Abstract
We estimate hedonic valuation models of local open space separately for 215 cities in the eastern United States and derive city-specific marginal willingness to pay (MWTP). We then examine variation in MWTP and city-level determinants. Valuation is largely local–relatively large changes in income or existing conservation lead to modest changes in MWTP–suggesting validity of benefit transfer across regions. However, geographic features that naturally limit development correlate with MWTP. As a result, we examine geographic features as instrumental variables and find that on average steep slope and water/wetlands yield valuation coefficients of opposite sign, consistent with a local average treatment effect interpretation
1. Introduction
Every day an estimated 6,000 acres of open space are converted for other uses, such as urban and suburban expansion (U.S. Forest Service n.d.). The forfeiture of these lands for development may assuage the needs of a growing population but lead to a loss of benefits to the individual and surrounding neighborhoods. Open space provides benefits from recreational and visual amenities as well as other ecosystem services, like improved air and water quality (U.S. Forest Service n.d.). As a result, permanently protecting open space is a policy priority for many local, state, and federal governments and nongovernmental organizations. However, conserving land can be costly and in many cases the benefits of conservation may be unknown and welfare measures imprecise.
There are three objectives to this article. First, we seek to estimate how the valuation of local open space varies using the hedonic housing price model applied to many markets across the Eastern United States. Although there are many papers in this vein of research (e.g., Geoghegan, Wainger, and Bockstael 1997; Irwin 2002; Song and Knaap 2004; Anderson and West 2006; Cho et al. 2009; Poudyal et al. 2009; Klaiber and Phaneuf 2010; Netusil, Chattopadhyay, and Kovacs 2010; Black 2018), these studies only use data from one city or state, which leaves open the question of how valuation varies across space and the validity of benefit transfer from one city/state to another.1 Second, we aim to understand how geographic features that naturally limit development affect valuation. Saiz (2010) shows that water features and steep slopes limit development, so we assess whether these features act as complements or substitutes to the valuation of conserved land. Third, we investigate how local geography can be used to instrument for preserved land. Irwin and Bockstael (2001), Irwin (2002), and Geoghegan, Lynch, and Bucholtz (2003) use steep slope to instrument for open space in Maryland. Using our more comprehensive database, we can assess the performance of this instrument, as well as one based on water and wetlands, across many cities and put prior results in context.
To achieve these objectives, we build a comprehensive data set of housing and neighborhood characteristics and land use across 215 metropolitan statistical areas (MSAs) in the eastern half of the United States. We use the American Community Survey at the block group level and measure conserved land, slope, and water and wetland features at the same level using GIS and the National Land Cover Database.
For each MSA in our data set, we estimate a separate hedonic valuation model regressing median home value on protected open space at the census block group level. Our specification controls for unprotected open space and numerous housing and demographic variables. We include county fixed effects and quadratic functions of latitude and longitude to guard against spatial unobservables that could be correlated with housing prices and open space. The local valuation results are consistent with previous literature and indicate positive housing premiums for proximity to open space. For example, a 1% increase in Proportion Protected Open Space is associated with a 0.14% increase in housing prices in the New York City MSA. We then calculate marginal willingness to pay (MWTP) for each MSA separately by multiplying the estimated premium by the average house price. The mean MWTP across all MSAs is $499, and the standard deviation is $284.
We seek to explain the variation in valuation by estimating a second model that regresses MWTP for protected open space on various geographic variables measured at the MSA level.2 We hypothesize that the amount of naturally occurring open space from undevelopable areas will decrease MWTP because it is a substitute. In contrast, our results suggest that a 1 standard deviation increase in the amount of Undevelopable Area increases MWTP by about $49, making it a complement. We break undevelopable area into its two components, steeply sloped land and water/wetlands, and find opposite results. Whereas water/wetland areas act as a complement to preservation, steeply sloped areas act as a substitute and reduce MWTP. Our model also includes the amount of protected open space in the MSA, which, consistent with expectations, is negatively relative to MWTP. However, while geography and the quantity of conservation affect MWTP, the changes are relatively small. Marginal changes in geography have essentially no effect on MWTP. We interpret this to mean that the majority of valuation is local and not greatly affected by conservation activities and geography in the larger MSA environment, which implies that benefit transfer is likely valid across many MSAs in this context. Our model includes average temperature and log per capita income. Income is positively correlated with MWTP, and temperature has no discernible effect. The latter is consistent with valuation being local and not based on the particular ecology of conserved land.
