1 /* PSPP - a program for statistical analysis.
2 Copyright (C) 1997-9, 2000, 2007, 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/>. */
19 #include <gsl/gsl_cdf.h>
24 #include <data/case.h>
25 #include <data/casegrouper.h>
26 #include <data/casereader.h>
27 #include <data/dictionary.h>
28 #include <data/procedure.h>
29 #include <data/value-labels.h>
30 #include <data/variable.h>
31 #include <language/command.h>
32 #include <language/dictionary/split-file.h>
33 #include <language/lexer/lexer.h>
34 #include <libpspp/compiler.h>
35 #include <libpspp/hash.h>
36 #include <libpspp/message.h>
37 #include <libpspp/misc.h>
38 #include <libpspp/str.h>
39 #include <libpspp/taint.h>
40 #include <math/group-proc.h>
41 #include <math/group.h>
42 #include <math/levene.h>
43 #include <output/manager.h>
44 #include <output/table.h>
45 #include "sort-criteria.h"
50 #define _(msgid) gettext (msgid)
57 missing=miss:!analysis/listwise,
58 incl:include/!exclude;
59 +contrast= double list;
60 +statistics[st_]=descriptives,homogeneity.
65 static struct cmd_oneway cmd;
67 /* The independent variable */
68 static const struct variable *indep_var;
70 /* Number of dependent variables */
73 /* The dependent variables */
74 static const struct variable **vars;
77 /* A hash table containing all the distinct values of the independent
79 static struct hsh_table *global_group_hash;
81 /* The number of distinct values of the independent variable, when all
82 missing values are disregarded */
83 static int ostensible_number_of_groups = -1;
86 static void run_oneway (struct cmd_oneway *, struct casereader *,
87 const struct dataset *);
90 /* Routines to show the output tables */
91 static void show_anova_table (void);
92 static void show_descriptives (void);
93 static void show_homogeneity (void);
95 static void show_contrast_coeffs (short *);
96 static void show_contrast_tests (short *);
99 enum stat_table_t {STAT_DESC = 1, STAT_HOMO = 2};
101 static enum stat_table_t stat_tables;
103 void output_oneway (void);
107 cmd_oneway (struct lexer *lexer, struct dataset *ds)
109 struct casegrouper *grouper;
110 struct casereader *group;
114 if ( !parse_oneway (lexer, ds, &cmd, NULL))
117 /* What statistics were requested */
118 if ( cmd.sbc_statistics)
121 for (i = 0; i < ONEWAY_ST_count; ++i)
123 if (! cmd.a_statistics[i]) continue;
127 case ONEWAY_ST_DESCRIPTIVES:
128 stat_tables |= STAT_DESC;
130 case ONEWAY_ST_HOMOGENEITY:
131 stat_tables |= STAT_HOMO;
137 /* Data pass. FIXME: error handling. */
138 grouper = casegrouper_create_splits (proc_open (ds), dataset_dict (ds));
139 while (casegrouper_get_next_group (grouper, &group))
140 run_oneway (&cmd, group, ds);
141 ok = casegrouper_destroy (grouper);
142 ok = proc_commit (ds) && ok;
147 return ok ? CMD_SUCCESS : CMD_CASCADING_FAILURE;
157 bad_contrast = xnmalloc (cmd.sbc_contrast, sizeof *bad_contrast);
