The Economic and Health Effects of the 2014 Chemical Spill in the Elk River, West Virginia

The Economic and Health Effects of the 2014 Chemical Spill in the Elk River, West Virginia Abstract In January 2014, Freedom Industries spilled 4-methylcyclohexylmethanol, a chemical foaming agent used in coal processing, from a storage facility into the Elk River in West Virginia. This chemical spill, one of the most significant in U.S. history, adversely affected the drinking water supply of over 300,000 individuals in the Charleston, West Virginia Metropolitan area. We use synthetic control methods to estimate the casual effects on macro-economic growth and infant health outcomes from this water crisis. We find a significant decrease in 5-minute Apgar Scores, a measure of how babies fare in the birthing process and outside of the womb, after the chemical spill. We do not find significant effects for infant birthweight or gestational age. We find a statistically insignificant decrease of per capita GDP in the Charlestown, West Virginia area compared to the synthetic control of 3% two years after the chemical spill. A safe and reliable supply of water is essential to human health (Hunter, MacDonld, and Carter 2010). Exposure to various organic or inorganic chemicals can lead to detrimental health effects, including nausea, vomiting, skin rashes, cancer, and fetal abnormalities (Hunter 1997). Water pollution exposure can also have indirect impacts to individuals, such as disease from malnutrition, hindered food production, reduced labor productivity, and increased risk of financial stress. Suffice it to say, historic water pollution events have had large effects on human communities (e.g., Schwabach 1989; Saha 2003; Shaban et al. 2009). Water insecurity can lead to increased pressure on poverty and social unrest (Sadoff and Grey 2007) and water-related shocks can negatively influence public health and economic stability. Researchers have found evidence that economic growth is affected by large environmental disasters (Cavallo et al. 2013). Large disturbances to economic activity through the destruction of infrastructure, death, disease, and financial losses can all contribute to lower growth in both the short and long term. Despite greater environmental protections and recent advancements in water treatment, there have been several mid- to large-scale contamination events in the United States. These include the Deepwater Horizon oil spill on the Gulf Coast (Camilli et al. 2010), the Gold King Mine wastewater spill in Colorado (Parker 2015), the Flint, Michigan water crisis in 2014 (Hanna-Attisha et al. 2016), and West Virginia’s Elk River Spill in early 2014. Water safety and security increasingly relies on evaluating the risks and causal effects of contamination events. A difficulty with assessing the local impacts of a water contamination event is that there is typically only one “treated” observation. In typical regression analyses, one compares average outcomes for a series of treated observations to outcomes from observationally-similar but untreated “control” observations. However, regression analyses contain certain statistical characteristics which may over-extrapolate effects1 1 Abadie et al. (2010, 2015) discuss the implications of weights in regression analysis, where weights not constrained to be positive and sum to one can extrapolate effects and yield less-accurate results. (Abadie, Diamond, and Hainmueller 2010; Abadie, Diamond, and Hainmueller 2015). For example, regression analysis will give equal weights to all control units regardless of how similar they are to the treated unit. In the economics literature, the synthetic control method has recently been used to assess treatment effects for single treated observations (Abadie, Diamond, and Hainmueller 2010; Coffman and Noy 2011; Abadie, Diamond, and Hainmueller 2015). This approach compares outcomes for a single treated observation to the outcomes of a “synthetic” control observation, constructed as a weighted average of the universe of potential control observations. The development of synthetic control methods provides a better statistical framework for analyzing singular events with aggregate level data. In this paper, we apply the synthetic control method to assess the economic and public health effects of a large water contamination event: The Elk River Chemical Spill in West Virginia. In January 2014, Freedom Industries released approximately 10,000 gallons of chemicals used for processing coal into the Elk River (U.S. Chemical Safety and Hazard Investigation Board 2016). The river provides drinking water to multiple counties in West Virginia, including the state capital of Charleston. The spill led to a shut-down of restaurants, hotels, and the local mall, and created a drinking water emergency involving responses from local, state, and national agencies. Concerns persisted about contact and consumption of the Elk River water. Toxicity levels of some of the chemicals were not well understood by the scientific community, which created additional concern about safe exposure limits. This paper is a part of a Teaching Case Study section of this journal, and a teaching note, data, and code are available to facilitate using this paper in the classroom.2 2 All these elements can be found online at Dr. Guilfoos and Dr. Hill’s personal websites. Available at: http://www.toddguilfoos.com/research.html and http://www.elainelhill.com/re. The primary county affected by the Elk River spill is Kanawha County, West Virginia. In our approach, we can make causal inferences by comparing post-spill economic and public health outcomes between a “synthetic” county and the real Kanawha County, West Virginia. The synthetic Kanawha County is made up of a weighted linear combination of counties unaffected by the chemical spill and has pre-spill characteristics that are very similar to Kanawha County. We argue that the weighted combination of control counties can act as a better control to measure causal effects of the chemical spill because the synthetic control better matches the treated unit (Abadie, Diamond, and Hainmueller 2010; Abadie, Diamond, and Hainmueller 2015; Coffman and Noy 2011). We use the synthetic control method to address two main questions: Was economic growth of the area impacted by the Elk River spill? Were birth outcomes affected by the Elk River spill? We examine these outcomes because they are readily available, are used in other studies, and represent social costs above those reported as a result of closed businesses and reported illness in the immediate aftermath of the chemical spill. The first question identifies if water-related shocks lead to dampened long-term economic growth. The second question addresses whether infant health is particularly vulnerable to in-utero exposure to water contamination (Galiani, Gertler, and Schargrodsky 2005; Gamper-Rabindran, Khan, and Timmins 2010; Currie and Walker 2011; Currie and Schwandt 2015). By focusing on pregnancy outcomes, we avoid the risk of attenuation due to avoidance behavior. We find weak evidence of a longer-term effect to aggregate economic outcomes to Charleston, West Virginia from the chemical spill, and we cannot rule out the null hypothesis of no effect. There is a statistically-insignificant 3% decline in GDP per capita in Charleston, West Virginia two years after the event. We find suggestive evidence of a negative effect on infant health outcomes for infants born in Kanawha County after January 2014. This effect is substantial for 5-minute Apgar Scores but relatively short-lived, dissipating for birth cohorts born approximately four to five months after the Elk River spill. Placebo test results are supportive of a strong post-January 2014 effect relative to unaffected counties elsewhere in Appalachia. However, we do not find significant effects for other infant birth outcomes, such as birthweight and gestational length. Background on the Elk River Spill On January 9, 2014, a chemical leak was discovered at a chemical distribution facility in Charleston, West Virginia (Trip 2014; Markham, Gianato, and Hoyer 2016). Approximately 10,000 gallons of 4-methylcyclohexanemethanol (MCHM) and Propylene glycol phenyl ether (PPH) were discharged from Freedom Industries into the nearby Elk River, quickly infiltrating the intake and distribution plant of West Virginia American Water Company (WVAWC). The WVAWC utilized a filtration system equipped with activated carbon to mitigate such incidences, but the filters became saturated and ineffective, exposing the drinking water supply to these chemicals (Howard 2014). Figure 1 illustrates the location of the spill in relation to the city of Charleston. Figure 1. View largeDownload slide Location of the spill in Charleston West Virginia Figure 1. View largeDownload slide Location of the spill in Charleston West Virginia Once filters could no longer handle the quantities of the chemicals in the river, WVAWC concluded that the water was unsafe to drink (Howard 2014; Watkins and Ellis 2016). Approximately 300,000 citizens of the Charleston metropolitan area were unable to use tap water for 4–9 days (Markam, Gianato, and Hoyer 2016). The chemicals spilled were used to clean and wash coal before processing and had relatively unknown health effects (Trip 2014). One day after the spill, 122 people had visited hospitals for symptoms of nausea and vomiting, and 4–6 were admitted (Trip 2014; Heyman and Fitzsimmons 2014). To cope with the absence of potable water, West Virginia Governor Early Ray Tomblin and President Barack Obama declared a state of emergency for nine affected counties, enabling the National Guard to bring in tanks of water for residents (Howard 2014). The Federal Emergency Management Agency (FEMA) provided additional aid, bringing over three million liters of water to the affected area and working closely with the state “to ensure there [were] no unmet needs in helping those impacted by the incident” (FEMA 2014). FEMA also set up Incident Management Assistance and Mobile Emergency Response teams on site in Charleston to help coordination. The National Guard began water testing on January 10, 2014, using the 1 part-per-million benchmark suggested by the Centers for Disease Control and Prevention as a safe screening standard; initial levels at the West Virginia American Water intake site were reported as high as 3.35 parts per million (Markam, Gianato, and Hoyer 2016). Authorities lifted the water use ban on January 18, almost two weeks after the spill. Government response to the spill was very involved, and a wide array of partnerships with local, state, federal, and private organizations and agencies were initiated to help mitigate the crisis.3 3 The list includes local health departments, the Kanawha-Charleston Board of Health, Kanawha County Board of Commission, Kanawha County Emergency Management, City of Charleston Emergency Management, the Mayor of Charleston, leaders of other townships, the West Virginia Department of Health and Human Resources, Governor’s office and Poison Control Center, along with U.S. Senators and Representatives, the National Guard and others (Snair 2014). The National Guard provided aid through troops helping staff reopen schools and businesses. A rapid-response team of National Guard troops, school officials, and health department staff was formed to address any complaints at schools across the county for several weeks (Snair 2014). Costs of the Spill Although the stated emergency resulting from the MCHM spill lasted less than two weeks, many effects of the spill were longer-lasting. According to a preliminary study by the Marshall University Center for Business and Economic Research (CBER), an estimated total of $61 million in damages was incurred by local businesses and residents (CBER 2014). Nearly 75,000 workers were unable to work during each day of the ban, representing over 40% of the working population in the area. The costs incurred by residents who bought bottled water, paid for extra childcare, and medical expenses were not included in this impact, nor were future health implications or economic disturbances. Within a week of the spill, twenty-five lawsuits had been filed against Freedom Industries and it subsequently filed for bankruptcy (White 2014). Lawsuits were also filed against the West Virginia American Water Company and chemical manufacturer, Eastman Chemical. One $151 million settlement—$126 million to be paid by W. Virginia Water and $25 million by Eastman—was reached in 2016 and will ultimately be distributed to affected businesses and residents (Raby 2016). Beyond the immediate costs from the spill are other important economic impacts, such as changes to economic activity where water is used in food preparation or is integrated into products, or longer-term health effects that may decrease productivity through increased absences or decreased output at work. Perceptions of unsafe water quality to visitors may also suppress growth in the economy if business is taken elsewhere. These effects could be longer-lasting than the period of the stated emergency as perceptions and illness persist. The CDC officials stated in a press conference on February 5, 2014, that the water was “appropriate” to drink, but five days later, multiple expert witnesses were unable to conclusively report on the water’s continued danger. “Because of the level of mistrust, the public is slow to return to using the water,” said Dr. Rahul Gupta, Executive Director of the Kanawha-Charleston Health Department. “Survey data from the time of the spill until March 1 shows that less than five percent of the population are drinking or cooking with our local water, and approximately 20 percent are not using the water for any purpose. There has been a secondary wave of health impacts after the flushing which has further exacerbated the frustrations of a very anxious and suspicious community,” (Snair 2014). Our model estimates would include some economic costs stated above, such as decreases in productivity and disturbances to businesses. But other costs would be in addition to our estimates, such as legal costs and averting expenditures borne by victims of the spill. The health costs we measure through birth outcomes would be in addition to any of the immediate costs through visits to the hospital, usually associated with stomach illnesses and nausea. Chemical Storage Policies and Regulations A key component of the costs of avoidance is evaluating risks and the costs of reducing risks of future contamination events. The U.S. Chemical and Hazard Investigation Board (CSB) conducted a thorough investigation of the Elk River spill. The CSB found a lack of effort by Freedom Industries to properly inspect and maintain equipment, though they were compliant with existing state and federal regulations. Regulations for chemical manufacturers and distributors are often difficult to navigate. The Occupational Safety & Health Administration (OSHA) classifies both crude-MCHM and stripped-PPH as “hazardous chemicals” (CSB 2016). Every manufacturer or distributor storing more than 10,000 pounds of MCHM or PPH stripped is