The housing and neighborhood conditions of America's children: patterns and trends over four decades

Abstract

This paper uses national and metropolitan area data from American Housing Surveys over four decades to examine the patterns and trends in the housing and neighborhood circumstances of children. Children across the income distribution have experienced dramatic improvements in the physical adequacy of their dwellings and in crowding but significant deterioration in housing affordability. Poor children are often in greatest jeopardy, with the rate of complaints about crime 25 percent higher in 2005 than in 1975, and the rate of school complaints twice as high in 2005 than 1975. Poor children also experience little payoff from residential mobility in terms of physical dwelling adequacy, crowding, affordability, or adequacy of schools, though moves are associated with fewer complaints about crime. However, it is the near poor – those between 101–200 percent of poverty – and not the poor who appear to be most affected by the tightness or looseness of the housing market.

Keywords:

• children
• housing
• neighborhoods

Acknowledgments

The authors gratefully acknowledge support for this research from the Center on Housing Policy, helpful insights and information from G. Duncan, F. Eggers, P. Emrath, T. Leventhal, and D. Vandenbroucke, programming assistance from D. Kantor, research assistance from A. Robie, and production assistance from L. Vernon-Russell.

Notes

¹The only examination of a range of features of children's housing we could find is a 1996 report published by the Low-Income Housing Information Service (Kaufman 1996).

²More than 20 years ago, Parke (1978, 73) advocated for descriptive studies of children's home environments, complaining that the literature provided “…only a scattered and fragmented assessment of a few aspects of children's home settings.”

³This review draws heavily on Leventhal and Newman (2010).

⁴This section is based on Eggers and Thackeray (2007) and D-HUD and US Census Bureau (2006).

⁵For comparison, roughly 54 percent of poor children and 27 percent of all children lived in single-parent households in 2005.

⁶Tests for differences in pattern were conducted with a general linear model (GLM) with year as the sole factor. Tests for differences between year pairs used Tukey's HSD post-hoc tests.

⁷The full detail of how this variable is defined is provided in the explanatory notes to Table 2. The data are collected via self-report from the survey respondent.

⁸At this writing, a revised standard has just been released indicating that there must be at least one bedroom or sleeping area for each two persons, though children of the opposite sex, except those who are very young, cannot be required to occupy the same bedroom or sleeping area (Code of Federal Regulations 2009).

⁹There is no standard definition of these categorizations. For this analysis, we used the following cutoffs: tight markets' ≤ 7 percent vacancy rate and ≥ 50 percent housing cost burden; moderate markets' 8–9 percent vacancy rate and 40–49 percent housing cost burden; and loose markets' ≥ 10 percent vacancy rate and ≥40 percent housing cost burden. Housing cost burden thresholds are based on moderate-income households. Housing markets in the 2004, 1995, and 1985 AHS metro that met these criteria define the sample for this analysis. (Results available from authors.)

¹⁰Although a Chow test also indicated significant differences by year, the results of models pooling the three years and including year dummy variables and separate models for each year are substantively quite similar. For simplicity, we show the pooled results and highlight important patterns across time. Results by year are available from authors. Note that models could not be tested with 1975 data because several of the housing and neighborhood outcomes are unavailable for that year.

¹¹Logits are used because the dependent variables are expressed in dummy variable form. Huber-White standard errors are used to address potential non-independence of child observations within families. To facilitate comparisons across variables and equations, odds ratios are reported. Estimates of model fit are given by McFadden's pseudo R² and adjusted R².

¹²Income and housing cost cannot both be included in the affordability regressions. We include housing cost only. However, in sensitivity tests that excluded housing costs and included income, income was statistically significant.

¹³American Community Survey data indicate that in 2008, the renter mobility rate was more than four times the rate for owners (32 percent versus 7 percent, respectively) (US Census Bureau 2009b).

¹⁴Ratings are based on a 10 point scale where 10 is best and 1 is worst.

¹⁵The exception here is the neighborhood rating model that includes housing measures and is estimated on the full sample. In this case, the housing cost burden measure is not significant, presumably because its contribution is captured by the other housing measures.

¹⁶For example, after taking the effect of degrees of freedom into account, the increased explanatory power of adding neighborhood indicators to the house rating model is statistically significant (F = 234.31; p<.001), as is the neighborhood rating model F = 57.103; p<.001) after including housing indicators.