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"
52 #include "debug-print.h"
57 format=cond:condense/onepage(*n:onepage_limit,"%s>=0")/!standard,
58 table:limit(n:limit,"%s>0")/notable/!table,
59 labels:!labels/nolabels,
60 sort:!avalue/dvalue/afreq/dfreq,
61 spaces:!single/double,
62 paging:newpage/!oldpage;
63 missing=miss:include/!exclude;
64 barchart(ba_)=:minimum(d:min),
66 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0");
67 piechart(pie_)=:minimum(d:min),
69 missing:missing/!nomissing;
70 histogram(hi_)=: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");
75 hbar(hb_)=:minimum(d:min),
77 scale:freq(*n:freq,"%s>0")/percent(*n:pcnt,"%s>0"),
78 norm:!nonormal/normal,
79 incr:increment(d:inc,"%s>0");
83 statistics[st_]=1|mean,2|semean,3|median,4|mode,5|stddev,6|variance,
84 7|kurtosis,8|skewness,9|range,10|minimum,11|maximum,12|sum,
85 13|default,14|seskewness,15|sekurtosis,all,none.
90 /* Description of a statistic. */
93 int st_indx; /* Index into a_statistics[]. */
94 const char *s10; /* Identifying string. */
97 /* Table of statistics, indexed by dsc_*. */
98 static struct frq_info st_name[frq_n_stats + 1] =
100 {FRQ_ST_MEAN, N_("Mean")},
101 {FRQ_ST_SEMEAN, N_("S.E. Mean")},
102 {FRQ_ST_MEDIAN, N_("Median")},
103 {FRQ_ST_MODE, N_("Mode")},
104 {FRQ_ST_STDDEV, N_("Std Dev")},
105 {FRQ_ST_VARIANCE, N_("Variance")},
106 {FRQ_ST_KURTOSIS, N_("Kurtosis")},
107 {FRQ_ST_SEKURTOSIS, N_("S.E. Kurt")},
108 {FRQ_ST_SKEWNESS, N_("Skewness")},
109 {FRQ_ST_SESKEWNESS, N_("S.E. Skew")},
110 {FRQ_ST_RANGE, N_("Range")},
111 {FRQ_ST_MINIMUM, N_("Minimum")},
112 {FRQ_ST_MAXIMUM, N_("Maximum")},
113 {FRQ_ST_SUM, N_("Sum")},
117 /* Percentiles to calculate. */
121 double p; /* the %ile to be calculated */
122 double value; /* the %ile's value */
123 double x1; /* The datum value <= the percentile */
124 double x2; /* The datum value >= the percentile */
126 int flag2; /* Set to 1 if this percentile value has been found */
129 static struct percentile *percentiles;
130 static int n_percentiles;
132 static int implicit_50th ;
134 /* Groups of statistics. */
136 #define frq_default \
137 (BI (frq_mean) | BI (frq_stddev) | BI (frq_min) | BI (frq_max))
139 (BI (frq_sum) | BI(frq_min) | BI(frq_max) \
140 | BI(frq_mean) | BI(frq_semean) | BI(frq_stddev) \
141 | BI(frq_variance) | BI(frq_kurt) | BI(frq_sekurt) \
142 | BI(frq_skew) | BI(frq_seskew) | BI(frq_range) \
143 | BI(frq_range) | BI(frq_mode) | BI(frq_median))
145 /* Statistics; number of statistics. */
146 static unsigned long stats;
149 /* Types of graphs. */
152 GFT_NONE, /* Don't draw graphs. */
153 GFT_BAR, /* Draw bar charts. */
154 GFT_HIST, /* Draw histograms. */
155 GFT_PIE, /* Draw piechart */
156 GFT_HBAR /* Draw bar charts or histograms at our discretion. */
159 /* Parsed command. */
160 static struct cmd_frequencies cmd;
162 /* Summary of the barchart, histogram, and hbar subcommands. */
163 /* FIXME: These should not be mututally exclusive */
164 static int chart; /* NONE/BAR/HIST/HBAR/PIE. */
165 static double min, max; /* Minimum, maximum on y axis. */
166 static int format; /* FREQ/PERCENT: Scaling of y axis. */
167 static double scale, incr; /* FIXME */
168 static int normal; /* FIXME */
170 /* Variables for which to calculate statistics. */
171 static int n_variables;
172 static struct variable **v_variables;
174 /* Arenas used to store semi-permanent storage. */
175 static struct pool *int_pool; /* Integer mode. */
176 static struct pool *gen_pool; /* General mode. */
178 /* Easier access to a_statistics. */
179 #define stat cmd.a_statistics
181 static void determine_charts (void);