159 /* Check the sanity of the given contrast values */
160 for (i = 0; i < cmd.sbc_contrast; ++i)
166 if (subc_list_double_count (&cmd.dl_contrast[i]) !=
167 ostensible_number_of_groups)
170 _("Number of contrast coefficients must equal the number of groups"));
175 for (j = 0; j < ostensible_number_of_groups; ++j)
176 sum += subc_list_double_at (&cmd.dl_contrast[i], j);
179 msg (SW, _("Coefficients for contrast %zu do not total zero"), i + 1);
182 if ( stat_tables & STAT_DESC )
183 show_descriptives ();
185 if ( stat_tables & STAT_HOMO )
190 if (cmd.sbc_contrast )
192 show_contrast_coeffs (bad_contrast);
193 show_contrast_tests (bad_contrast);
199 for (i = 0; i < n_vars; ++i )
201 struct hsh_table *group_hash = group_proc_get (vars[i])->group_hash;
203 hsh_destroy (group_hash);
206 hsh_destroy (global_group_hash);
210 /* Parser for the variables sub command */
212 oneway_custom_variables (struct lexer *lexer,
213 struct dataset *ds, struct cmd_oneway *cmd UNUSED,
216 struct dictionary *dict = dataset_dict (ds);
218 lex_match (lexer, '=');
220 if ((lex_token (lexer) != T_ID ||
221 dict_lookup_var (dict, lex_tokid (lexer)) == NULL)
222 && lex_token (lexer) != T_ALL)
225 if (!parse_variables_const (lexer, dict, &vars, &n_vars,
227 | PV_NUMERIC | PV_NO_SCRATCH) )
235 if ( ! lex_match (lexer, T_BY))
238 indep_var = parse_variable (lexer, dict);
242 msg (SE, _("`%s' is not a variable name"), lex_tokid (lexer));
250 /* Show the ANOVA table */
252 show_anova_table (void)
256 size_t n_rows = n_vars * 3 + 1;
261 t = tab_create (n_cols, n_rows, 0);
262 tab_headers (t, 2, 0, 1, 0);
263 tab_dim (t, tab_natural_dimensions);
270 n_cols - 1, n_rows - 1);
272 tab_hline (t, TAL_2, 0, n_cols - 1, 1 );
273 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
274 tab_vline (t, TAL_0, 1, 0, 0);
276 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Sum of Squares"));
277 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df"));
278 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Mean Square"));
279 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("F"));
280 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
283 for (i = 0; i < n_vars; ++i)
285 struct group_statistics *totals = &group_proc_get (vars[i])->ugs;
286 struct hsh_table *group_hash = group_proc_get (vars[i])->group_hash;
287 struct hsh_iterator g;
288 struct group_statistics *gs;
290 const char *s = var_to_string (vars[i]);
292 for (gs = hsh_first (group_hash, &g);
294 gs = hsh_next (group_hash, &g))
296 ssa += pow2 (gs->sum) / gs->n;
299 ssa -= pow2 (totals->sum) / totals->n;
301 tab_text (t, 0, i * 3 + 1, TAB_LEFT | TAT_TITLE, s);
302 tab_text (t, 1, i * 3 + 1, TAB_LEFT | TAT_TITLE, _("Between Groups"));
303 tab_text (t, 1, i * 3 + 2, TAB_LEFT | TAT_TITLE, _("Within Groups"));
304 tab_text (t, 1, i * 3 + 3, TAB_LEFT | TAT_TITLE, _("Total"));
