1 /* PSPP - computes sample statistics. -*-c-*-
3 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
4 Written by John Williams <johnr.williams@stonebow.otago.ac.nz>.
5 Almost completly re-written by John Darrington 2004
7 This program is free software; you can redistribute it and/or
8 modify it under the terms of the GNU General Public License as
9 published by the Free Software Foundation; either version 2 of the
10 License, or (at your option) any later version.
12 This program is distributed in the hope that it will be useful, but
13 WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
23 #include <gsl/gsl_cdf.h>
32 #include "dictionary.h"
39 #include "value-labels.h"
43 #include "group_proc.h"
48 #define _(msgid) gettext (msgid)
56 variables=varlist("PV_NO_SCRATCH | PV_NUMERIC");
58 +missing=miss:!analysis/listwise,
59 incl:include/!exclude;
60 format=fmt:!labels/nolabels;
61 criteria=:cin(d:criteria,"%s > 0. && %s < 1.").
69 /* Function to use for testing for missing values */
70 static is_missing_func *value_is_missing;
72 /* Variable for the GROUPS subcommand, if given. */
73 static struct variable *indep_var;
81 struct group_properties
83 /* The comparison criterion */
84 enum comparison criterion;
86 /* The width of the independent variable */
90 /* The value of the independent variable at which groups are determined to
91 belong to one group or the other */
92 double critical_value;
95 /* The values of the independent variable for each group */
96 union value g_value[2];
102 static struct group_properties gp ;
106 /* PAIRS: Number of pairs to be compared ; each pair. */
107 static int n_pairs = 0 ;
110 /* The variables comprising the pair */
111 struct variable *v[2];
113 /* The number of valid variable pairs */
116 /* The sum of the members */
119 /* sum of squares of the members */
122 /* Std deviation of the members */
126 /* Sample Std deviation of the members */
129 /* The means of the members */
132 /* The correlation coefficient between the variables */
135 /* The sum of the differences */
138 /* The sum of the products */
141 /* The mean of the differences */
144 /* The sum of the squares of the differences */
147 /* The std deviation of the differences */
151 static struct pair *pairs=0;
153 static int parse_value (union value * v, int type) ;
155 /* Structures and Functions for the Statistics Summary Box */
157 typedef void populate_ssbox_func(struct ssbox *ssb,
158 struct cmd_t_test *cmd);
159 typedef void finalize_ssbox_func(struct ssbox *ssb);
165 populate_ssbox_func *populate;
166 finalize_ssbox_func *finalize;
171 void ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode);
173 /* Populate a ssbox according to cmd */
174 void ssbox_populate(struct ssbox *ssb, struct cmd_t_test *cmd);
176 /* Submit and destroy a ssbox */
177 void ssbox_finalize(struct ssbox *ssb);
179 /* A function to create, populate and submit the Paired Samples Correlation
184 /* Structures and Functions for the Test Results Box */
187 typedef void populate_trbox_func(struct trbox *trb,
188 struct cmd_t_test *cmd);
189 typedef void finalize_trbox_func(struct trbox *trb);
193 populate_trbox_func *populate;
194 finalize_trbox_func *finalize;
198 void trbox_create(struct trbox *trb, struct cmd_t_test *cmd, int mode);
200 /* Populate a ssbox according to cmd */
201 void trbox_populate(struct trbox *trb, struct cmd_t_test *cmd);
203 /* Submit and destroy a ssbox */
204 void trbox_finalize(struct trbox *trb);
206 /* Which mode was T-TEST invoked */
214 static int common_calc (const struct ccase *, void *);
215 static void common_precalc (struct cmd_t_test *);
216 static void common_postcalc (struct cmd_t_test *);
218 static int one_sample_calc (const struct ccase *, void *);
219 static void one_sample_precalc (struct cmd_t_test *);
220 static void one_sample_postcalc (struct cmd_t_test *);
222 static int paired_calc (const struct ccase *, void *);
223 static void paired_precalc (struct cmd_t_test *);
224 static void paired_postcalc (struct cmd_t_test *);
226 static void group_precalc (struct cmd_t_test *);
227 static int group_calc (const struct ccase *, struct cmd_t_test *);
228 static void group_postcalc (struct cmd_t_test *);
231 static void calculate(const struct casefile *cf, void *_mode);
235 static struct cmd_t_test cmd;
237 static int bad_weight_warn;
240 static int compare_group_binary(const struct group_statistics *a,
241 const struct group_statistics *b,
242 const struct group_properties *p);
245 static unsigned hash_group_binary(const struct group_statistics *g,
246 const struct group_properties *p);
254 if ( !parse_t_test(&cmd) )
257 if (! cmd.sbc_criteria)
262 if (cmd.sbc_testval) ++m;
263 if (cmd.sbc_groups) ++m;
264 if (cmd.sbc_pairs) ++m;
269 _("TESTVAL, GROUPS and PAIRS subcommands are mutually exclusive.")
