1 /* PSPP - computes sample statistics.
2 Copyright (C) 1997-9, 2000 Free Software Foundation, Inc.
3 Written by Ben Pfaff <blp@gnu.org>.
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., 59 Temple Place - Suite 330, Boston, MA
23 * Remember that histograms, bar charts need mean, stddev.
31 #include "bitvector.h"
37 #include "algorithm.h"
45 #include "value-labels.h"
49 #include "debug-print.h"
54 format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
55 table:limit(n:limit,"%s>0")/notable/!table,
56 labels:!labels/nolabels,
57 sort:!avalue/dvalue/afreq/dfreq,
58 spaces:!single/double,
59 paging:newpage/!oldpage;
60 missing=miss:include/!exclude;
61 barchart(ba_)=:minimum(d:min),
63 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
64 histogram(hi_)=:minimum(d:min),
66 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
67 norm:!nonormal/normal,
68 incr:increment(d:inc,"%s>0");
69 hbar(hb_)=:minimum(d:min),
71 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
72 norm:!nonormal/normal,
73 incr:increment(d:inc,"%s>0");
77 statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
78 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
79 13|default,14|seskewness,15|sekurtosis,all,none.
84 /* Description of a statistic. */
87 int st_indx; /* Index into a_statistics[]. */
88 const char *s10; /* Identifying string. */
91 /* Table of statistics, indexed by dsc_*. */
92 static struct frq_info st_name[frq_n_stats + 1] =
94 {FRQ_ST_MEAN, N_("Mean")},
95 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
96 {FRQ_ST_MEDIAN, N_("Median")},
97 {FRQ_ST_MODE, N_("Mode")},
98 {FRQ_ST_STDDEV, N_("Std Dev")},
99 {FRQ_ST_VARIANCE, N_("Variance")},
100 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
101 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
102 {FRQ_ST_SKEWNESS, N_("Skewness")},
103 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
104 {FRQ_ST_RANGE, N_("Range")},
105 {FRQ_ST_MINIMUM, N_("Minimum")},
106 {FRQ_ST_MAXIMUM, N_("Maximum")},
107 {FRQ_ST_SUM, N_("Sum")},
111 /* Percentiles to calculate. */
112 static double *percentiles;
113 static double *percentile_values;
114 static int n_percentiles;
116 /* Groups of statistics. */
118 #define frq_default \
119 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
121 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
122 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
123 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
124 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
125 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
127 /* Statistics; number of statistics. */
128 static unsigned long stats;
131 /* Types of graphs. */
134 GFT_NONE, /* Don't draw graphs. */
135 GFT_BAR, /* Draw bar charts. */
136 GFT_HIST, /* Draw histograms. */
137 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
140 /* Parsed command. */
141 static struct cmd_frequencies cmd;
143 /* Summary of the barchart, histogram, and hbar subcommands. */
144 static int chart; /* NONE/BAR/HIST/HBAR. */
145 static double min, max; /* Minimum, maximum on y axis. */
146 static int format; /* FREQ/PERCENT: Scaling of y axis. */
147 static double scale, incr; /* FIXME */
148 static int normal; /* FIXME */
150 /* Variables for which to calculate statistics. */
151 static int n_variables;
152 static struct variable **v_variables;
154 /* Arenas used to store semi-permanent storage. */
155 static struct pool *int_pool; /* Integer mode. */
156 static struct pool *gen_pool; /* General mode. */
158 /* Easier access to a_statistics. */
159 #define stat cmd.a_statistics
161 static void determine_charts (void);
163 static void precalc (void *);
164 static int calc (struct ccase *, void *);
165 static void postcalc (void *);
167 static void postprocess_freq_tab (struct variable *);
168 static void dump_full (struct variable *);
169 static void dump_condensed (struct variable *);
170 static void dump_statistics (struct variable *, int show_varname);
171 static void cleanup_freq_tab (struct variable *);
173 static hsh_hash_func hash_value_numeric, hash_value_alpha;
174 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
175 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
176 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
177 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
179 /* Parser and outline. */
181 static int internal_cmd_frequencies (void);
184 cmd_frequencies (void)
188 int_pool = pool_create ();
189 result = internal_cmd_frequencies ();
190 pool_destroy (int_pool);
192 pool_destroy (gen_pool);
200 internal_cmd_frequencies (void)
205 percentile_values = NULL;
