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_data (c, (*vp)[0])->f - case_data (c, (*vp)[1])->f;
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, sizeof *ws);
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_destroy (&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);
128 case_data_rw (output, ws[i].sign)->f = 1.0;
133 case_data_rw (output, ws[i].sign)->f = -1.0;
139 w = case_data (c, weight)->f;
141 /* Central point values should be dropped */
147 case_data_rw (output, ws[i].absdiff)->f = fabs (d);
150 case_data_rw (output, weightx)->f = case_data (c, weight)->f;
152 casewriter_write (writer, output);
154 casereader_destroy (r);
155 ws[i].reader = casewriter_make_reader (writer);
157 caseproto_unref (proto);
159 for (i = 0 ; i < t2s->n_pairs; ++i)
161 struct casereader *rr ;
163 enum rank_error err = 0;
165 rr = casereader_create_append_rank (ws[i].reader, ws[i].absdiff,
166 weight ? weightx : NULL, &err,
167 distinct_callback, &ws[i]
170 for (; (c = casereader_read (rr)) != NULL; case_unref (c))
172 double sign = case_data (c, ws[i].sign)->f;
173 double rank = case_data_idx (c, weight ? 3 : 2)->f;
176 w = case_data (c, weightx)->f;
180 ws[i].positives.sum += rank * w;
181 ws[i].positives.n += w;
185 ws[i].negatives.sum += rank * w;
186 ws[i].negatives.n += w;
192 casereader_destroy (rr);
195 casereader_destroy (input);
197 dict_destroy_internal_var (weightx);
199 show_ranks_box (ws, t2s, dict);
200 show_tests_box (ws, t2s, exact, timer);
202 for (i = 0 ; i < t2s->n_pairs; ++i)
204 dict_destroy_internal_var (ws[i].sign);
205 dict_destroy_internal_var (ws[i].absdiff);
212 put_row (struct pivot_table *table, int var_idx, int sign_idx,
213 double n, double sum)
215 pivot_table_put3 (table, 0, sign_idx, var_idx, pivot_value_new_number (n));
218 pivot_table_put3 (table, 1, sign_idx, var_idx,
219 pivot_value_new_number (sum / n));
220 pivot_table_put3 (table, 2, sign_idx, var_idx,
221 pivot_value_new_number (sum));
226 add_pair_leaf (struct pivot_dimension *dimension, variable_pair *pair)
228 char *label = xasprintf ("%s - %s", var_to_string ((*pair)[0]),
229 var_to_string ((*pair)[1]));
230 return pivot_category_create_leaf (
232 pivot_value_new_user_text_nocopy (label));
236 show_ranks_box (const struct wilcoxon_state *ws,
237 const struct two_sample_test *t2s,
238 const struct dictionary *dict)
240 struct pivot_table *table = pivot_table_create (N_("Ranks"));
241 pivot_table_set_weight_var (table, dict_get_weight (dict));
243 pivot_dimension_create (table, PIVOT_AXIS_COLUMN, N_("Statistics"),
244 N_("N"), PIVOT_RC_COUNT,
245 N_("Mean Rank"), PIVOT_RC_OTHER,
246 N_("Sum of Ranks"), PIVOT_RC_OTHER);
248 pivot_dimension_create (table, PIVOT_AXIS_ROW, N_("Sign"),
249 N_("Negative Ranks"), N_("Positive Ranks"),
250 N_("Ties"), N_("Total"));
252 struct pivot_dimension *pairs = pivot_dimension_create (
253 table, PIVOT_AXIS_ROW, N_("Pairs"));
255 for (size_t i = 0 ; i < t2s->n_pairs; ++i)
257 variable_pair *vp = &t2s->pairs[i];
258 int pair_idx = add_pair_leaf (pairs, vp);
260 const struct wilcoxon_state *w = &ws[i];
261 put_row (table, pair_idx, 0, w->negatives.n, w->negatives.sum);
262 put_row (table, pair_idx, 1, w->positives.n, w->positives.sum);
263 put_row (table, pair_idx, 2, w->n_zeros, SYSMIS);
264 put_row (table, pair_idx, 3,
265 w->n_zeros + w->positives.n + w->negatives.n, SYSMIS);
268 pivot_table_submit (table);
273 show_tests_box (const struct wilcoxon_state *ws,
274 const struct two_sample_test *t2s,
279 struct pivot_table *table = pivot_table_create (N_("Test Statistics"));
281 struct pivot_dimension *statistics = pivot_dimension_create (
282 table, PIVOT_AXIS_ROW, N_("Statistics"),
283 N_("Z"), PIVOT_RC_OTHER,
284 N_("Asymp. Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE);
286 pivot_category_create_leaves (
288 N_("Exact Sig. (2-tailed)"), PIVOT_RC_SIGNIFICANCE,
289 N_("Exact Sig. (1-tailed)"), PIVOT_RC_SIGNIFICANCE);
291 struct pivot_dimension *pairs = pivot_dimension_create (
292 table, PIVOT_AXIS_COLUMN, N_("Pairs"));
294 struct pivot_footnote *too_many_pairs = pivot_table_create_footnote (
295 table, pivot_value_new_text (
296 N_("Too many pairs to calculate exact significance")));
298 for (size_t i = 0 ; i < t2s->n_pairs; ++i)
300 variable_pair *vp = &t2s->pairs[i];
301 int pair_idx = add_pair_leaf (pairs, vp);
303 double n = ws[i].positives.n + ws[i].negatives.n;
304 double z = MIN (ws[i].positives.sum, ws[i].negatives.sum);
305 z -= n * (n + 1)/ 4.0;
306 z /= sqrt (n * (n + 1) * (2*n + 1)/24.0 - ws[i].tiebreaker / 48.0);
310 entries[n_entries++] = z;
311 entries[n_entries++] = 2.0 * gsl_cdf_ugaussian_P (z);
313 int footnote_idx = -1;
316 double p = LevelOfSignificanceWXMPSR (ws[i].positives.sum, n);
319 footnote_idx = n_entries;
320 entries[n_entries++] = SYSMIS;
324 entries[n_entries++] = p;
325 entries[n_entries++] = p / 2.0;
329 for (int j = 0; j < n_entries; j++)
331 struct pivot_value *value = pivot_value_new_number (entries[j]);
332 if (j == footnote_idx)
333 pivot_value_add_footnote (value, too_many_pairs);
334 pivot_table_put2 (table, j, pair_idx, value);
338 pivot_table_submit (table);