Advanced environmental economics - Referee Report

Anonymous
timer Asked: Nov 26th, 2018
account_balance_wallet $25

Question description

Referee Report

You should write a report for each reading. All the reports should be printed, one and half spacing, each one should be no more than two pages. In the report, you should include the following sections:

• Motivation

• Research question

• Background and literature

• Data

• Methodology

• Findings and policy implication

• Your comments and critiques

Use your own understanding to write the report, do not copy and paste.

American Economic Association The Housing Market Impacts of Shale Gas Development Author(s): Lucija Muehlenbachs, Elisheba Spiller and Christopher Timmins Source: The American Economic Review, Vol. 105, No. 12 (DECEMBER 2015), pp. 3633-3659 Published by: American Economic Association Stable URL: https://www.jstor.org/stable/43821388 Accessed: 05-11-2018 02:50 UTC JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The American Economic Review This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms American Economic Review 2015, 105( 12): 3633-3659 http://dx.doi.org/10. 1257/aer.20 140079 The Housing Market Impacts of Shale Gas Development By Lucija Muehlenbachs, Elisheba Spiller, and Christopher Timmins* Using data from Pennsylvania and an array of empirical techniques to control for confounding factors, we recover hedonic estimates of property value impacts from nearby shale gas development that vary with water source, well productivity, and visibility. Results indicate large negative impacts on nearby groundwater-dependent homes, while piped-water-dependent homes exhibit smaller positive impacts, suggesting benefits from lease payments. Results have implications for the debate over regulation of shale gas development. ( JEL L71, Q35, Q53, R31) Technological improvements in the extraction of oil and natural gas from unconventional sources have transformed communities and landscapes and brought debate and controversy in the policy arena. Shale gas plays underlying the populated northeastern United States were thought to be uneconomical less than ten years ago, but now contribute a major share of US gas supply.1 Natural gas has been hailed as a bridge to energy independence and a clean future because of its domestic sourcing and, compared with coal and petroleum derivatives, its smaller carbon footprint and reduced emissions of other pollutants (e.g., particulates, sulfur dioxide, carbon monoxide, and nitrogen oxides). Yet, opposition to unconventional methods of natural gas extraction has emerged, citing the potential for damages from methane leakage, water contamination, and local air pollution (see Mason, Muehlenbachs, and Olmstead 2015 for a review). ♦Muehlenbachs: Department of Economics, University of Calgary, 2500 University Drive NW, Calgary, Alberta, Canada T2N 1N4, and Resources for the Future (e-mail: muehlenbachs@rff.org); Spiller: Environmental Defense Fund, 257 Park Avenue South, New York, NY 10010 (e-mail: espiller@edf.org); Timmins: Department of Economics, Duke University, 213 Social Sciences Building, 419 Chapel Drive, Box 90097, Durham, NC 27708 (e-mail: christopher.timmins@duke.edu). We thank Kelly Bishop, Yanyou Chen, Jessica Chu, Elaine Hill, Mark Fleming, Carolyn Kousky, Alan Krupnick, Nicolai Kuminoff, Corey Lang, Lala Ma, Jan Mares, Ralph Mastromonaco, Klaus Moeltner, Jaren Pope, Seth Sanders, Stefan Staubli, Randy Walsh, Zhongmin Wang, and Jackie Willwerth for their support. We thank seminar participants at Carnegie Mellon University, Georgia Institute of Technology, EPA-NCEE, Tinbergen Institute/Free University of Amsterdam, Toulouse School of Economics, University of Massachusetts Amherst, University of Michigan, University of Pittsburgh, West Virginia University, Colorado School of Mines, and University of Stirling for their helpful comments. All remaining errors and omissions are our own. We are grateful to CoreLogic for the data on property transactions. We thank the Bureau of Topographic and Geologic Survey in the Pennsylvania Department of Conservation and Natural Resources for data on well completions. The authors declare that they have no relevant or material financial interests that relate to the research described in this paper. +Go to http://dx.doi.org/10.1257/aer.20140079 to visit the article page for additional materials and author disclosure statement(s). 1 In 2000, shale gas accounted for 1 .6 percent of total US natural gas production; this rose to 4. 1 percent in 2005, and by 2010, it had reached 23. 1 percent (Wang and Krupnick 2013). Natural gas from the Marcellus formation currently accounts for the majority of this production (Rahm et al. 2013) and can be attributed to advances in hydraulic fracturing, horizontal drilling, and 3-D seismic imaging. 3633 This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3634 THE AMERICAN ECONOMIC REVIEW DECEMBER 201 5 Economic and environmental impacts may also nomenon, where local areas facing shale develop employment, business activity, and governmen may also suffer from negative social, economic such as increased crime rates, housing rental co Albrecht 1978; Freudenburg 1982). Furthermor a "bust" if benefits from shale gas developmen goods might be expanded during the boom at co underutilized, and sectors with better growth boom, leaving the area worse off in the long ru Properties surrounding shale gas development in value depending on whether the benefits o Moreover, benefits and costs may be heterogene ple, properties that rely on private water may suff confronted with shale gas development if there Access to a safe, reliable source of drinking wat property's value; even a perceived threat to that on housing prices. This is very important, as the to contaminate groundwater has been hotly deb benefits from drilling can vary with a variety of f ing activity, environmental activism, economic urban density (Theodoři 2009; Wynveen 201 1; valid arguments on both sides of the debate sur question of whether the benefits outweigh the c paper uses hedonic theory to better understand analysis exploits the trade-offs between propert neighborhood characteristics and amenities) and the former.4 Measuring the impacts of shale gas a convenient way to quantify its effects (wheth The impact of shale gas development on prope of a growing body of literature. One of the fir McMillan 2005), while not a study of shale ga wells emitting hydrogen sulfide (a lethal gas More recent studies have focused on shale gas (2014) and Muehlenbachs, Spiller, andTimmins ( County, Pennsylvania. Gopalakrishnan and Klaib shale gas wells diminishes property values across Muehlenbachs, Spiller, and Timmins (2013) find only for properties dependent on private-groun source. James and James (2014) find negative im 2 See, for example, Raimi and Newell (2014) and Wynveen (201 1 opment is an active area of research; for papers specific to shale g (2014); and Fetzer (2014). An example from Dimock, Pennsylvania, can be seen in these Contamination in Dimock," Riverkeeper.org, March 22, 2012, ve in Dimock," eidmarcellus.org, August 3, 2012. Under ambiguity of groundwater-dependent properties. 4 See online Appendix Section B for a deeper discussion of the h This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL: HOUSING AND SHALE GAS 3635 Delgado, Guilfoos, and Boslett (2014) only find weak evidence o and Bradford Counties, Pennsylvania. At the broader level, Bos Lang (2014) find that groundwater-dependent homes in New Yo the possibility of shale gas development. Weber, Burnett, and X property values in Texas are higher in zip codes with shale, conj by local public finances. A major obstacle to accurately estimating the impact of shale g surrounding homes is the presence of correlated unobservables identification. Shale gas wells are not located randomly, but ma with features that aid in the drilling process, such as near a roa servable property and neighborhood attributes may therefore be proximity to wells and with the property value. Providing evide wells are not randomly assigned (see Figures 4 and 5), we highl of using variation in the price of a property over time to estim new nearby shale gas well. We are able to conduct this estimati long panel of property transactions spanning 1995 to 2012; many mate the impacts of shale gas wells by comparing values across d Facilitated by data from across the Commonwealth of Pennsylv triple-difference (DDD) estimator, combined with a mix of fixe ment boundary techniques to deal with time invariant and tim ables that may be correlated with proximity to shale gas wells source. Moreover, we show that similar results are obtained by ence-in-differences-nearest-neighbor-matching (DDNNM) techn rely on panel data variation for identification. By using a geog dataset of properties, we are able to measure economic impacts local level while controlling for macroeconomic effects (e.g., th outsourcing of manufacturing) at the county level. Finally, our erty transactions creates a solid baseline for our DDD estimator shale gas development. Our results demonstrate that groundwater-dependent homes tively affected by nearby shale gas development, indicating that to groundwater contamination has indeed materialized into a re proximate homes that have access to publicly supplied piped hand, appear to receive small benefits from that development. H only comes from producing wells, suggesting that it reflects roy homeowner from natural gas production. Recently drilled wells the past year) do not contribute to this benefit, providing evide and hydraulic fracturing stages of shale gas development are th The burden of aesthetic disruptions is corroborated by the findi impacts are only driven by wells that are not in view of the pro These results are particularly representative of the economic i development in light of the fact that the Marcellus shale gas play country.5 Given the amount of extraction that may occur in this reg 5 See US Energy Information Administration (201 1) for a ranking of shale gas pla recoverable reserves. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3636 THE AMERICAN ECONOMIC REVIEW DECEMBER 201 5 the effect on property values may have important i the benefits and costs of a large scale shift toward d Our paper proceeds as follows. Section I describ II details our data, and Section III reports our emp results, with a summary of different property value concludes. Finally, we provide an online Appendix th made to our dataset; (ii) hedonic theory, the simplify most of the hedonic literature (including our analysis using panel data when the residential composition sh checks over space and time; (iv) the impact of shale g sociodemographics, the frequency of sales, and new c ical heterogeneity of the results. I. Methodology Our goal is to recover estimates of the nonmarket gas wells by measuring their capitalization into housi tiated by proximity to wells and by water source - e.g that are dependent upon their own private groundwat water versus houses at a similar range in public wat piped water. In this paper we identify the different proximity and drinking water source. A. Impact Categories We categorize the impacts of nearby shale gas expl housing values as follows: Adjacency Effects. - This category refers to all of ated with close proximity to a shale gas well that ar source. Costs in this category may include noise and lution (McKenzie et al. 2012; Litovitz et al. 2013), scape, and visual disamenities associated with drilling The most obvious benefit would be royalties and lease owner for the extraction of the natural gas beneath t 6 Given that property values could be negatively affected by proximit why a homeowner would be willing to lease their mineral rights to the ga important to note that refusing to lease out the mineral rights under one's p drilling on a neighbor's land, which would still expose the holdout-homeo for example, of groundwater contamination). Horizontal drilling requires contiguous area, which implies that a critical mass of homeowners need to occurs. Homeowners may form coalitions to prevent drilling; however, un neighbors, each homeowner may have an incentive to deviate and lease th This may be particularly true if there is the possibility of a large up-fron bor's decision to sign a lease, therefore, leasing one's mineral rights will and still being exposed to the impacts of shale development. We may the choose to lease their rights although it might have been optimal for none 7 In Pennsylvania, upon signing their mineral rights to a gas company, of dollars per acre as an upfront "bonus" payment, and then a 12.5 perce This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL : HOUSING AND SHALE GAS 3637 the mineral rights from the surface (property) rights, leaving fu ability to profit from lease and royalty payments. The extent to have been severed throughout our sample is impossible to know detailed data on leases and deeds, which we do not have. Thus, o find little to no positive impacts for homes located near shale g rights may have been severed, and without knowing which prope their mineral rights, we are unable to capture the positive impac Instead, our adjacency effect estimates an overall net effect: th payments for those households who may be receiving them unable to profit from the lease payments due to severed mineral ative externalities of being located near a drilling site (excludin associated with the property depending on groundwater). Groundwater Contamination Risk ( GWCR ). - This category r tional cost capitalized into adjacent properties that are dependent Our identification strategy assumes that this is the only additio cency associated with reliance on groundwater.8 If royalty rate water source, then this should not impact our estimate of GWCR In addition to these two direct impacts of shale gas activities there are broader Vicinity Effects that can also impact housing p the impacts of shale gas development on houses within a broadl 20 km) surrounding wells and may include increased traffic con damage from trucks delivering fresh water to wells and haulin wastewater disposal (to the extent that is done locally), increase and demand for local goods and services, and impacts on local p allow these vicinity effects to differ by drinking water source a reflect jurisdictional boundaries that determine the extent to wh benefit from, for example, an impact fee.9 Furthermore, there which are not specifically related to shale gas activity and are th be common to areas with and without a publicly provided drink Given the time period that we study, this impact category inclu ble, the subsequent housing bust and national recession, impa and jobs moving overseas, and other regional economic impacts. Figure 1 is useful in describing our identification strategy, and in more detail in Section HIB. Area A represents a buffer drawn defines adjacency. That buffer is located in an area dependent u (GW), i.e., outside the public water service area (PWSA). To choo buffer, we use two pieces of evidence. The first comes from Os who find that drinking water wells within 1 km of shale ga extracted. Natural Gas Forum for Landowners: Natural Gas Lease Offer Tracker, http com/lease_offers_tracker.php?action=resources (accessed September 17, 2015). 