Measuring Neighborhood Change: The Issue of Ex Post Borders

Do more populous neighborhoods grow less quickly than less populous areas? Is local housing price growth associated with initial population density? The Longitudinal Tract Data Base’s (LTDB) panel of Census tracts is the standard tool for measuring neighborhood change. The LTDB is based on 2010 Census tract boundaries, and Census tracts are partially designed so that they have a similar level of population. In this paper, we show that defining neighborhoods to equalize ex post population levels can significantly impact estimated coefficients in regressions in which population changes are regressed on initial population levels or with variables that are correlated with initial population levels. Most obviously, if neighborhood populations are ex post equalized, then a regression of population change on initial population must yield a coefficient of -1. We address this challenge by offering five alternative panels of tracts using 1970, 1980 and 1990 boundaries, which can be thought of as ‘reverse LTDBs’. The significant mean reversion of both population and housing units that appear in the LTDB before 2000 either dramatically ameliorates or reverses using the reverse LTDB. Comparing the LTDB with the reverse LTDB also finds that using tracts based on ex post borders also can influence estimated growth relationships where other tract-level attributes such as house price are correlated with initial population levels. This does not imply that using ex ante borders always is superior; earlier borders almost always means fewer observations, especially in rapidly growing areas.