183 static void calc_stats (struct variable * v, double d[frq_n_stats]);
185 static void precalc (void *);
186 static int calc (struct ccase *, void *);
187 static void postcalc (void *);
189 static void postprocess_freq_tab (struct variable *);
190 static void dump_full (struct variable *);
191 static void dump_condensed (struct variable *);
192 static void dump_statistics (struct variable *, int show_varname);
193 static void cleanup_freq_tab (struct variable *);
195 static hsh_hash_func hash_value_numeric, hash_value_alpha;
196 static hsh_compare_func compare_value_numeric_a, compare_value_alpha_a;
197 static hsh_compare_func compare_value_numeric_d, compare_value_alpha_d;
198 static hsh_compare_func compare_freq_numeric_a, compare_freq_alpha_a;
199 static hsh_compare_func compare_freq_numeric_d, compare_freq_alpha_d;
201 /* Parser and outline. */
203 static int internal_cmd_frequencies (void);
206 cmd_frequencies (void)
210 int_pool = pool_create ();
211 result = internal_cmd_frequencies ();
212 pool_destroy (int_pool);
214 pool_destroy (gen_pool);
222 internal_cmd_frequencies (void)
232 for (i = 0; i < dict_get_var_cnt (default_dict); i++)
233 dict_get_var(default_dict, i)->p.frq.used = 0;
235 if (!parse_frequencies (&cmd))
238 if (cmd.onepage_limit == NOT_LONG)
239 cmd.onepage_limit = 50;
241 /* Figure out statistics to calculate. */
243 if (stat[FRQ_ST_DEFAULT] || !cmd.sbc_statistics)
244 stats |= frq_default;
245 if (stat[FRQ_ST_ALL])
247 if (cmd.sort != FRQ_AVALUE && cmd.sort != FRQ_DVALUE)
248 stats &= ~frq_median;
249 for (i = 0; i < frq_n_stats; i++)
250 if (stat[st_name[i].st_indx])
251 stats |= BIT_INDEX (i);
252 if (stats & frq_kurt)
254 if (stats & frq_skew)
257 /* Calculate n_stats. */
259 for (i = 0; i < frq_n_stats; i++)
260 if ((stats & BIT_INDEX (i)))
265 if (chart != GFT_NONE || cmd.sbc_ntiles)
266 cmd.sort = FRQ_AVALUE;
269 procedure_with_splits (precalc, calc, postcalc, NULL);
274 /* Figure out which charts the user requested. */
276 determine_charts (void)
278 int count = (!!cmd.sbc_histogram) + (!!cmd.sbc_barchart) +
279 (!!cmd.sbc_hbar) + (!!cmd.sbc_piechart);
289 msg (SW, _("At most one of BARCHART, HISTOGRAM, or HBAR should be "
290 "given. HBAR will be assumed. Argument values will be "
291 "given precedence increasing along the order given."));
293 else if (cmd.sbc_histogram)
295 else if (cmd.sbc_barchart)
297 else if (cmd.sbc_piechart)
308 if (cmd.sbc_barchart)
310 if (cmd.ba_min != SYSMIS)
312 if (cmd.ba_max != SYSMIS)
314 if (cmd.ba_scale == FRQ_FREQ)
319 else if (cmd.ba_scale == FRQ_PERCENT)
321 format = FRQ_PERCENT;
326 if (cmd.sbc_histogram)
328 if (cmd.hi_min != SYSMIS)
330 if (cmd.hi_max != SYSMIS)
332 if (cmd.hi_scale == FRQ_FREQ)
337 else if (cmd.hi_scale == FRQ_PERCENT)
339 format = FRQ_PERCENT;
342 if (cmd.hi_norm != FRQ_NONORMAL )
344 if (cmd.hi_incr == FRQ_INCREMENT)
350 if (cmd.hb_min != SYSMIS)
352 if (cmd.hb_max != SYSMIS)
354 if (cmd.hb_scale == FRQ_FREQ)
359 else if (cmd.hb_scale == FRQ_PERCENT)
361 format = FRQ_PERCENT;
366 if (cmd.hb_incr == FRQ_INCREMENT)
370 if (min != SYSMIS && max != SYSMIS && min >= max)
372 msg (SE, _("MAX must be greater than or equal to MIN, if both are "
373 "specified. However, MIN was specified as %g and MAX as %g. "
374 "MIN and MAX will be ignored."), min, max);
379 /* Add data from case C to the frequency table. */
381 calc (struct ccase *c, void *aux UNUSED)
387 weight = dict_get_case_weight (default_dict, c, &bad_warn);
389 for (i = 0; i < n_variables; i++)
391 struct variable *v = v_variables[i];
392 const union value *val = case_data (c, v->fv);