8 Data on groundwater contamination resulting from shale gas development in Pennsy available to researchers or homeowners because there was no widespread testing of groun of drilling. What we are measuring is therefore the cost associated with the risk of c homeowners. 9 Impact fees are taxes levied on drilled wells. The total amount of impact fees collected in Pennsylvania through 2014 exceeded $850 million dollars, 60 percent of which is given to local counties and municipalities with wells. See https://stateimpact.npr.org/pennsylvania/tag/impact-fee/ (accessed September 17, 2015). This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3638 THE AMERICAN ECONOMIC REVIEW DECEMBER 2015 Figure 1 . Types of Areas Examined concentrations of methane. Although their findings are not causally identified, th study has received much press attention and to date is one of the most frequently cited studies on the environmental impacts of shale gas development. Second, the distance of the horizontal portion of the well is approximately 1 mile (or 1.6 km).1 This implies that lease payments would be provided to homeowners located within this distance of a well.1 1 We also vary the distance of the buffer to test our localized impact hypothesis, and find that distances less than 2 km are most affected by prox imity, thereby validating our hypothesis. Area B is located outside the adjacency buffer but is within the vicinity of a we and is located above the shale formation. Similarly defined regions of the PWSA area are labeled by C and D, respectively. II. Data We obtained transaction records of all properties sold in 36 count Pennsylvania between January 1995 and April 2012 from CoreLogic, a nati estate data provider. The data contain information on the transaction pri we convert into 2012 dollars), exact street address, parcel boundaries, squ age, year built, lot size, number of rooms, number of bathrooms, and nu stories.12 Figure 2 depicts the location of the Marcellus shale formation from the US Geological Survey) as well as the properties sold. 10 Although electronic records of the location of the horizontal segment of the wellbores are n anecdotal evidence suggests that wellbores are typically between 3,000 feet (0.9 km) and 5,000 feet Energy Information Administration 2013), but could be up to 10,684 feet (3.3 km) which is the longes well in the Marcellus shale (O'Brien 2013). Of course, payments would only be made to those households whose property is located above the while the pipes extend horizontally, they do not necessarily extend radially in all directions and theref of the homes located within 1 .6 km will not be entitled to a payment. Thus, the overall effect of proxi the combined impact on those houses that are eligible for payment and the remaining households eligible. See online Appendix Section A for a description of how we constructed our final samples. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL. 105 NO. 12 MUEHLENBACHS ETAL.: HOUSING AND SHALE GAS 3639 Figure 2. Property Sales Data from CoreLogic Mapped with GIS on Overlay of Marcellus Shale in Pennsylvania To determine the date that wells are drilled, we use the Pennsylvania Departm of Environmental Protection (PADEP) Spud Data as well as the Departme of Conservation and Natural Resources (DCNR) Well Information Sys (the Pennsylvania Internet Record Imaging System/Wells Information Sy [PA*IRIS/WIS]). Combining these two dataseis provides us with the most prehensive dataset on wells drilled in Pennsylvania that is available (for e ple, no other data distributors, such as IHS or Drillinginfo, would provide mo comprehensive data than this).13 The final dataset includes both vertical and zontal wells, both of which produce similar disamenities, including risks of gr water contamination. 14 Because operators are able to drill horizontally underground, they can locate tops of several wellbores close together at the surface, and radiate out the hori tal portion of the wellbore beneath the surface. Therefore, multiple wellbore be drilled within meters of one another on the same "well pad," concentratin surface disruption to a smaller space. Though the data do not group wellbores well pads, we believe this is important to consider when estimating the effec shale gas wells on nearby properties, as the impact from an additional wellbo likely different from the impact of an additional well pad. We therefore assum any wellbore within a short distance of another wellbore is located on the sam (specifically, any wellbore that is closer than 63 m, or the length of an acre, t other wellbore is designated to be in the same well pad).15 We start with 6,260 bores, which we group into 3,167 well pads (with an average of 2 bores per pad 13 We corroborated this by comparing our data with data from Drillinginfo, a credible third source; we 52 more wells than Drillinginfo and, because we have captured completion dates, we are able to use thes the "spud" dates are missing (which was the case for 847 wells). The spud date refers to the first day of dr Drillinginfo does not capture completion dates and thus provides a less complete dataset than that which we Risk of improper well casing or cementing would be present in both vertical and horizontal wells. 15 During completion, a multi- well pad, access road, and infrastructure are estimated to encompass 7.4 in size; after completion and partial reclamation, a multi-well pad averages 4.5 acres in size (New York Department of Environmental Conservation 201 1). This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3640 THE AMERICAN ECONOMIC REVIEW DECEMBER 2015 a maximum of 12). Using the geographic infor wells and the properties, we calculate counts o been drilled, within certain distances, at the t also provides information on the GIS location o to count the number of wells that have been pe (only about 60 percent of the wells that have b can also use the date that the well was permitte remained undrilled. We obtain the volume of na from the PADEP's Oil and Gas Reporting websi Pennsylvania has many hilly and mountainous depending on where the property is located, a h see all the wells within the adjacency buffer. F Kousky, and Chu (2013), who examine the p views, we count the number of wells that are To do so we use ArcGIS's Viewshed tool and a Elevation Dataset (at a 30 meter resolution) to p can see from all directions around the property visible wells within different radii at the time of the sale. To identify properties that do not have access to piped drinking water, we utilize data on public water service areas. We obtained the GIS boundaries of the public water suppliers' service areas in Pennsylvania from the PADEP and assume that any property outside these boundaries is groundwater dependent.18 Figure 3 shows the PWSA areas. The unshaded areas are assumed to depend on private groundwater wells for their drinking water source. This figure demonstrates that PWSAs are scattered throughout the state and that there are large areas without access to piped water, further illustrating the importance of estimating the impacts of shale development on groundwater-dependent homes. III. Empirical Strategy and Results In this section, we estimate the impacts of close proximity to shale gas wells on property values. These effects can be positive, such as in the case that the property owner receives royalty or other lease payments from the gas company for the natural gas extracted from their property, or negative, given perceived impacts of groundwater contamination, noise, light, and air pollution, or the alteration of the local landscape. The siting of shale gas wells can be strategic on the part of gas companies and must be agreed to on part of the property owner, so it is also important to account for a wide range of unobservable attributes that may be correlated with both the property and proximity to the shale well. We first provide some evidence that our adjacency buffer correctly identifies localized impacts. We then begin our 16The data are reported as annual quantities until 2009 and then semiannual from 2010 to 2012. 17 Of course, this technology has limitations. It does not tell us whether the homeowner would be able to see the well from the top floor of a home or from the edge of the property; it also does not take into account obstructing vegetation or other houses. Finally, a taller person may better be able to see the well. 18 There is not much financial assistance to households wishing to extend the piped water service area to their location. Doing so is a costly endeavor according to personal communication with the development manager at the Washington County Planning Commission, April 24, 2012. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL.: HOUSING AND SHALE GAS 3641 Figure 3. Public Water Service Areas in Pennsylvania estimation section with a triple-difference technique that also makes use of pr ties on the boundary of the public water supply area. Finally, we show that sim results can be obtained from a difference-in-differences technique combined w nearest-neighbor matching algorithm that does not rely on panel data variation identification. Comparing the effect over time we find it to be similarly sized in ferent time periods, though cutting by subperiod reduces sample size and stati significance. This points to our estimates being robust to the critique described Kuminoff and Pope (2014), though only weakly so due to low statistical power A. Descriptive Evidence of Adjacency Effects and Groundwater Contamination Risk Here we provide some evidence that the prices of groundwater-dependent houses are in fact affected by proximity to shale gas wells. We draw on a strategy similar to that employed by Linden and Rockoff (2008), which determines the point where a localized (dis) amenity no longer has localized impacts. For our application, this method compares the prices of properties sold after the drilling of a well to the prices of properties sold prior to drilling, and identifies the distance beyond which the well no longer has an additional effect. In order to conduct this test, we create a subsample of properties that have been sold more than once and with at least one sale starting after the placement of only one well pad within 10 km. 19 For each water source, we estimate two price functions based on distance to its nearest well pad: one using a sample of property sales that occurred prior to the well pad being drilled and the other using a sample of property 19 For this exercise, we choose to only look at homes that have one well pad within 10 km, as it would be difficult to separate the impact of the nearest well pad before and after the well pad is drilled if the home was already being impacted by another well pad drilled nearby. We chose 10 km because finding properties with only one well pad within farther distances would reduce our sample size, while we think it is a reasonable assumption that vicinity impacts that are felt at more than 10 km will be similar to those at 10 km. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3642 THE AMERICAN ECONOMIC REVIEW DECEMBER 2015 Figure 4. Price Gradient of Distance from Future/Curre Figure 5. Price Gradient of Distance from Future/Curren sales that occurred after the well pad was drill with local polynomial regressions using as depe regression controlling for county-year, quarter Figure 4 depicts the results from the local po on areas with access to piped water. This figu which depicts areas without access to piped w erty values of groundwater-dependent homes however, the prices for groundwater-dependent well remain the same before and after it is drilled. This exercise demonstrates that This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL : HOUSING AND SHALE GAS 3643 adjacency impacts differ by drinking water source within 2 km our usage of buffers less than 2 km in distance. It also demons of controlling for unobserved characteristics that might be cor of a well and the price of the property; in the case of public properties that are the closest to a well are priced lower ev drilled, while the opposite is true in groundwater-dependent ar B. Triple-Difference (DDD) Estimation ofGWCR Considering the impact categories described in Section IA begin by defining the components of the change in a particular time (A P) in each area: APa = A Adjacency + A GWCR + A Vicinity cw + A A Pß = AVicinityGW + AMacro A Pc = AAdjacency + AVicinityPWSA + AMacro A PD = AVicinityPWSA + AMacro, where, for example, A GWCR refers to the change in price a water contamination risk from new wells in area A. We differentiate vicin- ity effects by drinking water source: AVicinityGW refers to the vicinity impa on groundwater-dependent homes, while A Vicinity PWSA refers to the vicinit impact on PWSA homes. Our strategy for identifying adjacency effects us difference-in-differences (DD) estimator: A Adjacency DD = [APC-APD], where the first difference, "A," reflects the change in price of a particular house (e.g. accompanying the addition of a new well pad). The second difference compares change in prices for PWSA properties adjacent to shale gas development to change in prices of PWSA properties not adjacent to development. For the PWS homes, this differences away vicinity and macro effects that are common acro and D. Because vicinity effects may differ by drinking water source, we can on difference these away by looking within water sources; hence, our adjacency reg sions rely only on PWSA homes. Furthermore, note that the corresponding equat for GW homes results in both adjacency and groundwater contamination risk: (A Adjacency + A GWCR)DD = [APa - APB], Therefore, to estimate the effect of perceived groundwater contamination risk must then difference away the effects across PWSA and GW areas by implement the following triple-difference (DDD) estimator: A GWCRddd = [A PA - A PB] - [A Pc - A PD}. Similar to the expression for adjacency, in this expression, A reflects the fir difference, the change in the price of a particular house accompanying the addit This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3644 THE AMERICAN ECONOMIC REVIEW DECEMBER 2015 of a new well pad. The second difference co each adjacency buffer to the change in prices difference differences away relevant vicinit GWCR and adjacency effects. The third (and double-differences, eliminating adjacency effe new well pad. In order to conduct this estimation, we def results of our adjacency test in Section IIIA. In at well pads rather than wellbores to estimate a we focus on well pads because we are capturing When a pad is cleared and drilling begins, it is bore on that pad will have the same impact on Essentially, we assume that the perception of be the same regardless of the number of wellbo In deriving our empirical specification base begin by considering the price of house i at ti (k = 1, 2, ...K), a house fixed effect (/x,), a fixe raphy (i.e., either county or census tract) and y indicating the quarter ( q, ): K (2) In P j, = a0 + Pik wkt + fi¡ + uit + q, + e,„ k= i where k indexes the well and K is the total number of wells in Pennsylvania; wkt = 1 if well pad k has been drilled by time t (in a sensitivity analysis we differentiate between wells that are merely drilled and actually producing); and pik translates the presence of well wkt into an effect on house price based on its proximity. We can decompose equation (2) by dividing the well pads into those that are within 20 km and those outside of 20 km: K In k= K Pit i *=i = 365 days) in K km 0.021 0.023** 0.01 1** (0.018) (9.8e-03) (4.4e-03) New bores (drilled < 365 days) in K km - 4.4e-03 -9.7e-03 - 3.3e-04 (0.029) (0.013) (8.0e-03) Old undrilled permits (> 365 days) in K km 0.055** 0.022 0.01 1 (0.025) (0.014) (0.012) New undrilled permits (< 365 days) in K km 0.04* 7.2e-03 7.2e-03 (0.023) (0.014) (7.9e-03) Pads in 20 km - 6.0e-04* - 6.2e-04* -6.3e-04* (3.3e-04) (3.3e-04) (3.3e-04) Property effects Yes Yes Yes County-year effects Yes Yes Yes Quarter effects Yes Yes Yes Observations 212,207 212,207 212,207 Notes: Dependent variable is log sale price. Each pan per column. Regressors are the count of wells (or annu depending on the column. The sample used includes onl vice areas. Robust standard errors are clustered by cen *** Significant at the 1 percent level. ** Significant at the 5 percent level. ♦Significant at the 10 percent level. 