«Mixed Messages on Mixed incoMes Volume 15, Number 2 • 2013 U.S. Department of Housing and Urban Development | Office of Policy Development and ...»
Cityscape 231 Cityscape: A Journal of Policy Development and Research • Volume 15, Number 2 • 2013 U.S. Department of Housing and Urban Development • Office of Policy Development and Research Barker One of these factors is building type. Research on the relationship between the well-being of children and housing during the 1960s and 1970s was mostly about the effects of building type rather than ownership (Conway and Adams, 1977). Large post-war public housing projects prompted concerns about the effects on families of living in high-rise buildings. Several researchers found that living on the upper floors of tall buildings had detrimental effects on children. These results are important in any investigation of homeownership because of the correlation between building type and ownership— most single-family houses are owned, but most multifamily units are rented. If, as older research suggests, building type can influence child welfare, then this factor must be carefully controlled in any investigation of the effects of ownership. Recent papers finding positive effects of ownership on children do not control for building type.
Mobility is another factor that is correlated with homeownership and the well-being of children.
Most research on residential mobility shows that frequent moves have detrimental effects on the academic performance and the well-being of children. Because the costs of moving are higher for homeowners than for renters, families who expect to move in the near future are much more likely to live in rental housing. Although pushing families into ownership might reduce mobility, doing so might cause bigger problems, because families often move in response to loss of a job or because of crime or other problems associated with particular locations. Families might also move to place their children in better schools. Immobility resulting from subsidized homeownership can make such a move more difficult. Renting in a good school district, particularly with government rental assistance, is often the only affordable way for children in poor families to access high quality education.
Wealth is also an important factor in the well-being of children. Wealthy people tend to buy their homes, but poorer people tend to rent. Children obviously benefit in many ways from family wealth, and so any observed correlation between homeownership and the well-being of children might be because of these effects, rather than any direct effects of ownership. Having equity in a house provides a cushion against financial difficulties, but so does having money in the bank or in stocks, bonds, or a family business.
Many unobserved factors associated with child welfare are likely to also affect the propensity of families to purchase housing. Homeownership is something that many Americans desire for a variety of reasons. Capable people are more likely to attain goals such as homeownership, and they are also more likely to do a better job of raising their children. Observed correlations between homeownership and the well-being of children may be because of these unobserved characteristics of parents, rather than homeownership itself.
The first careful research to look specifically at the effect of homeownership on children was conducted by Green and White (1997). Controlling for several factors, they found that, on average, children of homeowners stay enrolled in school longer than children of renters. Aaronson (2000) found that controlling for mobility and self-selection into homeownership eliminated the statistical significance of Green and White’s results. Barker and Miller (2009) also reexamined Green and White’s data, finding that adding dwelling type, mobility, and wealth cast serious doubt on their results. In fact, we found that ownership of an automobile had a higher estimated effect on the well-being of children than did ownership of a home.
232 Point of Contention: Homeownership and Child Well-Being The Evidence Does Not Show That Homeownership Benefits Children Haurin, Parcel, and Haurin (2002), using different data and corrections for self-selection and mobility, found positive effects of homeownership on measures of children’s academic performance and well-being. Barker and Miller (2009) also reexamined their results, finding that, when families switched from renting to owning, these measures did not improve, and when they switched from owning to renting, they did not deteriorate. Examining switches such as these is useful because it is a way of controlling for unobserved characteristics of parents that do not change when the switch is made.
Using new data that were available neither to Green and White nor to Haurin, Parcel, and Haurin, Barker and Miller (2009) found that controlling for child and family characteristics eliminated the statistical significance of homeownership. In fact, we found limited evidence of a negative effect, with homeownership associated with lower reading scores for young children.
The most recent work on the effects of homeownership and child welfare is a paper by Holupka and Newman (2012). These authors find little evidence of beneficial effects of homeownership and conclude that selection effects probably explain earlier findings of statistically significant positive effects. They also find that the effects of homeownership are different for White and African-American families. There is weak evidence implying a beneficial effect of homeownership for White children, but none showing such an effect for African-American children. Different results for different groups suggests that the effects of homeownership are not well understood, and perhaps that a factor other than homeownership itself is responsible for the correlations that have been observed.
Papers from a decade ago that show positive effects of homeownership on the well-being of children have been cited prominently in support of policies favoring homeownership. Subsequent research has cast serious doubt on these results. In addition, the nation has conducted a large experiment in expanding homeownership, which ended very badly. Since 2007, more than 20 million foreclosures occurred, which suggests that many families have been uprooted physically and financially as a result of the push for homeownership, justified in part by published studies showing that children would benefit. Now that the evidence suggests that no causal relationship is probable between homeownership and the well-being of children, policy neutrality toward housing tenure seems more appropriate.
Author David R. Barker is an adjunct professor in the Finance Department at the University of Iowa.
References Aaronson, Daniel. 2000. “A Note on the Benefits of Homeownership,” Journal of Urban Economics 47: 356–369.
Barker, David, and Eric Miller. 2009. “Homeownership and Child Welfare,” Real Estate Economics 37: 279–303.
Conway, Jean, and Barbara Adams. 1977. “The Social Effects of Living off the Ground,” Habitat International 2 (5/6): 595–614.
Green, Richard, and Michelle White. 1997. “Measuring the Benefits of Homeowning: Effects on Children,” Journal of Urban Economics 41: 441–461.
Haurin, Donald, Toby L. Parcel, and Jean R. Haurin. 2002. “Does Homeownership Affect Child Outcomes?” Real Estate Economics 30: 635–666.