required by the Emergency Planning and Community Right-To-Know Act to submit a form to its Local Emergency Planning Committee regarding the stored amount (CSB 2016). At the time of the spill, Freedom Industries was subject “…to the West Virginia National Pollution Discharge Elimination System (NPDES),” a general permit for storm water discharge near industrial activity. Other regulators such as the Environmental Protection Agency (EPA) define “hazardous chemicals” and “hazardous substances” differently, and thus the chemicals are subject to different regulations. While OSHA classifies “hazardous chemicals” as any chemical that poses physical or health hazards, the EPA’s Clean Water Act defines “hazardous substances” as a substance where “the discharge of which may be harmful to the public health or the environment of the United States” (CSB 2016). While the EPA is required to establish regulations of these substances under the Clean Water Act, neither MCHM nor PPH stripped are listed. Storage container regulations are equally difficult to maneuver and enforce. At the time of the MCHM spill, aboveground storage tanks (ASTs) were inadequately regulated in West Virginia. No uniform regulation program exists for all ASTs, and states are charged with instituting regulations not addressed by the federal government. The 1984 state legislature established a “comprehensive statutory framework… regulating underground storage tanks, but it did not address ASTs,” (CSB 2016). The CSB investigation revealed that the three tanks storing MCHM and PPH were initially installed in 1938 and held glycerin or calcium chloride prior to 2009 (CSB 2016). American Petroleum Institute (API)-certified inspectors found the tank shells and roof were constructed with a now-obsolete construction; the bottoms appeared to be replacements of the originals (CSB 2016).4 4 According to the CSB, tank shells and roofs were constructed using a technique called lap-riveting; welding began to replace the process of riveting during the 1930s. The bottoms were lap-welded, but looked to be replacements of the originals. Two floor holes, 0.75 and 0.4 inches in diameter, were the source of the leak. Inspectors also found deep pits in the tank shell and floor, and determined the holes started as pits formed from corrosion. Examining Freedom Industries’ inspection protocols, the CSB determined the manufacturer did not have a program in place that would ensure the ASTs were maintained and inspected regularly. Freedom Industries also did not keep detailed history, maintenance, or inspection records for the failed tank as they were not forced to comply with regular inspections (CSB 2016). The CSB also reported on the lack of a leak detection system (LDS): “Freedom [Industries] did not have any level indication device, gauge system or measurement to capture the actual amount of MCHM leak, which contributed to the changing estimates of the spill amount. There was no West Virginia state or federal requirement that would have made the installation of an LDS mandatory for ASTs.” Compliance with additional federal regulations could have indirectly prevented the leak from entering the Elk River.5 5 The Spill, Prevention, Control and Countermeasure (SPCC) rule regulates various industrial chemicals and liquids; the CSB identified that Freedom Industries also stored an oil on-site, which was under regulation of the SPCC. Freedom Industries had not been complying with these requirements, specifically for secondary containment (CSB 2016). The proximity of the oil AST to the MCHM AST led CSB investigators to conclude that secondary containment may have prevented the spill had Freedom Industries been complying. Methodology Many of the direct costs and risks enumerated above are known, but to understand the larger costs to the economy and risks to public health we use the synthetic control method to analyze outcomes from this water crisis. The synthetic control method has had limited yet growing use in the economics literature (Abadie and Gardeazabal 2003; Abadie, Diamond, and Hainmueller 2010; Cavallo et al. 2013; Abadie, Diamond, and Hainmueller 2015). This method is used to compare a specific region or entity exposed to an intervention. In this case study, the intervention is the Elk River Chemical Spill. The main entity exposed is the Charleston Metropolitan Area or Kanawha County, West Virginia, depending on the outcome variable. The synthetic control is a weighted average of the control units; this allows for explicit inference of the “relative contribution” of the control units and explains the similarities (or differences) between the unit affected and the synthetic control. The weights of the controls can also be determined to be positive and sum to one, which can help guard against extrapolation errors. The synthetic control model relies on a series of constructed vectors and matrices to perform the analysis. Following Abadie, Diamond, and Hainmueller (2010), the outcome variable investigated is observed for T periods for the region exposed to the intervention, Y1t, where t=1,…, T and the synthetic control, Yjt, where j=2, …, J+1 and t=1,…, T. T1=T- T0 is the number of post-intervention periods, and Y1 is a (T1×1) vector of post-intervention outcomes for only the exposed region. This leaves Y0, a (T1×J) matrix, comprising the post-intervention outcomes for the control regions. Listing the treated region as the first of J regions is only done for convenience. We let a (T0×1) vector K=(k1,…,kT0)' be a weighting vector, and then define a linear combination of pre-exposure outcomes as Y-iK=∑s=1T0ksYis. There can be M values of K to form any linear combination, and therefore M linear combinations defined by the vectors K1,…KM. This allows for weights to be assigned to control units in a way that best fits the data. In our analysis we will rely on a single vector of weights and a linear combination of control unit outcomes. Next, we represent the pre-intervention characteristics of the treated region with a k×1 vector X1=(Z1',Y-1K1,…,Y-1K1)'. The vector Z is a set of explanatory variables that are used to predict outcomes, and which are not affected by the treatment. The Z vector is generally chosen through an iterative process, not unlike matching protocols, but is initially chosen based on the expected relationship to the outcome variable. The pre-intervention characteristics for the unaffected regions are represented by a similar k×J matrix containing the same variables for the untreated areas, X0=(Zj',Y-jK1,…,Y-jK1)'. We also let W be a (J×1) vector of positive weights, where W=(w2,…,wJ+1)' and w2+…+wJ+1=1; each value of W is a weighted average of all the available control regions. To measure the difference between the treated and untreated units, we take the distance X1-X0W=(X1-X0W)'V(X1-X0W). Here, V is some (k×k) symmetric and positive semidefinite matrix. Our synthetic control weight W* is chosen to minimize this distance, and the optimal choice for V minimizes the mean square error of the synthetic control estimator. The basic idea of this method is to construct a nearly identical county to our affected county in West Virginia in an effort to understand how outcomes like economic growth and infant health have changed because of the chemical spill. We compute the p-value of the effects found on an outcome variable through a permutation test using equations 1 and 2, following Cavallo et al. (2013). Equation (1) defines the estimate of the difference between outcomes from the treated unit, Y1t, and the synthetic control outcomes, Yjt, using the set of optimal weights, wj*, found by minimizing the distance between pretreatment observations and controls. Further, α^1t is the estimator of this difference for unit 1, which is the treated unit at time t,   α^1t=|Y1t-∑j=2J+1wj*Yjt| (1)  p-valuet=Pr⁡α^1tPL<α^1t=∑j=2J+1Iα^1tPL(j)<α^1tJ. (2) In equation (2), the term α^1lPL refers to the estimate for each placebo of a donor unit used to construct the synthetic control for unit 1. Each donor unit is a county (or other unit of observation) unaffected by the event of interest. There are J donor units which are used to construct a distribution of placebo estimates. The placebo estimate of α^1tPL(j) for each j donor unit, which are estimated with same choice of controls as our unit of interest, is used to determine significance of the estimate of α^1t. Moreover, I· is an indicator function which returns 1 if the interior argument is true and 0 if false. This procedure provides the rank of the estimate of α^1t compared to the distribution of placebos estimates α^1tPL. Economic Growth Data We apply the synthetic control method to the Charleston, West Virginia Metropolitan Statistical Area (MSA) for economic growth outcomes as defined by annual GDP per capita (Bureau of Economic Analysis 2015). We use a combination of demographics and economic characteristics as additional controls to match our treated MSA: the percentage who have attained a bachelor’s degree or higher, population count, and jobs by major industry (U.S. Census Bureau, American Community Survey 2015).6 6 We use 2013 1-year estimates for the demographic and industry data from the U.S. Census ACS. This choice was made because the definition of MSAs can change year-to-year. Importantly, in 2012 Charleston, WV, was re-defined. This complicates averaging across years as is typically done with control variables, so we choose 2013 as the year to match demographic and industry data to have a consistent definition for all MSAs. The GDP per capita data from the BEA has been adjusted for the change in definition of MSAs across all years. We also found that by using 5-Year estimates from ACS survey data, that our synthetic control is very similar to the results presented here. After eliminating MSAs that do not have a balanced panel over 2009 to 2015, we retain a total of 328 control units to construct the synthetic Charleston, MSA. The selection of variables is based on a backwards induction search that finds a good balance between Charleston, West Virginia, and the synthetic Charlestown, and has a good pre-treatment fit in predicting GDP per capita. The choice of variables is not made ad hoc; rather, we select variables that would be expected to have an effect on GDP per capita and improve pretreatment fit as measured by the root mean squared error (RMSE). The RMSE is a measure of fit that tracks the distance between prediction of synthetic control and the treated observation of the outcome variable. We choose 2009 as the starting point for the analysis because of the 2008 financial crisis, which may have caused structural shifts to regional economies. By using only post-2008 data, we do not restrict the synthetic control group to match the pre- and post-periods of the recession as this may be asking too much of the synthetic control method. By including pre-2008 data, we are asking the method to match pre-financial crisis trends, the decline, and recovery of economic growth. The treatment period is 2014 as the chemical spill occurred in January of 2014. The occurrence of the spill in January is advantageous since all variables are captured annually, making measurement of the post-treatment effects less likely to be attentuated. Infant Health Data We use a variety of monthly infant health outcomes from the National Vital Statistics System’s Birth Data files from 2011 to 2014. We received restricted-use data, which describes mother demographic information and health outcomes for the universe of births in the United States. The location of the birth is provided at the mother’s residential county-level. Mother’s home address-level data are available from the state, but we concluded that this level of disaggregation was not necessary given our identification strategy. We also restrict the data to singleton births for this analysis as is standard in the health literature. We estimate models for three main birth outcomes of interest. First, we use the Apgar score, a value ranging from 1–10 that indicates how well the baby performed through the birthing process and outside of the womb. Values are based on a series of respiratory, cardiovascular, muscular, reflexive, and skin color tests conducted within 5 minutes of birth. Values of 7 or above indicate that the baby is in good health; scores lower than 7 indicate that the baby needs medical attention. Second, we use the baby’s gestational age at birth. Typical full-term gestational age at birth ranges from 37 to 42 weeks. A gestational age at birth below 37 weeks is classified as premature. Third, we use the baby’s birthweight in grams. Babies are considered as having low birthweight if they weigh less than 2,500 grams. Both gestational age and birthweight are related to motor and social development (e.g., Hediger et al. 2002). We estimate both continuous and binary models of the variables described above. In the binary models, we define the dependent variable to equal 1 if the baby is born with an Apgar Score below 7, is born at less than 37 weeks in gestational age, and is born with a birthweight of less than 2,500 grams. These binary variables allow us to infer how the proportion of babies born with poor health outcomes changes in response to the Elk River Chemical Spell. We aggregate our outcomes and control variables to the county-month level. In order to create a balanced panel, we only use those counties that have monthly data from 2011 to 2014. We also only use those counties that are located within nearby states in the Appalachian region of the United States. These states include Kentucky, Maryland, Ohio, Pennsylvania, Tennessee, Virginia, and West Virginia. We use this additional restriction to lower the possibility of assigning positive weights to counties in regions of the states with unobserved characteristics that may lead to differences in birth outcomes. Additionally, we exclude nearby, more rural counties that were also impacted by the Elk River Chemical spill. These counties include Boone, Clay, Jackson, Lincoln, Logan, Putnam, Roane, and Cabell counties in West Virginia. We argue that the health effect in these counties may be mitigated by both distance and time. As additional controls in our infant health analyses, we use annual, county-level economic data from the U.S. Bureau of Labor Statistics and Census Bureau (Local Area Unemployment Statistics; American Community Survey 2015). We include median household income, poverty rate, and unemployment rate as controls for regional economic conditions. We include median household income, poverty rate, and unemployment rate as controls for regional economic conditions. We also use county-year means of the following mother characteristics from the NCHS: percentage of mothers who are black; percentage of mothers who are unmarried; and percentage of mothers that are at least 35 years old (National Center for Health Statistics 2009–2014). These variables are selected because they are general predictors of birth outcomes and health and should therefore be a part of the balance in constructing a synthetic Kanawha County, West Virginia.7 7 Models of infant health outcomes use a variety of mother