1 /* PSPP - EXAMINE data for normality . -*-c-*-
3 Copyright (C) 2004 Free Software Foundation, Inc.
5 This program is free software; you can redistribute it and/or
6 modify it under the terms of the GNU General Public License as
7 published by the Free Software Foundation; either version 2 of the
8 License, or (at your option) any later version.
10 This program is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with this program; if not, write to the Free Software
17 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
22 #include <gsl/gsl_cdf.h>
23 #include <libpspp/message.h>
28 #include <data/case.h>
29 #include <data/casefile.h>
30 #include <data/dictionary.h>
31 #include <data/procedure.h>
32 #include <data/value-labels.h>
33 #include <data/variable.h>
34 #include <language/command.h>
35 #include <language/dictionary/split-file.h>
36 #include <language/lexer/lexer.h>
37 #include <libpspp/alloc.h>
38 #include <libpspp/compiler.h>
39 #include <libpspp/hash.h>
40 #include <libpspp/magic.h>
41 #include <libpspp/message.h>
42 #include <libpspp/misc.h>
43 #include <libpspp/str.h>
44 #include <math/factor-stats.h>
45 #include <math/moments.h>
46 #include <math/percentiles.h>
47 #include <output/charts/box-whisker.h>
48 #include <output/charts/cartesian.h>
49 #include <output/manager.h>
50 #include <output/table.h>
55 #define _(msgid) gettext (msgid)
56 #define N_(msgid) msgid
59 #include <output/chart.h>
60 #include <output/charts/plot-hist.h>
61 #include <output/charts/plot-chart.h>
68 missing=miss:pairwise/!listwise,
70 incl:include/!exclude;
71 +compare=cmp:variables/!groups;
74 +plot[plt_]=stemleaf,boxplot,npplot,:spreadlevel(*d:n),histogram,all,none;
76 +statistics[st_]=descriptives,:extreme(*d:n),all,none.
85 static struct cmd_examine cmd;
87 static struct variable **dependent_vars;
89 static size_t n_dependent_vars;
94 /* The independent variable */
95 struct variable *indep_var[2];
98 /* Hash table of factor stats indexed by 2 values */
99 struct hsh_table *fstats;
101 /* The hash table after it has been crunched */
102 struct factor_statistics **fs;
108 /* Linked list of factors */
109 static struct factor *factors = 0;
111 static struct metrics *totals = 0;
113 /* Parse the clause specifying the factors */
114 static int examine_parse_independent_vars (struct lexer *lexer, const struct dictionary *dict, struct cmd_examine *cmd);
118 /* Output functions */
119 static void show_summary (struct variable **dependent_var, int n_dep_var,
120 const struct factor *f);
122 static void show_extremes (struct variable **dependent_var,
124 const struct factor *factor,
127 static void show_descriptives (struct variable **dependent_var,
129 struct factor *factor);
131 static void show_percentiles (struct variable **dependent_var,
133 struct factor *factor);
138 void np_plot (const struct metrics *m, const char *factorname);
141 void box_plot_group (const struct factor *fctr,
142 const struct variable **vars, int n_vars,
143 const struct variable *id
147 void box_plot_variables (const struct factor *fctr,
148 const struct variable **vars, int n_vars,
149 const struct variable *id
154 /* Per Split function */
155 static bool run_examine (const struct ccase *,
156 const struct casefile *cf, void *cmd_, const struct dataset *);
158 static void output_examine (void);
161 void factor_calc (const struct ccase *c, int case_no,
162 double weight, int case_missing);
165 /* Represent a factor as a string, so it can be
166 printed in a human readable fashion */
167 const char * factor_to_string (const struct factor *fctr,
168 const struct factor_statistics *fs,
169 const struct variable *var);
172 /* Represent a factor as a string, so it can be
173 printed in a human readable fashion,
174 but sacrificing some readablility for the sake of brevity */
175 const char *factor_to_string_concise (const struct factor *fctr,
176 struct factor_statistics *fs);
181 /* Function to use for testing for missing values */
182 static var_is_missing_func *value_is_missing;
187 static subc_list_double percentile_list;
189 static enum pc_alg percentile_algorithm;
191 static short sbc_percentile;
195 cmd_examine (struct lexer *lexer, struct dataset *ds)
199 subc_list_double_create (&percentile_list);
200 percentile_algorithm = PC_HAVERAGE;
202 if ( !parse_examine (lexer, ds, &cmd, NULL) )
204 subc_list_double_destroy (&percentile_list);
208 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
209 if (cmd.incl == XMN_INCLUDE )
210 value_is_missing = var_is_value_system_missing;
212 value_is_missing = var_is_value_missing;
214 if ( cmd.st_n == SYSMIS )
217 if ( ! cmd.sbc_cinterval)
218 cmd.n_cinterval[0] = 95.0;
220 /* If descriptives have been requested, make sure the
221 quartiles are calculated */
222 if ( cmd.a_statistics[XMN_ST_DESCRIPTIVES] )
224 subc_list_double_push (&percentile_list, 25);
