Abstract
PART 1 J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u This page intentionally left blank Chemical Hazardous Sites and Residential Prices: Determinants of Impact A u t h o r Perry Wisinger A b s t r a c t The Emergency Planning and Community Right-to-Know Act (EPCRA) requires reporting of potential chemical hazardous sites to the Environmental Protection Agency (EPA). The EPA discloses some sites on the Internet while others are not. I investigate whether Internet disclosure makes a difference on the impact a hazardous site has on nearby housing prices. I also investigate the relevance of EPA-hazard classifications to understand the residential market reaction to nearby chemical hazardous sites. Data from Lubbock, Texas confirm that housing values near registered chemical hazards are lower, ceteris paribus; however, Internet-listed hazardous sites do not have a bigger impact on housing prices than do hazards not listed on the Internet. But more importantly, hazard classifications other than EPA classification better define house price behavior. Neighborhood dangers lower residential values, but what hazards and by how much? Does Internet disclosure make a difference? Would hazardous site categorization differing from Environmental Protection Agency (EPA) definitions better explain the impact of chemical hazardous sites on house prices? I investigate public disclosure and labeling of chemical hazardous sites and the impact they have on nearby housing prices. The United States Congress responded to the deadly Union Carbide pesticide plant accident in Bhopal, India (and other similar disasters) by enacting Title III of the Superfund Amendment Reauthorization Act of 1986, which is also known as the Emergency Planning and Community Right-to-Know Act (EPCRA). To protect Americans, EPCRA mandates every public or private facility in the U.S. that routinely has a ‘‘threshold quantity’’ of any of 600 acutely hazardous chemicals to file Emergency and Hazardous Chemical Inventory (Tier One and Tier Two) forms containing the name, amount, and location of such chemicals with federal, state, and local emergency planners and responders. As expected, EPCRA requires reporting by most industrial facilities and waste treatment plants. But most municipal swimming pools, retail cleaners, auto-repair shops, traditional printers, and even some gas stations have to report, too. Far too often slow leaks and industrial accidents occur that endanger entire neighborhoods. Many factors, including gut feelings about neighborhood risk, figure into the final price a potential buyer is willing to pay for a home, and information for subjective J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 4 W i s i n g e r risk assessments comes from several sources. Besides visual cues, risk characteristics about commercial neighbors are quickly gleaned from business names, and additional information can be obtained by checking the phone book or from a cursory Internet search. Cautious buyers can gather still more details from specific Internet sites dedicated to documenting hazards associated with various businesses. And for the very suspicious, there are other sources that are not equally available to everybody including insider information and local gossip. Real estate market efficiency is the measure of the universal availability and use of information affecting land values. Air emissions facilities, water discharge facilities, hazardous waste handlers, and any facilities that previously reported a toxic release to the environment are all listed on the Internet, while information pertaining to Tier Two sites (sites that merely have substantial inventories of dangerous chemicals) must be requested by mail. Clearly, information available only by request is not as readily accessible as Internet information. While real estate professionals expect neighborhood risks to lower land values, does the readiness of public availability of such disclosure matter pricewise? In this paper, I confirm the expected negative correlation between EPCRA sites and nearby housing prices in a portion of Lubbock, Texas, a city with over 225,000 residents; however, I extend economic investigation into previously uncharted areas of real estate market efficiency. I compare the differential land-value impact of Internet disclosure with non-Internet disclosure. I also investigate the price impact of commercial risk stereotypes and measure the impact of visual risk impressions. Real estate professionals, mortgage lenders, government policymakers, and other stakeholders should find the results of interest. L i t e r a t u r e R e v i e w The Clean Air Act of 1970 requires listing sources of air pollution and estimating the amount each source discharges, and under the Clean Water Act of 1972, the EPA supervises direct discharges into rivers, streams, lakes, and other waterways. However, the roots of federal tracking of toxic chemicals begin with the Resource Conservation and Recovery Act of 1976, which provided for federal regulation of hazardous waste. Hazardous waste is legally defined as any by-product that potentially poses a substantial hazard to human health or the environment when improperly managed. This act makes generators, transporters, treaters, storers, and disposers report their activities to state environmental agencies, who relay this information to the EPA. To combat concerns over the health and environmental risks posed by past dumping of hazardous waste, congress added the Superfund Program in 1980 to locate, investigate, and cleanup these perilous dumpsites. The EPA organizes environmental toxin release information into the Toxic Release Inventory (TRI) database, which stores data by facility, by year, by chemical, and by medium of release whether air, water, underground injection, land disposal, or offsite. The TRI sites listed on the website (http: / / www.epa.gov / enviro / ) are those with a history of toxic chemical releases. Additional disclosures at this website C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 5 pertain to nearby hazardous waste handlers, Superfund