Abstract
Water quality regulation continues to be controversial, as demonstrated by recent litigation between the U.S. Environmental Protection Agency and the state of Florida over nutrient standards. While the costs of standards are usually known, benefits may be diverse and difficult to identify. This study investigates the effects of enhanced water quality on both waterfront and nonwaterfront property prices, using hedonic models within an urban market. Findings indicate (1) the value of increased water quality depends upon the property’s location and proximity to waterfront, and the surface area of the water body; and (2) aggregate benefits to nonwaterfront homes may dominate those realized by waterfront homeowners. (JEL Q51, Q53)
I. Introduction
The Clean Water Act (CWA) has directed billions of dollars into the management of U.S. surface waters since 1972. Despite these expenditures, nearly 50% of rivers and streams and 66% of lakes and reservoirs in the United States were classified as impaired for one or more uses in 2009 (US EPA 2010a). Section 303© of the Clean Water Act (CWA) directs states to adopt water quality regulations to protect designated uses of water bodies, but relatively little is known about the potential benefits because the impacts are diverse, nonmarket in nature, and require site-specific information. The benefits derived from recreational usage of water bodies (swimming, fishing, and boating) have typically been estimated with travel cost models and contingent valuation surveys (e.g., Morgan and Owens 2001; Iovanna and Griffiths 2006; Van Houtven, Powers, and Pattanayak 2007; Viscusi, Huber, and Bell 2008; Egan et al. 2009). While the open-space amenity value of proximity to surface waters in residential housing markets has been well documented (e.g., Palmquist and Fulcher 2006), the property price effects of improvements in the aesthetic, recreational and ecosystem services related to enhanced water quality have received limited attention in policy studies of water management programs.
The use of hedonic property models for measuring the price effects of changes in environmental quality has been extensive (Palmquist 2005). In the case of water resources, several studies over the years have reported that water quality significantly affects waterfront property prices. However, the methodology remains underutilized for analysis of water management programs, as noted by Palmquist and Smith (2001) nearly a decade ago. One conjecture is that the size and extent of the housing market impacts are small by comparison to the surplus effects realized by the population of recreational users (Olmstead 2010). Thus, it may suffice to consider the latter in assessing the benefits of water management programs, though there is little evidence supporting or rejecting this conjecture in the literature. While recreational use benefits may dominate in rural areas with low housing density, this relationship is more uncertain in urban housing markets.
Concerns regarding the spatial distribution of the welfare effects of pollution abatement in residential housing markets are analogous to those of market extent (or scope) in the recreation demand and contingent valuation literatures (Smith and Kopp 1980; Smith 1993). Smith (1993) observed that the definition of market extent is more important to environmental valuation than the refinement of per unit values. The stated preference literature has long considered the impact of distance on use and nonuse values (Sutherland and Walsh 1985). For instance, Pate and Loomis (1997) find that distance affects willingness to pay for public goods in a contingent valuation survey. Recent literature has explored benefit extent through distance-decay functions (Hanley, Schlapfer, and Spurgeon 2003; Bateman et al. 2006), the use of maps in surveys (Vajjhala, John, and Evans 2008), and referendum voting results (Deacon and Schiapfer 2010). Similarly, the spatial distribution of welfare effects has been an important theme in hedonic property studies of air quality (Boyle and Kiel 2001; Bayer, Keohane, and Timmins 2009) and proximity to hazardous waste sites in residential housing markets (Mendelsohn et al. 1992; Jackson 2001; Deaton and Hoehn 2004; Cameron 2006). If the property price effects of pollution abatement span beyond the waterfront, then measuring their spatial extent is critical for measuring the aggregate benefits of water management programs.
Several questions arise regarding the spatial extent of the benefits of water pollution abatement in residential housing markets. The present study formalizes three of these as hypotheses that are tested through hedonic price models. First, are there edge effects from improvements in water quality, whereby waterfront properties benefit from their unique location on the boundary separating the water body from developed areas? Second, are there proximity effects, whereby the marginal value of water quality diminishes as distance to the affected water body increases? And if so, then how far do the effects extend into surrounding areas? Third, are there area effects, whereby the price effects of abatement efforts are dependent upon the surface area of the water body? While these remain open issues in the literature, the generation of edge, proximity, and area effects through public expenditures on water quality management may have broader policy implications for state and federal efforts to establish water quality standards and the funding for these programs.