211 for (i = 0; i < dict_get_var_cnt (default_dict); i++)
212 dict_get_var(default_dict, i)->p.frq.used = 0;
214 if (!parse_frequencies (&cmd))
217 if (cmd.onepage_limit == NOT_LONG)
218 cmd.onepage_limit = 50;
220 /* Figure out statistics to calculate. */
222 if (stat[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
223 stats |= frq_default;
224 if (stat[FRQ_ST_ALL])
226 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
227 stats &= ~frq_median;
228 for (i = 0; i < frq_n_stats; i++)
229 if (stat[st_name[i].st_indx])
230 stats |= BIT_INDEX (i);
231 if (stats & frq_kurt)
233 if (stats & frq_skew)
236 /* Calculate n_stats. */
238 for (i = 0; i < frq_n_stats; i++)
239 if ((stats & BIT_INDEX (i)))
244 if (chart != GFT_NONE || cmd.sbc_ntiles)
245 cmd.sort = FRQ_AVALUE;
248 procedure_with_splits (precalc, calc, postcalc, NULL);
253 /* Figure out which charts the user requested. */
255 determine_charts (void)
257 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) + (!!cmd.sbc_hbar);
267 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
268 "given. HBAR will be assumed. Argument values will be "
269 "given precedence increasing along the order given."));
271 else if (cmd.sbc_histogram)
273 else if (cmd.sbc_barchart)
284 if (cmd.sbc_barchart)
286 if (cmd.ba_min != SYSMIS)
288 if (cmd.ba_max != SYSMIS)
290 if (cmd.ba_scale == FRQ_FREQ)
295 else if (cmd.ba_scale == FRQ_PERCENT)
297 format = FRQ_PERCENT;
302 if (cmd.sbc_histogram)
304 if (cmd.hi_min != SYSMIS)
306 if (cmd.hi_max != SYSMIS)
308 if (cmd.hi_scale == FRQ_FREQ)
313 else if (cmd.hi_scale == FRQ_PERCENT)
315 format = FRQ_PERCENT;
320 if (cmd.hi_incr == FRQ_INCREMENT)
326 if (cmd.hb_min != SYSMIS)
328 if (cmd.hb_max != SYSMIS)
330 if (cmd.hb_scale == FRQ_FREQ)
335 else if (cmd.hb_scale == FRQ_PERCENT)
337 format = FRQ_PERCENT;
342 if (cmd.hb_incr == FRQ_INCREMENT)
346 if (min != SYSMIS && max != SYSMIS && min >= max)
348 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
349 "specified. However, MIN was specified as %g and MAX as %g. "
350 "MIN and MAX will be ignored."), min, max);
355 /* Add data from case C to the frequency table. */
357 calc (struct ccase *c, void *aux UNUSED)
362 weight = dict_get_case_weight (default_dict, c);
364 for (i = 0; i < n_variables; i++)
366 struct variable *v = v_variables[i];
367 union value *val = &c->data[v->fv];
368 struct freq_tab *ft = &v->p.frq.tab;
370 switch (v->p.frq.tab.mode)
375 struct freq **fpp = (struct freq **) hsh_probe (ft->data, val);
381 struct freq *fp = *fpp = pool_alloc (gen_pool, sizeof *fp);
389 if (val->f == SYSMIS)
390 v->p.frq.tab.sysmis += weight;
391 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
394 if (i >= v->p.frq.tab.min && i <= v->p.frq.tab.max)
395 v->p.frq.tab.vector[i - v->p.frq.tab.min] += weight;
398 v->p.frq.tab.out_of_range += weight;
407 /* Prepares each variable that is the target of FREQUENCIES by setting
408 up its hash table. */
410 precalc (void *aux UNUSED)
414 pool_destroy (gen_pool);
415 gen_pool = pool_create ();
417 for (i = 0; i < n_variables; i++)
419 struct variable *v = v_variables[i];
421 if (v->p.frq.tab.mode == FRQM_GENERAL)
424 hsh_compare_func *compare;
426 if (v->type == NUMERIC)
428 hash = hash_value_numeric;
429 compare = compare_value_numeric_a;
433 hash = hash_value_alpha;
434 compare = compare_value_alpha_a;
436 v->p.frq.tab.data = hsh_create (16, compare, hash, NULL, v);
442 for (j = (v->p.frq.tab.max - v->p.frq.tab.min); j >= 0; j--)
443 v->p.frq.tab.vector[j] = 0.0;
444 v->p.frq.tab.out_of_range = 0.0;
445 v->p.frq.tab.sysmis = 0.0;
450 /* Finishes up with the variables after frequencies have been
451 calculated. Displays statistics, percentiles, ... */
453 postcalc (void *aux UNUSED)
457 for (i = 0; i < n_variables; i++)
459 struct variable *v = v_variables[i];
461 int dumped_freq_tab = 1;
463 postprocess_freq_tab (v);
465 /* Frequencies tables. */
466 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
467 if (cmd.table == FRQ_TABLE
468 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
478 if (n_categories > cmd.onepage_limit)
491 dump_statistics (v, !dumped_freq_tab);
493 cleanup_freq_tab (v);
497 /* Returns the comparison function that should be used for
498 sorting a frequency table by FRQ_SORT using VAR_TYPE
500 static hsh_compare_func *
501 get_freq_comparator (int frq_sort, int var_type)
503 /* Note that q2c generates tags beginning with 1000. */
504 switch (frq_sort | (var_type << 16))
506 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
507 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
508 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
