1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2006, 2009, 2010, 2011, 2012, 2013, 2014, 2016 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 - How to calculate significance of some symmetric and directional measures?
20 - How to calculate ASE for symmetric Somers ' d?
21 - How to calculate ASE for Goodman and Kruskal's tau?
22 - How to calculate approx. T of symmetric uncertainty coefficient?
30 #include <gsl/gsl_cdf.h>
34 #include "data/case.h"
35 #include "data/casegrouper.h"
36 #include "data/casereader.h"
37 #include "data/data-out.h"
38 #include "data/dataset.h"
39 #include "data/dictionary.h"
40 #include "data/format.h"
41 #include "data/value-labels.h"
42 #include "data/variable.h"
43 #include "language/command.h"
44 #include "language/stats/freq.h"
45 #include "language/dictionary/split-file.h"
46 #include "language/lexer/lexer.h"
47 #include "language/lexer/variable-parser.h"
48 #include "libpspp/array.h"
49 #include "libpspp/assertion.h"
50 #include "libpspp/compiler.h"
51 #include "libpspp/hash-functions.h"
52 #include "libpspp/hmap.h"
53 #include "libpspp/hmapx.h"
54 #include "libpspp/message.h"
55 #include "libpspp/misc.h"
56 #include "libpspp/pool.h"
57 #include "libpspp/str.h"
58 #include "output/pivot-table.h"
59 #include "output/chart-item.h"
60 #include "output/charts/barchart.h"
62 #include "gl/minmax.h"
63 #include "gl/xalloc.h"
67 #define _(msgid) gettext (msgid)
68 #define N_(msgid) msgid
76 missing=miss:!table/include/report;
77 count=roundwhat:asis/case/!cell,
78 roundhow:!round/truncate;
79 +write[wr_]=none,cells,all;
80 +format=val:!avalue/dvalue,
82 tabl:!tables/notables,
86 +cells[cl_]=count,expected,row,column,total,residual,sresidual,
88 +statistics[st_]=chisq,phi,cc,lambda,uc,none,btau,ctau,risk,gamma,d,
94 /* Number of chi-square statistics. */
97 /* Number of symmetric statistics. */
100 /* Number of directional statistics. */
101 #define N_DIRECTIONAL 13
104 /* Indexes into the 'vars' member of struct crosstabulation and
105 struct crosstab member. */
108 ROW_VAR = 0, /* Row variable. */
109 COL_VAR = 1 /* Column variable. */
110 /* Higher indexes cause multiple tables to be output. */
115 const struct variable *var;
120 /* A crosstabulation of 2 or more variables. */
121 struct crosstabulation
123 struct crosstabs_proc *proc;
124 struct fmt_spec weight_format; /* Format for weight variable. */
125 double missing; /* Weight of missing cases. */
127 /* Variables (2 or more). */
129 struct xtab_var *vars;
131 /* Constants (0 or more). */
133 struct xtab_var *const_vars;
134 size_t *const_indexes;
138 struct freq **entries;
141 /* Number of statistically interesting columns/rows
142 (columns/rows with data in them). */
143 int ns_cols, ns_rows;
145 /* Matrix contents. */
146 double *mat; /* Matrix proper. */
147 double *row_tot; /* Row totals. */
148 double *col_tot; /* Column totals. */
149 double total; /* Grand total. */
152 /* Integer mode variable info. */
155 struct hmap_node hmap_node; /* In struct crosstabs_proc var_ranges map. */
156 const struct variable *var; /* The variable. */
157 int min; /* Minimum value. */
158 int max; /* Maximum value + 1. */
159 int count; /* max - min. */
162 struct crosstabs_proc
164 const struct dictionary *dict;
165 enum { INTEGER, GENERAL } mode;
166 enum mv_class exclude;
170 struct fmt_spec weight_format;
172 /* Variables specifies on VARIABLES. */
173 const struct variable **variables;
175 struct hmap var_ranges;
178 struct crosstabulation *pivots;
182 int n_cells; /* Number of cells requested. */
183 unsigned int cells; /* Bit k is 1 if cell k is requested. */
184 int a_cells[CRS_CL_count]; /* 0...n_cells-1 are the requested cells. */
186 /* Rounding of cells. */
187 bool round_case_weights; /* Round case weights? */
188 bool round_cells; /* If !round_case_weights, round cells? */
189 bool round_down; /* Round down? (otherwise to nearest) */
192 unsigned int statistics; /* Bit k is 1 if statistic k is requested. */
194 bool descending; /* True if descending sort order is requested. */
197 const struct var_range *get_var_range (const struct crosstabs_proc *,
198 const struct variable *);
200 static bool should_tabulate_case (const struct crosstabulation *,
201 const struct ccase *, enum mv_class exclude);
202 static void tabulate_general_case (struct crosstabulation *, const struct ccase *,
204 static void tabulate_integer_case (struct crosstabulation *, const struct ccase *,
206 static void postcalc (struct crosstabs_proc *);
209 round_weight (const struct crosstabs_proc *proc, double weight)
211 return proc->round_down ? floor (weight) : floor (weight + 0.5);
214 #define FOR_EACH_POPULATED_COLUMN(C, XT) \
215 for (int C = next_populated_column (0, XT); \
216 C < (XT)->vars[COL_VAR].n_values; \
217 C = next_populated_column (C + 1, XT))
219 next_populated_column (int c, const struct crosstabulation *xt)
221 int n_columns = xt->vars[COL_VAR].n_values;
222 for (; c < n_columns; c++)
228 #define FOR_EACH_POPULATED_ROW(R, XT) \
229 for (int R = next_populated_row (0, XT); R < (XT)->vars[ROW_VAR].n_values; \
230 R = next_populated_row (R + 1, XT))
232 next_populated_row (int r, const struct crosstabulation *xt)
234 int n_rows = xt->vars[ROW_VAR].n_values;
235 for (; r < n_rows; r++)
241 /* Parses and executes the CROSSTABS procedure. */
243 cmd_crosstabs (struct lexer *lexer, struct dataset *ds)
245 struct var_range *range, *next_range;
246 struct crosstabs_proc proc;
247 struct casegrouper *grouper;
248 struct casereader *input, *group;
249 struct cmd_crosstabs cmd;
250 struct crosstabulation *xt;
255 proc.dict = dataset_dict (ds);
256 proc.bad_warn = true;
257 proc.variables = NULL;
258 proc.n_variables = 0;
259 hmap_init (&proc.var_ranges);
262 proc.descending = false;
263 proc.weight_format = *dict_get_weight_format (dataset_dict (ds));
265 if (!parse_crosstabs (lexer, ds, &cmd, &proc))
267 result = CMD_FAILURE;
271 proc.mode = proc.n_variables ? INTEGER : GENERAL;
272 proc.barchart = cmd.sbc_barchart > 0;
274 proc.descending = cmd.val == CRS_DVALUE;
276 proc.round_case_weights = cmd.sbc_count && cmd.roundwhat == CRS_CASE;
277 proc.round_cells = cmd.sbc_count && cmd.roundwhat == CRS_CELL;
278 proc.round_down = cmd.roundhow == CRS_TRUNCATE;
282 proc.cells = 1u << CRS_CL_COUNT;
283 else if (cmd.a_cells[CRS_CL_ALL])
284 proc.cells = UINT_MAX;
288 for (i = 0; i < CRS_CL_count; i++)
290 proc.cells |= 1u << i;
292 proc.cells = ((1u << CRS_CL_COUNT)
294 | (1u << CRS_CL_COLUMN)
295 | (1u << CRS_CL_TOTAL));
297 proc.cells &= ((1u << CRS_CL_count) - 1);
298 proc.cells &= ~((1u << CRS_CL_NONE) | (1u << CRS_CL_ALL));
300 for (i = 0; i < CRS_CL_count; i++)
301 if (proc.cells & (1u << i))
302 proc.a_cells[proc.n_cells++] = i;
305 if (cmd.a_statistics[CRS_ST_ALL])
306 proc.statistics = UINT_MAX;
307 else if (cmd.sbc_statistics)
312 for (i = 0; i < CRS_ST_count; i++)
313 if (cmd.a_statistics[i])
314 proc.statistics |= 1u << i;
315 if (proc.statistics == 0)
316 proc.statistics |= 1u << CRS_ST_CHISQ;
322 proc.exclude = (cmd.miss == CRS_TABLE ? MV_ANY
323 : cmd.miss == CRS_INCLUDE ? MV_SYSTEM
325 if (proc.mode == GENERAL && proc.exclude == MV_NEVER)
327 msg (SE, _("Missing mode %s not allowed in general mode. "
328 "Assuming %s."), "REPORT", "MISSING=TABLE");
329 proc.exclude = MV_ANY;
333 proc.pivot = cmd.pivot == CRS_PIVOT;
335 input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
337 grouper = casegrouper_create_splits (input, dataset_dict (ds));
338 while (casegrouper_get_next_group (grouper, &group))
342 /* Output SPLIT FILE variables. */
343 c = casereader_peek (group, 0);
346 output_split_file_values (ds, c);
350 /* Initialize hash tables. */
351 for (xt = &proc.pivots[0]; xt < &proc.pivots[proc.n_pivots]; xt++)
352 hmap_init (&xt->data);
355 for (; (c = casereader_read (group)) != NULL; case_unref (c))
356 for (xt = &proc.pivots[0]; xt < &proc.pivots[proc.n_pivots]; xt++)
358 double weight = dict_get_case_weight (dataset_dict (ds), c,
360 if (cmd.roundwhat == CRS_CASE)
362 weight = round_weight (&proc, weight);
366 if (should_tabulate_case (xt, c, proc.exclude))
368 if (proc.mode == GENERAL)
369 tabulate_general_case (xt, c, weight);
371 tabulate_integer_case (xt, c, weight);
374 xt->missing += weight;
376 casereader_destroy (group);
381 ok = casegrouper_destroy (grouper);
382 ok = proc_commit (ds) && ok;
384 result = ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