225 subc_list_double_push (&percentile_list, 50);
226 subc_list_double_push (&percentile_list, 75);
229 ok = multipass_procedure_with_splits (ds, run_examine, &cmd);
236 if ( dependent_vars )
237 free (dependent_vars);
240 struct factor *f = factors ;
243 struct factor *ff = f;
247 hsh_destroy ( ff->fstats ) ;
253 subc_list_double_destroy (&percentile_list);
255 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
260 /* Show all the appropriate tables */
262 output_examine (void)
266 /* Show totals if appropriate */
267 if ( ! cmd.sbc_nototal || factors == 0 )
269 show_summary (dependent_vars, n_dependent_vars, 0);
271 if ( cmd.sbc_statistics )
273 if ( cmd.a_statistics[XMN_ST_EXTREME])
274 show_extremes (dependent_vars, n_dependent_vars, 0, cmd.st_n);
276 if ( cmd.a_statistics[XMN_ST_DESCRIPTIVES])
277 show_descriptives (dependent_vars, n_dependent_vars, 0);
280 if ( sbc_percentile )
281 show_percentiles (dependent_vars, n_dependent_vars, 0);
286 if ( cmd.a_plot[XMN_PLT_STEMLEAF] )
287 msg (SW, _ ("%s is not currently supported."), "STEMLEAF");
289 if ( cmd.a_plot[XMN_PLT_SPREADLEVEL] )
290 msg (SW, _ ("%s is not currently supported."), "SPREADLEVEL");
292 if ( cmd.a_plot[XMN_PLT_NPPLOT] )
294 for ( v = 0 ; v < n_dependent_vars; ++v )
295 np_plot (&totals[v], var_to_string (dependent_vars[v]));
298 if ( cmd.a_plot[XMN_PLT_BOXPLOT] )
300 if ( cmd.cmp == XMN_GROUPS )
302 box_plot_group (0, (const struct variable **) dependent_vars,
303 n_dependent_vars, cmd.v_id);
306 box_plot_variables (0,
307 (const struct variable **) dependent_vars,
308 n_dependent_vars, cmd.v_id);
311 if ( cmd.a_plot[XMN_PLT_HISTOGRAM] )
313 for ( v = 0 ; v < n_dependent_vars; ++v )
315 struct normal_curve normal;
317 normal.N = totals[v].n;
318 normal.mean = totals[v].mean;
319 normal.stddev = totals[v].stddev;
321 histogram_plot (totals[v].histogram,
322 var_to_string (dependent_vars[v]),
332 /* Show grouped statistics as appropriate */
336 show_summary (dependent_vars, n_dependent_vars, fctr);
338 if ( cmd.sbc_statistics )
340 if ( cmd.a_statistics[XMN_ST_EXTREME])
341 show_extremes (dependent_vars, n_dependent_vars, fctr, cmd.st_n);
343 if ( cmd.a_statistics[XMN_ST_DESCRIPTIVES])
344 show_descriptives (dependent_vars, n_dependent_vars, fctr);
347 if ( sbc_percentile )
348 show_percentiles (dependent_vars, n_dependent_vars, fctr);
355 struct factor_statistics **fs = fctr->fs ;
357 if ( cmd.a_plot[XMN_PLT_BOXPLOT] )
359 if ( cmd.cmp == XMN_VARIABLES )
360 box_plot_variables (fctr,
361 (const struct variable **) dependent_vars,
362 n_dependent_vars, cmd.v_id);
364 box_plot_group (fctr,
365 (const struct variable **) dependent_vars,
366 n_dependent_vars, cmd.v_id);
369 for ( v = 0 ; v < n_dependent_vars; ++v )
372 for ( fs = fctr->fs ; *fs ; ++fs )
374 const char *s = factor_to_string (fctr, *fs, dependent_vars[v]);
376 if ( cmd.a_plot[XMN_PLT_NPPLOT] )
377 np_plot (& (*fs)->m[v], s);
379 if ( cmd.a_plot[XMN_PLT_HISTOGRAM] )
381 struct normal_curve normal;
383 normal.N = (*fs)->m[v].n;
384 normal.mean = (*fs)->m[v].mean;
385 normal.stddev = (*fs)->m[v].stddev;
387 histogram_plot ((*fs)->m[v].histogram,
391 } /* for ( fs .... */
393 } /* for ( v = 0 ..... */
403 /* Create a hash table of percentiles and their values from the list of
405 static struct hsh_table *
406 list_to_ptile_hash (const subc_list_double *l)
410 struct hsh_table *h ;
412 h = hsh_create (subc_list_double_count (l),
413 (hsh_compare_func *) ptile_compare,
414 (hsh_hash_func *) ptile_hash,
415 (hsh_free_func *) free,
419 for ( i = 0 ; i < subc_list_double_count (l) ; ++i )
421 struct percentile *p = xmalloc (sizeof *p);
423 p->p = subc_list_double_at (l,i);
434 /* Parse the PERCENTILES subcommand */
436 xmn_custom_percentiles (struct lexer *lexer, struct dataset *ds UNUSED,
437 struct cmd_examine *p UNUSED, void *aux UNUSED)
441 lex_match (lexer, '=');
443 lex_match (lexer, '(');
445 while ( lex_is_number (lexer) )
447 subc_list_double_push (&percentile_list, lex_number (lexer));
451 lex_match (lexer, ',') ;
453 lex_match (lexer, ')');
455 lex_match (lexer, '=');
457 if ( lex_match_id (lexer, "HAVERAGE"))
458 percentile_algorithm = PC_HAVERAGE;
460 else if ( lex_match_id (lexer, "WAVERAGE"))
461 percentile_algorithm = PC_WAVERAGE;
463 else if ( lex_match_id (lexer, "ROUND"))
464 percentile_algorithm = PC_ROUND;
466 else if ( lex_match_id (lexer, "EMPIRICAL"))
467 percentile_algorithm = PC_EMPIRICAL;
469 else if ( lex_match_id (lexer, "AEMPIRICAL"))
470 percentile_algorithm = PC_AEMPIRICAL;
472 else if ( lex_match_id (lexer, "NONE"))
473 percentile_algorithm = PC_NONE;
476 if ( 0 == subc_list_double_count (&percentile_list))
478 subc_list_double_push (&percentile_list, 5);
479 subc_list_double_push (&percentile_list, 10);
480 subc_list_double_push (&percentile_list, 25);
481 subc_list_double_push (&percentile_list, 50);