Estimating the marginal value of surface water quality and testing for edge, proximity, and area effects require data on a relatively large number of water bodies and sufficient variation in water quality over the study area and sample period. However, in moving beyond the waterfront there must also be sufficient variation in the proximity of properties to the waterfront and of the size of the lakes within the housing market. To satisfy these data requirements, the present study employs a unique dataset containing more than 54,000 property sales distributed around 146 monitored lakes in a metropolitan housing market (Orange County [Orlando], Florida) over the period 1996-2004.
II. The Hedonic Valuation of Water Resources
The hedonic pricing model was formalized in early work by Griliches (1971) and Rosen (1974) and has been refined over the years for characterizing the prices of competitively traded heterogenous goods (Ekeland, Heckman, and Nesheim 2004; Parmeter and Pope 2009). In the case of residential housing markets, the goods are a composite of the attributes of the physical property and surrounding landscape that can be exchanged at prices reflecting aggregate market conditions. Hedonic models have a long history for estimating the effects of spatially distributed environmental goods and bads on residential property values, spanning from air and water pollution to hazardous waste to the recreational access and open-space aesthetics provided by public lands and water resources; see Boyle and Kiel (2001), Palmquist and Smith (2001), and Palmquist (2005) for reviews.
Two themes have prevailed in hedonic studies of water resources. The first is that, similar to air pollution, the degradation of surface waters can depreciate surrounding property values. Research efforts have mostly focused on estimating the benefits of water quality for waterfront properties on lakes and rivers. David (1968) was the first to investigate the relation between the sales prices of lakefront properties and subjective, qualitative ratings of water quality (poor, moderate, and good) in Wisconsin and found mean prices were positively related to higher water quality. Subsequent studies by Brashares (1985), Michael, Boyle, and Bouchard (2000), Poor et al. (2001), Gibbs, Halstead, and Boyle (2002), Krysel et al. (2003), and Horsch and Lewis (2009) all reported that lakefront property prices were positively related to various measures of water quality, with water clarity (Secchi disk) the most commonly used indicator.1 Similar relationships were reported by Epp and Al-Ani (1979) and Leggett and Bockstael (2000) for riverfront property prices, although the water quality indicators were more diverse, reflecting the broader range of pollutants, sources, and political jurisdictions (Sigman 2002) that may impact these water bodies.
The second theme in the literature is that surface waters provide open-space, recreational use, and aesthetic values that are capitalized into the sales prices of residential properties in proximity to a water body. Brown and Pollokowski (1977), Milon, Gressel, and Mulkey (1984), Lansford and Jones (1995), and Anderson and West (2006) all report positive amenity values from proximity to a water body, and these values may extend several hundred meters into the surrounding neighborhood. Palmquist and Fulcher (2006) and Cho et al. (2009) also report positive price effects for proximity to a water body, and examine temporal changes in these prices. Palmquist and Fulcher estimated relatively consistent price effects over time, whereas Cho et al. found increasing prices.
Even though 40 years of literature has consistently shown a positive capitalization of surface water proximity into property prices, hedonic studies traditionally fail to examine the spatial distribution of the marginal value of water quality. The few exceptions relied on water quality data that were aggregated over sites so that individual differences across water bodies could not be identified. Dornbusch and Barrager (1973) evaluated the impact of pollution abatement programs on residential property values in communities adjacent to four highly polluted water bodies in four states over the period 1960-1970. Their empirical analysis indicated that the vast majority of benefits occurred within 600 to 900 m from the waterfront, but they suggested that benefits could extend up to 1,200 m. Poor, Pessagno, and Paul (2007) estimated a hedonic model using property sales in a county adjacent to the Chesapeake Bay. Marginal implicit prices were significantly related to ambient pollutant concentrations (total suspended solids and dissolved nitrogen) across the watershed. Phaneuf et al. (2008) developed a model combining stated and revealed preferences of residential homeowners for urban surface water quality using a “recreation access index” adjusted for ambient water quality within a watershed as an explanatory variable in the hedonic model. Findings for a sample of North Carolina homeowners indicated that the mean price of nonwaterfront properties was significantly related to recreational access. None of these studies, however, identified the differences in water quality effects for waterfront and nonwaterfront properties or how the water quality effects vary with proximity to specific water bodies within the same urban community.
While there is evidence that pollution abatement can have wide-reaching effects on residential property prices, the potential impacts of water quality changes have not been fully evaluated using the hedonic pricing model. The results from such applications could have significant practical information for benefit-cost analysis of water management programs in compromised watersheds. A timely example is the November 2010 rule by the U.S. Environmental Protection Agency (US EPA) to establish numeric nutrient standards for surface waters in the state of Florida to meet the requirements of Section 303© of the Clean Water Act (US EPA 2010b).2 The state of Florida, like many other states, has a large proportion of lakes, rivers, and estuaries that are impaired by nutrients that impact aquatic life, ecosystem health, and the various benefits provided by these water resources (FDEP 2008; State-EPA NITG 2009). Understanding the spatial extent and magnitude of benefits from water quality improvements in specific water bodies is necessary to evaluate alternative standards and to design strategies to fund implementation programs.
