1 /* PSPP - EXAMINE data for normality . -*-c-*-
3 Copyright (C) 2004 Free Software Foundation, Inc.
4 Author: John Darrington 2004, 2006
6 This program is free software; you can redistribute it and/or
7 modify it under the terms of the GNU General Public License as
8 published by the Free Software Foundation; either version 2 of the
9 License, or (at your option) any later version.
11 This program is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
23 #include <gsl/gsl_cdf.h>
24 #include <libpspp/message.h>
29 #include <data/case.h>
30 #include <data/casefile.h>
31 #include <data/dictionary.h>
32 #include <data/procedure.h>
33 #include <data/value-labels.h>
34 #include <data/variable.h>
35 #include <language/command.h>
36 #include <language/dictionary/split-file.h>
37 #include <language/lexer/lexer.h>
38 #include <libpspp/alloc.h>
39 #include <libpspp/compiler.h>
40 #include <libpspp/hash.h>
41 #include <libpspp/magic.h>
42 #include <libpspp/message.h>
43 #include <libpspp/misc.h>
44 #include <libpspp/str.h>
45 #include <math/factor-stats.h>
46 #include <math/moments.h>
47 #include <math/percentiles.h>
48 #include <output/charts/box-whisker.h>
49 #include <output/charts/cartesian.h>
50 #include <output/manager.h>
51 #include <output/table.h>
56 #define _(msgid) gettext (msgid)
57 #define N_(msgid) msgid
60 #include <output/chart.h>
61 #include <output/charts/plot-hist.h>
62 #include <output/charts/plot-chart.h>
69 missing=miss:pairwise/!listwise,
71 incl:include/!exclude;
72 +compare=cmp:variables/!groups;
75 +plot[plt_]=stemleaf,boxplot,npplot,:spreadlevel(*d:n),histogram,all,none;
77 +statistics[st_]=descriptives,:extreme(*d:n),all,none.
86 static struct cmd_examine cmd;
88 static struct variable **dependent_vars;
90 static size_t n_dependent_vars;
95 /* The independent variable */
96 struct variable *indep_var[2];
99 /* Hash table of factor stats indexed by 2 values */
100 struct hsh_table *fstats;
102 /* The hash table after it has been crunched */
103 struct factor_statistics **fs;
109 /* Linked list of factors */
110 static struct factor *factors = 0;
112 static struct metrics *totals = 0;
114 /* Parse the clause specifying the factors */
115 static int examine_parse_independent_vars (struct lexer *lexer, const struct dictionary *dict, struct cmd_examine *cmd);
119 /* Output functions */
120 static void show_summary (struct variable **dependent_var, int n_dep_var,
121 const struct factor *f);
123 static void show_extremes (struct variable **dependent_var,
125 const struct factor *factor,
128 static void show_descriptives (struct variable **dependent_var,
130 struct factor *factor);
132 static void show_percentiles (struct variable **dependent_var,
134 struct factor *factor);
139 void np_plot (const struct metrics *m, const char *factorname);
142 void box_plot_group (const struct factor *fctr,
143 const struct variable **vars, int n_vars,
144 const struct variable *id
148 void box_plot_variables (const struct factor *fctr,
149 const struct variable **vars, int n_vars,
150 const struct variable *id
155 /* Per Split function */
156 static bool run_examine (const struct ccase *,
157 const struct casefile *cf, void *cmd_, const struct dataset *);
159 static void output_examine (void);
162 void factor_calc (const struct ccase *c, int case_no,
163 double weight, int case_missing);
166 /* Represent a factor as a string, so it can be
167 printed in a human readable fashion */
168 const char * factor_to_string (const struct factor *fctr,
169 const struct factor_statistics *fs,
170 const struct variable *var);
173 /* Represent a factor as a string, so it can be
174 printed in a human readable fashion,
175 but sacrificing some readablility for the sake of brevity */
176 const char *factor_to_string_concise (const struct factor *fctr,
177 struct factor_statistics *fs);
182 /* Function to use for testing for missing values */
183 static var_is_missing_func *value_is_missing;
188 static subc_list_double percentile_list;
190 static enum pc_alg percentile_algorithm;
192 static short sbc_percentile;
196 cmd_examine (struct lexer *lexer, struct dataset *ds)
200 subc_list_double_create (&percentile_list);
201 percentile_algorithm = PC_HAVERAGE;
203 if ( !parse_examine (lexer, ds, &cmd, NULL) )
205 subc_list_double_destroy (&percentile_list);
209 /* If /MISSING=INCLUDE is set, then user missing values are ignored */
210 if (cmd.incl == XMN_INCLUDE )
211 value_is_missing = var_is_value_system_missing;
213 value_is_missing = var_is_value_missing;
215 if ( cmd.st_n == SYSMIS )
218 if ( ! cmd.sbc_cinterval)
219 cmd.n_cinterval[0] = 95.0;
221 /* If descriptives have been requested, make sure the
222 quartiles are calculated */
223 if ( cmd.a_statistics[XMN_ST_DESCRIPTIVES] )
225 subc_list_double_push (&percentile_list, 25);
226 subc_list_double_push (&percentile_list, 50);