509 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
510 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
511 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
512 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
513 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
520 /* Returns nonzero iff the value in struct freq F is non-missing
523 not_missing (const void *f_, void *v_)
525 const struct freq *f = f_;
526 struct variable *v = v_;
528 return !is_missing (&f->v, v);
531 /* Summarizes the frequency table data for variable V. */
533 postprocess_freq_tab (struct variable *v)
535 hsh_compare_func *compare;
539 struct freq *freqs, *f;
542 assert (v->p.frq.tab.mode == FRQM_GENERAL);
543 compare = get_freq_comparator (cmd.sort, v->type);
546 /* Extract data from hash table. */
547 count = hsh_count (ft->data);
548 data = hsh_data (ft->data);
550 /* Copy dereferenced data into freqs. */
551 freqs = xmalloc (count* sizeof *freqs);
552 for (i = 0; i < count; i++)
554 struct freq *f = data[i];
558 /* Put data into ft. */
560 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
561 ft->missing = freqs + ft->n_valid;
562 ft->n_missing = count - ft->n_valid;
565 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
566 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
568 /* Summary statistics. */
569 ft->total_cases = ft->valid_cases = 0.0;
570 for (f = ft->valid; f < ft->valid + ft->n_valid; f++)
572 ft->total_cases += f->c;
574 if ((v->type != NUMERIC || f->v.f != SYSMIS)
575 && (cmd.miss != FRQ_EXCLUDE || !is_user_missing (&f->v, v)))
576 ft->valid_cases += f->c;
580 /* Frees the frequency table for variable V. */
582 cleanup_freq_tab (struct variable *v)
584 assert (v->p.frq.tab.mode == FRQM_GENERAL);
585 free (v->p.frq.tab.valid);
586 hsh_destroy (v->p.frq.tab.data);
589 /* Parses the VARIABLES subcommand, adding to
590 {n_variables,v_variables}. */
592 frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
595 int min = 0, max = 0;
597 int old_n_variables = n_variables;
601 if (token != T_ALL && (token != T_ID
602 || dict_lookup_var (default_dict, tokid) == NULL))
605 if (!parse_variables (default_dict, &v_variables, &n_variables,
606 PV_APPEND | PV_NO_SCRATCH))
609 for (i = old_n_variables; i < n_variables; i++)
610 v_variables[i]->p.frq.tab.mode = FRQM_GENERAL;
612 if (!lex_match ('('))
617 if (!lex_force_int ())
619 min = lex_integer ();
621 if (!lex_force_match (','))
623 if (!lex_force_int ())
625 max = lex_integer ();
627 if (!lex_force_match (')'))
631 msg (SE, _("Upper limit of integer mode value range must be "
632 "greater than lower limit."));
637 for (i = old_n_variables; i < n_variables; i++)
639 struct variable *v = v_variables[i];
641 if (v->p.frq.used != 0)
643 msg (SE, _("Variable %s specified multiple times on VARIABLES "
644 "subcommand."), v->name);
648 v->p.frq.used = 1; /* Used simply as a marker. */
650 v->p.frq.tab.valid = v->p.frq.tab.missing = NULL;
652 if (mode == FRQM_INTEGER)
654 if (v->type != NUMERIC)
656 msg (SE, _("Integer mode specified, but %s is not a numeric "
657 "variable."), v->name);
661 v->p.frq.tab.min = min;
662 v->p.frq.tab.max = max;
663 v->p.frq.tab.vector = pool_alloc (int_pool,
664 sizeof (struct freq) * (max - min + 1));
667 v->p.frq.tab.vector = NULL;
669 v->p.frq.n_groups = 0;
670 v->p.frq.groups = NULL;
675 /* Parses the GROUPED subcommand, setting the frq.{n_grouped,grouped}
676 fields of specified variables. */
678 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
681 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
687 /* Max, current size of list; list itself. */
695 if (!parse_variables (default_dict, &v, &n,
696 PV_NO_DUPLICATE | PV_NUMERIC))
702 while (token == T_NUM)
707 dl = pool_realloc (int_pool, dl, ml * sizeof (double));
713 /* Note that nl might still be 0 and dl might still be
714 NULL. That's okay. */
715 if (!lex_match (')'))
718 msg (SE, _("`)' expected after GROUPED interval list."));
728 for (i = 0; i < n; i++)
730 if (v[i]->p.frq.used == 0)
731 msg (SE, _("Variables %s specified on GROUPED but not on "
732 "VARIABLES."), v[i]->name);
733 if (v[i]->p.frq.groups != NULL)
734 msg (SE, _("Variables %s specified multiple times on GROUPED "
735 "subcommand."), v[i]->name);
738 v[i]->p.frq.n_groups = nl;
739 v[i]->p.frq.groups = dl;
743 if (!lex_match ('/'))
745 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
756 /* Adds X to the list of percentiles, keeping the list in proper
759 add_percentile (double x)
763 for (i = 0; i < n_percentiles; i++)
764 if (x <= percentiles[i])
766 if (i >= n_percentiles || tokval != percentiles[i])