482 subc_list_double_push (&percentile_list, 75);
483 subc_list_double_push (&percentile_list, 90);
484 subc_list_double_push (&percentile_list, 95);
490 /* TOTAL and NOTOTAL are simple, mutually exclusive flags */
492 xmn_custom_total (struct lexer *lexer UNUSED, struct dataset *ds UNUSED, struct cmd_examine *p, void *aux UNUSED)
494 if ( p->sbc_nototal )
496 msg (SE, _ ("%s and %s are mutually exclusive"),"TOTAL","NOTOTAL");
504 xmn_custom_nototal (struct lexer *lexer UNUSED, struct dataset *ds UNUSED,
505 struct cmd_examine *p, void *aux UNUSED)
509 msg (SE, _ ("%s and %s are mutually exclusive"),"TOTAL","NOTOTAL");
518 /* Parser for the variables sub command
519 Returns 1 on success */
521 xmn_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_examine *cmd, void *aux UNUSED)
523 const struct dictionary *dict = dataset_dict (ds);
524 lex_match (lexer, '=');
526 if ( (lex_token (lexer) != T_ID || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
527 && lex_token (lexer) != T_ALL)
532 if (!parse_variables (lexer, dict, &dependent_vars, &n_dependent_vars,
533 PV_NO_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH) )
535 free (dependent_vars);
539 assert (n_dependent_vars);
541 totals = xnmalloc (n_dependent_vars, sizeof *totals);
543 if ( lex_match (lexer, T_BY))
546 success = examine_parse_independent_vars (lexer, dict, cmd);
547 if ( success != 1 ) {
548 free (dependent_vars);
559 /* Parse the clause specifying the factors */
561 examine_parse_independent_vars (struct lexer *lexer, const struct dictionary *dict, struct cmd_examine *cmd)
564 struct factor *sf = xmalloc (sizeof *sf);
566 if ( (lex_token (lexer) != T_ID || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
567 && lex_token (lexer) != T_ALL)
574 sf->indep_var[0] = parse_variable (lexer, dict);
575 sf->indep_var[1] = 0;
577 if ( lex_token (lexer) == T_BY )
580 lex_match (lexer, T_BY);
582 if ( (lex_token (lexer) != T_ID || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
583 && lex_token (lexer) != T_ALL)
589 sf->indep_var[1] = parse_variable (lexer, dict);
594 sf->fstats = hsh_create (4,
595 (hsh_compare_func *) factor_statistics_compare,
596 (hsh_hash_func *) factor_statistics_hash,
597 (hsh_free_func *) factor_statistics_free,
603 lex_match (lexer, ',');
605 if ( lex_token (lexer) == '.' || lex_token (lexer) == '/' )
608 success = examine_parse_independent_vars (lexer, dict, cmd);
619 void populate_percentiles (struct tab_table *tbl, int col, int row,
620 const struct metrics *m);
622 void populate_descriptives (struct tab_table *t, int col, int row,
623 const struct metrics *fs);
625 void populate_extremes (struct tab_table *t, int col, int row, int n,
626 const struct metrics *m);
628 void populate_summary (struct tab_table *t, int col, int row,
629 const struct metrics *m);
634 static bool bad_weight_warn = true;
637 /* Perform calculations for the sub factors */
639 factor_calc (const struct ccase *c, int case_no, double weight,
643 struct factor *fctr = factors;
647 struct factor_statistics **foo ;
648 union value *indep_vals[2] ;
650 indep_vals[0] = value_dup (
651 case_data (c, fctr->indep_var[0]),
652 var_get_width (fctr->indep_var[0])
655 if ( fctr->indep_var[1] )
656 indep_vals[1] = value_dup (
657 case_data (c, fctr->indep_var[1]),
658 var_get_width (fctr->indep_var[1])
662 const union value sm = {SYSMIS};
663 indep_vals[1] = value_dup (&sm, 0);
666 assert (fctr->fstats);
668 foo = ( struct factor_statistics ** )
669 hsh_probe (fctr->fstats, (void *) &indep_vals);
674 *foo = create_factor_statistics (n_dependent_vars,
678 for ( v = 0 ; v < n_dependent_vars ; ++v )
680 metrics_precalc ( & (*foo)->m[v] );
686 free (indep_vals[0]);
687 free (indep_vals[1]);
690 for ( v = 0 ; v < n_dependent_vars ; ++v )
692 const struct variable *var = dependent_vars[v];
693 union value *val = value_dup (
698 if ( value_is_missing (var, val) || case_missing )
704 metrics_calc ( & (*foo)->m[v], val, weight, case_no);
714 run_examine (const struct ccase *first, const struct casefile *cf,
715 void *cmd_, const struct dataset *ds)
717 struct dictionary *dict = dataset_dict (ds);
718 struct casereader *r;
722 const struct cmd_examine *cmd = (struct cmd_examine *) cmd_;
726 output_split_file_values (ds, first);
728 /* Make sure we haven't got rubbish left over from a
733 struct factor *next = fctr->next;
735 hsh_clear (fctr->fstats);
742 for ( v = 0 ; v < n_dependent_vars ; ++v )
743 metrics_precalc (&totals[v]);
745 for (r = casefile_get_reader (cf, NULL);
746 casereader_read (r, &c) ;
750 const int case_no = casereader_cnum (r);
752 const double weight =
753 dict_get_case_weight (dict, &c, &bad_weight_warn);
755 if ( cmd->miss == XMN_LISTWISE )
757 for ( v = 0 ; v < n_dependent_vars ; ++v )
759 const struct variable *var = dependent_vars[v];
760 union value *val = value_dup (
