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"
39 #include "algorithm.h"
46 #include "value-labels.h"
51 #include "debug-print.h"
56 format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
57 table:limit(n:limit,"%s>0")/notable/!table,
58 labels:!labels/nolabels,
59 sort:!avalue/dvalue/afreq/dfreq,
60 spaces:!single/double,
61 paging:newpage/!oldpage;
62 missing=miss:include/!exclude;
63 barchart(ba_)=:minimum(d:min),
65 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
66 histogram(hi_)=:minimum(d:min),
68 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
69 norm:!nonormal/normal,
70 incr:increment(d:inc,"%s>0");
71 hbar(hb_)=:minimum(d:min),
73 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
74 norm:!nonormal/normal,
75 incr:increment(d:inc,"%s>0");
79 statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
80 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
81 13|default,14|seskewness,15|sekurtosis,all,none.
86 /* Description of a statistic. */
89 int st_indx; /* Index into a_statistics[]. */
90 const char *s10; /* Identifying string. */
93 /* Table of statistics, indexed by dsc_*. */
94 static struct frq_info st_name[frq_n_stats + 1] =
96 {FRQ_ST_MEAN, N_("Mean")},
97 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
98 {FRQ_ST_MEDIAN, N_("Median")},
99 {FRQ_ST_MODE, N_("Mode")},
100 {FRQ_ST_STDDEV, N_("Std Dev")},
101 {FRQ_ST_VARIANCE, N_("Variance")},
102 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
103 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
104 {FRQ_ST_SKEWNESS, N_("Skewness")},
105 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
106 {FRQ_ST_RANGE, N_("Range")},
107 {FRQ_ST_MINIMUM, N_("Minimum")},
108 {FRQ_ST_MAXIMUM, N_("Maximum")},
109 {FRQ_ST_SUM, N_("Sum")},
113 /* Percentiles to calculate. */
117 double p; /* the %ile to be calculated */
118 double value; /* the %ile's value */
119 double x1; /* The datum value <= the percentile */
120 double x2; /* The datum value >= the percentile */
122 int flag2; /* Set to 1 if this percentile value has been found */
125 static struct percentile *percentiles;
126 static int n_percentiles;
128 static int implicit_50th ;
130 /* Groups of statistics. */
132 #define frq_default \
133 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
135 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
136 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
137 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
138 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
139 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
141 /* Statistics; number of statistics. */
142 static unsigned long stats;
145 /* Types of graphs. */
148 GFT_NONE, /* Don't draw graphs. */
149 GFT_BAR, /* Draw bar charts. */
150 GFT_HIST, /* Draw histograms. */
151 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
154 /* Parsed command. */
155 static struct cmd_frequencies cmd;
157 /* Summary of the barchart, histogram, and hbar subcommands. */
158 static int chart; /* NONE/BAR/HIST/HBAR. */
159 static double min, max; /* Minimum, maximum on y axis. */
160 static int format; /* FREQ/PERCENT: Scaling of y axis. */
161 static double scale, incr; /* FIXME */
162 static int normal; /* FIXME */
164 /* Variables for which to calculate statistics. */
165 static int n_variables;
166 static struct variable **v_variables;
168 /* Arenas used to store semi-permanent storage. */
169 static struct pool *int_pool; /* Integer mode. */
170 static struct pool *gen_pool; /* General mode. */
172 /* Easier access to a_statistics. */
173 #define stat cmd.a_statistics
175 static void determine_charts (void);
177 static void precalc (void *);
178 static int calc (struct ccase *, void *);
179 static void postcalc (void *);
181 static void postprocess_freq_tab (struct variable *);
182 static void dump_full (struct variable *);
