«A Journal of Policy Development and Research HoPe VI Volume 12, Number 1 • 2010 U.S. Department of Housing and Urban Development Office of Policy ...»
This study used the date the first unit became available for occupancy as the project completion date. Several different sources—newspapers, HOPE VI developer websites, the HABC website, and other HABC sources—reported different dates of project completion. This inconsistency likely occurred because of alternative definitions of project completion, ranging from the date that major construction is completed to the date that the last construction task is completed. This study defined completion date as the date the first unit became available for occupancy, reasoning that this date should best reflect the time that the HOPE VI project would begin to have its full effect, even if minor structural tasks were still in progress. In all three cases, these three dates coincided closely with completion announcements in local newspapers.
Methodology This study used a difference-in-differences approach to test for spillover effects of HOPE VI redevelopment on surrounding neighborhoods. It is based on the premise that the sales price of a property is a function of both its structural characteristics (for example, age and size) and its neighborhood characteristics (for example, crime rate and school quality) and therefore reflects neighborhood quality and desirability. This idea that both physical and neighborhood characteristics are capitalized in the price of a property is grounded in traditional hedonics pricing theory (Rosen, 1974). By controlling for physical characteristics, microneighborhood characteristics, and fixed effects of census tracts, this method attempts to isolate the part of the sales price that reflects the property’s proximity to the HOPE VI redevelopment.
Announcement effects happened early in the study period, and the number of sales before the announcement is limited.
For example, although demolition for all five buildings was announced for FY 1995, one Lexington Terraces building was already empty in March 1993. At this time, HABC started publicly deliberating options ranging from an $8.2 million renovation to total demolition. Approximately 25 sales were completed in the microneighborhood before 1993.
76 HOPE VI HOPE VI Neighborhood Spillover Effects in Baltimore Difference-in-differences methods have been used extensively to measure neighborhood spillover effects of subsidized housing. Briggs, Darden, and Aidala (1999) evaluated neighborhood effects of dispersed subsidized housing in Yonkers, New York, by comparing differences in sales prices between properties one-fourth of a mile from the subsidized housing and properties farther away but within the same census tract. Using larger data sets and more sophisticated extensions of this approach, Galster, Tatian, and Smith (1999) compared both the level and trend of property sales prices in the surrounding neighborhood before and after Section 813 occupancy, and Ellen et al.
(2001) measured the spillover effects—and their trajectories—of a homeownership program in New York City.
The intuition behind the difference-in-differences approach is that it compares changes in property values close to the HOPE VI site to changes in property values farther away from the site but in the same neighborhood before and after completion. The validity of the estimate hinges on the extent to which the change in property values before and after the redevelopment of the properties farther away from the site represents what the change in property values would have been for the properties closer to the site in the absence of HOPE VI redevelopment. It is important to note the possibility that the redevelopment affected property values outside the microneighborhood as well. If this were the case, the true spillover effect would be underestimated. One would expect that properties immediately surrounding the redevelopment, however, would be affected more directly than properties farther away from the site and that this difference would be reflected in the estimate. Equation 1 expresses a basic difference-in-differences model.
(1) where post represents postredevelopment, pre represents preredevelopment, micro represents property located in the surrounding (micro) neighborhood, and macro represents property located in the macroneighborhood but outside the microneighborhood. Finally, impact is the estimate of the spillover effect—that is, the effect of redevelopment on the average housing price. To obtain standard errors to test whether this estimate is statistically significantly different from zero, equation 2 uses ordinary least squares (OLS) regression to estimate the model.
(2) where micro is a dummy variable that equals 1 if the sale occurred inside the microneighborhood and 0 otherwise, post is a dummy that equals 1 if the sale occurred after project completion and 0 otherwise, and post*micro is an interaction term between the two dummies. The variable micro serves as a control variable, and its coefficient can be interpreted as the baseline difference in price levels between the microneighborhood and outside-of-micro neighborhood. The impact variable in this model is post*micro. The coefficient on post*micro indicates any deviation from the overall difference in prices of the two time periods that the microneighborhood experienced. A statistically significant positive coefficient signals a positive effect of the HOPE VI project on sales prices of Section 8 is a federally funded rental assistance program for low-income households in which recipients use vouchers to choose privately owned rental housing. The program subsidizes the difference between 30 percent of the household’s income and the total rent amount (determined by the public housing authority and the property owner based on Fair Market Rents).
surrounding property. Y is the property sales price, estimated in both linear and natural log form, adjusted for inflation to 2006 dollars using the Consumer Price Index. This model is referred to as the “basic difference-in-differences” model in the results section.
This basic model does not take into account any variation in the types of properties sold, either between those located in the microneighborhood and those located outside the microneighborhood, or over time. It also assumes that neighborhood characteristics—such as local crime and local services—do not differ between the microneighborhood and the macroneighborhood. The latter assumption seems plausible, given that each of the three analyses is limited to the larger neighborhood where the HOPE VI site is located, and property sales outside neighborhood boundaries (such as major roads and a highway ramp) are excluded from the analysis. Still, a second model, referred to as the “regression-adjusted difference-in-differences model,” includes a dummy variable for each census tract to serve as localized fixed effects, which control for differences in unmeasured factors that affect the entire census tract, such as school quality, local amenities, crime levels, and demographics. It also includes property characteristics to control for the variation in the type and quality of properties sold in the microneighborhood compared with the rest of the macroneighborhoods and before and after completion of the project. Equation 3 expresses this model.
