Short- and long-term effects of HOPE VI redevelopment on neighborhood housing conditions

Abstract

Recognizing the negative effects of concentrated poverty on access to quality housing and neighborhood investment, governments have pursued poverty deconcentration policies such as mixed-income housing, which promotes neighbourhood social mix to improve housing and neighborhood conditions. This study evaluated the effects of HOPE VI, an influential mixed-income housing policy in the U.S., on the housing context of surrounding neighbourhoods using quasi-experimental methods and geocoded indicators from 1990–2016. We matched HOPE VI and control census tracts, then used a flexible conditional difference-in-differences method to estimate average treatment effects on the treated, accounting for differing treatment start dates and durations.

Results demonstrate that HOPE VI increased housing values and rents, decreased vacancy rates, and decreased age of housing relative to control tracts. There were no significant effects on homeownership rates. Impacts were strongest in neighbourhoods that were high poverty, high vacancy, and predominantly Black prior to grant receipt. Results suggest that HOPE VI changed broader neighborhood housing contexts.

Introduction

Neighbourhood economic segregation and concentrated poverty have long been hallmarks of life in the United States, driving disparities in access to safe and healthy living environments and compounding the challenges faced by people living in poverty (Dreier et al., 2013). Recognizing the negative effects of concentrated poverty on individuals’ access to quality housing and opportunities for social mobility, as well as on neighbourhoods’ potential to draw investment in housing and services, many governments have implemented policies aimed at deconcentrating poverty, particularly in urban centers (Bolt et al., 2010; Joseph, 2018). One dominant approach, termed ‘social mixing,’ involves transforming high-poverty neighbourhoods into economically-mixed communities, often through redevelopment of public (social) housing (Levy et al., 2010).

Although specific policy approaches vary, the concept of social mix has gained traction globally (Bolt et al., 2010). For example, programs in Australia, Europe, and the U.S. have aimed to create mixed-income communities through programs such as ‘right to buy’ that provide pathways to home ownership for public/social housing residents, and ‘balanced communities’ approaches that seek to encourage mixing across economic, racial/ethnic, social position (e.g., students, families, elders) and tenure type (renters/owners) differences (Bolt et al., 2010; Costarelli et al., 2019). Many such policies rely on public-private cooperation to fund physical housing improvements in the face of government austerity measures, and most focus on socioeconomic integration across social class.

These policies often have similar goals and are driven by similar assumptions – that improved housing and community conditions and the influx of higher-resourced neighbors will benefit neighbourhoods and increase economic and social vitality (Chaskin et al., 2013; Costarelli et al., 2019; Joseph, 2019; Van Kempen et al., 2009). Proponents of mixed-income housing have theorized that benefits would accrue across community and interpersonal levels, improving housing conditions, neighbourhood safety, and access to social capital (Joseph, 2018). Such policies are broadly aimed at strengthening communities and forcing changes in social mix in an effort to spur broader neighbourhood-level investments in improved housing contexts, neighborhood amenities, and economic opportunities, thereby improving individual access to essential resources.

In the U.S., one of the largest policy efforts aimed at creating mixed-income communities was the federal Housing Opportunities for People Everywhere (HOPE) VI program which operated from 1993 to 2010 and has influenced many subsequent policies such as the Choice Neighbourhoods initiative. HOPE VI came about after decades of discriminatory policies and underinvestment in affordable housing had transformed many U.S. public housing developments into communities of severely concentrated poverty, isolation, and unstable housing and neighbourhood contexts (Nebbitt, 2015).

Research has demonstrated that while HOPE VI met some of its primary goals, which were to demolish the most severely distressed, high-rise public housing complexes and lower concentrated poverty at the neighbourhood level (Coley et al., 2022; Tach & Emory, 2017), it is less clear whether the policy met its broader revitalization goals, which aimed to create positive spillover effects in the housing contexts in surrounding neighbourhoods through improved housing conditions, increased services and amenities, and enhanced economic opportunities (Greene, 2008). Recent research using rigorous, quasi-experimental analytic techniques found that HOPE VI did indeed reduce concentrated poverty at the neighbourhood level, but did not lead to an uptick in community resources and amenities, suggesting that the policy was not wholly successful (Coley et al., 2022). However, no research to-date has used these rigorous analytic techniques to evaluate the policy impacts of HOPE VI on neighbourhood housing conditions- including factors such as additional housing stock, vacancy rates, and home values and costs- across all HOPE VI sites, an important shortcoming given the importance of the housing context to individual and neighbourhood wellbeing. In this study, we consider how HOPE VI contributed to changes in the neighbourhood housing context in the short- and long-term. Specifically, we examine the following research questions:

• How did the redevelopment of public housing through HOPE VI grants change the housing context of surrounding neighbourhoods (including vacancy rates, home ownership rates, median home values, median rents, and median structure age) and were changes sustained over time?

• Did patterns of change vary by level of disadvantage and disinvestment within the neighbourhood pre-redevelopment, or by the extent of income mixing in the redevelopment?

Background

Public housing in the United States and the HOPE VI policy

The federal public housing program in the U.S. was initiated in 1937, developed to assist families affected by the Great Depression. It was the dominant approach to housing assistance for low-income families until the 1970s, at which point severely declining physical and social conditions led to reduced political support and continued funding shortfalls (Goetz, 2010). By the 1980s and early 1990s, large, high-rise public housing complexes in U.S. cities had become visual markers of disinvestment, urban poverty, and social problems (Katz, 2009). The National Commission on Severely Distressed Public Housing was created to address these issues, which led to the development of the Urban Revitalization Demonstration, later renamed HOPE VI (Baron, 2009).

HOPE VI was initially proposed as a mechanism for demolishing and redeveloping severely distressed public housing. Over time, however, it was reconceived with broader neighbourhood revitalization goals through the Quality Housing and Work Opportunity Reconciliation Act (QHWRA) of 1998. The goals of HOPE VI grew to include poverty deconcentration, housing revitalization, and increases in community services and economic development within surrounding neighborhoods in the most distressed public housing communities (McCarty, 2012).

Rigorous research has found that HOPE VI was successful in lowering neighbourhood concentrated poverty, but did not lead to improved neighborhood amenities and services (such as institutional and commercial resources and social services; Coley, et al., 2022; Tach & Emory, 2017). Yet it remains unclear whether the policy met its goals of broader revitalization of neighborhood housing. Those that have attempted to measure connections between HOPE VI and neighborhood housing contexts have been hampered by a number of limitations, with prior studies often focusing on a small selection of HOPE VI developments (see, for example, Baird, 2020; Castells, 2010) or using descriptive methods that cannot isolate the impact of HOPE VI (see, for example, Cloud & Roll, 2011; Zielenbach & Voith, 2010). As such, there is a need for rigorous evaluation of the effects of HOPE VI on housing-related indicators which were a core component of HOPE VI’s neighbourhood revitalization goals.

Theoretical background

The community-level goals of HOPE VI were to revitalize neighborhoods by deconcentrating poverty and improving living conditions in public housing developments, thus promoting positive spillover effects for the environment of surrounding neighbourhoods. Though the policy goals related to neighborhood revitalization were not clearly operationalized and neighborhood-level indicators were not required components of HOPE VI grantee reporting requirements, some researchers have used measures of the housing context such as declining vacancy rates and increased home values to assess revitalization (Baird et al, 2020). We follow this approach to consider whether HOPE VI redevelopment did, in fact, contribute to neighborhood revitalization.

Ellen and colleagues (2007) theorize several potential mechanisms through which public housing demolition and redevelopment might affect the local housing context. First, physical structure mechanisms suggest that removing severely dilapidated housing and replacing it with higher quality, more attractive housing would be expected to have positive spillover effects on surrounding properties in arenas such as increased values and lowered turnover and vacancy rates. Second, market effects mechanisms suggest that the large-scale redevelopment efforts of HOPE VI projects, particularly those integrating new market rate units, signify that the community is worth investing in, spurring spillover investment effects such as new housing builds. Third, population growth and population mix mechanisms suggest that increasing income mix as well as population are expected to lead to heightened commercial activity and economic vitality, making the community more attractive to invest in and to live in, and thus improving the local housing market (Ellen et al., 2007). Together, policymakers hypothesized that these processes would drive changes in social conditions and economic activity in HOPE VI communities, making them more desirable and making the local housing market more competitive (Joseph et al., 2007; Zielenbach, 2003a). Based upon this theoretical framework, we hypothesized that HOPE VI redevelopment would lead to increasing housing values and rents in surrounding neighbourhoods, spur new housing development, and decrease housing vacancies.

