1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 2009, 2010, 2011, 2013, 2015, 2016 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/casegrouper.h"
22 #include "data/casereader.h"
23 #include "data/dataset.h"
24 #include "data/dictionary.h"
25 #include "data/format.h"
26 #include "data/missing-values.h"
27 #include "language/command.h"
28 #include "language/lexer/lexer.h"
29 #include "language/lexer/variable-parser.h"
30 #include "libpspp/message.h"
31 #include "libpspp/misc.h"
32 #include "libpspp/str.h"
33 #include "math/moments.h"
34 #include "output/tab.h"
35 #include "output/text-item.h"
38 #define _(msgid) gettext (msgid)
39 #define N_(msgid) msgid
43 const struct variable **items;
46 double sum_of_variances;
47 double variance_of_sums;
48 int totals_idx; /* Casereader index into the totals */
50 struct moments1 **m ; /* Moments of the items */
51 struct moments1 *total ; /* Moments of the totals */
56 dump_cronbach (const struct cronbach *s)
59 printf ("N items %d\n", s->n_items);
60 for (i = 0 ; i < s->n_items; ++i)
62 printf ("%s\n", var_get_name (s->items[i]));
65 printf ("Totals idx %d\n", s->totals_idx);
67 printf ("scale variance %g\n", s->variance_of_sums);
68 printf ("alpha %g\n", s->alpha);
82 SUMMARY_TOTAL = 0x0001,
88 const struct variable **variables;
90 enum mv_class exclude;
97 struct string scale_name;
103 enum summary_opts summary;
105 const struct variable *wv;
109 static bool run_reliability (struct dataset *ds, const struct reliability *reliability);
112 reliability_destroy (struct reliability *rel)
115 ds_destroy (&rel->scale_name);
117 for (j = 0; j < rel->n_sc ; ++j)
120 free (rel->sc[j].items);
121 moments1_destroy (rel->sc[j].total);
123 for (x = 0; x < rel->sc[j].n_items; ++x)
124 free (rel->sc[j].m[x]);
129 free (rel->variables);
133 cmd_reliability (struct lexer *lexer, struct dataset *ds)
135 const struct dictionary *dict = dataset_dict (ds);
137 struct reliability reliability;
138 reliability.n_variables = 0;
139 reliability.variables = NULL;
140 reliability.model = MODEL_ALPHA;
141 reliability.exclude = MV_ANY;
142 reliability.summary = 0;
143 reliability.n_sc = 0;
144 reliability.sc = NULL;
145 reliability.wv = dict_get_weight (dict);
146 reliability.total_start = 0;
147 ds_init_empty (&reliability.scale_name);
150 lex_match (lexer, T_SLASH);
152 if (!lex_force_match_id (lexer, "VARIABLES"))
157 lex_match (lexer, T_EQUALS);
159 if (!parse_variables_const (lexer, dict, &reliability.variables, &reliability.n_variables,
160 PV_NO_DUPLICATE | PV_NUMERIC))
163 if (reliability.n_variables < 2)
164 msg (MW, _("Reliability on a single variable is not useful."));
170 /* Create a default Scale */
172 reliability.n_sc = 1;
173 reliability.sc = xzalloc (sizeof (struct cronbach) * reliability.n_sc);
175 ds_assign_cstr (&reliability.scale_name, "ANY");
177 c = &reliability.sc[0];
178 c->n_items = reliability.n_variables;
179 c->items = xzalloc (sizeof (struct variable*) * c->n_items);
181 for (i = 0 ; i < c->n_items ; ++i)
182 c->items[i] = reliability.variables[i];
187 while (lex_token (lexer) != T_ENDCMD)
189 lex_match (lexer, T_SLASH);
191 if (lex_match_id (lexer, "SCALE"))
193 struct const_var_set *vs;
194 if ( ! lex_force_match (lexer, T_LPAREN))
197 if ( ! lex_force_string (lexer) )
200 ds_assign_substring (&reliability.scale_name, lex_tokss (lexer));
204 if ( ! lex_force_match (lexer, T_RPAREN))
207 lex_match (lexer, T_EQUALS);
209 vs = const_var_set_create_from_array (reliability.variables, reliability.n_variables);
