Implemented the McNemar test. Closes bug #33242
[pspp-builds.git] / src / language / stats / npar.c
index 3f2bae298ef45669cd6331f896c160d0988ff465..ce0c3d6d06419fecdad91ca79666de840dcbe42a 100644 (file)
@@ -1,5 +1,5 @@
 /* PSPP - a program for statistical analysis. -*-c-*-
-   Copyright (C) 2006, 2008, 2009, 2010 Free Software Foundation, Inc.
+   Copyright (C) 2006, 2008, 2009, 2010, 2011 Free Software Foundation, Inc.
 
    This program is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
 
 #include <config.h>
 
-#include <language/stats/npar.h>
-#include "npar-summary.h"
+#include "language/stats/npar.h"
 
 #include <stdlib.h>
 #include <math.h>
 
-#include "xalloc.h"
-
-#include <data/case.h>
-#include <data/casegrouper.h>
-#include <data/casereader.h>
-#include <data/dictionary.h>
-#include <data/procedure.h>
-#include <data/settings.h>
-#include <data/variable.h>
-#include <libpspp/array.h>
-#include <libpspp/assertion.h>
-#include <libpspp/cast.h>
-#include <libpspp/hmapx.h>
-#include <libpspp/hash-functions.h>
-#include <libpspp/message.h>
-#include <libpspp/pool.h>
-#include <libpspp/str.h>
-#include <libpspp/taint.h>
-#include <language/command.h>
-#include <language/lexer/lexer.h>
-#include <language/lexer/variable-parser.h>
-#include <language/lexer/value-parser.h>
-#include <language/stats/binomial.h>
-#include <language/stats/chisquare.h>
-#include <language/stats/cochran.h>
-#include <language/stats/runs.h>
-#include <language/stats/friedman.h>
-#include <language/stats/kruskal-wallis.h>
-#include <language/stats/mann-whitney.h>
-#include <language/stats/wilcoxon.h>
-#include <language/stats/sign.h>
-#include <math/moments.h>
+#include "data/case.h"
+#include "data/casegrouper.h"
+#include "data/casereader.h"
+#include "data/dataset.h"
+#include "data/dictionary.h"
+#include "data/settings.h"
+#include "data/variable.h"
+#include "language/command.h"
+#include "language/lexer/lexer.h"
+#include "language/lexer/value-parser.h"
+#include "language/lexer/variable-parser.h"
+#include "language/stats/binomial.h"
+#include "language/stats/chisquare.h"
+#include "language/stats/cochran.h"
+#include "language/stats/friedman.h"
+#include "language/stats/kruskal-wallis.h"
+#include "language/stats/mann-whitney.h"
+#include "language/stats/mcnemar.h"
+#include "language/stats/npar-summary.h"
+#include "language/stats/runs.h"
+#include "language/stats/sign.h"
+#include "language/stats/wilcoxon.h"
+#include "libpspp/array.h"
+#include "libpspp/assertion.h"
+#include "libpspp/cast.h"
+#include "libpspp/hash-functions.h"
+#include "libpspp/hmapx.h"
+#include "libpspp/message.h"
+#include "libpspp/pool.h"
+#include "libpspp/str.h"
+#include "libpspp/taint.h"
+#include "math/moments.h"
+
+#include "gl/xalloc.h"
 
 #include "gettext.h"
 #define _(msgid) gettext (msgid)
@@ -89,6 +90,7 @@ struct cmd_npar_tests
     int kendall;
     int kruskal_wallis;
     int mann_whitney;
+    int mcnemar;
     int missing;
     int method;
     int statistics;
@@ -132,6 +134,7 @@ static int npar_wilcoxon (struct lexer *, struct dataset *, struct npar_specs *)
 static int npar_sign (struct lexer *, struct dataset *, struct npar_specs *);
 static int npar_kruskal_wallis (struct lexer *, struct dataset *, struct npar_specs *);
 static int npar_mann_whitney (struct lexer *, struct dataset *, struct npar_specs *);
+static int npar_mcnemar (struct lexer *, struct dataset *, struct npar_specs *);
 static int npar_method (struct lexer *, struct npar_specs *);
 
