/* PSPP - a program for statistical analysis.
- Copyright (C) 1997-9, 2000, 2006, 2009 Free Software Foundation, Inc.
+ Copyright (C) 1997-9, 2000, 2006, 2009, 2010, 2011, 2012, 2013, 2014, 2016 Free Software Foundation, Inc.
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
/* FIXME:
- - Pearson's R (but not Spearman!) is off a little.
- - T values for Spearman's R and Pearson's R are wrong.
- - How to calculate significance of symmetric and directional measures?
- - Asymmetric ASEs and T values for lambda are wrong.
- - ASE of Goodman and Kruskal's tau is not calculated.
- - ASE of symmetric somers' d is wrong.
- - Approx. T of uncertainty coefficient is wrong.
+ - How to calculate significance of some symmetric and directional measures?
+ - How to calculate ASE for symmetric Somers ' d?
+ - How to calculate ASE for Goodman and Kruskal's tau?
+ - How to calculate approx. T of symmetric uncertainty coefficient?
*/
#include <config.h>
#include <ctype.h>
+#include <float.h>
#include <gsl/gsl_cdf.h>
#include <stdlib.h>
#include <stdio.h>
-#include <data/case.h>
-#include <data/casegrouper.h>
-#include <data/casereader.h>
-#include <data/data-out.h>
-#include <data/dictionary.h>
-#include <data/format.h>
-#include <data/procedure.h>
-#include <data/value-labels.h>
-#include <data/variable.h>
-#include <language/command.h>
-#include <language/dictionary/split-file.h>
-#include <language/lexer/lexer.h>
-#include <language/lexer/variable-parser.h>
-#include <libpspp/array.h>
-#include <libpspp/assertion.h>
-#include <libpspp/compiler.h>
-#include <libpspp/hash.h>
-#include <libpspp/message.h>
-#include <libpspp/misc.h>
-#include <libpspp/pool.h>
-#include <libpspp/str.h>
-#include <output/output.h>
-#include <output/table.h>
-
-#include "minmax.h"
-#include "xalloc.h"
-#include "xmalloca.h"
+#include "data/case.h"
+#include "data/casegrouper.h"
+#include "data/casereader.h"
+#include "data/data-out.h"
+#include "data/dataset.h"
+#include "data/dictionary.h"
+#include "data/format.h"
+#include "data/value-labels.h"
+#include "data/variable.h"
+#include "language/command.h"
+#include "language/stats/freq.h"
+#include "language/dictionary/split-file.h"
+#include "language/lexer/lexer.h"
+#include "language/lexer/variable-parser.h"
+#include "libpspp/array.h"
+#include "libpspp/assertion.h"
+#include "libpspp/compiler.h"
+#include "libpspp/hash-functions.h"
+#include "libpspp/hmap.h"
+#include "libpspp/hmapx.h"
+#include "libpspp/message.h"
+#include "libpspp/misc.h"
+#include "libpspp/pool.h"
+#include "libpspp/str.h"
+#include "output/pivot-table.h"
+#include "output/chart-item.h"
+#include "output/charts/barchart.h"
+
+#include "gl/minmax.h"
+#include "gl/xalloc.h"
+#include "gl/xsize.h"
#include "gettext.h"
#define _(msgid) gettext (msgid)
*^tables=custom;
+variables=custom;
missing=miss:!table/include/report;
+ count=roundwhat:asis/case/!cell,
+ roundhow:!round/truncate;
+write[wr_]=none,cells,all;
- +format=fmt:!labels/nolabels/novallabs,
- val:!avalue/dvalue,
+ +format=val:!avalue/dvalue,
indx:!noindex/index,
tabl:!tables/notables,
box:!box/nobox,
pivot:!pivot/nopivot;
- +cells[cl_]=count,none,expected,row,column,total,residual,sresidual,
- asresidual,all;
+ +barchart=;
+ +cells[cl_]=count,expected,row,column,total,residual,sresidual,
+ asresidual,all,none;
+statistics[st_]=chisq,phi,cc,lambda,uc,none,btau,ctau,risk,gamma,d,
kappa,eta,corr,all.
*/
/* Number of directional statistics. */
#define N_DIRECTIONAL 13
-/* A single table entry for general mode. */
-struct table_entry
+
+/* Indexes into the 'vars' member of struct crosstabulation and
+ struct crosstab member. */
+enum
{
- int table; /* Flattened table number. */
- union
- {
- double freq; /* Frequency count. */
- double *data; /* Crosstabulation table for integer mode. */
- }
- u;
- union value values[1]; /* Values. */
+ ROW_VAR = 0, /* Row variable. */
+ COL_VAR = 1 /* Column variable. */
+ /* Higher indexes cause multiple tables to be output. */
};
-/* A crosstabulation. */
-struct crosstab
+struct xtab_var
{
- int nvar; /* Number of variables. */
- double missing; /* Missing cases count. */
- int ofs; /* Integer mode: Offset into sorted_tab[]. */
- const struct variable *vars[2]; /* At least two variables; sorted by
- larger indices first. */
+ const struct variable *var;
+ union value *values;
+ size_t n_values;
+ };
+
+/* A crosstabulation of 2 or more variables. */
+struct crosstabulation
+ {
+ struct crosstabs_proc *proc;
+ struct fmt_spec weight_format; /* Format for weight variable. */
+ double missing; /* Weight of missing cases. */
+
+ /* Variables (2 or more). */
+ int n_vars;
+ struct xtab_var *vars;
+
+ /* Constants (0 or more). */
+ int n_consts;
+ struct xtab_var *const_vars;
+ size_t *const_indexes;
+
+ /* Data. */
+ struct hmap data;
+ struct freq **entries;
+ size_t n_entries;
+
+ /* Number of statistically interesting columns/rows
+ (columns/rows with data in them). */
+ int ns_cols, ns_rows;
+
+ /* Matrix contents. */
+ double *mat; /* Matrix proper. */
+ double *row_tot; /* Row totals. */
+ double *col_tot; /* Column totals. */
+ double total; /* Grand total. */
};
/* Integer mode variable info. */
struct var_range
{
+ struct hmap_node hmap_node; /* In struct crosstabs_proc var_ranges map. */
+ const struct variable *var; /* The variable. */
int min; /* Minimum value. */
int max; /* Maximum value + 1. */
int count; /* max - min. */
};
-static inline struct var_range *
-get_var_range (const struct variable *v)
-{
- return var_get_aux (v);
-}
-
-static int
-get_var_trimmed_width (const struct variable *v)
-{
- int width = var_get_width (v);
- return MIN (width, MAX_SHORT_STRING);
-}
-
-/* Indexes into crosstab.v. */
-enum
- {
- ROW_VAR = 0,
- COL_VAR = 1
- };
-
-/* General mode crosstabulation table. */
-static struct hsh_table *gen_tab; /* Hash table. */
-static int n_sorted_tab; /* Number of entries in sorted_tab. */
-static struct table_entry **sorted_tab; /* Sorted table. */
-
-/* Variables specifies on VARIABLES. */
-static const struct variable **variables;
-static size_t variables_cnt;
-
-/* TABLES. */
-static struct crosstab **xtab;
-static int nxtab;
-
-/* Integer or general mode? */
-enum
+struct crosstabs_proc
{
- INTEGER,
- GENERAL
+ const struct dictionary *dict;
+ enum { INTEGER, GENERAL } mode;
+ enum mv_class exclude;
+ bool pivot;
+ bool barchart;
+ bool bad_warn;
+ struct fmt_spec weight_format;
+
+ /* Variables specifies on VARIABLES. */
+ const struct variable **variables;
+ size_t n_variables;
+ struct hmap var_ranges;
+
+ /* TABLES. */
+ struct crosstabulation *pivots;
+ int n_pivots;
+
+ /* CELLS. */
+ int n_cells; /* Number of cells requested. */
+ unsigned int cells; /* Bit k is 1 if cell k is requested. */
+ int a_cells[CRS_CL_count]; /* 0...n_cells-1 are the requested cells. */
+
+ /* Rounding of cells. */
+ bool round_case_weights; /* Round case weights? */
+ bool round_cells; /* If !round_case_weights, round cells? */
+ bool round_down; /* Round down? (otherwise to nearest) */
+
+ /* STATISTICS. */
+ unsigned int statistics; /* Bit k is 1 if statistic k is requested. */
+
+ bool descending; /* True if descending sort order is requested. */
};
-static int mode;
-
-/* CELLS. */
-static int num_cells; /* Number of cells requested. */
-static int cells[8]; /* Cells requested. */
-
-/* WRITE. */
-static int write_style; /* One of WR_* that specifies the WRITE style. */
-/* Command parsing info. */
-static struct cmd_crosstabs cmd;
+const struct var_range *get_var_range (const struct crosstabs_proc *,
+ const struct variable *);
-/* Pools. */
-static struct pool *pl_tc; /* For table cells. */
-static struct pool *pl_col; /* For column data. */
+static bool should_tabulate_case (const struct crosstabulation *,
+ const struct ccase *, enum mv_class exclude);
+static void tabulate_general_case (struct crosstabulation *, const struct ccase *,
+ double weight);
+static void tabulate_integer_case (struct crosstabulation *, const struct ccase *,
+ double weight);
+static void postcalc (struct crosstabs_proc *);
-static int internal_cmd_crosstabs (struct lexer *lexer, struct dataset *ds);
-static void precalc (struct casereader *, const struct dataset *);
-static void calc_general (struct ccase *, const struct dataset *);
-static void calc_integer (struct ccase *, const struct dataset *);
-static void postcalc (const struct dataset *);
-static void submit (struct tab_table *);
-
-static void format_short (char *s, const struct fmt_spec *fp,
- const union value *v);
-
-/* Parse and execute CROSSTABS, then clean up. */
-int
-cmd_crosstabs (struct lexer *lexer, struct dataset *ds)
+static double
+round_weight (const struct crosstabs_proc *proc, double weight)
{
- int result = internal_cmd_crosstabs (lexer, ds);
- int i;
-
- free (variables);
- pool_destroy (pl_tc);
- pool_destroy (pl_col);
+ return proc->round_down ? floor (weight) : floor (weight + 0.5);
+}
- for (i = 0; i < nxtab; i++)
- free (xtab[i]);
- free (xtab);
+#define FOR_EACH_POPULATED_COLUMN(C, XT) \
+ for (int C = next_populated_column (0, XT); \
+ C < (XT)->vars[COL_VAR].n_values; \
+ C = next_populated_column (C + 1, XT))
+static int
+next_populated_column (int c, const struct crosstabulation *xt)
+{
+ int n_columns = xt->vars[COL_VAR].n_values;
+ for (; c < n_columns; c++)
+ if (xt->col_tot[c])
+ break;
+ return c;
+}
- return result;
+#define FOR_EACH_POPULATED_ROW(R, XT) \
+ for (int R = next_populated_row (0, XT); R < (XT)->vars[ROW_VAR].n_values; \
+ R = next_populated_row (R + 1, XT))
+static int
+next_populated_row (int r, const struct crosstabulation *xt)
+{
+ int n_rows = xt->vars[ROW_VAR].n_values;
+ for (; r < n_rows; r++)
+ if (xt->row_tot[r])
+ break;
+ return r;
}
/* Parses and executes the CROSSTABS procedure. */
-static int
-internal_cmd_crosstabs (struct lexer *lexer, struct dataset *ds)
+int
+cmd_crosstabs (struct lexer *lexer, struct dataset *ds)
{
+ struct var_range *range, *next_range;
+ struct crosstabs_proc proc;
struct casegrouper *grouper;
struct casereader *input, *group;
+ struct cmd_crosstabs cmd;
+ struct crosstabulation *xt;
+ int result;
bool ok;
int i;
- variables = NULL;
- variables_cnt = 0;
- xtab = NULL;
- nxtab = 0;
- pl_tc = pool_create ();
- pl_col = pool_create ();
+ proc.dict = dataset_dict (ds);
+ proc.bad_warn = true;
+ proc.variables = NULL;
+ proc.n_variables = 0;
+ hmap_init (&proc.var_ranges);
+ proc.pivots = NULL;
+ proc.n_pivots = 0;
+ proc.descending = false;
+ proc.weight_format = *dict_get_weight_format (dataset_dict (ds));
+
+ if (!parse_crosstabs (lexer, ds, &cmd, &proc))
+ {
+ result = CMD_FAILURE;
+ goto exit;
+ }
+
+ proc.mode = proc.n_variables ? INTEGER : GENERAL;
+ proc.barchart = cmd.sbc_barchart > 0;
- if (!parse_crosstabs (lexer, ds, &cmd, NULL))
- return CMD_FAILURE;
+ proc.descending = cmd.val == CRS_DVALUE;
- mode = variables ? INTEGER : GENERAL;
+ proc.round_case_weights = cmd.sbc_count && cmd.roundwhat == CRS_CASE;
+ proc.round_cells = cmd.sbc_count && cmd.roundwhat == CRS_CELL;
+ proc.round_down = cmd.roundhow == CRS_TRUNCATE;
/* CELLS. */
if (!cmd.sbc_cells)
- {
- cmd.a_cells[CRS_CL_COUNT] = 1;
- }
+ proc.cells = 1u << CRS_CL_COUNT;
+ else if (cmd.a_cells[CRS_CL_ALL])
+ proc.cells = UINT_MAX;
else
{
- int count = 0;
-
+ proc.cells = 0;
for (i = 0; i < CRS_CL_count; i++)
if (cmd.a_cells[i])
- count++;
- if (count == 0)
- {
- cmd.a_cells[CRS_CL_COUNT] = 1;
- cmd.a_cells[CRS_CL_ROW] = 1;
- cmd.a_cells[CRS_CL_COLUMN] = 1;
- cmd.a_cells[CRS_CL_TOTAL] = 1;
- }
- if (cmd.a_cells[CRS_CL_ALL])
- {
- for (i = 0; i < CRS_CL_count; i++)
- cmd.a_cells[i] = 1;
- cmd.a_cells[CRS_CL_ALL] = 0;
- }
- cmd.a_cells[CRS_CL_NONE] = 0;
+ proc.cells |= 1u << i;
+ if (proc.cells == 0)
+ proc.cells = ((1u << CRS_CL_COUNT)
+ | (1u << CRS_CL_ROW)
+ | (1u << CRS_CL_COLUMN)
+ | (1u << CRS_CL_TOTAL));
}
- for (num_cells = i = 0; i < CRS_CL_count; i++)
- if (cmd.a_cells[i])
- cells[num_cells++] = i;
+ proc.cells &= ((1u << CRS_CL_count) - 1);
+ proc.cells &= ~((1u << CRS_CL_NONE) | (1u << CRS_CL_ALL));
+ proc.n_cells = 0;
+ for (i = 0; i < CRS_CL_count; i++)
+ if (proc.cells & (1u << i))
+ proc.a_cells[proc.n_cells++] = i;
/* STATISTICS. */
- if (cmd.sbc_statistics)
+ if (cmd.a_statistics[CRS_ST_ALL])
+ proc.statistics = UINT_MAX;
+ else if (cmd.sbc_statistics)
{
int i;
- int count = 0;
+ proc.statistics = 0;
for (i = 0; i < CRS_ST_count; i++)
if (cmd.a_statistics[i])
- count++;
- if (count == 0)
- cmd.a_statistics[CRS_ST_CHISQ] = 1;
- if (cmd.a_statistics[CRS_ST_ALL])
- for (i = 0; i < CRS_ST_count; i++)
- cmd.a_statistics[i] = 1;
+ proc.statistics |= 1u << i;
+ if (proc.statistics == 0)
+ proc.statistics |= 1u << CRS_ST_CHISQ;
}
+ else
+ proc.statistics = 0;
/* MISSING. */
- if (cmd.miss == CRS_REPORT && mode == GENERAL)
- {
- msg (SE, _("Missing mode REPORT not allowed in general mode. "
- "Assuming MISSING=TABLE."));
- cmd.miss = CRS_TABLE;
+ proc.exclude = (cmd.miss == CRS_TABLE ? MV_ANY
+ : cmd.miss == CRS_INCLUDE ? MV_SYSTEM
+ : MV_NEVER);
+ if (proc.mode == GENERAL && proc.exclude == MV_NEVER)
+ {
+ msg (SE, _("Missing mode %s not allowed in general mode. "
+ "Assuming %s."), "REPORT", "MISSING=TABLE");
+ proc.exclude = MV_ANY;
}
- /* WRITE. */
- if (cmd.a_write[CRS_WR_ALL] && cmd.a_write[CRS_WR_CELLS])
- cmd.a_write[CRS_WR_ALL] = 0;
- if (cmd.a_write[CRS_WR_ALL] && mode == GENERAL)
- {
- msg (SE, _("Write mode ALL not allowed in general mode. "
- "Assuming WRITE=CELLS."));
- cmd.a_write[CRS_WR_CELLS] = 1;
- }
- if (cmd.sbc_write
- && (cmd.a_write[CRS_WR_NONE]
- + cmd.a_write[CRS_WR_ALL]
- + cmd.a_write[CRS_WR_CELLS] == 0))
- cmd.a_write[CRS_WR_CELLS] = 1;
- if (cmd.a_write[CRS_WR_CELLS])
- write_style = CRS_WR_CELLS;
- else if (cmd.a_write[CRS_WR_ALL])
- write_style = CRS_WR_ALL;
- else
- write_style = CRS_WR_NONE;
+ /* PIVOT. */
+ proc.pivot = cmd.pivot == CRS_PIVOT;
input = casereader_create_filter_weight (proc_open (ds), dataset_dict (ds),
NULL, NULL);
grouper = casegrouper_create_splits (input, dataset_dict (ds));
while (casegrouper_get_next_group (grouper, &group))
{
- struct ccase c;
+ struct ccase *c;
- precalc (group, ds);
-
- for (; casereader_read (group, &c); case_destroy (&c))
+ /* Output SPLIT FILE variables. */
+ c = casereader_peek (group, 0);
+ if (c != NULL)
{
- if (mode == GENERAL)
- calc_general (&c, ds);
- else
- calc_integer (&c, ds);
+ output_split_file_values (ds, c);
+ case_unref (c);
}
+
+ /* Initialize hash tables. */
+ for (xt = &proc.pivots[0]; xt < &proc.pivots[proc.n_pivots]; xt++)
+ hmap_init (&xt->data);
+
+ /* Tabulate. */
+ for (; (c = casereader_read (group)) != NULL; case_unref (c))
+ for (xt = &proc.pivots[0]; xt < &proc.pivots[proc.n_pivots]; xt++)
+ {
+ double weight = dict_get_case_weight (dataset_dict (ds), c,
+ &proc.bad_warn);
+ if (cmd.roundwhat == CRS_CASE)
+ {
+ weight = round_weight (&proc, weight);
+ if (weight == 0.)
