In addition, cities like Berkeley, Chicago, Madison, Philadelphia and Seattle have grappled with how to design programs to promote community gardens and neighborhood parks.
Under a TIF plan, a community designates a geographic area that is likely to benefit from the provision of open space, issues bonds to finance the purchase of that space, then pays for the debt service on the bonds from the additional property tax revenues resulting from the increased value of properties within the TIF district. For a TIF to be successful, the park or other amenity financed must spur private development and raise property values sufficiently to generate the increased tax receipts necessary for the debt service. Being able to accurately predict the property value impacts that different kinds of open space will have on neighboring property values accordingly is crucial to local governments considering TIF.
In preliminary research, we also investigated the effect of community gardens on neighboring commercial property values. Findings suggest that gardens do not have any significant immediate impacts on neighboring commercial property values, although some benefits may occur over time.
A recent working paper by Wachter and Wong (2006) focuses on tree plantings in Philadelphia and finds that they have positive and significant impacts on surrounding property values.
Specifically, for each case study park, the authors compare the average house price and price appreciation in the immediate neighborhood of the park with those farther away, in designated control areas, during the period of capital improvements.
Some studies attempt to address these limitations using case studies or other methodological approaches. See, for example, New Yorkers for Parks and Ernst & Young (2003). We discuss here only those studies using hedonic methods.
Note, however, that some studies claim that urban gardening may, in fact, reduce local crime (Warner and Hansi 1987, Hynes 1996, Murphy 1999).
A typical city block (North/South blocks between streets) in Midtown Manhattan is about 260 feet long. Thus, the 1,000-foot ring allows for impacts extending up to roughly four blocks away from the garden. The 1,000-foot radius was selected following Bolitzer and Netusil (2001) and Netusil (2005). The former assumed, based on discussions with park officials, that the impacts of urban parks may extend to up to 1,500 feet; the latter used a 1,300-foot (1/4 mile) radius to capture impacts of parks and other relatively large open areas (golf courses, wetlands, etc.). Because New York City community gardens are relatively small, we chose a somewhat smaller ring size. This choice of ring size is also supported by discussions with members of the garden community in New York. One gardener in the Bronx mentioned, for example, that the children from a school located about four blocks away from the garden did not participate in educational activities organized at the garden because they thought it was too far away.
Ellen et al. (2002) employ a similar methodology to evaluate the impact of investments in selected homeownership developments.
Community districts are political boundaries unique to New York City. The City is divided into a total of 59 community districts, each of which has a Community Board whose members are appointed by the Borough President, with half nominated by the City Council members who represent the district. The Community Boards review applications for zoning changes and make recommendations for budget priorities.
Most previous research has assumed that trends in housing prices are constant across a city or metropolitan area, but this seems particularly inappropriate in a city as large and diverse as New York. Schwartz, Susin and Voicu (2003), for instance, find considerable variation in price trends across community districts in New York City.
While specifying the time dummies using an even smaller geographic area—say a census tract—may seem preferable to the community districts, doing so comes at a considerable cost and adds little explanatory power. Put simply, census-tract specific time dummies would add approximately 150,000 more dummy variables to the specification, significantly increasing the number of parameters to be estimated, and greatly reducing degrees of freedom. Moreover, there is little variation in the time dummies within the community districts—an F test could not reject the hypothesis that census tract-quarter dummy variables were the same within a community district.
More precisely, the coefficient on a dummy variable should be interpreted as the difference in log price between properties that have the attribute and those that do not. Because the difference in log price closely approximates the percentage difference in price when the difference is small enough and because differences discussed in this article are generally smaller than 10%, we use this more intuitive interpretation throughout the article. The percentage effect of a difference in logs, b, is given by 100(eb– 1), although this formula is itself an approximation when b is a regression coefficient; see Halvorsen and Palmquist (1980) and Kennedy (1981).
Some sold properties are within 1,000 feet of two or more gardens, however, a majority (almost two-thirds of all sales within 1,000 feet of any garden) are within 1,000 feet of only one garden.
In preliminary research, we allowed baseline property values to vary also with both the size and the land ownership of the site; however, we did not find any statistically significant variations along these dimensions and settled on the more parsimonious specification shown here.
Of the gardens we studied, 95% are sited on publicly owned parcels that are leased to local communities for gardening; the rest are located on privately owned lots. Throughout the article we use the terms “gardens sited on publicly owned (privately owned) land” and “public (private) gardens” interchangeably.
Preliminary investigation of the data suggested that a third degree polynomial would be desirable to capture the general postopening time trend in house prices in rings. Nonetheless, results are robust to the choice of the power terms.
To be clear, Tpost equals 1/365 if a sale is located within the ring of a garden and occurs the day after its opening, it equals one if the sale occurs one year after the garden opening and so on. The environmental disamenities literature has explored alternative ways to specify the decay or acceleration of impacts over time. See Kiel and Zabel (2001) for a useful discussion.