769 = pool_realloc (int_pool, percentiles,
770 (n_percentiles + 1) * sizeof *percentiles);
772 = pool_realloc (int_pool, percentile_values,
773 (n_percentiles + 1) * sizeof *percentile_values);
774 if (i < n_percentiles)
776 memmove (&percentiles[i + 1], &percentiles[i],
777 (n_percentiles - i) * sizeof *percentiles);
778 memmove (&percentile_values[i + 1], &percentile_values[i],
779 (n_percentiles - i) * sizeof *percentile_values);
786 /* Parses the PERCENTILES subcommand, adding user-specified
787 percentiles to the list. */
789 frq_custom_percentiles (struct cmd_frequencies *cmd UNUSED)
794 msg (SE, _("Percentile list expected after PERCENTILES."));
800 if (tokval <= 0 || tokval >= 100)
802 msg (SE, _("Percentiles must be greater than "
803 "0 and less than 100."));
807 add_percentile (tokval / 100.0);
811 while (token == T_NUM);
815 /* Parses the NTILES subcommand, adding the percentiles that
816 correspond to the specified evenly-distributed ntiles. */
818 frq_custom_ntiles (struct cmd_frequencies *cmd UNUSED)
823 if (!lex_force_int ())
825 for (i = 1; i < lex_integer (); i++)
826 add_percentile (1.0 / lex_integer () * i);
831 /* Comparison functions. */
833 /* Hash of numeric values. */
835 hash_value_numeric (const void *value_, void *foo UNUSED)
837 const struct freq *value = value_;
838 return hsh_hash_double (value->v.f);
841 /* Hash of string values. */
843 hash_value_alpha (const void *value_, void *v_)
845 const struct freq *value = value_;
846 struct variable *v = v_;
848 return hsh_hash_bytes (value->v.s, v->width);
851 /* Ascending numeric compare of values. */
853 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
855 const struct freq *a = a_;
856 const struct freq *b = b_;
860 else if (a->v.f < b->v.f)
866 /* Ascending string compare of values. */
868 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
870 const struct freq *a = a_;
871 const struct freq *b = b_;
872 const struct variable *v = v_;
874 return memcmp (a->v.s, b->v.s, v->width);
877 /* Descending numeric compare of values. */
879 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
881 return -compare_value_numeric_a (a, b, foo);
884 /* Descending string compare of values. */
886 compare_value_alpha_d (const void *a, const void *b, void *v)
888 return -compare_value_alpha_a (a, b, v);
891 /* Ascending numeric compare of frequency;
892 secondary key on ascending numeric value. */
894 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
896 const struct freq *a = a_;
897 const struct freq *b = b_;
901 else if (a->c < b->c)
906 else if (a->v.f < b->v.f)
912 /* Ascending numeric compare of frequency;
913 secondary key on ascending string value. */
915 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
917 const struct freq *a = a_;
918 const struct freq *b = b_;
919 const struct variable *v = v_;
923 else if (a->c < b->c)
926 return memcmp (a->v.s, b->v.s, v->width);
929 /* Descending numeric compare of frequency;
930 secondary key on ascending numeric value. */
932 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
934 const struct freq *a = a_;
935 const struct freq *b = b_;
939 else if (a->c < b->c)
944 else if (a->v.f < b->v.f)
950 /* Descending numeric compare of frequency;
951 secondary key on ascending string value. */
953 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
955 const struct freq *a = a_;
956 const struct freq *b = b_;
957 const struct variable *v = v_;
961 else if (a->c < b->c)
964 return memcmp (a->v.s, b->v.s, v->width);
967 /* Frequency table display. */
969 /* Sets the widths of all the columns and heights of all the rows in
970 table T for driver D. */
972 full_dim (struct tab_table *t, struct outp_driver *d)
974 int lab = cmd.labels == FRQ_LABELS;
978 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
979 for (i = lab; i < lab + 5; i++)
980 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
981 for (i = 0; i < t->nr; i++)
982 t->h[i] = d->font_height;
985 /* Displays a full frequency table for variable V. */
987 dump_full (struct variable * v)
993 double cum_total = 0.0;
994 double cum_freq = 0.0;
1004 static struct init vec[] =
1006 {4, 0, N_("Valid")},
1008 {1, 1, N_("Value")},
1009 {2, 1, N_("Frequency")},
1010 {3, 1, N_("Percent")},
1011 {4, 1, N_("Percent")},
1012 {5, 1, N_("Percent")},
1020 int lab = cmd.labels == FRQ_LABELS;
1022 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1023 t = tab_create (5 + lab, n_categories + 3, 0);
1024 tab_headers (t, 0, 0, 2, 0);
1025 tab_dim (t, full_dim);
1028 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1029 for (p = vec; p->s; p++)
