Abstract
This study examines the impacts of animal agricultural facilities on the value of residential housing. Through a quantile regression framework we are able to show that the estimated price impacts are not uniform across the distribution of housing prices, and a statistically significant relationship exists only for houses at or above median price levels. These estimated price impacts also increase as the percentile increases. For comparison, the model also includes the proximity to three other waste facility types: industrial, solid, and septic. Our dataset features 14,785 single-family residential transactions in Tippecanoe County, Indiana, over the period 1993–2006. (JEL Q51)
I. INTRODUCTION
The increasing scale of operation in animal agriculture has generated significant interest in appropriate policy approaches to industry structure, animal welfare, and environmental endangerment. For each of these, intense debate continues over causes and consequences as they relate to the proliferation of confined feeding operations (CFOs). At the community level, debate has been particularly tense as citizenry expresses concerns over the potential for a large concentration of livestock waste to impact their well-being. A near endless stream of commentary persists on this issue, but only a meager amount of science compared to other agricultural externality problems such as nutrient runoff or soil erosion. Measurability and valuation of the flow of CFO externalities in the form of odor is the most likely culprit for the dearth of findings on this issue.
Hedonic models offer a reduced form approach to arriving at the equilibrium economic impact of CFO externalities, even if these models provide limited information for attributing costs to different external outputs (e.g., odor or water pollution). As such, this method has been used in several instances to investigate the value of negative externalities arising from CFOs by determining how characteristics of CFOs affect nearby property values. These studies have offered new and valuable information to contribute to the debate, but in large part have failed to converge to a set of key findings that allows discourse over CFO location and its impact on residents and communities to move appreciably forward.
We argue that this is a result of a singular focus in the literature on exploiting the variation in CFO attributes at the expense of exploring the segmentation that exists in all local housing markets. Our approach overcomes this by adopting a quantile regression approach to estimating the impacts of CFOs on residential home values. This allows us to gain information on the property value impact of proximity to a CFO across the distribution of home prices. Our findings show that CFO home price impacts are significantly negative only above the median sales price, and that the price impacts of other waste facilities that offer similar environmental damage potential follow a similar pattern of implicit pricing of externalities.
II. BACKGROUND
In his seminal paper, Lancaster (1966) outlines the characteristic approach to consumer theory in which utility is a function of the characteristics of goods. The characteristics approach was later formalized by Rosen (1974). In the context of the housing sector, Rosen’s hedonic price theory suggests that the value of a home is a function of its characteristics as given in equation [1]:
[1]
where Pi is the market value of house i. The characteristics of the property consist of S structural attributes, E environmental attributes, and L location attributes. The vector of environmental attributes is included to account for positive (“goods”) and negative (“bads”) externalities, the values of which are capitalized into the total market value of each property. With this flexibility and its consistency with an underlying theoretical equilibrium, the hedonic price method provides a useful alternative for estimating the implicit price effects of environmental externalities.
Kiel and Zabel (2008) offer distinctions on location as it affects housing prices, finding that local is a relative concept and environmental factors may have very different spheres of influence. They show with their hedonic estimation that appraisal considerations of nonmarket factors need consider extent or intensity of disamenity sources only at the most local level of location (e.g., on the same street). A recent assessment of the appraisal approach to CFO impacts identifies distance and its impact on property enjoyment and use as well as the potential for stigma costs (those above direct costs of redress for the negative externality) as important considerations in evaluating property value near CFOs (Kilpatrick 2001). These studies driven by appraisal practice and theory highlight the importance of market preferences and structure and provide contrast to most hedonic applications, which focus more readily on the characteristics of CFOs that might be correlated with intensity of environmental disamenity production.
Factors such as size of the operation (e.g., manure volume), dominant wind directions, and animal species have all been hypothesized to be important explanatory variables in determination of transaction prices near CFOs (Palmquist, Roka, and Vukina 1997; Bayoh, Irwin, and Roe 2004; Herriges, Secchi, and Babcock 2005; Ready and Abdalla 2005). A few consistent findings have emerged from this literature, some with relatively tenuous interpretations. For example, Herriges, Secchi, and Babcock (2005) use transaction data (1,145 observations on sales) from five rural Iowa counties and find that moderately sized operations have larger negative impacts than do the largest regulated facilities. Moreover, their results indicate that a high concentration of animal feeding operations within a three mile radius of a home increases mean property values. These two results are speculatively explained by larger facilities being more modern with better technology for odor abatement and increased demand for rural homes arising from labor-driven population growth in areas with many facilities.
