 "preceding WITH (%d) did not match the number "
496 n_before_WITH, n_after_WITH );
499 n_pairs_local=n_before_WITH;
501 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
503 n_pairs_local=n_before_WITH * n_after_WITH ;
505 else /* Neither WITH nor PAIRED keyword given */
510 msg (SE, _("At least two variables must be specified "
515 /* how many ways can you pick 2 from n_vars ? */
516 n_pairs_local = n_vars * (n_vars -1 ) /2 ;
520 /* Allocate storage for the pairs */
521 pairs = xrealloc(pairs, sizeof(struct pair) * (n_pairs + n_pairs_local) );
523 /* Populate the pairs with the appropriate variables */
528 assert(n_pairs_local == n_vars/2);
529 for (i = 0; i < n_pairs_local ; ++i)
531 pairs[i].v[n_pairs+0] = vars[i];
532 pairs[i].v[n_pairs+1] = vars[i+n_pairs_local];
535 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
540 for(i=0 ; i < n_before_WITH ; ++i )
542 for(j=0 ; j < n_after_WITH ; ++j)
544 pairs[p].v[0] = vars[i];
545 pairs[p].v[1] = vars[j+n_before_WITH];
550 else /* Neither WITH nor PAIRED given */
555 for(i=0 ; i < n_vars ; ++i )
557 for(j=i+1 ; j < n_vars ; ++j)
559 pairs[p].v[0] = vars[i];
560 pairs[p].v[1] = vars[j];
566 n_pairs+=n_pairs_local;
572 /* Parses the current token (numeric or string, depending on type)
573 value v and returns success. */
575 parse_value (union value * v, int type )
579 if (!lex_force_num ())
585 if (!lex_force_string ())
587 strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
596 /* Implementation of the SSBOX object */
598 void ssbox_base_init(struct ssbox *this, int cols,int rows);
600 void ssbox_base_finalize(struct ssbox *ssb);
602 void ssbox_one_sample_init(struct ssbox *this,
603 struct cmd_t_test *cmd );
605 void ssbox_independent_samples_init(struct ssbox *this,
606 struct cmd_t_test *cmd);
608 void ssbox_paired_init(struct ssbox *this,
609 struct cmd_t_test *cmd);
612 /* Factory to create an ssbox */
614 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
619 ssbox_one_sample_init(ssb,cmd);
622 ssbox_independent_samples_init(ssb,cmd);
625 ssbox_paired_init(ssb,cmd);
634 /* Despatcher for the populate method */
636 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
638 ssb->populate(ssb,cmd);
642 /* Despatcher for finalize */
644 ssbox_finalize(struct ssbox *ssb)
650 /* Submit the box and clear up */
652 ssbox_base_finalize(struct ssbox *ssb)
659 /* Initialize a ssbox struct */
661 ssbox_base_init(struct ssbox *this, int cols,int rows)
663 this->finalize = ssbox_base_finalize;
664 this->t = tab_create (cols, rows, 0);
666 tab_columns (this->t, SOM_COL_DOWN, 1);
667 tab_headers (this->t,0,0,1,0);
668 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
669 tab_hline(this->t, TAL_2,0,cols-1,1);
670 tab_dim (this->t, tab_natural_dimensions);
673 void ssbox_one_sample_populate(struct ssbox *ssb,
674 struct cmd_t_test *cmd);
676 /* Initialize the one_sample ssbox */
678 ssbox_one_sample_init(struct ssbox *this,
679 struct cmd_t_test *cmd )
682 const int vsize=cmd->n_variables+1;
684 this->populate = ssbox_one_sample_populate;
686 ssbox_base_init(this, hsize,vsize);
687 tab_title (this->t, 0, _("One-Sample Statistics"));
688 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
689 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
690 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
691 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
692 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
695 void ssbox_independent_samples_populate(struct ssbox *ssb,
696 struct cmd_t_test *cmd);
698 /* Initialize the independent samples ssbox */
700 ssbox_independent_samples_init(struct ssbox *this,
701 struct cmd_t_test *cmd)
704 int vsize = cmd->n_variables*2 +1;
706 this->populate = ssbox_independent_samples_populate;
708 ssbox_base_init(this, hsize,vsize);
709 tab_title (this->t, 0, _("Group Statistics"));
710 tab_vline(this->t,0,1,0,vsize - 1);
711 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