III. Hypotheses and Model Specification
This study focuses on the relationship between water quality in lakes and the related property price effects in the metropolitan Orange County, Florida, housing market. In this section, the hypotheses are presented, and hedonic property price models that nest the hypotheses are specified for estimation and inference in the next section. The hypotheses about the mean marginal implicit value of water quality are as follows:
H1: The edge effect. The value of water quality differs between waterfront properties and properties located off the waterfront, other factors held constant.
H2: The proximity effect. The value of water quality diminishes as the property distance to the waterfront increases, other factors held constant.
H3: The area effect. The value of water quality is an increasing function of the surface area of the water body, other factors held constant.
To specify the hedonic property model and related implicit prices, we assume households are freely mobile to choose housing bundles that maximize utility over structural characteristics, environmental amenities, and other locational attributes within the Orange County housing market (Kuminoff 2009). To allow the implicit price of water quality to vary between properties and between water bodies, water quality enters the property price model through multiple interaction variables. Also, water bodies are defined as naturally occurring lakes that vary in terms of surface area, private or public access, recreation facilities, or other lake-specific amenities. The model is written as
[1]where Price references the sales price; Lake-front is a dummy variable distinguishing lakefront from nonlakefront properties; WQ is a continuous measure of water quality in the nearest lake at the time of sale; Distance measures the property's proximity to the nearest lake; and Area is the surface water coverage of the nearest lake. The terms X and Z are vectors of property attributes and landscape attributes, and L and T are vectors of dummy variables to control, respectively, for unobserved fixed effects of individual lakes and time periods.
The model in equation [1] appears in double-log form. While there is no theoretical basis for the functional form, Cropper, Deck, and McConnell (1988) suggest that semilog and double-log models are superior to more complicated models in the face of misspecification and proxy variables. Both regular Box-Cox tests and spatially robust Box-Cox tests using double-length regression tests (Le and Li 2008; Le 2009) favored the double-log specification appearing in expression [1].
The included lake dummy variables (L) should capture lake-specific characteristics that do not change over time. The main tradeoff with these variables is that the lake area variable cannot be used alone in the regression (although it still appears as an interaction term), since it does not vary. The fixed effects will capture other lake-specific characteristics that do not vary, but there is a risk that the area-water quality interaction may overstate the impact of lake area on the value of water quality.
The marginal implicit values of the property attributes are obtained by partially differentiating [1]. In the case of water pollution, the desired value is the change in the expected property price caused by a change in water quality. The expected marginal value may be written as
[2]The implicit value is a function of the hypothesized edge, proximity, and area effects. The edge effect hypothesis involves a one-tailed test of the null H0: β3 = 0. The alternative hypothesis β3 > 0 indicates positive edge effects exist, and other factors constant, the lakefront premium of a unit increase in water quality is equal to β3Price∕WQ. The proximity and area effects depend, respectively, upon β5 and β7. Partially differentiating [2], the size of the proximity effect at a given distance from the nearest lake may be expressed as
[3]The proximity effect involves a one-tailed test of the null H0: β5 = 0. Positive proximity effects exist and the marginal value diminishes at a nonconstant rate as distance to the waterfront increases if β5 < 0. Similarly, the effect of surface water area on the marginal implicit value may be written as
[4]The area effect involves a one-tailed test of the null Ho: β7 = 0. Positive area effects exist and the marginal value increases with water body size if β7 > 0. Given available data, the parameters and covariance matrix may be estimated in order to test the three hypotheses and to estimate the relationships between water quality and property prices summarized in [2] through [4].
