1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2009, 2010 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
19 - Pearson's R (but not Spearman!) is off a little.
20 - T values for Spearman's R and Pearson's R are wrong.
21 - How to calculate significance of symmetric and directional measures?
22 - Asymmetric ASEs and T values for lambda are wrong.
23 - ASE of Goodman and Kruskal's tau is not calculated.
24 - ASE of symmetric somers' d is wrong.
25 - Approx. T of uncertainty coefficient is wrong.
32 #include <gsl/gsl_cdf.h>
36 #include <data/case.h>
37 #include <data/casegrouper.h>
38 #include <data/casereader.h>
39 #include <data/data-out.h>
40 #include <data/dictionary.h>
41 #include <data/format.h>
42 #include <data/procedure.h>
43 #include <data/value-labels.h>
44 #include <data/variable.h>
45 #include <language/command.h>
46 #include <language/dictionary/split-file.h>
47 #include <language/lexer/lexer.h>
48 #include <language/lexer/variable-parser.h>
49 #include <libpspp/array.h>
50 #include <libpspp/assertion.h>
51 #include <libpspp/compiler.h>
52 #include <libpspp/hash.h>
53 #include <libpspp/hmap.h>
54 #include <libpspp/hmapx.h>
55 #include <libpspp/message.h>
56 #include <libpspp/misc.h>
57 #include <libpspp/pool.h>
58 #include <libpspp/str.h>
59 #include <output/tab.h>
66 #define _(msgid) gettext (msgid)
67 #define N_(msgid) msgid
75 missing=miss:!table/include/report;
76 +write[wr_]=none,cells,all;
77 +format=fmt:!labels/nolabels/novallabs,
80 tabl:!tables/notables,
83 +cells[cl_]=count,expected,row,column,total,residual,sresidual,
85 +statistics[st_]=chisq,phi,cc,lambda,uc,none,btau,ctau,risk,gamma,d,
91 /* Number of chi-square statistics. */
94 /* Number of symmetric statistics. */
97 /* Number of directional statistics. */
98 #define N_DIRECTIONAL 13
100 /* A single table entry for general mode. */
103 struct hmap_node node; /* Entry in hash table. */
104 double freq; /* Frequency count. */
105 union value values[1]; /* Values. */
109 table_entry_size (size_t n_values)
111 return (offsetof (struct table_entry, values)
112 + n_values * sizeof (union value));
115 /* Indexes into the 'vars' member of struct pivot_table and
116 struct crosstab member. */
119 ROW_VAR = 0, /* Row variable. */
120 COL_VAR = 1 /* Column variable. */
121 /* Higher indexes cause multiple tables to be output. */
124 /* A crosstabulation of 2 or more variables. */
127 struct fmt_spec weight_format; /* Format for weight variable. */
128 double missing; /* Weight of missing cases. */
130 /* Variables (2 or more). */
132 const struct variable **vars;
134 /* Constants (0 or more). */
136 const struct variable **const_vars;
137 union value *const_values;
141 struct table_entry **entries;
144 /* Column values, number of columns. */
148 /* Row values, number of rows. */
152 /* Number of statistically interesting columns/rows
153 (columns/rows with data in them). */
154 int ns_cols, ns_rows;
156 /* Matrix contents. */
157 double *mat; /* Matrix proper. */
158 double *row_tot; /* Row totals. */
159 double *col_tot; /* Column totals. */
160 double total; /* Grand total. */
163 /* Integer mode variable info. */
166 int min; /* Minimum value. */
167 int max; /* Maximum value + 1. */
168 int count; /* max - min. */
171 static inline struct var_range *
172 get_var_range (const struct variable *v)
174 return var_get_aux (v);
177 struct crosstabs_proc
179 const struct dictionary *dict;
180 enum { INTEGER, GENERAL } mode;
181 enum mv_class exclude;
184 struct fmt_spec weight_format;
186 /* Variables specifies on VARIABLES. */
187 const struct variable **variables;
191 struct pivot_table *pivots;
195 int n_cells; /* Number of cells requested. */
196 unsigned int cells; /* Bit k is 1 if cell k is requested. */
197 int a_cells[CRS_CL_count]; /* 0...n_cells-1 are the requested cells. */
200 unsigned int statistics; /* Bit k is 1 if statistic k is requested. */
204 init_proc (struct crosstabs_proc *proc, struct dataset *ds)
206 const struct variable *wv = dict_get_weight (dataset_dict (ds));
207 proc->dict = dataset_dict (ds);
208 proc->bad_warn = true;
209 proc->variables = NULL;
210 proc->n_variables = 0;
213 proc->weight_format = wv ? *var_get_print_format (wv) : F_8_0;
217 free_proc (struct crosstabs_proc *proc)
219 struct pivot_table *pt;
221 free (proc->variables);
222 for (pt = &proc->pivots[0]; pt < &proc->pivots[proc->n_pivots]; pt++)
225 free (pt->const_vars);
226 /* We must not call value_destroy on const_values because
227 it is a wild pointer; it never pointed to anything owned
230 The rest of the data was allocated and destroyed at a
231 lower level already. */
236 static int internal_cmd_crosstabs (struct lexer *lexer, struct dataset *ds,
237 struct crosstabs_proc *);
238 static bool should_tabulate_case (const struct pivot_table *,
239 const struct ccase *, enum mv_class exclude);
240 static void tabulate_general_case (struct pivot_table *, const struct ccase *,
242 static void tabulate_integer_case (struct pivot_table *, const struct ccase *,
244 static void postcalc (struct crosstabs_proc *);
245 static void submit (struct pivot_table *, struct tab_table *);
247 /* Parse and execute CROSSTABS, then clean up. */
249 cmd_crosstabs (struct lexer *lexer, struct dataset *ds)
251 struct crosstabs_proc proc;
254 init_proc (&proc, ds);
255 result = internal_cmd_crosstabs (lexer, ds, &proc);
261 /* Parses and executes the CROSSTABS procedure. */
263 internal_cmd_crosstabs (struct lexer *lexer, struct dataset *ds,
264 struct crosstabs_proc *proc)
266 struct casegrouper *grouper;
267 struct casereader *input, *group;
268 struct cmd_crosstabs cmd;
269 struct pivot_table *pt;
273 if (!parse_crosstabs (lexer, ds, &cmd, proc))
276 proc->mode = proc->n_variables ? INTEGER : GENERAL;
280 proc->cells = 1u << CRS_CL_COUNT;
281 else if (cmd.a_cells[CRS_CL_ALL])
282 proc->cells = UINT_MAX;
286 for (i = 0; i < CRS_CL_count; i++)
288 proc->cells |= 1u << i;
289 if (proc->cells == 0)
290 proc->cells = ((1u << CRS_CL_COUNT)
292 | (1u << CRS_CL_COLUMN)
293 | (1u << CRS_CL_TOTAL));
295 proc->cells &= ((1u << CRS_CL_count) - 1);
296 proc->cells &= ~((1u << CRS_CL_NONE) | (1u << CRS_CL_ALL));
298 for (i = 0; i < CRS_CL_count; i++)
299 if (proc->cells & (1u << i))
300 proc->a_cells[proc->n_cells++] = i;
303 if (cmd.a_statistics[CRS_ST_ALL])
304 proc->statistics = UINT_MAX;
305 else if (cmd.sbc_statistics)
309 proc->statistics = 0;
310 for (i = 0; i < CRS_ST_count; i++)
311 if (cmd.a_statistics[i])
312 proc->statistics |= 1u << i;
313 if (proc->statistics == 0)
314 proc->statistics |= 1u << CRS_ST_CHISQ;
317 proc->statistics = 0;
320 proc->exclude = (cmd.miss == CRS_TABLE ? MV_ANY
321 : cmd.miss == CRS_INCLUDE ? MV_SYSTEM
323 if (proc->mode == GENERAL && proc->mode == MV_NEVER)
325 msg (SE, _("Missing mode REPORT not allowed in general mode. "
326 "Assuming MISSING=TABLE."));
331 proc->pivot = cmd.pivot == CRS_PIVOT;
333 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
335 grouper = casegrouper_create_splits (input, dataset_dict (ds));
336 while (casegrouper_get_next_group (grouper, &group))
340 /* Output SPLIT FILE variables. */
341 c = casereader_peek (group, 0);
344 output_split_file_values (ds, c);
348 /* Initialize hash tables. */
349 for (pt = &proc->pivots[0]; pt < &proc->pivots[proc->n_pivots]; pt++)
350 hmap_init (&pt->data);
353 for (; (c = casereader_read (group)) != NULL; case_unref (c))
354 for (pt = &proc->pivots[0]; pt < &proc->pivots[proc->n_pivots]; pt++)
356 double weight = dict_get_case_weight (dataset_dict (ds), c,
358 if (should_tabulate_case (pt, c, proc->exclude))
360 if (proc->mode == GENERAL)
361 tabulate_general_case (pt, c, weight);
363 tabulate_integer_case (pt, c, weight);
366 pt->missing += weight;
368 casereader_destroy (group);
373 ok = casegrouper_destroy (grouper);
374 ok = proc_commit (ds) && ok;
376 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
379 /* Parses the TABLES subcommand. */
381 crs_custom_tables (struct lexer *lexer, struct dataset *ds,
382 struct cmd_crosstabs *cmd UNUSED, void *proc_)
384 struct crosstabs_proc *proc = proc_;
385 struct const_var_set *var_set;
387 const struct variable ***by = NULL;
389 size_t *by_nvar = NULL;
394 /* Ensure that this is a TABLES subcommand. */
395 if (!lex_match_id (lexer, "TABLES")
396 && (lex_token (lexer) != T_ID ||
397 dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL)
398 && lex_token (lexer) != T_ALL)
400 lex_match (lexer, '=');
402 if (proc->variables != NULL)
403 var_set = const_var_set_create_from_array (proc->variables,
406 var_set = const_var_set_create_from_dict (dataset_dict (ds));
407 assert (var_set != NULL);
411 by = xnrealloc (by, n_by + 1, sizeof *by);
412 by_nvar = xnrealloc (by_nvar, n_by + 1, sizeof *by_nvar);
413 if (!parse_const_var_set_vars (lexer, var_set, &by[n_by], &by_nvar[n_by],
414 PV_NO_DUPLICATE | PV_NO_SCRATCH))
416 if (xalloc_oversized (nx, by_nvar[n_by]))
418 msg (SE, _("Too many cross-tabulation variables or dimensions."));
424 if (!lex_match (lexer, T_BY))
428 lex_error (lexer, _("expecting BY"));
436 by_iter = xcalloc (n_by, sizeof *by_iter);
437 proc->pivots = xnrealloc (proc->pivots,