393 struct freq_tab *ft = &v->p.frq.tab;
395 switch (v->p.frq.tab.mode)
401 struct freq **fpp = (struct freq **) hsh_probe (ft->data, val);
407 struct freq *fp = *fpp = pool_alloc (gen_pool, sizeof *fp);
415 if (val->f == SYSMIS)
416 v->p.frq.tab.sysmis += weight;
417 else if (val->f > INT_MIN+1 && val->f < INT_MAX-1)
420 if (i >= v->p.frq.tab.min && i <= v->p.frq.tab.max)
421 v->p.frq.tab.vector[i - v->p.frq.tab.min] += weight;
424 v->p.frq.tab.out_of_range += weight;
433 /* Prepares each variable that is the target of FREQUENCIES by setting
434 up its hash table. */
436 precalc (void *aux UNUSED)
440 pool_destroy (gen_pool);
441 gen_pool = pool_create ();
443 for (i = 0; i < n_variables; i++)
445 struct variable *v = v_variables[i];
447 if (v->p.frq.tab.mode == FRQM_GENERAL)
450 hsh_compare_func *compare;
452 if (v->type == NUMERIC)
454 hash = hash_value_numeric;
455 compare = compare_value_numeric_a;
459 hash = hash_value_alpha;
460 compare = compare_value_alpha_a;
462 v->p.frq.tab.data = hsh_create (16, compare, hash, NULL, v);
468 for (j = (v->p.frq.tab.max - v->p.frq.tab.min); j >= 0; j--)
469 v->p.frq.tab.vector[j] = 0.0;
470 v->p.frq.tab.out_of_range = 0.0;
471 v->p.frq.tab.sysmis = 0.0;
476 /* Finishes up with the variables after frequencies have been
477 calculated. Displays statistics, percentiles, ... */
479 postcalc (void *aux UNUSED)
483 for (i = 0; i < n_variables; i++)
485 struct variable *v = v_variables[i];
487 int dumped_freq_tab = 1;
489 postprocess_freq_tab (v);
491 /* Frequencies tables. */
492 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
493 if (cmd.table == FRQ_TABLE
494 || (cmd.table == FRQ_LIMIT && n_categories <= cmd.limit))
504 if (n_categories > cmd.onepage_limit)
517 dump_statistics (v, !dumped_freq_tab);
520 if ( chart == GFT_HIST)
523 double d[frq_n_stats];
524 struct frequencies_proc *frq = &v->p.frq;
526 struct normal_curve norm;
527 norm.N = frq->tab.total_cases ;
530 norm.mean = d[frq_mean];
531 norm.stddev = d[frq_stddev];
533 chart_initialise(&ch);
534 draw_histogram(&ch, v_variables[i], "HISTOGRAM",&norm,normal);
539 if ( chart == GFT_PIE)
543 chart_initialise(&ch);
545 draw_piechart(&ch, v_variables[i]);
551 cleanup_freq_tab (v);
556 /* Returns the comparison function that should be used for
557 sorting a frequency table by FRQ_SORT using VAR_TYPE
559 static hsh_compare_func *
560 get_freq_comparator (int frq_sort, int var_type)
562 /* Note that q2c generates tags beginning with 1000. */
563 switch (frq_sort | (var_type << 16))
565 case FRQ_AVALUE | (NUMERIC << 16): return compare_value_numeric_a;
566 case FRQ_AVALUE | (ALPHA << 16): return compare_value_alpha_a;
567 case FRQ_DVALUE | (NUMERIC << 16): return compare_value_numeric_d;
568 case FRQ_DVALUE | (ALPHA << 16): return compare_value_alpha_d;
569 case FRQ_AFREQ | (NUMERIC << 16): return compare_freq_numeric_a;
570 case FRQ_AFREQ | (ALPHA << 16): return compare_freq_alpha_a;
571 case FRQ_DFREQ | (NUMERIC << 16): return compare_freq_numeric_d;
572 case FRQ_DFREQ | (ALPHA << 16): return compare_freq_alpha_d;
579 /* Returns nonzero iff the value in struct freq F is non-missing
582 not_missing (const void *f_, void *v_)
584 const struct freq *f = f_;
585 struct variable *v = v_;
587 return !is_missing (&f->v, v);
590 /* Summarizes the frequency table data for variable V. */
592 postprocess_freq_tab (struct variable *v)
594 hsh_compare_func *compare;
598 struct freq *freqs, *f;
601 assert (v->p.frq.tab.mode == FRQM_GENERAL);
602 compare = get_freq_comparator (cmd.sort, v->type);
605 /* Extract data from hash table. */
606 count = hsh_count (ft->data);