278 else if (cmd.sbc_groups)
283 if ( mode == T_PAIRED)
285 if (cmd.sbc_variables)
287 msg(SE, _("VARIABLES subcommand is not appropriate with PAIRS"));
293 /* Iterate through the pairs and put each variable that is a
294 member of a pair into cmd.v_variables */
297 struct hsh_iterator hi;
298 struct hsh_table *hash;
301 hash = hsh_create (n_pairs, compare_var_names, hash_var_name, 0, 0);
303 for (i=0; i < n_pairs; ++i)
305 hsh_insert(hash,pairs[i].v[0]);
306 hsh_insert(hash,pairs[i].v[1]);
309 assert(cmd.n_variables == 0);
310 cmd.n_variables = hsh_count(hash);
312 cmd.v_variables = xrealloc(cmd.v_variables,
313 sizeof(struct variable) * cmd.n_variables);
314 /* Iterate through the hash */
315 for (i=0,v = (struct variable *) hsh_first(hash,&hi);
317 v=hsh_next(hash,&hi) )
318 cmd.v_variables[i++]=v;
323 else if ( !cmd.sbc_variables)
325 msg(SE, _("One or more VARIABLES must be specified."));
331 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
332 if (cmd.incl == TTS_INCLUDE )
333 value_is_missing = mv_is_value_system_missing;
335 value_is_missing = mv_is_value_missing;
339 multipass_procedure_with_splits (calculate, &cmd);
345 if ( mode == T_IND_SAMPLES)
348 /* Destroy any group statistics we created */
349 for (v = 0 ; v < cmd.n_variables ; ++v )
351 struct group_proc *grpp = group_proc_get (cmd.v_variables[v]);
352 hsh_destroy (grpp->group_hash);
361 tts_custom_groups (struct cmd_t_test *cmd UNUSED)
363 int n_group_values=0;
367 indep_var = parse_variable ();
370 lex_error ("expecting variable name in GROUPS subcommand");
374 if (indep_var->type == T_STRING && indep_var->width > MAX_SHORT_STRING)
376 msg (SE, _("Long string variable %s is not valid here."),
381 if (!lex_match ('('))
383 if (indep_var->type == NUMERIC)
385 gp.v.g_value[0].f = 1;
386 gp.v.g_value[1].f = 2;
388 gp.criterion = CMP_EQ;
396 msg (SE, _("When applying GROUPS to a string variable, two "
397 "values must be specified."));
402 if (!parse_value (&gp.v.g_value[0], indep_var->type))
408 if (indep_var->type != NUMERIC)
411 msg (SE, _("When applying GROUPS to a string variable, two "
412 "values must be specified."));
415 gp.criterion = CMP_LE;
416 gp.v.critical_value = gp.v.g_value[0].f;
422 if (!parse_value (&gp.v.g_value[1], indep_var->type))
426 if (!lex_force_match (')'))
429 if ( n_group_values == 2 )
430 gp.criterion = CMP_EQ ;
432 gp.criterion = CMP_LE ;
440 tts_custom_pairs (struct cmd_t_test *cmd UNUSED)
442 struct variable **vars;
447 int n_after_WITH = -1;
448 int paired ; /* Was the PAIRED keyword given ? */
453 if (!parse_variables (default_dict, &vars, &n_vars,
454 PV_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH))
462 if (lex_match (T_WITH))
464 n_before_WITH = n_vars;
465 if (!parse_variables (default_dict, &vars, &n_vars,
466 PV_DUPLICATE | PV_APPEND
467 | PV_NUMERIC | PV_NO_SCRATCH))
472 n_after_WITH = n_vars - n_before_WITH;
475 paired = (lex_match ('(') && lex_match_id ("PAIRED") && lex_match (')'));
477 /* Determine the number of pairs needed */
480 if (n_before_WITH != n_after_WITH)
483 msg (SE, _("PAIRED was specified but the number of variables "
484 "preceding WITH (%d) did not match the number "
486 n_before_WITH, n_after_WITH );
489 n_pairs_local=n_before_WITH;
491 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
493 n_pairs_local=n_before_WITH * n_after_WITH ;
495 else /* Neither WITH nor PAIRED keyword given */
500 msg (SE, _("At least two variables must be specified "
505 /* how many ways can you pick 2 from n_vars ? */
506 n_pairs_local = n_vars * (n_vars -1 ) /2 ;
510 /* Allocate storage for the pairs */
511 pairs = xrealloc(pairs, sizeof(struct pair) * (n_pairs + n_pairs_local) );