Prior research

Rigorous assessment of these hypotheses has been limited. Some research has focused only on shifts in the public housing communities themselves following HOPE VI redevelopment grants. In arguably the most comprehensive, HUD-sponsored evaluation of all 260 HOPE VI revitalization grants, the National Initiative on Mixed-Income Communities (Gress et al., 2016) provided a descriptive analysis of HOPE VI with a focus on units produced, income mix, and the processes of relocation, demolition, and construction. They found large-scale displacement of original public housing residents in HOPE VI communities between 1993 and 2014, with HOPE VI leading to an overall loss of public housing units, with nearly 80% of redeveloped units occupied by new rather than original public housing residents (Gress et al., 2016). They also found that HOPE VI sites with more market rate units leveraged more overall funding, while those with only public housing and affordable units were provided with less federal funding and were less able to leverage outside funds (Gress et al., 2016). However, the evaluation could not capture the policy’s effect on the neighbourhood housing context, which they cited as a key area for future research. Other research, summarized by topic below, has focused on the effects of HOPE VI on various indicators of the neighbourhood housing environment including home values, rental and sale prices, and housing vacancy rates.

HOPE VI effects on home values

Home values can be viewed as a measure of a neighbourhood’s desirability as well as its potential to offer educational, employment, and recreational opportunities for residents. Much research has documented that public housing is disparately located in neighbourhoods with low housing values, and low percentages of single family, owner occupied units (Zielenbach, 2003a). Ellen and colleagues’ (2007) model suggests that HOPE VI redevelopment would lead to increased home values, a hypothesis with some support in extant literature. For instance, Nguyen’s (2005) review of the literature on public housing found that well-designed redevelopment of affordable housing showed generally positive externalities for local home values (Nguyen, 2005). However, most studies that have examined links between redevelopment and housing values include numerous types of affordable housing (e.g., scattered site, Section 8, Section 202) and use single site or city-specific samples (Ellen et al., 2007). This evidence suggests that links between redevelopment and home values vary based upon the type of housing being redeveloped, the racial composition of the housing and community, and ownership of the development (e.g., non-profit versus privately owned) (Ellen et al., 2007; Santiago et al., 2001; Zielenbach, 2003a). Three quasi-experimental studies summarized below found that HOPE VI was associated with increases in local real estate prices, though results were heterogeneous based on neighbourhood features.

In a study focused on Pittsburgh’s Hill District, Baird and colleagues (2020) found that HOPE VI redevelopment drove significant increases in residential sales prices over the 16-year post-redevelopment period examined. Although limited by the single site analysis, the study was bolstered by the fact that it measured both HOPE VI investment and leveraged investments from other sources. Castells (2010) used a difference-in-differences approach to assess the spillover effects of 3 HOPE VI sites in Baltimore on surrounding property values. The study found significant positive effects in only one of the three sites, with a 10% increase (about $17,000) in the value of properties within 2,000 feet of the HOPE VI development. Further, the effects increased the most in the year after completion and remained durable, though smaller, in the 2 subsequent years measured (Castells, 2010). The author posited that the differing findings across sites, with only one showing ‘convincing evidence of a positive effect,’ was likely due to its location in an up-and-coming neighbourhood and their adherence to the mixed-income model which may have had a larger effect on the local housing market (Castells, 2010. p. 66). Brown (2009) examined HOPE VI developments across 4 cities. Using a quasi-experimental adjusted interrupted time series approach, the study found that the prices of units within .5 and 1 mile rings surrounding HOPE VI redevelopments significantly increased each year over a 10 year period after projects were completed, with higher growth in the closer versus outer rings. While this research points to positive effects of HOPE VI redevelopment on housing values in select cities, no research that we are aware of to-date has assessed these effects across the full national population of HOPE VI redevelopments.

HOPE VI effects on housing rental prices

Expectations regarding the effects of HOPE VI on housing rental prices are mixed, with some arguing that redevelopment should ease rent pressure by adding more rental units to the neighbourhood, while others suggest that improved housing conditions will make the neighbourhood more desirable and thus increase rental costs (Damiano & Frenier, 2020). Most evidence to date supports the latter hypothesis, though there is some variation across studies, likely driven by local housing conditions. (Cloud & Roll, 2011; Damiano & Frenier, 2020).

A number of studies have examined changes in rental prices following HOPE VI redevelopment, primarily using cross sectional and descriptive designs and small or single site samples. In the previously described study in Pittsburgh’s Hill District, Baird and colleagues (2020) found a 25.7% increase (equivalent to $131 higher average monthly rent) in rental prices following HOPE VI related investments. Similarly, in a study of four HOPE VI sites (selected due to their adherence to HOPE VI’s policy and implementation goals), Turbov & Piper (2005) found increases in property values and market rents in comparison to broader city and metro-area trends.

Several other descriptive studies have examined rental costs surrounding HOPE VI redevelopments using methods including single site case studies and comparisons between HOPE VI neighbourhoods and matched neighbourhoods in their respective cities (GAO, 2003; Holin et al., 2003; Zielenbach, 2003a; 2003b). Some found increases in rental prices in neighbourhoods surrounding HOPE VI sites, although the rents often remained well below city/metro averages (Zielenbach, 2003a; 2003b), while others reported no significant changes in local rents or found great variation across HOPE VI sites (GAO, 2003; Holin et al., 2003). Further research is needed to assess shifts in rental costs across all HOPE sites, and to more systematically assess how community characteristics may alter effects of redevelopment on these costs.

HOPE VI effects on home ownership and housing vacancy

Home ownership is another frequently assessed measure of neighbourhood stability and advantage, while high rates of housing vacancy, conversely, can be seen as a sign of instability and disinvestment. While limited theory informs the relationship between redevelopment and these housing indicators, HOPE VI redevelopment may have contributed to increases in home ownership and decreases in vacancies if investment catalyzed neighbourhood revitalization and improved neighbourhood desirability. Further, about 40% of HOPE VI developments provided new opportunities for home ownership for subsidized and unsubsidized residents, leveraging housing legislation that allowed Section 8 subsidies to be applied to mortgage payments (Gress et al., 2016). Finkel and colleagues (2000) note that this comports with the vision of HOPE VI to promote resident self-sufficiency and upward mobility, and to attract higher income residents spurring overall neighbourhood revitalization. However, there is limited rigorous empirical evidence of HOPE VI effects on local home ownership and housing vacancy rates.

In a pair of articles, Zielenbach (2003a; 2003b) focused on 8 HOPE VI communities that received awards in 1995 or earlier, comparing them to other high-poverty communities within their respective cities. Using residential loan rate information as a marker of levels of homeownership, Zielenbach (2003a; 2003b) found that HOPE VI neighbourhoods started off with lower residential loan rates and higher vacancy rates than other high-poverty communities in their respective cities. In absolute terms, HOPE VI neighbourhoods saw improvement in residential and commercial lending, with nearly a doubling of the residential loan rate (Zielenbach, 2003a). Housing vacancy rates, in contrast, rose to rates higher than other high-poverty neighbourhoods by 2000, perhaps due to incomplete HOPE VI redevelopments (Zielenbach, 2003a). Overall, Zielenbach (2003a; 2003b) found that HOPE VI neighbourhoods remained very distressed (behind city averages), but better off than other local high-poverty neighbourhoods in their respective cities.

Two other government-commissioned reports examined home ownership and housing vacancy rates. A Government Accountability Office (GAO) report (2003) compared four HOPE VI neighbourhoods to comparable neighbourhoods and found increased mortgage lending in three of the four HOPE VI communities. Holin and colleagues (2003) used Census data to assess neighbourhood conditions, including the percent of owner-occupied housing and vacancy rates, in 12 HOPE VI communities. They found that in 10 of the 12 HOPE VI communities assessed, the percent of owner-occupied homes increased between 1990 and 2000, while vacancy rates decreased in 9 out of the 12 HOPE VI neighbourhoods. Similar to Zielenbach (2003a), this report cautions that observed changes in vacancy rates may vary by the phase of HOPE VI redevelopment. In one site, for example, many residents had been displaced and the development had not been re-occupied by the final data point (Holin, et al., 2003). Additionally, as Popkin and colleagues (2004) caution, HOPE VI is but one of many possible factors that could have driven improvement or decline in the HOPE VI communities assessed in each of these studies, and the methods used do not allow causal interpretations.