1030 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1031 TAB_CENTER | TAT_TITLE, gettext (p->s));
1034 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1036 double percent, valid_percent;
1040 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1041 valid_percent = f->c / v->p.frq.tab.valid_cases * 100.0;
1042 cum_total += valid_percent;
1046 const char *label = val_labs_find (v->val_labs, f->v);
1048 tab_text (t, 0, r, TAB_LEFT, label);
1051 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1052 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1053 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1054 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1055 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1058 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1064 const char *label = val_labs_find (v->val_labs, f->v);
1066 tab_text (t, 0, r, TAB_LEFT, label);
1069 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1070 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1071 tab_float (t, 2 + lab, r, TAB_NONE,
1072 f->c / v->p.frq.tab.total_cases * 100.0, 5, 1);
1073 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1077 tab_box (t, TAL_1, TAL_1,
1078 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1080 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1081 tab_hline (t, TAL_2, 0, 4 + lab, r);
1082 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1083 tab_vline (t, TAL_0, 1, r, r);
1084 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1085 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1086 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1088 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1092 /* Sets the widths of all the columns and heights of all the rows in
1093 table T for driver D. */
1095 condensed_dim (struct tab_table *t, struct outp_driver *d)
1097 int cum_w = max (outp_string_width (d, _("Cum")),
1098 max (outp_string_width (d, _("Cum")),
1099 outp_string_width (d, "000")));
1103 for (i = 0; i < 2; i++)
1104 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1105 for (i = 2; i < 4; i++)
1107 for (i = 0; i < t->nr; i++)
1108 t->h[i] = d->font_height;
1111 /* Display condensed frequency table for variable V. */
1113 dump_condensed (struct variable * v)
1117 struct tab_table *t;
1119 double cum_total = 0.0;
1121 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1122 t = tab_create (4, n_categories + 2, 0);
1124 tab_headers (t, 0, 0, 2, 0);
1125 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1126 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1127 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1128 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1129 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1130 tab_dim (t, condensed_dim);
1133 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1137 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1138 cum_total += f->c / v->p.frq.tab.valid_cases * 100.0;
1140 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1141 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1142 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1143 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1146 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1148 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1149 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1150 tab_float (t, 2, r, TAB_NONE,
1151 f->c / v->p.frq.tab.total_cases * 100.0, 3, 0);
1155 tab_box (t, TAL_1, TAL_1,
1156 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1158 tab_hline (t, TAL_2, 0, 3, 2);
1159 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1160 tab_columns (t, SOM_COL_DOWN, 1);
1164 /* Statistical display. */
1166 /* Calculates all the pertinent statistics for variable V, putting
1167 them in array D[]. FIXME: This could be made much more optimal. */
1169 calc_stats (struct variable * v, double d[frq_n_stats])
1171 double W = v->p.frq.tab.valid_cases;
1172 double X_bar, X_mode, M2, M3, M4;
1178 double previous_value;
1180 /* Calculate the mean. */
1182 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1183 X_bar += f->v.f * f->c;
1186 /* Calculate percentiles. */
1188 previous_value = SYSMIS;
1189 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1192 for (; i < n_percentiles; i++)
1194 if (cum_total / v->p.frq.tab.valid_cases < percentiles[i])
1197 percentile_values[i] = previous_value;
1199 previous_value = f->v.f;
1202 /* Calculate the mode. */
1205 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1207 if (most_often < f->c)
1212 else if (most_often == f->c)
1214 /* A duplicate mode is undefined.