513 /* Populate the pairs with the appropriate variables */
518 assert(n_pairs_local == n_vars/2);
519 for (i = 0; i < n_pairs_local ; ++i)
521 pairs[i].v[n_pairs+0] = vars[i];
522 pairs[i].v[n_pairs+1] = vars[i+n_pairs_local];
525 else if (n_before_WITH > 0) /* WITH keyword given, but not PAIRED keyword */
530 for(i=0 ; i < n_before_WITH ; ++i )
532 for(j=0 ; j < n_after_WITH ; ++j)
534 pairs[p].v[0] = vars[i];
535 pairs[p].v[1] = vars[j+n_before_WITH];
540 else /* Neither WITH nor PAIRED given */
545 for(i=0 ; i < n_vars ; ++i )
547 for(j=i+1 ; j < n_vars ; ++j)
549 pairs[p].v[0] = vars[i];
550 pairs[p].v[1] = vars[j];
556 n_pairs+=n_pairs_local;
562 /* Parses the current token (numeric or string, depending on type)
563 value v and returns success. */
565 parse_value (union value * v, int type )
569 if (!lex_force_num ())
575 if (!lex_force_string ())
577 strncpy (v->s, ds_c_str (&tokstr), ds_length (&tokstr));
586 /* Implementation of the SSBOX object */
588 void ssbox_base_init(struct ssbox *this, int cols,int rows);
590 void ssbox_base_finalize(struct ssbox *ssb);
592 void ssbox_one_sample_init(struct ssbox *this,
593 struct cmd_t_test *cmd );
595 void ssbox_independent_samples_init(struct ssbox *this,
596 struct cmd_t_test *cmd);
598 void ssbox_paired_init(struct ssbox *this,
599 struct cmd_t_test *cmd);
602 /* Factory to create an ssbox */
604 ssbox_create(struct ssbox *ssb, struct cmd_t_test *cmd, int mode)
609 ssbox_one_sample_init(ssb,cmd);
612 ssbox_independent_samples_init(ssb,cmd);
615 ssbox_paired_init(ssb,cmd);
624 /* Despatcher for the populate method */
626 ssbox_populate(struct ssbox *ssb,struct cmd_t_test *cmd)
628 ssb->populate(ssb,cmd);
632 /* Despatcher for finalize */
634 ssbox_finalize(struct ssbox *ssb)
640 /* Submit the box and clear up */
642 ssbox_base_finalize(struct ssbox *ssb)
649 /* Initialize a ssbox struct */
651 ssbox_base_init(struct ssbox *this, int cols,int rows)
653 this->finalize = ssbox_base_finalize;
654 this->t = tab_create (cols, rows, 0);
656 tab_columns (this->t, SOM_COL_DOWN, 1);
657 tab_headers (this->t,0,0,1,0);
658 tab_box (this->t, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
659 tab_hline(this->t, TAL_2,0,cols-1,1);
660 tab_dim (this->t, tab_natural_dimensions);
663 void ssbox_one_sample_populate(struct ssbox *ssb,
664 struct cmd_t_test *cmd);
666 /* Initialize the one_sample ssbox */
668 ssbox_one_sample_init(struct ssbox *this,
669 struct cmd_t_test *cmd )
672 const int vsize=cmd->n_variables+1;
674 this->populate = ssbox_one_sample_populate;
676 ssbox_base_init(this, hsize,vsize);
677 tab_title (this->t, 0, _("One-Sample Statistics"));
678 tab_vline(this->t, TAL_2, 1,0,vsize - 1);
679 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, _("N"));
680 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
681 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
682 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
685 void ssbox_independent_samples_populate(struct ssbox *ssb,
686 struct cmd_t_test *cmd);
688 /* Initialize the independent samples ssbox */
690 ssbox_independent_samples_init(struct ssbox *this,
691 struct cmd_t_test *cmd)
694 int vsize = cmd->n_variables*2 +1;
696 this->populate = ssbox_independent_samples_populate;
698 ssbox_base_init(this, hsize,vsize);
699 tab_title (this->t, 0, _("Group Statistics"));
700 tab_vline(this->t,0,1,0,vsize - 1);
701 tab_text (this->t, 1, 0, TAB_CENTER | TAT_TITLE, indep_var->name);
702 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("N"));
703 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
704 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
705 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