183 static void dump_condensed (struct variable *);
184 static void dump_statistics (struct variable *, int show_varname);
185 static void cleanup_freq_tab (struct variable *);
187 static hsh_hash_func hash_value_numeric, hash_value_alpha;
188 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
189 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
190 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
191 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
193 /* Parser and outline. */
195 static int internal_cmd_frequencies (void);
198 cmd_frequencies (void)
202 int_pool = pool_create ();
203 result = internal_cmd_frequencies ();
204 pool_destroy (int_pool);
206 pool_destroy (gen_pool);
214 internal_cmd_frequencies (void)
224 for (i = 0; i < dict_get_var_cnt (default_dict); i++)
225 dict_get_var(default_dict, i)->p.frq.used = 0;
227 if (!parse_frequencies (&cmd))
230 if (cmd.onepage_limit == NOT_LONG)
231 cmd.onepage_limit = 50;
233 /* Figure out statistics to calculate. */
235 if (stat[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
236 stats |= frq_default;
237 if (stat[FRQ_ST_ALL])
239 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
240 stats &= ~frq_median;
241 for (i = 0; i < frq_n_stats; i++)
242 if (stat[st_name[i].st_indx])
243 stats |= BIT_INDEX (i);
244 if (stats & frq_kurt)
246 if (stats & frq_skew)
249 /* Calculate n_stats. */
251 for (i = 0; i < frq_n_stats; i++)
252 if ((stats & BIT_INDEX (i)))
257 if (chart != GFT_NONE || cmd.sbc_ntiles)
258 cmd.sort = FRQ_AVALUE;
261 procedure_with_splits (precalc, calc, postcalc, NULL);
266 /* Figure out which charts the user requested. */
268 determine_charts (void)
270 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) + (!!cmd.sbc_hbar);
280 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
281 "given. HBAR will be assumed. Argument values will be "
282 "given precedence increasing along the order given."));
284 else if (cmd.sbc_histogram)
286 else if (cmd.sbc_barchart)
297 if (cmd.sbc_barchart)
299 if (cmd.ba_min != SYSMIS)
301 if (cmd.ba_max != SYSMIS)
303 if (cmd.ba_scale == FRQ_FREQ)
308 else if (cmd.ba_scale == FRQ_PERCENT)
310 format = FRQ_PERCENT;
315 if (cmd.sbc_histogram)
317 if (cmd.hi_min != SYSMIS)
319 if (cmd.hi_max != SYSMIS)
321 if (cmd.hi_scale == FRQ_FREQ)
326 else if (cmd.hi_scale == FRQ_PERCENT)
328 format = FRQ_PERCENT;
333 if (cmd.hi_incr == FRQ_INCREMENT)
339 if (cmd.hb_min != SYSMIS)
341 if (cmd.hb_max != SYSMIS)
343 if (cmd.hb_scale == FRQ_FREQ)
348 else if (cmd.hb_scale == FRQ_PERCENT)
350 format = FRQ_PERCENT;
355 if (cmd.hb_incr == FRQ_INCREMENT)
359 if (min != SYSMIS && max != SYSMIS && min >= max)
361 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
362 "specified. However, MIN was specified as %g and MAX as %g. "
363 "MIN and MAX will be ignored."), min, max);
368 /* Add data from case C to the frequency table. */
370 calc (struct ccase *c, void *aux UNUSED)
376 weight = dict_get_case_weight (default_dict, c, &bad_warn);
378 for (i = 0; i < n_variables; i++)
380 struct variable *v = v_variables[i];
381 const union value *val = case_data (c, v->fv);
382 struct freq_tab *ft = &v->p.frq.tab;
384 switch (v->p.frq.tab.mode)
389 struct freq **fpp = (struct freq **) hsh_probe (ft->data, val);
395 struct freq *fp = *fpp = pool_alloc (gen_pool, sizeof *fp);
403 if (val->f == SYSMIS)
404 v->p.frq.tab.sysmis += weight;
405 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
408 if (i >= v->p.frq.tab.min && i <= v->p.frq.tab.max)
409 v->p.frq.tab.vector[i - v->p.frq.tab.min] += weight;
412 v->p.frq.tab.out_of_range += weight;
421 /* Prepares each variable that is the target of FREQUENCIES by setting
422 up its hash table. */
424 precalc (void *aux UNUSED)
428 pool_destroy (gen_pool);
429 gen_pool = pool_create ();
431 for (i = 0; i < n_variables; i++)
