X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=src%2Flanguage%2Fstats%2Fcrosstabs.q;h=aa7e2457ee24ae9d41d40bd989da985b2c822fe3;hb=5443c41f89a2b29cf8cf55f21eecb839fae08177;hp=b230702e28c36bd8517ebdc09a702b94ef5124ac;hpb=22d2bdfb26bf8df59a27191aa669090d57df56bd;p=pspp diff --git a/src/language/stats/crosstabs.q b/src/language/stats/crosstabs.q index b230702e28..aa7e2457ee 100644 --- a/src/language/stats/crosstabs.q +++ b/src/language/stats/crosstabs.q @@ -1,5 +1,5 @@ /* PSPP - a program for statistical analysis. - Copyright (C) 1997-9, 2000, 2006, 2009, 2010, 2011 Free Software Foundation, Inc. + Copyright (C) 1997-9, 2000, 2006, 2009, 2010, 2011, 2012, 2013, 2014 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 @@ -16,19 +16,17 @@ /* 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 #include +#include #include #include #include @@ -74,8 +72,7 @@ +variables=custom; missing=miss:!table/include/report; +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, @@ -124,6 +121,7 @@ enum /* A crosstabulation of 2 or more variables. */ struct pivot_table { + struct crosstabs_proc *proc; struct fmt_spec weight_format; /* Format for weight variable. */ double missing; /* Weight of missing cases. */ @@ -163,17 +161,13 @@ struct pivot_table /* 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); -} - struct crosstabs_proc { const struct dictionary *dict; @@ -186,6 +180,7 @@ struct crosstabs_proc /* Variables specifies on VARIABLES. */ const struct variable **variables; size_t n_variables; + struct hmap var_ranges; /* TABLES. */ struct pivot_table *pivots; @@ -202,6 +197,9 @@ struct crosstabs_proc bool descending; /* True if descending sort order is requested. */ }; +const struct var_range *get_var_range (const struct crosstabs_proc *, + const struct variable *); + static bool should_tabulate_case (const struct pivot_table *, const struct ccase *, enum mv_class exclude); static void tabulate_general_case (struct pivot_table *, const struct ccase *, @@ -216,6 +214,7 @@ int cmd_crosstabs (struct lexer *lexer, struct dataset *ds) { const struct variable *wv = dict_get_weight (dataset_dict (ds)); + struct var_range *range, *next_range; struct crosstabs_proc proc; struct casegrouper *grouper; struct casereader *input, *group; @@ -229,6 +228,7 @@ cmd_crosstabs (struct lexer *lexer, struct dataset *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; @@ -290,11 +290,11 @@ cmd_crosstabs (struct lexer *lexer, struct dataset *ds) proc.exclude = (cmd.miss == CRS_TABLE ? MV_ANY : cmd.miss == CRS_INCLUDE ? MV_SYSTEM : MV_NEVER); - if (proc.mode == GENERAL && proc.mode == MV_NEVER) + if (proc.mode == GENERAL && proc.exclude == MV_NEVER) { - msg (SE, _("Missing mode REPORT not allowed in general mode. " - "Assuming MISSING=TABLE.")); - proc.mode = MV_ANY; + msg (SE, _("Missing mode %s not allowed in general mode. " + "Assuming %s."), "REPORT", "MISSING=TABLE"); + proc.exclude = MV_ANY; } /* PIVOT. */ @@ -347,6 +347,12 @@ cmd_crosstabs (struct lexer *lexer, struct dataset *ds) 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 (pt = &proc.pivots[0]; pt < &proc.pivots[proc.n_pivots]; pt++) { free (pt->vars); @@ -428,6 +434,7 @@ crs_custom_tables (struct lexer *lexer, struct dataset *ds, struct pivot_table *pt = &proc->pivots[proc->n_pivots++]; int j; + pt->proc = proc; pt->weight_format = proc->weight_format; pt->missing = 0.; pt->n_vars = n_by; @@ -469,7 +476,7 @@ crs_custom_variables (struct lexer *lexer, struct dataset *ds, 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; } @@ -514,11 +521,15 @@ crs_custom_variables (struct lexer *lexer, struct dataset *ds, 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->count = max - min + 1; - var_attach_aux (proc->variables[i], vr, var_dtor_free); + hmap_insert (&proc->var_ranges, &vr->hmap_node, + hash_pointer (var, 0)); } if (lex_token (lexer) == T_SLASH) @@ -536,6 +547,22 @@ crs_custom_variables (struct lexer *lexer, struct dataset *ds, /* Data file processing. */ +const struct var_range * +get_var_range (const struct crosstabs_proc *proc, const struct variable *var) +{ + if (!hmap_is_empty (&proc->var_ranges)) + { + const struct var_range *range; + + 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; +} + static bool should_tabulate_case (const struct pivot_table *pt, const struct ccase *c, enum mv_class exclude) @@ -544,7 +571,7 @@ should_tabulate_case (const struct pivot_table *pt, const struct ccase *c, for (j = 0; j < pt->n_vars; j++) { const struct variable *var = pt->vars[j]; - struct var_range *range = get_var_range (var); + const struct var_range *range = get_var_range (pt->proc, var); if (var_is_value_missing (var, case_data (c, var), exclude)) return false; @@ -708,12 +735,25 @@ postcalc (struct crosstabs_proc *proc) pt->missing = 0.0; - /* Free only the members that were allocated in this - function. The other pointer members are either both - allocated and destroyed at a lower level (in - output_pivot_table), or both allocated and destroyed at - a higher level (in crs_custom_tables and free_proc, + /* Free the members that were allocated in this function(and the values + owned by the entries. + + The other pointer members are either both allocated and destroyed at a + lower level (in output_pivot_table), or both allocated and destroyed + at a higher level (in crs_custom_tables and free_proc, respectively). */ + for (i = 0; i < pt->n_vars; i++) + { + int width = var_get_width (pt->vars[i]); + if (value_needs_init (width)) + { + size_t j; + + for (j = 0; j < pt->n_entries; j++) + value_destroy (&pt->entries[j]->values[i], width); + } + } + for (i = 0; i < pt->n_entries; i++) free (pt->entries[i]); free (pt->entries); @@ -824,6 +864,7 @@ make_summary_table (struct crosstabs_proc *proc) int i; summary = tab_create (7, 3 + proc->n_pivots); + tab_set_format (summary, RC_WEIGHT, &proc->weight_format); tab_title (summary, _("Summary.")); tab_headers (summary, 1, 0, 3, 0); tab_joint_text (summary, 1, 0, 6, 0, TAB_CENTER, _("Cases")); @@ -868,8 +909,7 @@ make_summary_table (struct crosstabs_proc *proc) n[2] = n[0] + n[1]; for (i = 0; i < 3; i++) { - tab_double (summary, i * 2 + 1, 0, TAB_RIGHT, n[i], - &proc->weight_format); + tab_double (summary, i * 2 + 1, 