712 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
713 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
714 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
715 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
719 /* Populate the ssbox for independent samples */
721 ssbox_independent_samples_populate(struct ssbox *ssb,
722 struct cmd_t_test *cmd)
728 double indep_value[2];
730 char prefix[2][3]={"",""};
732 if ( indep_var->type == NUMERIC )
734 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
735 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
739 val_lab0 = gp.v.g_value[0].s;
740 val_lab1 = gp.v.g_value[1].s;
743 if (gp.criterion == CMP_LE )
745 strcpy(prefix[0],"< ");
746 strcpy(prefix[1],">=");
747 indep_value[0] = gp.v.critical_value;
748 indep_value[1] = gp.v.critical_value;
752 indep_value[0] = gp.v.g_value[0].f;
753 indep_value[1] = gp.v.g_value[1].f;
758 for (i=0; i < cmd->n_variables; ++i)
760 struct variable *var = cmd->v_variables[i];
761 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
764 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
767 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
768 "%s%s", prefix[0], val_lab0);
770 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
771 "%s%g", prefix[0], indep_value[0]);
775 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
776 "%s%s", prefix[1], val_lab1);
778 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
779 "%s%g", prefix[1], indep_value[1]);
782 /* Fill in the group statistics */
783 for ( count = 0 ; count < 2 ; ++count )
785 union value search_val;
787 struct group_statistics *gs;
789 if ( gp.criterion == CMP_LE )
793 /* less than ( < ) case */
794 search_val.f = gp.v.critical_value - 1.0;
799 search_val.f = gp.v.critical_value + 1.0;
804 search_val = gp.v.g_value[count];
807 gs = hsh_find(grp_hash, (void *) &search_val);
810 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
811 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
812 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
813 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
819 void ssbox_paired_populate(struct ssbox *ssb,
820 struct cmd_t_test *cmd);
822 /* Initialize the paired values ssbox */
824 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
828 int vsize = n_pairs*2+1;
830 this->populate = ssbox_paired_populate;
832 ssbox_base_init(this, hsize,vsize);
833 tab_title (this->t, 0, _("Paired Sample Statistics"));
834 tab_vline(this->t,TAL_0,1,0,vsize-1);
835 tab_vline(this->t,TAL_2,2,0,vsize-1);
836 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
837 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
838 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
839 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
843 /* Populate the ssbox for paired values */
845 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
851 for (i=0; i < n_pairs; ++i)
855 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
857 for (j=0 ; j < 2 ; ++j)
859 struct group_statistics *gs;
861 gs = &group_proc_get (pairs[i].v[j])->ugs;
865 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
868 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
869 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
870 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
871 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
877 /* Populate the one sample ssbox */
879 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
885 for (i=0; i < cmd->n_variables; ++i)
887 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
889 tab_text (ssb->t, 0, i+1, TAB_LEFT, cmd->v_variables[i]->name);
890 tab_float (ssb->t,1, i+1, TAB_RIGHT, gs->n, 2, 0);
891 tab_float (ssb->t,2, i+1, TAB_RIGHT, gs->mean, 8, 2);
892 tab_float (ssb->t,3, i+1, TAB_RIGHT, gs->std_dev, 8, 2);
893 tab_float (ssb->t,4, i+1, TAB_RIGHT, gs->se_mean, 8, 3);