307 tab_hline (t, TAL_1, 0, n_cols - 1, i * 3 + 1);
310 struct group_proc *gp = group_proc_get (vars[i]);
311 const double sst = totals->ssq - pow2 (totals->sum) / totals->n;
312 const double df1 = gp->n_groups - 1;
313 const double df2 = totals->n - gp->n_groups;
314 const double msa = ssa / df1;
316 gp->mse = (sst - ssa) / df2;
319 /* Sums of Squares */
320 tab_float (t, 2, i * 3 + 1, 0, ssa, 10, 2);
321 tab_float (t, 2, i * 3 + 3, 0, sst, 10, 2);
322 tab_float (t, 2, i * 3 + 2, 0, sst - ssa, 10, 2);
325 /* Degrees of freedom */
326 tab_float (t, 3, i * 3 + 1, 0, df1, 4, 0);
327 tab_float (t, 3, i * 3 + 2, 0, df2, 4, 0);
328 tab_float (t, 3, i * 3 + 3, 0, totals->n - 1, 4, 0);
331 tab_float (t, 4, i * 3 + 1, TAB_RIGHT, msa, 8, 3);
332 tab_float (t, 4, i * 3 + 2, TAB_RIGHT, gp->mse, 8, 3);
335 const double F = msa/gp->mse;
338 tab_float (t, 5, i * 3 + 1, 0, F, 8, 3);
340 /* The significance */
341 tab_float (t, 6, i * 3 + 1, 0, gsl_cdf_fdist_Q (F, df1, df2), 8, 3);
347 tab_title (t, _("ANOVA"));
352 /* Show the descriptives table */
354 show_descriptives (void)
361 const double confidence = 0.95;
362 const double q = (1.0 - confidence) / 2.0;
367 for ( v = 0; v < n_vars; ++v )
368 n_rows += group_proc_get (vars[v])->n_groups + 1;
370 t = tab_create (n_cols, n_rows, 0);
371 tab_headers (t, 2, 0, 2, 0);
372 tab_dim (t, tab_natural_dimensions);
375 /* Put a frame around the entire box, and vertical lines inside */
380 n_cols - 1, n_rows - 1);
382 /* Underline headers */
383 tab_hline (t, TAL_2, 0, n_cols - 1, 2);
384 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
386 tab_text (t, 2, 1, TAB_CENTER | TAT_TITLE, _("N"));
387 tab_text (t, 3, 1, TAB_CENTER | TAT_TITLE, _("Mean"));
388 tab_text (t, 4, 1, TAB_CENTER | TAT_TITLE, _("Std. Deviation"));
389 tab_text (t, 5, 1, TAB_CENTER | TAT_TITLE, _("Std. Error"));
392 tab_vline (t, TAL_0, 7, 0, 0);
393 tab_hline (t, TAL_1, 6, 7, 1);
394 tab_joint_text (t, 6, 0, 7, 0, TAB_CENTER | TAT_TITLE | TAT_PRINTF,
395 _("%g%% Confidence Interval for Mean"), confidence*100.0);
397 tab_text (t, 6, 1, TAB_CENTER | TAT_TITLE, _("Lower Bound"));
398 tab_text (t, 7, 1, TAB_CENTER | TAT_TITLE, _("Upper Bound"));
400 tab_text (t, 8, 1, TAB_CENTER | TAT_TITLE, _("Minimum"));
401 tab_text (t, 9, 1, TAB_CENTER | TAT_TITLE, _("Maximum"));
404 tab_title (t, _("Descriptives"));
408 for (v = 0; v < n_vars; ++v)
413 struct group_proc *gp = group_proc_get (vars[v]);
415 struct group_statistics *gs;
416 struct group_statistics *totals = &gp->ugs;
418 const char *s = var_to_string (vars[v]);
420 struct group_statistics *const *gs_array =
421 (struct group_statistics *const *) hsh_sort (gp->group_hash);
424 tab_text (t, 0, row, TAB_LEFT | TAT_TITLE, s);
426 tab_hline (t, TAL_1, 0, n_cols - 1, row);
428 for (count = 0; count < hsh_count (gp->group_hash); ++count)
431 ds_init_empty (&vstr);
432 gs = gs_array[count];
