The Local Benefits of Federal Mandates:Evidence from the Clean Water Act

Rhiannon L. Jerch∗Johns Hopkins University

Temple University

December 31, 2019For most recent draft, click here.

Abstract

A large component of local government spending is comprised of complying with a variety of federalmandates. However, empirical evidence on how local governments finance these mandates, such aswhether these expenditures crowd out other spending, and whether local residents value mandatedexpenditures above their local costs, is non-existent. This paper estimates how local governmentsfinanced a federal mandate and its impact on growth following passage of the 1972 Clean Water Act(CWA). I leverage the role of river networks in distributing pollutants across cities, combined withpre-1972 state regulatory intensity, to predict pre-CWA compliance with the infrastructure mandate.This paper has three main findings. First, cities financed substantial improvements to local waterquality primarily through an increase in resident fees. Second, mandate compliance did not crowd-outpublic spending on non-mandated items. Last, using housing prices as a metric, I find that residentsvalued the mandated infrastructure above their local costs. I employ a novel hydrological approach toshow that positive spillovers as well as complementarities in pollution abatement across jurisdictionsexplain part of this positive result. These findings imply that mandates can reduce inefficiencies tolocal public goods provision.

JEL Classification: H72, H76, Q51, Q53, Q58, R53Keywords: Local Government Budget, Environment, Revealed Preference, Water Pollution,Environmental Regulation, Local Public Infrastructure

∗Rhiannon Jerch is a postdoctoral researcher at Johns Hopkins University, and will be joining Temple University an AssistantProfessor of Economics in Spring of 2020. Email: rhiannon.jerch@temple.edu; Address: 824 Ritter Hall Annex, 1301 Cecil B. MooreAve, Philadelphia PA 19103. I thank Shanjun Li, Matthew E. Kahn, Nancy Brooks, Michael Lovenheim, and Nick J. Sanders fortheir guidance and support of this project. I also thank the many useful comments from Jorgen Harris, David Keiser, Eli Fenichel,Randy Walsh, Don Fullerton, William Gentry, Bob Inman, Stephen Coate, William Evans, Fernando Ferreira, Mariaflavia Harari,Ben Lockwood, Colin Sullivan, Caitlin Gorback, and participants of Heartland 2019 conference, European UEA 2019 conference,Cornell’s MARG and SEERE seminars and the Wharton School BEPP 900 seminar. I am indebted to Philadelphia and Ithaca civilengineers Drew Brown, Scott Gibson, and Brian Rahm for several discussions on municipal wastewater treatment operations. Lastly,I wish to thank Ted Czado for providing data and documentation on the EPA’s early Clean Watershed Needs Surveys. This projectwas supported by the Lincoln Institute’s C. Lowell Harriss Dissertation Fellowship.

1 Introduction

Federal spending mandates are a controversial component of US fiscal federalism. The bi-partisan

National Conference of State Legislatures recently declared that “The growth of federal mandates and

other costs that the federal government imposes on states and localities is one of the most serious fiscal

issues confronting state and local government officials (NCSL 2018).” Federally mandated programs

implemented at the local level include: surface water pollution control under the Clean Water Act,

drinking water treatment under the Safe Drinking Water Act, No Child Left Behind, the Americans

with Disabilities Act, lead-based paint abatement, and vehicle emissions control under the Clean Air Act,

among others (US Conference of Mayors 1993; Conlan 1994). In recent years, local governments have

allocated over 13% of their annual expenditures to such programs. This figure has tripled since the early

1970’s (Conlan 1994) and is larger than several more salient municipal-level budget categories such as

policing or public works maintenance.1 How have these federal mandates impacted local governments?

Opponents of mandated programs —both at state and local levels—argue that mandates infringe

upon local sovereignty, inhibit the ability of cities to tailor their spending to preferences of local taxpay-

ers, and place substantial cost burdens on local governments.2 These fiscal effects can have important

ramifications for the desirability of a municipality and its ability to attract residents (Gyourko & Tracy

1991). If local living costs increase without compensating improvements to amenities or wages, revealed

preference theory predicts that individuals will reduce their demand for that locality in favor of another

(Rosen 1974; Roback 1982; Albouy 2008).

Even though mandates present a fiscal burden to municipalities, local governments may not

necessarily be worse off after complying with mandates. Rather, the provision of mandated public goods

may improve the welfare of municipalities if the marginal social benefits of the mandated expenditure

outweigh its costs. In weighing the impacts of mandates on local governments, it is important to

understand not only how mandates impact local budgets, but whether goods or services provided under

a mandate are valued by local residents above their costs. This paper seeks to answer these questions.

There are several reasons why local governments may undersupply public goods relative to a

locally efficient level: large fixed costs, economies of scale, and credit constraints may prevent munici-

palities from investing in valued infrastructure projects (Haughwout 2002; Fisher 2015). Additionally,

spatial spillovers may lead to under provision of public goods. Local spending on beaches or river-1Based on author’s calculations using aggregate local cost needs reported in Conlan (1994) Tab. 2-2 and mean municipalannual expenditures sourced from US Census Bureau (2015). Conlan (1994) p.13 also cites national survey studies thatfind mandate compliance costs comprise between 11 and 12% of locally raised revenues.

2As recently as July 2018, members of Congress drafted a bill to limit federal mandates, citing its adverse impacts on localbusinesses and government budgets (Kasperowicz 2018). In 2017, the state of New York passed mandate-relief legislationin an effort to alleviate fiscal burdens on New York school districts (Seward 2017).

1

front paths will have low benefits if those same beaches and riverfronts suffer from upstream pollution.

Mandated programs may address these inefficiencies in at least two ways. First, several of the largest

mandated programs including, the Clean Water Act and the Safe Drinking Water Act, are accompanied

by federal grants and subsidies (Conlan 1994). These subsidies may relax credit constraints and enable

local infrastructure investment. Second, the national scale and uniformity of these programs may di-

rectly increase the returns to local spending by inducing coordination across local governments (Cooper

& John 1988; Fisher 2015). This is a noteworthy insight with water pollution abatement —the focus of

the 1972 Clean Water Act.

In this paper, I use the 1972 Clean Water Act (hereinafter “CWA”) requirements on local wastew-

ater infrastructure to test whether municipal growth is enhanced or depressed by compliance with federal

mandates. Prior work provides evidence that local public goods provision, including public works in-

frastructure provision (Haughwout 2002; Albouy & Farahani 2017), is an important determinant of

residential location decisions.3 Yet, whether the same is true of federally mandated public goods provi-

sion is not well understood, despite the growing prominence of federal mandates as a share of municipal

budgets. Recent work by Keiser & Shapiro (2018) finds the grants program associated with the CWA

generated little local benefits. However, the regulatory effect of the CWA —that is, the mandate effect

—may differ substantially from the grant effect because the set of grantee cities does not fully encompass

the set of regulated cities. Thus, the CWA mandate effect on municipal outcomes remains an heretofore

unknown parameter.

This paper contributes new findings that the CWA infrastructure mandate improved local growth

outcomes, as measured by changes to population and housing prices. I use variation in the hydrologic

flows of US river systems to show that existence of complementarities in pollution abatement across

jurisdictions is at least part of the reason for this result. I interpret this finding as consistent with

recent work by Albouy et al. (2018), which demonstrates that accounting for complementarities across

public goods can “unlock” their respective benefits, and Owens et al. (2019), which theorizes that in

the presence of spatial externalities, coordination across local governments and developers is integral

for optimal urban development.

I also provide an empirical examination, possibly the first, of how local governments finance

compliance with mandates on infrastructure. A collection of survey studies (Lake et al. 1979; EPA

1988; Conlan 1994; National League of Cities 2017) find that mandates displace local funding of existing3For some examples on a large literature concerned with how public goods improvements attract residents: Chay &Greenstone (2005), Bayer et al. (2009), and Lin (2018) study improvements to air quality; Black (1999) and Celliniet al. (2010) study public school quality; while Baum-Snow & Kahn (2000) and Duranton & Turner (2012) exploreimprovements to transit access.

2

public goods. However these studies are mainly descriptive in nature and do not consider counterfactual

fiscal outcomes without mandate compliance. In contrast to these prior studies, I exploit a natural

experiment that effectively randomizes the CWA mandate across cities. I then compare outcomes across

cities for which the CWA mandate was binding to counterfactual cities for which the CWA mandate

was nonbinding. I do not find evidence that cities displaced provision of other goods and services in

order to fund compliance with the CWA mandate. Rather, I find strong evidence that municipalities

financed local mandate costs through wastewater user fees. In other words, municipalities responded

to the mandate by increasing their overall budgets, rather than through austerity measures. These

findings highlight the potential regressivity of federal mandates as a form of fiscal federalism: while a

fee-for-service financing scheme may be less distortionary relative to displacing funding from existing

public goods or services, piped water is an inelastically demanded good.4 Consequently, lower income

households likely shouldered a greater share of local CWA compliance costs relative to higher income

households.

Work by Baicker (Baicker 2001; Baicker & Gordon 2006) provides the closet parallel to my

study. They find significant evidence of crowd-out in welfare payments at the state level as a result of

mandated Medicaid expansions. A likely reason my findings of no crowd-out at the local level contrast

from Baicker’s state level analysis is that wastewater treatment has few substitutes. A city cannot

readily reduce surface water pollution through means other than by adopting treatment technology.

Medicaid spending, in contrast, may be a substitute service for welfare payments, since both contribute

to health and quality-of-life among lower-income populations.5

More broadly, my finding that cities did not respond to the CWA mandate through austerity

measures contrasts with existing work on local budgetary responses following shocks to local tax revenues

(Lutz 2008; Skidmore & Scorsone 2011; Lutz et al. 2011; Alm et al. 2011; Feler & Senses 2017;

Cromwell et al. 2015; Melnik 2017; Shoag et al. 2019). This apparent asymmetry may, however,

reflect the “flypaper effect,” in which governments respond to taxpayer wealth shocks differently than

to proportional increases in expenditure obligations (Hines & Thaler 1995). The key takeaway is that

federal mandates that can be financed with a fee-for-service like water provision are unlikely to displace

funding of other local goods and services. In other words, the current focus within congressional mandate

reform on local fiscal burdens may be overly narrow.

Using variation across cities in pre-CWA infrastructure adoption, I provide new estimates of the4Per Foster & Beattie (1979) Tab. 2, the price elasticity of demand for water among US consumers is inelastic, at -0.1.5Other empirical work on federal mandates (unrelated to the CWA) mainly rely on isolated case studies (Hanford & Sokolow1987 and Weiland 1998) or focus on education outcomes following No Child Left Behind (Imazeki & Reschovsky 2004;Reback et al. 2014; Deming et al. 2016) without consideration for local government budgetary responses to mandatecompliance.

3

effect of federal mandate compliance on local budgets and growth. The 1972 CWA aimed to improve the

environmental health of US rivers and lakes by requiring that municipal governments operating a public

sewerage system treat their wastewater with a minimum level of pollution abatement. Cities with ex ante

noncompliant wastewater treatment were under regulatory pressure to invest in more rigorous abatement

technology following the 1972 legislation. My empirical approach compares differences before relative

to after 1972 in finances and growth across cities bound to the CWA’s infrastructure requirements

relative to cities that had adopted compliant infrastructure before the CWA was enacted into federal

law. To carry this out, I obtained data from the Environmental Protection Agency (EPA) on the census

of municipal wastewater treatment plants. Importantly, I obtained the earliest of these EPA surveys

which predates enforcement of the CWA regulations. These data provide crucial baseline information

that allow me to identify the mandate’s effects by comparing municipal outcomes before versus after

federal enforcement of the mandate. To the best of my knowledge, this paper is the first to categorize

cities according to their compliance status with the legal terms of the 1972 CWA, one of the most

ambitious and expensive environmental regulations in US history.

The primary challenge for my empirical analysis is that CWA mandate compliance and growth

outcomes are likely to be endogenous. Specifically, local governments differ widely with respect to

the preferences of their taxpayers and their ability to provide certain public goods, including wastew-

ater treatment infrastructure. Such underlying differences across cities in their pre-policy provision of

wastewater treatment infrastructure are likely deterministic of differences across these cities in their

local fiscal conditions and ability to attract taxpayers.

To solve this endogeneity problem, I construct an instrument that predicts wastewater treatment

adoption using city-level variation in riparian exposure to downstream populations and state-level vari-

ation in pre-policy water pollution regulation. The intuition behind my identification strategy is that

cities with historically large population centers downstream were more likely to be pressured by their

downstream neighbors to adopt stringent wastewater treatment, long before the CWA became legis-

lation. Furthermore, this inter-jurisdictional pressure was more likely to be enforced for cities within

states with more regulation of surface water pollution. Such early-adopter cities were unaffected by the

CWA infrastructure standard when the law passed in 1972 because they had already adopted secondary

treatment technology. By leveraging variation in infrastructure adoption driven by forces external to

the city, this instrument provides variation in ex ante CWA compliance that is plausibly exogenous to

local spending decisions or growth. Balance tests and event study analyses show that my instrument

captures a subset of cities that are similar in observed pre-1972 characteristics and trends, thus allowing

a better counterfactual than naıve comparisons across ex ante compliance status.

4

My instrumented difference-in-differences estimates show that cities did not displace provision of

other goods and services in order to fund compliance with the CWA mandate. While the infrastructure

requirements caused local governments to more than double their expenditures on wastewater from 6%

prior to the Act to over 16% of their total budgets, cities primarily funded these compulsory expenditures

through federal grants and by doubling fees on residents. After the CWA, the average household paid

over $200 per year on wastewater use fees, from a base of $90 per year. I find that the mandated

expenditures on wastewater treatment led to economically and statistically significant improvements in

water quality in the twenty years following the CWA. Cities under the burden of compliance with the

CWA experienced a 14-20% improvement to surface water quality, on average, as measured by dissolved

oxygen concentration.

As a consequence of mandate compliance, populations grew 9-16% faster and house prices 14-15%

faster on average in the twenty years following the CWA, suggesting that the mandated infrastructure

was at least valued at its marginal cost to local residents. These positive growth effects manifest mainly

in small communities with populations less than 10,000. I explicitly model the hydrological connections

across cities to show that spillover effects from upstream abatement efforts account for 5% of local

house price improvements and that abatement efforts may be complementary across municipalities. I

interpret these findings as evidence that federal mandates can correct for inefficiencies to local public

goods provision in the presence of market failures, and the benefits of these interventions are particularly

high for smaller local governments. Back-of-envelope calculations suggest that the benefits from the

CWA mandate, as implied by my hedonic estimates, are between 41% and 72% of the regulation’s

cost. This range is slightly higher than most prior estimates on the CWA,6 but are suggestive that the

benefits of the program are smaller than their costs.

Additional robustness analyses bolster my main findings about the local cost incidence on user

fees as opposed to other budget line items, and positive growth responses from mandate compliance.

These include: (1) cities predicted to be treatment and control based on my instrument show little

baseline differences in population or fiscal growth. Further industrial composition, water quality, and

population levels trend in parallel across the predicted treatment and control cities at least twenty years

leading up to 1972; (2) the robustness of my results to instrumental variation strictly stemming from

east-west rivers as well as inland cities; and (3) the significant and positive impact of upstream adoption

on local water quality, but lack of impact on local water quality stemming from adoption among cities

that are located on separate river systems.

The remainder of my paper proceeds as follows. Section 2 provides institutional background6Keiser & Shapiro (2018) find a ratio of 0.25. Keiser & Shapiro (2019) review prior studies on the CWA and find a meanbenefit-to-cost ratio of 0.57 and a median ratio of 0.37.

5

on the CWA regulations and determinants of wastewater treatment adoption. Section 3 describes the

data. Section 4 presents the naive difference-in-differences approach as well as the instrumental variable

approach I use to identify the effects of the CWA infrastructure mandate. Section 5 discusses my results

and provides several robustness checks. In Section 6, I identify spillover effects from the CWA and use

the estimates to conduct a back-of-envelope benefit-to-cost analysis of the CWA technology standard.