52 wells, or less than 1 percent of the wel abandoned; therefore, examining the marg appropriate than examining the margin of and abandoned. In panel B we show that th by producing wells. This result is intuitiv payments to the homeowner and the closer receive payments.28 Our final specification in panel C explores in particular, we estimate whether newly d 28 In another specification, not shown, the amount of na natural gas production in the year of sale) also increases pr This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3652 THE AMERICAN ECONOMIC REVIEW DECEMBER 201 5 12 months prior to the sale of the home) affect p bores. When examining timing we focus on wellbo tially added to well pads and therefore an old well drilled on it would look similar to a new well pad. impact from proximity only holds for old wellbore newer bores have an insignificant, negative impac disruptions from the drilling and hydraulic fract truck traffic and noise from drilling and hydrauli benefits associated with lease payments. At a very positive effect felt from old drilled wells; howeve permits, implying that expectations for drilling ha erty values in close proximity.29 D. D iff e re nce-irt- D iff e re > i c e s Nearest-Neigh In this section, we find similar GWCR and adjace that do not rely on panel data variation. In the DDD poral variation in price; however, as described by estimates would be biased if the hedonic gradient that argument is that methods based on using pan time-invariant unobserved property or neighborh rately describe the slope of the hedonic price function residential composition changes over time, causing move. Their argument is summarized in our discuss in the online Appendix. In this subsection, we descr relies on cross-sectional data but uses the logic of d junction with matching techniques to achieve ident be within the same year; although the estimate is a only relies on within-year variation. We focus on re within-year estimates over time as our sample s year-by-year estimates. However, dividing the sam and late) provides weak evidence of a stable gradie The fundamental problem of causal inference is th observation in its untreated state and vice-versa; i observe the price of a house located in close proxim house instead located farther away ("same," in this c and neighborhood attributes, both time-invariant a are frequently used to control for time-invariant u may be correlated with the (dis) amenity of interest 29This provides some evidence that homeowners expect future drilli be some attenuation bias given future expectations. However, formally outside the scope of this research, both in terms of data and computa for a description of the method and data needed to conduct such an est transparent) static hedonic framework in this paper, but note that it is lik muted to the extent that buyers of houses unexposed to wells consider This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL: HOUSING AND SHALE GAS 3653 Matching estimators impute counterfactual observations by with similar houses from a control group.30 The effect of trea averaging across the price differences for matched pairs. Mo niques involved in matching estimators can be found in Abad Abadie and Imbens (2006); Abadie and Imbens (2011); and (201 1); our main specification uses the nearest-neighbor mat The key to the success of this type of matching estimator is to so that unobservable house and neighborhood attributes are no ment status. We do so here by limiting the control sample in by requiring exact matches in other dimensions.31 In particula matching algorithm allows us to require exact matches in th sion (i.e., census tract) to control for neighborhood unobserv poral dimension (i.e., transaction year) to control for time-v Performing nearest-neighbor matching on house attributes, w to be exact in these dimensions to help control for various f that might otherwise bias our results. Moreover, we limit the houses that we expect to be in a relatively homogeneous neig census tract. Thus, we (i) limit our analysis to only houses th a well pad (defining the treatment buffer to be 1, 1.5, or 2 k small adjacency buffer found in Section IIIA); (ii) require exa tract; (iii) require exact matches by year of sale; and (iv) perf rately for groundwater and PWSA houses. The idea behind th houses within 6 km of a well pad in the same census tract that source will be located in similar neighborhoods, thereby r that may be correlated with the location of the property. Req by year of sale will further eliminate differences in unobser year to year at this level of the neighborhood. The nearest neighbor matching algorithm is used to recove average treatment effect on the treated (ATT), or the impact a nonadjacent house inside the adjacency buffer. In Figure 1, a move from B to A for groundwater houses, and from D to We now show that, by differencing these ATT estimates, we estimate of GWCR. We begin by defining the price of properties in each of the four areas in Figure 1 in a cross-sectional analogue of equation (1). Rather than using the change in price of a particular property over time (i.e., A), we focus on cross-sectional differences in prices. Our nearest neighbor matching algorithm applied to ground- water houses yields an estimate of the GWCR combined with the adjacency 30 For more background on the advantages of matching compared to parametric hedonic methods, see Cochran and Rubin (1973); Rubin (1974); Rosenbaum and Rubin (1983); Rubin and Thomas (1992); and Heckman, Ichimura, and Todd (1998). It is important to note that there may exist residual impacts of shale gas development for homes that are not immediately adjacent to a shale gas well. For example, homes that depend on piped water may face some level of drinking water contamination if the public water source is contaminated; while rivers and streams have been found to be affected by shale gas development (see Olmstead et al. 2013) there have yet to be any studies of the impacts on tap water. Key to our identification is that outside of a clearly defined adjacency buffer, the homes are not only less likely to be affected by shale gas development but also that these homes will be equally affected by development regardless of location (i.e., the contamination of publicly sourced piped drinking water is not correlated with adjacency). This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3654 THE AMERICAN ECONOMIC REVIEW DECEMBER 2015 Table 4 - log Sale Price on Groundwater Contamination Risk of Well Pads from a Matching Estimator Sample 1 Panel All A. km 1 .5 km 2 km years PWSA (n = 9,278) 0.002 0.024 -0.013 (-0.08,0.08) (-0.03,0.08) (-0.05,0.03) GW(n = 1,869) -0.070 -0.092 -0.030 (-0.18,0.04) (-0.18,-0.01) (-0.11,0.05) GWCRDD -0.072 -0.116 -0.016 (-0.21,0.06) (-0.22,-0.02) (-0.10,0.07) Panel B. Before 2010 PWSA (n = 3,541) 0.113 0.032 0.052 (-0.04,0.26) (-0.08,0.14) (-0.02,0.13) GW(n = 807) 0.046 -0.083 -0.040 ( -0. 1 2, 0.2 1 ) ( -0.2 1 , 0.05) ( GWCRDD -0.067 -0.115 -0.092 (-0.29,0.16) (-0.28,0.05) (-0.22,0.04) Panel C. 2010 and later PWSA {n = 5,737) -0.059 0.004 -0.046 (-0.15,0.03) (-0.06,0.06) (-0.09,0.00) GW(n = 1,062) -0.104 -0.082 -0.032 ( -0.24, 0.04) ( -0.20, 0.03) ( -0. 1 3, 0.07) GWCRdd -0.044 -0.087 0.014 (-0.21,0.12) (-0.21,0.04) (-0.10,0.13) Notes: Samples comprise all houses within 6 km of a well pad (panel A), within 6 km and sold before 2010 (panel B), and within 6 km and sold in 2010 or later (panel C). Each house in the treatment buffer is matched with four houses in the control sample. Exact match required on year of sale and census tract. Matching also based on house attributes (lot size, square footage, number of bedrooms, number of bathrooms, and year built). Treatment buffer size varies between 1 and 2 km. Bias adjustment equation contains all house attributes. 90 percent confidence intervals reported in parentheses. effect: PA - PB = GWCR + Adjacency (hence, PA is the price of a house A, etc.). Applied to PWSA houses, it yields an estimate of the adjacency alone: Pc - Pd = Adjacency. Differencing these two estimates leaves us estimate of the GWCR: GWCRddnnm = (PA - PB) - (Pc - PD)The results of the nearest neighbor matching procedure are reported in Table 4. The first two rows report the point estimates and 90 percent confidence intervals for PWSA houses using 1, 1.5, and 2 km treatment buffers. The next two rows report comparable figures for groundwater houses. In all cases, the difference-in-differences estimate of the GWCR effect based on these estimates is negative. In the case of the 1.5 km treatment buffer, the DD estimate is large (-11.6 percent) and significant at the 10 percent level. The Kuminoff and Pope critique emphasizes that the temporal average gradient may not always provide a policy-relevant measure of welfare. However, dividing the sample by properties sold before 2010 (panel B) and properties sold in 2010 or after (panel C), the coefficients are similar across time periods though statistically insignificant (potentially due to smaller sample sizes of treated wells in each distinct time period). Therefore, our results weakly address the This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL. : HOUSING AND SHALE GAS 3655 Kuminoff and Pope critique. Importantly, we also show that, rely variation yields an average effect over time that is similar to the get using intertemporal variation.32'33 IV. Summary of Impacts Using a variety of empirical methodologies, we demonstrat groundwater contamination negatively affects house values in th Although data are not available to measure the impact of actual tamination, the perception of these risks is large, causing importa on groundwater-dependent properties near wells. While it is clear that the perceived risk of groundwater contam impacts property values, homes that rely on piped water may in being adjacent to drilled and producing wells. These results appea royalty payments (or expectations of royalties) from productive w evident from how the results change when we use different sized that the positive impacts from being in close proximity to a we distance becomes very small. The overall positive impacts are net near a well; i.e., net of any negative environmental externality ( noise pollution from drilling) that is common to all properties r ing water source. Thus, even homes with piped water are better farther from a well, as long as they are able (i.e., not too far) to payments. Consistent with the increase in property values being and lease payments, we find that the property value increase is d wells. We also find that this positive finding is explained by wells over a year prior to the sale, most likely because disruptions suc the drilling rig, and hydraulic fracturing equipment are present in well's life. Coinciding with the visual disamenity of a shale gas w these positive effects for wells that are not visible from the prope Similarly, for groundwater-dependent homes, the negative im are large when the property is very close (1.5 km or closer) t and become more negative the closer a home gets to a shale gas w the costs of groundwater contamination risk are large and si from -9.9 percent to -16.5 percent), suggesting that there c to the housing market from regulations that reduce the risk. U net impact from adjacency and GWCR and data on the house recent year (April 2011 to April 2012), we calculate the avera groundwater-dependent homes within 1.5 km of a well to be $30, 32 While the DDNNM point estimate is larger than the DDD estimate, it is important to confidence intervals overlap the DDD estimate. Furthermore, it is unlikely that we would the same results, given that the DDD estimator utilizes property fixed effects and the bo DDNNM estimate does not. In further supporting evidence provided in the online Appendix, we show that neighborhood characteristics are not found to have changed in an economically significant manner with the introduction of shale gas. 34This value is calculated using all groundwater-dependent properties that are within 1.5 km of a well and sold between April 201 1 and April 2012. For these properties, the number of well pads in 1 km and between 1 and 1.5 km are combined with the adjacency and GWCR coefficients from our boundary sample (columns 2 and 4, in the first panel of Table 2). This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3656 THE AMERICAN ECONOMIC REVIEW DECEMBER 2015 annual loss for GW properties is larger than the properties within 1.5 km of a shale gas well ($4, to keep in mind that our estimates do not full with groundwater contamination risk. Owners may purchase expensive water filters to clean t a shale gas well nearby; whole home filters can extent that our estimates do not capture adapt lower bound to the actual costs incurred by wells, implying that contamination risk reduc nearby homes. The use of the properties in the band surroun to using the full sample of homes) demonstrat servable attributes that vary with location can negative impacts on groundwater-dependent ho water-dependent neighborhoods may be differ ways when compared with more urban PWSA n might vary over time. Using a sample containi specifically limited to be within the PWSA bou for these unobserved neighborhood differences ting comparison based on water source. V. Conclusion Development of shale deposits has become increasingly widespread due to advances in technology that allow for the inexpensive enhanced extraction of natural gas. This rapid expansion in development has generated ample debate about whether the benefits from a cleaner domestic fuel and the accompanying economic development outweigh the local negative impacts associated with the extraction technology. This paper addresses many of these questions by measuring the net capitalization of benefits and costs of shale gas development at various levels of proximity and water source exposure. The ability of shale gas development to impact nearby groundwater sources has been a major point of discussion. We estimate the local impacts on groundwaterdependent homes to be large and negative, which is not surprising given the attention the media has been placing on this potential risk. As groundwater contamination can cause severe economic hardship on homes without access to piped water, the perception that a nearby shale gas well will cause irreversible harm to an aquifer can have significant effects on nearby property values. These forces are beginning to show up in the way housing markets located on shale plays operate - e.g., recent evidence 35This is calculated using properties that have access to piped water, are within 1.5 km of a well, and are sold in the most recent year of our data. If we also include properties within 2 km of a well and include coefficients from column 6 for properties within 1.5 km and 2 km of a well, the groundwater losses are smaller on average while the piped-water properties have similar gains (i.e., the average loss for G W homes within 2 km of a well is $16,059 compared to gains for PWSA homes on average of $5,070). 