+ continue;
+ }
+ if (should_tabulate_case (xt, c, proc.exclude))
+ {
+ if (proc.mode == GENERAL)
+ tabulate_general_case (xt, c, weight);
+ else
+ tabulate_integer_case (xt, c, weight);
+ }
+ else
+ xt->missing += weight;
+ }
casereader_destroy (group);
- postcalc (ds);
+ /* Output. */
+ postcalc (&proc);
}
ok = casegrouper_destroy (grouper);
ok = proc_commit (ds) && ok;
- return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
+ result = ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
+
+exit:
+ free (proc.variables);
+ HMAP_FOR_EACH_SAFE (range, next_range, struct var_range, hmap_node,
+ &proc.var_ranges)
+ {
+ hmap_delete (&proc.var_ranges, &range->hmap_node);
+ free (range);
+ }
+ for (xt = &proc.pivots[0]; xt < &proc.pivots[proc.n_pivots]; xt++)
+ {
+ free (xt->vars);
+ free (xt->const_vars);
+ free (xt->const_indexes);
+ }
+ free (proc.pivots);
+
+ return result;
}
/* Parses the TABLES subcommand. */
static int
-crs_custom_tables (struct lexer *lexer, struct dataset *ds, struct cmd_crosstabs *cmd UNUSED, void *aux UNUSED)
+crs_custom_tables (struct lexer *lexer, struct dataset *ds,
+ struct cmd_crosstabs *cmd UNUSED, void *proc_)
{
+ struct crosstabs_proc *proc = proc_;
struct const_var_set *var_set;
int n_by;
const struct variable ***by = NULL;
+ int *by_iter;
size_t *by_nvar = NULL;
size_t nx = 1;
- int success = 0;
+ bool ok = false;
+ int i;
/* Ensure that this is a TABLES subcommand. */
if (!lex_match_id (lexer, "TABLES")
&& (lex_token (lexer) != T_ID ||
- dict_lookup_var (dataset_dict (ds), lex_tokid (lexer)) == NULL)
+ dict_lookup_var (dataset_dict (ds), lex_tokcstr (lexer)) == NULL)
&& lex_token (lexer) != T_ALL)
return 2;
- lex_match (lexer, '=');
+ lex_match (lexer, T_EQUALS);
- if (variables != NULL)
- var_set = const_var_set_create_from_array (variables, variables_cnt);
+ if (proc->variables != NULL)
+ var_set = const_var_set_create_from_array (proc->variables,
+ proc->n_variables);
else
var_set = const_var_set_create_from_dict (dataset_dict (ds));
assert (var_set != NULL);
by = xnrealloc (by, n_by + 1, sizeof *by);
by_nvar = xnrealloc (by_nvar, n_by + 1, sizeof *by_nvar);
if (!parse_const_var_set_vars (lexer, var_set, &by[n_by], &by_nvar[n_by],
- PV_NO_DUPLICATE | PV_NO_SCRATCH))
+ PV_NO_DUPLICATE | PV_NO_SCRATCH))
goto done;
if (xalloc_oversized (nx, by_nvar[n_by]))
{
if (!lex_match (lexer, T_BY))
{
if (n_by < 2)
- {
- lex_error (lexer, _("expecting BY"));
- goto done;
- }
+ goto done;
else
break;
}
}
- {
- int *by_iter = xcalloc (n_by, sizeof *by_iter);
- int i;
-
- xtab = xnrealloc (xtab, nxtab + nx, sizeof *xtab);
- for (i = 0; i < nx; i++)
- {
- struct crosstab *x;
-
- x = xmalloc (sizeof *x + sizeof (struct variable *) * (n_by - 2));
- x->nvar = n_by;
- x->missing = 0.;
-
- {
- int i;
-
- for (i = 0; i < n_by; i++)
- x->vars[i] = by[i][by_iter[i]];
- }
+ by_iter = xcalloc (n_by, sizeof *by_iter);
+ proc->pivots = xnrealloc (proc->pivots,
+ proc->n_pivots + nx, sizeof *proc->pivots);
+ for (i = 0; i < nx; i++)
+ {
+ struct crosstabulation *xt = &proc->pivots[proc->n_pivots++];
+ int j;
- {
- int i;
+ xt->proc = proc;
+ xt->weight_format = proc->weight_format;
+ xt->missing = 0.;
+ xt->n_vars = n_by;
+ xt->vars = xcalloc (n_by, sizeof *xt->vars);
+ xt->n_consts = 0;
+ xt->const_vars = NULL;
+ xt->const_indexes = NULL;
- for (i = n_by - 1; i >= 0; i--)
- {
- if (++by_iter[i] < by_nvar[i])
- break;
- by_iter[i] = 0;
- }
- }
+ for (j = 0; j < n_by; j++)
+ xt->vars[j].var = by[j][by_iter[j]];
- xtab[nxtab++] = x;
- }
- free (by_iter);
- }
- success = 1;
+ for (j = n_by - 1; j >= 0; j--)
+ {
+ if (++by_iter[j] < by_nvar[j])
+ break;
+ by_iter[j] = 0;
+ }
+ }
+ free (by_iter);
+ ok = true;
- done:
+done:
/* All return paths lead here. */
- {
- int i;
-
- for (i = 0; i < n_by; i++)
- free (by[i]);
- free (by);
- free (by_nvar);
- }
+ for (i = 0; i < n_by; i++)
+ free (by[i]);
+ free (by);
+ free (by_nvar);
const_var_set_destroy (var_set);
- return success;
+ return ok;
}
/* Parses the VARIABLES subcommand. */
static int
-crs_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_crosstabs *cmd UNUSED, void *aux UNUSED)
+crs_custom_variables (struct lexer *lexer, struct dataset *ds,
+ struct cmd_crosstabs *cmd UNUSED, void *proc_)
{
- if (nxtab)
+ struct crosstabs_proc *proc = proc_;
+ if (proc->n_pivots)
{
- msg (SE, _("VARIABLES must be specified before TABLES."));
+ msg (SE, _("%s must be specified before %s."), "VARIABLES", "TABLES");
return 0;
}
- lex_match (lexer, '=');
+ lex_match (lexer, T_EQUALS);
for (;;)
{
- size_t orig_nv = variables_cnt;
+ size_t orig_nv = proc->n_variables;
size_t i;
long min, max;
if (!parse_variables_const (lexer, dataset_dict (ds),
- &variables, &variables_cnt,
- (PV_APPEND | PV_NUMERIC
- | PV_NO_DUPLICATE | PV_NO_SCRATCH)))
+ &proc->variables, &proc->n_variables,
+ (PV_APPEND | PV_NUMERIC
+ | PV_NO_DUPLICATE | PV_NO_SCRATCH)))
return 0;
- if (lex_token (lexer) != '(')
- {
- lex_error (lexer, "expecting `('");
+ if (!lex_force_match (lexer, T_LPAREN))
goto lossage;
- }
- lex_get (lexer);
if (!lex_force_int (lexer))
goto lossage;
min = lex_integer (lexer);
lex_get (lexer);
- lex_match (lexer, ',');
+ lex_match (lexer, T_COMMA);
if (!lex_force_int (lexer))
goto lossage;
}
lex_get (lexer);
- if (lex_token (lexer) != ')')
- {
- lex_error (lexer, "expecting `)'");
- goto lossage;
- }
- lex_get (lexer);
+ if (!lex_force_match (lexer, T_RPAREN))
+ goto lossage;
- for (i = orig_nv; i < variables_cnt; i++)
+ for (i = orig_nv; i < proc->n_variables; i++)
{
+ const struct variable *var = proc->variables[i];
struct var_range *vr = xmalloc (sizeof *vr);
+
+ vr->var = var;
vr->min = min;
- vr->max = max + 1.;
+ vr->max = max;
vr->count = max - min + 1;
- var_attach_aux (variables[i], vr, var_dtor_free);
+ hmap_insert (&proc->var_ranges, &vr->hmap_node,
+ hash_pointer (var, 0));
}
- if (lex_token (lexer) == '/')
+ if (lex_token (lexer) == T_SLASH)
break;
}
return 1;
lossage:
- free (variables);
- variables = NULL;
+ free (proc->variables);
+ proc->variables = NULL;
+ proc->n_variables = 0;
return 0;
}
\f
/* Data file processing. */
-static int compare_table_entry (const void *, const void *, const void *);
-static unsigned hash_table_entry (const void *, const void *);
-
-/* Set up the crosstabulation tables for processing. */
-static void
-precalc (struct casereader *input, const struct dataset *ds)
+const struct var_range *
+get_var_range (const struct crosstabs_proc *proc, const struct variable *var)
{
- struct ccase c;
-
- if (casereader_peek (input, 0, &c))
- {
- output_split_file_values (ds, &c);
- case_destroy (&c);
- }
-
- if (mode == GENERAL)
+ if (!hmap_is_empty (&proc->var_ranges))
{
- gen_tab = hsh_create (512, compare_table_entry, hash_table_entry,
- NULL, NULL);
- }
- else
- {
- int i;
-
- sorted_tab = NULL;
- n_sorted_tab = 0;
-
- for (i = 0; i < nxtab; i++)
- {
- struct crosstab *x = xtab[i];
- int count = 1;
- int *v;
- int j;
-
- x->ofs = n_sorted_tab;
-
- for (j = 2; j < x->nvar; j++)
- count *= get_var_range (x->vars[j - 2])->count;
-
- sorted_tab = xnrealloc (sorted_tab,
- n_sorted_tab + count, sizeof *sorted_tab);
- v = xmalloca (sizeof *v * x->nvar);
- for (j = 2; j < x->nvar; j++)
- v[j] = get_var_range (x->vars[j])->min;
- for (j = 0; j < count; j++)
- {
- struct table_entry *te;
- int k;
+ const struct var_range *range;
- te = sorted_tab[n_sorted_tab++]
- = xmalloc (sizeof *te + sizeof (union value) * (x->nvar - 1));
- te->table = i;
-
- {
- int row_cnt = get_var_range (x->vars[0])->count;
- int col_cnt = get_var_range (x->vars[1])->count;
- const int mat_size = row_cnt * col_cnt;
- int m;
-
- te->u.data = xnmalloc (mat_size, sizeof *te->u.data);
- for (m = 0; m < mat_size; m++)
- te->u.data[m] = 0.;
- }
-
- for (k = 2; k < x->nvar; k++)
- te->values[k].f = v[k];
- for (k = 2; k < x->nvar; k++)
- {
- struct var_range *vr = get_var_range (x->vars[k]);
- if (++v[k] >= vr->max)
- v[k] = vr->min;
- else
- break;
- }
- }
- freea (v);
- }
-
- sorted_tab = xnrealloc (sorted_tab,
- n_sorted_tab + 1, sizeof *sorted_tab);
- sorted_tab[n_sorted_tab] = NULL;
+ HMAP_FOR_EACH_IN_BUCKET (range, struct var_range, hmap_node,
+ hash_pointer (var, 0), &proc->var_ranges)
+ if (range->var == var)
+ return range;
}
+ return NULL;
}
-/* Form crosstabulations for general mode. */
-static void
-calc_general (struct ccase *c, const struct dataset *ds)
+static bool
+should_tabulate_case (const struct crosstabulation *xt, const struct ccase *c,
+ enum mv_class exclude)
{
- /* Missing values to exclude. */
- enum mv_class exclude = (cmd.miss == CRS_TABLE ? MV_ANY
- : cmd.miss == CRS_INCLUDE ? MV_SYSTEM
- : MV_NEVER);
-
- /* Case weight. */
- double weight = dict_get_case_weight (dataset_dict (ds), c, NULL);
-
- /* Flattened current table index. */
- int t;
-
- for (t = 0; t < nxtab; t++)
+ int j;
+ for (j = 0; j < xt->n_vars; j++)
{
- struct crosstab *x = xtab[t];
- const size_t entry_size = (sizeof (struct table_entry)
- + sizeof (union value) * (x->nvar - 1));
- struct table_entry *te = xmalloca (entry_size);
-
- /* Construct table entry for the current record and table. */
- te->table = t;
- {
- int j;
-
- assert (x != NULL);
- for (j = 0; j < x->nvar; j++)
- {
- const union value *v = case_data (c, x->vars[j]);
- if (var_is_value_missing (x->vars[j], v, exclude))
- {
- x->missing += weight;
- goto next_crosstab;
- }
-
- if (var_is_numeric (x->vars[j]))
- te->values[j].f = case_num (c, x->vars[j]);
- else
- {
- size_t n = get_var_trimmed_width (x->vars[j]);
- memcpy (te->values[j].s, case_str (c, x->vars[j]), n);
-
- /* Necessary in order to simplify comparisons. */
- memset (&te->values[j].s[n], 0, sizeof (union value) - n);
- }
- }
- }
-
- /* Add record to hash table. */
- {
- struct table_entry **tepp
- = (struct table_entry **) hsh_probe (gen_tab, te);
- if (*tepp == NULL)
- {
- struct table_entry *tep = pool_alloc (pl_tc, entry_size);
+ const struct variable *var = xt->vars[j].var;
+ const struct var_range *range = get_var_range (xt->proc, var);
- te->u.freq = weight;
- memcpy (tep, te, entry_size);
-
- *tepp = tep;
- }
- else
- (*tepp)->u.freq += weight;
- }
+ if (var_is_value_missing (var, case_data (c, var), exclude))
+ return false;
- next_crosstab:
- freea (te);
+ if (range != NULL)
+ {
+ double num = case_num (c, var);
+ if (num < range->min || num >= range->max + 1.)
+ return false;
+ }
}
+ return true;
}
static void
-calc_integer (struct ccase *c, const struct dataset *ds)
+tabulate_integer_case (struct crosstabulation *xt, const struct ccase *c,
+ double weight)
{
- bool bad_warn = true;
+ struct freq *te;
+ size_t hash;
+ int j;
- /* Case weight. */
- double weight = dict_get_case_weight (dataset_dict (ds), c, &bad_warn);
-
- /* Flattened current table index. */
- int t;
-
- for (t = 0; t < nxtab; t++)
+ hash = 0;
+ for (j = 0; j < xt->n_vars; j++)
{
- struct crosstab *x = xtab[t];
- int i, fact, ofs;
-
- fact = i = 1;
- ofs = x->ofs;
- for (i = 0; i < x->nvar; i++)
- {
- const struct variable *const v = x->vars[i];
- struct var_range *vr = get_var_range (v);
- double value = case_num (c, v);
-
- /* Note that the first test also rules out SYSMIS. */
- if ((value < vr->min || value >= vr->max)
- || (cmd.miss == CRS_TABLE
- && var_is_num_missing (v, value, MV_USER)))
- {
- x->missing += weight;
- goto next_crosstab;
- }
-
- if (i > 1)
- {
- ofs += fact * ((int) value - vr->min);
- fact *= vr->count;
- }
- }
-
- {
- const struct variable *row_var = x->vars[ROW_VAR];
- const int row = case_num (c, row_var) - get_var_range (row_var)->min;
-
- const struct variable *col_var = x->vars[COL_VAR];
- const int col = case_num (c, col_var) - get_var_range (col_var)->min;
+ /* Throw away fractional parts of values. */
+ hash = hash_int (case_num (c, xt->vars[j].var), hash);
+ }
- const int col_dim = get_var_range (col_var)->count;
+ HMAP_FOR_EACH_WITH_HASH (te, struct freq, node, hash, &xt->data)
+ {
+ for (j = 0; j < xt->n_vars; j++)
+ if ((int) case_num (c, xt->vars[j].var) != (int) te->values[j].f)
+ goto no_match;
- sorted_tab[ofs]->u.data[col + row * col_dim] += weight;
- }
+ /* Found an existing entry. */
+ te->count += weight;
+ return;
- next_crosstab: ;
+ no_match: ;
}
+
+ /* No existing entry. Create a new one. */
+ te = xmalloc (table_entry_size (xt->n_vars));
+ te->count = weight;
+ for (j = 0; j < xt->n_vars; j++)
+ te->values[j].f = (int) case_num (c, xt->vars[j].var);
+ hmap_insert (&xt->data, &te->node, hash);
}
-/* Compare the table_entry's at A and B and return a strcmp()-type
- result. */
-static int
-compare_table_entry (const void *a_, const void *b_, const void *aux UNUSED)
+static void
+tabulate_general_case (struct crosstabulation *xt, const struct ccase *c,
+ double weight)
{
- const struct table_entry *a = a_;
- const struct table_entry *b = b_;
+ struct freq *te;
+ size_t hash;
+ int j;
- if (a->table > b->table)
- return 1;
- else if (a->table < b->table)
- return -1;
-
- {
- const struct crosstab *x = xtab[a->table];
- int i;
+ hash = 0;
+ for (j = 0; j < xt->n_vars; j++)
+ {
+ const struct variable *var = xt->vars[j].var;
+ hash = value_hash (case_data (c, var), var_get_width (var), hash);
+ }
- for (i = x->nvar - 1; i >= 0; i--)
- if (var_is_numeric (x->vars[i]))
- {
- const double diffnum = a->values[i].f - b->values[i].f;
- if (diffnum < 0)
- return -1;
- else if (diffnum > 0)
- return 1;
- }
- else
+ HMAP_FOR_EACH_WITH_HASH (te, struct freq, node, hash, &xt->data)
+ {
+ for (j = 0; j < xt->n_vars; j++)
{
- int width = get_var_trimmed_width (x->vars[i]);
- const int diffstr = strncmp (a->values[i].s, b->values[i].s, width);
- if (diffstr)
- return diffstr;
+ const struct variable *var = xt->vars[j].var;
+ if (!value_equal (case_data (c, var), &te->values[j],
+ var_get_width (var)))
+ goto no_match;
}
- }
-
- return 0;
-}
-/* Calculate a hash value from table_entry A. */
-static unsigned
-hash_table_entry (const void *a_, const void *aux UNUSED)
-{
- const struct table_entry *a = a_;
- unsigned long hash;
- int i;
+ /* Found an existing entry. */
+ te->count += weight;
+ return;
- hash = a->table;
- for (i = 0; i < xtab[a->table]->nvar; i++)
- hash ^= hsh_hash_bytes (&a->values[i], sizeof a->values[i]);
+ no_match: ;
+ }
- return hash;
+ /* No existing entry. Create a new one. */
+ te = xmalloc (table_entry_size (xt->n_vars));
+ te->count = weight;
+ for (j = 0; j < xt->n_vars; j++)
+ {
+ const struct variable *var = xt->vars[j].var;
+ value_clone (&te->values[j], case_data (c, var), var_get_width (var));
+ }
+ hmap_insert (&xt->data, &te->node, hash);
}
\f
/* Post-data reading calculations. */
-static struct table_entry **find_pivot_extent (struct table_entry **,
- int *cnt, int pivot);
-static void enum_var_values (struct table_entry **entries, int entry_cnt,
- int var_idx,
- union value **values, int *value_cnt);
-static void output_pivot_table (struct table_entry **, struct table_entry **,
- const struct dictionary *,
- double **, double **, double **,
- int *, int *, int *);
-static void make_summary_table (const struct dictionary *);
+static int compare_table_entry_vars_3way (const struct freq *a,
+ const struct freq *b,
+ const struct crosstabulation *xt,
+ int idx0, int idx1);
+static int compare_table_entry_3way (const void *ap_, const void *bp_,
+ const void *xt_);
+static int compare_table_entry_3way_inv (const void *ap_, const void *bp_,
+ const void *xt_);
+
+static void enum_var_values (const struct crosstabulation *, int var_idx,
+ bool descending);
+static void free_var_values (const struct crosstabulation *, int var_idx);
+static void output_crosstabulation (struct crosstabs_proc *,
+ struct crosstabulation *);
+static void make_crosstabulation_subset (struct crosstabulation *xt,
+ size_t row0, size_t row1,
+ struct crosstabulation *subset);
+static void make_summary_table (struct crosstabs_proc *);
+static bool find_crosstab (struct crosstabulation *, size_t *row0p,
+ size_t *row1p);
static void
-postcalc (const struct dataset *ds)
+postcalc (struct crosstabs_proc *proc)
{
- if (mode == GENERAL)
- {
- n_sorted_tab = hsh_count (gen_tab);
- sorted_tab = (struct table_entry **) hsh_sort (gen_tab);
- }
- make_summary_table (dataset_dict (ds));
+ /* Round hash table entries, if requested
- /* Identify all the individual crosstabulation tables, and deal with
- them. */
- {
- struct table_entry **pb = sorted_tab, **pe; /* Pivot begin, pivot end. */
- int pc = n_sorted_tab; /* Pivot count. */
+ If this causes any of the cell counts to fall to zero, delete those
+ cells. */
+ if (proc->round_cells)
+ for (struct crosstabulation *xt = proc->pivots;
+ xt < &proc->pivots[proc->n_pivots]; xt++)
+ {
+ struct freq *e, *next;
+ HMAP_FOR_EACH_SAFE (e, next, struct freq, node, &xt->data)
+ {
+ e->count = round_weight (proc, e->count);
+ if (e->count == 0.0)
+ {
+ hmap_delete (&xt->data, &e->node);
+ free (e);
+ }
+ }
+ }
+
+ /* Convert hash tables into sorted arrays of entries. */
+ for (struct crosstabulation *xt = proc->pivots;
+ xt < &proc->pivots[proc->n_pivots]; xt++)
+ {
+ struct freq *e;
- double *mat = NULL, *row_tot = NULL, *col_tot = NULL;
- int maxrows = 0, maxcols = 0, maxcells = 0;
+ xt->n_entries = hmap_count (&xt->data);
+ xt->entries = xnmalloc (xt->n_entries, sizeof *xt->entries);
+ size_t i = 0;
+ HMAP_FOR_EACH (e, struct freq, node, &xt->data)
+ xt->entries[i++] = e;
+ hmap_destroy (&xt->data);
- for (;;)
- {
- pe = find_pivot_extent (pb, &pc, cmd.pivot == CRS_PIVOT);
- if (pe == NULL)
- break;
+ sort (xt->entries, xt->n_entries, sizeof *xt->entries,
+ proc->descending ? compare_table_entry_3way_inv : compare_table_entry_3way,
+ xt);
- output_pivot_table (pb, pe, dataset_dict (ds),
- &mat, &row_tot, &col_tot,
- &maxrows, &maxcols, &maxcells);
+ }
- pb = pe;
- }
- free (mat);
- free (row_tot);
- free (col_tot);
- }
+ make_summary_table (proc);
- hsh_destroy (gen_tab);
- if (mode == INTEGER)
+ /* Output each pivot table. */
+ for (struct crosstabulation *xt = proc->pivots;
+ xt < &proc->pivots[proc->n_pivots]; xt++)
{
- int i;
- for (i = 0; i < n_sorted_tab; i++)
+ if (proc->pivot || xt->n_vars == 2)
+ output_crosstabulation (proc, xt);
+ else
{
- free (sorted_tab[i]->u.data);
- free (sorted_tab[i]);
+ size_t row0 = 0, row1 = 0;
+ while (find_crosstab (xt, &row0, &row1))
+ {
+ struct crosstabulation subset;
+ make_crosstabulation_subset (xt, row0, row1, &subset);
+ output_crosstabulation (proc, &subset);
+ free (subset.const_indexes);
+ }
+ }
+ if (proc->barchart)
+ {
+ const struct variable **vars = xcalloc (xt->n_vars, sizeof *vars);
+ for (size_t i = 0; i < xt->n_vars; i++)
+ vars[i] = xt->vars[i].var;
+ chart_item_submit (barchart_create (vars, xt->n_vars, _("Count"),
+ false,
+ xt->entries, xt->n_entries));
+ free (vars);
}
- free (sorted_tab);
}
-}
-
-static void insert_summary (struct tab_table *, int tab_index,
- const struct dictionary *,
- double valid);
-
-/* Output a table summarizing the cases processed. */
-static void
-make_summary_table (const struct dictionary *dict)
-{
- struct tab_table *summary;
-
- struct table_entry **pb = sorted_tab, **pe;
- int pc = n_sorted_tab;
- int cur_tab = 0;
-
- summary = tab_create (7, 3 + nxtab, 1);
- tab_title (summary, _("Summary."));
- tab_headers (summary, 1, 0, 3, 0);
- tab_joint_text (summary, 1, 0, 6, 0, TAB_CENTER, _("Cases"));
- tab_joint_text (summary, 1, 1, 2, 1, TAB_CENTER, _("Valid"));
- tab_joint_text (summary, 3, 1, 4, 1, TAB_CENTER, _("Missing"));
- tab_joint_text (summary, 5, 1, 6, 1, TAB_CENTER, _("Total"));
- tab_hline (summary, TAL_1, 1, 6, 1);
- tab_hline (summary, TAL_1, 1, 6, 2);
- tab_vline (summary, TAL_1, 3, 1, 1);
- tab_vline (summary, TAL_1, 5, 1, 1);
- {
- int i;
-
- for (i = 0; i < 3; i++)
- {
- tab_text (summary, 1 + i * 2, 2, TAB_RIGHT, _("N"));
- tab_text (summary, 2 + i * 2, 2, TAB_RIGHT, _("Percent"));
- }
- }
- tab_offset (summary, 0, 3);
- for (;;)
+ /* Free output and prepare for next split file. */
+ for (struct crosstabulation *xt = proc->pivots;
+ xt < &proc->pivots[proc->n_pivots]; xt++)
{
- double valid;
-
- pe = find_pivot_extent (pb, &pc, cmd.pivot == CRS_PIVOT);
- if (pe == NULL)
- break;
+ xt->missing = 0.0;
- while (cur_tab < (*pb)->table)
- insert_summary (summary, cur_tab++, dict, 0.);
+ /* Free the members that were allocated in this function(and the values
+ owned by the entries.
