BUG: fix groupby.quantile inconsistency when interpolation='nearest'

doc/source/whatsnew/v2.3.0.rst

@@ -188,7 +188,7 @@ Plotting
 
 Groupby/resample/rolling
 ^^^^^^^^^^^^^^^^^^^^^^^^
--
+- Bug in :meth:`.DataFrameGroupBy.quantile` when ``interpolation="nearest"`` is inconsistent with :meth:`DataFrame.quantile` (:issue:`47942`)
 -
 
 Reshaping

pandas/_libs/groupby.pyx

@@ -1286,7 +1286,9 @@ def group_quantile(
                     elif interp == INTERPOLATION_MIDPOINT:
                         out[i, k] = (val + next_val) / 2.0
                     elif interp == INTERPOLATION_NEAREST:
-                        if frac > .5 or (frac == .5 and q_val > .5):  # Always OK?
+                        if frac > .5 or (frac == .5 and idx % 2 == 1):
+                            # If quantile lies in the middle of two indexes,
+                            # take the even index, as np.quantile.
                             out[i, k] = next_val
                         else:
                             out[i, k] = val

pandas/tests/groupby/methods/test_quantile.py

@@ -38,19 +38,7 @@
     ],
 )
 @pytest.mark.parametrize("q", [0, 0.25, 0.5, 0.75, 1])
-def test_quantile(interpolation, a_vals, b_vals, q, request):
-    if (
-        interpolation == "nearest"
-        and q == 0.5
-        and isinstance(b_vals, list)
-        and b_vals == [4, 3, 2, 1]
-    ):
-        request.applymarker(
-            pytest.mark.xfail(
-                reason="Unclear numpy expectation for nearest "
-                "result with equidistant data"
-            )
-        )
+def test_quantile(interpolation, a_vals, b_vals, q):
     all_vals = pd.concat([pd.Series(a_vals), pd.Series(b_vals)])
 
     a_expected = pd.Series(a_vals).quantile(q, interpolation=interpolation)