36These water filters can cost about $l,480/year for a family of four (http://www.ezclearwater.com/ wordpress/tag/whole-house-water-filtration-system/, accessed September 17, 2015). Given the cost to adjacent groundwater-dependent homes is near $30,000, this implies a yearly cost of approximately $1,500 under a 20 year mortgage, which aligns with the price of installing a filter to clean the drinking water. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL: HOUSING AND SHALE GAS 3657 that major national mortgage lenders are refusing to make loan close proximity to shale gas wells, and that insurance providers policies on those houses.37 However, shale gas development can also bring positive impac through increased employment opportunities, economic exp tantly, lease payments to the holders of mineral rights. Our es there are localized benefits to homes that are adjacent to produ drilling stage is complete. We find that the negative impacts o during the active portion of drilling activities; minimizing con aspects of drilling (such as truck traffic and land clearing) may the benefits of shale gas development. Therefore, while we find small benefits from being in c shale gas wells, we find strong evidence of localized costs bo groundwater-dependent homes. As these negative impacts are ba groundwater contamination risk rather than actual risk or contam understanding the probability of groundwater contamination wo REFERENCES Abadie, Alberto, and Guido W. Imbens. 2002. "Simple and Bias-Corrected Matching Estim Average Treatment Effects." National Bureau of Economic Research Working Paper 283 Abadie, Alberto, and Guido W. Imbens. 2006. "Large Sample Properties of Matching Estim Average Treatment Effects." Econometrica 74 (1): 235-67. Abadie, Alberto, and Guido W. Imbens. 2011. "Bias-Corrected Matching Estimators fo Treatment Effects." Journal of Business & Economic Statistics 29 (1): 1-11. Abbott, Joshua K., and H.Allen Klaiber. 201 1 . "The Value Of Water As an Urban Club Good: ing Approach to HOA-Provided Lakes." Paper presented at the Agricultural and Applied Association Annual Meeting, Pittsburgh, PA. http://econpapers.repec.org/paper/agsaaeal 1/ htm (accessed September 17, 2015). Albrecht, Stan L. 1978. "Socio-cultural Factors and Energy Resource Development in Rur the West." Journal of Environmental Management 7: 73-90. Bayer, Patrick, Robert McMillan, Alvin Murphy, and Christopher Timmins. 2011. "A Model of Demand for Houses and Neighborhoods." National Bureau of Economic Resear ing Paper 17250. Boslett, Andrew, Todd Guilfoos, and Corey Lang. 2014. "Valuation of Expectations: A Hedo of Shale Gas and New York's Moratorium." http://works.bepress.com/corey_lang/18/ September 17, 2015). Boxali, Peter C., Wing H. Chan, and Melville L. McMillan. 2005. "The Impact of Oil and N Facilities on Rural Residential Property Values: A Spatial Hedonic Analysis." Resource an Economics 27 (3): 248-69. Brasier, Kathryn. J., Matthew R. Filteau, Diane K. McLaughlin, Jeffrey Jacquet, Richar man, Timothy W. Kelsey, and Stephan J. Goetz. 201 1. "Residents' Perceptions of Comm Environmental Impacts from Development of Natural Gas in the Marcellus Shale: A Com Pennsylvania and New York Cases." Journal of Rural Social Sciences 26 (1): 32-61. Cochran, William G., and Donald B. Rubin. 1973. "Controlling Bias in Observational S Review." Sankhya : The Indian Journal of Statistics 35 (4): 417^46. Delgado, Michael S, Todd Guilfoos, and Andrew Boslett. 2014. "The Cost of Hydraulic Fra A Hedonic Analysis." http://www.webmeets.com/wcere/20 14/m/viewpaper.asp?pid=890 September 17, 2015). Fetzer, Thiemo. 2014. "Fracking Growth." Center for Economic Performance Discussion P 37 For example, 'Tracking Boom Gives Banks Mortgage Headaches," American Ban "Couple Denied Mortgage Because of Gas Drilling," WTAE , Pittsburgh's Action News This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms 3658 THE AMERICAN ECONOMIC REVIEW DECEMBER 2015 Freudenburg, William R. 1982. "The Impacts of Rapid G ing of Local Community Residents." In Coping with Rap Boulder, CO: Westview Press. Gopalakrishnan, Sathya, and H. Allen Klaiber. 2014. "Is Residents? Evidence from Housing Values in Pennsylv Economics 96 (1): 43-66. Heckman, James, Hidehiko Ichimura, and Petra E. Todd ation Estimator." Review of Economic Studies 65 (2): 2 James, Alexander, and Jasmine James. 2014. "A Cana ing Market Busts in Colorado." http://alexandergjame hedonic.weld.manuscript.jeem.pdf (accessed September Kuminoff, Nicolai V., and Jaren C. Pope. 2014. "Do 'Cap the Public's Willingness to Pay." International Econom Linden, Leigh, and Jonah E. Rockoff. 2008. "Estimates of ues from Megan's Laws." American Economic Review Litovitz, Aviva, Aimee Curtright, Shmuel Abramzon, N 2013. "Estimation of Regional Air-Quality Damages fro in Pennsylvania." Environmental Research Letters 8 (1) Lovejoy, Stephen B. 1977. Local Perceptions of Energy D Plateau . Lake Powell Research Project Report PB-2873 Maniloff, Peter, and Ralph Mastromonaco. 2014. "The L Shale Development." http://www.researchgate.net/pub Impacts_of_Unconventional_Shale_Development (acce Mason, Charles F., Lūcija Muehlenbachs, and Sheila M Gas Development." Annual Review of Resource Econom McKenzie, Lisa M., Roxana Z. Witter, Lee S. Newman, Risk Assessment of Air Emissions from Development o Science of the Total Environment 424: 79-87. Muehlenbachs, Lucija. 2015. "A Dynamic Model of Clea Production." International Economic Review 56 (1): 15 Muehlenbachs, Lucija, Elisheba Spiller, and Christophe and the Costs of Groundwater Contamination Risk." R 12-40-REV. Muehlenbachs, Lucija, Elisheba Spiller, and Christopher Timmins. 2015. "The Housing Market Impacts of Shale Gas Development: Dataset." American Economic Review, http://dx.doi. org/10. 1257/aer.20140079. New York State Department of Environmental Conservation. 2011. "Revised Draft: Supplemental Generic Environmental Impact Statement on the Oil, Gas and Solution Mining Regulatory Program, Well Permit Issuance for Horizontal Drilling and High-Volume Hydraulic Fracturing to Develop the Marcellus Shale and Other Low-Permeability Gas Reservoirs." http://www.dec.ny.gov/data/ dmn/ogprdsgeisfull.pdf (accessed September 17, 2015). O'Brien, Dan. 2013. "Longest Lateral: Consol Innovates Efficiencies." http://oogeep.org/2013/ll/ longest-lateral-consol-innovates-efficiencies/ (accessed September 17, 2015). Olmstead, Sheila, Lucija Muehlenbachs, Jhih-Shyang Shih, Ziyan Chu, and Alan Krupnick. 2013. "Shale Gas Development Impacts on Surface Water Quality in Pennsylvania." Proceedings of the National Academy of Sciences 1 10 (13): 4962-67. Osborn, Stephen G., Avner Vengosh, Nathaniel R. Warner, and Robert B. Jackson. 201 1. "Methane Contamination of Drinking Water Accompanying Gas-Well Drilling and Hydraulic Fracturing." Proceedings of the National Academy of Sciences 108 (20): 8172-76. Rahm, Brian G., Josephine T. Bates, Lara R. Bertoia, Amy E. Galford, David A. Yoxtheimer, and Susan J. Riha. 2013. "Wastewater Management and Marcellus Shale Gas Development: Trends, Drivers, and Planning Implications." Journal of Environmental Management 120: 105-13. Raimi, Daniel, and Richard G. Newell. 2014. "Shale Public Finance: Local Government Revenues and Costs Associated with Oil and Gas Development." Duke University Energy Initiative Report. Rosenbaum, Paul R., and Donald B. Rubin. 1983. "The Central Role of the Propensity Score in Observational Studies for Causal Effects." Biometrika 70 (1): 41-55. Rubin, D.B. 1974. "Estimating Causal Effects of Treatments in Randomized and Nonrandomized studies." Journal of Educational Psychology 66 (5): 688-701. Rubin, Donald B., and Neal Thomas. 1992. "Characterizing the Effect of Matching Using Linear Propensity Score Methods with Normal Distributions." Biometrika 79 (4): 797-809. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms VOL 105 NO. 12 MUEHLENBACHS ETAL : HOUSING AND SHALE GAS 3659 Theodoři, G. L. 2009. "Paradoxical Perceptions of Problems Associated wit Gas Development." Southern Rural Sociology 24 (3): 97-1 17. Throupe, Ron, Robert A. Simons, and Xue Mao. 2013. "A Review of Hydro tial Effects on Real Estate." Journal of Real Estate Literature 21 (2): 205 US Energy Information Administration. 2011. Review of Emerging Res Shale Oil Plays. Washington, DC: EIA. http://www.eia.gov/analysis/stu September 17, 2015). US Energy Information Administration. 2013. Technically Recoverabl Resources: An Assessment of 137 Shale Formations in 41 Countries O Washington, DC: EIA. http://www.eia.gov/analysis/studies/worldshaleg 17, 2015). Walls, Margaret, Carolyn Kousky, and Ziyan Chu. 2013. "Is What You See What You Get? The Value of Natural Landscape Views." Resources for the Future Discussion Paper 13-25. Wang, Zhongmin, and Alan Krupnick. 2013. "A Retrospective Review of Shale Gas Development in the United States." Resources for the Future Discussion Paper 13-12. Weber, Jeremy G. 2012. "The Effects of a Natural Gas Boom on Employment and Income in Colorado, Texas, and Wyoming." Energy Economics 34 (5): 1580-88. Weber, Jeremy G., J. Burnett, and Irene M. Xiarchos. 2014. "Shale Gas Development and Housing Values over a Decade: Evidence from the Barnett Shale." United States Association for Energy Economics Research Paper Series 14-165. Wynveen, Brooklynn J. 2011. "A Thematic Analysis of Local Respondents' Perceptions of Barnett Shale Energy Development." Journal of Rural Social Sciences 26 (1): 8-31. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:50:44 UTC All use subject to https://about.jstor.org/terms
American Economic Association Traffic Congestion and Infant Health: Evidence from E-ZPass Author(s): Janet Currie and Reed Walker Source: American Economic Journal: Applied Economics, Vol. 3, No. 1 (January 2011), pp. 65-90 Published by: American Economic Association Stable URL: https://www.jstor.org/stable/25760246 Accessed: 05-11-2018 02:24 UTC JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to American Economic Journal: Applied Economics This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms American Economic Journal: Applied Economics 3 (January 2011): 65-90 http://www.aeaweb.org/articles.php7doi?10 J257/app.3.1.65 Traffic Congestion and Infant Health: Evidence from E-ZPass1 By Janet Currie and Reed Walker* We exploit the introduction of electronic toll collection, (E-ZPass), which greatly reduced both traffic congestion and vehicle emissions near highway toll plazas. We show that the introduction ofE-ZPass reduced prematurity and low birth weight among mothers within 2 kilometers (km) of a toll plaza by 10.8 percent and 11.8 percent, respectively, relative to mothers 2-10 km from a toll plaza. There were no immediate changes in the characteristics of mothers or in housing prices near toll plazas that could explain these changes. The results are robust to many changes in specification and suggest that traffic congestion contributes significantly to poor health among infants. (JEL112, J13, Q51, Q53, R41) Motor vehicles are50 apercent major source of(CO), air34pollution. Nationally they are respon sible for over of carbon monoxide percent of nitrogen dioxide (N02), and over 29 percent of hydrocarbon emissions, in addition to as much as 10 percent of fine particulate matter emissions (Michelle Ernst, James Corless, and Ryan Greene-Roesel 2003). In urban areas, vehicles are the dominant source of these emissions. Furthermore, between 1980 and 2003 total vehicle miles trav eled (VMT) in urban areas in the United States increased by 111 percent against an increase in urban lane-miles of only 51 percent (US Department of Transportation 2005). As a result, traffic congestion has steadily increased across the United States, causing 3.7 billion hours of delay by 2003 and wasting 2.3 billion gallons of motor fuel (David Schrank and Tim Lomax 2005). Traditional estimates of the cost of congestion typically include delay costs (William S. Vickrey 1969), but they rarely address other congestion externalities such as the health effects of congestion. This paper seeks to provide estimates of the health effects of traffic congestion by examining the effect of a policy change that caused a sharp drop in congestion (and therefore in the level of local motor vehicle emissions) within a relatively short time frame at different sites across the northeastern United States. Engineering studies * Currie: Department of Economics, Columbia University, 420 West 118th Street, New York, NY 10027 (e-mail: janet.currie@columbia.edu); Walker: Department of Economics, Columbia University, 420 West 118th Street, New York, NY 10027 (e-mail: rw2157@columbia.edu). We are grateful to the MacArthur Foundation for financial support. We thank Katherine Hempstead and Matthew Weinberg of the New Jersey Department of Health, and Craig Edelman of the Pennsylvania Department of Health for facilitating our access to the data. We are grate ful to James MacKinnon and seminar participants at Harvard University, Princeton University, Queens University, Tulane University, the University of Maryland, the University of Massachusetts-Amherst, the University of Rome, Uppsala University, Yale University, the City University of New York, the NBER Summer Institute, and the SOLE/ EALE 2010 meetings for helpful comments. All opinions and any errors are our own. t To comment on this article in the online discussion forum, or to view additional materials, visit the article page at http://www.aeaweb.org/articles.php?doi= 10.1257/app.3.1.65. 