«Moving to opportunity voluMe 14, nuMber 2 • 2012 U.S. Department of Housing and Urban Development | Office of Policy Development and Research ...»
MTO does not tell us anything about the effects of giving housing vouchers to people who are already living in the private housing market but without any sort of government subsidy. For that population, voucher receipt leads to large gains in disposable income for families because they can now spend much less out-of-pocket on rent, but it generates relatively little change in neighborhood conditions (see Jacob and Ludwig, 2012; Mills et al., 2006). The comparison of vouchers with living in the private housing market without a subsidy is relevant for the policy question of what happens when we expand the share of families receiving means-tested subsidies, which is important in its own right given that less than one-third of income-eligible families are in meanstested housing programs (Olsen, 2003).
MTO’s Effects on Neighborhood Conditions The logic model behind MTO is that assignment to the experimental or Section 8 group leads families to change their living conditions, which in turn leads to changes in their behavior and well-being. For there to be any value at all in looking at MTO impacts on behavioral outcomes, we need to first establish that the MTO demonstration did actually change the environments in which families were living. So that the articles in this Cityscape issue do not have to repeatedly replicate this material, I summarize MTO’s impacts on neighborhood conditions of participating families here. MTO also changed the housing conditions of families as well, which are carefully presented and discussed in the article in this symposium by Comey, Popkin, and Franks.
Exhibit 2 shows that, 1 year after the time of random assignment, even the ITT effects of MTO on neighborhood conditions were very large, despite the fact that many families who were offered MTO vouchers did not use them. The ITT estimates in exhibit 2 show that, 1 year after baseline, families assigned to the experimental or Section 8 group were living in census tracts with poverty rates that were 17 and 14 percentage points lower than the average census tract of the control group, which was 50 percent poor. (In what follows, I focus on the experimental-versus-control group contrast, which winds up providing the strongest test of “neighborhood effects,” although the contrast between the Section 8 and control groups is also of interest for what it can tell us about providing vouchers to public housing families and other key housing-policy questions about the right mix of housing program services.) Over time, MTO’s effect on neighborhood poverty rates diminishes. Exhibit 2 shows that the ITT effect on census tract poverty rates from being assigned to the experimental rather than control group was 10 percentage points measured 5 years after baseline, and about 5 percentage points measured 10 years after baseline.
What has not been widely appreciated is that most of this convergence in neighborhood conditions across randomized MTO groups is caused by improvements over time in the neighborhoods of control group families rather than by subsequent mobility (or “secondary moves”) by the experimental or Section 8 group families. The average census tract poverty rates for families assigned to the experimental group declined over the period from 1 to 10 years after baseline by 5 percentage points (from around 33 to 28 percent). The convergence in tract poverty rates between the experimental and control groups occurs because the control group experienced an even larger decline in tract poverty rates over this period, equal to fully 17 percentage points (from 50 to 33 percent).
Regardless of the cause, it is clearly true that the neighborhood conditions of the experimental and control groups became more similar over time. Rather than look at MTO’s impacts on tract poverty rates at a particular point in time, we can also average over the entire followup study period. Looking at MTO’s effects on average neighborhood conditions that families experience over the entire followup study period also fits with the common view that behavioral change may require accumulated exposure to neighborhood environments (see, for example, Clampet-Lundquist and Massey, 2008).
Exhibit 2 presents results that average the neighborhood conditions over all of the different addresses families report during the study period, giving more weight to those addresses at which people spent relatively more time. Over the course of the study period, the average control group family lived in a census tract that was 40 percent poor, compared with an average tract poverty rate for families assigned to the experimental group equal to 31 percent, for an ITT effect of 9 percentage points.
I have intentionally focused so far on the ITT effects of MTO on neighborhood environments to make it easier to see how much the changes over time in the control group neighborhoods are contributing to the convergence in neighborhood conditions between the experimental and control groups. As mentioned previously, however, it is also possible to calculate the effects of MTO on the neighborhood conditions of those who actually move through the program, or the TOT effects. Exhibit 2 shows the TOT effect on duration-weighted tract poverty rates was fully 18 percentage points, nearly one-half of the control group’s average tract poverty rate over the study period of 40 percent.
Exhibit 2 also shows that MTO had large impacts on an index of neighborhood disadvantage that Sampson, Sharkey, and Raudenbush (2008) argue may provide a better measure of the extent of neighborhood conditions compared with just looking at poverty alone. The index is a weighted average of census tract share poor, unemployed, share of households headed by a female, share receiving welfare, and share of the tract population that is under age 18.6 The logic behind this index is that some neighborhoods are considered low income because they are composed of twoparent families who are mostly working but have low earnings, whereas other neighborhoods are considered poor because they have a large share of single-parent households that are disconnected from the formal labor force. These two types of neighborhoods may have similar poverty rates but the social conditions in these two types of places will be quite different, which will be reflected in different values of the concentrated disadvantage index. Exhibit 2 shows that the average durationweighted tract disadvantage level of the control group in MTO over our study period was about
1.39. Those who move with an experimental group voucher experience a decline of 0.49 on this index, equal to around 35 percent of the control mean.
Although MTO focused explicitly on reducing economic rather than racial segregation for participating families, one might have expected there to be important changes in neighborhood racial segregation as a byproduct of the MTO moves, given that residents of high-poverty neighborhoods are very disproportionately likely to be Hispanic or African American (Jargowsky, 2003, 1997).