Finally, Section 7 concludes.

2 Regulation of Surface Water Pollution

In this section, I discuss the costs and benefits to municipalities of adopting wastewater treatment in

absence of a federal regulation. I then provide relevant institutional details on the 1972 CWA and its

providence in light of historic provision of urban wastewater treatment. This provides context for my

identification strategy, discussed in greater detail in Section 4.

2.1 Benefits & Costs of Wastewater Treatment

Wastewater treatment facilities protect environmental and public health by treating sewage, urban

debris, and pathogens from piped waters before they return to rivers and lakes. By removing oxygen-

consuming organic matter that damages aquatic ecosystems, wastewater treatment helps the environ-

ment as well as the aesthetic and recreational use value of surface waters.7 Prior work has shown that

consumers value improvements to recreational fishing, swimming, boating, and surface water clarity

(Bockstael et al. 1987; Boyle et al. 1999; Lipton 2004; Olmstead & Kuwayama 2015); and that water-

front (Leggett & Bockstael 2000) as well as non-waterfront property values (Poor et al. 2007; Walsh

et al. 2011) increase following local surface water pollution control. The latter of these findings suggests

that the benefits to local water quality extend beyond properties immediately adjacent to the affected

water body.8 Consequently, improvements to surface water following wastewater treatment have the

potential to increase surrounding property values or population levels if individuals value the water

quality amenity more than its marginal cost.

Yet, constructing and maintaining wastewater treatment facilities is costly and requires signif-

icant public financial investment. US municipalities allocate 10% of their total annual expenditures

toward sewerage and wastewater treatment (US Census Bureau 2015). Costs vary considerably with

the type and rigor of wastewater treatment technology. Primary treatment is a basic treatment process7Wastewater treatment also increases the supply of potable water and can help to prevent disease by removing harmfulbacteria and chemicals. In industrialized economies, however, health benefits from surface water pollution control arelikely to be dwarfed by recreational and ecosystem benefits because basic drinking water treatment methods are ubiquitousand have a long history, predating most federal environmental regulations (Olmstead 2010).

8Prior valuation studies show that people are willing to travel to access surface waters. In a 1977 study conducted by theDepartment of Interior, surveyed participants traveled over 126 miles on average to access 43 rivers and lakes throughoutthe US (Smith & Desvousges 1986).

6

that utilizes physical methods (gravity, settling tanks, or centrifuges) to separate waste from water.

Secondary treatment is a more advanced technology that uses biological processes to decompose the

organic matter in waste that can both spread disease and absorb oxygen in water. Secondary treatment

removes more than twice as much oxygen demand from wastewater as primary treatment and is thus

more effective at protecting aquatic life and reducing bacterial counts in surface water (Stoddard et al.

2003). However, secondary treatment is considerably more expensive to install and operate, ranging

between 2 and 10 times the cost of primary treatment (EPA 1976). Fig. 3 plots the engineering costs

required for secondary treatment technology based on a plant’s service population. For a city of 30,000

people to upgrade from primary to secondary, costs were roughly $6 million in 2012 dollars, equal to

the annual public safety operating budget of a similar-sized city.9

2.2 The 1972 Clean Water Act

The large investment costs and potentially diffuse benefits of secondary wastewater treatment con-

tributed to the need for federal regulation of surface water pollution. Prior to the CWA, over three-

quarters of municipal wastewater systems used the primitive, less expensive treatment technology of

primary treatment (see Fig. 1). The CWA addressed this low take-up of rigorous wastewater treatment

by establishing secondary treatment as the minimum technology standard for all wastewater re-entering

surface waters. For this reason, the Act constitutes a “technology-forcing statute” because of its focus

on pollution abatement technology (Copeland 1999).10 By directly removing bacteria and thus lowering

oxygen demand in wastewater effluent, secondary treatment provided the means to achieve the CWA’s

ultimate goal of making all US surface waters “fishable and swimmable.”

Congress enforced the secondary technology standard through a new monitoring system.11 Lo-

cal governments, firms, or individuals dumping untreated wastewater into surface waters through any

discrete conveyance could be fined up to $25,000 per day, sanctioned, sued, or imprisoned by the federal

government (Copeland 1999).12 The CWA also recognized the authority of citizens to bring civil suits

against their local governments for violating CWA standards (Andreen 2013).13

To assist in paying for these large infrastructure costs, the federal government distributed con-9Per Guo et al. (2014), the average wastewater flow per capita is 100 gallons per day. Public safety cost estimatecalculated by averaging Census of Governments data on annual police and fire expenditures for cities with populationsbetween 20,000 and 40,000.

10The original 1972 Act included requirements under Section 303(d) for states to monitor surface water pollution levelsand abate when pollution levels exceed state limits. However, the EPA did not begin enforcing the “control” approachof the CWA until 1992 (Copeland 2012).

11The monitoring system is called the National Pollutant Discharge Elimination System, and it requires any discreteconveyances of wastewater into surface waters, such as through pipes, sewers, conduits, ditches, or animal feedingoperations to obtain a permit (Andreen 2013).

12For examples of recent CWA enforcement, see Fields & Emshwiller (2011) or Westerling (2011).13Earnhart (2004) analyzes the impact of different regulatory instruments (permits, inspections, and enforcement actions)

on water pollution abatement among municipal plants in Kansas in the mid 1990s.

7

struction grants to state and municipal governments. Approximately two-thirds of existing wastewater

treatment plants received at least some funding (Keiser & Shapiro 2018, hereinafter “K&S”). At their

peak, these grants were intended to support up to 75% of total capital costs. However, by 1981, Congress

changed the matching rate to 55%, and by 1987 the grant program was phased out.

While a substantial portion of these construction costs for CWA compliance were supported by

federal grants, there are several reasons to expect that the unfunded costs placed a significant burden

on local budgets. First, municipalities were obligated to fund at least 25% of their capital needs for

secondary treatment, equivalent to approximately 4% of the average noncompliant city’s budget prior to

the CWA. Second, operating costs, which on average are 60% of total annual wastewater treatment costs

(US Census Bureau 2015), were not eligible for grant assistance and operations costs are likely to increase

with secondary treatment.14 Lastly, while no study has comprehensively tested how compliance with

the CWA technology standard impacted local finances, early case studies found that some communities

were unable to provide the required finances for infrastructure adoption without burdening taxpayers

or displacing other services (EPA 1973; Lake et al. 1979; GAO 1980; Hanford & Sokolow 1987; US

Conference of Mayors 1993).

Figs. 1 and 2 show aggregate, national-level effects of the CWA technology standard. Fig.

1 shows that between 1972 and 1977, the number of treatment plants with noncompliant “primary”

technology fell by 50%, and steadily declined thereafter. The contemporaneous changes to municipal

budgets are apparent in Fig. 2; while the share of total spending in wastewater trended with overall

spending in years 1967 and 1972, there is a divergence after the CWA in which the share of wastewater

spending increased by 3 percentage points on average, despite a downward trend in total spending per

capita. An important goal of this paper is to causally estimate the magnitude of this policy-induced

expenditure change.

The 1972 CWA marked a major shift in the nation’s approach to surface water regulation. Prior

to the CWA, state governments had de facto autonomy over their surface water regulations. In contrast,

the 1972 CWA gave the federal government substantial power to respond directly to violations of the

Act through administrative actions, civil actions, and criminal sanctions (Andreen 2013).15 The prior

federal approach to surface water regulation was more passive, relying on voluntary subsidies to local14Secondary treatment requires more energy input relative to primary treatment to operate aeration pumps and added

personnel to monitor electrical and mechanical processes. Also, monitoring of the secondary treatment biological digestionprocess often requires skilled labor from environmental and civil engineers, unlike primary treatment operations (Brown2018).

15Of the six major federal Acts between 1948 and 1970 related to surface waters that preceded the 1972 CWA, all maintainedthat regulatory authority was mainly in the hands of the states (Fairfax & Hamilton 2000). Two Acts in 1966 and 1970had attempted to impose wastewater treatment standards, but these actions were legally challenged by the States, andwere never enforced (Stoddard et al. 2003).

8

governments in order to promote national programs (Dilger 2013). Additionally, disputes between the

executive and legislative branch nearly handicapped the enforcement of the Act.

The unique providence of the CWA legislation is important for this paper’s empirical design. I

leverage the unanticipated nature of the CWA to compare changes in outcomes after the Act across

ex ante compliant and noncompliant cities. Identification of causal estimates under this approach

would be invalidated if cities had been able to anticipate the CWA regulations or if broader changes to

wastewater treatment were taking hold prior to the CWA. However, for cities to have anticipated the

CWA regulations, they would need to have foreseen a substantial deviation from historical precedent

on state rights to self-regulate and strong collective action on the part of the legislative and judicial

branches to counteract a presidential veto.

2.3 History of Wastewater Treatment

Although secondary treatment is costly, as of 1972 over 20% of municipal plants had secondary treat-

ment already installed. What motivated these municipalities to adopt prior to federal enforcement under

the 1972 CWA? Litigious downstream neighbors suffering from pollution externalities were one poten-

tial mechanism inducing cities to adopt secondary treatment technologies. My identification strategy

exploits this determinant of infrastructure adoption to predict which cities were constrained to comply

with the CWA and which cities were not because they had already adopted compliant treatment.

In the early development of modern water infrastructure, methods of wastewater treatment

were mainly for aesthetic as opposed to direct health benefits. Urban water systems were designed

such that drinking water intakes were upstream of wastewater outfall locations and protected from

wastewater contaminants (Okun 1996). Primary treatment could provide a localized improvement in

surface water quality by reducing accumulation of solid materials, debris, and pervasive odors, but

wastewater treatment was not considered necessary for cleaning a city’s drinking water supply (Tarr

2016).16 Any health benefits from sewage treatment were perceived by early city planners to accrue

mainly to downstream neighbors (Metcalf & Eddy 1922).

Urban population growth during the twentieth century combined with improved understanding of

riparian biochemical cycles in the scientific community brought about water pollution disputes between

upper and lower riparian cities. Development of the metric “biochemical oxygen demand” provided a

method to directly link the existence of organic wastes in wastewater to bacterial and oxygen levels

in natural waters (Fairfax & Hamilton 2000; Melosi 2000), enabling downstream cities to pinpoint16In reference to primary treatment development, Stoddard et al. (2003) states: “In many cases, this construction was

promoted by city officials and entrepreneurs, who were rapidly learning that unsightly urban debris and a delightfulgrowing phenomenon, tourists with leisure dollars to spend, did not mix.”

9

sources of pollution in their own drinking and surface waters. Consequently, pressure from downstream

cities could induce upstream polluters to adopt secondary treatment (Melosi 2000).17 Various historic

anecdotes illustrate this pattern. For example, the city of Chicago—which shares the Mississippi basin

with St. Louis—invested in secondary treatment beginning in 1916, only after the state of Missouri

enacted a lawsuit against Chicago in 1901 for polluting the drinking water of St. Louis (Missouri 1901;

Cain 2005; Stoddard et al. 2003). Similarly, Melosi (2000) discusses how city adoption of wastewater

treatment methods were “born amid the unhealthy background of injunctions and court orders” between

cities.18 Several state and federal court cases at the turn of the century set the precedent for individual

protections against water pollution. These cases provided the legal framework for individuals and local

governments to prevent both public and private entities from polluting surface waters.19

In summary, a city’s riparian exposure to downstream populations may have a positive impact on

the likelihood of ex ante compliance with the CWA technology standard. Prior to the 1972 CWA, cities

demanding aesthetic improvements to nearby surface waters generally adopted the low-cost primary

technology. More costly secondary treatment adoption, on the other hand, could follow litigation

disputes from downstream neighbors suffering from pollution externalities. Cities facing little pressure

from downstream were, in contrast, less likely to adopt the more costly treatment. My identification

strategy, discussed in greater detail in Section 4.1, estimates the causal effect of the CWA using cities

that were induced to adopt (or not) as a function of their exposure to downstream populations.

3 Data

Ninety percent of publicly owned wastewater treatment plants in the US are financed, operated, and

managed by local governments.20 Consequently, I consider the local government, including cities, vil-

lages, boroughs, towns, and townships as my primary unit of observation.21 Throughout this paper,

I refer to “local governments,” “municipalities,” and “cities” interchangeably. Local governments in17The first recorded biological treatment of wastewater occurred in Medford, MA in 1887, however secondary treatment

technologies advanced between 1901 and 1916 to include trickling filters, Imhoff tanks, and activated sludge (RMQAA2015).

18Melosi (2000) also states: “Conflict between upstream and downstream cities over the dumping of sewage and industrialwaste had been fought in the courts and addressed through interstate sanitation compacts.”

19Example cases include: Storley v. Armour & Co., 107 F.2d 499 (1939), Sammons v. City of Gloversville, 67 N.E. 622(NY 1903); Butler v. White Plains, 69 N.Y.S. 193 (1901); Gould v. City of Rochester, 105 N.Y. 46 (1887). McQuillin(1912) provides a thorough law review of municipal responsibilities for water pollution control. These cases set theprecedent that “a city has no right to gather its sewage and cast it into a stream so as to injure the lower proprietor”and further, the “power of a municipal corporation to construct sewers or to use a natural stream as a sewer does notauthorize it to so construct the sewers or to use the stream as to create a nuisance to the damage of a lower riparianowner. (p. 3051)”

20Roughly 11% of the 22,500 publicly-owned plants in the US are managed by counties, states, or other non-municipalauthorities such as universities, national parks, or correction facilities (EPA Clean Watershed Needs Survey, 1973-2004).

21Local governments are defined by the US Census Bureau as political entities authorized by state constitution to providegovernment for a specific population in a defined area. The geographic boundaries of local governments are endogenouslydetermined and can vary over time. I standardize the political jurisdiction of local governments over time through FIPSplace codes rather than geographic boundaries because jurisdictional borders may respond to fiscal shocks.

10

the US are small. Eighty percent of all municipalities have populations less than 20,000 and a mean

population of about 5,000 people (see Appendix Fig. A1). The analysis in this paper, therefore, is dis-

tinct from much prior literature concerned with urban sorting responses because I estimate the average

treatment effect of a representative US municipality as opposed to metropolitan urban centers.

I construct a dataset of municipal CWA compliance using records obtained from the Clean Wa-

tershed Needs Survey (CWNS) Team of the EPA. The CWNS is a census of approximately 22,500

publicly-owned wastewater treatment facilities. These data provide detailed facility-level information

including unique facility identifier codes, treatment technology characteristics, operating status, and

identifying information on the facility’s managing authority including name, county, state, and govern-

ment type. The CWNS surveys began in 1972 and have since been administered roughly once every

two years.22

Plant-specific treatment technology variables from the 1972 survey provide the crucial infor-

mation I use to observe compliance status of a city’s wastewater treatment plant before the CWA

regulations came into effect. While the EPA has not maintained code books associated with the 1972

computer-readable survey file, I was able to identify the survey’s compliance status information using

a copy of the original 1972 survey questionnaire, found in the appendix of EPA (1973). I define a plant

as ex ante noncompliant if its effluent discharge is recorded as not meeting secondary treatment levels

at the time of the survey. A plant is ex ante compliant if its effluent discharge is recorded as meeting

secondary or more stringent treatment levels. To the best of my knowledge, this is the first study to

codify this early CWNS survey and to categorize municipalities according to their ex ante compliance

status.

While I use only the 1972 survey to designate municipal treatment compliance status, I utilize the

full survey panel to construct a sample suitable for analyzing the CWA technology standard. Appendix

B provides additional detail on my sample restrictions.

As of 1972, over three-quarters of publicly owned wastewater treatment plants lacked the sec-

ondary treatment technology mandated by the CWA (Fig. 1). Between 1975 and 1977, the number of

plants with only primary treatment fell by over 55%, from roughly 10,000 plants to under 4,500, and

declined steadily thereafter. Compliance with the CWA’s technology standard had a substantial time

lag. Part of this lag is mechanical. Primary to secondary treatment upgrades require several years for

engineers to execute planning and construction, as well as for the municipality to secure financing and

apply for federal aid. Another part of this lag is political. The CWA passed amid substantial backlash22Data reported in each CWNS report are representative of the prior calendar year. For example, the first available CWNS

survey titled the “1973 Clean Watershed Needs Survey” describes plant technology as of 1972.