- if (mode == GENERAL)
- for (valid = 0.; pb < pe; pb++)
- valid += (*pb)->u.freq;
- else
- {
- const struct crosstab *const x = xtab[(*pb)->table];
- const int n_cols = get_var_range (x->vars[COL_VAR])->count;
- const int n_rows = get_var_range (x->vars[ROW_VAR])->count;
- const int count = n_cols * n_rows;
-
- for (valid = 0.; pb < pe; pb++)
- {
- const double *data = (*pb)->u.data;
- int i;
-
- for (i = 0; i < count; i++)
- valid += *data++;
- }
- }
- insert_summary (summary, cur_tab++, dict, valid);
+ The other pointer members are either both allocated and destroyed at a
+ lower level (in output_crosstabulation), or both allocated and
+ destroyed at a higher level (in crs_custom_tables and free_proc,
+ respectively). */
+ for (size_t i = 0; i < xt->n_vars; i++)
+ {
+ int width = var_get_width (xt->vars[i].var);
+ if (value_needs_init (width))
+ {
+ size_t j;
+
+ for (j = 0; j < xt->n_entries; j++)
+ value_destroy (&xt->entries[j]->values[i], width);
+ }
+ }
- pb = pe;
+ for (size_t i = 0; i < xt->n_entries; i++)
+ free (xt->entries[i]);
+ free (xt->entries);
}
-
- while (cur_tab < nxtab)
- insert_summary (summary, cur_tab++, dict, 0.);
-
- submit (summary);
}
-/* Inserts a line into T describing the crosstabulation at index
- TAB_INDEX, which has VALID valid observations. */
static void
-insert_summary (struct tab_table *t, int tab_index,
- const struct dictionary *dict,
- double valid)
+make_crosstabulation_subset (struct crosstabulation *xt, size_t row0,
+ size_t row1, struct crosstabulation *subset)
{
- struct crosstab *x = xtab[tab_index];
-
- const struct variable *wv = dict_get_weight (dict);
- const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
-
- tab_hline (t, TAL_1, 0, 6, 0);
-
- /* Crosstabulation name. */
- {
- char *buf = xmalloca (128 * x->nvar);
- char *cp = buf;
- int i;
-
- for (i = 0; i < x->nvar; i++)
- {
- if (i > 0)
- cp = stpcpy (cp, " * ");
-
- cp = stpcpy (cp, var_to_string (x->vars[i]));
- }
- tab_text (t, 0, 0, TAB_LEFT, buf);
+ *subset = *xt;
+ if (xt->n_vars > 2)
+ {
+ assert (xt->n_consts == 0);
+ subset->n_vars = 2;
+ subset->vars = xt->vars;
+
+ subset->n_consts = xt->n_vars - 2;
+ subset->const_vars = xt->vars + 2;
+ subset->const_indexes = xcalloc (subset->n_consts,
+ sizeof *subset->const_indexes);
+ for (size_t i = 0; i < subset->n_consts; i++)
+ {
+ const union value *value = &xt->entries[row0]->values[2 + i];
- freea (buf);
- }
+ for (size_t j = 0; j < xt->vars[2 + i].n_values; j++)
+ if (value_equal (&xt->vars[2 + i].values[j], value,
+ var_get_width (xt->vars[2 + i].var)))
+ {
+ subset->const_indexes[i] = j;
+ goto found;
+ }
+ NOT_REACHED ();
+ found: ;
+ }
+ }
+ subset->entries = &xt->entries[row0];
+ subset->n_entries = row1 - row0;
+}
- /* Counts and percentages. */
- {
- double n[3];
- int i;
+static int
+compare_table_entry_var_3way (const struct freq *a,
+ const struct freq *b,
+ const struct crosstabulation *xt,
+ int idx)
+{
+ return value_compare_3way (&a->values[idx], &b->values[idx],
+ var_get_width (xt->vars[idx].var));
+}
- n[0] = valid;
- n[1] = x->missing;
- n[2] = n[0] + n[1];
+static int
+compare_table_entry_vars_3way (const struct freq *a,
+ const struct freq *b,
+ const struct crosstabulation *xt,
+ int idx0, int idx1)
+{
+ int i;
+ for (i = idx1 - 1; i >= idx0; i--)
+ {
+ int cmp = compare_table_entry_var_3way (a, b, xt, i);
+ if (cmp != 0)
+ return cmp;
+ }
+ return 0;
+}
- for (i = 0; i < 3; i++)
- {
- tab_double (t, i * 2 + 1, 0, TAB_RIGHT, n[i], wfmt);
- tab_text (t, i * 2 + 2, 0, TAB_RIGHT | TAT_PRINTF, "%.1f%%",
- n[i] / n[2] * 100.);
- }
- }
+/* Compare the struct freq at *AP to the one at *BP and
+ return a strcmp()-type result. */
+static int
+compare_table_entry_3way (const void *ap_, const void *bp_, const void *xt_)
+{
+ const struct freq *const *ap = ap_;
+ const struct freq *const *bp = bp_;
+ const struct freq *a = *ap;
+ const struct freq *b = *bp;
+ const struct crosstabulation *xt = xt_;
+ int cmp;
+
+ cmp = compare_table_entry_vars_3way (a, b, xt, 2, xt->n_vars);
+ if (cmp != 0)
+ return cmp;
+
+ cmp = compare_table_entry_var_3way (a, b, xt, ROW_VAR);
+ if (cmp != 0)
+ return cmp;
+
+ return compare_table_entry_var_3way (a, b, xt, COL_VAR);
+}
- tab_next_row (t);
+/* Inverted version of compare_table_entry_3way */
+static int
+compare_table_entry_3way_inv (const void *ap_, const void *bp_, const void *xt_)
+{
+ return -compare_table_entry_3way (ap_, bp_, xt_);
}
-\f
-/* Output. */
-/* Tables. */
-static struct tab_table *table; /* Crosstabulation table. */
-static struct tab_table *chisq; /* Chi-square table. */
-static struct tab_table *sym; /* Symmetric measures table. */
-static struct tab_table *risk; /* Risk estimate table. */
-static struct tab_table *direct; /* Directional measures table. */
-
-/* Statistics. */
-static int chisq_fisher; /* Did any rows include Fisher's exact test? */
-
-/* Column values, number of columns. */
-static union value *cols;
-static int n_cols;
-
-/* Row values, number of rows. */
-static union value *rows;
-static int n_rows;
-
-/* Number of statistically interesting columns/rows (columns/rows with
- data in them). */
-static int ns_cols, ns_rows;
-
-/* Crosstabulation. */
-static const struct crosstab *x;
-
-/* Number of variables from the crosstabulation to consider. This is
- either x->nvar, if pivoting is on, or 2, if pivoting is off. */
-static int nvar;
-
-/* Matrix contents. */
-static double *mat; /* Matrix proper. */
-static double *row_tot; /* Row totals. */
-static double *col_tot; /* Column totals. */
-static double W; /* Grand total. */
-
-static void display_dimensions (struct tab_table *, int first_difference,
- struct table_entry *);
-static void display_crosstabulation (void);
-static void display_chisq (const struct dictionary *);
-static void display_symmetric (const struct dictionary *);
-static void display_risk (const struct dictionary *);
-static void display_directional (void);
-static void crosstabs_dim (struct tab_table *, struct outp_driver *);
-static void table_value_missing (struct tab_table *table, int c, int r,
- unsigned char opt, const union value *v,
- const struct variable *var);
-static void delete_missing (void);
-
-/* Output pivot table beginning at PB and continuing until PE,
- exclusive. For efficiency, *MATP is a pointer to a matrix that can
- hold *MAXROWS entries. */
+/* Output a table summarizing the cases processed. */
static void
-output_pivot_table (struct table_entry **pb, struct table_entry **pe,
- const struct dictionary *dict,
- double **matp, double **row_totp, double **col_totp,
- int *maxrows, int *maxcols, int *maxcells)
+make_summary_table (struct crosstabs_proc *proc)
{
- /* Subtable. */
- struct table_entry **tb = pb, **te; /* Table begin, table end. */
- int tc = pe - pb; /* Table count. */
+ struct pivot_table *table = pivot_table_create (N_("Summary"));
+ pivot_table_set_weight_var (table, dict_get_weight (proc->dict));
- /* Table entry for header comparison. */
- struct table_entry *cmp = NULL;
+ pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
+ N_("N"), PIVOT_RC_COUNT,
+ N_("Percent"), PIVOT_RC_PERCENT);
- x = xtab[(*pb)->table];
- enum_var_values (pb, pe - pb, COL_VAR, &cols, &n_cols);
+ struct pivot_dimension *cases = pivot_dimension_create (
+ table, PIVOT_AXIS_COLUMN, N_("Cases"),
+ N_("Valid"), N_("Missing"), N_("Total"));
+ cases->root->show_label = true;
- nvar = cmd.pivot == CRS_PIVOT ? x->nvar : 2;
-
- /* Crosstabulation table initialization. */
- if (num_cells)
+ struct pivot_dimension *tables = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Crosstabulation"));
+ for (struct crosstabulation *xt = &proc->pivots[0];
+ xt < &proc->pivots[proc->n_pivots]; xt++)
{
- table = tab_create (nvar + n_cols,
- (pe - pb) / n_cols * 3 / 2 * num_cells + 10, 1);
- tab_headers (table, nvar - 1, 0, 2, 0);
-
- /* First header line. */
- tab_joint_text (table, nvar - 1, 0, (nvar - 1) + (n_cols - 1), 0,
- TAB_CENTER | TAT_TITLE, var_get_name (x->vars[COL_VAR]));
-
- tab_hline (table, TAL_1, nvar - 1, nvar + n_cols - 2, 1);
-
- /* Second header line. */
- {
- int i;
-
- for (i = 2; i < nvar; i++)
- tab_joint_text (table, nvar - i - 1, 0, nvar - i - 1, 1,
- TAB_RIGHT | TAT_TITLE, var_to_string (x->vars[i]));
- tab_text (table, nvar - 2, 1, TAB_RIGHT | TAT_TITLE,
- var_get_name (x->vars[ROW_VAR]));
- for (i = 0; i < n_cols; i++)
- table_value_missing (table, nvar + i - 1, 1, TAB_RIGHT, &cols[i],
- x->vars[COL_VAR]);
- tab_text (table, nvar + n_cols - 1, 1, TAB_CENTER, _("Total"));
- }
-
- tab_hline (table, TAL_1, 0, nvar + n_cols - 1, 2);
- tab_vline (table, TAL_1, nvar + n_cols - 1, 0, 1);
-
- /* Title. */
- {
- char *title = xmalloca (x->nvar * 64 + 128);
- char *cp = title;
- int i;
-
- if (cmd.pivot == CRS_PIVOT)
- for (i = 0; i < nvar; i++)
- {
- if (i)
- cp = stpcpy (cp, " by ");
- cp = stpcpy (cp, var_get_name (x->vars[i]));
- }
- else
- {
- cp = spprintf (cp, "%s by %s for",
- var_get_name (x->vars[0]),
- var_get_name (x->vars[1]));
- for (i = 2; i < nvar; i++)
- {
- char buf[64], *bufp;
-
- if (i > 2)
- *cp++ = ',';
- *cp++ = ' ';
- cp = stpcpy (cp, var_get_name (x->vars[i]));
- *cp++ = '=';
- format_short (buf, var_get_print_format (x->vars[i]),
- &(*pb)->values[i]);
- for (bufp = buf; isspace ((unsigned char) *bufp); bufp++)
- ;
- cp = stpcpy (cp, bufp);
- }
- }
-
- cp = stpcpy (cp, " [");
- for (i = 0; i < num_cells; i++)
- {
- struct tuple
- {
- int value;
- const char *name;
- };
-
- static const struct tuple cell_names[] =
- {
- {CRS_CL_COUNT, N_("count")},
- {CRS_CL_ROW, N_("row %")},
- {CRS_CL_COLUMN, N_("column %")},
- {CRS_CL_TOTAL, N_("total %")},
- {CRS_CL_EXPECTED, N_("expected")},
- {CRS_CL_RESIDUAL, N_("residual")},
- {CRS_CL_SRESIDUAL, N_("std. resid.")},
- {CRS_CL_ASRESIDUAL, N_("adj. resid.")},
- {-1, NULL},
- };
-
- const struct tuple *t;
-
- for (t = cell_names; t->value != cells[i]; t++)
- assert (t->value != -1);
- if (i)
- cp = stpcpy (cp, ", ");
- cp = stpcpy (cp, gettext (t->name));
- }
- strcpy (cp, "].");
-
- tab_title (table, "%s", title);
- freea (title);
- }
+ struct string name = DS_EMPTY_INITIALIZER;
+ for (size_t i = 0; i < xt->n_vars; i++)
+ {
+ if (i > 0)
+ ds_put_cstr (&name, " × ");
+ ds_put_cstr (&name, var_to_string (xt->vars[i].var));
+ }
- tab_offset (table, 0, 2);
- }
- else
- table = NULL;
+ int row = pivot_category_create_leaf (
+ tables->root,
+ pivot_value_new_user_text_nocopy (ds_steal_cstr (&name)));
- /* Chi-square table initialization. */
- if (cmd.a_statistics[CRS_ST_CHISQ])
- {
- chisq = tab_create (6 + (nvar - 2),
- (pe - pb) / n_cols * 3 / 2 * N_CHISQ + 10, 1);
- tab_headers (chisq, 1 + (nvar - 2), 0, 1, 0);
-
- tab_title (chisq, _("Chi-square tests."));
-
- tab_offset (chisq, nvar - 2, 0);
- tab_text (chisq, 0, 0, TAB_LEFT | TAT_TITLE, _("Statistic"));
- tab_text (chisq, 1, 0, TAB_RIGHT | TAT_TITLE, _("Value"));
- tab_text (chisq, 2, 0, TAB_RIGHT | TAT_TITLE, _("df"));
- tab_text (chisq, 3, 0, TAB_RIGHT | TAT_TITLE,
- _("Asymp. Sig. (2-sided)"));
- tab_text (chisq, 4, 0, TAB_RIGHT | TAT_TITLE,
- _("Exact. Sig. (2-sided)"));
- tab_text (chisq, 5, 0, TAB_RIGHT | TAT_TITLE,
- _("Exact. Sig. (1-sided)"));
- chisq_fisher = 0;
- tab_offset (chisq, 0, 1);
- }
- else
- chisq = NULL;
+ double valid = 0.;
+ for (size_t i = 0; i < xt->n_entries; i++)
+ valid += xt->entries[i]->count;
- /* Symmetric measures. */
- if (cmd.a_statistics[CRS_ST_PHI] || cmd.a_statistics[CRS_ST_CC]
- || cmd.a_statistics[CRS_ST_BTAU] || cmd.a_statistics[CRS_ST_CTAU]
- || cmd.a_statistics[CRS_ST_GAMMA] || cmd.a_statistics[CRS_ST_CORR]
- || cmd.a_statistics[CRS_ST_KAPPA])
- {
- sym = tab_create (6 + (nvar - 2), (pe - pb) / n_cols * 7 + 10, 1);
- tab_headers (sym, 2 + (nvar - 2), 0, 1, 0);
- tab_title (sym, _("Symmetric measures."));
-
- tab_offset (sym, nvar - 2, 0);
- tab_text (sym, 0, 0, TAB_LEFT | TAT_TITLE, _("Category"));
- tab_text (sym, 1, 0, TAB_LEFT | TAT_TITLE, _("Statistic"));
- tab_text (sym, 2, 0, TAB_RIGHT | TAT_TITLE, _("Value"));
- tab_text (sym, 3, 0, TAB_RIGHT | TAT_TITLE, _("Asymp. Std. Error"));
- tab_text (sym, 4, 0, TAB_RIGHT | TAT_TITLE, _("Approx. T"));
- tab_text (sym, 5, 0, TAB_RIGHT | TAT_TITLE, _("Approx. Sig."));
- tab_offset (sym, 0, 1);
+ double n[3];
+ n[0] = valid;
+ n[1] = xt->missing;
+ n[2] = n[0] + n[1];
+ for (int i = 0; i < 3; i++)
+ {
+ pivot_table_put3 (table, 0, i, row, pivot_value_new_number (n[i]));
+ pivot_table_put3 (table, 1, i, row,
+ pivot_value_new_number (n[i] / n[2] * 100.0));
+ }
}
- else
- sym = NULL;
- /* Risk estimate. */
- if (cmd.a_statistics[CRS_ST_RISK])
- {
- risk = tab_create (4 + (nvar - 2), (pe - pb) / n_cols * 4 + 10, 1);
- tab_headers (risk, 1 + nvar - 2, 0, 2, 0);
- tab_title (risk, _("Risk estimate."));
-
- tab_offset (risk, nvar - 2, 0);
- tab_joint_text (risk, 2, 0, 3, 0, TAB_CENTER | TAT_TITLE | TAT_PRINTF,
- _("95%% Confidence Interval"));
- tab_text (risk, 0, 1, TAB_LEFT | TAT_TITLE, _("Statistic"));
- tab_text (risk, 1, 1, TAB_RIGHT | TAT_TITLE, _("Value"));
- tab_text (risk, 2, 1, TAB_RIGHT | TAT_TITLE, _("Lower"));
- tab_text (risk, 3, 1, TAB_RIGHT | TAT_TITLE, _("Upper"));
- tab_hline (risk, TAL_1, 2, 3, 1);
- tab_vline (risk, TAL_1, 2, 0, 1);
- tab_offset (risk, 0, 2);
- }
- else
- risk = NULL;
+ pivot_table_submit (table);
+}
+\f
+/* Output. */
- /* Directional measures. */
- if (cmd.a_statistics[CRS_ST_LAMBDA] || cmd.a_statistics[CRS_ST_UC]
- || cmd.a_statistics[CRS_ST_D] || cmd.a_statistics[CRS_ST_ETA])
- {
- direct = tab_create (7 + (nvar - 2), (pe - pb) / n_cols * 7 + 10, 1);
- tab_headers (direct, 3 + (nvar - 2), 0, 1, 0);
- tab_title (direct, _("Directional measures."));
-
- tab_offset (direct, nvar - 2, 0);
- tab_text (direct, 0, 0, TAB_LEFT | TAT_TITLE, _("Category"));
- tab_text (direct, 1, 0, TAB_LEFT | TAT_TITLE, _("Statistic"));
- tab_text (direct, 2, 0, TAB_LEFT | TAT_TITLE, _("Type"));
- tab_text (direct, 3, 0, TAB_RIGHT | TAT_TITLE, _("Value"));
- tab_text (direct, 4, 0, TAB_RIGHT | TAT_TITLE, _("Asymp. Std. Error"));
- tab_text (direct, 5, 0, TAB_RIGHT | TAT_TITLE, _("Approx. T"));
- tab_text (direct, 6, 0, TAB_RIGHT | TAT_TITLE, _("Approx. Sig."));
- tab_offset (direct, 0, 1);
- }
- else
- direct = NULL;
+static struct pivot_table *create_crosstab_table (
+ struct crosstabs_proc *, struct crosstabulation *,
+ size_t crs_leaves[CRS_CL_count]);
+static struct pivot_table *create_chisq_table (struct crosstabulation *);
+static struct pivot_table *create_sym_table (struct crosstabulation *);
+static struct pivot_table *create_risk_table (
+ struct crosstabulation *, struct pivot_dimension **risk_statistics);
+static struct pivot_table *create_direct_table (struct crosstabulation *);
+static void display_crosstabulation (struct crosstabs_proc *,
+ struct crosstabulation *,
+ struct pivot_table *,
+ size_t crs_leaves[CRS_CL_count]);
+static void display_chisq (struct crosstabulation *, struct pivot_table *);
+static void display_symmetric (struct crosstabs_proc *,
+ struct crosstabulation *, struct pivot_table *);
+static void display_risk (struct crosstabulation *, struct pivot_table *,
+ struct pivot_dimension *risk_statistics);
+static void display_directional (struct crosstabs_proc *,
+ struct crosstabulation *,
+ struct pivot_table *);
+static void delete_missing (struct crosstabulation *);
+static void build_matrix (struct crosstabulation *);
+
+/* Output pivot table XT in the context of PROC. */
+static void
+output_crosstabulation (struct crosstabs_proc *proc, struct crosstabulation *xt)
+{
+ for (size_t i = 0; i < xt->n_vars; i++)
+ enum_var_values (xt, i, proc->descending);
- for (;;)
+ if (xt->vars[COL_VAR].n_values == 0)
{
- /* Find pivot subtable if applicable. */
- te = find_pivot_extent (tb, &tc, 0);
- if (te == NULL)
- break;
-
- /* Find all the row variable values. */
- enum_var_values (tb, te - tb, ROW_VAR, &rows, &n_rows);
-
- /* Allocate memory space for the column and row totals. */
- if (n_rows > *maxrows)
- {
- *row_totp = xnrealloc (*row_totp, n_rows, sizeof **row_totp);
- row_tot = *row_totp;
- *maxrows = n_rows;
- }
- if (n_cols > *maxcols)
- {
- *col_totp = xnrealloc (*col_totp, n_cols, sizeof **col_totp);
- col_tot = *col_totp;
- *maxcols = n_cols;
- }
-
- /* Allocate table space for the matrix. */
- if (table && tab_row (table) + (n_rows + 1) * num_cells > tab_nr (table))
- tab_realloc (table, -1,
- MAX (tab_nr (table) + (n_rows + 1) * num_cells,
- tab_nr (table) * (pe - pb) / (te - tb)));
-
- if (mode == GENERAL)
- {
- /* Allocate memory space for the matrix. */
- if (n_cols * n_rows > *maxcells)
- {
- *matp = xnrealloc (*matp, n_cols * n_rows, sizeof **matp);
- *maxcells = n_cols * n_rows;
- }
-
- mat = *matp;
-
- /* Build the matrix and calculate column totals. */
- {
- union value *cur_col = cols;
- union value *cur_row = rows;
- double *mp = mat;
- double *cp = col_tot;
- struct table_entry **p;
-
- *cp = 0.;
- for (p = &tb[0]; p < te; p++)
- {
- for (; memcmp (cur_col, &(*p)->values[COL_VAR], sizeof *cur_col);
- cur_row = rows)
- {
- *++cp = 0.;
- for (; cur_row < &rows[n_rows]; cur_row++)
- {
- *mp = 0.;
- mp += n_cols;
- }
- cur_col++;
- mp = &mat[cur_col - cols];
- }
-
- for (; memcmp (cur_row, &(*p)->values[ROW_VAR], sizeof *cur_row);
- cur_row++)
- {
- *mp = 0.;
- mp += n_cols;
- }
-
- *cp += *mp = (*p)->u.freq;
- mp += n_cols;
- cur_row++;
- }
-
- /* Zero out the rest of the matrix. */
- for (; cur_row < &rows[n_rows]; cur_row++)
- {
- *mp = 0.;
- mp += n_cols;
- }
- cur_col++;
- if (cur_col < &cols[n_cols])
- {
- const int rem_cols = n_cols - (cur_col - cols);
- int c, r;
-
- for (c = 0; c < rem_cols; c++)
- *++cp = 0.;
- mp = &mat[cur_col - cols];
- for (r = 0; r < n_rows; r++)
- {
- for (c = 0; c < rem_cols; c++)
- *mp++ = 0.;
- mp += n_cols - rem_cols;
- }
- }
- }
- }
- else
- {
- int r, c;
- double *tp = col_tot;
-
- assert (mode == INTEGER);
- mat = (*tb)->u.data;
- ns_cols = n_cols;
-
- /* Calculate column totals. */
- for (c = 0; c < n_cols; c++)
- {
- double cum = 0.;
- double *cp = &mat[c];
-
- for (r = 0; r < n_rows; r++)
- cum += cp[r * n_cols];
- *tp++ = cum;
- }
- }
-
- {
- double *cp;
-
- for (ns_cols = 0, cp = col_tot; cp < &col_tot[n_cols]; cp++)
- ns_cols += *cp != 0.;
- }
-
- /* Calculate row totals. */
- {
- double *mp = mat;
- double *rp = row_tot;
- int r, c;
+ struct string vars;
+ int i;
- for (ns_rows = 0, r = n_rows; r--; )
- {
- double cum = 0.;
- for (c = n_cols; c--; )
- cum += *mp++;
- *rp++ = cum;
- if (cum != 0.)