433 struct variable *v = v_variables[i];
435 if (v->p.frq.tab.mode == FRQM_GENERAL)
438 hsh_compare_func *compare;
440 if (v->type == NUMERIC)
442 hash = hash_value_numeric;
443 compare = compare_value_numeric_a;
447 hash = hash_value_alpha;
448 compare = compare_value_alpha_a;
450 v->p.frq.tab.data = hsh_create (16, compare, hash, NULL, v);
456 for (j = (v->p.frq.tab.max - v->p.frq.tab.min); j >= 0; j--)
457 v->p.frq.tab.vector[j] = 0.0;
458 v->p.frq.tab.out_of_range = 0.0;
459 v->p.frq.tab.sysmis = 0.0;
464 /* Finishes up with the variables after frequencies have been
465 calculated. Displays statistics, percentiles, ... */
467 postcalc (void *aux UNUSED)
471 for (i = 0; i < n_variables; i++)
473 struct variable *v = v_variables[i];
475 int dumped_freq_tab = 1;
477 postprocess_freq_tab (v);
479 /* Frequencies tables. */
480 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
481 if (cmd.table == FRQ_TABLE
482 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
492 if (n_categories > cmd.onepage_limit)
505 dump_statistics (v, !dumped_freq_tab);
507 cleanup_freq_tab (v);
511 /* Returns the comparison function that should be used for
512 sorting a frequency table by FRQ_SORT using VAR_TYPE
514 static hsh_compare_func *
515 get_freq_comparator (int frq_sort, int var_type)
517 /* Note that q2c generates tags beginning with 1000. */
518 switch (frq_sort | (var_type << 16))
520 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
521 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
522 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
523 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
524 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
525 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
526 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
527 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
534 /* Returns nonzero iff the value in struct freq F is non-missing
537 not_missing (const void *f_, void *v_)
539 const struct freq *f = f_;
540 struct variable *v = v_;
542 return !is_missing (&f->v, v);
545 /* Summarizes the frequency table data for variable V. */
547 postprocess_freq_tab (struct variable *v)
549 hsh_compare_func *compare;
553 struct freq *freqs, *f;
556 assert (v->p.frq.tab.mode == FRQM_GENERAL);
557 compare = get_freq_comparator (cmd.sort, v->type);
560 /* Extract data from hash table. */
561 count = hsh_count (ft->data);
562 data = hsh_data (ft->data);
564 /* Copy dereferenced data into freqs. */
565 freqs = xmalloc (count * sizeof *freqs);
566 for (i = 0; i < count; i++)
568 struct freq *f = data[i];
572 /* Put data into ft. */
574 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
575 ft->missing = freqs + ft->n_valid;
576 ft->n_missing = count - ft->n_valid;
579 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
580 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
582 /* Summary statistics. */
583 ft->valid_cases = 0.0;
584 for(i = 0 ; i < ft->n_valid ; ++i )
587 ft->valid_cases += f->c;
591 ft->total_cases = ft->valid_cases ;
592 for(i = 0 ; i < ft->n_missing ; ++i )
595 ft->total_cases += f->c;
600 /* Frees the frequency table for variable V. */
602 cleanup_freq_tab (struct variable *v)
604 assert (v->p.frq.tab.mode == FRQM_GENERAL);
605 free (v->p.frq.tab.valid);
606 hsh_destroy (v->p.frq.tab.data);
609 /* Parses the VARIABLES subcommand, adding to
610 {n_variables,v_variables}. */
612 frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
615 int min = 0, max = 0;
617 int old_n_variables = n_variables;
621 if (token != T_ALL && (token != T_ID
622 || dict_lookup_var (default_dict, tokid) == NULL))
625 if (!parse_variables (default_dict, &v_variables, &n_variables,
626 PV_APPEND | PV_NO_SCRATCH))
629 for (i = old_n_variables; i < n_variables; i++)
630 v_variables[i]->p.frq.tab.mode = FRQM_GENERAL;