387 free (proc.variables);
388 HMAP_FOR_EACH_SAFE (range, next_range, struct var_range, hmap_node,
391 hmap_delete (&proc.var_ranges, &range->hmap_node);
394 for (xt = &proc.pivots[0]; xt < &proc.pivots[proc.n_pivots]; xt++)
397 free (xt->const_vars);
398 free (xt->const_indexes);
405 /* Parses the TABLES subcommand. */
407 crs_custom_tables (struct lexer *lexer, struct dataset *ds,
408 struct cmd_crosstabs *cmd UNUSED, void *proc_)
410 struct crosstabs_proc *proc = proc_;
411 struct const_var_set *var_set;
413 const struct variable ***by = NULL;
415 size_t *by_nvar = NULL;
420 /* Ensure that this is a TABLES subcommand. */
421 if (!lex_match_id (lexer, "TABLES")
422 && (lex_token (lexer) != T_ID ||
423 dict_lookup_var (dataset_dict (ds), lex_tokcstr (lexer)) == NULL)
424 && lex_token (lexer) != T_ALL)
426 lex_match (lexer, T_EQUALS);
428 if (proc->variables != NULL)
429 var_set = const_var_set_create_from_array (proc->variables,
432 var_set = const_var_set_create_from_dict (dataset_dict (ds));
433 assert (var_set != NULL);
437 by = xnrealloc (by, n_by + 1, sizeof *by);
438 by_nvar = xnrealloc (by_nvar, n_by + 1, sizeof *by_nvar);
439 if (!parse_const_var_set_vars (lexer, var_set, &by[n_by], &by_nvar[n_by],
440 PV_NO_DUPLICATE | PV_NO_SCRATCH))
442 if (xalloc_oversized (nx, by_nvar[n_by]))
444 msg (SE, _("Too many cross-tabulation variables or dimensions."));
450 if (!lex_match (lexer, T_BY))
459 by_iter = xcalloc (n_by, sizeof *by_iter);
460 proc->pivots = xnrealloc (proc->pivots,
461 proc->n_pivots + nx, sizeof *proc->pivots);
462 for (i = 0; i < nx; i++)
464 struct crosstabulation *xt = &proc->pivots[proc->n_pivots++];
468 xt->weight_format = proc->weight_format;
471 xt->vars = xcalloc (n_by, sizeof *xt->vars);
473 xt->const_vars = NULL;
474 xt->const_indexes = NULL;
476 for (j = 0; j < n_by; j++)
477 xt->vars[j].var = by[j][by_iter[j]];
479 for (j = n_by - 1; j >= 0; j--)
481 if (++by_iter[j] < by_nvar[j])
490 /* All return paths lead here. */
491 for (i = 0; i < n_by; i++)
496 const_var_set_destroy (var_set);
501 /* Parses the VARIABLES subcommand. */
503 crs_custom_variables (struct lexer *lexer, struct dataset *ds,
504 struct cmd_crosstabs *cmd UNUSED, void *proc_)
506 struct crosstabs_proc *proc = proc_;
509 msg (SE, _("%s must be specified before %s."), "VARIABLES", "TABLES");
513 lex_match (lexer, T_EQUALS);
517 size_t orig_nv = proc->n_variables;
522 if (!parse_variables_const (lexer, dataset_dict (ds),
523 &proc->variables, &proc->n_variables,
524 (PV_APPEND | PV_NUMERIC
525 | PV_NO_DUPLICATE | PV_NO_SCRATCH)))
528 if (!lex_force_match (lexer, T_LPAREN))
531 if (!lex_force_int (lexer))
533 min = lex_integer (lexer);
536 lex_match (lexer, T_COMMA);
538 if (!lex_force_int (lexer))
540 max = lex_integer (lexer);
543 msg (SE, _("Maximum value (%ld) less than minimum value (%ld)."),
549 if (!lex_force_match (lexer, T_RPAREN))
552 for (i = orig_nv; i < proc->n_variables; i++)
554 const struct variable *var = proc->variables[i];
555 struct var_range *vr = xmalloc (sizeof *vr);
560 vr->count = max - min + 1;
561 hmap_insert (&proc->var_ranges, &vr->hmap_node,
562 hash_pointer (var, 0));
565 if (lex_token (lexer) == T_SLASH)
572 free (proc->variables);
573 proc->variables = NULL;
574 proc->n_variables = 0;
578 /* Data file processing. */
580 const struct var_range *
581 get_var_range (const struct crosstabs_proc *proc, const struct variable *var)
583 if (!hmap_is_empty (&proc->var_ranges))
585 const struct var_range *range;
587 HMAP_FOR_EACH_IN_BUCKET (range, struct var_range, hmap_node,
588 hash_pointer (var, 0), &proc->var_ranges)
589 if (range->var == var)
597 should_tabulate_case (const struct crosstabulation *xt, const struct ccase *c,
598 enum mv_class exclude)
601 for (j = 0; j < xt->n_vars; j++)
603 const struct variable *var = xt->vars[j].var;
604 const struct var_range *range = get_var_range (xt->proc, var);
606 if (var_is_value_missing (var, case_data (c, var), exclude))
611 double num = case_num (c, var);
612 if (num < range->min || num >= range->max + 1.)
620 tabulate_integer_case (struct crosstabulation *xt, const struct ccase *c,
628 for (j = 0; j < xt->n_vars; j++)
630 /* Throw away fractional parts of values. */
631 hash = hash_int (case_num (c, xt->vars[j].var), hash);
634 HMAP_FOR_EACH_WITH_HASH (te, struct freq, node, hash, &xt->data)
636 for (j = 0; j < xt->n_vars; j++)
637 if ((int) case_num (c, xt->vars[j].var) != (int) te->values[j].f)
640 /* Found an existing entry. */
647 /* No existing entry. Create a new one. */
648 te = xmalloc (table_entry_size (xt->n_vars));
650 for (j = 0; j < xt->n_vars; j++)
651 te->values[j].f = (int) case_num (c, xt->vars[j].var);
652 hmap_insert (&xt->data, &te->node, hash);
656 tabulate_general_case (struct crosstabulation *xt, const struct ccase *c,
664 for (j = 0; j < xt->n_vars; j++)
666 const struct variable *var = xt->vars[j].var;
667 hash = value_hash (case_data (c, var), var_get_width (var), hash);
670 HMAP_FOR_EACH_WITH_HASH (te, struct freq, node, hash, &xt->data)
672 for (j = 0; j < xt->n_vars; j++)
674 const struct variable *var = xt->vars[j].var;
675 if (!value_equal (case_data (c, var), &te->values[j],
676 var_get_width (var)))
680 /* Found an existing entry. */
687 /* No existing entry. Create a new one. */
688 te = xmalloc (table_entry_size (xt->n_vars));
690 for (j = 0; j < xt->n_vars; j++)
692 const struct variable *var = xt->vars[j].var;
693 value_clone (&te->values[j], case_data (c, var), var_get_width (var));
695 hmap_insert (&xt->data, &te->node, hash);
698 /* Post-data reading calculations. */
700 static int compare_table_entry_vars_3way (const struct freq *a,
701 const struct freq *b,
702 const struct crosstabulation *xt,
704 static int compare_table_entry_3way (const void *ap_, const void *bp_,
706 static int compare_table_entry_3way_inv (const void *ap_, const void *bp_,
709 static void enum_var_values (const struct crosstabulation *, int var_idx,
711 static void free_var_values (const struct crosstabulation *, int var_idx);
712 static void output_crosstabulation (struct crosstabs_proc *,
713 struct crosstabulation *);
714 static void make_crosstabulation_subset (struct crosstabulation *xt,
715 size_t row0, size_t row1,
716 struct crosstabulation *subset);
717 static void make_summary_table (struct crosstabs_proc *);
718 static bool find_crosstab (struct crosstabulation *, size_t *row0p,
722 postcalc (struct crosstabs_proc *proc)
725 /* Round hash table entries, if requested
727 If this causes any of the cell counts to fall to zero, delete those
729 if (proc->round_cells)
730 for (struct crosstabulation *xt = proc->pivots;
731 xt < &proc->pivots[proc->n_pivots]; xt++)
733 struct freq *e, *next;
734 HMAP_FOR_EACH_SAFE (e, next, struct freq, node, &xt->data)
736 e->count = round_weight (proc, e->count);
739 hmap_delete (&xt->data, &e->node);
745 /* Convert hash tables into sorted arrays of entries. */
746 for (struct crosstabulation *xt = proc->pivots;
747 xt < &proc->pivots[proc->n_pivots]; xt++)
751 xt->n_entries = hmap_count (&xt->data);
752 xt->entries = xnmalloc (xt->n_entries, sizeof *xt->entries);
754 HMAP_FOR_EACH (e, struct freq, node, &xt->data)
755 xt->entries[i++] = e;
756 hmap_destroy (&xt->data);
758 sort (xt->entries, xt->n_entries, sizeof *xt->entries,
759 proc->descending ? compare_table_entry_3way_inv : compare_table_entry_3way,
764 make_summary_table (proc);
766 /* Output each pivot table. */
767 for (struct crosstabulation *xt = proc->pivots;
768 xt < &proc->pivots[proc->n_pivots]; xt++)
770 if (proc->pivot || xt->n_vars == 2)
771 output_crosstabulation (proc, xt);
774 size_t row0 = 0, row1 = 0;
775 while (find_crosstab (xt, &row0, &row1))
777 struct crosstabulation subset;
778 make_crosstabulation_subset (xt, row0, row1, &subset);
779 output_crosstabulation (proc, &subset);
780 free (subset.const_indexes);
785 int n_vars = (xt->n_vars > 2 ? 2 : xt->n_vars);
786 const struct variable **vars = xcalloc (n_vars, sizeof *vars);
787 for (size_t i = 0; i < n_vars; i++)
788 vars[i] = xt->vars[i].var;
789 chart_item_submit (barchart_create (vars, n_vars, _("Count"),
791 xt->entries, xt->n_entries));
796 /* Free output and prepare for next split file. */
797 for (struct crosstabulation *xt = proc->pivots;
798 xt < &proc->pivots[proc->n_pivots]; xt++)
802 /* Free the members that were allocated in this function(and the values
803 owned by the entries.