438 proc->n_pivots + nx, sizeof *proc->pivots);
439 for (i = 0; i < nx; i++)
441 struct pivot_table *pt = &proc->pivots[proc->n_pivots++];
444 pt->weight_format = proc->weight_format;
447 pt->vars = xmalloc (n_by * sizeof *pt->vars);
449 pt->const_vars = NULL;
450 pt->const_values = NULL;
452 for (j = 0; j < n_by; j++)
453 pt->vars[j] = by[j][by_iter[j]];
455 for (j = n_by - 1; j >= 0; j--)
457 if (++by_iter[j] < by_nvar[j])
466 /* All return paths lead here. */
467 for (i = 0; i < n_by; i++)
472 const_var_set_destroy (var_set);
477 /* Parses the VARIABLES subcommand. */
479 crs_custom_variables (struct lexer *lexer, struct dataset *ds,
480 struct cmd_crosstabs *cmd UNUSED, void *proc_)
482 struct crosstabs_proc *proc = proc_;
485 msg (SE, _("VARIABLES must be specified before TABLES."));
489 lex_match (lexer, '=');
493 size_t orig_nv = proc->n_variables;
498 if (!parse_variables_const (lexer, dataset_dict (ds),
499 &proc->variables, &proc->n_variables,
500 (PV_APPEND | PV_NUMERIC
501 | PV_NO_DUPLICATE | PV_NO_SCRATCH)))
504 if (lex_token (lexer) != '(')
506 lex_error (lexer, "expecting `('");
511 if (!lex_force_int (lexer))
513 min = lex_integer (lexer);
516 lex_match (lexer, ',');
518 if (!lex_force_int (lexer))
520 max = lex_integer (lexer);
523 msg (SE, _("Maximum value (%ld) less than minimum value (%ld)."),
529 if (lex_token (lexer) != ')')
531 lex_error (lexer, "expecting `)'");
536 for (i = orig_nv; i < proc->n_variables; i++)
538 struct var_range *vr = xmalloc (sizeof *vr);
541 vr->count = max - min + 1;
542 var_attach_aux (proc->variables[i], vr, var_dtor_free);
545 if (lex_token (lexer) == '/')
552 free (proc->variables);
553 proc->variables = NULL;
554 proc->n_variables = 0;
558 /* Data file processing. */
561 should_tabulate_case (const struct pivot_table *pt, const struct ccase *c,
562 enum mv_class exclude)
565 for (j = 0; j < pt->n_vars; j++)
567 const struct variable *var = pt->vars[j];
568 struct var_range *range = get_var_range (var);
570 if (var_is_value_missing (var, case_data (c, var), exclude))
575 double num = case_num (c, var);
576 if (num < range->min || num > range->max)
584 tabulate_integer_case (struct pivot_table *pt, const struct ccase *c,
587 struct table_entry *te;
592 for (j = 0; j < pt->n_vars; j++)
594 /* Throw away fractional parts of values. */
595 hash = hash_int (case_num (c, pt->vars[j]), hash);
598 HMAP_FOR_EACH_WITH_HASH (te, struct table_entry, node, hash, &pt->data)
600 for (j = 0; j < pt->n_vars; j++)
601 if ((int) case_num (c, pt->vars[j]) != (int) te->values[j].f)
604 /* Found an existing entry. */
611 /* No existing entry. Create a new one. */
612 te = xmalloc (table_entry_size (pt->n_vars));
614 for (j = 0; j < pt->n_vars; j++)
615 te->values[j].f = (int) case_num (c, pt->vars[j]);
616 hmap_insert (&pt->data, &te->node, hash);
620 tabulate_general_case (struct pivot_table *pt, const struct ccase *c,
623 struct table_entry *te;
628 for (j = 0; j < pt->n_vars; j++)
630 const struct variable *var = pt->vars[j];
631 hash = value_hash (case_data (c, var), var_get_width (var), hash);
634 HMAP_FOR_EACH_WITH_HASH (te, struct table_entry, node, hash, &pt->data)
636 for (j = 0; j < pt->n_vars; j++)
638 const struct variable *var = pt->vars[j];
639 if (!value_equal (case_data (c, var), &te->values[j],
640 var_get_width (var)))
644 /* Found an existing entry. */
651 /* No existing entry. Create a new one. */
652 te = xmalloc (table_entry_size (pt->n_vars));
654 for (j = 0; j < pt->n_vars; j++)
656 const struct variable *var = pt->vars[j];
657 value_clone (&te->values[j], case_data (c, var), var_get_width (var));
659 hmap_insert (&pt->data, &te->node, hash);
662 /* Post-data reading calculations. */
664 static int compare_table_entry_vars_3way (const struct table_entry *a,
665 const struct table_entry *b,
666 const struct pivot_table *pt,
668 static int compare_table_entry_3way (const void *ap_, const void *bp_,
670 static void enum_var_values (const struct pivot_table *, int var_idx,
671 union value **valuesp, int *n_values);
672 static void output_pivot_table (struct crosstabs_proc *,
673 struct pivot_table *);
674 static void make_pivot_table_subset (struct pivot_table *pt,
675 size_t row0, size_t row1,
676 struct pivot_table *subset);
677 static void make_summary_table (struct crosstabs_proc *);
678 static bool find_crosstab (struct pivot_table *, size_t *row0p, size_t *row1p);
681 postcalc (struct crosstabs_proc *proc)
683 struct pivot_table *pt;
685 /* Convert hash tables into sorted arrays of entries. */
686 for (pt = &proc->pivots[0]; pt < &proc->pivots[proc->n_pivots]; pt++)
688 struct table_entry *e;
691 pt->n_entries = hmap_count (&pt->data);
692 pt->entries = xnmalloc (pt->n_entries, sizeof *pt->entries);
694 HMAP_FOR_EACH (e, struct table_entry, node, &pt->data)
695 pt->entries[i++] = e;
696 hmap_destroy (&pt->data);
698 sort (pt->entries, pt->n_entries, sizeof *pt->entries,
699 compare_table_entry_3way, pt);
702 make_summary_table (proc);
704 /* Output each pivot table. */
705 for (pt = &proc->pivots[0]; pt < &proc->pivots[proc->n_pivots]; pt++)
707 if (proc->pivot || pt->n_vars == 2)
708 output_pivot_table (proc, pt);
711 size_t row0 = 0, row1 = 0;
712 while (find_crosstab (pt, &row0, &row1))
714 struct pivot_table subset;
715 make_pivot_table_subset (pt, row0, row1, &subset);
716 output_pivot_table (proc, &subset);
721 /* Free output and prepare for next split file. */
722 for (pt = &proc->pivots[0]; pt < &proc->pivots[proc->n_pivots]; pt++)
728 /* Free only the members that were allocated in this
729 function. The other pointer members are either both
730 allocated and destroyed at a lower level (in
731 output_pivot_table), or both allocated and destroyed at
732 a higher level (in crs_custom_tables and free_proc,
734 for (i = 0; i < pt->n_entries; i++)
735 free (pt->entries[i]);
741 make_pivot_table_subset (struct pivot_table *pt, size_t row0, size_t row1,
742 struct pivot_table *subset)
747 assert (pt->n_consts == 0);
748 subset->missing = pt->missing;
750 subset->vars = pt->vars;
751 subset->n_consts = pt->n_vars - 2;
752 subset->const_vars = pt->vars + 2;
753 subset->const_values = &pt->entries[row0]->values[2];
755 subset->entries = &pt->entries[row0];
756 subset->n_entries = row1 - row0;
760 compare_table_entry_var_3way (const struct table_entry *a,
761 const struct table_entry *b,
762 const struct pivot_table *pt,
765 return value_compare_3way (&a->values[idx], &b->values[idx],
766 var_get_width (pt->vars[idx]));
770 compare_table_entry_vars_3way (const struct table_entry *a,
771 const struct table_entry *b,
772 const struct pivot_table *pt,
777 for (i = idx1 - 1; i >= idx0; i--)
779 int cmp = compare_table_entry_var_3way (a, b, pt, i);
786 /* Compare the struct table_entry at *AP to the one at *BP and
787 return a strcmp()-type result. */
789 compare_table_entry_3way (const void *ap_, const void *bp_, const void *pt_)
791 const struct table_entry *const *ap = ap_;
792 const struct table_entry *const *bp = bp_;
793 const struct table_entry *a = *ap;
794 const struct table_entry *b = *bp;
795 const struct pivot_table *pt = pt_;
798 cmp = compare_table_entry_vars_3way (a, b, pt, 2, pt->n_vars);
802 cmp = compare_table_entry_var_3way (a, b, pt, ROW_VAR);
806 return compare_table_entry_var_3way (a, b, pt, COL_VAR);
810 find_first_difference (const struct pivot_table *pt, size_t row)
813 return pt->n_vars - 1;
816 const struct table_entry *a = pt->entries[row];
817 const struct table_entry *b = pt->entries[row - 1];
820 for (col = pt->n_vars - 1; col >= 0; col--)
821 if (compare_table_entry_var_3way (a, b, pt, col))
827 /* Output a table summarizing the cases processed. */
829 make_summary_table (struct crosstabs_proc *proc)
831 struct tab_table *summary;
832 struct pivot_table *pt;
836 summary = tab_create (7, 3 + proc->n_pivots);
837 tab_title (summary, _("Summary."));
838 tab_headers (summary, 1, 0, 3, 0);
839 tab_joint_text (summary, 1, 0, 6, 0, TAB_CENTER, _("Cases"));
840 tab_joint_text (summary, 1, 1, 2, 1, TAB_CENTER, _("Valid"));
841 tab_joint_text (summary, 3, 1, 4, 1, TAB_CENTER, _("Missing"));
842 tab_joint_text (summary, 5, 1, 6, 1, TAB_CENTER, _("Total"));
843 tab_hline (summary, TAL_1, 1, 6, 1);
844 tab_hline (summary, TAL_1, 1, 6, 2);
845 tab_vline (summary, TAL_1, 3, 1, 1);
846 tab_vline (summary, TAL_1, 5, 1, 1);
847 for (i = 0; i < 3; i++)
849 tab_text (summary, 1 + i * 2, 2, TAB_RIGHT, _("N"));
850 tab_text (summary, 2 + i * 2, 2, TAB_RIGHT, _("Percent"));
852 tab_offset (summary, 0, 3);
854 ds_init_empty (&name);
855 for (pt = &proc->pivots[0]; pt < &proc->pivots[proc->n_pivots]; pt++)
861 tab_hline (summary, TAL_1, 0, 6, 0);
864 for (i = 0; i < pt->n_vars; i++)
867 ds_put_cstr (&name, " * ");
868 ds_put_cstr (&name, var_to_string (pt->vars[i]));
870 tab_text (summary, 0, 0, TAB_LEFT, ds_cstr (&name));
873 for (i = 0; i < pt->n_entries; i++)
874 valid += pt->entries[i]->freq;
879 for (i = 0; i < 3; i++)
881 tab_double (summary, i * 2 + 1, 0, TAB_RIGHT, n[i],
882 &proc->weight_format);
883 tab_text_format (summary, i * 2 + 2, 0, TAB_RIGHT, "%.1f%%",
887 tab_next_row (summary);
