Fixed the trimmed mean calculation under various special cases.

[pspp-builds.git] / src / factor_stats.c

diff --git a/src/factor_stats.c b/src/factor_stats.c
index 2ba785a7d8e8c7044b2f42fa3ff2c94f96124d02..3a88ab44a4a234a8f64dec6d1d9674cc2ee79caf 100644 (file)
--- a/src/factor_stats.c
+++ b/src/factor_stats.c
@@ -144,20 +144,26 @@ metrics_postcalc(struct metrics *m)
   m->wvp = (struct weighted_value **) hsh_sort(m->ordered_data);
   m->n_data = hsh_count(m->ordered_data);
 
-  if ( m->n_data == 0 ) 
+  m->histogram = histogram_create(10, m->min, m->max);
+
+  for ( i = 0 ; i < m->n_data ; ++i ) 
     {
-      m->trimmed_mean = m->mean;
-      return;
+      struct weighted_value **wv = (m->wvp) ;
+      gsl_histogram_accumulate(m->histogram, wv[i]->v.f, wv[i]->w);
     }
 
 
   /* Trimmed mean calculation */
+  if ( m->n_data <= 1 ) 
+    {
+      m->trimmed_mean = m->mean;
+      return;
+    }
 
   tc = m->n * 0.05 ;
   k1 = -1;
   k2 = -1;
 
-
   for ( i = 0 ; i < m->n_data ; ++i ) 
     {
       cc += m->wvp[i]->w;
@@ -179,6 +185,12 @@ metrics_postcalc(struct metrics *m)
        k2 = i;
     }
 
+  /* Special case here */
+  if ( k1 + 1 == k2 ) 
+    {
+      m->trimmed_mean = m->wvp[k2]->v.f;
+      return;
+    }
 
   m->trimmed_mean = 0;
   for ( i = k1 + 2 ; i <= k2 - 1 ; ++i ) 
@@ -191,14 +203,6 @@ metrics_postcalc(struct metrics *m)
   m->trimmed_mean += (m->wvp[k1 + 1]->cc - tc) * m->wvp[k1 + 1]->v.f ;
   m->trimmed_mean /= 0.9 * m->n ;
 
-  m->histogram = histogram_create(10, m->min, m->max);
-
-  for ( i = 0 ; i < m->n_data ; ++i ) 
-    {
-      struct weighted_value **wv = (m->wvp) ;
-      gsl_histogram_accumulate(m->histogram, wv[i]->v.f, wv[i]->w);
-    }
-
 }