434 var_append_value_name (indep_var, &gs->id, &vstr);
436 tab_text (t, 1, row + count,
437 TAB_LEFT | TAT_TITLE,
442 /* Now fill in the numbers ... */
444 tab_float (t, 2, row + count, 0, gs->n, 8, 0);
446 tab_float (t, 3, row + count, 0, gs->mean, 8, 2);
448 tab_float (t, 4, row + count, 0, gs->std_dev, 8, 2);
450 std_error = gs->std_dev/sqrt (gs->n);
451 tab_float (t, 5, row + count, 0,
454 /* Now the confidence interval */
456 T = gsl_cdf_tdist_Qinv (q, gs->n - 1);
458 tab_float (t, 6, row + count, 0,
459 gs->mean - T * std_error, 8, 2);
461 tab_float (t, 7, row + count, 0,
462 gs->mean + T * std_error, 8, 2);
465 tab_float (t, 8, row + count, 0, gs->minimum, 8, 2);
466 tab_float (t, 9, row + count, 0, gs->maximum, 8, 2);
469 tab_text (t, 1, row + count,
470 TAB_LEFT | TAT_TITLE, _("Total"));
472 tab_float (t, 2, row + count, 0, totals->n, 8, 0);
474 tab_float (t, 3, row + count, 0, totals->mean, 8, 2);
476 tab_float (t, 4, row + count, 0, totals->std_dev, 8, 2);
478 std_error = totals->std_dev/sqrt (totals->n);
480 tab_float (t, 5, row + count, 0, std_error, 8, 2);
482 /* Now the confidence interval */
484 T = gsl_cdf_tdist_Qinv (q, totals->n - 1);
486 tab_float (t, 6, row + count, 0,
487 totals->mean - T * std_error, 8, 2);
489 tab_float (t, 7, row + count, 0,
490 totals->mean + T * std_error, 8, 2);
493 tab_float (t, 8, row + count, 0, totals->minimum, 8, 2);
494 tab_float (t, 9, row + count, 0, totals->maximum, 8, 2);
496 row += gp->n_groups + 1;
502 /* Show the homogeneity table */
504 show_homogeneity (void)
508 size_t n_rows = n_vars + 1;
513 t = tab_create (n_cols, n_rows, 0);
514 tab_headers (t, 1, 0, 1, 0);
515 tab_dim (t, tab_natural_dimensions);
517 /* Put a frame around the entire box, and vertical lines inside */
522 n_cols - 1, n_rows - 1);
525 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
526 tab_vline (t, TAL_2, 1, 0, n_rows - 1);
529 tab_text (t, 1, 0, TAB_CENTER | TAT_TITLE, _("Levene Statistic"));
530 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("df1"));
531 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("df2"));
532 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Significance"));
534 tab_title (t, _("Test of Homogeneity of Variances"));
536 for (v = 0; v < n_vars; ++v)
539 const struct variable *var = vars[v];
540 const struct group_proc *gp = group_proc_get (vars[v]);
541 const char *s = var_to_string (var);
542 const struct group_statistics *totals = &gp->ugs;
544 const double df1 = gp->n_groups - 1;
545 const double df2 = totals->n - gp->n_groups;
547 tab_text (t, 0, v + 1, TAB_LEFT | TAT_TITLE, s);
550 tab_float (t, 1, v + 1, TAB_RIGHT, F, 8, 3);
551 tab_float (t, 2, v + 1, TAB_RIGHT, df1, 8, 0);
552 tab_float (t, 3, v + 1, TAB_RIGHT, df2, 8, 0);
554 /* Now the significance */
555 tab_float (t, 4, v + 1, TAB_RIGHT, gsl_cdf_fdist_Q (F, df1, df2), 8, 3);