227 subc_list_double_push (&percentile_list, 75);
230 ok = multipass_procedure_with_splits (ds, run_examine, &cmd);
237 if ( dependent_vars )
238 free (dependent_vars);
241 struct factor *f = factors ;
244 struct factor *ff = f;
248 hsh_destroy ( ff->fstats ) ;
254 subc_list_double_destroy (&percentile_list);
256 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
261 /* Show all the appropriate tables */
263 output_examine (void)
267 /* Show totals if appropriate */
268 if ( ! cmd.sbc_nototal || factors == 0 )
270 show_summary (dependent_vars, n_dependent_vars, 0);
272 if ( cmd.sbc_statistics )
274 if ( cmd.a_statistics[XMN_ST_EXTREME])
275 show_extremes (dependent_vars, n_dependent_vars, 0, cmd.st_n);
277 if ( cmd.a_statistics[XMN_ST_DESCRIPTIVES])
278 show_descriptives (dependent_vars, n_dependent_vars, 0);
281 if ( sbc_percentile )
282 show_percentiles (dependent_vars, n_dependent_vars, 0);
287 if ( cmd.a_plot[XMN_PLT_STEMLEAF] )
288 msg (SW, _ ("%s is not currently supported."), "STEMLEAF");
290 if ( cmd.a_plot[XMN_PLT_SPREADLEVEL] )
291 msg (SW, _ ("%s is not currently supported."), "SPREADLEVEL");
293 if ( cmd.a_plot[XMN_PLT_NPPLOT] )
295 for ( v = 0 ; v < n_dependent_vars; ++v )
296 np_plot (&totals[v], var_to_string (dependent_vars[v]));
299 if ( cmd.a_plot[XMN_PLT_BOXPLOT] )
301 if ( cmd.cmp == XMN_GROUPS )
303 box_plot_group (0, (const struct variable **) dependent_vars,
304 n_dependent_vars, cmd.v_id);
307 box_plot_variables (0,
308 (const struct variable **) dependent_vars,
309 n_dependent_vars, cmd.v_id);
312 if ( cmd.a_plot[XMN_PLT_HISTOGRAM] )
314 for ( v = 0 ; v < n_dependent_vars; ++v )
316 struct normal_curve normal;
318 normal.N = totals[v].n;
319 normal.mean = totals[v].mean;
320 normal.stddev = totals[v].stddev;
322 histogram_plot (totals[v].histogram,
323 var_to_string (dependent_vars[v]),
333 /* Show grouped statistics as appropriate */
337 show_summary (dependent_vars, n_dependent_vars, fctr);
339 if ( cmd.sbc_statistics )
341 if ( cmd.a_statistics[XMN_ST_EXTREME])
342 show_extremes (dependent_vars, n_dependent_vars, fctr, cmd.st_n);
344 if ( cmd.a_statistics[XMN_ST_DESCRIPTIVES])
345 show_descriptives (dependent_vars, n_dependent_vars, fctr);
348 if ( sbc_percentile )
349 show_percentiles (dependent_vars, n_dependent_vars, fctr);
356 struct factor_statistics **fs = fctr->fs ;
358 if ( cmd.a_plot[XMN_PLT_BOXPLOT] )
360 if ( cmd.cmp == XMN_VARIABLES )
361 box_plot_variables (fctr,
362 (const struct variable **) dependent_vars,
363 n_dependent_vars, cmd.v_id);
365 box_plot_group (fctr,
366 (const struct variable **) dependent_vars,
367 n_dependent_vars, cmd.v_id);
370 for ( v = 0 ; v < n_dependent_vars; ++v )
373 for ( fs = fctr->fs ; *fs ; ++fs )
375 const char *s = factor_to_string (fctr, *fs, dependent_vars[v]);
377 if ( cmd.a_plot[XMN_PLT_NPPLOT] )
378 np_plot (& (*fs)->m[v], s);
380 if ( cmd.a_plot[XMN_PLT_HISTOGRAM] )
382 struct normal_curve normal;
384 normal.N = (*fs)->m[v].n;
385 normal.mean = (*fs)->m[v].mean;
386 normal.stddev = (*fs)->m[v].stddev;
388 histogram_plot ((*fs)->m[v].histogram,
392 } /* for ( fs .... */
394 } /* for ( v = 0 ..... */
404 /* Create a hash table of percentiles and their values from the list of
406 static struct hsh_table *
407 list_to_ptile_hash (const subc_list_double *l)
411 struct hsh_table *h ;
413 h = hsh_create (subc_list_double_count (l),
414 (hsh_compare_func *) ptile_compare,
415 (hsh_hash_func *) ptile_hash,
416 (hsh_free_func *) free,
420 for ( i = 0 ; i < subc_list_double_count (l) ; ++i )
422 struct percentile *p = xmalloc (sizeof *p);
424 p->p = subc_list_double_at (l,i);
435 /* Parse the PERCENTILES subcommand */
437 xmn_custom_percentiles (struct lexer *lexer, struct dataset *ds UNUSED,
438 struct cmd_examine *p UNUSED, void *aux UNUSED)
442 lex_match (lexer, '=');
444 lex_match (lexer, '(');
446 while ( lex_is_number (lexer) )
448 subc_list_double_push (&percentile_list, lex_number (lexer));
452 lex_match (lexer, ',') ;
454 lex_match (lexer, ')');
456 lex_match (lexer, '=');
458 if ( lex_match_id (lexer, "HAVERAGE"))
459 percentile_algorithm = PC_HAVERAGE;
461 else if ( lex_match_id (lexer, "WAVERAGE"))
462 percentile_algorithm = PC_WAVERAGE;
464 else if ( lex_match_id (lexer, "ROUND"))
465 percentile_algorithm = PC_ROUND;
467 else if ( lex_match_id (lexer, "EMPIRICAL"))
468 percentile_algorithm = PC_EMPIRICAL;
470 else if ( lex_match_id (lexer, "AEMPIRICAL"))
471 percentile_algorithm = PC_AEMPIRICAL;
473 else if ( lex_match_id (lexer, "NONE"))
474 percentile_algorithm = PC_NONE;
477 if ( 0 == subc_list_double_count (&percentile_list))
479 subc_list_double_push (&percentile_list, 5);
480 subc_list_double_push (&percentile_list, 10);
481 subc_list_double_push (&percentile_list, 25);
482 subc_list_double_push (&percentile_list, 50);