765 if ( value_is_missing (var, val))
772 for ( v = 0 ; v < n_dependent_vars ; ++v )
774 const struct variable *var = dependent_vars[v];
775 union value *val = value_dup (
780 if ( value_is_missing (var, val) || case_missing )
786 metrics_calc (&totals[v], val, weight, case_no);
791 factor_calc (&c, case_no, weight, case_missing);
794 for ( v = 0 ; v < n_dependent_vars ; ++v)
799 struct hsh_iterator hi;
800 struct factor_statistics *fs;
802 for ( fs = hsh_first (fctr->fstats, &hi);
804 fs = hsh_next (fctr->fstats, &hi))
807 fs->m[v].ptile_hash = list_to_ptile_hash (&percentile_list);
808 fs->m[v].ptile_alg = percentile_algorithm;
809 metrics_postcalc (&fs->m[v]);
815 totals[v].ptile_hash = list_to_ptile_hash (&percentile_list);
816 totals[v].ptile_alg = percentile_algorithm;
817 metrics_postcalc (&totals[v]);
821 /* Make sure that the combination of factors are complete */
826 struct hsh_iterator hi;
827 struct hsh_iterator hi0;
828 struct hsh_iterator hi1;
829 struct factor_statistics *fs;
831 struct hsh_table *idh0=0;
832 struct hsh_table *idh1=0;
836 idh0 = hsh_create (4, (hsh_compare_func *) compare_values,
837 (hsh_hash_func *) hash_value,
840 idh1 = hsh_create (4, (hsh_compare_func *) compare_values,
841 (hsh_hash_func *) hash_value,
845 for ( fs = hsh_first (fctr->fstats, &hi);
847 fs = hsh_next (fctr->fstats, &hi))
849 hsh_insert (idh0, (void *) &fs->id[0]);
850 hsh_insert (idh1, (void *) &fs->id[1]);
853 /* Ensure that the factors combination is complete */
854 for ( val0 = hsh_first (idh0, &hi0);
856 val0 = hsh_next (idh0, &hi0))
858 for ( val1 = hsh_first (idh1, &hi1);
860 val1 = hsh_next (idh1, &hi1))
862 struct factor_statistics **ffs;
867 ffs = (struct factor_statistics **)
868 hsh_probe (fctr->fstats, (void *) &key );
872 (*ffs) = create_factor_statistics (n_dependent_vars,
874 for ( i = 0 ; i < n_dependent_vars ; ++i )
875 metrics_precalc ( & (*ffs)->m[i]);
883 fctr->fs = (struct factor_statistics **) hsh_sort_copy (fctr->fstats);
894 for ( i = 0 ; i < n_dependent_vars ; ++i )
896 metrics_destroy (&totals[i]);
905 show_summary (struct variable **dependent_var, int n_dep_var,
906 const struct factor *fctr)
908 static const char *subtitle[]=
916 int heading_columns ;
918 const int heading_rows = 3;
919 struct tab_table *tbl;
927 n_factors = hsh_count (fctr->fstats);
928 n_rows = n_dep_var * n_factors ;
930 if ( fctr->indep_var[1] )
939 n_rows += heading_rows;
941 n_cols = heading_columns + 6;
943 tbl = tab_create (n_cols,n_rows,0);
944 tab_headers (tbl, heading_columns, 0, heading_rows, 0);
946 tab_dim (tbl, tab_natural_dimensions);
948 /* Outline the box */
953 n_cols - 1, n_rows - 1);
955 /* Vertical lines for the data only */
960 n_cols - 1, n_rows - 1);
963 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows );
964 tab_hline (tbl, TAL_1, heading_columns, n_cols - 1, 1 );
965 tab_hline (tbl, TAL_1, heading_columns, n_cols - 1, heading_rows -1 );
967 tab_vline (tbl, TAL_2, heading_columns, 0, n_rows - 1);
970 tab_title (tbl, _ ("Case Processing Summary"));
973 tab_joint_text (tbl, heading_columns, 0,
975 TAB_CENTER | TAT_TITLE,
978 /* Remove lines ... */
985 for ( i = 0 ; i < 3 ; ++i )
987 tab_text (tbl, heading_columns + i*2 , 2, TAB_CENTER | TAT_TITLE,
990 tab_text (tbl, heading_columns + i*2 + 1, 2, TAB_CENTER | TAT_TITLE,
993 tab_joint_text (tbl, heading_columns + i*2 , 1,
994 heading_columns + i*2 + 1, 1,
995 TAB_CENTER | TAT_TITLE,
998 tab_box (tbl, -1, -1,
1000 heading_columns + i*2, 1,
1001 heading_columns + i*2 + 1, 1);
1006 /* Titles for the independent variables */
1009 tab_text (tbl, 1, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1010 var_to_string (fctr->indep_var[0]));
1012 if ( fctr->indep_var[1] )
1014 tab_text (tbl, 2, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1015 var_to_string (fctr->indep_var[1]));
1021 for ( i = 0 ; i < n_dep_var ; ++i )
1025 n_factors = hsh_count (fctr->fstats);
1029 tab_hline (tbl, TAL_1, 0, n_cols -1 , i * n_factors + heading_rows);
1032 0, i * n_factors + heading_rows,
1033 TAB_LEFT | TAT_TITLE,
1034 var_to_string (dependent_var[i])
1039 populate_summary (tbl, heading_columns,
1040 (i * n_factors) + heading_rows,
1046 struct factor_statistics **fs = fctr->fs;
1048 const union value *prev = NULL;
1053 0 != compare_values (prev, (*fs)->id[0],
1054 var_get_width (fctr->indep_var[0])))
1058 (i * n_factors ) + count +
1060 TAB_LEFT | TAT_TITLE,
1061 var_get_value_name (fctr->indep_var[0],
1065 if (fctr->indep_var[1] && count > 0 )
1066 tab_hline (tbl, TAL_1, 1, n_cols - 1,
1067 (i * n_factors ) + count + heading_rows);
1071 prev = (*fs)->id[0];
1074 if ( fctr->indep_var[1])
1077 (i * n_factors ) + count +
1079 TAB_LEFT | TAT_TITLE,
1080 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
1083 populate_summary (tbl, heading_columns,
1084 (i * n_factors) + count