562 /* Show the contrast coefficients table */
564 show_contrast_coeffs (short *bad_contrast)
566 int n_cols = 2 + ostensible_number_of_groups;
567 int n_rows = 2 + cmd.sbc_contrast;
568 union value *group_value;
570 void *const *group_values;
574 t = tab_create (n_cols, n_rows, 0);
575 tab_headers (t, 2, 0, 2, 0);
576 tab_dim (t, tab_natural_dimensions);
578 /* Put a frame around the entire box, and vertical lines inside */
583 n_cols - 1, n_rows - 1);
597 tab_hline (t, TAL_1, 2, n_cols - 1, 1);
598 tab_hline (t, TAL_2, 0, n_cols - 1, 2);
600 tab_vline (t, TAL_2, 2, 0, n_rows - 1);
602 tab_title (t, _("Contrast Coefficients"));
604 tab_text (t, 0, 2, TAB_LEFT | TAT_TITLE, _("Contrast"));
607 tab_joint_text (t, 2, 0, n_cols - 1, 0, TAB_CENTER | TAT_TITLE,
608 var_to_string (indep_var));
610 group_values = hsh_sort (global_group_hash);
612 count < hsh_count (global_group_hash);
617 group_value = group_values[count];
619 ds_init_empty (&vstr);
621 var_append_value_name (indep_var, group_value, &vstr);
623 tab_text (t, count + 2, 1, TAB_CENTER | TAT_TITLE,
629 for (i = 0; i < cmd.sbc_contrast; ++i )
631 tab_text (t, 1, i + 2, TAB_CENTER | TAT_PRINTF, "%d", i + 1);
633 if ( bad_contrast[i] )
634 tab_text (t, count + 2, i + 2, TAB_RIGHT, "?" );
636 tab_text (t, count + 2, i + 2, TAB_RIGHT | TAT_PRINTF, "%g",
637 subc_list_double_at (&cmd.dl_contrast[i], count)
646 /* Show the results of the contrast tests */
648 show_contrast_tests (short *bad_contrast)
652 size_t n_rows = 1 + n_vars * 2 * cmd.sbc_contrast;
656 t = tab_create (n_cols, n_rows, 0);
657 tab_headers (t, 3, 0, 1, 0);
658 tab_dim (t, tab_natural_dimensions);
660 /* Put a frame around the entire box, and vertical lines inside */
665 n_cols - 1, n_rows - 1);
673 tab_hline (t, TAL_2, 0, n_cols - 1, 1);
674 tab_vline (t, TAL_2, 3, 0, n_rows - 1);
677 tab_title (t, _("Contrast Tests"));
679 tab_text (t, 2, 0, TAB_CENTER | TAT_TITLE, _("Contrast"));
680 tab_text (t, 3, 0, TAB_CENTER | TAT_TITLE, _("Value of Contrast"));
681 tab_text (t, 4, 0, TAB_CENTER | TAT_TITLE, _("Std. Error"));
682 tab_text (t, 5, 0, TAB_CENTER | TAT_TITLE, _("t"));
683 tab_text (t, 6, 0, TAB_CENTER | TAT_TITLE, _("df"));
684 tab_text (t, 7, 0, TAB_CENTER | TAT_TITLE, _("Sig. (2-tailed)"));
686 for (v = 0; v < n_vars; ++v)
689 int lines_per_variable = 2 * cmd.sbc_contrast;
692 tab_text (t, 0, (v * lines_per_variable) + 1, TAB_LEFT | TAT_TITLE,
693 var_to_string (vars[v]));
695 for (i = 0; i < cmd.sbc_contrast; ++i)
698 double contrast_value = 0.0;
699 double coef_msq = 0.0;
700 struct group_proc *grp_data = group_proc_get (vars[v]);
701 struct hsh_table *group_hash = grp_data->group_hash;
703 void *const *group_stat_array;
706 double std_error_contrast;
708 double sec_vneq = 0.0;
711 /* Note: The calculation of the degrees of freedom in the
712 "variances not equal" case is painfull!!