The final modeling issue is potential spatial autocorrelation, which can be induced by unobserved neighborhood and landscape attributes common to property sales located within close proximity to one another. This concern has been raised in several hedonic property studies, with spatial lag and spatial error econometric models estimated to test and correct for spatial dependence in the errors.3 The findings are mixed, with some studies reporting improvements in fit (Gelfand et al. 2004; Case et al. 2004) and others reporting results that are robust to the error specification (Leggett and Bockstael 2000; Kim, Phipps, and Anselin 2003; Muller and Loomis 2008). For completeness, estimation is performed under the null of zero spatial correlation and with the spatial lag specification. An inverse distance-based spatial weights matrix required for estimation of the spatial lag model is defined for property sale i as all sales located within 200 m of i that occurred 6 months before or 3 months after the sale of i.4
IV. Data and Estimation Results Data
The urban housing market is defined as Orange County (Orlando), Florida. The county contains more than 200 natural lakes, and more than 100,000 property sales occurred over the 1996-2004 sample period.5 The sales data were obtained from the Orange County Property Appraiser (OCPA) and were restricted to single-family residences. The data include detailed property information, such as historic sales prices, dates of construction and permitted improvements, the sizes of the structure and parcel, and the number of bedrooms and bathrooms. The data were geocoded and overlaid in GIS with a shape file from the Orange County Environmental Protection Division (OCEPD) defining all lakes in the county. Lakefront and nonlakefront properties were identified, and the distance between the centroid of each property and the edge of the nearest lake was measured.
Due to the prevalence of water bodies in Orange County, lakes are expected to have a relatively local impact. There is a trade-off between including all homes potentially affected by water quality changes and including irrelevant homes that contribute noise to the analysis (Ihlanfelt and Taylor 2004). Palmquist and Fulcher (2006) assert that property values farther than a half mile (750 m) from a lake probably depend on more general amenity considerations than access to a distinct lake, and should not be considered. To address this boundary issue, the present paper examined several lake distance thresholds. Results indicated that the precision of the estimates improved as the boundary expanded up to 900 to 1,000 m.6 Furthermore, the estimated implicit price of water quality declined continuously over this distance. Consequently, all single-family residential property sales located within 1,000 m (1 km) of a lake (35% of all residential sales) were retained for estimation, yielding a final dataset containing 1,496 lakefront and53,216 nonlakefront sales around 146 natural lakes.
The large number of lakes in the study area raises the possibility that lakes other than the nearest one may also influence the value of a property. A lake count variable was constructed that indexed the number of lakes within a particular distance from each property. Similar count variables have been used to control for proximity to multiple sites in the hazardous waste literature (see Ihlanfeldt and Taylor 2004). This variable was constructed for boundaries of 1,000 and 1,500 m. In both cases the variable was insignificant in all specifications and was omitted from the final estimation.7
The water quality data were obtained from the OCEDP and three municipalities. The data contain annual average concentrations of nitrogen, phosphorous, and dissolved oxygen and the Secchi depth of the lakes. Consistent with several studies discussed in Section II and for ease of interpretation, water quality is proxied by Secchi depth.8 The water quality data were merged with the sales data, and properties that sold in a given calendar year were assigned the mean annual Secchi depth at the nearest lake.9 Although education districts are sometimes included in hedonic models, direct measures of school quality or districting were not available (particularly since local “school choice” programs allow mobility, and grade and high school boundaries differ significantly). Lastly, the data were merged with block-level sociodemographic variables from the 2000 Census to control for attributes of the neighborhood and surrounding landscape.
Summary statistics on the property sales data are reported in Table 1. Lakefront and nonlakefront properties on average had similar numbers of bedrooms and bathrooms, but the lakefront properties were larger and older and sold for greater amounts. The mean waterfront sale was located on a lake covering about 520 acres, whereas the mean nonwaterfront sale was located about 470 m from a lake covering about 280 acres.10 In both cases the mean sale was located about 5.5 miles from downtown Orlando. The geo-coordinates of the properties are also included as regressors. As demonstrated by Cameron (2006), these control for property location within the housing market and also allow location effects to be tested. Neighborhood controls include the block-level median household income and proportions of the population that were Caucasian, African-American, and over 65 years of age. Lastly, Table 1 summarizes the distribution of sales over the sample period; overall, similar proportions of lakefront and nonlakefront properties sold annually.
Summary Statistics for Residential Property Sales in Orange County, Florida: Lakefront and Nonlakefront Homes (1996-2004)
Summary statistics on the lake and water quality data are summarized in Table 2. The number of lakes varies annually from 125 to 146, reflecting missing water quality records at several lakes.11 The majority of the lakes are public, as indicated by local fishing web sites, but the line between public and private access is not easily determined, due to stream and canal connections, private and public boat ramps, neighborhood access easements, and other factors. Both public and private lakes allow a range of recreational activities including water skiing, fishing, and boating. The mean annual Secchi depth in the sample lakes is consistently about 5 ft over the period and ranges from about 1 ft to more than 18 ft. Table 2 also summarizes the composition of the lakes by their surface water acreage. Similar to Secchi depth, the area covered by the lakes varies considerably. The average lake covered about 250 acres, and the lakes range in size from about 1 acre to more than 1,800 acres.