11

from the executive branch. After a Congressional override on Nixon’s initial veto of the Act, Nixon

impounded half of the funding Congress had appropriated for plant construction costs. It was not until

after 1975, when the Supreme Court ruled against presidential power to impound funds (Train 1975),

that appropriations for the CWA ramped up and several cities received construction grants (Copeland

2015). These lags justify assigning 1972 as a pre-policy year even though this was the year that the Act

became law.

To gauge the impact of the CWA technology standard on municipal finances, I use the US

Census Bureau’s “Historical Finances of Individual Governments” database. These data provide detailed

information on annual revenues, expenditures, and debt for the census of local governments every five

years, starting from 1967. I merge the municipal finance data with the CWNS plant technology data

based on the name, state, county, and government type (i.e., “city”, “village”, “township”, or “borough”)

of the plant’s managing authority. To ensure the accuracy of this merge, I exclude plants managed by

counties, districts, universities, or corrections facilities. Additionally, I exclude cities with non-unique

name–government type combinations within their county. Under this criteria, I am able to match 3,593

municipalities to the Census finance data from approximately 4,000 municipalities in the CWNS data.

Because the Census finance data are self-reported, they may be prone to measurement error. I exclude

municipalities reporting zero total expenditures or property taxes, which eliminates approximately 6%

of the municipalities. Lastly, I restrict the sample to cities that appear in each decade of the “Historical

Finances of Individual Governments” database, which eliminates approximately 5% of municipalities.

These data restrictions yield a sample of 2,975 cities.

Data on municipal growth outcomes, including population, median housing prices, and education

levels are sourced from the Decennial Census. The IPUMS National Historical Geographic Information

System provides these data at the relevant FIPS place and county subdivision levels from 1970. IPUMS

also provides shapefiles, which I use to calculate the centroid of each FIPS place and county subdivision

in GIS (Schroeder et al. 2019). Information on local labor markets and industrial composition are

sourced from County Business Patterns, available for years 1956 and 1964, and annually from 1965. I

gauge pre-CWA support for environmental issues among state senators using the League of Conservation

Voters score card from 1971 and 1972, and state-level variation in municipal balanced budget rules as of

1970 from Bohn & Inman (1996). I obtain distance from counties to major waterbodies using data from

Rappaport & Sachs (2003). Lastly, I source water quality data back to 1962 from the EPA STORET

Legacy database, as well as the National Water Information System Water Quality Portal. These data

provide water quality readings from over 740,000 monitoring locations across the US as far back as the

1920s, although monitoring frequency is limited prior to 1970. I calculate ambient water quality as

12

the annual average dissolved oxygen level within 25 miles of a city centroid, where monitor readings

are inversely weighted by their distance from the city centroid.23 I focus on dissolved oxygen as my

preferred measure of water quality because it is directly impacted by secondary treatment and because

it provides a holistic measure of aquatic ecosystem health. Appendix D provides further discussion on

dissolved oxygen and its relevance as a measure of water quality.

Tab. 1 compares descriptive statistics across ex ante compliant and noncompliant cities (i.e.,

compliant or not with the CWA technology standard of secondary treatment prior to 1972). Voluntary

secondary treatment is correlated with both ability to pay for, and propensity to benefit from, water

pollution abatement. Secondary treatment adoption is positively correlated with wealth (e.g., share

of population with a college degree, median housing prices, and revenues per capita), preference for

environmental protection (e.g., conservation score), higher levels of manufacturing industry, receipt of

intergovernmental funding, and proximity to waterbodies. Expenditures appear overall balanced, with

the noticeable exception of wastewater expenditures: cites that already adopted secondary treatment

spent nearly double per capita on wastewater treatment prior to the CWA. Fig. 4 shows the spatial

distribution of ex ante compliance aggregated to the county-level for exposition purposes. Ex ante com-

pliant cities are more likely to be near large lakes, population centers, or manufacturing-intensive areas,

such as Tennessee, Michigan, Pennsylvania, and New York. These substantial differences underscore

the importance of using an empirical approach that eliminates potential confounding factors correlated

with outcome differences across compliant and noncompliant cities.

4 Empirical Strategy

The goal of my analysis is to estimate the impact of the CWA infrastructure mandate on local govern-

ment budgets and growth. The relationship of interest is:

yirt = β(Pi × POSTt) + Xiθt + (γr × t) + τt + νi + εirt (1)

where yirt is one of several outcomes related to municipal expenditures (e.g., wastewater expenditures)

or growth (e.g., population) for city i in geographic region r in year t.24 Pi is an indicator variable equal

to 1 if a city is ex ante noncompliant, meaning it had only primary treatment technology at the start

of the CWA in 1972, and 0 if a city is ex ante compliant, meaning it had at least secondary treatment

as of 1972. POSTt is an indicator equal to 1 for all post-CWA years (e.g., years 1977 and later because

my panel structure is quinquennial).

As ex ante compliant and noncompliant cities exhibit substantial differences in observable char-23I focus on a distance of 25 miles following Keiser & Shapiro (2018). This is likely a conservative delineation, as Smith &

Desvousges 1986 found people traveled over 126 miles on average to recreate on rivers and lakes throughout the US.24I follow the Bureau of Economic Analysis (BEA) definition to categorize states into one of eight US regions: New

England, Mideast, Great Lakes, Plains, Southeast, Southwest, Rocky Mountain, and Far West.

13

acteristics (see Tab. 1), I include a vector of pre–CWA city characteristics, Xi, whose effects are allowed

to vary by year (Lechner 2011). The vector Xi includes income per capita, share of employment in all

manufacturing as well as water-polluting manufacturing all of which are measured at the county-level.

These control for changes in public spending or growth that are driven by local industry or labor de-

mand trends. The control vector also includes a city’s overall river population size, watershed, and

distance to coast, which collectively account for differential growth patterns stemming from riparian

connections to urban markets and ports and changes in demand for coastal relative to inland cities

over time (Bleakley & Lin 2012; Rappaport & Sachs 2003). Xi lastly includes time-varying trends in

baseline receipt of intergovernmental grants, which accounts for differences across cities in their fiscal

managerial capabilities to obtain federal or state funding. The year fixed effects τt control for macroe-

conomic time-varying determinants of yirt common to all cities, such as federal budget cycles, while νi

captures all unobserved, time-invariant differences across cities that affect yirt such as distance from a

water body or soil and landscape attributes that affect infrastructure construction costs. (γr × t) is a

vector of region-specific linear time trends, which ensures estimated differences in yirt across cities are

not driven by divergent patterns of growth across US regions. Lastly, εirt is an error term. I cluster

standard errors at the city level to account for city-specific correlations in unobserved components of

spending and growth over time. The coefficient of interest is β, which measures the differential change,

conditional on controls, in outcome yirt between noncompliant and compliant cities before versus after

the CWA.

For the difference-in-differences estimate of the CWA mandate effect to be unbiased, (Pi×POSTt)

must be uncorrelated with the error term εirt. That is, potential outcomes yirt would have trended

similarly for ex ante compliant and noncompliant cities in absence of the CWA technology standard.

There are several reasons why this assumption is problematic in this context. First, the architecture

of the CWA legislation included not only enforcement of a technology standard, but also substantial

federal financial assistance in the form of capital construction grants. Congress distributed over $153

million in outlays from 1973 through 1986 for the construction and modification of municipal wastewater

treatment plants (Copeland 2015).25 These construction grants provided funding not only for ex ante

noncompliant cities to upgrade from primary to secondary, but also for ex ante compliant cities to

invest beyond the minimum requirements in advanced treatment technologies. Tab. 1 suggests that

compliant cities are historically more capable of obtaining intergovernmental funding, therefore the

availability of CWA federal grants is likely to have impacted ex ante compliant cities differently than

ex ante noncompliant cities. This means that simply controlling for baseline differences in grant receipt

will not remove bias induced by the contemporaneous federal grants shock. In particular, difference-in-25See Keiser & Shapiro (2018) for an empirical analysis on the cost-effectiveness of the CWA grants program.

14

differences estimates of β are likely to be attenuated toward zero for fiscal and water quality outcomes

because control cities may respond to the federal grants program by further investing in pollution

abatement technology.

Second, the early 1970’s witnessed several major environmental regulations, including the Safe

Drinking Water Act in 1974 and the Clean Air Act in 1970, the latter of which affected the economic

development of regulated counties through its impacts on industry, labor markets, and local amenities

(Greenstone 2002; Kahn 2001; Lin 2018). If violations across these Acts were correlated within cities,

deviations from trend that appear after the CWA may be spurious. Lastly, differences in baseline wealth,

surface water quality, and environmental preferences shown in Tab. 1 suggest that ex ante compliant

cities may be on a more positive growth path and, therefore, may not provide a valid counterfactual

to the ex ante noncompliant cities. If ex ante compliant cities were more competitive in attracting

taxpayers, difference-in-differences estimates of β will be biased away from zero toward a negative

growth effect.

To address these various sources of bias, I employ an instrumental variable approach that uses

variation in downstream population across cities combined with variation in pre-CWA water pollution

abatement across states to predict CWA compliance status. The key to my identification strategy is

that I exploit variation that is external to the city; I predict CWA compliance from factors that are

unlikely to be correlated with local taxpayer preferences for public goods or local abilities to obtain

federal grants.

4.1 Instrumental Variable Approach

The positive relationship between downstream population size as well as state environmental regulation

and pre-CWA secondary treatment adoption (as described in Section 2.3) forms the basis of my identi-

fication strategy. Specifically, cities situated upstream of population centers were more likely to adopt

secondary treatment technology prior to the CWA regulations relative to low-downstream-population

cities. Further, cities located in states with stronger regulation of surface waters prior to the CWA were

more likely to adopt. The CWA secondary technology standard was, therefore, more likely to bind for

cities with smaller populations downstream and weaker state regulation of water pollution.

I construct the downstream population component of the instrument using digital spatial maps

sourced from the National Hydrography Dataset Plus of the US Geological Survey (USGS). These

maps contain hydrologic information for over 2.6 million stream segments averaging 1 kilometer in

length. Every river segment possesses identifying attributes that allow me to identify upstream versus

downstream relationships across cities located on the same major river (e.g., the Mississippi) as well

15

as across cities on differing tributaries sharing the same major river basin (e.g., the Illinois and Ohio

rivers, which both feed into the Mississippi).

I assign each city centroid to its closest stream segment using GIS software. My criteria for

matching cities to a stream segment is to select the six closest stream segments to a city centroid and

assign the city to the stream segment with the lowest branching level. This approach accounts for

the tendency of cities to divert wastewater effluent into the main river segment closest to their city as

opposed to a small tributary. I then calculate each city’s cumulative downstream population through

a recursive algorithm that treats river branching points as “river mouths” and again cumulatively

sums city populations until another branch point occurs. Appendix C provides further details on the

downstream population calculation.

The final result is a downstream population value for every river segment. This recursive algo-

rithm provides important precision in my measure of downstream population by explicitly accounting

for tributary branching within river networks. A naıve reliance on distance to river mouth across cities

without accounting for branching would induce substantial measurement error into the downstream

population calculation.

Fig. 5 shows downstream populations, aggregated as county means for exposition purposes.

Cities with a higher downstream population are generally located near the headwaters of populous river

networks, such as the upper Missouri, the upper Mississippi, and the upper Ohio rivers. Importantly,

downstream population is not only a function of river length. For example, cities along the Columbia

river in Washington state have high downstream populations despite a shorter river length. Cities along

the similarly-sized Colorado River in Arizona, in contrast, have relatively low downstream populations

owing to the relatively low population density in the American southwest.

The second component of the instrument exploits pre-existing differences in water pollution

regulation across states to predict municipal ex ante compliance. My instrument includes the share of

wastewater treatment plants within a state that had secondary treatment technology prior to the CWA.

Fig. 6 shows variation across states in pre-CWA municipal secondary treatment adoption. States with

historically more environmental legislation, such as Pennsylvania and New York, had higher levels of

secondary treatment prior to the CWA.26,27

26As of 1970, twenty-nine state constitutions included standards on water quality (Andreen 2013), though enforcement ofthese regulations was generally infrequent (Rechtschaffen 2003). Pennsylvania began regulating water quality in 1937with the Clean Streams Law, and several requirements of the CWA were enacted in Pennsylvania prior to 1968 (Walters2017). New York state laws on watershed regulations date back to the early 1900s (Hudson River Watershed Alliance2015.)

27The pre-CWA state composition of treatment plants can be equivalently interpreted as the predicted CWA technologyincidence in the state. In their identification strategy, Duflo & Pande (2007) employ a similar approach by predictingdistrict-level dam adoption in India using a state’s baseline share of all national dams.

16

My identification strategy thus exploits two sources of variation to predict ex ante compliance:

pre-CWA downstream population size and pre-CWA state composition of compliant wastewater treat-

ment plant technology. In my empirical approach, these fixed characteristics predict differences in

outcomes across cities as a function of their differential effects prior to versus after the CWA. The first

and second stage estimation equations are shown below in Eqs. 2 and 3. Let i index cities and t index

years. Pi is an indicator equal to 1 if a city is ex ante noncompliant and yist is one of several outcomes

of interest (e.g., wastewater expenditures, population, etc.).

yist = βIV

∧

(Pi × POSTt) + Xiθt + (γr × t) + τt + νi + εist (2)

Pi × POSTt =α1(Di × Ss × POSTt) + α2(Ss × POSTt) + α3(Di × POSTt) (3)

+ Xiλt + (γr × t) + τt + νi + µist

In Eq. 3, the first three terms are the excluded instruments, where Di is a city’s downstream population

as of 1970 and Ss is pre-CWA state technology composition, measured as the share of all wastewater

treatment plants in state s with secondary treatment as of 1972. As before, Xi is a vector of pre-CWA

determinants of yist whose effects are allowed to vary with time, including intergovernmental grants

per capita, industry mix, distance to coast, watershed, and river population size to account for time

varying effects of historic exposure to urban markets via river-based trade routes (Bleakley & Lin 2012,

Rappaport & Sachs 2003). This river population control is important because this ensures that ex

ante compliance is predicted from a city’s relative positioning on a river system, and not by the overall

population size of their river. Standard errors are clustered at the city level.

The remaining variation that powers my instrument is the population size downstream, state

secondary treatment share, and its interaction with population downstream. Fig. 7 shows the variation

I exploit. For a given downstream population, there is a range of state technology composition. Similarly,

states with similar technology composition harbor cities with varying degrees of downstream population

sizes. The interaction of these two terms is important because the marginal impact of downstream

population varies with pre-existing, state-level regulations of water pollution. Specifically, cities located

in states with more regulation of water pollution were more likely to adopt wastewater treatment as

a result of downstream population pressure relative to cities located in states with less regulation of

water pollution. The state compliant plant share term (Ss ×POSTt) enters into the instrument to take

advantage of variation in pre-existing state environmental regulations.

I motivate the reduced form variation driving my identification in Fig. 8. Each panel plots

mean wastewater spending per capita relative to 1972 separately for cities with high relative to low

downstream populations (Panel A) as well as high relative to low state compliance share (Panel B). For

17

example, Panel A plots the linear combination of estimates δt×50 + δt in black and δt in gray from the

following equation:

yit =∑

t

δt×50(I50 × Ss ×Dt) +∑

t

δt(Ss ×Dt) + (NR ×Dt) + νi + εit (4)

where I50 is an indicator for a city having downstream population size in the bottom 50th percentile and

NR is river population. An estimate of δt×50 > 0 indicates higher expenditures in year t relative to 1972

for cities in the bottom 50th percentile of the downstream population distribution relative to cities in

the top 50th percentile. Prior to the CWA, cities with high and low downstream populations had similar

wastewater expenditures per capita, suggesting that cities with differing downstream population sizes

had similar potential outcomes in absence of the CWA. However, their expenditures per capita diverge

beginning after 1972. Cities with lower downstream populations incur larger wastewater expenditures

per capita following the CWA relative to cities with downstream populations in the top 50th percentile.28

Panel B similarly shows how cities with 1972 state compliance share in the bottom 50th percentile of the

distribution incur larger wastewater expenditures per capita after the CWA relative to cities in states

with higher compliance shares. In both panels, cities with historically less pressure to abate, either from

downstream neighbors or state regulators, ramp up their wastewater spending comparatively more in the

post-period, suggesting that the 1972 CWA mandate was more likely to bind for these low-downstream

population and low-state compliance cities.