A similar finding that the size of the livestock operation has no incremental impact on property value is reported by Kim and Goldsmith (2009). Here the authors indicate that there are scale effects in abatement that require a hog operation to be large enough to afford odor-reducing technologies. Bayoh, Irwin, and Roe (2004) identify a positive marginal impact on house prices from nearby livestock density, raising the possibility that areas with significant livestock numbers see a net increase in property values with the addition of animal numbers through location of a CFO.
The most intuitive finding observed across studies is that all else equal, close proximity to a large animal operation will have some negative impact on residential property value. In some instances the interactions between wind and distance have been used to account for the odor diffusion process, with one study finding wind direction to be key in explaining CFO externalities (Herriges, Secchi, and Babcock 2005) in Iowa, and another finding no significant explanatory power in North Carolina (Kim and Goldsmith 2009).
While all of these studies suffer the same limitation that leads to the hedonic specification, a consistent conclusion across studies emerges for better measurement at the source of the externality, the CFO. While this is certainly important for our understanding of many issues, it is unlikely to be the case for statistical explanation of housing price patterns in proximity to CFOs. Previous research has shown information asymmetry about the local real estate market between parties to be an important determinant in the final sale price of a home (Levitt and Syverson 2008). A similar asymmetry likely exists between the seller and buyer of a home, whereby the potential buyer formulates a bid based on public information (e.g., the location of a regulated CFO), whereas the seller knows how the externality is actually realized on the property.
The seller in this situation has little incentive to disclose this information to lower the ask price, though it may lower the threshold offer he will accept. Bidders on the home are typically comparing a diverse set of homes and locations themselves and are unlikely to consider the bevy of differences in attributes of individual nearby CFO facilities with significant weight. With more complete transaction data, we could ideally test how winning bidder attributes such as age, education, or wealth impact sales price. Absent this, we might logically conclude that significant information exists in the home price distribution itself that is not being fully exploited. We adopt this approach in our analysis, estimating implicit prices of environmental attributes at different points in the home price distribution. This reflects our view that the distributional impacts of CFO siting are realized not only spatially as measured by the distance to the operation, but also according to differential purchaser attributes that allow them to participate at different points of the house price distribution.
III. DATA AND METHODS
The hedonic price function of equation [1] can be written as the stochastic equation given in [2] below, where y is a vector of observed or assessed property values, X is a matrix of K quantifiable characteristics of each property, β is a vector of unknown parameters, and ε is a vector of regression disturbances with typical properties of zero mean and constant variance:
[2]
The unknown regression parameters of [2] are estimated via ordinary least squares (OLS) typically, but recent studies adopt alternative estimation procedures to account for systematic patterns in the distribution of housing values across space. In particular for housing markets, neighboring observations will share a number of location amenities and may also serve as a proxy for similar socioeconomic factors such as income and the occupational status of homeowners (Basu and Thibodeau 1998; Gelfand et al. 1998). To address this in estimation, we account for potential spatial price spillovers using a spatial lag model as in [3], where λ is a spatial lag parameter, W is an exogenous spatial weights matrix giving Wy as a spatial lag of the dependent variable:
[3]
The spatial weights matrix defines a relevant neighborhood for each observation, with zeroes on the diagonal and off-diagonal nonzero elements wij indicative of observations i and j sharing a spatial relationship. The sign of the estimated spatial parameter indicates the nature of the spatial process. For example, when λ > 0, the model suggests property values are, on average, positively impacted by neighboring property values. Unlike temporal lags in time-series analysis, the spatial lag characterizes simultaneous feedback in space such that a home is impacted by the value of its neighbor while it simultaneously impacts the value of its neighbors. Because of this, the spatial lag term Wy is correlated with the disturbance term, introducing biased estimates of regression parameters using OLS. We adopt an augmented version of two-stage least squares (2SLS) estimation procedure for the spatial lag model (for a recent discussion of 2SLS spatial lag estimation, see Anselin 2006). The spatial lag hedonic model has been previously used to generate estimates of the price impacts of environmental amenities such as public parks (Munroe 2007), school quality (Brasington and Haurin 2006), traffic volume (Kawamura and Mahajan 2005), and land cover (Paterson and Boyle 2001).
With our goal of not only explaining environmental impacts on prices at the means but at various points of the house price distribution, we adopt the spatial lag specification and implement it in a two-step quantile regression construction, as done by Zietz, Zietz, and Sirmans (2008). Quantile regression estimates allow for parameters that vary by quantile (τ) of the dependent variable’s conditional distribution. Thus, our specification allows for differing degrees of spatial dependence across the conditional distribution of y. Similar to the mean regression approach, the spatial lag term is endogenously determined, such that we require an instrumental variable technique for identification. We follow Kim and Muller’s (2004) estimation strategy for endogenous quantile regression, with the linearly independent columns of X and their spatial lags, WX serving as instruments for the spatial lag term.