891 submit (NULL, summary);
896 static struct tab_table *create_crosstab_table (struct crosstabs_proc *,
897 struct pivot_table *);
898 static struct tab_table *create_chisq_table (struct pivot_table *);
899 static struct tab_table *create_sym_table (struct pivot_table *);
900 static struct tab_table *create_risk_table (struct pivot_table *);
901 static struct tab_table *create_direct_table (struct pivot_table *);
902 static void display_dimensions (struct crosstabs_proc *, struct pivot_table *,
903 struct tab_table *, int first_difference);
904 static void display_crosstabulation (struct crosstabs_proc *,
905 struct pivot_table *,
907 static void display_chisq (struct pivot_table *, struct tab_table *,
908 bool *showed_fisher);
909 static void display_symmetric (struct crosstabs_proc *, struct pivot_table *,
911 static void display_risk (struct pivot_table *, struct tab_table *);
912 static void display_directional (struct crosstabs_proc *, struct pivot_table *,
914 static void table_value_missing (struct crosstabs_proc *proc,
915 struct tab_table *table, int c, int r,
916 unsigned char opt, const union value *v,
917 const struct variable *var);
918 static void delete_missing (struct pivot_table *);
919 static void build_matrix (struct pivot_table *);
921 /* Output pivot table beginning at PB and continuing until PE,
922 exclusive. For efficiency, *MATP is a pointer to a matrix that can
923 hold *MAXROWS entries. */
925 output_pivot_table (struct crosstabs_proc *proc, struct pivot_table *pt)
927 struct tab_table *table = NULL; /* Crosstabulation table. */
928 struct tab_table *chisq = NULL; /* Chi-square table. */
929 bool showed_fisher = false;
930 struct tab_table *sym = NULL; /* Symmetric measures table. */
931 struct tab_table *risk = NULL; /* Risk estimate table. */
932 struct tab_table *direct = NULL; /* Directional measures table. */
935 enum_var_values (pt, COL_VAR, &pt->cols, &pt->n_cols);
938 table = create_crosstab_table (proc, pt);
939 if (proc->statistics & (1u << CRS_ST_CHISQ))
940 chisq = create_chisq_table (pt);
941 if (proc->statistics & ((1u << CRS_ST_PHI) | (1u << CRS_ST_CC)
942 | (1u << CRS_ST_BTAU) | (1u << CRS_ST_CTAU)
943 | (1u << CRS_ST_GAMMA) | (1u << CRS_ST_CORR)
944 | (1u << CRS_ST_KAPPA)))
945 sym = create_sym_table (pt);
946 if (proc->statistics & (1u << CRS_ST_RISK))
947 risk = create_risk_table (pt);
948 if (proc->statistics & ((1u << CRS_ST_LAMBDA) | (1u << CRS_ST_UC)
949 | (1u << CRS_ST_D) | (1u << CRS_ST_ETA)))
950 direct = create_direct_table (pt);
953 while (find_crosstab (pt, &row0, &row1))
955 struct pivot_table x;
956 int first_difference;
958 make_pivot_table_subset (pt, row0, row1, &x);
960 /* Find all the row variable values. */
961 enum_var_values (&x, ROW_VAR, &x.rows, &x.n_rows);
963 if (size_overflow_p (xtimes (xtimes (x.n_rows, x.n_cols),
966 x.row_tot = xmalloc (x.n_rows * sizeof *x.row_tot);
967 x.col_tot = xmalloc (x.n_cols * sizeof *x.col_tot);
968 x.mat = xmalloc (x.n_rows * x.n_cols * sizeof *x.mat);
970 /* Allocate table space for the matrix. */
972 && tab_row (table) + (x.n_rows + 1) * proc->n_cells > tab_nr (table))
973 tab_realloc (table, -1,
974 MAX (tab_nr (table) + (x.n_rows + 1) * proc->n_cells,
975 tab_nr (table) * pt->n_entries / x.n_entries));
979 /* Find the first variable that differs from the last subtable. */
980 first_difference = find_first_difference (pt, row0);
983 display_dimensions (proc, &x, table, first_difference);
984 display_crosstabulation (proc, &x, table);
987 if (proc->exclude == MV_NEVER)
992 display_dimensions (proc, &x, chisq, first_difference);
993 display_chisq (&x, chisq, &showed_fisher);
997 display_dimensions (proc, &x, sym, first_difference);
998 display_symmetric (proc, &x, sym);
1002 display_dimensions (proc, &x, risk, first_difference);
1003 display_risk (&x, risk);
1007 display_dimensions (proc, &x, direct, first_difference);
1008 display_directional (proc, &x, direct);
1011 /* Free the parts of x that are not owned by pt. In
1012 particular we must not free x.cols, which is the same as
1013 pt->cols, which is freed at the end of this function. */
1021 submit (NULL, table);
1026 tab_resize (chisq, 4 + (pt->n_vars - 2), -1);
1032 submit (pt, direct);
1038 build_matrix (struct pivot_table *x)
1040 const int col_var_width = var_get_width (x->vars[COL_VAR]);
1041 const int row_var_width = var_get_width (x->vars[ROW_VAR]);
1044 struct table_entry **p;
1048 for (p = x->entries; p < &x->entries[x->n_entries]; p++)
1050 const struct table_entry *te = *p;
1052 while (!value_equal (&x->rows[row], &te->values[ROW_VAR], row_var_width))
1054 for (; col < x->n_cols; col++)
1060 while (!value_equal (&x->cols[col], &te->values[COL_VAR], col_var_width))
1067 if (++col >= x->n_cols)
1073 while (mp < &x->mat[x->n_cols * x->n_rows])
1075 assert (mp == &x->mat[x->n_cols * x->n_rows]);
1077 /* Column totals, row totals, ns_rows. */
1079 for (col = 0; col < x->n_cols; col++)
1080 x->col_tot[col] = 0.0;
1081 for (row = 0; row < x->n_rows; row++)
1082 x->row_tot[row] = 0.0;
1084 for (row = 0; row < x->n_rows; row++)
1086 bool row_is_empty = true;
1087 for (col = 0; col < x->n_cols; col++)
1091 row_is_empty = false;
1092 x->col_tot[col] += *mp;
1093 x->row_tot[row] += *mp;
1100 assert (mp == &x->mat[x->n_cols * x->n_rows]);
1104 for (col = 0; col < x->n_cols; col++)
1105 for (row = 0; row < x->n_rows; row++)
1106 if (x->mat[col + row * x->n_cols] != 0.0)
1114 for (col = 0; col < x->n_cols; col++)
1115 x->total += x->col_tot[col];
1118 static struct tab_table *
1119 create_crosstab_table (struct crosstabs_proc *proc, struct pivot_table *pt)
1126 static const struct tuple names[] =
1128 {CRS_CL_COUNT, N_("count")},
1129 {CRS_CL_ROW, N_("row %")},
1130 {CRS_CL_COLUMN, N_("column %")},
1131 {CRS_CL_TOTAL, N_("total %")},
1132 {CRS_CL_EXPECTED, N_("expected")},
1133 {CRS_CL_RESIDUAL, N_("residual")},
1134 {CRS_CL_SRESIDUAL, N_("std. resid.")},
1135 {CRS_CL_ASRESIDUAL, N_("adj. resid.")},
1137 const int n_names = sizeof names / sizeof *names;
1138 const struct tuple *t;
1140 struct tab_table *table;
1141 struct string title;
1142 struct pivot_table x;
1146 make_pivot_table_subset (pt, 0, 0, &x);
1148 table = tab_create (x.n_consts + 1 + x.n_cols + 1,
1149 (x.n_entries / x.n_cols) * 3 / 2 * proc->n_cells + 10);
1150 tab_headers (table, x.n_consts + 1, 0, 2, 0);
1152 /* First header line. */
1153 tab_joint_text (table, x.n_consts + 1, 0,
1154 (x.n_consts + 1) + (x.n_cols - 1), 0,
1155 TAB_CENTER | TAT_TITLE, var_get_name (x.vars[COL_VAR]));
1157 tab_hline (table, TAL_1, x.n_consts + 1,
1158 x.n_consts + 2 + x.n_cols - 2, 1);
1160 /* Second header line. */
1161 for (i = 2; i < x.n_consts + 2; i++)
1162 tab_joint_text (table, x.n_consts + 2 - i - 1, 0,
1163 x.n_consts + 2 - i - 1, 1,
1164 TAB_RIGHT | TAT_TITLE, var_to_string (x.vars[i]));
1165 tab_text (table, x.n_consts + 2 - 2, 1, TAB_RIGHT | TAT_TITLE,
1166 var_get_name (x.vars[ROW_VAR]));
1167 for (i = 0; i < x.n_cols; i++)
1168 table_value_missing (proc, table, x.n_consts + 2 + i - 1, 1, TAB_RIGHT,
1169 &x.cols[i], x.vars[COL_VAR]);
1170 tab_text (table, x.n_consts + 2 + x.n_cols - 1, 1, TAB_CENTER, _("Total"));
1172 tab_hline (table, TAL_1, 0, x.n_consts + 2 + x.n_cols - 1, 2);
1173 tab_vline (table, TAL_1, x.n_consts + 2 + x.n_cols - 1, 0, 1);
1176 ds_init_empty (&title);
1177 for (i = 0; i < x.n_consts + 2; i++)
1180 ds_put_cstr (&title, " * ");
1181 ds_put_cstr (&title, var_get_name (x.vars[i]));
1183 for (i = 0; i < pt->n_consts; i++)
1185 const struct variable *var = pt->const_vars[i];
1189 ds_put_format (&title, ", %s=", var_get_name (var));
1191 /* Insert the formatted value of the variable, then trim
1192 leading spaces in what was just inserted. */
1193 ofs = ds_length (&title);
1194 s = data_out (&pt->const_values[i], dict_get_encoding (proc->dict), var_get_print_format (var));
1195 ds_put_cstr (&title, s);
1197 ds_remove (&title, ofs, ss_cspan (ds_substr (&title, ofs, SIZE_MAX),
1201 ds_put_cstr (&title, " [");
1203 for (t = names; t < &names[n_names]; t++)
1204 if (proc->cells & (1u << t->value))
1207 ds_put_cstr (&title, ", ");
1208 ds_put_cstr (&title, gettext (t->name));
1210 ds_put_cstr (&title, "].");
1212 tab_title (table, "%s", ds_cstr (&title));
1213 ds_destroy (&title);
1215 tab_offset (table, 0, 2);
1219 static struct tab_table *
1220 create_chisq_table (struct pivot_table *pt)
1222 struct tab_table *chisq;
1224 chisq = tab_create (6 + (pt->n_vars - 2),
1225 pt->n_entries / pt->n_cols * 3 / 2 * N_CHISQ + 10);
1226 tab_headers (chisq, 1 + (pt->n_vars - 2), 0, 1, 0);
1228 tab_title (chisq, _("Chi-square tests."));
1230 tab_offset (chisq, pt->n_vars - 2, 0);
1231 tab_text (chisq, 0, 0, TAB_LEFT | TAT_TITLE, _("Statistic"));
1232 tab_text (chisq, 1, 0, TAB_RIGHT | TAT_TITLE, _("Value"));
1233 tab_text (chisq, 2, 0, TAB_RIGHT | TAT_TITLE, _("df"));
1234 tab_text (chisq, 3, 0, TAB_RIGHT | TAT_TITLE,
1235 _("Asymp. Sig. (2-sided)"));
1236 tab_text (chisq, 4, 0, TAB_RIGHT | TAT_TITLE,
1237 _("Exact Sig. (2-sided)"));
1238 tab_text (chisq, 5, 0, TAB_RIGHT | TAT_TITLE,
1239 _("Exact Sig. (1-sided)"));
1240 tab_offset (chisq, 0, 1);