805 The other pointer members are either both allocated and destroyed at a
806 lower level (in output_crosstabulation), or both allocated and
807 destroyed at a higher level (in crs_custom_tables and free_proc,
809 for (size_t i = 0; i < xt->n_vars; i++)
811 int width = var_get_width (xt->vars[i].var);
812 if (value_needs_init (width))
816 for (j = 0; j < xt->n_entries; j++)
817 value_destroy (&xt->entries[j]->values[i], width);
821 for (size_t i = 0; i < xt->n_entries; i++)
822 free (xt->entries[i]);
828 make_crosstabulation_subset (struct crosstabulation *xt, size_t row0,
829 size_t row1, struct crosstabulation *subset)
834 assert (xt->n_consts == 0);
836 subset->vars = xt->vars;
838 subset->n_consts = xt->n_vars - 2;
839 subset->const_vars = xt->vars + 2;
840 subset->const_indexes = xcalloc (subset->n_consts,
841 sizeof *subset->const_indexes);
842 for (size_t i = 0; i < subset->n_consts; i++)
844 const union value *value = &xt->entries[row0]->values[2 + i];
846 for (size_t j = 0; j < xt->vars[2 + i].n_values; j++)
847 if (value_equal (&xt->vars[2 + i].values[j], value,
848 var_get_width (xt->vars[2 + i].var)))
850 subset->const_indexes[i] = j;
857 subset->entries = &xt->entries[row0];
858 subset->n_entries = row1 - row0;
862 compare_table_entry_var_3way (const struct freq *a,
863 const struct freq *b,
864 const struct crosstabulation *xt,
867 return value_compare_3way (&a->values[idx], &b->values[idx],
868 var_get_width (xt->vars[idx].var));
872 compare_table_entry_vars_3way (const struct freq *a,
873 const struct freq *b,
874 const struct crosstabulation *xt,
879 for (i = idx1 - 1; i >= idx0; i--)
881 int cmp = compare_table_entry_var_3way (a, b, xt, i);
888 /* Compare the struct freq at *AP to the one at *BP and
889 return a strcmp()-type result. */
891 compare_table_entry_3way (const void *ap_, const void *bp_, const void *xt_)
893 const struct freq *const *ap = ap_;
894 const struct freq *const *bp = bp_;
895 const struct freq *a = *ap;
896 const struct freq *b = *bp;
897 const struct crosstabulation *xt = xt_;
900 cmp = compare_table_entry_vars_3way (a, b, xt, 2, xt->n_vars);
904 cmp = compare_table_entry_var_3way (a, b, xt, ROW_VAR);
908 return compare_table_entry_var_3way (a, b, xt, COL_VAR);
911 /* Inverted version of compare_table_entry_3way */
913 compare_table_entry_3way_inv (const void *ap_, const void *bp_, const void *xt_)
915 return -compare_table_entry_3way (ap_, bp_, xt_);
918 /* Output a table summarizing the cases processed. */
920 make_summary_table (struct crosstabs_proc *proc)
922 struct pivot_table *table = pivot_table_create (N_("Summary"));
923 pivot_table_set_weight_var (table, dict_get_weight (proc->dict));
925 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
926 N_("N"), PIVOT_RC_COUNT,
927 N_("Percent"), PIVOT_RC_PERCENT);
929 struct pivot_dimension *cases = pivot_dimension_create (
930 table, PIVOT_AXIS_COLUMN, N_("Cases"),
931 N_("Valid"), N_("Missing"), N_("Total"));
932 cases->root->show_label = true;
934 struct pivot_dimension *tables = pivot_dimension_create (
935 table, PIVOT_AXIS_ROW, N_("Crosstabulation"));
936 for (struct crosstabulation *xt = &proc->pivots[0];
937 xt < &proc->pivots[proc->n_pivots]; xt++)
939 struct string name = DS_EMPTY_INITIALIZER;
940 for (size_t i = 0; i < xt->n_vars; i++)
943 ds_put_cstr (&name, " × ");
944 ds_put_cstr (&name, var_to_string (xt->vars[i].var));
947 int row = pivot_category_create_leaf (
949 pivot_value_new_user_text_nocopy (ds_steal_cstr (&name)));
952 for (size_t i = 0; i < xt->n_entries; i++)
953 valid += xt->entries[i]->count;
959 for (int i = 0; i < 3; i++)
961 pivot_table_put3 (table, 0, i, row, pivot_value_new_number (n[i]));
962 pivot_table_put3 (table, 1, i, row,
963 pivot_value_new_number (n[i] / n[2] * 100.0));
967 pivot_table_submit (table);
972 static struct pivot_table *create_crosstab_table (
973 struct crosstabs_proc *, struct crosstabulation *,
974 size_t crs_leaves[CRS_CL_count]);
975 static struct pivot_table *create_chisq_table (struct crosstabulation *);
976 static struct pivot_table *create_sym_table (struct crosstabulation *);
977 static struct pivot_table *create_risk_table (
978 struct crosstabulation *, struct pivot_dimension **risk_statistics);
979 static struct pivot_table *create_direct_table (struct crosstabulation *);
980 static void display_crosstabulation (struct crosstabs_proc *,
981 struct crosstabulation *,
982 struct pivot_table *,
983 size_t crs_leaves[CRS_CL_count]);
984 static void display_chisq (struct crosstabulation *, struct pivot_table *);
985 static void display_symmetric (struct crosstabs_proc *,
986 struct crosstabulation *, struct pivot_table *);
987 static void display_risk (struct crosstabulation *, struct pivot_table *,
988 struct pivot_dimension *risk_statistics);
989 static void display_directional (struct crosstabs_proc *,
990 struct crosstabulation *,
991 struct pivot_table *);
992 static void delete_missing (struct crosstabulation *);
993 static void build_matrix (struct crosstabulation *);
995 /* Output pivot table XT in the context of PROC. */
997 output_crosstabulation (struct crosstabs_proc *proc, struct crosstabulation *xt)
999 for (size_t i = 0; i < xt->n_vars; i++)
1000 enum_var_values (xt, i, proc->descending);
1002 if (xt->vars[COL_VAR].n_values == 0)
1007 ds_init_cstr (&vars, var_to_string (xt->vars[0].var));
1008 for (i = 1; i < xt->n_vars; i++)
1009 ds_put_format (&vars, " × %s", var_to_string (xt->vars[i].var));
1011 /* TRANSLATORS: The %s here describes a crosstabulation. It takes the
1012 form "var1 * var2 * var3 * ...". */
1013 msg (SW, _("Crosstabulation %s contained no non-missing cases."),
1017 for (size_t i = 0; i < xt->n_vars; i++)
1018 free_var_values (xt, i);
1022 size_t crs_leaves[CRS_CL_count];
1023 struct pivot_table *table = (proc->cells
1024 ? create_crosstab_table (proc, xt, crs_leaves)
1026 struct pivot_table *chisq = (proc->statistics & (1u << CRS_ST_CHISQ)
1027 ? create_chisq_table (xt)
1029 struct pivot_table *sym
1030 = (proc->statistics & ((1u << CRS_ST_PHI) | (1u << CRS_ST_CC)
1031 | (1u << CRS_ST_BTAU) | (1u << CRS_ST_CTAU)
1032 | (1u << CRS_ST_GAMMA) | (1u << CRS_ST_CORR)
1033 | (1u << CRS_ST_KAPPA))
1034 ? create_sym_table (xt)
1036 struct pivot_dimension *risk_statistics = NULL;
1037 struct pivot_table *risk = (proc->statistics & (1u << CRS_ST_RISK)
1038 ? create_risk_table (xt, &risk_statistics)
1040 struct pivot_table *direct
1041 = (proc->statistics & ((1u << CRS_ST_LAMBDA) | (1u << CRS_ST_UC)
1042 | (1u << CRS_ST_D) | (1u << CRS_ST_ETA))
1043 ? create_direct_table (xt)
1048 while (find_crosstab (xt, &row0, &row1))
1050 struct crosstabulation x;
1052 make_crosstabulation_subset (xt, row0, row1, &x);
1054 size_t n_rows = x.vars[ROW_VAR].n_values;
1055 size_t n_cols = x.vars[COL_VAR].n_values;
1056 if (size_overflow_p (xtimes (xtimes (n_rows, n_cols), sizeof (double))))
1058 x.row_tot = xmalloc (n_rows * sizeof *x.row_tot);
1059 x.col_tot = xmalloc (n_cols * sizeof *x.col_tot);
1060 x.mat = xmalloc (n_rows * n_cols * sizeof *x.mat);
1064 /* Find the first variable that differs from the last subtable. */
1066 display_crosstabulation (proc, &x, table, crs_leaves);
1068 if (proc->exclude == MV_NEVER)
1069 delete_missing (&x);
1072 display_chisq (&x, chisq);
1075 display_symmetric (proc, &x, sym);
1077 display_risk (&x, risk, risk_statistics);
1079 display_directional (proc, &x, direct);
1084 free (x.const_indexes);
1088 pivot_table_submit (table);
1091 pivot_table_submit (chisq);
1094 pivot_table_submit (sym);
1098 if (!pivot_table_is_empty (risk))
1099 pivot_table_submit (risk);
1101 pivot_table_unref (risk);
1105 pivot_table_submit (direct);
1107 for (size_t i = 0; i < xt->n_vars; i++)
1108 free_var_values (xt, i);
1112 build_matrix (struct crosstabulation *x)
1114 const int col_var_width = var_get_width (x->vars[COL_VAR].var);
1115 const int row_var_width = var_get_width (x->vars[ROW_VAR].var);
1116 size_t n_rows = x->vars[ROW_VAR].n_values;
1117 size_t n_cols = x->vars[COL_VAR].n_values;
1124 for (p = x->entries; p < &x->entries[x->n_entries]; p++)
1126 const struct freq *te = *p;
1128 while (!value_equal (&x->vars[ROW_VAR].values[row],
1129 &te->values[ROW_VAR], row_var_width))
1131 for (; col < n_cols; col++)
1137 while (!value_equal (&x->vars[COL_VAR].values[col],
1138 &te->values[COL_VAR], col_var_width))
1145 if (++col >= n_cols)
1151 while (mp < &x->mat[n_cols * n_rows])
1153 assert (mp == &x->mat[n_cols * n_rows]);
1155 /* Column totals, row totals, ns_rows. */
1157 for (col = 0; col < n_cols; col++)
1158 x->col_tot[col] = 0.0;
1159 for (row = 0; row < n_rows; row++)
1160 x->row_tot[row] = 0.0;
1162 for (row = 0; row < n_rows; row++)
1164 bool row_is_empty = true;
1165 for (col = 0; col < n_cols; col++)
1169 row_is_empty = false;
1170 x->col_tot[col] += *mp;
1171 x->row_tot[row] += *mp;
1178 assert (mp == &x->mat[n_cols * n_rows]);
1182 for (col = 0; col < n_cols; col++)
1183 for (row = 0; row < n_rows; row++)
1184 if (x->mat[col + row * n_cols] != 0.0)
1192 for (col = 0; col < n_cols; col++)
1193 x->total += x->col_tot[col];
1197 add_var_dimension (struct pivot_table *table, const struct xtab_var *var,
1198 enum pivot_axis_type axis_type, bool total)
1200 struct pivot_dimension *d = pivot_dimension_create__ (
1201 table, axis_type, pivot_value_new_variable (var->var));
1203 struct pivot_footnote *missing_footnote = pivot_table_create_footnote (
1204 table, pivot_value_new_text (N_("Missing value")));
1206 struct pivot_category *group = pivot_category_create_group__ (
1207 d->root, pivot_value_new_variable (var->var));
1208 for (size_t j = 0; j < var->n_values; j++)
1210 struct pivot_value *value = pivot_value_new_var_value (
1211 var->var, &var->values[j]);
1212 if (var_is_value_missing (var->var, &var->values[j], MV_ANY))