900 /* Implementation of the Test Results box struct */
902 void trbox_base_init(struct trbox *self,int n_vars, int cols);
903 void trbox_base_finalize(struct trbox *trb);
905 void trbox_independent_samples_init(struct trbox *trb,
906 struct cmd_t_test *cmd );
908 void trbox_independent_samples_populate(struct trbox *trb,
909 struct cmd_t_test *cmd);
911 void trbox_one_sample_init(struct trbox *self,
912 struct cmd_t_test *cmd );
914 void trbox_one_sample_populate(struct trbox *trb,
915 struct cmd_t_test *cmd);
917 void trbox_paired_init(struct trbox *self,
918 struct cmd_t_test *cmd );
920 void trbox_paired_populate(struct trbox *trb,
921 struct cmd_t_test *cmd);
925 /* Create a trbox according to mode*/
927 trbox_create(struct trbox *trb,
928 struct cmd_t_test *cmd, int mode)
933 trbox_one_sample_init(trb,cmd);
936 trbox_independent_samples_init(trb,cmd);
939 trbox_paired_init(trb,cmd);
946 /* Populate a trbox according to cmd */
948 trbox_populate(struct trbox *trb, struct cmd_t_test *cmd)
950 trb->populate(trb,cmd);
953 /* Submit and destroy a trbox */
955 trbox_finalize(struct trbox *trb)
960 /* Initialize the independent samples trbox */
962 trbox_independent_samples_init(struct trbox *self,
963 struct cmd_t_test *cmd UNUSED)
966 const int vsize=cmd->n_variables*2+3;
969 self->populate = trbox_independent_samples_populate;
971 trbox_base_init(self,cmd->n_variables*2,hsize);
972 tab_title(self->t,0,_("Independent Samples Test"));
973 tab_hline(self->t,TAL_1,2,hsize-1,1);
974 tab_vline(self->t,TAL_2,2,0,vsize-1);
975 tab_vline(self->t,TAL_1,4,0,vsize-1);
976 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,hsize-2,vsize-1);
977 tab_hline(self->t,TAL_1, hsize-2,hsize-1,2);
978 tab_box(self->t,-1,-1,-1,TAL_1, hsize-2,2,hsize-1,vsize-1);
979 tab_joint_text(self->t, 2, 0, 3, 0,
980 TAB_CENTER,_("Levene's Test for Equality of Variances"));
981 tab_joint_text(self->t, 4,0,hsize-1,0,
982 TAB_CENTER,_("t-test for Equality of Means"));
984 tab_text(self->t,2,2, TAB_CENTER | TAT_TITLE,_("F"));
985 tab_text(self->t,3,2, TAB_CENTER | TAT_TITLE,_("Sig."));
986 tab_text(self->t,4,2, TAB_CENTER | TAT_TITLE,_("t"));
987 tab_text(self->t,5,2, TAB_CENTER | TAT_TITLE,_("df"));
988 tab_text(self->t,6,2, TAB_CENTER | TAT_TITLE,_("Sig. (2-tailed)"));
989 tab_text(self->t,7,2, TAB_CENTER | TAT_TITLE,_("Mean Difference"));
990 tab_text(self->t,8,2, TAB_CENTER | TAT_TITLE,_("Std. Error Difference"));
991 tab_text(self->t,9,2, TAB_CENTER | TAT_TITLE,_("Lower"));
992 tab_text(self->t,10,2, TAB_CENTER | TAT_TITLE,_("Upper"));
994 tab_joint_text(self->t, 9, 1, 10, 1, TAB_CENTER | TAT_PRINTF,
995 _("%g%% Confidence Interval of the Difference"),
996 cmd->criteria*100.0);
1000 /* Populate the independent samples trbox */
1002 trbox_independent_samples_populate(struct trbox *self,
1003 struct cmd_t_test *cmd )
1008 for (i=0; i < cmd->n_variables; ++i)
1017 double pooled_variance;
1018 double std_err_diff;
1021 struct variable *var = cmd->v_variables[i];
1022 struct group_proc *grp_data = group_proc_get (var);
1024 struct hsh_table *grp_hash = grp_data->group_hash;
1026 struct group_statistics *gs0 ;
1027 struct group_statistics *gs1 ;
1029 union value search_val;
1031 if ( gp.criterion == CMP_LE )
1032 search_val.f = gp.v.critical_value - 1.0;
1034 search_val = gp.v.g_value[0];
1036 gs0 = hsh_find(grp_hash, (void *) &search_val);
1039 if ( gp.criterion == CMP_LE )
1040 search_val.f = gp.v.critical_value + 1.0;
1042 search_val = gp.v.g_value[1];
1044 gs1 = hsh_find(grp_hash, (void *) &search_val);
1048 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1050 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1053 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1055 /* Now work out the significance of the Levene test */
1056 df1 = 1; df2 = grp_data->ugs.n - 2;
1057 q = gsl_cdf_fdist_Q(grp_data->levene, df1, df2);
1059 tab_float(self->t, 3, i*2+3, TAB_CENTER, q, 8,3 );
1061 df = gs0->n + gs1->n - 2.0 ;