- ns_rows++;
- }
- }
+ ds_init_cstr (&vars, var_to_string (xt->vars[0].var));
+ for (i = 1; i < xt->n_vars; i++)
+ ds_put_format (&vars, " × %s", var_to_string (xt->vars[i].var));
- /* Calculate grand total. */
- {
- double *tp;
- double cum = 0.;
- int n;
+ /* TRANSLATORS: The %s here describes a crosstabulation. It takes the
+ form "var1 * var2 * var3 * ...". */
+ msg (SW, _("Crosstabulation %s contained no non-missing cases."),
+ ds_cstr (&vars));
- if (n_rows < n_cols)
- tp = row_tot, n = n_rows;
- else
- tp = col_tot, n = n_cols;
- while (n--)
- cum += *tp++;
- W = cum;
- }
+ ds_destroy (&vars);
+ for (size_t i = 0; i < xt->n_vars; i++)
+ free_var_values (xt, i);
+ return;
+ }
- /* Find the first variable that differs from the last subtable,
- then display the values of the dimensioning variables for
- each table that needs it. */
- {
- int first_difference = nvar - 1;
+ size_t crs_leaves[CRS_CL_count];
+ struct pivot_table *table = (proc->cells
+ ? create_crosstab_table (proc, xt, crs_leaves)
+ : NULL);
+ struct pivot_table *chisq = (proc->statistics & (1u << CRS_ST_CHISQ)
+ ? create_chisq_table (xt)
+ : NULL);
+ struct pivot_table *sym
+ = (proc->statistics & ((1u << CRS_ST_PHI) | (1u << CRS_ST_CC)
+ | (1u << CRS_ST_BTAU) | (1u << CRS_ST_CTAU)
+ | (1u << CRS_ST_GAMMA) | (1u << CRS_ST_CORR)
+ | (1u << CRS_ST_KAPPA))
+ ? create_sym_table (xt)
+ : NULL);
+ struct pivot_dimension *risk_statistics = NULL;
+ struct pivot_table *risk = (proc->statistics & (1u << CRS_ST_RISK)
+ ? create_risk_table (xt, &risk_statistics)
+ : NULL);
+ struct pivot_table *direct
+ = (proc->statistics & ((1u << CRS_ST_LAMBDA) | (1u << CRS_ST_UC)
+ | (1u << CRS_ST_D) | (1u << CRS_ST_ETA))
+ ? create_direct_table (xt)
+ : NULL);
+
+ size_t row0 = 0;
+ size_t row1 = 0;
+ while (find_crosstab (xt, &row0, &row1))
+ {
+ struct crosstabulation x;
+
+ make_crosstabulation_subset (xt, row0, row1, &x);
+
+ size_t n_rows = x.vars[ROW_VAR].n_values;
+ size_t n_cols = x.vars[COL_VAR].n_values;
+ if (size_overflow_p (xtimes (xtimes (n_rows, n_cols), sizeof (double))))
+ xalloc_die ();
+ x.row_tot = xmalloc (n_rows * sizeof *x.row_tot);
+ x.col_tot = xmalloc (n_cols * sizeof *x.col_tot);
+ x.mat = xmalloc (n_rows * n_cols * sizeof *x.mat);
+
+ build_matrix (&x);
+
+ /* Find the first variable that differs from the last subtable. */
+ if (table)
+ display_crosstabulation (proc, &x, table, crs_leaves);
- if (tb != pb)
- for (; ; first_difference--)
- {
- assert (first_difference >= 2);
- if (memcmp (&cmp->values[first_difference],
- &(*tb)->values[first_difference],
- sizeof *cmp->values))
- break;
- }
- cmp = *tb;
-
- if (table)
- display_dimensions (table, first_difference, *tb);
- if (chisq)
- display_dimensions (chisq, first_difference, *tb);
- if (sym)
- display_dimensions (sym, first_difference, *tb);
- if (risk)
- display_dimensions (risk, first_difference, *tb);
- if (direct)
- display_dimensions (direct, first_difference, *tb);
- }
+ if (proc->exclude == MV_NEVER)
+ delete_missing (&x);
- if (table)
- display_crosstabulation ();
- if (cmd.miss == CRS_REPORT)
- delete_missing ();
if (chisq)
- display_chisq (dict);
+ display_chisq (&x, chisq);
+
if (sym)
- display_symmetric (dict);
+ display_symmetric (proc, &x, sym);
if (risk)
- display_risk (dict);
+ display_risk (&x, risk, risk_statistics);
if (direct)
- display_directional ();
+ display_directional (proc, &x, direct);
- tb = te;
- free (rows);
+ free (x.mat);
+ free (x.row_tot);
+ free (x.col_tot);
+ free (x.const_indexes);
}
- submit (table);
+ if (table)
+ pivot_table_submit (table);
if (chisq)
+ pivot_table_submit (chisq);
+
+ if (sym)
+ pivot_table_submit (sym);
+
+ if (risk)
{
- if (!chisq_fisher)
- tab_resize (chisq, 4 + (nvar - 2), -1);
- submit (chisq);
+ if (!pivot_table_is_empty (risk))
+ pivot_table_submit (risk);
+ else
+ pivot_table_unref (risk);
}
- submit (sym);
- submit (risk);
- submit (direct);
+ if (direct)
+ pivot_table_submit (direct);
- free (cols);
+ for (size_t i = 0; i < xt->n_vars; i++)
+ free_var_values (xt, i);
}
-/* Delete missing rows and columns for statistical analysis when
- /MISSING=REPORT. */
static void
-delete_missing (void)
+build_matrix (struct crosstabulation *x)
{
- {
- int r;
-
- for (r = 0; r < n_rows; r++)
- if (var_is_num_missing (x->vars[ROW_VAR], rows[r].f, MV_USER))
- {
- int c;
+ const int col_var_width = var_get_width (x->vars[COL_VAR].var);
+ const int row_var_width = var_get_width (x->vars[ROW_VAR].var);
+ size_t n_rows = x->vars[ROW_VAR].n_values;
+ size_t n_cols = x->vars[COL_VAR].n_values;
+ int col, row;
+ double *mp;
+ struct freq **p;
+
+ mp = x->mat;
+ col = row = 0;
+ for (p = x->entries; p < &x->entries[x->n_entries]; p++)
+ {
+ const struct freq *te = *p;
+
+ while (!value_equal (&x->vars[ROW_VAR].values[row],
+ &te->values[ROW_VAR], row_var_width))
+ {
+ for (; col < n_cols; col++)
+ *mp++ = 0.0;
+ col = 0;
+ row++;
+ }
- for (c = 0; c < n_cols; c++)
- mat[c + r * n_cols] = 0.;
- ns_rows--;
- }
- }
+ while (!value_equal (&x->vars[COL_VAR].values[col],
+ &te->values[COL_VAR], col_var_width))
+ {
+ *mp++ = 0.0;
+ col++;
+ }
- {
- int c;
+ *mp++ = te->count;
+ if (++col >= n_cols)
+ {
+ col = 0;
+ row++;
+ }
+ }
+ while (mp < &x->mat[n_cols * n_rows])
+ *mp++ = 0.0;
+ assert (mp == &x->mat[n_cols * n_rows]);
+
+ /* Column totals, row totals, ns_rows. */
+ mp = x->mat;
+ for (col = 0; col < n_cols; col++)
+ x->col_tot[col] = 0.0;
+ for (row = 0; row < n_rows; row++)
+ x->row_tot[row] = 0.0;
+ x->ns_rows = 0;
+ for (row = 0; row < n_rows; row++)
+ {
+ bool row_is_empty = true;
+ for (col = 0; col < n_cols; col++)
+ {
+ if (*mp != 0.0)
+ {
+ row_is_empty = false;
+ x->col_tot[col] += *mp;
+ x->row_tot[row] += *mp;
+ }
+ mp++;
+ }
+ if (!row_is_empty)
+ x->ns_rows++;
+ }
+ assert (mp == &x->mat[n_cols * n_rows]);
- for (c = 0; c < n_cols; c++)
- if (var_is_num_missing (x->vars[COL_VAR], cols[c].f, MV_USER))
- {
- int r;
+ /* ns_cols. */
+ x->ns_cols = 0;
+ for (col = 0; col < n_cols; col++)
+ for (row = 0; row < n_rows; row++)
+ if (x->mat[col + row * n_cols] != 0.0)
+ {
+ x->ns_cols++;
+ break;
+ }
- for (r = 0; r < n_rows; r++)
- mat[c + r * n_cols] = 0.;
- ns_cols--;
- }
- }
+ /* Grand total. */
+ x->total = 0.0;
+ for (col = 0; col < n_cols; col++)
+ x->total += x->col_tot[col];
}
-/* Prepare table T for submission, and submit it. */
static void
-submit (struct tab_table *t)
+add_var_dimension (struct pivot_table *table, const struct xtab_var *var,
+ enum pivot_axis_type axis_type, bool total)
{
- int i;
+ struct pivot_dimension *d = pivot_dimension_create__ (
+ table, axis_type, pivot_value_new_variable (var->var));
- if (t == NULL)
- return;
+ struct pivot_footnote *missing_footnote = pivot_table_create_footnote (
+ table, pivot_value_new_text (N_("Missing value")));
- tab_resize (t, -1, 0);
- if (tab_nr (t) == tab_t (t))
+ struct pivot_category *group = pivot_category_create_group__ (
+ d->root, pivot_value_new_variable (var->var));
+ for (size_t j = 0; j < var->n_values; j++)
{
- tab_destroy (t);
- return;
+ struct pivot_value *value = pivot_value_new_var_value (
+ var->var, &var->values[j]);
+ if (var_is_value_missing (var->var, &var->values[j], MV_ANY))
+ pivot_value_add_footnote (value, missing_footnote);
+ pivot_category_create_leaf (group, value);
}
- tab_offset (t, 0, 0);
- if (t != table)
- for (i = 2; i < nvar; i++)
- tab_text (t, nvar - i - 1, 0, TAB_RIGHT | TAT_TITLE,
- var_to_string (x->vars[i]));
- tab_box (t, TAL_2, TAL_2, -1, -1, 0, 0, tab_nc (t) - 1, tab_nr (t) - 1);
- tab_box (t, -1, -1, -1, TAL_1, tab_l (t), tab_t (t) - 1, tab_nc (t) - 1,
- tab_nr (t) - 1);
- tab_box (t, -1, -1, -1, TAL_GAP, 0, tab_t (t), tab_l (t) - 1,
- tab_nr (t) - 1);
- tab_vline (t, TAL_2, tab_l (t), 0, tab_nr (t) - 1);
- tab_dim (t, crosstabs_dim);
- tab_submit (t);
+
+ if (total)
+ pivot_category_create_leaf (d->root, pivot_value_new_text (N_("Total")));
}
-/* Sets the widths of all the columns and heights of all the rows in
- table T for driver D. */
-static void
-crosstabs_dim (struct tab_table *t, struct outp_driver *d)
+static struct pivot_table *
+create_crosstab_table (struct crosstabs_proc *proc, struct crosstabulation *xt,
+ size_t crs_leaves[CRS_CL_count])
{
- int i;
-
- /* Width of a numerical column. */
- int c = outp_string_width (d, "0.000000", OUTP_PROPORTIONAL);
- if (cmd.miss == CRS_REPORT)
- c += outp_string_width (d, "M", OUTP_PROPORTIONAL);
-
- /* Set width for header columns. */
- if (t->l != 0)
+ /* Title. */
+ struct string title = DS_EMPTY_INITIALIZER;
+ for (size_t i = 0; i < xt->n_vars; i++)
{
- size_t i;
- int w;
-
- w = d->width - c * (t->nc - t->l);
- for (i = 0; i <= t->nc; i++)
- w -= t->wrv[i];
- w /= t->l;
-
- if (w < d->prop_em_width * 8)
- w = d->prop_em_width * 8;
+ if (i)
+ ds_put_cstr (&title, " × ");
+ ds_put_cstr (&title, var_to_string (xt->vars[i].var));
+ }
+ for (size_t i = 0; i < xt->n_consts; i++)
+ {
+ const struct variable *var = xt->const_vars[i].var;
+ const union value *value = &xt->entries[0]->values[2 + i];
+ char *s;
- if (w > d->prop_em_width * 15)
- w = d->prop_em_width * 15;
+ ds_put_format (&title, ", %s=", var_to_string (var));
- for (i = 0; i < t->l; i++)
- t->w[i] = w;
+ /* Insert the formatted value of VAR without any leading spaces. */
+ s = data_out (value, var_get_encoding (var), var_get_print_format (var));
+ ds_put_cstr (&title, s + strspn (s, " "));
+ free (s);
}
+ struct pivot_table *table = pivot_table_create__ (
+ pivot_value_new_user_text_nocopy (ds_steal_cstr (&title)));
+ pivot_table_set_weight_format (table, &proc->weight_format);
+ table->omit_empty = true;
- for (i = t->l; i < t->nc; i++)
- t->w[i] = c;
+ struct pivot_dimension *statistics = pivot_dimension_create (
+ table, PIVOT_AXIS_ROW, N_("Statistics"));
- for (i = 0; i < t->nr; i++)
- t->h[i] = tab_natural_height (t, d, i);
+ struct statistic
+ {
+ const char *label;
+ const char *rc;
+ };
+ static const struct statistic stats[CRS_CL_count] =
+ {
+ [CRS_CL_COUNT] = { N_("Count"), PIVOT_RC_COUNT },
+ [CRS_CL_ROW] = { N_("Row %"), PIVOT_RC_PERCENT },
+ [CRS_CL_COLUMN] = { N_("Column %"), PIVOT_RC_PERCENT },
+ [CRS_CL_TOTAL] = { N_("Total %"), PIVOT_RC_PERCENT },
+ [CRS_CL_EXPECTED] = { N_("Expected"), PIVOT_RC_OTHER },
+ [CRS_CL_RESIDUAL] = { N_("Residual"), PIVOT_RC_RESIDUAL },
+ [CRS_CL_SRESIDUAL] = { N_("Std. Residual"), PIVOT_RC_RESIDUAL },
+ [CRS_CL_ASRESIDUAL] = { N_("Adjusted Residual"), PIVOT_RC_RESIDUAL },
+ };
+ for (size_t i = 0; i < CRS_CL_count; i++)
+ if (proc->cells & (1u << i) && stats[i].label)
+ crs_leaves[i] = pivot_category_create_leaf_rc (
+ statistics->root, pivot_value_new_text (stats[i].label),
+ stats[i].rc);
+
+ for (size_t i = 0; i < xt->n_vars; i++)
+ add_var_dimension (table, &xt->vars[i],
+ i == COL_VAR ? PIVOT_AXIS_COLUMN : PIVOT_AXIS_ROW,
+ true);
+
+ return table;
}
-static struct table_entry **find_pivot_extent_general (struct table_entry **tp,
- int *cnt, int pivot);
-static struct table_entry **find_pivot_extent_integer (struct table_entry **tp,
- int *cnt, int pivot);
-
-/* Calls find_pivot_extent_general or find_pivot_extent_integer, as
- appropriate. */
-static struct table_entry **
-find_pivot_extent (struct table_entry **tp, int *cnt, int pivot)
+static struct pivot_table *
+create_chisq_table (struct crosstabulation *xt)
{
- return (mode == GENERAL
- ? find_pivot_extent_general (tp, cnt, pivot)
- : find_pivot_extent_integer (tp, cnt, pivot));
+ struct pivot_table *chisq = pivot_table_create (N_("Chi-Square Tests"));
+ pivot_table_set_weight_format (chisq, &xt->weight_format);
+ chisq->omit_empty = true;
+
+ pivot_dimension_create (
+ chisq, PIVOT_AXIS_ROW, N_("Statistics"),
+ N_("Pearson Chi-Square"),
+ N_("Likelihood Ratio"),
+ N_("Fisher's Exact Test"),
+ N_("Continuity Correction"),
+ N_("Linear-by-Linear Association"),
+ N_("N of Valid Cases"), PIVOT_RC_COUNT);
+
+ pivot_dimension_create (
+ chisq, PIVOT_AXIS_COLUMN, N_("Statistics"),
+ N_("Value"), PIVOT_RC_OTHER,
+ N_("df"), PIVOT_RC_COUNT,
+ N_("Asymptotic Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE,
+ N_("Exact Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE,
+ N_("Exact Sig. (1-tailed)"), PIVOT_RC_SIGNIFICANCE);
+
+ for (size_t i = 2; i < xt->n_vars; i++)
+ add_var_dimension (chisq, &xt->vars[i], PIVOT_AXIS_ROW, false);
+
+ return chisq;
}
-/* Find the extent of a region in TP that contains one table. If
- PIVOT != 0 that means a set of table entries with identical table
- number; otherwise they also have to have the same values for every
- dimension after the row and column dimensions. The table that is
- searched starts at TP and has length CNT. Returns the first entry
- after the last one in the table; sets *CNT to the number of
- remaining values. If there are no entries in TP at all, returns
- NULL. A yucky interface, admittedly, but it works. */
-static struct table_entry **
-find_pivot_extent_general (struct table_entry **tp, int *cnt, int pivot)
+/* Symmetric measures. */
+static struct pivot_table *
+create_sym_table (struct crosstabulation *xt)
{
- struct table_entry *fp = *tp;
- struct crosstab *x;
-
- if (*cnt == 0)
- return NULL;
- x = xtab[(*tp)->table];
- for (;;)
- {
- tp++;
- if (--*cnt == 0)
- break;
- assert (*cnt > 0);
-
- if ((*tp)->table != fp->table)
- break;
- if (pivot)
- continue;
-
- if (memcmp (&(*tp)->values[2], &fp->values[2], sizeof (union value) * (x->nvar - 2)))
- break;
- }
-
- return tp;
+ struct pivot_table *sym = pivot_table_create (N_("Symmetric Measures"));
+ pivot_table_set_weight_format (sym, &xt->weight_format);
+ sym->omit_empty = true;
+
+ pivot_dimension_create (
+ sym, PIVOT_AXIS_COLUMN, N_("Values"),
+ N_("Value"), PIVOT_RC_OTHER,
+ N_("Asymp. Std. Error"), PIVOT_RC_OTHER,
+ N_("Approx. T"), PIVOT_RC_OTHER,
+ N_("Approx. Sig."), PIVOT_RC_SIGNIFICANCE);
+
+ struct pivot_dimension *statistics = pivot_dimension_create (
+ sym, PIVOT_AXIS_ROW, N_("Statistics"));
+ pivot_category_create_group (
+ statistics->root, N_("Nominal by Nominal"),
+ N_("Phi"), N_("Cramer's V"), N_("Contingency Coefficient"));
+ pivot_category_create_group (
+ statistics->root, N_("Ordinal by Ordinal"),
+ N_("Kendall's tau-b"), N_("Kendall's tau-c"),
+ N_("Gamma"), N_("Spearman Correlation"));
+ pivot_category_create_group (
+ statistics->root, N_("Interval by Interval"),
+ N_("Pearson's R"));
+ pivot_category_create_group (
+ statistics->root, N_("Measure of Agreement"),
+ N_("Kappa"));
+ pivot_category_create_leaves (statistics->root, N_("N of Valid Cases"),
+ PIVOT_RC_COUNT);
+
+ for (size_t i = 2; i < xt->n_vars; i++)
+ add_var_dimension (sym, &xt->vars[i], PIVOT_AXIS_ROW, false);
+
+ return sym;
}
-/* Integer mode correspondent to find_pivot_extent_general(). This
- could be optimized somewhat, but I just don't give a crap about
- CROSSTABS performance in integer mode, which is just a
- CROSSTABS wart as far as I'm concerned.