In Tab. 2, I formally examine the first stage relationship between a city’s downstream population,

state plant composition, and CWA compliance status. These results confirm the importance of water

pollution externalities: a city with higher likelihood of inflicting pollution on its neighbors was more

likely to have adopted secondary treatment prior to 1972. Columns (1) and (2) use cross-sectional

variation. A one standard deviation increase in downstream population reduces the likelihood that a

city had only primary treatment as of 1972 by between 7 and 9% from a mean compliance rate of 75%.

These results are robust to employing within-river variation, as well as including controls for distance

to river mouth, which supports that the relative position of a city to other cities along a river provides

the relevant variation in compliance, as opposed to regionally-determined differences.

Columns (3) through (7) show estimates of the first stage (Eq. 3) where downstream population is

interacted with state pre-CWA share of secondary treatment plants. A one standard deviation increase

in downstream population reduces the likelihood that a city had only primary treatment as of 1972

by between 4.2 and 8.8 percentage points, depending on the specification, from a mean noncompliance

rate of 75%. Similarly, a one standard deviation increase in baseline state compliance share reduces the28Appendix Fig. A2 shows the same figure after including year fixed effects in Eq. 4. Consequently, δt×50 + δt and δt

capture the difference across low relative to high downstream population cities conditional on being located in a highcompliance state. The same pattern holds; however, the coefficients capture the effect of low versus high downstreampopulation relative to all cities in low compliant states.

18

likelihood that a city is ex ante noncompliant by between 9.9 and 11.6 percentage points. The negative

sign on the interaction of both instruments in the first row demonstrate that the marginal impact

of downstream population for ex ante compliance was stronger in states with more water pollution

abatement. Moving from column (3) to column (7), the interaction term becomes less significant, and

the level downstream population term becomes more significant with additional controls that limit

variation to be within-state. Since most rivers flow across state borders, adding time varying trends

in baseline state debt rules or state-by-year fixed effects absorbs this cross-state variation. Because

the instruments become substantially weaker in columns (6) and (7),29 I use column (5) as my main

specification for Eq. 3 throughout this paper. Results are qualitatively similar using either specification

(6) or (7) for Eq. 3, however the standard errors become larger after introducing state-by-year fixed

effects.30 In all specifications, I fail to reject that the instruments are uncorrelated with the error εist.

Tab. 2 shows that my instruments are sufficiently strong, even after controlling for time-varying trends

of geography, industry, and wealth differences across cities.

4.2 Validity of Instrumental Variable Approach

A causal interpretation of the instrumented difference-in-differences parameter, βIV , requires two as-

sumptions (Hudson et al. 2015). First, the evolution of outcomes across cities with different pre-existing

state compliance shares should have trended similarly; and —conditional on state compliance—the evo-

lution of outcomes across cities with high- versus low- downstream populations should have trended

similarly absent the CWA technology mandate. Second, the exclusion restriction requires that shocks

which co-vary with the CWA do not differentially impact cities with high- versus low-downstream pop-

ulation sizes, or cities in high- versus low- compliance states. In other words, the exclusion restriction

requires that downstream population size and state compliance share explain post-CWA differences in

outcomes only through their influence on pre-CWA wastewater treatment plant technology adoption.

I assess the plausibility of these assumptions by testing for the presence of pre-trends of city

characteristics across municipalities exposed to above median versus below median exposure to the in-

strument. My ability to assess pre-trends is somewhat limited because most of my sources for municipal-

level city characteristics extend back only a few periods prior to the CWA. The Census of Governments,

for example, only reports its quinquennial local spending data back to 1967, while decennial Census

data on house prices or skill composition extends back only to 1970 at the municipal level. Nonetheless,

this allows me the ability to explore evidence of differential growth across treated and control cities in29The Kleibergen-Paap first stage F-statistic (which is robust to non-i.i.d. errors) falls well below the Stock & Yogo (2005)

critical values for 10% maximum relative bias after employing time-varying state controls in Tab. 2 column (4).30Appendix A3 and Tab. 5 column (5) show results with state balanced budget rules x year fixed effects. Appendices A2

and A9, column (1) show results with state-by-year fixed effects.

19

one period prior to the CWA, 1967 to 1972, for several variables of interest.

I test whether growth in city characteristics differs significantly across treated versus control cities

in the five years leading up to the CWA. Tab. 3 results suggest that potential outcomes in absence of

the CWA were more likely to trend in parallel under the instrumental variable approach compared to

a basic difference-in-differences approach. Columns (1) and (2) show apparent differences in pre-CWA

growth trends across observed treated versus control cities prior to the CWA for population, wastewater

expenditures, user fees, and receipt of federal grants. In column (4), pre-CWA growth trends in city

characteristics are more comparable across treated and control cities on the basis of the instrument.

The fact that the instrumental variable approach invites comparable pre-trends in federal grant receipt

is particularly important; this reduces concern that the CWA grant program will have differentially

impacted cities predicted to be control relative to those predicted to be treated.

Historic data on municipal-level population, county-level industrial composition, and water qual-

ity are available further back in time. I test the assumption of common potential outcomes among

these three variables by estimating a dynamic effect specification that allows for visual examination of

pre-trends in Fig. 9. Each plot shows a flexible version of Eq. 2, where the impact of noncompliance is

allowed to vary in each year:

yirt =′02∑

t=′67(3′72)δt

(Pi ×Dt) + Xiθt + (γr × t) + τt + νi + εirt. (5)

The coefficient δt measures the difference, conditional on controls, in outcome yirt between noncompliant

and compliant cities in year t relative to 1972. Fig. 9 plots estimates of δt. The bars show 95% confidence

intervals and the dashed line denotes the start of the CWA. An estimate of δt > 0 indicates higher values

of yirt in noncompliant cities relative to compliant cities in year t relative to 1972. The left column

(Panels A, C and E) utilizes the specification in column (5) of Tab. 2 for the first stage, whereas

the right column (Panels B, D, and F) uses the specification in column (6) of Tab. 2, which includes

time-varying affects of 1972 state debt rules.

Visual examination of pre-trends in Fig. 9 suggests that cities predicted to be ex ante compliant

versus cities predicted to be ex ante noncompliant on the basis of the instrument experienced similar

growth patterns leading up to the CWA. Panel A shows some indication of pre-trends in population

growth leading up to 1972, however the rate of growth appears to accelerate following 1972. Controlling

for trends in state debt rules in Panel B removes most of the population pre-trend. After including these

controls, most of the identifying variation is driven by within-state differences in downstream population,

consequently the predictive power of the instrument is substantially weaker. This is why my preferred

specification throughout this paper allows for cross-state comparisons in downstream population size,

20

and includes only the controls listed in Tab. 2 column (5).31

In Fig. 9, employment in manufacturing and water quality do not appear to grow significantly

differently for ex-ante compliant relative to noncompliant cities in the years leading up to the CWA. In

Panel D, manufacturing actually appears to be in decline in counties with noncompliant cities, and then

recovers following the CWA. These results should be interpreted with caution, however, as they rely

on municipal-level variation to explain county-level outcomes. Water quality, as measured by dissolved

oxygen levels, exhibits parallel trends since 1957, with a significant trend break following 1972. The

water quality results are discussed in more detail in the following section. Each of these plots suggests

that treated and control cities had similar growth trends leading up to the 1972 CWA, but experienced

a break in trend following the mandate.

The local average treatment effect (LATE) identified by my instrument is the effect of the CWA

infrastructure mandate among cities with low riparian exposure to downstream populations and low

levels of pre-CWA state regulation. The CWA was binding for these cities because they faced little

pressure from downstream to abate. The comparison group of cities are those that adopted compliant

infrastructure prior to the CWA as a consequence of downstream pressure. Recall that ex ante compliant

cities could apply for and use federal infrastructure grants under the CWA to further upgrade their

secondary treatment plants. The correlated shock of the CWA grants program that may have impacted

both the treatment and control cities under the difference-in-differences approach is less likely to manifest

with the IV approach. My instrument exploits variation across cities in the degree of external pressure

they faced to abate surface water pollution. Consequently, the cities induced by the instrument to be ex

ante compliant are less likely, relative to the self-selected ex ante compliant cities, to undertake federal

infrastructure grants and inframarginally consume additional wastewater treatment. My IV estimates

do not capture the effect of the CWA infrastructure mandate due to alternative reasons that would

cause the policy to bind, such as insufficient finances or local taxpayer indifference to environmental

protection. Consequently, the LATE identified by my instrument may differ from the impact experienced

by the average city bound by the infrastructure mandate.

5 Results

My empirical tasks are threefold: first, I identify the magnitude of direct compliance costs by examining

changes in wastewater expenditures. Second, I test how cities financed those direct costs by estimating

differences in expenditures of non-wastewater public goods and municipal revenue sources. Finally, I

estimate the indirect, non-pecuniary impacts of compliance by testing how water quality, population,31Appendices A3 and A9 column (2) show that my results are qualitatively similar after controlling for trends in state

debt rules. Inclusion of state-by-year fixed effects in Appendices A2 and A9 column (1) however, generally reduce thepredictive power of the instrument and increase standard errors substantially.

21

and housing prices change differently among noncompliant cities relative to compliant cities. I present

results from both the difference-in-differences approach and the instrumental variable approach.

5.1 Local Government Budgets

Panel A of Tab. 4 provides difference-in-differences estimates of Eq. 1; and Part B, instrumental

variable estimates of Eq 2.32 Wastewater expenditures increased substantially more from pre-CWA

levels relative to all other expenditure categories. The difference-in-differences results in Panel A show

minor increases in general and administrative spending, suggesting potential crowd-in effects of the

CWA mandate. Revenue-side responses to the increased wastewater expenditures are comprised of

federal grant and user fees. Overall tax revenues did not significantly change, as decreases in property

taxes offset increases in sales & license tax revenues.

The IV estimates in Panel B show results with substantially larger magnitudes. Attenuation of

the difference-in-differences estimates is consistent with uptake of CWA federal grants and increased

expenditures on wastewater treatment among the control group. Ex ante compliant cities may be dif-

ferentially impacted by the grants program because of their pre-existing advantage in obtaining federal

funding. Thus, the IV estimates provide a more accurate estimate of the fiscal costs required to comply

with CWA infrastructure mandate because the LATE group of cities are more comparable with respect

to pre-existing wealth and intergovernmental funding, as shown in Tab. 3. Wastewater expenditures

increased by $155 per capita, or over 200 percentage points after the CWA. Both wastewater capital

and operating costs increased. The estimated increase to total wastewater expenditures closely aligns

with engineering cost estimates from the EPA CWNS. These surveys show that additional expendi-

tures required for secondary treatment adoption for a city of 30,000—the mean population size in my

sample—is approximately $200 per capita per year, which is approximately 129% the aggregate cost

increase implied by my per capita expenditure estimate of $155 per year.

Importantly, Panel B shows no evidence of crowd-out in the funding of other goods. In other

words, cities did not respond to the federal mandate through austerity measures. Rather, total city

expenditures increased by approximately 32%, driven mainly by increased wastewater expenditures.

Positive contemporaneous increases in public safety and welfare suggest that the CWA mandate actually

crowded-in other public spending, however, as shown in Appendices A2 through A6, these crowd-in32Appendix Tab. A1 show the sensitivity of wastewater expenditures and growth outcomes to various specifications.

Inclusion of city-level fixed effects tends to increase the CWA effect substantially, suggesting that unobserved fixeddifferences across cities are highly correlated with expenditure decisions, water quality, and growth outcomes. In column(3), region time trends and county-level income time trends increase the effect size for wastewater and house prices, butdecrease the effect sizes for population and dissolved oxygen. K&S find a steady improvement in national water qualitystarting from the 1960’s, thus it is not surprising that a linear trend reduces the measured effect of the CWA mandate onwater dissolved oxygen. The point estimates generally increase in magnitude with the city-specific controls, suggestingtime-varying effects of baseline differences are predictive of spending and growth outcomes decades later even amongsmaller local governments.

22

results are less robust after adding state-level controls relative to the wastewater expenditure result.

To meet these increased expenditures, cities relied on receipts of intergovernmental grants and

increased user fees. The estimated increase in wastewater user fees of $38.5 per capita aligns remarkably

well with the increase in wastewater operating costs. Debt issuance does not significantly change,

indicating local governments relied mainly on federal grants for the capital-portion of their infrastructure

investments. Some subsequent robustness checks show short term debt actually falls significantly.

Since user fees are indexed directly to voluntary consumption of the public good, user fees are

an efficient way to fund public goods. However, the $39 per capita (or, $117 per household) annual

increase could place a nontrivial burden on local taxpayers, particularly those with lower incomes that

cannot easily substitute away from consuming water. The fact that federal grant receipts per capita

increased more than wastewater expenditures per capita is suggestive that the CWA grants program

may have crowded in additional federal funds. That is, it is possible that cities responded to the CWA

federal grants program by not only applying for treatment plant construction grants but for other

federal grant programs as well. Case studies of 16 communities, collectively, in Hanford & Sokolow

(1987) and Weiland (1998), found that municipal CWA compliance served to improve their financial

positions and organizational skills for acquiring intergovernmental grants. The estimated magnitudes

seem plausible. Total federal outlays implied by my estimate in Panel B of $173 per capita per year is

equal to approximately 74% of the $205.4 billion in congressional funds actually distributed to all forms

of government from 1972 through 1992 for secondary treatment adoption.33

In summary, Tab. 4 demonstrates that wastewater expenditures per capita tripled after the CWA,

with little change in expenditures on other goods and services. Lack of displaced funding contrasts with

prior work that largely finds local governments reduce spending on goods and services in response to

fiscal shocks. However, the asymmetric response to expenditure liabilities relative to tax revenue loss

may be additional evidence of the “flypaper effect” (Hines & Thaler 1995), whereby governments act

as though money is not fungible and respond to taxpayer wealth shocks differently than proportional

shocks to local fiscal obligations.

5.2 Water Quality & Local Government Growth

Do these budgetary changes, in turn, lead to improved water quality or municipal growth? To answer

this question, I employ dependent variables drawn from the EPA STORET database and the census to

measure city growth through pre- versus post- CWA changes in water quality, population, and housing33Estimated aggregate federal grants increase from CWA calculated as $173 times the total population of my sample as

of 1972 (or roughly 55 million people) times 20 years. Information on cumulative funds awarded from 1972-1992 forsecondary treatment sourced from the EPA Grants Information and Control System (GICS) database, adjusted to 2012USD.

23

prices. Unlike the municipal finance data which I observe once every five years from 1967, Census

outcomes of population and housing prices are observed only once per decade. Consequently, results in

Tab. 5 Panel B and C are estimated from a more limited panel where observations from 1972 provide

pre–CWA outcomes, and observations from 1982 and 1992 provide post–CWA outcomes. Difference-in-

differences estimates in column (1) show that the noncompliant cities experienced no significant change

in water quality, and declines relative to compliant cities in both measures of growth. Compared to

ex ante compliant cities, ex ante noncompliant cities grew 2% slower with respect to median housing

prices and 3% slower with respect to population.