Summary Statistics for Orange County Lakes over the Sample Period
Estimation Results
Maximum likelihood estimates of three versions of the hedonic property price model specified in [1] are reported in Table 3.12 The first (Model 1) includes water quality and property distance as separate regressors and imposes exclusion restrictions on the proximity and area effect coefficients (β5 = β7 = 0). The second (Model 2) relaxes the proximity effect restriction by adding the interaction variable ln(Secchi Depth)*ln(Distance) but maintains the restriction upon the area effect (β7 = 0). And the third specification (Model 3) adds the area effect interaction variable ln(Secchi Depth)*ln(Area) for the unrestricted version of the model expressed in [1]. The respective specifications of the spatial lag models are referenced by Model 1S, Model 2S, and Model 3S in Table 3.13Likelihood ratio tests (LeSage 1999) rejected the hypothesis of no spatial dependence.
The results in Table 3 indicate that the coefficient estimates are robust to the error specification, and the spatial lag coefficient (ρ) is small but significant in Models 1S-3S. The estimated Lakefront coefficients indicate a significant price premium for location on the waterfront. Considering the interaction variable ln(Secchi Depth)*ln(Lakefront), the null of zero edge effect H0 β3 = 0 is rejected at the 1% level in all cases, indicating that the mean marginal value of surface water quality differs significantly between lakefront and nonlakefront properties. The proximity effect hypothesis entails a test of the null H0 β5 = 0 associated with the interaction variable ln(Secchi Depth)*ln(Distance) in Models 2 and 3. As shown in Table 3, the estimated coefficients are negative across all model specifications, and the null of zero proximity effect is rejected at the 1% level in all cases. Note also that the distance coefficients alone are negative and significant in Models 2 and 3, indicating that both proximity to a lake and the water quality in the lake are important. Finally, the area effect hypothesis for the null H0: β7 = 0 associated with the variable ln(Secchi Depth)*ln(Area) is rejected at the 1% level.14
Selected Hedonic Estimation Results
In addition, information at the bottom of Table 3 indicates the importance of controlling for individual lake- and time-specific effects across model specifications. For example, 112 (91) of the 145 lake-dummy coefficients and all eight of the time-dummy coefficients were significant in Model 3 (Model 3S) at the 1% level.15 Table 3 also shows that property location within the county can affect property prices, as indicated by the significant ln(Latitude) and ln(Longitude) coefficients. The location variables reflect the presence of non-central business district commercial nodes in the western portion (Disney World) of the county and in the north of the county (Winter Park/Maitland). Collectively, these results strongly indicate that water quality effects on property prices are dependent on the locational and physical characteristics of each water body.
V. The Implicit Value of Urban Water Quality
The estimation results indicate that the amenity value of water quality depends on the type and location of the property and the area of the surface water. In this section, the models are used to estimate the implicit value (marginal willingness to pay [MWTP]) of increased water quality for representative lakefront and nonlakefront properties and the aggregate benefits for all properties surrounding three lakes. In the process, the lakefront premium and the proximity and area effects of water quality are estimated.
Estimates of the expected marginal value16 of a 1 ft increase in Secchi depth (approximately 17% from the mean) are obtained from Models 3 and 3S by substituting the sample means17 of the independent variables and estimates of the respective coefficients and of E(Price∖X) into expression [2]. Given the double-log specification, 
. Estimates are obtained using the smearing approach proposed by Duan (1983).18
Beginning with the edge effect in Table 4, a unit increase in Secchi depth is associated with about a $5,500 (or 1 .24%) increase in the price of the mean lakefront property versus about a $700 (or 0.36%) increase in the price of the mean nonlakefront sale. Thus, whether viewed in absolute or relative terms, enhanced surface water quality has a notably larger impact on lakefront property prices.
The mean proximity effect of a 1 m increase in distance to the waterfront on the marginal implicit price was estimated in 200 m increments from the waterfront. As reported in Table 4, the mean effect realized by properties located immediately beyond the waterfront was small in absolute and relative terms compared to waterfront properties and diminishes rapidly as distance from the waterfront increases. For example, the mean implicit price decreases by about one-fourth in moving from 100 to 200 m from the waterfront. At 600 m the implicit value has fallen by more than 50%, and at 1,000 m it is about one-sixth of the value. Thus, while there is a significant distance gradient, positive price effects extend hundreds of meters into the surrounding community.19
Estimated Mean Marginal Implicit Prices of Increased Water Clarity (2000 dollars)
The mean area effect of a 1 acre increase in lake area on the marginal implicit price can also be evaluated. The final section of Table 4 compares the estimated mean implicit price between lakefront and nonlakefront properties surrounding lakes covering 100 acres and 1,000 acres. Results for waterfront properties indicate that a marginal change in lake area has a small positive effect on the implicit value of water quality: a 10-fold increase in lake size is associated with about a $1,000 (or 20%) increase in the marginal implicit price. In contrast, relatively larger area effects are found to be realized by nonwaterfront properties, with the mean implicit price differing about $700 (or 300%) between the 100 acre and 1,000 acre lakes.