1062 tab_float (self->t, 5, i*2+3, TAB_RIGHT, df, 2, 0);
1064 pooled_variance = ( (gs0->n )*pow2(gs0->s_std_dev)
1066 (gs1->n )*pow2(gs1->s_std_dev)
1069 t = (gs0->mean - gs1->mean) / sqrt(pooled_variance) ;
1070 t /= sqrt((gs0->n + gs1->n)/(gs0->n*gs1->n));
1072 tab_float (self->t, 4, i*2+3, TAB_RIGHT, t, 8, 3);
1074 p = gsl_cdf_tdist_P(t, df);
1075 q = gsl_cdf_tdist_Q(t, df);
1077 tab_float(self->t, 6, i*2+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1079 mean_diff = gs0->mean - gs1->mean;
1080 tab_float(self->t, 7, i*2+3, TAB_RIGHT, mean_diff, 8, 3);
1083 std_err_diff = sqrt( pow2(gs0->se_mean) + pow2(gs1->se_mean));
1084 tab_float(self->t, 8, i*2+3, TAB_RIGHT, std_err_diff, 8, 3);
1087 /* Now work out the confidence interval */
1088 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1090 t = gsl_cdf_tdist_Qinv(q,df);
1091 tab_float(self->t, 9, i*2+3, TAB_RIGHT,
1092 mean_diff - t * std_err_diff, 8, 3);
1094 tab_float(self->t, 10, i*2+3, TAB_RIGHT,
1095 mean_diff + t * std_err_diff, 8, 3);
1100 /* Now for the \sigma_1 != \sigma_2 case */
1101 tab_text (self->t, 1, i*2+3+1,
1102 TAB_LEFT, _("Equal variances not assumed"));
1105 se2 = (pow2(gs0->s_std_dev)/(gs0->n -1) ) +
1106 (pow2(gs1->s_std_dev)/(gs1->n -1) );
1108 t = mean_diff / sqrt(se2) ;
1109 tab_float (self->t, 4, i*2+3+1, TAB_RIGHT, t, 8, 3);
1112 (pow2(pow2(gs0->s_std_dev)/(gs0->n - 1 ))
1116 (pow2(pow2(gs1->s_std_dev)/(gs1->n - 1 ))
1120 tab_float (self->t, 5, i*2+3+1, TAB_RIGHT, df, 8, 3);
1122 p = gsl_cdf_tdist_P(t, df);
1123 q = gsl_cdf_tdist_Q(t, df);
1125 tab_float(self->t, 6, i*2+3+1, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1127 /* Now work out the confidence interval */
1128 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1130 t = gsl_cdf_tdist_Qinv(q, df);
1132 tab_float(self->t, 7, i*2+3+1, TAB_RIGHT, mean_diff, 8, 3);
1135 tab_float(self->t, 8, i*2+3+1, TAB_RIGHT, std_err_diff, 8, 3);
1138 tab_float(self->t, 9, i*2+3+1, TAB_RIGHT,
1139 mean_diff - t * std_err_diff, 8, 3);
1141 tab_float(self->t, 10, i*2+3+1, TAB_RIGHT,
1142 mean_diff + t * std_err_diff, 8, 3);
1148 /* Initialize the paired samples trbox */
1150 trbox_paired_init(struct trbox *self,
1151 struct cmd_t_test *cmd UNUSED)
1155 const int vsize=n_pairs+3;
1157 self->populate = trbox_paired_populate;
1159 trbox_base_init(self,n_pairs,hsize);
1160 tab_title (self->t, 0, _("Paired Samples Test"));
1161 tab_hline(self->t,TAL_1,2,6,1);
1162 tab_vline(self->t,TAL_2,2,0,vsize - 1);
1163 tab_joint_text(self->t,2,0,6,0,TAB_CENTER,_("Paired Differences"));
1164 tab_box(self->t,-1,-1,-1,TAL_1, 2,1,6,vsize-1);
1165 tab_box(self->t,-1,-1,-1,TAL_1, 6,0,hsize-1,vsize-1);
1166 tab_hline(self->t,TAL_1,5,6, 2);
1167 tab_vline(self->t,TAL_0,6,0,1);
1169 tab_joint_text(self->t, 5, 1, 6, 1, TAB_CENTER | TAT_PRINTF,
1170 _("%g%% Confidence Interval of the Difference"),
1171 cmd->criteria*100.0);
1173 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("Mean"));
1174 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
1175 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Std. Error Mean"));
1176 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1177 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1178 tab_text (self->t, 7, 2, TAB_CENTER | TAT_TITLE, _("t"));
1179 tab_text (self->t, 8, 2, TAB_CENTER | TAT_TITLE, _("df"));
1180 tab_text (self->t, 9, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1183 /* Populate the paired samples trbox */
1185 trbox_paired_populate(struct trbox *trb,
1186 struct cmd_t_test *cmd UNUSED)
1190 for (i=0; i < n_pairs; ++i)
1195 double n = pairs[i].n;
1199 tab_text (trb->t, 0, i+3, TAB_LEFT | TAT_PRINTF, _("Pair %d"),i);
1201 tab_text (trb->t, 1, i+3, TAB_LEFT | TAT_PRINTF, "%s - %s",
1202 pairs[i].v[0]->name, pairs[i].v[1]->name);
1204 tab_float(trb->t, 2, i+3, TAB_RIGHT, pairs[i].mean_diff, 8, 4);
1206 tab_float(trb->t, 3, i+3, TAB_RIGHT, pairs[i].std_dev_diff, 8, 5);