709 /* Populate the ssbox for independent samples */
711 ssbox_independent_samples_populate(struct ssbox *ssb,
712 struct cmd_t_test *cmd)
718 double indep_value[2];
720 char prefix[2][3]={"",""};
722 if ( indep_var->type == NUMERIC )
724 val_lab0 = val_labs_find( indep_var->val_labs,gp.v.g_value[0]);
725 val_lab1 = val_labs_find( indep_var->val_labs,gp.v.g_value[1]);
729 val_lab0 = gp.v.g_value[0].s;
730 val_lab1 = gp.v.g_value[1].s;
733 if (gp.criterion == CMP_LE )
735 strcpy(prefix[0],"< ");
736 strcpy(prefix[1],">=");
737 indep_value[0] = gp.v.critical_value;
738 indep_value[1] = gp.v.critical_value;
742 indep_value[0] = gp.v.g_value[0].f;
743 indep_value[1] = gp.v.g_value[1].f;
748 for (i=0; i < cmd->n_variables; ++i)
750 struct variable *var = cmd->v_variables[i];
751 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
754 tab_text (ssb->t, 0, i*2+1, TAB_LEFT, cmd->v_variables[i]->name);
757 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
758 "%s%s", prefix[0], val_lab0);
760 tab_text (ssb->t, 1, i*2+1, TAB_LEFT | TAT_PRINTF,
761 "%s%g", prefix[0], indep_value[0]);
765 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
766 "%s%s", prefix[1], val_lab1);
768 tab_text (ssb->t, 1, i*2+1+1, TAB_LEFT | TAT_PRINTF,
769 "%s%g", prefix[1], indep_value[1]);
772 /* Fill in the group statistics */
773 for ( count = 0 ; count < 2 ; ++count )
775 union value search_val;
777 struct group_statistics *gs;
779 if ( gp.criterion == CMP_LE )
783 /* less than ( < ) case */
784 search_val.f = gp.v.critical_value - 1.0;
789 search_val.f = gp.v.critical_value + 1.0;
794 search_val = gp.v.g_value[count];
797 gs = hsh_find(grp_hash, (void *) &search_val);
800 tab_float(ssb->t, 2 ,i*2+count+1, TAB_RIGHT, gs->n, 2, 0);
801 tab_float(ssb->t, 3 ,i*2+count+1, TAB_RIGHT, gs->mean, 8, 2);
802 tab_float(ssb->t, 4 ,i*2+count+1, TAB_RIGHT, gs->std_dev, 8, 3);
803 tab_float(ssb->t, 5 ,i*2+count+1, TAB_RIGHT, gs->se_mean, 8, 3);
809 void ssbox_paired_populate(struct ssbox *ssb,
810 struct cmd_t_test *cmd);
812 /* Initialize the paired values ssbox */
814 ssbox_paired_init(struct ssbox *this, struct cmd_t_test *cmd UNUSED)
818 int vsize = n_pairs*2+1;
820 this->populate = ssbox_paired_populate;
822 ssbox_base_init(this, hsize,vsize);
823 tab_title (this->t, 0, _("Paired Sample Statistics"));
824 tab_vline(this->t,TAL_0,1,0,vsize-1);
825 tab_vline(this->t,TAL_2,2,0,vsize-1);
826 tab_text (this->t, 2, 0, TAB_CENTER | TAT_TITLE, _("Mean"));
827 tab_text (this->t, 3, 0, TAB_CENTER | TAT_TITLE, _("N"));
828 tab_text (this->t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
829 tab_text (this->t, 5, 0, TAB_CENTER | TAT_TITLE, _("SE. Mean"));
833 /* Populate the ssbox for paired values */
835 ssbox_paired_populate(struct ssbox *ssb,struct cmd_t_test *cmd UNUSED)
841 for (i=0; i < n_pairs; ++i)
845 tab_text (ssb->t, 0, i*2+1, TAB_LEFT | TAT_PRINTF , _("Pair %d"),i);
847 for (j=0 ; j < 2 ; ++j)
849 struct group_statistics *gs;
851 gs = &group_proc_get (pairs[i].v[j])->ugs;
855 tab_text (ssb->t, 1, i*2+j+1, TAB_LEFT, pairs[i].v[j]->name);
858 tab_float (ssb->t,2, i*2+j+1, TAB_RIGHT, pairs[i].mean[j], 8, 2);
859 tab_float (ssb->t,3, i*2+j+1, TAB_RIGHT, pairs[i].n, 2, 0);
860 tab_float (ssb->t,4, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j], 8, 3);
861 tab_float (ssb->t,5, i*2+j+1, TAB_RIGHT, pairs[i].std_dev[j]/sqrt(pairs[i].n), 8, 3);
867 /* Populate the one sample ssbox */
869 ssbox_one_sample_populate(struct ssbox *ssb, struct cmd_t_test *cmd)
875 for (i=0; i < cmd->n_variables; ++i)
877 struct group_statistics *gs = &group_proc_get (cmd->v_variables[i])->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];