-
- That said, feel free to send optimization patches to me. */
-static struct table_entry **
-find_pivot_extent_integer (struct table_entry **tp, int *cnt, int pivot)
+/* Risk estimate. */
+static struct pivot_table *
+create_risk_table (struct crosstabulation *xt,
+ struct pivot_dimension **risk_statistics)
{
- struct table_entry *fp = *tp;
- struct crosstab *x;
+ struct pivot_table *risk = pivot_table_create (N_("Risk Estimate"));
+ pivot_table_set_weight_format (risk, &xt->weight_format);
+ risk->omit_empty = true;
- if (*cnt == 0)
- return NULL;
- x = xtab[(*tp)->table];
- for (;;)
- {
- tp++;
- if (--*cnt == 0)
- break;
- assert (*cnt > 0);
+ struct pivot_dimension *values = pivot_dimension_create (
+ risk, PIVOT_AXIS_COLUMN, N_("Values"),
+ N_("Value"), PIVOT_RC_OTHER);
+ pivot_category_create_group (
+ values->root, N_("95% Confidence Interval"),
+ N_("Lower"), PIVOT_RC_OTHER,
+ N_("Upper"), PIVOT_RC_OTHER);
- if ((*tp)->table != fp->table)
- break;
- if (pivot)
- continue;
+ *risk_statistics = pivot_dimension_create (
+ risk, PIVOT_AXIS_ROW, N_("Statistics"));
- if (memcmp (&(*tp)->values[2], &fp->values[2],
- sizeof (union value) * (x->nvar - 2)))
- break;
- }
+ for (size_t i = 2; i < xt->n_vars; i++)
+ add_var_dimension (risk, &xt->vars[i], PIVOT_AXIS_ROW, false);
- return tp;
+ return risk;
}
-/* Compares `union value's A_ and B_ and returns a strcmp()-like
- result. WIDTH_ points to an int which is either 0 for a
- numeric value or a string width for a string value. */
-static int
-compare_value (const void *a_, const void *b_, const void *width_)
+static void
+create_direct_stat (struct pivot_category *parent,
+ const struct crosstabulation *xt,
+ const char *name, bool symmetric)
{
- const union value *a = a_;
- const union value *b = b_;
- const int *pwidth = width_;
- const int width = *pwidth;
+ struct pivot_category *group = pivot_category_create_group (
+ parent, name);
+ if (symmetric)
+ pivot_category_create_leaf (group, pivot_value_new_text (N_("Symmetric")));
+
+ char *row_label = xasprintf (_("%s Dependent"),
+ var_to_string (xt->vars[ROW_VAR].var));
+ pivot_category_create_leaf (group, pivot_value_new_user_text_nocopy (
+ row_label));
+
+ char *col_label = xasprintf (_("%s Dependent"),
+ var_to_string (xt->vars[COL_VAR].var));
+ pivot_category_create_leaf (group, pivot_value_new_user_text_nocopy (
+ col_label));
+}
- if (width == 0)
- return (a->f < b->f) ? -1 : (a->f > b->f);
- else
- return strncmp (a->s, b->s, width);
+/* Directional measures. */
+static struct pivot_table *
+create_direct_table (struct crosstabulation *xt)
+{
+ struct pivot_table *direct = pivot_table_create (N_("Directional Measures"));
+ pivot_table_set_weight_format (direct, &xt->weight_format);
+ direct->omit_empty = true;
+
+ pivot_dimension_create (
+ direct, PIVOT_AXIS_COLUMN, N_("Values"),
+ N_("Value"), PIVOT_RC_OTHER,
+ N_("Asymp. Std. Error"), PIVOT_RC_OTHER,
+ N_("Approx. T"), PIVOT_RC_OTHER,
+ N_("Approx Sig."), PIVOT_RC_SIGNIFICANCE);
+
+ struct pivot_dimension *statistics = pivot_dimension_create (
+ direct, PIVOT_AXIS_ROW, N_("Statistics"));
+ struct pivot_category *nn = pivot_category_create_group (
+ statistics->root, N_("Nominal by Nominal"));
+ create_direct_stat (nn, xt, N_("Lambda"), true);
+ create_direct_stat (nn, xt, N_("Goodman and Kruskal tau"), false);
+ create_direct_stat (nn, xt, N_("Uncertainty Coefficient"), true);
+ struct pivot_category *oo = pivot_category_create_group (
+ statistics->root, N_("Ordinal by Ordinal"));
+ create_direct_stat (oo, xt, N_("Somers' d"), true);
+ struct pivot_category *ni = pivot_category_create_group (
+ statistics->root, N_("Nominal by Interval"));
+ create_direct_stat (ni, xt, N_("Eta"), false);
+
+ for (size_t i = 2; i < xt->n_vars; i++)
+ add_var_dimension (direct, &xt->vars[i], PIVOT_AXIS_ROW, false);
+
+ return direct;
}
-/* Given an array of ENTRY_CNT table_entry structures starting at
- ENTRIES, creates a sorted list of the values that the variable
- with index VAR_IDX takes on. The values are returned as a
- malloc()'darray stored in *VALUES, with the number of values
- stored in *VALUE_CNT.
- */
+/* Delete missing rows and columns for statistical analysis when
+ /MISSING=REPORT. */
static void
-enum_var_values (struct table_entry **entries, int entry_cnt, int var_idx,
- union value **values, int *value_cnt)
+delete_missing (struct crosstabulation *xt)
{
- const struct variable *v = xtab[(*entries)->table]->vars[var_idx];
+ size_t n_rows = xt->vars[ROW_VAR].n_values;
+ size_t n_cols = xt->vars[COL_VAR].n_values;
+ int r, c;
- if (mode == GENERAL)
- {
- int width = get_var_trimmed_width (v);
- int i;
+ for (r = 0; r < n_rows; r++)
+ if (var_is_num_missing (xt->vars[ROW_VAR].var,
+ xt->vars[ROW_VAR].values[r].f, MV_USER))
+ {
+ for (c = 0; c < n_cols; c++)
+ xt->mat[c + r * n_cols] = 0.;
+ xt->ns_rows--;
+ }
- *values = xnmalloc (entry_cnt, sizeof **values);
- for (i = 0; i < entry_cnt; i++)
- (*values)[i] = entries[i]->values[var_idx];
- *value_cnt = sort_unique (*values, entry_cnt, sizeof **values,
- compare_value, &width);
- }
- else
- {
- struct var_range *vr = get_var_range (v);
- int i;
- assert (mode == INTEGER);
- *values = xnmalloc (vr->count, sizeof **values);
- for (i = 0; i < vr->count; i++)
- (*values)[i].f = i + vr->min;
- *value_cnt = vr->count;
- }
+ for (c = 0; c < n_cols; c++)
+ if (var_is_num_missing (xt->vars[COL_VAR].var,
+ xt->vars[COL_VAR].values[c].f, MV_USER))
+ {
+ for (r = 0; r < n_rows; r++)
+ xt->mat[c + r * n_cols] = 0.;
+ xt->ns_cols--;
+ }
}
-/* Sets cell (C,R) in TABLE, with options OPT, to have a value taken
- from V, displayed with print format spec from variable VAR. When
- in REPORT missing-value mode, missing values have an M appended. */
-static void
-table_value_missing (struct tab_table *table, int c, int r, unsigned char opt,
- const union value *v, const struct variable *var)
+static bool
+find_crosstab (struct crosstabulation *xt, size_t *row0p, size_t *row1p)
{
- struct substring s;
- const struct fmt_spec *print = var_get_print_format (var);
+ size_t row0 = *row1p;
+ size_t row1;
- const char *label = var_lookup_value_label (var, v);
- if (label)
- {
- tab_text (table, c, r, TAB_LEFT, label);
- return;
- }
+ if (row0 >= xt->n_entries)
+ return false;
- s.string = tab_alloc (table, print->w);
- format_short (s.string, print, v);
- s.length = strlen (s.string);
- if (cmd.miss == CRS_REPORT && var_is_num_missing (var, v->f, MV_USER))
- s.string[s.length++] = 'M';
- while (s.length && *s.string == ' ')
+ for (row1 = row0 + 1; row1 < xt->n_entries; row1++)
{
- s.length--;
- s.string++;
+ struct freq *a = xt->entries[row0];
+ struct freq *b = xt->entries[row1];
+ if (compare_table_entry_vars_3way (a, b, xt, 2, xt->n_vars) != 0)
+ break;
}
- tab_raw (table, c, r, opt, &s);
+ *row0p = row0;
+ *row1p = row1;
+ return true;
}
-/* Draws a line across TABLE at the current row to indicate the most
- major dimension variable with index FIRST_DIFFERENCE out of NVAR
- that changed, and puts the values that changed into the table. TB
- and X must be the corresponding table_entry and crosstab,
- respectively. */
-static void
-display_dimensions (struct tab_table *table, int first_difference, struct table_entry *tb)
+/* Compares `union value's A_ and B_ and returns a strcmp()-like
+ result. WIDTH_ points to an int which is either 0 for a
+ numeric value or a string width for a string value. */
+static int
+compare_value_3way (const void *a_, const void *b_, const void *width_)
{
- tab_hline (table, TAL_1, nvar - first_difference - 1, tab_nc (table) - 1, 0);
+ const union value *a = a_;
+ const union value *b = b_;
+ const int *width = width_;
- for (; first_difference >= 2; first_difference--)
- table_value_missing (table, nvar - first_difference - 1, 0,
- TAB_RIGHT, &tb->values[first_difference],
- x->vars[first_difference]);
+ return value_compare_3way (a, b, *width);
}
-/* Put VALUE into cell (C,R) of TABLE, suffixed with character
- SUFFIX if nonzero. If MARK_MISSING is true the entry is
- additionally suffixed with a letter `M'. */
-static void
-format_cell_entry (struct tab_table *table, int c, int r, double value,
- char suffix, bool mark_missing)
+/* Inverted version of the above */
+static int
+compare_value_3way_inv (const void *a_, const void *b_, const void *width_)
{
- const struct fmt_spec f = {FMT_F, 10, 1};
- union value v;
- struct substring s;
-
- s.length = 10;
- s.string = tab_alloc (table, 16);
- v.f = value;
- data_out (&v, &f, s.string);
- while (*s.string == ' ')
- {
- s.length--;
- s.string++;
- }
- if (suffix != 0)
- s.string[s.length++] = suffix;
- if (mark_missing)
- s.string[s.length++] = 'M';
-
- tab_raw (table, c, r, TAB_RIGHT, &s);
+ return -compare_value_3way (a_, b_, width_);
}
-/* Displays the crosstabulation table. */
+
+/* Given an array of ENTRY_CNT table_entry structures starting at
+ ENTRIES, creates a sorted list of the values that the variable
+ with index VAR_IDX takes on. Stores the array of the values in
+ XT->values and the number of values in XT->n_values. */
static void
-display_crosstabulation (void)
+enum_var_values (const struct crosstabulation *xt, int var_idx,
+ bool descending)
{
- {
- int r;
-
- for (r = 0; r < n_rows; r++)
- table_value_missing (table, nvar - 2, r * num_cells,
- TAB_RIGHT, &rows[r], x->vars[ROW_VAR]);
- }
- tab_text (table, nvar - 2, n_rows * num_cells,
- TAB_LEFT, _("Total"));
-
- /* Put in the actual cells. */
- {
- double *mp = mat;
- int r, c, i;
-
- tab_offset (table, nvar - 1, -1);
- for (r = 0; r < n_rows; r++)
- {
- if (num_cells > 1)
- tab_hline (table, TAL_1, -1, n_cols, 0);
- for (c = 0; c < n_cols; c++)
- {
- bool mark_missing = false;
- double expected_value = row_tot[r] * col_tot[c] / W;
- if (cmd.miss == CRS_REPORT
- && (var_is_num_missing (x->vars[COL_VAR], cols[c].f, MV_USER)
- || var_is_num_missing (x->vars[ROW_VAR], rows[r].f,
- MV_USER)))
- mark_missing = true;
- for (i = 0; i < num_cells; i++)
- {
- double v;
- int suffix = 0;
-
- switch (cells[i])
- {
- case CRS_CL_COUNT:
- v = *mp;
- break;
- case CRS_CL_ROW:
- v = *mp / row_tot[r] * 100.;
- suffix = '%';
- break;
- case CRS_CL_COLUMN:
- v = *mp / col_tot[c] * 100.;
- suffix = '%';
- break;
- case CRS_CL_TOTAL:
- v = *mp / W * 100.;
- suffix = '%';
- break;
- case CRS_CL_EXPECTED:
- v = expected_value;
- break;
- case CRS_CL_RESIDUAL:
- v = *mp - expected_value;
- break;
- case CRS_CL_SRESIDUAL:
- v = (*mp - expected_value) / sqrt (expected_value);
- break;
- case CRS_CL_ASRESIDUAL:
- v = ((*mp - expected_value)
- / sqrt (expected_value
- * (1. - row_tot[r] / W)
- * (1. - col_tot[c] / W)));
- break;
- default:
- NOT_REACHED ();
- }
-
- format_cell_entry (table, c, i, v, suffix, mark_missing);
- }
-
- mp++;
- }
-
- tab_offset (table, -1, tab_row (table) + num_cells);
- }
- }
+ struct xtab_var *xv = &xt->vars[var_idx];
+ const struct var_range *range = get_var_range (xt->proc, xv->var);
- /* Row totals. */
- {
- int r, i;
+ if (range)
+ {
+ xv->values = xnmalloc (range->count, sizeof *xv->values);
+ xv->n_values = range->count;
+ for (size_t i = 0; i < range->count; i++)
+ xv->values[i].f = range->min + i;
+ }
+ else
+ {
+ int width = var_get_width (xv->var);
+ struct hmapx_node *node;
+ const union value *iter;
+ struct hmapx set;
- tab_offset (table, -1, tab_row (table) - num_cells * n_rows);
- for (r = 0; r < n_rows; r++)
- {
- bool mark_missing = false;
+ hmapx_init (&set);
+ for (size_t i = 0; i < xt->n_entries; i++)
+ {
+ const struct freq *te = xt->entries[i];
+ const union value *value = &te->values[var_idx];
+ size_t hash = value_hash (value, width, 0);
- if (cmd.miss == CRS_REPORT
- && var_is_num_missing (x->vars[ROW_VAR], rows[r].f, MV_USER))
- mark_missing = true;
+ HMAPX_FOR_EACH_WITH_HASH (iter, node, hash, &set)
+ if (value_equal (iter, value, width))
+ goto next_entry;
- for (i = 0; i < num_cells; i++)
- {
- char suffix = 0;
- double v;
+ hmapx_insert (&set, (union value *) value, hash);
- switch (cells[i])
- {
- case CRS_CL_COUNT:
- v = row_tot[r];
- break;
- case CRS_CL_ROW:
- v = 100.0;
- suffix = '%';
- break;
- case CRS_CL_COLUMN:
- v = row_tot[r] / W * 100.;
- suffix = '%';
- break;
- case CRS_CL_TOTAL:
- v = row_tot[r] / W * 100.;
- suffix = '%';
- break;
- case CRS_CL_EXPECTED:
- case CRS_CL_RESIDUAL:
- case CRS_CL_SRESIDUAL:
- case CRS_CL_ASRESIDUAL:
- v = 0.;
- break;
- default:
- NOT_REACHED ();
- }
+ next_entry: ;
+ }
- format_cell_entry (table, n_cols, 0, v, suffix, mark_missing);
- tab_next_row (table);
- }
- }
- }
+ xv->n_values = hmapx_count (&set);
+ xv->values = xnmalloc (xv->n_values, sizeof *xv->values);
+ size_t i = 0;
+ HMAPX_FOR_EACH (iter, node, &set)
+ xv->values[i++] = *iter;
+ hmapx_destroy (&set);
- /* Column totals, grand total. */
- {
- int c;
- int last_row = 0;
+ sort (xv->values, xv->n_values, sizeof *xv->values,
+ descending ? compare_value_3way_inv : compare_value_3way,
+ &width);
+ }
+}
- if (num_cells > 1)
- tab_hline (table, TAL_1, -1, n_cols, 0);
- for (c = 0; c <= n_cols; c++)
- {
- double ct = c < n_cols ? col_tot[c] : W;
- bool mark_missing = false;
- int i;
+static void
+free_var_values (const struct crosstabulation *xt, int var_idx)
+{
+ struct xtab_var *xv = &xt->vars[var_idx];
+ free (xv->values);
+ xv->values = NULL;
+ xv->n_values = 0;
+}
- if (cmd.miss == CRS_REPORT && c < n_cols
- && var_is_num_missing (x->vars[COL_VAR], cols[c].f, MV_USER))
- mark_missing = true;
+/* Displays the crosstabulation table. */
+static void
+display_crosstabulation (struct crosstabs_proc *proc,
+ struct crosstabulation *xt, struct pivot_table *table,
+ size_t crs_leaves[CRS_CL_count])
+{
+ size_t n_rows = xt->vars[ROW_VAR].n_values;
+ size_t n_cols = xt->vars[COL_VAR].n_values;
- for (i = 0; i < num_cells; i++)
- {
- char suffix = 0;
- double v;
+ size_t *indexes = xnmalloc (table->n_dimensions, sizeof *indexes);
+ assert (xt->n_vars == 2);
+ for (size_t i = 0; i < xt->n_consts; i++)
+ indexes[i + 3] = xt->const_indexes[i];
- switch (cells[i])
- {
- case CRS_CL_COUNT:
- v = ct;
- break;
- case CRS_CL_ROW:
- v = ct / W * 100.;
- suffix = '%';
- break;
- case CRS_CL_COLUMN:
- v = 100.;
- suffix = '%';
- break;
- case CRS_CL_TOTAL:
- v = ct / W * 100.;
- suffix = '%';
- break;
- case CRS_CL_EXPECTED:
- case CRS_CL_RESIDUAL:
- case CRS_CL_SRESIDUAL:
- case CRS_CL_ASRESIDUAL:
- continue;
- default:
- NOT_REACHED ();
- }
+ /* Put in the actual cells. */
+ double *mp = xt->mat;
+ for (size_t r = 0; r < n_rows; r++)
+ {
+ if (!xt->row_tot[r] && proc->mode != INTEGER)
+ continue;
- format_cell_entry (table, c, i, v, suffix, mark_missing);
- }
- last_row = i;
- }
+ indexes[ROW_VAR + 1] = r;
+ for (size_t c = 0; c < n_cols; c++)
+ {
+ if (!xt->col_tot[c] && proc->mode != INTEGER)
+ continue;
+
+ indexes[COL_VAR + 1] = c;
+
+ double expected_value = xt->row_tot[r] * xt->col_tot[c] / xt->total;
+ double residual = *mp - expected_value;
+ double sresidual = residual / sqrt (expected_value);
+ double asresidual = (sresidual
+ * (1. - xt->row_tot[r] / xt->total)
+ * (1. - xt->col_tot[c] / xt->total));
+ double entries[] = {
+ [CRS_CL_COUNT] = *mp,
+ [CRS_CL_ROW] = *mp / xt->row_tot[r] * 100.,
+ [CRS_CL_COLUMN] = *mp / xt->col_tot[c] * 100.,
+ [CRS_CL_TOTAL] = *mp / xt->total * 100.,
+ [CRS_CL_EXPECTED] = expected_value,
+ [CRS_CL_RESIDUAL] = residual,
+ [CRS_CL_SRESIDUAL] = sresidual,
+ [CRS_CL_ASRESIDUAL] = asresidual,
+ };
+ for (size_t i = 0; i < proc->n_cells; i++)
+ {
+ int cell = proc->a_cells[i];
+ indexes[0] = crs_leaves[cell];
+ pivot_table_put (table, indexes, table->n_dimensions,
+ pivot_value_new_number (entries[cell]));
+ }
+
+ mp++;
+ }
+ }
- tab_offset (table, -1, tab_row (table) + last_row);
- }
+ /* Row totals. */
+ for (size_t r = 0; r < n_rows; r++)
+ {
+ if (!xt->row_tot[r] && proc->mode != INTEGER)
+ continue;
+
+ double expected_value = xt->row_tot[r] / xt->total;
+ double entries[] = {
+ [CRS_CL_COUNT] = xt->row_tot[r],
+ [CRS_CL_ROW] = 100.0,
+ [CRS_CL_COLUMN] = expected_value * 100.,
+ [CRS_CL_TOTAL] = expected_value * 100.,
+ [CRS_CL_EXPECTED] = expected_value,
+ [CRS_CL_RESIDUAL] = SYSMIS,
+ [CRS_CL_SRESIDUAL] = SYSMIS,
+ [CRS_CL_ASRESIDUAL] = SYSMIS,
+ };
+ for (size_t i = 0; i < proc->n_cells; i++)
+ {
+ int cell = proc->a_cells[i];
+ double entry = entries[cell];
+ if (entry != SYSMIS)
+ {
+ indexes[ROW_VAR + 1] = r;
+ indexes[COL_VAR + 1] = n_cols;
+ indexes[0] = crs_leaves[cell];
+ pivot_table_put (table, indexes, table->n_dimensions,
+ pivot_value_new_number (entry));
+ }
+ }
+ }
- tab_offset (table, 0, -1);
+ for (size_t c = 0; c <= n_cols; c++)
+ {
+ if (c < n_cols && !xt->col_tot[c] && proc->mode != INTEGER)
+ continue;
+
+ double ct = c < n_cols ? xt->col_tot[c] : xt->total;
+ double expected_value = ct / xt->total;
+ double entries[] = {
+ [CRS_CL_COUNT] = ct,