607 data = hsh_data (ft->data);
609 /* Copy dereferenced data into freqs. */
610 freqs = xmalloc (count * sizeof *freqs);
611 for (i = 0; i < count; i++)
613 struct freq *f = data[i];
617 /* Put data into ft. */
619 ft->n_valid = partition (freqs, count, sizeof *freqs, not_missing, v);
620 ft->missing = freqs + ft->n_valid;
621 ft->n_missing = count - ft->n_valid;
624 sort (ft->valid, ft->n_valid, sizeof *ft->valid, compare, v);
625 sort (ft->missing, ft->n_missing, sizeof *ft->missing, compare, v);
627 /* Summary statistics. */
628 ft->valid_cases = 0.0;
629 for(i = 0 ; i < ft->n_valid ; ++i )
632 ft->valid_cases += f->c;
636 ft->total_cases = ft->valid_cases ;
637 for(i = 0 ; i < ft->n_missing ; ++i )
640 ft->total_cases += f->c;
645 /* Frees the frequency table for variable V. */
647 cleanup_freq_tab (struct variable *v)
649 assert (v->p.frq.tab.mode == FRQM_GENERAL);
650 free (v->p.frq.tab.valid);
651 hsh_destroy (v->p.frq.tab.data);
654 /* Parses the VARIABLES subcommand, adding to
655 {n_variables,v_variables}. */
657 frq_custom_variables (struct cmd_frequencies *cmd UNUSED)
660 int min = 0, max = 0;
662 int old_n_variables = n_variables;
666 if (token != T_ALL && (token != T_ID
667 || dict_lookup_var (default_dict, tokid) == NULL))
670 if (!parse_variables (default_dict, &v_variables, &n_variables,
671 PV_APPEND | PV_NO_SCRATCH))
674 for (i = old_n_variables; i < n_variables; i++)
675 v_variables[i]->p.frq.tab.mode = FRQM_GENERAL;
677 if (!lex_match ('('))
682 if (!lex_force_int ())
684 min = lex_integer ();
686 if (!lex_force_match (','))
688 if (!lex_force_int ())
690 max = lex_integer ();
692 if (!lex_force_match (')'))
696 msg (SE, _("Upper limit of integer mode value range must be "
697 "greater than lower limit."));
702 for (i = old_n_variables; i < n_variables; i++)
704 struct variable *v = v_variables[i];
706 if (v->p.frq.used != 0)
708 msg (SE, _("Variable %s specified multiple times on VARIABLES "
709 "subcommand."), v->name);
713 v->p.frq.used = 1; /* Used simply as a marker. */
715 v->p.frq.tab.valid = v->p.frq.tab.missing = NULL;
717 if (mode == FRQM_INTEGER)
719 if (v->type != NUMERIC)
721 msg (SE, _("Integer mode specified, but %s is not a numeric "
722 "variable."), v->name);
726 v->p.frq.tab.min = min;
727 v->p.frq.tab.max = max;
728 v->p.frq.tab.vector = pool_alloc (int_pool,
729 sizeof (struct freq) * (max - min + 1));
732 v->p.frq.tab.vector = NULL;
734 v->p.frq.n_groups = 0;
735 v->p.frq.groups = NULL;
740 /* Parses the GROUPED subcommand, setting the frq.{n_grouped,grouped}
741 fields of specified variables. */
743 frq_custom_grouped (struct cmd_frequencies *cmd UNUSED)
746 if ((token == T_ID && dict_lookup_var (default_dict, tokid) != NULL)
752 /* Max, current size of list; list itself. */
760 if (!parse_variables (default_dict, &v, &n,
761 PV_NO_DUPLICATE | PV_NUMERIC))
767 while (token == T_NUM)
772 dl = pool_realloc (int_pool, dl, ml * sizeof (double));
778 /* Note that nl might still be 0 and dl might still be
779 NULL. That's okay. */
780 if (!lex_match (')'))
783 msg (SE, _("`)' expected after GROUPED interval list."));
793 for (i = 0; i < n; i++)
795 if (v[i]->p.frq.used == 0)
796 msg (SE, _("Variables %s specified on GROUPED but not on "
797 "VARIABLES."), v[i]->name);
798 if (v[i]->p.frq.groups != NULL)
799 msg (SE, _("Variables %s specified multiple times on GROUPED "
800 "subcommand."), v[i]->name);
803 v[i]->p.frq.n_groups = nl;
804 v[i]->p.frq.groups = dl;
808 if (!lex_match ('/'))
810 if ((token != T_ID || dict_lookup_var (default_dict, tokid) != NULL)
821 /* Adds X to the list of percentiles, keeping the list in proper
