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"
38 #include "algorithm.h"
45 #include "value-labels.h"
50 #include "debug-print.h"
55 format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
56 table:limit(n:limit,"%s>0")/notable/!table,
57 labels:!labels/nolabels,
58 sort:!avalue/dvalue/afreq/dfreq,
59 spaces:!single/double,
60 paging:newpage/!oldpage;
61 missing=miss:include/!exclude;
62 barchart(ba_)=:minimum(d:min),
64 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
65 histogram(hi_)=:minimum(d:min),
67 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
68 norm:!nonormal/normal,
69 incr:increment(d:inc,"%s>0");
70 hbar(hb_)=:minimum(d:min),
72 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
73 norm:!nonormal/normal,
74 incr:increment(d:inc,"%s>0");
78 statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
79 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
80 13|default,14|seskewness,15|sekurtosis,all,none.
85 /* Description of a statistic. */
88 int st_indx; /* Index into a_statistics[]. */
89 const char *s10; /* Identifying string. */
92 /* Table of statistics, indexed by dsc_*. */
93 static struct frq_info st_name[frq_n_stats + 1] =
95 {FRQ_ST_MEAN, N_("Mean")},
96 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
97 {FRQ_ST_MEDIAN, N_("Median")},
98 {FRQ_ST_MODE, N_("Mode")},
99 {FRQ_ST_STDDEV, N_("Std Dev")},
100 {FRQ_ST_VARIANCE, N_("Variance")},
101 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
102 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
103 {FRQ_ST_SKEWNESS, N_("Skewness")},
104 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
105 {FRQ_ST_RANGE, N_("Range")},
106 {FRQ_ST_MINIMUM, N_("Minimum")},
107 {FRQ_ST_MAXIMUM, N_("Maximum")},
108 {FRQ_ST_SUM, N_("Sum")},
112 /* Percentiles to calculate. */
116 double p; /* the %ile to be calculated */
117 double value; /* the %ile's value */
118 double x1; /* The datum value <= the percentile */
119 double x2; /* The datum value >= the percentile */
121 int flag2; /* Set to 1 if this percentile value has been found */
124 static struct percentile *percentiles;
125 static int n_percentiles;
127 static int implicit_50th ;
129 /* Groups of statistics. */
131 #define frq_default \
132 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
134 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
135 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
136 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
137 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
138 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
140 /* Statistics; number of statistics. */
141 static unsigned long stats;
144 /* Types of graphs. */
147 GFT_NONE, /* Don't draw graphs. */
148 GFT_BAR, /* Draw bar charts. */
149 GFT_HIST, /* Draw histograms. */
150 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
153 /* Parsed command. */
154 static struct cmd_frequencies cmd;
156 /* Summary of the barchart, histogram, and hbar subcommands. */
157 static int chart; /* NONE/BAR/HIST/HBAR. */
158 static double min, max; /* Minimum, maximum on y axis. */
159 static int format; /* FREQ/PERCENT: Scaling of y axis. */
160 static double scale, incr; /* FIXME */
161 static int normal; /* FIXME */
163 /* Variables for which to calculate statistics. */
164 static int n_variables;
165 static struct variable **v_variables;
167 /* Arenas used to store semi-permanent storage. */
168 static struct pool *int_pool; /* Integer mode. */
169 static struct pool *gen_pool; /* General mode. */
171 /* Easier access to a_statistics. */
172 #define stat cmd.a_statistics
174 static void determine_charts (void);
176 static void precalc (void *);
177 static int calc (struct ccase *, void *);
178 static void postcalc (void *);
180 static void postprocess_freq_tab (struct variable *);
181 static void dump_full (struct variable *);