1245 /* Symmetric measures. */
1246 static struct tab_table *
1247 create_sym_table (struct pivot_table *pt)
1249 struct tab_table *sym;
1251 sym = tab_create (6 + (pt->n_vars - 2),
1252 pt->n_entries / pt->n_cols * 7 + 10);
1253 tab_headers (sym, 2 + (pt->n_vars - 2), 0, 1, 0);
1254 tab_title (sym, _("Symmetric measures."));
1256 tab_offset (sym, pt->n_vars - 2, 0);
1257 tab_text (sym, 0, 0, TAB_LEFT | TAT_TITLE, _("Category"));
1258 tab_text (sym, 1, 0, TAB_LEFT | TAT_TITLE, _("Statistic"));
1259 tab_text (sym, 2, 0, TAB_RIGHT | TAT_TITLE, _("Value"));
1260 tab_text (sym, 3, 0, TAB_RIGHT | TAT_TITLE, _("Asymp. Std. Error"));
1261 tab_text (sym, 4, 0, TAB_RIGHT | TAT_TITLE, _("Approx. T"));
1262 tab_text (sym, 5, 0, TAB_RIGHT | TAT_TITLE, _("Approx. Sig."));
1263 tab_offset (sym, 0, 1);
1268 /* Risk estimate. */
1269 static struct tab_table *
1270 create_risk_table (struct pivot_table *pt)
1272 struct tab_table *risk;
1274 risk = tab_create (4 + (pt->n_vars - 2), pt->n_entries / pt->n_cols * 4 + 10);
1275 tab_headers (risk, 1 + pt->n_vars - 2, 0, 2, 0);
1276 tab_title (risk, _("Risk estimate."));
1278 tab_offset (risk, pt->n_vars - 2, 0);
1279 tab_joint_text_format (risk, 2, 0, 3, 0, TAB_CENTER | TAT_TITLE,
1280 _("95%% Confidence Interval"));
1281 tab_text (risk, 0, 1, TAB_LEFT | TAT_TITLE, _("Statistic"));
1282 tab_text (risk, 1, 1, TAB_RIGHT | TAT_TITLE, _("Value"));
1283 tab_text (risk, 2, 1, TAB_RIGHT | TAT_TITLE, _("Lower"));
1284 tab_text (risk, 3, 1, TAB_RIGHT | TAT_TITLE, _("Upper"));
1285 tab_hline (risk, TAL_1, 2, 3, 1);
1286 tab_vline (risk, TAL_1, 2, 0, 1);
1287 tab_offset (risk, 0, 2);
1292 /* Directional measures. */
1293 static struct tab_table *
1294 create_direct_table (struct pivot_table *pt)
1296 struct tab_table *direct;
1298 direct = tab_create (7 + (pt->n_vars - 2),
1299 pt->n_entries / pt->n_cols * 7 + 10);
1300 tab_headers (direct, 3 + (pt->n_vars - 2), 0, 1, 0);
1301 tab_title (direct, _("Directional measures."));
1303 tab_offset (direct, pt->n_vars - 2, 0);
1304 tab_text (direct, 0, 0, TAB_LEFT | TAT_TITLE, _("Category"));
1305 tab_text (direct, 1, 0, TAB_LEFT | TAT_TITLE, _("Statistic"));
1306 tab_text (direct, 2, 0, TAB_LEFT | TAT_TITLE, _("Type"));
1307 tab_text (direct, 3, 0, TAB_RIGHT | TAT_TITLE, _("Value"));
1308 tab_text (direct, 4, 0, TAB_RIGHT | TAT_TITLE, _("Asymp. Std. Error"));
1309 tab_text (direct, 5, 0, TAB_RIGHT | TAT_TITLE, _("Approx. T"));
1310 tab_text (direct, 6, 0, TAB_RIGHT | TAT_TITLE, _("Approx. Sig."));
1311 tab_offset (direct, 0, 1);
1317 /* Delete missing rows and columns for statistical analysis when
1320 delete_missing (struct pivot_table *pt)
1324 for (r = 0; r < pt->n_rows; r++)
1325 if (var_is_num_missing (pt->vars[ROW_VAR], pt->rows[r].f, MV_USER))
1327 for (c = 0; c < pt->n_cols; c++)
1328 pt->mat[c + r * pt->n_cols] = 0.;
1333 for (c = 0; c < pt->n_cols; c++)
1334 if (var_is_num_missing (pt->vars[COL_VAR], pt->cols[c].f, MV_USER))
1336 for (r = 0; r < pt->n_rows; r++)
1337 pt->mat[c + r * pt->n_cols] = 0.;
1342 /* Prepare table T for submission, and submit it. */
1344 submit (struct pivot_table *pt, struct tab_table *t)
1351 tab_resize (t, -1, 0);
1352 if (tab_nr (t) == tab_t (t))
1354 table_unref (&t->table);
1357 tab_offset (t, 0, 0);
1359 for (i = 2; i < pt->n_vars; i++)
1360 tab_text (t, pt->n_vars - i - 1, 0, TAB_RIGHT | TAT_TITLE,
1361 var_to_string (pt->vars[i]));
1362 tab_box (t, TAL_2, TAL_2, -1, -1, 0, 0, tab_nc (t) - 1, tab_nr (t) - 1);
1363 tab_box (t, -1, -1, -1, TAL_1, tab_l (t), tab_t (t) - 1, tab_nc (t) - 1,
1365 tab_box (t, -1, -1, -1, TAL_GAP, 0, tab_t (t), tab_l (t) - 1,
1367 tab_vline (t, TAL_2, tab_l (t), 0, tab_nr (t) - 1);
1373 find_crosstab (struct pivot_table *pt, size_t *row0p, size_t *row1p)
1375 size_t row0 = *row1p;
1378 if (row0 >= pt->n_entries)
1381 for (row1 = row0 + 1; row1 < pt->n_entries; row1++)
1383 struct table_entry *a = pt->entries[row0];
1384 struct table_entry *b = pt->entries[row1];
1385 if (compare_table_entry_vars_3way (a, b, pt, 2, pt->n_vars) != 0)
1393 /* Compares `union value's A_ and B_ and returns a strcmp()-like
1394 result. WIDTH_ points to an int which is either 0 for a
1395 numeric value or a string width for a string value. */
1397 compare_value_3way (const void *a_, const void *b_, const void *width_)
1399 const union value *a = a_;
1400 const union value *b = b_;
1401 const int *width = width_;
1403 return value_compare_3way (a, b, *width);
1406 /* Given an array of ENTRY_CNT table_entry structures starting at
1407 ENTRIES, creates a sorted list of the values that the variable
1408 with index VAR_IDX takes on. The values are returned as a
1409 malloc()'d array stored in *VALUES, with the number of values
1410 stored in *VALUE_CNT.
1413 enum_var_values (const struct pivot_table *pt, int var_idx,
1414 union value **valuesp, int *n_values)
1416 const struct variable *var = pt->vars[var_idx];
1417 struct var_range *range = get_var_range (var);
1418 union value *values;
1423 values = *valuesp = xnmalloc (range->count, sizeof *values);
1424 *n_values = range->count;
1425 for (i = 0; i < range->count; i++)
1426 values[i].f = range->min + i;
1430 int width = var_get_width (var);
1431 struct hmapx_node *node;
1432 const union value *iter;
1436 for (i = 0; i < pt->n_entries; i++)
1438 const struct table_entry *te = pt->entries[i];
1439 const union value *value = &te->values[var_idx];
1440 size_t hash = value_hash (value, width, 0);
1442 HMAPX_FOR_EACH_WITH_HASH (iter, node, hash, &set)
1443 if (value_equal (iter, value, width))
1446 hmapx_insert (&set, (union value *) value, hash);
1451 *n_values = hmapx_count (&set);
1452 values = *valuesp = xnmalloc (*n_values, sizeof *values);
1454 HMAPX_FOR_EACH (iter, node, &set)
1455 values[i++] = *iter;
1456 hmapx_destroy (&set);
1458 sort (values, *n_values, sizeof *values, compare_value_3way, &width);
1462 /* Sets cell (C,R) in TABLE, with options OPT, to have a value taken
1463 from V, displayed with print format spec from variable VAR. When
1464 in REPORT missing-value mode, missing values have an M appended. */
1466 table_value_missing (struct crosstabs_proc *proc,
1467 struct tab_table *table, int c, int r, unsigned char opt,
1468 const union value *v, const struct variable *var)
1470 const char *label = var_lookup_value_label (var, v);
1472 tab_text (table, c, r, TAB_LEFT, label);
1475 const struct fmt_spec *print = var_get_print_format (var);
1476 if (proc->exclude == MV_NEVER && var_is_value_missing (var, v, MV_USER))
1478 char *s = data_out (v, dict_get_encoding (proc->dict), print);
1479 tab_text_format (table, c, r, opt, "%sM", s + strspn (s, " "));
1483 tab_value (table, c, r, opt, v, proc->dict, print);
1487 /* Draws a line across TABLE at the current row to indicate the most
1488 major dimension variable with index FIRST_DIFFERENCE out of N_VARS
1489 that changed, and puts the values that changed into the table. TB
1490 and PT must be the corresponding table_entry and crosstab,
1493 display_dimensions (struct crosstabs_proc *proc, struct pivot_table *pt,
1494 struct tab_table *table, int first_difference)
1496 tab_hline (table, TAL_1, pt->n_consts + pt->n_vars - first_difference - 1, tab_nc (table) - 1, 0);
1498 for (; first_difference >= 2; first_difference--)
1499 table_value_missing (proc, table, pt->n_consts + pt->n_vars - first_difference - 1, 0,
1500 TAB_RIGHT, &pt->entries[0]->values[first_difference],
1501 pt->vars[first_difference]);
1504 /* Put VALUE into cell (C,R) of TABLE, suffixed with character
1505 SUFFIX if nonzero. If MARK_MISSING is true the entry is
1506 additionally suffixed with a letter `M'. */
1508 format_cell_entry (struct tab_table *table, int c, int r, double value,
1509 char suffix, bool mark_missing, const struct dictionary *dict)
1511 const struct fmt_spec f = {FMT_F, 10, 1};
1518 s = data_out (&v, dict_get_encoding (dict), &f);
1522 suffixes[suffix_len++] = suffix;
1524 suffixes[suffix_len++] = 'M';
1525 suffixes[suffix_len] = '\0';
1527 tab_text_format (table, c, r, TAB_RIGHT, "%s%s",
1528 s + strspn (s, " "), suffixes);
1531 /* Displays the crosstabulation table. */
1533 display_crosstabulation (struct crosstabs_proc *proc, struct pivot_table *pt,
1534 struct tab_table *table)
1540 for (r = 0; r < pt->n_rows; r++)
1541 table_value_missing (proc, table, pt->n_consts + pt->n_vars - 2,
1542 r * proc->n_cells, TAB_RIGHT, &pt->rows[r],
1545 tab_text (table, pt->n_vars - 2, pt->n_rows * proc->n_cells,
1546 TAB_LEFT, _("Total"));
1548 /* Put in the actual cells. */
1550 tab_offset (table, pt->n_consts + pt->n_vars - 1, -1);
1551 for (r = 0; r < pt->n_rows; r++)
1553 if (proc->n_cells > 1)
1554 tab_hline (table, TAL_1, -1, pt->n_cols, 0);
1555 for (c = 0; c < pt->n_cols; c++)
1557 bool mark_missing = false;
1558 double expected_value = pt->row_tot[r] * pt->col_tot[c] / pt->total;
1559 if (proc->exclude == MV_NEVER
1560 && (var_is_num_missing (pt->vars[COL_VAR], pt->cols[c].f, MV_USER)
1561 || var_is_num_missing (pt->vars[ROW_VAR], pt->rows[r].f,
1563 mark_missing = true;
1564 for (i = 0; i < proc->n_cells; i++)
1569 switch (proc->a_cells[i])
1575 v = *mp / pt->row_tot[r] * 100.;