[4]
Our data consists of 14,785 single-family residential real estate transactions in Tippecanoe County, Indiana, over the years 1993–2006. These data were obtained from the local multiple listing service, and we report summary statistics in Table 1. The “average house” is approximately 32 years old, with three bedrooms, two bathrooms, and a lot size of 0.68 acres, and features a sales price of approximately $129,000.
Data Summary
The model also includes several variables to control for location effects, including several distance-based measures. The mean distance to the city center is 3.7 miles, defined by the distance from the county courthouse located at the center of the city square. The mean distance to industrial parks is 12.2 miles, and the mean distance to state highways is 0.6 miles. The additional locational variable indicates whether the home is located in the 100-year flood plain—which applies to 1% of the observations.
Table 1 reports the means of four variables that proxy the presence of environmental externalities. Distances to three waste sites are included (industrial waste, septic waste, and solid waste), along with the variable measuring the distance to the nearest CFO. These distances to the nearest potential externality source are measured without respect to administrative boundaries, meaning some cases identify the nearest CFO in a neighboring county. The mean distances to the nearest industrial, septic, and solid waste facilities are 1.2, 4.7, and 2.7 miles, respectively, while the mean distance to the nearest CFO is 4.1 miles. The location of each waste facility and CFO was obtained from the Indiana Department of Environmental Management. The locations of home sales, the three types of waste facilities, and CFOs are shown in Figure 1.
Transactions, Waste Facilities, and Confined Feeding Operations
IV. RESULTS
We estimate the hedonic price function from [4] using a semilog functional form with price expressed as its natural log and explanatory variables in their levels. The predominance of the semilog form in hedonic studies arises from its ability to mitigate problems with heteroskedasticity (Green and Malpezzi 2003), as well as its ease of interpretation with coefficients describing the percentage change in the sale price of the house for a unit change in the characteristic measured in the independent variable. Estimated coefficients are reported in Table 2. The spatial lag parameter is specified with a row-normalized binary near est neighbors spatial weights matrix in which the elements take the value of 1 for each observation’s 10 nearest neighbors. Subsequent normalization is applied to leave all rows of this matrix summing to 1. The resulting spatial lag term is therefore the average house price at the ten nearest neighbors with each neighbor receiving an equal weight.
Coefficient Estimates
The first column of Table 2 includes the coefficient estimates of the spatial lag mean regression using 2SLS (Anselin 1988). All of the coefficient estimates have the expected signs and are significantly different from zero with the exception of the squared age term, implying that on average the impact of age on house prices is linear and negative. The results suggest that on average a location that is close to one of the three waste site locations or a CFO will have a lower housing value, all else constant. The spatial coefficient is statistically significant and positive with a value of 0.512.
The remaining columns of Table 2 report the quantile regression estimates, also augmented to include a spatial lag component. We first note the degree of variation exhibited by coefficient estimates across the conditional distribution of house prices. For example, as the house price increases (i.e., as the conditional quantile is increased) the negative impact on sales price for the house’s age is more pronounced.
An additional advantage of the quantile regression with spatial lag is that the spatial coefficient is allowed to vary over the conditional distribution of y. We plot the spatial lag coefficients by quantile in Figure 2, with the x axis indicating the quantile of the house price distribution and the y axis showing the estimated value of the spatial lag coefficient (shaded area indicates the 95% confidence interval). The (horizontal line) mean regression estimate and its confidence interval are plotted on this same graph. We see that the degree of spatial interaction is higher than the mean regression confidence interval at τ = 0.20 to τ = 0.50. This indicates that the degree of spatial interaction across the conditional price distribution differs from that suggested by the conditional mean.