1099 populate_summary (struct tab_table *t, int col, int row,
1100 const struct metrics *m)
1103 const double total = m->n + m->n_missing ;
1105 tab_float (t, col + 0, row + 0, TAB_RIGHT, m->n, 8, 0);
1106 tab_float (t, col + 2, row + 0, TAB_RIGHT, m->n_missing, 8, 0);
1107 tab_float (t, col + 4, row + 0, TAB_RIGHT, total, 8, 0);
1111 tab_text (t, col + 1, row + 0, TAB_RIGHT | TAT_PRINTF, "%2.0f%%",
1112 100.0 * m->n / total );
1114 tab_text (t, col + 3, row + 0, TAB_RIGHT | TAT_PRINTF, "%2.0f%%",
1115 100.0 * m->n_missing / total );
1117 /* This seems a bit pointless !!! */
1118 tab_text (t, col + 5, row + 0, TAB_RIGHT | TAT_PRINTF, "%2.0f%%",
1119 100.0 * total / total );
1130 show_extremes (struct variable **dependent_var, int n_dep_var,
1131 const struct factor *fctr, int n_extremities)
1134 int heading_columns ;
1136 const int heading_rows = 1;
1137 struct tab_table *tbl;
1144 heading_columns = 2;
1145 n_factors = hsh_count (fctr->fstats);
1147 n_rows = n_dep_var * 2 * n_extremities * n_factors;
1149 if ( fctr->indep_var[1] )
1150 heading_columns = 3;
1154 heading_columns = 1;
1155 n_rows = n_dep_var * 2 * n_extremities;
1158 n_rows += heading_rows;
1160 heading_columns += 2;
1161 n_cols = heading_columns + 2;
1163 tbl = tab_create (n_cols,n_rows,0);
1164 tab_headers (tbl, heading_columns, 0, heading_rows, 0);
1166 tab_dim (tbl, tab_natural_dimensions);
1168 /* Outline the box, No internal lines*/
1173 n_cols - 1, n_rows - 1);
1175 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows );
1177 tab_title (tbl, _ ("Extreme Values"));
1179 tab_vline (tbl, TAL_2, n_cols - 2, 0, n_rows -1);
1180 tab_vline (tbl, TAL_1, n_cols - 1, 0, n_rows -1);
1184 tab_text (tbl, 1, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1185 var_to_string (fctr->indep_var[0]));
1187 if ( fctr->indep_var[1] )
1188 tab_text (tbl, 2, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1189 var_to_string (fctr->indep_var[1]));
1192 tab_text (tbl, n_cols - 1, 0, TAB_CENTER | TAT_TITLE, _ ("Value"));
1193 tab_text (tbl, n_cols - 2, 0, TAB_CENTER | TAT_TITLE, _ ("Case Number"));
1195 for ( i = 0 ; i < n_dep_var ; ++i )
1199 tab_hline (tbl, TAL_1, 0, n_cols -1 ,
1200 i * 2 * n_extremities * n_factors + heading_rows);
1203 i * 2 * n_extremities * n_factors + heading_rows,
1204 TAB_LEFT | TAT_TITLE,
1205 var_to_string (dependent_var[i])
1210 populate_extremes (tbl, heading_columns - 2,
1211 i * 2 * n_extremities * n_factors + heading_rows,
1212 n_extremities, &totals[i]);
1216 struct factor_statistics **fs = fctr->fs;
1218 const union value *prev = NULL;
1222 const int row = heading_rows + ( 2 * n_extremities ) *
1223 ( ( i * n_factors ) + count );
1226 if ( !prev || 0 != compare_values (prev, (*fs)->id[0],
1227 var_get_width (fctr->indep_var[0])))
1231 tab_hline (tbl, TAL_1, 1, n_cols - 1, row);
1235 TAB_LEFT | TAT_TITLE,
1236 var_get_value_name (fctr->indep_var[0],
1241 prev = (*fs)->id[0];
1243 if (fctr->indep_var[1] && count > 0 )
1244 tab_hline (tbl, TAL_1, 2, n_cols - 1, row);
1246 if ( fctr->indep_var[1])
1247 tab_text (tbl, 2, row,
1248 TAB_LEFT | TAT_TITLE,
1249 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
1252 populate_extremes (tbl, heading_columns - 2,
1267 /* Fill in the extremities table */
1269 populate_extremes (struct tab_table *t,
1270 int col, int row, int n, const struct metrics *m)
1276 tab_text (t, col, row,
1277 TAB_RIGHT | TAT_TITLE ,
1281 tab_text (t, col, row + n ,
1282 TAB_RIGHT | TAT_TITLE ,
1287 tab_hline (t, TAL_1, col, col + 3, row + n );
1289 for (extremity = 0; extremity < n ; ++extremity )
1292 tab_float (t, col + 1, row + extremity,
1294 extremity + 1, 8, 0);
1298 tab_float (t, col + 1, row + extremity + n,
1300 extremity + 1, 8, 0);
1306 for (idx = 0, extremity = 0; extremity < n && idx < m->n_data ; ++idx )
1309 const struct weighted_value *wv = m->wvp[idx];
1310 struct case_node *cn = wv->case_nos;
1313 for (j = 0 ; j < wv->w ; ++j )
1315 if ( extremity + j >= n )
1318 tab_float (t, col + 3, row + extremity + j + n,
1322 tab_float (t, col + 2, row + extremity + j + n,
1331 extremity += wv->w ;
1336 for (idx = m->n_data - 1, extremity = 0; extremity < n && idx >= 0; --idx )
1339 const struct weighted_value *wv = m->wvp[idx];
1340 struct case_node *cn = wv->case_nos;
1342 for (j = 0 ; j < wv->w ; ++j )
1344 if ( extremity + j >= n )
1347 tab_float (t, col + 3, row + extremity + j,
1351 tab_float (t, col + 2, row + extremity + j,
1360 extremity += wv->w ;
1365 /* Show the descriptives table */
1367 show_descriptives (struct variable **dependent_var,
1369 struct factor *fctr)
1372 int heading_columns ;
1374 const int n_stat_rows = 13;
1376 const int heading_rows = 1;
1378 struct tab_table *tbl;
1385 heading_columns = 4;
1386 n_factors = hsh_count (fctr->fstats);
1388 n_rows = n_dep_var * n_stat_rows * n_factors;
1390 if ( fctr->indep_var[1] )