1215 FIXME: keep track of *all* the modes. */
1220 /* Calculate moments about the mean. */
1222 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1224 double dev = f->v.f - X_bar;
1234 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1235 d[frq_min] = v->p.frq.tab.valid[0].v.f;
1236 d[frq_max] = v->p.frq.tab.valid[v->p.frq.tab.n_valid - 1].v.f;
1237 d[frq_mode] = X_mode;
1238 d[frq_range] = d[frq_max] - d[frq_min];
1239 d[frq_median] = SYSMIS;
1240 d[frq_mean] = X_bar;
1241 d[frq_sum] = X_bar * W;
1242 d[frq_variance] = M2 / (W - 1);
1243 d[frq_stddev] = sqrt (d[frq_variance]);
1244 d[frq_semean] = d[frq_stddev] / sqrt (W);
1245 if (W >= 3.0 && d[frq_variance] > 0)
1247 double S = d[frq_stddev];
1248 d[frq_skew] = (W * M3 / ((W - 1.0) * (W - 2.0) * S * S * S));
1249 d[frq_seskew] = sqrt (6.0 * W * (W - 1.0)
1250 / ((W - 2.0) * (W + 1.0) * (W + 3.0)));
1254 d[frq_skew] = d[frq_seskew] = SYSMIS;
1256 if (W >= 4.0 && d[frq_variance] > 0)
1258 double S2 = d[frq_variance];
1259 double SE_g1 = d[frq_seskew];
1261 d[frq_kurt] = ((W * (W + 1.0) * M4 - 3.0 * M2 * M2 * (W - 1.0))
1262 / ((W - 1.0) * (W - 2.0) * (W - 3.0) * S2 * S2));
1263 d[frq_sekurt] = sqrt ((4.0 * (W * W - 1.0) * SE_g1 * SE_g1)
1264 / ((W - 3.0) * (W + 5.0)));
1268 d[frq_kurt] = d[frq_sekurt] = SYSMIS;
1272 /* Displays a table of all the statistics requested for variable V. */
1274 dump_statistics (struct variable * v, int show_varname)
1276 double stat_value[frq_n_stats];
1277 struct tab_table *t;
1280 if (v->type == ALPHA)
1282 if (v->p.frq.tab.n_valid == 0)
1284 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1288 calc_stats (v, stat_value);
1290 t = tab_create (2, n_stats + n_percentiles, 0);
1291 tab_dim (t, tab_natural_dimensions);
1292 tab_vline (t, TAL_1 | TAL_SPACING, 1, 0, n_stats - 1);
1293 for (i = r = 0; i < frq_n_stats; i++)
1294 if (stats & BIT_INDEX (i))
1296 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1297 gettext (st_name[i].s10));
1298 tab_float (t, 1, r, TAB_NONE, stat_value[i], 11, 3);
1302 for (i = 0; i < n_percentiles; i++, r++)
1306 ds_init (gen_pool, &ds, 20);
1307 ds_printf (&ds, "%s %d", _("Percentile"), (int) (percentiles[i] * 100));
1309 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, ds.string);
1310 tab_float (t, 1, r, TAB_NONE, percentile_values[i], 11, 3);
1315 tab_columns (t, SOM_COL_DOWN, 1);
1319 tab_title (t, 1, "%s: %s", v->name, v->label);
1321 tab_title (t, 0, v->name);
1324 tab_flags (t, SOMF_NO_TITLE);