(3) where [tract] is a series of dummy variables indicating the census tract in which the sale is located and [structure] is a vector of structural property characteristics that controls for structural and tenure characteristics, including the building’s age, lot size, structure size, number of stories, presence or absence of a basement, construction type (brick, wood, or other), building type (rowhouse, detached, or semidetached), quality of construction, and previous housing tenure (rented or owner occupied).
Exhibit 7 shows that average sales prices vary from year to year. This variation could yield misleading results if the volume of sales were not constant over time. If, for example, there were more sales in the microneighborhood in the earliest years of the study period before project completion than there were in the following years just before completion and if prices in the entire macroneighborhood increased steadily over time, if this ratio were reversed for the rest of the macroneighborhood (that is, if there were fewer sales in the earlier years of the study period when prices were lower than they were in following years just before project completion), the impact would be underestimated. Equation 4 accounts for this variation.
(4) where [year] is a vector of dummy variables representing the year of the sale. Results from this third model represent the most reliable estimates of HOPE VI spillover effects in each site. These [year] dummies after completion replace the post dummy, which captured the aggregate change in property values after project completion that would have increased property values even in the absence of HOPE VI redevelopment, and, in this third model, each year is captured separately.
For one site, Broadway Overlook, which showed evidence of positive spillover effects, this post*micro interaction was replaced with a series of year*micro dummies for each year after completion in a fourth model (equation 5), to examine whether the effect increased or decreased with time.
where [year*micro] is a series of interaction variables between each postcompletion year dummy variable and the micro dummy variable.
The credibility of these estimates relies on the assumption that sales prices followed the same trend over time in the microneighborhoods and macroneighborhoods before HOPE VI redevelopment. Spillover effects could be overestimated if prices in the microneighborhood were already rising at a faster rate than prices in the macroneighborhood, or underestimated if prices in the microneighborhood were increasing more slowly than prices in the rest of the macroneighborhood. Recent research measuring spillover effects of subsidized housing have used sophisticated methods to account for differing price trends of properties immediately surrounding subsidized housing compared with properties farther away. This method was first used in Galster, Tatian, and Smith (1999) to compare both the level and the trend of property sales prices in the surrounding neighborhood before and after Section 8 occupancy, and adapted in later studies that examined spillover effects of subsidized housing (for example, Santiago, Galster, and Tatian 2001).
In the context of the three Baltimore HOPE VI sites, the number of sales in each of the three samples is too few to reliably estimate different trends in each part of the neighborhoods separately, or to test the developments’ impacts on trends. Estimating separate trends for the microneighborhood and macroneighborhood is important only if there are preexisting differences.
Basing trend differences on the mean sales prices in exhibit 7, it seems very unlikely that these differences existed before HOPE VI redevelopment, because no difference in yearly sales price trajectories in the microneighborhoods and macroneighborhoods is evident before completion.
Results and Discussion The regression results of the three main models described in the previous section are shown in exhibit 8. The first pair of columns presents the results of the basic difference-in-differences model, the second pair presents the results of the difference-in-differences model controlling for structural characteristics and localized fixed effects, and the third pair presents the results of this model replacing the post variable with dummy variables for year fixed effects. Full results are presented by site in appendix exhibits B1 through B3.
The micro estimates represent baseline differences in price levels between the microneighborhood and the rest of the macroneighborhood. In the models using the linear form of sales price as the outcome, the coefficients on micro can be interpreted directly as estimates of this difference. In the models using the natural log of price as the outcome variable, where estimates are small (approximately 0.25 or less), the coefficients on micro multiplied by 100 can be interpreted approximately as the percent by which properties in the microneighborhood deviate from comparable properties
outside the microneighborhood but within the same macroneighborhood.14 The coefficients on the post*micro variables can be interpreted as the amount by which property values in the microneighborhood increased (if the coefficient is positive) or decreased (if the coefficient is negative) for the models using price as the outcome, and multiplying the coefficient by 100 for the models using the log of price as the outcome gives the percent of increase or decrease.
The evidence weakly supports the hypothesis that properties in the microneighborhood surrounding the Townes at the Terraces redevelopment significantly increased in value after the project’s completion.
The coefficients on micro indicate that properties within the microneighborhood were already of higher value than similar properties outside the microneighborhood. Property values increased overall by about $20,000 after HOPE VI completion (40 percent, according the model specification using the log of sales price, translated using the Halvorsen-Palmquist equation shown in footnote 14).
None of the coefficients on the post*micro interaction are significant, and the large difference in magnitude between the basic difference-in-differences model and the second model controlling for physical characteristics and localized fixed effects suggests that the composition of properties sold in this neighborhood before and after the HOPE VI redevelopment is not similar. It is possible that properties sold after redevelopment were of higher quality because of competition from the new HOPE VI units, but no such definitive conclusions can be drawn from these results. Note that the Townes at the Terraces site had the smallest sample size, with only a handful of sales in some years.
It is therefore unclear if the positive coefficients on post*micro should be interpreted as an indication of positive spillover effects, or if the observed differences in property values are due to chance.