0, TAB_RIGHT, n[i], NULL, RC_WEIGHT); tab_text_format (summary, i * 2 + 2, 0, TAB_RIGHT, "%.1f%%", n[i] / n[2] * 100.); } @@ -885,10 +925,10 @@ make_summary_table (struct crosstabs_proc *proc) static struct tab_table *create_crosstab_table (struct crosstabs_proc *, struct pivot_table *); -static struct tab_table *create_chisq_table (struct pivot_table *); -static struct tab_table *create_sym_table (struct pivot_table *); -static struct tab_table *create_risk_table (struct pivot_table *); -static struct tab_table *create_direct_table (struct pivot_table *); +static struct tab_table *create_chisq_table (struct crosstabs_proc *proc, struct pivot_table *); +static struct tab_table *create_sym_table (struct crosstabs_proc *proc, struct pivot_table *); +static struct tab_table *create_risk_table (struct crosstabs_proc *proc, struct pivot_table *); +static struct tab_table *create_direct_table (struct crosstabs_proc *proc, struct pivot_table *); static void display_dimensions (struct crosstabs_proc *, struct pivot_table *, struct tab_table *, int first_difference); static void display_crosstabulation (struct crosstabs_proc *, @@ -927,9 +967,9 @@ output_pivot_table (struct crosstabs_proc *proc, struct pivot_table *pt) struct string vars; int i; - ds_init_cstr (&vars, var_get_name (pt->vars[0])); + ds_init_cstr (&vars, var_to_string (pt->vars[0])); for (i = 1; i < pt->n_vars; i++) - ds_put_format (&vars, " * %s", var_get_name (pt->vars[i])); + ds_put_format (&vars, " * %s", var_to_string (pt->vars[i])); /* TRANSLATORS: The %s here describes a crosstabulation. It takes the form "var1 * var2 * var3 * ...". */ @@ -937,23 +977,24 @@ output_pivot_table (struct crosstabs_proc *proc, struct pivot_table *pt) ds_cstr (&vars)); ds_destroy (&vars); + free (pt->cols); return; } if (proc->cells) table = create_crosstab_table (proc, pt); if (proc->statistics & (1u << CRS_ST_CHISQ)) - chisq = create_chisq_table (pt); + chisq = create_chisq_table (proc, pt); if (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))) - sym = create_sym_table (pt); + sym = create_sym_table (proc, pt); if (proc->statistics & (1u << CRS_ST_RISK)) - risk = create_risk_table (pt); + risk = create_risk_table (proc, pt); if (proc->statistics & ((1u << CRS_ST_LAMBDA) | (1u << CRS_ST_UC) | (1u << CRS_ST_D) | (1u << CRS_ST_ETA))) - direct = create_direct_table (pt); + direct = create_direct_table (proc, pt); row0 = row1 = 0; while (find_crosstab (pt, &row0, &row1)) @@ -1154,11 +1195,12 @@ create_crosstab_table (struct crosstabs_proc *proc, struct pivot_table *pt) table = tab_create (x.n_consts + 1 + x.n_cols + 1, (x.n_entries / x.n_cols) * 3 / 2 * proc->n_cells + 10); tab_headers (table, x.n_consts + 1, 0, 2, 0); + tab_set_format (table, RC_WEIGHT, &proc->weight_format); /* First header line. */ tab_joint_text (table, x.n_consts + 1, 0, (x.n_consts + 1) + (x.n_cols - 1), 0, - TAB_CENTER | TAT_TITLE, var_get_name (x.vars[COL_VAR])); + TAB_CENTER | TAT_TITLE, var_to_string (x.vars[COL_VAR])); tab_hline (table, TAL_1, x.n_consts + 1, x.n_consts + 2 + x.n_cols - 2, 1); @@ -1169,7 +1211,7 @@ create_crosstab_table (struct crosstabs_proc *proc, struct pivot_table *pt) x.n_consts + 2 - i - 1, 1, TAB_RIGHT | TAT_TITLE, var_to_string (x.vars[i])); tab_text (table, x.n_consts + 2 - 2, 1, TAB_RIGHT | TAT_TITLE, - var_get_name (x.vars[ROW_VAR])); + var_to_string (x.vars[ROW_VAR])); for (i = 0; i < x.n_cols; i++) table_value_missing (proc, table, x.n_consts + 2 + i - 1, 1, TAB_RIGHT, &x.cols[i], x.vars[COL_VAR]); @@ -1184,14 +1226,14 @@ create_crosstab_table (struct crosstabs_proc *proc, struct pivot_table *pt) { if (i) ds_put_cstr (&title, " * "); - ds_put_cstr (&title, var_get_name (x.vars[i])); + ds_put_cstr (&title, var_to_string (x.vars[i])); } for (i = 0; i < pt->n_consts; i++) { const struct variable *var = pt->const_vars[i]; char *s; - ds_put_format (&title, ", %s=", var_get_name (var)); + ds_put_format (&title, ", %s=", var_to_string (var)); /* Insert the formatted value of VAR without any leading spaces. */ s = data_out (&pt->const_values[i], var_get_encoding (var), @@ -1219,13 +1261,14 @@ create_crosstab_table (struct crosstabs_proc *proc, struct pivot_table *pt) } static struct tab_table * -create_chisq_table (struct pivot_table *pt) +create_chisq_table (struct crosstabs_proc *proc, struct pivot_table *pt) { struct tab_table *chisq; chisq = tab_create (6 + (pt->n_vars - 2), pt->n_entries / pt->n_cols * 3 / 2 * N_CHISQ + 10); tab_headers (chisq, 1 + (pt->n_vars - 2), 0, 1, 0); + tab_set_format (chisq, RC_WEIGHT, &proc->weight_format); tab_title (chisq, _("Chi-square tests.")); @@ -1246,12 +1289,15 @@ create_chisq_table (struct pivot_table *pt) /* Symmetric measures. */ static struct tab_table * -create_sym_table (struct pivot_table *pt) +create_sym_table (struct crosstabs_proc *proc, struct pivot_table *pt) { struct tab_table *sym; sym = tab_create (6 + (pt->n_vars - 2), pt->n_entries / pt->n_cols * 7 + 10); + + tab_set_format (sym, RC_WEIGHT, &proc->weight_format); + tab_headers (sym, 2 + (pt->n_vars - 2), 0, 1, 0); tab_title (sym, _("Symmetric measures.")); @@ -1269,13 +1315,14 @@ create_sym_table (struct pivot_table *pt) /* Risk estimate. */ static struct tab_table * -create_risk_table (struct pivot_table *pt) +create_risk_table (struct crosstabs_proc *proc, struct pivot_table *pt) { struct tab_table *risk; risk = tab_create (4 + (pt->n_vars - 2), pt->n_entries / pt->n_cols * 4 + 10); tab_headers (risk, 1 + pt->n_vars - 2, 0, 2, 0); tab_title (risk, _("Risk estimate.")); + tab_set_format (risk, RC_WEIGHT, &proc->weight_format); tab_offset (risk, pt->n_vars - 2, 0); tab_joint_text_format (risk, 2, 0, 3, 0, TAB_CENTER | TAT_TITLE, @@ -1293,7 +1340,7 @@ create_risk_table (struct pivot_table *pt) /* Directional measures. */ static struct tab_table * -create_direct_table (struct pivot_table *pt) +create_direct_table (struct crosstabs_proc *proc, struct pivot_table *pt) { struct tab_table *direct; @@ -1301,6 +1348,7 @@ create_direct_table (struct pivot_table *pt) pt->n_entries / pt->n_cols * 7 + 10); tab_headers (direct, 3 + (pt->n_vars - 2), 0, 1, 0); tab_title (direct, _("Directional measures.")); + tab_set_format (direct, RC_WEIGHT, &proc->weight_format); tab_offset (direct, pt->n_vars - 2, 0); tab_text (direct, 0, 0, TAB_LEFT | TAT_TITLE, _("Category")); @@ -1418,13 +1466,15 @@ compare_value_3way_inv (const void *a_, const void *b_, const void *width_) with index VAR_IDX takes on. The values are returned as a malloc()'d array stored in *VALUES, with the number of values stored in *VALUE_CNT. - */ + + The caller must eventually free *VALUES, but each pointer in *VALUES points + to existing data not owned by *VALUES itself. */ static void enum_var_values (const struct pivot_table *pt, int var_idx, union value **valuesp, int *n_values, bool descending) { const struct variable *var = pt->vars[var_idx]; - struct var_range *range = get_var_range (var); + const struct var_range *range = get_var_range (pt->proc, var); union value *values; size_t i; @@ -1520,14 +1570,13 @@ static void format_cell_entry (struct tab_table *table, int c, int r, double value, char suffix, bool mark_missing, const struct dictionary *dict) { - const struct fmt_spec f = {FMT_F, 10, 1}; union value v; char suffixes[3]; int suffix_len; char *s; v.f = value; - s = data_out (&v, dict_get_encoding (dict), &f); + s = data_out (&v, dict_get_encoding (dict), settings_get_format ()); suffix_len = 0; if (suffix != 0) @@ -1751,7 +1800,7 @@ display_chisq (struct pivot_table *pt, struct tab_table *chisq, calc_chisq (pt, chisq_v, df, &fisher1, &fisher2); - tab_offset (chisq, pt->n_vars - 2, -1); + tab_offset (chisq, pt->n_consts + pt->n_vars - 2, -1); for (i = 0; i < N_CHISQ; i++) { @@ -1762,22 +1811,22 @@ display_chisq (struct pivot_table *pt, struct tab_table *chisq, 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], &pt->weight_format); + tab_double (chisq, 1, 0, TAB_RIGHT, chisq_v[i], NULL, RC_OTHER); + tab_double (chisq, 2, 0, TAB_RIGHT, df[i], NULL, RC_WEIGHT); tab_double (chisq, 3, 0, TAB_RIGHT, - gsl_cdf_chisq_Q (chisq_v[i], df[i]), NULL); + gsl_cdf_chisq_Q (chisq_v[i], df[i]), NULL, RC_PVALUE); } else { *showed_fisher = true; - tab_double (chisq, 4, 0, TAB_RIGHT, fisher2, NULL); - tab_double (chisq, 5, 0, TAB_RIGHT, fisher1, NULL); + tab_double (chisq, 4, 0, TAB_RIGHT, fisher2, NULL, RC_PVALUE); + tab_double (chisq, 5, 0, TAB_RIGHT, fisher1, NULL, RC_PVALUE); } tab_next_row (chisq); } tab_text (chisq, 0, 0, TAB_LEFT, _("N of Valid Cases")); - tab_double (chisq, 1, 0, TAB_RIGHT, pt->total, &pt->weight_format); + tab_double (chisq, 1, 0, TAB_RIGHT, pt->total, NULL, RC_WEIGHT); tab_next_row (chisq); tab_offset (chisq, 0, -1); @@ -1828,7 +1877,7 @@ display_symmetric (struct crosstabs_proc *proc, struct pivot_table *pt, somers_d_v, somers_d_ase, somers_d_t)) return; - tab_offset (sym, pt->n_vars - 2, -1); + tab_offset (sym, pt->n_consts + pt->n_vars - 2, -1); for (i = 0; i < N_SYMMETRIC; i++) { @@ -1842,17 +1891,17 @@ display_symmetric (struct crosstabs_proc *proc, struct pivot_table *pt, } tab_text (sym, 1, 0, TAB_LEFT, gettext (stats[i])); - tab_double (sym, 2, 0, TAB_RIGHT, sym_v[i], NULL); + tab_double (sym, 2, 0, TAB_RIGHT, sym_v[i], NULL, RC_OTHER); if (sym_ase[i] != SYSMIS) - tab_double (sym, 3, 0, TAB_RIGHT, sym_ase[i], NULL); + tab_double (sym, 3, 