sites, and sites requiring either an air release permit or water discharge permit. However, if there is no history of toxic release, stored toxic chemical inventory site information may not be readily available to the public. Instead, a specific request for Tier One or Tier Two information is required. While Tier One information may be vague, Tier Two reports contain the exact name, quantity, method of storage, and specific location of each toxic chemical (Abell, 1994). Generally, Tier Two reports are the state repository for both one-time emergency planning letters notifying the state that certain hazardous chemicals in specified amounts are stored at a facility and annual hazardous chemical inventory reports. The only public reporting requirement is that Local Emergency Planning Committees must merely publish annually in local newspapers a notice that Tier Two forms have been received (Skillern, 1995). The negative impact of hazards on real estate values is well established in the literature as evidenced by the meta-analysis by Simons and Saginor (2006) of 75 peer-reviewed articles and case studies. Perhaps the first to investigate the impact of manmade hazards on land values was Ridker and Henning (1967), who discovered that ambient air pollution lowered property values. Numerous follow- up studies confirmed the negative impact air pollution has on land values including the meta-analysis by Smith and Huang (1993) of 37 previous studies. The negative impact of water pollution on land values was documented early by Epp and Al- Ani (1979), Rich and Moffitt (1982), Mendelsohn, et al. (1992), and more recently by Michael, Boyle, and Bouchard (2000). The suspected negative impact on real estate values caused by nearby waste disposal sites was confirmed by Smith and Desvousges (1986a, 1986b), Kohlhase (1991), Ketkar (1992), and Smolen, Moore, and Conway (1992a, 1992b). Thayer, Albers, and Rahmatian (1992) noted the distance from a designated hazardous waste site has more impact than the distance from a nonhazardous waste site does. Market efficiency theory posits that markets incorporate all reasonably available information into prices (Fama, 1970). However, Wisinger (2006) found no immediate housing market response to EPA reporting either toxic leaks or protective regulation violations. Based on their findings of little impact on housing prices following toxic releases, Bui and Mayer (2003) questioned whether the public is capable of understanding the complex implications of chemical risk from TRI reporting. And while Decker, Nielsen, and Sindt (2005) did find that housing prices declined following TRI reporting, they too noted the public seemed unable to properly rate the degree of danger from current hazard reporting practices. Minguez, Montero, and Frenandez-Aviles (2013) suggested that housing prices respond to subjective public perception of pollution dangers rather than scientific risk assessments. In keeping with the above, Greenstone and Gallagher (2008) found that land values surrounding Superfund sites did not improve following site cleanup. T h e D a t a Totally within Lubbock County, Texas, and mostly within the city of Lubbock, the study area comprises four contiguous ZIP Code areas: 79404, 79405, 79411, J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 6 W i s i n g e r and 79412. Together they cover about 31 square miles of Lubbock County. Exhibit 1 is a map of Lubbock, Texas, with the study area and Texas Tech University indicated. Together the four zones contain 38,044 people and 13,728 housing units according to the 2010 census. A minority-rich area, it is composed of 56.8% Hispanic and 15.8% black residents. This small area is selected for study because it has an adequate number of house sales within the study period, as well as an adequate number of chemical hazardous facilities. Using the Lubbock MLS database, 254 house sales were identified during the 18- month period ended December 31, 2005 within the study area. Data on the sales price of each house, along with other characteristics are used for estimation of a hedonic model. The sale prices of the houses vary from $8,500 to $163,000, with a mean value of $61,950, a median of $55,000, a mode of $65,000, and a standard deviation of $32,076. Data on chemical hazard sites were obtained from the EPA Envirofacts Data Warehouse Internet site and the Texas Tier Two Chemical Reporting Program. MapQuest (http: / / www.mapquest.com / ), provides the longitude and latitude of the houses and business sites used in the study. A similar approach was followed by Hunter and Sutton (1995) to locate hazardous waste generators. The initial focal independent variables are a) whether hazard information is available on the EPA website, b) whether the site is a Tier-Two storage site with no disclosure on the EPA website, or c) whether the site is both a TRI Internet- listed site and a Tier-Two site. Along with this category information, the inverse distance (proximity) of the site to each of the house sales is also used. All of these hazard sites within the study area plus those within one mile of the study area constitute the study zone. Hazard sites within a mile of the study area are included because they might strongly influence the sales price of homes within the study area. Along with the 254 houses sold, there are 147 hazard sites with at least five houses within 2.5 miles of each site. The 2.5-mile distance is based on the results of Ihlanfeldt and Taylor (2004) and Smolen, Moore, and Conway (1992a, 1992b). The hazard sites consist of four air-release sites, three water-discharge sites, seven TRI sites, 95 hazardous-waste handlers, and 57 Tier Two sites. Some sites fit more than one hazard category. To ensure the residential market had time to adjust, all of the hazardous sites had been designated as such by government officials for at least two years prior to the beginning of the study period. An additional focal variable is risk reputation. News sources consistently report explosions and other neighborhood dangers by industrial activity. To assess the impact of commercial