824 add_percentile (double x)
828 for (i = 0; i < n_percentiles; i++)
829 if (x <= percentiles[i].p)
832 if (i >= n_percentiles || tokval != percentiles[i].p)
835 = pool_realloc (int_pool, percentiles,
836 (n_percentiles + 1) * sizeof (struct percentile ));
838 if (i < n_percentiles)
839 memmove (&percentiles[i + 1], &percentiles[i],
840 (n_percentiles - i) * sizeof (struct percentile) );
842 percentiles[i].p = x;
847 /* Parses the PERCENTILES subcommand, adding user-specified
848 percentiles to the list. */
850 frq_custom_percentiles (struct cmd_frequencies *cmd UNUSED)
855 msg (SE, _("Percentile list expected after PERCENTILES."));
861 if (tokval < 0 || tokval > 100)
863 msg (SE, _("Percentiles must be between 0 and 100."));
867 add_percentile (tokval / 100.0);
871 while (token == T_NUM);
875 /* Parses the NTILES subcommand, adding the percentiles that
876 correspond to the specified evenly-distributed ntiles. */
878 frq_custom_ntiles (struct cmd_frequencies *cmd UNUSED)
883 if (!lex_force_int ())
885 for (i = 1; i < lex_integer (); i++)
886 add_percentile (1.0 / lex_integer () * i);
891 /* Comparison functions. */
893 /* Hash of numeric values. */
895 hash_value_numeric (const void *value_, void *foo UNUSED)
897 const struct freq *value = value_;
898 return hsh_hash_double (value->v.f);
901 /* Hash of string values. */
903 hash_value_alpha (const void *value_, void *v_)
905 const struct freq *value = value_;
906 struct variable *v = v_;
908 return hsh_hash_bytes (value->v.s, v->width);
911 /* Ascending numeric compare of values. */
913 compare_value_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
915 const struct freq *a = a_;
916 const struct freq *b = b_;
920 else if (a->v.f < b->v.f)
926 /* Ascending string compare of values. */
928 compare_value_alpha_a (const void *a_, const void *b_, void *v_)
930 const struct freq *a = a_;
931 const struct freq *b = b_;
932 const struct variable *v = v_;
934 return memcmp (a->v.s, b->v.s, v->width);
937 /* Descending numeric compare of values. */
939 compare_value_numeric_d (const void *a, const void *b, void *foo UNUSED)
941 return -compare_value_numeric_a (a, b, foo);
944 /* Descending string compare of values. */
946 compare_value_alpha_d (const void *a, const void *b, void *v)
948 return -compare_value_alpha_a (a, b, v);
951 /* Ascending numeric compare of frequency;
952 secondary key on ascending numeric value. */
954 compare_freq_numeric_a (const void *a_, const void *b_, void *foo UNUSED)
956 const struct freq *a = a_;
957 const struct freq *b = b_;
961 else if (a->c < b->c)
966 else if (a->v.f < b->v.f)
972 /* Ascending numeric compare of frequency;
973 secondary key on ascending string value. */
975 compare_freq_alpha_a (const void *a_, const void *b_, void *v_)
977 const struct freq *a = a_;
978 const struct freq *b = b_;
979 const struct variable *v = v_;
983 else if (a->c < b->c)
986 return memcmp (a->v.s, b->v.s, v->width);
989 /* Descending numeric compare of frequency;
990 secondary key on ascending numeric value. */
992 compare_freq_numeric_d (const void *a_, const void *b_, void *foo UNUSED)
994 const struct freq *a = a_;
995 const struct freq *b = b_;
999 else if (a->c < b->c)
1002 if (a->v.f > b->v.f)
1004 else if (a->v.f < b->v.f)
1010 /* Descending numeric compare of frequency;
1011 secondary key on ascending string value. */
1013 compare_freq_alpha_d (const void *a_, const void *b_, void *v_)
1015 const struct freq *a = a_;
1016 const struct freq *b = b_;
1017 const struct variable *v = v_;
1021 else if (a->c < b->c)
1024 return memcmp (a->v.s, b->v.s, v->width);