1213 pivot_value_add_footnote (value, missing_footnote);
1214 pivot_category_create_leaf (group, value);
1218 pivot_category_create_leaf (d->root, pivot_value_new_text (N_("Total")));
1221 static struct pivot_table *
1222 create_crosstab_table (struct crosstabs_proc *proc, struct crosstabulation *xt,
1223 size_t crs_leaves[CRS_CL_count])
1226 struct string title = DS_EMPTY_INITIALIZER;
1227 for (size_t i = 0; i < xt->n_vars; i++)
1230 ds_put_cstr (&title, " × ");
1231 ds_put_cstr (&title, var_to_string (xt->vars[i].var));
1233 for (size_t i = 0; i < xt->n_consts; i++)
1235 const struct variable *var = xt->const_vars[i].var;
1236 const union value *value = &xt->entries[0]->values[2 + i];
1239 ds_put_format (&title, ", %s=", var_to_string (var));
1241 /* Insert the formatted value of VAR without any leading spaces. */
1242 s = data_out (value, var_get_encoding (var), var_get_print_format (var),
1243 settings_get_fmt_settings ());
1244 ds_put_cstr (&title, s + strspn (s, " "));
1247 struct pivot_table *table = pivot_table_create__ (
1248 pivot_value_new_user_text_nocopy (ds_steal_cstr (&title)),
1250 pivot_table_set_weight_format (table, &proc->weight_format);
1252 struct pivot_dimension *statistics = pivot_dimension_create (
1253 table, PIVOT_AXIS_ROW, N_("Statistics"));
1260 static const struct statistic stats[CRS_CL_count] =
1262 [CRS_CL_COUNT] = { N_("Count"), PIVOT_RC_COUNT },
1263 [CRS_CL_ROW] = { N_("Row %"), PIVOT_RC_PERCENT },
1264 [CRS_CL_COLUMN] = { N_("Column %"), PIVOT_RC_PERCENT },
1265 [CRS_CL_TOTAL] = { N_("Total %"), PIVOT_RC_PERCENT },
1266 [CRS_CL_EXPECTED] = { N_("Expected"), PIVOT_RC_OTHER },
1267 [CRS_CL_RESIDUAL] = { N_("Residual"), PIVOT_RC_RESIDUAL },
1268 [CRS_CL_SRESIDUAL] = { N_("Std. Residual"), PIVOT_RC_RESIDUAL },
1269 [CRS_CL_ASRESIDUAL] = { N_("Adjusted Residual"), PIVOT_RC_RESIDUAL },
1271 for (size_t i = 0; i < CRS_CL_count; i++)
1272 if (proc->cells & (1u << i) && stats[i].label)
1273 crs_leaves[i] = pivot_category_create_leaf_rc (
1274 statistics->root, pivot_value_new_text (stats[i].label),
1277 for (size_t i = 0; i < xt->n_vars; i++)
1278 add_var_dimension (table, &xt->vars[i],
1279 i == COL_VAR ? PIVOT_AXIS_COLUMN : PIVOT_AXIS_ROW,
1285 static struct pivot_table *
1286 create_chisq_table (struct crosstabulation *xt)
1288 struct pivot_table *chisq = pivot_table_create (N_("Chi-Square Tests"));
1289 pivot_table_set_weight_format (chisq, &xt->weight_format);
1291 pivot_dimension_create (
1292 chisq, PIVOT_AXIS_ROW, N_("Statistics"),
1293 N_("Pearson Chi-Square"),
1294 N_("Likelihood Ratio"),
1295 N_("Fisher's Exact Test"),
1296 N_("Continuity Correction"),
1297 N_("Linear-by-Linear Association"),
1298 N_("N of Valid Cases"), PIVOT_RC_COUNT);
1300 pivot_dimension_create (
1301 chisq, PIVOT_AXIS_COLUMN, N_("Statistics"),
1302 N_("Value"), PIVOT_RC_OTHER,
1303 N_("df"), PIVOT_RC_COUNT,
1304 N_("Asymptotic Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE,
1305 N_("Exact Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE,
1306 N_("Exact Sig. (1-tailed)"), PIVOT_RC_SIGNIFICANCE);
1308 for (size_t i = 2; i < xt->n_vars; i++)
1309 add_var_dimension (chisq, &xt->vars[i], PIVOT_AXIS_ROW, false);
1314 /* Symmetric measures. */
1315 static struct pivot_table *
1316 create_sym_table (struct crosstabulation *xt)
1318 struct pivot_table *sym = pivot_table_create (N_("Symmetric Measures"));
1319 pivot_table_set_weight_format (sym, &xt->weight_format);
1321 pivot_dimension_create (
1322 sym, PIVOT_AXIS_COLUMN, N_("Values"),
1323 N_("Value"), PIVOT_RC_OTHER,
1324 N_("Asymp. Std. Error"), PIVOT_RC_OTHER,
1325 N_("Approx. T"), PIVOT_RC_OTHER,
1326 N_("Approx. Sig."), PIVOT_RC_SIGNIFICANCE);
1328 struct pivot_dimension *statistics = pivot_dimension_create (
1329 sym, PIVOT_AXIS_ROW, N_("Statistics"));
1330 pivot_category_create_group (
1331 statistics->root, N_("Nominal by Nominal"),
1332 N_("Phi"), N_("Cramer's V"), N_("Contingency Coefficient"));
1333 pivot_category_create_group (
1334 statistics->root, N_("Ordinal by Ordinal"),
1335 N_("Kendall's tau-b"), N_("Kendall's tau-c"),
1336 N_("Gamma"), N_("Spearman Correlation"));
1337 pivot_category_create_group (
1338 statistics->root, N_("Interval by Interval"),
1340 pivot_category_create_group (
1341 statistics->root, N_("Measure of Agreement"),
1343 pivot_category_create_leaves (statistics->root, N_("N of Valid Cases"),
1346 for (size_t i = 2; i < xt->n_vars; i++)
1347 add_var_dimension (sym, &xt->vars[i], PIVOT_AXIS_ROW, false);
1352 /* Risk estimate. */
1353 static struct pivot_table *
1354 create_risk_table (struct crosstabulation *xt,
1355 struct pivot_dimension **risk_statistics)
1357 struct pivot_table *risk = pivot_table_create (N_("Risk Estimate"));
1358 pivot_table_set_weight_format (risk, &xt->weight_format);
1360 struct pivot_dimension *values = pivot_dimension_create (
1361 risk, PIVOT_AXIS_COLUMN, N_("Values"),
1362 N_("Value"), PIVOT_RC_OTHER);
1363 pivot_category_create_group (
1364 /* xgettext:no-c-format */
1365 values->root, N_("95% Confidence Interval"),
1366 N_("Lower"), PIVOT_RC_OTHER,
1367 N_("Upper"), PIVOT_RC_OTHER);
1369 *risk_statistics = pivot_dimension_create (
1370 risk, PIVOT_AXIS_ROW, N_("Statistics"));
1372 for (size_t i = 2; i < xt->n_vars; i++)
1373 add_var_dimension (risk, &xt->vars[i], PIVOT_AXIS_ROW, false);
1379 create_direct_stat (struct pivot_category *parent,
1380 const struct crosstabulation *xt,
1381 const char *name, bool symmetric)
1383 struct pivot_category *group = pivot_category_create_group (
1386 pivot_category_create_leaf (group, pivot_value_new_text (N_("Symmetric")));
1388 char *row_label = xasprintf (_("%s Dependent"),
1389 var_to_string (xt->vars[ROW_VAR].var));
1390 pivot_category_create_leaf (group, pivot_value_new_user_text_nocopy (
1393 char *col_label = xasprintf (_("%s Dependent"),
1394 var_to_string (xt->vars[COL_VAR].var));
1395 pivot_category_create_leaf (group, pivot_value_new_user_text_nocopy (
1399 /* Directional measures. */
1400 static struct pivot_table *
1401 create_direct_table (struct crosstabulation *xt)
1403 struct pivot_table *direct = pivot_table_create (N_("Directional Measures"));
1404 pivot_table_set_weight_format (direct, &xt->weight_format);
1406 pivot_dimension_create (
1407 direct, PIVOT_AXIS_COLUMN, N_("Values"),
1408 N_("Value"), PIVOT_RC_OTHER,
1409 N_("Asymp. Std. Error"), PIVOT_RC_OTHER,
1410 N_("Approx. T"), PIVOT_RC_OTHER,
1411 N_("Approx. Sig."), PIVOT_RC_SIGNIFICANCE);
1413 struct pivot_dimension *statistics = pivot_dimension_create (
1414 direct, PIVOT_AXIS_ROW, N_("Statistics"));
1415 struct pivot_category *nn = pivot_category_create_group (
1416 statistics->root, N_("Nominal by Nominal"));
1417 create_direct_stat (nn, xt, N_("Lambda"), true);
1418 create_direct_stat (nn, xt, N_("Goodman and Kruskal tau"), false);
1419 create_direct_stat (nn, xt, N_("Uncertainty Coefficient"), true);
1420 struct pivot_category *oo = pivot_category_create_group (
1421 statistics->root, N_("Ordinal by Ordinal"));
1422 create_direct_stat (oo, xt, N_("Somers' d"), true);
1423 struct pivot_category *ni = pivot_category_create_group (
1424 statistics->root, N_("Nominal by Interval"));
1425 create_direct_stat (ni, xt, N_("Eta"), false);
1427 for (size_t i = 2; i < xt->n_vars; i++)
1428 add_var_dimension (direct, &xt->vars[i], PIVOT_AXIS_ROW, false);
1433 /* Delete missing rows and columns for statistical analysis when
1436 delete_missing (struct crosstabulation *xt)
1438 size_t n_rows = xt->vars[ROW_VAR].n_values;
1439 size_t n_cols = xt->vars[COL_VAR].n_values;
1442 for (r = 0; r < n_rows; r++)
1443 if (var_is_num_missing (xt->vars[ROW_VAR].var,
1444 xt->vars[ROW_VAR].values[r].f, MV_USER))
1446 for (c = 0; c < n_cols; c++)
1447 xt->mat[c + r * n_cols] = 0.;
1452 for (c = 0; c < n_cols; c++)
1453 if (var_is_num_missing (xt->vars[COL_VAR].var,
1454 xt->vars[COL_VAR].values[c].f, MV_USER))
1456 for (r = 0; r < n_rows; r++)
1457 xt->mat[c + r * n_cols] = 0.;
1463 find_crosstab (struct crosstabulation *xt, size_t *row0p, size_t *row1p)
1465 size_t row0 = *row1p;
1468 if (row0 >= xt->n_entries)
1471 for (row1 = row0 + 1; row1 < xt->n_entries; row1++)
1473 struct freq *a = xt->entries[row0];
1474 struct freq *b = xt->entries[row1];
1475 if (compare_table_entry_vars_3way (a, b, xt, 2, xt->n_vars) != 0)
1483 /* Compares `union value's A_ and B_ and returns a strcmp()-like
1484 result. WIDTH_ points to an int which is either 0 for a
1485 numeric value or a string width for a string value. */
1487 compare_value_3way (const void *a_, const void *b_, const void *width_)
1489 const union value *a = a_;
1490 const union value *b = b_;
1491 const int *width = width_;
1493 return value_compare_3way (a, b, *width);
1496 /* Inverted version of the above */
1498 compare_value_3way_inv (const void *a_, const void *b_, const void *width_)
1500 return -compare_value_3way (a_, b_, width_);
1504 /* Given an array of ENTRY_CNT table_entry structures starting at
1505 ENTRIES, creates a sorted list of the values that the variable
1506 with index VAR_IDX takes on. Stores the array of the values in
1507 XT->values and the number of values in XT->n_values. */
1509 enum_var_values (const struct crosstabulation *xt, int var_idx,
1512 struct xtab_var *xv = &xt->vars[var_idx];
1513 const struct var_range *range = get_var_range (xt->proc, xv->var);
1517 xv->values = xnmalloc (range->count, sizeof *xv->values);
1518 xv->n_values = range->count;
1519 for (size_t i = 0; i < range->count; i++)
1520 xv->values[i].f = range->min + i;
1524 int width = var_get_width (xv->var);
1525 struct hmapx_node *node;
1526 const union value *iter;
1530 for (size_t i = 0; i < xt->n_entries; i++)
1532 const struct freq *te = xt->entries[i];
1533 const union value *value = &te->values[var_idx];
1534 size_t hash = value_hash (value, width, 0);