1012 struct group_proc *grp_data = group_proc_get (var);
1014 struct hsh_table *grp_hash = grp_data->group_hash;
1016 struct group_statistics *gs0 ;
1017 struct group_statistics *gs1 ;
1019 union value search_val;
1021 if ( gp.criterion == CMP_LE )
1022 search_val.f = gp.v.critical_value - 1.0;
1024 search_val = gp.v.g_value[0];
1026 gs0 = hsh_find(grp_hash, (void *) &search_val);
1029 if ( gp.criterion == CMP_LE )
1030 search_val.f = gp.v.critical_value + 1.0;
1032 search_val = gp.v.g_value[1];
1034 gs1 = hsh_find(grp_hash, (void *) &search_val);
1038 tab_text (self->t, 0, i*2+3, TAB_LEFT, cmd->v_variables[i]->name);
1040 tab_text (self->t, 1, i*2+3, TAB_LEFT, _("Equal variances assumed"));
1043 tab_float(self->t, 2, i*2+3, TAB_CENTER, grp_data->levene, 8,3);
1045 /* Now work out the significance of the Levene test */
1046 df1 = 1; df2 = grp_data->ugs.n - 2;
1047 q = gsl_cdf_fdist_Q(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[0]);
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 = &group_proc_get (cmd->v_variables[i])->ugs;
1286 tab_text (trb->t, 0, i+3, TAB_LEFT, cmd->v_variables[i]->name);
1288 t = (gs->mean - cmd->n_testval[0] ) * sqrt(gs->n) / gs->std_dev ;
1290 tab_float (trb->t, 1, i+3, TAB_RIGHT, t, 8,3);
1292 /* degrees of freedom */
1295 tab_float (trb->t, 2, i+3, TAB_RIGHT, df, 8,0);
1297 p = gsl_cdf_tdist_P(t, df);
1298 q = gsl_cdf_tdist_Q(t, df);
1300 /* Multiply by 2 to get 2-tailed significance, makeing sure we've got
1302 tab_float (trb->t, 3, i+3, TAB_RIGHT, 2.0*(t>0?q:p), 8,3);
1304 tab_float (trb->t, 4, i+3, TAB_RIGHT, gs->mean_diff, 8,3);
1307 q = (1 - cmd->criteria)/2.0; /* 2-tailed test */
1308 t = gsl_cdf_tdist_Qinv(q, df);
1310 tab_float (trb->t, 5, i+3, TAB_RIGHT,
1311 gs->mean_diff - t * gs->se_mean, 8,4);
1313 tab_float (trb->t, 6, i+3, TAB_RIGHT,
1314 gs->mean_diff + t * gs->se_mean, 8,4);
1318 /* Base initializer for the generalized trbox */
1320 trbox_base_init(struct trbox *self, int data_rows, int cols)
1322 const int rows = 3 + data_rows;
1324 self->finalize = trbox_base_finalize;
1325 self->t = tab_create (cols, rows, 0);
1326 tab_headers (self->t,0,0,3,0);
1327 tab_box (self->t, TAL_2, TAL_2, TAL_0, TAL_0, 0, 0, cols -1, rows -1);
1328 tab_hline(self->t, TAL_2,0,cols-1,3);
1329 tab_dim (self->t, tab_natural_dimensions);
1333 /* Base finalizer for the trbox */
1335 trbox_base_finalize(struct trbox *trb)
1341 /* Create , populate and submit the Paired Samples Correlation box */
1345 const int rows=1+n_pairs;
1349 struct tab_table *table;
1351 table = tab_create (cols,rows,0);
1353 tab_columns (table, SOM_COL_DOWN, 1);
1354 tab_headers (table,0,0,1,0);
1355 tab_box (table, TAL_2, TAL_2, TAL_0, TAL_1, 0, 0, cols -1, rows -1 );
1356 tab_hline(table, TAL_2, 0, cols - 1, 1);
1357 tab_vline(table, TAL_2, 2, 0, rows - 1);
1358 tab_dim(table, tab_natural_dimensions);
1359 tab_title(table, 0, _("Paired Samples Correlations"));
1361 /* column headings */
1362 tab_text(table, 2,0, TAB_CENTER | TAT_TITLE, _("N"));
1363 tab_text(table, 3,0, TAB_CENTER | TAT_TITLE, _("Correlation"));
1364 tab_text(table, 4,0, TAB_CENTER | TAT_TITLE, _("Sig."));
1366 for (i=0; i < n_pairs; ++i)
1370 double df = pairs[i].n -2;
1372 double correlation_t =
1373 pairs[i].correlation * sqrt(df) /
1374 sqrt(1 - pow2(pairs[i].correlation));
1378 tab_text(table, 0,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1381 tab_text(table, 1,i+1, TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1382 _("%s & %s"), pairs[i].v[0]->name, pairs[i].v[1]->name);
1386 tab_float(table, 2, i+1, TAB_RIGHT, pairs[i].n, 4, 0);
1387 tab_float(table, 3, i+1, TAB_RIGHT, pairs[i].correlation, 8, 3);
1389 p = gsl_cdf_tdist_P(correlation_t, df);
