1 /* Pspp - a program for statistical analysis.
2 Copyright (C) 2008, 2009 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <http://www.gnu.org/licenses/>. */
21 #include <data/variable.h>
22 #include <data/casereader.h>
23 #include <data/casewriter.h>
24 #include <data/subcase.h>
25 #include <math/sort.h>
26 #include <libpspp/message.h>
28 #include <output/table.h>
29 #include <data/procedure.h>
30 #include <data/dictionary.h>
31 #include <math/wilcoxon-sig.h>
32 #include <gsl/gsl_cdf.h>
35 #include <libpspp/assertion.h>
36 #include <data/format.h>
39 append_difference (const struct ccase *c, casenumber n UNUSED, void *aux)
41 const variable_pair *vp = aux;
43 return case_data (c, (*vp)[0])->f - case_data (c, (*vp)[1])->f;
46 static void show_ranks_box (const struct wilcoxon_state *,
47 const struct two_sample_test *,
48 const struct dictionary *);
50 static void show_tests_box (const struct wilcoxon_state *,
51 const struct two_sample_test *,
52 bool exact, double timer);
57 distinct_callback (double v UNUSED, casenumber n, double w UNUSED, void *aux)
59 struct wilcoxon_state *ws = aux;
61 ws->tiebreaker += pow3 (n) - n;
67 wilcoxon_execute (const struct dataset *ds,
68 struct casereader *input,
69 enum mv_class exclude,
70 const struct npar_test *test,
76 const struct dictionary *dict = dataset_dict (ds);
77 const struct two_sample_test *t2s = (struct two_sample_test *) test;
79 struct wilcoxon_state *ws = xcalloc (sizeof (*ws), t2s->n_pairs);
80 const struct variable *weight = dict_get_weight (dict);
81 struct variable *weightx = var_create_internal (WEIGHT_IDX);
84 casereader_create_filter_weight (input, dict, &warn, NULL);
86 for (i = 0 ; i < t2s->n_pairs; ++i )
88 struct casereader *r = casereader_clone (input);
89 struct casewriter *writer;
91 struct subcase ordering;
92 variable_pair *vp = &t2s->pairs[i];
94 const int reader_width = weight ? 3 : 2;
96 ws[i].sign = var_create_internal (0);
97 ws[i].absdiff = var_create_internal (1);
99 r = casereader_create_filter_missing (r, *vp, 2,
103 subcase_init_var (&ordering, ws[i].absdiff, SC_ASCEND);
104 writer = sort_create_writer (&ordering, reader_width);
105 subcase_destroy (&ordering);
107 for (; (c = casereader_read (r)) != NULL; case_unref (c))
109 struct ccase *output = case_create (reader_width);
110 double d = append_difference (c, 0, vp);
114 case_data_rw (output, ws[i].sign)->f = 1.0;
119 case_data_rw (output, ws[i].sign)->f = -1.0;
125 w = case_data (c, weight)->f;
127 /* Central point values should be dropped */
133 case_data_rw (output, ws[i].absdiff)->f = fabs (d);
136 case_data_rw (output, weightx)->f = case_data (c, weight)->f;
138 casewriter_write (writer, output);
140 casereader_destroy (r);
141 ws[i].reader = casewriter_make_reader (writer);
144 for (i = 0 ; i < t2s->n_pairs; ++i )
146 struct casereader *rr ;
148 enum rank_error err = 0;
150 rr = casereader_create_append_rank (ws[i].reader, ws[i].absdiff,
151 weight ? weightx : NULL, &err,
152 distinct_callback, &ws[i]
155 for (; (c = casereader_read (rr)) != NULL; case_unref (c))
157 double sign = case_data (c, ws[i].sign)->f;
158 double rank = case_data_idx (c, weight ? 3 : 2)->f;
161 w = case_data (c, weightx)->f;
165 ws[i].positives.sum += rank * w;
166 ws[i].positives.n += w;
170 ws[i].negatives.sum += rank * w;
171 ws[i].negatives.n += w;
177 casereader_destroy (rr);
180 casereader_destroy (input);
182 var_destroy (weightx);
184 show_ranks_box (ws, t2s, dict);
185 show_tests_box (ws, t2s, exact, timer);
187 for (i = 0 ; i < t2s->n_pairs; ++i )
189 var_destroy (ws[i].sign);
190 var_destroy (ws[i].absdiff);
200 #define _(msgid) gettext (msgid)
203 show_ranks_box (const struct wilcoxon_state *ws,
204 const struct two_sample_test *t2s,
205 const struct dictionary *dict)
209 const struct variable *wv = dict_get_weight (dict);
210 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
212 struct tab_table *table = tab_create (5, 1 + 4 * t2s->n_pairs, 0);