1536 HMAPX_FOR_EACH_WITH_HASH (iter, node, hash, &set)
1537 if (value_equal (iter, value, width))
1540 hmapx_insert (&set, (union value *) value, hash);
1545 xv->n_values = hmapx_count (&set);
1546 xv->values = xnmalloc (xv->n_values, sizeof *xv->values);
1548 HMAPX_FOR_EACH (iter, node, &set)
1549 xv->values[i++] = *iter;
1550 hmapx_destroy (&set);
1552 sort (xv->values, xv->n_values, sizeof *xv->values,
1553 descending ? compare_value_3way_inv : compare_value_3way,
1559 free_var_values (const struct crosstabulation *xt, int var_idx)
1561 struct xtab_var *xv = &xt->vars[var_idx];
1567 /* Displays the crosstabulation table. */
1569 display_crosstabulation (struct crosstabs_proc *proc,
1570 struct crosstabulation *xt, struct pivot_table *table,
1571 size_t crs_leaves[CRS_CL_count])
1573 size_t n_rows = xt->vars[ROW_VAR].n_values;
1574 size_t n_cols = xt->vars[COL_VAR].n_values;
1576 size_t *indexes = xnmalloc (table->n_dimensions, sizeof *indexes);
1577 assert (xt->n_vars == 2);
1578 for (size_t i = 0; i < xt->n_consts; i++)
1579 indexes[i + 3] = xt->const_indexes[i];
1581 /* Put in the actual cells. */
1582 double *mp = xt->mat;
1583 for (size_t r = 0; r < n_rows; r++)
1585 if (!xt->row_tot[r] && proc->mode != INTEGER)
1588 indexes[ROW_VAR + 1] = r;
1589 for (size_t c = 0; c < n_cols; c++)
1591 if (!xt->col_tot[c] && proc->mode != INTEGER)
1594 indexes[COL_VAR + 1] = c;
1596 double expected_value = xt->row_tot[r] * xt->col_tot[c] / xt->total;
1597 double residual = *mp - expected_value;
1598 double sresidual = residual / sqrt (expected_value);
1599 double asresidual = (sresidual
1600 * (1. - xt->row_tot[r] / xt->total)
1601 * (1. - xt->col_tot[c] / xt->total));
1602 double entries[] = {
1603 [CRS_CL_COUNT] = *mp,
1604 [CRS_CL_ROW] = *mp / xt->row_tot[r] * 100.,
1605 [CRS_CL_COLUMN] = *mp / xt->col_tot[c] * 100.,
1606 [CRS_CL_TOTAL] = *mp / xt->total * 100.,
1607 [CRS_CL_EXPECTED] = expected_value,
1608 [CRS_CL_RESIDUAL] = residual,
1609 [CRS_CL_SRESIDUAL] = sresidual,
1610 [CRS_CL_ASRESIDUAL] = asresidual,
1612 for (size_t i = 0; i < proc->n_cells; i++)
1614 int cell = proc->a_cells[i];
1615 indexes[0] = crs_leaves[cell];
1616 pivot_table_put (table, indexes, table->n_dimensions,
1617 pivot_value_new_number (entries[cell]));
1625 for (size_t r = 0; r < n_rows; r++)
1627 if (!xt->row_tot[r] && proc->mode != INTEGER)
1630 double expected_value = xt->row_tot[r] / xt->total;
1631 double entries[] = {
1632 [CRS_CL_COUNT] = xt->row_tot[r],
1633 [CRS_CL_ROW] = 100.0,
1634 [CRS_CL_COLUMN] = expected_value * 100.,
1635 [CRS_CL_TOTAL] = expected_value * 100.,
1636 [CRS_CL_EXPECTED] = expected_value,
1637 [CRS_CL_RESIDUAL] = SYSMIS,
1638 [CRS_CL_SRESIDUAL] = SYSMIS,
1639 [CRS_CL_ASRESIDUAL] = SYSMIS,
1641 for (size_t i = 0; i < proc->n_cells; i++)
1643 int cell = proc->a_cells[i];
1644 double entry = entries[cell];
1645 if (entry != SYSMIS)
1647 indexes[ROW_VAR + 1] = r;
1648 indexes[COL_VAR + 1] = n_cols;
1649 indexes[0] = crs_leaves[cell];
1650 pivot_table_put (table, indexes, table->n_dimensions,
1651 pivot_value_new_number (entry));
1656 for (size_t c = 0; c <= n_cols; c++)
1658 if (c < n_cols && !xt->col_tot[c] && proc->mode != INTEGER)
1661 double ct = c < n_cols ? xt->col_tot[c] : xt->total;
1662 double expected_value = ct / xt->total;
1663 double entries[] = {
1664 [CRS_CL_COUNT] = ct,
1665 [CRS_CL_ROW] = expected_value * 100.0,
1666 [CRS_CL_COLUMN] = 100.0,
1667 [CRS_CL_TOTAL] = expected_value * 100.,
1668 [CRS_CL_EXPECTED] = expected_value,
1669 [CRS_CL_RESIDUAL] = SYSMIS,
1670 [CRS_CL_SRESIDUAL] = SYSMIS,
1671 [CRS_CL_ASRESIDUAL] = SYSMIS,
1673 for (size_t i = 0; i < proc->n_cells; i++)
1675 int cell = proc->a_cells[i];
1676 double entry = entries[cell];
1677 if (entry != SYSMIS)
1679 indexes[ROW_VAR + 1] = n_rows;
1680 indexes[COL_VAR + 1] = c;
1681 indexes[0] = crs_leaves[cell];
1682 pivot_table_put (table, indexes, table->n_dimensions,
1683 pivot_value_new_number (entry));
1691 static void calc_r (struct crosstabulation *,
1692 double *XT, double *Y, double *, double *, double *);
1693 static void calc_chisq (struct crosstabulation *,
1694 double[N_CHISQ], int[N_CHISQ], double *, double *);
1696 /* Display chi-square statistics. */
1698 display_chisq (struct crosstabulation *xt, struct pivot_table *chisq)
1700 double chisq_v[N_CHISQ];
1701 double fisher1, fisher2;
1703 calc_chisq (xt, chisq_v, df, &fisher1, &fisher2);
1705 size_t *indexes = xnmalloc (chisq->n_dimensions, sizeof *indexes);
1706 assert (xt->n_vars == 2);
1707 for (size_t i = 0; i < xt->n_consts; i++)
1708 indexes[i + 2] = xt->const_indexes[i];
1709 for (int i = 0; i < N_CHISQ; i++)
1713 double entries[5] = { SYSMIS, SYSMIS, SYSMIS, SYSMIS, SYSMIS };
1716 entries[3] = fisher2;
1717 entries[4] = fisher1;
1719 else if (chisq_v[i] != SYSMIS)
1721 entries[0] = chisq_v[i];
1723 entries[2] = gsl_cdf_chisq_Q (chisq_v[i], df[i]);
1726 for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
1727 if (entries[j] != SYSMIS)
1730 pivot_table_put (chisq, indexes, chisq->n_dimensions,
1731 pivot_value_new_number (entries[j]));
1737 pivot_table_put (chisq, indexes, chisq->n_dimensions,
1738 pivot_value_new_number (xt->total));
1743 static int calc_symmetric (struct crosstabs_proc *, struct crosstabulation *,
1744 double[N_SYMMETRIC], double[N_SYMMETRIC],
1745 double[N_SYMMETRIC],
1746 double[3], double[3], double[3]);
1748 /* Display symmetric measures. */
1750 display_symmetric (struct crosstabs_proc *proc, struct crosstabulation *xt,
1751 struct pivot_table *sym)
1753 double sym_v[N_SYMMETRIC], sym_ase[N_SYMMETRIC], sym_t[N_SYMMETRIC];
1754 double somers_d_v[3], somers_d_ase[3], somers_d_t[3];
1756 if (!calc_symmetric (proc, xt, sym_v, sym_ase, sym_t,
1757 somers_d_v, somers_d_ase, somers_d_t))
1760 size_t *indexes = xnmalloc (sym->n_dimensions, sizeof *indexes);
1761 assert (xt->n_vars == 2);
1762 for (size_t i = 0; i < xt->n_consts; i++)
1763 indexes[i + 2] = xt->const_indexes[i];
1765 for (int i = 0; i < N_SYMMETRIC; i++)
1767 if (sym_v[i] == SYSMIS)
1772 double entries[] = { sym_v[i], sym_ase[i], sym_t[i] };
1773 for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
1774 if (entries[j] != SYSMIS)
1777 pivot_table_put (sym, indexes, sym->n_dimensions,
1778 pivot_value_new_number (entries[j]));
1782 indexes[1] = N_SYMMETRIC;
1784 struct pivot_value *total = pivot_value_new_number (xt->total);
1785 pivot_value_set_rc (sym, total, PIVOT_RC_COUNT);
1786 pivot_table_put (sym, indexes, sym->n_dimensions, total);
1791 static bool calc_risk (struct crosstabulation *,
1792 double[], double[], double[], union value *,
1795 /* Display risk estimate. */
1797 display_risk (struct crosstabulation *xt, struct pivot_table *risk,
1798 struct pivot_dimension *risk_statistics)
1800 double risk_v[3], lower[3], upper[3], n_valid;
1802 if (!calc_risk (xt, risk_v, upper, lower, c, &n_valid))
1805 size_t *indexes = xnmalloc (risk->n_dimensions, sizeof *indexes);
1806 assert (xt->n_vars == 2);
1807 for (size_t i = 0; i < xt->n_consts; i++)
1808 indexes[i + 2] = xt->const_indexes[i];
1810 for (int i = 0; i < 3; i++)
1812 const struct variable *cv = xt->vars[COL_VAR].var;
1813 const struct variable *rv = xt->vars[ROW_VAR].var;
1815 if (risk_v[i] == SYSMIS)
1818 struct string label = DS_EMPTY_INITIALIZER;
1822 ds_put_format (&label, _("Odds Ratio for %s"), var_to_string (rv));
1823 ds_put_cstr (&label, " (");
1824 var_append_value_name (rv, &c[0], &label);
1825 ds_put_cstr (&label, " / ");
1826 var_append_value_name (rv, &c[1], &label);
1827 ds_put_cstr (&label, ")");
1831 ds_put_format (&label, _("For cohort %s = "), var_to_string (cv));
1832 var_append_value_name (cv, &xt->vars[ROW_VAR].values[i - 1], &label);
1836 indexes[1] = pivot_category_create_leaf (
1837 risk_statistics->root,
1838 pivot_value_new_user_text_nocopy (ds_steal_cstr (&label)));
1840 double entries[] = { risk_v[i], lower[i], upper[i] };
1841 for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
1844 pivot_table_put (risk, indexes, risk->n_dimensions,
1845 pivot_value_new_number (entries[i]));
1848 indexes[1] = pivot_category_create_leaf (
1849 risk_statistics->root,
1850 pivot_value_new_text (N_("N of Valid Cases")));
1852 pivot_table_put (risk, indexes, risk->n_dimensions,
1853 pivot_value_new_number (n_valid));
1857 static int calc_directional (struct crosstabs_proc *, struct crosstabulation *,
1858 double[N_DIRECTIONAL], double[N_DIRECTIONAL],
1859 double[N_DIRECTIONAL], double[N_DIRECTIONAL]);
1861 /* Display directional measures. */
1863 display_directional (struct crosstabs_proc *proc,
1864 struct crosstabulation *xt, struct pivot_table *direct)
1866 double direct_v[N_DIRECTIONAL];
1867 double direct_ase[N_DIRECTIONAL];
1868 double direct_t[N_DIRECTIONAL];
1869 double sig[N_DIRECTIONAL];
1870 if (!calc_directional (proc, xt, direct_v, direct_ase, direct_t, sig))
1873 size_t *indexes = xnmalloc (direct->n_dimensions, sizeof *indexes);
1874 assert (xt->n_vars == 2);
1875 for (size_t i = 0; i < xt->n_consts; i++)
1876 indexes[i + 2] = xt->const_indexes[i];
1878 for (int i = 0; i < N_DIRECTIONAL; i++)
1880 if (direct_v[i] == SYSMIS)
1885 double entries[] = {
1886 direct_v[i], direct_ase[i], direct_t[i], sig[i],
1888 for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
1889 if (entries[j] != SYSMIS)
1892 pivot_table_put (direct, indexes, direct->n_dimensions,
1893 pivot_value_new_number (entries[j]));
1900 /* Statistical calculations. */
1902 /* Returns the value of the logarithm of gamma (factorial) function for an integer
1905 log_gamma_int (double xt)
1910 for (i = 2; i < xt; i++)
1916 /* Calculate P_r as specified in _SPSS Statistical Algorithms_,
1918 static inline double
1919 Pr (int a, int b, int c, int d)
1921 return exp (log_gamma_int (a + b + 1.) - log_gamma_int (a + 1.)