1579 v = *mp / pt->col_tot[c] * 100.;
1583 v = *mp / pt->total * 100.;
1586 case CRS_CL_EXPECTED:
1589 case CRS_CL_RESIDUAL:
1590 v = *mp - expected_value;
1592 case CRS_CL_SRESIDUAL:
1593 v = (*mp - expected_value) / sqrt (expected_value);
1595 case CRS_CL_ASRESIDUAL:
1596 v = ((*mp - expected_value)
1597 / sqrt (expected_value
1598 * (1. - pt->row_tot[r] / pt->total)
1599 * (1. - pt->col_tot[c] / pt->total)));
1604 format_cell_entry (table, c, i, v, suffix, mark_missing, proc->dict);
1610 tab_offset (table, -1, tab_row (table) + proc->n_cells);
1614 tab_offset (table, -1, tab_row (table) - proc->n_cells * pt->n_rows);
1615 for (r = 0; r < pt->n_rows; r++)
1617 bool mark_missing = false;
1619 if (proc->exclude == MV_NEVER
1620 && var_is_num_missing (pt->vars[ROW_VAR], pt->rows[r].f, MV_USER))
1621 mark_missing = true;
1623 for (i = 0; i < proc->n_cells; i++)
1628 switch (proc->a_cells[i])
1638 v = pt->row_tot[r] / pt->total * 100.;
1642 v = pt->row_tot[r] / pt->total * 100.;
1645 case CRS_CL_EXPECTED:
1646 case CRS_CL_RESIDUAL:
1647 case CRS_CL_SRESIDUAL:
1648 case CRS_CL_ASRESIDUAL:
1655 format_cell_entry (table, pt->n_cols, 0, v, suffix, mark_missing, proc->dict);
1656 tab_next_row (table);
1660 /* Column totals, grand total. */
1662 if (proc->n_cells > 1)
1663 tab_hline (table, TAL_1, -1, pt->n_cols, 0);
1664 for (c = 0; c <= pt->n_cols; c++)
1666 double ct = c < pt->n_cols ? pt->col_tot[c] : pt->total;
1667 bool mark_missing = false;
1670 if (proc->exclude == MV_NEVER && c < pt->n_cols
1671 && var_is_num_missing (pt->vars[COL_VAR], pt->cols[c].f, MV_USER))
1672 mark_missing = true;
1674 for (i = 0; i < proc->n_cells; i++)
1679 switch (proc->a_cells[i])
1685 v = ct / pt->total * 100.;
1693 v = ct / pt->total * 100.;
1696 case CRS_CL_EXPECTED:
1697 case CRS_CL_RESIDUAL:
1698 case CRS_CL_SRESIDUAL:
1699 case CRS_CL_ASRESIDUAL:
1705 format_cell_entry (table, c, i, v, suffix, mark_missing, proc->dict);
1710 tab_offset (table, -1, tab_row (table) + last_row);
1711 tab_offset (table, 0, -1);
1714 static void calc_r (struct pivot_table *,
1715 double *PT, double *Y, double *, double *, double *);
1716 static void calc_chisq (struct pivot_table *,
1717 double[N_CHISQ], int[N_CHISQ], double *, double *);
1719 /* Display chi-square statistics. */
1721 display_chisq (struct pivot_table *pt, struct tab_table *chisq,
1722 bool *showed_fisher)
1724 static const char *chisq_stats[N_CHISQ] =
1726 N_("Pearson Chi-Square"),
1727 N_("Likelihood Ratio"),
1728 N_("Fisher's Exact Test"),
1729 N_("Continuity Correction"),
1730 N_("Linear-by-Linear Association"),
1732 double chisq_v[N_CHISQ];
1733 double fisher1, fisher2;
1738 calc_chisq (pt, chisq_v, df, &fisher1, &fisher2);
1740 tab_offset (chisq, pt->n_vars - 2, -1);
1742 for (i = 0; i < N_CHISQ; i++)
1744 if ((i != 2 && chisq_v[i] == SYSMIS)
1745 || (i == 2 && fisher1 == SYSMIS))
1748 tab_text (chisq, 0, 0, TAB_LEFT, gettext (chisq_stats[i]));
1751 tab_double (chisq, 1, 0, TAB_RIGHT, chisq_v[i], NULL);
1752 tab_double (chisq, 2, 0, TAB_RIGHT, df[i], &pt->weight_format);
1753 tab_double (chisq, 3, 0, TAB_RIGHT,
1754 gsl_cdf_chisq_Q (chisq_v[i], df[i]), NULL);
1758 *showed_fisher = true;
1759 tab_double (chisq, 4, 0, TAB_RIGHT, fisher2, NULL);
1760 tab_double (chisq, 5, 0, TAB_RIGHT, fisher1, NULL);
1762 tab_next_row (chisq);
1765 tab_text (chisq, 0, 0, TAB_LEFT, _("N of Valid Cases"));
1766 tab_double (chisq, 1, 0, TAB_RIGHT, pt->total, &pt->weight_format);
1767 tab_next_row (chisq);
1769 tab_offset (chisq, 0, -1);
1772 static int calc_symmetric (struct crosstabs_proc *, struct pivot_table *,
1773 double[N_SYMMETRIC], double[N_SYMMETRIC],
1774 double[N_SYMMETRIC],
1775 double[3], double[3], double[3]);
1777 /* Display symmetric measures. */
1779 display_symmetric (struct crosstabs_proc *proc, struct pivot_table *pt,
1780 struct tab_table *sym)
1782 static const char *categories[] =
1784 N_("Nominal by Nominal"),
1785 N_("Ordinal by Ordinal"),
1786 N_("Interval by Interval"),
1787 N_("Measure of Agreement"),
1790 static const char *stats[N_SYMMETRIC] =
1794 N_("Contingency Coefficient"),
1795 N_("Kendall's tau-b"),
1796 N_("Kendall's tau-c"),
1798 N_("Spearman Correlation"),
1803 static const int stats_categories[N_SYMMETRIC] =
1805 0, 0, 0, 1, 1, 1, 1, 2, 3,
1809 double sym_v[N_SYMMETRIC], sym_ase[N_SYMMETRIC], sym_t[N_SYMMETRIC];
1810 double somers_d_v[3], somers_d_ase[3], somers_d_t[3];
1813 if (!calc_symmetric (proc, pt, sym_v, sym_ase, sym_t,
1814 somers_d_v, somers_d_ase, somers_d_t))
1817 tab_offset (sym, pt->n_vars - 2, -1);
1819 for (i = 0; i < N_SYMMETRIC; i++)
1821 if (sym_v[i] == SYSMIS)
1824 if (stats_categories[i] != last_cat)
1826 last_cat = stats_categories[i];
1827 tab_text (sym, 0, 0, TAB_LEFT, gettext (categories[last_cat]));
1830 tab_text (sym, 1, 0, TAB_LEFT, gettext (stats[i]));
1831 tab_double (sym, 2, 0, TAB_RIGHT, sym_v[i], NULL);
1832 if (sym_ase[i] != SYSMIS)
1833 tab_double (sym, 3, 0, TAB_RIGHT, sym_ase[i], NULL);
1834 if (sym_t[i] != SYSMIS)
1835 tab_double (sym, 4, 0, TAB_RIGHT, sym_t[i], NULL);
1836 /*tab_double (sym, 5, 0, TAB_RIGHT, normal_sig (sym_v[i]), NULL);*/
1840 tab_text (sym, 0, 0, TAB_LEFT, _("N of Valid Cases"));
1841 tab_double (sym, 2, 0, TAB_RIGHT, pt->total, &pt->weight_format);
1844 tab_offset (sym, 0, -1);
1847 static int calc_risk (struct pivot_table *,
1848 double[], double[], double[], union value *);
1850 /* Display risk estimate. */
1852 display_risk (struct pivot_table *pt, struct tab_table *risk)
1855 double risk_v[3], lower[3], upper[3];
1859 if (!calc_risk (pt, risk_v, upper, lower, c))
1862 tab_offset (risk, pt->n_vars - 2, -1);
1864 for (i = 0; i < 3; i++)
1866 const struct variable *cv = pt->vars[COL_VAR];
1867 const struct variable *rv = pt->vars[ROW_VAR];
1868 int cvw = var_get_width (cv);
1869 int rvw = var_get_width (rv);
1871 if (risk_v[i] == SYSMIS)
1877 if (var_is_numeric (cv))
1878 sprintf (buf, _("Odds Ratio for %s (%g / %g)"),
1879 var_get_name (cv), c[0].f, c[1].f);
1881 sprintf (buf, _("Odds Ratio for %s (%.*s / %.*s)"),
1883 cvw, value_str (&c[0], cvw),
1884 cvw, value_str (&c[1], cvw));
1888 if (var_is_numeric (rv))
1889 sprintf (buf, _("For cohort %s = %g"),
1890 var_get_name (rv), pt->rows[i - 1].f);
1892 sprintf (buf, _("For cohort %s = %.*s"),
1894 rvw, value_str (&pt->rows[i - 1], rvw));
1898 tab_text (risk, 0, 0, TAB_LEFT, buf);
1899 tab_double (risk, 1, 0, TAB_RIGHT, risk_v[i], NULL);
1900 tab_double (risk, 2, 0, TAB_RIGHT, lower[i], NULL);
1901 tab_double (risk, 3, 0, TAB_RIGHT, upper[i], NULL);
1902 tab_next_row (risk);
1905 tab_text (risk, 0, 0, TAB_LEFT, _("N of Valid Cases"));
1906 tab_double (risk, 1, 0, TAB_RIGHT, pt->total, &pt->weight_format);
1907 tab_next_row (risk);
1909 tab_offset (risk, 0, -1);
1912 static int calc_directional (struct crosstabs_proc *, struct pivot_table *,
1913 double[N_DIRECTIONAL], double[N_DIRECTIONAL],
1914 double[N_DIRECTIONAL]);
1916 /* Display directional measures. */
1918 display_directional (struct crosstabs_proc *proc, struct pivot_table *pt,
1919 struct tab_table *direct)
1921 static const char *categories[] =
1923 N_("Nominal by Nominal"),
1924 N_("Ordinal by Ordinal"),
1925 N_("Nominal by Interval"),
1928 static const char *stats[] =
1931 N_("Goodman and Kruskal tau"),
1932 N_("Uncertainty Coefficient"),
1937 static const char *types[] =
1944 static const int stats_categories[N_DIRECTIONAL] =
1946 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 2, 2,
1949 static const int stats_stats[N_DIRECTIONAL] =
1951 0, 0, 0, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4,
1954 static const int stats_types[N_DIRECTIONAL] =
1956 0, 1, 2, 1, 2, 0, 1, 2, 0, 1, 2, 1, 2,
1959 static const int *stats_lookup[] =
1966 static const char **stats_names[] =
1978 double direct_v[N_DIRECTIONAL];
1979 double direct_ase[N_DIRECTIONAL];
1980 double direct_t[N_DIRECTIONAL];
1984 if (!calc_directional (proc, pt, direct_v, direct_ase, direct_t))
1987 tab_offset (direct, pt->n_vars - 2, -1);
1989 for (i = 0; i < N_DIRECTIONAL; i++)
1991 if (direct_v[i] == SYSMIS)
1997 for (j = 0; j < 3; j++)
1998 if (last[j] != stats_lookup[j][i])
2001 tab_hline (direct, TAL_1, j, 6, 0);
2006 int k = last[j] = stats_lookup[j][i];
2011 string = var_get_name (pt->vars[0]);
2013 string = var_get_name (pt->vars[1]);
2015 tab_text_format (direct, j, 0, TAB_LEFT,
2016 gettext (stats_names[j][k]), string);
2021 tab_double (direct, 3, 0, TAB_RIGHT, direct_v[i], NULL);
2022 if (direct_ase[i] != SYSMIS)
2023 tab_double (direct, 4, 0, TAB_RIGHT, direct_ase[i], NULL);
2024 if (direct_t[i] != SYSMIS)
2025 tab_double (direct, 5, 0, TAB_RIGHT, direct_t[i], NULL);
2026 /*tab_double (direct, 6, 0, TAB_RIGHT, normal_sig (direct_v[i]), NULL);*/
2027 tab_next_row (direct);
2030 tab_offset (direct, 0, -1);
2033 /* Statistical calculations. */
2035 /* Returns the value of the gamma (factorial) function for an integer
2038 gamma_int (double pt)
2043 for (i = 2; i < pt; i++)
2048 /* Calculate P_r as specified in _SPSS Statistical Algorithms_,
2050 static inline double
2051 Pr (int a, int b, int c, int d)
2053 return (gamma_int (a + b + 1.) / gamma_int (a + 1.)