65 This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 66 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JANUARY 20 suggest that the introduction of electronic toll collection (ETC) technology, c E-ZPass in the Northeast, sharply reduced delays at toll plazas and pollution ca by idling, decelerating, and accelerating. We study the effect of E-ZPass, and the sharp reductions in local traffic congestion, on the health of infants born to mo ers living near toll plazas. This question is of interest for three reasons. First, there is increasing evide the long-term effects of poor health at birth on future outcomes. For example birth weight has been linked to future health problems and lower educational a ment (see Currie 2009 for a summary of this research). The debate over the cos benefits of emission controls and traffic congestion policies could be signific impacted by evidence that traffic congestion has a deleterious effect on fetal h Second, the study of newborns overcomes several difficulties in making the c nection between pollution and health because, unlike adult diseases that may r pollution exposure that occurred many years ago, the link between cause and ef is immediate. Third, E-ZPass is an interesting policy experiment because, whi lution control was an important consideration for policy makers, the main m for consumers to sign up for E-ZPass is to reduce travel time. Hence, E-ZPass o an example of achieving reductions in pollution by bundling emissions reduct with something consumers perhaps value more highly, such as reduced travel t Our analysis improves upon much of the previous research linking air pollu to fetal health as well as on the somewhat smaller literature focusing specifica the relationship between residential proximity to busy roadways and poor preg outcomes. Since air pollution is not randomly assigned, studies that attempt t pare health outcomes for populations exposed to differing pollution levels ma be adequately controlling for confounding determinants of health. Since air qua capitalized into housing prices (see Kenneth Y. Chay and Michael Greenstone 2 families with higher incomes or preferences for cleaner air are likely to sort in tions with better air quality, and failure to account for this sorting will lead to ove timates of the effects of pollution. Alternatively, pollution levels are higher in areas where there are often more educated individuals with better access to h care, which can cause underestimates of the true effects of pollution on health. In the absence of a randomized trial, we exploit a policy change that created local and persistent reductions in traffic congestion and traffic related air em for certain segments along a highway. We compare the infant health outcom those living near an electronic toll plaza before and after implementation of Eto those living near a major highway but further away from a toll plaza. Specif we compare mothers within 2 km of a toll plaza to mothers who are between and 10 km from a toll plaza, but still within 3 km, of a major highway befor after the adoption of E-ZPass in New Jersey and Pennsylvania. New Jersey and Pennsylvania provide a compelling setting for our part research design. First, both New Jersey and Pennsylvania are heavily populat with New Jersey being the most densely populated state in the United States Pennsylvania being the sixth most populous state in the country. As a result, th states have some of the busiest Interstate systems in the country, systems that also pen to be densely surrounded by residential housing. Furthermore, we know th addresses of mothers, in contrast to many observational studies which approx This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. 1 CURRIE AND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 67 the individual's location as the centroid of a geographic area or by computing average pollution levels within the geographic area. This information enables us to improve on the assignment of pollution exposure. Lastly, E-ZPass adoption and take up was extremely quick, and the reductions in congestion spillover to all automobiles, not just those registered with E-ZPass (New Jersey Turnpike Authority 2001). Our difference-in-differences research design relies on the assumption that the characteristics of mothers near a toll plaza change over time in a way that is com parable to those of other mothers who live further away from a plaza, but still close to a major highway. We test this assumption by examining the way that observable characteristics of the two groups of mothers and housing prices change before and after E-ZPass adoption. We also estimate a range of alternative specifications in an effort to control for unobserved characteristics of mothers and neighborhoods that could confound our estimates. We find significant effects on infant health. The difference-in-difference models suggest that prematurity fell by 6.7-9.16 percent among mothers within 2 km of a toll plaza, while the incidence of low birth weight fell by 8.5-11.3 percent. We argue that these are large but not implausible effects given previous studies. In contrast, we find that there are no significant effects of E-ZPass adoption on the demographic characteristics of mothers in the vicinity of a toll plaza. We also find no immediate effect on housing prices, suggesting that the composition of women giving birth near toll plazas shows little change in the immediate aftermath of E-ZPass adoption (though of course it might change more over time). The rest of the paper is laid out as follows. Section I provides necessary back ground. Section II describes our methods, while data are described in Section III. Section IV presents our results. Section VI discusses the magnitude of the effects we find, and Section V details our conclusions. I. Background Many studies suggest an association between air pollution and fetal health.1 Donald R. Mattison et al. (2003) and Svetlana V. Glinianaia et al. (2004a) sum marize much of the literature. For more recent papers see, for example, Currie, Matthew Neidell, and Johannes F. Schmeider (2009); Rose Dugandzic et al. (2006); Mary Huynh et al. (2006); Catherine J. Karr et al. (2009); Sue J. Lee et al. (2008); Jong-Han Leem et al. (2006); Shiliang Liu et al. (2007); Jennifer D. Parker, Pauline Mendola, and Tracey Woodruff (2008); Muhammad T. Salam et al. (2005); Beate Ritz, Michelle Wilhelm, and Yingxu Zhao (2006); Wilhelm and Ritz (2005); Tracey J. Woodruff, Lyndsey A. Darrow, and Parker (2008). Since traffic is a major con tributor to air pollution, several studies have focused specifically on the effects of exposure to motor vehicle exhaust (see Wilhelm and Ritz 2003; Ninez A. Ponce et al. 2005; Michael Brauer et al. 2003; Remy Slama et al. 2007; Timothy K. M. Beatty and Jay P. Shimshack 2009; Christopher R. Knittel, Douglas Miller, and Nicholas J. Sanders 2009). 1 There is also a large literature linking air pollution and child health, some of it focusing on the effects of traffic on child health. See Joel Schwartz (2004) and Glinianaia et al. (2004b) for reviews. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 68 AMERICAN ECONOMIC JOURNAL- APPLIED ECONOMICS JANUARY 2011 At the same time, researchers have documented many differences between peo ple who are exposed to high volumes of traffic and those who are not (Robert B. Gunier et al. 2003). A correlational study cannot demonstrate that the effect of pol lution is causal. Women living close to busy roadways are more likely to have other characteristics that are linked to poor pregnancy outcomes, such as lower income, education, probabilities of being married, and a higher probability of being a teen mother. This is partly because wealthier people are more likely to move away from pollution. Brooks Depro and Chris Timmins (2008) show that gains in wealth from appreciating housing values during the 1990s allowed households in San Francisco to move to cleaner areas. H. Spencer Banzhaf and Randall R Walsh (2008) show that neighborhoods experiencing improvements in environmental quality tend to gain population while the converse is also true. Most previous studies include a minimal set of controls for potential confound ers. Families with higher incomes or greater preferences for cleaner air may be more likely to sort into neighborhoods with better air quality. These families are also likely to provide other investments in their children, so that fetuses exposed to lower levels of pollution also receive more family inputs, such as better quality prenatal care or less maternal stress. If these factors are unaccounted for, then the estimated effects of pollution may be biased upward. Alternatively, emission sources tend to be located in urban areas, and individuals in urban areas may be more educated and have better access to health care, factors that may improve health. Omitting these factors would lead to a downward bias in the estimated effects of pollution, suggest ing that the overall direction of bias from confounding is unclear. Several previous studies are especially relevant to our work because they address the problem of omitted confounders by focusing on "natural experiments." Chay and Greenstone (2003a, 2003b) examine the implementation of the Clean Air Act of 1970 and the recession of the early 1980s. Both events induced sharper reductions in par ticulates in some counties than they do in others, and they use this exogenous variation in pollution at the county-year level to identify its effects. They estimate that a one unit decline in particulates caused by the implementation of the Clean Air Act (or by recession) led to between 5 and 8 (4 and 7) fewer infant deaths per 100,000 live births. They also find some evidence that declines in total suspended particles (TSPs) led to reductions in the incidence of low birth weight. However, the levels of particulates studied by Chay and Greenstone (2003a, 2003b) are much higher than those prevalent today. For example, PM10 levels have fallen by nearly 50 percent from 1980 to 2000. Furthermore, only TSPs were measured during the time period they examine, which precludes the examination of other pollutants that are found in motor vehicle exhaust. Other studies that are similar in spirit include a sequence of papers by C. Arden Pope and his collaborators, who investigated the health effects of the temporary closing of a Utah steel mill (Pope 1989; Michael R. Ransom and Pope 1992; Pope, Schwartz, and Ransom 1992) and Michael S. Friedman et al. (2001) who examine the effect of changes in traffic patterns in Atlanta due to the 1996 Olympic Games. However, these studies did not look at fetal health. Parker, Mendola, and Woodruff (2008) examine the effect of the Utah steel mill closure on preterm births and find that exposure to pollu tion from the mill increased the probability of preterm birth. This study, however, does not speak to the issue of effects of traffic congestion on infant health. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. 1 CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 69 Currie, Neidell, and Schmeider (2009) examine the effects of several pollutants on fetal health in New Jersey using models that include maternal fixed effects to control for potential confounders. They find that CO is particularly implicated in negative birth outcomes. In pregnant women, exposure to CO reduces the avail ability of oxygen to be transported to the fetus. Carbon monoxide readily crosses the placenta and binds to fetal haemoglobin more readily than to maternal haemo globin. It is cleared from fetal blood more slowly than from maternal blood, leading to concentrations that may be 10-15 percent higher in the fetus's blood than in the mother's. Indeed, much of the negative effect of smoking on infant health is believed to be due to the CO contained in cigarette smoke (World Health Organization 2000). Hence, a significant effect of E-ZPass on CO alone would be expected to have a significant positive effect on fetal health. E-ZPass is an electronic toll collection system that allows vehicles equipped with a special windshield-mounted tag to drive through designated toll lanes without stopping to manually pay a toll. The benefits include time saved, reduced fuel con sumption, and reductions in harmful emissions caused by idling and acceleration at toll plazas. In addition, the air quality benefits are thought to be large enough that some counties have introduced ETC explicitly in order to meet pollution migitation requirements under the Clean Air Act (Anthony A. Saka et al. 2000). Engineering estimates of the reduction in pollution with E-ZPass adoption vary. They are typically based on a combination of traffic count data, and measures of the extent to which reducing the idling, deceleration, and acceleration around toll plazas would reduce emissions for a given vehicle mix. For example, Saka et al. (2000) compared data on traffic flows through manned toll lanes and electronic toll collection lanes at one toll plaza at a single point in time and estimated that reductions in queuing, decelerations, and accelerations in the ETC lanes resulted in reductions of 11 percent for N02 and a decrease of more than 40 percent for hydrocarbons and CO relative to emissions in the manned lanes. A similar study of the George Washington Bridge toll plaza, one of those included in this study, by Mohan Venigalla and Michael Krimmer (2006), estimated that VOC, CO, and N02 emissions from trucks were reduced in the E-ZPass lanes by 30.8 percent, 23.5 percent, and 5.8 percent. Although these studies suggest that E-ZPass could lead to substantial reductions in ambient pollution, these studies may overestimate or underestimate the extent of that reduction. For example, if reducing toll plaza delays encourages more people to drive rather than take public transit, then this may offset the reduction in pollution per vehicle to some extent. Conversely, to the extent that drivers in non-E-ZPass lanes also benefit from reduced congestion, comparing delays at E-ZPass and manual lanes will understate the benefits of E-ZPass. We were unable to find a study that measured pol lution in the radius of a toll plaza before and after the introduction of ETC. However, the New Jersey Turnpike Authority commissioned a study of the extent to which E-ZPass reduced total delays at toll plazas (New Jersey Turnpike Authority 2001). This study used before and after data on traffic counts at each toll plaza, and measured the delays at toll plazas using video cameras. Evidently, the total delay is given by (number of vehicles) x (delay per vehicle). This study concluded that total delay at toll plazas dropped by 85 percent after the implementation of E-ZPass, sav ing 1.8 million hours of delay for cars and 231,000 hours of delay for trucks in the This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 70 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JANUARY 2011 year after