1 /* Pspp - a program for statistical analysis.
2 Copyright (C) 2008, 2009, 2010, 2011 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 "language/stats/wilcoxon.h"
23 #include <gsl/gsl_cdf.h>
26 #include "data/casereader.h"
27 #include "data/casewriter.h"
28 #include "data/dataset.h"
29 #include "data/dictionary.h"
30 #include "data/format.h"
31 #include "data/subcase.h"
32 #include "data/variable.h"
33 #include "libpspp/assertion.h"
34 #include "libpspp/message.h"
35 #include "libpspp/misc.h"
36 #include "math/sort.h"
37 #include "math/wilcoxon-sig.h"
38 #include "output/pivot-table.h"
40 #include "gl/minmax.h"
41 #include "gl/xalloc.h"
44 #define N_(msgid) msgid
45 #define _(msgid) gettext (msgid)
48 append_difference (const struct ccase *c, casenumber n UNUSED, void *aux)
50 const variable_pair *vp = aux;
52 return case_num (c, (*vp)[0]) - case_num (c, (*vp)[1]);
55 static void show_ranks_box (const struct wilcoxon_state *,
56 const struct two_sample_test *,
57 const struct dictionary *);
59 static void show_tests_box (const struct wilcoxon_state *,
60 const struct two_sample_test *,
61 bool exact, double timer);
66 distinct_callback (double v UNUSED, casenumber n, double w UNUSED, void *aux)
68 struct wilcoxon_state *ws = aux;
70 ws->tiebreaker += pow3 (n) - n;
76 wilcoxon_execute (const struct dataset *ds,
77 struct casereader *input,
78 enum mv_class exclude,
79 const struct npar_test *test,
85 const struct dictionary *dict = dataset_dict (ds);
86 const struct two_sample_test *t2s = UP_CAST (test, const struct two_sample_test, parent);
88 struct wilcoxon_state *ws = XCALLOC (t2s->n_pairs, struct wilcoxon_state);
89 const struct variable *weight = dict_get_weight (dict);
90 struct variable *weightx = dict_create_internal_var (WEIGHT_IDX, 0);
91 struct caseproto *proto;
94 casereader_create_filter_weight (input, dict, &warn, NULL);
96 proto = caseproto_create ();
97 proto = caseproto_add_width (proto, 0);
98 proto = caseproto_add_width (proto, 0);
100 proto = caseproto_add_width (proto, 0);
102 for (i = 0 ; i < t2s->n_pairs; ++i)
104 struct casereader *r = casereader_clone (input);
105 struct casewriter *writer;
107 struct subcase ordering;
108 variable_pair *vp = &t2s->pairs[i];
110 ws[i].sign = dict_create_internal_var (0, 0);
111 ws[i].absdiff = dict_create_internal_var (1, 0);
113 r = casereader_create_filter_missing (r, *vp, 2,
117 subcase_init_var (&ordering, ws[i].absdiff, SC_ASCEND);
118 writer = sort_create_writer (&ordering, proto);
119 subcase_uninit (&ordering);
121 for (; (c = casereader_read (r)) != NULL; case_unref (c))
123 struct ccase *output = case_create (proto);
124 double d = append_difference (c, 0, vp);
127 *case_num_rw (output, ws[i].sign) = 1.0;
129 *case_num_rw (output, ws[i].sign) = -1.0;
134 w = case_num (c, weight);
136 /* Central point values should be dropped */
142 *case_num_rw (output, ws[i].absdiff) = fabs (d);
145 *case_num_rw (output, weightx) = case_num (c, weight);
147 casewriter_write (writer, output);
149 casereader_destroy (r);
150 ws[i].reader = casewriter_make_reader (writer);
152 caseproto_unref (proto);
154 for (i = 0 ; i < t2s->n_pairs; ++i)
156 struct casereader *rr ;
158 enum rank_error err = 0;
160 rr = casereader_create_append_rank (ws[i].reader, ws[i].absdiff,
161 weight ? weightx : NULL, &err,
162 distinct_callback, &ws[i]
165 for (; (c = casereader_read (rr)) != NULL; case_unref (c))
167 double sign = case_num (c, ws[i].sign);
168 double rank = case_num_idx (c, weight ? 3 : 2);
169 double w = weight ? case_num (c, weightx) : 1.0;
173 ws[i].positives.sum += rank * w;
174 ws[i].positives.n += w;
178 ws[i].negatives.sum += rank * w;
179 ws[i].negatives.n += w;
185 casereader_destroy (rr);
188 casereader_destroy (input);