Once I correct for the correlated shock and negative selection biases systemic to the OLS es-

timates, the instrumental variable results in Tab. 5, column (2) suggest a more positive effect of the

CWA mandate for city growth. Cities predicted to be noncompliant as a function of their downstream

population and pre-CWA state adoption share experienced 1 mg/l (or 14%) improvements to ambient

water quality, 11% higher housing price growth and approximately 20% greater changes to population

relative to control cities in the twenty years following the CWA. In columns (3) through (5) I include

controls for the time-varying effects of proximate (within a 50 mile radius) and upstream population

concentrations. These controls account for ways in which the CWA mandate effect on water quality

and subsequent resorting behavior may depend upon the population density of areas surrounding and

upstream of city i. Such general equilibrium and spillover feedback effects are more likely to affect

hedonic and sorting outcomes relative to the direct budgetary outcomes estimated in Tab. 4, thus they

are included here.34 These additional controls do not substantially change the instrumental variable es-

timates, although the water quality result attenuates and the population response increases slightly. In

columns (4) and (5), I allow growth responses to vary across terciles of municipal population. Notably,

the house price and population responses manifest mainly for smaller municipalities with populations

less than 10,000 people. House prices increased as a result of the mandate between 9 and 16% among

small communities, but did not change significantly for larger cities. Population grew between 14.1 and

15.8% across all sizes of municipalities, though the effect is comparatively weaker for the largest cities.

Lastly, column (5) includes time varying effects of 1970 state balanced budget rules to account for the

slight pre-trends in population evident in Panel A of Fig. 9. After netting out this cross-state variation,

the mandate effect on water quality increases considerably with little effect on the growth outcomes.

These estimated growth effects are large, but not inconsistent with prior literature. Column (5)

suggests that water quality improved over 20% in the twenty years following the CWA mandate. K&S

find that the average infrastructure grant improved dissolved oxygen deficit by 7% in the year following34Section 6 as well as Appendix C explains how I calculate the number of municipalities and populations upstream of cityi using spatial data of the US river network.

24

grant receipt and improvements grew in magnitude over time. They also find positive dose-response

effects to additional grants. Since the average municipality received at least three infrastructure grants

(K&S), my city-level estimate is within range of their estimate. Further, the 14-15% increase in treated

city populations following a 20% improvement to water quality is consistent with prior literature that

tests for migration effects from environmental regulation. Banzhaf & Walsh (2008) find population

loss effects ranging from 8-12% over a decade among communities that gain exposure to TRI chemical

emissions, Gamper-Rabindran & Timmins (2011) find an 18% increase over 10 years in population

density among census tracts within three kilometers of Superfund site remediation, while Kahn (2000)

estimates a population increase of 8% over 15 years among California counties that experienced an

average improvement to ozone exposure of 21%. Between 1967 and 1992, ex ante compliant cities grew

in population by an average of 77% whereas ex ante noncompliant cities grew an average of 48%. My

results imply that CWA compliance decreased the growth gap between compliant and noncompliant

cities by approximately one third.

My hedonic estimates are substantially larger than that of K&S, who find house prices increased

0.25% within a 25 mile radius of wastewater treatment plants that received construction grants. The

average hedonic estimate of 10.6% shown in column (3) suggests that CWA mandate compliance ac-

counted for roughly 9% of the total change in housing stock value among treated cities from 1967

through 1992.35 While the treatment effect of interest in this paper is not directly comparable to that

of K&S, I provide several potential reasons for the significant difference in this paper’s estimated he-

donic effect size. First, K&S measure the effect of a grant on house prices whereas the present paper

measures the effect of the CWA infrastructure mandate. This is an important distinction because the

mapping of ex ante noncompliance status and “grant receipt” were not necessarily one-for-one; not all

ex ante noncompliant cities received grants, while some ex ante compliant cities did receive grants.36

Consequently, their estimation is more likely to assign an ex ante compliant city as “treated” by the

grants program, thus attenuating their hedonic effect toward a null result. Second, because I employ

an instrumental variable approach, my treatment effect is relevant for the subset of “complier” cities

who are shifted toward the treatment or control group only as a consequence of my instrument. As

discussed in Section 4.2, this “LATE” group may be more fiscally successful and therefore may benefit

more from the CWA mandate than the average city treated by the CWA technology standard. Third,

the house price sample in K&S, sourced from Geolytics, is more representative of large urban areas

relative to my sample from the Census of Governments.37 My estimated hedonic effect for the largest35Source: US Census house prices and housing unit counts among ex ante noncompliant cities in 1967 and 1992.36The Construction Grants Program allotted funding to wastewater treatment plants for both secondary treatment, as

required by the 1972 Act, and treatment more stringent than secondary (i.e., tertiary) (Copeland 2016)37K&S rely on the balanced panel of census tracts appearing in each decade from 1970 through 2000 in the Geolytics

25

tercile of municipalities (Tab. 5, columns 4 and 5) is statistically indistinguishable from K&S’ main

hedonic result.38

The growth results collectively suggest that mandated treatment infrastructure significantly im-

proved ambient water quality between 14 and 20%. Despite a two-fold increase in resident user fees to

over $200 per year, this amenity improvement capitalized positively into housing values and attracted

residents, particularly among smaller cities.

5.3 Robustness Checks

The validity of the instrumented difference-in-differences design requires that outcomes across cities

with high versus low exposures to the instrument would trend in parallel absent the CWA technology

standard. One concern suggested by Fig. 5 and 6 is that, both, downstream population size and state

compliance is regionally determined. Northern cities, for example, have higher downstream populations

on average relative to southern cities because several major US rivers begin in the North and terminate

along coastlines in the South. Additionally, coastal cities generally have lower downstream populations

relative to interior cities. The regional correlations generated by my instrument may be problematic

because coastal cities are unlikely to be comparable controls to interior cities,39 and likewise northern

cities are unlikely to be comparable controls to southern cities. Appendix Tabs. A5, A6, and A9

show that my instrumental variable results are qualitatively similar after excluding coastal cities, and

excluding hydrologic regions with the largest downstream populations (the upper Mississippi and the

Ohio river watersheds). This suggests that my findings are not driven strictly by divergent growth

trends across coastal relative to interior cities, or by northern relative to southern cities.

I lastly show that results are robust to using the full, unbalanced panel of municipal wastewater

treatment plants, including plants that were built after the CWA in Appendix Tabs. A7 and A9 column

(5). First stage results with the full sample are shown in Appendix Tab. A8. This final robustness

check provides qualitatively similar results to that of my main balanced sample. However, significant

increases in other goods expenditures, population and housing price growth due to CWA noncompliance

are suggestive that the unbalanced panel of cities likely includes those that adopt wastewater treatment

Neighborhood Change Database. According to Geolytics, “The 1970 and 1980 censuses did not have full tract coverageof the US. The tracts are predominantly located in urban areas (Census 1994).” While their paper does not providedescriptive statistics on census tracts, it is likely their set of sample tracts are those in urban areas of the US, and arenot representative of the average municipality in the Census of Governments.

38To further place my hedonic estimates in the context of prior literature on large-scale changes to environmental goods,Gamper-Rabindran & Timmins (2013) find house prices within a 3 mile radius of Superfund sites increase 18-24% inthe decade following site remediation. Currie et al. (2015) find house prices decline 10-11% after a toxic plant openingamong houses within half a mile from a plant.

39A small number of coastal cities face different treatment technology standards from interior cities. Known as 301h WavierRecipient facilities, some treatment plants that discharge into coastal waters are exempt from secondary treatmentrequirements. As of 1994, only nineteen facilities in the lower 48 states had this exemption (EPA 1994). Most arelocated in Maine. I remove such facilities as a robustness check in Appendix Tab. A5 and find consistent results.

26

due to unobserved local demand shocks, and not purely due to the binding constraints of the mandate.

This result supports my selection criterion used for my main results, whereby only cities that have had

an operating wastewater treatment plant since 1972 are included in the sample.

5.4 Limitations & Interpretation

I consider some of the potential limitations associated with these data and my experimental design.

These potential limitations will generally tend to bias my results toward a null effect. First, I am

comparing treated cities to control cities, whereby the treated cities may receive an exogenous im-

provement to surface water quality after complying with mandated investment in wastewater treatment

infrastructure. However, municipal boundaries may not provide the correct spatial extent of pollution

abatement effects from wastewater treatment. If the infrastructure’s actual impact is more narrow than

municipal borders, any perceived benefits of wastewater pollution measured at the this level will be

diluted. Conversely, if the actual impacts of surface water pollution extend beyond municipal borders

- to downstream cities, for example - the mandated infrastructure will have some impact on control

cities, again diluting the identified differential across these communities.

Second, multiple municipalities may share a single wastewater treatment plant, particularly if

those municipalities are located close together. I can identify the municipality that manages a publicly

owned plant from the CWNS data, but I cannot distinguish whether other municipalities are serviced

by that plant. This will not compromise the diagnosis of treatment versus control cities in my design, as

I consider only municipalities that are, themselves, the managing authority of a plant. However, to the

extent that there is cost sharing of mandate compliance across unobserved communities, the estimated

differential on expenditure changes will be diluted. In Appendix Fig. A3, I compare the plant service

population of each plant reported in CWNS to the Census population estimate for the plant’s managing

municipality and find a correlation very close to unity. This suggests that mis-measurement of the per

capita compliance costs borne by municipal residents is likely to be minimal.

Finally, the interpretation of my hedonic estimates on property values are best interpreted as a

capitalization effect of mandated wastewater treatment infrastructure rather than the exact willingness-

to-pay. This is because, first, the impacts on water quality following the CWA are non-marginal and

national in scope and, second, my identification strategy exploits a long panel. Over this time period,

preferences among treated city residents likely changed as treated cities attracted individuals with higher

educational attainment.40 When hedonic analysis is used to estimate “large” changes in public goods,40In Appendix Tab. A10 I test for compositional changes in skill following CWA compliance. Aggregate number of

individuals with at least a college degree increased in smaller municipalities (Panel A columns (4) and (5), but theshare of the population with a college degree did not change significantly (Panel B), suggesting the CWA did not causegentrification or displacement of lower skill individuals. This is in contrast to prior work by Sieg et al. (2004) andGamper-Rabindran & Timmins (2011) that demonstrates the marginal claimant of environmental improvements tends

27

Sieg et al. (2004), Kuminoff & Pope (2014), and Banzhaf (2018) suggest that the resulting partial

equilibrium estimates - as in my empirical design - will likely understate residents’ willingness to pay

and are better interpreted as a lower bound on the Hicksian equivalent surplus.

6 Spillovers & Decomposition of Benefits

Water and water pollution flow across jurisdictional borders. Less self-evident, however, is how much

local residents value upstream pollution abatement relative to local abatement. The answers to these

questions are important primarily because spillover benefits should be included in any efficiency as-

sessments of federal mandates. Cities treated by the CWA mandate may generate benefits to other

downstream cities unaffected directly by the CWA.

One complication to isolating spillover benefits generated by mandated public goods is that

treated cities can be, themselves, influenced by other treated cities through general equilibrium re-

sorting. The scale of the CWA mandate caused several communities to simultaneously experience

improvements in water quality. Such large-scale changes can adjust the shadow price for water quality,

change the demographic composition of households in treated cities, and/or change people’s willingness

to pay for amenities complementary to water quality such as waterfront parks or restaurants (Kuminoff

& Pope 2014). Each of these forces can alter a marginal mover’s rank ordering of the tax-amenity

bundles provided by various municipalities (Oates 1969). The result of these general equilibrium effects

is that two treated cities can exert competitive pressure on each other, and compress the competitor

city’s housing prices, all else equal.

In this final section, I conduct additional analyses to decompose the CWA treatment effect into

local versus spillover component parts while accounting for general equilibrium resorting effects. First,

I test how much of the CWA mandate effects on dissolved oxygen are due to local relative to upstream

abatement. I find that most of the water quality improvements experienced by the average city following

the CWA are due to local efforts, but 5-6% are due to efforts of upstream neighboring cities. Details of

this analysis are discussed in Appendix F. I then revisit the hedonic estimates after controlling explicitly

for CWA-induced upstream abatement and general equilibrium competition effects across cities. I find

significant positive effects of upstream abatement on local housing prices, but conclude that most CWA

compliance benefits arise from local abatement efforts. Third, I find evidence that upstream and local

abatement efforts are strategic complements, a stylized fact that may explain ex ante low adoption,

but ex post high compliance. Finally, I apply my hedonic estimates in a back-of-envelop calculation of

benefits versus costs of the CWA program. I find benefit-to-cost ratios less than one, but higher than

prior literature. Sieg et al. (2004) demonstrates that partial equilibrium estimates are significantly lower

to be those with higher income.

28

than general equilibrium estimates of willingness to pay for region-wide change in air quality. While

the present exercise is reduced form in nature, my results are thematically similar and underscore how

ignoring general equilibrium effects from nonmarginal changes in an amenity can underestimate true

benefits of environmental programs (Kuminoff & Pope 2014; Keiser et al. 2019).

6.1 Empirical Approach for Decomposition

I quantify the portion of local benefits generated from local versus upstream abatement efforts by

estimating a model that accounts for both effects as follows:yijt =δ1(Pi × dt) + δ2(

∑j 1(Pj)NUS × dt) + δ3(

∑j 1(Pj)N50mi × dt)

+ ZiΠt + (γr × t) + νi + τt + µit

(6)

where yijt denotes either water quality or house prices in city i with upstream cities j at time t, P is

an indicator for treatment status (primary treatment as of 1972), and dt is a post-CWA indicator. The

second and third terms capture the sum of treated upstream city populations (NUS) as well as treated

city populations within a 50 mile radius (N50mi) of city i. Lastly, ZiΠt is a vector of trends in city

baseline characteristics, including a control for time-varying effects of local labor market concentration

on ambient water pollution, measured as the overall population within a 50 mile radius (N50mi) of city

i. In Eq. (6), δ1 captures the change in yijt from a city’s own abatement effort under the CWA—the

“local effect”. δ2 is the change from upstream abatement efforts, while δ3 is the change from abatement

efforts within a 50 mile radius of city i—the “general equilibrium” (GE) effect.

To capture exposure to populations upstream (NUS) I, again, use digital maps on river networks

from the National Hydrography Dataset of the USGS to observe, for every stream segment i in the

contiguous US, which stream segment is immediately upstream and immediately downstream of that

segment i. This process requires that I reverse the recursive approach outlined in Section 4.1. Appendix

C provides further details on the upstream population calculation. The result of this process is that

I can identify both the aggregate population of all cities, as well as the aggregate population among

treated cities upstream of any city.

Fig. 10 shows upstream populations, aggregated as county means. The spatial distribution

exhibits a marked checker board pattern relative to that of mean downstream populations shown in

Fig. 5. The greater heterogeneity in upstream population within a small geographic area is a result

of the positive correlation between branching and being positioned further upstream on a network. If

two cities are on separate branches of the same network, those two cities are likely to have very similar

downstream populations (since their respective branches will converge), but can have vastly different

population counts upstream. Fig. 10 shows how counties located along major rivers are generally in

the darkest (largest) category of upstream population count, but upstream population declines with

29

distance in any direction from a major river.

6.2 Decomposition of Hedonic Effects

I show empirically in Appendix F that upstream abatement efforts from the CWA improved water quality

among cities downstream, though most water quality improvements among treated cities were due to

their own local adoption of secondary treatment. I find support that these spillover effects were due to

the CWA itself, as opposed to unobserved secular trends; placebo tests show that local water quality

was unaffected by mandate compliance from nearby cities located on different river networks. That is,

the GE term (δ3) in Eq. 6 has an effect on local water quality that is statistically indistinguishable from

zero. This is the result we would expect if two cities must be on the same river system to impact each

other’s water quality through the CWA mandate. Similarly, a second placebo shows that local water

quality was unaffected by exposure to non-treated cities upstream.

Next, I revisit the hedonic estimates from Section 5.2. I estimate Eq. 6 in order to separately

identify how local, upstream, and proximate exposure to CWA mandate compliance impacted local

housing prices. Unlike the water quality decomposition case, the hedonic GE effect (δ3) can impact

housing prices directly as people resort following the CWA. As resorting behavior will differ based

on availability of “substitute” municipalities, I include controls for the total upstream population in

addition to local labor market concentration in Zi. By estimating the GE effect in this reduced form

approach, I can quantify a lower bound on the hedonic spillover effect. δ2 is a combination of the

GE effect and a possible spillover effect. While upstream treated cities may improve water quality and

increase housing prices downstream, they also exert competitive pressure on housing prices downstream.