adoption. If pollution around the toll plaza is proportional to these delay then it is reasonable to conclude that it was also reduced considerably. The report estimated that E-ZPass reduced emissions of N02 by 0.056 tons per day, or 20 tons per year. In 2002, mobile on-road sources emitted approximately 300 tons of N02 per year (New Jersey Department of Environmental Protection 2005). Hence, a crude estimate is that E-ZPass reduced N02 emissions from traffic by about 6.8 percent. Unfortunately, the EPA's air quality monitors are placed throughout the state such that there is only one monitor located near a toll plaza in our study area Furthermore, this particular monitor only measures N02 and sulfer dioxide (S02). Nevertheless, we show evidence that suggests a sharp decline in N02 levels fo lowing E-ZPass adoption. This is in contrast to S02 levels at the same monitor, fo which we see no noticeable decline. This is consistent with the fact that cars produc a large percentage of local N02 emissions, while they are responsible for a ve small fraction of S02 emissions. An important unresolved question is how far elevated pollution levels extend from highways or toll plazas? Most studies have focused on areas 100-500 meters from roadway. However, Shishan Hu et al. (2009) find evidence that pollution from the 4 Freeway in Los Angeles is found up to 2,600 meters from the roadway. Moreover, their study was conducted in the hours before sunrise, when traffic volumes are re tively light and most people are in their homes. We investigate this issue below. We focus on the implementation of E-ZPass on three major state tollways i New Jersey and Pennsylvania: the Pennsylvania Turnpike, the New Jersey Turnpike and the Garden State Parkway. Portions of all three of these state highways rank nationally as some of the busiest in the country. In addition to these state tollways we also use the major bridge and tunnel tolls connecting New Jersey to New York (George Washington Bridge, Lincoln Tunnel, and Holland Tunnel). Each of the bridges and tunnels are extremely well traveled, transporting around 105 million 42 million, and 35 million vehicles, respectively. New Jersey has 38 toll plazas, 3 bridge/tunnel entrances to New York City, 11 along the Garden State Parkway, 2 along the New Jersey Turnpike, and 2 along the Atlantic City Expressway. There a 60 toll plazas in Pennsylvania. Figure 1 shows the toll plazas and major highwa that we use. Our research design exploits the fact that E-ZPass was installed at different time and in different locations across the two states. The Port Authority of New York and New Jersey implemented E-ZPass at the bridge and tunnels entering New Yor City in 1997. Soon after, New Jersey installed its first E-ZPass toll plazas on t Atlantic City Expressway. Starting in December 1999, New Jersey began installing E-ZPass on the Garden State Parkway. Throughout the course of the following year toll plazas were added at the rate of one per month (working from North to Sou on the Garden State Parkway), with the final plaza installed in August of 2000. In September 2000, the New Jersey Turnpike installed E-ZPass at all their toll collec tion terminals throughout the system. Similarly, the PA Turnpike installed most o their toll plazas with E-ZPass in December 2000, with a major addition occurring in December of 2001. E-ZPass adoption and take up was extremely rapid. By early 2001 (1 year after implementation of the Garden State Parkway and NJ Turnpike 1.3 million cars had been registered with E-ZPass in New Jersey. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. 1 CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 71 Figure 1. Locations of Toll Plazas and Major Roadways in New Jersey and Pennsylvania II. Data Our main source of data for this study are Vital Statistics Natality records from Pennsylvania for 1997 to 2002 and for New Jersey for the years 1994 to 2003. Vital Statistics records are a very rich source of data that cover all births in the two states. They have both detailed information about health at birth and background informa tion about the mother, including race, education, and marital status. We were able to make use of a confidential version of the data with the mother's address, and we were also able to match births to the same mother over time using information about the mother's name, race, and birth date. Like most previous studies of infant health, we focus on two birth outcomes: prematurity (defined as gestation less than 38 weeks) and low birth weight (defined as birth weight less than 2,500 grams).2 Using this information, we first divided mothers into three groups: those liv ing within 2 km of a toll plaza; those living within 3 km of a major highway, but between 2 km and 10 kilometers of a toll plaza; and those who lived 10 km or more away from a toll plaza. Our treatment group in the difference-in-difference design is the mothers living within 2 km of a toll plaza, while the control group is those who live close to a highway, but between 2 km and 10 km of a toll plaza. We drop mothers who live more than 10 km away from a toll plaza. In total, we have 98 toll plazas that adopted electronic tolling in our sample, and thus we have 98 separate sample regions. We also drop births that occurred more than three years before or after the E-ZPass conversion of the nearest plaza, in an effort to focus on births that occurred around the changes. All of the mothers in the sample are assigned to their nearest toll plaza. 2 Outcomes such as infant deaths and congenital anomalies are much rarer, and when we restrict the dataset to those who are within 2 km of a toll plaza, there are insufficient cases in our data for us to be able to expect to see an effect. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 72 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JANUARY 2011 Figure 2. Research Design Showing 1.5 km and 2 km Treatment Radii and 3 km from Highway Control Group Figure 2 illustrates the way that we created the treatment and control groups for each of our toll plaza sample regions. As one can see from the figure, there are many homes within the relevant radius of the toll plaza. Moreover, housing tends to follow the highway. The areas more than 2 km away from either a toll plaza or the highway are somewhat less dense. We also repeat this procedure using mothers less than 1.5 km from a toll plaza as the treatment group, comparing them to mothers who live within 3 km of a highway but between 1.5 and 10 km of a toll plaza. In the analysis including mother fixed effects, we select the sample differently. Specifically, we keep only mothers with more than one birth in our data. We then restrict the sample to only mothers who have had at least one child born within 2 km of a toll plaza, since only these mothers can help to identify the effects of E-ZPass. (The other mothers could in principal identify some of the other coefficients in the model, but as we show below, they have quite different average characteristics so we prefer to exclude them). We use all available years of sample data, in order to maximize the number of women we observe with two or more children. We obtained data on housing prices in New Jersey from 1989 to 2009 by submit ting an open access records request. In addition to the sales date and price, these data include information about address, square footage, age of structures, whether the unit is a condominium, assessed value of the land, and assessed value of the structures. We will use these data to see if housing prices changed in the neighbor hood of toll plazas in response to amenity benefits generated from reduced traffic congestion and increased air quality surrounding E-ZPass implementation. Means of the outcomes we examine (prematurity and low birth weight) and of the independent variables are shown in Table 1 for all of these groups. Panel A shows means for the treatment and control group used in the difference-in-differences This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. 1 CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 73 analysis. For the control group, "before" and "after" are assigned on the basis of when the closest toll plaza converted to E-ZPass. The last column of panel A shows means for mothers who live more than 10 km from a toll plaza. They are less likely to have a premature birth, and their babies are less likely to be low birth weight. They are also less likely to be black or Hispanic. These mothers are omitted from our difference-in-difference analysis. The treatment and control groups are similar to each other before the adoption of E-ZPass except in terms of racial composition. Mothers close to toll plazas are much more likely to be Hispanic and somewhat less likely to be African American than other mothers. Mothers close to toll plazas are also less likely to have smoked during the pregnancy. These differences have potentially important implications for our analysis, since other things being equal, African Americans and smokers tend to have worse birth outcomes than others. Hence, it is important to control for these differences, and we will also examine these subgroups separately. In terms of before and after trends, both areas show increases in the fraction of births to Hispanic and African American mothers, and decreases in the fraction of births to smokers and teen mothers over time. The fraction of births that were pre mature rose over time, especially in the control areas. The fraction of births that were low birth weight showed a slight decrease in the treatment area near toll plazas, but an increase in the control areas. These patterns reflect national time trends in the demo graphic characteristics of new mothers and in birth outcomes. We can use these means tables to do a crude difference-in-difference comparison. Such a comparison suggests that prematurity and low birth weight fell by about 7 percent in areas less than 2 km from a toll plaza after E-ZPass. Appendix Table 1 shows changes in mean outcomes when the treatment group is restricted to those who were within 1.5 km of a toll plaza. Panel B of Table 1 shows means for the sample that we use in the mother fixed effects analysis. Panel B shows that, in general, the mothers with more than one birth in the sample have somewhat better birth outcomes?their children are less likely to be premature or low birth weight than in the full sample of children (panel A). The sample of women who have more than one birth and who ever had a child within 2 km of a toll plaza changes over time. Comparing columns 1 and 2 shows that over time this population has become more Hispanic, less educated, and some what more likely to be having a higher order birth. Columns 3 and 4 of panel B show that the population of women who never had a birth within 2 km of a toll plaza are quite different?they are less likely to be Hispanic, the sample tends to gain educa tion over time, and (not surprisingly) live further from a highway. Panel C shows means from the housing sales data. All prices were deflated by the consumer price index (CPI) into 1993 dollars. Comparing columns 1 and 3 suggests that sales prices were similar in areas close to toll plazas and a little fur ther away from toll plazas before E-ZPass, but that prices increased faster near toll plazas after adoption. The same comparison is shown for the area within 1.5 km of a toll plaza and areas 1.5-10 km away from toll plazas in Appendix Table 1. We show below that controlling for a fairly minimal set of covariates (month and year of sale, square footage, age of structure, municipality, and whether it is a condominium) reduces this estimate to statistical insignificance. Still, the idea that prices may have increased, thereby changing the composition of mothers in the This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 74 AMERICAN ECONOMIC JOURNAL: APPUED ECONOMICS JANUARY 2011 Table 1?Summary Statistics >2 km and >2 km and <2kmE-ZPass <2kmE-ZPass <10kmE-ZPass <10kmE-ZPass >10km before after before _after_Toll plaza Panel A. Difference-in-difference sample Outcomes Premature Low birth weight Controls Mother Hispanic Mother black Mother education Mother HS dropout Mother smoked Teen mother Birth order Multiple birth Child male Distance to roadway Observations NJ observations PA observations 0.095 0.095 0.078 0.102 0.089 0.109 0.092 0.085 0.078 0.291 0.16 13.12 0.169 0.089 0.073 1.3 0.028 0.51 1.099 0.332 0.173 13.2 0.164 0.075 0.061 1.37 0.033 0.165 0.233 0.229 0.264 13.24 0.163 0.086 0.069 0.054 0.047 12.92 0.173 0.152 0.079 1.68 0.033 0.512 21 33,758 26,415 7,343 29,677 26,563 3,114 0.082 0.512 1.074 Ever birth <2km 13.276 0.154 0.109 0.082 1.39 0.032 0.514 1.507 1.46 0.037 0.512 1.482 190,904 128,547 161,145 133,560 Ever birth Never birth 62,357 <2km E-ZPass plaza E-ZPass plaza before after 27,585 <2 km E-ZPass plaza before 185,795 70,484 115,311 Never birth<2km E-ZPass plaza after Panel B. Mothers with more than one birth in sample Outcomes Premature 0.088 0.099 Low birth weight 0.081 0.077 Controls Mother Hispanic 0.167 0.29 Mother black 0.145 0.157 0.092 0.086 0.103 0.086 0.088 0.169 0.161 0.171 Mother education 12.78 12.6 12.75 Mother smoked 0.113 0.076 Teen mother 0.041 0.044 Birth order 1.575 1.708 0.072 Mother HS dropout 0.168 0.201 Multiple birth 0.03 0.037 Child male 0.513 0.512 Distance to highway 3.702 Observations 179,537 58,180 NJ observations 85,565 47,012 <2km E-ZPass before Panel C Summary statistics for housing sales data (New Jersey only) 1.598 0.033 0.512 5.598 2.561 <2km E-ZPass after Sales price 94,883 126,006 >2 km and before after <10km E-ZPass 95,518 115,129 1951 1,646 105,341 Square footage 1,573 1,569 5.3 >2 km and Total assessed value 119,166 123,640 Observations 22,350 22,604 0.512 678,025 46,551 Year built 1952 1954 1.735 0.046 962,093 Assessed land value 42,146 43,219 Assessed building value 78,234 81,437 0.095 0.047 485,351 352,751 132,600 1,640,118 PA observations 93,972 11,168 13.13 0.162 0.178 0.135 70,093 <10km E-ZPass 116,691 46,126 69,752 114,403 1950 1,675 102,048 Notes: All observations in panels A and C are selected to be within 3 km of a busy roadway. Housing price data is only for New Jersey and pertains to housing units, not mothers, as described in the text. The housing price data has been deflated by the CPI (base year = 1993). This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. I CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 75 neighborhood provides a motivation for the models we estimate below, including mother fixed effects. Figures 3-6 provide more nuanced pictures of the relationship between E-ZPass adoption, birth weight, and prematurity. Figures 3 and 4 focus on mothers within 2 km of a toll plaza and take the average values over 0.1 km bins before and after E-ZPass. Figure 3 shows that there is a dramatic reduction in low birth weight after E-ZPass in the area closest to the toll plaza. The reduction tapers off and the lines cross a little after 1 km. Figure 4 shows a similar pattern for prematurity, although