The present study

Many limitations remain in our understanding of how HOPE VI affected neighbourhood housing contexts in the communities surrounding redeveloped public housing sites. Prior research suffers from limitations in external validity and generalizability due to assessment of only subsets of HOPE VI sites, typically the earlier sites funded. Numerous internal validity threats are apparent due to descriptive methods and an inability to adequately control for potential confounders. Finally, although some research has suggested that effects of HOPE VI on local housing markets may vary depending upon neighbourhood characteristics or characteristics of the redevelopment itself, prior research has not systematically evaluated such moderation effects across HOPE VI sites.

This study seeks to address these limitations and expand our understanding of the effects of HOPE VI on the local housing context. Specifically, we use a national sample of HOPE VI sites funded throughout the program from 1993 through 2010, improving the external validity and generalizability of results. In addition, using a rigorous new quasi-experimental analytic technique designed specifically to evaluate the effects of time-varying interventions, we selected matched control sites and rigorously evaluated effects of HOPE VI redevelopment using a flexible conditional difference-in-differences estimation model, which incorporates variability in both the start date and the length of the intervention in order to address selection bias and improve the internal validity of our estimates (Coley et al., 2022; Dettman et al., 2021). This analytic technique also allowed us to assess both short and long-term effects of HOPE VI. Finally, we considered how the effects of HOPE VI on local housing market indicators varied by community and redevelopment features. Through these methods, we sought to provide insights into whether HOPE VI contributed to changes in neighbourhood housing contexts.

Methods

Analytic sample and data sources

To address our research questions, we compiled a unique neighbourhood-level dataset composed of information on U.S. public housing developments, receipt and implementation of HOPE VI revitalization grants, and neighbourhood characteristics from 1990 to 2016. HUD’s ‘Picture of Subsidized Households’ database provided data on the locations and characteristics of public housing developments. Data were available in 1993, 1996–1998, 2000, and 2004–2016, allowing us to track changes in the characteristics of specific developments over time. One complication was that HUD shifted to a new Project ID numbering system between 2007 and 2008, making it difficult to match cases between these years. We used several non-exact matching programs in Stata 15 (i.e., matchit and reclink) followed by manual assessments, when necessary, to match developments based on their name, location, size, and resident characteristics. Because the HOPE VI program targeted large urban public housing developments, we excluded public housing developments that were outside of metropolitan counties, as well as scattered sites, vacant sites, elder-only developments, and developments with 25 or fewer units. This left us with 7,202 developments across the continental U.S.

We next drew data on HOPE VI grant implementation from the National Initiative on Mixed-Income Communities (NIMC) Mixed-Income Database which included start and end dates for different phases of redevelopment, the size and income mix of redeveloped projects, and financing information for HOPE VI revitalization grants (Gress et al., 2016; NIMC, 2021). These data were merged with our public housing dataset using project ID numbers. We thus had information on which public housing developments received HOPE VI revitalization grants, when those grants were received, and the duration of redevelopment.

Finally, census tract level data on neighbourhood characteristics were drawn from the Decennial Census in 1990 and 2000, and annually from American Community Survey (ACS) 5-year estimates between 2007 and 2016. We coded ACS 5-years estimates to represent the midpoint of each 5-year span (e.g., 2005–2009 estimates were coded as 2007 data). All tract-level data were normalized to 2010 census tract boundaries using the Longitudinal Tract Database (Logan et al., 2016) to allow for comparison of neighbourhood characteristics over time. Because neither multiple imputation nor full information maximum likelihood were compatible with our analytic approach, we used linear interpolation (ipolate command in Stata) to create a long form dataset with annual estimates from 1990 to 2016. Linear interpolation is a common strategy used to estimate neighbourhood conditions in years between data releases, including with Decennial Census data (Do, 2009; Ludwig et al., 2012).

We then used ArcGIS 10.6 to geolocate each public housing development based on its latitude and longitude within its census tract (using 2010 boundaries). When multiple public housing developments were located within the same census tract, we combined data across all developments. These analytic steps resulted in a dataset of 3,884 urban census tracts with data from 1990 through 2016 that contained public housing, with an average of 2.86 developments per tract.

Measures

Our treatment indicator, HOPE VI redevelopment, was created using data from the NIMC. Each census tract was categorized as “HOPE VI redevelopment in progress” (1) or “No HOPE VI redevelopment in progress” (0) each year. Tracts in which at least one development had received a HOPE VI grant were marked “in progress” from the start of redevelopment until redevelopment completion. In tracts in which multiple developments received a HOPE VI grant (n = 14), we considered redevelopment to start with the earliest grant start date and to end when the first redevelopment was completed.

To consider how HOPE VI affected local housing contexts, we focused on housing measures drawn from the Decennial Census and the ACS. These included vacancy rate, ownership rate, median home value, median rent, and median structure age. Vacancy and ownership rates were coded as proportion of the housing units in each census tract that were vacant and owner-occupied, respectively. Median home value and median rent served as proxies for property values and costs of living in the neighbourhood and were adjusted for inflation to 2010 dollars, assessed in $10,000 units, and logged to correct for non-normality. Finally, median structure age served as a proxy for new housing development. Additional neighbourhood measures, also drawn from Decennial Census and ACS data, were used in our matching process. These variables included the household poverty rate, percent Black residents, percent non-Black people of color residents, and median household income (adjusted for inflation) of households in each census tract.

We were also interested in assessing whether effects of HOPE VI redevelopment varied by neighbourhood and redevelopment characteristics. To this end, we considered how patterns differed by pre-redevelopment economic composition, racial composition, and vacancy rate, and by type of redevelopment that was undertaken. All neighbourhood moderators, drawn from Decennial Census and ACS, were dichotomized around the median. As a result, we compared effects of HOPE VI on neighbourhoods with a moderate (<40%) versus high (≥40%) poverty rate; with moderate (<70%) versus high (≥70%) percent Black residents; and with a moderate (<13%) versus high (≥13%) vacancy rate, assessing these three characteristics two years prior to the initiation of the HOPE VI grant. These indicators tap into the level of economic isolation, Black racial isolation, and housing distress that communities were experiencing prior to receiving HOPE VI grants. Finally, we considered variation by redevelopment type, comparing redevelopments that included only affordable units to those that were mixed-income (i.e., including both affordable and market rate units). This information was drawn from the NIMC database.

Data cleaning was conducted in Stata 15. Following linear interpolation, missing data were minimal, ranging from .03% to 1.32% across analytic variables. Due to skewness, median home values, median rent, median household income, and percent non-Black people of color were transformed using a natural log to improve normality.

Table 1 presents descriptive statistics of HOPE VI and all non-HOPE VI tracts (termed potential control tracts), assessed at two years prior to redevelopment for each HOPE VI tract. These descriptive results show that HOPE VI redevelopments were located in census tracts with higher poverty rates, lower median incomes, and more Black residents than potential control tracts, with little difference in the percent of residents who were non-Black people of color. HOPE VI redevelopments also were located in tracts with higher vacancy rates and lower ownership rates compared to potential control tracts. Median home values and median rent were also lower, and structures were older in HOPE VI tracts compared to control tracts. These differences suggest selection bias in the neighbourhoods in which HOPE VI sites were located, and highlight the need for analytic techniques to correct for such bias.

Table 1.

Descriptives for HOPE VI and potential control tracts at matching.^a

	HOPE VI Tracts		Potential Control Tracts
	N	Mean (SD)	N	Mean (SD)
Tract Characteristics
Percent in Poverty	309	40.10 (17.56)	44,498	25.50 (14.97)
Percent Black	309	60.08 (33.88)	44,498	31.15 (32.38)
Percent non-Black People of Colorc	309	2.12 (1.18)	44,498	2.33 (1.17)
Median Household Income (per $10,000)c	309	10.20 (0.50)	44,487	10.58 (0.46)
Median Home Values (per $10,000)c	307	11.66 (0.61)	43,996	11.81 (0.62)
Median Rentc	309	6.13 (0.43)	44,478	6.36 (0.40)
Ownership Rate	309	30.44 (19.44)	44,498	45.73 (22.02)
Vacancy Rate	309	14.84 (8.14)	44,490	9.93 (6.23)
Median Structure Age (years)	309	62.51 (10.77)	44,499	59.00 (13.60)
Mixed-income post-redevelopmentb	309	0.33 (0.47)

Note:

^aMatching occurred 2 years prior to grant receipt.