2054 * gamma_int (c + d + 1.) / gamma_int (b + 1.)
2055 * gamma_int (a + c + 1.) / gamma_int (c + 1.)
2056 * gamma_int (b + d + 1.) / gamma_int (d + 1.)
2057 / gamma_int (a + b + c + d + 1.));
2060 /* Swap the contents of A and B. */
2062 swap (int *a, int *b)
2069 /* Calculate significance for Fisher's exact test as specified in
2070 _SPSS Statistical Algorithms_, Appendix 5. */
2072 calc_fisher (int a, int b, int c, int d, double *fisher1, double *fisher2)
2076 if (MIN (c, d) < MIN (a, b))
2077 swap (&a, &c), swap (&b, &d);
2078 if (MIN (b, d) < MIN (a, c))
2079 swap (&a, &b), swap (&c, &d);
2083 swap (&a, &b), swap (&c, &d);
2085 swap (&a, &c), swap (&b, &d);
2089 for (pt = 0; pt <= a; pt++)
2090 *fisher1 += Pr (a - pt, b + pt, c + pt, d - pt);
2092 *fisher2 = *fisher1;
2093 for (pt = 1; pt <= b; pt++)
2094 *fisher2 += Pr (a + pt, b - pt, c - pt, d + pt);
2097 /* Calculates chi-squares into CHISQ. MAT is a matrix with N_COLS
2098 columns with values COLS and N_ROWS rows with values ROWS. Values
2099 in the matrix sum to pt->total. */
2101 calc_chisq (struct pivot_table *pt,
2102 double chisq[N_CHISQ], int df[N_CHISQ],
2103 double *fisher1, double *fisher2)
2107 chisq[0] = chisq[1] = 0.;
2108 chisq[2] = chisq[3] = chisq[4] = SYSMIS;
2109 *fisher1 = *fisher2 = SYSMIS;
2111 df[0] = df[1] = (pt->ns_cols - 1) * (pt->ns_rows - 1);
2113 if (pt->ns_rows <= 1 || pt->ns_cols <= 1)
2115 chisq[0] = chisq[1] = SYSMIS;
2119 for (r = 0; r < pt->n_rows; r++)
2120 for (c = 0; c < pt->n_cols; c++)
2122 const double expected = pt->row_tot[r] * pt->col_tot[c] / pt->total;
2123 const double freq = pt->mat[pt->n_cols * r + c];
2124 const double residual = freq - expected;
2126 chisq[0] += residual * residual / expected;
2128 chisq[1] += freq * log (expected / freq);
2139 /* Calculate Yates and Fisher exact test. */
2140 if (pt->ns_cols == 2 && pt->ns_rows == 2)
2142 double f11, f12, f21, f22;
2148 for (i = j = 0; i < pt->n_cols; i++)
2149 if (pt->col_tot[i] != 0.)
2158 f11 = pt->mat[nz_cols[0]];
2159 f12 = pt->mat[nz_cols[1]];
2160 f21 = pt->mat[nz_cols[0] + pt->n_cols];
2161 f22 = pt->mat[nz_cols[1] + pt->n_cols];
2166 const double pt_ = fabs (f11 * f22 - f12 * f21) - 0.5 * pt->total;
2169 chisq[3] = (pt->total * pow2 (pt_)
2170 / (f11 + f12) / (f21 + f22)
2171 / (f11 + f21) / (f12 + f22));
2179 if (f11 < 5. || f12 < 5. || f21 < 5. || f22 < 5.)
2180 calc_fisher (f11 + .5, f12 + .5, f21 + .5, f22 + .5, fisher1, fisher2);
2183 /* Calculate Mantel-Haenszel. */
2184 if (var_is_numeric (pt->vars[ROW_VAR]) && var_is_numeric (pt->vars[COL_VAR]))
2186 double r, ase_0, ase_1;
2187 calc_r (pt, (double *) pt->rows, (double *) pt->cols, &r, &ase_0, &ase_1);
2189 chisq[4] = (pt->total - 1.) * r * r;
2194 /* Calculate the value of Pearson's r. r is stored into R, ase_1 into
2195 ASE_1, and ase_0 into ASE_0. The row and column values must be
2196 passed in PT and Y. */
2198 calc_r (struct pivot_table *pt,
2199 double *PT, double *Y, double *r, double *ase_0, double *ase_1)
2201 double SX, SY, S, T;
2203 double sum_XYf, sum_X2Y2f;
2204 double sum_Xr, sum_X2r;
2205 double sum_Yc, sum_Y2c;
2208 for (sum_X2Y2f = sum_XYf = 0., i = 0; i < pt->n_rows; i++)
2209 for (j = 0; j < pt->n_cols; j++)
2211 double fij = pt->mat[j + i * pt->n_cols];
2212 double product = PT[i] * Y[j];
2213 double temp = fij * product;
2215 sum_X2Y2f += temp * product;
2218 for (sum_Xr = sum_X2r = 0., i = 0; i < pt->n_rows; i++)
2220 sum_Xr += PT[i] * pt->row_tot[i];
2221 sum_X2r += pow2 (PT[i]) * pt->row_tot[i];
2223 Xbar = sum_Xr / pt->total;
2225 for (sum_Yc = sum_Y2c = 0., i = 0; i < pt->n_cols; i++)
2227 sum_Yc += Y[i] * pt->col_tot[i];
2228 sum_Y2c += Y[i] * Y[i] * pt->col_tot[i];
2230 Ybar = sum_Yc / pt->total;
2232 S = sum_XYf - sum_Xr * sum_Yc / pt->total;
2233 SX = sum_X2r - pow2 (sum_Xr) / pt->total;
2234 SY = sum_Y2c - pow2 (sum_Yc) / pt->total;
2237 *ase_0 = sqrt ((sum_X2Y2f - pow2 (sum_XYf) / pt->total) / (sum_X2r * sum_Y2c));
2242 for (s = c = 0., i = 0; i < pt->n_rows; i++)
2243 for (j = 0; j < pt->n_cols; j++)
2245 double Xresid, Yresid;
2248 Xresid = PT[i] - Xbar;
2249 Yresid = Y[j] - Ybar;
2250 temp = (T * Xresid * Yresid
2252 * (Xresid * Xresid * SY + Yresid * Yresid * SX)));
2253 y = pt->mat[j + i * pt->n_cols] * temp * temp - c;
2258 *ase_1 = sqrt (s) / (T * T);
2262 /* Calculate symmetric statistics and their asymptotic standard
2263 errors. Returns 0 if none could be calculated. */
2265 calc_symmetric (struct crosstabs_proc *proc, struct pivot_table *pt,
2266 double v[N_SYMMETRIC], double ase[N_SYMMETRIC],
2267 double t[N_SYMMETRIC],
2268 double somers_d_v[3], double somers_d_ase[3],
2269 double somers_d_t[3])
2273 q = MIN (pt->ns_rows, pt->ns_cols);
2277 for (i = 0; i < N_SYMMETRIC; i++)
2278 v[i] = ase[i] = t[i] = SYSMIS;
2280 /* Phi, Cramer's V, contingency coefficient. */
2281 if (proc->statistics & ((1u << CRS_ST_PHI) | (1u << CRS_ST_CC)))
2283 double Xp = 0.; /* Pearson chi-square. */
2286 for (r = 0; r < pt->n_rows; r++)
2287 for (c = 0; c < pt->n_cols; c++)
2289 const double expected = pt->row_tot[r] * pt->col_tot[c] / pt->total;
2290 const double freq = pt->mat[pt->n_cols * r + c];
2291 const double residual = freq - expected;
2293 Xp += residual * residual / expected;
2296 if (proc->statistics & (1u << CRS_ST_PHI))
2298 v[0] = sqrt (Xp / pt->total);
2299 v[1] = sqrt (Xp / (pt->total * (q - 1)));
2301 if (proc->statistics & (1u << CRS_ST_CC))
2302 v[2] = sqrt (Xp / (Xp + pt->total));
2305 if (proc->statistics & ((1u << CRS_ST_BTAU) | (1u << CRS_ST_CTAU)
2306 | (1u << CRS_ST_GAMMA) | (1u << CRS_ST_D)))
2311 double btau_cum, ctau_cum, gamma_cum, d_yx_cum, d_xy_cum;
2315 Dr = Dc = pow2 (pt->total);
2316 for (r = 0; r < pt->n_rows; r++)
2317 Dr -= pow2 (pt->row_tot[r]);
2318 for (c = 0; c < pt->n_cols; c++)
2319 Dc -= pow2 (pt->col_tot[c]);
2321 cum = xnmalloc (pt->n_cols * pt->n_rows, sizeof *cum);
2322 for (c = 0; c < pt->n_cols; c++)
2326 for (r = 0; r < pt->n_rows; r++)