here the lines cross at about 1.5 km from the toll plaza. Figures 5 and 6 compare low birth weight and prematurity in households more than 1.5 km from a toll plaza and households less than 1.5 km from a toll plaza in the days before and after E-ZPass. These figures indicate a higher incidence of low birth weight in the 500 days prior to E-ZPass adoption in the area near the toll plaza. Around the time of E-ZPass adoption, the incidence of low birth weight near toll plazas begins to decline dramatically, and falls below the control rate soon after adoption. Figure 6 shows increasing rates of prematurity in both mothers near toll plazas and mothers further away from toll plazas. Around the time of E-ZPass adop tion, the rate of prematurity begins to fall for the near toll plaza group. It is noticeable that in both figures, the incidence of poor outcomes begins to decline slightly before the official date of E-ZPass adoption. We believe that this slight dis crepancy in the timing may be explained by E-ZPass construction. Prior to the official opening date, each plaza had to be adapted for E-ZPass. The New Jersey E-ZPass contract included the installation of fiber optic communications networks, patron fare displays, E-ZPass toll plaza signs, and road stripping at a cost of $500 million (New Jersey Department of Transportation 1998). In one recent example, the toll plaza for the 1-78 Toll Bridge is being upgraded to E-ZPass. Construction took place between early January 2010 and Memorial Day, approximately 5 months.3 During that time, commuters were advised to use an alternative route so that traffic would be lighter than usual near this plaza (Warren Reporter 2010). III. Methods To implement our difference-in-difference estimator, we begin by testing the assumptions for the estimator to be valid, namely that any trends in the observable characteristics of mothers are the same across both treatment and control groups. The models for these specification checks take the following form: (1) Mom_Charit = a + bxE-ZPasslt + b2Closeit + b3Plazait + b4 E-ZPass x Closeit + b5Year + b6Month + b7 Distanceit + eit, 3 The construction included: partial demolition and removal of the canopy over a portion of the toll plaza; new overhead sign structures, construction of a canopy over the new open road tolling lanes to house the ETC array; the construction of a concrete barrier to separate the ETC lanes from the others; restriping; and the construction of electrical systems to support the ETC equipment (Delaware River Joint Toll Bridge Commission 2009). This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 76 AMERICAN ECONOMIC JOURNAL: APPUED ECONOMICS JANUARY 2011 Low birth weight by distance before and after E-ZPass Figure 3 Notes: Smoothed plots of treatment and control groups using locally weighted regression. To facilitate computation, observations are first grouped into 0.1-mile bins by treatment and control and averaged. The weights are applied using a tricube weighting function (William S. Cleveland 1979) with a bandwidth of 1. Figure 4 Notes: Smoothed plots of treatment and control groups using locally weighted regression. To facilitate computation, observations are first grouped into 0.1-mile bins by treatment and control and averaged. The weights are applied using a tricube weighting function (Cleveland 1979) with a bandwidth of 1. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL 3 NO. 1 CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH Figure 5 Notes: Smoothed plots of treatment and control groups using locally weighted regression. The weights are applied using a tricube weighting function (Cleveland 1979) with a bandwidth of 1. Figure 6 Notes: Smoothed plots of treatment and control groups using locally weighted regression. The weights are applied using a tricube weighting function (Cleveland 1979) with a bandwidth of 1. where Mom_Charit are indicators for mother f s race or ethnicity, her education, teen motherhood, and whether she smoked during pregnancy t. E-ZPass is an indicator equal to one if the closest toll plaza has implemented E-ZPass; Closeit is an indicator equal to one if the mother lived within 2 km (or 1.5 km) of a toll plaza; and Plazait is a series of indicators for the closest toll plaza. This indicator is designed to capture any unobserved, time-invariant characteristics of each toll plaza sample region. The coefficient of interest is on the interaction between E-ZPassit and Closeit. We also This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 77 78 AMERICAN ECONOMIC JO URNAL: APPLIED ECONOMICS JANUARY 2011 include indicators for the year and month to allow for systematic trends, such as the increase in minority mothers. Finally, we control for linear distance from a busy roadway. Standard errors are clustered at the level of the toll plaza, to allow for cor relations in the errors of mothers around each plaza. If we saw that maternal char acteristics changed in some systematic way following the introduction of E-ZPass, then we would need to take account of this selection when assessing the effects of E-ZPass on health outcomes. We also estimate models of the effects of E-ZPass on housing prices. These mod els are similar to equation (1) except that they control for whether it is a condomin ium, age (in categories, including missing), square footage (in categories, including missing), fixed effects for the municipality, and year and month of sale. We have also estimated models that control for the ratio of assessed structure to land values, with similar results. Our baseline models examining the effects of E-ZPass on the probabilities of low birth weight and prematurity are similar to the models from equation (1). The estimated equation takes the following form: (2) Outcome it = a + bxE-ZPassit + b2Closeit + b3Plazait + b4 E-ZPassit x Closeit + b5Year + b6Month + b7Xit + bsDistanceit + eit, where Outcome is either prematurity or low birth weight; and the vector Xit of mother and child characteristics includes indicators for whether the mother is black or Hispanic; four mother education categories (< 12, high school, some college, and college or more; missing is the left out category); mother age categories (19-24, 25-24, 35 +); an indicator for smoking during pregnancy; indicators for birth order (second, third, or fourth or higher order); an indicator for multiple birth; and an indi cator for male child. Indicators for missing data on each of these variables were also included. Again, the main coefficient of interest is b4 which can be interpreted as the difference-in-differences coefficient comparing births that are closer or further from a toll plaza, before and after adoption of E-ZPass. We perform a series of robustness checks. First, we estimate models that restrict the sample to mothers within 5 km of a toll plaza. Second, we include interactions of Closeit and a linear time trend. It is possible that areas close to toll plazas are gener ally evolving in some way that is different from other areas (e.g., racial composition), but, as we shall see, this does not seem to affect our estimates. Third, we estimated models of the propensity to live close to a toll plaza to see whether mothers were more or less likely to live near a toll plaza before or after E-ZPass adoption. The propensity models are estimated using all of the maternal and child characteristics listed above, the interactions of these variables, as well as zip code fixed effects.4 We then excluded 4 We obtained similar results using models that controlled for county fixed effects instead of zip code fixed effects. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. 1 CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 79 all observations with a propensity less than 0.1 or greater than 0.9 as suggested by Richard K. Crump et al. (2009). We estimated separate models for African Americans and non-African Americans since these groups tend to have very different average birth outcomes. We also looked separately at estimates for nonsmokers. As we show below, our difference-in-difference results are robust to these changes, though we do find larger effects for African Americans and for smokers. The estimates from (2) reflect an average effect of E-ZPass on people anywhere within the 2 km (or 1.5 km) window. We have also experimented with allowing the effect to vary with distance from the toll plaza. To do this requires that some assumption be made about the rate at which the effects decay with distance from the toll plaza. The engineering literature is not particularly helpful in this respect, since most studies focus on areas very close to roadways. As we show below, the esti mates are somewhat sensitive to these assumptions, but are qualitatively consistent with the results from the simple difference-in-difference models. One possible threat to identification is that new mothers with better predicted birth outcomes could select into areas around toll plazas after E-ZPass is adopted. Although we do not find evidence of changes in the average demographic charac teristics of those living near toll plazas after E-ZPass, an arguably better way to control for possible changes in the composition of mothers is to estimate models with mother fixed effects. These models take the following form: (3) Outcomeit ? at + bxE-ZPassit + b2Closeit + b3Plazait + b4 E-ZPassit x Closeit + b5Year + b6Month + b7Zit + bs Distanceit + eit, where at is a fixed effect for each mother /, and Z is a vector including child gen der and birth order, and potentially time varying maternal characteristics including mother's age, education, and an indicator for smoking. Although all the mothers are selected to have had at least one child while residing within 2 km of a toll plaza, we alternatively define the indicator for Close either as less than 2 km from a toll plaza or less than 1.5 km from a toll plaza.5 5 One difficulty with the interpretation of these models is that they are identified primarily from movers (there are few mothers with two or more births, both within 2 km of a toll plaza). This would be a problem if we thought that women systematically moved closer to toll plazas when their circumstances improved, and that improved circumstances led to better birth outcomes. The birth certificates do not record income, but marital status is likely to be correlated with maternal well-being and does change over time. We have estimated placebo models similar to (3) using an indicator for married as the dependent variable, and find a negative coefficient on the interaction of close x E-ZPass which is not statistically significant. This suggests that if anything, women are less likely rather than more likely to be married when they live near toll plazas post E-ZPass so that any bias due to movers probably causes an underestimate of the effects of E-ZPass in the mother fixed effects models. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 80 AMERICAN ECONOMIC JOURNAL: APPUED ECONOMICS JANUARY 2011 Table 2?Testing the Validity of the Research Design: Regressions of Maternal Characteristic on E-ZPass Adoption (Difference-in-Difference Specification) Mother Teen Mother Housing Black Hispanic yrs. ed Dropout mother smoked sale price Panel 1 _(1) (2) (3) (4) (5) (6) (7) <2 km toll x after E-ZPass -0.011 -0.01 0.037 -0.007 -0.001 0.005* 0.149 [0.011] [0.010] [0.040] [0.005] [0.005] [0.003] [0.103] Observations 397,201 406,641 406,198 397,201 412,884 402,590 252,343 Panel 2 <1.5 km toll x after E-ZPass -0.014 -0.01 0.013 -0.003 0.001 0.007** 0.031 [0.055] [0.011] [0.010] [0.006] [0.003] [0.003] [0.106] Observations 397,201 406,641 406,198 397,201 412,884 402,590 252,343 Notes: Each coefficient is from a separate regression. Each coefficient in columns 1-6 is from a regression that also included controls for being within 2 km (or 1.5 km) of a toll plaza, year of birth, month of birth, indicators for each toll plaza, an indicator for post E-ZPass at nearest toll plaza, and distance to highway. Housing sale price regressions in column 7 include year and month of sale, indicators for nearest toll plaza, an indicator for condo units, distance to highway, municipality fixed effects, square footage (in categories including dummies for missing), and age (in categories, including dummies for missing). Standard errors are in brackets. ** Indicates that the estimate is statistically significant at the 95 percent level of confidence. * Indicates significance at the 90 percent level of confidence. IV. Results Table 2 shows the results of estimating equation (1), the effects of E-ZPass on the characteristics of mothers who live near toll plazas and on housing prices. Each coefficient represents an estimate of &4from a separate regression. The only maternal characteristic to show any significant changes with E-ZPass adoption is smoking, where it is estimated that E-ZPass has a positive effect. Note that if more smokers move to areas after E-ZPass adoption (or if mothers smoke more) this will tend to work against finding any net benefit of E-ZPass on birth outcomes. The last col umn shows that there is no immediate significant effect on housing prices (although the coefficient is positive), suggesting that it takes time for any effects through the housing market to be felt. These results suggest that the estimated health effects of E-ZPass are not due to changes in the composition of mothers who live close to toll plazas. Table 3 shows our estimates of (2). Again, each coefficient is an estimate of b4 from a separate regression. The first and third columns show a model that controls only for month and year of birth, toll plaza fixed effects, and distance to highway. These estimates are somewhat higher than the raw difference-in-difference esti mates implied by Table 1, suggesting that it is important to control for time trends and regional differences. The second and fourth columns add maternal characteris tics as in equation (2). Assuming our research design is valid, adding controls for mother's characteristics should only reduce the sampling variance while leaving the coefficient estimates unchanged. The results in columns 2 and 4, are consistent with the validity of the research design, since adding maternal characteristics has little This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. 