BUG: to_string using wrong na_rep for ea dtype in multiindex

doc/source/whatsnew/v1.5.0.rst

@@ -958,6 +958,7 @@ I/O
 - Bug in :func:`read_sas` with certain types of compressed SAS7BDAT files (:issue:`35545`)
 - Bug in :func:`read_excel` not forward filling :class:`MultiIndex` when no names were given (:issue:`47487`)
 - Bug in :func:`read_sas` returned ``None`` rather than an empty DataFrame for SAS7BDAT files with zero rows (:issue:`18198`)
+- Bug in :meth:`DataFrame.to_string` using wrong missing value with extension arrays in :class:`MultiIndex` (:issue:`47986`)
 - Bug in :class:`StataWriter` where value labels were always written with default encoding (:issue:`46750`)
 - Bug in :class:`StataWriterUTF8` where some valid characters were removed from variable names (:issue:`47276`)
 - Bug in :meth:`DataFrame.to_excel` when writing an empty dataframe with :class:`MultiIndex` (:issue:`19543`)

pandas/core/indexes/multi.py

@@ -54,6 +54,7 @@
     ensure_int64,
     ensure_platform_int,
     is_categorical_dtype,
+    is_extension_array_dtype,
     is_hashable,
     is_integer,
     is_iterator,
@@ -1370,7 +1371,7 @@ def format(
 
         stringified_levels = []
         for lev, level_codes in zip(self.levels, self.codes):
-            na = na_rep if na_rep is not None else _get_na_rep(lev.dtype.type)
+            na = na_rep if na_rep is not None else _get_na_rep(lev.dtype)
 
             if len(lev) > 0:
 
@@ -3889,6 +3890,11 @@ def sparsify_labels(label_list, start: int = 0, sentinel=""):
 
 
 def _get_na_rep(dtype) -> str:
+    if is_extension_array_dtype(dtype):
+        return f"{dtype.na_value}"
+    else:
+        dtype = dtype.type
+
     return {np.datetime64: "NaT", np.timedelta64: "NaT"}.get(dtype, "NaN")
 
 

pandas/tests/frame/test_repr_info.py

@@ -9,6 +9,7 @@
 import pytest
 
 from pandas import (
+    NA,
     Categorical,
     DataFrame,
     MultiIndex,
@@ -342,6 +343,19 @@ def test_frame_to_string_with_periodindex(self):
         # it works!
         frame.to_string()
 
+    def test_to_string_ea_na_in_multiindex(self):
+        # GH#47986
+        df = DataFrame(
+            {"a": [1, 2]},
+            index=MultiIndex.from_arrays([Series([NA, 1], dtype="Int64")]),
+        )
+
+        result = df.to_string()
+        expected = """      a
+<NA>  1
+1     2"""
+        assert result == expected
+
     def test_datetime64tz_slice_non_truncate(self):
         # GH 30263
         df = DataFrame({"x": date_range("2019", periods=10, tz="UTC")})