Previous studies (Irwin 2002; Saiz 2010) and intuition suggest that geographic features that limit the developability of land may provide observable open space free from endogeneity problems. Slope has been commonly used as an instrumental variable in the open space valuation literature because steeply sloped land is difficult to develop and is arguably exogenous open space (Irwin and Bockstael 2001; Irwin 2002; Geoghegan, Lynch, and Bucholtz 2003). Given our findings of a negative relationship between sloped land in an MSA and MWTP, and the complementary relationship between water/wetlands and preservation, we revisit the use of these geographic variables as instruments. The large geographic scope of our data allows us to examine how slope and water/wetlands perform as an instrument across many housing markets. In the first stage, we find that across most MSAs, slope and water/wetlands are typically strong positive predictors of protected open space, indicated by positive and statistically significant coefficients and large F-statistics. However, there is considerable variation in the second-stage valuation results. Despite qualitatively replicating the IV valuation results for Maryland, the average second-stage coefficient when using slope as the instrument is −0.84, indicating a counterintuitive negative valuation of protected open space. In contrast, the average second-stage coefficient when using water/wetland as the instrument is 0.20, implying positive valuation of protected open space. We conclude that it is necessary to interpret the IV results as a local average treatment effect (LATE); when there is heterogeneity in treatment effects, an IV identification strategy will estimate the treatment effect associated with the specific variation caused by the instrument (Angrist and Pischke 2008). In this context, the IV valuation coefficient does not reflect valuation for all protected open space, just valuation for protected open space that is associated with steeply sloped land or water/wetland features. Ironically, IV estimates are likely less useful for benefit transfer.
2. Data
Our area of study includes data from the 26 U.S. states east of the Mississippi River. We exclude the western United States because of the large variation in precipitation, land cover, and the prevalence of national parks. Our study area contains over 129,000 census block groups, 215 MSAs, 240 million acres of land, and 148 million people. We eliminate any block groups that fall outside of an MSA boundary or are missing data.3 Our final data set includes 103,052 block groups in 215 MSAs.
We obtain census block group level housing and socioeconomic variables from the 2009–2013 American Community Survey (ACS) from the U.S. Census Bureau.4 These data include several structural housing characteristics, such as number of bedrooms, construction year, and median home value. In addition, the socioeconomic data available include age, employment, proportion of renters, housing vacancy, income, race, and education.
There are multiple sources of housing data available for hedonic modeling, and there are benefits and costs of each. Prior research on valuation of open space has typically used individual housing transactions (Irwin 2002; Anderson and West 2006). Although these data allow for fine detail about proximate land use characteristics, they are typically proprietary and difficult or expensive to collect, which limits the geographic scope of analysis. In contrast, housing data from the U.S. Census (or another aggregate source like Zillow) allows for analysis of the entire United States, and this feature has been used by research evaluating national environmental programs (Greenstone and Gallagher 2008; Bento, Freedman, and Lang 2015) or disamenities or events that infrequently occur in a given area (Davis 2011; Lang 2018). Because the focus and contribution of this article come from comparing valuation across geographies, we choose to use census data. However, this choice restricts the details about proximate land use we can include in our local valuation model, and we acknowledge this as a limitation.
Land use data come from the U.S. Geological Survey (USGS) 2011 National Land Cover Database (NLCD). This data set includes land cover types, locations, and topographic features. For the purposes of our study, we define open space as the sum of water, forests, barren land, developed open space (parks, golf courses, etc.), agricultural lands, and wetlands. We then distinguish our definition of open space in two separate ways. Using USGS Protected Area Database (PAD) shapefiles in GIS, we are able to identify which areas are conserved and who owns them. For our study we define Proportion Protected Open Space as any open space that falls within a PAD conserved area. Our definition for Proportion Unprotected Open Space is all remaining open space or developed open space. We use GIS to calculate Proportion Protected Open Space and Proportion Unprotected Open Space in each census block group for our local valuation analysis.