Overall, enhanced surface water quality positively impacts the price of properties located throughout the metropolitan area. Although the price effect for the mean sale occurring off the waterfront is smaller relative to the mean waterfront sale, residential housing surrounding water bodies can be relatively dense in urban watersheds. It follows that the aggregate benefits could in fact exceed those realized by waterfront properties. To illustrate this possibility, the aggregate effects of a 1 ft increase in Secchi depth are estimated for three representative lakes that vary in size and water clarity: in2004, Lake Silver covered 70 acres and had a Secchi depth of 5.8 ft; Lake Mann covered 271 acres and had a Secchi depth of 2.2 ft; and Lake Conway covered 1,625 acres and had a Secchi depth of10.5 ft. The location of all lakefront properties and nonlakefront properties located within the 1,000 m boundary around each lake was identified, and the distance between each property and the respective lake was measured using the geo-coded tax role described in Section III. The marginal implicit price of a 1 ft increase in Secchi depth was estimated for each property using Model 3S and then summed across properties to estimate the aggregate benefits.20
Aggregate Property Price Effects of a 1 ft Increase in Secchi Depth at Three Selected Lakes
Table 5 describes the characteristics of the lakes and the number of homes in the surrounding community and reports estimates of the aggregate price effects. For comparison, estimates are reported separately for lakefront properties and nonlakefront properties located within 500 m and 1,000 m of the respective lake. The number of lakefront properties around each lake is a small fraction of the total properties; for example, about 5%, 8%, and 16% of all properties located within 500 m of Lakes Mann, Silver, and Conway, respectively, are lakefront properties. Despite the small fraction of total properties, the aggregate benefits for lakefront properties are comparable to the cumulative benefits for all properties within 500 m at Lake Silver and greater in Lake Conway, but they are only one-third of the aggregate benefits within 500 m at Lake Mann. At 1,000 m, however, the estimated benefits derived collectively by the nonlakefront properties exceed the lakefront benefits in all three lakes. Thus, the individual benefits of water quality improvements vary within the urban community, and the total benefits could be considerably underestimated if only waterfront properties are defined as the beneficiaries of pollution abatement programs.
VI. Discussion and Conclusions
Water quality remains a significant public concern in the United States despite longstanding federal and state laws to regulate and control pollution from point and nonpoint sources. A sizable percentage of all water bodies are classified as impaired, but the full extent of the problem is not known because only one-quarter of all rivers and streams and approximately 40% of lakes and reservoirs have been assessed (US EPA 2010a). Recent efforts to expand and strengthen water quality improvement programs using total maximum daily loads and numeric water quality standards (e.g., State-EPA NITG 2009) are hindered by the limited information on the potential benefits for specific water bodies. The benefits of improved water quality accrue in the form of enhanced recreation, aesthetic enjoyment, and ecosystem services. In past literature, it was assumed that water quality benefits were enjoyed primarily by waterfront property owners and recreational users. This article explored the extent of property owner benefits by including both lakefront and nonlakefront homes in a large spatial hedonic analysis across multiple lakes in a metropolitan setting.
Three hypotheses about the effect of water quality on surrounding residential properties were tested using water clarity (Secchi disk) as the indicator of water quality. First, a waterfront edge effect supported previous studies that identified an increasing price premium for residing next to a water body with cleaner water. Second, the proximity effect established that property prices around a water body reflect the value of residing both closer to the water body and closer to water with a higher quality. This result should not be surprising, but previous research has not directly tested the effect of water quality on the value of proximity. Third, the area of a lake also affects the implicit price, with water quality in larger lakes valued more than water quality in smaller lakes. Overall, the results of this article provide considerable evidence that nonlakefront property owners are impacted by water quality changes in nearby water bodies, and these impacts should be integrated in future hedonic property analyses in which proximity to surface waters is an environmental amenity.
These results also provide several important policy implications. One of the main objectives of environmental amenity valuation is to inform benefit-cost analysis. In two of the three examples of the total benefits of water quality improvements (Table 5), including the gains to nonlakefront properties would more than double the estimate of water quality benefits. The nonlakefront component represents a considerable share of the total benefits and indicates that restricting the analysis to waterfront homes alone would understate the value of water quality improvements for the surrounding community.