182 static void dump_condensed (struct variable *);
183 static void dump_statistics (struct variable *, int show_varname);
184 static void cleanup_freq_tab (struct variable *);
186 static hsh_hash_func hash_value_numeric, hash_value_alpha;
187 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
188 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
189 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
190 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
192 /* Parser and outline. */
194 static int internal_cmd_frequencies (void);
197 cmd_frequencies (void)
201 int_pool = pool_create ();
202 result = internal_cmd_frequencies ();
203 pool_destroy (int_pool);
205 pool_destroy (gen_pool);
213 internal_cmd_frequencies (void)
223 for (i = 0; i < dict_get_var_cnt (default_dict); i++)
224 dict_get_var(default_dict, i)->p.frq.used = 0;
226 if (!parse_frequencies (&cmd))
229 if (cmd.onepage_limit == NOT_LONG)
230 cmd.onepage_limit = 50;
232 /* Figure out statistics to calculate. */
234 if (stat[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
235 stats |= frq_default;
236 if (stat[FRQ_ST_ALL])
238 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
239 stats &= ~frq_median;
240 for (i = 0; i < frq_n_stats; i++)
241 if (stat[st_name[i].st_indx])
242 stats |= BIT_INDEX (i);
243 if (stats & frq_kurt)
245 if (stats & frq_skew)
248 /* Calculate n_stats. */
250 for (i = 0; i < frq_n_stats; i++)
251 if ((stats & BIT_INDEX (i)))
256 if (chart != GFT_NONE || cmd.sbc_ntiles)
257 cmd.sort = FRQ_AVALUE;
260 procedure_with_splits (precalc, calc, postcalc, NULL);
265 /* Figure out which charts the user requested. */
267 determine_charts (void)
269 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) + (!!cmd.sbc_hbar);
279 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
280 "given. HBAR will be assumed. Argument values will be "
281 "given precedence increasing along the order given."));
283 else if (cmd.sbc_histogram)
285 else if (cmd.sbc_barchart)
296 if (cmd.sbc_barchart)
298 if (cmd.ba_min != SYSMIS)
300 if (cmd.ba_max != SYSMIS)
302 if (cmd.ba_scale == FRQ_FREQ)
307 else if (cmd.ba_scale == FRQ_PERCENT)
309 format = FRQ_PERCENT;
314 if (cmd.sbc_histogram)
316 if (cmd.hi_min != SYSMIS)
318 if (cmd.hi_max != SYSMIS)
320 if (cmd.hi_scale == FRQ_FREQ)
325 else if (cmd.hi_scale == FRQ_PERCENT)
327 format = FRQ_PERCENT;
332 if (cmd.hi_incr == FRQ_INCREMENT)
338 if (cmd.hb_min != SYSMIS)
340 if (cmd.hb_max != SYSMIS)
342 if (cmd.hb_scale == FRQ_FREQ)
347 else if (cmd.hb_scale == FRQ_PERCENT)
349 format = FRQ_PERCENT;
354 if (cmd.hb_incr == FRQ_INCREMENT)
358 if (min != SYSMIS && max != SYSMIS && min >= max)
360 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
361 "specified. However, MIN was specified as %g and MAX as %g. "
362 "MIN and MAX will be ignored."), min, max);
367 /* Add data from case C to the frequency table. */
369 calc (struct ccase *c, void *aux UNUSED)
375 weight = dict_get_case_weight (default_dict, c, &bad_warn);
377 for (i = 0; i < n_variables; i++)
379 struct variable *v = v_variables[i];
380 union value *val = &c->data[v->fv];
381 struct freq_tab *ft = &v->p.frq.tab;
383 switch (v->p.frq.tab.mode)
388 struct freq **fpp = (struct freq **) hsh_probe (ft->data, val);
394 struct freq *fp = *fpp = pool_alloc (gen_pool, sizeof *fp);
402 if (val->f == SYSMIS)
403 v->p.frq.tab.sysmis += weight;
404 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
407 if (i >= v->p.frq.tab.min && i <= v->p.frq.tab.max)
408 v->p.frq.tab.vector[i - v->p.frq.tab.min] += weight;
411 v->p.frq.tab.out_of_range += weight;
420 /* Prepares each variable that is the target of FREQUENCIES by setting
421 up its hash table. */
423 precalc (void *aux UNUSED)
427 pool_destroy (gen_pool);
428 gen_pool = pool_create ();
430 for (i = 0; i < n_variables; i++)