We also use GIS to combine wetland features and water feature layers from the NLCD. We then sum their levels within each census block group to define our Water/Wetlands variable. Additionally, we incorporate the USGS’s 3D Elevation Program data set, which maps elevation levels across the United States, to define our variable Area with Slope Over 15%.
There are additional variables that are unavailable directly from the ACS or USGS, such as lot size, longitude and latitude, and distance to nearest central business district (CBD). We calculate a measure of average lot size by using data on total developed land and dividing by number of housing units in a census block group. Using GIS, we obtain longitude and latitude coordinates for each census block group centroid and calculate the Euclidean distance between each census block group and its CBD.
For estimating models that compare MWTP across MSAs, all data must be aggregated to the MSA level. To calculate Protected Open Space, we sum the total acres of protected open space for all block groups in an MSA and divide that by the total acreage in that MSA. Undevelopable Area is similarly calculated by summing the total acres in a MSA with slope over 15% and water and wetlands land cover types over the total acreage for that MSA. Mean annual temperature is collected from Weather Underground. Last, we gather county-level aggregate 2012 presidential voting data from Election Atlas.
3. Variation in Valuation
Methods
The traditional theory behind the hedonic pricing model (Rosen 1974) relates the value of a good to its bundled attributes. Applied in the housing market, the hedonic model uses a property’s value to reveal preferences for different structural and locational characteristics, including environmental amenities. The market equilibrium is portrayed by a hedonic price function that is the tangency of bids from buyers and offers from sellers. Differentiating the price function with respect to an observed trait reveals the individual’s marginal willingness to pay for that attribute. The hedonic price function can be represented as
[1]
where Pi is the sale price of the property i, Hi is a vector of a property’s structural characteristics, Ni is a vector of neighborhood and locational characteristics, and Ei is a vector of environmental amenities.
Using census and land use data detailed in Section 2, the specific local valuation model we estimate is
[2]
where Picm is the log median house price for census block group i in county c in MSA m, Proportion Protected Open Space is the proportion of acres in census block group i that is protected or conserved open space. We include Proportion Undevelopable Open Space as the proportion of acres in block group i that is unprotected or developable open space. X is a vector of housing and neighborhood characteristics. We also include county fixed effects (πc) and spatial coordinate controls, specifically longitude, longitude squared, latitude, and latitude squared, which mitigate bias stemming from uneven and nonrandom distribution of open space.
β1m is the coefficient of interest and is interpreted as a 1 percentage point increase in protected open space is associated with a β1m% change in housing prices. We expect the coefficient on Proportion Protected Open Space to be positive, due to the stream of benefits received from sustained accessibility and use of land, β1m > 0. For the coefficient on Proportion Undevelopable Open Space we also expect a positive effect but of smaller magnitude than β1m. Because of the possibility of future development, it may not remain open space and the stream of benefits received by properties will likely be discounted, β1m > β2m > 0.
Importantly, we estimate equation [2] separately for each MSA in our data set. Thus, we are estimating MWTP for local open space across many different housing markets, each with its own population and geographical characteristics.
Using these estimates, we then develop a secondary model that examines geographic factors that affect MWTP estimates for protected open space. The dependent variable in our second model, MWTP for Protected Open Spacem, is the estimated MWTP for open space for MSA m and is calculated by multiplying 
by the average median house price from that MSA. We estimate the following model:
[3]
where Protected Open Space is the percentage of total area that is protected open space in MSA m. We predict that this coefficient will be negative, γ1 < 0, suggesting that as the amount of protected open space increases, the less people are willing to pay for more of it. Undevelopable Area is the percent of total area that is undevelopable in MSA m. We follow the logic of Saiz (2010), who establishes a measure of “undevelopable” area where development of residential property is improbable. We define “undevelopable” as any area of wetlands, rivers, lakes, oceans or other water features, or land area with slope over 15% within the boundaries of the MSA as set forth by the census GIS shapefile. We hypothesize that γ1 < 0 because naturally occurring open space will act as a substitute to preserved land. Mean Temperature is average annual temperature. We hypothesize that γ3 = 0 because the value of proximate preserved land is unlikely to reflect specific ecology but views, access, and character. However, if γ3 ≠ 0, this may have implications for an additional effect of climate change. Log Per Capita Income is the logged per capita income averaged across all census block groups in MSA m. We expect the coefficient on income to be positive, γ4 < 0, as areas with more income are able and willing to spend more on environmental amenities like open space.