category stereotyping on neighborhood risk assessment and home prices, each hazardous site is classified into one of 13 industry types based on information easily available. Visual impression is important to forming emotional risk assessment. Paterson and Boyle (2002) stated that omitting important visibility attributes could lead to erroneous ideas about environment variables in a hedonic real estate model. C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 7 J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u Exhibit 1 u Lubbock Study Area with 1 ⁄2 Mile Circle around Texas Tech University 8 W i s i n g e r Accordingly, for this study, information pertaining to the appearance of a hazard site was gathered by visually inspecting each of the 147 chemical hazard sites. During the visual inspection of each site, the following focal-variable information was collected: a) percentage of neighborhood residential usage; b) geographic size of each site; and c) the presence of danger signs. M e t h o d o l o g y This study began by questioning whether Internet disclosure of neighborhood chemical hazards had a major impact on housing prices; however, the overarching question investigated is the usefulness of various chemical hazardous site classifications to understanding and predicting real estate market behavior. The approach here is to model the impact of each site individually and then combine these different impacts in new ways testing the resulting patterns for usefulness. Traditionally hazards are grouped first and then measured and tested for significance. Here, hazards are first measured individually, then grouped and tested for significance. Statistical test results are considered significant at the 90% level. The analysis is performed in two stages. In the first stage, a hedonic regression model is fitted with the dependent variable being the house price and the individual predictor variables being the structural attributes, location attributes, and the inverse distance from the house sold to each of the hazard sites. The estimated coefficient for the inverse-distance predictor represents the marginal impact of each individual hazard site on the price of surrounding houses. It should not be implied from the ‘‘impact’’ terminology that the hazard causes the full reduction in the house price observed. In some cases, hazards may be in their locations because of low neighborhood house prices and not the other way around. In the second stage, the relationship between the marginal price impact of hazard sites and various hazard-information sets are investigated. Based on results from the first stage model, the three information sets investigated in the second stage are the price impact of: (1) whether the hazard is reported on the EPA website, the site is Tier Two and not listed on the EPA website, or the site is both Tier Two and reported on the EPA website; (2) the industrial classification for the commercial activities of the hazard site; and (3) visual inspection variables. The hedonic price model, first introduced by Court (1939) and refined by Rosen (1974), is a very useful, flexible tool commonly used in econometric analysis. In this study, I use structural variables consistent with those suggested and validated by Sirmans, Macpherson, and Zietz (2005) for studies covering the southwest U.S. The structural variables selected for use are: size of house in square feet, age of house, number of fireplaces, presence of central air conditioning indicator, number of cars garage, and presence of brick exterior indicator. For his neighborhood characteristics, Hwang (2003) selected the percentage of whites and household income and both were statistically significant at the 99% confidence level. To reduce the co-linearity between percentage of whites and C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 9 household income, only one of these two variables is used in this study, that being the census-tract median value of the surrounding property. While commonly included, an independent variable for the distance to a central business district requires the assumption of a monocentric city, and the literature indicates this is a safe assumption only for a large metropolitan area, which Lubbock, Texas is not. However, Wisinger (2006) concluded that proximity to Texas Tech University has a positive impact on housing values. The influence of Texas Tech University on housing prices is limited to 1.5 miles: TTUAdj 5 (1.5 2 (Distance to TTU)) / 1.5. (1) The study area along with the impact area of Texas Tech University is shown in Exhibit 1. Over 100 houses sold were near the university. In stage one of the analyses, least square estimates for the parameters of the models: 7 2 P 5 b 1 b ID 1 u S 1 l L 1 « (2) O O h 0,s 1,s h,s i,s h,i j,s h, j h,s i51 j51 are found for each hazard site, s 5 1 to 147. In equation (2): P 5 Selling price of house h, h 5 1, ..., 254; ID 5 1 / (Distance of house, h, from hazard site s, s 5 1, ..., 147); h,s S 5 Structural variable i for house h; h,i L 5 Location variable j for house h; h, j b, u, l 5 The resultant estimators; and « 5 The error term. Regressions are calculated for each of the hazardous sites yielding 147 results. The main parameter estimates of interest are the 147 b . These estimate the 1,s hazardous sites marginal impact of each hazardous site on housing prices. Regression analysis does not show or prove causation. In particular, a large negative b does not imply the site causes a decline in the house prices; it only 1,s indicates that houses near the site are associated with lower values. In particular, no distinction can be made between a business choosing its location because of low house values and house values being low because of the presence of the business. One or both conditions could be present. In Exhibit 2, I summarize the structural and location variables (including descriptive statistics) used in equation (2). Stage one of the analysis results in a b , for each hazard site. Grouping 1,s these individual estimators allows examination by whatever categorical group the researcher