Consequently, I recover the spillover effect by subtracting estimates of δ3 from δ2.

I report results of this decomposition exercise in Tab. 6. Column (1) reports the baseline mean

hedonic estimate before accounting for spillover or GE resorting effects (also reported in column (3)

Tab. 5). Consistent with Sieg et al. (2004), the local effect increases by nearly 3 percentage points from

10.6 to 13.5% after accounting for spillover and GE effects in column (2). As before, most of the local

housing price responses are driven by small or medium-sized cities of less than 10,000 people (column

3).

In the case of the CWA, the GE effect (δ3) from proximate competitor cities adopting secondary

treatment suppresses local housing values by 1.5%. Subtracting δ3 from δ2 provides an hedonic estimate

of the spillover effect; compliance with the CWA mandate increased housing values by approximately

5% among communities downstream of treated cities in the 20 years following the CWA, all else equal.

30

6.3 Mechanisms Driving Mandate Benefits

The positive effects of the CWA mandate on revealed preference outcomes presents a puzzle. If water

quality improvements from secondary treatment adoption were mostly local, and revealed preference

outcomes respond positively to these local investments, why was federal regulation necessary for adop-

tion? A priori, a city may under-provide pollution abatement services due to several potential market

failures, including inaccess to credit needed for the fixed costs of infrastructure or failures of coordina-

tion across municipalities to collectively abate. Both explanations may be applicable here. For instance,

Fig. 1 suggests that cities were slow to adopt prior to the availability of federal construction grants

in 1975. It is outside the scope of this paper to disentangle the credit access effect from the com-

pulsory adoption effect in understanding which aspect of the CWA legislation was more instrumental

in generating positive house price growth.41 However, I provide evidence that pollution abatement is

complementary, rather than substitutable, across jurisdictions, suggesting that coordinated efforts in

pollution abatement are an integral part of the reason the mandate generated local benefits.

In Fig. 11, I show nonparametric estimates of median house price changes caused by the CWA

mandate, distinguished by baseline water quality. The housing price response is nonconvex: residents

value water quality improvements to surface waters that are high quality at baseline greater than similar

improvements to surface waters that are worse quality at baseline. Secondary treatment adoption

renders a positive impact on housing prices above baseline dissolved oxygen levels of 6-8 mg/l, when

surface water quality becomes acceptable for swimming. Water with dissolved oxygen above 7 mg/l is

acceptable for drinking (Vaughan 1986), and generates even greater benefits from the CWA mandate.

These findings suggest that cities may be more incentivized to abate their own pollution if the

water quality entering their city from upstream is, all else equal, cleaner. I interpret these findings as

consistent with recent work by Albouy et al. (2018). Their paper shows that estimates of resident

value for public goods can critically depend on their complementarities with other public goods. Public

park access, for example, is valuable to residents only if accompanied by improvements to public safety.

In the context of wastewater treatment, residents may value pollution abatement of outgoing piped

water only if the surface waters entering the city limits are of high quality. That is, only if upstream

neighboring cities coordinate in their efforts to improve surface waters. The existence of nonconvex

price responses from the CWA provides support that the CWA - by mandating uniform adoption of

pollution abatement - may have helped correct for prior failures of coordination.41Such an analysis would require an identification strategy for grant receipt, in addition to the identification strategy

proposed here for ex ante CWA noncompliance.

31

6.4 Benefit-to-Cost Calculation

In this section, I calculate the ratio of benefits implied by the effect of the CWA secondary treatment

technology standard on housing values to its cost. I calculate aggregate changes in housing values using

estimates from Eq. 5 column (3) combined with municipality-specific median housing values and housing

unit counts as of 1970 from the census. Fig. 12 shows the spatial distribution of net benefits aggregated

to the county-level. Areas with the greatest benefits are those with several upstream cities forced to

comply with the mandate and few nearby compliers that would otherwise exert competitive pressure.

For example, St. Louis and New Orleans have some of the highest benefits of all cities because they

both have a large housing stock and a large number of upstream complying cities. Smaller cities like

Cincinnati, OH or Henderson, KY also experience positive net benefits because several cities upstream

along the Ohio river complied with the CWA. Areas with the greatest losses tend to be near large,

treated urban centers. For instance, municipalities in New Jersey that surround NYC or areas around

Philadelphia (both of which adopted secondary treatment under the CWA) experience a net loss due to

predicted out migration into these treated cities. In aggregate, net benefits from the CWA are $206.4

billion (in 2012 USD).

I obtain cost estimates from prior work by K&S, Keiser et al. (2019), and Anderson (2010). I

also obtain a measure of federal costs by directly calculating the cost of federal grant outlays under

the Construction Grants Program, sourced from the EPA Grants Information and Control System

database. All cost measures are representative of costs from 1972 to 1992 to be consistent with my

hedonic estimates. Tab. 7 shows that ratios of benefits-to-costs range from 0.41 to 0.72 depending on

the cost measure. Prior work on the CWA finds a median ratio of 0.37 (Keiser & Shapiro 2019), thus

this paper’s ratios are slightly higher than what prior literature suggests. However, my results concede

that even after accounting for spillover effects to nontreated cities, the benefits of the CWA technology

standard appear to be smaller than its cost. These ratios are likely to represent a lower bound, however,

since several potentially important sources of benefits, such as health impacts, labor market growth in

the public works sector, or technological improvements in wastewater are excluded from this analysis.

7 Conclusion

The American Society of Civil Engineers estimates that infrastructure in the United States requires an

investment of at least $3.2 trillion to prevent deterioration of the country’s aging roadways, electrical

grids, transit, and waterworks (ASCE 2016). To meet some of these needs, the Trump administration

proposed in January of 2018 to use local user fees and state and local financing to fund a $1.5 trillion

infrastructure renewal plan. While there is bipartisan consensus that infrastructure renewal is necessary,

we know very little about how localized financing for federally-mandated projects affects local economies.

32

The striking gap in our knowledge of these effects matters because understanding who ultimately bears

the burden of federal spending mandates and whether mandates are valued locally can alter conclusions

of the cost effectiveness of federal policies.

This paper is the first empirical effort to assess the impact of federal mandates on local gov-

ernment budgets and to determine whether mandated provision of goods and services are valued by

local residents. Further, this is the first paper to categorize compliance with the CWA’s legal terms on

infrastructure, and to test the effectiveness and consequences of that regulation for local governments.

I find that a mandate on wastewater treatment induced cities to spend over 200% more per capita

than they otherwise would have on the mandated good. Extrapolating these results to all mandates

implies that cities allocate nearly 9.5% more of their budgets annually—or approximately $49 million

in aggregate as of 2012—on mandated goods and services than they would without federal enforcement.

While opponents of mandated programs argue that mandates crowd out local spending, I do not

find evidence that local governments displaced funding from other public goods and services in order

to fund mandated infrastructure, even several decades following the CWA legislation. Rather, local

governments relied mainly on a two-fold increase in user fees to finance the non-subsidized portions

of mandate compliance. Several of the largest federal mandates - including regulations on solid waste

management and drinking water quality - are funded with a fee for service, suggesting that mandate

compliance is unlikely to generate distortions in the menu of goods and services offered by local gov-

ernments. Yet, reliance on user fees for essential, demand-inelastic goods like piped water presents

important distributional concerns. In Baltimore, for example, federal demands on renewal of the city’s

water infrastructure have increased resident water bills so much that the city has repossessed several

homes for unpaid water bills (Baltimore Sun Editorial Board 2019). Future research should assess the

equity implications of federal public works regulations.

These mandated expenditures do not necessarily leave municipalities worse off, however. The

local impacts of the CWA infrastructure mandate were associated with statistically significant positive

increases to local municipal population and housing prices, particularly among smaller communities

of less than 10,000 people which represent the majority of municipal governments in the US. The

networked nature of local water pollution abatement services is integral for quantifying and interpreting

these effects. I find that local benefits increase with exposure to abatement efforts upstream, and such

abatement efforts appear complementary across jurisdictions. While national level benefit-to-cost ratios

are less than one, my results suggest that the externality corrections induced by the CWA mandate are

at least valued above their cost by taxpayers at the local level.

33

ReferencesAlbouy, David. 2008. Are big cities bad places to live? Estimating quality of life across metropolitan

areas. Tech. rept. National Bureau of Economic Research.Albouy, David, & Farahani, Arash. 2017. Valuing public goods more generally: the case of infrastructure.

Upjohn Institute working paper.Albouy, David, Christensen, Peter, & Sarmiento-Barbieri, Ignacio. 2018. Unlocking Amenities: Esti-

mating Public-Good Complementarity. Tech. rept. National Bureau of Economic Research.Alm, James, Buschman, Robert D, & Sjoquist, David L. 2011. Rethinking local government reliance

on the property tax. Regional Science and Urban Economics, 41(4), 320–331.Anderson, Richard F. 2010 (February). Trends in Local Government Expenditures on Public Water and

Wastewater Services and Infrastructure: Past, Present and Future.Andreen, William L. 2013. Success and Backlash: The remarkable (continuing) story of the Clean

Water Act. George Washington Journal of Energy & Environmental Law, 4, 25.ASCE. 2016. Failure to Act: Closing the Infrastructure Investment Gap for America’s Economic Future.

Tech. rept. American Society of Civil Engineers, Washington, DC.Baicker, Katherine. 2001. Government decision-making and the incidence of federal mandates. Journal

of Public Economics, 82(2), 147–194.Baicker, Katherine, & Gordon, Nora. 2006. The effect of state education finance reform on total local

resources. Journal of Public Economics, 90(8-9), 1519–1535.Baltimore Sun Editorial Board. 2019. How to help with escalating water bills. The Baltimore Sun, Jan

22.Banzhaf, H Spencer. 2018. Difference-in-Differences Hedonics. Working Paper.Banzhaf, H Spencer, & Walsh, Randall P. 2008. Do people vote with their feet? An empirical test of

Tiebout. American Economic Review, 98(3), 843–63.Baum-Snow, Nathaniel, & Kahn, Matthew E. 2000. The effects of new public projects to expand urban

rail transit. Journal of Public Economics, 77(2), 241–263.Bayer, Patrick, Keohane, Nathaniel, & Timmins, Christopher. 2009. Migration and hedonic valuation:

The case of air quality. Journal of Environmental Economics and Management, 58(1), 1–14.Black, Sandra E. 1999. Do better schools matter? Parental valuation of elementary education. The

Quarterly Journal of Economics, 114(2), 577–599.Bleakley, Hoyt, & Lin, Jeffrey. 2012. Portage and path dependence. The Quarterly Journal of Economics,

127(2), 587–644.Bockstael, Nancy E, Hanemann, W Michael, & Kling, Catherine L. 1987. Estimating the value of

water quality improvements in a recreational demand framework. Water Resources Research, 23(5),951–960.

Bohn, Henning, & Inman, Robert P. 1996. Balanced-budget rules and public deficits: evidence from theUS states. Pages 13–76 of: Carnegie-Rochester conference series on public policy, vol. 45. Elsevier.

Boyle, Kevin J, Poor, P Joan, & Taylor, Laura O. 1999. Estimating the demand for protecting freshwaterlakes from eutrophication. American Journal of Agricultural Economics, 81(5), 1118–1122.

Brown, Drew. 2018 (August). Personal interview. Environmental Engineer and Manager, City ofPhiladelphia Water Department.

Cain, Louis. 2005. Sanitation in Chicago: A Strategy for a Lakefront Metropolis. Tech. rept. The

34

Encyclopedia of Chicago, Chicago, IL.Cellini, Stephanie Riegg, Ferreira, Fernando, & Rothstein, Jesse. 2010. The value of school facility

investments: Evidence from a dynamic regression discontinuity design. The Quarterly Journal ofEconomics, 125(1), 215–261.

Census, US. 1994 (August). Products-Geography. https://www.geolytics.com/USCensus,Neighborhood-Change-Database-1970-2000,Data,Geography,Products.asp. Accessed: 2019-10-18.

Chay, Kenneth Y., & Greenstone, Michael. 2005. Does air quality matter? Evidence from the housingmarket. Journal of Political Economy, 113(21), 376–424.

Conlan, Timothy J. 1994. Federally induced costs affecting state and local governments. US AdvisoryCommission on Intergovernmental Relations.

Cooper, Russell, & John, Andrew. 1988. Coordinating coordination failures in Keynesian models. TheQuarterly Journal of Economics, 103(3), 441–463.

Copeland, Claudia. 1999. Clean Water Act: a summary of the law. Congressional Research Service,Library of Congress Washington, DC.

Copeland, Claudia. 2012. Clean Water Act and Pollutant Total Maximum Daily Loads (TMDLs).Congressional Research Service, Library of Congress Washington, DC.

Copeland, Claudia. 2015. Funding for EPA Water Infrastructure: A Fact Sheet. Congressional ResearchService, Library of Congress Washington, DC.

Copeland, Claudia. 2016. Allocation of Wastewater Treatment Assistance: Formula and Other Changes.Congressional Research Service, Library of Congress Washington, DC.

Cromwell, Erich, Ihlanfeldt, Keith, et al. . 2015. Local government responses to exogenous shocks inrevenue sources: Evidence from Florida. National Tax Journal, 68(2), 339–376.

Currie, Janet, Davis, Lucas, Greenstone, Michael, & Walker, Reed. 2015. Environmental health risksand housing values: evidence from 1,600 toxic plant openings and closings. American EconomicReview, 105(2), 678–709.

Dahlby, Bev. 2008. The marginal cost of public funds: Theory and applications. MIT press.Deming, David J, Cohodes, Sarah, Jennings, Jennifer, & Jencks, Christopher. 2016. School accountabil-

ity, postsecondary attainment, and earnings. Review of Economics and Statistics, 98(5), 848–862.Dilger, Robert Jay. 2013. Unfunded Mandates Reform Act: History, Impact, and Issues. Congressional

Research Service, Library of Congress Washington, DC.Duflo, Esther, & Pande, Rohini. 2007. Dams. The Quarterly Journal of Economics, 122(2), 601–646.Duranton, Gilles, & Turner, Matthew A. 2012. Urban growth and transportation. Review of Economic

Studies, 79(4), 1407–1440.Earnhart, Dietrich. 2004. Regulatory factors shaping environmental performance at publicly-owned

treatment plants. Journal of Environmental Economics and Management, 48(1), 655–681.EPA. 1973. Estimate of Municipal Wastewater Treatment Facility Requirements. Tech. rept. Clean

Watershed Needs Survey.EPA. 1976 (February). An Analysis of Construction Cost Experience for Wastewater Treatment Plants.

Tech. rept. 430/9-76-002. US Environmental Protection Agency Office of Water Program Operations,Washington, DC.

EPA. 1988 (September). The Municipal Sector Study: Impacts of Environmental Regulations on Munic-

35

ipalities. Tech. rept. MCDEPA 230-09/88-038. US Environmental Protection Agency, Washington,DC.

EPA. 1994 (August). Current 301(h) Waiver Recipients AND 301(h) Applications Pending FinalDecision. https://web.archive.org/web/20150227140444/http:/water.epa.gov/type/oceb/assessmonitor/301list.cfm. Accessed: 2019-04-18.

Fairfax, V, & Hamilton, V. 2000. Progress in water quality: an evaluation of the national investmentin municipal wastewater treatment. Washington, DC: US EPA.

Feler, Leo, & Senses, Mine Z. 2017. Trade shocks and the provision of local public goods. AmericanEconomic Journal: Economic Policy, 9(4), 101–43.

Fields, Gary, & Emshwiller, John R. 2011. Federal Offenses: A Sewage Blunder Earns Engineer aCriminal Record. Wall Street Journal, Dec 12.

Fisher, Ronald C. 2015. State and local public finance. Routledge.Foster, Henry S, & Beattie, Bruce R. 1979. Urban residential demand for water in the United States.