^bThis variable is applicable only to developments that received a HOPE VI grant and was thus not used in the matching process. In contrast to other indicators, it is time invariant.

^cThese variables were transformed using a natural log.

Analytic approach

We addressed our two research questions using the flexpaneldid toolbox in Stata (Dettman et al., 2021). An extension of Heckman and colleagues’ (1998) conditional difference-in-differences approach, this empirical approach incorporates individual variation in treatment timing- both start date and duration- to estimate average treatment on the treated (ATT) effects. Sharing similarities with a host of new analytic techniques designed to estimate causal treatment effects of time-varying treatments (e.g., Callaway & Sant-Anna, 2021; Imai et al., 2021), flexible conditional difference-in-differences provides numerous advantages over prior analytic techniques.

First, the flexpaneldid technique matches ‘treatment cases’ (defined as a census tract including a public housing development which received a HOPE VI redevelopment grant) with one or more control cases (census tracts with public housing not receiving a HOPE VI grant), matching based on the individual start date of treatment, which varied across sites. We matched two years prior to the start of each HOPE VI grant to address the possibility of Ashenfelter’s dip (1978) and to assess characteristics of census tracts at the approximate time such characteristics would have affected HOPE VI granting decisions. Our matching algorithm used a combination of exact and nearest neighbor matching. We used exact matching on year and geographic region to control for macroeconomic and political contexts, while using nearest neighbor matching that used a combined statistical distance function to match on percent poor residents, percent Black residents, percent non-Black people of color residents, median income, and our five housing measures: vacancy rate, homeownership rate, median home value, median rent, and median structure age. Only treatment cases with control cases identified with common support—that is, when a strong match was attained— were retained, supporting internal validity. This matching procedure was highly successful in creating equalized samples of treatment and control cases and erasing the initial differences between HOPE VI and potential control cases. As seen in Table 2, after matching, no significant differences in means (assessed through t-tests) or variances (assessed through Kolmogorov-Smirnov tests) remained across any of the housing and neighborhood variables.

Table 2.

Nearest neighbor matching results.

		% in Poverty	% Black	% non-Black POC	Median Household Income	Median Home Values	Vacancy Rate	Ownership Rate	Median Rent	Median Structure Age
Median Home Values	Control (N=265)	38.54	58.62	2.07	10.22	11.64	13.61	31.91	6.15	63.12
	Treated (N=307)	39.91	59.86	2.12	10.21	11.66	14.78	30.63	6.13	62.56
	t-test	1.00	0.45	0.54	−0.30	0.47	1.88+	−0.84	−0.74	−0.66
	K-S Test	0.06	0.04	0.05	0.06	0.07	0.09	0.08	0.07	0.08
Median Rent	Control (N=268)	38.60	58.59	2.05	10.21	11.60	14.01	32.14	6.13	62.88
	Treated (N=307)	39.91	59.86	2.12	10.21	11.66	14.78	30.63	6.13	62.56
	t-test	0.95	0.46	0.73	−0.21	1.16	1.22	−0.99	−0.25	−0.38
	K-S Test	0.06	0.05	0.05	0.04	0.08	0.07	0.09	0.05	0.07
Ownership Rate	Control (N=267)	38.56	58.65	2.07	10.22	11.62	14.07	31.82	6.14	62.90
	Treated (N=307)	39.91	59.86	2.12	10.21	11.66	14.78	30.63	6.13	62.56
	t-test	0.98	0.44	0.57	−0.31	0.82	1.12	−0.77	−0.44	−0.39
	K-S Test	0.07	0.05	0.05	0.05	0.08	0.07	0.07	0.05	0.07
Vacancy Rate	Control (N=260)	38.64	58.59	2.05	10.22	11.62	14.22	31.89	6.15	63.00
	Treated (N=307)	39.91	59.86	2.12	10.21	11.66	14.78	30.63	6.13	62.56
	t-test	0.93	0.46	0.81	−0.24	0.88	0.89	−0.83	−0.62	−0.52
	K-S Test	0.06	0.05	0.05	0.05	0.07	0.06	0.09	0.06	0.06
Median Structure Age	Control (N=266)	38.10	58.45	2.06	10.22	11.63	13.86	32.02	6.15	62.86
	Treated (N=307)	39.91	59.86	2.12	10.21	11.66	14.78	30.63	6.13	62.56
	t-test	1.33	0.51	0.64	−0.47	0.67	1.47	−0.92	−0.59	−0.35
	K-S Test	0.07	0.06	0.05	0.06	0.07	0.07	0.09	0.06	0.05

Note. T-tests assess significant differences in means. K-S tests assess significant differences in variances.

⁺p<.10;

^*p<.05;

^**p<.01.

The second stage of the flexible conditional difference-in-differences approach estimates average treatment effects on the treated (ATT) with a non-parametric approach that incorporates variation in both treatment start dates and treatment duration across treated tracts and their matched controls. In addition, the technique can incorporate different lengths of assessment, allowing assessment of both shorter- and longer-term effects. We estimated ATT effects at the conclusion of treatment (the year of redevelopment completion), as well as two years, five years, and ten years after the conclusion of treatment in order to assess short-term effects of HOPE VI and whether effects were sustained over time. Equation 1 presents our analytic model, in which outcomes, Y, for treatment i and control j tracts are assessed from the start date, t0i, through varying numbers of time units, with βi signifying the end of treatment for each case and a the end of the outcome observation. The ATT was calculated as the weighted average of different years of treatment observations, with differences in outcomes between each treatment year and the start date twice differenced for each pair of treatment and control tracts. Significance was assessed using a t-test with corrected standard errors (Dettmann et al., 2021).

$$\mathrm{ATT}=\frac{1}{I}\sum _{i=1}^I\left[\left({\mathrm{Y}}_{\mathrm{i},\mathrm{t}0\mathrm{i}+\mathrm{\beta }\mathrm{i}+a}-{\mathrm{Y}}_{\mathrm{i},\mathrm{t}0\mathrm{i}}\right)-\left({\mathrm{Y}}_{\mathrm{j},\mathrm{t}0\mathrm{i}+\mathrm{\beta }\mathrm{i}+a}-{\mathrm{Y}}_{\mathrm{j},\mathrm{t}0\mathrm{i}}\right)\right]$$

We first estimated models to assess effects of HOPE VI zero, two, five, and ten years after the conclusion of treatment for the full sample of treatment and control sites. We then estimated a series of moderation models to assess whether effects varied across our four moderators. Because the analytic technique does not allow interactions, we first split the sample by each moderator, matched the HOPE VI and control sites within each subgroup, and then estimated ATTs within each subgroup. Moderators included (1) neighbourhood poverty, comparing neighbourhoods with <40% versus ≥40% poverty; (2) neighbourhood percent Black residents, comparing <70% versus ≥70% Black; (3) neighbourhood vacancy rates, comparing neighbourhoods with <13% versus ≥13% vacancy rates; and 4) whether HOPE VI development retained all affordable units versus introducing market rate units. The first three moderators were assessed two years prior to the initiation of each HOPE VI grant, while affordable versus mixed-income reflected the status of each HOPE VI site following redevelopment.

We also estimated a set of alternate models to test the robustness of our results. First, to help address concerns that some public housing developments may have acquired public funding for similar efforts outside of the HOPE VI redevelopment grants, we excluded tracts which had received federal Choice Neighbourhoods grants (N=22) or HOPE VI demolition-only grants (N= 205) from the population of census tracts with public housing developments prior to the matching step.

A second set of sensitivity analyses incorporated a more rigorous matching process. In these, we included in the matching step an additional requirement to match on the trend in the outcome variable from 7 years to 2 years prior to the initiation of treatment. By matching on trends rather than only on a single point in time, these alternate models provide an extra layer of rigor to our selection of control neighbourhoods with similar characteristics to treatment neighbourhoods (Dettman et al., 2021).