1390 q = gsl_cdf_tdist_Q(correlation_t, df);
1392 tab_float(table, 4, i+1, TAB_RIGHT, 2.0*(correlation_t>0?q:p), 8, 3);
1401 /* Calculation Implementation */
1403 /* Per case calculations common to all variants of the T test */
1405 common_calc (const struct ccase *c, void *_cmd)
1408 struct cmd_t_test *cmd = (struct cmd_t_test *)_cmd;
1410 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1413 /* Skip the entire case if /MISSING=LISTWISE is set */
1414 if ( cmd->miss == TTS_LISTWISE )
1416 for(i=0; i< cmd->n_variables ; ++i)
1418 struct variable *v = cmd->v_variables[i];
1419 const union value *val = case_data (c, v->fv);
1421 if (value_is_missing(&v->miss, val) )
1428 /* Listwise has to be implicit if the independent variable is missing ?? */
1429 if ( cmd->sbc_groups )
1431 const union value *gv = case_data (c, indep_var->fv);
1432 if ( value_is_missing(&indep_var->miss, gv) )
1439 for(i=0; i< cmd->n_variables ; ++i)
1441 struct group_statistics *gs;
1442 struct variable *v = cmd->v_variables[i];
1443 const union value *val = case_data (c, v->fv);
1445 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1447 if (! value_is_missing(&v->miss, val) )
1450 gs->sum+=weight * val->f;
1451 gs->ssq+=weight * val->f * val->f;
1457 /* Pre calculations common to all variants of the T test */
1459 common_precalc ( struct cmd_t_test *cmd )
1463 for(i=0; i< cmd->n_variables ; ++i)
1465 struct group_statistics *gs;
1466 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1475 /* Post calculations common to all variants of the T test */
1477 common_postcalc ( struct cmd_t_test *cmd )
1482 for(i=0; i< cmd->n_variables ; ++i)
1484 struct group_statistics *gs;
1485 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1487 gs->mean=gs->sum / gs->n;
1488 gs->s_std_dev= sqrt(
1489 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1494 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1497 gs->se_mean = gs->std_dev / sqrt(gs->n);
1498 gs->mean_diff= gs->sum_diff / gs->n;
1502 /* Per case calculations for one sample t test */
1504 one_sample_calc (const struct ccase *c, void *cmd_)
1507 struct cmd_t_test *cmd = (struct cmd_t_test *)cmd_;
1510 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1512 /* Skip the entire case if /MISSING=LISTWISE is set */
1513 if ( cmd->miss == TTS_LISTWISE )
1515 for(i=0; i< cmd->n_variables ; ++i)
1517 struct variable *v = cmd->v_variables[i];
1518 const union value *val = case_data (c, v->fv);
1520 if (value_is_missing(&v->miss, val) )
1527 for(i=0; i< cmd->n_variables ; ++i)
1529 struct group_statistics *gs;
1530 struct variable *v = cmd->v_variables[i];
1531 const union value *val = case_data (c, v->fv);
1533 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1535 if ( ! value_is_missing(&v->miss, val))
1536 gs->sum_diff += weight * (val->f - cmd->n_testval[0]);
1542 /* Pre calculations for one sample t test */
1544 one_sample_precalc ( struct cmd_t_test *cmd )
1548 for(i=0; i< cmd->n_variables ; ++i)
1550 struct group_statistics *gs;
1551 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1557 /* Post calculations for one sample t test */
1559 one_sample_postcalc (struct cmd_t_test *cmd)
1563 for(i=0; i< cmd->n_variables ; ++i)
1565 struct group_statistics *gs;
1566 gs= &group_proc_get (cmd->v_variables[i])->ugs;
1568 gs->mean_diff = gs->sum_diff / gs->n ;
1575 paired_precalc (struct cmd_t_test *cmd UNUSED)
1579 for(i=0; i < n_pairs ; ++i )
1582 pairs[i].sum[0] = 0; pairs[i].sum[1] = 0;
1583 pairs[i].ssq[0] = 0; pairs[i].ssq[1] = 0;
1584 pairs[i].sum_of_prod = 0;
1585 pairs[i].correlation = 0;
1586 pairs[i].sum_of_diffs = 0;
1587 pairs[i].ssq_diffs = 0;
1594 paired_calc (const struct ccase *c, void *cmd_)
