Document View

This study tests the hypothesis that the acquisition of existing property by the public housing authority and its subsequent rehabilitation and occupancy by subsidized tenants significantly reduced the property values of surrounding single-family homes in Denver during the 1990s.

INTRODUCTION

Policymakers have long harbored concerns over the location of low-income households who receive housing assistance. Traditionally, these concerns have been articulated in several themes related to social problems. Some have worried that if subsidized households are concentrated in a neighborhood, a variety of social maladies-violence, crime, substance abuse, alienation, out-of-wedlock childbearing-will be intensified disproportionately (Coulton and Pandey, 1992; Crane, 1991; Polikoff, 1994; Quercia and Galster, 1998). Others believe that concentrating subsidized households facilitates their stigmatization and the withdrawal of private and public capital from their neighborhood (Leavitt and Loukaitou-Sideris, 1995; Massey and Kanaiapuni, 1995). Still others see a social cost in the form of perpetuated racial and ethnic segregation and isolation (Goering and Couilibably, 1989; Massey and Denton, 1993; Massey, Gross, and Eggers, 1991).

Recently, however, the locational issue has been framed more positively. Momentum has grown for structuring housing subsidy programs to give low-income households more spatial options than they have had previously. This enrichment of residential alternatives not only improves the freedom and well-being of recipients in the short run, but, by enhancing their access to employment and job information networks and better-quality education, also increases their prospects for economic selfsufficiency in the long run. It also exposes them to community social norms more supportive of education and employment (Cisneros, 1995; Polikoff, 1994; Rosenbaum, 1995; Rosenbaum, Deluca, and Miller, 1999).

Nonetheless, efforts to disperse tenants receiving housing subsidies outside concentrated poverty neighborhoods must confront a common challenge: potential hostility from communities into which subsidized households move (Goetz, Lam, and Heitlinger, 1996; Hogan, 1996, ch. 3; Schill, 1992). Although policymakers sometimes dismiss such hostility founded on indefensible racial or class prejudices, public policy concerns that the introduction of subsidized tenants may seriously harm the quality of life in a neighborhood may be legitimate. These concerns often focus on possible erosion of civil behavior, upsurge in crime and violence, accelerated physical decay, and a consequent drop in property values. As such, subsidized housing may be considered another type of locally unwanted land use, an issue upon which considerable literature has focused (e.g., Freudenberg and Pastor, 1992; Graham and Logan, 1990; Lake, 1993; Takahashi and Dear, 1997).

These fears manifested themselves in the form of local resident protest in Denver, Colorado, in 1989, where the public housing authority's plans to acquire additional single-family and small multi-family properties for use as scattered-site public housing resulted in the City Council's imposing siting restrictions. This study examines the issues that arose around the Dispersed Housing Program in Denver and tests the hypothesis that the acquisition of existing property by the Housing Authority of the City and County of Denver (DHA) and its subsequent rehabilitation and occupancy by DHA tenants significantly reduced the sales prices of surrounding single-family homes during the 1990s.

The authors employed a mixed-method approach, combining econometric and qualitative analyses. Specifically, multiple regression analyses were performed to ascertain whether the sales prices of single-family homes were adversely affected by proximity to dispersed public housing units. In addition, focus groups were conducted with homeowners from a variety of neighborhoods in which dispersed housing was located, to collect qualitative information on the possible interactions among housing prices, neighborhood quality, and the dispersed sites.

PREVIOUS EMPIRICAL STUDIES: RESULTS AND METHODOLOGICAL SHORTCOMINGS

Through the 1980s, at least a dozen scholarly studies had investigated the question of whether subsidized housing generates a negative effect on neighboring single-family property values.1 The preponderant conclusion reached by these studies was that there was no sizable or statistically significant impact, while a few studies (De Salvo, 1974; Nourse, 1963; Warren, Aduddell, and Tatalovich, 1983) even concluded that there was a positive effect.2

Recently, however, the conventional wisdom of no impact has been shaken by four, sophisticated statistical studies that have emphasized the contextuality of impacts (Cummings and Landis, 1993; Goetz, Lam, and Heitlinger, 1996; Lee, Culhane, and Wachter, forthcoming; Lyons and Loverage, 1993). These studies have concluded that, in certain circumstances and certain kinds of developments, subsidized housing can create harmful effects on nearby property values.3

One possible explanation as to why the forgoing analyses have come to such variant, non-generalizable conclusions is because they employ different methodologies, each of which suffers from serious, if somewhat different, shortcomings.4 Prior studies share two crucial weaknesses. The first is that they cannot convincingly distinguish the direction of causation between trends in neighborhood property values and the siting of assisted housing.5 Put differently, because they do not control for the quality and market strength of the micro-neighborhood into which assisted housing is placed relative to the larger universe of potential sites, they cannot ascertain whether subsidized sites lead to neighborhood decline or whether subsidized sites are systematically developed in areas having property values that are low and expected to depreciate in the future. One suspects that subsidized housing developers may select neighborhoods with low property values to preserve scare resources by acquiring the least-expensive properties available. A second, overarching criticism of all extant empirical work related to the property value effects of subsidized housing is the failure to account for spatial econometric issues (Can, 1997; Can and Megbolugbe, 1997).

The authors' econometric approach overcomes the shortcomings of prior approaches. By employing a "pre-/post-" design involving localized fixed effects before and after occupancy of a DHA dispersed site, it controls for microneighborhood characteristics unrelated to such occupancy. By relating these localized fixed effects to property value trends and levels in larger geographic areas it distinguishes the self-selection of dispersed housing into weak neighborhood submarkets from their ultimate consequences on these neighborhoods. By carefully controlling for the characteristics of sold properties, regional and seasonal trends in values, and spatial econometric effects, it purges several additional confounding elements that have plagued earlier analyses.