Results

Average HOPE VI treatment effects over time

Results of the flexible conditional difference-in-difference models are displayed in Table 3. Results show significant effects of HOPE VI redevelopment on median home values, median rents, vacancy rates, and median structure age, but not on home ownership. HOPE VI led to an increase in median home values and median rents relative to control tracts, with effects emerging upon completion of redevelopment and persisting five years post-redevelopment. Specifically, given similar housing characteristics at matching (2 years prior to redevelopment), median home values rose 0.07, 0.09, and 0.10 log units more in HOPE VI than control tracts through zero, two, and five years after redevelopment, respectively, and median rents increased 0.06–0.07 log units more in HOPE VI tracts than in matched controls over the same period. Ten years post-redevelopment, however, there were no longer significant differences in median home values or median rents between HOPE VI and control tracts.

Table 3.

Flexible conditional difference-in-differences model results assessing effects of HOPE VI on housing indicators.

	0 years after end			2 years after end			5 years after end			10 years after end
	(n=278–574)			(n=266–556)			(n=233–480)			(n=103–219)
	Treated	Control	Estimate(SE)	Treated	Control	Estimate(SE)	Treated	Control	Estimate(SE)	Treated	Control	Estimate(SE)
Median Home Value	0.19	0.12	0.07(0.03)**	0.21	0.13	0.09(0.03)**	0.21	0.11	0.10(0.04)**	0.26	0.24	0.02(0.04)
Median Rent	0.17	0.11	0.07(0.02)**	0.17	0.10	0.07(0.02)**	0.19	0.13	0.06(0.02)*	0.22	0.19	0.03(0.03)
Ownership Rate	1.12	0.61	0.50(0.65)	0.15	0.23	−0.08(0.79)	−0.85	−0.05	−0.80(0.85)	−0.53	−1.84	1.31(1.18)
Vacancy Rate	0.41	2.05	−1.64(0.55)**	−0.47	2.37	−2.84(0.69)**	−0.63	2.25	−2.88(0.66)**	−0.28	2.89	−3.17(1.01)**
Median Structure Age	−7.36	−2.93	−4.43(0.8)**	−9.62	−2.89	−6.74(0.96)**	−11.26	−4.38	−6.88(1.08)**	−8.26	−3.57	−4.69(1.65)**

Note: Median Home Value and Median Rent were transformed using a natural log.

⁺p<.10;

^*p<.05;

^**p<.01. The treated and control columns contain point estimates for the outcome variable between treated and control census tracts. For example, 0 years after HOPE VI redevelopment, Vacancy Rates were estimated to be 0.41 in HOPE VI tracts and 2.05 in control tracts, resulting in a difference-in-difference estimate of −1.64, which was statistically significant at the .05 level.

HOPE VI redevelopment also led to significant decreases in vacancy rates and age of housing relative to control tracts, with effects persisting 10 years post redevelopment. Given similar housing characteristics at baseline, HOPE VI tracts experienced a 1.64 percent greater decline in vacancy rates relative to control tracts immediately following redevelopment, a 2.84 percent decline two years out, a 2.88 percent decline five years out, and a 3.17 decline 10 years out. Additionally, the median age of housing in HOPE VI tracts decreased 4.43 years more than control tracts upon completion of development, a gap that grew to nearly 7 years difference five years post-redevelopment. By ten years out, the median age of housing in HOPE VI tracts was 4.69 years lower than the median age of housing in control tracts. In contrast, there were no significant effects of HOPE VI redevelopment on homeownership rates.

Because difference-in-difference estimates compare change over time for HOPE VI tracts versus control tracts, it is useful to consider results alongside general time trends, which tap into broader patterns of urban neighbourhood change. While our analytic technique does not test for significant changes in housing characteristics over time, time trend point estimates (presented for treated and control tracts in Table 3) show the direction of change for HOPE VI and control tracts. Results suggest that, adjusted for inflation, median home values and median rents were generally increasing over time for HOPE VI and control tracts, with HOPE VI prompting elevated increases. Conversely, age of housing appears to have been decreasing in both HOPE VI and control tracts over time, with HOPE VI leading to more dramatic decreases. Interpretation is slightly more complex in relation to vacancy rates. Immediately following redevelopment, vacancy rates appear to have increased in both HOPE VI and control tracts from baseline, with redevelopment slowing the increase for HOPE VI tracts. By two years post-redevelopment, however, vacancy rates began declining in HOPE VI tracts relative to baseline, while they were still increasing in control tracts.

Variation in effects based on tract and redevelopment characteristics

To consider our second research question, we compared effects of HOPE VI redevelopment across subsets of census tracts. Model results are presented in Tables 4–7.

Table 4.

Flexible conditional difference-in-differences model results assessing HOPE VI effects in lower (<40%) versus higher (≥ 40%) poverty tracts.

		0 years after end	2 years after end	5 years after end	10 years after end
		(n=123–312)	(n=116–305)	(n=105–264)	(n=45–118)
		Estimate (SE)	Estimate (SE)	Estimate (SE)	Estimate (SE)
Median Home Value	Low Poverty	0.02 (0.03)	0.01 (0.03)	0.07 (0.03) *	0.04 (0.05)
	High Poverty	0.12 (0.06) *	0.17 (0.06) **	0.15 (0.06) *	−0.01 (0.08)
Median Rent	Low Poverty	0.04 (0.02) +	0.05 (0.02) *	0.04 (0.03) +	−0.01 (0.03)
	High Poverty	0.10 (0.04) *	0.12 (0.04) **	0.11 (0.04) *	0.10 (0.05) *
Ownership Rate	Low Poverty	−0.17 (0.84)	−0.28 (1.04)	−1.07 (1.14)	3.20 (1.77) +
	High Poverty	0.82 (1.03)	−0.12 (1.16)	−0.48 (1.24)	−0.47 (1.6)
Vacancy Rate	Low Poverty	−1.76 (0.56) **	−2.14 (0.67) **	−1.92 (0.87) *	−2.41 (1.19) *
	High Poverty	−1.48 (1.09)	−3.60 (1.17) **	−3.95 (1.16) **	−3.29 (1.78) +
Median Structure Age	Low Poverty	−3.01 (1) **	−3.87 (1.15) **	−4.33 (1.34) **	−3.03 (1.9)
	High Poverty	−5.39 (1.47) **	−9.84 (1.51) **	−9.67 (1.78) **	−6.18 (2.7) *

Note: Median Home Value and Median Rent were transformed using a natural log.

⁺p<.10;

^*p<.05;

^**p<.01.

Table 7.

Flexible conditional difference-in-differences model results assessing HOPE VI effects in tracts with affordable versus mixed-income public housing.

		0 years after end	2 years after end	5 years after end	10 years after end
		(n=68–396)	(n=66–379)	(n=62–328)	(n=20–164)
		Estimate (SE)	Estimate (SE)	Estimate (SE)	Estimate (SE)
Median Home Value	Affordable	0.04 (0.04)	0.06 (0.03) +	0.09 (0.04) *	0.01 (0.05)
	Mixed-Income	0.14 (0.05) *	0.16 (0.06) **	0.12 (0.05) *	0.01 (0.11)
Median Rent	Affordable	0.04 (0.02) *	0.04 (0.02) +	0.02 (0.03)	0.05 (0.03)
	Mixed-Income	0.12 (0.05) *	0.14 (0.05) **	0.15 (0.05) **	0.02 (0.07)
Ownership Rate	Affordable	0.83 (0.76)	0.51 (0.89)	−0.68 (1.07)	0.13 (1.32)
	Mixed-Income	−0.54 (1.4)	−1.07 (1.5)	−1.05 (1.69)	4.45 (2.92)
Vacancy Rate	Affordable	−1.83 (0.59) **	−2.55 (0.7) **	−2.73 (0.8) **	−3.18 (1.08) **
	Mixed-Income	−1.49 (1.07)	−3.52 (1.12) **	−3.19 (1.3) *	−5.13 (2.38) *
Median Structure Age	Affordable	−3.72 (0.91) **	−4.88 (0.98) **	−4.53 (1.08) **	−4.42 (1.57) **
	Mixed-Income	−5.48 (1.98) **	−10.43 (2.04) **	−11.61 (2.54) **	−5.41 (4.57)

Note: Median Home Value and Median Rent were transformed using a natural log.

⁺p<.10;

^*p<.05;

^**p<.01.