Spatial Lag Coefficient
Although the spatial lag term offers valuable information by accounting for spatial price spillovers, it complicates interpretation of the marginal effects of the model. In OLS regression, the marginal benefit (in percentage terms) of each regressor xk is obtained by the partial derivative of the regression equation. The marginal implicit price is simply the constant βk multiplied by the average price. However, the marginal implicit price of the spatial-lag hedonic model is impacted by the “global spillovers” attributed to what Anselin (2003) calls the spatial multiplier effect. The spatial multiplier stems from the notion that the price of each home is tied to all other locations through the spatial lag term. The multiplier can be seen from the reduced form of the spatial lag model in [5], where the derivative with respect to xk yields the expression: βk(I−λW)−1. Kim, Phipps, and Anselin (2003) show that when the weights matrix W is row-normalized as we have done here, the spatial multiplier reduces to 1/(1−λ):
[5]
The spatial lag hedonic model therefore captures the induced effects of a change in a neighborhood’s housing characteristics, which would be omitted in least squares estimates (Kim, Phipps, and Anselin 2003). For example, the semielasticity (proportionate change due to a unit change) of housing price from a small change in lot size in the semilog functional form mean regression is calculated as in [6]:
[6]
Accounting for the differing spatial multiplier effects across the conditional distribution of y, the total effect semielasticity estimates for the four environmental characteristics are shown in Figure 3. Each pane depicts the total effect semielasticity with a 95% confidence interval (shaded), as well as the complementary mean regression results (horizontal dashed lines). The graphs show that with the exception of septic waste facilities, there is a strong tendency for increasing distance from an externality source to have stronger proportionate impacts on the sale price for a home. To wit, higher-value homes exist in a segment of the market where prices are considerably more sensitive to the nearness of potential environmental externality sources. For the distance to solid wastes the results are pronounced, with an increase from 0.03 at τ = 0.10 to 0.52 at τ = 0.90.
Total Effect Elasticities
For our estimates of CFO distance and its impact on prices, we first note that coefficient estimates are statistically significant only at τ ≥ 0.60. Thus, for the home sales in our data, a home would need to sell at a price above the median home to realize a significant sales penalty for being nearer a CFO. In comparison to the estimated impact from the mean regression, which indicated a 10% increase in the sale price of a home for every mile removed from a CFO all else constant, we now see that the least expensive homes are not impacted significantly by proximity, while more expensive homes are profoundly impacted. For example, at the 70th percentile of the conditional house price distribution, if a home were moved one mile further from a CFO, then its price would be expected to increase by approximately 31%.
In Table 3, we report the estimated implicit price for a marginal change in each environmental variable included in the regression and for which our results were statistically different from zero. The implicit price calculations are based on the mean predicted value of the conditional house price distribution at each quantile. The smallest implicit price is associated with industrial waste facilities for low-value homes (τ = 0.10), and the greatest implicit price is associated with solid waste facilities for high value homes (τ = 0.90). The pattern of increasing external costs (as measured by the implicit price of a mile further removed from the site) is consistently increasing for all types of sources. Of particular note are the sharply increasing patterns (beyond the median home price) for both solid waste and CFOs, implying high willingness to avoid these disamenities by purchasers of homes in this neighborhood of the price distribution. For example the implicit price for the distance to solid waste or CFO are nearly twelve and four times respectively that of the sixtieth percentile value at the ninetieth percentile holding all else constant, while the sales price only doubles between the median and the ninetieth percentile.
Estimated Implicit Prices
V. CONCLUSIONS
There is a growing concern about the impacts of animal agriculture on the value of neighboring residential properties. A number of existing studies estimate the price effects of production facilities as measured by the associated negative attributes, such as odor. The existing literature is inconclusive on the exact nature of the relationship between confined animal operations and residential property values.
This study is motivated by the belief that existing studies may represent an incomplete picture of the underlying relationship between animal agriculture and residential property values. Through a quantile regression framework, we estimate the implicit price effects across the conditional distribution of house prices. We find that the semielasticity of distance to the nearest CFO is not statistically significant at all points on the conditional distribution. Instead, the price impacts are significant only for houses at or above the conditional median. Further, the estimated implicit price increases as the percentile increases.
Our hedonic price model also includes three physical waste facilities: industrial, solid, and septic waste facilities. The variables allow for additional comparisons of the price impacts for similar environmental hazards. Unlike CFOs, the industrial and septic waste facilities suggest a statistically significant relationship at each point of the conditional distribution. The implicit price of each hazard also increases as the percentile under examination increases.
In sum, the results suggest that estimates of the conditional mean (traditional linear regression) may not be reflective of the true economic phenomenon in segmented markets. In the case of waste facilities, the impacts are statistically significant at each point on the conditional distribution of house prices, yet the price effects are statistically significant only for higher-value homes in the case of CFOs. As the debate over the local economic consequences of livestock agriculture and appropriate land use policy planning continues, the need for quantitative results will continue to grow. Our results indicate that an important element that has been overlooked in previous estimates is the distribution of home values itself. To the extent that policy objectives favor fair compensation or minimizing external economic impacts, it is imperative that this element is given due attention in determining the cost-benefit balance of CFO placement.
Acknowledgments
We wish to thank the Lafayette Regional Association of Realtors for providing the real estate transactions data. The views expressed are those of the authors and should not be attributed to the Economic Research Service or USDA.
Footnotes
The authors are, respectively, economist, Economic Research Service, U.S. Department of Agriculture, Washington, D.C.; and associate professor, Department of Agricultural Economics, Purdue University, West Lafayette, Indiana.
