Comments

Jeffrey Kling: Most social scientists believe that the social environment influences children's educational achievement. The importance of residential location in affecting children's education, however, is a subject on which there is conflicting evidence.⁵⁴ Given the increase in the spatial concentration of poverty, however, this question has taken on increasing urgency. Study of this issue has been hampered for at least two main reasons. First, the range of variation in residential location is often limited among otherwise similar children. Second, the fact that observably similar children (for example, same age and demographics) are living in different locations may reflect choices made by their families that are indicative of factors unobserved by an analyst. It is difficult to credibly identify the importance of residential location as distinct from those unobserved factors.

The research design in this paper uses a randomized demonstration program known as Moving to Opportunity (MTO) to address both of these issues. Section 8 rental vouchers from the Department of Housing and Urban Development (HUD) were provided by lottery to public housing residents. Some vouchers were valid only in low-poverty neighborhoods and were bundled with counseling services (the experimental group), while others could be used to move to any location (the Section 8 comparison group). Those who did not receive a voucher (the control group) retained their eligibility to continue to live in their original public housing project unit. The consequences of this design for research are that similar groups of families are living in vastly different neighborhoods, and that the randomization implies that, on average, the groups offered and not offered the vouchers through the lottery will be similar in observable and unobservable characteristics. This is a tremendous advantage for these authors in having potentially convincing results that could identify the impact of residential location on educational outcomes.

Other results using MTO data, collectively summarized in a volume edited by John Goering and Judie Feins, indicate that the impact of residential location on children may be particularly important. Research from the New York site indicates reductions in child behavior problems, in addition to the research cited by the authors on reductions in teen criminal behavior and children's physical health and behavior.⁵⁵ None of the other existing studies using the MTO demonstration effectively studies educational outcomes. This novel and innovative paper addresses a critical gap in our knowledge.

Besides the randomization of the housing voucher offers, the other principal element of this research design is the linkage of MTO participants to administrative data on educational outcomes, such as test scores, grade retention, absences, and disciplinary actions. When Jens Ludwig first discussed the idea for this project with me in 1995, I was frankly skeptical that these data could be produced, mainly because of the multitude of government agencies involved. So I am pleased to be able to congratulate this team of researchers for truly creating new knowledge through their combination of data entrepreneurship and analytical skill.

Use of administrative records in this research does have some fundamental limitations, however. For example, the data indicate increased grade retention and suspensions among teens in families offered vouchers through the MTO program. This may accurately describe the experiences of these teens. Yet the schools attended by those who have moved through the MTO program have changed at the same time as their residential location. It is not clear whether administrative records are indicative of different behavior by the youth or of different standards being applied to the same behavior in different schools. My prior intuition was that the role of different standards could be important, and the authors provide some evidence in support of this idea. Thus, I agree with the authors' conclusion that the results on grade retention and suspensions for teens are difficult to interpret.

Another limitation is that many children do not have records of test scores. The authors are very clear about the reasons for missing data, including problems linked to test records using...