211 free (reliability.sc->items);
212 if (!parse_const_var_set_vars (lexer, vs, &reliability.sc->items, &reliability.sc->n_items, 0))
214 const_var_set_destroy (vs);
218 const_var_set_destroy (vs);
220 else if (lex_match_id (lexer, "MODEL"))
222 lex_match (lexer, T_EQUALS);
223 if (lex_match_id (lexer, "ALPHA"))
225 reliability.model = MODEL_ALPHA;
227 else if (lex_match_id (lexer, "SPLIT"))
229 reliability.model = MODEL_SPLIT;
230 reliability.split_point = -1;
232 if ( lex_match (lexer, T_LPAREN)
233 && lex_force_num (lexer))
235 reliability.split_point = lex_number (lexer);
237 if (! lex_force_match (lexer, T_RPAREN))
244 else if (lex_match_id (lexer, "SUMMARY"))
246 lex_match (lexer, T_EQUALS);
247 if (lex_match_id (lexer, "TOTAL"))
249 reliability.summary |= SUMMARY_TOTAL;
251 else if (lex_match (lexer, T_ALL))
253 reliability.summary = 0xFFFF;
258 else if (lex_match_id (lexer, "MISSING"))
260 lex_match (lexer, T_EQUALS);
261 while (lex_token (lexer) != T_ENDCMD && lex_token (lexer) != T_SLASH)
263 if (lex_match_id (lexer, "INCLUDE"))
265 reliability.exclude = MV_SYSTEM;
267 else if (lex_match_id (lexer, "EXCLUDE"))
269 reliability.exclude = MV_ANY;
273 lex_error (lexer, NULL);
278 else if (lex_match_id (lexer, "STATISTICS"))
280 lex_match (lexer, T_EQUALS);
281 msg (SW, _("The STATISTICS subcommand is not yet implemented. "
282 "No statistics will be produced."));
283 while (lex_match (lexer, T_ID))
288 lex_error (lexer, NULL);
293 if ( reliability.model == MODEL_SPLIT)
296 const struct cronbach *s;
298 if ( reliability.split_point >= reliability.n_variables)
300 msg (ME, _("The split point must be less than the number of variables"));
304 reliability.n_sc += 2 ;
305 reliability.sc = xrealloc (reliability.sc, sizeof (struct cronbach) * reliability.n_sc);
307 s = &reliability.sc[0];
309 reliability.sc[1].n_items =
310 (reliability.split_point == -1) ? s->n_items / 2 : reliability.split_point;
312 reliability.sc[2].n_items = s->n_items - reliability.sc[1].n_items;
313 reliability.sc[1].items = xzalloc (sizeof (struct variable *)
314 * reliability.sc[1].n_items);
316 reliability.sc[2].items = xzalloc (sizeof (struct variable *) *
317 reliability.sc[2].n_items);
319 for (i = 0; i < reliability.sc[1].n_items ; ++i)
320 reliability.sc[1].items[i] = s->items[i];
322 while (i < s->n_items)
324 reliability.sc[2].items[i - reliability.sc[1].n_items] = s->items[i];
329 if ( reliability.summary & SUMMARY_TOTAL)
332 const int base_sc = reliability.n_sc;
334 reliability.total_start = base_sc;
336 reliability.n_sc += reliability.sc[0].n_items ;
337 reliability.sc = xrealloc (reliability.sc, sizeof (struct cronbach) * reliability.n_sc);
340 for (i = 0 ; i < reliability.sc[0].n_items; ++i )
344 struct cronbach *s = &reliability.sc[i + base_sc];
346 s->n_items = reliability.sc[0].n_items - 1;
347 s->items = xzalloc (sizeof (struct variable *) * s->n_items);
348 for (v_src = 0 ; v_src < reliability.sc[0].n_items ; ++v_src)
351 s->items[v_dest++] = reliability.sc[0].items[v_src];
357 if ( ! run_reliability (ds, &reliability))
360 reliability_destroy (&reliability);
364 reliability_destroy (&reliability);
370 do_reliability (struct casereader *group, struct dataset *ds,
371 const struct reliability *rel);
374 static void reliability_summary_total (const struct reliability *rel);
376 static void reliability_statistics (const struct reliability *rel);
380 run_reliability (struct dataset *ds, const struct reliability *reliability)