2327 cum[c + r * pt->n_cols] = ct += pt->mat[c + r * pt->n_cols];
2336 for (i = 0; i < pt->n_rows; i++)
2340 for (j = 1; j < pt->n_cols; j++)
2341 Cij += pt->col_tot[j] - cum[j + i * pt->n_cols];
2344 for (j = 1; j < pt->n_cols; j++)
2345 Dij += cum[j + (i - 1) * pt->n_cols];
2349 double fij = pt->mat[j + i * pt->n_cols];
2353 if (++j == pt->n_cols)
2355 assert (j < pt->n_cols);
2357 Cij -= pt->col_tot[j] - cum[j + i * pt->n_cols];
2358 Dij += pt->col_tot[j - 1] - cum[j - 1 + i * pt->n_cols];
2362 Cij += cum[j - 1 + (i - 1) * pt->n_cols];
2363 Dij -= cum[j + (i - 1) * pt->n_cols];
2369 if (proc->statistics & (1u << CRS_ST_BTAU))
2370 v[3] = (P - Q) / sqrt (Dr * Dc);
2371 if (proc->statistics & (1u << CRS_ST_CTAU))
2372 v[4] = (q * (P - Q)) / (pow2 (pt->total) * (q - 1));
2373 if (proc->statistics & (1u << CRS_ST_GAMMA))
2374 v[5] = (P - Q) / (P + Q);
2376 /* ASE for tau-b, tau-c, gamma. Calculations could be
2377 eliminated here, at expense of memory. */
2382 btau_cum = ctau_cum = gamma_cum = d_yx_cum = d_xy_cum = 0.;
2383 for (i = 0; i < pt->n_rows; i++)
2387 for (j = 1; j < pt->n_cols; j++)
2388 Cij += pt->col_tot[j] - cum[j + i * pt->n_cols];
2391 for (j = 1; j < pt->n_cols; j++)
2392 Dij += cum[j + (i - 1) * pt->n_cols];
2396 double fij = pt->mat[j + i * pt->n_cols];
2398 if (proc->statistics & (1u << CRS_ST_BTAU))
2400 const double temp = (2. * sqrt (Dr * Dc) * (Cij - Dij)
2401 + v[3] * (pt->row_tot[i] * Dc
2402 + pt->col_tot[j] * Dr));
2403 btau_cum += fij * temp * temp;
2407 const double temp = Cij - Dij;
2408 ctau_cum += fij * temp * temp;
2411 if (proc->statistics & (1u << CRS_ST_GAMMA))
2413 const double temp = Q * Cij - P * Dij;
2414 gamma_cum += fij * temp * temp;
2417 if (proc->statistics & (1u << CRS_ST_D))
2419 d_yx_cum += fij * pow2 (Dr * (Cij - Dij)
2420 - (P - Q) * (pt->total - pt->row_tot[i]));
2421 d_xy_cum += fij * pow2 (Dc * (Dij - Cij)
2422 - (Q - P) * (pt->total - pt->col_tot[j]));
2425 if (++j == pt->n_cols)
2427 assert (j < pt->n_cols);
2429 Cij -= pt->col_tot[j] - cum[j + i * pt->n_cols];
2430 Dij += pt->col_tot[j - 1] - cum[j - 1 + i * pt->n_cols];
2434 Cij += cum[j - 1 + (i - 1) * pt->n_cols];
2435 Dij -= cum[j + (i - 1) * pt->n_cols];
2441 btau_var = ((btau_cum
2442 - (pt->total * pow2 (pt->total * (P - Q) / sqrt (Dr * Dc) * (Dr + Dc))))
2444 if (proc->statistics & (1u << CRS_ST_BTAU))
2446 ase[3] = sqrt (btau_var);
2447 t[3] = v[3] / (2 * sqrt ((ctau_cum - (P - Q) * (P - Q) / pt->total)
2450 if (proc->statistics & (1u << CRS_ST_CTAU))
2452 ase[4] = ((2 * q / ((q - 1) * pow2 (pt->total)))
2453 * sqrt (ctau_cum - (P - Q) * (P - Q) / pt->total));
2454 t[4] = v[4] / ase[4];
2456 if (proc->statistics & (1u << CRS_ST_GAMMA))
2458 ase[5] = ((4. / ((P + Q) * (P + Q))) * sqrt (gamma_cum));
2459 t[5] = v[5] / (2. / (P + Q)
2460 * sqrt (ctau_cum - (P - Q) * (P - Q) / pt->total));
2462 if (proc->statistics & (1u << CRS_ST_D))
2464 somers_d_v[0] = (P - Q) / (.5 * (Dc + Dr));
2465 somers_d_ase[0] = 2. * btau_var / (Dr + Dc) * sqrt (Dr * Dc);
2466 somers_d_t[0] = (somers_d_v[0]
2468 * sqrt (ctau_cum - pow2 (P - Q) / pt->total)));
2469 somers_d_v[1] = (P - Q) / Dc;
2470 somers_d_ase[1] = 2. / pow2 (Dc) * sqrt (d_xy_cum);
2471 somers_d_t[1] = (somers_d_v[1]
2473 * sqrt (ctau_cum - pow2 (P - Q) / pt->total)));
2474 somers_d_v[2] = (P - Q) / Dr;
2475 somers_d_ase[2] = 2. / pow2 (Dr) * sqrt (d_yx_cum);
2476 somers_d_t[2] = (somers_d_v[2]
2478 * sqrt (ctau_cum - pow2 (P - Q) / pt->total)));
2484 /* Spearman correlation, Pearson's r. */
2485 if (proc->statistics & (1u << CRS_ST_CORR))
2487 double *R = xmalloc (sizeof *R * pt->n_rows);
2488 double *C = xmalloc (sizeof *C * pt->n_cols);
2491 double y, t, c = 0., s = 0.;
2496 R[i] = s + (pt->row_tot[i] + 1.) / 2.;
2497 y = pt->row_tot[i] - c;
2501 if (++i == pt->n_rows)
2503 assert (i < pt->n_rows);
2508 double y, t, c = 0., s = 0.;
2513 C[j] = s + (pt->col_tot[j] + 1.) / 2;
2514 y = pt->col_tot[j] - c;
2518 if (++j == pt->n_cols)
2520 assert (j < pt->n_cols);
2524 calc_r (pt, R, C, &v[6], &t[6], &ase[6]);
2530 calc_r (pt, (double *) pt->rows, (double *) pt->cols, &v[7], &t[7], &ase[7]);
2534 /* Cohen's kappa. */
2535 if (proc->statistics & (1u << CRS_ST_KAPPA) && pt->ns_rows == pt->ns_cols)
2537 double sum_fii, sum_rici, sum_fiiri_ci, sum_fijri_ci2, sum_riciri_ci;
2540 for (sum_fii = sum_rici = sum_fiiri_ci = sum_riciri_ci = 0., i = j = 0;
2541 i < pt->ns_rows; i++, j++)
2545 while (pt->col_tot[j] == 0.)
2548 prod = pt->row_tot[i] * pt->col_tot[j];
2549 sum = pt->row_tot[i] + pt->col_tot[j];
2551 sum_fii += pt->mat[j + i * pt->n_cols];
2553 sum_fiiri_ci += pt->mat[j + i * pt->n_cols] * sum;
2554 sum_riciri_ci += prod * sum;
2556 for (sum_fijri_ci2 = 0., i = 0; i < pt->ns_rows; i++)
2557 for (j = 0; j < pt->ns_cols; j++)
2559 double sum = pt->row_tot[i] + pt->col_tot[j];
2560 sum_fijri_ci2 += pt->mat[j + i * pt->n_cols] * sum * sum;
2563 v[8] = (pt->total * sum_fii - sum_rici) / (pow2 (pt->total) - sum_rici);
2565 ase[8] = sqrt ((pow2 (pt->total) * sum_rici
2566 + sum_rici * sum_rici
2567 - pt->total * sum_riciri_ci)
2568 / (pt->total * (pow2 (pt->total) - sum_rici) * (pow2 (pt->total) - sum_rici)));
2570 t[8] = v[8] / sqrt (pt->total * (((sum_fii * (pt->total - sum_fii))
2571 / pow2 (pow2 (pt->total) - sum_rici))
2572 + ((2. * (pt->total - sum_fii)
2573 * (2. * sum_fii * sum_rici
2574 - pt->total * sum_fiiri_ci))
2575 / cube (pow2 (pt->total) - sum_rici))
2576 + (pow2 (pt->total - sum_fii)
2577 * (pt->total * sum_fijri_ci2 - 4.
2578 * sum_rici * sum_rici)
2579 / pow4 (pow2 (pt->total) - sum_rici))));
2581 t[8] = v[8] / ase[8];
2588 /* Calculate risk estimate. */
2590 calc_risk (struct pivot_table *pt,
2591 double *value, double *upper, double *lower, union value *c)
2593 double f11, f12, f21, f22;
2599 for (i = 0; i < 3; i++)
2600 value[i] = upper[i] = lower[i] = SYSMIS;
2603 if (pt->ns_rows != 2 || pt->ns_cols != 2)
2610 for (i = j = 0; i < pt->n_cols; i++)
2611 if (pt->col_tot[i] != 0.)