desires. Here, the estimators are first grouped into Tier Two and non- Tier Two groups and t-tested for statistical difference between the group means J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 0 W i s i n g e r Exhibit 2 u Structural and Location Variables and Descriptive Statistics Mean Median Mode Max. Min. Std. Dev. Structural Variables Age of house 49.8 50 45 76 3 14.9 Size of house in square feet 1437 1276 1026 3369 560 539 Size of house squared Size of garage (number of cars) 1 1 1 3 0 0.76 Number of fireplaces 0.3 0 0 2 0 0.49 Central air indicator 0.64 1 1 1 0 0.48 Brick indicator 0.4 0 0 1 0 0.49 Location Variables Adjacent to Texas Tech Univ. 0.16 0 0 0.8 0 0.22 distance adjustment Census-tract median value of 38266 39800 40100 76600 23700 6192 surrounding property to determine whether the Internet disclosure group has a bigger impact than the non-Internet disclosure group. In stage two, the investigation is confined to the relationship between the 147 b , and the three additional information sets (Internet disclosure, commercial 1,s activity, and visual cues). Each of these information sets has a slightly different structure and requires a slightly different regression model to obtain the desired model fit. There are three Internet disclosure categories to compare in the model investigating the Internet disclosure information sets. The model used to investigate this information set is: b 5 m D , (3) 1,s i i i51 where i 5 1 represents Internet disclosure but not a Tier Two reporting, i 5 2 represents Tier Two reporting, but no Internet disclosure, and i 5 3 represents both Internet disclosure and Tier Two reporting. The D are indicator variables for each of these categories and m are the three resultant estimators. An R-square for the model and t-tests for H : m 5 0, i 5 1, 2, 3 and for H : m 5 m for i, j 5 1, 0 i 0 i j 2, 3, i Þ j are performed to again compare sites with Internet disclosure against sites without Internet disclosure. Next, because I question whether the EPA hazardous site classification scheme is optimal for generalizing about the impact of chemical hazardous sites on housing C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 1 1 prices, an alternative grouping is tested. There is no literature guidance on alternative EPA site classifications for measuring the impact on housing, so for testing purposes the hazardous sites are divided into the following commercial 13 groups: agricultural, automotive / trucking dealers, automotive / trucking repair, city/ government, cleaners, commercial bakery / food production, communications, convenience / gas and oil change, industrial, large wholesale / warehouse, printing, transportation, and other. The commercial activity information set places each hazard site into one and only one of the 13 industry categories. The regression model: b 5 m ID (4) 1,s i i i51 is fit where ID is the distance indicator variable for the ith industry category and m are the 13 resultant estimators. An R-square for the model and tests for H : m 5 0, i 5 1, ..., 13 are performed to t-test estimator means for statistical significance from zero. Possibly there are other factors more relevant, such as visual stigma, to understanding the impact of EPA sites on housing prices. Again there is little guidance for visual variable specification. While preliminary investigation included more, I test the synergistic visual impact of three: the percentage of residential versus nonresidential land use, the size of the facility, and whether there is a danger sign present. In visiting the individual hazardous sites, it became apparent that the most verifiable division was between those sites located in areas where the land use was either more or less than 30% residential. Another variable is facility size. Facilities are arbitrarily divided into the following three sizes: less than 25% of a typical city block, 25% to one block, and facilities larger than one typical city block. The last variable investigated is whether a physical danger sign is posted. An example would be a sign posted on a surrounding fence warning people of hazards within the facility. The model to investigate this information set is: 2 3 2 b 5 m VD , (5) O O O 1,s ijk ijk i51 j51 k51 with i 5 1 indicating less than 30% residential and i 5 2 indicating more than 30% residential. j 5 1, 2 or 3 indicates less than 0.25 of a city block, between 0.25 and under 1 city block, and 1 block in size or more, respectively. k 5 1 indicates no danger signs visible, while k 5 2 indicates visible danger sign. An J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 2 W i s i n g e r R-square for the model and t-tests of the null hypotheses that the following marginal means are equal to zero are summarized: m 5 (m 1 m 1 m 1 m 1 m 1 m ) / 6 1.. 111 112 121 122 131 132 m 5 (m 1 m 1 m 1 m 1 m 1 m ) / 6 2.. 211 212 221 222 231 232 m 5 (m 1 m 1 m 1 m ) / 4 .1. 111 211 112 212 m 5 (m 1 m 1 m 1 m ) / 4 .2. 121 221 122 222 m 5 (m 1 m 1 m 1 m ) / 4 .3. 131 231 132 232 m 5 (m 1 m 1 m 1 m 1 m 1 m ) / 6 ..1 111 121 131 211 221 231 m 5 (m 1 m 1 m 1 m 1 m 1 m ) / 6. (6) ..2 112 122 132 212 222 232 For the investigations associated with models (3) and (4), the individual means are the quantities of interest since a site fits into one and only one of the categories. However, in model (5) the three visual dimensions represent 12 distinct combinations so it is the means for the levels of the individual dimensions that are t-tested for significance. R e s u l t s During the first stage of the analysis, the hedonic model (2) is fitted for each of the 147 hazard sites. The goal of this stage is to obtain stable estimates of hazardous site impact on the house prices. To increase the stability of the estimates, houses sales are removed from the analysis if they can be verified to have unusual characteristics. All houses with residuals greater than $35,000 are identified and nine house sales removed because of either disproportionate lot size or suspicious sales data. This represented 3.5% of the original house sales in the sample. The estimated regression