+ [CRS_CL_ROW] = expected_value * 100.0,
+ [CRS_CL_COLUMN] = 100.0,
+ [CRS_CL_TOTAL] = expected_value * 100.,
+ [CRS_CL_EXPECTED] = expected_value,
+ [CRS_CL_RESIDUAL] = SYSMIS,
+ [CRS_CL_SRESIDUAL] = SYSMIS,
+ [CRS_CL_ASRESIDUAL] = SYSMIS,
+ };
+ for (size_t i = 0; i < proc->n_cells; i++)
+ {
+ int cell = proc->a_cells[i];
+ double entry = entries[cell];
+ if (entry != SYSMIS)
+ {
+ indexes[ROW_VAR + 1] = n_rows;
+ indexes[COL_VAR + 1] = c;
+ indexes[0] = crs_leaves[cell];
+ pivot_table_put (table, indexes, table->n_dimensions,
+ pivot_value_new_number (entry));
+ }
+ }
+ }
+
+ free (indexes);
}
-static void calc_r (double *X, double *Y, double *, double *, double *);
-static void calc_chisq (double[N_CHISQ], int[N_CHISQ], double *, double *);
+static void calc_r (struct crosstabulation *,
+ double *XT, double *Y, double *, double *, double *);
+static void calc_chisq (struct crosstabulation *,
+ double[N_CHISQ], int[N_CHISQ], double *, double *);
/* Display chi-square statistics. */
static void
-display_chisq (const struct dictionary *dict)
+display_chisq (struct crosstabulation *xt, struct pivot_table *chisq)
{
- const struct variable *wv = dict_get_weight (dict);
- const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
-
- static const char *chisq_stats[N_CHISQ] =
- {
- N_("Pearson Chi-Square"),
- N_("Likelihood Ratio"),
- N_("Fisher's Exact Test"),
- N_("Continuity Correction"),
- N_("Linear-by-Linear Association"),
- };
double chisq_v[N_CHISQ];
double fisher1, fisher2;
int df[N_CHISQ];
- int s = 0;
-
- int i;
-
- calc_chisq (chisq_v, df, &fisher1, &fisher2);
+ calc_chisq (xt, chisq_v, df, &fisher1, &fisher2);
- tab_offset (chisq, nvar - 2, -1);
-
- for (i = 0; i < N_CHISQ; i++)
+ size_t *indexes = xnmalloc (chisq->n_dimensions, sizeof *indexes);
+ assert (xt->n_vars == 2);
+ for (size_t i = 0; i < xt->n_consts; i++)
+ indexes[i + 2] = xt->const_indexes[i];
+ for (int i = 0; i < N_CHISQ; i++)
{
- if ((i != 2 && chisq_v[i] == SYSMIS)
- || (i == 2 && fisher1 == SYSMIS))
- continue;
- s = 1;
+ indexes[0] = i;
- tab_text (chisq, 0, 0, TAB_LEFT, gettext (chisq_stats[i]));
- if (i != 2)
- {
- tab_double (chisq, 1, 0, TAB_RIGHT, chisq_v[i], NULL);
- tab_double (chisq, 2, 0, TAB_RIGHT, df[i], wfmt);
- tab_double (chisq, 3, 0, TAB_RIGHT,
- gsl_cdf_chisq_Q (chisq_v[i], df[i]), NULL);
- }
- else
- {
- chisq_fisher = 1;
- tab_double (chisq, 4, 0, TAB_RIGHT, fisher2, NULL);
- tab_double (chisq, 5, 0, TAB_RIGHT, fisher1, NULL);
- }
- tab_next_row (chisq);
+ double entries[5] = { SYSMIS, SYSMIS, SYSMIS, SYSMIS, SYSMIS };
+ if (i == 2)
+ {
+ entries[3] = fisher2;
+ entries[4] = fisher1;
+ }
+ else if (chisq_v[i] != SYSMIS)
+ {
+ entries[0] = chisq_v[i];
+ entries[1] = df[i];
+ entries[2] = gsl_cdf_chisq_Q (chisq_v[i], df[i]);
+ }
+
+ for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
+ if (entries[j] != SYSMIS)
+ {
+ indexes[1] = j;
+ pivot_table_put (chisq, indexes, chisq->n_dimensions,
+ pivot_value_new_number (entries[j]));
+ }
}
- tab_text (chisq, 0, 0, TAB_LEFT, _("N of Valid Cases"));
- tab_double (chisq, 1, 0, TAB_RIGHT, W, wfmt);
- tab_next_row (chisq);
+ indexes[0] = 5;
+ indexes[1] = 0;
+ pivot_table_put (chisq, indexes, chisq->n_dimensions,
+ pivot_value_new_number (xt->total));
- tab_offset (chisq, 0, -1);
+ free (indexes);
}
-static int calc_symmetric (double[N_SYMMETRIC], double[N_SYMMETRIC],
- double[N_SYMMETRIC]);
+static int calc_symmetric (struct crosstabs_proc *, struct crosstabulation *,
+ double[N_SYMMETRIC], double[N_SYMMETRIC],
+ double[N_SYMMETRIC],
+ double[3], double[3], double[3]);
/* Display symmetric measures. */
static void
-display_symmetric (const struct dictionary *dict)
+display_symmetric (struct crosstabs_proc *proc, struct crosstabulation *xt,
+ struct pivot_table *sym)
{
- const struct variable *wv = dict_get_weight (dict);
- const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
-
- static const char *categories[] =
- {
- N_("Nominal by Nominal"),
- N_("Ordinal by Ordinal"),
- N_("Interval by Interval"),
- N_("Measure of Agreement"),
- };
-
- static const char *stats[N_SYMMETRIC] =
- {
- N_("Phi"),
- N_("Cramer's V"),
- N_("Contingency Coefficient"),
- N_("Kendall's tau-b"),
- N_("Kendall's tau-c"),
- N_("Gamma"),
- N_("Spearman Correlation"),
- N_("Pearson's R"),
- N_("Kappa"),
- };
-
- static const int stats_categories[N_SYMMETRIC] =
- {
- 0, 0, 0, 1, 1, 1, 1, 2, 3,
- };
-
- int last_cat = -1;
double sym_v[N_SYMMETRIC], sym_ase[N_SYMMETRIC], sym_t[N_SYMMETRIC];
- int i;
+ double somers_d_v[3], somers_d_ase[3], somers_d_t[3];
- if (!calc_symmetric (sym_v, sym_ase, sym_t))
+ if (!calc_symmetric (proc, xt, sym_v, sym_ase, sym_t,
+ somers_d_v, somers_d_ase, somers_d_t))
return;
- tab_offset (sym, nvar - 2, -1);
+ size_t *indexes = xnmalloc (sym->n_dimensions, sizeof *indexes);
+ assert (xt->n_vars == 2);
+ for (size_t i = 0; i < xt->n_consts; i++)
+ indexes[i + 2] = xt->const_indexes[i];
- for (i = 0; i < N_SYMMETRIC; i++)
+ for (int i = 0; i < N_SYMMETRIC; i++)
{
if (sym_v[i] == SYSMIS)
continue;
- if (stats_categories[i] != last_cat)
- {
- last_cat = stats_categories[i];
- tab_text (sym, 0, 0, TAB_LEFT, gettext (categories[last_cat]));
- }
+ indexes[1] = i;
- tab_text (sym, 1, 0, TAB_LEFT, gettext (stats[i]));
- tab_double (sym, 2, 0, TAB_RIGHT, sym_v[i], NULL);
- if (sym_ase[i] != SYSMIS)
- tab_double (sym, 3, 0, TAB_RIGHT, sym_ase[i], NULL);
- if (sym_t[i] != SYSMIS)
- tab_double (sym, 4, 0, TAB_RIGHT, sym_t[i], NULL);
- /*tab_double (sym, 5, 0, TAB_RIGHT, normal_sig (sym_v[i]), NULL);*/
- tab_next_row (sym);
+ double entries[] = { sym_v[i], sym_ase[i], sym_t[i] };
+ for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
+ if (entries[j] != SYSMIS)
+ {
+ indexes[0] = j;
+ pivot_table_put (sym, indexes, sym->n_dimensions,
+ pivot_value_new_number (entries[j]));
+ }
}
- tab_text (sym, 0, 0, TAB_LEFT, _("N of Valid Cases"));
- tab_double (sym, 2, 0, TAB_RIGHT, W, wfmt);
- tab_next_row (sym);
+ indexes[1] = N_SYMMETRIC;
+ indexes[0] = 0;
+ struct pivot_value *total = pivot_value_new_number (xt->total);
+ pivot_value_set_rc (sym, total, PIVOT_RC_COUNT);
+ pivot_table_put (sym, indexes, sym->n_dimensions, total);
- tab_offset (sym, 0, -1);
+ free (indexes);
}
-static int calc_risk (double[], double[], double[], union value *);
+static bool calc_risk (struct crosstabulation *,
+ double[], double[], double[], union value *,
+ double *);
/* Display risk estimate. */
static void
-display_risk (const struct dictionary *dict)
+display_risk (struct crosstabulation *xt, struct pivot_table *risk,
+ struct pivot_dimension *risk_statistics)
{
- const struct variable *wv = dict_get_weight (dict);
- const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
-
- char buf[256];
- double risk_v[3], lower[3], upper[3];
+ double risk_v[3], lower[3], upper[3], n_valid;
union value c[2];
- int i;
-
- if (!calc_risk (risk_v, upper, lower, c))
+ if (!calc_risk (xt, risk_v, upper, lower, c, &n_valid))
return;
- tab_offset (risk, nvar - 2, -1);
+ size_t *indexes = xnmalloc (risk->n_dimensions, sizeof *indexes);
+ assert (xt->n_vars == 2);
+ for (size_t i = 0; i < xt->n_consts; i++)
+ indexes[i + 2] = xt->const_indexes[i];
- for (i = 0; i < 3; i++)
+ for (int i = 0; i < 3; i++)
{
+ const struct variable *cv = xt->vars[COL_VAR].var;
+ const struct variable *rv = xt->vars[ROW_VAR].var;
+
if (risk_v[i] == SYSMIS)
continue;
+ struct string label = DS_EMPTY_INITIALIZER;
switch (i)
{
case 0:
- if (var_is_numeric (x->vars[COL_VAR]))
- sprintf (buf, _("Odds Ratio for %s (%g / %g)"),
- var_get_name (x->vars[COL_VAR]), c[0].f, c[1].f);
- else
- sprintf (buf, _("Odds Ratio for %s (%.*s / %.*s)"),
- var_get_name (x->vars[COL_VAR]),
- get_var_trimmed_width (x->vars[COL_VAR]), c[0].s,
- get_var_trimmed_width (x->vars[COL_VAR]), c[1].s);
+ ds_put_format (&label, _("Odds Ratio for %s"), var_to_string (rv));
+ ds_put_cstr (&label, " (");
+ var_append_value_name (rv, &c[0], &label);
+ ds_put_cstr (&label, " / ");
+ var_append_value_name (rv, &c[1], &label);
+ ds_put_cstr (&label, ")");
break;
case 1:
case 2:
- if (var_is_numeric (x->vars[ROW_VAR]))
- sprintf (buf, _("For cohort %s = %g"),
- var_get_name (x->vars[ROW_VAR]), rows[i - 1].f);
- else
- sprintf (buf, _("For cohort %s = %.*s"),
- var_get_name (x->vars[ROW_VAR]),
- get_var_trimmed_width (x->vars[ROW_VAR]), rows[i - 1].s);
+ ds_put_format (&label, _("For cohort %s = "), var_to_string (cv));
+ var_append_value_name (cv, &xt->vars[ROW_VAR].values[i - 1], &label);
break;
}
- tab_text (risk, 0, 0, TAB_LEFT, buf);
- tab_double (risk, 1, 0, TAB_RIGHT, risk_v[i], NULL);
- tab_double (risk, 2, 0, TAB_RIGHT, lower[i], NULL);
- tab_double (risk, 3, 0, TAB_RIGHT, upper[i], NULL);
- tab_next_row (risk);
- }
-
- tab_text (risk, 0, 0, TAB_LEFT, _("N of Valid Cases"));
- tab_double (risk, 1, 0, TAB_RIGHT, W, wfmt);
- tab_next_row (risk);
+ indexes[1] = pivot_category_create_leaf (
+ risk_statistics->root,
+ pivot_value_new_user_text_nocopy (ds_steal_cstr (&label)));
- tab_offset (risk, 0, -1);
+ double entries[] = { risk_v[i], lower[i], upper[i] };
+ for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
+ {
+ indexes[0] = j;
+ pivot_table_put (risk, indexes, risk->n_dimensions,
+ pivot_value_new_number (entries[i]));
+ }
+ }
+ indexes[1] = pivot_category_create_leaf (
+ risk_statistics->root,
+ pivot_value_new_text (N_("N of Valid Cases")));
+ indexes[0] = 0;
+ pivot_table_put (risk, indexes, risk->n_dimensions,
+ pivot_value_new_number (n_valid));
+ free (indexes);
}
-static int calc_directional (double[N_DIRECTIONAL], double[N_DIRECTIONAL],
- double[N_DIRECTIONAL]);
+static int calc_directional (struct crosstabs_proc *, struct crosstabulation *,
+ double[N_DIRECTIONAL], double[N_DIRECTIONAL],
+ double[N_DIRECTIONAL], double[N_DIRECTIONAL]);
/* Display directional measures. */
static void
-display_directional (void)
+display_directional (struct crosstabs_proc *proc,
+ struct crosstabulation *xt, struct pivot_table *direct)
{
- static const char *categories[] =
- {
- N_("Nominal by Nominal"),
- N_("Ordinal by Ordinal"),
- N_("Nominal by Interval"),
- };
-
- static const char *stats[] =
- {
- N_("Lambda"),
- N_("Goodman and Kruskal tau"),
- N_("Uncertainty Coefficient"),
- N_("Somers' d"),
- N_("Eta"),
- };
-
- static const char *types[] =
- {
- N_("Symmetric"),
- N_("%s Dependent"),
- N_("%s Dependent"),
- };
-
- static const int stats_categories[N_DIRECTIONAL] =
- {
- 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 2, 2,
- };
-
- static const int stats_stats[N_DIRECTIONAL] =
- {
- 0, 0, 0, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4,
- };
-
- static const int stats_types[N_DIRECTIONAL] =
- {
- 0, 1, 2, 1, 2, 0, 1, 2, 0, 1, 2, 1, 2,
- };
-
- static const int *stats_lookup[] =
- {
- stats_categories,
- stats_stats,
- stats_types,
- };
-
- static const char **stats_names[] =
- {
- categories,
- stats,
- types,
- };
-
- int last[3] =
- {
- -1, -1, -1,
- };
-
double direct_v[N_DIRECTIONAL];
double direct_ase[N_DIRECTIONAL];
double direct_t[N_DIRECTIONAL];
-
- int i;
-
- if (!calc_directional (direct_v, direct_ase, direct_t))
+ double sig[N_DIRECTIONAL];
+ if (!calc_directional (proc, xt, direct_v, direct_ase, direct_t, sig))
return;
- tab_offset (direct, nvar - 2, -1);
+ size_t *indexes = xnmalloc (direct->n_dimensions, sizeof *indexes);
+ assert (xt->n_vars == 2);
+ for (size_t i = 0; i < xt->n_consts; i++)
+ indexes[i + 2] = xt->const_indexes[i];
- for (i = 0; i < N_DIRECTIONAL; i++)
+ for (int i = 0; i < N_DIRECTIONAL; i++)
{
if (direct_v[i] == SYSMIS)
continue;
- {
- int j;
-
- for (j = 0; j < 3; j++)
- if (last[j] != stats_lookup[j][i])
- {
- if (j < 2)
- tab_hline (direct, TAL_1, j, 6, 0);
-
- for (; j < 3; j++)
- {
- const char *string;
- int k = last[j] = stats_lookup[j][i];
-
- if (k == 0)
- string = NULL;
- else if (k == 1)
- string = var_get_name (x->vars[0]);
- else
- string = var_get_name (x->vars[1]);
-
- tab_text (direct, j, 0, TAB_LEFT | TAT_PRINTF,
- gettext (stats_names[j][k]), string);
- }
- }
- }
+ indexes[1] = i;
- tab_double (direct, 3, 0, TAB_RIGHT, direct_v[i], NULL);
- if (direct_ase[i] != SYSMIS)
- tab_double (direct, 4, 0, TAB_RIGHT, direct_ase[i], NULL);
- if (direct_t[i] != SYSMIS)
- tab_double (direct, 5, 0, TAB_RIGHT, direct_t[i], NULL);
- /*tab_double (direct, 6, 0, TAB_RIGHT, normal_sig (direct_v[i]), NULL);*/
- tab_next_row (direct);
+ double entries[] = {
+ direct_v[i], direct_ase[i], direct_t[i], sig[i],
+ };
+ for (size_t j = 0; j < sizeof entries / sizeof *entries; j++)
+ if (entries[j] != SYSMIS)
+ {
+ indexes[0] = j;
+ pivot_table_put (direct, indexes, direct->n_dimensions,
+ pivot_value_new_number (entries[j]));
+ }
}
- tab_offset (direct, 0, -1);
+ free (indexes);
}
\f
/* Statistical calculations. */
-/* Returns the value of the gamma (factorial) function for an integer
- argument X. */
+/* Returns the value of the logarithm of gamma (factorial) function for an integer
+ argument XT. */
static double
-gamma_int (double x)
+log_gamma_int (double xt)
{
- double r = 1;
+ double r = 0;
int i;
- for (i = 2; i < x; i++)
- r *= i;
+ for (i = 2; i < xt; i++)
+ r += log(i);
+
return r;
}
static inline double
Pr (int a, int b, int c, int d)
{
- return (gamma_int (a + b + 1.) / gamma_int (a + 1.)
- * gamma_int (c + d + 1.) / gamma_int (b + 1.)
- * gamma_int (a + c + 1.) / gamma_int (c + 1.)
- * gamma_int (b + d + 1.) / gamma_int (d + 1.)
- / gamma_int (a + b + c + d + 1.));
+ return exp (log_gamma_int (a + b + 1.) - log_gamma_int (a + 1.)
+ + log_gamma_int (c + d + 1.) - log_gamma_int (b + 1.)
+ + log_gamma_int (a + c + 1.) - log_gamma_int (c + 1.)
+ + log_gamma_int (b + d + 1.) - log_gamma_int (d + 1.)
+ - log_gamma_int (a + b + c + d + 1.));
}
/* Swap the contents of A and B. */
static void
calc_fisher (int a, int b, int c, int d, double *fisher1, double *fisher2)
{
- int x;
+ int xt;
+ double pn1;
if (MIN (c, d) < MIN (a, b))
swap (&a, &c), swap (&b, &d);
swap (&a, &c), swap (&b, &d);
}
- *fisher1 = 0.;
- for (x = 0; x <= a; x++)
- *fisher1 += Pr (a - x, b + x, c + x, d - x);
+ pn1 = Pr (a, b, c, d);
+ *fisher1 = pn1;
+ for (xt = 1; xt <= a; xt++)
+ {
+ *fisher1 += Pr (a - xt, b + xt, c + xt, d - xt);
+ }
*fisher2 = *fisher1;
- for (x = 1; x <= b; x++)
- *fisher2 += Pr (a + x, b - x, c - x, d + x);
+
+ for (xt = 1; xt <= b; xt++)
+ {
+ double p = Pr (a + xt, b - xt, c - xt, d + xt);
+ if (p < pn1)
+ *fisher2 += p;
+ }
}
/* Calculates chi-squares into CHISQ. MAT is a matrix with N_COLS
columns with values COLS and N_ROWS rows with values ROWS. Values
- in the matrix sum to W. */
+ in the matrix sum to xt->total. */
static void
-calc_chisq (double chisq[N_CHISQ], int df[N_CHISQ],
+calc_chisq (struct crosstabulation *xt,
+ double chisq[N_CHISQ], int df[N_CHISQ],
double *fisher1, double *fisher2)
{
- int r, c;
-
chisq[0] = chisq[1] = 0.;
chisq[2] = chisq[3] = chisq[4] = SYSMIS;
*fisher1 = *fisher2 = SYSMIS;
- df[0] = df[1] = (ns_cols - 1) * (ns_rows - 1);
+ df[0] = df[1] = (xt->ns_cols - 1) * (xt->ns_rows - 1);
- if (ns_rows <= 1 || ns_cols <= 1)
+ if (xt->ns_rows <= 1 || xt->ns_cols <= 1)
{
chisq[0] = chisq[1] = SYSMIS;
return;
}
- for (r = 0; r < n_rows; r++)
- for (c = 0; c < n_cols; c++)
+ size_t n_cols = xt->vars[COL_VAR].n_values;
+ FOR_EACH_POPULATED_ROW (r, xt)
+ FOR_EACH_POPULATED_COLUMN (c, xt)
{
- const double expected = row_tot[r] * col_tot[c] / W;
- const double freq = mat[n_cols * r + c];
- const double residual = freq - expected;
+ const double expected = xt->row_tot[r] * xt->col_tot[c] / xt->total;
+ const double freq = xt->mat[n_cols * r + c];
+ const double residual = freq - expected;
chisq[0] += residual * residual / expected;
- if (freq)
- chisq[1] += freq * log (expected / freq);
+ if (freq)
+ chisq[1] += freq * log (expected / freq);
}
if (chisq[0] == 0.)
chisq[1] = SYSMIS;
/* Calculate Yates and Fisher exact test. */
- if (ns_cols == 2 && ns_rows == 2)
+ if (xt->ns_cols == 2 && xt->ns_rows == 2)
{
double f11, f12, f21, f22;
{
int nz_cols[2];
- int i, j;
-
- for (i = j = 0; i < n_cols; i++)
- if (col_tot[i] != 0.)
- {
- nz_cols[j++] = i;
- if (j == 2)
- break;
- }
+ int j = 0;
+ FOR_EACH_POPULATED_COLUMN (c, xt)
+ {
+ nz_cols[j++] = c;
+ if (j == 2)
+ break;
+ }
assert (j == 2);
- f11 = mat[nz_cols[0]];
- f12 = mat[nz_cols[1]];
- f21 = mat[nz_cols[0] + n_cols];
- f22 = mat[nz_cols[1] + n_cols];
+ f11 = xt->mat[nz_cols[0]];
+ f12 = xt->mat[nz_cols[1]];
+ f21 = xt->mat[nz_cols[0] + n_cols];
+ f22 = xt->mat[nz_cols[1] + n_cols];
}
/* Yates. */
{
- const double x = fabs (f11 * f22 - f12 * f21) - 0.5 * W;
+ const double xt_ = fabs (f11 * f22 - f12 * f21) - 0.5 * xt->total;
- if (x > 0.)
- chisq[3] = (W * x * x
+ if (xt_ > 0.)
+ chisq[3] = (xt->total * pow2 (xt_)
/ (f11 + f12) / (f21 + f22)
/ (f11 + f21) / (f12 + f22));
else
}
/* Fisher. */
- if (f11 < 5. || f12 < 5. || f21 < 5. || f22 < 5.)