In preliminary work, we also allowed for nonlinear distance gradients by including distance squared terms; because their coefficients were not statistically significant and results are robust to their exclusion we opted to exclude them from the models shown here.
Although not shown in Equation (1), our specification also includes a set of control variables that capture proximity to other gardens that could not be used for impact estimation either due to missing opening date or because the opening date was outside of the period covered by our sales data. If we did not include these controls and the location of these other gardens were correlated with that of our sample of gardens with valid data, our impact estimates would be biased.
In earlier models, an F-test rejected the hypothesis that the coefficients on property characteristics are similar across neighborhoods.
To create submarkets, we matched census tract-level data to community districts.
To be clear, if the property is within 1,000 feet of only one garden, the share variable takes on either value 0 (if the garden quality is unacceptable) or value 1 (if the garden quality is acceptable); if the property is within 1,000 feet of two or more gardens, the share variable takes on values between 0 and 1 (e.g., if there are two equally large gardens within 1,000 feet of a given property, one of which is of acceptable quality and the other is of unacceptable quality, the share variable equals 0.5).
In sensitivity analysis using the alternative measures of overall condition, we replace these variables either with the shares of garden area of low and excellent quality or with the number of favorable answers (or points in surveyors' overall assessment).
In an additional extension, we allow impacts to vary with the garden's rating on each of the six criteria described above. For this, we replace share of existing garden area with acceptable overall quality with six variables representing the shares of existing garden area with acceptable rating on each of the criteria of community access (PostRing_ShareACA), fencing attractiveness/permanence (PostRing_ShareAF), cleanliness (PostRing_ShareAC), landscaping (PostRing_ShareAL), decorations (PostRing_SharePD) and social spaces (PostRing_SharePS).
Experts on community gardens in New York City advised us that it takes between several months and two years to set up a community garden. Thus, one year seems a reasonable assumption for the average length of time needed to create a garden.
Note, however, that this specification may be overly conservative if the planning of the garden started well in advance of the actual work. In this case, the anticipation of the effect the garden would have on the surrounding community may have resulted in price changes before the start of the garden development and thus, the price appreciation in the years immediately preceding this start may have been caused, at least in part, by the garden itself.
One could also imagine that a community opening a garden might at the same time undertake a broader range of neighborhood improvement initiatives that would confound any estimates of the property value impact of the garden. However, experts on community gardens in New York City informed us that while garden communities may embark on other neighborhood improvement activities, such activities usually lag behind the formation of the gardens. Gardens become catalysts of community development, as the networks and other social capital formed over gardens are deployed to start fixing schools and housing, organizing neighborhood watches and serving other community needs.
InRingSHsg indicates whether the property sold is within 1,000 feet of any existing or future subsidized housing project; PostRingSHsg indicates whether the property sold is within 1,000 feet of any completed subsidized units; PostRingSHsg Units represents the number of completed subsidized units within 1,000 feet of the sale; TPostSHsg equals the number of years between the date of sale and the project completion date for properties in the 1,000-foot ring.
We obtained these data in 2004. CENYC updates the data periodically, however, different gardens may have different dates of last update depending on when CENYC last received information from the gardeners or some other source. For example, the year of last update for a majority (55%) of the gardens in our main estimation sample is 2002 and the information on the remaining gardens was updated in 2003 and 2004.
For a majority of the gardens, the opening date recorded in the CENYC data actually represents the date when the first lease for the site was issued by Green Thumb, the city agency in charge of leasing city-owned land at no charge to neighborhood groups for gardening. According to CENYC staff, the actual opening may have occurred either before, or, more often, after the first lease was secured. To test the sensitivity of our results to changes in the opening date, we estimated our baseline models using alternative opening dates: the date recorded in the data and the date recorded in the data ± one year. Results differ little in response to different dates. We present in the article the results based on the date recorded in the data (the positive impact estimates are actually slightly smaller when using this date).
The CENYC raw data included 783 gardens. Out of these, we focused on the 636 gardens established between 1977 and 2000. The reason for this selection criterion is that our sales data only covers the period 1974–2003, and it is generally desirable to match a minimum of two to three years of property sales data both before the earliest garden opening date and after the latest opening date. This approach ensures that the estimates of both pre- and post-opening levels and trends in prices in the micro-neighborhoods around the garden sites will be representative of all gardens included in the sample. Nonetheless, we also included in the analysis the rest of 147 gardens which were established outside of the 1977–2000 interval or had missing foundation year—but only to control for proximity to them and thus to obtain accurate impact estimates for the gardens on which we focus.