1027 /* Frequency table display. */
1029 /* Sets the widths of all the columns and heights of all the rows in
1030 table T for driver D. */
1032 full_dim (struct tab_table *t, struct outp_driver *d)
1034 int lab = cmd.labels == FRQ_LABELS;
1038 t->w[0] = min (tab_natural_width (t, d, 0), d->prop_em_width * 15);
1039 for (i = lab; i < lab + 5; i++)
1040 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1041 for (i = 0; i < t->nr; i++)
1042 t->h[i] = d->font_height;
1045 /* Displays a full frequency table for variable V. */
1047 dump_full (struct variable * v)
1051 struct tab_table *t;
1053 double cum_total = 0.0;
1054 double cum_freq = 0.0;
1064 static struct init vec[] =
1066 {4, 0, N_("Valid")},
1068 {1, 1, N_("Value")},
1069 {2, 1, N_("Frequency")},
1070 {3, 1, N_("Percent")},
1071 {4, 1, N_("Percent")},
1072 {5, 1, N_("Percent")},
1080 int lab = cmd.labels == FRQ_LABELS;
1082 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1083 t = tab_create (5 + lab, n_categories + 3, 0);
1084 tab_headers (t, 0, 0, 2, 0);
1085 tab_dim (t, full_dim);
1088 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value Label"));
1089 for (p = vec; p->s; p++)
1090 tab_text (t, p->c - (p->r ? !lab : 0), p->r,
1091 TAB_CENTER | TAT_TITLE, gettext (p->s));
1094 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1096 double percent, valid_percent;
1100 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1101 valid_percent = f->c / v->p.frq.tab.valid_cases * 100.0;
1102 cum_total += valid_percent;
1106 const char *label = val_labs_find (v->val_labs, f->v);
1108 tab_text (t, 0, r, TAB_LEFT, label);
1111 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1112 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1113 tab_float (t, 2 + lab, r, TAB_NONE, percent, 5, 1);
1114 tab_float (t, 3 + lab, r, TAB_NONE, valid_percent, 5, 1);
1115 tab_float (t, 4 + lab, r, TAB_NONE, cum_total, 5, 1);
1118 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1124 const char *label = val_labs_find (v->val_labs, f->v);
1126 tab_text (t, 0, r, TAB_LEFT, label);
1129 tab_value (t, 0 + lab, r, TAB_NONE, &f->v, &v->print);
1130 tab_float (t, 1 + lab, r, TAB_NONE, f->c, 8, 0);
1131 tab_float (t, 2 + lab, r, TAB_NONE,
1132 f->c / v->p.frq.tab.total_cases * 100.0, 5, 1);
1133 tab_text (t, 3 + lab, r, TAB_NONE, _("Missing"));
1137 tab_box (t, TAL_1, TAL_1,
1138 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1140 tab_hline (t, TAL_2, 0, 4 + lab, 2);
1141 tab_hline (t, TAL_2, 0, 4 + lab, r);
1142 tab_joint_text (t, 0, r, 0 + lab, r, TAB_RIGHT | TAT_TITLE, _("Total"));
1143 tab_vline (t, TAL_0, 1, r, r);
1144 tab_float (t, 1 + lab, r, TAB_NONE, cum_freq, 8, 0);
1145 tab_float (t, 2 + lab, r, TAB_NONE, 100.0, 5, 1);
1146 tab_float (t, 3 + lab, r, TAB_NONE, 100.0, 5, 1);
1148 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1153 /* Sets the widths of all the columns and heights of all the rows in
1154 table T for driver D. */
1156 condensed_dim (struct tab_table *t, struct outp_driver *d)
1158 int cum_w = max (outp_string_width (d, _("Cum")),
1159 max (outp_string_width (d, _("Cum")),
1160 outp_string_width (d, "000")));
1164 for (i = 0; i < 2; i++)
1165 t->w[i] = max (tab_natural_width (t, d, i), d->prop_em_width * 8);
1166 for (i = 2; i < 4; i++)
1168 for (i = 0; i < t->nr; i++)
1169 t->h[i] = d->font_height;
1172 /* Display condensed frequency table for variable V. */
1174 dump_condensed (struct variable * v)
1178 struct tab_table *t;
1180 double cum_total = 0.0;
1182 n_categories = v->p.frq.tab.n_valid + v->p.frq.tab.n_missing;
1183 t = tab_create (4, n_categories + 2, 0);