We also estimate a variant of equation [3] that splits the measure of undevelopable area into a water and wetlands component and a slope component to assess if these geographic factors correlate with MWTP differently:
[4]
Area with Slope Over 15% is the proportion of land with a slope gradient over 15% in MSA m, and Water/Wetlands is the proportion of water bodies and wetlands in MSA m. We expect both variables to be negatively related to MWTP because they act as natural substitutes to preserved land.
Equations [3] and [4] are estimated using weighted least squares, with MSA observations weighted by their total population.
Assumptions
Several econometric identification problems arise in the hedonic literature relating to the endogeneity of open space variables. The quantities of open space in a census block group or MSA is not random and is heavily influenced by factors such as developability and the spatial characteristics related to home values in different areas. Many of these factors are unobservable and may be correlated with both open space and housing prices. For these reasons, bivariate regression almost certainly leads to biased estimates of MWTP in our local valuation model.
Several clever strategies have been used to mitigate this endogeneity issue. Anderson and West (2006) mitigate bias from omitted spatial variables by including block group–level fixed effects in their model, and Lang (2018) uses a regression discontinuity in conservation referendum voting outcomes. Other studies use an IV approach. Research that uses geographic features like slope and soil quality (e.g., Irwin and Bockstael 2001; Irwin 2002; Geoghegan, Lynch, and Bucholtz 2003) are most credible.5 In Section 4, we explore replicating the use of slope as an IV, but we find it is not a viable strategy because steeply sloped protected open space is not universally valued across MSAs.
The necessary assumption we make is that protected open space is exogenous after conditioning on unprotected open space, housing attributes, neighborhood socioeconomic characteristics, and spatial controls. In terms of identification, our study is most similar to that of Anderson and West (2006). We argue that this rich set of parametric and nonparametric controls mitigates bias from unobservables, and we proceed cautiously from there.
Furthermore, we argue that even if some bias remains in our valuation estimates, our MWTP determinants model results will still hold. Bias in the first model will affect the absolute magnitude of MWTP estimates, but if the bias is similar across MSAs, then the relative ordering of MWTP will be the same and our analysis of MWTP shifters will be valid.
Local Valuation Results
We report a selected portion of our local valuation model results because it is not feasible to present results for all 215 MSAs. Table 1 presents results from estimating equation [2] for three sample MSAs, which were chosen because they are well known and vary regionally and socioeconomically. Each column displays the results from a different MSA. Our key independent variable, Proportion Protected Open Space, is positive and statistically significant across all three columns, which is consistent with other findings in the literature (Irwin 2002; Anderson and West 2006). For example, the coefficient of 0.14 in column (1) suggests that a 1 percentage point increase in Proportion Protected Open Space in a block group is associated with a 0.14% increase in median home value in the New York City MSA. Similarly, Proportion Unprotected Open Space coefficients are positive, statistically significant across all three columns, and smaller in two of the three cases, as expected.
Housing and demographic covariates are mostly consistent with expectations. Distance to CBD and its polynomial are statistically significant and indicate a U-shape relationship. Other variables like Proportion College Educated, Proportion High School Dropout, and Log Per Capita Income are all statistically significant and fall in line with previous literature on neighborhood determinants of property value (Irwin 2002; Poudyal et al. 2009).
Only a few MSAs’ coefficients are observed in Table 1, but to visualize the varying estimates and their geographic locations, we map them using GIS. Figure 1 plots estimated MWTP for all 215 MSAs in our study area. The population-weighted mean MWTP for protected open space is $499 but ranges to over $2,500. To get a sense of scope, we graph the distribution of all MWTP estimates in Figure 2. More than 85% of recovered estimates across all MSAs were positive. A majority of the positive MWTP estimates were greater than $250. We report the 25th and 75th percentiles as $171 and $581, respectively.6 We note an interesting pattern in Figure 1: the coastal MSAs seem to have higher MWTP than some of the more inland MSAs. Research has suggested that increasingly dense urban areas are more likely to support preservation efforts than are less populated spaces (Altonji, Lang, and Puggioni 2016). This might help explain why larger coastal cities, which have overtaken much of the remaining open space with urban development, may value conserved lands more than other areas do with abundant land cover.