Land Economics, 55(1), 43–58.Gamper-Rabindran, Shanti, & Timmins, Christopher. 2011. Hazardous waste cleanup, neighborhood

gentrification, and environmental justice: Evidence from restricted access census block data. AmericanEconomic Review, 101(3), 620–24.

Gamper-Rabindran, Shanti, & Timmins, Christopher. 2013. Does cleanup of hazardous waste sites raisehousing values? Evidence of spatially localized benefits. Journal of Environmental Economics andManagement, 65(3), 345–360.

GAO. 1980. EPA Should Help Small Communities Cope with Federal Pollution Control Requirements,Report to Congress by the Comptroller General of the United States. Tech. rept. US General Account-ing Office.

Greenstone, Michael. 2002. The Impacts of Environmental Regulations on Industrial Activity: Evidencefrom the 1970 and 1977 Clean Air Act Amendments and the of Manufactures. Journal of PoliticalEconomy, 110(6), 1175–1219.

Guo, Tianjiao, Englehardt, James, & Wu, Tingting. 2014. Review of cost versus scale: water andwastewater treatment and reuse processes. Water Science and Technology, 69(2), 223–234.

Gyourko, Joseph, & Tracy, Joseph. 1991. The structure of local public finance and the quality of life.Journal of Political Economy, 99(4), 774–806.

Hanford, Priscilla L, & Sokolow, Alvin D. 1987. Mandates as Both Hardship and Benefit: The CleanWater Program in Small Communities. Publius: The Journal of Federalism, 17(4), 131–146.

Haughwout, Andrew F. 2002. Public infrastructure investments, productivity and welfare in fixedgeographic areas. Journal of Public Economics, 83(3), 405–428.

Hines, James R, & Thaler, Richard H. 1995. Anomalies: The flypaper effect. The Journal of EconomicPerspectives, 9(4), 217–226.

Hudson, S, Hull, P, & Liebersohn, CJ. 2015. Interpreting Instrumented Difference-in-Differences. Tech.rept. Working Paper (available upon request).

Hudson River Watershed Alliance. 2015 (November). Watershed Rules and Regulations for Protectionand Drinking Water in New York. New York State Department of Environmental Conservation.

Imazeki, Jennifer, & Reschovsky, Andrew. 2004. Is No Child Left Behind an un (or under) fundedfederal mandate? Evidence from Texas. National Tax Journal, 571–588.

36

Kahn, Matthew E. 2000. Smog reduction’s impact on California county growth. Journal of Regionalscience, 40(3), 565–582.

Kahn, Matthew E. 2001. The Beneficiaries of Clean Air Act Regulation. Regulation Magazine, 24, 34.Kasperowicz, Pete. 2018. GOP to attack unfunded federal mandates next week. Washington Examiner,

July 5.Keiser, David A, & Shapiro, Joseph S. 2018. Consequences of the Clean Water Act and the demand for

water quality. The Quarterly Journal of Economics, 134(1), 349–396.Keiser, David A, & Shapiro, Joseph S. 2019. Burning Waters to Crystal Springs? US Water Pollution

Regulation over the Last Half Century. Tech. rept. National Bureau of Economic Research.Keiser, David A, Kling, Catherine L, & Shapiro, Joseph S. 2019. The low but uncertain measured

benefits of US water quality policy. Proceedings of the National Academy of Sciences, 116(12), 5262–5269.

Kuminoff, Nicolai V, & Pope, Jaren C. 2014. Do “capitalization effects” for public goods reveal thepublic’s willingness to pay? International Economic Review, 55(4), 1227–1250.

Lake, Elizabeth E Hanneman, Oster, William M, et al. . 1979. Who pays for clean water? the distributionof Water pollution control costsan urban systems research report.

Lechner, Michael. 2011. The estimation of causal effects by difference-in-difference methods. Foundationsand Trends in Econometrics, 4(3), 165–224.

Leggett, Christopher G, & Bockstael, Nancy E. 2000. Evidence of the effects of water quality onresidential land prices. Journal of Environmental Economics and Management, 39(2), 121–144.

Lin, Yatang. 2018 (November). The Hidden Cost of Industrial Pollution: Environmental Amenities andthe Location of Service Jobs.

Lipton, Douglas. 2004. The value of improved water quality to Chesapeake Bay boaters. MarineResource Economics, 19(2), 265–270.

Lutz, Byron, Molloy, Raven, & Shan, Hui. 2011. The housing crisis and state and local government taxrevenue: Five channels. Regional Science and Urban Economics, 41(4), 306–319.

Lutz, Byron F. 2008. The connection between house price appreciation and property tax revenues.National Tax Journal, 555–572.

McQuillin, Eugene. 1912. A Treatise on the law of municipal corporations. Vol. 4. Callaghan & Company.Melnik, Walter. 2017. Municipal Government Reaction to Mass Layoffs in Ohio. Working Paper.Melosi, Martin V. 2000. Sanitary City. the University of Pittsburgh Press.Metcalf, Leonard, & Eddy, Harrison P. 1922. Sewerage and Sewage Disposal: A Textbook. McGraw Hill.Minnesota Pollution Control Agency. 2009. Low Dissolved Oxygen in Water: Causes, Impact on Aquatic

Life-An Overview. Tech. rept. St. Paul, MN.Missouri. 1901. Missouri v. Illinois and Sanitary District of Chicago, 180 U.S. 208.National League of Cities. 2017. Energy, Environment and Natural Resources 2018. Tech. rept.NCSL. 2018. Policies for the Jurisdiction of the Budgets and Revenue Committee. Tech. rept. National

Conference of State Legislatures, Washington, DC.Oates, Wallace E. 1969. The effects of property taxes and local public spending on property values:

An empirical study of tax capitalization and the Tiebout hypothesis. Journal of Political Economy,77(6), 957–971.

Okun, Daniel A. 1996. From cholera to cancer to cryptosporidiosis. Journal of Environmental engi-

37

neering, 122(6), 453–458.Olmstead, Sheila M. 2010. The economics of water quality. Review of Environmental Economics and

Policy, 4(1), 44–62.Olmstead, Sheila M, & Kuwayama, Yusuke. 2015. Water quality and economics: willingness to pay,

efficiency, cost-effectiveness, and new research frontiers. Handbook on the Economics of NaturalResources, 474.

Owens, Raymond E, Rossi-Hansberg, Esteban, & Sarte, Pierre-Daniel G. 2019. Rethinking Detroit.American Economic Journal: Economc Policy.

Phelps, Earle B. 1914. Studies on the self-purification of streams. Public Health Reports (1896-1970),2128–2132.

Poor, P Joan, Pessagno, Keri L, & Paul, Robert W. 2007. Exploring the hedonic value of ambient waterquality: A local watershed-based study. Ecological Economics, 60(4), 797–806.

Rappaport, Jordan, & Sachs, Jeffrey D. 2003. The United States as a coastal nation. Journal ofEconomic growth, 8(1), 5–46.

Reback, Randall, Rockoff, Jonah, & Schwartz, Heather L. 2014. Under pressure: Job security, resourceallocation, and productivity in schools under No Child Left Behind. American Economic Journal:Economic Policy, 6(3), 207–41.

Rechtschaffen, Clifford. 2003. Enforcing the Clean Water Act in the twenty-first century: Harnessingthe power of the public spotlight. Alabama Law Review, 55, 775.

RMQAA. 2015. On the Water Front: US Wastewater Treatment History. http://www.rmwqaa.org/Resources/Documents/Newsletters/RMWQAA_2015_Q2_Final.pdf. Rocky Mountain Water QualityAnalysts Association.

Roback, Jennifer. 1982. Wages, rents, and the quality of life. Journal of political Economy, 90(6),1257–1278.

Rosen, Sherwin. 1974. Hedonic prices and implicit markets: product differentiation in pure competition.Journal of Political Economy, 82(1), 34–55.

Schroeder, Jonathan, Riper, David Van, & Ruggles, Steven. 2019. IPUMS National Historical Geo-graphic Information System: Version 14.0. http://doi.org/10.18128/D050.V14.0.

Seward, James. 2017 (June). Senate Passes Comprehensive Mandate Relief Legislation. The New YorkState Senate.

Shoag, Daniel, Tuttle, Cody, & Veuger, Stan. 2019. Rules versus home rule: Local government responsesto negative revenue shocks. National Tax Journal, 72(3), 543–547.

Sieg, Holger, Smith, V Kerry, Banzhaf, H Spencer, & Walsh, Randy. 2004. Estimating the generalequilibrium benefits of large changes in spatially delineated public goods. International EconomicReview, 45(4), 1047–1077.

Skidmore, Mark, & Scorsone, Eric. 2011. Causes and consequences of fiscal stress in Michigan cities.Regional Science and Urban Economics, 41(4), 360–371.

Smith, V Kerry, & Desvousges, William H. 1986. Measuring water quality benefits. Vol. 3. SpringerScience & Business Media.

Stets, Edward G, Kelly, Valerie J, Broussard III, Whitney P, Smith, Thor E, & Crawford, Charles G.2012. Century-scale perspective on water quality in selected river basins of the conterminous UnitedStates. Tech. rept. US Geological Survey.

38

Stock, James, & Yogo, Motohiro. 2005. Asymptotic distributions of instrumental variables statisticswith many instruments. Vol. 6. Chapter.

Stoddard, Andrew, Harcum, Jon B, Simpson, Jonathan T, Pagenkopf, James R, & Bastian, Robert K.2003. Municipal wastewater treatment: evaluating improvements in national water quality. JohnWiley & Sons.

Tarr, Joel. 2016. Infrastructure: Water and Sewers. In: Oxford Research Encyclopedia of AmericanHistory.

Train. 1975. Train versus City of New York, 420 U.S. 35.US Census Bureau. 2015. Historical Finances of Individual Governments, 1967-2012.US Conference of Mayors. 1993. Impact of Unfunded Federal Mandates on US Cities: A 314-City

Survey. Second Printing, October, 4.Vaughan, William J. 1986. The RFF Water Quality Ladder (Appendix B of ”The Use of Contingent

Valuation Data for Benefit/Cost Analysis in Water Pollution Control”). Tech. rept. EPA.Walsh, Patrick J, Milon, J Walter, & Scrogin, David O. 2011. The spatial extent of water quality

benefits in urban housing markets. Land Economics, 87(4), 628–644.Walters, Gary. 2017. Personal interview. Pennsylvania Department of Environmental Protection.Weiland, Paul S. 1998. Environmental regulations and local government institutional capacity. Public

Administration Quarterly, 176–203.Westerling, Kevin ed. 2011. Owner and operator of wastewater treatment facilities sentenced for vio-

lating the Clean Water Act. Water Online, June 10.

39

Figure 1: Inventory of Publicly-Owned Wastewater Treatment Plant Technology. Source: EPA Clean Watershed NeedsSurveys (1973-2004). Figure plots the total number of publicly-owned wastewater treatment plants by technology type.Secondary treatment technology was mandated under the 1972 CWA for all surface water-bound effluent. Treatmentplants with only primary treatment were required to either cease operations or upgrade to secondary treatment. See textfor further details on treatment technology characteristics.

Figure 2: Municipal Expenditure Trends. Source: Census of Governments. Figure plots raw means for annual directexpenditures (total expenditures net of intergovernmental payments) in thousands of 2012 USD in gray; and annual shareof expenditures in wastewater in black. Dashed line indicates the start of the Clean Water Act in 1972.

40

Figure 3: Capital Cost of Secondary Treatment by Plant Service Population. Source: EPA (1973). Figure plots thebinned scatterplot and quadratic fit of cost needs for compliance with the secondary treatment standard relative to aplant’s service population. Plot divided into 100 equal-sized bins. Residualized by year fixed effects. Sample is based on48,115 observations, which includes 5,884 treatment facilities from 1975-2003 reporting non-zero secondary treatment costneeds and non-zero service population. Sample excludes the top and bottom 5% of treatment capacity, and plants thatappear less than 7 years over the 28-year panel. Plant service population is calculated as plant capacity in gallons per daydivided by 100 (Guo et al. 2014). Cost values are in thousands of 2012 dollars.

41

Figure 4: Distribution of Pre-CWA Secondary Treatment Adoption. Source: EPA (1973), Census. Figure shows thecounty share of municipal wastewater treatment plants with secondary treatment as of 1972.

Figure 5: Mean Municipal Downstream Population Size by County, 1970. Source: USGS, Census, author’s own calcula-tion. Figure shows county-level averages of city downstream population as of 1970. Only major rivers shown for expositionpurposes.

42

Figure 6: State Composition of Wastewater Treatment Technology, 1972. Source: EPA 1973 Clean Watershed NeedsSurvey. Figure shows the distribution of wastewater treatment plant technology composition by state. Each shade of graycorresponds to a quartile.

Figure 7: State Compliant Plant Share (1972) vs Downstream Population. Source: EPA 1973 clean watershed needssurvey, USGS, author’s calculations. Figure shows the distribution of downstream population (25th-75th percentile) bystate share of compliant plants as of 1972.

43

Figure 8: Wastewater Expenditures per capita and Instrumental Variable Variation. Source: USGS, EPA (1973), Censusof Governments, author’s calculations. Panel A plots δt×50 + δt and δt from equation: yit =

∑tδt×50(I50 × Ss ×Dt) +∑

tδt(Ss × Dt) + (Diσt) + νi + εit where the dependent variable is wastewater expenditures per capita for city i in

year t; I50 is an indicator for a city having downstream population in the bottom 50th percentile, Ss is state share ofcompliant plants as of 1972, Dt is an indicator for year t, and Di is downstream population. Bands show 95% confidenceintervals. All coefficients are evaluated at the mean state share of secondary treatment plants as of 1972. The referenceyear is t=1972. Figure depicts city variation only within states with non-zero compliance share. Black triangles show theestimated difference in expenditures in year t relative to 1972 for cities with downstream population size in the bottom50th percentile. Gray estimates show the difference in expenditures in year t relative to 1972 for cities with downstreampopulation size in the top 50th percentile. Robust standard errors are clustered at the city level. Panel B plots σt×50 +σt

and σt from equation: yit =∑

tσt×50(I50 × Dt) +

∑tσt(Dt) + νi + εit where the dependent variable is wastewater

expenditures per capita for city i in year t; I50 is an indicator equal to 1 if a city’s state has share of compliant treatmentplants as of 1972 in the bottom 50th percentile, and Dt is an indicator for year t. Bands show 95% confidence intervals.The reference year is t=1972. Black triangles show the estimated difference in expenditures in year t relative to 1972 forcities with state compliant plant share in the bottom 50th percentile. Gray estimates show the difference in expendituresin year t relative to 1972 for cities with state compliant plant share in the top 50th percentile. Robust standard errors areclustered at the city level.

44

Figure 9: Pre-trends of selected city characteristics. Figure plots estimates of δt from Eq. 5: yirt =∑′02t=′67(3′72) δt

(Pi ×Dt) + Xiθt + (γr × t) + τt + νi + εirt. δt is the difference between control and treated cities ineach of three outcomes (yirt=ln(population), county share of employment in manufacturing, and dissolved oxygen) fromyear t relative to 1972. Treatment status Pi is instrumented using downstream population, state baseline compliance share,and their interaction. Regressions include all controls listed in column 5 of Tab. 2. Panels B, D, and E include time trendsin state debt rules. Bands show 95% confidence intervals. Robust standard errors clustered at the city level. Dissolvedoxygen (DO2) measured as the five-year annual average from years t to t − 5. DO2 in 1957 is the 10-year average from1957 to 1947.

45

Figure 10: Mean Municipal Upstream Population Size by County, 1970. Source: USGS, Census, author’s calculations.Figure shows county-level averages of city upstream population as of 1970. Only major rivers shown for exposition purposes.