coefficients, R-squares, and Moran’s I values for model (2) for all 147 hazard sites are summarized in Exhibit 3. The models can be used to explain a large percentage of the variability in house prices, with an average R-square of 83.7%. All of the hedonic model coefficients have stable signs from site to site. According to Anselin (1992), Moran’s I is the most common test for spatial autocorrelation errors. No serial correlation problem is detected with an average Moran’s I of 0.0388. The estimated b coefficients of 1,s model (2), representing the impact of a hazardous site on the house price, show considerable variation from site to site. The mean estimated b is 262.6, but the 1,s values range from 2603.7 to 1035.4. In Exhibit 4, I summarize descriptive statistics for the b coefficients for the sites 1,s in the six different EPA classifications, as well as how many are positive and significant, not significant at 90% level, and those negative and significant in the hedonic model (2). All categories have a negative average coefficient. The Tier C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 1 3 Exhibit 3 u Summary of the Regression Coefficients, R-squares, and Moran’s I for the Hedonic Models (2) Coefficient Estimate P-value Mean Min. Mean Min. Variable (Std. Dev.) (Max.) (Std. Dev.) (Max.) Intercept $13,434.3 25,870.3 0.2625 0.0114 (5,529.6) (31,416.5) (0.1660) (0.9971) Age of house (years) 2$263.6 2333.9 0.0048 0.0000 (37.0) (2176.6) (0.0091) (0.0520) Size of house (sq. ft.) $11.4 8.2 0.2433 0.1418 (1.51) (14.1) (0.0658) (0.3996) Size of house $0.0074 0.0062 0.0111 0.0038 (0.0005) (0.0084) (0.0052) (0.0299) Garage size (# of cars) $4,046.5 3,688.7 0.0029 0.0014 (158.4) (4,367.5) (0.0011) (0.0060) Fireplace indicator $5,900.2 5,540.9 0.0108 0.0067 (150.9) (6,255.7) (0.0019) (0.0160) Central air indicator $11,227.0 10,973.1 0.0000 0.0000 (128.4) (11,684.8) (0.0000) (0.0000) Brick indicator $2,471.3 1,914.7 0.2397 0.1089 (237.0) (3,429.2) (0.0430) (0.3580) Adjacent to Texas Tech University $17,833.5 6,644.4 0.0065 0.0000 (4,388.7) (27,146.0) (0.0295) (0.3474) Census-tract median value of 0.3284 0.0542 0.1213 0.0078 surrounding property (0.0844) (0.4889) (0.1448) (0.8025) Inverse distance of house to 2$62.6 2603.7 0.2656 0.0020 hazard site (175.1) (1,035.4) (0.2604) (0.9874) R 83.66 83.49 (0.0014) (84.14) Moran’s I 0.0388 0.0067 (0.0030) (0.0408) Two sites have an average coefficient 30 points larger in magnitude than sites that are not Tier Two sites: 289.79 versus 259.72. A two-sample t-test comparing the mean coefficient for these two groups of sites results in a 21.40 calculated test statistic, with a P-value of 0.164. This finding does not support the hypothesis that hazard sites whose information is on the Internet will have a larger negative impact on house prices than those sites whose information is not on the Internet. The categories summarized in Exhibit 4 are not disjoint. In Exhibit 5, I summarize the mean estimated coefficient for the three disjoint categories based on model (3). According to the R-square, the different site classifications represented by government sources explains only 3.06% of the J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 4 W i s i n g e r Exhibit 4 u Summary Statistics for the Estimated b Coefficients 1,s Mean Min. Significance* of b 1,s Category N (Std. Dev.) (Max.) Negative, Not, Positive Tier Two site 57 289.79 2329.51 25, 29, 3 (107.00) (259.19) Not Tier Two site 90 259.72 2603.66 31, 51, 8 (153.01) (464.87) Hazardous waste handlers 101 267.41 2603.66 37, 56, 8 (151.02) (464.87) TRI sites 7 2104.49 2258.04 2, 5, 0 (86.69) (225.23) Air emissions facilities 4 2146.07 2258.04 2, 2, 0 (108.71) (248.65) Water discharge facilities 2 2106.08 2145.15 0, 2, 0 (55.24) (267.02) Overall 147 270.46 2620.49 56, 80, 11 (155.80) (530.88) Note: Coefficients represent the marginal impact of a site on the surrounding house values for the different EPA Internet hazard classification and for the Tier Two hazard sites. *Significance is based on a 90% test. Exhibit 5 u Summary Statistics for the Mean Estimated Coefficients for Discrete and Overlapping Categories Based on Model (3) Average (Std. Dev.) Internet Information Impact to Surrounding T Significance* of b 1,s Categories N House Prices (P-value)* Negative, Not, Positive Level 1: Site listed on EPA 90 259.72 24.16 31, 51, 8 website; not Tier Two site (153.01) (,0.0001) Level 2: Site not listed on 36 266.55 22.93 16, 17, 3 EPA website; Tier Two site (100.58) (0.0039) Level 3: Site listed on EPA 21 2129.61 24.36 9, 12, 0 website; Tier Two site (108.25) (0.0000) Notes: R 5 3.06%; no significant difference between any of the mean coefficients. Based on test of null hypotheses that mean equals zero based on equation (3). *Significance is based on a 90% test. C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 1 5 variation in the marginal impact on house prices. The estimated impact of the sites that are listed on the EPA website, but are not Tier Two sites, have an average coefficient of 259.72 compared to an average coefficient of 266.55 for Tier Two sites that are not listed on the EPA website. While there is not a significant difference between these two means, the direction of the evidence is in the opposite direction of what would support the hypothesis that sites with information on the Internet have a larger negative price impact. But sites that are both listed on the EPA website and are Tier Two sites seemingly have the greatest impact, with an average coefficient of 2129.61. The results summarized in Exhibits 3 and 4 show that many sites have a negative impact on the nearby house prices that is not explained by EPA classification. In an attempt to better generalize about the impact of hazard sites on house prices, two other information sets are investigated. First, the businesses associated with the hazard sites are classified into one of 13 categories. Information used in this classification is obtained from the name of the business. If the name is not adequate to classify the business, the Internet is searched for the business name and / or