632 if (!lex_match ('('))
637 if (!lex_force_int ())
639 min = lex_integer ();
641 if (!lex_force_match (','))
643 if (!lex_force_int ())
645 max = lex_integer ();
647 if (!lex_force_match (')'))
651 msg (SE, _("Upper limit of integer mode value range must be "
652 "greater than lower limit."));
657 for (i = old_n_variables; i < n_variables; i++)
659 struct variable *v = v_variables[i];
661 if (v->p.frq.used != 0)
663 msg (SE, _("Variable %s specified multiple times on VARIABLES "
664 "subcommand."), v->name);
668 v->p.frq.used = 1; /* Used simply as a marker. */
670 v->p.frq.tab.valid = v->p.frq.tab.missing = NULL;
672 if (mode == FRQM_INTEGER)
674 if (v->type != NUMERIC)
676 msg (SE, _("Integer mode specified, but %s is not a numeric "
677 "variable."), v->name);
681 v->p.frq.tab.min = min;
682 v->p.frq.tab.max = max;
683 v->p.frq.tab.vector = pool_alloc (int_pool,
684 sizeof (struct freq) * (max - min + 1));
687 v->p.frq.tab.vector = NULL;
689 v->p.frq.n_groups = 0;
690 v->p.frq.groups = NULL;
695 /* Parses the GROUPED subcommand, setting the frq.{n_grouped,grouped}
696 fields of specified variables. */
698 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
701 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
707 /* Max, current size of list; list itself. */
715 if (!parse_variables (default_dict, &v, &n,
716 PV_NO_DUPLICATE | PV_NUMERIC))
722 while (token == T_NUM)
727 dl = pool_realloc (int_pool, dl, ml * sizeof (double));
733 /* Note that nl might still be 0 and dl might still be
734 NULL. That's okay. */
735 if (!lex_match (')'))
738 msg (SE, _("`)' expected after GROUPED interval list."));
748 for (i = 0; i < n; i++)
750 if (v[i]->p.frq.used == 0)
751 msg (SE, _("Variables %s specified on GROUPED but not on "
752 "VARIABLES."), v[i]->name);
753 if (v[i]->p.frq.groups != NULL)
754 msg (SE, _("Variables %s specified multiple times on GROUPED "
755 "subcommand."), v[i]->name);
758 v[i]->p.frq.n_groups = nl;
759 v[i]->p.frq.groups = dl;
763 if (!lex_match ('/'))
765 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
776 /* Adds X to the list of percentiles, keeping the list in proper
779 add_percentile (double x)
783 for (i = 0; i < n_percentiles; i++)
784 if (x <= percentiles[i].p)
787 if (i >= n_percentiles || tokval != percentiles[i].p)
790 = pool_realloc (int_pool, percentiles,
791 (n_percentiles + 1) * sizeof (struct percentile ));
793 if (i < n_percentiles)
794 memmove (&percentiles[i + 1], &percentiles[i],
795 (n_percentiles - i) * sizeof (struct percentile) );
797 percentiles[i].p = x;
802 /* Parses the PERCENTILES subcommand, adding user-specified
803 percentiles to the list. */
805 frq_custom_percentiles (struct cmd_frequencies *cmd UNUSED)
810 msg (SE, _("Percentile list expected after PERCENTILES."));
816 if (tokval < 0 || tokval > 100)
818 msg (SE, _("Percentiles must be between 0 and 100."));
822 add_percentile (tokval / 100.0);
826 while (token == T_NUM);
830 /* Parses the NTILES subcommand, adding the percentiles that
831 correspond to the specified evenly-distributed ntiles. */
833 frq_custom_ntiles (struct cmd_frequencies *cmd UNUSED)
838 if (!lex_force_int ())
840 for (i = 1; i < lex_integer (); i++)
841 add_percentile (1.0 / lex_integer () * i);
846 /* Comparison functions. */
848 /* Hash of numeric values. */
850 hash_value_numeric (const void *value_, void *foo UNUSED)
852 const struct freq *value = value_;
853 return hsh_hash_double (value->v.f);
856 /* Hash of string values. */
858 hash_value_alpha (const void *value_, void *v_)
860 const struct freq *value = value_;
861 struct variable *v = v_;
863 return hsh_hash_bytes (value->v.s, v->width);
866 /* Ascending numeric compare of values. */
868 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
870 const struct freq *a = a_;
871 const struct freq *b = b_;
875 else if (a->v.f < b->v.f)