1391 heading_columns = 5;
1395 heading_columns = 3;
1396 n_rows = n_dep_var * n_stat_rows;
1399 n_rows += heading_rows;
1401 n_cols = heading_columns + 2;
1404 tbl = tab_create (n_cols, n_rows, 0);
1406 tab_headers (tbl, heading_columns + 1, 0, heading_rows, 0);
1408 tab_dim (tbl, tab_natural_dimensions);
1410 /* Outline the box and have no internal lines*/
1415 n_cols - 1, n_rows - 1);
1417 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows );
1419 tab_vline (tbl, TAL_1, heading_columns, 0, n_rows - 1);
1420 tab_vline (tbl, TAL_2, n_cols - 2, 0, n_rows - 1);
1421 tab_vline (tbl, TAL_1, n_cols - 1, 0, n_rows - 1);
1423 tab_text (tbl, n_cols - 2, 0, TAB_CENTER | TAT_TITLE, _ ("Statistic"));
1424 tab_text (tbl, n_cols - 1, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Error"));
1426 tab_title (tbl, _ ("Descriptives"));
1429 for ( i = 0 ; i < n_dep_var ; ++i )
1431 const int row = heading_rows + i * n_stat_rows * n_factors ;
1434 tab_hline (tbl, TAL_1, 0, n_cols - 1, row );
1437 i * n_stat_rows * n_factors + heading_rows,
1438 TAB_LEFT | TAT_TITLE,
1439 var_to_string (dependent_var[i])
1445 const union value *prev = NULL;
1447 struct factor_statistics **fs = fctr->fs;
1450 tab_text (tbl, 1, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1451 var_to_string (fctr->indep_var[0]));
1454 if ( fctr->indep_var[1])
1455 tab_text (tbl, 2, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1456 var_to_string (fctr->indep_var[1]));
1460 const int row = heading_rows + n_stat_rows *
1461 ( ( i * n_factors ) + count );
1464 if ( !prev || 0 != compare_values (prev, (*fs)->id[0],
1465 var_get_width (fctr->indep_var[0])))
1469 tab_hline (tbl, TAL_1, 1, n_cols - 1, row);
1473 TAB_LEFT | TAT_TITLE,
1474 var_get_value_name (fctr->indep_var[0],
1479 prev = (*fs)->id[0];
1481 if (fctr->indep_var[1] && count > 0 )
1482 tab_hline (tbl, TAL_1, 2, n_cols - 1, row);
1484 if ( fctr->indep_var[1])
1485 tab_text (tbl, 2, row,
1486 TAB_LEFT | TAT_TITLE,
1487 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
1490 populate_descriptives (tbl, heading_columns - 2,
1491 row, & (*fs)->m[i]);
1502 populate_descriptives (tbl, heading_columns - 2,
1503 i * n_stat_rows * n_factors + heading_rows,
1515 /* Fill in the descriptives data */
1517 populate_descriptives (struct tab_table *tbl, int col, int row,
1518 const struct metrics *m)
1521 const double t = gsl_cdf_tdist_Qinv (1 - cmd.n_cinterval[0]/100.0/2.0, \
1527 TAB_LEFT | TAT_TITLE,
1530 tab_float (tbl, col + 2,
1536 tab_float (tbl, col + 3,
1545 TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1546 _ ("%g%% Confidence Interval for Mean"), cmd.n_cinterval[0]);
1549 tab_text (tbl, col + 1,
1551 TAB_LEFT | TAT_TITLE,
1554 tab_float (tbl, col + 2,
1557 m->mean - t * m->se_mean,
1560 tab_text (tbl, col + 1,
1562 TAB_LEFT | TAT_TITLE,
1566 tab_float (tbl, col + 2,
1569 m->mean + t * m->se_mean,
1574 TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1575 _ ("5%% Trimmed Mean"));
1577 tab_float (tbl, col + 2,
1585 TAB_LEFT | TAT_TITLE,
1589 struct percentile *p;
1592 p = hsh_find (m->ptile_hash, &d);
1597 tab_float (tbl, col + 2,
1607 TAB_LEFT | TAT_TITLE,
1610 tab_float (tbl, col + 2,
1619 TAB_LEFT | TAT_TITLE,
1620 _ ("Std. Deviation"));
1623 tab_float (tbl, col + 2,
1632 TAB_LEFT | TAT_TITLE,
1635 tab_float (tbl, col + 2,
1643 TAB_LEFT | TAT_TITLE,
1646 tab_float (tbl, col + 2,
1655 TAB_LEFT | TAT_TITLE,
1659 tab_float (tbl, col + 2,
1667 TAB_LEFT | TAT_TITLE,
1668 _ ("Interquartile Range"));
1671 struct percentile *p1;
1672 struct percentile *p2;
1675 p1 = hsh_find (m->ptile_hash, &d);
1678 p2 = hsh_find (m->ptile_hash, &d);
1683 tab_float (tbl, col + 2,
1694 TAB_LEFT | TAT_TITLE,
1698 tab_float (tbl, col + 2,
1704 /* stderr of skewness */
1705 tab_float (tbl, col + 3,
1714 TAB_LEFT | TAT_TITLE,
1718 tab_float (tbl, col + 2,
1724 /* stderr of kurtosis */
1725 tab_float (tbl, col + 3,
1737 box_plot_variables (const struct factor *fctr,
1738 const struct variable **vars, int n_vars,
1739 const struct variable *id)
1743 struct factor_statistics **fs ;
1747 box_plot_group (fctr, vars, n_vars, id);
1751 for ( fs = fctr->fs ; *fs ; ++fs )
1753 double y_min = DBL_MAX;
1754 double y_max = -DBL_MAX;
1755 struct chart *ch = chart_create ();
1756 const char *s = factor_to_string (fctr, *fs, 0 );
1758 chart_write_title (ch, s);
1760 for ( i = 0 ; i < n_vars ; ++i )
1762 y_max = MAX (y_max, (*fs)->m[i].max);
1763 y_min = MIN (y_min, (*fs)->m[i].min);
1766 boxplot_draw_yscale (ch, y_max, y_min);
1768 for ( i = 0 ; i < n_vars ; ++i )
1771 const double box_width = (ch->data_right - ch->data_left)
1774 const double box_centre = ( i * 2 + 1) * box_width
1777 boxplot_draw_boxplot (ch,
1778 box_centre, box_width,
1780 var_to_string (vars[i]));
1792 /* Do a box plot, grouping all factors into one plot ;
1793 each dependent variable has its own plot.