483 subc_list_double_push (&percentile_list, 75);
484 subc_list_double_push (&percentile_list, 90);
485 subc_list_double_push (&percentile_list, 95);
491 /* TOTAL and NOTOTAL are simple, mutually exclusive flags */
493 xmn_custom_total (struct lexer *lexer UNUSED, struct dataset *ds UNUSED, struct cmd_examine *p, void *aux UNUSED)
495 if ( p->sbc_nototal )
497 msg (SE, _ ("%s and %s are mutually exclusive"),"TOTAL","NOTOTAL");
505 xmn_custom_nototal (struct lexer *lexer UNUSED, struct dataset *ds UNUSED,
506 struct cmd_examine *p, void *aux UNUSED)
510 msg (SE, _ ("%s and %s are mutually exclusive"),"TOTAL","NOTOTAL");
519 /* Parser for the variables sub command
520 Returns 1 on success */
522 xmn_custom_variables (struct lexer *lexer, struct dataset *ds, struct cmd_examine *cmd, void *aux UNUSED)
524 const struct dictionary *dict = dataset_dict (ds);
525 lex_match (lexer, '=');
527 if ( (lex_token (lexer) != T_ID || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
528 && lex_token (lexer) != T_ALL)
533 if (!parse_variables (lexer, dict, &dependent_vars, &n_dependent_vars,
534 PV_NO_DUPLICATE | PV_NUMERIC | PV_NO_SCRATCH) )
536 free (dependent_vars);
540 assert (n_dependent_vars);
542 totals = xnmalloc (n_dependent_vars, sizeof *totals);
544 if ( lex_match (lexer, T_BY))
547 success = examine_parse_independent_vars (lexer, dict, cmd);
548 if ( success != 1 ) {
549 free (dependent_vars);
560 /* Parse the clause specifying the factors */
562 examine_parse_independent_vars (struct lexer *lexer, const struct dictionary *dict, struct cmd_examine *cmd)
565 struct factor *sf = xmalloc (sizeof *sf);
567 if ( (lex_token (lexer) != T_ID || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
568 && lex_token (lexer) != T_ALL)
575 sf->indep_var[0] = parse_variable (lexer, dict);
576 sf->indep_var[1] = 0;
578 if ( lex_token (lexer) == T_BY )
581 lex_match (lexer, T_BY);
583 if ( (lex_token (lexer) != T_ID || dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
584 && lex_token (lexer) != T_ALL)
590 sf->indep_var[1] = parse_variable (lexer, dict);
595 sf->fstats = hsh_create (4,
596 (hsh_compare_func *) factor_statistics_compare,
597 (hsh_hash_func *) factor_statistics_hash,
598 (hsh_free_func *) factor_statistics_free,
604 lex_match (lexer, ',');
606 if ( lex_token (lexer) == '.' || lex_token (lexer) == '/' )
609 success = examine_parse_independent_vars (lexer, dict, cmd);
620 void populate_percentiles (struct tab_table *tbl, int col, int row,
621 const struct metrics *m);
623 void populate_descriptives (struct tab_table *t, int col, int row,
624 const struct metrics *fs);
626 void populate_extremes (struct tab_table *t, int col, int row, int n,
627 const struct metrics *m);
629 void populate_summary (struct tab_table *t, int col, int row,
630 const struct metrics *m);
635 static bool bad_weight_warn = true;
638 /* Perform calculations for the sub factors */
640 factor_calc (const struct ccase *c, int case_no, double weight,
644 struct factor *fctr = factors;
648 struct factor_statistics **foo ;
649 union value *indep_vals[2] ;
651 indep_vals[0] = value_dup (
652 case_data (c, fctr->indep_var[0]),
653 var_get_width (fctr->indep_var[0])
656 if ( fctr->indep_var[1] )
657 indep_vals[1] = value_dup (
658 case_data (c, fctr->indep_var[1]),
659 var_get_width (fctr->indep_var[1])
663 const union value sm = {SYSMIS};
664 indep_vals[1] = value_dup (&sm, 0);
667 assert (fctr->fstats);
669 foo = ( struct factor_statistics ** )
670 hsh_probe (fctr->fstats, (void *) &indep_vals);
675 *foo = create_factor_statistics (n_dependent_vars,
679 for ( v = 0 ; v < n_dependent_vars ; ++v )
681 metrics_precalc ( & (*foo)->m[v] );
687 free (indep_vals[0]);
688 free (indep_vals[1]);
691 for ( v = 0 ; v < n_dependent_vars ; ++v )
693 const struct variable *var = dependent_vars[v];
694 union value *val = value_dup (
699 if ( value_is_missing (var, val) || case_missing )
705 metrics_calc ( & (*foo)->m[v], val, weight, case_no);
715 run_examine (const struct ccase *first, const struct casefile *cf,
716 void *cmd_, const struct dataset *ds)
718 struct dictionary *dict = dataset_dict (ds);
719 struct casereader *r;
723 const struct cmd_examine *cmd = (struct cmd_examine *) cmd_;
727 output_split_file_values (ds, first);
729 /* Make sure we haven't got rubbish left over from a
734 struct factor *next = fctr->next;
736 hsh_clear (fctr->fstats);
743 for ( v = 0 ; v < n_dependent_vars ; ++v )
744 metrics_precalc (&totals[v]);
746 for (r = casefile_get_reader (cf, NULL);
747 casereader_read (r, &c) ;
751 const int case_no = casereader_cnum (r);
753 const double weight =
754 dict_get_case_weight (dict, &c, &bad_weight_warn);
756 if ( cmd->miss == XMN_LISTWISE )
758 for ( v = 0 ; v < n_dependent_vars ; ++v )
760 const struct variable *var = dependent_vars[v];
761 union value *val = value_dup (