characteristics, including percentages of mothers in different groups based on race, age, educational attainment, marital status, level of cigarette consumption, and the receipt of Women, Infants, and Children (WIC), as well as percentage of babies that were female. Data on educational attainment and cigarette smoking were missing for West Virginia during our study period. Our results are robust to the inclusion of other variables in the synthetic control formation, including percentage of mothers who are Hispanic and percentage of mothers who are < 20 years old, as well as percentage of babies who are female. Economic Growth Results The makeup of the synthetic Charleston, West Virginia, MSA is listed in table 2 as the weights given to each MSA based on pretreatment matches to the variables in table 1. Our results, shown in figure 2, suggest a divergence between the synthetic Charleston, West Virginia and the real Charleston, West Virginia MSA, suggesting that the chemical spill may have a longer-term effect on economic growth. The synthetic control shows a high degree of tracking Charleston MSA before the chemical spill and there is a high degree of balance between predictors. We find a decrease of 1% of GDP per capita in 2014 and a decrease of 3% in 2015. Table 1. Means of Predictors for GDP per capita Variables  Charleston, WV  Synthetic Charleston  Rest of United States  % of Bachelor degree or higher  22.8  22.8  26  Count of jobs in public sector  8,715  7,555  78,893  Count of jobs in health care  23,245  22,993  16,863  Total population estimate  224,727  227,342  713,147  Count of white population  202,050  195,285  521,080  GDP per capita 2009  56,765  56,742  40,485  GDP per capita 2011  59,847  59,820  41,011  GDP per capita 2013  56,163  56,135  41,432  Variables  Charleston, WV  Synthetic Charleston  Rest of United States  % of Bachelor degree or higher  22.8  22.8  26  Count of jobs in public sector  8,715  7,555  78,893  Count of jobs in health care  23,245  22,993  16,863  Total population estimate  224,727  227,342  713,147  Count of white population  202,050  195,285  521,080  GDP per capita 2009  56,765  56,742  40,485  GDP per capita 2011  59,847  59,820  41,011  GDP per capita 2013  56,163  56,135  41,432  Table 2. Weights to MSAs for Synthetic Charleston, West Virginia MSA  State  Weight  Casper  WY  0.538  Elkhart-Goshen  IN  0.165  Ogden-Clearfield  UT  0.111  Duluth  MN-WI  0.086  Hanford-Corcoran  CA  0.052  Peoria  IL  0.023  McAllen-Edison-Mission  TX  0.021  Washington-Arlington-Alexandria  DC-VA-MD-WV  0.004  MSA  State  Weight  Casper  WY  0.538  Elkhart-Goshen  IN  0.165  Ogden-Clearfield  UT  0.111  Duluth  MN-WI  0.086  Hanford-Corcoran  CA  0.052  Peoria  IL  0.023  McAllen-Edison-Mission  TX  0.021  Washington-Arlington-Alexandria  DC-VA-MD-WV  0.004  Figure 2. View largeDownload slide Trends in GDP per capita Figure 2. View largeDownload slide Trends in GDP per capita To infer statistical significance of these impacts, we apply a permutation test, sometimes called a placebo test, on unaffected MSAs to see if the change in post-treatment GDP per capita is large compared to the Charleston synthetic analysis. The placebo test posits that unaffected MSAs should not have significant differences in post-treatment outcomes since they did not experience a water crisis, and if there are many placebos with larger differences from their synthetic controls, but good balance pre-treatment, then our measured effect is not statistically significant but rather within the statistical variation of the data. Figure 3 shows that we cannot reject the null hypothesis that the water crisis had no effect on longer-term economic growth since there are many placebos with larger post-treatment variations than the Charleston analysis suggests. We have trimmed the placebo data to only include placebo MSAs with a RMSE within 4 times of the Charleston, West Virginia RMSE, with the goal of excluding placebos that did not have good pretreatment matches with their synthetic controls. Figure 3. View largeDownload slide GDP placebo test Note: This figure uses the remaining MSAs and calculates the distance from their synthetic control. We trim the placebo MSAs to only show placebos with a RMSE within 4 times of the Charleston, West Virginia RMSE pretreatment. The black-dashed line represents the Charleston, West Virginia MSA. Figure 3. View largeDownload slide GDP placebo test Note: This figure uses the remaining MSAs and calculates the distance from their synthetic control. We trim the placebo MSAs to only show placebos with a RMSE within 4 times of the Charleston, West Virginia RMSE pretreatment. The black-dashed line represents the Charleston, West Virginia MSA. We follow the procedure in equations (1) and (2) to find the p-value of our estimates from the placebo tests. We find the estimated p-value is 0.372 one-year post-treatment, and 0.199 two years post-treatment, which indicates that the estimated effect of the water crisis on economic growth in Charleston, West Virginia is not statistically significant. The strength of the synthetic control method is its ability to define a better statistical counterfactual for a singular event such as a water crisis, and allows the effects of confounding unobserved characteristics to vary with time, unlike a fixed effect regression model. The weaknesses of this approach for our given question is the short timeframe of pretreatment periods due to the concerns generated from the housing market collapse and 2008 financial crisis. Another challenge of our data is the size of shifts in GDP per capita in the Charleston, West Virginia pretreatment period, which have the potential of masking the significance of the estimate through a type II error. We rely on few control MSAs to construct the synthetic control, though there are no obvious disadvantages to relying on few control units if they are good matches and predictors for the treated unit. Infant Health Results We apply the same methodology to county-month-level infant health outcomes from the National Center for Health Statistics. We estimate synthetic control models for 5-minute Apgar Score, birthweight (in grams), and gestational age (in weeks). We also estimate models using binary variables equal to 1 if those variables are indicative of poor infant health: 5-minute Apgar Score < 7; birthweight < 2,500 grams; and gestational age < 37 weeks. Our synthetic control model selects control counties using a function of 2011 to 2013 annual values of median household income, poverty rate, unemployment rate, the health outcome of interest, and mother characteristics, including the percentages of mothers who are black, over the age of 35, and unmarried. Table 3 shows the predictor variables used when constructing the synthetic controls. We choose the pretreatment average annual median household income, poverty percentage, unemployment rate, average county-year mother characteristics, and the respective health outcome as the predictor variables in the health analysis. This table shows a strong balance of the variables for all three synthetic control analyses. It also highlights the value of synthetic control analyses. Kanawha County, West Virginia is different from other counties in Appalachia; relative to the rest of Appalachia, it has a lower unemployment rate, a slightly-higher proportion of the mother population that is black, and a significantly higher proportion of the mother population that is unmarried.8 8 Similar results are found with slightly less restrictions on averaging covariates across years, but the presented combination of controls provided the best pretreatment fits to the data. Table 3. Predictor Balance between Appalachia Counties, Kanawha County, and Synthetic Kanawha Counties Variable  Appalachia  Kanawha County  Synthetic Kanawha County   Apgar Score  Birth weight  Gestational age  Mean Apgar Score            2011  8.9  8.9  8.9      2012  8.9  9.0  9.0      2013  8.9  9.0  9.0      Birth weight            2011  3.3  3.2    3.3    2012  3.3  3.3    3.3    2013  3.3  3.3    3.2    Gestational Age            2011  38.6  38.3      38.2  2012  38.6  38.3      38.3  2013  38.6  38.3      38.3  Med. Household Income            2011  43.9  40.4  42.1  41.6  42.9  2012  44.8  45.8  45.0  44.7  45.3  2013  45.8  45.9  45.1  45.2  45.9  Poverty %            2011  17.4  17.2  15.7  15.8  15.6  2012  17.4  14.4  15.7  15.7  15.7  2013  17.2  15.3  15.2  15.5  16.1  Unemployment Rate            2011  9.3  7.1  7.6  7.5  7.9  2012  8.3  6.6  6.7  6.7  6.8  2013  8.2  5.9  6.1  6.2  6.2  % Mothers: Black            2011  7.9  7.9  8.2  8.5  8.4  2012  7.9  9.1  9.0  8.8  8.9  2013  8.0  9.2  8.7  8.9  9.1  % Mothers: Unmarried            2011  41.0  46.7  45.9  46.6  44.8  2012  41.4  48.6  47.2  47.3  48.0  2013  42.0  49.2  49.3  48.6  48.1  % Mothers: Age > = 35            2011  10.1  9.6  9.7  10.1  10.2  2012  10.0  8.8  9.5  9.5  9.1  2013  10.0  9.6  9.3  9.3  8.9  Variable  Appalachia  Kanawha County  Synthetic Kanawha County   Apgar Score  Birth weight  Gestational age  Mean Apgar Score            2011  8.9  8.9  8.9      2012  8.9  9.0  9.0      2013  8.9  9.0  9.0      Birth weight            2011  3.3  3.2    3.3    2012  3.3  3.3    3.3    2013  3.3  3.3    3.2    Gestational Age            2011  38.6  38.3      38.2  2012  38.6  38.3      38.3  2013  38.6  38.3      38.3  Med. Household Income            2011  43.9  40.4  42.1  41.6  42.9  2012  44.8  45.8  45.0  44.7  45.3  2013  45.8  45.9  45.1  45.2  45.9  Poverty %            2011  17.4  17.2  15.7  15.8  15.6  2012  17.4  14.4  15.7  15.7  15.7  2013  17.2  15.3  15.2  15.5  16.1  Unemployment Rate            2011  9.3  7.1  7.6  7.5  7.9  2012  8.3  6.6  6.7  6.7  6.8  2013  8.2  5.9  6.1  6.2  6.2  % Mothers: Black            2011  7.9  7.9  8.2  8.5  8.4  2012  7.9  9.1  9.0  8.8  8.9  2013  8.0  9.2  8.7  8.9  9.1  % Mothers: Unmarried            2011  41.0  46.7  45.9  46.6  44.8  2012  41.4  48.6  47.2  47.3  48.0  2013  42.0  49.2  49.3  48.6  48.1  % Mothers: Age > = 35            2011  10.1  9.6  9.7  10.1  10.2  2012  10.0  8.8  9.5  9.5  9.1  2013  10.0  9.6  9.3  9.3  8.9  Note: Annual economic data is taken from the U.S. Census Bureau. Mother birth and socio-economic characteristics data are from the National Center of Health Statistics' Vital Statistics Natality Birth Database. Birthweight is measured in thousands of grams. Median Household Income is measured in thousands of dollars. Table 4 lists the weights used in determining Kanawha County’s synthetic control for each continuous infant health outcome. Most counties selected as components within the Kanawha County synthetic control are semi-rural counties containing towns with a population size ranging from 10,000 to 100,000. The two counties that form large proportions of Kanawha County’s synthetic control across all three health outcomes—Ohio and Wood Counties in West Virginia—contain cities that are similar in size to Kanawha County’s Charleston (Wheeling and Parkersburg, respectively). Table 4. Weights for Synthetic Kanawha County, West Virginia County  State  Synthetic control weight   Apgar Score  Birth weight  Gestational age  Henderson  KY      0.173  Woodford  KY      0.231  Howard  MD  0.033  0.033    Mercer  OH  0.034      Elk  PA  0.103  0.120    Lawrence  PA  0.050      Hamilton  TN    0.073    Lincoln  TN    0.129    Culpepper  VA    0.022    Cumberland  VA  0.028      Essex  VA    0.033  0.121  Gloucester  VA    0.012    Lancaster  VA    0.029    Mathews  VA  0.069      Nottoway  VA  0.068      Rappahannock  VA  0.004      Surry  VA  0.039      Hampshire  WV  0.034      Mercer  WV    0.031    Ohio  WV  0.288  0.315  0.074  Upshur  WV  0.060      Wood  WV  0.189  0.203  0.399  County  State  Synthetic control weight   Apgar Score  Birth weight  Gestational age  Henderson  KY      0.173  Woodford  KY      0.231  Howard  MD  0.033  0.033    Mercer  OH  0.034      Elk  PA  0.103  0.120    Lawrence  PA  0.050      Hamilton  TN    0.073    Lincoln  TN    0.129    Culpepper  VA    0.022    Cumberland  VA  0.028      Essex  VA    0.033  0.121  Gloucester  VA    0.012    Lancaster  VA    0.029    Mathews  VA  0.069      Nottoway  VA  0.068      Rappahannock  VA  0.004      Surry  VA  0.039      Hampshire  WV  0.034      Mercer  WV    0.031    Ohio  WV  0.288  0.315  0.074  Upshur  WV  0.060      Wood  WV  0.189  0.203  0.399  Note: Synthetic controls were selected from all counties in West Virginia and nearby Appalachia states, including Kentucky, Maryland, Ohio, Pennsylvania, Tennessee, and Virginia. We used only those observations that had (a) monthly birth data for all months from 2011 to 2014 and (b) annual economic data from the U.S. Census Bureau over the same time period. Independent cities in Virginia were excluded from the analysis. Our results suggest that Kanawha County, West Virginia, suffered a large and significant decrease in 5-minute Apgar Scores after January 2014 (see figure 4). This drop is pronounced and much larger than the decrease experienced by the synthetic control. Apgar Score trends for Kanawha County, West Virginia match up reasonably well with the synthetic control’s trends, though the data do show monthly variation. We do not find a similar drop in birthweight or gestational age. These results are available in the supplementary online appendix figures 1A and 2A. Although the pre-spill outcomes match up well, there is not a sustained drop in either birthweight or gestational age after the spill. Figure 4. View largeDownload slide Trends in 5-minute Apgar Scores Figure 4. View largeDownload slide Trends in 5-minute Apgar Scores Using the binary outcome of Apgar Score, we can estimate the proportion of poor health outcomes caused by the chemical spill. There is an approximate 6% increase of adverse birth outcomes (as indicated by Apgar Score < 7) one month after the spill. Figure 3A in the supplementary online appendix highlights this result. We also estimate synthetic control models using binary versions of the birthweight and gestational age. We find no post-spill difference between Kanawha County and its synthetic control for either variable (see figures 4A-7A in the supplementary online appendix). To determine the statistical significance of the change in Apgar Score, we use the same method as applied to the MSA-level GDP data. In this case, we estimate synthetic control models for all counties not impacted by the Elk River Chemical Spill to see how changes in 5-minute Apgar Scores after January 2014 relate to those for Kanawha County. As before, the test is based on the premise that unaffected counties will not have casual changes in the outcome-of-interest since their water supplies were unaffected by the spill. Figure 5 highlights the results of these separate models. Each line represents the difference in outcome between each unaffected county and its synthetic control. We have again trimmed the data to only include those placebo counties with a RMSE within 4 times the Kanawha County RMSE, which leaves us with 464 placebo counties. It is clear that the post-January 2014 difference in 5-minute Apgar Scores between Kanawha