382 struct dictionary *dict = dataset_dict (ds);
384 struct casereader *group;
386 struct casegrouper *grouper = casegrouper_create_splits (proc_open (ds), dict);
389 for (si = 0 ; si < reliability->n_sc; ++si)
391 struct cronbach *s = &reliability->sc[si];
394 s->m = xzalloc (sizeof *s->m * s->n_items);
395 s->total = moments1_create (MOMENT_VARIANCE);
397 for (i = 0 ; i < s->n_items ; ++i )
398 s->m[i] = moments1_create (MOMENT_VARIANCE);
402 while (casegrouper_get_next_group (grouper, &group))
404 do_reliability (group, ds, reliability);
406 reliability_statistics (reliability);
408 if (reliability->summary & SUMMARY_TOTAL )
409 reliability_summary_total (reliability);
412 ok = casegrouper_destroy (grouper);
413 ok = proc_commit (ds) && ok;
422 /* Return the sum of all the item variables in S */
424 append_sum (const struct ccase *c, casenumber n UNUSED, void *aux)
427 const struct cronbach *s = aux;
430 for (v = 0 ; v < s->n_items; ++v)
432 sum += case_data (c, s->items[v])->f;
439 case_processing_summary (casenumber n_valid, casenumber n_missing,
440 const struct dictionary *dict);
444 alpha (int k, double sum_of_variances, double variance_of_sums)
446 return k / ( k - 1.0) * ( 1 - sum_of_variances / variance_of_sums);
450 do_reliability (struct casereader *input, struct dataset *ds,
451 const struct reliability *rel)
456 casenumber n_missing ;
457 casenumber n_valid = 0;
460 for (si = 0 ; si < rel->n_sc; ++si)
462 struct cronbach *s = &rel->sc[si];
464 moments1_clear (s->total);
466 for (i = 0 ; i < s->n_items ; ++i )
467 moments1_clear (s->m[i]);
470 input = casereader_create_filter_missing (input,
477 for (si = 0 ; si < rel->n_sc; ++si)
479 struct cronbach *s = &rel->sc[si];
482 s->totals_idx = caseproto_get_n_widths (casereader_get_proto (input));
484 casereader_create_append_numeric (input, append_sum,
488 for (; (c = casereader_read (input)) != NULL; case_unref (c))
493 for (si = 0; si < rel->n_sc; ++si)
495 struct cronbach *s = &rel->sc[si];
497 for (i = 0 ; i < s->n_items ; ++i )
498 moments1_add (s->m[i], case_data (c, s->items[i])->f, weight);
500 moments1_add (s->total, case_data_idx (c, s->totals_idx)->f, weight);
503 casereader_destroy (input);
505 for (si = 0; si < rel->n_sc; ++si)
507 struct cronbach *s = &rel->sc[si];
509 s->sum_of_variances = 0;
510 for (i = 0 ; i < s->n_items ; ++i )
512 double weight, mean, variance;
513 moments1_calculate (s->m[i], &weight, &mean, &variance, NULL, NULL);
515 s->sum_of_variances += variance;
518 moments1_calculate (s->total, NULL, NULL, &s->variance_of_sums,
522 alpha (s->n_items, s->sum_of_variances, s->variance_of_sums);
525 text_item_submit (text_item_create_format (TEXT_ITEM_PARAGRAPH, _("Scale: %s"),
526 ds_cstr (&rel->scale_name)));
528 case_processing_summary (n_valid, n_missing, dataset_dict (ds));
536 case_processing_summary (casenumber n_valid, casenumber n_missing,
537 const struct dictionary *dict)
539 const struct variable *wv = dict_get_weight (dict);
540 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
545 int heading_columns = 2;
546 int heading_rows = 1;
547 struct tab_table *tbl;
548 tbl = tab_create (n_cols, n_rows);
549 tab_set_format (tbl, RC_WEIGHT, wfmt);
550 tab_headers (tbl, heading_columns, 0, heading_rows, 0);
552 tab_title (tbl, _("Case Processing Summary"));
554 /* Vertical lines for the data only */
559 n_cols - 1, n_rows - 1);
561 /* Box around table */
566 n_cols - 1, n_rows - 1);
569 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows);
