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"
44 #include "value-labels.h"
48 #include "debug-print.h"
53 format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
54 table:limit(n:limit,"%s>0")/notable/!table,
55 labels:!labels/nolabels,
56 sort:!avalue/dvalue/afreq/dfreq,
57 spaces:!single/double,
58 paging:newpage/!oldpage;
59 missing=miss:include/!exclude;
60 barchart(ba_)=:minimum(d:min),
62 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
63 histogram(hi_)=:minimum(d:min),
65 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
66 norm:!nonormal/normal,
67 incr:increment(d:inc,"%s>0");
68 hbar(hb_)=:minimum(d:min),
70 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
71 norm:!nonormal/normal,
72 incr:increment(d:inc,"%s>0");
76 statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
77 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
78 13|default,14|seskewness,15|sekurtosis,all,none.
83 /* Description of a statistic. */
86 int st_indx; /* Index into a_statistics[]. */
87 const char *s10; /* Identifying string. */
90 /* Table of statistics, indexed by dsc_*. */
91 static struct frq_info st_name[frq_n_stats + 1] =
93 {FRQ_ST_MEAN, N_("Mean")},
94 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
95 {FRQ_ST_MEDIAN, N_("Median")},
96 {FRQ_ST_MODE, N_("Mode")},
97 {FRQ_ST_STDDEV, N_("Std Dev")},
98 {FRQ_ST_VARIANCE, N_("Variance")},
99 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
100 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
101 {FRQ_ST_SKEWNESS, N_("Skewness")},
102 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
103 {FRQ_ST_RANGE, N_("Range")},
104 {FRQ_ST_MINIMUM, N_("Minimum")},
105 {FRQ_ST_MAXIMUM, N_("Maximum")},
106 {FRQ_ST_SUM, N_("Sum")},
110 /* Percentiles to calculate. */
111 static double *percentiles;
112 static double *percentile_values;
113 static int n_percentiles;
115 /* Groups of statistics. */
117 #define frq_default \
118 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
120 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
121 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
122 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
123 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
124 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
126 /* Statistics; number of statistics. */
127 static unsigned long stats;
130 /* Types of graphs. */
133 GFT_NONE, /* Don't draw graphs. */
134 GFT_BAR, /* Draw bar charts. */
135 GFT_HIST, /* Draw histograms. */
136 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
139 /* Parsed command. */
140 static struct cmd_frequencies cmd;
142 /* Summary of the barchart, histogram, and hbar subcommands. */
143 static int chart; /* NONE/BAR/HIST/HBAR. */
144 static double min, max; /* Minimum, maximum on y axis. */
145 static int format; /* FREQ/PERCENT: Scaling of y axis. */
146 static double scale, incr; /* FIXME */
147 static int normal; /* FIXME */
149 /* Variables for which to calculate statistics. */
150 static int n_variables;
151 static struct variable **v_variables;
153 /* Arenas used to store semi-permanent storage. */
154 static struct pool *int_pool; /* Integer mode. */
155 static struct pool *gen_pool; /* General mode. */
157 /* Easier access to a_statistics. */
158 #define stat cmd.a_statistics
160 static void determine_charts (void);
162 static void precalc (void *);
163 static int calc (struct ccase *, void *);
164 static void postcalc (void *);
166 static void postprocess_freq_tab (struct variable *);
167 static void dump_full (struct variable *);
168 static void dump_condensed (struct variable *);
169 static void dump_statistics (struct variable *, int show_varname);
170 static void cleanup_freq_tab (struct variable *);
172 static hsh_hash_func hash_value_numeric, hash_value_alpha;
173 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
174 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
175 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
176 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
178 /* Parser and outline. */
180 static int internal_cmd_frequencies (void);
183 cmd_frequencies (void)
187 int_pool = pool_create ();
188 result = internal_cmd_frequencies ();
189 pool_destroy (int_pool);
191 pool_destroy (gen_pool);
199 internal_cmd_frequencies (void)
204 percentile_values = NULL;
210 for (i = 0; i < dict_get_var_cnt (default_dict); i++)
211 dict_get_var(default_dict, i)->p.frq.used = 0;
213 if (!parse_frequencies (&cmd))
216 if (cmd.onepage_limit == NOT_LONG)
217 cmd.onepage_limit = 50;