2620 f11 = pt->mat[nz_cols[0]];
2621 f12 = pt->mat[nz_cols[1]];
2622 f21 = pt->mat[nz_cols[0] + pt->n_cols];
2623 f22 = pt->mat[nz_cols[1] + pt->n_cols];
2625 c[0] = pt->cols[nz_cols[0]];
2626 c[1] = pt->cols[nz_cols[1]];
2629 value[0] = (f11 * f22) / (f12 * f21);
2630 v = sqrt (1. / f11 + 1. / f12 + 1. / f21 + 1. / f22);
2631 lower[0] = value[0] * exp (-1.960 * v);
2632 upper[0] = value[0] * exp (1.960 * v);
2634 value[1] = (f11 * (f21 + f22)) / (f21 * (f11 + f12));
2635 v = sqrt ((f12 / (f11 * (f11 + f12)))
2636 + (f22 / (f21 * (f21 + f22))));
2637 lower[1] = value[1] * exp (-1.960 * v);
2638 upper[1] = value[1] * exp (1.960 * v);
2640 value[2] = (f12 * (f21 + f22)) / (f22 * (f11 + f12));
2641 v = sqrt ((f11 / (f12 * (f11 + f12)))
2642 + (f21 / (f22 * (f21 + f22))));
2643 lower[2] = value[2] * exp (-1.960 * v);
2644 upper[2] = value[2] * exp (1.960 * v);
2649 /* Calculate directional measures. */
2651 calc_directional (struct crosstabs_proc *proc, struct pivot_table *pt,
2652 double v[N_DIRECTIONAL], double ase[N_DIRECTIONAL],
2653 double t[N_DIRECTIONAL])
2658 for (i = 0; i < N_DIRECTIONAL; i++)
2659 v[i] = ase[i] = t[i] = SYSMIS;
2663 if (proc->statistics & (1u << CRS_ST_LAMBDA))
2665 double *fim = xnmalloc (pt->n_rows, sizeof *fim);
2666 int *fim_index = xnmalloc (pt->n_rows, sizeof *fim_index);
2667 double *fmj = xnmalloc (pt->n_cols, sizeof *fmj);
2668 int *fmj_index = xnmalloc (pt->n_cols, sizeof *fmj_index);
2669 double sum_fim, sum_fmj;
2671 int rm_index, cm_index;
2674 /* Find maximum for each row and their sum. */
2675 for (sum_fim = 0., i = 0; i < pt->n_rows; i++)
2677 double max = pt->mat[i * pt->n_cols];
2680 for (j = 1; j < pt->n_cols; j++)
2681 if (pt->mat[j + i * pt->n_cols] > max)
2683 max = pt->mat[j + i * pt->n_cols];
2687 sum_fim += fim[i] = max;
2688 fim_index[i] = index;
2691 /* Find maximum for each column. */
2692 for (sum_fmj = 0., j = 0; j < pt->n_cols; j++)
2694 double max = pt->mat[j];
2697 for (i = 1; i < pt->n_rows; i++)
2698 if (pt->mat[j + i * pt->n_cols] > max)
2700 max = pt->mat[j + i * pt->n_cols];
2704 sum_fmj += fmj[j] = max;
2705 fmj_index[j] = index;
2708 /* Find maximum row total. */
2709 rm = pt->row_tot[0];
2711 for (i = 1; i < pt->n_rows; i++)
2712 if (pt->row_tot[i] > rm)
2714 rm = pt->row_tot[i];
2718 /* Find maximum column total. */
2719 cm = pt->col_tot[0];
2721 for (j = 1; j < pt->n_cols; j++)
2722 if (pt->col_tot[j] > cm)
2724 cm = pt->col_tot[j];
2728 v[0] = (sum_fim + sum_fmj - cm - rm) / (2. * pt->total - rm - cm);
2729 v[1] = (sum_fmj - rm) / (pt->total - rm);
2730 v[2] = (sum_fim - cm) / (pt->total - cm);
2732 /* ASE1 for Y given PT. */
2736 for (accum = 0., i = 0; i < pt->n_rows; i++)
2737 for (j = 0; j < pt->n_cols; j++)
2739 const int deltaj = j == cm_index;
2740 accum += (pt->mat[j + i * pt->n_cols]
2741 * pow2 ((j == fim_index[i])
2746 ase[2] = sqrt (accum - pt->total * v[0]) / (pt->total - cm);
2749 /* ASE0 for Y given PT. */
2753 for (accum = 0., i = 0; i < pt->n_rows; i++)
2754 if (cm_index != fim_index[i])
2755 accum += (pt->mat[i * pt->n_cols + fim_index[i]]
2756 + pt->mat[i * pt->n_cols + cm_index]);
2757 t[2] = v[2] / (sqrt (accum - pow2 (sum_fim - cm) / pt->total) / (pt->total - cm));
2760 /* ASE1 for PT given Y. */
2764 for (accum = 0., i = 0; i < pt->n_rows; i++)
2765 for (j = 0; j < pt->n_cols; j++)
2767 const int deltaj = i == rm_index;
2768 accum += (pt->mat[j + i * pt->n_cols]
2769 * pow2 ((i == fmj_index[j])
2774 ase[1] = sqrt (accum - pt->total * v[0]) / (pt->total - rm);
2777 /* ASE0 for PT given Y. */
2781 for (accum = 0., j = 0; j < pt->n_cols; j++)
2782 if (rm_index != fmj_index[j])
2783 accum += (pt->mat[j + pt->n_cols * fmj_index[j]]
2784 + pt->mat[j + pt->n_cols * rm_index]);
2785 t[1] = v[1] / (sqrt (accum - pow2 (sum_fmj - rm) / pt->total) / (pt->total - rm));
2788 /* Symmetric ASE0 and ASE1. */
2793 for (accum0 = accum1 = 0., i = 0; i < pt->n_rows; i++)
2794 for (j = 0; j < pt->n_cols; j++)
2796 int temp0 = (fmj_index[j] == i) + (fim_index[i] == j);
2797 int temp1 = (i == rm_index) + (j == cm_index);
2798 accum0 += pt->mat[j + i * pt->n_cols] * pow2 (temp0 - temp1);
2799 accum1 += (pt->mat[j + i * pt->n_cols]
2800 * pow2 (temp0 + (v[0] - 1.) * temp1));
2802 ase[0] = sqrt (accum1 - 4. * pt->total * v[0] * v[0]) / (2. * pt->total - rm - cm);
2803 t[0] = v[0] / (sqrt (accum0 - pow2 ((sum_fim + sum_fmj - cm - rm) / pt->total))
2804 / (2. * pt->total - rm - cm));
2813 double sum_fij2_ri, sum_fij2_ci;
2814 double sum_ri2, sum_cj2;
2816 for (sum_fij2_ri = sum_fij2_ci = 0., i = 0; i < pt->n_rows; i++)
2817 for (j = 0; j < pt->n_cols; j++)
2819 double temp = pow2 (pt->mat[j + i * pt->n_cols]);
2820 sum_fij2_ri += temp / pt->row_tot[i];
2821 sum_fij2_ci += temp / pt->col_tot[j];
2824 for (sum_ri2 = 0., i = 0; i < pt->n_rows; i++)
2825 sum_ri2 += pow2 (pt->row_tot[i]);
2827 for (sum_cj2 = 0., j = 0; j < pt->n_cols; j++)
2828 sum_cj2 += pow2 (pt->col_tot[j]);
2830 v[3] = (pt->total * sum_fij2_ci - sum_ri2) / (pow2 (pt->total) - sum_ri2);
2831 v[4] = (pt->total * sum_fij2_ri - sum_cj2) / (pow2 (pt->total) - sum_cj2);
2835 if (proc->statistics & (1u << CRS_ST_UC))
2837 double UX, UY, UXY, P;
2838 double ase1_yx, ase1_xy, ase1_sym;
2841 for (UX = 0., i = 0; i < pt->n_rows; i++)
2842 if (pt->row_tot[i] > 0.)
2843 UX -= pt->row_tot[i] / pt->total * log (pt->row_tot[i] / pt->total);
2845 for (UY = 0., j = 0; j < pt->n_cols; j++)
2846 if (pt->col_tot[j] > 0.)
2847 UY -= pt->col_tot[j] / pt->total * log (pt->col_tot[j] / pt->total);
2849 for (UXY = P = 0., i = 0; i < pt->n_rows; i++)
2850 for (j = 0; j < pt->n_cols; j++)
2852 double entry = pt->mat[j + i * pt->n_cols];
2857 P += entry * pow2 (log (pt->col_tot[j] * pt->row_tot[i] / (pt->total * entry)));
2858 UXY -= entry / pt->total * log (entry / pt->total);
2861 for (ase1_yx = ase1_xy = ase1_sym = 0., i = 0; i < pt->n_rows; i++)
2862 for (j = 0; j < pt->n_cols; j++)
2864 double entry = pt->mat[j + i * pt->n_cols];
2869 ase1_yx += entry * pow2 (UY * log (entry / pt->row_tot[i])
2870 + (UX - UXY) * log (pt->col_tot[j] / pt->total));
2871 ase1_xy += entry * pow2 (UX * log (entry / pt->col_tot[j])
2872 + (UY - UXY) * log (pt->row_tot[i] / pt->total));
2873 ase1_sym += entry * pow2 ((UXY
2874 * log (pt->row_tot[i] * pt->col_tot[j] / pow2 (pt->total)))
2875 - (UX + UY) * log (entry / pt->total));
2878 v[5] = 2. * ((UX + UY - UXY) / (UX + UY));
2879 ase[5] = (2. / (pt->total * pow2 (UX + UY))) * sqrt (ase1_sym);
2880 t[5] = v[5] / ((2. / (pt->total * (UX + UY)))
2881 * sqrt (P - pow2 (UX + UY - UXY) / pt->total));
2883 v[6] = (UX + UY - UXY) / UX;
2884 ase[6] = sqrt (ase1_xy) / (pt->total * UX * UX);
2885 t[6] = v[6] / (sqrt (P - pt->total * pow2 (UX + UY - UXY)) / (pt->total * UX));
2887 v[7] = (UX + UY - UXY) / UY;
2888 ase[7] = sqrt (ase1_yx) / (pt->total * UY * UY);
2889 t[7] = v[7] / (sqrt (P - pt->total * pow2 (UX + UY - UXY)) / (pt->total * UY));
2893 if (proc->statistics & (1u << CRS_ST_D))
2895 double v_dummy[N_SYMMETRIC];
2896 double ase_dummy[N_SYMMETRIC];
2897 double t_dummy[N_SYMMETRIC];
2898 double somers_d_v[3];
2899 double somers_d_ase[3];
2900 double somers_d_t[3];
2902 if (calc_symmetric (proc, pt, v_dummy, ase_dummy, t_dummy,
2903 somers_d_v, somers_d_ase, somers_d_t))
2906 for (i = 0; i < 3; i++)
2908 v[8 + i] = somers_d_v[i];
2909 ase[8 + i] = somers_d_ase[i];
2910 t[8 + i] = somers_d_t[i];
2916 if (proc->statistics & (1u << CRS_ST_ETA))
2919 double sum_Xr, sum_X2r;
2923 for (sum_Xr = sum_X2r = 0., i = 0; i < pt->n_rows; i++)
2925 sum_Xr += pt->rows[i].f * pt->row_tot[i];
2926 sum_X2r += pow2 (pt->rows[i].f) * pt->row_tot[i];
2928 SX = sum_X2r - pow2 (sum_Xr) / pt->total;
2930 for (SXW = 0., j = 0; j < pt->n_cols; j++)
2934 for (cum = 0., i = 0; i < pt->n_rows; i++)
2936 SXW += pow2 (pt->rows[i].f) * pt->mat[j + i * pt->n_cols];
2937 cum += pt->rows[i].f * pt->mat[j + i * pt->n_cols];
2940 SXW -= cum * cum / pt->col_tot[j];
2942 v[11] = sqrt (1. - SXW / SX);
2946 double sum_Yc, sum_Y2c;
2950 for (sum_Yc = sum_Y2c = 0., i = 0; i < pt->n_cols; i++)
2952 sum_Yc += pt->cols[i].f * pt->col_tot[i];
2953 sum_Y2c += pow2 (pt->cols[i].f) * pt->col_tot[i];
2955 SY = sum_Y2c - sum_Yc * sum_Yc / pt->total;
2957 for (SYW = 0., i = 0; i < pt->n_rows; i++)
2961 for (cum = 0., j = 0; j < pt->n_cols; j++)
2963 SYW += pow2 (pt->cols[j].f) * pt->mat[j + i * pt->n_cols];
2964 cum += pt->cols[j].f * pt->mat[j + i * pt->n_cols];
2967 SYW -= cum * cum / pt->row_tot[i];
2969 v[12] = sqrt (1. - SYW / SY);