Map of Estimated MWTP for Protected Open Space for MSAs
Density of MWTP Estimates
Determinants of MWTP Model Results
Table 2 presents our results from estimating equation [3] in column (1) and equation [4] in column (2). To compare coefficients across variables with different units, we transform all independent variables into z-scores. For reference, the table displays the standard deviation of each variable.
The coefficient on Protected Open Space is negative and statistically significant in both specifications. As the amount of protected open space increases MWTP decreases. The column (2) results indicate that a 1 standard deviation increase in protected open space decreases MWTP by $40.72. Given the immense effort and cost that would be needed to increase protected areas 1 standard deviation (or 25%), the decline in MWTP is remarkably small. If interpreted as a demand curve, the slope is quite elastic.
In column (1), the sign on the Undevelopable Area coefficient contradicts our hypothesis, as the results suggest that exogenous open space acts as complement to protected open space. A 1 standard deviation increase in undevelopable land is associated with a $48.62 increase in MWTP. In column (2), we split Undevelopable Area into its two geographic components: Water/Wetlands and Area with Slope Over 15%, which sheds some light on this complement relationship. We find that the coefficient on Area with Slope Over 15% is negative and statistically significant, indicating a substitute relationship, which is closer to our original expectation. On average, a 1 standard deviation increase in Area with Slope Over 15% decreases MWTP by $42.08. In contrast, the coefficient on Water/Wetlands is positive and statistically significant, indicating a complementary relationship. For a 1 standard deviation increase in Water/Wetlands, MWTP increases by $53.95. From a statistical perspective, it is clear that the positive coefficient on Undevelopable Area in column (1) is driven by the effect of Water/Wetlands. The results seem to suggest that water features enhance the value of protected land, perhaps through new recreation opportunities or pleasing views, whereas steeply sloped land detracts from the value of protected land, acting as a substitute or providing recreation or views that fewer people enjoy. Consistent with the coefficient on Protected Open Space, these coefficients are small relative to mean MWTP. Hence although geography can affect MWTP, the changes are small, which we interpret to mean that the majority of valuation is local and not greatly affected by conservation activities and geography in the larger MSA environment.
There is little research to compare these results with, the one exception being Lang (2018), who examines the impact of referendum-authorized land conservation spending on home prices across the United States. He finds that capitalization is lower in areas with higher levels of undevelopable area. The author reports that steeply sloped lands make up 72% of the undevelopable area in his data set, so we can infer that most of that smaller price effect is driven by slope. He proposes that there may be diminishing marginal benefits to conserving areas with more undevelopable open space, which lends itself to our story of steeply sloped lands acting as a substitute.
Mean Temperature has no statistical effect on MWTP in either specification. This aligns with our hypothesis and bolsters the idea that valuation is primarily local and not based much on surrounding geography.
The coefficient on Log Per Capita Income is positive and statistically significant in both columns, indicating that protected open space behaves as a normal good, as income increases people are willing (and able) to spend more on protected open space. The results of column (2) suggest that a 1 standard deviation increase in Log Per Capita Income increases MWTP for protected open space by $109.60. The magnitude of this coefficient is the largest of any in the model, suggesting that an area’s wealth has a larger effect than geography on valuation.7
4. Geographic Instrumental Variables
Methods
As previously discussed, geographic features that limit development can be used as instruments to identify exogenous open space. Irwin (2002) and Saiz (2010) argue that steeply sloped land and water features exogenously increase the cost of development or prohibit it entirely and provide exogenous variation in open space. Irwin and Bockstael (2001), Irwin (2002), and Geoghegan, Lynch, and Bucholtz (2003) exploit this idea by including slope as one of several instrumental variables in their analyses of open space valuation in suburban and exurban Maryland and find positive, statistically significant housing premiums in proximity to open space.
The results found in Section 3 suggest that steeply sloped lands are negatively correlated with MWTP and water/wetlands are positively correlated with MWTP. Together, these findings lead us to question the applicability of geographic instruments across different markets. Given the large geographic scope of data used in this article, we are in a unique position to examine the performance of these instruments and any disparities in estimated valuation across geographies.