Furthermore, the shape and magnitude of the implicit price gradient for water quality could be used to design more effective funding mechanisms for water quality management programs. These programs typically require sources to incur costs to limit emissions into a water body or rely on general revenues to provide financial incentives (State-EPA NITG 2009). In the case of water bodies with surrounding residential communities, property tax levies that reflect underlying implicit prices for water quality improvements could be used to fund management programs. For example, one taxing mechanism for lake improvement activities in Orange County, Florida, is a municipal service taxing unit (MSTU) fee as part of the property taxes for each home in a neighborhood surrounding a lake. These MSTU fees are similar to recycling collection and fire protection service fees and are instituted through a majority vote by the neighborhood. If approved, every home in the MSTU faces the same ad valorem property tax increase.21 If the MSTU is smaller than the surrounding community receiving benefits from water quality improvements, there is the problem of free-riding by those outside the MSTU. If the MSTU is large, those at the periphery may be disproportionally taxed in relation to the differential that could be attributed to property price appreciation from improved water quality. It would be more efficient to tax homes proportional to the underlying implicit values for water quality improvements based on lake proximity and waterfront status. A more efficient tax structure may also increase participation in MSTU programs.
Alternatively, property price appreciation from improved water quality will yield additional municipal property tax revenues if appraisal values reflect market conditions and tax rates do not change. The incremental tax revenues provide a funding source for water quality management programs and suggest that these programs may be self-financing in metropolitan settings where the welfare gains are spread across the community.
While this article provides new evidence on the spatial distribution of water quality benefits in an urban setting, two concerns remain that should be addressed by future research. First, this study used water clarity as the only indicator of water quality. Other indicators that are more often used for setting water quality standards, such as nutrient levels or trophic state, may imply different implicit price gradients and benefits (e.g., Walsh 2009, Ch. 2). Second, urban housing markets are complex, multidimensional surfaces that can be approximated only by parametric functional forms and price gradients. Future research should consider alternative estimation methods that address the interplay between spatial amenities and housing preferences, as well as the potential sorting of home owners across neighborhoods as a response to the presence of water bodies with different levels of water quality.
APPENDIX A
Distribution of N = 53,216 Nonlakefront Property Sales within 1,000 m of a Lake
APPENDIX B
Coefficient t-Statistics and Alternative Boundaries
Footnotes
↵1 Secchi depth is a measure of surface water clarity (or transparency) obtained when a trained technician lowers a Secchi disk into the water body and records the level at which the disk disappears from sight.
↵2 Separate water quality criteria were proposed for lakes, streams, springs and clear streams, and canals in four Florida regions. Criteria for lakes include limits on nitrogen, phosphorous, and chlorophyll a (US EPA 2010b).
↵1 The spatial lag model is written Y=ρWY + Xβ + u, and the spatial error model is written Y=Xβ + ε, where Y is the dependent variable, W is the “inverse distance” spatial weights matrix, X is a vector of regressors, u is iid normal random error, ε=λWε + u, and β, ρ, and λ are parameters. Estimates of ρ and λ may be used to test for spatial dependence in the errors. See Anselin (1988) and LeSage (1999) for technical details on spatial econometric models.
↵1 Several definitions of the weights matrix were investigated by varying the spatial (distance) and temporal components; changing either affects the number of nearest neighbors around each property. For the weights matrix used in the empirical application, the mean property has 5.6 nearest neighbors. This specification approximates local appraisal practices.
↵1 Orange County covers 627,723 acres, of which 168,276 acres (or 26.8%) were designated as residential development and 60,069 acres (or 9.6%) were designated as public waters (lakes, ponds, and rivers) in 2004 (see Milon, Scrogin, and Weishampel 2009).
↵6 For instance, Appendix B contains a graph of the estimated t-statistics of the relevant variables as the boundary expands.
↵6 We thank an anonymous reviewer for suggesting this procedure.
↵6 Water pollution in the lakes is the product of nonpoint pollution sources such as stormwater runoff because there are no point sources, such as adjacent sewage treatment plants or factories, that emit into any of the 146 lakes (FDEP 2008).