Holupka, Scott, and Sandra J. Newman. 2012. “The Effects of Homeownership on Children’s Outcomes: Real Effects or Self-Selection?” Real Estate Economics 40: 566–602.
Thernstrom, Stephan. 1964. Poverty and Progress: Social Mobility in a Nineteenth Century City. New York: Atheneum.
234 Point of Contention: Homeownership and Child Well-Being Looking Back To Move Forward in Homeownership Research Sandra J. Newman C. Scott Holupka Johns Hopkins University For those interested in the effects of homeownership on children’s well-being, the empirical literature offers good news and bad news. The good news is that the topic has generated a sizable body of serious research by highly respected researchers. The existence of multiple studies examining the same basic question is the coin of the realm in the hard sciences but is all too rare in housing policy research. The bad news is that this research has not produced consistent evidence about whether the effects of homeownership are positive, negative, or nonexistent. Because the studies are not replications, differences in a host of features from sample composition to the approach for addressing selectivity bias could account for discrepant results. Until we begin to take stock of the state of homeownership research and the tedious task of sorting through the sources of these divergent results, we are unlikely to make much progress in understanding whether a “homeownership effect” exists. In this commentary, we start this stocktaking and sorting process. We briefly discuss four topics: addressing selection; specifying models; treating income, race, and ethnicity; and handling residential stability.
Selectivity Bias Researchers agree that a (or perhaps the) major challenge in estimating the net effects of homeownership on child well-being is separating the effects of the characteristics of parents who select into homeownership, which are highly correlated with child outcomes, from the effects of the homeownership per se. Most studies of homeownership address selection with an instrumental variable (IV) strategy.
Exhibit 1 summarizes key characteristics of 10 prominent studies of homeownership, nearly all of which use nationally representative panel data. The first column indicates that 9 studies use an IV, and 1 study (Engelhardt et al., 2010) uses an experimental design. The first 7 studies pertain to shorter term child outcomes or the effects of homeownership during childhood and adolescence on longer term outcomes in early adulthood. We include the final 3 studies, which pertain to effects on citizen engagement, because the selection problem applies to all homeownership outcomes (not only child outcomes), all IV studies must estimate a first-stage model predicting homeownership regardless of the outcome being predicted in the second-stage model, and this group of studies includes the only experimental evidence available.
We draw three main conclusions from reviewing the first column listing the IVs and the second column showing the strength of the IVs. First, the range of IVs tested to date and their weak performance indicate that we have not been successful in identifying theoretically valid and empirically strong instruments.1 This set of results may indicate that we need to think harder, but it may also indicate that we need to consider other strategies for addressing selection beyond the IV approach.
Second, while there is some overlap in IVs, only one identical replication exists: Holupka and Newman (2012) intentionally replicate Aaronson (2000) to allow for the comparison of results from the second-stage outcome models. Comparing results when the IV is different (or even when it is similar but not identical) could account for differences in both IV strength and outcome findings. For example, studies by Green and White (1997) and Galster et al. (2007) both use the ratio of the cost of owning to renting, measured in somewhat different ways, as their IV; both analyze national Panel Study of Income Dynamics (PSID) data for all income groups combined; and both examine education and teen birth outcomes in late adolescence. Yet Green and White report that their IV has no effect on their probit models showing effects of homeownership on both outcomes, and they conclude that no evidence of selectivity bias exists. By contrast, Galster et al. use the ratio of owning to renting as one of several IVs, report that their first-stage homeownership prediction model performed “only moderately well,” and find no effects of homeownership on these two sets of outcomes. Another example of inconsistent results arises in analyses using variants on the state homeownership rate, a second IV that has been used in several studies. Focusing only on studies that consider all income groups combined, this IV is relatively weak in DiPasquale and Glaeser (1999)2 and Galster et al. (2007) but is very strong in Aaronson (2000).3 Finally, although the IV approach has yielded mixed results to date, two other analytic approaches are consistent with the IV approach in not finding a homeownership effect. The strongest evidence comes from the sole homeownership experiment (Engelhardt et al., 2010), and additional evidence comes from propensity score matching (Holupka and Newman, 2012). In both cases, after selection is accounted for, homeownership has no effect on community engagement (Engelhardt et al.,
2010) or on child cognitive achievement, behavior, or health (Holupka and Newman, 2012).
Model Specification The third column of the exhibit lists the covariates included in the models. Although substantial overlap exists in demographic and socioeconomic background variables, substantial differences also exist across studies, which could account for discrepant results. Noteworthy for this brief commentary are the treatment of assets and wealth, neighborhood characteristics, and community characteristics. Key issues are whether controls for any of these indicators should be included, Some of the IVs used to date do not appear to meet the exclusion principle. For example, state homeownership rates could plausibly affect children’s cognitive achievement through their effect on property taxes, which are the main source of revenue for public schools.
DiPasquale and Glaeser rely on their uninstrumented ordinary least squares model because they describe the homeownership coefficient in the outcome model with the instrument as “implausibly large.” Although Aaronson reports large F-test results, the partial R2 for each model is small.
given endogeneity concerns, how they should be accounted for, and, if included, which measures should be used. Another issue is the inclusion of mediator variables in reduced form models, which will lessen the estimated effect of homeownership.
Treatment of Income, Race, and Ethnicity Most homeownership research includes income, race, and ethnicity among the array of independent variables in the first-stage homeownership prediction model and the second-stage outcome models.