OVERVIEW OF THE DISPERSED PUBLIC HOUSING PROGRAM IN DENVER

According to Hogan (1996, ch. 3), national policies aimed at deconcentrating public housing emerged in the 1950s and 1960s in response to public concerns regarding the deleterious effects of large-scale public housing projects on inner-city neighborhoods. Increased criminal activity, concentrated poverty, increased social and spatial isolation of the poor, and the concomitant rise in negative behavior (i.e. high school desertion, out-of-wedlock childbearing, withdrawal from the labor force) previously have been attributed to the presence of large public housing projects. To address these purported social ills, policymakers advocated the development of low-density, geographically dispersed assisted housing, also known as scattered-site housing.

Denver was an early respondent to these national debates regarding public housing, and began operating a public housing "dispersal" program in 1969.6 The program initially involved 100 single-family and duplex units acquired at foreclosure sales, which were then renovated and occupied by DHA tenants.7 In 1988, HUD ordered DHA to publicly notify the Denver City Council about the site-by-site details and obtain its approval for any dispersed housing plans. Before that ruling, DHA housing acquisitions required only the Mayor's consent, accompanied by a description in general terms to the Council. When DHA proposed its second-phase dispersal plan in 1989, which involved purchase of more than 400 additional homes in middle-class neighborhoods, an inflammatory political skirmish erupted, centering on concerns of the local citizenry regarding the perceived deleterious effects of moving public housing residents into middle-class neighborhoods (Galster, 1989).8 In response to these concerns, a Council-appointed task force drafted a set of guidelines regulating further DHA acquisitions. These guidelines stipulated that the DHA could not acquire more than one unit per block face and no more than 1 percent of the units in any census tract. Moreover, DHA was to target the "non-impacted" areas of Denver for these purchases.9 The Denver City Council approved the plan and formulated an intergovernmental agreement, which has been in operation since 1989.

In the 1990s, DHA acquired approximately 500 units of dispersed housing, primarily purchasing existing homes with an emphasis on duplexes and smaller, multifamily buildings. To identify appropriate properties, staff at the DHA work with local realtors, scour MLS listings, and look for "For Sale" signs in neighborhoods where DHA would like to expand their dispersed housing stock. In sharp contrast to 1989, the most recent intergovernmental agreement for continuing this policy was reached with limited public fanfare.

DATA AND METHODS

In this study, a series of related data were used, including, archival materials, key informant interviews, statistical databases, and focus group discussions with homeowners living near the DHA dispersed sites being analyzed. Each of these data sources is described at length in Galster et al. (1999).

Data Sources for the Quantitative Analyses

Data on single-family home sales for Denver were obtained from a private data vendor, Experian, for the period beginning in the first quarter of 1987 through the third quarter of 1997. These included the street address of the house, the amount and date of the sale, and characteristics of the house (square footage, lot size, number of rooms, and type of construction).10 Data on conventional and dispersed public housing programs were obtained from the DHA. These included information on project address, year of construction, DHA acquisition date, and the number of units. In this study only the most recent "vintage" of DHA dispersed sites are investigated: the 167 that opened between the first quarter, 1989, through the third quarter, 1995. This permits matching a minimum of at least two years of sales data both before and after the first and last DHA dispersed sites in this vintage.

Both home sales files and the DHA site data files were geo-coded to match street addresses with latitude and longitude coordinates, Census geographic identifiers (i.e., state, county, tract, and block), and U.S. Postal Service ZIP+4 codes.11 Of these property addresses, 98 percent were geo-coded to an exact street address or to a ZIP+4 centroid.12 It was possible to geo-code 97 percent of the records to an exact street address and an additional 1 percent to a ZIP+4 centroid. Sales and DHA records that could not be geo-coded to at least this level of precision were excluded from the analysis.

Analysis Sites

Econometric analysis of property value effects were conducted on a subset of dispersed housing sites referred to as analysis sites. To operationalize the pre- and posteconometric specification described below, the analysis was restricted to those subsidized locations having: no DHA conventional or other dispersed sites within 2000 feet when it opened; and having an average annual rate of single-family homes sales of at least 2.0 in each of the concentric circle ranges of 0-500 feet, 501-1000 feet, and 1001-2000 feet both before and after first occupancy. These restrictions reduced the analysis sample size from 167 to 92 DHA dispersed sites.13

The subset of home sales to be used in the econometric analysis was chosen in relation to these analysis sites. All single-family home sales were used that either were not within 2000 feet of any occupied DHA conventional or dispersed site, or were not within 2000 feet of one (or more) of the analysis sites after occupancy. Sales that were within 2000 feet of any other occupied DHA site(s) but did not qualify as an analysis site(s) were also omitted. This yielded a sample of 43,361 sales and permitted unambiguous tests based on our pre and post principles of deciphering effects.

A Conceptual Framework for Analyzing the Determinants of House Prices

The conventional assumption is adopted that each house may be described as a package of various characteristics, which describe numerous attributes of the structure [S], neighborhood [N], and local public services [L]. Symbolically:

H = f([S], [N], [L]) (1)

where H can be thought of as the "quality" of that house or its "hedonic value" (Rothenberg et al., 1991, ch. 3). The price of the housing package is a function of its embodied quality:

P = g(H) (2)

The hedonic price function represented by Equation 2 represents, according to Rosen (1974), "a joint envelope of a family of 'value functions' [of sellers]." The partial derivative of P in Equation 2 with respect to a particular attribute of the house yields the implicit price of that attribute. Rosen suggests that if hedonic relationships in part reflect sellers' pricing strategies, there will be a problem in identifying household preferences. This is less of a concern in the case of housing because, as Muellbauer (1974) demonstrates, household preferences dominate "in second-hand durables markets where aggregate supply is fairly stable and particular supplies are usually held in decentralized fashion." Thus, the sign and magnitude of the implicit price can be interpreted as a measure of the degree to which households in the market prefer (or are averse to) that attribute (Muellbauer, 1974). Should proximity to a DHA dispersed site prove to have a negative implicit price in the estimated hedonic index, it would imply that such was imposing a social cost upon its neighbors. Based on previous work (Rothenberg et al., 1991, ch. 13), a semi-log form for Equation 1 is use, that is, it expresses the logarithm of sales price as a linear function of the house and neighborhood characteristics and other independent variables.

