Estimating Neighborhood Choice Models: Lessons from a Housing Assistance Experiment

Abstract

We use data from a housing assistance experiment to estimate a model of neighborhood choice. The experimental variation, which effectively randomizes the rents that households face, helps us to identify a key structural parameter of the model. Access to two randomly-selected treatment groups, in addition to a control group, allows for the out-of-sample validation of the model using a group of households who were not used in estimation and who faced a separate set of incentives. We use our estimated model to simulate the effects of changing the subsidy-use constraints implemented in the actual experiment and find that restricting subsidies to even lower poverty neighborhoods substantially reduces take-up. As a result, average exposure to poverty actually increases under these more restrictive subsidies. We also simulate the effect of adding additional subsidy restrictions based on neighborhood racial composition and find that this policy does not change the average household exposure to either neighborhood racial composition or poverty.

Keywords: Neighborhood Choice Models, Experimental Data, MTO

JEL Classification: R23, I38