County and its synthetic control is much larger and more negative than the difference for the vast majority of untreated counties across Appalachia. Additionally, the same placebo tests were conducted for birthweight and gestational age. For birthweight and gestational age, post-spill differences in outcomes between Kanawha County and its synthetic control are not significantly larger than those found in other counties. Figure 5. View largeDownload slide 5-Minute Apgar Score placebo test Note: We trim the placebo data to only include placebo MSAs with a RMSE within 4 times of the Kanawha County, West Virginia RMSE. The black line represents Kanawha County, West Virginia. Figure 5. View largeDownload slide 5-Minute Apgar Score placebo test Note: We trim the placebo data to only include placebo MSAs with a RMSE within 4 times of the Kanawha County, West Virginia RMSE. The black line represents Kanawha County, West Virginia. Table 5 highlights monthly p-values of our estimates from the placebo tests, per equations (1) and (2). Given a statistical significance level of 0.05, we find that the significance of the effect on Apgar Scores extends out to April of 2014, which suggests that the impact of the spill was primarily significant for babies in the last trimester of gestation. The effect is significant in May at the 10% level, so there is slight evidence that the chemical spill’s effect reaches into the middle part of the second trimester. In this table, we also include p-values for the same placebo test analyses conducted for the birthweight and gestational age models. For these models, we find no p-values that are lower than the conventional statistical significance level of 0.05. Table 5. Post-Spill Monthly p-values for Synthetic Control Placebo Tests Month in 2014  P-Values   Apgar Score  Birthweight  Gestational age  January  0.011  0.517  0.262  February  0.019  0.987  0.897  March  0.022  0.662  0.499  April  0.037  0.818  0.153  May  0.080  0.873  0.274  June  0.534  0.630  0.769  July  0.291  0.579  0.412  August  0.116  0.823  0.541  September  0.203  0.299  0.463  October  0.338  0.857  0.843  November  0.366  0.303  0.258  December  0.812  0.406  0.829  Month in 2014  P-Values   Apgar Score  Birthweight  Gestational age  January  0.011  0.517  0.262  February  0.019  0.987  0.897  March  0.022  0.662  0.499  April  0.037  0.818  0.153  May  0.080  0.873  0.274  June  0.534  0.630  0.769  July  0.291  0.579  0.412  August  0.116  0.823  0.541  September  0.203  0.299  0.463  October  0.338  0.857  0.843  November  0.366  0.303  0.258  December  0.812  0.406  0.829  In summary, using the synthetic control methodology, we find suggestive evidence that the Elk River spill had a statistically-significant negative impact on 5-minute Apgar Scores for infants born in the four months after the spill. This indicates that the health impact was felt in the later months of the pregnancy, which supports work that has also found significant health impacts of late-pregnancy exposure to pollution (e.g., Rich et al. 2015). This decrease may have long-term impacts for those affected given evidence in the literature that low Apgar Scores are connected with future health (e.g., Li et al. 2011) and education (Stuart, Otterblad Olausson, and Kallen 2011) outcomes. However, there are some limitations to our analyses. Most prominently, the mechanism of impact is unclear. The decrease in Apgar Score could be related to chemical ingestion by the mother or maternal stress related to the spill itself, among potential other issues. Since the mechanism is unclear, it is difficult to understand whether the (large) magnitude of our estimated effect is realistic. This is especially pertinent given our null results for birthweight and gestational age. These are surprising given the strong effect seen in Apgar Scores. Low birthweight and prematurity can lead to low Apgar Scores (e.g., Hegyi et al. 1998) so it is unclear why we do not see any strong effect of the spill on either variable. Future work should address the potential mechanisms between the spill and the effects we see in our results. Conclusion Comparative case studies provide opportunities for learning about agricultural and environmental policies, risks, and impacts from unexpected events. We employ the synthetic control method to a water contamination tragedy to understand larger causal effects than those reported immediately after the event. Case studies often display outcomes as aggregate statistics with few treated units, and this method can be very beneficial under these circumstances. In applying this method to the Elk River spill in West Virginia, we hope to provide exposure to the methodology as well as a rigorous analysis of this important water contamination event. We find evidence for a sharp decrease in birth outcomes as measured by 5-minute Apgar Scores. Since exposure to toxins for babies in-utero occur at the end of pregnancy, the timing of these effects are consistent with exposure from the chemical spill. It is unclear why the exposure to toxins through the water supply would only contribute to lower Apgar Scores but is undetectable in other birth outcomes. We do not find support for a long-run effect on economic growth. Although there was undoubtedly an immediate economic effect from closed businesses after the spill, the size of this effect is not large enough to distinguish it from the stochastic components of economic growth for Charleston, West Virginia. Other lessons can be learned from this case study about how the public perceived water risks and policy measures to mitigate the probability of future contamination events. First, there are clear loopholes in chemical storage policies that need to address above-ground storage containers so that sites similar to Freedom Industries are adequately monitored. A second lesson is that the perceptions of water quality can persist if the agencies responsible are not coordinated and transparent about risks. Supplementary Material Supplementary material are available at American Journal of Agricultural Economics online. References Abadie A., Diamond A., Hainmueller J.. 2010. Synthetic Control Methods for Comparitive Case Studies: Estimating the Effect of California’s Tobacco Control Program. Journal of the American Statistical Association  105: 493– 505. Google Scholar CrossRef Search ADS   Abadie A., Diamond A., Hainmueller J.. 2015. Comparative Politics and the Synthetic Control Method. American Journal of Political Science  59 2: 495– 510. http://dx.doi.org/10.1111/ajps.12116 Google Scholar CrossRef Search ADS   Abadie A., Gardeazabal J.. 2003. The Economic Costs of Conflict: A Case Study of the Basque Country. The American Economic Review  93 1: 113– 32. http://dx.doi.org/10.1257/000282803321455188 Google Scholar CrossRef Search ADS   Abdalla C.W., Roach B.A., Epp D.J.. 1992. Valuing Environmental Quality Changes Using Averting Expenditures: An Application to Groundwater Contamination. Land Economics  68 2: 163– 169. Google Scholar CrossRef Search ADS   Camilli R., Reddy C.M., Yoerger D.R., Van Mooy B.A.S., Jakuba M.V., Kinsey J.C., McIntyre C.P., Sylva S.P., Maloney J.V.. 2010. Tracking Hydrocarbon Plume Transport and Biodegradation at Deepwater Horizon. Science  330 6001: 201– 4. Google Scholar CrossRef Search ADS PubMed  Cavallo E., Galiani S., Noy I., Pantano J.. 2013. Catastrophic Natural Disasters and Economic Growth. Review of Economics and Statistics  95 5: 1549– 61. http://dx.doi.org/10.1162/REST_a_00413 Google Scholar CrossRef Search ADS   Collins A.R., Steinback S.. 1993. Rural Households Response to Water Contamination in West Virginia. JAWRA Journal of the American Water Resources Association  29 2: 199– 209. Google Scholar CrossRef Search ADS   Coffman M., Noy I.. 2011. Hurricane Iniki: Measuring the Long-Term Economic Impact of a Natural Disaster Using Synthetic Control. Environment and Development Economics  17: 187– 205. Google Scholar CrossRef Search ADS   Federal Emergency Management Agency. 2014. FEMA Continues to Support Response Efforts in West Virginia. Available at: https://www.fema.gov/news-release/2014/01/13/fema-continues-support-response-efforts-west-virginia. Gabriel T. 2014. Thousands without Water After Spill in West Virginia. The New York Times.  Available at: http://www.nytimes.com/2014/01/11/us/west-virginia-chemical-spill.html. Grey D., Sadoff C.W.. 2007. Sink or Swim? Water Security for Growth and Development. Water Policy  9 6: 545– 71. http://dx.doi.org/10.2166/wp.2007.021 Google Scholar CrossRef Search ADS   Hanna-Attisha M., LaChance J., Sadler R.C., Schnepp A.C.. 2016. Elevation Blood Levels in Children Associated with the Flint Drinking Water Crisis: A Spatial Analysis of Risk and Public Health Response. American Journal of Public Health  106 2: 283– 90. Google Scholar CrossRef Search ADS PubMed  Hediger M.L., Overpeck M.D., Ruan W.J., Troendle J.F.. 2002. Birthweight and Gestational Age Effects on Motor and Social Development. Pediatric and Perinatal Epidemiology  16 1: 33– 46. Google Scholar CrossRef Search ADS PubMed  Hegyi T., Carbone T., Anwar M., Ostfeld B., Hiatt M., Koons A., Pinto-Martin J., Paneth N.. 1998. The Apgar score and its components in the preterm infant. Pediatrics  101 1: 77– 81. Google Scholar CrossRef Search ADS PubMed  Heyman D., Fitzsimmons E.G.. 2014. Wait Continues for Safe Tap Water in West Virginia. The New York Times . Available at: http://www.nytimes.com/2014/01/12/us/the-wait-continues-for-safe-tap-water-in-west-virginia.html. Howard B.C. 2014. What’s the Chemical Behind West Virginia’s River Spill? National Geographic News.  Available at: http://news.nationalgeographic.com/news/2014/01/140110-4-methylcyclohexane-methanol-chemical-spill-west-virginia-science/. Hunter P. 1997. Waterborne Disease: Epidemiology and Ecology . John Wiley and Sons. Hunter P.R., MacDonald A.M., Carter R.C.. 2010. Water Supply and Health. PLoS Medicine  7 11: e1000361. https://doi.org/10.1371/journal.pmed.1000361. Johnson C. 2014. Spill Affects Some Local Areas. The Herald Dispatch . Li J., Olsen J., Vestergaard M., Obel C.. 2011. Low Apgar Scores and Risk of Childhood Attention Deficit Hyperactivity Disorder. The Journal of Pediatrics  158 5: 775– 9. http://dx.doi.org/10.1016/j.jpeds.2010.10.041 Google Scholar CrossRef Search ADS PubMed  Markham P., Gianato J., Hoyer J.. 2016. After Action Review: Emergency Response to January 9, 2014. Freedom Industries Chemical Leak. Available at: http://www.governor.wv.gov/SiteCollectionDocuments/After%20Action%20Review%20Final.pdf?utm_source=1-9-14+AAR+review&utm_campaign=1-9-15+AAR+review&utm_medium=email. Marshall University Center for Business and Economic Research. 2014. CBER Calculates Impact from Chemical Spill into Elk River. Marshall University Center for Business and Economic Research. Available at: http://mediad.publicbroadcasting.net/p/wvpn/files/201402/CBER_PR_WaterUsageBan_02.04.14.pdf. B.A. Miller, Reidinger R.A., eds. 1998. Comprehensive River Basin Development: The Tennessee Valley Authority. Vol. 23. National Center for Health Statistics. 2009-2014. Birth Data. National Center for Health Statistics. World Bank Publications, Hyattsville, Maryland. Parker R. 2015. A River Runs Yellow. The Atlantic.  August 21. Available at: https://www.theatlantic.com/national/archive/2015/08/a-river-runs-yellow/401966/. Raby J. 2016. Judge Tentatively OKs $151M Settlement in Chemical Leak Suit. The Associated Press . Available at: http://bigstory.ap.org/article/12cef7cdaa79477eae2a7a7afd29980b/judge-tentatively-oks-151m-settlement-chemical-leak-suit. Rich D.Q., Liu K., Zhang J., Thurston S.W., Stevens T.P., Pan Y., Kane C., et al.   2015. Differences in Birth Weight Associated with the 2008 Beijing Olympics Air Pollution Reduction: Results from a Natural Experiment. Environmental Health Perspectives  123 9: 880– 8. Google Scholar PubMed  Saha K.C. 2003. Review of Arsenicosis in West Bengal, India--A Clinical Perspective. Critical Reviews in Environmental Science and Technology  33 2: 127– 63. http://dx.doi.org/10.1080/10643380390814514 Google Scholar CrossRef Search ADS   Savoia E., Stoto M.A., Gupta R., Wright N., Viswanath K.. 2015. Public Response to the 2014 Chemical Spill in West Virginia: Knowledge, Opinions and Behaviors. BMC Public Health  17: 790. doi:10.1186/s12889-015-2134-2. Google Scholar CrossRef Search ADS   Shaban A., Hamze M., El-Baz F., Ghoneim E.. 2009. Characterization of an Oil Spill Along the Lebanese Coast by Satellite Images. Environmental Forensics  10 1: 51– 9. http://dx.doi.org/10.1080/15275920802659735 Google Scholar CrossRef Search ADS   Siegel M. 2002. The Effectiveness of State-Level Tobacco Control Interventions: A Review of Program Implementation and Behavioral Outcomes. Annual Review of Public Health  23 1: 45– 71. http://dx.doi.org/10.1146/annurev.publhealth.23.092601.095916 Google Scholar CrossRef Search ADS PubMed  Snair J. 2014. Challenges in Responding to the Elk River Chemical Spill: An Interview with Dr. Rahul Gupta. National Association of County and City Health Officials.  Available at: http://nacchopreparedness.org/response-to-the-january-9th-chemical-spill-into-the-elk-river-west-virginia-an-interview-with-dr-gupta/. Schwabach A. 1989. The Sandoz Spill: The Failure of International Law to Protect the Rhine from Pollution. Ecology Law Quarterly  16: 443– 80. Stuart A., Otterblad Olausson P., Kallen K.. 2011. Apgar Scores at 5 Minutes after Birth in Relation to School Performance at 16 Years of Age. Obstetrics and Gynecology  118 2: 201– 8. http://dx.doi.org/10.1097/AOG.0b013e31822200eb Google Scholar CrossRef Search ADS PubMed  U.S. Census Bureau, American Community Survey 2006-2015. 1-Year Estimates. Using American FactFinder. Generated April 20, 2017. Available at: http://www.bea.gov/iTable/iTableHtml.cfm?reqid=70&step=10&isuri=1&7003=1000&7035=-1&7004=naics&7005=1&7006=xx&7036=-1&7001=21000&7002=2&7090=70&7007=-1&7093=levels. U.S. Chemical Safety and Hazard Investigation Board. 2016. Investigation Report: Chemical Spill Contaminates Public Water Supply in Charleston, West Virginia. 2014-01-I-WV. U.S. Chemical Safety and Hazard Investigation Board. Available at: http://www.csb.gov/assets/1/19/Freedom_Industries_Report.pdf. Watkins T., Ellis R.. 2016. Official: No Quick Fix for West Virginia Water Woes. CNN.com.  Available at: http://www.cnn.com/2014/01/11/us/west-virginia-contaminated-water/index.html. Whelton A.J., McMillan L., Connell M., Kelley K.M., Gill J.P., White K.D., Gupta R., Dey R., Novy C.. 2014. Residential Tap Water Contamination Following the Freedom Industries Chemical Spill: Perceptions, Water Quality, and Health Impacts. Environmental Science and Technology  49: 813– 23. Google Scholar CrossRef Search ADS   White K. 2014a. Chemical, Water Companies Quickly Face Lawsuits. Charleston Gazette-Mail . White K.. 2014b. Freedom Industries Files for Bankruptcy. Charleston Gazette-Mail . © The Authors 2017. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png American Journal of Agricultural Economics Oxford University Press

The Economic and Health Effects of the 2014 Chemical Spill in the Elk River, West Virginia

Loading next page...