↵6 Past papers have used alternatives to mean water quality, for instance Gibbs, Halstead, and Boyle (2002) use the minimum annual clarity. Michael, Boyle, and Bouchard (2000) investigated several different measures of water clarity representing a variety of assumptions about perceptions, including historical minimums and maximums. Their results do not strongly support one measure over the others. For the present paper, mean annual clarity was compared against maximum and minimum annual values in the analysis. The mean values performed the best in terms of variable significance and goodness of fit of the regressions. Furthermore, there was inconsistent frequency and timing in sampling between lakes, preventing the use of an indicator with duration shorter than one year. Also, it is important to be mindful that individual preferences may differ over clarity measures, so any one will be an imperfect measure.
↵10 Distance was not set equal to zero in lakefront properties to account for differences in the length of waterfront plots. Some waterfront plots are long and narrow, with the home as far as 100 m from the lake, while others are right next to the lake. The distribution of the 53,216 nonlakefront property sales located within 1,000 m of a lake are summarized in Appendix A.
↵11 Missing water quality data for individual lakes were most often due to irregular sampling schedules, budget cuts, or incomplete records. For comparison, estimation was also performed with the subset of 100 lakes that appears in all years of the sample. The results were similar to those obtained with the complete set of lakes. Estimation results for the two samples are available in an appendix upon request.
↵12 In all models, waterfront and nonwaterfront properties were pooled. Properties in both groups share many unobservable neighborhood characteristics, so separating the data would severely restrict the spatial regression analysis. The models were estimated in Stata (StataCorp 2009) and Matlab (LeSage 1999).
↵13 The normal approach for testing between the spatial error and lag models involves normal and robust versions of the Lagrange multiplier (LM) test. The size of the present data, however, does not allow the LM test to be used. Instead, out-of-sample forecasting was used to choose between models. Seven error statistics from Case et al. (2004) were used to evaluate the fit of both models. The spatial lag had lower error in six of the seven error statistics and was therefore chosen. The error variables include mean and median prediction error, the standard deviation of the errors, root mean and median square error, and mean and median absolute value percent error. See Walsh (2009) for additional details.
↵1 Similar results were found for the edge, proximity, and area effects, with other water quality indicators including the trophic state index, total nitrogen, and chlorophyll a (Walsh 2009). A full discussion of other water quality measures is beyond the scope of the current paper.
↵6 The full set of estimation results for the hedonic property price models is available in an appendix upon request.
↵1 The marginal implicit prices from the spatial lag model are obtained by scaling [2] by 1/(1 — p); see Kim, Phipps, and Anselin (2003) for technical details. In all cases, the standard errors of the estimated implicit prices were obtained with the delta method. For comparison, estimates from Models 1 and 2 and Models 1S and 2S are reported in Appendix B.
↵6 Mean values are used to compute implicit prices in the majority of past hedonic water quality studies, including those of Poor et al. (2001), Gibbs, Halstead, and Boyle (2002) (who use minimum Secchi disk but mean lake area in the implicit price), Krysel et al. (2003), and Poor, Pessagno, and Paul (2007). An alternative would be the use of the median sale price, which is smaller for both the waterfront homes and nonwaterfront homes. The Duan (1983) predicted values, however, fall between the mean and median values, reducing the impact of outliers.
↵6 Duan (1983) demonstrated that if ε is assumed to be independent and identically distributed, then E(eε ) may be consistently estimated by
, where are the residuals.↵19 Several alternative functional forms were also estimated to evaluate the shape of the implicit price function, but the qualitative results were the same.
↵20 Given the double-log specification, the implicit price of a marginal increase in water quality is a function of property price, as shown in [2]. To calculate the implicit price for each property within the 1,000 m boundary, the 2004 assessed values from the OCPA were used. The state of Florida requires that assessed value be an annual determination of the fair market value (Florida Statutes, Ch. 192.001).
↵21 As of 2004, only 10 of the 146 lakes included in this study had MSTUs.
References
In this issue
Jump to section
Related Articles
Cited By...
- Spatial Heterogeneity in Hedonic Price Effects for Lake Water Quality
- The Impact of Water Body Restoration on the Sales and Rental Markets in Beijing
- Property Values, Water Quality, and Benefit Transfer: A Nationwide Meta-analysis
- Using Auxiliary Population Samples for Sample-Selection Correction in Models Based on Crowd-Sourced Volunteered Geographic Information
- Convergent Validity of Satellite and Secchi Disk Measures of Water Clarity in Hedonic Models
- Convergent Validity of Satellite and Secchi Disk Measures of Water Clarity in Hedonic Models
- The Impacts of Harmful Algal Blooms and E. coli on Recreational Behavior in Lake Erie
- An Alternative to the Standard Spatial Econometric Approaches in Hedonic House Price Models
- What Do Property Values Really Tell Us? A Hedonic Study of Underground Storage Tanks