219 /* Figure out statistics to calculate. */
221 if (stat[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
222 stats |= frq_default;
223 if (stat[FRQ_ST_ALL])
225 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
226 stats &= ~frq_median;
227 for (i = 0; i < frq_n_stats; i++)
228 if (stat[st_name[i].st_indx])
229 stats |= BIT_INDEX (i);
230 if (stats & frq_kurt)
232 if (stats & frq_skew)
235 /* Calculate n_stats. */
237 for (i = 0; i < frq_n_stats; i++)
238 if ((stats & BIT_INDEX (i)))
243 if (chart != GFT_NONE || cmd.sbc_ntiles)
244 cmd.sort = FRQ_AVALUE;
247 procedure_with_splits (precalc, calc, postcalc, NULL);
252 /* Figure out which charts the user requested. */
254 determine_charts (void)
256 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) + (!!cmd.sbc_hbar);
266 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
267 "given. HBAR will be assumed. Argument values will be "
268 "given precedence increasing along the order given."));
270 else if (cmd.sbc_histogram)
272 else if (cmd.sbc_barchart)
283 if (cmd.sbc_barchart)
285 if (cmd.ba_min != SYSMIS)
287 if (cmd.ba_max != SYSMIS)
289 if (cmd.ba_scale == FRQ_FREQ)
294 else if (cmd.ba_scale == FRQ_PERCENT)
296 format = FRQ_PERCENT;
301 if (cmd.sbc_histogram)
303 if (cmd.hi_min != SYSMIS)
305 if (cmd.hi_max != SYSMIS)
307 if (cmd.hi_scale == FRQ_FREQ)
312 else if (cmd.hi_scale == FRQ_PERCENT)
314 format = FRQ_PERCENT;
319 if (cmd.hi_incr == FRQ_INCREMENT)
325 if (cmd.hb_min != SYSMIS)
327 if (cmd.hb_max != SYSMIS)
329 if (cmd.hb_scale == FRQ_FREQ)
334 else if (cmd.hb_scale == FRQ_PERCENT)
336 format = FRQ_PERCENT;
341 if (cmd.hb_incr == FRQ_INCREMENT)
345 if (min != SYSMIS && max != SYSMIS && min >= max)
347 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
348 "specified. However, MIN was specified as %g and MAX as %g. "
349 "MIN and MAX will be ignored."), min, max);
354 /* Add data from case C to the frequency table. */
356 calc (struct ccase *c, void *aux UNUSED)
361 weight = dict_get_case_weight (default_dict, c);
363 for (i = 0; i < n_variables; i++)
365 struct variable *v = v_variables[i];
366 union value *val = &c->data[v->fv];
367 struct freq_tab *ft = &v->p.frq.tab;
369 switch (v->p.frq.tab.mode)
374 struct freq **fpp = (struct freq **) hsh_probe (ft->data, val);
380 struct freq *fp = *fpp = pool_alloc (gen_pool, sizeof *fp);
388 if (val->f == SYSMIS)
389 v->p.frq.tab.sysmis += weight;
390 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
393 if (i >= v->p.frq.tab.min && i <= v->p.frq.tab.max)
394 v->p.frq.tab.vector[i - v->p.frq.tab.min] += weight;
397 v->p.frq.tab.out_of_range += weight;
406 /* Prepares each variable that is the target of FREQUENCIES by setting
407 up its hash table. */
409 precalc (void *aux UNUSED)
413 pool_destroy (gen_pool);
414 gen_pool = pool_create ();
416 for (i = 0; i < n_variables; i++)
418 struct variable *v = v_variables[i];
420 if (v->p.frq.tab.mode == FRQM_GENERAL)
423 hsh_compare_func *compare;
425 if (v->type == NUMERIC)
427 hash = hash_value_numeric;
428 compare = compare_value_numeric_a;
432 hash = hash_value_alpha;
433 compare = compare_value_alpha_a;
435 v->p.frq.tab.data = hsh_create (16, compare, hash, NULL, v);
441 for (j = (v->p.frq.tab.max - v->p.frq.tab.min); j >= 0; j--)
442 v->p.frq.tab.vector[j] = 0.0;
443 v->p.frq.tab.out_of_range = 0.0;
444 v->p.frq.tab.sysmis = 0.0;
449 /* Finishes up with the variables after frequencies have been
450 calculated. Displays statistics, percentiles, ... */
452 postcalc (void *aux UNUSED)
456 for (i = 0; i < n_variables; i++)
458 struct variable *v = v_variables[i];
460 int dumped_freq_tab = 1;
462 postprocess_freq_tab (v);
464 /* Frequencies tables. */
465 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
466 if (cmd.table == FRQ_TABLE
467 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
477 if (n_categories > cmd.onepage_limit)
490 dump_statistics (v, !dumped_freq_tab);
492 cleanup_freq_tab (v);
496 /* Returns the comparison function that should be used for
497 sorting a frequency table by FRQ_SORT using VAR_TYPE
499 static hsh_compare_func *
500 get_freq_comparator (int frq_sort, int var_type)
502 /* Note that q2c generates tags beginning with 1000. */
503 switch (frq_sort | (var_type << 16))
505 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
506 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
507 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
508 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
509 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
510 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