1209 se_mean = pairs[i].std_dev_diff / sqrt(n) ;
1210 tab_float(trb->t, 4, i+3, TAB_RIGHT, se_mean, 8,5 );
1212 /* Now work out the confidence interval */
1213 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1215 t = gsl_cdf_tdist_Qinv(q, df);
1217 tab_float(trb->t, 5, i+3, TAB_RIGHT,
1218 pairs[i].mean_diff - t * se_mean , 8, 4);
1220 tab_float(trb->t, 6, i+3, TAB_RIGHT,
1221 pairs[i].mean_diff + t * se_mean , 8, 4);
1223 t = (pairs[i].mean[0] - pairs[i].mean[1])
1225 ( pow2 (pairs[i].s_std_dev[0]) + pow2 (pairs[i].s_std_dev[1]) -
1226 2 * pairs[i].correlation *
1227 pairs[i].s_std_dev[0] * pairs[i].s_std_dev[1] )
1231 tab_float(trb->t, 7, i+3, TAB_RIGHT, t , 8,3 );
1233 /* Degrees of freedom */
1234 tab_float(trb->t, 8, i+3, TAB_RIGHT, df , 2, 0 );
1236 p = gsl_cdf_tdist_P(t,df);
1237 q = gsl_cdf_tdist_P(t,df);
1239 tab_float(trb->t, 9, i+3, TAB_RIGHT, 2.0*(t>0?q:p) , 8, 3);
1244 /* Initialize the one sample trbox */
1246 trbox_one_sample_init(struct trbox *self, struct cmd_t_test *cmd )
1249 const int vsize=cmd->n_variables+3;
1251 self->populate = trbox_one_sample_populate;
1253 trbox_base_init(self, cmd->n_variables,hsize);
1254 tab_title (self->t, 0, _("One-Sample Test"));
1255 tab_hline(self->t, TAL_1, 1, hsize - 1, 1);
1256 tab_vline(self->t, TAL_2, 1, 0, vsize - 1);
1258 tab_joint_text(self->t, 1, 0, hsize-1,0, TAB_CENTER | TAT_PRINTF,
1259 _("Test Value = %f"), cmd->n_testval[0]);
1261 tab_box(self->t, -1, -1, -1, TAL_1, 1,1,hsize-1,vsize-1);
1264 tab_joint_text(self->t,5,1,6,1,TAB_CENTER | TAT_PRINTF,
1265 _("%g%% Confidence Interval of the Difference"),
1266 cmd->criteria*100.0);
1268 tab_vline(self->t,TAL_0,6,1,1);
1269 tab_hline(self->t,TAL_1,5,6,2);
1270 tab_text (self->t, 1, 2, TAB_CENTER | TAT_TITLE, _("t"));
1271 tab_text (self->t, 2, 2, TAB_CENTER | TAT_TITLE, _("df"));
1272 tab_text (self->t, 3, 2, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
1273 tab_text (self->t, 4, 2, TAB_CENTER | TAT_TITLE, _("Mean Difference"));
1274 tab_text (self->t, 5, 2, TAB_CENTER | TAT_TITLE, _("Lower"));
1275 tab_text (self->t, 6, 2, TAB_CENTER | TAT_TITLE, _("Upper"));
1280 /* Populate the one sample trbox */
1282 trbox_one_sample_populate(struct trbox *trb, struct cmd_t_test *cmd)
1288 for (i=0; i < cmd->n_variables; ++i)
1293 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->ugs;
1296 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1298 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1300 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1302 /* degrees of freedom */
1305 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1307 p = gsl_cdf_tdist_P(t, df);
1308 q = gsl_cdf_tdist_Q(t, df);
1310 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1312 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1314 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1317 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1318 t = gsl_cdf_tdist_Qinv(q, df);
1320 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1321 gs->mean_diff - t * gs->se_mean, 8,4);
1323 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1324 gs->mean_diff + t * gs->se_mean, 8,4);
1328 /* Base initializer for the generalized trbox */
1330 trbox_base_init(struct trbox *self, int data_rows, int cols)
1332 const int rows = 3 + data_rows;
1334 self->finalize = trbox_base_finalize;
1335 self->t = tab_create (cols, rows, 0);
1336 tab_headers (self->t,0,0,3,0);
1337 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1338 tab_hline(self->t, TAL_2,0,cols-1,3);
1339 tab_dim (self->t, tab_natural_dimensions);
1343 /* Base finalizer for the trbox */
1345 trbox_base_finalize(struct trbox *trb)
1351 /* Create , populate and submit the Paired Samples Correlation box */
1355 const int rows=1+n_pairs;
1359 struct tab_table *table;
1361 table = tab_create (cols,rows,0);
1363 tab_columns (table, SOM_COL_DOWN, 1);
1364 tab_headers (table,0,0,1,0);