We eliminated from the analysis 55 garden sites (37% of the Bronx total) which, upon inspection, turned out to not actually host an active community garden (e.g., the site was vacant or abandoned, or hosted a building, a school playground or a school garden), as well as six active community gardens (4% of the total) for which the surveyor could not obtain reliable information on one or more items of interest. One possible reason for the high percentage of sites that turned out not to be active gardens is that the uses of some of the sites included in the CENYC database may have changed between the time of the latest data update and the date of our survey (for a large majority of these sites, the latest data update was beginning of 2002, whereas our survey took place towards the end of 2004). Because we do not know when the use of the lot changed, we did not exclude these gardens from our baseline model.
Note that sales of cooperative apartments are not considered to be sales of real property and are not included in the DOF data set. Most of the apartment buildings in our sample are rent stabilized. Given that legally allowable rents were typically above market rents outside of affluent neighborhoods in Manhattan and Brooklyn during the period of our study, we do not believe that their inclusion biases our results (see Pollakowski 1997).
We limited the analysis to properties that are located within the 51 community districts (of the total 59) with community gardens.
Most of the RPAD data we use were collected in 1999, and it is conceivable that some building characteristics may have changed between the time of sale and 1999. However, most of the characteristics that we use in the regressions are fairly immutable (e.g., corner location, square feet, presence of garage), and when we merged RPAD data from 1990 to 1999, we found that characteristics changed very rarely. Even among these apparent changes, we suspect that a majority are corrections, rather than true changes.
See Ellen et al. (2002) for more detail on the data and parameter estimates on the building characteristics in a similar model.
For more details on the subsidized housing data sets used in this article, see Ellen et al. (2007).
As all tax lots in New York City have been geocoded by the New York City Department of City Planning we used a “cross-walk” (the “Geosupport File”) which associates each tax lot with an x, y coordinate (i.e., latitude, longitude using the U.S. State Plane 1927 projection), community district and census tract. In the case of physical structures, a tax lot is usually a building and is an identifier available to the property sales and RPAD data. We are able to assign x, y coordinates and other geographic variables to over 98% of the sales using this method. The community gardens data also include the tax lot(s) occupied by each garden, so we were able to assign x, y coordinates to each garden lot by the same method. If a garden encompassed multiple lots, we calculated the coordinates of the center of the garden.
We only use subboroughs with gardens for these statistics because our subsequent estimation of the garden impacts on property values is based only on these areas (recall from footnote 35 that we limited the analysis to properties that are located within the community districts with community gardens; subborough area boundaries are, in general, fairly similar to community district boundaries, and, in addition, they match exactly with census tract boundaries).
The census tract data is taken from the 1980 Census. Tracts are characterized as including gardens even if these gardens did not open until later in the decade.
We use 1980 tract characteristics because Table 2 shows that a vast majority of the gardens in our study were built during the last two decades. Thus, the table largely captures characteristics of the tracts before the gardens were opened.
Briefly, results indicate that sales price is higher if a building is larger or newer, located on a corner, or includes a garage. Sales price is lower if the building is vandalized or abandoned. The building class dummies are also consistent with expectations. Sales prices per unit for most of the building types are lower than those for single-family attached homes (the omitted category), and sales prices for single-family detached homes are higher than those for single-family attached homes. The coefficient on the dummy variable indicating that the building has undergone a major alteration prior to sale is positive and significant. Statistically significant coefficients on dummy variables indicating missing values for the age or size of a building indicate that these buildings missing age data are less valuable than others (perhaps because they are older) and buildings missing square footage data are more valuable (perhaps because they are larger).
These derelict vacant lots were the result of urban decay, landlord disinvestment and abandonment during the fiscal crisis of the 1970s. Most of them were owned by the City and were temporarily leased to community gardeners for a symbolic fee.
The coefficient for the InRing * D variable is positive and significant, indicating a sharp price gradient such that the pregarden price-depressing effects of the site (the disamenity) decline with distance. For example, at a distance of 1,000 feet, residential prices are only 1% lower, meaning that the price differential falls at a rate of about one percentage point per 100 feet.
We thank an anonymous referee for suggesting this additional explanation.
As noted before, the PostRing coefficient provides an estimate of the fixed component of the public garden effect—that is, the portion of the impact that is independent of the garden area. Increasing garden size appears to bring a smaller benefit, perhaps because larger gardens tend to be noisier and to generate more congestion. However, this negative marginal effect of garden area is fairly small and declines as the area increases. To take a concrete example, a 10,000 sq. ft. increase in garden area, from 1,000 to 11,000 sq. ft., reduces the external benefit by only 0.2 percentage points. Spillover benefits are also negatively affected by the private ownership of the land on which the garden is sited. The benefit of a privately owned garden is 2.8 percentage points lower than that of similar garden sited on publicly owned land, and it is not statistically significant. Finally, note that the coefficients on PostRing * D and PostRing_GArea * D are not statistically significant, suggesting that impacts vary little with distance from the site right after garden opening.