Localized Fixed Effects

As shown in Equations 1 and 2, the sales price of a home will be affected by numerous neighborhood attributes (including physical and occupancy characteristics of neighboring properties, environmental conditions, and potentially the proximity of subsidized tenant-occupied apartments) and attributes of the local public sector (including schools, police protection, taxes, and zoning). The challenge facing the analyst is to gather complete data on this array of neighborhood attributes so that results will not be biased by omitted variables. This challenge has two facets: not only must one gather a comprehensive, dauntingly large set of attributes, but one must also ascertain the geographic area over which these attributes are most appropriately measured for each site.

The approach chosen responds to this challenge by specifying a spatial fixed effects model. That is, dummy variables were specified that denote a particular geographic area ranging in scale from a census tract down to the area within 500 feet of a subsidized site. These variables control, in summary form, for the idiosyncratic bundle of attributes present in the corresponding space. The effect on sales price of individual attributes in this bundle cannot be determined, however.

This procedure is distinguished by its specification of the "neighborhood." A set of fixed-boundary, mutually exclusive areas (census tracts) were employed to define one set of spatial fixed effect variables. However, a different procedure was used to measure fixed effects in smaller, micro-neighborhoods. Essentially, a series of "neighborhoods" centered on each dispersed housing site was defined, each one comprising one of several concentric rings: 0-500 feet, 501-1001 feet, and 1001-2000 feet from a site.

Alternative Model Specifications and Their Assumptions

Three alternative specifications of the hedonic price function (Equation 1) were estimated that did not distinguish among subsidized sites. Each specification is founded on particular assumptions discussed as follows. These alternatives are expressed symbolically as:

Model 1 (proximity to any subsidized site model):

LnP = c + [Struct][b] + [Quarter][n] + [Tract][m] + dDAll500 + eDAll1k + fDAll2k + gDPost500 + hDPost1k + jDPost2k + qTime500 + rTime1k + sTime2k + tTrPost500 + uTrPost1k + vTrPost2k +(3)

Model 2 (proximity to number of subsidized sites interaction model):

LnP = c + [Struct][b] + [Quarter][n] + [Tract][m] + dDAll500 + eDAll1k + fDAll2k + gPost500 + hPost1k + jPost2k + qTime500 + rTime1k + sTime2k + tTrPost500 + uTrPost1k + vTrPost2k + t'(TrPost500 * Post500) + u'(TrPost1k * Post1k) + v'(TrPost2k * Post2k) +(4)

Model 3 (proximity to number of subsidized units interaction model):

LnP = c + [Struct][b] + [Quarter][n] + [Tract][m] + dDAll500 + eDAll1k + fDAll2k + gUPost500 + hUPost1k + jUPost2k + qTime500 + rTime1k + sTime2k + tTrPost500 + uTrPost1k + vTrPost2k + t'(TrPost500 * UPost500) + u'(TrPost1k * UPost1k) + v'(TrPost2k * UPost2k) +(5)

Where the components of the models are defined as:

LnP = Log of the single-family home sales price

c = Constant term

[Struct] = Vector of structural characteristics of home, including home and lot size, age, building materials and type, and numerous amenities; for details, see the Appendix; this controls for the home's size, quality, and amenities

[Quarter] = Vector of dummies indicating the time (year and quarter) of sale; this controls for regional and seasonal price trends

[Tract] = Vector of census tract dummies indicating the location of home; this controls for tract-scale neighborhood characteristics

Dpostx = Post-occupancy dummy for distance ring x; equals 1 if sale occurs within x feet of one or more occupied dispersed housing sites; zero otherwise; tests for change in level of prices in ring after any DHA developments sited there

Dallx = Dummy for distance ring x; equals 1 if sale occurs within x feet of current or future dispersed site, whether occupied or not; zero otherwise; controls for price level in ring before DHA development

Postx = Number of post-occupancy dispersed sites within distance ring x at time of home sale; tests for change in level of prices in ring after N number of DHA sites developed

Upostx = Number of occupied dispersed units for distance ring x at time of sale tests for change in level of prices in ring after N number of DHA units developed

TrPostx = Post-occupancy trend variable for distance ring x; equals 0 if sale is preoccupancy for all dispersed sites in distance ring; if sale is post-occupancy of a site in ring x, then equals 1 if sale occurs in first quarter after site was occupied, equals 2 if sale occurs in second quarter after site was occupied, etc.; tests for change in trend of prices in ring after any DHA sites developed in ring

Timex = Trend variable for distance ring x; equals 0 if no dispersed sites are in distance ring x of the sale; otherwise, equals 1 if sale occurs in first quarter of study period (i.e., 1st quarter 1987), equals 2 if sale occurs in second quarter of study period, and sale is in distance ring x, etc.; controls for price trend in ring before DHA development

- = A random error term with statistical properties as discussed below.

All lowercase letters in the equations represent coefficients to be estimated.

Model 1 tests both for price level shift and for price trend slope alteration effects in impact areas near dispersed sites, and thus makes relatively few assumptions about what form any effect might take. Model 1 permits us to identify unambiguously the effect of proximity to dispersed housing, purged of self-selection effects.

The set of variables [Quarter] measures quarterly changes in the overall county house price levels associated to seasonality and general market trends. The set [Tract] measures the fixed effect on house prices due to location in the area defined by the Census tract. Variable set Dallx measures the fixed effect throughout the county of being in the area defined as within distance x of one or more dispersed site(s), regardless of whether occupied yet, whereas Dpostx measures the fixed effect throughout the county of being in the area defined as within distance x of one or more dispersed site(s) after occupancy. Timex measures the trend in house prices during the study period in the area throughout the county defined as within distance x of one or more dispersed site(s), regardless of whether occupied yet, whereas TrPostx measures the trend in house prices during the study period in the area throughout the county defined as within distance x of one or more dispersed site(s) after occupancy.

