Introduction

Residential cluster development is a form of land development in which principal buildings and structures are grouped together on a site, thus saving the remaining land area for common open space, conservation, agriculture, recreation, and public and semi-public uses (Whyte 1964; Unterman and Small 1977; Arendt 1996; Sanders 1980).

In the United States, the development of Radburn, New Jersey, in 1928 represented the first formal introduction of the cluster development concept. In Radburn, single-family homes and garden apartments are sited in “superblocks” of 35 to 50 acres (Stein 1957). Clustering also became the basic site design concept in such contemporary new towns as Reston, Virginia, and Columbia, Maryland (APA 2006). It drew on English town planning principles, notably those of the so-called “garden city” movement.

The “garden city” movement is an approach to urban planning that was founded in 1898 by Sir Ebenezer Howard in the United Kingdom. Garden cities were intended to be planned, self-contained, communities surrounded by greenbelts, containing carefully balanced areas of residences, industry, and agriculture (see for example Godall 1987).Footnote 1 Notable “garden city” examples in the United States include: the Woodbourne neighborhood of Boston; Newport News, Virginia’s Hilton Village; Pittsburgh’s Chatham Village; Garden City, New York; Sunnyside, Queens; Jackson Heights, Queens; Forest Hills Gardens, also in the borough of Queens, New York; Radburn, New Jersey; Greenbelt, Maryland; the Lake Vista neighborhood in New Orleans; Norris, Tennessee; Baldwin Hills Village in Los Angeles; and the Cleveland suburb of Shaker Heights.

Today, there are many garden cities in the world. Most of them, however, have devolved to exist as just dormitory suburbs, which completely differ from what Howard set out to create. Contemporary town-planning charters like New Urbanism, Principles of Intelligent Urbanism, and Cluster Residential Districts (the subject of this study) find their origins in this movement.

The typical planning goals of cluster development are as follows:

  1. a)

    preservation of open space to serve recreational and scenic purposes;

  2. b)

    improved living environments which with a variety of housing that permits more economical housing to be constructed;

  3. c)

    provide a pattern of development in harmony with the natural features of land; and,

  4. d)

    provide an economical subdivision layout, efficient use of the land, with smaller networks of utilities and streets.

Many studies have apparently tied “open space” to value [see for example, Correll et al. (1978); Bolitzer and Netusil (2000); Luttik (2000); Smith et al. (2002); Geoghegan (2002); Irwin (2002); Lindsey et al. (2004); Evenson et al. (2005); Earnhart (2006); Krizek (2006); and Asabere and Huffman (2009)]. Open spaces, greenways, trails (green amenities already tied to value enhancement) are common in cluster developments. If clusters provide a package of these value added enhancements, one would expect to see some evidence in hedonic pricing modeling that supports the positive impact of cluster on price.

Conversely, as noted above, one of the goals of clustering is to produce high density and economical housing. The cramped living spaces associated with high density developments, usually the trade-off for clustering, could create negative externalities. However, the relationship between density and property values is not that simple. An increase in legally-permitted density will probably increase the value of a property if its market density is higher than the permitted density. Studies showing the effects of land use and environmental regulation on housing costs include: Courant (1976); Dowall (1979); and Katz and Rosen (1987).

Thus, we have an empirical question that needs to be resolved by this work. Albeit, will the effect of the competing equilibrium forces due to clustering be positive or negative? The objective of this study is to resolve this empirical question.

The next section presents a brief description of residential clusters in the study area and the study framework.

Cluster Residential Districts, the Study Area and the Data

There are two types of cluster development residential zoning districts in the township of Lower Gwynedd. These districts are classified by the Lower Gwynedd Township Zoning Ordinance,Footnote 2 as “AA-1” and “A-1” residential districts. These forms of development may permit a reduction in lot area requirements, frontage and setbacks to allow development on the most appropriate portions of a parcel of land in return for provision of a compensatory amount of permanently protected open space within the development. In effect, a developer of a tract in a cluster district may request as a conditional use, in accordance with Section 1298.07 and 1258.10 of the zoning ordinance that the tract be permitted to be developed at higher density if there is preservation of open space.

Among other things, cluster residential developments in the “A-1” district must have a minimum of 10 acres and shall be in a single and separate ownership or shall be the subject of an application filed jointly by all the owners of the entire tract, who shall stipulate that the entire tract will be developed in accordance with the approved plan. The corresponding minimum for the “AA-1” district is 5 acres.Footnote 3 The existence of these residential clusters in our study area presents a unique opportunity for a study of their potential impacts on home values. To the best of my knowledge, there exists no empirical evidence on the impacts of clusters.