1365 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1366 tab_hline(table, TAL_2, 0, cols - 1, 1);
1367 tab_vline(table, TAL_2, 2, 0, rows - 1);
1368 tab_dim(table, tab_natural_dimensions);
1369 tab_title(table, 0, _("Paired Samples Correlations"));
1371 /* column headings */
1372 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1373 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1374 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1376 for (i=0; i < n_pairs; ++i)
1380 double df = pairs[i].n -2;
1382 double correlation_t =
1383 pairs[i].correlation * sqrt(df) /
1384 sqrt(1 - pow2(pairs[i].correlation));
1388 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1391 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1392 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1396 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1397 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1399 p = gsl_cdf_tdist_P(correlation_t, df);
1400 q = gsl_cdf_tdist_Q(correlation_t, df);
1402 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1411 /* Calculation Implementation */
1413 /* Per case calculations common to all variants of the T test */
1415 common_calc (const struct ccase *c, void *_cmd)
1418 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1420 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1423 /* Skip the entire case if /MISSING=LISTWISE is set */
1424 if ( cmd->miss == TTS_LISTWISE )
1426 for(i=0; i< cmd->n_variables ; ++i)
1428 struct variable *v = cmd->v_variables[i];
1429 const union value *val = case_data (c, v->fv);
1431 if (value_is_missing(val,v) )
1438 /* Listwise has to be implicit if the independent variable is missing ?? */
1439 if ( cmd->sbc_groups )
1441 const union value *gv = case_data (c, indep_var->fv);
1442 if ( value_is_missing(gv,indep_var) )
1449 for(i=0; i< cmd->n_variables ; ++i)
1451 struct group_statistics *gs;
1452 struct variable *v = cmd->v_variables[i];
1453 const union value *val = case_data (c, v->fv);
1455 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1457 if (! value_is_missing(val,v) )
1460 gs->sum+=weight * val->f;
1461 gs->ssq+=weight * val->f * val->f;
1467 /* Pre calculations common to all variants of the T test */
1469 common_precalc ( struct cmd_t_test *cmd )
1473 for(i=0; i< cmd->n_variables ; ++i)
1475 struct group_statistics *gs;
1476 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1485 /* Post calculations common to all variants of the T test */
1487 common_postcalc ( struct cmd_t_test *cmd )
1492 for(i=0; i< cmd->n_variables ; ++i)
1494 struct group_statistics *gs;
1495 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1497 gs->mean=gs->sum / gs->n;
1498 gs->s_std_dev= sqrt(
1499 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1504 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1507 gs->se_mean = gs->std_dev / sqrt(gs->n);
1508 gs->mean_diff= gs->sum_diff / gs->n;
1512 /* Per case calculations for one sample t test */
1514 one_sample_calc (const struct ccase *c, void *cmd_)
1517 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1520 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1522 /* Skip the entire case if /MISSING=LISTWISE is set */
1523 if ( cmd->miss == TTS_LISTWISE )
1525 for(i=0; i< cmd->n_variables ; ++i)
1527 struct variable *v = cmd->v_variables[i];
1528 const union value *val = case_data (c, v->fv);
1530 if (value_is_missing(val,v) )
1537 for(i=0; i< cmd->n_variables ; ++i)
1539 struct group_statistics *gs;
1540 struct variable *v = cmd->v_variables[i];
1541 const union value *val = case_data (c, v->fv);
1543 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1545 if ( ! value_is_missing(val,v))
1546 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1552 /* Pre calculations for one sample t test */
1554 one_sample_precalc ( struct cmd_t_test *cmd )
1558 for(i=0; i< cmd->n_variables ; ++i)
1560 struct group_statistics *gs;