The test for statistical significance of the post-occupancy shift coefficients (g, h, j) of the DPostx variables is equivalent to testing that there is a discontinuous change in the price levels in the micro-neighborhoods (defined by a particular distance ring) around dispersed units post-occupancy. The test for statistical significance of the post-occupancy trend coefficients (t, u, v) of TrPostx is equivalent to testing that there is a change in the price trends in the micro-neighborhoods around dispersed units post-occupancy. Should both the shift and trend post-occupancy coefficients prove not to be significantly different from zero, it would reject the hypothesis of effect. Should one or both be statistically significant, however, the magnitude of dispersed housing impact across all sites involves assessing whether (d+qTime*) - (g+tTrPost), (e+rTime*) - (h+uTrPost), or (f+sTime*) - (j+vTrPost) 0, where Time* represents the latest quarter prior to occupancy of the site by a dispersed household. Should the alterations in shift and trend terms yield contrary implications (such as a downward shift but increased slope in the price gradient), it will be necessary to calculate net effects at different quarters post-occupancy. In any event, the implicit comparison being made in these statistical tests is the price of a single-family home within a specified distance from an occupied DHA dispersed site at the time of sale to the price of an identical home within the same geographic area selling before any DHA dispersed sites have been occupied there (controlling for temporal and seasonal price differences).

Note that, regardless of the coefficient signs of the Dpost and TrPost variables, their implied impact post-DHA occupancy is measured against the baseline price level and trend for the corresponding geographic areas delineated by the Dall and Time variables. Thus, even if such areas prove to have systematically different price profiles than elsewhere in their surrounding census tracts (measured by the [Tract] variables), the post-occupancy effect can be discerned. This is the way in which this specification avoids the locational self-selection ambiguity that has plagued earlier models.

Models 2 and 3 build upon the foundation specification of Model 1 but differ from it in two important ways. Model 1 implicitly assumes that the measured effect of proximity to any dispersed site(s) is invariant to the number of such proximate sites. Model 2 relaxes this assumption and allows the post-occupancy shift variable to assume the number of occupied dispersed sites at the given distance at the time of sale. Model 3 does the same, but uses the number of occupied subsidized units instead of sites. The maximum number of sites in the three distance rings is 6, 11, and 16, and the corresponding maximum number of dispersed units is 26, 34, and 55. Models 2 and 3 also test for the possible effects of the number of sites or units on the post-occupancy price trends by use of the multiplicative interaction variables. That is, the model measures whether the decline (or appreciation) in house prices is magnified by the number of subsidized sites or units present.

Econometric and Data Issues

Before estimating the price effect models, highly idiosyncratic sales and those that did not represent arms-length transactions were excluded from the database. In this vein the top and bottom 2 percent of all sales according to sales price and land area were eliminated.14

Heteroskedasticity of the model error terms was tested for in the house price regressions by performing the Goldfeld-Quandt test (Pyndick and Rubinfeld, 1981). The F statistic produced by this test was barely statistically significant at the 10 percent level, indicating that heteroskedasticity was not likely present. Nevertheless, to be certain that consistent estimates of parameter standard errors had been employed, the White (1980) covariance matrix was used to correct the standard errors.

To test for spatial autocorrelation, a specification that Can and Megbolugbe (1997) found to be robust was employed. The spatial lag of the dependent variable (house price) was calculated and included in the model as an independent variable. The spatial lag is a weighted average of all sales prices of homes within a certain distance from the reference sale. The average is weighted by the spatial weight, which is some function of the distance between sales. Consistent with the approach of Can and Megbolugbe, the inverse of the distance (1/d) was used as the spatial weight. The formula for the spatial lag is:

SpLag (Pi) = -j(1/dij) Pj -j1/dij (6)

where Pi is the sale for which we are calculating the spatial lag, di j is the distance between sales i and j, and Pj is one of the set of all sales within distance D of Pi and that occurred within the six months before the date of Pi. Spatial lags were calculated with distance D of 2000, 5000 and 10,000 feet, to examine the possibility that spatial dependence may exist over a larger area. None of the spatial lag variables improved the model fit by any substantial amount and were excluded from the final models because of computational costs.

Spatial heterogeneity, sometimes known as spatial submarket segmentation, refers to whether the parameters of the hedonic price equation are invariant across space. If such were the case, the error term would be heteroskedastic. To deal with this issue we employed the "spatial contextual expansion with quadratic trend" specification as suggested by Can (1997). This method involves adding to the models above the latitude (X) and longitude (Y) coordinates of each observed home sale in the following variables (normalized so that zero values represent the center of Denver County): X, Y, XY, X2, and Y2. Higher numerical values of X (Y) signify increasing distance from the center of the county heading west (north). These variables typically proved statistically significant in aggregate specifications, suggesting that various controls for local fixed effects needed further supplementation from these spatial coordinates.

RESULTS

Locations of Denver Dispersed Housing Sites

The locations of the 1989-1995 vintage of DHA dispersed public housing sites across census tracts are shown in Figure 1, with the total sites and the analysis subset of sites noted. Each dot on this map represents a single site. One can see that, in general, the sites are fairly evenly spread throughout the city, with some small clusters in a few locations.15 Selected characteristics of these analysis sites are presented in Table 1.

The significance of this generally deconcentrated pattern of DHA dispersed housing sites should not be minimized. Through a combination of DHA choices and clustering restrictions imposed by Denver City Council after 1989, Denver now evinces a remarkably uniform distribution of dispersed public housing units across the majority of census tracts.16 Although the primary exceptions to the uniform pattern are in the predominately white-occupied areas in the south-central and eastern portion of Denver, these also correspond to areas with little rental housing. Moreover, the dispersed units were located not only in many of the highest-valued (1990) tracts, but ones that appreciated the most during the housing boom of the 1990s.