432 struct variable *v = v_variables[i];
434 if (v->p.frq.tab.mode == FRQM_GENERAL)
437 hsh_compare_func *compare;
439 if (v->type == NUMERIC)
441 hash = hash_value_numeric;
442 compare = compare_value_numeric_a;
446 hash = hash_value_alpha;
447 compare = compare_value_alpha_a;
449 v->p.frq.tab.data = hsh_create (16, compare, hash, NULL, v);
455 for (j = (v->p.frq.tab.max - v->p.frq.tab.min); j >= 0; j--)
456 v->p.frq.tab.vector[j] = 0.0;
457 v->p.frq.tab.out_of_range = 0.0;
458 v->p.frq.tab.sysmis = 0.0;
463 /* Finishes up with the variables after frequencies have been
464 calculated. Displays statistics, percentiles, ... */
466 postcalc (void *aux UNUSED)
470 for (i = 0; i < n_variables; i++)
472 struct variable *v = v_variables[i];
474 int dumped_freq_tab = 1;
476 postprocess_freq_tab (v);
478 /* Frequencies tables. */
479 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
480 if (cmd.table == FRQ_TABLE
481 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
491 if (n_categories > cmd.onepage_limit)
504 dump_statistics (v, !dumped_freq_tab);
506 cleanup_freq_tab (v);
510 /* Returns the comparison function that should be used for
511 sorting a frequency table by FRQ_SORT using VAR_TYPE
513 static hsh_compare_func *
514 get_freq_comparator (int frq_sort, int var_type)
516 /* Note that q2c generates tags beginning with 1000. */
517 switch (frq_sort | (var_type << 16))
519 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
520 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
521 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
522 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
523 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
524 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
525 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
526 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
533 /* Returns nonzero iff the value in struct freq F is non-missing
536 not_missing (const void *f_, void *v_)
538 const struct freq *f = f_;
539 struct variable *v = v_;
541 return !is_missing (&f->v, v);
544 /* Summarizes the frequency table data for variable V. */
546 postprocess_freq_tab (struct variable *v)
548 hsh_compare_func *compare;
552 struct freq *freqs, *f;
555 assert (v->p.frq.tab.mode == FRQM_GENERAL);
556 compare = get_freq_comparator (cmd.sort, v->type);
559 /* Extract data from hash table. */
560 count = hsh_count (ft->data);
561 data = hsh_data (ft->data);
563 /* Copy dereferenced data into freqs. */
564 freqs = xmalloc (count* sizeof *freqs);
565 for (i = 0; i < count; i++)
567 struct freq *f = data[i];
571 /* Put data into ft. */
573 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
574 ft->missing = freqs + ft->n_valid;
575 ft->n_missing = count - ft->n_valid;
578 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
579 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
581 /* Summary statistics. */
582 ft->valid_cases = 0.0;
583 for(i = 0 ; i < ft->n_valid ; ++i )
586 ft->valid_cases += f->c;
590 ft->total_cases = ft->valid_cases ;
591 for(i = 0 ; i < ft->n_missing ; ++i )
594 ft->total_cases += f->c;
599 /* Frees the frequency table for variable V. */
601 cleanup_freq_tab (struct variable *v)
603 assert (v->p.frq.tab.mode == FRQM_GENERAL);
604 free (v->p.frq.tab.valid);
605 hsh_destroy (v->p.frq.tab.data);
608 /* Parses the VARIABLES subcommand, adding to
609 {n_variables,v_variables}. */
611 frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
614 int min = 0, max = 0;
616 int old_n_variables = n_variables;
620 if (token != T_ALL && (token != T_ID
621 || dict_lookup_var (default_dict, tokid) == NULL))
624 if (!parse_variables (default_dict, &v_variables, &n_variables,
625 PV_APPEND | PV_NO_SCRATCH))
628 for (i = old_n_variables; i < n_variables; i++)
629 v_variables[i]->p.frq.tab.mode = FRQM_GENERAL;
631 if (!lex_match ('('))