1561 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1567 /* Post calculations for one sample t test */
1569 one_sample_postcalc (struct cmd_t_test *cmd)
1573 for(i=0; i< cmd->n_variables ; ++i)
1575 struct group_statistics *gs;
1576 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1578 gs->mean_diff = gs->sum_diff / gs->n ;
1585 compare_var_name (const void *a_, const void *b_, void *v_ UNUSED)
1587 const struct variable *a = a_;
1588 const struct variable *b = b_;
1590 return strcmp(a->name,b->name);
1594 hash_var_name (const void *a_, void *v_ UNUSED)
1596 const struct variable *a = a_;
1598 return hsh_hash_bytes (a->name, strlen(a->name));
1604 paired_precalc (struct cmd_t_test *cmd UNUSED)
1608 for(i=0; i < n_pairs ; ++i )
1611 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1612 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1613 pairs[i].sum_of_prod = 0;
1614 pairs[i].correlation = 0;
1615 pairs[i].sum_of_diffs = 0;
1616 pairs[i].ssq_diffs = 0;
1623 paired_calc (const struct ccase *c, void *cmd_)
1627 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1629 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1631 /* Skip the entire case if /MISSING=LISTWISE is set ,
1632 AND one member of a pair is missing */
1633 if ( cmd->miss == TTS_LISTWISE )
1635 for(i=0; i < n_pairs ; ++i )
1637 struct variable *v0 = pairs[i].v[0];
1638 struct variable *v1 = pairs[i].v[1];
1640 const union value *val0 = case_data (c, v0->fv);
1641 const union value *val1 = case_data (c, v1->fv);
1643 if ( value_is_missing(val0,v0) ||
1644 value_is_missing(val1,v1) )
1651 for(i=0; i < n_pairs ; ++i )
1653 struct variable *v0 = pairs[i].v[0];
1654 struct variable *v1 = pairs[i].v[1];
1656 const union value *val0 = case_data (c, v0->fv);
1657 const union value *val1 = case_data (c, v1->fv);
1659 if ( ( !value_is_missing(val0,v0) && !value_is_missing(val1,v1) ) )
1661 pairs[i].n += weight;
1662 pairs[i].sum[0] += weight * val0->f;
1663 pairs[i].sum[1] += weight * val1->f;
1665 pairs[i].ssq[0] += weight * pow2(val0->f);
1666 pairs[i].ssq[1] += weight * pow2(val1->f);
1668 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1670 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1671 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1679 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1683 for(i=0; i < n_pairs ; ++i )
1686 const double n = pairs[i].n;
1688 for (j=0; j < 2 ; ++j)
1690 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1691 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1692 pow2(pairs[i].mean[j]))
1695 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1696 pow2(pairs[i].mean[j]))
1700 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1701 pairs[i].mean[0] * pairs[i].mean[1] ;
1702 /* correlation now actually contains the covariance */
1704 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1705 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1707 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1709 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1710 ( pairs[i].ssq_diffs / n )
1712 pow2(pairs[i].mean_diff )
1718 group_precalc (struct cmd_t_test *cmd )
1723 for(i=0; i< cmd->n_variables ; ++i)
1725 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1727 /* There's always 2 groups for a T - TEST */
1730 gp.indep_width = indep_var->width;
1732 ttpr->group_hash = hsh_create(2,
1733 (hsh_compare_func *) compare_group_binary,
1734 (hsh_hash_func *) hash_group_binary,
1735 (hsh_free_func *) free_group,
1738 for (j=0 ; j < 2 ; ++j)
1741 struct group_statistics *gs = (struct group_statistics *)
1742 xmalloc (sizeof(struct group_statistics));
1748 if ( gp.criterion == CMP_EQ )
1750 gs->id = gp.v.g_value[j];
1755 gs->id.f = gp.v.critical_value - 1.0 ;
1757 gs->id.f = gp.v.critical_value + 1.0 ;
1760 hsh_insert ( ttpr->group_hash, (void *) gs );
1768 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1772 const union value *gv = case_data (c, indep_var->fv);