1185 tab_headers (t, 0, 0, 2, 0);
1186 tab_text (t, 0, 1, TAB_CENTER | TAT_TITLE, _("Value"));
1187 tab_text (t, 1, 1, TAB_CENTER | TAT_TITLE, _("Freq"));
1188 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1189 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Cum"));
1190 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Pct"));
1191 tab_dim (t, condensed_dim);
1194 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1198 percent = f->c / v->p.frq.tab.total_cases * 100.0;
1199 cum_total += f->c / v->p.frq.tab.valid_cases * 100.0;
1201 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1202 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1203 tab_float (t, 2, r, TAB_NONE, percent, 3, 0);
1204 tab_float (t, 3, r, TAB_NONE, cum_total, 3, 0);
1207 for (; f < &v->p.frq.tab.valid[n_categories]; f++)
1209 tab_value (t, 0, r, TAB_NONE, &f->v, &v->print);
1210 tab_float (t, 1, r, TAB_NONE, f->c, 8, 0);
1211 tab_float (t, 2, r, TAB_NONE,
1212 f->c / v->p.frq.tab.total_cases * 100.0, 3, 0);
1216 tab_box (t, TAL_1, TAL_1,
1217 cmd.spaces == FRQ_SINGLE ? -1 : (TAL_1 | TAL_SPACING), TAL_1,
1219 tab_hline (t, TAL_2, 0, 3, 2);
1220 tab_title (t, 1, "%s: %s", v->name, v->label ? v->label : "");
1221 tab_columns (t, SOM_COL_DOWN, 1);
1225 /* Statistical display. */
1227 /* Calculates all the pertinent statistics for variable V, putting
1228 them in array D[]. FIXME: This could be made much more optimal. */
1230 calc_stats (struct variable * v, double d[frq_n_stats])
1232 double W = v->p.frq.tab.valid_cases;
1241 double *median_value;
1243 /* Calculate percentiles. */
1245 /* If the 50th percentile was not explicitly requested then we must
1246 calculate it anyway --- it's the median */
1248 for (i = 0; i < n_percentiles; i++)
1250 if (percentiles[i].p == 0.5)
1252 median_value = &percentiles[i].value;
1257 if ( 0 == median_value )
1259 add_percentile (0.5);
1263 for (i = 0; i < n_percentiles; i++)
1265 percentiles[i].flag = 0;
1266 percentiles[i].flag2 = 0;
1270 for (idx = 0; idx < v->p.frq.tab.n_valid; ++idx)
1272 static double prev_value = SYSMIS;
1273 f = &v->p.frq.tab.valid[idx];
1275 for (i = 0; i < n_percentiles; i++)
1278 if ( percentiles[i].flag2 ) continue ;
1280 if ( get_algorithm() != COMPATIBLE )
1282 (v->p.frq.tab.valid_cases - 1) * percentiles[i].p;
1285 (v->p.frq.tab.valid_cases + 1) * percentiles[i].p - 1;
1287 if ( percentiles[i].flag )
1289 percentiles[i].x2 = f->v.f;
1290 percentiles[i].x1 = prev_value;
1291 percentiles[i].flag2 = 1;
1297 if ( f->c > 1 && rank - (f->c - 1) > tp )
1299 percentiles[i].x2 = percentiles[i].x1 = f->v.f;
1300 percentiles[i].flag2 = 1;
1304 percentiles[i].flag=1;
1310 prev_value = f->v.f;
1313 for (i = 0; i < n_percentiles; i++)
1315 /* Catches the case when p == 100% */
1316 if ( ! percentiles[i].flag2 )
1317 percentiles[i].x1 = percentiles[i].x2 = f->v.f;
1320 printf("percentile %d (p==%.2f); X1 = %g; X2 = %g\n",
1321 i,percentiles[i].p,percentiles[i].x1,percentiles[i].x2);
1325 for (i = 0; i < n_percentiles; i++)
1327 struct freq_tab *ft = &v->p.frq.tab;
1331 if ( get_algorithm() != COMPATIBLE )
1333 s = modf((ft->valid_cases - 1) * percentiles[i].p , &dummy);
1337 s = modf((ft->valid_cases + 1) * percentiles[i].p -1, &dummy);
1340 percentiles[i].value = percentiles[i].x1 +
1341 ( percentiles[i].x2 - percentiles[i].x1) * s ;
1343 if ( percentiles[i].p == 0.50)
1344 median_value = &percentiles[i].value;
1348 /* Calculate the mode. */
1351 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1353 if (most_often < f->c)
1358 else if (most_often == f->c)
1360 /* A duplicate mode is undefined.