- calc_fisher (f11 + .5, f12 + .5, f21 + .5, f22 + .5, fisher1, fisher2);
+ calc_fisher (f11 + .5, f12 + .5, f21 + .5, f22 + .5, fisher1, fisher2);
}
/* Calculate Mantel-Haenszel. */
- if (var_is_numeric (x->vars[ROW_VAR]) && var_is_numeric (x->vars[COL_VAR]))
+ if (var_is_numeric (xt->vars[ROW_VAR].var)
+ && var_is_numeric (xt->vars[COL_VAR].var))
{
double r, ase_0, ase_1;
- calc_r ((double *) rows, (double *) cols, &r, &ase_0, &ase_1);
+ calc_r (xt, (double *) xt->vars[ROW_VAR].values,
+ (double *) xt->vars[COL_VAR].values,
+ &r, &ase_0, &ase_1);
- chisq[4] = (W - 1.) * r * r;
+ chisq[4] = (xt->total - 1.) * r * r;
df[4] = 1;
}
}
-/* Calculate the value of Pearson's r. r is stored into R, ase_1 into
- ASE_1, and ase_0 into ASE_0. The row and column values must be
- passed in X and Y. */
+/* Calculate the value of Pearson's r. r is stored into R, its T value into
+ T, and standard error into ERROR. The row and column values must be
+ passed in XT and Y. */
static void
-calc_r (double *X, double *Y, double *r, double *ase_0, double *ase_1)
+calc_r (struct crosstabulation *xt,
+ double *XT, double *Y, double *r, double *t, double *error)
{
+ size_t n_rows = xt->vars[ROW_VAR].n_values;
+ size_t n_cols = xt->vars[COL_VAR].n_values;
double SX, SY, S, T;
double Xbar, Ybar;
double sum_XYf, sum_X2Y2f;
for (sum_X2Y2f = sum_XYf = 0., i = 0; i < n_rows; i++)
for (j = 0; j < n_cols; j++)
{
- double fij = mat[j + i * n_cols];
- double product = X[i] * Y[j];
+ double fij = xt->mat[j + i * n_cols];
+ double product = XT[i] * Y[j];
double temp = fij * product;
sum_XYf += temp;
sum_X2Y2f += temp * product;
for (sum_Xr = sum_X2r = 0., i = 0; i < n_rows; i++)
{
- sum_Xr += X[i] * row_tot[i];
- sum_X2r += X[i] * X[i] * row_tot[i];
+ sum_Xr += XT[i] * xt->row_tot[i];
+ sum_X2r += pow2 (XT[i]) * xt->row_tot[i];
}
- Xbar = sum_Xr / W;
+ Xbar = sum_Xr / xt->total;
for (sum_Yc = sum_Y2c = 0., i = 0; i < n_cols; i++)
{
- sum_Yc += Y[i] * col_tot[i];
- sum_Y2c += Y[i] * Y[i] * col_tot[i];
+ sum_Yc += Y[i] * xt->col_tot[i];
+ sum_Y2c += Y[i] * Y[i] * xt->col_tot[i];
}
- Ybar = sum_Yc / W;
+ Ybar = sum_Yc / xt->total;
- S = sum_XYf - sum_Xr * sum_Yc / W;
- SX = sum_X2r - sum_Xr * sum_Xr / W;
- SY = sum_Y2c - sum_Yc * sum_Yc / W;
+ S = sum_XYf - sum_Xr * sum_Yc / xt->total;
+ SX = sum_X2r - pow2 (sum_Xr) / xt->total;
+ SY = sum_Y2c - pow2 (sum_Yc) / xt->total;
T = sqrt (SX * SY);
*r = S / T;
- *ase_0 = sqrt ((sum_X2Y2f - (sum_XYf * sum_XYf) / W) / (sum_X2r * sum_Y2c));
+ *t = *r / sqrt (1 - pow2 (*r)) * sqrt (xt->total - 2);
{
double s, c, y, t;
double Xresid, Yresid;
double temp;
- Xresid = X[i] - Xbar;
+ Xresid = XT[i] - Xbar;
Yresid = Y[j] - Ybar;
temp = (T * Xresid * Yresid
- ((S / (2. * T))
* (Xresid * Xresid * SY + Yresid * Yresid * SX)));
- y = mat[j + i * n_cols] * temp * temp - c;
+ y = xt->mat[j + i * n_cols] * temp * temp - c;
t = s + y;
c = (t - s) - y;
s = t;
}
- *ase_1 = sqrt (s) / (T * T);
+ *error = sqrt (s) / (T * T);
}
}
-static double somers_d_v[3];
-static double somers_d_ase[3];
-static double somers_d_t[3];
-
/* Calculate symmetric statistics and their asymptotic standard
errors. Returns 0 if none could be calculated. */
static int
-calc_symmetric (double v[N_SYMMETRIC], double ase[N_SYMMETRIC],
- double t[N_SYMMETRIC])
+calc_symmetric (struct crosstabs_proc *proc, struct crosstabulation *xt,
+ double v[N_SYMMETRIC], double ase[N_SYMMETRIC],
+ double t[N_SYMMETRIC],
+ double somers_d_v[3], double somers_d_ase[3],
+ double somers_d_t[3])
{
- int q = MIN (ns_rows, ns_cols);
+ size_t n_rows = xt->vars[ROW_VAR].n_values;
+ size_t n_cols = xt->vars[COL_VAR].n_values;
+ int q, i;
+ q = MIN (xt->ns_rows, xt->ns_cols);
if (q <= 1)
return 0;
- {
- int i;
-
- if (v)
- for (i = 0; i < N_SYMMETRIC; i++)
- v[i] = ase[i] = t[i] = SYSMIS;
- }
+ for (i = 0; i < N_SYMMETRIC; i++)
+ v[i] = ase[i] = t[i] = SYSMIS;
/* Phi, Cramer's V, contingency coefficient. */
- if (cmd.a_statistics[CRS_ST_PHI] || cmd.a_statistics[CRS_ST_CC])
+ if (proc->statistics & ((1u << CRS_ST_PHI) | (1u << CRS_ST_CC)))
{
double Xp = 0.; /* Pearson chi-square. */
- {
- int r, c;
-
- for (r = 0; r < n_rows; r++)
- for (c = 0; c < n_cols; c++)
- {
- const double expected = row_tot[r] * col_tot[c] / W;
- const double freq = mat[n_cols * r + c];
- const double residual = freq - expected;
+ FOR_EACH_POPULATED_ROW (r, xt)
+ FOR_EACH_POPULATED_COLUMN (c, xt)
+ {
+ double expected = xt->row_tot[r] * xt->col_tot[c] / xt->total;
+ double freq = xt->mat[n_cols * r + c];
+ double residual = freq - expected;
- Xp += residual * residual / expected;
- }
- }
+ Xp += residual * residual / expected;
+ }
- if (cmd.a_statistics[CRS_ST_PHI])
+ if (proc->statistics & (1u << CRS_ST_PHI))
{
- v[0] = sqrt (Xp / W);
- v[1] = sqrt (Xp / (W * (q - 1)));
+ v[0] = sqrt (Xp / xt->total);
+ v[1] = sqrt (Xp / (xt->total * (q - 1)));
}
- if (cmd.a_statistics[CRS_ST_CC])
- v[2] = sqrt (Xp / (Xp + W));
+ if (proc->statistics & (1u << CRS_ST_CC))
+ v[2] = sqrt (Xp / (Xp + xt->total));
}
- if (cmd.a_statistics[CRS_ST_BTAU] || cmd.a_statistics[CRS_ST_CTAU]
- || cmd.a_statistics[CRS_ST_GAMMA] || cmd.a_statistics[CRS_ST_D])
+ if (proc->statistics & ((1u << CRS_ST_BTAU) | (1u << CRS_ST_CTAU)
+ | (1u << CRS_ST_GAMMA) | (1u << CRS_ST_D)))
{
double *cum;
double Dr, Dc;
double P, Q;
double btau_cum, ctau_cum, gamma_cum, d_yx_cum, d_xy_cum;
double btau_var;
+ int r, c;
- {
- int r, c;
-
- Dr = Dc = W * W;
- for (r = 0; r < n_rows; r++)
- Dr -= row_tot[r] * row_tot[r];
- for (c = 0; c < n_cols; c++)
- Dc -= col_tot[c] * col_tot[c];
- }
-
- {
- int r, c;
+ Dr = Dc = pow2 (xt->total);
+ for (r = 0; r < n_rows; r++)
+ Dr -= pow2 (xt->row_tot[r]);
+ for (c = 0; c < n_cols; c++)
+ Dc -= pow2 (xt->col_tot[c]);
- cum = xnmalloc (n_cols * n_rows, sizeof *cum);
- for (c = 0; c < n_cols; c++)
- {
- double ct = 0.;
+ cum = xnmalloc (n_cols * n_rows, sizeof *cum);
+ for (c = 0; c < n_cols; c++)
+ {
+ double ct = 0.;
- for (r = 0; r < n_rows; r++)
- cum[c + r * n_cols] = ct += mat[c + r * n_cols];
- }
- }
+ for (r = 0; r < n_rows; r++)
+ cum[c + r * n_cols] = ct += xt->mat[c + r * n_cols];
+ }
/* P and Q. */
{
Cij = Dij = 0.;
for (j = 1; j < n_cols; j++)
- Cij += col_tot[j] - cum[j + i * n_cols];
+ Cij += xt->col_tot[j] - cum[j + i * n_cols];
if (i > 0)
for (j = 1; j < n_cols; j++)
for (j = 0;;)
{
- double fij = mat[j + i * n_cols];
+ double fij = xt->mat[j + i * n_cols];
P += fij * Cij;
Q += fij * Dij;
break;
assert (j < n_cols);
- Cij -= col_tot[j] - cum[j + i * n_cols];
- Dij += col_tot[j - 1] - cum[j - 1 + i * n_cols];
+ Cij -= xt->col_tot[j] - cum[j + i * n_cols];
+ Dij += xt->col_tot[j - 1] - cum[j - 1 + i * n_cols];
if (i > 0)
{
}
}
- if (cmd.a_statistics[CRS_ST_BTAU])
+ if (proc->statistics & (1u << CRS_ST_BTAU))
v[3] = (P - Q) / sqrt (Dr * Dc);
- if (cmd.a_statistics[CRS_ST_CTAU])
- v[4] = (q * (P - Q)) / ((W * W) * (q - 1));
- if (cmd.a_statistics[CRS_ST_GAMMA])
+ if (proc->statistics & (1u << CRS_ST_CTAU))
+ v[4] = (q * (P - Q)) / (pow2 (xt->total) * (q - 1));
+ if (proc->statistics & (1u << CRS_ST_GAMMA))
v[5] = (P - Q) / (P + Q);
/* ASE for tau-b, tau-c, gamma. Calculations could be
Cij = Dij = 0.;
for (j = 1; j < n_cols; j++)
- Cij += col_tot[j] - cum[j + i * n_cols];
+ Cij += xt->col_tot[j] - cum[j + i * n_cols];
if (i > 0)
for (j = 1; j < n_cols; j++)
for (j = 0;;)
{
- double fij = mat[j + i * n_cols];
+ double fij = xt->mat[j + i * n_cols];
- if (cmd.a_statistics[CRS_ST_BTAU])
+ if (proc->statistics & (1u << CRS_ST_BTAU))
{
const double temp = (2. * sqrt (Dr * Dc) * (Cij - Dij)
- + v[3] * (row_tot[i] * Dc
- + col_tot[j] * Dr));
+ + v[3] * (xt->row_tot[i] * Dc
+ + xt->col_tot[j] * Dr));
btau_cum += fij * temp * temp;
}
ctau_cum += fij * temp * temp;
}
- if (cmd.a_statistics[CRS_ST_GAMMA])
+ if (proc->statistics & (1u << CRS_ST_GAMMA))
{
const double temp = Q * Cij - P * Dij;
gamma_cum += fij * temp * temp;
}
- if (cmd.a_statistics[CRS_ST_D])
+ if (proc->statistics & (1u << CRS_ST_D))
{
d_yx_cum += fij * pow2 (Dr * (Cij - Dij)
- - (P - Q) * (W - row_tot[i]));
+ - (P - Q) * (xt->total - xt->row_tot[i]));
d_xy_cum += fij * pow2 (Dc * (Dij - Cij)
- - (Q - P) * (W - col_tot[j]));
+ - (Q - P) * (xt->total - xt->col_tot[j]));
}
if (++j == n_cols)
break;
assert (j < n_cols);
- Cij -= col_tot[j] - cum[j + i * n_cols];
- Dij += col_tot[j - 1] - cum[j - 1 + i * n_cols];
+ Cij -= xt->col_tot[j] - cum[j + i * n_cols];
+ Dij += xt->col_tot[j - 1] - cum[j - 1 + i * n_cols];
if (i > 0)
{
}
btau_var = ((btau_cum
- - (W * pow2 (W * (P - Q) / sqrt (Dr * Dc) * (Dr + Dc))))
+ - (xt->total * pow2 (xt->total * (P - Q) / sqrt (Dr * Dc) * (Dr + Dc))))
/ pow2 (Dr * Dc));
- if (cmd.a_statistics[CRS_ST_BTAU])
+ if (proc->statistics & (1u << CRS_ST_BTAU))
{
ase[3] = sqrt (btau_var);
- t[3] = v[3] / (2 * sqrt ((ctau_cum - (P - Q) * (P - Q) / W)
+ t[3] = v[3] / (2 * sqrt ((ctau_cum - (P - Q) * (P - Q) / xt->total)
/ (Dr * Dc)));
}
- if (cmd.a_statistics[CRS_ST_CTAU])
+ if (proc->statistics & (1u << CRS_ST_CTAU))
{
- ase[4] = ((2 * q / ((q - 1) * W * W))
- * sqrt (ctau_cum - (P - Q) * (P - Q) / W));
+ ase[4] = ((2 * q / ((q - 1) * pow2 (xt->total)))
+ * sqrt (ctau_cum - (P - Q) * (P - Q) / xt->total));
t[4] = v[4] / ase[4];
}
- if (cmd.a_statistics[CRS_ST_GAMMA])
+ if (proc->statistics & (1u << CRS_ST_GAMMA))
{
ase[5] = ((4. / ((P + Q) * (P + Q))) * sqrt (gamma_cum));
t[5] = v[5] / (2. / (P + Q)
- * sqrt (ctau_cum - (P - Q) * (P - Q) / W));
+ * sqrt (ctau_cum - (P - Q) * (P - Q) / xt->total));
}
- if (cmd.a_statistics[CRS_ST_D])
+ if (proc->statistics & (1u << CRS_ST_D))
{
somers_d_v[0] = (P - Q) / (.5 * (Dc + Dr));
- somers_d_ase[0] = 2. * btau_var / (Dr + Dc) * sqrt (Dr * Dc);
+ somers_d_ase[0] = SYSMIS;
somers_d_t[0] = (somers_d_v[0]
/ (4 / (Dc + Dr)
- * sqrt (ctau_cum - pow2 (P - Q) / W)));
+ * sqrt (ctau_cum - pow2 (P - Q) / xt->total)));
somers_d_v[1] = (P - Q) / Dc;
somers_d_ase[1] = 2. / pow2 (Dc) * sqrt (d_xy_cum);
somers_d_t[1] = (somers_d_v[1]
/ (2. / Dc
- * sqrt (ctau_cum - pow2 (P - Q) / W)));
+ * sqrt (ctau_cum - pow2 (P - Q) / xt->total)));
somers_d_v[2] = (P - Q) / Dr;
somers_d_ase[2] = 2. / pow2 (Dr) * sqrt (d_yx_cum);
somers_d_t[2] = (somers_d_v[2]
/ (2. / Dr
- * sqrt (ctau_cum - pow2 (P - Q) / W)));
+ * sqrt (ctau_cum - pow2 (P - Q) / xt->total)));
}
free (cum);
}
/* Spearman correlation, Pearson's r. */
- if (cmd.a_statistics[CRS_ST_CORR])
+ if (proc->statistics & (1u << CRS_ST_CORR))
{
- double *R = xmalloca (sizeof *R * n_rows);
- double *C = xmalloca (sizeof *C * n_cols);
+ double *R = xmalloc (sizeof *R * n_rows);
+ double *C = xmalloc (sizeof *C * n_cols);
{
double y, t, c = 0., s = 0.;
for (;;)
{
- R[i] = s + (row_tot[i] + 1.) / 2.;
- y = row_tot[i] - c;
+ R[i] = s + (xt->row_tot[i] + 1.) / 2.;
+ y = xt->row_tot[i] - c;
t = s + y;
c = (t - s) - y;
s = t;
for (;;)
{
- C[j] = s + (col_tot[j] + 1.) / 2;
- y = col_tot[j] - c;
+ C[j] = s + (xt->col_tot[j] + 1.) / 2;
+ y = xt->col_tot[j] - c;
t = s + y;
c = (t - s) - y;
s = t;
}
}
- calc_r (R, C, &v[6], &t[6], &ase[6]);
- t[6] = v[6] / t[6];
+ calc_r (xt, R, C, &v[6], &t[6], &ase[6]);
- freea (R);
- freea (C);
+ free (R);
+ free (C);
- calc_r ((double *) rows, (double *) cols, &v[7], &t[7], &ase[7]);
- t[7] = v[7] / t[7];
+ calc_r (xt, (double *) xt->vars[ROW_VAR].values,
+ (double *) xt->vars[COL_VAR].values,
+ &v[7], &t[7], &ase[7]);
}
/* Cohen's kappa. */
- if (cmd.a_statistics[CRS_ST_KAPPA] && ns_rows == ns_cols)
+ if (proc->statistics & (1u << CRS_ST_KAPPA) && xt->ns_rows == xt->ns_cols)
{
+ double ase_under_h0;
double sum_fii, sum_rici, sum_fiiri_ci, sum_fijri_ci2, sum_riciri_ci;
int i, j;
for (sum_fii = sum_rici = sum_fiiri_ci = sum_riciri_ci = 0., i = j = 0;
- i < ns_rows; i++, j++)
+ i < xt->ns_rows; i++, j++)
{
double prod, sum;
- while (col_tot[j] == 0.)
+ while (xt->col_tot[j] == 0.)
j++;
- prod = row_tot[i] * col_tot[j];
- sum = row_tot[i] + col_tot[j];
+ prod = xt->row_tot[i] * xt->col_tot[j];
+ sum = xt->row_tot[i] + xt->col_tot[j];
- sum_fii += mat[j + i * n_cols];
+ sum_fii += xt->mat[j + i * n_cols];
sum_rici += prod;
- sum_fiiri_ci += mat[j + i * n_cols] * sum;
+ sum_fiiri_ci += xt->mat[j + i * n_cols] * sum;
sum_riciri_ci += prod * sum;
}
- for (sum_fijri_ci2 = 0., i = 0; i < ns_rows; i++)
- for (j = 0; j < ns_cols; j++)
+ for (sum_fijri_ci2 = 0., i = 0; i < xt->ns_rows; i++)
+ for (j = 0; j < xt->ns_cols; j++)
{
- double sum = row_tot[i] + col_tot[j];
- sum_fijri_ci2 += mat[j + i * n_cols] * sum * sum;
+ double sum = xt->row_tot[i] + xt->col_tot[j];
+ sum_fijri_ci2 += xt->mat[j + i * n_cols] * sum * sum;
}
- v[8] = (W * sum_fii - sum_rici) / (W * W - sum_rici);
+ v[8] = (xt->total * sum_fii - sum_rici) / (pow2 (xt->total) - sum_rici);
- ase[8] = sqrt ((W * W * sum_rici
- + sum_rici * sum_rici
- - W * sum_riciri_ci)
- / (W * (W * W - sum_rici) * (W * W - sum_rici)));
-#if 0
- t[8] = v[8] / sqrt (W * (((sum_fii * (W - sum_fii))
- / pow2 (W * W - sum_rici))
- + ((2. * (W - sum_fii)
+ ase_under_h0 = sqrt ((pow2 (xt->total) * sum_rici
+ + sum_rici * sum_rici
+ - xt->total * sum_riciri_ci)
+ / (xt->total * (pow2 (xt->total) - sum_rici) * (pow2 (xt->total) - sum_rici)));
+
+ ase[8] = sqrt (xt->total * (((sum_fii * (xt->total - sum_fii))
+ / pow2 (pow2 (xt->total) - sum_rici))
+ + ((2. * (xt->total - sum_fii)
* (2. * sum_fii * sum_rici
- - W * sum_fiiri_ci))
- / cube (W * W - sum_rici))
- + (pow2 (W - sum_fii)
- * (W * sum_fijri_ci2 - 4.
+ - xt->total * sum_fiiri_ci))
+ / pow3 (pow2 (xt->total) - sum_rici))
+ + (pow2 (xt->total - sum_fii)
+ * (xt->total * sum_fijri_ci2 - 4.
* sum_rici * sum_rici)
- / pow4 (W * W - sum_rici))));
-#else
- t[8] = v[8] / ase[8];
-#endif
+ / pow4 (pow2 (xt->total) - sum_rici))));
+
+ t[8] = v[8] / ase_under_h0;
}
return 1;
}
/* Calculate risk estimate. */
-static int
-calc_risk (double *value, double *upper, double *lower, union value *c)
+static bool
+calc_risk (struct crosstabulation *xt,
+ double *value, double *upper, double *lower, union value *c,
+ double *n_valid)
{
+ size_t n_cols = xt->vars[COL_VAR].n_values;
double f11, f12, f21, f22;
double v;
- {
- int i;
-
- for (i = 0; i < 3; i++)
- value[i] = upper[i] = lower[i] = SYSMIS;
- }
+ for (int i = 0; i < 3; i++)
+ value[i] = upper[i] = lower[i] = SYSMIS;
- if (ns_rows != 2 || ns_cols != 2)
- return 0;
+ if (xt->ns_rows != 2 || xt->ns_cols != 2)
+ return false;
{
+ /* Find populated columns. */
int nz_cols[2];
- int i, j;
-
- for (i = j = 0; i < n_cols; i++)
- if (col_tot[i] != 0.)