511 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
512 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
519 /* Returns nonzero iff the value in struct freq F is non-missing
522 not_missing (const void *f_, void *v_)
524 const struct freq *f = f_;
525 struct variable *v = v_;
527 return !is_missing (&f->v, v);
530 /* Summarizes the frequency table data for variable V. */
532 postprocess_freq_tab (struct variable *v)
534 hsh_compare_func *compare;
538 struct freq *freqs, *f;
541 assert (v->p.frq.tab.mode == FRQM_GENERAL);
542 compare = get_freq_comparator (cmd.sort, v->type);
545 /* Extract data from hash table. */
546 count = hsh_count (ft->data);
547 data = hsh_data (ft->data);
549 /* Copy dereferenced data into freqs. */
550 freqs = xmalloc (count* sizeof *freqs);
551 for (i = 0; i < count; i++)
553 struct freq *f = data[i];
557 /* Put data into ft. */
559 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
560 ft->missing = freqs + ft->n_valid;
561 ft->n_missing = count - ft->n_valid;
564 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
565 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
567 /* Summary statistics. */
568 ft->total_cases = ft->valid_cases = 0.0;
569 for (f = ft->valid; f < ft->valid + ft->n_valid; f++)
571 ft->total_cases += f->c;
573 if ((v->type != NUMERIC || f->v.f != SYSMIS)
574 && (cmd.miss != FRQ_EXCLUDE || !is_user_missing (&f->v, v)))
575 ft->valid_cases += f->c;
579 /* Frees the frequency table for variable V. */
581 cleanup_freq_tab (struct variable *v)
583 assert (v->p.frq.tab.mode == FRQM_GENERAL);
584 free (v->p.frq.tab.valid);
585 hsh_destroy (v->p.frq.tab.data);
588 /* Parses the VARIABLES subcommand, adding to
589 {n_variables,v_variables}. */
591 frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
594 int min = 0, max = 0;
596 int old_n_variables = n_variables;
600 if (token != T_ALL && (token != T_ID
601 || dict_lookup_var (default_dict, tokid) == NULL))
604 if (!parse_variables (default_dict, &v_variables, &n_variables,
605 PV_APPEND | PV_NO_SCRATCH))
608 for (i = old_n_variables; i < n_variables; i++)
609 v_variables[i]->p.frq.tab.mode = FRQM_GENERAL;
611 if (!lex_match ('('))
616 if (!lex_force_int ())
618 min = lex_integer ();
620 if (!lex_force_match (','))
622 if (!lex_force_int ())
624 max = lex_integer ();
626 if (!lex_force_match (')'))
630 msg (SE, _("Upper limit of integer mode value range must be "
631 "greater than lower limit."));
636 for (i = old_n_variables; i < n_variables; i++)
638 struct variable *v = v_variables[i];
640 if (v->p.frq.used != 0)
642 msg (SE, _("Variable %s specified multiple times on VARIABLES "
643 "subcommand."), v->name);
647 v->p.frq.used = 1; /* Used simply as a marker. */
649 v->p.frq.tab.valid = v->p.frq.tab.missing = NULL;
651 if (mode == FRQM_INTEGER)
653 if (v->type != NUMERIC)
655 msg (SE, _("Integer mode specified, but %s is not a numeric "
656 "variable."), v->name);
660 v->p.frq.tab.min = min;
661 v->p.frq.tab.max = max;
662 v->p.frq.tab.vector = pool_alloc (int_pool,
663 sizeof (struct freq) * (max - min + 1));
666 v->p.frq.tab.vector = NULL;
668 v->p.frq.n_groups = 0;
669 v->p.frq.groups = NULL;
674 /* Parses the GROUPED subcommand, setting the frq.{n_grouped,grouped}
675 fields of specified variables. */
677 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
680 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
686 /* Max, current size of list; list itself. */
694 if (!parse_variables (default_dict, &v, &n,
695 PV_NO_DUPLICATE | PV_NUMERIC))
701 while (token == T_NUM)
706 dl = pool_realloc (int_pool, dl, ml * sizeof (double));
712 /* Note that nl might still be 0 and dl might still be
713 NULL. That's okay. */
714 if (!lex_match (')'))
717 msg (SE, _("`)' expected after GROUPED interval list."));
727 for (i = 0; i < n; i++)
729 if (v[i]->p.frq.used == 0)
730 msg (SE, _("Variables %s specified on GROUPED but not on "
731 "VARIABLES."), v[i]->name);
732 if (v[i]->p.frq.groups != NULL)
733 msg (SE, _("Variables %s specified multiple times on GROUPED "
734 "subcommand."), v[i]->name);
737 v[i]->p.frq.n_groups = nl;
738 v[i]->p.frq.groups = dl;
742 if (!lex_match ('/'))
744 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
755 /* Adds X to the list of percentiles, keeping the list in proper
758 add_percentile (double x)
762 for (i = 0; i < n_percentiles; i++)
763 if (x <= percentiles[i])
765 if (i >= n_percentiles || tokval != percentiles[i])
768 = pool_realloc (int_pool, percentiles,
769 (n_percentiles + 1) * sizeof *percentiles);