address. In Exhibit 6, I summarize the average coefficient and tests based on fitting regression model (4). The business category information set explains 20.17% of the estimated site impact. Nine of the categories (Printing, Commercial Bakery / Food Producers, Agricultural, Automotive / Trucking Repair, Transportation, Industrial, Large Wholesale / Warehouse, Communications, and Other) have average coefficients significantly less than zero using a 90% test. These categories contain only 69% of the total sites, but 84% of the sites with a significantly negative coefficient. Two of the categories (Cleaners and Automotive/ Trucking Dealers) have an average coefficient greater than zero and contained 55% of the sites with significantly positive coefficients. The remaining two categories (Convenience / gas or oil change and City / Government) have negative average coefficients not significantly different from zero. This business classification information set explains 6.6 times the variation in hazard site impact on nearby housing prices than do the Internet disclosure categories. The second information set is based on a visual inspection of the hazard sites. The sites are classified based on whether the neighborhood surrounding the site was less than 30% residential versus more than 30% residential; whether the site occupied less than or equal to 25% of a city block, occupied more than 25% but less than one city block, or occupied at least one city block; and whether danger signs where present or not. In Exhibit 7, I summarize the average coefficient and tests based on fitting regression model (5). The three factors of this information set explain 29.46% of the variation in the coefficient estimates, which is 9.6 times the Internet disclosure categorization. When the neighborhood surrounding the site was ,30% residential, the average coefficient estimate of 2124.8 is significantly different from zero. Fifty-three percent of the sites in this category have negative coefficients. When the neighborhood surrounding the site was .30% residential, the average coefficient estimate of 10.62 is not significantly different from zero. One hundred percent of the sites with a positive and significant coefficient are in this category. The average coefficients for site sizes ‘‘#0.25 city block,’’ ‘‘.0.25 but ,1 city block,’’ and ‘‘at least 1 city block’’ are 244.15, 295.84 and 281.59, J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 6 W i s i n g e r Exhibit 6 u Summary Statistics for the Estimated b Coefficients 1,s Mean Min. T Significance* of b 1,s Business Type N (Std. Dev.) (Max.) (P-Value) Negative, Not, Positive Agricultural 8 2143.3 2258 23.16 5, 3, 0 (100.3) (18.3) (0.0019) Automotive/Trucking Dealers 11 24 270.8 0.62 0, 9, 2 (147.9) (448.5) (0.5349) Automotive/Trucking Repair 18 294 2603.7 23.11 5, 12, 1 (167.7) (309.3) (0.0023) City/Government 12 252.4 2278.1 21.42 5, 7, 0 (100.9) (90.3) (0.1593) Cleaners 12 68.8 2132.3 1.86 2, 6, 4 (195.4) (464.9) (0.0651) Commercial Bakery/Food 11 2120.5 2329.5 23.12 5, 6, 0 Production (113) (225.2) (0.0022) Communications 6 296.6 2184.4 21.85 5, 0, 1 (133.2) (148.8) (0.0672) Convenience/Gas and Oil Change 11 225.2 2217.2 20.65 2, 8, 1 (128.5) (259.2) (0.5158) Industrial 19 272.5 2352.5 22.47 4, 14, 1 (105.1) (40.7) (0.0149) Large Wholesale/Warehouse 10 2108.9 2219.7 22.69 6, 4, 0 (77.3) (27.8) (0.0081) Printing 11 2156 2274.4 24.04 8, 3, 0 (72.5) (237.1) (0.0000) Transportation 8 276.7 2205.4 21.69 4, 3, 1 (138.3) (204.5) (0.0929) Other 10 2114 2290.5 22.81 5, 5, 0 (94.7) (20.3) (0.0057) Notes: The coefficients represent the marginal impact of a site on the surrounding house values. The business type categories were determined from the business name and/or an Internet search based on the business name and/or the business address. a 2 T, P-value, and R are from model (4). The tests summarized are for the null hypothesis that the mean coefficient is equal to zero. R 5 20.17%. **Based on a 90% test of null hypothesis b 5 0 based on hedonic model (2). 1,s respectively. The main pattern is seen going from the smallest size to the next smallest, where the average coefficient estimate decreases by 52 units. The sites with danger signs have negative mean coefficients approximately 28 units larger then sites with no danger signs (267.00 for sites with no signs and 294.82 for sites with danger signs). Both the business category and the visual information sets are found to explain a significant percentage of the variation in the sites estimated impact on the sale C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 1 7 Exhibit 7 u Summary Statistics for the Estimated b Coefficients 1,s Mean Min. T Significance* of b 1,s N (Std. Dev.) (Max.) (P-value) Negative, Not, Positive Residential ,30% 89 2124.81 2352.5 27.1 47, 42, 0 (86.9) (18.3) (,0.0000) .30% 58 10.62 2603.7 20.5 9, 38, 11 (159.3) (464.9) (0.6200) Size #0.25 block 58 244.15 2603.7 22.2 16, 36, 6 (158.4) (464.9) (0.0285) .0.25 and ,1 block 47 295.84 2352.5 23.25 19, 26, 2 (96.1) (100.3) (0.0014) 1 block or more 42 281.59 2329.6 22.98 21, 18, 3 (141.9) (448.5) (0.0034) Danger Signs Not present 125 267.00 2352.5 24.4 34, 83, 8 (132.0) (464.9) (,0.0000) Present 22 294.82 2603.7 23.5 9, 11, 2 (166.3) (259.2) (0.0007) Notes: Coefficients represent the marginal impact of a site on the surrounding house values. The values for Residential, Size, and Danger sign was determined for each site by visually assessing the information with a drive by of the site. a 2 T, P-value, and R are from model (5). The tests summarized are for the null hypothesis that the corresponding mean defined in equation (6) is equal to zero. R 5 29.46%. *Based on a 90% test of null hypothesis b 5 0 based on hedonic model (2). 