 
/lp/ou_press/the-economic-and-health-effects-of-the-2014-chemical-spill-in-the-elk-yFdtxtDd29
Publisher
Oxford University Press
Copyright
© The Authors 2017. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
0002-9092
eISSN
1467-8276
D.O.I.
10.1093/ajae/aax089
Publisher site
See Article on Publisher Site

Abstract

Abstract In January 2014, Freedom Industries spilled 4-methylcyclohexylmethanol, a chemical foaming agent used in coal processing, from a storage facility into the Elk River in West Virginia. This chemical spill, one of the most significant in U.S. history, adversely affected the drinking water supply of over 300,000 individuals in the Charleston, West Virginia Metropolitan area. We use synthetic control methods to estimate the casual effects on macro-economic growth and infant health outcomes from this water crisis. We find a significant decrease in 5-minute Apgar Scores, a measure of how babies fare in the birthing process and outside of the womb, after the chemical spill. We do not find significant effects for infant birthweight or gestational age. We find a statistically insignificant decrease of per capita GDP in the Charlestown, West Virginia area compared to the synthetic control of 3% two years after the chemical spill. A safe and reliable supply of water is essential to human health (Hunter, MacDonld, and Carter 2010). Exposure to various organic or inorganic chemicals can lead to detrimental health effects, including nausea, vomiting, skin rashes, cancer, and fetal abnormalities (Hunter 1997). Water pollution exposure can also have indirect impacts to individuals, such as disease from malnutrition, hindered food production, reduced labor productivity, and increased risk of financial stress. Suffice it to say, historic water pollution events have had large effects on human communities (e.g., Schwabach 1989; Saha 2003; Shaban et al. 2009). Water insecurity can lead to increased pressure on poverty and social unrest (Sadoff and Grey 2007) and water-related shocks can negatively influence public health and economic stability. Researchers have found evidence that economic growth is affected by large environmental disasters (Cavallo et al. 2013). Large disturbances to economic activity through the destruction of infrastructure, death, disease, and financial losses can all contribute to lower growth in both the short and long term. Despite greater environmental protections and recent advancements in water treatment, there have been several mid- to large-scale contamination events in the United States. These include the Deepwater Horizon oil spill on the Gulf Coast (Camilli et al. 2010), the Gold King Mine wastewater spill in Colorado (Parker 2015), the Flint, Michigan water crisis in 2014 (Hanna-Attisha et al. 2016), and West Virginia’s Elk River Spill in early 2014. Water safety and security increasingly relies on evaluating the risks and causal effects of contamination events. A difficulty with assessing the local impacts of a water contamination event is that there is typically only one “treated” observation. In typical regression analyses, one compares average outcomes for a series of treated observations to outcomes from observationally-similar but untreated “control” observations. However, regression analyses contain certain statistical characteristics which may over-extrapolate effects1 1 Abadie et al. (2010, 2015) discuss the implications of weights in regression analysis, where weights not constrained to be positive and sum to one can extrapolate effects and yield less-accurate results. (Abadie, Diamond, and Hainmueller 2010; Abadie, Diamond, and Hainmueller 2015). For example, regression analysis will give equal weights to all control units regardless of how similar they are to the treated unit. In the economics literature, the synthetic control method has recently been used to assess treatment effects for single treated observations (Abadie, Diamond, and Hainmueller 2010; Coffman and Noy 2011; Abadie, Diamond, and Hainmueller 2015). This approach compares outcomes for a single treated observation to the outcomes of a “synthetic” control observation, constructed as a weighted average of the universe of potential control observations. The development of synthetic control methods provides a better statistical framework for analyzing singular events with aggregate level data. In this paper, we apply the synthetic control method to assess the economic and public health effects of a large water contamination event: The Elk River Chemical Spill in West Virginia. In January 2014, Freedom Industries released approximately 10,000 gallons of chemicals used for processing coal into the Elk River (U.S. Chemical Safety and Hazard Investigation Board 2016). The river provides drinking water to multiple counties in West Virginia, including the state capital of Charleston. The spill led to a shut-down of restaurants, hotels, and the local mall, and created a drinking water emergency involving responses from local, state, and national agencies. Concerns persisted about contact and consumption of the Elk River water. Toxicity levels of some of the chemicals were not well understood by the scientific community, which created additional concern about safe exposure limits. This paper is a part of a Teaching Case Study section of this journal, and a teaching note, data, and code are available to facilitate using this paper in the classroom.2 2 All these elements can be found online at Dr. Guilfoos and Dr. Hill’s personal websites. Available at: http://www.toddguilfoos.com/research.html and http://www.elainelhill.com/re. The primary county affected by the Elk River spill is Kanawha County, West Virginia. In our approach, we can make causal inferences by comparing post-spill economic and public health outcomes between a “synthetic” county and the real Kanawha County, West Virginia. The synthetic Kanawha County is made up of a weighted linear combination of counties unaffected by the chemical spill and has pre-spill characteristics that are very similar to Kanawha County. We argue that the weighted combination of control counties can act as a better control to measure causal effects of the chemical spill because the synthetic control better matches the treated unit (Abadie, Diamond, and Hainmueller 2010; Abadie, Diamond, and Hainmueller 2015; Coffman and Noy 2011). We use the synthetic control method to address two main questions: Was economic growth of the area impacted by the Elk River spill? Were birth outcomes affected by the Elk River spill? We examine these outcomes because they are readily available, are used in other studies, and represent social costs above those reported as a result of closed businesses and reported illness in the immediate aftermath of the chemical spill. The first question identifies if water-related shocks lead to dampened long-term economic growth. The second question addresses whether infant health is particularly vulnerable to in-utero exposure to water contamination (Galiani, Gertler, and Schargrodsky 2005; Gamper-Rabindran, Khan, and Timmins 2010; Currie and Walker 2011; Currie and Schwandt 2015). By focusing on pregnancy outcomes, we avoid the risk of attenuation due to avoidance behavior. We find weak evidence of a longer-term effect to aggregate economic outcomes to Charleston, West Virginia from the chemical spill, and we cannot rule out the null hypothesis of no effect. There is a statistically-insignificant 3% decline in GDP per capita in Charleston, West Virginia two years after the event. We find suggestive evidence of a negative effect on infant health outcomes for infants born in Kanawha County after January 2014. This effect is substantial for 5-minute Apgar Scores but relatively short-lived, dissipating for birth cohorts born approximately four to five months after the Elk River spill. Placebo test results are supportive of a strong post-January 2014 effect relative to unaffected counties elsewhere in Appalachia. However, we do not find significant effects for other infant birth outcomes, such as birthweight and gestational length. Background on the Elk River Spill On January 9, 2014, a chemical leak was discovered at a chemical distribution facility in Charleston, West Virginia (Trip 2014; Markham, Gianato, and Hoyer 2016). Approximately 10,000 gallons of 4-methylcyclohexanemethanol (MCHM) and Propylene glycol phenyl ether (PPH) were discharged from Freedom Industries into the nearby Elk River, quickly infiltrating the intake and distribution plant of West Virginia American Water Company (WVAWC). The WVAWC utilized a filtration system equipped with activated carbon to mitigate such incidences, but the filters became saturated and ineffective, exposing the drinking water supply to these chemicals (Howard 2014). Figure 1 illustrates the location of the spill in relation to the city of Charleston. Figure 1. View largeDownload slide Location of the spill in Charleston West Virginia Figure 1. View largeDownload slide Location of the spill in Charleston West Virginia Once filters could no longer handle the quantities of the chemicals in the river, WVAWC concluded that the water was unsafe to drink (Howard 2014; Watkins and Ellis 2016). Approximately 300,000 citizens of the Charleston metropolitan area were unable to use tap water for 4–9 days (Markam, Gianato, and Hoyer 2016). The chemicals spilled were used to clean and wash coal before processing and had relatively unknown health effects (Trip 2014). One day after the spill, 122 people had visited hospitals for symptoms of nausea and vomiting, and 4–6 were admitted (Trip 2014; Heyman and Fitzsimmons 2014). To cope with the absence of potable water, West Virginia Governor Early Ray Tomblin and President Barack Obama declared a state of emergency for nine affected counties, enabling the National Guard to bring in tanks of water for residents (Howard 2014). The Federal Emergency Management Agency (FEMA) provided additional aid, bringing over three million liters of water to the affected area and working closely with the state “to ensure there [were] no unmet needs in helping those impacted by the incident” (FEMA 2014). FEMA also set up Incident Management Assistance and Mobile Emergency Response teams on site in Charleston to help coordination. The National Guard began water testing on January 10, 2014, using the 1 part-per-million benchmark suggested by the Centers for Disease Control and Prevention as a safe screening standard; initial levels at the West Virginia American Water intake site were reported as high as 3.35 parts per million (Markam, Gianato, and Hoyer 2016). Authorities lifted the water use ban on January 18, almost two weeks after the spill. Government response to the spill was very involved, and a wide array of partnerships with local, state, federal, and private organizations and agencies were initiated to help mitigate the crisis.3 3 The list includes local health departments, the Kanawha-Charleston Board of Health, Kanawha County Board of Commission, Kanawha County Emergency Management, City of Charleston Emergency Management, the Mayor of Charleston, leaders of other townships, the West Virginia Department of Health and Human Resources, Governor’s office and Poison Control Center, along with U.S. Senators and Representatives, the National Guard and others (Snair 2014). The National Guard provided aid through troops helping staff reopen schools and businesses. A rapid-response team of National Guard troops, school officials, and health department staff was formed to address any complaints at schools across the county for several weeks (Snair 2014). Costs of the Spill Although the stated emergency resulting from the MCHM spill lasted less than two weeks, many effects of the spill were longer-lasting. According to a preliminary study by the Marshall University Center for Business and Economic Research (CBER), an estimated total of $61 million in damages was incurred by local businesses and residents (CBER 2014). Nearly 75,000 workers were unable to work during each day of the ban, representing over 40% of the working population in the area. The costs incurred by residents who bought bottled water, paid for extra childcare, and medical expenses were not included in this impact, nor were future health implications or economic disturbances. Within a week of the spill, twenty-five lawsuits had been filed against Freedom Industries and it subsequently filed for bankruptcy (White 2014). Lawsuits were also filed against the West Virginia American Water Company and chemical manufacturer, Eastman Chemical. One $151 million settlement—$126 million to be paid by W. Virginia Water and $25 million by Eastman—was reached in 2016 and will ultimately be distributed to affected businesses and residents (Raby 2016). Beyond the immediate costs from the spill are other important economic impacts, such as changes to economic activity where water is used in food preparation or is integrated into products, or longer-term health effects that may decrease productivity through increased absences or decreased output at work. Perceptions of unsafe water quality to visitors may also suppress growth in the economy if business is taken elsewhere. These effects could be longer-lasting than the period of the stated emergency as perceptions and illness persist. The CDC officials stated in a press conference on February 5, 2014, that the water was “appropriate” to drink, but five days later, multiple expert witnesses were unable to conclusively report on the water’s continued danger. “Because of the level of mistrust, the public is slow to return to using the water,” said Dr. Rahul Gupta, Executive Director of the Kanawha-Charleston Health Department. “Survey data from the time of the spill until March 1 shows that less than five percent of the population are drinking or cooking with our local water, and approximately 20 percent are not using the water for any purpose. There has been a secondary wave of health impacts after the flushing which has further exacerbated the frustrations of a very anxious and suspicious community,” (Snair 2014). Our model estimates would include some economic costs stated above, such as decreases in productivity and disturbances to businesses. But other costs would be in addition to our estimates, such as legal costs and averting expenditures borne by victims of the spill. The health costs we measure through birth outcomes would be in addition to any of the immediate costs through visits to the hospital, usually associated with stomach illnesses and nausea. Chemical Storage Policies and Regulations A key component of the costs of avoidance is evaluating risks and the costs of reducing risks of future contamination events. The U.S. Chemical and Hazard Investigation Board (CSB) conducted a thorough investigation of the Elk River spill. The CSB found a lack of effort by Freedom Industries to properly inspect and maintain equipment, though they were compliant with existing state and federal regulations. Regulations for chemical manufacturers and distributors are often difficult to navigate. The Occupational Safety & Health Administration (OSHA) classifies both crude-MCHM and stripped-PPH as “hazardous chemicals” (CSB 