Sales Price Trends before Dispersed Sites Are Occupied

The regression results are summarized in graphical form for selected models and distance rings. The graphs in Figures 2 and 3 show the relative percentage differences in prices over time in single-family home sales prices in proximity to DHA dispersed housing sites, compared with prices for similar dwellings elsewhere in the same census tracts but not within 2000 feet of any dispersed (or other public housing) units. The vertical axis on the graph indicates the percentage differences in house prices over the baseline. The horizontal axis indicates time, starting with the beginning of the study period, the first quarter of 1987. The first dotted line indicates a representative starting date chosen as the point of first occupancy of the typical dispersed unit. Therefore, the section of the graph to the left of the dotted line is the relative price pattern before the dispersed housing site was occupied, and the section to the right of the dotted line is the pattern after the site was occupied.17

The results show that in Denver there was a systematic tendency for dispersed housing sites to be acquired in declining, lower-priced pockets within census tracts.18 The negative and significant coefficients on the time trend variables indicate that areas within 2000 feet of sites acquired for dispersed housing developments evinced price trends in the late 1980s and early 1990s that were falling relative to other areas within their same census tracts. These declines can also be seen in the downward sloping trend lines in the left-hand sides of the price trend graphs (Figures 2 and 3). These declines were roughly constant across all three distance rings. Over a five-year period 1989-1993, prices within 2000 feet of future dispersed sites fell about 3 to 4 percentage points relative to other areas within the same tracts.19

Key informant interviews and subsequent discussions with DHA operational staff confirmed and further explained this empirical finding. First, DHA typically acquired vacant property for their dispersed units. Insofar as these units arguably had been generating negative externalities for the surrounding neighborhood for oftentimes considerable periods before DHA acquisition, the micro-neighborhoods defined by proximity to these units would tend to have lower values. Second, there were two sources of potential self-selection bias in DHA's purchasing strategy. Because DHA was required to do a variety of time-consuming property inspections before purchase, private interests would often purchase buildings in stronger housing submarkets before DHA could acquire them. Moreover, DHA itself was likely to search more intensively for buildings to purchase in areas where they could get the most building for the money, and thereby stretch their scarce programmatic resources as far as possible.

Property Value Impacts of Dispersed Public Housing for Denver Overall

Overall, the models performed extremely well. The adjusted R2s were 0.81 in the regressions and did not vary significantly across the three model specifications. Not surprisingly given the exceptional sample sizes, virtually all of the [Struct], [Tract], and [Quarter] control variables evinced coefficients that were significantly different from zero. All the coefficients of the [Struct] characteristics of homes proved to have the expected signs. Results of the [Struct], [Quarter] and [Tract] control variables are presented in the Appendix.

The estimated parameters of the variables of interest for the three Denver models are shown in Table 2. The regressions using all analysis sites showed clear evidence of positive property price effects associated with nearby DHA dispersed housing.20 During the late 1980s to mid-1990s, overall increases in property values were observed as a result of proximity to DHA dispersed public housing sites, with more of proximate sites magnifying the beneficial effects. As shown in the first panel of Figure 2, after a dispersed housing project was occupied, sales prices within 500 feet reversed their pre-occupancy downward trend. Fourteen quarters after occupancy, prices at this distance were only 2.6 percent less than the baseline; immediately preceding occupancy they were 3.4 percent less. Thus, for all intents and purposes, the opening of a DHA dispersed development significantly revitalized the surrounding neighborhood within 500 feet, on average for these analysis sites.

Within 1001-2000 feet, proximity to a dispersed site apparently provided an additional upward boost of 1.7 percentage points to house prices relative to other areas (Figure 2, second panel).21 The conclusion of positive effect is reinforced by consideration of various numbers of dispersed sites nearby (Figure 3). The greater the number of sites within this distance, the more positive was the initial effect, though there was no subsequent change in the relative trend of prices. At 12 sites within 1001-2000 feet of a sale, the fixed effect was to boost price levels 4 percentage points, which initially was sufficient to boost the area above the baseline price for such areas.

The price-enhancing influence observed for the Denver dispersed housing program can most likely be attributed to the structure of the program itself. Under this program, DHA acquired vacant properties and invested, on average, $21,432 per unit for rehabilitation.22 The presumed improvement of the physical appearance of these properties and the return of vacant buildings to occupancy may serve to stem the decline of their areas. As these activities are carried out, the declining relative price trends within a small area of the neighborhood (i.e., within 500 feet of the site) can apparently be reversed.23 Meanwhile, at farther from the site (i.e., 1001-2000 feet), a fixed positive price effect was observed that increased with the number of DHA sites within the same area. Unfortunately, this upward shift in home values was eventually overcome by the continuing downward trend in prices. It would therefore appear that, while the housing market perceives a short-term boost in the desirability of an area from the act of rehabilitating units, such activity is insufficient to reverse overall negative trends in a larger neighborhood.

Property Value Effects of Dispersed Public Housing in Different Sorts of Denver Neighborhoods

Each of the specifications of aggregated impact models was replicated for different clusters of census tracts in Denver. Census tracts were stratified according to the racial and ethnic composition,24 median 1990 home values, and real changes in median home values from 1990-1995. Some cross-neighborhood variations in property value effects consistent with the hypothesis of countervailing externalities were identified. This hypothesis suggests two sorts of externalities generated by DHA dispersed housing sites: positive externalities emanating from the rehabilitation (and subsequent maintenance) of the structure, and negative externalities associated with behaviors of residents of DHA sites or with public stigmatization of areas near such (whether justified or not). The magnitude and spatial extent of positive and negative externalities need not be identical, and they likely differ throughout neighborhoods.