1598 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1600 double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1602 /* Skip the entire case if /MISSING=LISTWISE is set ,
1603 AND one member of a pair is missing */
1604 if ( cmd->miss == TTS_LISTWISE )
1606 for(i=0; i < n_pairs ; ++i )
1608 struct variable *v0 = pairs[i].v[0];
1609 struct variable *v1 = pairs[i].v[1];
1611 const union value *val0 = case_data (c, v0->fv);
1612 const union value *val1 = case_data (c, v1->fv);
1614 if ( value_is_missing(&v0->miss, val0) ||
1615 value_is_missing(&v1->miss, val1) )
1622 for(i=0; i < n_pairs ; ++i )
1624 struct variable *v0 = pairs[i].v[0];
1625 struct variable *v1 = pairs[i].v[1];
1627 const union value *val0 = case_data (c, v0->fv);
1628 const union value *val1 = case_data (c, v1->fv);
1630 if ( ( !value_is_missing(&v0->miss, val0)
1631 && !value_is_missing(&v1->miss, val1) ) )
1633 pairs[i].n += weight;
1634 pairs[i].sum[0] += weight * val0->f;
1635 pairs[i].sum[1] += weight * val1->f;
1637 pairs[i].ssq[0] += weight * pow2(val0->f);
1638 pairs[i].ssq[1] += weight * pow2(val1->f);
1640 pairs[i].sum_of_prod += weight * val0->f * val1->f ;
1642 pairs[i].sum_of_diffs += weight * ( val0->f - val1->f ) ;
1643 pairs[i].ssq_diffs += weight * pow2(val0->f - val1->f);
1651 paired_postcalc (struct cmd_t_test *cmd UNUSED)
1655 for(i=0; i < n_pairs ; ++i )
1658 const double n = pairs[i].n;
1660 for (j=0; j < 2 ; ++j)
1662 pairs[i].mean[j] = pairs[i].sum[j] / n ;
1663 pairs[i].s_std_dev[j] = sqrt((pairs[i].ssq[j] / n -
1664 pow2(pairs[i].mean[j]))
1667 pairs[i].std_dev[j] = sqrt(n/(n-1)*(pairs[i].ssq[j] / n -
1668 pow2(pairs[i].mean[j]))
1672 pairs[i].correlation = pairs[i].sum_of_prod / pairs[i].n -
1673 pairs[i].mean[0] * pairs[i].mean[1] ;
1674 /* correlation now actually contains the covariance */
1676 pairs[i].correlation /= pairs[i].std_dev[0] * pairs[i].std_dev[1];
1677 pairs[i].correlation *= pairs[i].n / ( pairs[i].n - 1 );
1679 pairs[i].mean_diff = pairs[i].sum_of_diffs / n ;
1681 pairs[i].std_dev_diff = sqrt ( n / (n - 1) * (
1682 ( pairs[i].ssq_diffs / n )
1684 pow2(pairs[i].mean_diff )
1690 group_precalc (struct cmd_t_test *cmd )
1695 for(i=0; i< cmd->n_variables ; ++i)
1697 struct group_proc *ttpr = group_proc_get (cmd->v_variables[i]);
1699 /* There's always 2 groups for a T - TEST */
1702 gp.indep_width = indep_var->width;
1704 ttpr->group_hash = hsh_create(2,
1705 (hsh_compare_func *) compare_group_binary,
1706 (hsh_hash_func *) hash_group_binary,
1707 (hsh_free_func *) free_group,
1710 for (j=0 ; j < 2 ; ++j)
1713 struct group_statistics *gs = (struct group_statistics *)
1714 xmalloc (sizeof(struct group_statistics));
1720 if ( gp.criterion == CMP_EQ )
1722 gs->id = gp.v.g_value[j];
1727 gs->id.f = gp.v.critical_value - 1.0 ;
1729 gs->id.f = gp.v.critical_value + 1.0 ;
1732 hsh_insert ( ttpr->group_hash, (void *) gs );
1740 group_calc (const struct ccase *c, struct cmd_t_test *cmd)
1744 const union value *gv = case_data (c, indep_var->fv);
1746 const double weight = dict_get_case_weight(default_dict,c,&bad_weight_warn);
1748 if ( value_is_missing(&indep_var->miss, gv) )
1753 if ( cmd->miss == TTS_LISTWISE )
1755 for(i=0; i< cmd->n_variables ; ++i)
1757 struct variable *v = cmd->v_variables[i];
1758 const union value *val = case_data (c, v->fv);
1760 if (value_is_missing(&v->miss, val) )
1767 gv = case_data (c, indep_var->fv);
1769 for(i=0; i< cmd->n_variables ; ++i)
1771 struct variable *var = cmd->v_variables[i];
1772 const union value *val = case_data (c, var->fv);
1773 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1774 struct group_statistics *gs;
1776 gs = hsh_find(grp_hash, (void *) gv);