Variation by neighbourhood poverty rate

We first considered whether the effects of HOPE VI varied by the level of poverty present in neighbourhoods two years prior to redevelopment. Results indicate that effects of redevelopment on median home values, median rent, and median structure age were largely driven by effects in high poverty neighbourhoods (greater than or equal to 40%). In relation to median home values, high poverty neighbourhoods that received HOPE VI redevelopment grants had home values 0.12–0.17 log units higher than control tracts from zero to five years post-redevelopment. In contrast, low/moderate poverty neighbourhoods that received HOPE VI redevelopment grants experienced significant effects of redevelopment only five years after completion, and effects were half the size of those that emerged in high poverty tracts. In terms of median rent, high poverty neighbourhoods that received HOPE VI redevelopment grants had rents 0.10 to 0.12 log units higher than control tracts in all years considered, whereas low/moderate poverty neighbourhoods that received HOPE grants had significantly higher rent than control tracts only two years after redevelopment. As with median home values, effects on rent were roughly twice as large in high versus low/moderate poverty neighbourhoods.

While effects of HOPE VI redevelopment on median age of housing were present in both high and low/moderate poverty neighbourhoods across most years, effects were stronger in high poverty neighbourhoods. Two years after redevelopment, for instance, the median age of housing in low/moderate poverty HOPE VI tracts was 3.87 years lower than that of control tracts, whereas in high poverty HOPE VI tracts, the median age of housing was 9.84 years lower than that of control tracts. In short, median structure age declined more drastically for HOPE VI tracts that started with higher levels of poverty. Patterns were less consistent for vacancy rates.

Variation by neighbourhood racial composition

We next considered how patterns varied for neighbourhoods that had higher versus lower concentrations of Black residents prior to redevelopment. Neighbourhoods in which at least 70% of residents were Black were considered highly concentrated Black neighbourhoods. Findings were parallel to findings for poverty rate. Effects of HOPE VI redevelopment on median home values and median rent were driven mainly by neighbourhoods with high concentrations of Black residents prior to redevelopment. These predominantly Black neighbourhoods that received HOPE VI redevelopment grants had median home values 0.10 to 0.14 log units higher than control tracts up to five years after redevelopment, as well as median rents 0.10 to 0.12 log units higher than control tracts over the same time frame. In contrast, neighbourhoods with a low-to-moderate concentration of Black residents that received HOPE VI grants had home values and median rents that were not statistically different from control tracts in most years considered.

Effects of HOPE VI redevelopment on vacancy rates and median age of housing were present in all neighbourhoods at most time points, regardless of pre-redevelopment racial composition. The size of impact was slightly larger in neighbourhoods with a high concentration of Black residents in nearly all time periods.

Variation by neighbourhood vacancy rate

Models were also estimated to consider whether patterns differed for neighbourhoods that had higher versus lower vacancy rates prior to redevelopment. Results show that effects of HOPE VI redevelopment on median home values, median rent, and vacancy rates were largely driven by neighbourhoods with high vacancy rates (≥13%) prior to redevelopment. For both median home values and median rent, significant effects of HOPE VI redevelopment were present only for neighbourhoods that had high vacancy rates prior to redevelopment, with effects emerging zero, two, and five years post-redevelopment, but dissipating ten years out. Neighbourhoods with high vacancy rates pre-redevelopment also saw elevated effects on vacancy rates. After redevelopment, these neighbourhoods had vacancy rates 2.19% to 7.03% lower than control tracts in all years considered, with effects growing in magnitude over time. In contrast, neighbourhoods that had comparatively low vacancy rates at matching had significantly lower vacancy rates only two- and five-years post-redevelopment, and the magnitude of these differences was much smaller, with HOPE VI tracts seeing vacancy rates just 1.30% to 1.43% lower than control tracts. In contrast, effects of redevelopment on median structure age did not follow a clear pattern. Overall, except for age of housing, effects of HOPE VI appear to have been stronger in neighbourhoods with higher initial vacancy rates.

Variation by redevelopment type

Our final moderation analyses considered whether HOPE VI treatment effects varied by redevelopment type, comparing mixed-income redevelopments that introduced market rate units to those that retained only subsidized and affordable units. Results suggest that HOPE VI effects on median home values and median rents were driven largely by mixed-income redevelopments. Tracts wherein HOPE VI redevelopment transitioned to mixed-income had rents and home values that were significantly higher than matched control tracts zero, two, and five years after redevelopment. In contrast, tracts wherein HOPE VI redevelopment remained fully affordable saw significantly higher home values and median rents than control tracts at only one time point, with smaller effects. Similarly, effects on age of housing were stronger in HOPE VI sites that redeveloped into mixed-income developments, whereas effects on vacancy rates were relatively consistent across redevelopments that did and did not transition to mixed-income.

In sum, effects of HOPE VI redevelopment on neighborhood housing conditions were generally strongest and most persistent in neighbourhoods with a poverty rate exceeding 40% prior to redevelopment, neighbourhoods with more than 70% Black residents prior to redevelopment, neighbourhoods with vacancy rates exceeding 13% prior to redevelopment, and neighbourhoods in which mixed-income redevelopment was undertaken. Low-to-modest tetrachoric correlations between each of these indicators (ranging from a low of 0.16 between redevelopment type and vacancy rate to a high of 0.60 between poverty rate and percent Black residents) indicates that while these neighbourhoods overlapped to some degree, each analysis captured a unique constellation of neighborhoods. As such, our results suggest that each of these factors - poverty rate, racial isolation, vacancy rate, and redevelopment type - uniquely influenced the degree to which HOPE VI reshaped the neighbourhood housing context.

Sensitivity analyses

Several sensitivity analyses were undertaken to evaluate the robustness of our primary findings. While our main models matched on all covariates and outcomes two years prior to redevelopment, they did not account for parallel trends prior to grant receipt. It is possible that HOPE VI tracts were already on a different trajectory of change than control tracts prior to grant receipt, which could produce spurious effects. To account for this possibility, we estimated an alternative set of models that accounted for parallel trends within the flexpaneldid framework. This allowed us to compare HOPE VI tracts to control tracts that experienced similar changes in the given outcome variable from seven to two years prior to redevelopment. While more rigorous than our primary models, this approach resulted in reductions in our analytic sample. Because our first year of data was from 1990, all tracts that received a HOPE VI grant prior to 1997 were dropped from the sample, as they did not have sufficient data for parallel trends analyses. Additionally, accounting for parallel trends resulted in a smaller sample of potential control cases. As such, we elected to present these as a sensitivity check, rather than as our primary models.

Results (presented in Table 8) are parallel to but generally stronger than those from our primary models, despite the aforementioned reductions in sample size. As with our main models, significant effects of HOPE VI redevelopment emerged for median home values, median rent, vacancy rates, and age of housing. Effects on vacancy rates and age of housing were larger in magnitude at each time point when accounting for parallel trends. Effects on home values were similar in magnitude to those in our primary models zero to five years post-redevelopment, but were substantially stronger ten years out (0.13 log units in parallel trends models, versus 0.02 log units in primary models). In contrast, effects on median rent were similar to our primary models zero and two years post redevelopment, but weakened slightly five years post-redevelopment, falling to just under statistical significance. Considered holistically, results provide compelling evidence that the effects of HOPE VI redevelopment observed in our primary models were not driven by uncontrolled parallel trends, and in fact may underestimate the true effects of HOPE VI on neighbourhood housing conditions.

Table 8.

Alternative model: Flexible conditional difference-in-differences model results assessing effects of HOPE VI on housing, controlling for parallel trends from 7 to 2 years prior to redevelopment

	0 years after end	2 years after end	5 years after end	10 years after end
	(n=278–574)	(n=266–556)	(n=233–480)	(n=103–219)
	Estimate (SE)	Estimate (SE)	Estimate (SE)	Estimate (SE)
Median Home Value	0.07 (0.03) *	0.08 (0.03) **	0.10 (0.04) **	0.13 (0.06) *
Median Rent	0.09 (0.02) **	0.07 (0.02) **	0.05 (0.03) +	−0.01 (0.04)
Ownership Rate	0.48 (0.74)	0.15 (0.86)	−0.45 (0.93)	0.9 (1.66)
Vacancy Rate	−1.68 (0.57) **	−2.99 (0.74) **	−3.16 (0.83) **	−3.88 (1.06) **
Median Structure Age	−5.05 (0.86) **	−7.31 (1.07) **	−8.32 (1.28) **	−4.75 (1.7) **

Note: Median Home Value and Median Rent were transformed using a natural log.

