REGR: Groupby first/last/nth treats None as an observation

pandas/_libs/groupby.pyx

@@ -928,9 +928,7 @@ def group_last(rank_t[:, :] out,
             for j in range(K):
                 val = values[i, j]
 
-                # None should not be treated like other NA-like
-                # so that it won't be converted to nan
-                if not checknull(val) or val is None:
+                if not checknull(val):
                     # NB: use _treat_as_na here once
                     #  conditional-nogil is available.
                     nobs[lab, j] += 1
@@ -939,7 +937,7 @@ def group_last(rank_t[:, :] out,
         for i in range(ncounts):
             for j in range(K):
                 if nobs[i, j] < min_count:
-                    out[i, j] = NAN
+                    out[i, j] = None
                 else:
                     out[i, j] = resx[i, j]
     else:
@@ -1023,9 +1021,7 @@ def group_nth(rank_t[:, :] out,
             for j in range(K):
                 val = values[i, j]
 
-                # None should not be treated like other NA-like
-                # so that it won't be converted to nan
-                if not checknull(val) or val is None:
+                if not checknull(val):
                     # NB: use _treat_as_na here once
                     #  conditional-nogil is available.
                     nobs[lab, j] += 1
@@ -1035,7 +1031,7 @@ def group_nth(rank_t[:, :] out,
         for i in range(ncounts):
             for j in range(K):
                 if nobs[i, j] < min_count:
-                    out[i, j] = NAN
+                    out[i, j] = None
                 else:
                     out[i, j] = resx[i, j]
 

pandas/tests/groupby/test_nth.py

@@ -91,14 +91,27 @@ def test_nth_with_na_object(index, nulls_fixture):
 
 
 @pytest.mark.parametrize("method", ["first", "last"])
-def test_first_last_with_None(method):
-    # https://github.com/pandas-dev/pandas/issues/32800
-    # None should be preserved as object dtype
-    df = DataFrame.from_dict({"id": ["a"], "value": [None]})
-    groups = df.groupby("id", as_index=False)
-    result = getattr(groups, method)()
-
-    tm.assert_frame_equal(result, df)
+@pytest.mark.parametrize(
+    "df, expected",
+    [
+        (
+            DataFrame({"id": ["a"], "value": [None]}),
+            DataFrame({"value": [None]}, index=Index(["a"], name="id")),
+        ),
+        (
+            DataFrame({"id": "a", "value": [None, "foo", np.nan]}),
+            DataFrame({"value": ["foo"]}, index=Index(["a"], name="id")),
+        ),
+        (
+            DataFrame({"id": "a", "value": [np.nan]}, dtype=object),
+            DataFrame({"value": [None]}, index=Index(["a"], name="id")),
+        ),
+    ],
+)
+def test_first_last_with_none(method, df, expected):
+    # GH 32800, 38286
+    result = getattr(df.groupby("id"), method)()
+    tm.assert_frame_equal(result, expected)
 
 
 def test_first_last_nth_dtypes(df_mixed_floats):