 /* Command parsing functions. */
@@ -148,6 +151,7 @@ parse_npar_tests (struct lexer *lexer, struct dataset *ds, struct cmd_npar_tests
   npt->friedman = 0;
   npt->kruskal_wallis = 0;
   npt->mann_whitney = 0;
+  npt->mcnemar = 0;
   npt->runs = 0;
   npt->sign = 0;
   npt->wilcoxon = 0;
@@ -258,8 +262,8 @@ parse_npar_tests (struct lexer *lexer, struct dataset *ds, struct cmd_npar_tests
               NOT_REACHED ();
             }
         }
-      else if (lex_match_hyphenated_word (lexer, "K-W") ||
-              lex_match_hyphenated_word (lexer, "KRUSKAL-WALLIS"))
+      else if (lex_match_phrase (lexer, "K-W") ||
+              lex_match_phrase (lexer, "KRUSKAL-WALLIS"))
         {
           lex_match (lexer, T_EQUALS);
           npt->kruskal_wallis++;
@@ -276,8 +280,25 @@ parse_npar_tests (struct lexer *lexer, struct dataset *ds, struct cmd_npar_tests
               NOT_REACHED ();
             }
         }
-      else if (lex_match_hyphenated_word (lexer, "M-W") ||
-              lex_match_hyphenated_word (lexer, "MANN-WHITNEY"))
+      else if (lex_match_phrase (lexer, "MCNEMAR"))
+        {
+          lex_match (lexer, T_EQUALS);
+          npt->mcnemar++;
+          switch (npar_mcnemar (lexer, ds, nps))
+            {
+            case 0:
+              goto lossage;
+            case 1:
+              break;
+            case 2:
+              lex_error (lexer, NULL);
+              goto lossage;
+            default:
+              NOT_REACHED ();
+            }
+        }
+      else if (lex_match_phrase (lexer, "M-W") ||
+              lex_match_phrase (lexer, "MANN-WHITNEY"))
         {
           lex_match (lexer, T_EQUALS);
           npt->mann_whitney++;
@@ -759,44 +780,39 @@ npar_chisquare (struct lexer *lexer, struct dataset *ds,
 
   cstp->n_expected = 0;
   cstp->expected = NULL;
-  if ( lex_match (lexer, T_SLASH) )
+  if (lex_match_phrase (lexer, "/EXPECTED"))
     {
-      if ( lex_match_id (lexer, "EXPECTED") )
-       {
-         lex_force_match (lexer, T_EQUALS);
-         if ( ! lex_match_id (lexer, "EQUAL") )
-           {
-             double f;
-             int n;
-             while ( lex_is_number (lexer) )
-               {
-                 int i;
-                 n = 1;
-                 f = lex_number (lexer);
-                 lex_get (lexer);
-                 if ( lex_match (lexer, T_ASTERISK))
-                   {
-                     n = f;
-                     f = lex_number (lexer);
-                     lex_get (lexer);
-                   }
-                 lex_match (lexer, T_COMMA);
-
-                 cstp->n_expected += n;
-                 cstp->expected = pool_realloc (specs->pool,
-                                                cstp->expected,
-                                                sizeof (double) *
-                                                cstp->n_expected);
-                 for ( i = cstp->n_expected - n ;
-                       i < cstp->n_expected;
-                       ++i )
-                   cstp->expected[i] = f;
+      lex_force_match (lexer, T_EQUALS);
+      if ( ! lex_match_id (lexer, "EQUAL") )
+        {
+          double f;
+          int n;
+          while ( lex_is_number (lexer) )
+            {
+              int i;
+              n = 1;
+              f = lex_number (lexer);
+              lex_get (lexer);
+              if ( lex_match (lexer, T_ASTERISK))
+                {
+                  n = f;
+                  f = lex_number (lexer);
+                  lex_get (lexer);
+                }
+              lex_match (lexer, T_COMMA);
 
-               }
-           }
-       }
-      else
-        retval = 3;
+              cstp->n_expected += n;
+              cstp->expected = pool_realloc (specs->pool,
+                                             cstp->expected,
+                                             sizeof (double) *
+                                             cstp->n_expected);
+              for ( i = cstp->n_expected - n ;
+                    i < cstp->n_expected;
+                    ++i )
+                cstp->expected[i] = f;
+
+            }
+        }
     }
 
   if ( cstp->ranged && cstp->n_expected > 0 &&
@@ -828,7 +844,7 @@ npar_binomial (struct lexer *lexer, struct dataset *ds,
   struct binomial_test *btp = pool_alloc (specs->pool, sizeof (*btp));
   struct one_sample_test *tp = &btp->parent;
   struct npar_test *nt = &tp->parent;
-  bool equals;
+  bool equals = false;
 
   nt->execute = binomial_execute;
   nt->insert_variables = one_sample_insert_variables;
@@ -893,14 +909,6 @@ npar_binomial (struct lexer *lexer, struct dataset *ds,
 }
 