1,s price of nearby houses. The units of the b estimates are the change in price of 1,s the house for a unit change in the inverse distance of the house to the site. The distance is in the units defined by the Euclidean distance between house the site based on the longitude and latitude of the house and site. One mile is approximately 0.0144927 of one of these distance units. Additionally the inverse difference has a nonlinear relationship to the distance. In Exhibit 8, I summarize the estimated impact coefficients converted to dollars per specific changes in the distance in terms of miles. The average amount of decrease in house price (in dollars) for distance changes of 0.25 mile to 0.5 mile, 0.5 mile to 1 mile, and from 1 mile to 2 miles are summarized for the different business categories and for the three factors of the visual information set. The larger impact categories (Printing, Agriculture, Large Wholesale / Retail, and ,30 residential) result in around a $5,000 decrease as you approach the chemical hazardous site from 2 miles away to 1 mile, around a $10,000 decrease in price going from 1 mile to 0.5 mile, and around a $20,000 decrease in price going from 0.5 mile to 0.25 mile. J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 1 8 W i s i n g e r Exhibit 8 u Average Decrease in House Price by Distance and Characteristic Category Mean Coeff. 0.25 Mile–0.5 Mile 0.5 Mile–1 Mile 1 Mile–2 Miles Agricultural 2143.3 $19,776.16 $9,887.91 $5,142.66 Automotive/Trucking Dealers 24 2$3,312.13 2$1,656.03 2$861.30 Automotive/Trucking Repair 294 $12,972.49 $6,486.14 $3,373.42 City/Government 252.4 $7,231.47 $3,615.67 $1,880.51 Cleaners 68.8 2$9,494.77 2$4,747.30 2$2,469.05 Commercial Bakery/Food 2120.5 $16,629.63 $8,314.68 $4,324.43 Production Communications 296.6 $13,331.31 $6,665.53 $3,466.72 Convenience/Gas and Oil Change 225.2 $3,477.73 $1,738.84 $904.36 Industrial 272.5 $10,005.38 $5,002.60 $2,601.84 Large Wholesale/Warehouse 2108.9 $15,028.78 $7,514.26 $3,908.14 Printing 2156 $21,528.82 $10,764.22 $5,598.44 Transportation 276.7 $10,585.00 $5,292.41 $2,752.56 Other 2114 $15,732.60 $7,866.17 $4,091.17 ,30% residential 2124.8 $17,224.43 $8,612.07 $4,479.11 .30% residential 10.6 2$1,465.61 2$732.79 2$381.12 #0.25 block 244.2 $6,092.93 $3,046.41 $1,584.44 .0.25 and ,1 block 295.8 $13,226.43 $6,613.10 $3,439.45 1 block or more 281.6 $11,259.85 $5,629.83 $2,928.05 Danger signs not present 267 $9,246.36 $4,623.10 $2,404.46 Danger signs present 294.8 $13,085.66 $6,542.72 $3,402.85 As shown in Exhibit 4, I find that only 41 out of the 114 individual Internet- reported hazard sites studied correlate with reduced house sales prices. Perhaps more importantly, only 56 of the 147 EPA-designated sites actually have a negative impact with no individual site within any of the EPA categories more likely than not to produce a significant impact. Clearly, a better classification scheme is needed to generalize about the negative impact of chemical hazardous sites on housing prices. Congress enacted EPCRA to reduce community health risk posed by potentially hazardous commercial activities. But are EPCRA Internet disclosures responsible for lowering housing values near hazardous commercial activities? To increase the understanding of the economic impact this legislation has on the housing market, this study started with two questions. First, does the presence of an EPA- designated environmental hazard listed on the Internet correlate with lower nearby property values? The results of statistical analysis shown in Exhibit 4 indicate that housing values near EPA-Internet-listed hazards are lower, ceteris paribus. Second, do hazardous sites listed on the Internet have a bigger adverse impact on housing values than do hazardous sites not listed on the Internet? The results of this study indicate that EPA-Internet-listed hazards do not have a bigger impact than do non- listed hazards—even after two years of Internet disclosure. If anything, Tier Two C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 1 9 sites seem to have a bigger impact than Internet-listed hazardous sites. In other words, the passage of EPCRA and Internet reporting does not appear to impact community housing value patterns, at least not in the short-run. There are five possible ways to reconcile market efficiency theory with the statistical analysis of this study: (1) the housing market does not care about hazards; (2) the housing market is extremely slow to react to information; (3) because of market restrictions, housing prices are unable to react; (4) the housing market uses information sources other than the EPA to capture and factor hazard data into prices; or (5) the housing market is either unaware of or incapable of understanding the Internet disclosures. The literature indicates the first possibility is unlikely and a two-year delay in price reaction would not be consistent with an efficient market. So the only remaining explanations are either the market is unable to react, possibility due to lack of housing options, the market uses different knowledge for pricing homes, or that Internet disclosures are not fully integrated into housing markets. The broader research question developed during this study is whether EPA classifications are optimal for understanding and predicting residential market behavior. The clear answer is that other groupings yield more insight. C o n c l u s i o n The findings indicate the housing market does not seem efficient because EPA Internet disclosures do not appear to translate into residential price adjustments (i.e., Internet disclosure of apparent neighborhood danger does not seem to significantly influence residential consumer behavior). However, this finding does not dispute that homes near hazard sites have lower values. On the contrary, the results of statistical analysis shown in Exhibit 8 demonstrate rather startling impacts in terms of dollars so the public must be relying on other information sources for drawing conclusions about hazardous sites. As shown in Exhibit 6, printers, commercial food producers, agricultural organizations, vehicle repair centers, industrial sites, and large wholesalers and warehouses are clearly a locally undesirable land use. On the other hand, dry cleaners seemingly enhance a community. The statistical significance of whether the neighborhood is less than 30% residential, the size of the site, and the importance of danger signs all provides important evidence the public responds