The overall positive property value influence attributed to proximity to DHA dispersed housing units, as revealed in the aforementioned aggregate models, apparently do not occur uniformly across all types of neighborhoods in Denver. Rather, they are manifest most noticeably in predominantly white-occupied neighborhoods and in the most affluent neighborhoods. This suggests that the market so strongly evaluates the rehabilitation of property in such areas that it swamps, at all distances within 2000 feet, any negative externalities that might be produced. Positive effects also occurred in low- and moderate-value neighborhoods, but only in the 1001- to 2000-feet range, suggesting that some more substantial negative externalities localized around the DHA sites counteracted the positive externalities there.

The only consistent negative effects occurred in substantially black-occupied neighborhoods at all distances within 2000 feet from the dispersed sites. Many black neighborhoods in Denver are characterized by high poverty rates, high rates of outofwedlock birth, and higher concentrations of publicly subsidized housing (including all DHA, HUD, Section 8, and low-income housing tax credit-assisted units).25 Thus, these neighborhoods may be exhibiting the effects of concentrated poverty and concentrated subsidized housing-more DHA units in such a vulnerable market context apparently only contribute further to the negative effects of such concentration. Put differently, the potential positive externalities associated with spot rehabilitation in poor, black-occupied neighborhoods are small compared with the accompanying negative externalities from adding more low-income households to the area.26 This finding provides a provocative implication for "impaction standards."

Key Insights of the Focus Groups Regarding Effects

To enrich understanding of the quantitative results, focus groups were conducted with homeowners in six of the analysis sites demonstrating different types of property value effects (i.e., none, positive, or negative), as well as variations in the ethnic and socioeconomic composition of the neighborhoods. Using information available in the public domain, focus group participants were recruited among homeowners residing within 2000 feet of each specified subsidized housing address selected from the post-1987 analysis sites. Homeowners were contacted by mail; then the research team contacted the 37 homeowners interested in participating in the focus group discussions; and the focus groups were conducted during the spring of 1998. In the focus group discussions, resident perceptions of quality of life in the neighborhood and perceptions of changes in quality of life were probed. The perceived influence of subsidized housing sites was explored when raised by the participants.27

The econometric results underscore that the real estate market in Denver is receiving consistent and accurate information regarding the location of dispersed housing units and that house pricing systematically reflects this information. Indeed, since the enactment of the intergovernmental agreement between the City Council and the DHA in late 1989 that stipulates public hearings in regard to the purchase of dispersed units, both the market as well as homeowners are formally made aware of the purchase of units in their neighborhoods. Further, according to key informants, DHA works directly with realtors to identify and purchase potential dispersed units. This would suggest that the market is responding to the rehabilitation and management of these units by DHA. Key informant interviews suggest that this, indeed, is probable.

It is notable, then, that none of the homeowner focus group participants specifically mentioned the dispersed public housing units in their neighborhoods.28 This suggests that the DHA has been successful, through acquisition of existing units and subsequent maintenance and tenant monitoring efforts, in blending their dispersed projects into the larger community.29 The focus groups did, however, consistently emphasize elements of neighborhood quality that are relevant to subsidized housing policymakers: the physical condition of the neighborhood, the presence of vacant and (especially deteriorated) properties, social cohesion, and crime and public safety.

One of the primary concerns raised during the 1989 controversy regarding dispersed housing was the potential for physical degradation of the neighborhood and how that would negatively affect property values. Focus group participants underscored the importance of this issue. Participants were very interested in the upkeep not only of residential units, but also of green space, streets, commercial establishments, and other infrastructural features of the neighborhood. The consensus among participants was that poor upkeep contributed to the decline in property values in a neighborhood: "There's trash everywhere. People aren't cleaning up their front yards. It gives kids the impression that people don't care," according to one northwest Denver neighborhood respondent.

DHA has responded (with apparent success) to such fears of neighborhood degradation by enacting a strict maintenance and inspection schedule for all of its dispersed properties. Housing managers reputedly respond promptly to complaints, including those made by neighbors. DHA staff told the authors that their properties are often some of the best-maintained on the block, and visual inspections confirmed this. The focus group results suggest that these proactive maintenance policies have helped DHA achieve greater acceptance of their developments in neighborhoods. Numerous other studies have cited this factor as a key to program success (Briggs, Darden, and Aidala, 1999; Hogan, 1996, chs. 3, 7).

Related to residents' worries about the physical upkeep of the neighborhood was their apprehension about abandoned or vacant properties. Focus group participants expressed concern about having homes left vacant for extended periods, especially multiple vacant units on a single block, which were thought to lower property values. It is therefore significant that a key element of the DHA dispersed housing program is the acquisition and rehabilitation of vacant properties. The act of returning these housing units to active occupancy apparently signals a beneficial change to the community.

Homeowners also expressed concerns regarding the high number of rental units in their neighborhoods and the potential problems that might result from having a large renter population. One respondent from a south Denver neighborhood stated: "I think it makes a huge difference about how many rental houses there are. That was a major problem here-there were so many rentals. There was no stability here." The general consensus was that neighborhood life, particularly a sense of community and commitment to the neighborhood, was enhanced by the presence of homeowners. Although participants acknowledged that some fraction of the housing in the neighborhood must be set aside for rental use, they were very concerned that after a certain threshold, the number of rental units in the area would adversely affect property values.

The presence of numerous rental properties was seen to weaken social cohesion in a neighborhood, which was another issue raised by the focus group participants. Homeowners in all six groups indicated that the most important factor affecting their quality of life was having good neighbors and feeling some sense of connection with others in the neighborhood. Who you lived with in the neighborhood was equally, if not more, important than where you lived. The focus group discussions underscored how important it was for participants to feel a sense of community with neighbors. This sense of community was described in terms of knowing one another, looking out for one another, interacting with each other, and protecting one another. However, being neighborly did not necessarily mean being totally immersed in the lives of one's neighbors. Rather, it reflected a commitment to others as well as to the neighborhood. Apparently, DHA's dispersed housing did not harm such "non-financial" investments in neighborhoods, similar to the Yonkers situation (Briggs, Darden, and Aidala, 1999).