766 if ( value_is_missing (var, val))
773 for ( v = 0 ; v < n_dependent_vars ; ++v )
775 const struct variable *var = dependent_vars[v];
776 union value *val = value_dup (
781 if ( value_is_missing (var, val) || case_missing )
787 metrics_calc (&totals[v], val, weight, case_no);
792 factor_calc (&c, case_no, weight, case_missing);
795 for ( v = 0 ; v < n_dependent_vars ; ++v)
800 struct hsh_iterator hi;
801 struct factor_statistics *fs;
803 for ( fs = hsh_first (fctr->fstats, &hi);
805 fs = hsh_next (fctr->fstats, &hi))
808 fs->m[v].ptile_hash = list_to_ptile_hash (&percentile_list);
809 fs->m[v].ptile_alg = percentile_algorithm;
810 metrics_postcalc (&fs->m[v]);
816 totals[v].ptile_hash = list_to_ptile_hash (&percentile_list);
817 totals[v].ptile_alg = percentile_algorithm;
818 metrics_postcalc (&totals[v]);
822 /* Make sure that the combination of factors are complete */
827 struct hsh_iterator hi;
828 struct hsh_iterator hi0;
829 struct hsh_iterator hi1;
830 struct factor_statistics *fs;
832 struct hsh_table *idh0=0;
833 struct hsh_table *idh1=0;
837 idh0 = hsh_create (4, (hsh_compare_func *) compare_values,
838 (hsh_hash_func *) hash_value,
841 idh1 = hsh_create (4, (hsh_compare_func *) compare_values,
842 (hsh_hash_func *) hash_value,
846 for ( fs = hsh_first (fctr->fstats, &hi);
848 fs = hsh_next (fctr->fstats, &hi))
850 hsh_insert (idh0, (void *) &fs->id[0]);
851 hsh_insert (idh1, (void *) &fs->id[1]);
854 /* Ensure that the factors combination is complete */
855 for ( val0 = hsh_first (idh0, &hi0);
857 val0 = hsh_next (idh0, &hi0))
859 for ( val1 = hsh_first (idh1, &hi1);
861 val1 = hsh_next (idh1, &hi1))
863 struct factor_statistics **ffs;
868 ffs = (struct factor_statistics **)
869 hsh_probe (fctr->fstats, (void *) &key );
873 (*ffs) = create_factor_statistics (n_dependent_vars,
875 for ( i = 0 ; i < n_dependent_vars ; ++i )
876 metrics_precalc ( & (*ffs)->m[i]);
884 fctr->fs = (struct factor_statistics **) hsh_sort_copy (fctr->fstats);
895 for ( i = 0 ; i < n_dependent_vars ; ++i )
897 metrics_destroy (&totals[i]);
906 show_summary (struct variable **dependent_var, int n_dep_var,
907 const struct factor *fctr)
909 static const char *subtitle[]=
917 int heading_columns ;
919 const int heading_rows = 3;
920 struct tab_table *tbl;
928 n_factors = hsh_count (fctr->fstats);
929 n_rows = n_dep_var * n_factors ;
931 if ( fctr->indep_var[1] )
940 n_rows += heading_rows;
942 n_cols = heading_columns + 6;
944 tbl = tab_create (n_cols,n_rows,0);
945 tab_headers (tbl, heading_columns, 0, heading_rows, 0);
947 tab_dim (tbl, tab_natural_dimensions);
949 /* Outline the box */
954 n_cols - 1, n_rows - 1);
956 /* Vertical lines for the data only */
961 n_cols - 1, n_rows - 1);
964 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows );
965 tab_hline (tbl, TAL_1, heading_columns, n_cols - 1, 1 );
966 tab_hline (tbl, TAL_1, heading_columns, n_cols - 1, heading_rows -1 );
968 tab_vline (tbl, TAL_2, heading_columns, 0, n_rows - 1);
971 tab_title (tbl, _ ("Case Processing Summary"));
974 tab_joint_text (tbl, heading_columns, 0,
976 TAB_CENTER | TAT_TITLE,
979 /* Remove lines ... */
986 for ( i = 0 ; i < 3 ; ++i )
988 tab_text (tbl, heading_columns + i*2 , 2, TAB_CENTER | TAT_TITLE,
991 tab_text (tbl, heading_columns + i*2 + 1, 2, TAB_CENTER | TAT_TITLE,
994 tab_joint_text (tbl, heading_columns + i*2 , 1,
995 heading_columns + i*2 + 1, 1,
996 TAB_CENTER | TAT_TITLE,
999 tab_box (tbl, -1, -1,
1001 heading_columns + i*2, 1,
1002 heading_columns + i*2 + 1, 1);
1007 /* Titles for the independent variables */
1010 tab_text (tbl, 1, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1011 var_to_string (fctr->indep_var[0]));
1013 if ( fctr->indep_var[1] )
1015 tab_text (tbl, 2, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1016 var_to_string (fctr->indep_var[1]));
1022 for ( i = 0 ; i < n_dep_var ; ++i )
1026 n_factors = hsh_count (fctr->fstats);
1030 tab_hline (tbl, TAL_1, 0, n_cols -1 , i * n_factors + heading_rows);
1033 0, i * n_factors + heading_rows,
1034 TAB_LEFT | TAT_TITLE,
1035 var_to_string (dependent_var[i])
1040 populate_summary (tbl, heading_columns,
1041 (i * n_factors) + heading_rows,
1047 struct factor_statistics **fs = fctr->fs;
1049 const union value *prev = NULL;
1054 0 != compare_values (prev, (*fs)->id[0],
1055 var_get_width (fctr->indep_var[0])))
1059 (i * n_factors ) + count +
1061 TAB_LEFT | TAT_TITLE,
1062 var_get_value_name (fctr->indep_var[0],
1066 if (fctr->indep_var[1] && count > 0 )
1067 tab_hline (tbl, TAL_1, 1, n_cols - 1,
1068 (i * n_factors ) + count + heading_rows);
1072 prev = (*fs)->id[0];
1075 if ( fctr->indep_var[1])
1078 (i * n_factors ) + count +
1080 TAB_LEFT | TAT_TITLE,
1081 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
1084 populate_summary (tbl, heading_columns,
1085 (i * n_factors) + count
