Initial Findings




Ludwig, Jens, Greg J. Duncan, and Paul Hirschfield, “Urban Poverty and Juvenile Crime: Evidence from a Randomized Housing-Mobility Experiment.” Final version published in the Quarterly Journal of Economics (May 2001), 655-679.


This paper uses data from a randomized housing-mobility experiment to study the effects of relocating families from high- to low-poverty neighborhoods on juvenile crime. Outcome measures come from juvenile arrest records taken from government administrative data. Our findings seem to suggest that providing families with the opportunity to move to lower-poverty neighborhoods reduces violent criminal behavior by teens.


This study uses

  • the HUD MTO baseline survey, completed by all participants in Section 8-only, Experimental, and Control groups of the MTO program, and
  • juvenile arrest records obtained from the Maryland Department of Juvenile Justice (DJJ). The DJJ records include three categories of crime: violent crimes, property crimes, and "other" crimes.  Data are used for those MTO teens who are between the ages of 11 and 16 at the time of assignment, excluding person-quarters that follow a participant's 18th birthday.


These data are used to show the effect of assignment in the experimental group on a dependent variable (the “supply” of criminal offenses), controlling for pre-program characteristics, and including dummy variables for calendar quarter and quarters since randomization. The use of panel data allows for the control of common trends in crime. The researchers also gauge the effect of “Treatment on the Treated” (TOT), assuming that assignment to the experimental or Section 8-only groups has no effect on families who did not relocate through MTO.


  • Findings indicate that starting 4 to 6 quarters after assignment to the experimental or Section 8-only groups, juvenile arrests for violent crime decline relative to the control group. Reductions in robbery account for about half this decline. The figures also suggest that teens in the experimental group may have higher rates of property crime arrest relative to the control group. This latter result is ambiguous, however, and is not statistically significant when controlling for pre-program characteristics.
  • The effects of treatment on the treated are slightly greater for the experimental group than for the Section 8-only group. For both these groups, the post-program arrest rates for violent crime are slightly higher for compliers than non-compliers. However, pre-program arrest rates also have a positive affect on a family's probability of making an MTO move, suggesting that it is not simply families with “better” outcomes that chose to relocate to low-poverty areas.
  • The evidence in this paper suggests that the offer to relocate families from high- to low-poverty neighborhoods reduces juvenile arrests for violent offenses by 30 to 50 percent, with some indications that there is a concomitant increase in property crimes in these groups.
  • Though the results presented here reflect both criminal behavior and the behavior of the local justice systems, the authors persuasively argue that the latter would tend to understate reductions in offending and overstate increases, thus adding weight to the finding of lower levels of violent crime among the treated groups.
  • Although it should be kept in mind that MTO participants are a self-selected group, these results at least suggest that policies which reduce the spatial concentration of poverty may affect overall levels of violent crime.

Ludwig, Jens, Greg J. Duncan, and Joshua C. Pinkston, “Neighborhood Effects on Economic Self-Sufficiency: Evidence from a Randomized Housing-Mobility Experiment.” Draft date: January 31, 2000.


The paper examines whether residence within high-poverty urban neighborhoods affects individual economic outcomes. Our data are generated by a randomized housing-mobility experiment, with measures of economic self-sufficiency taken from state administrative records. We find that providing low-income families living in public housing units with private-market rental subsidies that can only be redeemed in very low-poverty neighborhoods reduces rates of welfare use by around 15 percent. Most of this reduction appears to be explained by differences in welfare-to-work transitions. We also find that providing families with unrestricted housing vouchers has little effect on economic outcomes beyond the first year.


  • Do the characteristics of one’s neighbors affect behavior?
  • Does the location of a neighborhood affect labor market outcomes; that is, is there a spatial mismatch between inner city dwellers and suburban job opportunities?



Four sources of Baltimore area data are used in this paper:

  • The HUD MTO baseline survey, completed by all participants in Section 8-only, Experimental, and Control groups of the MTO program.
  • Post-program addresses, as collected by Abt Associates in phone surveys between July and December of 1997 (91% response rate).
  • Administrative data on public assistance from the Maryland Department of Human Resources. These data are current as of August 1998, an average of 3.2 years after random assignment. There was some disagreement between self-reported welfare receipt at the time of entry and these state public assistance records.  The authors find the match rate to be on the order of 80-90%.
  • Quarterly employment and earnings histories from the Maryland Department of Labor, Licensing and Regulation, covering under the Unemployment Insurance (UI) program those employers who pay more than $1,500 in a quarter to one or more employees. These data are less susceptible to misreporting problems such as recall error or self-presentation bias than self-reported data, but they also miss income from off-the-books work. This may bias findings, particularly if opportunity for such off-the-books work varies across neighborhoods.



MTO allows the authors to test the joint hypothesis that neither the location nor the social composition of inner-city neighborhoods affects economic self-sufficiency. Controlling for family and neighborhood characteristics, this study compares mean economic outcomes for families assigned to each of the three treatment groups, known as "Intent To Treat " (ITT). The random assignment to the MTO groups also allows for the inference of the effects of "treatment on the treated" (TOT), by comparing economic outcomes of the MTO families who complied with the treatment to those assigned to the control group who would have complied with the treatment.