771 = pool_realloc (int_pool, percentile_values,
772 (n_percentiles + 1) * sizeof *percentile_values);
773 if (i < n_percentiles)
775 memmove (&percentiles[i + 1], &percentiles[i],
776 (n_percentiles - i) * sizeof *percentiles);
777 memmove (&percentile_values[i + 1], &percentile_values[i],
778 (n_percentiles - i) * sizeof *percentile_values);
785 /* Parses the PERCENTILES subcommand, adding user-specified
786 percentiles to the list. */
788 frq_custom_percentiles (struct cmd_frequencies *cmd UNUSED)
793 msg (SE, _("Percentile list expected after PERCENTILES."));
799 if (tokval <= 0 || tokval >= 100)
801 msg (SE, _("Percentiles must be greater than "
802 "0 and less than 100."));
806 add_percentile (tokval / 100.0);
810 while (token == T_NUM);
814 /* Parses the NTILES subcommand, adding the percentiles that
815 correspond to the specified evenly-distributed ntiles. */
817 frq_custom_ntiles (struct cmd_frequencies *cmd UNUSED)
822 if (!lex_force_int ())
824 for (i = 1; i < lex_integer (); i++)
825 add_percentile (1.0 / lex_integer () * i);
830 /* Comparison functions. */
832 /* Hash of numeric values. */
834 hash_value_numeric (const void *value_, void *foo UNUSED)
836 const struct freq *value = value_;
837 return hsh_hash_double (value->v.f);
840 /* Hash of string values. */
842 hash_value_alpha (const void *value_, void *v_)
844 const struct freq *value = value_;
845 struct variable *v = v_;
847 return hsh_hash_bytes (value->v.s, v->width);
850 /* Ascending numeric compare of values. */
852 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
854 const struct freq *a = a_;
855 const struct freq *b = b_;
859 else if (a->v.f < b->v.f)
865 /* Ascending string compare of values. */
867 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
869 const struct freq *a = a_;
870 const struct freq *b = b_;
871 const struct variable *v = v_;
873 return memcmp (a->v.s, b->v.s, v->width);
876 /* Descending numeric compare of values. */
878 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
880 return -compare_value_numeric_a (a, b, foo);
883 /* Descending string compare of values. */
885 compare_value_alpha_d (const void *a, const void *b, void *v)
887 return -compare_value_alpha_a (a, b, v);
890 /* Ascending numeric compare of frequency;
891 secondary key on ascending numeric value. */
893 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
895 const struct freq *a = a_;
896 const struct freq *b = b_;
900 else if (a->c < b->c)
905 else if (a->v.f < b->v.f)
911 /* Ascending numeric compare of frequency;
912 secondary key on ascending string value. */
914 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
916 const struct freq *a = a_;
917 const struct freq *b = b_;
918 const struct variable *v = v_;
922 else if (a->c < b->c)
925 return memcmp (a->v.s, b->v.s, v->width);
928 /* Descending numeric compare of frequency;
929 secondary key on ascending numeric value. */
931 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
933 const struct freq *a = a_;
934 const struct freq *b = b_;
938 else if (a->c < b->c)
943 else if (a->v.f < b->v.f)
949 /* Descending numeric compare of frequency;
950 secondary key on ascending string value. */
952 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
954 const struct freq *a = a_;
955 const struct freq *b = b_;
956 const struct variable *v = v_;
960 else if (a->c < b->c)
963 return memcmp (a->v.s, b->v.s, v->width);
966 /* Frequency table display. */
968 /* Sets the widths of all the columns and heights of all the rows in
969 table T for driver D. */
971 full_dim (struct tab_table *t, struct outp_driver *d)
973 int lab = cmd.labels == FRQ_LABELS;
977 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
978 for (i = lab; i < lab + 5; i++)
979 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
980 for (i = 0; i < t->nr; i++)
981 t->h[i] = d->font_height;
984 /* Displays a full frequency table for variable V. */
986 dump_full (struct variable * v)
992 double cum_total = 0.0;
993 double cum_freq = 0.0;
1003 static struct init vec[] =
1005 {4, 0, N_("Valid")},
1007 {1, 1, N_("Value")},
1008 {2, 1, N_("Frequency")},
1009 {3, 1, N_("Percent")},
1010 {4, 1, N_("Percent")},
1011 {5, 1, N_("Percent")},
1019 int lab = cmd.labels == FRQ_LABELS;
1021 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1022 t = tab_create (5 + lab, n_categories + 3, 0);
1023 tab_headers (t, 0, 0, 2, 0);
1024 tab_dim (t, full_dim);
1027 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1028 for (p = vec; p->s; p++)
1029 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1030 TAB_CENTER | TAT_TITLE, gettext (p->s));