1100 populate_summary (struct tab_table *t, int col, int row,
1101 const struct metrics *m)
1104 const double total = m->n + m->n_missing ;
1106 tab_float (t, col + 0, row + 0, TAB_RIGHT, m->n, 8, 0);
1107 tab_float (t, col + 2, row + 0, TAB_RIGHT, m->n_missing, 8, 0);
1108 tab_float (t, col + 4, row + 0, TAB_RIGHT, total, 8, 0);
1112 tab_text (t, col + 1, row + 0, TAB_RIGHT | TAT_PRINTF, "%2.0f%%",
1113 100.0 * m->n / total );
1115 tab_text (t, col + 3, row + 0, TAB_RIGHT | TAT_PRINTF, "%2.0f%%",
1116 100.0 * m->n_missing / total );
1118 /* This seems a bit pointless !!! */
1119 tab_text (t, col + 5, row + 0, TAB_RIGHT | TAT_PRINTF, "%2.0f%%",
1120 100.0 * total / total );
1131 show_extremes (struct variable **dependent_var, int n_dep_var,
1132 const struct factor *fctr, int n_extremities)
1135 int heading_columns ;
1137 const int heading_rows = 1;
1138 struct tab_table *tbl;
1145 heading_columns = 2;
1146 n_factors = hsh_count (fctr->fstats);
1148 n_rows = n_dep_var * 2 * n_extremities * n_factors;
1150 if ( fctr->indep_var[1] )
1151 heading_columns = 3;
1155 heading_columns = 1;
1156 n_rows = n_dep_var * 2 * n_extremities;
1159 n_rows += heading_rows;
1161 heading_columns += 2;
1162 n_cols = heading_columns + 2;
1164 tbl = tab_create (n_cols,n_rows,0);
1165 tab_headers (tbl, heading_columns, 0, heading_rows, 0);
1167 tab_dim (tbl, tab_natural_dimensions);
1169 /* Outline the box, No internal lines*/
1174 n_cols - 1, n_rows - 1);
1176 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows );
1178 tab_title (tbl, _ ("Extreme Values"));
1180 tab_vline (tbl, TAL_2, n_cols - 2, 0, n_rows -1);
1181 tab_vline (tbl, TAL_1, n_cols - 1, 0, n_rows -1);
1185 tab_text (tbl, 1, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1186 var_to_string (fctr->indep_var[0]));
1188 if ( fctr->indep_var[1] )
1189 tab_text (tbl, 2, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1190 var_to_string (fctr->indep_var[1]));
1193 tab_text (tbl, n_cols - 1, 0, TAB_CENTER | TAT_TITLE, _ ("Value"));
1194 tab_text (tbl, n_cols - 2, 0, TAB_CENTER | TAT_TITLE, _ ("Case Number"));
1196 for ( i = 0 ; i < n_dep_var ; ++i )
1200 tab_hline (tbl, TAL_1, 0, n_cols -1 ,
1201 i * 2 * n_extremities * n_factors + heading_rows);
1204 i * 2 * n_extremities * n_factors + heading_rows,
1205 TAB_LEFT | TAT_TITLE,
1206 var_to_string (dependent_var[i])
1211 populate_extremes (tbl, heading_columns - 2,
1212 i * 2 * n_extremities * n_factors + heading_rows,
1213 n_extremities, &totals[i]);
1217 struct factor_statistics **fs = fctr->fs;
1219 const union value *prev = NULL;
1223 const int row = heading_rows + ( 2 * n_extremities ) *
1224 ( ( i * n_factors ) + count );
1227 if ( !prev || 0 != compare_values (prev, (*fs)->id[0],
1228 var_get_width (fctr->indep_var[0])))
1232 tab_hline (tbl, TAL_1, 1, n_cols - 1, row);
1236 TAB_LEFT | TAT_TITLE,
1237 var_get_value_name (fctr->indep_var[0],
1242 prev = (*fs)->id[0];
1244 if (fctr->indep_var[1] && count > 0 )
1245 tab_hline (tbl, TAL_1, 2, n_cols - 1, row);
1247 if ( fctr->indep_var[1])
1248 tab_text (tbl, 2, row,
1249 TAB_LEFT | TAT_TITLE,
1250 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
1253 populate_extremes (tbl, heading_columns - 2,
1268 /* Fill in the extremities table */
1270 populate_extremes (struct tab_table *t,
1271 int col, int row, int n, const struct metrics *m)
1277 tab_text (t, col, row,
1278 TAB_RIGHT | TAT_TITLE ,
1282 tab_text (t, col, row + n ,
1283 TAB_RIGHT | TAT_TITLE ,
1288 tab_hline (t, TAL_1, col, col + 3, row + n );
1290 for (extremity = 0; extremity < n ; ++extremity )
1293 tab_float (t, col + 1, row + extremity,
1295 extremity + 1, 8, 0);
1299 tab_float (t, col + 1, row + extremity + n,
1301 extremity + 1, 8, 0);
1307 for (idx = 0, extremity = 0; extremity < n && idx < m->n_data ; ++idx )
1310 const struct weighted_value *wv = m->wvp[idx];
1311 struct case_node *cn = wv->case_nos;
1314 for (j = 0 ; j < wv->w ; ++j )
1316 if ( extremity + j >= n )
1319 tab_float (t, col + 3, row + extremity + j + n,
1323 tab_float (t, col + 2, row + extremity + j + n,
1332 extremity += wv->w ;
1337 for (idx = m->n_data - 1, extremity = 0; extremity < n && idx >= 0; --idx )
1340 const struct weighted_value *wv = m->wvp[idx];
1341 struct case_node *cn = wv->case_nos;
1343 for (j = 0 ; j < wv->w ; ++j )
1345 if ( extremity + j >= n )
1348 tab_float (t, col + 3, row + extremity + j,
1352 tab_float (t, col + 2, row + extremity + j,
1361 extremity += wv->w ;
1366 /* Show the descriptives table */
1368 show_descriptives (struct variable **dependent_var,
1370 struct factor *fctr)
1373 int heading_columns ;
1375 const int n_stat_rows = 13;
1377 const int heading_rows = 1;
1379 struct tab_table *tbl;
1386 heading_columns = 4;
1387 n_factors = hsh_count (fctr->fstats);
1389 n_rows = n_dep_var * n_stat_rows * n_factors;