0, TAB_RIGHT, sym_ase[i], NULL, RC_OTHER); 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_double (sym, 4, 0, TAB_RIGHT, sym_t[i], NULL, RC_OTHER); + /*tab_double (sym, 5, 0, TAB_RIGHT, normal_sig (sym_v[i]), NULL, RC_PVALUE);*/ tab_next_row (sym); } tab_text (sym, 0, 0, TAB_LEFT, _("N of Valid Cases")); - tab_double (sym, 2, 0, TAB_RIGHT, pt->total, &pt->weight_format); + tab_double (sym, 2, 0, TAB_RIGHT, pt->total, NULL, RC_WEIGHT); tab_next_row (sym); tab_offset (sym, 0, -1); @@ -1873,7 +1922,7 @@ display_risk (struct pivot_table *pt, struct tab_table *risk) if (!calc_risk (pt, risk_v, upper, lower, c)) return; - tab_offset (risk, pt->n_vars - 2, -1); + tab_offset (risk, pt->n_consts + pt->n_vars - 2, -1); for (i = 0; i < 3; i++) { @@ -1890,34 +1939,34 @@ display_risk (struct pivot_table *pt, struct tab_table *risk) case 0: if (var_is_numeric (cv)) sprintf (buf, _("Odds Ratio for %s (%g / %g)"), - var_get_name (cv), c[0].f, c[1].f); + var_to_string (cv), c[0].f, c[1].f); else sprintf (buf, _("Odds Ratio for %s (%.*s / %.*s)"), - var_get_name (cv), + var_to_string (cv), cvw, value_str (&c[0], cvw), cvw, value_str (&c[1], cvw)); break; case 1: case 2: if (var_is_numeric (rv)) - sprintf (buf, _("For cohort %s = %g"), - var_get_name (rv), pt->rows[i - 1].f); + sprintf (buf, _("For cohort %s = %.*g"), + var_to_string (rv), DBL_DIG + 1, pt->rows[i - 1].f); else sprintf (buf, _("For cohort %s = %.*s"), - var_get_name (rv), + var_to_string (rv), rvw, value_str (&pt->rows[i - 1], rvw)); 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_double (risk, 1, 0, TAB_RIGHT, risk_v[i], NULL, RC_OTHER); + tab_double (risk, 2, 0, TAB_RIGHT, lower[i], NULL, RC_OTHER); + tab_double (risk, 3, 0, TAB_RIGHT, upper[i], NULL, RC_OTHER); tab_next_row (risk); } tab_text (risk, 0, 0, TAB_LEFT, _("N of Valid Cases")); - tab_double (risk, 1, 0, TAB_RIGHT, pt->total, &pt->weight_format); + tab_double (risk, 1, 0, TAB_RIGHT, pt->total, NULL, RC_WEIGHT); tab_next_row (risk); tab_offset (risk, 0, -1); @@ -1925,7 +1974,7 @@ display_risk (struct pivot_table *pt, struct tab_table *risk) static int calc_directional (struct crosstabs_proc *, struct pivot_table *, double[N_DIRECTIONAL], double[N_DIRECTIONAL], - double[N_DIRECTIONAL]); + double[N_DIRECTIONAL], double[N_DIRECTIONAL]); /* Display directional measures. */ static void @@ -1992,13 +2041,14 @@ display_directional (struct crosstabs_proc *proc, struct pivot_table *pt, double direct_v[N_DIRECTIONAL]; double direct_ase[N_DIRECTIONAL]; double direct_t[N_DIRECTIONAL]; + double sig[N_DIRECTIONAL]; int i; - if (!calc_directional (proc, pt, direct_v, direct_ase, direct_t)) + if (!calc_directional (proc, pt, direct_v, direct_ase, direct_t, sig)) return; - tab_offset (direct, pt->n_vars - 2, -1); + tab_offset (direct, pt->n_consts + pt->n_vars - 2, -1); for (i = 0; i < N_DIRECTIONAL; i++) { @@ -2022,9 +2072,9 @@ display_directional (struct crosstabs_proc *proc, struct pivot_table *pt, if (k == 0) string = NULL; else if (k == 1) - string = var_get_name (pt->vars[0]); + string = var_to_string (pt->vars[0]); else - string = var_get_name (pt->vars[1]); + string = var_to_string (pt->vars[1]); tab_text_format (direct, j, 0, TAB_LEFT, gettext (stats_names[j][k]), string); @@ -2032,12 +2082,12 @@ display_directional (struct crosstabs_proc *proc, struct pivot_table *pt, } } - tab_double (direct, 3, 0, TAB_RIGHT, direct_v[i], NULL); + tab_double (direct, 3, 0, TAB_RIGHT, direct_v[i], NULL, RC_OTHER); if (direct_ase[i] != SYSMIS) - tab_double (direct, 4, 0, TAB_RIGHT, direct_ase[i], NULL); + tab_double (direct, 4, 0, TAB_RIGHT, direct_ase[i], NULL, RC_OTHER); 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_double (direct, 5, 0, TAB_RIGHT, direct_t[i], NULL, RC_OTHER); + tab_double (direct, 6, 0, TAB_RIGHT, sig[i], NULL, RC_PVALUE); tab_next_row (direct); } @@ -2046,16 +2096,17 @@ display_directional (struct crosstabs_proc *proc, struct pivot_table *pt, /* Statistical calculations. */ -/* Returns the value of the gamma (factorial) function for an integer +/* Returns the value of the logarithm of gamma (factorial) function for an integer argument PT. */ static double -gamma_int (double pt) +log_gamma_int (double pt) { - double r = 1; + double r = 0; int i; for (i = 2; i < pt; i++) - r *= i; + r += log(i); + return r; } @@ -2064,11 +2115,11 @@ gamma_int (double pt) 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. */ @@ -2086,6 +2137,7 @@ static void calc_fisher (int a, int b, int c, int d, double *fisher1, double *fisher2) { int pt; + double pn1; if (MIN (c, d) < MIN (a, b)) swap (&a, &c), swap (&b, &d); @@ -2099,13 +2151,21 @@ calc_fisher (int a, int b, int c, int d, double *fisher1, double *fisher2) swap (&a, &c), swap (&b, &d); } - *fisher1 = 0.; - for (pt = 0; pt <= a; pt++) - *fisher1 += Pr (a - pt, b + pt, c + pt, d - pt); + pn1 = Pr (a, b, c, d); + *fisher1 = pn1; + for (pt = 1; pt <= a; pt++) + { + *fisher1 += Pr (a - pt, b + pt, c + pt, d - pt); + } *fisher2 = *fisher1; + for (pt = 1; pt <= b; pt++) - *fisher2 += Pr (a + pt, b - pt, c - pt, d + pt); + { + double p = Pr (a + pt, b - pt, c - pt, d + pt); + if (p < pn1) + *fisher2 += p; + } } /* Calculates chi-squares into CHISQ. MAT is a matrix with N_COLS @@ -2190,8 +2250,7 @@ calc_chisq (struct pivot_table *pt, } /* 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. */ @@ -2205,12 +2264,12 @@ calc_chisq (struct pivot_table *pt, } } -/* 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 +/* 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 PT and Y. */ static void calc_r (struct pivot_table *pt, - double *PT, double *Y, double *r, double *ase_0, double *ase_1) + double *PT, double *Y, double *r, double *t, double *error) { double SX, SY, S, T; double Xbar, Ybar; @@ -2248,7 +2307,7 @@ calc_r (struct pivot_table *pt, SY = sum_Y2c - pow2 (sum_Yc) / pt->total; T = sqrt (SX * SY); *r = S / T; - *ase_0 = sqrt ((sum_X2Y2f - pow2 (sum_XYf) / pt->total) / (sum_X2r * sum_Y2c)); + *t = *r / sqrt (1 - pow2 (*r)) * sqrt (pt->total - 2); { double s, c, y, t; @@ -2269,7 +2328,7 @@ calc_r (struct pivot_table *pt, c = (t - s) - y; s = t; } - *ase_1 = sqrt (s) / (T * T); + *error = sqrt (s) / (T * T); } } @@ -2476,7 +2535,7 @@ calc_symmetric (struct