1033 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1035 double percent, valid_percent;
1039 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1040 valid_percent = f->c / v->p.frq.tab.valid_cases * 100.0;
1041 cum_total += valid_percent;
1045 const char *label = val_labs_find (v->val_labs, f->v);
1047 tab_text (t, 0, r, TAB_LEFT, label);
1050 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1051 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1052 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1053 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1054 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1057 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1063 const char *label = val_labs_find (v->val_labs, f->v);
1065 tab_text (t, 0, r, TAB_LEFT, label);
1068 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1069 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1070 tab_float (t, 2 + lab, r, TAB_NONE,
1071 f->c / v->p.frq.tab.total_cases * 100.0, 5, 1);
1072 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1076 tab_box (t, TAL_1, TAL_1,
1077 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1079 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1080 tab_hline (t, TAL_2, 0, 4 + lab, r);
1081 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1082 tab_vline (t, TAL_0, 1, r, r);
1083 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1084 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1085 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1087 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1091 /* Sets the widths of all the columns and heights of all the rows in
1092 table T for driver D. */
1094 condensed_dim (struct tab_table *t, struct outp_driver *d)
1096 int cum_w = max (outp_string_width (d, _("Cum")),
1097 max (outp_string_width (d, _("Cum")),
1098 outp_string_width (d, "000")));
1102 for (i = 0; i < 2; i++)
1103 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1104 for (i = 2; i < 4; i++)
1106 for (i = 0; i < t->nr; i++)
1107 t->h[i] = d->font_height;
1110 /* Display condensed frequency table for variable V. */
1112 dump_condensed (struct variable * v)
1116 struct tab_table *t;
1118 double cum_total = 0.0;
1120 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1121 t = tab_create (4, n_categories + 2, 0);
1123 tab_headers (t, 0, 0, 2, 0);
1124 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1125 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1126 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1127 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1128 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1129 tab_dim (t, condensed_dim);
1132 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1136 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1137 cum_total += f->c / v->p.frq.tab.valid_cases * 100.0;
1139 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1140 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1141 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1142 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1145 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1147 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1148 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1149 tab_float (t, 2, r, TAB_NONE,
1150 f->c / v->p.frq.tab.total_cases * 100.0, 3, 0);
1154 tab_box (t, TAL_1, TAL_1,
1155 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1157 tab_hline (t, TAL_2, 0, 3, 2);
1158 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1159 tab_columns (t, SOM_COL_DOWN, 1);
1163 /* Statistical display. */
1165 /* Calculates all the pertinent statistics for variable V, putting
1166 them in array D[]. FIXME: This could be made much more optimal. */
1168 calc_stats (struct variable * v, double d[frq_n_stats])
1170 double W = v->p.frq.tab.valid_cases;
1171 double X_bar, X_mode, M2, M3, M4;
1177 double previous_value;
1179 /* Calculate the mean. */
1181 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1182 X_bar += f->v.f * f->c;
1185 /* Calculate percentiles. */
1187 previous_value = SYSMIS;
1188 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1191 for (; i < n_percentiles; i++)
1193 if (cum_total / v->p.frq.tab.valid_cases < percentiles[i])
1196 percentile_values[i] = previous_value;
1198 previous_value = f->v.f;
1201 /* Calculate the mode. */
1204 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1206 if (most_often < f->c)
1211 else if (most_often == f->c)
1213 /* A duplicate mode is undefined.
