BUG: Create empty dataframe with string dtype fails

Author: kotamatsuoka

pandas/core/internals/construction.py

@@ -242,7 +242,10 @@ def init_dict(data, index, columns, dtype=None):
 
         # no obvious "empty" int column
         if missing.any() and not is_integer_dtype(dtype):
-            if dtype is None or np.issubdtype(dtype, np.flexible):
+            if is_dtype_equal(dtype, "string"):
+                # GH 33623
+                nan_dtype = dtype
+            elif dtype is None or np.issubdtype(dtype, np.flexible):
                 # GH#1783
                 nan_dtype = object
             else:

pandas/tests/frame/test_constructors.py

@@ -2679,3 +2679,7 @@ def test_construction_from_set_raises(self):
         msg = "Set type is unordered"
         with pytest.raises(TypeError, match=msg):
             pd.DataFrame({"a": {1, 2, 3}})
+
+    def test_construct_empty_dataframe_with_string_dtype(self):
+        # GH 33623
+        pd.DataFrame(columns=["a"], dtype="string")