We modify our approach to local valuation of protected open space by building on Irwin’s IV strategy to identify exogenous open space across all 215 MSAs.8 Given the opposite relationships found in Section 3, we use two separate geographic instruments in our analysis: slope and water/wetlands. We define our first- and second-stage IV models as
[5]
[6]
where Geographic IV will be defined as either Proportion of Area Water/Wetlands or Proportion of Area with Slope Over 15%, and these variables are now defined at the block group level. All other variables are as defined above in equation [2]. Again, we estimate these equations for all MSAs separately. We also estimate the model for the state of Maryland, while using slope as our geographic instrumental variable, to serve as a better comparison to prior research.
Results
Tables 3 and 4 report the first- and second-stage estimates from our IV models for slope and water/wetlands, respectively. Table 3 includes three sample MSAs (same as Table 1) and the state of Maryland. Table 4 reports the same three sample MSAs and includes Tampa, an arbitrarily chosen MSA with lots of water features. The first-stage IV coefficients in both tables are all positive and statistically significant, which shows that our instrumental variables are highly correlated with land conservation.
The second-stage results in Table 3, column (4) are consistent with the findings in Irwin and Bockstael (2001), Irwin (2002), and Geoghegan, Lynch, and Bucholtz (2003) for Maryland, finding positive and statistically significant valuation for protected open space. In Maryland, a 1% increase in protected open space is associated with a 0.31% increase in median home value. However, as we compare results across the sample MSAs, we find that this is not always the case. For New York City, a 1% increase in Proportion Protected Open Space is associated with a −0.16% decrease in median home value. Results for Chicago are statistically insignificant but negative, and the results for Atlanta match more closely to Maryland’s.
In Table 4, the second-stage results across all columns suggest a positive or statistically insignificant valuation for protected open space when using water/wetlands as an instrument. In column (4) we see that for Tampa, a 1% increase in Proportion Protected Open Space, on average, leads to a 1.08% increase in median home value. The New York City MSA is likewise positive and statistically significant, while Chicago and Atlanta MSAs are not statistically different from zero.
To further understand the heterogeneous results of our IV analysis, Figure 3 presents density plots of our first- and second-stage IV coefficients for both instruments. The weighted mean for the second-stage slope IV coefficients is −0.84, which suggests that the use of sloped lands as an instrument for protected open space on average estimates a negative effect on median home values for most MSAs in our study area. The weighted mean for the second-stage water/wetlands IV coefficients is 0.20, conversely indicating positive valuations on average for median home values across MSAs. These overall results suggest that steeply sloped conserved land is less valued than water/wetland conserved land and is even seen as a disamenity in some places. These results are consistent with our findings in Table 2.
Density Plots of First- and Second-Stage IV Coefficients
It is necessary to interpret these results in the framework of LATEs. When the instrument is slope, the analysis isolates variation in preserved land due to steep slopes, and hence the second-stage valuation estimate is a valuation of steeply sloped preserved land, not all preserved land. Similarly for the water/wetlands instrument, the second-stage valuation estimate is a valuation of protected land with water/wetland features. We find tremendous differences across space and across instruments in the valuation coefficients, meaning that all protected land is not created equal, and the instruments are identifying protected lands that are valued differently. To take the case of Maryland, the results are specific to that state and may be driven by the geography of that state. Maryland is relatively flat, with only 5% of the total area designated as steeply sloped. The scarcity of steeply sloped land may drive the positive valuation. Other areas with more abundant steeply sloped areas or just idiosyncrasies of where steeply sloped land is distributed in an MSA may have very different valuation of proximity to those lands. For these reasons, we stress caution when using geographic instrumental variables. They are not a panacea for endogeneity concerns because the LATE estimated depends on local factors.
5. Conclusion
This article seeks to examine the heterogeneity in protected open space valuation across the Eastern United States and whether natural geography can explain some of that variation. We ground our analysis in the hedonic price method, and using census block group data, we estimate housing premiums for proximity to conserved open space. Importantly, we estimate hedonic valuation models for each of 215 MSAs separately, which yields 215 estimates of MWTP. We seek to explain variation in MWTP across MSAs and regress it on existing conservation, exogenous open space via steeply sloped lands and water features, mean temperature, and income.