1796 box_plot_group (const struct factor *fctr,
1797 const struct variable **vars,
1799 const struct variable *id UNUSED)
1804 for ( i = 0 ; i < n_vars ; ++i )
1806 struct factor_statistics **fs ;
1809 ch = chart_create ();
1811 boxplot_draw_yscale (ch, totals[i].max, totals[i].min);
1817 for ( fs = fctr->fs ; *fs ; ++fs )
1820 chart_write_title (ch, _ ("Boxplot of %s vs. %s"),
1821 var_to_string (vars[i]), var_to_string (fctr->indep_var[0]) );
1823 for ( fs = fctr->fs ; *fs ; ++fs )
1826 const char *s = factor_to_string_concise (fctr, *fs);
1828 const double box_width = (ch->data_right - ch->data_left)
1829 / (n_factors * 2.0 ) ;
1831 const double box_centre = ( f++ * 2 + 1) * box_width
1834 boxplot_draw_boxplot (ch,
1835 box_centre, box_width,
1842 const double box_width = (ch->data_right - ch->data_left) / 3.0;
1843 const double box_centre = (ch->data_right + ch->data_left) / 2.0;
1845 chart_write_title (ch, _ ("Boxplot"));
1847 boxplot_draw_boxplot (ch,
1848 box_centre, box_width,
1850 var_to_string (vars[i]) );
1859 /* Plot the normal and detrended normal plots for m
1860 Label the plots with factorname */
1862 np_plot (const struct metrics *m, const char *factorname)
1865 double yfirst=0, ylast=0;
1868 struct chart *np_chart;
1870 /* Detrended Normal Plot */
1871 struct chart *dnp_chart;
1873 /* The slope and intercept of the ideal normal probability line */
1874 const double slope = 1.0 / m->stddev;
1875 const double intercept = - m->mean / m->stddev;
1877 /* Cowardly refuse to plot an empty data set */
1878 if ( m->n_data == 0 )
1881 np_chart = chart_create ();
1882 dnp_chart = chart_create ();
1884 if ( !np_chart || ! dnp_chart )
1887 chart_write_title (np_chart, _ ("Normal Q-Q Plot of %s"), factorname);
1888 chart_write_xlabel (np_chart, _ ("Observed Value"));
1889 chart_write_ylabel (np_chart, _ ("Expected Normal"));
1892 chart_write_title (dnp_chart, _ ("Detrended Normal Q-Q Plot of %s"),
1894 chart_write_xlabel (dnp_chart, _ ("Observed Value"));
1895 chart_write_ylabel (dnp_chart, _ ("Dev from Normal"));
1897 yfirst = gsl_cdf_ugaussian_Pinv (m->wvp[0]->rank / ( m->n + 1));
1898 ylast = gsl_cdf_ugaussian_Pinv (m->wvp[m->n_data-1]->rank / ( m->n + 1));
1902 /* Need to make sure that both the scatter plot and the ideal fit into the
1904 double x_lower = MIN (m->min, (yfirst - intercept) / slope) ;
1905 double x_upper = MAX (m->max, (ylast - intercept) / slope) ;
1906 double slack = (x_upper - x_lower) * 0.05 ;
1908 chart_write_xscale (np_chart, x_lower - slack, x_upper + slack, 5);
1910 chart_write_xscale (dnp_chart, m->min, m->max, 5);
1914 chart_write_yscale (np_chart, yfirst, ylast, 5);
1917 /* We have to cache the detrended data, beacause we need to
1918 find its limits before we can plot it */
1919 double *d_data = xnmalloc (m->n_data, sizeof *d_data);
1920 double d_max = -DBL_MAX;
1921 double d_min = DBL_MAX;
1922 for ( i = 0 ; i < m->n_data; ++i )
1924 const double ns = gsl_cdf_ugaussian_Pinv (m->wvp[i]->rank / ( m->n + 1));
1926 chart_datum (np_chart, 0, m->wvp[i]->v.f, ns);
1928 d_data[i] = (m->wvp[i]->v.f - m->mean) / m->stddev - ns;
1930 if ( d_data[i] < d_min ) d_min = d_data[i];
1931 if ( d_data[i] > d_max ) d_max = d_data[i];
1933 chart_write_yscale (dnp_chart, d_min, d_max, 5);
1935 for ( i = 0 ; i < m->n_data; ++i )
1936 chart_datum (dnp_chart, 0, m->wvp[i]->v.f, d_data[i]);
1941 chart_line (np_chart, slope, intercept, yfirst, ylast , CHART_DIM_Y);
1942 chart_line (dnp_chart, 0, 0, m->min, m->max , CHART_DIM_X);
1944 chart_submit (np_chart);
1945 chart_submit (dnp_chart);
1951 /* Show the percentiles */
1953 show_percentiles (struct variable **dependent_var,
1955 struct factor *fctr)
1957 struct tab_table *tbl;
1963 struct hsh_table *ptiles ;
1965 int n_heading_columns;
1966 const int n_heading_rows = 2;
1967 const int n_stat_rows = 2;
1973 struct factor_statistics **fs = fctr->fs ;
1974 n_heading_columns = 3;
1975 n_factors = hsh_count (fctr->fstats);
