Building a National Housing Permits Database

• In Event: New Tools for Understanding the Effects of State Housing Reforms

Abstract

Existing data on building permits is incomplete and inadequate for advanced research. Some cities maintain high-quality records, but most do not. Moreover, even if two cities both maintain good records, they may store information in different ways that are difficult to harmonize. In this project, we use a machine learning approach to construct a national data set on new residential building construction. Our prediction model identifies new buildings using USPS mailing addresses and cell phone GPS data, allowing us to estimate where new buildings are completed nationally, even in cities that do not maintain high-quality permit records. We validate the model on a held-out sample of data and show that it produces reliable estimates.

Author

• Michael Hankinson, George Washington University
  Presenting Author