713 The following formula may help to understand it:
714 \frac{\left (\sum_{i=1}^k{c_i^2\frac{s_i^2}{n_i}}\right)^2}
717 \frac{\left (c_i^2\frac{s_i^2}{n_i}\right)^2} {n_i-1}
722 double df_denominator = 0.0;
723 double df_numerator = 0.0;
726 tab_text (t, 1, (v * lines_per_variable) + i + 1,
727 TAB_LEFT | TAT_TITLE,
728 _("Assume equal variances"));
730 tab_text (t, 1, (v * lines_per_variable) + i + 1 + cmd.sbc_contrast,
731 TAB_LEFT | TAT_TITLE,
732 _("Does not assume equal"));
735 tab_text (t, 2, (v * lines_per_variable) + i + 1,
736 TAB_CENTER | TAT_TITLE | TAT_PRINTF, "%d", i + 1);
739 tab_text (t, 2, (v * lines_per_variable) + i + 1 + cmd.sbc_contrast,
740 TAB_CENTER | TAT_TITLE | TAT_PRINTF, "%d", i + 1);
743 if ( bad_contrast[i])
746 group_stat_array = hsh_sort (group_hash);
748 for (ci = 0; ci < hsh_count (group_hash); ++ci)
750 const double coef = subc_list_double_at (&cmd.dl_contrast[i], ci);
751 struct group_statistics *gs = group_stat_array[ci];
753 const double winv = pow2 (gs->std_dev) / gs->n;
755 contrast_value += coef * gs->mean;
757 coef_msq += (coef * coef) / gs->n;
759 sec_vneq += (coef * coef) * pow2 (gs->std_dev) /gs->n;
761 df_numerator += (coef * coef) * winv;
762 df_denominator += pow2((coef * coef) * winv) / (gs->n - 1);
764 sec_vneq = sqrt (sec_vneq);
766 df_numerator = pow2(df_numerator);
768 tab_float (t, 3, (v * lines_per_variable) + i + 1,
769 TAB_RIGHT, contrast_value, 8, 2);
771 tab_float (t, 3, (v * lines_per_variable) + i + 1 +
773 TAB_RIGHT, contrast_value, 8, 2);
775 std_error_contrast = sqrt (grp_data->mse * coef_msq);
778 tab_float (t, 4, (v * lines_per_variable) + i + 1,
779 TAB_RIGHT, std_error_contrast,
782 T = fabs (contrast_value / std_error_contrast);
786 tab_float (t, 5, (v * lines_per_variable) + i + 1,
790 df = grp_data->ugs.n - grp_data->n_groups;
792 /* Degrees of Freedom */
793 tab_float (t, 6, (v * lines_per_variable) + i + 1,
798 /* Significance TWO TAILED !!*/
799 tab_float (t, 7, (v * lines_per_variable) + i + 1,
800 TAB_RIGHT, 2 * gsl_cdf_tdist_Q (T, df),
804 /* Now for the Variances NOT Equal case */
808 (v * lines_per_variable) + i + 1 + cmd.sbc_contrast,
813 T = contrast_value / sec_vneq;
815 (v * lines_per_variable) + i + 1 + cmd.sbc_contrast,
820 df = df_numerator / df_denominator;
823 (v * lines_per_variable) + i + 1 + cmd.sbc_contrast,
827 /* The Significance */
829 tab_float (t, 7, (v * lines_per_variable) + i + 1 + cmd.sbc_contrast,
830 TAB_RIGHT, 2 * gsl_cdf_tdist_Q (T, df),
837 tab_hline (t, TAL_1, 0, n_cols - 1, (v * lines_per_variable) + 1);
845 /* ONEWAY ANOVA Calculations */
847 static void postcalc (struct cmd_oneway *cmd UNUSED);
849 static void precalc (struct cmd_oneway *cmd UNUSED);
853 /* Pre calculations */
855 precalc (struct cmd_oneway *cmd UNUSED)
859 for (i = 0; i < n_vars; ++i)
861 struct group_proc *gp = group_proc_get (vars[i]);
862 struct group_statistics *totals = &gp->ugs;
864 /* Create a hash for each of the dependent variables.