881 /* Ascending string compare of values. */
883 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
885 const struct freq *a = a_;
886 const struct freq *b = b_;
887 const struct variable *v = v_;
889 return memcmp (a->v.s, b->v.s, v->width);
892 /* Descending numeric compare of values. */
894 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
896 return -compare_value_numeric_a (a, b, foo);
899 /* Descending string compare of values. */
901 compare_value_alpha_d (const void *a, const void *b, void *v)
903 return -compare_value_alpha_a (a, b, v);
906 /* Ascending numeric compare of frequency;
907 secondary key on ascending numeric value. */
909 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
911 const struct freq *a = a_;
912 const struct freq *b = b_;
916 else if (a->c < b->c)
921 else if (a->v.f < b->v.f)
927 /* Ascending numeric compare of frequency;
928 secondary key on ascending string value. */
930 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
932 const struct freq *a = a_;
933 const struct freq *b = b_;
934 const struct variable *v = v_;
938 else if (a->c < b->c)
941 return memcmp (a->v.s, b->v.s, v->width);
944 /* Descending numeric compare of frequency;
945 secondary key on ascending numeric value. */
947 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
949 const struct freq *a = a_;
950 const struct freq *b = b_;
954 else if (a->c < b->c)
959 else if (a->v.f < b->v.f)
965 /* Descending numeric compare of frequency;
966 secondary key on ascending string value. */
968 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
970 const struct freq *a = a_;
971 const struct freq *b = b_;
972 const struct variable *v = v_;
976 else if (a->c < b->c)
979 return memcmp (a->v.s, b->v.s, v->width);
982 /* Frequency table display. */
984 /* Sets the widths of all the columns and heights of all the rows in
985 table T for driver D. */
987 full_dim (struct tab_table *t, struct outp_driver *d)
989 int lab = cmd.labels == FRQ_LABELS;
993 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
994 for (i = lab; i < lab + 5; i++)
995 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
996 for (i = 0; i < t->nr; i++)
997 t->h[i] = d->font_height;
1000 /* Displays a full frequency table for variable V. */
1002 dump_full (struct variable * v)
1006 struct tab_table *t;
1008 double cum_total = 0.0;
1009 double cum_freq = 0.0;
1019 static struct init vec[] =
1021 {4, 0, N_("Valid")},
1023 {1, 1, N_("Value")},
1024 {2, 1, N_("Frequency")},
1025 {3, 1, N_("Percent")},
1026 {4, 1, N_("Percent")},
1027 {5, 1, N_("Percent")},
1035 int lab = cmd.labels == FRQ_LABELS;
1037 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1038 t = tab_create (5 + lab, n_categories + 3, 0);
1039 tab_headers (t, 0, 0, 2, 0);
1040 tab_dim (t, full_dim);
1043 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1044 for (p = vec; p->s; p++)
1045 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1046 TAB_CENTER | TAT_TITLE, gettext (p->s));
1049 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1051 double percent, valid_percent;
1055 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1056 valid_percent = f->c / v->p.frq.tab.valid_cases * 100.0;
1057 cum_total += valid_percent;
1061 const char *label = val_labs_find (v->val_labs, f->v);
1063 tab_text (t, 0, r, TAB_LEFT, label);
1066 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1067 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1068 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1069 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1070 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1073 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1079 const char *label = val_labs_find (v->val_labs, f->v);
1081 tab_text (t, 0, r, TAB_LEFT, label);
