Choosing the Bandwidth for Decomposing the Redistributive Effect: Evidence from Milan Using AMeRIcA Data