636 if (!lex_force_int ())
638 min = lex_integer ();
640 if (!lex_force_match (','))
642 if (!lex_force_int ())
644 max = lex_integer ();
646 if (!lex_force_match (')'))
650 msg (SE, _("Upper limit of integer mode value range must be "
651 "greater than lower limit."));
656 for (i = old_n_variables; i < n_variables; i++)
658 struct variable *v = v_variables[i];
660 if (v->p.frq.used != 0)
662 msg (SE, _("Variable %s specified multiple times on VARIABLES "
663 "subcommand."), v->name);
667 v->p.frq.used = 1; /* Used simply as a marker. */
669 v->p.frq.tab.valid = v->p.frq.tab.missing = NULL;
671 if (mode == FRQM_INTEGER)
673 if (v->type != NUMERIC)
675 msg (SE, _("Integer mode specified, but %s is not a numeric "
676 "variable."), v->name);
680 v->p.frq.tab.min = min;
681 v->p.frq.tab.max = max;
682 v->p.frq.tab.vector = pool_alloc (int_pool,
683 sizeof (struct freq) * (max - min + 1));
686 v->p.frq.tab.vector = NULL;
688 v->p.frq.n_groups = 0;
689 v->p.frq.groups = NULL;
694 /* Parses the GROUPED subcommand, setting the frq.{n_grouped,grouped}
695 fields of specified variables. */
697 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
700 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
706 /* Max, current size of list; list itself. */
714 if (!parse_variables (default_dict, &v, &n,
715 PV_NO_DUPLICATE | PV_NUMERIC))
721 while (token == T_NUM)
726 dl = pool_realloc (int_pool, dl, ml * sizeof (double));
732 /* Note that nl might still be 0 and dl might still be
733 NULL. That's okay. */
734 if (!lex_match (')'))
737 msg (SE, _("`)' expected after GROUPED interval list."));
747 for (i = 0; i < n; i++)
749 if (v[i]->p.frq.used == 0)
750 msg (SE, _("Variables %s specified on GROUPED but not on "
751 "VARIABLES."), v[i]->name);
752 if (v[i]->p.frq.groups != NULL)
753 msg (SE, _("Variables %s specified multiple times on GROUPED "
754 "subcommand."), v[i]->name);
757 v[i]->p.frq.n_groups = nl;
758 v[i]->p.frq.groups = dl;
762 if (!lex_match ('/'))
764 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
775 /* Adds X to the list of percentiles, keeping the list in proper
778 add_percentile (double x)
782 for (i = 0; i < n_percentiles; i++)
783 if (x <= percentiles[i].p)
786 if (i >= n_percentiles || tokval != percentiles[i].p)
789 = pool_realloc (int_pool, percentiles,
790 (n_percentiles + 1) * sizeof (struct percentile ));
792 if (i < n_percentiles)
793 memmove (&percentiles[i + 1], &percentiles[i],
794 (n_percentiles - i) * sizeof (struct percentile) );
796 percentiles[i].p = x;
801 /* Parses the PERCENTILES subcommand, adding user-specified
802 percentiles to the list. */
804 frq_custom_percentiles (struct cmd_frequencies *cmd UNUSED)
809 msg (SE, _("Percentile list expected after PERCENTILES."));
815 if (tokval < 0 || tokval > 100)
817 msg (SE, _("Percentiles must be between 0 and 100."));
821 add_percentile (tokval / 100.0);
825 while (token == T_NUM);
829 /* Parses the NTILES subcommand, adding the percentiles that
830 correspond to the specified evenly-distributed ntiles. */
832 frq_custom_ntiles (struct cmd_frequencies *cmd UNUSED)
837 if (!lex_force_int ())
839 for (i = 1; i < lex_integer (); i++)
840 add_percentile (1.0 / lex_integer () * i);
845 /* Comparison functions. */
847 /* Hash of numeric values. */
849 hash_value_numeric (const void *value_, void *foo UNUSED)
851 const struct freq *value = value_;
852 return hsh_hash_double (value->v.f);
855 /* Hash of string values. */
857 hash_value_alpha (const void *value_, void *v_)
859 const struct freq *value = value_;
860 struct variable *v = v_;
862 return hsh_hash_bytes (value->v.s, v->width);
865 /* Ascending numeric compare of values. */
867 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
869 const struct freq *a = a_;
870 const struct freq *b = b_;
874 else if (a->v.f < b->v.f)