We use the median rather than the mean garden size to define the typical garden because our sample distribution of garden area is highly skewed to the right (e.g., the mean is larger than the 80th percentile).
The difference between the 3.6 percentage point impact reported in Table 5 and the 4.1 percentage point impact reported in this simulation results partly from our using a garden of the median size in our sample for the simulation. An additional reason for this discrepancy is that the former number represents the impact immediately after garden opening whereas the latter represents the impact one year after opening.
Note, however, that these estimates likely understate impacts at 1,000 feet because the coefficients on the interactions of PostRing and GardenArea with distance are not statistically significant, and yet we included them in predictions. Discarding the insignificant coefficients yields impact estimates at 1,000 feet that vary between 3.3 percentage points one year after opening to 3.8 percentage points five years after opening.
We use median rather than mean price in our simulations because the citywide mean house price is driven up by the hot submarkets in Manhattan and certain areas of Brooklyn. Nonetheless, we also computed the less conservative dollar-value impact estimates based on the mean housing price in rings ($125,275). Right next to the garden, these estimates range from $5,134 one year after opening to $9,323 five years after opening (more detailed estimates are available upon request from the authors).
Impact also declines with private ownership of the land on which the garden is sited.
Additionally, in low-income neighborhoods, the negative marginal effect of additional garden area is significantly smaller than in higher-income neighborhoods.
Note, however, that even in higher-income areas benefits may arise over time, as suggested by the positive and significant TPost2 coefficient.
We alternatively hypothesized at the outset that homeowners may have stronger economic incentives than renters to create higher-quality gardens, but we did not find a statistically significant correlation between garden quality and the homeownership rate in the garden neighborhood.
See Schill et al. (2002) and Ellen et al. (2006) for comprehensive evidence that subsidized housing is concentrated in poor neighborhoods.
Table A2 in the Appendix presents results of a specification that focuses on specific aspects of garden quality identified in the survey—community access, fencing attractiveness/permanence, cleanliness, landscaping, decorations and social spaces. We find that features that matter most are fencing attractiveness/permanence, cleanliness and landscaping.
By relative price, we mean the percentage price differential between properties in the ring and those outside the ring (but still in the same census tract).
In this discussion, we are using interchangeably the terms “acceptable (unacceptable) condition” and “good (low)-quality.”
This assumption is also realistic, given that a majority of property sales are within 1,000 feet of only one garden (see footnote 14).
These estimates are computed at one year after garden opening.
We also find a significant increase in house prices in the year immediately preceding the garden opening, however, as noted above, this increase is likely caused by the garden itself.
These estimates are available upon request from the authors.
These estimates are available upon request from the authors.
The repeat sales estimates are available upon request from the authors.
The average garden size in Fox, Koeppel and Kellam (1985) is 5,225 sq. ft., which is very similar to the median garden size in our study (6,000 sq. ft.)
These initial, one-time expenses cover the clearing of the site, providing a fence, plant materials, lumber of raised beds and benches, signs, technical assistance and the like.
This dollar value represents 1982 dollars. In 2003 dollars, the cost is $31,500.
Again, this dollar value represents 1982 dollars. In 2003 dollars, the annual maintenance cost is $16,700.
For the present value calculation, we assume a growth rate in maintenance costs of 4.2% (equal to the average inflation rate over the study period) and a discount rate of 5.35% (the average for recent city general obligation debt issues). We compute maintenance costs over a 20-year period because this is the time interval used to estimate tax benefits (see below).
The RPAD database is an annual census of all New York City properties, described above.
For this calculation, we assume, again, a temporal discount rate of 5.35%.
A comparison between the 1998 assessed values of properties in rings that sold in 1999 and the assessed values of all properties in rings suggests that transacting properties may have higher values than other properties. We thus correct for this selection bias when estimating total benefits.
Assessment ratios and tax rates vary by property type. The Department of Finance groups residential properties into three classes—class 1 (1–3 family houses), class 2 (4–6 family buildings) and class 2A (all the others)—and sets an assessment ratio and a tax rate for each class. The assessment ratio for class 1 is 0.08, for class 2 is 0.25 and for class 2 A is 0.45. The tax rate for class 1 is 0.116 and for the other two classes is 0.108. Due to this variation in assessment and tax rates, we also estimate the total benefits from gardens separately for each of the three building classes and then apply the corresponding rates to the benefit for each class.
Specifically, we assume an annual growth of 2.5% in property taxes over 20 years, based on average increases in assessments and a discount rate of 5.35%. Note that during the first five years after garden completion, the assumed 2.5% annual increase in taxes is over and above the increase due to the growth in garden impacts.