⁺p<.10;

^*p<.05;

^**p<.01.

Recognizing that HOPE VI revitalization grants constituted one of several mechanisms through which the federal government reshaped public housing between 1990 and 2016, we ran a second set of sensitivity analyses to differentiate tracts that were redeveloped via HOPE VI revitalization grants from tracts that were redeveloped through other mechanisms. Specifically, using data from the NIMC Database, we were able to drop census tracts that received HOPE VI demolition grants and Choice Neighbourhoods grants from our pool of potential control cases. Excluding these cases from our pool of control tracts generally strengthened our findings, despite slight reductions in our analytic sample (results available upon request). This suggests that our primary models may present lower-bound estimates of the effects of HOPE VI redevelopment, as select control tracts were undergoing redevelopment through alternate mechanisms.

Discussion

Public and social housing schemes across the globe operate under conditions of austerity, and these conditions have driven a turn to mixed-income models that leverage the resources of for-profit housing developers and/or higher income residents to improve housing conditions in areas of concentrated poverty (Bolt et al., 2010). Despite research suggesting that mixed-income models have shown varied success in terms of their broader neighbourhood revitalization goals, such goals have been touted by advocates who seek to garner the funding and political support necessary to address deteriorated public housing (Hochstenbach, 2017; Vale, 2015). HOPE VI, one such program in the U.S., operated from 1993 to 2010 and has served as a model of this approach for many subsequent housing policies.

HOPE VI had ambitious goals, aiming to support economic integration of low-income public housing residents with higher income residents and to spur investment in surrounding neighbourhoods. They sought to do this by improving the physical design of public housing to blend better with and revitalize housing in surrounding neighbourhoods, deconcentrating poverty, and promoting local economic development (McCarty, 2012). We considered HOPE VI’s neighbourhood revitalization-related goals, specifically seeking to understand whether and how HOPE VI affected the housing context in neighbourhoods (operationalized as census tracts) within which HOPE VI sites were located. We used a new analytic approach that addresses limitations of the prior research, matching HOPE VI neighbourhoods with control neighbourhoods that contained public housing developments but did not receive HOPE VI revitalization grants, and then estimating impacts of HOPE VI across different time spans. This analytic technique allowed us to address two key limitations in prior analytic techniques assessing effects of HOPE VI - variation in the start time and length of redevelopment across HOPE VI sites and non-random selection of HOPE VI sites - and to provide results generalizable to HOPE VI sites across the U.S.

HOPE VI Effects on neighborhood housing contexts

Our findings provide compelling evidence that HOPE VI redevelopment affected the housing context surrounding public housing developments. According to DID estimates, HOPE VI redevelopment led to increased housing costs and rents, decreased vacancy rates, and decreased age of housing in HOPE VI tracts relative to control tracts, but did not alter homeownership rates. These results extend the findings of prior descriptive and quasi-experimental studies focused on smaller and more regional samples which found that HOPE VI redevelopment drove the development of new housing units (Baird, et al., 2020) and was associated with significant increases in residential sale prices and rents (Baird et al., 2020; Castells, 2010), and that, in some cases, increases in home values were sustained over time (Brown, 2009). Our findings also extend prior research on rental prices, which found that for “gold standard” mixed-income HOPE VI sites, surrounding rents generally increased (Turbov et al., 2005; Baird, 2020). Interestingly, our findings that HOPE VI redevelopment caused decreases in median age but did not cause increases in home ownership suggests that the new development that came in the wake of HOPE VI investments focused on rental housing.

In this study, HOPE VI redevelopment caused a short-term bump in residential home values and rental prices, which was expected given that such redevelopments removed seriously dilapidated housing, brought the promise of reinvestment, and, at least temporarily, removed public housing units. Prior research suggests that HOPE VI neighbourhoods in some promising urban housing markets represented a comparative bargain and a way for people to get into the housing market in affordable, yet up-and-coming urban neighbourhoods, pushing up property values and rental demand (Baird et al., 2020; Zielenbach, 2003a). These studies are also reflective of findings from qualitative and descriptive literature on HOPE VI that suggests that key stakeholders including residents and housing service providers observed generally positive community changes following HOPE VI redevelopment, including reduced crime and disorder and improved physical conditions, that made neighbourhoods more desirable and in turn stimulated interest in the local real estate market (Curley, 2010; Holin, 2003).

We found no significant effects of HOPE VI on homeownership rates, which is somewhat surprising given some compelling, albeit limited, evidence of increasing homeownership in prior literature. For example, prior studies of HOPE VI communities found increases in residential and commercial lending (Zielenbach, 2003a), increases in mortgage lending and increases in percent of owner-occupied homes (GAO, 2003; Holin, 2003) in many, but not all of the HOPE VI communities assessed. Further, although much HOPE VI development was focused on rental units (approximately 85% of units produced were for rental), about 40% of HOPE VI developments provided at least some new opportunities for home ownership (Gress et al., 2016). The absence of spillover effects of HOPE VI onto neighbourhood-level homeownership rates in our study suggests that rising homeownership was not a consistent effect of HOPE VI redevelopment across sites and over time, or perhaps that such changes were driven by selection effects rather than effects of HOPE VI policies.

We also found that HOPE VI led to declines in vacancy rates, with effects growing over time following completion of HOPE VI redevelopment. This aligns with some prior work; for example, Holin and colleagues (2003) found that vacancy rates decreased in 9 of 12 HOPE VI neighbourhoods examined. However, it contrasts with other descriptive research that found that HOPE VI neighbourhoods had above average vacancy rates compared to similar non-HOPE VI tracts (Zielenbach, 2003a). This may be reflective of the phase of HOPE VI redevelopment included in each study. Whereas the present work assessed effects through and following completion of redevelopment, Zielenbach and colleagues (2010) assessed effects in 2000 and 2010, when developments were largely completed but not in all cases. One interpretation is that vacancy rates were initially inflated in HOPE VI tracts due to the relocation process prior to demolition, leaving units empty during the redevelopment process. The subsequent decrease in vacancy could be explained by reoccupation of new units, or by the fact that redevelopment signals that a neighbourhood is in-demand (Ding et al., 2000).

While these changes can be viewed as a sign of successful policy implementation, the international literature has raised concerns about the impact of redevelopment of social housing on rents and home ownership opportunities. In some policy efforts in Europe, for example, the restructuring of social housing focused heavily on disadvantaged communities, with some concerns that such policies led to a reduction in affordable rental units and a rise in gentrification in cities like Amsterdam (Hochstenbach, 2017). This increased neighbourhood desirability that is thought to accompany an influx of government and leveraged funds also raises concerns about the ability of non-subsidized low-income renters to weather the increased rental prices in the neighbourhood (Baird et al., 2020). The present research illustrates that HOPE VI contributed to increased rent and housing values as well as lowered vacancy rates and reduced age of housing, changes which have potential to create benefits and pitfalls for current and potential residents.

Variation in housing impacts across neighbourhood and development context

A second aim of this study was to understand whether the effects of HOPE VI on neighbourhood housing contexts varied systematically across neighbourhood and public housing characteristics. Our results suggest that impacts on housing-related indicators were strongest in neighbourhoods that were the most disadvantaged at the time of grant receipt in terms of concentrated poverty, racial isolation, and vacancy rates. Effects were also driven by developments that involved market rate housing, thus adhering to the “gold standard” HOPE VI mixed-income model of redevelopment.

Effects on home values and median rent observed in the full sample of HOPE VI sites were driven by neighbourhoods that started with heightened poverty and high levels of Black racial isolation, wherein effects were generally larger in size and longer-lasting than in other neighbourhoods. Effects of HOPE VI on home values and rents were also driven largely by redevelopments that were mixed-income. The rise in these housing indicators is likely reflective of a combination of dispersal of low-income residents in the early stages, increased perceived desirability, and increased income mix of the neighbourhood (Greene, 2008). Indeed, prior research has found that HOPE VI led to decreased neighbourhood poverty and increased median household incomes, but did not alter the proportion of wealthy residents, suggesting that economic shifts were due to greater integration of middle-income neighbors rather than to gentrification by high-income residents (Coley et al., 2022). Still, there remains some uncertainty regarding how much these shifts were due to displacement of the most disadvantaged residents versus to increasing economic integration and desirability of the neighbourhoods of HOPE VI sites (Greene, 2008; McCarty, 2012).