 
-static bool
-parse_two_sample_related_test (struct lexer *lexer,
-                                   const struct dictionary *dict,
-                                   struct two_sample_test *test_parameters,
-                                   struct pool *pool
-                                   );
-
-
 static bool
 parse_two_sample_related_test (struct lexer *lexer,
                               const struct dictionary *dict,
@@ -970,8 +978,8 @@ parse_two_sample_related_test (struct lexer *lexer,
          assert (n_vlist1 == n_vlist2);
          for ( i = 0 ; i < n_vlist1; ++i )
            {
-             test_parameters->pairs[n][1] = vlist1[i];
-             test_parameters->pairs[n][0] = vlist2[i];
+             test_parameters->pairs[n][0] = vlist1[i];
+             test_parameters->pairs[n][1] = vlist2[i];
              n++;
            }
        }
@@ -982,8 +990,8 @@ parse_two_sample_related_test (struct lexer *lexer,
            {
              for ( j = 0 ; j < n_vlist2; ++j )
                {
-                 test_parameters->pairs[n][1] = vlist1[i];
-                 test_parameters->pairs[n][0] = vlist2[j];
+                 test_parameters->pairs[n][0] = vlist1[i];
+                 test_parameters->pairs[n][1] = vlist2[j];
                  n++;
                }
            }
@@ -997,8 +1005,8 @@ parse_two_sample_related_test (struct lexer *lexer,
          for ( j = i + 1 ; j < n_vlist1; ++j )
            {
              assert ( n < test_parameters->n_pairs);
-             test_parameters->pairs[n][1] = vlist1[i];
-             test_parameters->pairs[n][0] = vlist1[j];
+             test_parameters->pairs[n][0] = vlist1[i];
+             test_parameters->pairs[n][1] = vlist1[j];
              n++;
            }
        }
@@ -1032,7 +1040,7 @@ parse_n_sample_related_test (struct lexer *lexer,
     return false;
 
   value_init (&nst->val1, var_get_width (nst->indep_var));
-  if ( ! parse_value (lexer, &nst->val1, var_get_width (nst->indep_var)))
+  if ( ! parse_value (lexer, &nst->val1, nst->indep_var))
     {
       value_destroy (&nst->val1, var_get_width (nst->indep_var));
       return false;
@@ -1041,7 +1049,7 @@ parse_n_sample_related_test (struct lexer *lexer,
   lex_match (lexer, T_COMMA);
 
   value_init (&nst->val2, var_get_width (nst->indep_var));
-  if ( ! parse_value (lexer, &nst->val2, var_get_width (nst->indep_var)))
+  if ( ! parse_value (lexer, &nst->val2, nst->indep_var))
     {
       value_destroy (&nst->val2, var_get_width (nst->indep_var));
       return false;
@@ -1058,8 +1066,6 @@ npar_wilcoxon (struct lexer *lexer,
               struct dataset *ds,
               struct npar_specs *specs )
 {
-
-
   struct two_sample_test *tp = pool_alloc (specs->pool, sizeof (*tp));
   struct npar_test *nt = &tp->parent;
   nt->execute = wilcoxon_execute;
@@ -1103,6 +1109,8 @@ npar_mann_whitney (struct lexer *lexer,
 }
 
 
+
+
 static int
 npar_sign (struct lexer *lexer, struct dataset *ds,
           struct npar_specs *specs)
@@ -1125,6 +1133,30 @@ npar_sign (struct lexer *lexer, struct dataset *ds,
   return 1;
 }
 
+
+static int
+npar_mcnemar (struct lexer *lexer, struct dataset *ds,
+          struct npar_specs *specs)
+{
+  struct two_sample_test *tp = pool_alloc (specs->pool, sizeof (*tp));
+  struct npar_test *nt = &tp->parent;
+
+  nt->execute = mcnemar_execute;
+
+  if (!parse_two_sample_related_test (lexer, dataset_dict (ds),
+                                     tp, specs->pool) )
+    return 0;
+
+  specs->n_tests++;
+  specs->test = pool_realloc (specs->pool,
+                             specs->test,
+                             sizeof (*specs->test) * specs->n_tests);
+  specs->test[specs->n_tests - 1] = nt;
+
+  return 1;
+}
+
+
 static int
 npar_kruskal_wallis (struct lexer *lexer, struct dataset *ds,
                      struct npar_specs *specs)