214 tab_dim (table, tab_natural_dimensions);
216 tab_title (table, _("Ranks"));
218 tab_headers (table, 2, 0, 1, 0);
220 /* Vertical lines inside the box */
221 tab_box (table, 0, 0, -1, TAL_1,
222 1, 0, table->nc - 1, tab_nr (table) - 1 );
224 /* Box around entire table */
225 tab_box (table, TAL_2, TAL_2, -1, -1,
226 0, 0, table->nc - 1, tab_nr (table) - 1 );
229 tab_text (table, 2, 0, TAB_CENTER, _("N"));
230 tab_text (table, 3, 0, TAB_CENTER, _("Mean Rank"));
231 tab_text (table, 4, 0, TAB_CENTER, _("Sum of Ranks"));
234 for (i = 0 ; i < t2s->n_pairs; ++i)
236 variable_pair *vp = &t2s->pairs[i];
238 struct string pair_name;
239 ds_init_cstr (&pair_name, var_to_string ((*vp)[0]));
240 ds_put_cstr (&pair_name, " - ");
241 ds_put_cstr (&pair_name, var_to_string ((*vp)[1]));
243 tab_text (table, 1, 1 + i * 4, TAB_LEFT, _("Negative Ranks"));
244 tab_text (table, 1, 2 + i * 4, TAB_LEFT, _("Positive Ranks"));
245 tab_text (table, 1, 3 + i * 4, TAB_LEFT, _("Ties"));
246 tab_text (table, 1, 4 + i * 4, TAB_LEFT, _("Total"));
248 tab_hline (table, TAL_1, 0, table->nc - 1, 1 + i * 4);
251 tab_text (table, 0, 1 + i * 4, TAB_LEFT, ds_cstr (&pair_name));
252 ds_destroy (&pair_name);
256 tab_double (table, 2, 1 + i * 4, TAB_RIGHT, ws[i].negatives.n, wfmt);
257 tab_double (table, 2, 2 + i * 4, TAB_RIGHT, ws[i].positives.n, wfmt);
258 tab_double (table, 2, 3 + i * 4, TAB_RIGHT, ws[i].n_zeros, wfmt);
260 tab_double (table, 2, 4 + i * 4, TAB_RIGHT,
261 ws[i].n_zeros + ws[i].positives.n + ws[i].negatives.n, wfmt);
264 tab_double (table, 4, 1 + i * 4, TAB_RIGHT, ws[i].negatives.sum, NULL);
265 tab_double (table, 4, 2 + i * 4, TAB_RIGHT, ws[i].positives.sum, NULL);
269 tab_double (table, 3, 1 + i * 4, TAB_RIGHT,
270 ws[i].negatives.sum / (double) ws[i].negatives.n, NULL);
272 tab_double (table, 3, 2 + i * 4, TAB_RIGHT,
273 ws[i].positives.sum / (double) ws[i].positives.n, NULL);
277 tab_hline (table, TAL_2, 0, table->nc - 1, 1);
278 tab_vline (table, TAL_2, 2, 0, table->nr - 1);
286 show_tests_box (const struct wilcoxon_state *ws,
287 const struct two_sample_test *t2s,
293 struct tab_table *table = tab_create (1 + t2s->n_pairs, exact ? 5 : 3, 0);
295 tab_dim (table, tab_natural_dimensions);
297 tab_title (table, _("Test Statistics"));
299 tab_headers (table, 1, 0, 1, 0);
301 /* Vertical lines inside the box */
302 tab_box (table, 0, 0, -1, TAL_1,
303 0, 0, table->nc - 1, tab_nr (table) - 1 );
305 /* Box around entire table */
306 tab_box (table, TAL_2, TAL_2, -1, -1,
307 0, 0, table->nc - 1, tab_nr (table) - 1 );
310 tab_text (table, 0, 1, TAB_LEFT, _("Z"));
311 tab_text (table, 0, 2, TAB_LEFT, _("Asymp. Sig (2-tailed)"));
315 tab_text (table, 0, 3, TAB_LEFT, _("Exact Sig (2-tailed)"));
316 tab_text (table, 0, 4, TAB_LEFT, _("Exact Sig (1-tailed)"));
319 tab_text (table, 0, 5, TAB_LEFT, _("Point Probability"));
323 for (i = 0 ; i < t2s->n_pairs; ++i)
326 double n = ws[i].positives.n + ws[i].negatives.n;
327 variable_pair *vp = &t2s->pairs[i];
329 struct string pair_name;
330 ds_init_cstr (&pair_name, var_to_string ((*vp)[0]));
331 ds_put_cstr (&pair_name, " - ");
332 ds_put_cstr (&pair_name, var_to_string ((*vp)[1]));
335 tab_text (table, 1 + i, 0, TAB_CENTER, ds_cstr (&pair_name));
336 ds_destroy (&pair_name);
338 z = MIN (ws[i].positives.sum, ws[i].negatives.sum);
339 z -= n * (n + 1)/ 4.0;
341 z /= sqrt (n * (n + 1) * (2*n + 1)/24.0 - ws[i].tiebreaker / 48.0);
343 tab_double (table, 1 + i, 1, TAB_RIGHT, z, NULL);
345 tab_double (table, 1 + i, 2, TAB_RIGHT,
346 2.0 * gsl_cdf_ugaussian_P (z),
351 double p = LevelOfSignificanceWXMPSR (ws[i].positives.sum, n);
354 msg (MW, ("Too many pairs to calculate exact significance."));
358 tab_double (table, 1 + i, 3, TAB_RIGHT, p, NULL);
359 tab_double (table, 1 + i, 4, TAB_RIGHT, p / 2.0, NULL);
364 tab_hline (table, TAL_2, 0, table->nc - 1, 1);
365 tab_vline (table, TAL_2, 1, 0, table->nr - 1);