865 The hash contains a group_statistics structure,
866 and is keyed by value of the independent variable */
868 gp->group_hash = hsh_create (4, compare_group, hash_group,
869 (hsh_free_func *) free_group,
875 totals->sum_diff = 0;
876 totals->maximum = -DBL_MAX;
877 totals->minimum = DBL_MAX;
882 free_value (void *value_, const void *aux UNUSED)
884 union value *value = value_;
889 run_oneway (struct cmd_oneway *cmd,
890 struct casereader *input,
891 const struct dataset *ds)
894 struct dictionary *dict = dataset_dict (ds);
895 enum mv_class exclude;
896 struct casereader *reader;
899 c = casereader_peek (input, 0);
902 casereader_destroy (input);
905 output_split_file_values (ds, c);
908 taint = taint_clone (casereader_get_taint (input));
910 global_group_hash = hsh_create (4,
911 compare_values_short,
918 exclude = cmd->incl != ONEWAY_INCLUDE ? MV_ANY : MV_SYSTEM;
919 input = casereader_create_filter_missing (input, &indep_var, 1,
920 exclude, NULL, NULL);
921 if (cmd->miss == ONEWAY_LISTWISE)
922 input = casereader_create_filter_missing (input, vars, n_vars,
923 exclude, NULL, NULL);
924 input = casereader_create_filter_weight (input, dict, NULL, NULL);
926 reader = casereader_clone (input);
927 for (; (c = casereader_read (reader)) != NULL; case_unref (c))
931 const double weight = dict_get_case_weight (dict, c, NULL);
933 const union value *indep_val = case_data (c, indep_var);
934 void **p = hsh_probe (global_group_hash, indep_val);
936 *p = value_dup (indep_val, var_get_width (indep_var));
938 for (i = 0; i < n_vars; ++i)
940 const struct variable *v = vars[i];
942 const union value *val = case_data (c, v);
944 struct group_proc *gp = group_proc_get (vars[i]);
945 struct hsh_table *group_hash = gp->group_hash;
947 struct group_statistics *gs;
949 gs = hsh_find (group_hash, indep_val );
953 gs = xmalloc (sizeof *gs);
959 gs->minimum = DBL_MAX;
960 gs->maximum = -DBL_MAX;
962 hsh_insert ( group_hash, gs );
965 if (!var_is_value_missing (v, val, exclude))
967 struct group_statistics *totals = &gp->ugs;
970 totals->sum += weight * val->f;
971 totals->ssq += weight * pow2 (val->f);
973 if ( val->f * weight < totals->minimum )
974 totals->minimum = val->f * weight;
976 if ( val->f * weight > totals->maximum )
977 totals->maximum = val->f * weight;
980 gs->sum += weight * val->f;
981 gs->ssq += weight * pow2 (val->f);
983 if ( val->f * weight < gs->minimum )
984 gs->minimum = val->f * weight;
986 if ( val->f * weight > gs->maximum )
987 gs->maximum = val->f * weight;
990 gp->n_groups = hsh_count ( group_hash );
994 casereader_destroy (reader);
999 if ( stat_tables & STAT_HOMO )
1000 levene (dict, casereader_clone (input), indep_var, n_vars, vars, exclude);
1002 casereader_destroy (input);
1004 ostensible_number_of_groups = hsh_count (global_group_hash);
1006 if (!taint_has_tainted_successor (taint))
1008 taint_destroy (taint);
1012 /* Post calculations for the ONEWAY command */
1014 postcalc ( struct cmd_oneway *cmd UNUSED )
1018 for (i = 0; i < n_vars; ++i)
1020 struct group_proc *gp = group_proc_get (vars[i]);
1021 struct hsh_table *group_hash = gp->group_hash;
1022 struct group_statistics *totals = &gp->ugs;
1024 struct hsh_iterator g;
1025 struct group_statistics *gs;
1027 for (gs = hsh_first (group_hash, &g);
1029 gs = hsh_next (group_hash, &g))
1031 gs->mean = gs->sum / gs->n;
1032 gs->s_std_dev = sqrt (gs->ssq / gs->n - pow2 (gs->mean));
1034 gs->std_dev = sqrt (
1035 gs->n / (gs->n - 1) *
1036 ( gs->ssq / gs->n - pow2 (gs->mean))
1039 gs->se_mean = gs->std_dev / sqrt (gs->n);
1040 gs->mean_diff = gs->sum_diff / gs->n;
1043 totals->mean = totals->sum / totals->n;
1044 totals->std_dev = sqrt (
1045 totals->n / (totals->n - 1) *
1046 (totals->ssq / totals->n - pow2 (totals->mean))
1049 totals->se_mean = totals->std_dev / sqrt (totals->n);