880 /* Ascending string compare of values. */
882 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
884 const struct freq *a = a_;
885 const struct freq *b = b_;
886 const struct variable *v = v_;
888 return memcmp (a->v.s, b->v.s, v->width);
891 /* Descending numeric compare of values. */
893 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
895 return -compare_value_numeric_a (a, b, foo);
898 /* Descending string compare of values. */
900 compare_value_alpha_d (const void *a, const void *b, void *v)
902 return -compare_value_alpha_a (a, b, v);
905 /* Ascending numeric compare of frequency;
906 secondary key on ascending numeric value. */
908 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
910 const struct freq *a = a_;
911 const struct freq *b = b_;
915 else if (a->c < b->c)
920 else if (a->v.f < b->v.f)
926 /* Ascending numeric compare of frequency;
927 secondary key on ascending string value. */
929 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
931 const struct freq *a = a_;
932 const struct freq *b = b_;
933 const struct variable *v = v_;
937 else if (a->c < b->c)
940 return memcmp (a->v.s, b->v.s, v->width);
943 /* Descending numeric compare of frequency;
944 secondary key on ascending numeric value. */
946 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
948 const struct freq *a = a_;
949 const struct freq *b = b_;
953 else if (a->c < b->c)
958 else if (a->v.f < b->v.f)
964 /* Descending numeric compare of frequency;
965 secondary key on ascending string value. */
967 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
969 const struct freq *a = a_;
970 const struct freq *b = b_;
971 const struct variable *v = v_;
975 else if (a->c < b->c)
978 return memcmp (a->v.s, b->v.s, v->width);
981 /* Frequency table display. */
983 /* Sets the widths of all the columns and heights of all the rows in
984 table T for driver D. */
986 full_dim (struct tab_table *t, struct outp_driver *d)
988 int lab = cmd.labels == FRQ_LABELS;
992 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
993 for (i = lab; i < lab + 5; i++)
994 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
995 for (i = 0; i < t->nr; i++)
996 t->h[i] = d->font_height;
999 /* Displays a full frequency table for variable V. */
1001 dump_full (struct variable * v)
1005 struct tab_table *t;
1007 double cum_total = 0.0;
1008 double cum_freq = 0.0;
1018 static struct init vec[] =
1020 {4, 0, N_("Valid")},
1022 {1, 1, N_("Value")},
1023 {2, 1, N_("Frequency")},
1024 {3, 1, N_("Percent")},
1025 {4, 1, N_("Percent")},
1026 {5, 1, N_("Percent")},
1034 int lab = cmd.labels == FRQ_LABELS;
1036 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1037 t = tab_create (5 + lab, n_categories + 3, 0);
1038 tab_headers (t, 0, 0, 2, 0);
1039 tab_dim (t, full_dim);
1042 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1043 for (p = vec; p->s; p++)
1044 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1045 TAB_CENTER | TAT_TITLE, gettext (p->s));
1048 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1050 double percent, valid_percent;
1054 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1055 valid_percent = f->c / v->p.frq.tab.valid_cases * 100.0;
1056 cum_total += valid_percent;
1060 const char *label = val_labs_find (v->val_labs, f->v);
1062 tab_text (t, 0, r, TAB_LEFT, label);
1065 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1066 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1067 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1068 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1069 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1072 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1078 const char *label = val_labs_find (v->val_labs, f->v);
1080 tab_text (t, 0, r, TAB_LEFT, label);