Another concern closely related to stability and cohesion, and one that was expressed during the 1989 controversy, is the perceived threat of increased criminal activity. While none of the focus group participants directly attributed increased crime with the presence of subsidized or unsubsidized renters, they nonetheless expressed ongoing concern about crime and safety issues in their neighborhoods. In particular, participants were concerned about safety on the streets and in the schools: "It used to be when you saw kids you could ask them why they weren't in school. In this day and age you can't do that because they might have a gun or knife or will track you down" (northwest Denver neighborhood respondent). Other ongoing issues in the neighborhoods included youth crime (mainly vandalism and petty theft, graffiti, unsupervised children, and truancy), gang violence, and drugs.

Even though DHA cannot directly affect the distribution of rental properties in a neighborhood or the problems that may be caused by occupants of other types of rental housing, DHA's tenant screening and monitoring procedures have tried to diminish these concerns in the Dispersed Housing Program. Prospective DHA tenants (in both conventional and dispersed housing programs) must have an acceptable rent payment history and no record of criminal activity. They are also expected to exhibit a high degree of motivation toward self-sufficiency and community involvement. While participating in the dispersed housing program, the tenants are expected to cooperate with DHA in maintaining the interior and exterior of the property and must be able to pay for snow removal and lawn care. DHA staff informed us that they respond promptly to any complaints regarding their properties or tenants, and will not hesitate to remove a tenant who cannot meet the required behavioral standards. Such effective management practices have been widely cited as crucial elements of a successful program (Briggs, Darden, and Aidala, 1999; Hogan, 1996, chs. 3, 7).

While participant homeowners did not identify dispersed housing developments as the source of neighborhood problems, they were very aware of the existence of conventional public housing units, as well as Section 8-occupied units, in their neighborhoods. For example, residents in both focus groups held in northwest Denver mentioned disparagingly the conventional public housing projects in their neighborhood. Participants in all six focus groups discussed at length their concerns about the maintenance and management of rental properties in their neighborhoods, particularly those owned by Section 8 landlords. Their knowledge of Section 8 landlords was not based on mere speculation. Homeowners identified "problem" landlords by calling the City Property Assessor's office directly. Although focus group respondents did not directly mention dispersed housing, homeowners still approach the DHA to check whether the problematic property is owned by the Authority. According to our DHA key informants, most of the inquiries they receive prove to be for privately owned units.

Based on this discussion, it is important to emphasize that the positive property value impacts observed in aggregate here cannot necessarily be generalized to public housing dispersal programs run by other housing authorities that may not possess the features of the Denver program. It is the particular characteristics of DHA's program-the acquisition and rehabilitation of small-scale, vacant properties, the effective ongoing maintenance practices, and the strict screening and monitoring of tenants-that arguably contribute to the beneficial, or at least non-harmful, effects that have been found. One might expect that comparable programs in other cities would have similar effects, but this is a question that can only be answered by further research. Indeed, the foregoing program features have risen to the stage of conventional wisdom in some circles (Briggs, Darden, and Aidala, 1999; Hogan, 1996, chs. 3, 7), although rigorous statistical studies have rarely produced these claims.

CONCLUSIONS AND POLICY IMPLICATIONS

Housing subsidy programs for deconcentrating poverty are a major policy thrust both nationally and in many local areas (Hogan, 1996). Much evidence suggests that such programs, whether the subsidies are attached to sites or to tenants, can have more positive consequences for participants than residence in traditional public housing in concentrated poverty neighborhoods. The main question raised about these programs has concerned their social costs, typically framed in terms of their impacts on receiving neighborhoods. This study analyzed the issue of neighborhood impacts by reporting on a quantitative and qualitative reconnaissance of Denver, where fears about neighborhood impacts of a scattered-site public housing program led to major political controversies and subsequent program modifications a decade ago.

These political controversies appear especially intriguing in light of our primary statistical findings. The general pattern during the late 1980s through the mid-1990s was that 500-foot proximity to a DHA dispersed public housing unit opening in the period was associated with an increase in single-family home prices, contrary to conventional wisdom. The positive effects extended to a range of 2000 feet, especially in highest-value, white-occupied areas. However, additional amounts of dispersed subsidized housing proved harmful to proximate home values if the neighborhood was already "vulnerable," as measured either by low property values or appreciation rates, high percentages of black residents, or high poverty rates.

The authors believe that facility, management, and client characteristics are responsible for this effect, as has been found elsewhere (Dear, 1992; Hogan, 1996, chs. 3, 7). It is manifest that the acquisition of vacant, small-scale properties by DHA and their subsequent rehabilitation and occupation (followed apparently by consistently good management, tenant monitoring, and property upkeep) were viewed by the neighborhood and the market as the replacement of a negative externality generator with a positive externality generator.

The negative effect finding associated with siting additional dispersed units in more "vulnerable" neighborhoods suggests that there may, however, be a maximum threshold of poor or subsidized households beyond which this negative impact is triggered, irrespective of the distance or characteristics of DHA housing or occupants. This implies that further additions of any sort of subsidized housing units should be discouraged in these affected neighborhoods.

A crucial, if incidental, finding of the statistical analyses was that dispersed subsidized sites systematically tended to be located in the lowest-valued, slowestappreciating sectors within any given census tract. Perhaps of more import, however, are two implications from this finding. First, from a research perspective, it implies that statistical models of house price effects must be specified carefully to avoid erroneous conclusions. For example, if one merely does a cross-sectional comparison of prices near subsidized sites with those less proximate, one will tend to observe lower prices in the former area, but this cannot necessarily be traced to an independent impact from the subsidized sites. As another example, if one merely compares levels of prices before and after occupancy of a subsidized site, there will be a bias toward observing a lower than predicted post-occupancy level because of a preexisting trend in the area, not because of subsidized housing.