Limitations and future directions

While this research has many strengths, it is not without limitations. First among these is the possibility that some unmeasured factor may differentiate HOPE VI tracts from matched controls, biasing results. Although the flexpaneldid approach allows for sophisticated matching and provides more accurate estimates of HOPE VI effects than most prior work, and we found that results were even more robust when we matched on parallel trends rather than point-in-time differences in HOPE VI and control neighborhoods, the possibility of unexamined third variables cannot be eliminated totally without random assignment.

Another key issue to consider is the whether the quality and value of housing shifted in HOPE VI neighborhoods. Our results finding that HOPE VI led to new housing units, increased housing prices, and lowered vacancy rates suggest increased housing availability and desirability, yet it remains unclear whether increased prices and occupancy reflect improved housing quality or simply greater competition. These are important issues for future research to assess.

Finally, it is important to note that it remains unclear whether HOPE VI-related shifts in neighbourhood housing conditions are wholly beneficial for residents and communities. Housing indicators act as proxies for many things, including costs of living, level of investment, availability and competition, and desirability. Given the extreme disinvestment that public housing and surrounding neighbourhoods faced in the decades prior to HOPE VI, changes like rising home values and declining vacancy rates may be experienced within the community as positive, contributing to an improved physical environment, new resources, and potentially reductions in neighbourhood stigma (Joseph, 2006). However, prior research on HOPE VI found that the policy did not lead to increases in neighbourhood institutional resources (such as medical and educational services), social services, or basic amenities like restaurants and grocery stores (Coley et al., 2022). Moreover, there may be a tipping point at which the benefits of redevelopment for low-income residents are offset by rising costs, increasing housing demand, and cultural shifts within the neighbourhood. While a plethora of research documents these challenges (Lees, 2008), more work is needed to identify this tipping point, and to explore strategies for mitigating direct and indirect forms of displacement (Davidson, 2008). Although our results address changes at the neighbourhood level, national data limitations prohibit an assessment of the experiences of individual residents and the benefits or detriments to their well-being caused by HOPE VI.

Conclusion

A primary goal of HOPE VI was to spur revitalization beyond the immediate public housing development, creating positive spillover into the surrounding neighbourhood (this was, in fact, one of the ranking criteria for awarding grants) (Greene, 2008). Decision-makers hoped that the improved housing stock and increased social capital and purchasing power of mixed-income communities would help spur the local market, bringing amenities and housing revitalization along with improved social conditions and property values (Joseph et al., 2007; Zielenbach, 2003a). However, HOPE VI’s revitalization goals have been critiqued, with some suggesting that the policy did not set out with primarily good intentions but, instead, aimed to leverage private developers’ interest in prime city development sites at the expense of existing low-income residents (Vale & Shamsuddin, 2018).

This research provides compelling evidence that HOPE VI changed the neighbourhood housing context surrounding public housing developments. Effects were felt most strongly in communities that were the most economically disadvantaged, racially isolated, and lowest demand when redevelopment began, and where HOPE VI redevelopment involved the introduction of market rate units. Along with prior results showing decreases in concentrated poverty and increased median incomes in HOPE VI communities (Coley et al., 2022), results indicate that the HOPE VI program accomplished several of its aims, contributing to the deconcentration of poverty and revitalization of neighbourhood housing contexts. With an increased focus on prompting neighborhood change, new generations of public housing redevelopment policies seek to carry these goals forward.

Table 5.

Flexible conditional difference-in-differences model results assessing HOPE VI effects in lower (<70%) versus higher (≥70%) percent Black tracts.

		0 years after end	2 years after end	5 years after end	10 years after end
		(n=115–298)	(n=105–295)	(n=88–257)	(n=32–122)
		Estimate (SE)	Estimate (SE)	Estimate (SE)	Estimate (SE)
Median Home Value	<70% Black	0.06 (0.03) +	0.05 (0.04)	0.08 (0.04) *	0.04 (0.05)
	>70% Black	0.10 (0.04) *	0.14 (0.05) **	0.14 (0.05) *	−0.01 (0.08)
Median Rent	<70% Black	0.02 (0.02)	0.04 (0.03)	0.02 (0.03)	0.00 (0.04)
	>70% Black	0.12 (0.03) **	0.11 (0.04) **	0.10 (0.04) *	0.08 (0.05) +
Ownership Rate	<70% Black	0.22 (0.85)	−0.22 (1.05)	−1.41 (1.26)	1.8 (1.69)
	>70% Black	1.14 (0.99)	0.20 (1.13)	0.05 (1.24)	1.23 (1.77)
Vacancy Rate	<70% Black	−1.35 (0.57) *	−2.02 (0.74) **	−2.37 (0.84) **	−1.99 (1.01) +
	>70% Black	−2.08 (0.86) *	−3.74 (0.91) **	−3.30 (1.07) **	−4.87 (1.8) **
Median Structure Age	<70% Black	−3.84 (1.07) **	−5.57 (1.27) **	−6.39 (1.55) **	−5.97 (2.39) *
	>70% Black	−5.60 (1.24) **	−8.44 (1.37) **	−7.55 (1.57) **	−5.14 (2.17) *

Note: Median Home Value and Median Rent were transformed using a natural log.

⁺p<.10;

^*p<.05;

^**p<.01.

Table 6.

Flexible conditional difference-in-differences model results assessing HOPE VI effects in lower (< 13%) versus higher (≥13%) vacancy tracts.

		0 years after end	2 years after end	5 years after end	10 years after end
		(n=139–295)	(n=131–289)	(n=117–254)	(n=50–120)
		Estimate (SE)	Estimate (SE)	Estimate (SE)	Estimate (SE)
Median Home Value	Vacancy rate <13%	0.03 (0.04)	0.05 (0.03)	0.05 (0.04)	0.00 (0.06)
	Vacancy rate >13%	0.09 (0.04) *	0.11 (0.05) *	0.15 (0.06) *	0.04 (0.08)
Median Rent	Vacancy rate <13%	0.05 (0.03)	0.05 (0.03)	0.02 (0.03)	0.00 (0.04)
	Vacancy rate >13%	0.08 (0.03) **	0.10 (0.03) **	0.11 (0.04) **	0.06 (0.05)
Ownership Rate	Vacancy rate <13%	1.26 (0.94)	1.01 (0.99)	−0.48 (1.1)	−0.07 (1.52)
	Vacancy rate >13%	−0.49 (1.15)	−1.82 (1.13)	−1.82 (1.34)	2.32 (2.01)
Vacancy Rate	Vacancy rate <13%	−0.75 (0.55)	−1.30 (0.6) *	−1.43 (0.68) *	−0.60 (0.99)
	Vacancy rate >13%	−2.19 (0.89) *	−3.90 (0.96) **	−4.09 (1.12) **	−7.03 (1.68) **
Median Structure Age	Vacancy rate <13%	−5.01 (1.09) **	−5.80 (1.25) **	−5.88 (1.43) **	−5.66 (2.28) *
	Vacancy rate >13%	−3.62 (1.35) **	−7.49 (1.54) **	−7.77 (1.67) **	−3.71 (2.35)

Note: Median Home Value and Median Rent were transformed using a natural log.

⁺p<.10;

^*p<.05;

^**p<.01.

Biographies

Samantha Teixeira is an Associate Professor at the Boston College School of Social Work. She holds a PhD in social work and her current research focuses on how neighborhood and housing environments shapes health disparities, with an emphasis on community-based research approaches.

Bryn Spielvogel is a Postdoctoral Researcher at the University of Utah. Her research is broadly focused on how the contexts that people live in and interact with shape their development and wellbeing.

Dabin Hwang is a Research Implementation Specialist at the Connell School of Nursing at Boston College. His research focuses on how inequities in children’s developmental contexts, including schools, neighborhoods, and housing contributes to disparities in educational outcomes.

Josh Lown is a Postdoctoral Research Fellow at Northeastern University. His research uses community-based and participatory approaches to understand the intersection of gentrification and residential experiences of social control and chronic urban trauma.

Rebekah Levine Coley is a Professor of Applied Developmental and Educational Psychology and Director of the Institute of Early Childhood Policy at Boston College. Her work focuses on identifying and interrupting economic and racial disparities in development and wellbeing.