The Influence of Neighborhood Poverty During Childhood on Fertility, Education, and Earnings Outcomes

Abstract

Previous studies attempting to estimate the relative importance of family, neighborhood, residential stability, and homeownership status characteristics of childhood environments on young adult outcomes have: (1) treated these variables as though they were independent, and (2) were limited in their ability to control for household selection effects. This study offers advances in both areas. First, it treats the key explanatory variables above as endogenously determined (sometimes simultaneously so). Second, to deal both with this endogeneity and the selection problem, instrumental variable estimates are computed for how childhood average values of neighborhood poverty rate relate to fertility, education and labor market outcomes in later life. The paper analyzes data from the US Panel Study of Income Dynamics (PSID) that are matched with Census tract data, thereby permitting documentation of a wide range of family background and contextual characteristics. For children born between 1968 and 1974, data are analyzed on their first 18 years and various outcomes in 1999 when they are between 25 and 31 years of age. The application of instrumental variables substantially attenuates the apparent neighborhood effects. Nevertheless, support is found for the proposition that cumulative neighborhood poverty effects averaged over childhood have an independent, non-trivial causal effect on high school attainment and earnings.

Keywords:

• Childhood
• neighbourhood effects
• socioeconomic success
• young adult outcomes
• educational attainment
• earnings
• homeownership
• family background
• PSID

Acknowledgements

This research is supported by a grant from the Ford Foundation. The authors wish to thank Jorg Blasius, Jurgen Friedrichs, Harry Holzer, Alex Marsh and anonymous referees for their helpful suggestions on earlier drafts. Seminar participants at the Universities of Southern California, Cologne and Cambridge also provided constructive suggestions. The research assistance of Jackie Cutsinger and Ying Wang and clerical assistance of Caitlin Malloy is gratefully acknowledged. The opinions expressed herein are the authors', and do not necessarily reflect those of the Boards of Trustees of the Ford Foundation or our respective Universities.

Notes

¹ The direction of the bias has been the subject of debate, with Jencks & Mayer (1990) and Tienda (1991) arguing that neighborhood impacts are biased upwards, and Brooks-Gunn et al. (1997) arguing the opposite.

² While other studies have discussed this issue, it has been in the context of the reflection problem (Manski, 1995) of people in the neighborhood tautologically cause the aggregate neighborhood characteristics to be what they are as well as the neighborhood causes constituent residents' behaviors (Duncan & Raudenbush, 1999).

³ Other recent research has employed natural experiments where the selection bias was minimized through geographic assignment of households through governmental housing program auspices (Aslund & Fredriksson, 2005; Edin et al., 2003; Oreopolis, 2003).

⁴ Such a longitudinal analysis has been strongly recommended as the vehicle for overcoming the reflection problem (Duncan & Raudenbsuh, 1999; Manski, 1995).

⁵ A database from Geolytics is used, the ‘Neighborhood Change Database’, that adjusts data in 1970, 1980 and 1990 tracts that have changed their boundary definitions over the years to values that would appertain had boundaries remained at their 1990 specifications.

⁶ For details, see Galster (2003a).

⁷ There are two exceptions to this. First, for those years in which the family lived in a rural area, the observed value of county characteristics is used. Second, for child age zero the observed value is used since a first-stage equation for year zero (due to unavailability of lagged variables) cannot be estimated.

⁸ Details of the first-stage regressions are available upon request.

⁹ However, some were asked again in 1975 and a question about union membership was collected from 1968 through to 1981.

¹⁰ The estimates here also differ substantially from those finding no statistically significant impacts from neighborhood poverty and associated measures of disadvantage; e.g. see Corcoran et al. (1992); Ensminger et al. (1996); Plotnick & Hoffman (1999).

¹¹ Wheaton & Clarke (2003) find cumulative neighborhood conditions much more powerful in explaining various child developmental outcomes than contemporaneous conditions.