1922 + log_gamma_int (c + d + 1.) - log_gamma_int (b + 1.)
1923 + log_gamma_int (a + c + 1.) - log_gamma_int (c + 1.)
1924 + log_gamma_int (b + d + 1.) - log_gamma_int (d + 1.)
1925 - log_gamma_int (a + b + c + d + 1.));
1928 /* Swap the contents of A and B. */
1930 swap (int *a, int *b)
1937 /* Calculate significance for Fisher's exact test as specified in
1938 _SPSS Statistical Algorithms_, Appendix 5. */
1940 calc_fisher (int a, int b, int c, int d, double *fisher1, double *fisher2)
1945 if (MIN (c, d) < MIN (a, b))
1946 swap (&a, &c), swap (&b, &d);
1947 if (MIN (b, d) < MIN (a, c))
1948 swap (&a, &b), swap (&c, &d);
1952 swap (&a, &b), swap (&c, &d);
1954 swap (&a, &c), swap (&b, &d);
1957 pn1 = Pr (a, b, c, d);
1959 for (xt = 1; xt <= a; xt++)
1961 *fisher1 += Pr (a - xt, b + xt, c + xt, d - xt);
1964 *fisher2 = *fisher1;
1966 for (xt = 1; xt <= b; xt++)
1968 double p = Pr (a + xt, b - xt, c - xt, d + xt);
1974 /* Calculates chi-squares into CHISQ. MAT is a matrix with N_COLS
1975 columns with values COLS and N_ROWS rows with values ROWS. Values
1976 in the matrix sum to xt->total. */
1978 calc_chisq (struct crosstabulation *xt,
1979 double chisq[N_CHISQ], int df[N_CHISQ],
1980 double *fisher1, double *fisher2)
1982 chisq[0] = chisq[1] = 0.;
1983 chisq[2] = chisq[3] = chisq[4] = SYSMIS;
1984 *fisher1 = *fisher2 = SYSMIS;
1986 df[0] = df[1] = (xt->ns_cols - 1) * (xt->ns_rows - 1);
1988 if (xt->ns_rows <= 1 || xt->ns_cols <= 1)
1990 chisq[0] = chisq[1] = SYSMIS;
1994 size_t n_cols = xt->vars[COL_VAR].n_values;
1995 FOR_EACH_POPULATED_ROW (r, xt)
1996 FOR_EACH_POPULATED_COLUMN (c, xt)
1998 const double expected = xt->row_tot[r] * xt->col_tot[c] / xt->total;
1999 const double freq = xt->mat[n_cols * r + c];
2000 const double residual = freq - expected;
2002 chisq[0] += residual * residual / expected;
2004 chisq[1] += freq * log (expected / freq);
2015 /* Calculate Yates and Fisher exact test. */
2016 if (xt->ns_cols == 2 && xt->ns_rows == 2)
2018 double f11, f12, f21, f22;
2024 FOR_EACH_POPULATED_COLUMN (c, xt)
2032 f11 = xt->mat[nz_cols[0]];
2033 f12 = xt->mat[nz_cols[1]];
2034 f21 = xt->mat[nz_cols[0] + n_cols];
2035 f22 = xt->mat[nz_cols[1] + n_cols];
2040 const double xt_ = fabs (f11 * f22 - f12 * f21) - 0.5 * xt->total;
2043 chisq[3] = (xt->total * pow2 (xt_)
2044 / (f11 + f12) / (f21 + f22)
2045 / (f11 + f21) / (f12 + f22));
2053 calc_fisher (f11 + .5, f12 + .5, f21 + .5, f22 + .5, fisher1, fisher2);
2056 /* Calculate Mantel-Haenszel. */
2057 if (var_is_numeric (xt->vars[ROW_VAR].var)
2058 && var_is_numeric (xt->vars[COL_VAR].var))
2060 double r, ase_0, ase_1;
2061 calc_r (xt, (double *) xt->vars[ROW_VAR].values,
2062 (double *) xt->vars[COL_VAR].values,
2063 &r, &ase_0, &ase_1);
2065 chisq[4] = (xt->total - 1.) * r * r;
2070 /* Calculate the value of Pearson's r. r is stored into R, its T value into
2071 T, and standard error into ERROR. The row and column values must be
2072 passed in XT and Y. */
2074 calc_r (struct crosstabulation *xt,
2075 double *XT, double *Y, double *r, double *t, double *error)
2077 size_t n_rows = xt->vars[ROW_VAR].n_values;
2078 size_t n_cols = xt->vars[COL_VAR].n_values;
2079 double SX, SY, S, T;
2081 double sum_XYf, sum_X2Y2f;
2082 double sum_Xr, sum_X2r;
2083 double sum_Yc, sum_Y2c;
2086 for (sum_X2Y2f = sum_XYf = 0., i = 0; i < n_rows; i++)
2087 for (j = 0; j < n_cols; j++)
2089 double fij = xt->mat[j + i * n_cols];
2090 double product = XT[i] * Y[j];
2091 double temp = fij * product;
2093 sum_X2Y2f += temp * product;
2096 for (sum_Xr = sum_X2r = 0., i = 0; i < n_rows; i++)
2098 sum_Xr += XT[i] * xt->row_tot[i];
2099 sum_X2r += pow2 (XT[i]) * xt->row_tot[i];
2101 Xbar = sum_Xr / xt->total;
2103 for (sum_Yc = sum_Y2c = 0., i = 0; i < n_cols; i++)
2105 sum_Yc += Y[i] * xt->col_tot[i];
2106 sum_Y2c += Y[i] * Y[i] * xt->col_tot[i];
2108 Ybar = sum_Yc / xt->total;
2110 S = sum_XYf - sum_Xr * sum_Yc / xt->total;
2111 SX = sum_X2r - pow2 (sum_Xr) / xt->total;
2112 SY = sum_Y2c - pow2 (sum_Yc) / xt->total;
2115 *t = *r / sqrt (1 - pow2 (*r)) * sqrt (xt->total - 2);
2120 for (s = c = 0., i = 0; i < n_rows; i++)
2121 for (j = 0; j < n_cols; j++)
2123 double Xresid, Yresid;
2126 Xresid = XT[i] - Xbar;
2127 Yresid = Y[j] - Ybar;
2128 temp = (T * Xresid * Yresid
2130 * (Xresid * Xresid * SY + Yresid * Yresid * SX)));
2131 y = xt->mat[j + i * n_cols] * temp * temp - c;
2136 *error = sqrt (s) / (T * T);
2140 /* Calculate symmetric statistics and their asymptotic standard
2141 errors. Returns 0 if none could be calculated. */
2143 calc_symmetric (struct crosstabs_proc *proc, struct crosstabulation *xt,
2144 double v[N_SYMMETRIC], double ase[N_SYMMETRIC],
2145 double t[N_SYMMETRIC],
2146 double somers_d_v[3], double somers_d_ase[3],
2147 double somers_d_t[3])
2149 size_t n_rows = xt->vars[ROW_VAR].n_values;
2150 size_t n_cols = xt->vars[COL_VAR].n_values;
2153 q = MIN (xt->ns_rows, xt->ns_cols);
2157 for (i = 0; i < N_SYMMETRIC; i++)
2158 v[i] = ase[i] = t[i] = SYSMIS;
2160 /* Phi, Cramer's V, contingency coefficient. */
2161 if (proc->statistics & ((1u << CRS_ST_PHI) | (1u << CRS_ST_CC)))
2163 double Xp = 0.; /* Pearson chi-square. */
2165 FOR_EACH_POPULATED_ROW (r, xt)
2166 FOR_EACH_POPULATED_COLUMN (c, xt)
2168 double expected = xt->row_tot[r] * xt->col_tot[c] / xt->total;
2169 double freq = xt->mat[n_cols * r + c];
2170 double residual = freq - expected;
2172 Xp += residual * residual / expected;
2175 if (proc->statistics & (1u << CRS_ST_PHI))
2177 v[0] = sqrt (Xp / xt->total);
2178 v[1] = sqrt (Xp / (xt->total * (q - 1)));
2180 if (proc->statistics & (1u << CRS_ST_CC))
2181 v[2] = sqrt (Xp / (Xp + xt->total));
2184 if (proc->statistics & ((1u << CRS_ST_BTAU) | (1u << CRS_ST_CTAU)