1778 /* If the independent variable doesn't match either of the values
1779 for this case then move on to the next case */
1783 if ( !value_is_missing(&var->miss, val) )
1786 gs->sum+=weight * val->f;
1787 gs->ssq+=weight * pow2(val->f);
1796 group_postcalc ( struct cmd_t_test *cmd )
1800 for(i=0; i< cmd->n_variables ; ++i)
1802 struct variable *var = cmd->v_variables[i];
1803 struct hsh_table *grp_hash = group_proc_get (var)->group_hash;
1804 struct hsh_iterator g;
1805 struct group_statistics *gs;
1808 for (gs = hsh_first (grp_hash,&g);
1810 gs = hsh_next(grp_hash,&g))
1812 gs->mean = gs->sum / gs->n;
1814 gs->s_std_dev= sqrt(
1815 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1820 ( (gs->ssq / gs->n ) - gs->mean * gs->mean )
1823 gs->se_mean = gs->std_dev / sqrt(gs->n);
1833 calculate(const struct casefile *cf, void *cmd_)
1835 struct ssbox stat_summary_box;
1836 struct trbox test_results_box;
1838 struct casereader *r;
1841 struct cmd_t_test *cmd = (struct cmd_t_test *) cmd_;
1843 common_precalc(cmd);
1844 for(r = casefile_get_reader (cf);
1845 casereader_read (r, &c) ;
1848 common_calc(&c,cmd);
1850 casereader_destroy (r);
1851 common_postcalc(cmd);
1856 one_sample_precalc(cmd);
1857 for(r = casefile_get_reader (cf);
1858 casereader_read (r, &c) ;
1861 one_sample_calc(&c,cmd);
1863 casereader_destroy (r);
1864 one_sample_postcalc(cmd);
1868 paired_precalc(cmd);
1869 for(r = casefile_get_reader (cf);
1870 casereader_read (r, &c) ;
1873 paired_calc(&c,cmd);
1875 casereader_destroy (r);
1876 paired_postcalc(cmd);
1882 for(r = casefile_get_reader (cf);
1883 casereader_read (r, &c) ;
1888 casereader_destroy (r);
1889 group_postcalc(cmd);
1891 levene(cf, indep_var, cmd->n_variables, cmd->v_variables,
1892 (cmd->miss == TTS_LISTWISE)?LEV_LISTWISE:LEV_ANALYSIS ,
1897 ssbox_create(&stat_summary_box,cmd,mode);
1898 ssbox_populate(&stat_summary_box,cmd);
1899 ssbox_finalize(&stat_summary_box);
1901 if ( mode == T_PAIRED)
1904 trbox_create(&test_results_box,cmd,mode);
1905 trbox_populate(&test_results_box,cmd);
1906 trbox_finalize(&test_results_box);
1910 short which_group(const struct group_statistics *g,
1911 const struct group_properties *p);
1913 /* Return -1 if the id of a is less than b; +1 if greater than and
1916 compare_group_binary(const struct group_statistics *a,
1917 const struct group_statistics *b,
1918 const struct group_properties *p)
1923 if ( p->criterion == CMP_LE )
1925 /* less-than-or-equal comparision is not meaningfull for
1926 alpha variables, so we shouldn't ever arrive here */
1927 assert(p->indep_width == 0 ) ;
1929 flag_a = ( a->id.f < p->v.critical_value ) ;
1930 flag_b = ( b->id.f < p->v.critical_value ) ;
1934 flag_a = which_group(a, p);
1935 flag_b = which_group(b, p);
1938 if (flag_a < flag_b )
1941 return (flag_a > flag_b);
1944 /* This is a degenerate case of a hash, since it can only return three possible
1945 values. It's really a comparison, being used as a hash function */
1948 hash_group_binary(const struct group_statistics *g,
1949 const struct group_properties *p)
1953 if ( p->criterion == CMP_LE )
1955 /* Not meaningfull to do a less than compare for alpha values ? */
1956 assert(p->indep_width == 0 ) ;
1957 flag = ( g->id.f < p->v.critical_value ) ;
1959 else if ( p->criterion == CMP_EQ)
1961 flag = which_group(g,p);
1969 /* return 0 if G belongs to group 0,
1970 1 if it belongs to group 1,
1971 2 if it belongs to neither group */
1973 which_group(const struct group_statistics *g,
1974 const struct group_properties *p)
1977 if ( 0 == compare_values (&g->id, &p->v.g_value[0], p->indep_width))
1980 if ( 0 == compare_values (&g->id, &p->v.g_value[1], p->indep_width))