571 tab_vline (tbl, TAL_2, heading_columns, 0, n_rows - 1);
574 tab_text (tbl, 0, heading_rows, TAB_LEFT | TAT_TITLE,
577 tab_text (tbl, 1, heading_rows, TAB_LEFT | TAT_TITLE,
580 tab_text (tbl, 1, heading_rows + 1, TAB_LEFT | TAT_TITLE,
583 tab_text (tbl, 1, heading_rows + 2, TAB_LEFT | TAT_TITLE,
586 tab_text (tbl, heading_columns, 0, TAB_CENTER | TAT_TITLE,
589 tab_text (tbl, heading_columns + 1, 0, TAB_CENTER | TAT_TITLE, _("%"));
591 total = n_missing + n_valid;
593 tab_double (tbl, 2, heading_rows, TAB_RIGHT,
594 n_valid, NULL, RC_WEIGHT);
597 tab_double (tbl, 2, heading_rows + 1, TAB_RIGHT,
598 n_missing, NULL, RC_WEIGHT);
601 tab_double (tbl, 2, heading_rows + 2, TAB_RIGHT,
602 total, NULL, RC_WEIGHT);
605 tab_double (tbl, 3, heading_rows, TAB_RIGHT,
606 100 * n_valid / (double) total, NULL, RC_OTHER);
609 tab_double (tbl, 3, heading_rows + 1, TAB_RIGHT,
610 100 * n_missing / (double) total, NULL, RC_OTHER);
613 tab_double (tbl, 3, heading_rows + 2, TAB_RIGHT,
614 100 * total / (double) total, NULL, RC_OTHER);
623 reliability_summary_total (const struct reliability *rel)
626 const int n_cols = 5;
627 const int heading_columns = 1;
628 const int heading_rows = 1;
629 const int n_rows = rel->sc[0].n_items + heading_rows ;
630 const struct variable *wv = rel->wv;
631 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
632 struct tab_table *tbl = tab_create (n_cols, n_rows);
633 tab_set_format (tbl, RC_WEIGHT, wfmt);
634 tab_headers (tbl, heading_columns, 0, heading_rows, 0);
636 tab_title (tbl, _("Item-Total Statistics"));
638 /* Vertical lines for the data only */
643 n_cols - 1, n_rows - 1);
645 /* Box around table */
650 n_cols - 1, n_rows - 1);
653 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows);
655 tab_vline (tbl, TAL_2, heading_columns, 0, n_rows - 1);
657 tab_text (tbl, 1, 0, TAB_CENTER | TAT_TITLE,
658 _("Scale Mean if Item Deleted"));
660 tab_text (tbl, 2, 0, TAB_CENTER | TAT_TITLE,
661 _("Scale Variance if Item Deleted"));
663 tab_text (tbl, 3, 0, TAB_CENTER | TAT_TITLE,
664 _("Corrected Item-Total Correlation"));
666 tab_text (tbl, 4, 0, TAB_CENTER | TAT_TITLE,
667 _("Cronbach's Alpha if Item Deleted"));
670 for (i = 0 ; i < rel->sc[0].n_items; ++i)
672 double cov, item_to_total_r;
673 double mean, weight, var;
675 const struct cronbach *s = &rel->sc[rel->total_start + i];
676 tab_text (tbl, 0, heading_rows + i, TAB_LEFT| TAT_TITLE,
677 var_to_string (rel->sc[0].items[i]));
679 moments1_calculate (s->total, &weight, &mean, &var, 0, 0);
681 tab_double (tbl, 1, heading_rows + i, TAB_RIGHT,
682 mean, NULL, RC_OTHER);
684 tab_double (tbl, 2, heading_rows + i, TAB_RIGHT,
685 s->variance_of_sums, NULL, RC_OTHER);
687 tab_double (tbl, 4, heading_rows + i, TAB_RIGHT,
688 s->alpha, NULL, RC_OTHER);
691 moments1_calculate (rel->sc[0].m[i], &weight, &mean, &var, 0,0);
692 cov = rel->sc[0].variance_of_sums + var - s->variance_of_sums;
695 item_to_total_r = (cov - var) / (sqrt(var) * sqrt (s->variance_of_sums));
698 tab_double (tbl, 3, heading_rows + i, TAB_RIGHT,
699 item_to_total_r, NULL, RC_OTHER);
707 static void reliability_statistics_model_alpha (struct tab_table *tbl,
708 const struct reliability *rel);
710 static void reliability_statistics_model_split (struct tab_table *tbl,
711 const struct reliability *rel);
714 struct reliability_output_table
720 void (*populate) (struct tab_table *, const struct reliability *);
724 static struct reliability_output_table rol[2] =
726 { 2, 2, 1, 1, reliability_statistics_model_alpha},