1774 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1776 if ( value_is_missing(gv,indep_var) )
1781 if ( cmd->miss == TTS_LISTWISE )
1783 for(i=0; i< cmd->n_variables ; ++i)
1785 struct variable *v = cmd->v_variables[i];
1786 const union value *val = case_data (c, v->fv);
1788 if (value_is_missing(val,v) )
1795 gv = case_data (c, indep_var->fv);
1797 for(i=0; i< cmd->n_variables ; ++i)
1799 struct variable *var = cmd->v_variables[i];
1800 const union value *val = case_data (c, var->fv);
1801 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1802 struct group_statistics *gs;
1804 gs = hsh_find(grp_hash, (void *) gv);
1806 /* If the independent variable doesn't match either of the values
1807 for this case then move on to the next case */
1811 if ( !value_is_missing(val,var) )
1814 gs->sum+=weight * val->f;
1815 gs->ssq+=weight * pow2(val->f);
1824 group_postcalc ( struct cmd_t_test *cmd )
1828 for(i=0; i< cmd->n_variables ; ++i)
1830 struct variable *var = cmd->v_variables[i];
1831 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1832 struct hsh_iterator g;
1833 struct group_statistics *gs;
1836 for (gs = hsh_first (grp_hash,&g);
1838 gs = hsh_next(grp_hash,&g))
1840 gs->mean = gs->sum / gs->n;
1842 gs->s_std_dev= sqrt(
1843 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1848 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1851 gs->se_mean = gs->std_dev / sqrt(gs->n);
1861 calculate(const struct casefile *cf, void *cmd_)
1863 struct ssbox stat_summary_box;
1864 struct trbox test_results_box;
1866 struct casereader *r;
1869 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1871 common_precalc(cmd);
1872 for(r = casefile_get_reader (cf);
1873 casereader_read (r, &c) ;
1876 common_calc(&c,cmd);
1878 casereader_destroy (r);
1879 common_postcalc(cmd);
1884 one_sample_precalc(cmd);
1885 for(r = casefile_get_reader (cf);
1886 casereader_read (r, &c) ;
1889 one_sample_calc(&c,cmd);
1891 casereader_destroy (r);
1892 one_sample_postcalc(cmd);
1896 paired_precalc(cmd);
1897 for(r = casefile_get_reader (cf);
1898 casereader_read (r, &c) ;
1901 paired_calc(&c,cmd);
1903 casereader_destroy (r);
1904 paired_postcalc(cmd);
1910 for(r = casefile_get_reader (cf);
1911 casereader_read (r, &c) ;
1916 casereader_destroy (r);
1917 group_postcalc(cmd);
1919 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1920 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1925 ssbox_create(&stat_summary_box,cmd,mode);
1926 ssbox_populate(&stat_summary_box,cmd);
1927 ssbox_finalize(&stat_summary_box);
1929 if ( mode == T_PAIRED)
1932 trbox_create(&test_results_box,cmd,mode);
1933 trbox_populate(&test_results_box,cmd);
1934 trbox_finalize(&test_results_box);
1939 /* Return -1 if the id of a is less than b; +1 if greater than and
1942 compare_group_binary(const struct group_statistics *a,
1943 const struct group_statistics *b,
1944 const struct group_properties *p)
1950 if ( p->criterion == CMP_LE )
1952 assert(p->indep_width == 0 ) ;
1953 flag_a = ( a->id.f < p->v.critical_value ) ;
1954 flag_b = ( b->id.f < p->v.critical_value ) ;
1958 flag_a = ( a->id.f == p->v.critical_value ) ;
1959 flag_b = ( b->id.f == p->v.critical_value ) ;
1962 if ( flag_a == flag_b)
1965 return ( flag_a < flag_b);
1968 /* This is a degenerate case of a hash, since it can only return three possible
1969 values. It's really a comparison, being used as a hash function */
1972 hash_group_binary(const struct group_statistics *g,
1973 const struct group_properties *p)
1977 if ( p->criterion == CMP_LE )
1979 /* Not meaningfull to do a less than compare for alpha values ? */
1980 assert(p->indep_width == 0 ) ;
1981 flag = ( g->id.f < p->v.critical_value ) ;
1983 else if ( p->criterion == CMP_EQ)
1985 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width ))
1987 else if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width ))