2016). Every manufacturer or distributor storing more than 10,000 pounds of MCHM or PPH stripped is required by the Emergency Planning and Community Right-To-Know Act to submit a form to its Local Emergency Planning Committee regarding the stored amount (CSB 2016). At the time of the spill, Freedom Industries was subject “…to the West Virginia National Pollution Discharge Elimination System (NPDES),” a general permit for storm water discharge near industrial activity. Other regulators such as the Environmental Protection Agency (EPA) define “hazardous chemicals” and “hazardous substances” differently, and thus the chemicals are subject to different regulations. While OSHA classifies “hazardous chemicals” as any chemical that poses physical or health hazards, the EPA’s Clean Water Act defines “hazardous substances” as a substance where “the discharge of which may be harmful to the public health or the environment of the United States” (CSB 2016). While the EPA is required to establish regulations of these substances under the Clean Water Act, neither MCHM nor PPH stripped are listed. Storage container regulations are equally difficult to maneuver and enforce. At the time of the MCHM spill, aboveground storage tanks (ASTs) were inadequately regulated in West Virginia. No uniform regulation program exists for all ASTs, and states are charged with instituting regulations not addressed by the federal government. The 1984 state legislature established a “comprehensive statutory framework… regulating underground storage tanks, but it did not address ASTs,” (CSB 2016). The CSB investigation revealed that the three tanks storing MCHM and PPH were initially installed in 1938 and held glycerin or calcium chloride prior to 2009 (CSB 2016). American Petroleum Institute (API)-certified inspectors found the tank shells and roof were constructed with a now-obsolete construction; the bottoms appeared to be replacements of the originals (CSB 2016).4 4 According to the CSB, tank shells and roofs were constructed using a technique called lap-riveting; welding began to replace the process of riveting during the 1930s. The bottoms were lap-welded, but looked to be replacements of the originals. Two floor holes, 0.75 and 0.4 inches in diameter, were the source of the leak. Inspectors also found deep pits in the tank shell and floor, and determined the holes started as pits formed from corrosion. Examining Freedom Industries’ inspection protocols, the CSB determined the manufacturer did not have a program in place that would ensure the ASTs were maintained and inspected regularly. Freedom Industries also did not keep detailed history, maintenance, or inspection records for the failed tank as they were not forced to comply with regular inspections (CSB 2016). The CSB also reported on the lack of a leak detection system (LDS): “Freedom [Industries] did not have any level indication device, gauge system or measurement to capture the actual amount of MCHM leak, which contributed to the changing estimates of the spill amount. There was no West Virginia state or federal requirement that would have made the installation of an LDS mandatory for ASTs.” Compliance with additional federal regulations could have indirectly prevented the leak from entering the Elk River.5 5 The Spill, Prevention, Control and Countermeasure (SPCC) rule regulates various industrial chemicals and liquids; the CSB identified that Freedom Industries also stored an oil on-site, which was under regulation of the SPCC. Freedom Industries had not been complying with these requirements, specifically for secondary containment (CSB 2016). The proximity of the oil AST to the MCHM AST led CSB investigators to conclude that secondary containment may have prevented the spill had Freedom Industries been complying. Methodology Many of the direct costs and risks enumerated above are known, but to understand the larger costs to the economy and risks to public health we use the synthetic control method to analyze outcomes from this water crisis. The synthetic control method has had limited yet growing use in the economics literature (Abadie and Gardeazabal 2003; Abadie, Diamond, and Hainmueller 2010; Cavallo et al. 2013; Abadie, Diamond, and Hainmueller 2015). This method is used to compare a specific region or entity exposed to an intervention. In this case study, the intervention is the Elk River Chemical Spill. The main entity exposed is the Charleston Metropolitan Area or Kanawha County, West Virginia, depending on the outcome variable. The synthetic control is a weighted average of the control units; this allows for explicit inference of the “relative contribution” of the control units and explains the similarities (or differences) between the unit affected and the synthetic control. The weights of the controls can also be determined to be positive and sum to one, which can help guard against extrapolation errors. The synthetic control model relies on a series of constructed vectors and matrices to perform the analysis. Following Abadie, Diamond, and Hainmueller (2010), the outcome variable investigated is observed for T periods for the region exposed to the intervention, Y1t, where t=1,…, T and the synthetic control, Yjt, where j=2, …, J+1 and t=1,…, T. T1=T- T0 is the number of post-intervention periods, and Y1 is a (T1×1) vector of post-intervention outcomes for only the exposed region. This leaves Y0, a (T1×J) matrix, comprising the post-intervention outcomes for the control regions. Listing the treated region as the first of J regions is only done for convenience. We let a (T0×1) vector K=(k1,…,kT0)' be a weighting vector, and then define a linear combination of pre-exposure outcomes as Y-iK=∑s=1T0ksYis. There can be M values of K to form any linear combination, and therefore M linear combinations defined by the vectors K1,…KM. This allows for weights to be assigned to control units in a way that best fits the data. In our analysis we will rely on a single vector of weights and a linear combination of control unit outcomes. Next, we represent the pre-intervention characteristics of the treated region with a k×1 vector X1=(Z1',Y-1K1,…,Y-1K1)'. The vector Z is a set of explanatory variables that are used to predict outcomes, and which are not affected by the treatment. The Z vector is generally chosen through an iterative process, not unlike matching protocols, but is initially chosen based on the expected relationship to the outcome variable. The pre-intervention characteristics for the unaffected regions are represented by a similar k×J matrix containing the same variables for the untreated areas, X0=(Zj',Y-jK1,…,Y-jK1)'. We also let W be a (J×1) vector of positive weights, where W=(w2,…,wJ+1)' and w2+…+wJ+1=1; each value of W is a weighted average of all the available control regions. To measure the difference between the treated and untreated units, we take the distance X1-X0W=(X1-X0W)'V(X1-X0W). Here, V is some (k×k) symmetric and positive semidefinite matrix. Our synthetic control weight W* is chosen to minimize this distance, and the optimal choice for V minimizes the mean square error of the synthetic control estimator. The basic idea of this method is to construct a nearly identical county to our affected county in West Virginia in an effort to understand how outcomes like economic growth and infant health have changed because of the chemical spill. We compute the p-value of the effects found on an outcome variable through a permutation test using equations 1 and 2, following Cavallo et al. (2013). Equation (1) defines the estimate of the difference between outcomes from the treated unit, Y1t, and the synthetic control outcomes, Yjt, using the set of optimal weights, wj*, found by minimizing the distance between pretreatment observations and controls. Further, α^1t is the estimator of this difference for unit 1, which is the treated unit at time t,   α^1t=|Y1t-∑j=2J+1wj*Yjt| (1)  p-valuet=Pr⁡α^1tPL<α^1t=∑j=2J+1Iα^1tPL(j)<α^1tJ. (2) In equation (2), the term α^1lPL refers to the estimate for each placebo of a donor unit used to construct the synthetic control for unit 1. Each donor unit is a county (or other unit of observation) unaffected by the event of interest. There are J donor units which are used to construct a distribution of placebo estimates. The placebo estimate of α^1tPL(j) for each j donor unit, which are estimated with same choice of controls as our unit of interest, is used to determine significance of the estimate of α^1t. Moreover, I· is an indicator function which returns 1 if the interior argument is true and 0 if false. This procedure provides the rank of the estimate of α^1t compared to the distribution of placebos estimates α^1tPL. Economic Growth Data We apply the synthetic control method to the Charleston, West Virginia Metropolitan Statistical Area (MSA) for economic growth outcomes as defined by annual GDP per capita (Bureau of Economic Analysis 2015). We use a combination of demographics and economic characteristics as additional controls to match our treated MSA: the percentage who have attained a bachelor’s degree or higher, population count, and jobs by major industry (U.S. Census Bureau, American Community Survey 2015).6 6 We use 2013 1-year estimates for the demographic and industry data from the U.S. Census ACS. This choice was made because the definition of MSAs can change year-to-year. Importantly, in 2012 Charleston, WV, was re-defined. This complicates averaging across years as is typically done with control variables, so we choose 2013 as the year to match demographic and industry data to have a consistent definition for all MSAs. The GDP per capita data from the BEA has been adjusted for the change in definition of MSAs across all years. We also found that by using 5-Year estimates from ACS survey data, that our synthetic control is very similar to the results presented here. After eliminating MSAs that do not have a balanced panel over 2009 to 2015, we retain a total of 328 control units to construct the synthetic Charleston, MSA. The selection of variables is based on a backwards induction search that finds a good balance between Charleston, West Virginia, and the synthetic Charlestown, and has a good pre-treatment fit in predicting GDP per capita. The choice of variables is not made ad hoc; rather, we select variables that would be expected to have an effect on GDP per capita and improve pretreatment fit as measured by the root mean squared error (RMSE). The RMSE is a measure of fit that tracks the distance between prediction of synthetic control and the treated observation of the outcome variable. We choose 2009 as the starting point for the analysis because of the 2008 financial crisis, which may have caused structural shifts to regional economies. By using only post-2008 data, we do not restrict the synthetic control group to match the pre- and post-periods of the recession as this may be asking too much of the synthetic control method. By including pre-2008 data, we are asking the method to match pre-financial crisis trends, the decline, and recovery of economic growth. The treatment period is 2014 as the chemical spill occurred in January of 2014. The occurrence of the spill in January is advantageous since all variables are captured annually, making measurement of the post-treatment effects less likely to be attentuated. Infant Health Data We use a variety of monthly infant health outcomes from the National Vital Statistics System’s Birth Data files from 2011 to 2014. We received restricted-use data, which describes mother demographic information and health outcomes for the universe of births in the United States. The location of the birth is provided at the mother’s residential county-level. Mother’s home address-level data are available from the state, but we concluded that this level of disaggregation was not necessary given our identification strategy. We also restrict the data to singleton births for this analysis as is standard in the health literature. We estimate models for three main birth outcomes of interest. First, we use the Apgar score, a value ranging from 1–10 that indicates how well the baby performed through the birthing process and outside of the womb. Values are based on a series of respiratory, cardiovascular, muscular, reflexive, and skin color tests conducted within 5 minutes of birth. Values of 7 or above indicate that the baby is in good health; scores lower than 7 indicate that the baby needs medical attention. Second, we use the baby’s gestational age at birth. Typical full-term gestational age at birth ranges from 37 to 42 weeks. A gestational age at birth below 37 weeks is classified as premature. Third, we use the baby’s birthweight in grams. Babies are considered as having low birthweight if they weigh less than 2,500 grams. Both gestational age and birthweight are related to motor and social development (e.g., Hediger et al. 2002). We estimate both continuous and binary models of the variables described above. In the binary models, we define the dependent variable to equal 1 if the baby is born with an Apgar Score below 7, is born at less than 37 weeks in gestational age, and is born with a birthweight of less than 2,500 grams. These binary variables allow us to infer how the proportion of babies born with poor health outcomes changes in response to the Elk River Chemical Spell. We aggregate our outcomes and control variables to the county-month level. In order to create a balanced panel, we only use those counties that have monthly data from 2011 to 2014. We also only use those counties that are located within nearby states in the Appalachian region of the United States. These states include Kentucky, Maryland, Ohio, Pennsylvania, Tennessee, Virginia, and West Virginia. We use this additional restriction to lower the possibility of assigning positive weights to counties in regions of the states with unobserved characteristics that may lead to differences in birth outcomes. Additionally, we exclude nearby, more rural counties that were also impacted by the Elk River Chemical spill. These counties include Boone, Clay, Jackson, Lincoln, Logan, Putnam, Roane, and Cabell counties in West Virginia. We argue that the health effect in these counties may be mitigated by both distance and time. As additional controls in our infant health analyses, we use annual, county-level economic data from the U.S. Bureau of Labor Statistics and Census Bureau (Local Area Unemployment Statistics; American Community Survey 2015). We include median household income, poverty rate, and unemployment rate as controls for regional economic conditions. We include median household income, poverty rate, and unemployment rate as controls for regional economic conditions. We also use county-year means of the following mother characteristics from the NCHS: percentage of mothers who are black; percentage of mothers who are unmarried; and percentage of mothers that are at least 35 years old (National Center for Health Statistics 2009–2014). These variables are selected because they are general predictors of birth outcomes and health and should therefore be a part of the balance in constructing a synthetic Kanawha County, West Virginia.7 7 Models