2185 | (1u << CRS_ST_GAMMA) | (1u << CRS_ST_D)))
2190 double btau_cum, ctau_cum, gamma_cum, d_yx_cum, d_xy_cum;
2194 Dr = Dc = pow2 (xt->total);
2195 for (r = 0; r < n_rows; r++)
2196 Dr -= pow2 (xt->row_tot[r]);
2197 for (c = 0; c < n_cols; c++)
2198 Dc -= pow2 (xt->col_tot[c]);
2200 cum = xnmalloc (n_cols * n_rows, sizeof *cum);
2201 for (c = 0; c < n_cols; c++)
2205 for (r = 0; r < n_rows; r++)
2206 cum[c + r * n_cols] = ct += xt->mat[c + r * n_cols];
2215 for (i = 0; i < n_rows; i++)
2219 for (j = 1; j < n_cols; j++)
2220 Cij += xt->col_tot[j] - cum[j + i * n_cols];
2223 for (j = 1; j < n_cols; j++)
2224 Dij += cum[j + (i - 1) * n_cols];
2228 double fij = xt->mat[j + i * n_cols];
2234 assert (j < n_cols);
2236 Cij -= xt->col_tot[j] - cum[j + i * n_cols];
2237 Dij += xt->col_tot[j - 1] - cum[j - 1 + i * n_cols];
2241 Cij += cum[j - 1 + (i - 1) * n_cols];
2242 Dij -= cum[j + (i - 1) * n_cols];
2248 if (proc->statistics & (1u << CRS_ST_BTAU))
2249 v[3] = (P - Q) / sqrt (Dr * Dc);
2250 if (proc->statistics & (1u << CRS_ST_CTAU))
2251 v[4] = (q * (P - Q)) / (pow2 (xt->total) * (q - 1));
2252 if (proc->statistics & (1u << CRS_ST_GAMMA))
2253 v[5] = (P - Q) / (P + Q);
2255 /* ASE for tau-b, tau-c, gamma. Calculations could be
2256 eliminated here, at expense of memory. */
2261 btau_cum = ctau_cum = gamma_cum = d_yx_cum = d_xy_cum = 0.;
2262 for (i = 0; i < n_rows; i++)
2266 for (j = 1; j < n_cols; j++)
2267 Cij += xt->col_tot[j] - cum[j + i * n_cols];
2270 for (j = 1; j < n_cols; j++)
2271 Dij += cum[j + (i - 1) * n_cols];
2275 double fij = xt->mat[j + i * n_cols];
2277 if (proc->statistics & (1u << CRS_ST_BTAU))
2279 const double temp = (2. * sqrt (Dr * Dc) * (Cij - Dij)
2280 + v[3] * (xt->row_tot[i] * Dc
2281 + xt->col_tot[j] * Dr));
2282 btau_cum += fij * temp * temp;
2286 const double temp = Cij - Dij;
2287 ctau_cum += fij * temp * temp;
2290 if (proc->statistics & (1u << CRS_ST_GAMMA))
2292 const double temp = Q * Cij - P * Dij;
2293 gamma_cum += fij * temp * temp;
2296 if (proc->statistics & (1u << CRS_ST_D))
2298 d_yx_cum += fij * pow2 (Dr * (Cij - Dij)
2299 - (P - Q) * (xt->total - xt->row_tot[i]));
2300 d_xy_cum += fij * pow2 (Dc * (Dij - Cij)
2301 - (Q - P) * (xt->total - xt->col_tot[j]));
2306 assert (j < n_cols);
2308 Cij -= xt->col_tot[j] - cum[j + i * n_cols];
2309 Dij += xt->col_tot[j - 1] - cum[j - 1 + i * n_cols];
2313 Cij += cum[j - 1 + (i - 1) * n_cols];
2314 Dij -= cum[j + (i - 1) * n_cols];
2320 btau_var = ((btau_cum
2321 - (xt->total * pow2 (xt->total * (P - Q) / sqrt (Dr * Dc) * (Dr + Dc))))
2323 if (proc->statistics & (1u << CRS_ST_BTAU))
2325 ase[3] = sqrt (btau_var);
2326 t[3] = v[3] / (2 * sqrt ((ctau_cum - (P - Q) * (P - Q) / xt->total)
2329 if (proc->statistics & (1u << CRS_ST_CTAU))
2331 ase[4] = ((2 * q / ((q - 1) * pow2 (xt->total)))
2332 * sqrt (ctau_cum - (P - Q) * (P - Q) / xt->total));
2333 t[4] = v[4] / ase[4];
2335 if (proc->statistics & (1u << CRS_ST_GAMMA))
2337 ase[5] = ((4. / ((P + Q) * (P + Q))) * sqrt (gamma_cum));
2338 t[5] = v[5] / (2. / (P + Q)
2339 * sqrt (ctau_cum - (P - Q) * (P - Q) / xt->total));
2341 if (proc->statistics & (1u << CRS_ST_D))
2343 somers_d_v[0] = (P - Q) / (.5 * (Dc + Dr));
2344 somers_d_ase[0] = SYSMIS;
2345 somers_d_t[0] = (somers_d_v[0]
2347 * sqrt (ctau_cum - pow2 (P - Q) / xt->total)));
2348 somers_d_v[1] = (P - Q) / Dc;
2349 somers_d_ase[1] = 2. / pow2 (Dc) * sqrt (d_xy_cum);
2350 somers_d_t[1] = (somers_d_v[1]
2352 * sqrt (ctau_cum - pow2 (P - Q) / xt->total)));
2353 somers_d_v[2] = (P - Q) / Dr;
2354 somers_d_ase[2] = 2. / pow2 (Dr) * sqrt (d_yx_cum);
2355 somers_d_t[2] = (somers_d_v[2]
2357 * sqrt (ctau_cum - pow2 (P - Q) / xt->total)));
2363 /* Spearman correlation, Pearson's r. */
2364 if (proc->statistics & (1u << CRS_ST_CORR))
2366 double *R = xmalloc (sizeof *R * n_rows);
2367 double *C = xmalloc (sizeof *C * n_cols);
2370 double y, t, c = 0., s = 0.;
2375 R[i] = s + (xt->row_tot[i] + 1.) / 2.;
2376 y = xt->row_tot[i] - c;
2382 assert (i < n_rows);
2387 double y, t, c = 0., s = 0.;
2392 C[j] = s + (xt->col_tot[j] + 1.) / 2;
2393 y = xt->col_tot[j] - c;
2399 assert (j < n_cols);
2403 calc_r (xt, R, C, &v[6], &t[6], &ase[6]);
2408 calc_r (xt, (double *) xt->vars[ROW_VAR].values,
2409 (double *) xt->vars[COL_VAR].values,
2410 &v[7], &t[7], &ase[7]);
2413 /* Cohen's kappa. */
2414 if (proc->statistics & (1u << CRS_ST_KAPPA) && xt->ns_rows == xt->ns_cols)
2416 double ase_under_h0;
2417 double sum_fii, sum_rici, sum_fiiri_ci, sum_fijri_ci2, sum_riciri_ci;
2420 for (sum_fii = sum_rici = sum_fiiri_ci = sum_riciri_ci = 0., i = j = 0;
2421 i < xt->ns_rows; i++, j++)
2425 while (xt->col_tot[j] == 0.)
2428 prod = xt->row_tot[i] * xt->col_tot[j];
2429 sum = xt->row_tot[i] + xt->col_tot[j];
2431 sum_fii += xt->mat[j + i * n_cols];
2433 sum_fiiri_ci += xt->mat[j + i * n_cols] * sum;
2434 sum_riciri_ci += prod * sum;
2436 for (sum_fijri_ci2 = 0., i = 0; i < xt->ns_rows; i++)
2437 for (j = 0; j < xt->ns_cols; j++)
2439 double sum = xt->row_tot[i] + xt->col_tot[j];
2440 sum_fijri_ci2 += xt->mat[j + i * n_cols] * sum * sum;
2443 v[8] = (xt->total * sum_fii - sum_rici) / (pow2 (xt->total) - sum_rici);
2445 ase_under_h0 = sqrt ((pow2 (xt->total) * sum_rici
2446 + sum_rici * sum_rici
2447 - xt->total * sum_riciri_ci)
2448 / (xt->total * (pow2 (xt->total) - sum_rici) * (pow2 (xt->total) - sum_rici)));
2450 ase[8] = sqrt (xt->total * (((sum_fii * (xt->total - sum_fii))
2451 / pow2 (pow2 (xt->total) - sum_rici))
2452 + ((2. * (xt->total - sum_fii)
2453 * (2. * sum_fii * sum_rici
2454 - xt->total * sum_fiiri_ci))
2455 / pow3 (pow2 (xt->total) - sum_rici))
2456 + (pow2 (xt->total - sum_fii)
2457 * (xt->total * sum_fijri_ci2 - 4.