1084 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1085 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1086 tab_float (t, 2 + lab, r, TAB_NONE,
1087 f->c / v->p.frq.tab.total_cases * 100.0, 5, 1);
1088 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1092 tab_box (t, TAL_1, TAL_1,
1093 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1095 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1096 tab_hline (t, TAL_2, 0, 4 + lab, r);
1097 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1098 tab_vline (t, TAL_0, 1, r, r);
1099 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1100 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1101 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1103 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1107 /* Sets the widths of all the columns and heights of all the rows in
1108 table T for driver D. */
1110 condensed_dim (struct tab_table *t, struct outp_driver *d)
1112 int cum_w = max (outp_string_width (d, _("Cum")),
1113 max (outp_string_width (d, _("Cum")),
1114 outp_string_width (d, "000")));
1118 for (i = 0; i < 2; i++)
1119 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1120 for (i = 2; i < 4; i++)
1122 for (i = 0; i < t->nr; i++)
1123 t->h[i] = d->font_height;
1126 /* Display condensed frequency table for variable V. */
1128 dump_condensed (struct variable * v)
1132 struct tab_table *t;
1134 double cum_total = 0.0;
1136 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1137 t = tab_create (4, n_categories + 2, 0);
1139 tab_headers (t, 0, 0, 2, 0);
1140 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1141 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1142 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1143 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1144 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1145 tab_dim (t, condensed_dim);
1148 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1152 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1153 cum_total += f->c / v->p.frq.tab.valid_cases * 100.0;
1155 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1156 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1157 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1158 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1161 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1163 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1164 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1165 tab_float (t, 2, r, TAB_NONE,
1166 f->c / v->p.frq.tab.total_cases * 100.0, 3, 0);
1170 tab_box (t, TAL_1, TAL_1,
1171 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1173 tab_hline (t, TAL_2, 0, 3, 2);
1174 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1175 tab_columns (t, SOM_COL_DOWN, 1);
1179 /* Statistical display. */
1181 /* Calculates all the pertinent statistics for variable V, putting
1182 them in array D[]. FIXME: This could be made much more optimal. */
1184 calc_stats (struct variable * v, double d[frq_n_stats])
1186 double W = v->p.frq.tab.valid_cases;
1195 double *median_value;
1197 /* Calculate percentiles. */
1199 /* If the 50th percentile was not explicitly requested then we must
1200 calculate it anyway --- it's the median */
1202 for (i = 0; i < n_percentiles; i++)
1204 if (percentiles[i].p == 0.5)
1206 median_value = &percentiles[i].value;
1211 if ( 0 == median_value )
1213 add_percentile (0.5);
1217 for (i = 0; i < n_percentiles; i++)
1219 percentiles[i].flag = 0;
1220 percentiles[i].flag2 = 0;
1224 for (idx = 0; idx < v->p.frq.tab.n_valid; ++idx)
1226 static double prev_value = SYSMIS;
1227 f = &v->p.frq.tab.valid[idx];
1229 for (i = 0; i < n_percentiles; i++)
1232 if ( percentiles[i].flag2 ) continue ;
1234 if ( get_algorithm() != COMPATIBLE )
1236 (v->p.frq.tab.valid_cases - 1) * percentiles[i].p;
1239 (v->p.frq.tab.valid_cases + 1) * percentiles[i].p - 1;
1241 if ( percentiles[i].flag )
