Local Landscapes of Assisted Housing: Reconciling Layered and Imprecise Administrative Data for Research Purposes

Shiloh Deitz
Will B. Payne
Eric Seymour
Kathe Newman
Lauren Nolan
Rutgers University

Understanding the stock of rental housing affordable to lower-income households is a crucial task for local governments aiming to meet rising demand and inform policy priorities. However, enumerating the number of units with public housing, Project Based Section 8, and Low-Income Housing Tax Credit (LIHTC) assistance and identifying precisely where those units are located is deceptively challenging. Although federal datasets with that information are easily accessible, development and building location information may be unavailable or imprecise. Critically, identifying units that receive more than one form of assistance is hard, especially units with LIHTC. To address these challenges in New Jersey, the authors developed a largely automated and replicable process for precisely placing subsidized housing units into tax parcels. Doing so enables linking units across federal programs and with state and local data and to more accurately aggregate counts to integrate with decennial census and American Community Survey (ACS) data from the U.S. Census Bureau. Within New Jersey, the research team re-geocoded records in three datasets using two commercial geocoding services, assigned them confidence scores, designated records for manual handling, and then assigned them to parcels. Following those steps, they identified more than 15,000 units statewide with overlapping federal subsidies, which would lead to a 12-percent overcount of subsidized units in the state if the three datasets were used as given (and up to a 40-percent overcount in individual municipalities). By reusing and reconciling those datasets at the parcel level, researchers can more accurately enumerate rental units associated with different levels of subsidy depth and duration, a crucial task for identifying housing needs within and beyond the assisted rental stock.