1391 if ( fctr->indep_var[1] )
1392 heading_columns = 5;
1396 heading_columns = 3;
1397 n_rows = n_dep_var * n_stat_rows;
1400 n_rows += heading_rows;
1402 n_cols = heading_columns + 2;
1405 tbl = tab_create (n_cols, n_rows, 0);
1407 tab_headers (tbl, heading_columns + 1, 0, heading_rows, 0);
1409 tab_dim (tbl, tab_natural_dimensions);
1411 /* Outline the box and have no internal lines*/
1416 n_cols - 1, n_rows - 1);
1418 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows );
1420 tab_vline (tbl, TAL_1, heading_columns, 0, n_rows - 1);
1421 tab_vline (tbl, TAL_2, n_cols - 2, 0, n_rows - 1);
1422 tab_vline (tbl, TAL_1, n_cols - 1, 0, n_rows - 1);
1424 tab_text (tbl, n_cols - 2, 0, TAB_CENTER | TAT_TITLE, _ ("Statistic"));
1425 tab_text (tbl, n_cols - 1, 0, TAB_CENTER | TAT_TITLE, _ ("Std. Error"));
1427 tab_title (tbl, _ ("Descriptives"));
1430 for ( i = 0 ; i < n_dep_var ; ++i )
1432 const int row = heading_rows + i * n_stat_rows * n_factors ;
1435 tab_hline (tbl, TAL_1, 0, n_cols - 1, row );
1438 i * n_stat_rows * n_factors + heading_rows,
1439 TAB_LEFT | TAT_TITLE,
1440 var_to_string (dependent_var[i])
1446 const union value *prev = NULL;
1448 struct factor_statistics **fs = fctr->fs;
1451 tab_text (tbl, 1, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1452 var_to_string (fctr->indep_var[0]));
1455 if ( fctr->indep_var[1])
1456 tab_text (tbl, 2, heading_rows - 1, TAB_CENTER | TAT_TITLE,
1457 var_to_string (fctr->indep_var[1]));
1461 const int row = heading_rows + n_stat_rows *
1462 ( ( i * n_factors ) + count );
1465 if ( !prev || 0 != compare_values (prev, (*fs)->id[0],
1466 var_get_width (fctr->indep_var[0])))
1470 tab_hline (tbl, TAL_1, 1, n_cols - 1, row);
1474 TAB_LEFT | TAT_TITLE,
1475 var_get_value_name (fctr->indep_var[0],
1480 prev = (*fs)->id[0];
1482 if (fctr->indep_var[1] && count > 0 )
1483 tab_hline (tbl, TAL_1, 2, n_cols - 1, row);
1485 if ( fctr->indep_var[1])
1486 tab_text (tbl, 2, row,
1487 TAB_LEFT | TAT_TITLE,
1488 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
1491 populate_descriptives (tbl, heading_columns - 2,
1492 row, & (*fs)->m[i]);
1503 populate_descriptives (tbl, heading_columns - 2,
1504 i * n_stat_rows * n_factors + heading_rows,
1516 /* Fill in the descriptives data */
1518 populate_descriptives (struct tab_table *tbl, int col, int row,
1519 const struct metrics *m)
1522 const double t = gsl_cdf_tdist_Qinv (1 - cmd.n_cinterval[0]/100.0/2.0, \
1528 TAB_LEFT | TAT_TITLE,
1531 tab_float (tbl, col + 2,
1537 tab_float (tbl, col + 3,
1546 TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1547 _ ("%g%% Confidence Interval for Mean"), cmd.n_cinterval[0]);
1550 tab_text (tbl, col + 1,
1552 TAB_LEFT | TAT_TITLE,
1555 tab_float (tbl, col + 2,
1558 m->mean - t * m->se_mean,
1561 tab_text (tbl, col + 1,
1563 TAB_LEFT | TAT_TITLE,
1567 tab_float (tbl, col + 2,
1570 m->mean + t * m->se_mean,
1575 TAB_LEFT | TAT_TITLE | TAT_PRINTF,
1576 _ ("5%% Trimmed Mean"));
1578 tab_float (tbl, col + 2,
1586 TAB_LEFT | TAT_TITLE,
1590 struct percentile *p;
1593 p = hsh_find (m->ptile_hash, &d);
1598 tab_float (tbl, col + 2,
1608 TAB_LEFT | TAT_TITLE,
1611 tab_float (tbl, col + 2,
1620 TAB_LEFT | TAT_TITLE,
1621 _ ("Std. Deviation"));
1624 tab_float (tbl, col + 2,
1633 TAB_LEFT | TAT_TITLE,
1636 tab_float (tbl, col + 2,
1644 TAB_LEFT | TAT_TITLE,
1647 tab_float (tbl, col + 2,
1656 TAB_LEFT | TAT_TITLE,
1660 tab_float (tbl, col + 2,
1668 TAB_LEFT | TAT_TITLE,
1669 _ ("Interquartile Range"));
1672 struct percentile *p1;
1673 struct percentile *p2;
1676 p1 = hsh_find (m->ptile_hash, &d);
1679 p2 = hsh_find (m->ptile_hash, &d);
1684 tab_float (tbl, col + 2,
1695 TAB_LEFT | TAT_TITLE,
1699 tab_float (tbl, col + 2,
1705 /* stderr of skewness */
1706 tab_float (tbl, col + 3,
1715 TAB_LEFT | TAT_TITLE,
1719 tab_float (tbl, col + 2,
1725 /* stderr of kurtosis */
1726 tab_float (tbl, col + 3,
1738 box_plot_variables (const struct factor *fctr,
1739 const struct variable **vars, int n_vars,
1740 const struct variable *id)
1744 struct factor_statistics **fs ;
1748 box_plot_group (fctr, vars, n_vars, id);
1752 for ( fs = fctr->fs ; *fs ; ++fs )
1754 double y_min = DBL_MAX;
1755 double y_max = -DBL_MAX;
1756 struct chart *ch = chart_create ();
1757 const char *s = factor_to_string (fctr, *fs, 0 );
1759 chart_write_title (ch, s);
1761 for ( i = 0 ; i < n_vars ; ++i )
1763 y_max = MAX (y_max, (*fs)->m[i].max);
1764 y_min = MIN (y_min, (*fs)->m[i].min);
1767 boxplot_draw_yscale (ch, y_max, y_min);
1769 for ( i = 0 ; i < n_vars ; ++i )
1772 const double box_width = (ch->data_right - ch->data_left)
1775 const double box_centre = ( i * 2 + 1) * box_width
1778 boxplot_draw_boxplot (ch,
1779 box_centre, box_width,
1781 var_to_string (vars[i]));
1793 /* Do a box plot, grouping all factors into one plot ;
1794 each dependent variable has its own plot.
