diff --git a/tests/unit/column_pairs/statistical/test_referential_integrity.py b/tests/unit/column_pairs/statistical/test_referential_integrity.py
index 6214542c..a68cc77d 100644
--- a/tests/unit/column_pairs/statistical/test_referential_integrity.py
+++ b/tests/unit/column_pairs/statistical/test_referential_integrity.py
@@ -68,32 +68,56 @@ def test_compute(self, compute_breakdown_mock):
         assert result == 0.6
 
     def test_compute_with_nan_foreign_keys_real_data(self):
-        """Test the ``compute`` method with NaN foreign keys inside the real data."""
+        """Test the ``compute`` method with NaN foreign keys inside the real data.
+
+        Here, the score should be 1.0, whether or not the synthetic data have NaN values
+        values, as the real data have null foreign keys.
+        """
         # Setup
         parent_keys = pd.Series(['a', 'b', 'c'])
-        foreign_keys = pd.Series(['a', 'a', 'b', 'c', np.nan])
+        real_fk = pd.Series(['a', 'a', 'b', 'c', np.nan])
+        synthetic_fk = pd.Series(['a', 'a', 'b', 'c', 'a'])
+        synthetic_fk_with_nan = pd.Series(['a', 'a', 'b', 'c', np.nan])
         metric = ReferentialIntegrity()
 
         # Run
         result = metric.compute(
-            real_data=(parent_keys, foreign_keys), synthetic_data=(parent_keys, foreign_keys)
+            real_data=(parent_keys, real_fk), synthetic_data=(parent_keys, synthetic_fk)
+        )
+        result_with_nan = metric.compute(
+            real_data=(parent_keys, real_fk), synthetic_data=(parent_keys, synthetic_fk_with_nan)
         )
 
         # Assert
         assert result == 1.0
+        assert result_with_nan == 1.0
 
     def test_compute_with_nan_foreign_keys_only_synthetic_data(self):
-        """Test the ``compute`` method with NaN foreign keys inside the synthetic data."""
+        """Test the ``compute`` method with NaN foreign keys inside the synthetic data.
+
+        Here, the real data have no null foreign keys, so the score should decrease as
+        the number of NaN values in the synthetic data increases.
+        """
         # Setup
         parent_keys = pd.Series(['a', 'b', 'c'])
-        foreign_keys = pd.Series(['a', 'a', 'b', 'c', 'a'])
-        synth_foreign_keys = pd.Series(['a', 'a', 'b', 'c', np.nan])
+        real_fk = pd.Series(['a', 'a', 'b', 'c', 'a'])
+        synth_fk_0_nan = pd.Series(['a', 'a', 'b', 'c'])
+        synth_fk_1_nan = pd.Series(['a', 'a', 'b', 'c', np.nan])
+        synth_fk_2_nan = pd.Series(['a', 'a', 'b', 'c', np.nan, np.nan])
         metric = ReferentialIntegrity()
 
         # Run
-        result = metric.compute(
-            real_data=(parent_keys, foreign_keys), synthetic_data=(parent_keys, synth_foreign_keys)
+        result_0 = metric.compute(
+            real_data=(parent_keys, real_fk), synthetic_data=(parent_keys, synth_fk_0_nan)
+        )
+        result_1 = metric.compute(
+            real_data=(parent_keys, real_fk), synthetic_data=(parent_keys, synth_fk_1_nan)
+        )
+        result_2 = metric.compute(
+            real_data=(parent_keys, real_fk), synthetic_data=(parent_keys, synth_fk_2_nan)
         )
 
         # Assert
-        assert result == 0.8
+        assert result_0 == 1.0
+        assert result_1 == 0.8
+        assert result_2 == 2 / 3