of infant health outcomes use a variety of mother characteristics, including percentages of mothers in different groups based on race, age, educational attainment, marital status, level of cigarette consumption, and the receipt of Women, Infants, and Children (WIC), as well as percentage of babies that were female. Data on educational attainment and cigarette smoking were missing for West Virginia during our study period. Our results are robust to the inclusion of other variables in the synthetic control formation, including percentage of mothers who are Hispanic and percentage of mothers who are < 20 years old, as well as percentage of babies who are female. Economic Growth Results The makeup of the synthetic Charleston, West Virginia, MSA is listed in table 2 as the weights given to each MSA based on pretreatment matches to the variables in table 1. Our results, shown in figure 2, suggest a divergence between the synthetic Charleston, West Virginia and the real Charleston, West Virginia MSA, suggesting that the chemical spill may have a longer-term effect on economic growth. The synthetic control shows a high degree of tracking Charleston MSA before the chemical spill and there is a high degree of balance between predictors. We find a decrease of 1% of GDP per capita in 2014 and a decrease of 3% in 2015. Table 1. Means of Predictors for GDP per capita Variables  Charleston, WV  Synthetic Charleston  Rest of United States  % of Bachelor degree or higher  22.8  22.8  26  Count of jobs in public sector  8,715  7,555  78,893  Count of jobs in health care  23,245  22,993  16,863  Total population estimate  224,727  227,342  713,147  Count of white population  202,050  195,285  521,080  GDP per capita 2009  56,765  56,742  40,485  GDP per capita 2011  59,847  59,820  41,011  GDP per capita 2013  56,163  56,135  41,432  Variables  Charleston, WV  Synthetic Charleston  Rest of United States  % of Bachelor degree or higher  22.8  22.8  26  Count of jobs in public sector  8,715  7,555  78,893  Count of jobs in health care  23,245  22,993  16,863  Total population estimate  224,727  227,342  713,147  Count of white population  202,050  195,285  521,080  GDP per capita 2009  56,765  56,742  40,485  GDP per capita 2011  59,847  59,820  41,011  GDP per capita 2013  56,163  56,135  41,432  Table 2. Weights to MSAs for Synthetic Charleston, West Virginia MSA  State  Weight  Casper  WY  0.538  Elkhart-Goshen  IN  0.165  Ogden-Clearfield  UT  0.111  Duluth  MN-WI  0.086  Hanford-Corcoran  CA  0.052  Peoria  IL  0.023  McAllen-Edison-Mission  TX  0.021  Washington-Arlington-Alexandria  DC-VA-MD-WV  0.004  MSA  State  Weight  Casper  WY  0.538  Elkhart-Goshen  IN  0.165  Ogden-Clearfield  UT  0.111  Duluth  MN-WI  0.086  Hanford-Corcoran  CA  0.052  Peoria  IL  0.023  McAllen-Edison-Mission  TX  0.021  Washington-Arlington-Alexandria  DC-VA-MD-WV  0.004  Figure 2. View largeDownload slide Trends in GDP per capita Figure 2. View largeDownload slide Trends in GDP per capita To infer statistical significance of these impacts, we apply a permutation test, sometimes called a placebo test, on unaffected MSAs to see if the change in post-treatment GDP per capita is large compared to the Charleston synthetic analysis. The placebo test posits that unaffected MSAs should not have significant differences in post-treatment outcomes since they did not experience a water crisis, and if there are many placebos with larger differences from their synthetic controls, but good balance pre-treatment, then our measured effect is not statistically significant but rather within the statistical variation of the data. Figure 3 shows that we cannot reject the null hypothesis that the water crisis had no effect on longer-term economic growth since there are many placebos with larger post-treatment variations than the Charleston analysis suggests. We have trimmed the placebo data to only include placebo MSAs with a RMSE within 4 times of the Charleston, West Virginia RMSE, with the goal of excluding placebos that did not have good pretreatment matches with their synthetic controls. Figure 3. View largeDownload slide GDP placebo test Note: This figure uses the remaining MSAs and calculates the distance from their synthetic control. We trim the placebo MSAs to only show placebos with a RMSE within 4 times of the Charleston, West Virginia RMSE pretreatment. The black-dashed line represents the Charleston, West Virginia MSA. Figure 3. View largeDownload slide GDP placebo test Note: This figure uses the remaining MSAs and calculates the distance from their synthetic control. We trim the placebo MSAs to only show placebos with a RMSE within 4 times of the Charleston, West Virginia RMSE pretreatment. The black-dashed line represents the Charleston, West Virginia MSA. We follow the procedure in equations (1) and (2) to find the p-value of our estimates from the placebo tests. We find the estimated p-value is 0.372 one-year post-treatment, and 0.199 two years post-treatment, which indicates that the estimated effect of the water crisis on economic growth in Charleston, West Virginia is not statistically significant. The strength of the synthetic control method is its ability to define a better statistical counterfactual for a singular event such as a water crisis, and allows the effects of confounding unobserved characteristics to vary with time, unlike a fixed effect regression model. The weaknesses of this approach for our given question is the short timeframe of pretreatment periods due to the concerns generated from the housing market collapse and 2008 financial crisis. Another challenge of our data is the size of shifts in GDP per capita in the Charleston, West Virginia pretreatment period, which have the potential of masking the significance of the estimate through a type II error. We rely on few control MSAs to construct the synthetic control, though there are no obvious disadvantages to relying on few control units if they are good matches and predictors for the treated unit. Infant Health Results We apply the same methodology to county-month-level infant health outcomes from the National Center for Health Statistics. We estimate synthetic control models for 5-minute Apgar Score, birthweight (in grams), and gestational age (in weeks). We also estimate models using binary variables equal to 1 if those variables are indicative of poor infant health: 5-minute Apgar Score < 7; birthweight < 2,500 grams; and gestational age < 37 weeks. Our synthetic control model selects control counties using a function of 2011 to 2013 annual values of median household income, poverty rate, unemployment rate, the health outcome of interest, and mother characteristics, including the percentages of mothers who are black, over the age of 35, and unmarried. Table 3 shows the predictor variables used when constructing the synthetic controls. We choose the pretreatment average annual median household income, poverty percentage, unemployment rate, average county-year mother characteristics, and the respective health outcome as the predictor variables in the health analysis. This table shows a strong balance of the variables for all three synthetic control analyses. It also highlights the value of synthetic control analyses. Kanawha County, West Virginia is different from other counties in Appalachia; relative to the rest of Appalachia, it has a lower unemployment rate, a slightly-higher proportion of the mother population that is black, and a significantly higher proportion of the mother population that is unmarried.8 8 Similar results are found with slightly less restrictions on averaging covariates across years, but the presented combination of controls provided the best pretreatment fits to the data. Table 3. Predictor Balance between Appalachia Counties, Kanawha County, and Synthetic Kanawha Counties Variable  Appalachia  Kanawha County  Synthetic Kanawha County   Apgar Score  Birth weight  Gestational age  Mean Apgar Score            2011  8.9  8.9  8.9      2012  8.9  9.0  9.0      2013  8.9  9.0  9.0      Birth weight            2011  3.3  3.2    3.3    2012  3.3  3.3    3.3    2013  3.3  3.3    3.2    Gestational Age            2011  38.6  38.3      38.2  2012  38.6  38.3      38.3  2013  38.6  38.3      38.3  Med. Household Income            2011  43.9  40.4  42.1  41.6  42.9  2012  44.8  45.8  45.0  44.7  45.3  2013  45.8  45.9  45.1  45.2  45.9  Poverty %            2011  17.4  17.2  15.7  15.8  15.6  2012  17.4  14.4  15.7  15.7  15.7  2013  17.2  15.3  15.2  15.5  16.1  Unemployment Rate            2011  9.3  7.1  7.6  7.5  7.9  2012  8.3  6.6  6.7  6.7  6.8  2013  8.2  5.9  6.1  6.2  6.2  % Mothers: Black            2011  7.9  7.9  8.2  8.5  8.4  2012  7.9  9.1  9.0  8.8  8.9  2013  8.0  9.2  8.7  8.9  9.1  % Mothers: Unmarried            2011  41.0  46.7  45.9  46.6  44.8  2012  41.4  48.6  47.2  47.3  48.0  2013  42.0  49.2  49.3  48.6  48.1  % Mothers: Age > = 35            2011  10.1  9.6  9.7  10.1  10.2  2012  10.0  8.8  9.5  9.5  9.1  2013  10.0  9.6  9.3  9.3  8.9  Variable  Appalachia  Kanawha County  Synthetic Kanawha County   Apgar Score  Birth weight  Gestational age  Mean Apgar Score            2011  8.9  8.9  8.9      2012  8.9  9.0  9.0      2013  8.9  9.0  9.0      Birth weight            2011  3.3  3.2    3.3    2012  3.3  3.3    3.3    2013  3.3  3.3    3.2    Gestational Age            2011  38.6  38.3      38.2  2012  38.6  38.3      38.3  2013  38.6  38.3      38.3  Med. Household Income            2011  43.9  40.4  42.1  41.6  42.9  2012  44.8  45.8  45.0  44.7  45.3  2013  45.8  45.9  45.1  45.2  45.9  Poverty %            2011  17.4  17.2  15.7  15.8  15.6  2012  17.4  14.4  15.7  15.7  15.7  2013  17.2  15.3  15.2  15.5  16.1  Unemployment Rate            2011  9.3  7.1  7.6  7.5  7.9  2012  8.3  6.6  6.7  6.7  6.8  2013  8.2  5.9  6.1  6.2  6.2  % Mothers: Black            2011  7.9  7.9  8.2  8.5  8.4  2012  7.9  9.1  9.0  8.8  8.9  2013  8.0  9.2  8.7  8.9  9.1  % Mothers: Unmarried            2011  41.0  46.7  45.9  46.6  44.8  2012  41.4  48.6  47.2  47.3  48.0  2013  42.0  49.2  49.3  48.6  48.1  % Mothers: Age > = 35            2011  10.1  9.6  9.7  10.1  10.2  2012  10.0  8.8  9.5  9.5  9.1  2013  10.0  9.6  9.3  9.3  8.9  Note: Annual economic data is taken from the U.S. Census Bureau. Mother birth and socio-economic characteristics data are from the National Center of Health Statistics' Vital Statistics Natality Birth Database. Birthweight is measured in thousands of grams. Median Household Income is measured in thousands of dollars. Table 4 lists the weights used in determining Kanawha County’s synthetic control for each continuous infant health outcome. Most counties selected as components within the Kanawha County synthetic control are semi-rural counties containing towns with a population size ranging from 10,000 to 100,000. The two counties that form large proportions of Kanawha County’s synthetic control across all three health outcomes—Ohio and Wood Counties in West Virginia—contain cities that are similar in size to Kanawha County’s Charleston (Wheeling and Parkersburg, respectively). Table 4. Weights for Synthetic Kanawha County, West Virginia County  State  Synthetic control weight   Apgar Score  Birth weight  Gestational age  Henderson  KY      0.173  Woodford  KY      0.231  Howard  MD  0.033  0.033    Mercer  OH  0.034      Elk  PA  0.103  0.120    Lawrence  PA  0.050      Hamilton  TN    0.073    Lincoln  TN    0.129    Culpepper  VA    0.022    Cumberland  VA  0.028      Essex  VA    0.033  0.121  Gloucester  VA    0.012    Lancaster  VA    0.029    Mathews  VA  0.069      Nottoway  VA  0.068      Rappahannock  VA  0.004      Surry  VA  0.039      Hampshire  WV  0.034      Mercer  WV    0.031    Ohio  WV  0.288  0.315  0.074  Upshur  WV  0.060      Wood  WV  0.189  0.203  0.399  County  State  Synthetic control weight   Apgar Score  Birth weight  Gestational age  Henderson  KY      0.173  Woodford  KY      0.231  Howard  MD  0.033  0.033    Mercer  OH  0.034      Elk  PA  0.103  0.120    Lawrence  PA  0.050      Hamilton  TN    0.073    Lincoln  TN    0.129    Culpepper  VA    0.022    Cumberland  VA  0.028      Essex  VA    0.033  0.121  Gloucester  VA    0.012    Lancaster  VA    0.029    Mathews  VA  0.069      Nottoway  VA  0.068      Rappahannock  VA  0.004      Surry  VA  0.039      Hampshire  WV  0.034      Mercer  WV    0.031    Ohio  WV  0.288  0.315  0.074  Upshur  WV  0.060      Wood  WV  0.189  0.203  0.399  Note: Synthetic controls were selected from all counties in West Virginia and nearby Appalachia states, including Kentucky, Maryland, Ohio, Pennsylvania, Tennessee, and Virginia. We used only those observations that had (a) monthly birth data for all months from 2011 to 2014 and (b) annual economic data from the U.S. Census Bureau over the same time period. Independent cities in Virginia were excluded from the analysis. Our results suggest that Kanawha County, West Virginia, suffered a large and significant decrease in 5-minute Apgar Scores after January 2014 (see figure 4). This drop is pronounced and much larger than the decrease experienced by the synthetic control. Apgar Score trends for Kanawha County, West Virginia match up reasonably well with the synthetic control’s trends, though the data do show monthly variation. We do not find a similar drop in birthweight or gestational age. These results are available in the supplementary online appendix figures 1A and 2A. Although the pre-spill outcomes match up well, there is not a sustained drop in either birthweight or gestational age after the spill. Figure 4. View largeDownload slide Trends in 5-minute Apgar Scores Figure 4. View largeDownload slide Trends in 5-minute Apgar Scores Using the binary outcome of Apgar Score, we can estimate the proportion of poor health outcomes caused by the chemical spill. There is an approximate 6% increase of adverse birth outcomes (as indicated by Apgar Score < 7) one month after the spill. Figure 3A in the supplementary online appendix highlights this result. We also estimate synthetic control models using binary versions of the birthweight and gestational age. We find no post-spill difference between Kanawha County and its synthetic control for either variable (see figures 4A-7A in the supplementary online appendix). To determine the statistical significance of the change in Apgar Score, we use the same method as applied to the MSA-level GDP data. In this case, we estimate synthetic control models for all counties not impacted by the Elk River Chemical Spill to see how changes in 5-minute Apgar Scores after January 2014 relate to those for Kanawha County. As before, the test is based on the premise that unaffected counties will not have casual changes in the outcome-of-interest since their water supplies were unaffected by the spill. Figure 5 highlights the results of these separate models. Each line represents the difference in outcome between each unaffected county and its synthetic control. We have again trimmed the data to only include those placebo counties with a RMSE within 4 times the Kanawha County RMSE, which leaves us with 464 placebo counties. It is clear that the post-January 2014 difference in 5-minute Apgar