1361 FIXME: keep track of *all* the modes. */
1366 /* Calculate moments. */
1367 m = moments_create (MOMENT_KURTOSIS);
1368 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1369 moments_pass_one (m, f->v.f, f->c);
1370 for (f = v->p.frq.tab.valid; f < v->p.frq.tab.missing; f++)
1371 moments_pass_two (m, f->v.f, f->c);
1372 moments_calculate (m, NULL, &d[frq_mean], &d[frq_variance],
1373 &d[frq_skew], &d[frq_kurt]);
1374 moments_destroy (m);
1376 /* Formulas below are taken from _SPSS Statistical Algorithms_. */
1377 d[frq_min] = v->p.frq.tab.valid[0].v.f;
1378 d[frq_max] = v->p.frq.tab.valid[v->p.frq.tab.n_valid - 1].v.f;
1379 d[frq_mode] = X_mode;
1380 d[frq_range] = d[frq_max] - d[frq_min];
1381 d[frq_median] = *median_value;
1382 d[frq_sum] = d[frq_mean] * W;
1383 d[frq_stddev] = sqrt (d[frq_variance]);
1384 d[frq_semean] = d[frq_stddev] / sqrt (W);
1385 d[frq_seskew] = calc_seskew (W);
1386 d[frq_sekurt] = calc_sekurt (W);
1389 /* Displays a table of all the statistics requested for variable V. */
1391 dump_statistics (struct variable * v, int show_varname)
1393 double stat_value[frq_n_stats];
1394 struct tab_table *t;
1397 int n_explicit_percentiles = n_percentiles;
1399 if ( implicit_50th && n_percentiles > 0 )
1402 if (v->type == ALPHA)
1404 if (v->p.frq.tab.n_valid == 0)
1406 msg (SW, _("No valid data for variable %s; statistics not displayed."),
1410 calc_stats (v, stat_value);
1412 t = tab_create (3, n_stats + n_explicit_percentiles + 2, 0);
1413 tab_dim (t, tab_natural_dimensions);
1415 tab_box (t, TAL_1, TAL_1, -1, -1 , 0 , 0 , 2, tab_nr(t) - 1) ;
1418 tab_vline (t, TAL_1 , 2, 0, tab_nr(t) - 1);
1419 tab_vline (t, TAL_1 | TAL_SPACING , 1, 0, tab_nr(t) - 1 ) ;
1421 r=2; /* N missing and N valid are always dumped */
1423 for (i = 0; i < frq_n_stats; i++)
1424 if (stats & BIT_INDEX (i))
1426 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE,
1427 gettext (st_name[i].s10));
1428 tab_float (t, 2, r, TAB_NONE, stat_value[i], 11, 3);
1432 tab_text (t, 0, 0, TAB_LEFT | TAT_TITLE, _("N"));
1433 tab_text (t, 1, 0, TAB_LEFT | TAT_TITLE, _("Valid"));
1434 tab_text (t, 1, 1, TAB_LEFT | TAT_TITLE, _("Missing"));
1436 tab_float(t, 2, 0, TAB_NONE, v->p.frq.tab.valid_cases, 11, 0);
1437 tab_float(t, 2, 1, TAB_NONE,
1438 v->p.frq.tab.total_cases - v->p.frq.tab.valid_cases, 11, 0);
1441 for (i = 0; i < n_explicit_percentiles; i++, r++)
1445 tab_text (t, 0, r, TAB_LEFT | TAT_TITLE, _("Percentiles"));
1448 tab_float (t, 1, r, TAB_LEFT, percentiles[i].p * 100, 3, 0 );
1449 tab_float (t, 2, r, TAB_NONE, percentiles[i].value, 11, 3);
1453 tab_columns (t, SOM_COL_DOWN, 1);
1457 tab_title (t, 1, "%s: %s", v->name, v->label);
1459 tab_title (t, 0, v->name);
1462 tab_flags (t, SOMF_NO_TITLE);