1243 percentiles[i].x2 = f->v.f;
1244 percentiles[i].x1 = prev_value;
1245 percentiles[i].flag2 = 1;
1251 if ( f->c > 1 && rank - (f->c - 1) > tp )
1253 percentiles[i].x2 = percentiles[i].x1 = f->v.f;
1254 percentiles[i].flag2 = 1;
1258 percentiles[i].flag=1;
1264 prev_value = f->v.f;
1267 for (i = 0; i < n_percentiles; i++)
1269 /* Catches the case when p == 100% */
1270 if ( ! percentiles[i].flag2 )
1271 percentiles[i].x1 = percentiles[i].x2 = f->v.f;
1274 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1275 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1279 for (i = 0; i < n_percentiles; i++)
1281 struct freq_tab *ft = &v->p.frq.tab;
1285 if ( get_algorithm() != COMPATIBLE )
1287 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1291 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1294 percentiles[i].value = percentiles[i].x1 +
1295 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1297 if ( percentiles[i].p == 0.50)
1298 median_value = &percentiles[i].value;
1302 /* Calculate the mode. */
1305 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1307 if (most_often < f->c)
1312 else if (most_often == f->c)
1314 /* A duplicate mode is undefined.
1315 FIXME: keep track of *all* the modes. */
1320 /* Calculate moments. */
1321 m = moments_create (MOMENT_KURTOSIS);
1322 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1323 moments_pass_one (m, f->v.f, f->c);
1324 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1325 moments_pass_two (m, f->v.f, f->c);
1326 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1327 &d[frq_skew], &d[frq_kurt]);
1328 moments_destroy (m);
1330 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1331 d[frq_min] = v->p.frq.tab.valid[0].v.f;
1332 d[frq_max] = v->p.frq.tab.valid[v->p.frq.tab.n_valid - 1].v.f;
1333 d[frq_mode] = X_mode;
1334 d[frq_range] = d[frq_max] - d[frq_min];
1335 d[frq_median] = *median_value;
1336 d[frq_sum] = d[frq_mean] * W;
1337 d[frq_stddev] = sqrt (d[frq_variance]);
1338 d[frq_semean] = d[frq_stddev] / sqrt (W);
1339 d[frq_seskew] = calc_seskew (W);
1340 d[frq_sekurt] = calc_sekurt (W);
1343 /* Displays a table of all the statistics requested for variable V. */
1345 dump_statistics (struct variable * v, int show_varname)
1347 double stat_value[frq_n_stats];
1348 struct tab_table *t;
1351 int n_explicit_percentiles = n_percentiles;
1353 if ( implicit_50th && n_percentiles > 0 )
1356 if (v->type == ALPHA)
1358 if (v->p.frq.tab.n_valid == 0)
1360 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1364 calc_stats (v, stat_value);
1366 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1367 tab_dim (t, tab_natural_dimensions);
1369 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1372 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1373 tab_vline (t, TAL_1 | TAL_SPACING , 1, 0, tab_nr(t) - 1 ) ;
1375 r=2; /* N missing and N valid are always dumped */
1377 for (i = 0; i < frq_n_stats; i++)
1378 if (stats & BIT_INDEX (i))
1380 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1381 gettext (st_name[i].s10));
1382 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1386 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1387 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1388 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1390 tab_float(t, 2, 0, TAB_NONE, v->p.frq.tab.valid_cases, 11, 0);
1391 tab_float(t, 2, 1, TAB_NONE,
1392 v->p.frq.tab.total_cases - v->p.frq.tab.valid_cases, 11, 0);
1395 for (i = 0; i < n_explicit_percentiles; i++, r++)
1399 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1402 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1403 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1407 tab_columns (t, SOM_COL_DOWN, 1);
1411 tab_title (t, 1, "%s: %s", v->name, v->label);
1413 tab_title (t, 0, v->name);
1416 tab_flags (t, SOMF_NO_TITLE);