1 CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 81 Table 3?Regressions of Birth Outcomes on E-ZPass Adoption (Difference-in-Difference Specification) Prematurity Prematurity LBW LBW Panel 1 _(1)_(2) (3) (4) <2km toll x after E-ZPass -0.0085 -0.0086 -0.0094 -0.0093 [0.0039]** [0.0034]** [0.0032]** [0.0028]** R2 0.0044 0.0034 0.0032 0.0028 Panel 2 < 1.5km toll x after E-ZPass -0.0088 -0.0098 -0.0077 -0.0084 [0.0051]* [0.0048]** [0.0035]** [0.0032]* R2 0.0042 0.0048 0.0035 0.0032 Maternal characteristics No Yes No Yes Observations 405,802 405,802 409,673 409,673 Notes: Each coefficient is from a different regression. All regressions also included controls f being within 2 km (or 1.5 km) of a toll plaza, year of birth, month of birth, toll plaza indic tors, an indicator for post E-ZPass, and distance to highway. Maternal characteristics includ mother black, mother Hispanic, mother education ( 1.5 km and before after < 10 km before < 10 km after >10km Toll plaza Panel A. Difference-in-Difference sample Outcomes Premature 0.096 0.096 0.102 0.108 0.085 Low birth weight 0.082 0.08 0.089 0.091 0.078 Controls Mother Hispanic 0.272 0.309 0.176 0.239 0.054 Mother black 0.159 0.174 0.227 0.256 0.047 Mother education 13.25 13.31 13.25 13.23 12.92 Mother HS dropout 0.152 0.152 0.156 0.164 0.173 Mother smoked 0.088 0.078 0.107 0.085 0.152 Teen mother 0.067 0.058 0.082 0.069 0.079 Birth order 1.3 1.37 1.38 1.45 1.68 Multiple birth 0.029 0.034 0.031 0.036 0.033 Child male 0.511 0.518 0.513 0.512 0.512 Distance to roadway 0.976 0.939 1.484 1.459 21 Observations 16,934 14,856 207,728 175,966 185,795 NJ observations 12,980 13,175 141,982 146,948 70,484 PA observations 3,954 1,681 65,746 29,018 115,311 Ever birth Ever birth Never birth Never birth < 1.5 km <1.5 km <1.5km <1.5km E-ZPass plaza E-ZPass plaza E-ZPass plaza E-ZPass plaza before before after after Panel B. Mothers with more than one birth in sample Outcomes Premature 0.0883 0.0988 0.0914 0.103 Low birth weight 0.0803 0.0755 0.0862 0.0857 Controls Mother Hispanic 0.164 0.286 0.0916 0.168 Mother black 0.144 0.156 0.168 0.17 Mother education 12.81 12.54 12.75 13.11 Mother HS dropout 0.163 0.202 0.178 0.164 Mother smoked 0.113 0.0756 0.134 0.0939 Teen mother 0.0414 0.0417 0.07 0.0464 Birth order 1.581 1.723 1.596 1.733 Multiple birth 0.0306 0.0382 0.0331 0.0451 Child male 0.512 0.512 0.512 0.512 Distance to highway 3.612 2.502 5.509 5.159 Total observations 94,473 31,188 1,725,182 512,343 NJ observations 45,215 25,376 718,375 PA observations 49,258 5,812 1,006,807 <1.5 km <1.5 km E-ZPass before E-ZPass after > 1.5 km and <10km E-ZPass before >1.5 km and <10 km E-ZPass after Panel C. Summary statistics for housing sales data (New Jersey only) Sales price 95,033 125,567 95,444 117,600 Assessed land value 45,270 45,462 45,825 45,608 Assessed building value 84,445 87,394 70,219 70,186 Total assessed value 128,899 131,867 114,531 114,363 Year built 1953 1955 1951 1950 Square footage 1,593 1,551 1,639 1,670 Observations 11,586 12,214 116,1 This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 88 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JANUARY 2011 REFERENCES American Housing Survey. 2003. United States Census Bureau, http://www.census.gov/hhes/www/ housing/ahs/ahs03/tab28.htm. (accessed September 12, 2009). Banzhaf, H. Spencer, and Randall P. Walsh. 2008. "Do People Vote with Their Feet? An Empirical Test of Tiebout's Mechanism." American Economic Review, 98(3): 843-63. Beatty, Timothy K. M., and Jay P. Shimshack. 2009. "School Buses, Diesel Emissions, and Respira tory Health." https://greenspace.tulane.edu/jshimsha/buses_healthjune09.pdf. Behrman, Richard E., and Adrienne Stith Butler. 2007. Preterm Birth: Causes, Consequences, and Prevention. Washington, DC: National Academies Press. Brauer, Michael, Gerard Hoek, Patricia van Vliet, Kees Meliefste, Paul Fischer, Ulrike Gehring, Joachim Heinrich, et al. 2003. "Estimating Long-Term Average Particulate Air Pollution Concen trations: Application of Traffic Indicators and Geographic Information Systems." Epidemiology, 14(2): 228-39. Chay, Kenneth Y., and Michael Greenstone. 2005. "Does Air Quality Matter? Evidence from the Hous ing Market." Journal of Political Economy, 113(2): 376-424. Chay, Kenneth Y., and Michael Greenstone. 2003a. "The Impact of Air Pollution on Infant Mortality: Evidence from Geographic Variation in Pollution Shocks Induced by a Recession." Quarterly Jour nal of Economics, 118(3): 1121-67. Chay, Kenneth Y., and Michael Greenstone. 2003b. "Air Quality, Infant Mortality, and the Clean Air Act of 1970." National Bureau of Economic Research Working Paper 10053. Cleveland, William S. 1979. "Robust Locally Weighted Regression and Smoothing Scatterplots." Jour nal of the American Statistical Association, 74(368): 829-36. Crump, Richard K., V. Joseph Hotz, Guido Imbens, Oscar A. Mitnik. 2009. "Dealing with Limited Overlap in Estimation of Average Treatment Effects." Biometrika, 96(1): 187-99. Currie, Janet. 2009. "Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood, and Human Capital Development." Journal of Economic Literature, 47(1): 87-122. Currie, Janet, Matthew Neidell, and Johannes F. Schmieder. 2009. "Air Pollution and Infant Health: Lessons from New Jersey." Journal of Health Economics, 28(3): 688-703. Delaware River Joint Toll Bridge Commission. 2009. "Commission Awards Design-Build Contract for Open Road Tolling at 1-78 Bridge." http://www.drjtbc.org/default.aspx?pageid=1558. (accessed September 30, 2009). Depro, Brooks, and Christopher Timmins. 2008. "Mobility and Environmental Equity: Do Housing Choices Determine Exposure to Air Pollution?" http://www.ncsu.edu/cenrep/workshops/docu ments/DeproTimmins8_7_08CampResources.pdf. Dugandzic, Rose, Linda Dodds, David Stieb, and Marc Smith-Doiron. 2006. "The Association between Low Level Exposures to Ambient Air Pollution and Term Low Birth Weight: A Retrospective Cohort Study." Environmental Health, 5(3). Environmental Protection Agency (EPA). 1993. "Automobiles and Carbon Monoxide." Office of Mobile Sources Fact Sheet OMS-3. January. Ernst, Michelle, James Corless, and Ryan Greene-Roesel. 2003. "Clearing the Air: Public Health Threats from Cars and Heavy Duty Vehicles?Why We Need to Protect Federal Clean Air Laws." Surface Transportation Policy Partnership. Washington, DC. http://www.transact.org/report. asp?id=227. Friedman, Michael S., Kenneth E. Powell, Lori Hutwagner, LeRoy M. Graham, and W. Gerald Teague. 2001. "Impact of Changes in Transportation and Commuting Behaviors during the 1996 Summer Olympic Games in Atlanta on Air Quality and Childhood Asthma." Journal of the American Medi cal Association, 285(7): 897-905. Glinianaia, Svetlana V., Judith Rankin, Ruth Bell, Tanja Pless-Mulloli, and Denise Howel. 2004a. "Par ticulate Air Pollution and Fetal Health: A Systematic Review of the Epidemiologic Evidence." Epi demiology, 15(1): 36-45. Glinianaia, Svetlana V., Judith Rankin, Ruth Bell, Tanja Pless-Mulloli, and Denise Howel. 2004b. "Does Particulate Air Pollution Contribute to Infant Death? A Systematic Review." Environmental Health Perspectives, 112(14): 1365-70. Gunier, Robert B., Andrew Hertz, Julie von Behren, and Peggy Reynolds. 2003. "Traffic Density in California: Socioeconomic and Ethnic Differences Among Potentially Exposed Children." Journal of Exposure Analysis and Environmental Epidemiology, 13(3): 240-46. Hu, Shishan, Scott Fruin, Kathleen Kozawa, Steve Mara, Suzanne E. Paulson, and Arthur Winer. 2009. "A Wide Area of Air Pollutant Impact Downwind of a Freeway During Pre-Sunrise Hours." Atmospheric Environment, 43(16): 2541-49. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms VOL. 3 NO. 1 CURRIEAND WALKER: TRAFFIC CONGESTION AND INFANT HEALTH 89 Huynh, Mary, Tracey J. Woodruff, Jennifer D. Parker, and Kenneth C. Schoendorf. 2006. "Relation ships between Air Pollution and Preterm Birth in California." Paediatric and Perinatal Epidemiol ogy, 20(6): 454-61. Karr, Catherine J., Carole B. Rudra, Kristin A. Miller, Timothy R. Gould, Timothy Larson, Sheela Sathyanarayana, and Jane Q. Koenig. 2009. "Infant Exposure to Fine Paniculate Matter and Traf fic and Risk of Hospitalization for RSV Bronchiolitis in a Region with Lower Ambient Air Pollu tion." Environmental Research, 109(3): 321-27. Knittel, Christopher R., Douglas Miller, and Nicholas J. Sanders. 2009. "Caution, Drivers! Children Present. Traffic, Pollution and Infant Health." http://www.econ.ucdavis.edu/faculty/knittel/papers/ kms_latest.pdf. Lee, Sue J., Shakoor Hajat, Philip J. Steer, and Veronique Filippi. 2008. "A Time-Series Analysis of Any Short-Term Effects of Meteorological and Air Pollution Factors on Preterm Births in London, UK." Environmental Research, 106(2): 185-94. Leem, Jong-Han, Brian M. Kaplan, Youn K. Shim, Hana R. Pohl, Carol A. Gotway, Stevan M. Bul lard, J. Felix Rogers, Melissa M. Smith, and Carolyn A. Tylenda. 2006. "Exposures to Air Pol lutants During Pregnancy and Preterm Delivery." Environmental Health Perspectives, 114(6): 905-10. Lin, Jie, and Dan Yu. 2008. "Traffic-Related Air Quality Assessment for Open Road Tolling Highway Facility." Journal of Environmental Management, 88(4): 962-69. Liu, Shiliang, Daniel Krewski, Yuanli Shi, Yue Chen, and Richard T. Burnett. 2007. "Association between Maternal Exposure to Ambient Air Pollutants During Pregnancy and Fetal Growth Restric tion." Journal of Exposure Science and Environmental Epidemiology, 17(5): 426-32. Mattison, Donald R., Samuel Wilson, Christine Coussens, and Dalia Gilbert, ed. 2003. The Role of Environmental Hazards in Premature Birth: Workshop Summary. Washington DC: National Acad emies Press. New Jersey Department of Environmental Protection. 2005. "NOx and VOCs." In New Jersey's Envi ronmental Trends. Trenton, NJ. http://www.nj.gov/dep/dsr/trends2005/pdfs/nox-voc.pdf. New Jersey Department of Transportation. 1998. "Agreement on E-ZPass Contract Reached," news release, March 11, 1998. http://www.state.nj.us/transportation/about/press/1998/031198.shtm. New Jersey Turnpike Authority. 2001. Operational and Traffic Benefits of E-ZPass Deployment to the New Jersey Turnpike. New Brunswick: Wilbur Smith Associates. Parker, Jennifer D., Pauline Mendola, and Tracey Woodruff. 2008. "Preterm Birth after the Utah Val ley Steel Mill Closure: A Natural Experiment." Epidemiology, 19(6): 820-23. Parker, Jennifer D., Tracey J. Woodruff, Rupa Basu, and Kenneth C. Schoendorf. 2005. "Air Pollution and Birth Weight among Term Infants in California." Pediatrics, 115(1): 121-28. Ponce, Ninez A., Katherine J. Hogatt, Michelle Wilhelm, and Beate Ritz. 2005. "Preterm Birth: The Interaction of Traffic-Related Air Pollution with Economic Hardship in Los Angeles Neighbor hoods." American Journal of Epidemiology, 162(2): 140-48. Pope, C. Arden. 1989. "Respiratory Disease Associated with Community Air Pollution and a Steel Mill, Utah Valley." American Journal of Public Health, 79(5): 623-28. Pope, C. Arden, Joel Schwartz, and Michael R. Ransom. 1992. "Daily Mortality and PM10 Pollution in Utah Valley." Archives of Environmental Health, 47(3): 211-17. Ransom, Michael R., and C. Arden Pope. 1992. "Elementary School Absences and PM10 Pollution in Utah Valley." Environmental Research, 58(2): 204-19. Ritz, Beate, Michelle Wilhelm, Katherine J. Hoggatt, and Jo Kay C. Ghosh. 2007. "Ambient Air Pol lution and Preterm Birth in the Environment and Pregnancy Outcomes Study at the University of California, Los Angeles." American Journal of Epidemiology, 166(9): 1045-52. Ritz, Beate, Michelle Wilhelm, and Yingxu Zhao. 2006. "Air Pollution and Infant Death in Southern California, 1989-2000." Pediatrics, 118(2): 493-502. Saka, Anthony A., Dennis K. Agboh, Simon Ndiritu, and Richard A. Glassco. 2000. An Estimation of Mobile Emissions Reduction from Using Electronic Toll Collection in the Baltimore Metropolitan Area: A Case Study of the Fort McHenry Tunnel Toll Plaza. National Transportation Center. Balti more, March. Salam, Muhammad T., Joshua Millstein, Yu-Fen Li, Frederick W. Lurmann, Helene G. Margolis, and Frank D. Gilliland. 2005. "Birth Outcomes and Prenatal Exposure to Ozone, Carbon Monoxide, and Paniculate Matter: Results from the Children's Health Study." Environmental Health Perspec tives, 113(11): 1638-44. Schlenker, Wolfram, and Michael J. Roberts. 2009. "Nonlinear Temperature Effects Indicate Severe Damages to U.S. Crop Yields under Climate Change." Proceedings of the National Academy of Sci ences, 106(37): 15594-98. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms 90 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JANUARY 2011 Schrank, David, and Tim Lomax. 2005. The 2005 Urban Mobility Report. Texas Transportation Insti tute, Texas A&M University. College Station, TX, May. Schrank, David, and Tim Lomax. 2009. The 2009 Urban Mobility Report. Texas Transportation Insti tute, Texas A&M University. College Station, TX, July. Schwartz, Joel. 2004. "Air Pollution and Children's Health." Pediatrics, 113(4): 1037-43. Slama, Remy, Verena Morgenstern, Josef Cyrys, Anne Zutavern, Olf Herbarth, Heinz-Erich Wich mann, and Joachim Heinrich. 2007. "Traffic-Related Atmospheric Pollutants Levels during Preg nancy and Offspring's Term Birth Weight: A Study Relying on a Land-Use Regression Exposure Model." Environmental Health Perspectives, 115(9): 1283-92. U. S. Department of Transportation. 2005. National Transportation Statistics 2004. Bureau of Trans portation Statistics. Washington, DC: U.S Government Printing Office. Venigalla, Mohan, and Michael Krimmer. 2006. "Impact of Electronic Toll Collection and Electronic Screening on Heavy-Duty Vehicle Emissions." Transportation Research Record: Journal of the Transportation Research Board, 1987: 11-20. Vickrey, William S. 1969. "Congestion Theory and Transport Investment." American Economic Review, 59(2): 251-60. Warren Reporter. 2010. "Motorists Could Encounter Delays during 1-78 Toll Plaza Construction." Janu ary 8. http://www.nj.com/warrenreporter/index.ssf/201 %l/motorists_could_encounter_dela.html. Wilhelm, Michelle, and Beate Ritz. 2003. "Residential Proximity to Traffic and Adverse Birth Out comes in Los Angeles County, California, 1994-1996." Environmental Health Perspectives, 111(2): 207-16. Wilhelm, Michelle, and Beate Ritz. 2005. "Local Variations in CO and Particulate Air Pollution and Adverse Birth Outcomes in Los Angeles County, California, USA." Environmental Health Perspec tives, 113(9): 1212-21. Woodruff, Tracey J., Lyndsey A. Darrow, and Jennifer D. Parker. 2008. "Air Pollution and Postneona tal Infant Mortality in the United States, 1999-2002." Environmental Health Perspectives, 116(1): 110-15. World Health Organization (WHO). 2000. "Carbon monoxide." In WHO Air Quality Guidelines, 2nd ed. Copenhagen: WHO Regional Office for Europe. This content downloaded from 149.125.250.159 on Mon, 05 Nov 2018 02:24:40 UTC All use subject to https://about.jstor.org/terms

Tutor Answer

Tutor_Booth
School: Carnegie Mellon University

Hey! Kindly f...

flag Report DMCA
Review

Anonymous
Totally impressed with results!! :-)

Similar Questions
Hot Questions
Related Tags

Brown University





1271 Tutors

California Institute of Technology




2131 Tutors

Carnegie Mellon University




982 Tutors

Columbia University





1256 Tutors

Dartmouth University





2113 Tutors

Emory University





2279 Tutors

Harvard University





599 Tutors

Massachusetts Institute of Technology



2319 Tutors

New York University





1645 Tutors

Notre Dam University





1911 Tutors

Oklahoma University





2122 Tutors

Pennsylvania State University





932 Tutors

Princeton University





1211 Tutors

Stanford University





983 Tutors

University of California





1282 Tutors

Oxford University





123 Tutors

Yale University





2325 Tutors