The study area is the township of Lower Gwynedd located in Montgomery County, Pennsylvania. Lower Gwynedd is an affluent township with residents having a median household income of $74,351. In comparison, the State of Pennsylvania has a median household income of $50,713 (2000 census). The census also recorded a total of 10,442 residents (an estimate for 2005 was 10,920 residents). The total housing stock in 2008 was 4,784 with an average home value of $252,344 (MCPC 2008).Footnote 4 Inter neighborhood comparisons show that the residential neighborhoods of Lower Gwynedd are quite uniform in quality with median housing prices between $350,000–$545,000(MCPC 2008). There is only one school district in Lower Gwynedd Township, and there are no discernible variations in the demographics and the quality of public goods across the subject cluster versus non-cluster neighborhoods.

The sample consists of a sample of 1,502 Single-family home sales that occurred from January 2005 to December 2009 in the Township of Lower Gwynedd. All single family detached sales in Lower Gwynedd during the five-year sample period were used, except for a number of distressed sales and sales with missing data. Information about the transactions was obtained from the Montgomery County Board of Assessment (BOA).Footnote 5 The database provides information on the sales price, and a set of variables describing property characteristics such as amenities, location, and date of sale, age, square footage, lot size, property address, rooms, baths, and so forth.

Information on cluster residential development in the Lower Gwynedd Township is provided by the Building and Zoning Department as summarized in Table 1. Among other things, the cluster development data contain a list of all cluster residential developments. As can be seen in Table 1, there are a total of 19 cluster developments in the township. Five out of the 19 are in “AA-1” cluster development districts while the remaining 14 are in “A-1” districts. About 11 % of the transactions in our database are cluster housing. The 19 clusters together cover a total area of 537.5 acres with an average density of 0.76 dwellings per acre with 31 % preserved as open space. The detailed nature of the information on the clusters made it relatively easy to distinguish cluster developments from non-cluster developments. Table 2, presents the data and summary statistics of the five-year database used for this study. The next section presents “The Empirical Framework and the Estimation Results”.

Table 1 Cluster Development in Lower Gwynedd Township
Full size table
Table 2 Descriptive statistics
Full size table

The Empirical Framework and the Estimation Results

In explaining house prices, the real estate literature has typically used the hedonic framework to identify the marginal effect on house prices of various housing characteristics. The empirical framework for this study is the hedonic model (Rosen 1974). Sirmans et al. (2005) examines hedonic pricing models for over 125 empirical studies and finds that these studies have examined a vast number of variables. However, the impacts of cluster on house price were not identified as one of the variables previously studied.

The well-known hedonic framework is employed as shown by Eq. 1, below:

$$ {\text{Ln}}\left( {\text{SP}} \right) = {\text{Ln}}\left( {{\beta_0}} \right) + {\beta_1}\left( {\text{CLUS}} \right) + {\beta_2}\left( {\text{OPEN}} \right) + {\beta_3}\left( {\text{DENSITY}} \right) + \sum\nolimits_{j = 4}^n {\beta jXij + ej} $$
(1)

Where:

Ln (SP):

The natural log of sales price

DENSITY:

Dwelling units per developable acre permitted

CLUS:

Dummy variable for cluster residential development

OPEN:

Acres of open space in the subdivision

Xij:

Conventional hedonic

ej :

Error term.

In addition the traditional OLS hedonic, this study also utilizes the spatial autoregression (SAR) estimator on the hedonic model. The WLS (plus SAR) procedure is employed owing to the fact that hedonic housing price studies of this type are prone to the typical problem of spatial autocorrelation. Basu and Thibodeau (1998), for instance, argue that spatial dependence exists because nearby properties will often have similar structural features and also share locational amenities. This is likely to be true in this case given that clusters were often developed at the same time and share the same location-specific amenities.

The WLS (plus SAR) procedure uses the same variables as the OLS to estimate the regression. However, this technique uses the correlated errors of the geographic information present in the data to improve prediction (see Pace and Gilley 1977; and Carter and Haloupek 2000 for detailed treatment of procedure).

As shown in Table 2, control variables for the hedonic analyses include physical property characteristics and time of sale. The study also includes several control variables for location including: distance from the township center (TBD); dummy variables for MLS specified areas of Gwynedd Village (GWYDD), Gwynedd Valley(GWYVL), Penllyn(PENLYN), Spring House(SPRNHS), and OTHER for all other locations). Also included are a dummy variable (STREET) for location on a major township road (this is a catch-all dummy variable assigning 1 for location on any of the major roads: Welsh Road, Norristown Road, Sumneytown Pike, 309 Expressway, Bethlehem Pike, Township road, Tennis Road, Pennlyn Pike, Dekalb Pike, Swedesford Road, and Gypsy Hill Road); proximity to Ambler Borough (AMBLER); and another dummy variable (R5TRAIN) for proximity to R-5 suburban train stations at Pennlyn and Gwynedd Valley.