190 dict_destroy_internal_var (weightx);
192 show_ranks_box (ws, t2s, dict);
193 show_tests_box (ws, t2s, exact, timer);
195 for (i = 0 ; i < t2s->n_pairs; ++i)
197 dict_destroy_internal_var (ws[i].sign);
198 dict_destroy_internal_var (ws[i].absdiff);
205 put_row (struct pivot_table *table, int var_idx, int sign_idx,
206 double n, double sum)
208 pivot_table_put3 (table, 0, sign_idx, var_idx, pivot_value_new_number (n));
211 pivot_table_put3 (table, 1, sign_idx, var_idx,
212 pivot_value_new_number (sum / n));
213 pivot_table_put3 (table, 2, sign_idx, var_idx,
214 pivot_value_new_number (sum));
219 add_pair_leaf (struct pivot_dimension *dimension, variable_pair *pair)
221 char *label = xasprintf ("%s - %s", var_to_string ((*pair)[0]),
222 var_to_string ((*pair)[1]));
223 return pivot_category_create_leaf (
225 pivot_value_new_user_text_nocopy (label));
229 show_ranks_box (const struct wilcoxon_state *ws,
230 const struct two_sample_test *t2s,
231 const struct dictionary *dict)
233 struct pivot_table *table = pivot_table_create (N_("Ranks"));
234 pivot_table_set_weight_var (table, dict_get_weight (dict));
236 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
237 N_("N"), PIVOT_RC_COUNT,
238 N_("Mean Rank"), PIVOT_RC_OTHER,
239 N_("Sum of Ranks"), PIVOT_RC_OTHER);
241 pivot_dimension_create (table, PIVOT_AXIS_ROW, N_("Sign"),
242 N_("Negative Ranks"), N_("Positive Ranks"),
243 N_("Ties"), N_("Total"));
245 struct pivot_dimension *pairs = pivot_dimension_create (
246 table, PIVOT_AXIS_ROW, N_("Pairs"));
248 for (size_t i = 0 ; i < t2s->n_pairs; ++i)
250 variable_pair *vp = &t2s->pairs[i];
251 int pair_idx = add_pair_leaf (pairs, vp);
253 const struct wilcoxon_state *w = &ws[i];
254 put_row (table, pair_idx, 0, w->negatives.n, w->negatives.sum);
255 put_row (table, pair_idx, 1, w->positives.n, w->positives.sum);
256 put_row (table, pair_idx, 2, w->n_zeros, SYSMIS);
257 put_row (table, pair_idx, 3,
258 w->n_zeros + w->positives.n + w->negatives.n, SYSMIS);
261 pivot_table_submit (table);
266 show_tests_box (const struct wilcoxon_state *ws,
267 const struct two_sample_test *t2s,
272 struct pivot_table *table = pivot_table_create (N_("Test Statistics"));
274 struct pivot_dimension *statistics = pivot_dimension_create (
275 table, PIVOT_AXIS_ROW, N_("Statistics"),
276 N_("Z"), PIVOT_RC_OTHER,
277 N_("Asymp. Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE);
279 pivot_category_create_leaves (
281 N_("Exact Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE,
282 N_("Exact Sig. (1-tailed)"), PIVOT_RC_SIGNIFICANCE);
284 struct pivot_dimension *pairs = pivot_dimension_create (
285 table, PIVOT_AXIS_COLUMN, N_("Pairs"));
287 struct pivot_footnote *too_many_pairs = pivot_table_create_footnote (
288 table, pivot_value_new_text (
289 N_("Too many pairs to calculate exact significance")));
291 for (size_t i = 0 ; i < t2s->n_pairs; ++i)
293 variable_pair *vp = &t2s->pairs[i];
294 int pair_idx = add_pair_leaf (pairs, vp);
296 double n = ws[i].positives.n + ws[i].negatives.n;
297 double z = MIN (ws[i].positives.sum, ws[i].negatives.sum);
298 z -= n * (n + 1)/ 4.0;
299 z /= sqrt (n * (n + 1) * (2*n + 1)/24.0 - ws[i].tiebreaker / 48.0);
303 entries[n_entries++] = z;
304 entries[n_entries++] = 2.0 * gsl_cdf_ugaussian_P (z);
306 int footnote_idx = -1;
309 double p = LevelOfSignificanceWXMPSR (ws[i].positives.sum, n);
312 footnote_idx = n_entries;
313 entries[n_entries++] = SYSMIS;
317 entries[n_entries++] = p;
318 entries[n_entries++] = p / 2.0;
322 for (int j = 0; j < n_entries; j++)
324 struct pivot_value *value = pivot_value_new_number (entries[j]);
325 if (j == footnote_idx)
326 pivot_value_add_footnote (value, too_many_pairs);
327 pivot_table_put2 (table, j, pair_idx, value);
331 pivot_table_submit (table);