crosstabs_proc *proc, struct pivot_table *pt, 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) / pt->total))); @@ -2536,13 +2595,11 @@ calc_symmetric (struct crosstabs_proc *proc, struct pivot_table *pt, } calc_r (pt, R, C, &v[6], &t[6], &ase[6]); - t[6] = v[6] / t[6]; free (R); free (C); calc_r (pt, (double *) pt->rows, (double *) pt->cols, &v[7], &t[7], &ase[7]); - t[7] = v[7] / t[7]; } /* Cohen's kappa. */ @@ -2664,13 +2721,13 @@ calc_risk (struct pivot_table *pt, static int calc_directional (struct crosstabs_proc *proc, struct pivot_table *pt, double v[N_DIRECTIONAL], double ase[N_DIRECTIONAL], - double t[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; + v[i] = ase[i] = t[i] = sig[i] = SYSMIS; } /* Lambda. */ @@ -2745,19 +2802,14 @@ calc_directional (struct crosstabs_proc *proc, struct pivot_table *pt, /* ASE1 for Y given PT. */ { - double accum; + double accum; - for (accum = 0., i = 0; i < pt->n_rows; i++) - for (j = 0; j < pt->n_cols; j++) - { - const int deltaj = j == cm_index; - accum += (pt->mat[j + i * pt->n_cols] - * pow2 ((j == fim_index[i]) - - deltaj - + v[0] * deltaj)); - } - - ase[2] = sqrt (accum - pt->total * v[0]) / (pt->total - cm); + accum = 0.; + for (i = 0; i < pt->n_rows; i++) + if (cm_index == fim_index[i]) + accum += fim[i]; + ase[2] = sqrt ((pt->total - sum_fim) * (sum_fim + cm - 2. * accum) + / pow3 (pt->total - cm)); } /* ASE0 for Y given PT. */ @@ -2773,19 +2825,14 @@ calc_directional (struct crosstabs_proc *proc, struct pivot_table *pt, /* ASE1 for PT given Y. */ { - double accum; + double accum; - for (accum = 0., i = 0; i < pt->n_rows; i++) - for (j = 0; j < pt->n_cols; j++) - { - const int deltaj = i == rm_index; - accum += (pt->mat[j + i * pt->n_cols] - * pow2 ((i == fmj_index[j]) - - deltaj - + v[0] * deltaj)); - } - - ase[1] = sqrt (accum - pt->total * v[0]) / (pt->total - rm); + accum = 0.; + for (j = 0; j < pt->n_cols; j++) + if (rm_index == fmj_index[j]) + accum += fmj[j]; + ase[1] = sqrt ((pt->total - sum_fmj) * (sum_fmj + rm - 2. * accum) + / pow3 (pt->total - rm)); } /* ASE0 for PT given Y. */ @@ -2814,15 +2861,19 @@ calc_directional (struct crosstabs_proc *proc, struct pivot_table *pt, * pow2 (temp0 + (v[0] - 1.) * temp1)); } ase[0] = sqrt (accum1 - 4. * pt->total * v[0] * v[0]) / (2. * pt->total - rm - cm); - t[0] = v[0] / (sqrt (accum0 - pow2 ((sum_fim + sum_fmj - cm - rm) / pt->total)) + t[0] = v[0] / (sqrt (accum0 - pow2 (sum_fim + sum_fmj - cm - rm) / pt->total) / (2. * pt->total - rm - cm)); } + for (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; @@ -2891,8 +2942,7 @@ calc_directional (struct crosstabs_proc *proc, struct pivot_table *pt, v[5] = 2. * ((UX + UY - UXY) / (UX + UY)); ase[5] = (2. / (pt->total * pow2 (UX + UY))) * sqrt (ase1_sym); - t[5] = v[5] / ((2. / (pt->total * (UX + UY))) - * sqrt (P - pow2 (UX + UY - UXY) / pt->total)); + t[5] = SYSMIS; v[6] = (UX + UY - UXY) / UX; ase[6] = sqrt (ase1_xy) / (pt->total * UX * UX); @@ -2922,6 +2972,7 @@ calc_directional (struct crosstabs_proc *proc, struct pivot_table *pt, 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])); } } }