1083 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1084 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1085 tab_float (t, 2 + lab, r, TAB_NONE,
1086 f->c / v->p.frq.tab.total_cases * 100.0, 5, 1);
1087 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1091 tab_box (t, TAL_1, TAL_1,
1092 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1094 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1095 tab_hline (t, TAL_2, 0, 4 + lab, r);
1096 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1097 tab_vline (t, TAL_0, 1, r, r);
1098 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1099 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1100 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1102 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1106 /* Sets the widths of all the columns and heights of all the rows in
1107 table T for driver D. */
1109 condensed_dim (struct tab_table *t, struct outp_driver *d)
1111 int cum_w = max (outp_string_width (d, _("Cum")),
1112 max (outp_string_width (d, _("Cum")),
1113 outp_string_width (d, "000")));
1117 for (i = 0; i < 2; i++)
1118 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1119 for (i = 2; i < 4; i++)
1121 for (i = 0; i < t->nr; i++)
1122 t->h[i] = d->font_height;
1125 /* Display condensed frequency table for variable V. */
1127 dump_condensed (struct variable * v)
1131 struct tab_table *t;
1133 double cum_total = 0.0;
1135 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1136 t = tab_create (4, n_categories + 2, 0);
1138 tab_headers (t, 0, 0, 2, 0);
1139 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1140 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1141 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1142 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1143 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1144 tab_dim (t, condensed_dim);
1147 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1151 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1152 cum_total += f->c / v->p.frq.tab.valid_cases * 100.0;
1154 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1155 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1156 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1157 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1160 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1162 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1163 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1164 tab_float (t, 2, r, TAB_NONE,
1165 f->c / v->p.frq.tab.total_cases * 100.0, 3, 0);
1169 tab_box (t, TAL_1, TAL_1,
1170 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1172 tab_hline (t, TAL_2, 0, 3, 2);
1173 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1174 tab_columns (t, SOM_COL_DOWN, 1);
1178 /* Statistical display. */
1180 /* Calculates all the pertinent statistics for variable V, putting
1181 them in array D[]. FIXME: This could be made much more optimal. */
1183 calc_stats (struct variable * v, double d[frq_n_stats])
1185 double W = v->p.frq.tab.valid_cases;
1194 double *median_value;
1196 /* Calculate percentiles. */
1198 /* If the 50th percentile was not explicitly requested then we must
1199 calculate it anyway --- it's the median */
1201 for (i = 0; i < n_percentiles; i++)
1203 if (percentiles[i].p == 0.5)
1205 median_value = &percentiles[i].value;
1210 if ( 0 == median_value )
1212 add_percentile (0.5);
1216 for (i = 0; i < n_percentiles; i++)
1218 percentiles[i].flag = 0;
1219 percentiles[i].flag2 = 0;
1223 for (idx = 0; idx < v->p.frq.tab.n_valid; ++idx)
1225 static double prev_value = SYSMIS;
1226 f = &v->p.frq.tab.valid[idx];
1228 for (i = 0; i < n_percentiles; i++)
1231 if ( percentiles[i].flag2 ) continue ;
1233 if ( get_algorithm() != COMPATIBLE )
1235 (v->p.frq.tab.valid_cases - 1) * percentiles[i].p;
1238 (v->p.frq.tab.valid_cases + 1) * percentiles[i].p - 1;
1240 if ( percentiles[i].flag )