1214 FIXME: keep track of *all* the modes. */
1219 /* Calculate moments about the mean. */
1221 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1223 double dev = f->v.f - X_bar;
1233 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1234 d[frq_min] = v->p.frq.tab.valid[0].v.f;
1235 d[frq_max] = v->p.frq.tab.valid[v->p.frq.tab.n_valid - 1].v.f;
1236 d[frq_mode] = X_mode;
1237 d[frq_range] = d[frq_max] - d[frq_min];
1238 d[frq_median] = SYSMIS;
1239 d[frq_mean] = X_bar;
1240 d[frq_sum] = X_bar * W;
1241 d[frq_variance] = M2 / (W - 1);
1242 d[frq_stddev] = sqrt (d[frq_variance]);
1243 d[frq_semean] = d[frq_stddev] / sqrt (W);
1244 if (W >= 3.0 && d[frq_variance] > 0)
1246 double S = d[frq_stddev];
1247 d[frq_skew] = (W * M3 / ((W - 1.0) * (W - 2.0) * S * S * S));
1248 d[frq_seskew] = sqrt (6.0 * W * (W - 1.0)
1249 / ((W - 2.0) * (W + 1.0) * (W + 3.0)));
1253 d[frq_skew] = d[frq_seskew] = SYSMIS;
1255 if (W >= 4.0 && d[frq_variance] > 0)
1257 double S2 = d[frq_variance];
1258 double SE_g1 = d[frq_seskew];
1260 d[frq_kurt] = ((W * (W + 1.0) * M4 - 3.0 * M2 * M2 * (W - 1.0))
1261 / ((W - 1.0) * (W - 2.0) * (W - 3.0) * S2 * S2));
1262 d[frq_sekurt] = sqrt ((4.0 * (W * W - 1.0) * SE_g1 * SE_g1)
1263 / ((W - 3.0) * (W + 5.0)));
1267 d[frq_kurt] = d[frq_sekurt] = SYSMIS;
1271 /* Displays a table of all the statistics requested for variable V. */
1273 dump_statistics (struct variable * v, int show_varname)
1275 double stat_value[frq_n_stats];
1276 struct tab_table *t;
1279 if (v->type == ALPHA)
1281 if (v->p.frq.tab.n_valid == 0)
1283 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1287 calc_stats (v, stat_value);
1289 t = tab_create (2, n_stats + n_percentiles, 0);
1290 tab_dim (t, tab_natural_dimensions);
1291 tab_vline (t, TAL_1 | TAL_SPACING, 1, 0, n_stats - 1);
1292 for (i = r = 0; i < frq_n_stats; i++)
1293 if (stats & BIT_INDEX (i))
1295 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1296 gettext (st_name[i].s10));
1297 tab_float (t, 1, r, TAB_NONE, stat_value[i], 11, 3);
1301 for (i = 0; i < n_percentiles; i++, r++)
1305 ds_init (gen_pool, &ds, 20);
1306 ds_printf (&ds, "%s %d", _("Percentile"), (int) (percentiles[i] * 100));
1308 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, ds.string);
1309 tab_float (t, 1, r, TAB_NONE, percentile_values[i], 11, 3);
1314 tab_columns (t, SOM_COL_DOWN, 1);
1318 tab_title (t, 1, "%s: %s", v->name, v->label);
1320 tab_title (t, 0, v->name);
1323 tab_flags (t, SOMF_NO_TITLE);