There are two main takeaways from our valuation results. First, geography matters for valuation, but in a nuanced way. Consistent with expectations we find that steeply sloped land is substitute for conserved open space. However, wetlands and water features are a complement that enhance the value of conservation. Second, all determinants of MWTP that we test have a relatively small effects on valuation. Large changes in exogenous open space or existing protected open space only have small changes in the valuation of local open space. For example, suppose the Nashville MSA has a MWTP estimate of $535. If the local government or land trusts chose to conserve an additional 1,000 acres in the MSA, which is a 0.16 percentage point increase, we estimate that MWTP would decline by only $0.27. We interpret this to mean that the majority of valuation is local and not dependent on the larger MSA’s geography. In turn, this implies a high likelihood that benefit transfer will yield reasonable estimates.
Using our comprehensive data set, we revisit the idea of using geographic development constraints as instruments for protected land. Although we are able to produce similar findings as Irwin and Bockstael (2001), Irwin (2002), and Geoghegan, Lynch, and Bucholtz (2003) using steeply sloped land as an instrument in Maryland, we find that these positive valuation results are more an exception than the norm. The average local valuation coefficient across the whole sample is negative. Importantly, these results must be interpreted in LATE framework, which suggests that people in many MSAs may not value proximate, steeply sloped conserved land. In contrast, using water/wetland features as an instrument tends to produce local valuation results that are positive, indicating that people do value proximate, conserved land with water features.
There are a number of possible research directions that stem from these findings. We see likely value in future contingent valuation research that investigates preference for conserved land with attributes related to slope, water features, and surrounding conservation. In addition, hedonic valuation would be advanced by more research with large geographic scope and a focus on heterogeneity instead of a single treatment effect. Last, we are struck by disparities between our study that finds the average partisanship of a MSA has no effect on MWTP and many voting studies that find Republicans far less likely to vote for land conservation. Future research should work to better understand how partisanship affects valuation.
Acknowledgments
For useful feedback, we thank conference participants at NAREA and W4133. This work was supported by the U.S. Department of Agriculture’s National Institute of Food and Agriculture, Agricultural and Food Research Initiative Competitive Program, Agriculture Economics and Rural Communities, grant 2018-67024-27695.
Footnotes
Appendix materials are freely available at http://le.uwpress.org and via the links in the electronic version of this article.
↵1 Lang (2018) uses data from across the United States, but assumes homogeneous treatment effects.
↵2 This multimarket approach is similar to those of Boyle, Poor, and Taylor (1999) and Zabel and Kiel (2000).
↵3 The Cape Girardeau-Jackson MO-IL MSA was dropped because it only holds nine block groups, which does not provide enough variation to estimate our model without substantial bias.
↵4 The ACS is published annually in five-year rolling averages. We chose to use the 2009–2013 version because it is centered on 2011, which is the year of our land cover data.
↵5 Some studies use instruments that do not pass the exclusion restriction. E.g., Poudyal et al. (2009) use neighborhood income as an instrument, even though this surely influences housing prices.
↵6 Summary statistics for Figure 1: (< $0): 30 observations, mean –$423, population 6,251,074; ($0–$250): 44 observations, mean $150, population 12,807,559; ($250–$500): 67 observations, mean $368, population 43,048,462; (> $500): 74 observations, mean $783, population 85,960,006. Sample mean: $356; population weighted sample mean: $499; 25th percentile: $171; 75th percentile: $581.
↵7 Although geography is our main focus, we also estimate extensions of equation [4] that include various socioeconomic variables that have been previously shown to relate to preferences for open space (political partisanship, education, and homeownership) in Appendix Table A1. Among these variables, only % homeowner is correlated with MWTP. This makes sense because home buyers are making a long-term investment in an asset and location and are likely to be more cognizant of local surroundings than renters. This result is consistent with Bento, Freedman, and Lang (2015) that find housing values are more responsive of amenity changes than rental rates. Despite typically strong results between partisanship and willingness to vote for land conservation (Altonji, Lang, and Puggioni 2016; Prendergast, Pearson-Merkowitz, and Lang 2019), no such relationship exists between % Democrat and MWTP. Perhaps the partisan split on voting is based on ideological stances of government spending and not on valuation.
↵8 Irwin (2002) used a set of instrumental variables in the model including soil type and other agricultural variables that we do not account for.
References
This article requires a subscription to view the full text. If you have a subscription you may use the login form below to view the article. Access to this article can also be purchased.