1242 percentiles[i].x2 = f->v.f;
1243 percentiles[i].x1 = prev_value;
1244 percentiles[i].flag2 = 1;
1250 if ( f->c > 1 && rank - (f->c - 1) > tp )
1252 percentiles[i].x2 = percentiles[i].x1 = f->v.f;
1253 percentiles[i].flag2 = 1;
1257 percentiles[i].flag=1;
1263 prev_value = f->v.f;
1266 for (i = 0; i < n_percentiles; i++)
1268 /* Catches the case when p == 100% */
1269 if ( ! percentiles[i].flag2 )
1270 percentiles[i].x1 = percentiles[i].x2 = f->v.f;
1273 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1274 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1278 for (i = 0; i < n_percentiles; i++)
1280 struct freq_tab *ft = &v->p.frq.tab;
1284 if ( get_algorithm() != COMPATIBLE )
1286 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1290 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1293 percentiles[i].value = percentiles[i].x1 +
1294 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1296 if ( percentiles[i].p == 0.50)
1297 median_value = &percentiles[i].value;
1301 /* Calculate the mode. */
1304 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1306 if (most_often < f->c)
1311 else if (most_often == f->c)
1313 /* A duplicate mode is undefined.
1314 FIXME: keep track of *all* the modes. */
1319 /* Calculate moments. */
1320 m = moments_create (MOMENT_KURTOSIS);
1321 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1322 moments_pass_one (m, f->v.f, f->c);
1323 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1324 moments_pass_two (m, f->v.f, f->c);
1325 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1326 &d[frq_skew], &d[frq_kurt]);
1327 moments_destroy (m);
1329 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1330 d[frq_min] = v->p.frq.tab.valid[0].v.f;
1331 d[frq_max] = v->p.frq.tab.valid[v->p.frq.tab.n_valid - 1].v.f;
1332 d[frq_mode] = X_mode;
1333 d[frq_range] = d[frq_max] - d[frq_min];
1334 d[frq_median] = *median_value;
1335 d[frq_sum] = d[frq_mean] * W;
1336 d[frq_stddev] = sqrt (d[frq_variance]);
1337 d[frq_semean] = d[frq_stddev] / sqrt (W);
1338 d[frq_seskew] = calc_seskew (W);
1339 d[frq_sekurt] = calc_sekurt (W);
1342 /* Displays a table of all the statistics requested for variable V. */
1344 dump_statistics (struct variable * v, int show_varname)
1346 double stat_value[frq_n_stats];
1347 struct tab_table *t;
1350 int n_explicit_percentiles = n_percentiles;
1352 if ( implicit_50th && n_percentiles > 0 )
1355 if (v->type == ALPHA)
1357 if (v->p.frq.tab.n_valid == 0)
1359 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1363 calc_stats (v, stat_value);
1365 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1366 tab_dim (t, tab_natural_dimensions);
1368 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1371 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1372 tab_vline (t, TAL_1 | TAL_SPACING , 1, 0, tab_nr(t) - 1 ) ;
1374 r=2; /* N missing and N valid are always dumped */
1376 for (i = 0; i < frq_n_stats; i++)
1377 if (stats & BIT_INDEX (i))
1379 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1380 gettext (st_name[i].s10));
1381 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1385 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1386 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1387 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1389 tab_float(t, 2, 0, TAB_NONE, v->p.frq.tab.valid_cases, 11, 0);
1390 tab_float(t, 2, 1, TAB_NONE,
1391 v->p.frq.tab.total_cases - v->p.frq.tab.valid_cases, 11, 0);
1394 for (i = 0; i < n_explicit_percentiles; i++, r++)
1398 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1401 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1402 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1406 tab_columns (t, SOM_COL_DOWN, 1);
1410 tab_title (t, 1, "%s: %s", v->name, v->label);
1412 tab_title (t, 0, v->name);
1415 tab_flags (t, SOMF_NO_TITLE);