727 { 4, 9, 3, 0, reliability_statistics_model_split}
731 reliability_statistics (const struct reliability *rel)
733 int n_cols = rol[rel->model].n_cols;
734 int n_rows = rol[rel->model].n_rows;
735 int heading_columns = rol[rel->model].heading_cols;
736 int heading_rows = rol[rel->model].heading_rows;
737 const struct variable *wv = rel->wv;
738 const struct fmt_spec *wfmt = wv ? var_get_print_format (wv) : & F_8_0;
739 struct tab_table *tbl = tab_create (n_cols, n_rows);
740 tab_set_format (tbl, RC_WEIGHT, wfmt);
742 tab_headers (tbl, heading_columns, 0, heading_rows, 0);
744 tab_title (tbl, _("Reliability Statistics"));
746 /* Vertical lines for the data only */
751 n_cols - 1, n_rows - 1);
753 /* Box around table */
758 n_cols - 1, n_rows - 1);
761 tab_hline (tbl, TAL_2, 0, n_cols - 1, heading_rows);
763 tab_vline (tbl, TAL_2, heading_columns, 0, n_rows - 1);
765 if ( rel->model == MODEL_ALPHA )
766 reliability_statistics_model_alpha (tbl, rel);
767 else if (rel->model == MODEL_SPLIT )
768 reliability_statistics_model_split (tbl, rel);
775 reliability_statistics_model_alpha (struct tab_table *tbl,
776 const struct reliability *rel)
778 const struct cronbach *s = &rel->sc[0];
780 tab_text (tbl, 0, 0, TAB_CENTER | TAT_TITLE,
781 _("Cronbach's Alpha"));
783 tab_text (tbl, 1, 0, TAB_CENTER | TAT_TITLE,
786 tab_double (tbl, 0, 1, TAB_RIGHT, s->alpha, NULL, RC_OTHER);
788 tab_double (tbl, 1, 1, TAB_RIGHT, s->n_items, NULL, RC_WEIGHT);
793 reliability_statistics_model_split (struct tab_table *tbl,
794 const struct reliability *rel)
796 tab_text (tbl, 0, 0, TAB_LEFT,
797 _("Cronbach's Alpha"));
799 tab_text (tbl, 1, 0, TAB_LEFT,
802 tab_text (tbl, 2, 0, TAB_LEFT,
805 tab_text (tbl, 2, 1, TAB_LEFT,
808 tab_text (tbl, 1, 2, TAB_LEFT,
811 tab_text (tbl, 2, 2, TAB_LEFT,
814 tab_text (tbl, 2, 3, TAB_LEFT,
817 tab_text (tbl, 1, 4, TAB_LEFT,
818 _("Total N of Items"));
820 tab_text (tbl, 0, 5, TAB_LEFT,
821 _("Correlation Between Forms"));
823 tab_text (tbl, 0, 6, TAB_LEFT,
824 _("Spearman-Brown Coefficient"));
826 tab_text (tbl, 1, 6, TAB_LEFT,
829 tab_text (tbl, 1, 7, TAB_LEFT,
830 _("Unequal Length"));
833 tab_text (tbl, 0, 8, TAB_LEFT,
834 _("Guttman Split-Half Coefficient"));
838 tab_double (tbl, 3, 0, TAB_RIGHT, rel->sc[1].alpha, NULL, RC_OTHER);
839 tab_double (tbl, 3, 2, TAB_RIGHT, rel->sc[2].alpha, NULL, RC_OTHER);
841 tab_double (tbl, 3, 1, TAB_RIGHT, rel->sc[1].n_items, NULL, RC_WEIGHT);
842 tab_double (tbl, 3, 3, TAB_RIGHT, rel->sc[2].n_items, NULL, RC_WEIGHT);
844 tab_double (tbl, 3, 4, TAB_RIGHT,
845 rel->sc[1].n_items + rel->sc[2].n_items, NULL, RC_WEIGHT);
848 /* R is the correlation between the two parts */
849 double r = rel->sc[0].variance_of_sums -
850 rel->sc[1].variance_of_sums -
851 rel->sc[2].variance_of_sums ;
853 /* Guttman Split Half Coefficient */
854 double g = 2 * r / rel->sc[0].variance_of_sums;
856 /* Unequal Length Spearman Brown Coefficient, and
857 intermediate value used in the computation thereof */
860 r /= sqrt (rel->sc[1].variance_of_sums);
861 r /= sqrt (rel->sc[2].variance_of_sums);
864 tab_double (tbl, 3, 5, TAB_RIGHT, r, NULL, RC_OTHER);
866 /* Equal length Spearman-Brown Coefficient */
867 tab_double (tbl, 3, 6, TAB_RIGHT, 2 * r / (1.0 + r), NULL, RC_OTHER);
869 tab_double (tbl, 3, 8, TAB_RIGHT, g, NULL, RC_OTHER);
871 tmp = (1.0 - r*r) * rel->sc[1].n_items * rel->sc[2].n_items /
872 pow2 (rel->sc[0].n_items);
874 uly = sqrt( pow4 (r) + 4 * pow2 (r) * tmp);
878 tab_double (tbl, 3, 7, TAB_RIGHT, uly, NULL, RC_OTHER);