The second implication relates to politics and the public support than can be mustered for a dispersed subsidized housing program in the current context, factors that are essential for program success (Briggs, Darden, and Aidala, 1999; Hogan, 1996: ch. 7). Inasmuch as such housing currently has a tendency to be located in lower-valued, lower-appreciation neighborhoods, local residents and the market as a whole will more likely have their anxieties about the neighborhood's future abetted. Moreover, local residents and the market are unlikely to be able to make the subtle distinctions in causality that our statistical analyses permitted here. From their perspective, subsidized housing will tend to be seen as highly correlated with neighborhood depreciation, and this probably is sufficient to attribute causation to the former. Of course, this analysis showed that the attribution may be correct if the census tract in question is in a "vulnerable" market situation.

Finally, consider the focus group finding that many residents in neighborhoods containing dispersed subsidized housing tend to equate "bad landlords of bad properties housing bad tenants" with government housing subsidy programs. Although clearly not generalizable to all programs, this conventional wisdom erodes support for such initiatives and provides another "factual" justification for recipient neighborhood NIMBY-ism. In any event, the empirical and perceptual conditions surrounding the geography of dispersed subsidized housing appears primed for public opposition by recipient neighborhoods, needing only the spark of political opportunism to set off an explosion.

It follows that a cornerstone for reestablishing a constituency for dispersed subsidized housing and defusing potential local opposition must be an attack on the stereotypes surrounding such housing. The authors believe that this attack requires changing both the objective conditions associated with subsidized housing and the public's perceptions of these conditions. This will necessitate comprehensive attention to the local housing authority's dispersed subsidized housing program design and operations, including siting, management, tenant selection and monitoring, dwelling monitoring, and public relations (Briggs, Darden, and Aidala, 1999; Hogan, 1996, ch. 7).

Again, the authors caution that their results do not necessarily apply to all other scattered-site, subsidized housing programs. Denver is distinguished by: a highperformance housing authority; mandated site separation and density requirements; an acquisition-rehabilitation program of small-scale properties; and a rapidly inflating housing market. Nevertheless, it us valuable to discuss what these findings imply for policy were they to prove generally applicable after further study and replication. In this spirit the authors offer the following recommendations for those who design and administer dispersed subsidized housing policies.

Renovate Dwellings. To the extent feasible, site-based dispersed subsidized housing programs should attempt to acquire and rehabilitate small-scale, vacant (especially poorly maintained) properties, given the positive externalities associated with such activities evinced in Denver. Of course, as the efforts of DHA make manifest, the conscientious management and maintenance of such properties is required if this positive effect is to persist.

Establish Impaction Standards. Dispersed subsidized dwelling sites should not be permitted in "vulnerable" neighborhoods and concentrations of such units should not be permitted in any neighborhood. This recommendation is similar in principle to the longstanding HUD impaction standards for privately owned, subsidized complexes or to those imposed on DHA by the Denver City Council. In some contexts, the deconcentration of preexisting clusters of subsidized and public housing that give rise to "vulnerable" neighborhoods should be considered, such as the "dedensification" programs for conventional public housing sites now being pursued in Denver. In either event, a policy of opportunistic acquisitions should be reconsidered.

Monitor Tenants. Once subsidized tenants are in residence, local housing authorities should ensure that the tenants obey all the financial and behavioral conditions of the lease, and are swiftly evicted once these conditions are violated. For instance, DHA claims that vigilant monitoring and lease enforcement result in fewer than five percent of their dispersed tenants creating any problems.

Maintain Dwellings. Once subsidized tenants are in residence, local housing authorities must ensure that the property is maintained at a level superior to the general upkeep of the neighborhood, to confound public stereotypes and make the unit less likely to be identified as subsidized. This means the conscientious investment of housing authority resources in building upkeep and semi-annual inspections, as is done in Denver.

Beyond the aforementioned programmatic and operational changes, however, our findings lead the authors to recommend initiatives aimed at altering the perceptions of dispersed subsidized housing programs held by residents in potential recipient neighborhoods and the public at large. Although changing the functional reality of dispersed housing programs may be a necessary condition for changing public perceptions and opposition, it may not be sufficient. The authors recommend two sets of initiatives in this regard (cf. Dear, 1992; Hogan, 1996: ch. 7).

Collaboration with Neighborhood Groups. The lessons of 1989 in Denver (and elsewhere) demonstrate how a campaign of misinformation and fear-mongering can mobilize powerful forces in opposition to dispersed housing. The authors thus recommend that local housing authorities develop constructive, ongoing relationships with neighborhood groups, homeowner associations, and other local opinion leaders. For example, it now appears that DHA has learned to work more openly with neighborhoods receiving dispersed sites, listening to the neighborhood's concerns and committing resources to avoid "problem properties." The fact that none of the focus group participants cited DHA as a source of their neighborhood's problems, should be grounds for optimism in light of the vitriol of a decade ago in Denver.

Burnish the Image of Dispersed Housing. Evidence is suggestive that scattered, wellmaintained, well-managed buildings are ruled out as "subsidized housing" in the public's perceptual scheme. Local housing authorities should undertake a concerted, ongoing public relations campaign to convince the public that dispersed housing is good for neighborhoods and offers appropriate aid to low-income households. Many developers of supportive housing for special needs, for example, offer tours of their operating developments to leaders in neighborhoods where new developments are being proposed. The DHA blanketed the media with information on successful dispersed sites, though it only did so reactively once the 1989 furor had erupted. The typical resentment against the "undeserving poor" might well be mollified were local housing authorities effective in providing poignant vignettes of past subsidized dispersed households (preferably, with a variety of races and ethnicities portrayed) who had moved to economic independence. Finally, the authors believe that local and federal policymakers should seriously consider renaming the set of dispersed housing programs. Inasmuch as "public housing" remains stigmatized, the aforementioned programmatic improvements face an uphill struggle to alter public opinion. A new symbolic umbrella under which these reforms could flourish might yield dramatically better public relations payoffs.