1797 box_plot_group (const struct factor *fctr,
1798 const struct variable **vars,
1800 const struct variable *id UNUSED)
1805 for ( i = 0 ; i < n_vars ; ++i )
1807 struct factor_statistics **fs ;
1810 ch = chart_create ();
1812 boxplot_draw_yscale (ch, totals[i].max, totals[i].min);
1818 for ( fs = fctr->fs ; *fs ; ++fs )
1821 chart_write_title (ch, _ ("Boxplot of %s vs. %s"),
1822 var_to_string (vars[i]), var_to_string (fctr->indep_var[0]) );
1824 for ( fs = fctr->fs ; *fs ; ++fs )
1827 const char *s = factor_to_string_concise (fctr, *fs);
1829 const double box_width = (ch->data_right - ch->data_left)
1830 / (n_factors * 2.0 ) ;
1832 const double box_centre = ( f++ * 2 + 1) * box_width
1835 boxplot_draw_boxplot (ch,
1836 box_centre, box_width,
1843 const double box_width = (ch->data_right - ch->data_left) / 3.0;
1844 const double box_centre = (ch->data_right + ch->data_left) / 2.0;
1846 chart_write_title (ch, _ ("Boxplot"));
1848 boxplot_draw_boxplot (ch,
1849 box_centre, box_width,
1851 var_to_string (vars[i]) );
1860 /* Plot the normal and detrended normal plots for m
1861 Label the plots with factorname */
1863 np_plot (const struct metrics *m, const char *factorname)
1866 double yfirst=0, ylast=0;
1869 struct chart *np_chart;
1871 /* Detrended Normal Plot */
1872 struct chart *dnp_chart;
1874 /* The slope and intercept of the ideal normal probability line */
1875 const double slope = 1.0 / m->stddev;
1876 const double intercept = - m->mean / m->stddev;
1878 /* Cowardly refuse to plot an empty data set */
1879 if ( m->n_data == 0 )
1882 np_chart = chart_create ();
1883 dnp_chart = chart_create ();
1885 if ( !np_chart || ! dnp_chart )
1888 chart_write_title (np_chart, _ ("Normal Q-Q Plot of %s"), factorname);
1889 chart_write_xlabel (np_chart, _ ("Observed Value"));
1890 chart_write_ylabel (np_chart, _ ("Expected Normal"));
1893 chart_write_title (dnp_chart, _ ("Detrended Normal Q-Q Plot of %s"),
1895 chart_write_xlabel (dnp_chart, _ ("Observed Value"));
1896 chart_write_ylabel (dnp_chart, _ ("Dev from Normal"));
1898 yfirst = gsl_cdf_ugaussian_Pinv (m->wvp[0]->rank / ( m->n + 1));
1899 ylast = gsl_cdf_ugaussian_Pinv (m->wvp[m->n_data-1]->rank / ( m->n + 1));
1903 /* Need to make sure that both the scatter plot and the ideal fit into the
1905 double x_lower = MIN (m->min, (yfirst - intercept) / slope) ;
1906 double x_upper = MAX (m->max, (ylast - intercept) / slope) ;
1907 double slack = (x_upper - x_lower) * 0.05 ;
1909 chart_write_xscale (np_chart, x_lower - slack, x_upper + slack, 5);
1911 chart_write_xscale (dnp_chart, m->min, m->max, 5);
1915 chart_write_yscale (np_chart, yfirst, ylast, 5);
1918 /* We have to cache the detrended data, beacause we need to
1919 find its limits before we can plot it */
1920 double *d_data = xnmalloc (m->n_data, sizeof *d_data);
1921 double d_max = -DBL_MAX;
1922 double d_min = DBL_MAX;
1923 for ( i = 0 ; i < m->n_data; ++i )
1925 const double ns = gsl_cdf_ugaussian_Pinv (m->wvp[i]->rank / ( m->n + 1));
1927 chart_datum (np_chart, 0, m->wvp[i]->v.f, ns);
1929 d_data[i] = (m->wvp[i]->v.f - m->mean) / m->stddev - ns;
1931 if ( d_data[i] < d_min ) d_min = d_data[i];
1932 if ( d_data[i] > d_max ) d_max = d_data[i];
1934 chart_write_yscale (dnp_chart, d_min, d_max, 5);
1936 for ( i = 0 ; i < m->n_data; ++i )
1937 chart_datum (dnp_chart, 0, m->wvp[i]->v.f, d_data[i]);
1942 chart_line (np_chart, slope, intercept, yfirst, ylast , CHART_DIM_Y);
1943 chart_line (dnp_chart, 0, 0, m->min, m->max , CHART_DIM_X);
1945 chart_submit (np_chart);
1946 chart_submit (dnp_chart);
1952 /* Show the percentiles */
1954 show_percentiles (struct variable **dependent_var,
1956 struct factor *fctr)
1958 struct tab_table *tbl;
1964 struct hsh_table *ptiles ;
1966 int n_heading_columns;
1967 const int n_heading_rows = 2;
1968 const int n_stat_rows = 2;
1974 struct factor_statistics **fs = fctr->fs ;
1975 n_heading_columns = 3;
1976 n_factors = hsh_count (fctr->fstats);