1977 ptiles = (*fs)->m[0].ptile_hash;
1979 if ( fctr->indep_var[1] )
1980 n_heading_columns = 4;
1985 n_heading_columns = 2;
1987 ptiles = totals[0].ptile_hash;
1990 n_ptiles = hsh_count (ptiles);
1992 n_rows = n_heading_rows + n_dep_var * n_stat_rows * n_factors;
1994 n_cols = n_heading_columns + n_ptiles ;
1996 tbl = tab_create (n_cols, n_rows, 0);
1998 tab_headers (tbl, n_heading_columns + 1, 0, n_heading_rows, 0);
2000 tab_dim (tbl, tab_natural_dimensions);
2002 /* Outline the box and have no internal lines*/
2007 n_cols - 1, n_rows - 1);
2009 tab_hline (tbl, TAL_2, 0, n_cols - 1, n_heading_rows );
2011 tab_vline (tbl, TAL_2, n_heading_columns, 0, n_rows - 1);
2014 tab_title (tbl, _ ("Percentiles"));
2017 tab_hline (tbl, TAL_1, n_heading_columns, n_cols - 1, 1 );
2024 n_heading_columns - 1, n_rows - 1);
2030 n_heading_columns, n_heading_rows - 1,
2031 n_cols - 1, n_rows - 1);
2033 tab_joint_text (tbl, n_heading_columns + 1, 0,
2035 TAB_CENTER | TAT_TITLE ,
2040 /* Put in the percentile break points as headings */
2042 struct percentile **p = (struct percentile **) hsh_sort (ptiles);
2047 tab_float (tbl, n_heading_columns + i++ , 1,
2056 for ( i = 0 ; i < n_dep_var ; ++i )
2058 const int n_stat_rows = 2;
2059 const int row = n_heading_rows + i * n_stat_rows * n_factors ;
2062 tab_hline (tbl, TAL_1, 0, n_cols - 1, row );
2065 i * n_stat_rows * n_factors + n_heading_rows,
2066 TAB_LEFT | TAT_TITLE,
2067 var_to_string (dependent_var[i])
2072 const union value *prev = NULL ;
2073 struct factor_statistics **fs = fctr->fs;
2076 tab_text (tbl, 1, n_heading_rows - 1,
2077 TAB_CENTER | TAT_TITLE,
2078 var_to_string (fctr->indep_var[0]));
2081 if ( fctr->indep_var[1])
2082 tab_text (tbl, 2, n_heading_rows - 1, TAB_CENTER | TAT_TITLE,
2083 var_to_string (fctr->indep_var[1]));
2087 const int row = n_heading_rows + n_stat_rows *
2088 ( ( i * n_factors ) + count );
2091 if ( !prev || 0 != compare_values (prev, (*fs)->id[0],
2092 var_get_width (fctr->indep_var[0])))
2096 tab_hline (tbl, TAL_1, 1, n_cols - 1, row);
2100 TAB_LEFT | TAT_TITLE,
2101 var_get_value_name (fctr->indep_var[0],
2108 prev = (*fs)->id[0];
2110 if (fctr->indep_var[1] && count > 0 )
2111 tab_hline (tbl, TAL_1, 2, n_cols - 1, row);
2113 if ( fctr->indep_var[1])
2114 tab_text (tbl, 2, row,
2115 TAB_LEFT | TAT_TITLE,
2116 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
2120 populate_percentiles (tbl, n_heading_columns - 1,
2121 row, & (*fs)->m[i]);
2132 populate_percentiles (tbl, n_heading_columns - 1,
2133 i * n_stat_rows * n_factors + n_heading_rows,
2150 populate_percentiles (struct tab_table *tbl, int col, int row,
2151 const struct metrics *m)
2155 struct percentile **p = (struct percentile **) hsh_sort (m->ptile_hash);
2159 TAB_LEFT | TAT_TITLE,
2160 _ ("Tukey\'s Hinges")
2165 TAB_LEFT | TAT_TITLE,
2166 ptile_alg_desc[m->ptile_alg]
2173 tab_float (tbl, col + i + 1 , row,
2176 if ( (*p)->p == 25 )
2177 tab_float (tbl, col + i + 1 , row + 1,
2181 if ( (*p)->p == 50 )
2182 tab_float (tbl, col + i + 1 , row + 1,
2186 if ( (*p)->p == 75 )
2187 tab_float (tbl, col + i + 1 , row + 1,
2202 factor_to_string (const struct factor *fctr,
2203 const struct factor_statistics *fs,
2204 const struct variable *var)
2207 static char buf1[100];
2213 sprintf (buf1, "%s (",var_to_string (var) );
2216 snprintf (buf2, 100, "%s = %s",
2217 var_to_string (fctr->indep_var[0]),
2218 var_get_value_name (fctr->indep_var[0], fs->id[0]));
2220 strcat (buf1, buf2);
2222 if ( fctr->indep_var[1] )
2224 sprintf (buf2, "; %s = %s)",
2225 var_to_string (fctr->indep_var[1]),
2226 var_get_value_name (fctr->indep_var[1], fs->id[1]));
2227 strcat (buf1, buf2);
2241 factor_to_string_concise (const struct factor *fctr,
2242 struct factor_statistics *fs)
2246 static char buf[100];
2250 snprintf (buf, 100, "%s",
2251 var_get_value_name (fctr->indep_var[0], fs->id[0]));
2253 if ( fctr->indep_var[1] )
2255 sprintf (buf2, ",%s)", var_get_value_name (fctr->indep_var[1],