Of the location variables, it is expected that TBD will carry a negative sign indicating preference for location in proximity to the center of economic activities. Relative to PENLYN, which is relatively not so affluent, it is expected that the location variables GWYDD, GWYVL, SPRNHS, and OTHER will carry positive signs consistent with local wisdom. The dummy variables STREET and R5TRAIN are expected to carry positive signs indicating universal preference for access. The dummy variable AMBLER for proximity to Ambler is expected to carry a negative sign granted that the Ambler Borough is not as wealthy as Lower Gwynedd township, relatively speaking.

Based on the Lower Gwynedd described above, several estimates are made using the OLS and WLS procedures. As to be expected, the results based on WLS are slightly qualitatively superior to the results based on the OLS. For the sake of brevity the WLS results are reported in Table 3 (the OLS results are not reported). A detailed discussion of the WLS results based on Table 3 is provided below.

Table 3 The WLS Regression Results with correction for spatial autocorrelation
Full size table

The regression coefficients of the WLS regression results with correction for spatial autocorrelation are reported in Table 3 with their t-statistics (next to them). As can be seen in Table 3, the adjusted coefficient of determination (R2) for Models 1 and 2 are 0.79 and 0.81, respectively. These are reasonable compared with much of the hedonic literature. An examination of variance inflation factors, tolerance levels and the correlation matrix (not reported in Table 3) reveal no obvious signs of multi-colinearity.

First on the control variables for property characteristics, the following variables; Ln(LOT); SQFT; (SQFT)2; AGE; (AGE)2; EXCL; GOOD; NSTORS; POOL; GARG; BSMT; FIREPL; DECK; AC, and NBATHS; are all significantly different from zero at conventional levels with expected signs. The variables NROOMS; and POOR are statistically insignificant. It must, however, be noted that the partial effects due to living area is already accounted for with the inclusion of NSTORS, SQFT, (SQFT) 2 and BATHS.

Of the control variables for location, only the dummy variable GWYVL for Gwynedd Village is significantly positive at conventional levels. The other location dummy variables: GWYDD; SPRNHS; OTHER, and R5TRAIN; are all statistically insignificant. The variable for distance to the township center (TBD) is also not statistically significant. The STREET dummy variable, however, is significantly positive as to be expected at conventional levels. Relative to the first quarter of 2005 (QRT01), all the quarterly variables from QRT02 through QRT13 (at the start of 2008) are statistically insignificant at conventional levels. However, the estimated coefficients of QRT14, 16, 18, 19 are significantly negative. These generally negative coefficients towards the end of the study period are consistent with the overall negative outlook of US real estate markets towards the end of the study period (after 2008).

Now turning on the variables of interest, the estimated coefficient of cluster (CLUS) is significantly positive as expected at the 99 % level of confidence. The magnitude of the estimated coefficient on CLUS in Models 1(without the open space variable (OPEN)) is 0.038. However, part of this price premium is attributable to the permanent open spaces which are parts and parcels of clusters. When the variable for open space (OPEN) is introduced as shown in Model 2, the cluster (CLUS) premium drops from 3.9 % to 2.02 % suggesting the relative importance the permanent open spaces. The open space variable (OPEN) is associated with a premium of as much as 5.2 %, on average.

The dummy variables for the two types of clusters {CLUS (AA-1) and CLUS (A-1)} were employed at earlier runs of the model (not reported). However, they both proved to be statistically insignificant at conventional levels. The estimated coefficient on development density (DENSITY) is significantly negative at conventional levels. In other words, density of development, per se, has adverse impacts on value as to be expected. The next section presents the “Conclusions” of this study.

Conclusions

The major finding of this study is that cluster developments will produce higher home values, ceteris paribus. The estimated coefficient on CLUS is significantly positive at the 99 % level of confidence with magnitude of 0.038. However, when the open space variable (OPEN) is introduced (as shown in Models 2) the magnitude of the estimated coefficient of cluster (CLUS) drops from 0.038 to 0.020 suggesting the relative importance of the permanent open spaces in cluster developments. The estimated coefficient on open space (OPEN) is significantly positive at the 99 % level of confidence as shown in Model 2 with a magnitude of 0.051. The estimated coefficient on density of development (DENSITY) is also statistically significant in both models with negative signs suggesting the adverse nature of raising densities, per se, in a sub-urban setting where market densities are lower than permitted densities. All the control variables work as expected with predictable results. The findings of this study provide empirical support for organic, green-by-design, residential development.