2458 * sum_rici * sum_rici)
2459 / pow4 (pow2 (xt->total) - sum_rici))));
2461 t[8] = v[8] / ase_under_h0;
2467 /* Calculate risk estimate. */
2469 calc_risk (struct crosstabulation *xt,
2470 double *value, double *upper, double *lower, union value *c,
2473 size_t n_cols = xt->vars[COL_VAR].n_values;
2474 double f11, f12, f21, f22;
2477 for (int i = 0; i < 3; i++)
2478 value[i] = upper[i] = lower[i] = SYSMIS;
2480 if (xt->ns_rows != 2 || xt->ns_cols != 2)
2484 /* Find populated columns. */
2487 FOR_EACH_POPULATED_COLUMN (c, xt)
2491 /* Find populated rows. */
2494 FOR_EACH_POPULATED_ROW (r, xt)
2498 f11 = xt->mat[nz_cols[0] + n_cols * nz_rows[0]];
2499 f12 = xt->mat[nz_cols[1] + n_cols * nz_rows[0]];
2500 f21 = xt->mat[nz_cols[0] + n_cols * nz_rows[1]];
2501 f22 = xt->mat[nz_cols[1] + n_cols * nz_rows[1]];
2502 *n_valid = f11 + f12 + f21 + f22;
2504 c[0] = xt->vars[COL_VAR].values[nz_cols[0]];
2505 c[1] = xt->vars[COL_VAR].values[nz_cols[1]];
2508 value[0] = (f11 * f22) / (f12 * f21);
2509 v = sqrt (1. / f11 + 1. / f12 + 1. / f21 + 1. / f22);
2510 lower[0] = value[0] * exp (-1.960 * v);
2511 upper[0] = value[0] * exp (1.960 * v);
2513 value[1] = (f11 * (f21 + f22)) / (f21 * (f11 + f12));
2514 v = sqrt ((f12 / (f11 * (f11 + f12)))
2515 + (f22 / (f21 * (f21 + f22))));
2516 lower[1] = value[1] * exp (-1.960 * v);
2517 upper[1] = value[1] * exp (1.960 * v);
2519 value[2] = (f12 * (f21 + f22)) / (f22 * (f11 + f12));
2520 v = sqrt ((f11 / (f12 * (f11 + f12)))
2521 + (f21 / (f22 * (f21 + f22))));
2522 lower[2] = value[2] * exp (-1.960 * v);
2523 upper[2] = value[2] * exp (1.960 * v);
2528 /* Calculate directional measures. */
2530 calc_directional (struct crosstabs_proc *proc, struct crosstabulation *xt,
2531 double v[N_DIRECTIONAL], double ase[N_DIRECTIONAL],
2532 double t[N_DIRECTIONAL], double sig[N_DIRECTIONAL])
2534 size_t n_rows = xt->vars[ROW_VAR].n_values;
2535 size_t n_cols = xt->vars[COL_VAR].n_values;
2536 for (int i = 0; i < N_DIRECTIONAL; i++)
2537 v[i] = ase[i] = t[i] = sig[i] = SYSMIS;
2540 if (proc->statistics & (1u << CRS_ST_LAMBDA))
2542 /* Find maximum for each row and their sum. */
2543 double *fim = xnmalloc (n_rows, sizeof *fim);
2544 int *fim_index = xnmalloc (n_rows, sizeof *fim_index);
2545 double sum_fim = 0.0;
2546 for (int i = 0; i < n_rows; i++)
2548 double max = xt->mat[i * n_cols];
2551 for (int j = 1; j < n_cols; j++)
2552 if (xt->mat[j + i * n_cols] > max)
2554 max = xt->mat[j + i * n_cols];
2560 fim_index[i] = index;
2563 /* Find maximum for each column. */
2564 double *fmj = xnmalloc (n_cols, sizeof *fmj);
2565 int *fmj_index = xnmalloc (n_cols, sizeof *fmj_index);
2566 double sum_fmj = 0.0;
2567 for (int j = 0; j < n_cols; j++)
2569 double max = xt->mat[j];
2572 for (int i = 1; i < n_rows; i++)
2573 if (xt->mat[j + i * n_cols] > max)
2575 max = xt->mat[j + i * n_cols];
2581 fmj_index[j] = index;
2584 /* Find maximum row total. */
2585 double rm = xt->row_tot[0];
2587 for (int i = 1; i < n_rows; i++)
2588 if (xt->row_tot[i] > rm)
2590 rm = xt->row_tot[i];
2594 /* Find maximum column total. */
2595 double cm = xt->col_tot[0];
2597 for (int j = 1; j < n_cols; j++)
2598 if (xt->col_tot[j] > cm)
2600 cm = xt->col_tot[j];
2604 v[0] = (sum_fim + sum_fmj - cm - rm) / (2. * xt->total - rm - cm);
2605 v[1] = (sum_fmj - rm) / (xt->total - rm);
2606 v[2] = (sum_fim - cm) / (xt->total - cm);
2608 /* ASE1 for Y given XT. */
2611 for (int i = 0; i < n_rows; i++)
2612 if (cm_index == fim_index[i])
2614 ase[2] = sqrt ((xt->total - sum_fim) * (sum_fim + cm - 2. * accum)
2615 / pow3 (xt->total - cm));
2618 /* ASE0 for Y given XT. */
2621 for (int i = 0; i < n_rows; i++)
2622 if (cm_index != fim_index[i])
2623 accum += (xt->mat[i * n_cols + fim_index[i]]
2624 + xt->mat[i * n_cols + cm_index]);
2625 t[2] = v[2] / (sqrt (accum - pow2 (sum_fim - cm) / xt->total) / (xt->total - cm));
2628 /* ASE1 for XT given Y. */
2631 for (int j = 0; j < n_cols; j++)
2632 if (rm_index == fmj_index[j])
2634 ase[1] = sqrt ((xt->total - sum_fmj) * (sum_fmj + rm - 2. * accum)
2635 / pow3 (xt->total - rm));
2638 /* ASE0 for XT given Y. */
2641 for (int j = 0; j < n_cols; j++)
2642 if (rm_index != fmj_index[j])
2643 accum += (xt->mat[j + n_cols * fmj_index[j]]
2644 + xt->mat[j + n_cols * rm_index]);
2645 t[1] = v[1] / (sqrt (accum - pow2 (sum_fmj - rm) / xt->total) / (xt->total - rm));
2648 /* Symmetric ASE0 and ASE1. */
2650 double accum0 = 0.0;
2651 double accum1 = 0.0;
2652 for (int i = 0; i < n_rows; i++)
2653 for (int j = 0; j < n_cols; j++)
2655 int temp0 = (fmj_index[j] == i) + (fim_index[i] == j);
2656 int temp1 = (i == rm_index) + (j == cm_index);
2657 accum0 += xt->mat[j + i * n_cols] * pow2 (temp0 - temp1);
2658 accum1 += (xt->mat[j + i * n_cols]
2659 * pow2 (temp0 + (v[0] - 1.) * temp1));
2661 ase[0] = sqrt (accum1 - 4. * xt->total * v[0] * v[0]) / (2. * xt->total - rm - cm);
2662 t[0] = v[0] / (sqrt (accum0 - pow2 (sum_fim + sum_fmj - cm - rm) / xt->total)
2663 / (2. * xt->total - rm - cm));
2666 for (int i = 0; i < 3; i++)
2667 sig[i] = 2 * gsl_cdf_ugaussian_Q (t[i]);
2676 double sum_fij2_ri = 0.0;
2677 double sum_fij2_ci = 0.0;
2678 FOR_EACH_POPULATED_ROW (i, xt)
2679 FOR_EACH_POPULATED_COLUMN (j, xt)
2681 double temp = pow2 (xt->mat[j + i * n_cols]);
2682 sum_fij2_ri += temp / xt->row_tot[i];
2683 sum_fij2_ci += temp / xt->col_tot[j];
2686 double sum_ri2 = 0.0;
2687 for (int i = 0; i < n_rows; i++)
2688 sum_ri2 += pow2 (xt->row_tot[i]);
2690 double sum_cj2 = 0.0;
2691 for (int j = 0; j < n_cols; j++)
2692 sum_cj2 += pow2 (xt->col_tot[j]);
2694 v[3] = (xt->total * sum_fij2_ci - sum_ri2) / (pow2 (xt->total) - sum_ri2);
2695 v[4] = (xt->total * sum_fij2_ri - sum_cj2) / (pow2 (xt->total) - sum_cj2);
2699 if (proc->statistics & (1u << CRS_ST_UC))
2702 FOR_EACH_POPULATED_ROW (i, xt)
2703 UX -= xt->row_tot[i] / xt->total * log (xt->row_tot[i] / xt->total);
2706 FOR_EACH_POPULATED_COLUMN (j, xt)
2707 UY -= xt->col_tot[j] / xt->total * log (xt->col_tot[j] / xt->total);
2711 for (int i = 0; i < n_rows; i++)
2712 for (int j = 0; j < n_cols; j++)
2714 double entry = xt->mat[j + i * n_cols];
2719 P += entry * pow2 (log (xt->col_tot[j] * xt->row_tot[i] / (xt->total * entry)));
2720 UXY -= entry / xt->total * log (entry / xt->total);
2723 double ase1_yx = 0.0;
2724 double ase1_xy = 0.0;
2725 double ase1_sym = 0.0;
2726 for (int i = 0; i < n_rows; i++)
2727 for (int j = 0; j < n_cols; j++)
2729 double entry = xt->mat[j + i * n_cols];
2734 ase1_yx += entry * pow2 (UY * log (entry / xt->row_tot[i])
2735 + (UX - UXY) * log (xt->col_tot[j] / xt->total));
2736 ase1_xy += entry * pow2 (UX * log (entry / xt->col_tot[j])
2737 + (UY - UXY) * log (xt->row_tot[i] / xt->total));
2738 ase1_sym += entry * pow2 ((UXY
2739 * log (xt->row_tot[i] * xt->col_tot[j] / pow2 (xt->total)))
2740 - (UX + UY) * log (entry / xt->total));
2743 v[5] = 2. * ((UX + UY - UXY) / (UX + UY));
2744 ase[5] = (2. / (xt->total * pow2 (UX + UY))) * sqrt (ase1_sym);
2747 v[6] = (UX + UY - UXY) / UX;
2748 ase[6] = sqrt (ase1_xy) / (xt->total * UX * UX);
2749 t[6] = v[6] / (sqrt (P - xt->total * pow2 (UX + UY - UXY)) / (xt->total * UX));
2751 v[7] = (UX + UY - UXY) / UY;
2752 ase[7] = sqrt (ase1_yx) / (xt->total * UY * UY);
2753 t[7] = v[7] / (sqrt (P - xt->total * pow2 (UX + UY - UXY)) / (xt->total * UY));
2757 if (proc->statistics & (1u << CRS_ST_D))
2759 double v_dummy[N_SYMMETRIC];
2760 double ase_dummy[N_SYMMETRIC];
2761 double t_dummy[N_SYMMETRIC];
2762 double somers_d_v[3];
2763 double somers_d_ase[3];
2764 double somers_d_t[3];
2766 if (calc_symmetric (proc, xt, v_dummy, ase_dummy, t_dummy,
2767 somers_d_v, somers_d_ase, somers_d_t))
2769 for (int i = 0; i < 3; i++)
2771 v[8 + i] = somers_d_v[i];
2772 ase[8 + i] = somers_d_ase[i];
2773 t[8 + i] = somers_d_t[i];
2774 sig[8 + i] = 2 * gsl_cdf_ugaussian_Q (fabs (somers_d_t[i]));
2780 if (proc->statistics & (1u << CRS_ST_ETA))
2783 double sum_Xr = 0.0;
2784 double sum_X2r = 0.0;
2785 for (int i = 0; i < n_rows; i++)
2787 sum_Xr += xt->vars[ROW_VAR].values[i].f * xt->row_tot[i];
2788 sum_X2r += pow2 (xt->vars[ROW_VAR].values[i].f) * xt->row_tot[i];
2790 double SX = sum_X2r - pow2 (sum_Xr) / xt->total;
2793 FOR_EACH_POPULATED_COLUMN (j, xt)
2797 for (int i = 0; i < n_rows; i++)
2799 SXW += (pow2 (xt->vars[ROW_VAR].values[i].f)
2800 * xt->mat[j + i * n_cols]);
2801 cum += (xt->vars[ROW_VAR].values[i].f
2802 * xt->mat[j + i * n_cols]);
2805 SXW -= cum * cum / xt->col_tot[j];
2807 v[11] = sqrt (1. - SXW / SX);
2810 double sum_Yc = 0.0;
2811 double sum_Y2c = 0.0;
2812 for (int i = 0; i < n_cols; i++)
2814 sum_Yc += xt->vars[COL_VAR].values[i].f * xt->col_tot[i];
2815 sum_Y2c += pow2 (xt->vars[COL_VAR].values[i].f) * xt->col_tot[i];
2817 double SY = sum_Y2c - pow2 (sum_Yc) / xt->total;
2820 FOR_EACH_POPULATED_ROW (i, xt)
2823 for (int j = 0; j < n_cols; j++)
2825 SYW += (pow2 (xt->vars[COL_VAR].values[j].f)
2826 * xt->mat[j + i * n_cols]);
2827 cum += (xt->vars[COL_VAR].values[j].f
2828 * xt->mat[j + i * n_cols]);
2831 SYW -= cum * cum / xt->row_tot[i];
2833 v[12] = sqrt (1. - SYW / SY);