Whereas Sampson, Sharkey, and Raudenbush (2008) calculated the index using share African American as an additional component, we discuss MTO impacts on tract minority share separately and so do not include that variable in our own calculation of the index. The weights we use in exhibit 2 are based on a principal components analysis that Sampson, Sharkey, and Raudenbush (2008) calculated using tract-level data for Chicago from the 2000 decennial census and equal.90 for tract share receiving welfare,.88 for tract share poor,.86 for tract share unemployed,.87 for tract share households headed by female, and.73 for tract share under age 18.
Exhibit 2 makes clear, however, that MTO’s impacts on racial segregation for participants were fairly modest. The average control group family spent the study period in a census tract that was 88 percent minority. The tract share minority for those who moved with an experimental voucher was lower by a statistically significant amount, but the TOT effect of about 13 percent means that, over the study period, even the experimental group movers were living in census tracts in which fully three-quarters of all residents were members of racial and ethnic minority groups.
Despite the lack of major MTO impact on neighborhood racial composition, MTO moves led to sizable changes in neighborhood social processes that a growing body of sociological research suggests might be particularly important in affecting people’s life outcomes (Sampson, Morenoff, and Gannon-Rowley, 2002; Sampson, 2012). Note that exhibit 2 focuses on the self-reports of MTO adults about their social networks and neighborhood social processes measured 10 to 12 years after random assignment—that is, after the convergence in neighborhood poverty rates between the two treatment groups and the control group has occurred.
Exhibit 2 shows that, 10 to 12 years after baseline, the experimental group TOT effect on the likelihood of having at least one college-educated friend was nearly 15 percentage points, or about one-third of the control mean of 53 percent. The experimental TOT effect on the likelihood that neighbors would do something if local youth were spraying graffiti (intended to measure what Sampson, Raudenbush, and Earls, 1997, call “collective efficacy”) was over 15 percentage points, about one-quarter of the control group’s value of 59 percent.
MTO also delivered in terms of changing the neighborhood condition that was the main reason most MTO families signed up for the program originally—safety. Moving with an experimental group voucher reduces the local violent-crime rate (as measured by police data) by 876 violent crimes per 100,000 residents, equal to more than one-third the control group average of 2,420 violent crimes per 100,000.7 Self-reported data about neighborhood safety from MTO participants show similarly large effects. The experimental TOT effect on the likelihood that adults report feeling unsafe in their neighborhood during the day equals 7 percentage points, over one-third of the control group’s rate of 20 percent, and reduces the likelihood of having seen drugs used or sold in the neighborhood over the past month by 13 percentage points, over two-fifths of the control group value of 31 percent.
What Can MTO Tell Us About Neighborhood Effects?
If it had turned out that there were few differences in average neighborhood conditions between the two treatment groups and the control group in MTO, then the MTO demonstration will not have much useful to say about the existence of any “neighborhood effects” on families. In the previous section, however, we showed that MTO moves generate changes in neighborhood disadvantage and social processes that are, during the period initially after random assignment, extremely large.
These administrative records might understate MTO’s effects on safety, because the geographic resolution of the local area crime data we can get from police departments varies greatly across cities and is quite large in some places. Moreover, only about one-half of all violent crimes nationwide are reported to police (Truman and Rand, 2011), and we might worry that reporting rates are even lower in distressed areas where people tend to distrust the police.
14 Moving to Opportunity
Guest Editor’s IntroductionThese effects are still sizable when averaged over the entire study period, viewed in either absolute terms or as a share of the control group’s average neighborhood attributes. Why, then, do many people argue that MTO is a “weak treatment” that is of limited value for answering the social science question of whether and how neighborhood environments affect behavior?
One concern that I think is legitimate is that some potentially important neighborhood attributes were not changed very much by MTO, and in particular neighborhood racial composition. It is worth reiterating that many of the leading theories about why neighborhood environments might affect the well-being of residents focus on neighborhood attributes other than racial composition.
For example, the seminal work of Wilson (1987), which helped stimulate the sizable neighborhoodeffects research literature that has developed over the past 25 years, focused on the consequences for low-income African Americans from having middle-class African Americans move out to other areas. Wilson’s hypothesis is about the importance of neighborhood socioeconomic disadvantage, not racial segregation.
Some people have expressed the view that MTO is a weak treatment even with respect to the sorts of socioeconomic measures that I have argued in the previous section were strongly affected. Why is that? One reason is a frequent tendency to focus exclusively on the ITT effect on neighborhood conditions, even though the TOT effect can also be identified from the experimental data so long as we are willing to assume that assignment to one of the two voucher groups has little effect on those families who do not actually move with a voucher. Both types of estimates are of interest. ITT estimates are relevant for public policy because most housing-mobility programs in the real world would be voluntary, and so compliance will inevitably be less than perfect. The TOT estimates are of interest because they help extrapolate MTO results to other mobility interventions that might have different voucher compliance rates, and they are of scientific interest because relative to ITT estimates, the TOT more directly identifies the effects of changing neighborhood contexts on people’s outcomes.
A second reason MTO can look like a weak treatment is if one focuses on how far families change their rank in the national census tract poverty distribution. For example, Quigley and Raphael (2008) note that the low-poverty voucher ITT effect moves families from the 96th percentile to the 88th percentile within the census tract poverty rate distribution for the five MTO cities. As a share of all census tracts in the United States as a whole, however, there are just not all that many census tracts that have very high poverty rates. This means that large absolute changes in tract poverty rates can lead to relatively small changes in rank order at the top end of the distribution.