- {
- nz_cols[j++] = i;
- if (j == 2)
- break;
- }
-
- assert (j == 2);
-
- f11 = mat[nz_cols[0]];
- f12 = mat[nz_cols[1]];
- f21 = mat[nz_cols[0] + n_cols];
- f22 = mat[nz_cols[1] + n_cols];
-
- c[0] = cols[nz_cols[0]];
- c[1] = cols[nz_cols[1]];
+ int n = 0;
+ FOR_EACH_POPULATED_COLUMN (c, xt)
+ nz_cols[n++] = c;
+ assert (n == 2);
+
+ /* Find populated rows. */
+ int nz_rows[2];
+ n = 0;
+ FOR_EACH_POPULATED_ROW (r, xt)
+ nz_rows[n++] = r;
+ assert (n == 2);
+
+ f11 = xt->mat[nz_cols[0] + n_cols * nz_rows[0]];
+ f12 = xt->mat[nz_cols[1] + n_cols * nz_rows[0]];
+ f21 = xt->mat[nz_cols[0] + n_cols * nz_rows[1]];
+ f22 = xt->mat[nz_cols[1] + n_cols * nz_rows[1]];
+ *n_valid = f11 + f12 + f21 + f22;
+
+ c[0] = xt->vars[COL_VAR].values[nz_cols[0]];
+ c[1] = xt->vars[COL_VAR].values[nz_cols[1]];
}
value[0] = (f11 * f22) / (f12 * f21);
lower[2] = value[2] * exp (-1.960 * v);
upper[2] = value[2] * exp (1.960 * v);
- return 1;
+ return true;
}
/* Calculate directional measures. */
static int
-calc_directional (double v[N_DIRECTIONAL], double ase[N_DIRECTIONAL],
- double t[N_DIRECTIONAL])
+calc_directional (struct crosstabs_proc *proc, struct crosstabulation *xt,
+ double v[N_DIRECTIONAL], double ase[N_DIRECTIONAL],
+ double t[N_DIRECTIONAL], double sig[N_DIRECTIONAL])
{
- {
- int i;
-
- for (i = 0; i < N_DIRECTIONAL; i++)
- v[i] = ase[i] = t[i] = SYSMIS;
- }
+ size_t n_rows = xt->vars[ROW_VAR].n_values;
+ size_t n_cols = xt->vars[COL_VAR].n_values;
+ for (int i = 0; i < N_DIRECTIONAL; i++)
+ v[i] = ase[i] = t[i] = sig[i] = SYSMIS;
/* Lambda. */
- if (cmd.a_statistics[CRS_ST_LAMBDA])
+ if (proc->statistics & (1u << CRS_ST_LAMBDA))
{
+ /* Find maximum for each row and their sum. */
double *fim = xnmalloc (n_rows, sizeof *fim);
int *fim_index = xnmalloc (n_rows, sizeof *fim_index);
- double *fmj = xnmalloc (n_cols, sizeof *fmj);
- int *fmj_index = xnmalloc (n_cols, sizeof *fmj_index);
- double sum_fim, sum_fmj;
- double rm, cm;
- int rm_index, cm_index;
- int i, j;
-
- /* Find maximum for each row and their sum. */
- for (sum_fim = 0., i = 0; i < n_rows; i++)
+ double sum_fim = 0.0;
+ for (int i = 0; i < n_rows; i++)
{
- double max = mat[i * n_cols];
+ double max = xt->mat[i * n_cols];
int index = 0;
- for (j = 1; j < n_cols; j++)
- if (mat[j + i * n_cols] > max)
+ for (int j = 1; j < n_cols; j++)
+ if (xt->mat[j + i * n_cols] > max)
{
- max = mat[j + i * n_cols];
+ max = xt->mat[j + i * n_cols];
index = j;
}
- sum_fim += fim[i] = max;
+ fim[i] = max;
+ sum_fim += max;
fim_index[i] = index;
}
/* Find maximum for each column. */
- for (sum_fmj = 0., j = 0; j < n_cols; j++)
+ double *fmj = xnmalloc (n_cols, sizeof *fmj);
+ int *fmj_index = xnmalloc (n_cols, sizeof *fmj_index);
+ double sum_fmj = 0.0;
+ for (int j = 0; j < n_cols; j++)
{
- double max = mat[j];
+ double max = xt->mat[j];
int index = 0;
- for (i = 1; i < n_rows; i++)
- if (mat[j + i * n_cols] > max)
+ for (int i = 1; i < n_rows; i++)
+ if (xt->mat[j + i * n_cols] > max)
{
- max = mat[j + i * n_cols];
+ max = xt->mat[j + i * n_cols];
index = i;
}
- sum_fmj += fmj[j] = max;
+ fmj[j] = max;
+ sum_fmj += max;
fmj_index[j] = index;
}
/* Find maximum row total. */
- rm = row_tot[0];
- rm_index = 0;
- for (i = 1; i < n_rows; i++)
- if (row_tot[i] > rm)
+ double rm = xt->row_tot[0];
+ int rm_index = 0;
+ for (int i = 1; i < n_rows; i++)
+ if (xt->row_tot[i] > rm)
{
- rm = row_tot[i];
+ rm = xt->row_tot[i];
rm_index = i;
}
/* Find maximum column total. */
- cm = col_tot[0];
- cm_index = 0;
- for (j = 1; j < n_cols; j++)
- if (col_tot[j] > cm)
+ double cm = xt->col_tot[0];
+ int cm_index = 0;
+ for (int j = 1; j < n_cols; j++)
+ if (xt->col_tot[j] > cm)
{
- cm = col_tot[j];
+ cm = xt->col_tot[j];
cm_index = j;
}
- v[0] = (sum_fim + sum_fmj - cm - rm) / (2. * W - rm - cm);
- v[1] = (sum_fmj - rm) / (W - rm);
- v[2] = (sum_fim - cm) / (W - cm);
+ v[0] = (sum_fim + sum_fmj - cm - rm) / (2. * xt->total - rm - cm);
+ v[1] = (sum_fmj - rm) / (xt->total - rm);
+ v[2] = (sum_fim - cm) / (xt->total - cm);
- /* ASE1 for Y given X. */
+ /* ASE1 for Y given XT. */
{
- double accum;
-
- for (accum = 0., i = 0; i < n_rows; i++)
- for (j = 0; j < n_cols; j++)
- {
- const int deltaj = j == cm_index;
- accum += (mat[j + i * n_cols]
- * pow2 ((j == fim_index[i])
- - deltaj
- + v[0] * deltaj));
- }
-
- ase[2] = sqrt (accum - W * v[0]) / (W - cm);
+ double accum = 0.0;
+ for (int i = 0; i < n_rows; i++)
+ if (cm_index == fim_index[i])
+ accum += fim[i];
+ ase[2] = sqrt ((xt->total - sum_fim) * (sum_fim + cm - 2. * accum)
+ / pow3 (xt->total - cm));
}
- /* ASE0 for Y given X. */
+ /* ASE0 for Y given XT. */
{
- double accum;
-
- for (accum = 0., i = 0; i < n_rows; i++)
+ double accum = 0.0;
+ for (int i = 0; i < n_rows; i++)
if (cm_index != fim_index[i])
- accum += (mat[i * n_cols + fim_index[i]]
- + mat[i * n_cols + cm_index]);
- t[2] = v[2] / (sqrt (accum - pow2 (sum_fim - cm) / W) / (W - cm));
+ accum += (xt->mat[i * n_cols + fim_index[i]]
+ + xt->mat[i * n_cols + cm_index]);
+ t[2] = v[2] / (sqrt (accum - pow2 (sum_fim - cm) / xt->total) / (xt->total - cm));
}
- /* ASE1 for X given Y. */
+ /* ASE1 for XT given Y. */
{
- double accum;
-
- for (accum = 0., i = 0; i < n_rows; i++)
- for (j = 0; j < n_cols; j++)
- {
- const int deltaj = i == rm_index;
- accum += (mat[j + i * n_cols]
- * pow2 ((i == fmj_index[j])
- - deltaj
- + v[0] * deltaj));
- }
-
- ase[1] = sqrt (accum - W * v[0]) / (W - rm);
+ double accum = 0.0;
+ for (int j = 0; j < n_cols; j++)
+ if (rm_index == fmj_index[j])
+ accum += fmj[j];
+ ase[1] = sqrt ((xt->total - sum_fmj) * (sum_fmj + rm - 2. * accum)
+ / pow3 (xt->total - rm));
}
- /* ASE0 for X given Y. */
+ /* ASE0 for XT given Y. */
{
- double accum;
-
- for (accum = 0., j = 0; j < n_cols; j++)
+ double accum = 0.0;
+ for (int j = 0; j < n_cols; j++)
if (rm_index != fmj_index[j])
- accum += (mat[j + n_cols * fmj_index[j]]
- + mat[j + n_cols * rm_index]);
- t[1] = v[1] / (sqrt (accum - pow2 (sum_fmj - rm) / W) / (W - rm));
+ accum += (xt->mat[j + n_cols * fmj_index[j]]
+ + xt->mat[j + n_cols * rm_index]);
+ t[1] = v[1] / (sqrt (accum - pow2 (sum_fmj - rm) / xt->total) / (xt->total - rm));
}
/* Symmetric ASE0 and ASE1. */
{
- double accum0;
- double accum1;
-
- for (accum0 = accum1 = 0., i = 0; i < n_rows; i++)
- for (j = 0; j < n_cols; j++)
+ double accum0 = 0.0;
+ double accum1 = 0.0;
+ for (int i = 0; i < n_rows; i++)
+ for (int j = 0; j < n_cols; j++)
{
int temp0 = (fmj_index[j] == i) + (fim_index[i] == j);
int temp1 = (i == rm_index) + (j == cm_index);
- accum0 += mat[j + i * n_cols] * pow2 (temp0 - temp1);
- accum1 += (mat[j + i * n_cols]
+ accum0 += xt->mat[j + i * n_cols] * pow2 (temp0 - temp1);
+ accum1 += (xt->mat[j + i * n_cols]
* pow2 (temp0 + (v[0] - 1.) * temp1));
}
- ase[0] = sqrt (accum1 - 4. * W * v[0] * v[0]) / (2. * W - rm - cm);
- t[0] = v[0] / (sqrt (accum0 - pow2 ((sum_fim + sum_fmj - cm - rm) / W))
- / (2. * W - rm - cm));
+ ase[0] = sqrt (accum1 - 4. * xt->total * v[0] * v[0]) / (2. * xt->total - rm - cm);
+ t[0] = v[0] / (sqrt (accum0 - pow2 (sum_fim + sum_fmj - cm - rm) / xt->total)
+ / (2. * xt->total - rm - cm));
}
+ for (int i = 0; i < 3; i++)
+ sig[i] = 2 * gsl_cdf_ugaussian_Q (t[i]);
+
free (fim);
free (fim_index);
free (fmj);
free (fmj_index);
+ /* Tau. */
{
- double sum_fij2_ri, sum_fij2_ci;
- double sum_ri2, sum_cj2;
-
- for (sum_fij2_ri = sum_fij2_ci = 0., i = 0; i < n_rows; i++)
- for (j = 0; j < n_cols; j++)
+ double sum_fij2_ri = 0.0;
+ double sum_fij2_ci = 0.0;
+ FOR_EACH_POPULATED_ROW (i, xt)
+ FOR_EACH_POPULATED_COLUMN (j, xt)
{
- double temp = pow2 (mat[j + i * n_cols]);
- sum_fij2_ri += temp / row_tot[i];
- sum_fij2_ci += temp / col_tot[j];
+ double temp = pow2 (xt->mat[j + i * n_cols]);
+ sum_fij2_ri += temp / xt->row_tot[i];
+ sum_fij2_ci += temp / xt->col_tot[j];
}
- for (sum_ri2 = 0., i = 0; i < n_rows; i++)
- sum_ri2 += row_tot[i] * row_tot[i];
+ double sum_ri2 = 0.0;
+ for (int i = 0; i < n_rows; i++)
+ sum_ri2 += pow2 (xt->row_tot[i]);
- for (sum_cj2 = 0., j = 0; j < n_cols; j++)
- sum_cj2 += col_tot[j] * col_tot[j];
+ double sum_cj2 = 0.0;
+ for (int j = 0; j < n_cols; j++)
+ sum_cj2 += pow2 (xt->col_tot[j]);
- v[3] = (W * sum_fij2_ci - sum_ri2) / (W * W - sum_ri2);
- v[4] = (W * sum_fij2_ri - sum_cj2) / (W * W - sum_cj2);
+ v[3] = (xt->total * sum_fij2_ci - sum_ri2) / (pow2 (xt->total) - sum_ri2);
+ v[4] = (xt->total * sum_fij2_ri - sum_cj2) / (pow2 (xt->total) - sum_cj2);
}
}
- if (cmd.a_statistics[CRS_ST_UC])
+ if (proc->statistics & (1u << CRS_ST_UC))
{
- double UX, UY, UXY, P;
- double ase1_yx, ase1_xy, ase1_sym;
- int i, j;
+ double UX = 0.0;
+ FOR_EACH_POPULATED_ROW (i, xt)
+ UX -= xt->row_tot[i] / xt->total * log (xt->row_tot[i] / xt->total);
- for (UX = 0., i = 0; i < n_rows; i++)
- if (row_tot[i] > 0.)
- UX -= row_tot[i] / W * log (row_tot[i] / W);
+ double UY = 0.0;
+ FOR_EACH_POPULATED_COLUMN (j, xt)
+ UY -= xt->col_tot[j] / xt->total * log (xt->col_tot[j] / xt->total);
- for (UY = 0., j = 0; j < n_cols; j++)
- if (col_tot[j] > 0.)
- UY -= col_tot[j] / W * log (col_tot[j] / W);
-
- for (UXY = P = 0., i = 0; i < n_rows; i++)
- for (j = 0; j < n_cols; j++)
+ double UXY = 0.0;
+ double P = 0.0;
+ for (int i = 0; i < n_rows; i++)
+ for (int j = 0; j < n_cols; j++)
{
- double entry = mat[j + i * n_cols];
+ double entry = xt->mat[j + i * n_cols];
if (entry <= 0.)
continue;
- P += entry * pow2 (log (col_tot[j] * row_tot[i] / (W * entry)));
- UXY -= entry / W * log (entry / W);
+ P += entry * pow2 (log (xt->col_tot[j] * xt->row_tot[i] / (xt->total * entry)));
+ UXY -= entry / xt->total * log (entry / xt->total);
}
- for (ase1_yx = ase1_xy = ase1_sym = 0., i = 0; i < n_rows; i++)
- for (j = 0; j < n_cols; j++)
+ double ase1_yx = 0.0;
+ double ase1_xy = 0.0;
+ double ase1_sym = 0.0;
+ for (int i = 0; i < n_rows; i++)
+ for (int j = 0; j < n_cols; j++)
{
- double entry = mat[j + i * n_cols];
+ double entry = xt->mat[j + i * n_cols];
if (entry <= 0.)
continue;
- ase1_yx += entry * pow2 (UY * log (entry / row_tot[i])
- + (UX - UXY) * log (col_tot[j] / W));
- ase1_xy += entry * pow2 (UX * log (entry / col_tot[j])
- + (UY - UXY) * log (row_tot[i] / W));
+ ase1_yx += entry * pow2 (UY * log (entry / xt->row_tot[i])
+ + (UX - UXY) * log (xt->col_tot[j] / xt->total));
+ ase1_xy += entry * pow2 (UX * log (entry / xt->col_tot[j])
+ + (UY - UXY) * log (xt->row_tot[i] / xt->total));
ase1_sym += entry * pow2 ((UXY
- * log (row_tot[i] * col_tot[j] / (W * W)))
- - (UX + UY) * log (entry / W));
+ * log (xt->row_tot[i] * xt->col_tot[j] / pow2 (xt->total)))
+ - (UX + UY) * log (entry / xt->total));
}
v[5] = 2. * ((UX + UY - UXY) / (UX + UY));
- ase[5] = (2. / (W * pow2 (UX + UY))) * sqrt (ase1_sym);
- t[5] = v[5] / ((2. / (W * (UX + UY)))
- * sqrt (P - pow2 (UX + UY - UXY) / W));
+ ase[5] = (2. / (xt->total * pow2 (UX + UY))) * sqrt (ase1_sym);
+ t[5] = SYSMIS;
v[6] = (UX + UY - UXY) / UX;
- ase[6] = sqrt (ase1_xy) / (W * UX * UX);
- t[6] = v[6] / (sqrt (P - W * pow2 (UX + UY - UXY)) / (W * UX));
+ ase[6] = sqrt (ase1_xy) / (xt->total * UX * UX);
+ t[6] = v[6] / (sqrt (P - xt->total * pow2 (UX + UY - UXY)) / (xt->total * UX));
v[7] = (UX + UY - UXY) / UY;
- ase[7] = sqrt (ase1_yx) / (W * UY * UY);
- t[7] = v[7] / (sqrt (P - W * pow2 (UX + UY - UXY)) / (W * UY));
+ ase[7] = sqrt (ase1_yx) / (xt->total * UY * UY);
+ t[7] = v[7] / (sqrt (P - xt->total * pow2 (UX + UY - UXY)) / (xt->total * UY));
}
/* Somers' D. */
- if (cmd.a_statistics[CRS_ST_D])
+ if (proc->statistics & (1u << CRS_ST_D))
{
- int i;
+ double v_dummy[N_SYMMETRIC];
+ double ase_dummy[N_SYMMETRIC];
+ double t_dummy[N_SYMMETRIC];
+ double somers_d_v[3];
+ double somers_d_ase[3];
+ double somers_d_t[3];
- if (!sym)
- calc_symmetric (NULL, NULL, NULL);
- for (i = 0; i < 3; i++)
- {
- v[8 + i] = somers_d_v[i];
- ase[8 + i] = somers_d_ase[i];
- t[8 + i] = somers_d_t[i];
- }
+ if (calc_symmetric (proc, xt, v_dummy, ase_dummy, t_dummy,
+ somers_d_v, somers_d_ase, somers_d_t))
+ {
+ for (int i = 0; i < 3; i++)
+ {
+ v[8 + i] = somers_d_v[i];
+ ase[8 + i] = somers_d_ase[i];
+ t[8 + i] = somers_d_t[i];
+ sig[8 + i] = 2 * gsl_cdf_ugaussian_Q (fabs (somers_d_t[i]));
+ }
+ }
}
/* Eta. */
- if (cmd.a_statistics[CRS_ST_ETA])
+ if (proc->statistics & (1u << CRS_ST_ETA))
{
- {
- double sum_Xr, sum_X2r;
- double SX, SXW;
- int i, j;
-
- for (sum_Xr = sum_X2r = 0., i = 0; i < n_rows; i++)
- {
- sum_Xr += rows[i].f * row_tot[i];
- sum_X2r += rows[i].f * rows[i].f * row_tot[i];
- }
- SX = sum_X2r - sum_Xr * sum_Xr / W;
-
- for (SXW = 0., j = 0; j < n_cols; j++)
- {
- double cum;
-
- for (cum = 0., i = 0; i < n_rows; i++)
- {
- SXW += rows[i].f * rows[i].f * mat[j + i * n_cols];
- cum += rows[i].f * mat[j + i * n_cols];
- }
-
- SXW -= cum * cum / col_tot[j];
- }
- v[11] = sqrt (1. - SXW / SX);
- }
+ /* X dependent. */
+ double sum_Xr = 0.0;
+ double sum_X2r = 0.0;
+ for (int i = 0; i < n_rows; i++)
+ {
+ sum_Xr += xt->vars[ROW_VAR].values[i].f * xt->row_tot[i];
+ sum_X2r += pow2 (xt->vars[ROW_VAR].values[i].f) * xt->row_tot[i];
+ }
+ double SX = sum_X2r - pow2 (sum_Xr) / xt->total;
- {
- double sum_Yc, sum_Y2c;
- double SY, SYW;
- int i, j;
+ double SXW = 0.0;
+ FOR_EACH_POPULATED_COLUMN (j, xt)
+ {
+ double cum = 0.0;
- for (sum_Yc = sum_Y2c = 0., i = 0; i < n_cols; i++)
- {
- sum_Yc += cols[i].f * col_tot[i];
- sum_Y2c += cols[i].f * cols[i].f * col_tot[i];
- }
- SY = sum_Y2c - sum_Yc * sum_Yc / W;
+ for (int i = 0; i < n_rows; i++)
+ {
+ SXW += (pow2 (xt->vars[ROW_VAR].values[i].f)
+ * xt->mat[j + i * n_cols]);
+ cum += (xt->vars[ROW_VAR].values[i].f
+ * xt->mat[j + i * n_cols]);
+ }
- for (SYW = 0., i = 0; i < n_rows; i++)
- {
- double cum;
+ SXW -= cum * cum / xt->col_tot[j];
+ }
+ v[11] = sqrt (1. - SXW / SX);
- for (cum = 0., j = 0; j < n_cols; j++)
- {
- SYW += cols[j].f * cols[j].f * mat[j + i * n_cols];
- cum += cols[j].f * mat[j + i * n_cols];
- }
+ /* Y dependent. */
+ double sum_Yc = 0.0;
+ double sum_Y2c = 0.0;
+ for (int i = 0; i < n_cols; i++)
+ {
+ sum_Yc += xt->vars[COL_VAR].values[i].f * xt->col_tot[i];
+ sum_Y2c += pow2 (xt->vars[COL_VAR].values[i].f) * xt->col_tot[i];
+ }
+ double SY = sum_Y2c - pow2 (sum_Yc) / xt->total;
- SYW -= cum * cum / row_tot[i];
- }
- v[12] = sqrt (1. - SYW / SY);
- }
+ double SYW = 0.0;
+ FOR_EACH_POPULATED_ROW (i, xt)
+ {
+ double cum = 0.0;
+ for (int j = 0; j < n_cols; j++)
+ {
+ SYW += (pow2 (xt->vars[COL_VAR].values[j].f)
+ * xt->mat[j + i * n_cols]);
+ cum += (xt->vars[COL_VAR].values[j].f
+ * xt->mat[j + i * n_cols]);
+ }
+
+ SYW -= cum * cum / xt->row_tot[i];
+ }
+ v[12] = sqrt (1. - SYW / SY);
}
return 1;
}
-/* A wrapper around data_out() that limits string output to short
- string width and null terminates the result. */
-static void
-format_short (char *s, const struct fmt_spec *fp, const union value *v)
-{
- struct fmt_spec fmt_subst;
-
- /* Limit to short string width. */
- if (fmt_is_string (fp->type))
- {
- fmt_subst = *fp;
-
- assert (fmt_subst.type == FMT_A || fmt_subst.type == FMT_AHEX);
- if (fmt_subst.type == FMT_A)
- fmt_subst.w = MIN (8, fmt_subst.w);
- else
- fmt_subst.w = MIN (16, fmt_subst.w);
-
- fp = &fmt_subst;
- }
-
- /* Format. */
- data_out (v, fp, s);
-
- /* Null terminate. */
- s[fp->w] = '\0';
-}
-
/*
Local Variables:
mode: c
projects / pspp / blobdiff