Scores between Kanawha County and its synthetic control is much larger and more negative than the difference for the vast majority of untreated counties across Appalachia. Additionally, the same placebo tests were conducted for birthweight and gestational age. For birthweight and gestational age, post-spill differences in outcomes between Kanawha County and its synthetic control are not significantly larger than those found in other counties. Figure 5. View largeDownload slide 5-Minute Apgar Score placebo test Note: We trim the placebo data to only include placebo MSAs with a RMSE within 4 times of the Kanawha County, West Virginia RMSE. The black line represents Kanawha County, West Virginia. Figure 5. View largeDownload slide 5-Minute Apgar Score placebo test Note: We trim the placebo data to only include placebo MSAs with a RMSE within 4 times of the Kanawha County, West Virginia RMSE. The black line represents Kanawha County, West Virginia. Table 5 highlights monthly p-values of our estimates from the placebo tests, per equations (1) and (2). Given a statistical significance level of 0.05, we find that the significance of the effect on Apgar Scores extends out to April of 2014, which suggests that the impact of the spill was primarily significant for babies in the last trimester of gestation. The effect is significant in May at the 10% level, so there is slight evidence that the chemical spill’s effect reaches into the middle part of the second trimester. In this table, we also include p-values for the same placebo test analyses conducted for the birthweight and gestational age models. For these models, we find no p-values that are lower than the conventional statistical significance level of 0.05. Table 5. Post-Spill Monthly p-values for Synthetic Control Placebo Tests Month in 2014  P-Values   Apgar Score  Birthweight  Gestational age  January  0.011  0.517  0.262  February  0.019  0.987  0.897  March  0.022  0.662  0.499  April  0.037  0.818  0.153  May  0.080  0.873  0.274  June  0.534  0.630  0.769  July  0.291  0.579  0.412  August  0.116  0.823  0.541  September  0.203  0.299  0.463  October  0.338  0.857  0.843  November  0.366  0.303  0.258  December  0.812  0.406  0.829  Month in 2014  P-Values   Apgar Score  Birthweight  Gestational age  January  0.011  0.517  0.262  February  0.019  0.987  0.897  March  0.022  0.662  0.499  April  0.037  0.818  0.153  May  0.080  0.873  0.274  June  0.534  0.630  0.769  July  0.291  0.579  0.412  August  0.116  0.823  0.541  September  0.203  0.299  0.463  October  0.338  0.857  0.843  November  0.366  0.303  0.258  December  0.812  0.406  0.829  In summary, using the synthetic control methodology, we find suggestive evidence that the Elk River spill had a statistically-significant negative impact on 5-minute Apgar Scores for infants born in the four months after the spill. This indicates that the health impact was felt in the later months of the pregnancy, which supports work that has also found significant health impacts of late-pregnancy exposure to pollution (e.g., Rich et al. 2015). This decrease may have long-term impacts for those affected given evidence in the literature that low Apgar Scores are connected with future health (e.g., Li et al. 2011) and education (Stuart, Otterblad Olausson, and Kallen 2011) outcomes. However, there are some limitations to our analyses. Most prominently, the mechanism of impact is unclear. The decrease in Apgar Score could be related to chemical ingestion by the mother or maternal stress related to the spill itself, among potential other issues. Since the mechanism is unclear, it is difficult to understand whether the (large) magnitude of our estimated effect is realistic. This is especially pertinent given our null results for birthweight and gestational age. These are surprising given the strong effect seen in Apgar Scores. Low birthweight and prematurity can lead to low Apgar Scores (e.g., Hegyi et al. 1998) so it is unclear why we do not see any strong effect of the spill on either variable. Future work should address the potential mechanisms between the spill and the effects we see in our results. Conclusion Comparative case studies provide opportunities for learning about agricultural and environmental policies, risks, and impacts from unexpected events. We employ the synthetic control method to a water contamination tragedy to understand larger causal effects than those reported immediately after the event. Case studies often display outcomes as aggregate statistics with few treated units, and this method can be very beneficial under these circumstances. In applying this method to the Elk River spill in West Virginia, we hope to provide exposure to the methodology as well as a rigorous analysis of this important water contamination event. We find evidence for a sharp decrease in birth outcomes as measured by 5-minute Apgar Scores. Since exposure to toxins for babies in-utero occur at the end of pregnancy, the timing of these effects are consistent with exposure from the chemical spill. It is unclear why the exposure to toxins through the water supply would only contribute to lower Apgar Scores but is undetectable in other birth outcomes. We do not find support for a long-run effect on economic growth. Although there was undoubtedly an immediate economic effect from closed businesses after the spill, the size of this effect is not large enough to distinguish it from the stochastic components of economic growth for Charleston, West Virginia. Other lessons can be learned from this case study about how the public perceived water risks and policy measures to mitigate the probability of future contamination events. First, there are clear loopholes in chemical storage policies that need to address above-ground storage containers so that sites similar to Freedom Industries are adequately monitored. A second lesson is that the perceptions of water quality can persist if the agencies responsible are not coordinated and transparent about risks. Supplementary Material Supplementary material are available at American Journal of Agricultural Economics online. References Abadie A., Diamond A., Hainmueller J.. 2010. Synthetic Control Methods for Comparitive Case Studies: Estimating the Effect of California’s Tobacco Control Program. Journal of the American Statistical Association  105: 493– 505. Google Scholar CrossRef Search ADS   Abadie A., Diamond A., Hainmueller J.. 2015. Comparative Politics and the Synthetic Control Method. American Journal of Political Science  59 2: 495– 510. http://dx.doi.org/10.1111/ajps.12116 Google Scholar CrossRef Search ADS   Abadie A., Gardeazabal J.. 2003. The Economic Costs of Conflict: A Case Study of the Basque Country. The American Economic Review  93 1: 113– 32. http://dx.doi.org/10.1257/000282803321455188 Google Scholar CrossRef Search ADS   Abdalla C.W., Roach B.A., Epp D.J.. 1992. Valuing Environmental Quality Changes Using Averting Expenditures: An Application to Groundwater Contamination. Land Economics  68 2: 163– 169. Google Scholar CrossRef Search ADS   Camilli R., Reddy C.M., Yoerger D.R., Van Mooy B.A.S., Jakuba M.V., Kinsey J.C., McIntyre C.P., Sylva S.P., Maloney J.V.. 2010. Tracking Hydrocarbon Plume Transport and Biodegradation at Deepwater Horizon. Science  330 6001: 201– 4. Google Scholar CrossRef Search ADS PubMed  Cavallo E., Galiani S., Noy I., Pantano J.. 2013. Catastrophic Natural Disasters and Economic Growth. Review of Economics and Statistics  95 5: 1549– 61. http://dx.doi.org/10.1162/REST_a_00413 Google Scholar CrossRef Search ADS   Collins A.R., Steinback S.. 1993. Rural Households Response to Water Contamination in West Virginia. JAWRA Journal of the American Water Resources Association  29 2: 199– 209. Google Scholar CrossRef Search ADS   Coffman M., Noy I.. 2011. Hurricane Iniki: Measuring the Long-Term Economic Impact of a Natural Disaster Using Synthetic Control. Environment and Development Economics  17: 187– 205. Google Scholar CrossRef Search ADS   Federal Emergency Management Agency. 2014. FEMA Continues to Support Response Efforts in West Virginia. Available at: https://www.fema.gov/news-release/2014/01/13/fema-continues-support-response-efforts-west-virginia. Gabriel T. 2014. Thousands without Water After Spill in West Virginia. The New York Times.  Available at: http://www.nytimes.com/2014/01/11/us/west-virginia-chemical-spill.html. Grey D., Sadoff C.W.. 2007. Sink or Swim? Water Security for Growth and Development. Water Policy  9 6: 545– 71. http://dx.doi.org/10.2166/wp.2007.021 Google Scholar CrossRef Search ADS   Hanna-Attisha M., LaChance J., Sadler R.C., Schnepp A.C.. 2016. Elevation Blood Levels in Children Associated with the Flint Drinking Water Crisis: A Spatial Analysis of Risk and Public Health Response. American Journal of Public Health  106 2: 283– 90. Google Scholar CrossRef Search ADS PubMed  Hediger M.L., Overpeck M.D., Ruan W.J., Troendle J.F.. 2002. Birthweight and Gestational Age Effects on Motor and Social Development. Pediatric and Perinatal Epidemiology  16 1: 33– 46. Google Scholar CrossRef Search ADS PubMed  Hegyi T., Carbone T., Anwar M., Ostfeld B., Hiatt M., Koons A., Pinto-Martin J., Paneth N.. 1998. The Apgar score and its components in the preterm infant. Pediatrics  101 1: 77– 81. Google Scholar CrossRef Search ADS PubMed  Heyman D., Fitzsimmons E.G.. 2014. Wait Continues for Safe Tap Water in West Virginia. The New York Times . Available at: http://www.nytimes.com/2014/01/12/us/the-wait-continues-for-safe-tap-water-in-west-virginia.html. Howard B.C. 2014. What’s the Chemical Behind West Virginia’s River Spill? National Geographic News.  Available at: http://news.nationalgeographic.com/news/2014/01/140110-4-methylcyclohexane-methanol-chemical-spill-west-virginia-science/. Hunter P. 1997. Waterborne Disease: Epidemiology and Ecology . John Wiley and Sons. Hunter P.R., MacDonald A.M., Carter R.C.. 2010. Water Supply and Health. PLoS Medicine  7 11: e1000361. https://doi.org/10.1371/journal.pmed.1000361. Johnson C. 2014. Spill Affects Some Local Areas. The Herald Dispatch . Li J., Olsen J., Vestergaard M., Obel C.. 2011. Low Apgar Scores and Risk of Childhood Attention Deficit Hyperactivity Disorder. The Journal of Pediatrics  158 5: 775– 9. http://dx.doi.org/10.1016/j.jpeds.2010.10.041 Google Scholar CrossRef Search ADS PubMed  Markham P., Gianato J., Hoyer J.. 2016. After Action Review: Emergency Response to January 9, 2014. Freedom Industries Chemical Leak. Available at: http://www.governor.wv.gov/SiteCollectionDocuments/After%20Action%20Review%20Final.pdf?utm_source=1-9-14+AAR+review&utm_campaign=1-9-15+AAR+review&utm_medium=email. Marshall University Center for Business and Economic Research. 2014. CBER Calculates Impact from Chemical Spill into Elk River. Marshall University Center for Business and Economic Research. Available at: http://mediad.publicbroadcasting.net/p/wvpn/files/201402/CBER_PR_WaterUsageBan_02.04.14.pdf. B.A. Miller, Reidinger R.A., eds. 1998. Comprehensive River Basin Development: The Tennessee Valley Authority. Vol. 23. National Center for Health Statistics. 2009-2014. Birth Data. National Center for Health Statistics. World Bank Publications, Hyattsville, Maryland. Parker R. 2015. A River Runs Yellow. The Atlantic.  August 21. Available at: https://www.theatlantic.com/national/archive/2015/08/a-river-runs-yellow/401966/. Raby J. 2016. Judge Tentatively OKs $151M Settlement in Chemical Leak Suit. The Associated Press . Available at: http://bigstory.ap.org/article/12cef7cdaa79477eae2a7a7afd29980b/judge-tentatively-oks-151m-settlement-chemical-leak-suit. Rich D.Q., Liu K., Zhang J., Thurston S.W., Stevens T.P., Pan Y., Kane C., et al.   2015. Differences in Birth Weight Associated with the 2008 Beijing Olympics Air Pollution Reduction: Results from a Natural Experiment. Environmental Health Perspectives  123 9: 880– 8. Google Scholar PubMed  Saha K.C. 2003. Review of Arsenicosis in West Bengal, India--A Clinical Perspective. Critical Reviews in Environmental Science and Technology  33 2: 127– 63. http://dx.doi.org/10.1080/10643380390814514 Google Scholar CrossRef Search ADS   Savoia E., Stoto M.A., Gupta R., Wright N., Viswanath K.. 2015. Public Response to the 2014 Chemical Spill in West Virginia: Knowledge, Opinions and Behaviors. BMC Public Health  17: 790. doi:10.1186/s12889-015-2134-2. Google Scholar CrossRef Search ADS   Shaban A., Hamze M., El-Baz F., Ghoneim E.. 2009. Characterization of an Oil Spill Along the Lebanese Coast by Satellite Images. Environmental Forensics  10 1: 51– 9. http://dx.doi.org/10.1080/15275920802659735 Google Scholar CrossRef Search ADS   Siegel M. 2002. The Effectiveness of State-Level Tobacco Control Interventions: A Review of Program Implementation and Behavioral Outcomes. Annual Review of Public Health  23 1: 45– 71. http://dx.doi.org/10.1146/annurev.publhealth.23.092601.095916 Google Scholar CrossRef Search ADS PubMed  Snair J. 2014. Challenges in Responding to the Elk River Chemical Spill: An Interview with Dr. Rahul Gupta. National Association of County and City Health Officials.  Available at: http://nacchopreparedness.org/response-to-the-january-9th-chemical-spill-into-the-elk-river-west-virginia-an-interview-with-dr-gupta/. Schwabach A. 1989. The Sandoz Spill: The Failure of International Law to Protect the Rhine from Pollution. Ecology Law Quarterly  16: 443– 80. Stuart A., Otterblad Olausson P., Kallen K.. 2011. Apgar Scores at 5 Minutes after Birth in Relation to School Performance at 16 Years of Age. Obstetrics and Gynecology  118 2: 201– 8. http://dx.doi.org/10.1097/AOG.0b013e31822200eb Google Scholar CrossRef Search ADS PubMed  U.S. Census Bureau, American Community Survey 2006-2015. 1-Year Estimates. Using American FactFinder. Generated April 20, 2017. Available at: http://www.bea.gov/iTable/iTableHtml.cfm?reqid=70&step=10&isuri=1&7003=1000&7035=-1&7004=naics&7005=1&7006=xx&7036=-1&7001=21000&7002=2&7090=70&7007=-1&7093=levels. U.S. Chemical Safety and Hazard Investigation Board. 2016. Investigation Report: Chemical Spill Contaminates Public Water Supply in Charleston, West Virginia. 2014-01-I-WV. U.S. Chemical Safety and Hazard Investigation Board. Available at: http://www.csb.gov/assets/1/19/Freedom_Industries_Report.pdf. Watkins T., Ellis R.. 2016. Official: No Quick Fix for West Virginia Water Woes. CNN.com.  Available at: http://www.cnn.com/2014/01/11/us/west-virginia-contaminated-water/index.html. Whelton A.J., McMillan L., Connell M., Kelley K.M., Gill J.P., White K.D., Gupta R., Dey R., Novy C.. 2014. Residential Tap Water Contamination Following the Freedom Industries Chemical Spill: Perceptions, Water Quality, and Health Impacts. Environmental Science and Technology  49: 813– 23. Google Scholar CrossRef Search ADS   White K. 2014a. Chemical, Water Companies Quickly Face Lawsuits. Charleston Gazette-Mail . White K.. 2014b. Freedom Industries Files for Bankruptcy. Charleston Gazette-Mail . © The Authors 2017. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For Permissions, please email: journals.permissions@oup.com

Journal

American Journal of Agricultural EconomicsOxford University Press

Published: Mar 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off