1978 ptiles = (*fs)->m[0].ptile_hash;
1980 if ( fctr->indep_var[1] )
1981 n_heading_columns = 4;
1986 n_heading_columns = 2;
1988 ptiles = totals[0].ptile_hash;
1991 n_ptiles = hsh_count (ptiles);
1993 n_rows = n_heading_rows + n_dep_var * n_stat_rows * n_factors;
1995 n_cols = n_heading_columns + n_ptiles ;
1997 tbl = tab_create (n_cols, n_rows, 0);
1999 tab_headers (tbl, n_heading_columns + 1, 0, n_heading_rows, 0);
2001 tab_dim (tbl, tab_natural_dimensions);
2003 /* Outline the box and have no internal lines*/
2008 n_cols - 1, n_rows - 1);
2010 tab_hline (tbl, TAL_2, 0, n_cols - 1, n_heading_rows );
2012 tab_vline (tbl, TAL_2, n_heading_columns, 0, n_rows - 1);
2015 tab_title (tbl, _ ("Percentiles"));
2018 tab_hline (tbl, TAL_1, n_heading_columns, n_cols - 1, 1 );
2025 n_heading_columns - 1, n_rows - 1);
2031 n_heading_columns, n_heading_rows - 1,
2032 n_cols - 1, n_rows - 1);
2034 tab_joint_text (tbl, n_heading_columns + 1, 0,
2036 TAB_CENTER | TAT_TITLE ,
2041 /* Put in the percentile break points as headings */
2043 struct percentile **p = (struct percentile **) hsh_sort (ptiles);
2048 tab_float (tbl, n_heading_columns + i++ , 1,
2057 for ( i = 0 ; i < n_dep_var ; ++i )
2059 const int n_stat_rows = 2;
2060 const int row = n_heading_rows + i * n_stat_rows * n_factors ;
2063 tab_hline (tbl, TAL_1, 0, n_cols - 1, row );
2066 i * n_stat_rows * n_factors + n_heading_rows,
2067 TAB_LEFT | TAT_TITLE,
2068 var_to_string (dependent_var[i])
2073 const union value *prev = NULL ;
2074 struct factor_statistics **fs = fctr->fs;
2077 tab_text (tbl, 1, n_heading_rows - 1,
2078 TAB_CENTER | TAT_TITLE,
2079 var_to_string (fctr->indep_var[0]));
2082 if ( fctr->indep_var[1])
2083 tab_text (tbl, 2, n_heading_rows - 1, TAB_CENTER | TAT_TITLE,
2084 var_to_string (fctr->indep_var[1]));
2088 const int row = n_heading_rows + n_stat_rows *
2089 ( ( i * n_factors ) + count );
2092 if ( !prev || 0 != compare_values (prev, (*fs)->id[0],
2093 var_get_width (fctr->indep_var[0])))
2097 tab_hline (tbl, TAL_1, 1, n_cols - 1, row);
2101 TAB_LEFT | TAT_TITLE,
2102 var_get_value_name (fctr->indep_var[0],
2109 prev = (*fs)->id[0];
2111 if (fctr->indep_var[1] && count > 0 )
2112 tab_hline (tbl, TAL_1, 2, n_cols - 1, row);
2114 if ( fctr->indep_var[1])
2115 tab_text (tbl, 2, row,
2116 TAB_LEFT | TAT_TITLE,
2117 var_get_value_name (fctr->indep_var[1], (*fs)->id[1])
2121 populate_percentiles (tbl, n_heading_columns - 1,
2122 row, & (*fs)->m[i]);
2133 populate_percentiles (tbl, n_heading_columns - 1,
2134 i * n_stat_rows * n_factors + n_heading_rows,
2151 populate_percentiles (struct tab_table *tbl, int col, int row,
2152 const struct metrics *m)
2156 struct percentile **p = (struct percentile **) hsh_sort (m->ptile_hash);
2160 TAB_LEFT | TAT_TITLE,
2161 _ ("Tukey\'s Hinges")
2166 TAB_LEFT | TAT_TITLE,
2167 ptile_alg_desc[m->ptile_alg]
2174 tab_float (tbl, col + i + 1 , row,
2177 if ( (*p)->p == 25 )
2178 tab_float (tbl, col + i + 1 , row + 1,
2182 if ( (*p)->p == 50 )
2183 tab_float (tbl, col + i + 1 , row + 1,
2187 if ( (*p)->p == 75 )
2188 tab_float (tbl, col + i + 1 , row + 1,
2203 factor_to_string (const struct factor *fctr,
2204 const struct factor_statistics *fs,
2205 const struct variable *var)
2208 static char buf1[100];
2214 sprintf (buf1, "%s (",var_to_string (var) );
2217 snprintf (buf2, 100, "%s = %s",
2218 var_to_string (fctr->indep_var[0]),
2219 var_get_value_name (fctr->indep_var[0], fs->id[0]));
2221 strcat (buf1, buf2);
2223 if ( fctr->indep_var[1] )
2225 sprintf (buf2, "; %s = %s)",
2226 var_to_string (fctr->indep_var[1]),
2227 var_get_value_name (fctr->indep_var[1], fs->id[1]));
2228 strcat (buf1, buf2);
2242 factor_to_string_concise (const struct factor *fctr,
2243 struct factor_statistics *fs)
2247 static char buf[100];
2251 snprintf (buf, 100, "%s",
2252 var_get_value_name (fctr->indep_var[0], fs->id[0]));
2254 if ( fctr->indep_var[1] )
2256 sprintf (buf2, ",%s)", var_get_value_name (fctr->indep_var[1],