  • On average, the proportion of experimental group families on welfare during the post-program period is 6 percentage points lower than for the control group (equal to about 15 percent of the control group’s welfare receipt rate). The Section 8-only group rate of welfare receipt is 5 percentage points lower than the control group in the first program year. This margin dissipates in subsequent years, while the gap between the experimental and control groups actually grows to nearly 10 percentage points in the third year. The central finding is that assignment to the experimental group reduces welfare receipt relative to controls, but assignment to the Section 8-only group has little effect beyond the first year.
  • This decline in welfare caseloads appears to be due to increased earnings in the formal labor market: the differences in welfare-to-work transitions between experimentals and controls appears to account for most of the differences in welfare receipts. The UI program does not show consistent differences in employment levels between the experimental and controls groups. The authors suggest that this is because income increases enjoyed by the experimental group may be "under the table", or otherwise unreported through the UI program.
  • Neighborhoods matter for economic self-sufficiency, at least for the Baltimore MTO population. The specific mechanisms through which this happens are unclear. They may involve changes in social capital, increased proximity to job opportunities, or improvements in social services or other local institutions.

Ludwig, Jens, Helen F. Ladd, and Greg J. Duncan, “Urban Poverty and Educational Outcomes.”  Brookings-Wharton Papers on Urban Affairs (2001), 147-201.


Using data from a randomized housing-mobility experiment, this paper examines the effects of neighborhood poverty on the educational outcomes of children as reported in school administrative records.  We find that the opportunity for public housing residents to move from high-poverty to low-poverty neighborhoods improves reading and math scores by about one- quarter of a standard deviation compared to their control group counterparts. The opportunity for residents of public housing to move using traditional Section 8 housing vouchers also generates positive impacts on some reading outcomes for young children, but no impact on math outcomes. Data limitations kept us from concluding much about the impacts of residential mobility on the academic achievement of older children. However, compared to the control group, adolescents in the two experimental groups appear to be retained in grade more frequently, and may experience greater disciplinary problems and higher dropout rates.


Though it is theorized that neighborhood conditions are directly related to children's educational outcomes, currently, there exists little empirical data to support this claim. Because most families have at least some degree of choice over where they live, observed correlations between neighborhood characteristics and child outcomes may be due either to the causal effects of the neighborhood environment, or to unmeasured family attributes that affect both residential choices and children's outcomes.

In this paper the authors estimate the effects of neighborhood conditions on children’s educational outcomes using data from the U.S. Department of Housing and Urban Development’s (HUD) Moving to Opportunity (MTO) residential-mobility program. Families applying for the program were randomly assigned into one of three treatment groups. Members of the experimental group were offered rental subsidies which could be used only for private-market housing in census tracts with very low poverty rates, as well as counseling services and assistance in their housing search from a local non-profit agency. Members of the Section 8-only comparison group were also offered rental subsidies, but were not required to move to a low-poverty census tract and were not provided any additional services. Members of the control group received no rental subsidies.

Impacts of the MTO program on children's educational outcomes were estimated in regressions using a panel of person-year observations for MTO children and controlling for a vector of pre-program child and family characteristics. The randomized experimental design of MTO breaks the link between family preferences and neighborhood conditions, helping to overcome the self-selection problem inherent in previous studies of neighborhood effects.


The study uses three sources of data, as follows:

  • The HUD MTO baseline survey, completed at the time of enrollment by household heads of all families in the program.
  • Follow-up addresses at two time points: during the immediate post-program period (reflecting the initial post-program moves of families), and during the second half of 1997. The address data were obtained from local housing authorities, change-of-address registries, credit bureaus, and a brief follow-up survey of MTO families.
  • Educational outcome data obtained from school administrative records. The main outcome measures for the study are based on student performance on two sets of standardized achievement test scores, the Comprehensive Test of Basic Skills (CTBS) for elementary and middle-school students, and the Maryland Functional Tests (MFT) for middle- and high-school students. Other outcome measures include the number of school absences, an indicator for disciplinary actions, and indicators for whether the student received special education services, dropped out of school, or was retained in grade.


Please note that there is a substantial amount of missing data for some students in some years.  This includes missing information for children who are absent from school on standardized achievement testing days or who are classified as having significant learning disability as well as for children whose records the school system could not find either because of misreporting of their name, date of birth, or social security or because they were enrolled in a private school. Data attrition might have also occurred because many school data-management systems are better at identifying current-year information than historical student data.  All of these things could have certainly affected the study's results.


  • The control group's educational outcomes generally deteriorate either absolutely or relative to the national average as children age. The MTO experimental and Section 8-only comparison treatments seem to slow the rate of relative decline in children’s test scores as they age, at least for younger children.
  • Assignment to the Section 8-only comparison treatment appears to improve young children’s CTBS reading scores by about 6 percentile points relative to controls, equal to about one-quarter of either the control group average or the standard deviation for reading scores. The Section 8-only group also appears to pass the MFT reading test at a rate that is around 6 percentage points higher than that of controls, but the difference is not statistically significant. The authors also observe no statistically significant difference between Section 8-only and control children in CTBS math scores.
  • For older children (age 12 and up at random assignment) the MFT reading test reveals no statistically significant differences across MTO treatment groups.
  • The study observe a substantial increase in grade retentions for experimental and Section 8-only teens relative to controls. The data are insufficient to allow authors to conclude whether the difference in grade retentions across groups changes over time.
  • Finally, the study observes relatively large increases in the proportion of experimental and Section 8-only teens who are suspended or expelled in school relative to control teens, although only the experimental intent-to-treat effect is statistically significant (and at only the 10 percent level).



Brennan, Brian and Manisha Modi.  “Baltimore.” Moving To Opportunity Research.  Created August 30, 2000.  Last modified February 25, 2004.

Return to MTO Main Page