to visual cues in forming opinions about locally undesirable land uses. And although the Envirofacts website does not appear to be a major source of this information, these and previous research results reveal the public has strong aversion to living near environmental hazards. Suspected public information sources included newspapers, TV, radio, neighborhood gossip, employee inside information, and gut hunches formed after viewing or sometimes even smelling a suspect site and its surroundings. Thus, chemical hazardous sites should be grouped differently than EPA classifications for real estate investigations of house price impacts. The continuing challenge is to determine what the individual sites of major impact have in common. J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u 2 0 W i s i n g e r E n d n o t e Other visual measurements included whether the site was gated, whether there was a security fence, and whether the site was clearly dilapidated. None of these are significantly related to the impact of the hazard on surrounding housing prices. R e f e r e n c e s Abell, D. Emergency Planning and Community Right to Know: The Toxics Release Inventory. Southern Methodist University Law Review, 1994, 47:3, 581–96. Anselin, L. Spatial Data Analysis with GIS: An Introduction to Application in the Social Sciences. Technical Report 92-10, August 1992. Bui, L. and C., Mayer. Regulation and Capitalization of Environmental Amenities: Evidence from the Toxic Release Inventory in Massachusetts. Review of Economics and Statistics, 2003, 85:3, 693–708. Court, A. The Dynamics of Automobile Demand. New York, NY: General Motors Corporation, 1939. Decker, C., D. Nielsen, and R. Sindt. Is Pollution a Homogeneous Determinant of Value? Appraisal Journal, 2005, 73:2, 183–97. Epp, D. and K. Al-Ani. The Effect of Water Quality on Rural Nonfarm Residential Property Values. American Journal of Agricultural Economics, 1979, 61:3, 529–34. Fama, E. Efficient Capital Markets: A Review of Theory and Empirical Work. Journal of Finance, 1970, 25, 383–420. Greenstone, M. and J. Gallagher. Does Hazardous Waste Matter? Evidence from the Housing Market and the Superfund Program. Quarterly Journal of Economics, 2008, 123: 3, 951–1003. Hunter, L. and J. Sutton. Examining the Association between Hazardous Waste Facilities and Rural ‘‘Brain Drain.’’ Rural Sociology, 1995, 69:2, 197–212. Hwang, S. Environmental Amenities and Disamenities, and Housing Prices: Using GIS Techniques. Ph.D. Dissertation. Texas A&M University, 2003. Ihlanfeldt, K. and L. Taylor. Externality Effects of Small-scale Hazardous Waste Sites: Evidence from Urban Commercial Property Markets. Journal of Environmental Economics and Management, 2004, 47,117–39. Ketkar, K. Hazardous Waste Sites and Property Values in the State of New Jersey. Applied Economics, 1992, 24, 647–59. Kohlhase, J. The Impact of Toxic Waste Sites on Housing Values. Journal of Urban Economics, 1991, 30, 1–26. Mendelsohn, R., D. Hellerstein, M. Huguenin, R. Unsworth, and R. Brazee. Measuring Hazardous Waste Damages with Panel Modes. Journal of Environmental Economics and Management, 1992, 22, 259–71. Michael, H., K. Boyle, and R. Bouchard. Does the Measurement of Environmental Quality Affect Implicit Prices Estimated from Hedonic Models. Land Economics, 2000, 76:2, 283– Minguez, R., J. Montero, and G. Frenandez-Aviles. Measuring the Impact of Pollution on Property Prices in Madrid: Objective versus Subjective Pollution Indicators in Spatial Models. Journal of Geographic Systems, 2013, 15:2, 169–91. C h e m i c a l H a z a r d o u s S i t e s a n d R e s i d e n t i a l P r i c e s 2 1 Paterson, R. and K. Boyle. Out of Sight, Out of Mind? Using GIS to Incorporate Visibility in Hedonic Property Value Models. Land Economics, 2002, 78:3, 417–25. Rich, P. and J. Moffitt. Benefits of Pollution Control on Massachusetts’ Housatonic River: A Hedonic Pricing Approach. Water Resources Bulletin, 1982, 18, 1033–37. Ridker, R. and J. Henning. The Determinants of Residential Property Values with Special Reference to Air Pollution. Review of Economics and Statistics, 1967, 49, 246–57. Rosen, S. Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition. Journal of Political Economy, 1974, 82, 34–55. Simons, R. and J. Saginor. A Meta-Analysis of the Effect of Environmental Contamination and Positive Amenities on Residential Real Estate Values. Journal of Real Estate Research, 2006, 28:1, 71–104. Sirmans, G., D. Macpherson, and E. Zietz. The Composition of Hedonic Pricing Models. Journal of Real Estate Literature, 2005, 13:1, 3–43. Skillern, F. Environmental Protection Deskbook. Second edition. Deerfield, IL: McGraw- Hill, Inc., 1995. Smith, V. and W. Desvousges. The Value of Avoiding a LULU: Hazardous Waste Disposal Sites. The Review of Economics and Statistics, 1986a, 68, 293–99. ——. Asymmetries in the Valuation of Risk and the Siting of Hazardous Waste Disposal Facilities. American Economic Review, 1986b, 76:2, 291–94. Smith, V. and J. Huang. Hedonic Models and Air Pollution: Twenty-Five Years and Counting. Environmental and Resource Economics, 1993, 3, 381–94. Smolen, G., G. Moore, and L. Conway. Hazardous Waste Landfill Impacts on Local Property Values. Real Estate Appraiser, 1992a, 58:1, 4–11. ——. Economic Effects of Hazardous Chemical and Proposed Radioactive Waste Landfills on Surrounding Real Estate Values. Journal of Real Estate Research, 1992b, 7:3, 283–95. Thayer, M., H. Albers, and M. Rahmatian. The Benefits of Reducing Exposure to Waste Disposal Sites: A Hedonic Housing Value Approach. Journal of Real Estate Research, 1992, 7:3, 265–82. Wisinger, P. The Impact of Chemical Hazardous Sites on Residential Values. Ph.D. Dissertation, Texas Tech University, 2006. I would like to gratefully acknowledge the contributions of Paul Goebel and Ronald Bremer. Their time and thoughts are greatly appreciated. Perry Wisinger, Regis University, Denver, CO 80221 or pwisinger@regis.edu. J O S R E V o l . 6 N o . 1 – 2 0 1 4 u u This page intentionally left blank
Journal
Journal of Sustainable Real Estate
– Taylor & Francis
Published: Jan 1, 2014