Figure 11: Average Effect of CWA Mandate on Housing Prices by Baseline Water Quality Bin. Figure plots estimatesof αb, the heterogeneous effects by baseline water quality of the CWA on municipal ln(housing prices). The estimatingequation is a version of Eq. 5: yirt =

∑6b=2 αb

(Pi × POSTt × Ib) + Xiθt + (γr × t) + τt + νi + εirt. Ib is an indicator forone of six 2-mg/l-bins of baseline local water quality measured by dissolved oxygen. The coefficients αb are the averagechange in housing prices following the CWA for bin Ib relative to the reference bin of cities with baseline water quality lessthan 2mg/l.Treatment status Pi is instrumented using downstream population, state baseline compliance share, and theirinteraction. The instruments are interacted with Ib in the first stage. Dashed lines represent 95% confidence intervals.Controls include time trends in baseline city characteristics and watershed (all controls listed in column (5) of Tab. 2), aswell as spillover (upstream treated population x post) and general equilibrium (treated population within 50km radius xpost) effects of CWA compliance. Standard errors clustered by county.

46

Net Benefits ($mn)Less than -0.15(-0.14) - ( -0.01)0.00 - 0.010.02 - 0.100.11 - 10.00Over 10.01No Data

Figure 12: Distribution of Net Benefits from CWA mandate. Figure shows county-level aggregates of net changes tomunicipal housing stock value due to the CWA secondary treatment mandate. Net Benefits based on multiplying the local(δ1), spillover (δ2−δ3), and general equilibrium (δ3) hedonic effects from Tab. 6 column(2) by 1970 values of: (i) municipaltreatment status; (ii) median local house price (census), (iii) local housing stock (census) (iii) upstream treated populationsize (details on calculation in Appendix C.2) and (iv) population size of cities within a 50km radius (calculated in GISusing census population data. Municipal net benefits aggregated to county level for exposition purposes.)

47

Table 1: Pre-Clean Water Act Descriptive Statistics

Primary Secondary P-value forTreatment Treatment difference in means

City CharacteristicsPopulation 36,676.641 30,076.018 0.451Median House Price ($) 95,576.984 103,378.320 0.000Share of population with a college degree 0.110 0.115 0.101Dissolved oxygen (mg/l) 7.772 7.814 0.518League of Conservation Voters score 43.379 50.499 0.000

County-level labor marketCounty income per capita ($) 23,100.982 24,110.404 0.000County employment share in manufacturing 0.361 0.388 0.000County employment share in water-polluting manufacturing 0.146 0.153 0.214Manufacturing employment growth (1956-1970) -0.347 -0.325 0.265

Expenditures per capitaTotal expenditures ($) 1,026.562 1,204.099 0.000

Wastewater 66.654 116.813 0.000Total other 648.340 672.364 0.223

Public works 132.459 134.291 0.557Public safety 382.878 387.104 0.795General & admin. 61.354 75.748 0.000Health & welfare 28.631 31.231 0.500Recreation 43.018 43.990 0.657

Revenues per capitaTotal revenues pc ($) 1,013.392 1,149.182 0.000

Intergovernment revenues 168.738 218.858 0.000Revenues from own sources 844.692 930.360 0.003

Total taxes 389.035 507.524 0.000Property taxes 301.137 430.637 0.000Sales & License taxes 87.902 76.844 0.012

Total user fees 111.038 102.130 0.209Wastewater user fees 31.131 30.595 0.741

Long-term debt outstanding 1,388.240 1,385.920 0.982Short-term debt outstanding 88.716 114.147 0.027

GeographyRiver Population as of 1970 (th.) 8,521.381 5,636.650 0.000Distance to waterbody (km) 0.057 0.021 0.004Distance to river mouth (’000 km) 1,267.481 839.875 0.000Distance to navigable river (km) 207.404 192.960 0.082Distance to Great Lake (km) 734.905 664.471 0.006Distance to Ocean (km) 522.979 392.917 0.000

Number of Cities 2,312 663Panel Frequency 5.2 5.4Observations 11,419 3,447Note: All variables measured as means in 1967 and 1972. P-value denotes significance of difference in means.Dollars in USD 2012 values.

48

Table 2: Determinants of Ex Ante CWA Mandate Compliance Status

Cross Section (1972) Panel

(1) (2) (3) (4) (5) (6) (7)

Downstream Population -0.053∗∗∗ -0.068∗∗∗(0.012) (0.014)

Dowstream Population x StateShare’72 x Post -0.668∗∗ -0.584∗ -0.272 -0.244 -0.035(0.321) (0.329) (0.372) (0.374) (0.431)

Dowstream Population x Post -0.008 -0.015 -0.038∗ -0.041∗ -0.065∗∗(0.019) (0.019) (0.022) (0.022) (0.026)

StateShare’72 x Post -2.072∗∗∗ -2.476∗∗∗ -2.942∗∗∗ -2.912∗∗∗(0.343) (0.366) (0.522) (0.699)

River FE Y YBaseline Controls YCity & YearFE Y Y Y Y YRiverPopulation x YearFE Y Y Y Y YBaseline Controls x YearFE Y Y Y YIncome Trend Y Y Y YRegion Trend Y Y Y YWatershed x Year FE Y YState Debt Rule x Year FE YState x Year FE Y

Change in P(noncompliance) given:1 SD increase DSpop -7.10% -9.06% -4.20% -4.72% -6.36% -6.55% -8.79%1 SD increase StateShare -9.86% -11.01% -11.57% -11.36% -0.13%

Kleibergen-Paap F-statistic 20.145 23.161 17.436 11.185 10.356Stock-Yogo critical value 22.30 22.30 22.30 22.30 19.93Observations 2151 2151 14866 14866 14866 14866 14866

Note: The dependent variable in (1) and (2) is an indicator for primary treatment as of 1972. The dependent variable in(3)-(7) is an indicator for primary treatment as of 1972 interacted with a post CWA indicator. The “River” fixed effect isdenoted by the terminating point of the city’s nearest river. The Baseline Controls in specifications (1) and (2) include:distance to the river terminating point, pre-CWA (averaged from 1967-1972) intergovernmental grant receipt, growthfrom 1956 to 1970 in manufacturing employment, share of pre-CWA employment in water pollution industries, anddistance to ocean. River Population is size of total river network population as of 1970. Baseline Controls in columns(3)-(7) include time trends of several pre-CWA characteristics (averaged from 1967-1972): share of employment inwater-polluting industries, growth from 1956 to 1970 in manufacturing employment; and annual federal, state, localintergovernmental grants; and distance from city centroid to coastline. Income trend is county-level average income percapita, averaged from 1967 and 1972, interacted with a linear time trend. Region trend consists of 8 indicators based onBEA US regions: New England, Mideast, Great Lakes, Plains, Southeast, Southwest, Rocky Mountain, and Far Westeach interacted with a linear time trend. Watershed includes one of 18 hydrologic units across the coterminous US. StateDebt Rule takes one of three values for none, statutory, or constitutional state requirements as of 1970 that localgovernments balance their budget at the end of each fiscal year (source: Bohn & Inman 1996). Downstream population isnormalized by its standard deviation. Stock-Yogo critical values based on the weak identification values for 10% maximalIV size. Marginal effects evaluated at the baseline mean of noncompliance (75%). Standard errors clustered by city. ∗

(p<0.10), ∗∗ (p<0.05), ∗∗∗ (p<0.01).

49

Table 3: Pretrends of City Characteristics by Treatment and Instrument

Secondary Treatment Above Median Mean ofas of 1972 Exposure to Instrument Secondary Treatment

(1) (2) (3) (4) (5)

Ln(Population)† 0.035** 0.036** -0.009 0.001 30176(0.016) (0.016) (0.012) (0.025)

Manufacturing Employment Share‡ -0.010 0.003 0.012 0.051*** 0.38(0.009) (0.009) (0.008) (0.012)

Dissolved Oxygen (mg/l)∓ 0.113 0.012 0.240** 0.221 7.814(0.106) (0.115) (0.102) (0.166)

Total federal grants pc ($) 15.450 18.897 5.618 7.257 35.67(10.282) (11.675) (9.225) (9.805)

Federal Infrastructure grants pc ($) 22.570*** 24.164*** 0.311 3.458 25.41(8.462) (8.266) (5.642) (6.698)

Total Expenditures pc ($) 66.126 49.429 0.827 -89.521 1204.1(55.710) (54.730) (41.100) (63.051)

Sewerage Expenditures pc ($) 73.062*** 65.848*** 36.122*** 7.209 116.81(22.132) (21.982) (13.546) (17.930)

Other Expenditures pc ($) -22.101 -21.534 -58.494** -98.731** 672.36(36.622) (37.122) (26.963) (42.422)

Total Revenues pc ($) -33.093 -27.144 -73.670 -82.554 1149.18(47.144) (39.035) (50.549) (57.764)

Total User Fees pc ($) -9.253 -6.503 -5.835 1.093 102.13(9.332) (9.235) (8.546) (12.246)

Wastewater User Fees pc ($) 7.161** 7.696*** -0.202 0.375 30.6(2.830) (2.918) (2.228) (3.921)

Long Term Debt pc ($) 101.313 162.409 -33.417 77.547 1385.92(118.578) (122.834) (134.979) (197.709)

Short Term Debt pc ($) 36.285* 30.404 24.645 -36.030 1385.92(20.427) (22.233) (18.495) (22.131)

Controls Y YObservations 2578 2578 2578 2578Note: Table provides estimates of β from: fit = β(Controli× ˜POST t) +µi + τt + εit where fit is a characteristic for city i in year t; ˜POST is anindicator equal to 0 in 1967 and 1 in 1972; and β is the mean difference in pre-CWA growth from 1967 to 1972 among control cities relative totreated cities. Column (1) reports estimates of β when “Control” equals 1 if a city had secondary treatment as of 1972 (i.e., compliant treatmenttechnology). Column (2) reports estimates of β when “Control” equals 1 if a city has a higher than median probability of having a complianttreatment plant prior to the CWA’s adoption on the basis of the instruments: Downstream Population, StateShare’72, and their interaction.Sample includes only pre-CWA years, 1967 and 1972. Federal Construction Grants pc include federal grants for wastewater treatment as wellas disaster relief, homeland security, and miscellaneous goods. “Controls” include all controls listed in Table 2, column (5). Standard errorsclustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01†Regressions include census years 1960 and 1970. Includes 4354 observations. Controls include only characteristics observed prior to 1967,including city and year fixed effects, region linear time trends, time trends in distance to ocean, time trends in river network population, andtime trends in hydrologic region.‡Regressions include years 1956 and 1970. Includes 4710 observations. Controls are same as the population regression.∓ Includes 1820 observations, due to limited sampling in STORET and NWIS prior to 1970. Includes all controls in Table 2, column (5).

50

Table 4: Effect of CWA Mandate on Local Government Budgets

PANEL A: DIFFERENCE-IN-DIFFERENCESExpenditures Per Capita Total Wastewater Other

Total Capital Operating Total Public Safety Public Works Gen Admin Welfare Rec

Primary’72xPost 69.242∗ 63.274∗∗∗ 59.907∗∗∗ 2.855 -24.309 5.320 -42.580 10.866∗ 1.601 0.483(41.948) (9.951) (9.549) (2.192) (35.456) (3.238) (33.593) (5.624) (4.539) (2.342)

Marginal effect (%) 6.72% 94.89% 155.04% 10.31% -3.74% 4.01% -11.12% 17.66% 5.45% 1.12%Baseline mean 1031.13 66.68 38.64 27.70 649.23 132.55 382.75 61.53 29.36 43.03

Revenues Per Capita Total Revenues User Fees Taxes Debt

Total Federal Grants Own Total Wastewater Total Property Sales & License Long Term Short Term

Primary’72xPost 49.880 20.832∗∗∗ -11.867 18.470∗ 5.169∗∗ 2.720 -5.712 8.367∗ 53.529 12.382(30.808) (4.882) (27.817) (10.571) (2.563) (9.944) (8.927) (4.488) (101.029) (13.986)

Marginal effect (%) 4.90% 71.10% -1.40% 16.62% 16.59% 0.69% -1.89% 9.39% 3.85% 13.93%Baseline mean 1017.71 29.30 847.01 111.15 31.15 391.54 302.42 89.12 1388.86 88.87

PANEL B: INSTRUMENTED DIFFERENCE-IN-DIFFERENCESExpenditures Per Capita Total Wastewater Other

Total Capital Operating Total Public Safety Public Works Gen Admin Welfare Rec

Primary’72xPost 326.412 155.215∗∗∗ 115.533∗∗ 38.849∗∗ 217.011 42.309∗∗ 89.206 15.409 59.863∗ 10.225(245.763) (54.807) (49.069) (18.527) (186.409) (19.167) (168.414) (28.727) (31.331) (12.351)

Marginal effect (%) 32% 233% 299% 140% 33% 32% 23% 25% 204% 24%Baseline mean 1031.13 66.68 38.64 27.70 649.23 132.55 382.75 61.53 29.36 43.03

Revenues Per Capita Total Revenues User Fees Taxes Debt

Total Federal Grants Own Total Wastewater Total Property Sales & License Long Term Short Term

Primary’72xPost 718.791∗∗∗ 173.506∗∗∗ 477.192∗∗ -34.366 38.582∗∗ 183.602 133.478 50.212∗ 1103.274 -37.146(237.873) (45.065) (214.666) (74.402) (19.424) (132.225) (127.521) (27.127) (963.795) (84.825)

Marginal effect (%) 71% 592% 56% -31% 124% 47% 44% 56% 79% -42%Baseline mean 1017.71 29.30 847.01 111.15 31.15 391.54 302.42 89.12 1388.86 88.87

First Stage F-statistic 17.44 17.44 17.44 17.44 17.44 17.44 17.44 17.44 17.44 17.44

Controls Y Y Y Y Y Y Y Y Y YClusters 2975 2975 2975 2975 2975 2975 2975 2975 2975 2975Observations 14866 14866 14866 14866 14866 14866 14866 14866 14866 14866Note: Dependent variables are in 2012 dollars per capita. Table reports estimates of β from Eq. 1 in Panel A and βIV from Eq. 2 in Panel B. “Controls” include all controls listed in Table 2,column(5). Baseline means are the average budget line outcome among treated cities from 1967 to 1972. Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

51

Table 5: Effect of CWA Mandate on Local Government Growth

OLS Two Stage Least Squares

(1) (2) (3) (4) (5)

Panel A: Dissolved Oxygen (mg/l) (N=13,964)

Primary’72xPost 0.088 1.005∗∗ 0.857∗∗ 0.824∗∗ 1.700∗∗∗(0.055) (0.391) (0.388) (0.335) (0.450)

x (1st pop tercile) -0.067 -0.033(0.122) (0.135)

x (2nd pop tercile) -0.085 -0.026(0.099) (0.112)

First Stage F-statistic 14.611 14.206 5.266 3.300Baseline mean: 7.77mg/l

Panel B: Ln(Median House Price) (N=8,247)

Primary’72xPost -0.021∗∗∗ 0.119∗∗ 0.106∗ -0.008 -0.070(0.007) (0.060) (0.059) (0.050) (0.058)

x (1st pop tercile) 0.161∗∗∗ 0.161∗∗∗(0.016) (0.015)

x (2nd pop tercile) 0.094∗∗∗ 0.092∗∗∗(0.014) (0.013)

First Stage F-statistic 16.959 16.474 6.157 4.083Baseline mean: $95,603

Panel C: Ln(Population) (N=8,247)

Primary’72xPost -0.033∗∗∗ 0.206∗∗∗ 0.232∗∗∗ 0.125∗ 0.158∗(0.012) (0.079) (0.082) (0.073) (0.089)

x (1st pop tercile) 0.140∗∗∗ 0.141∗∗∗(0.025) (0.026)

x (2nd pop tercile) 0.151∗∗∗ 0.153∗∗∗(0.023) (0.023)

First Stage F-statistic 16.959 16.474 6.157 4.083Baseline mean: 36,635

Controls Y Y Y Y YUpstream & Within 50mi Population x Year FE Y Y YState Debt Rule x YearFE YClusters 2959 2959 2959 2959 2959Note: Dependent variables are dissolved oxygen in Panel A; ln(median house prices) in Panel B; andln(population) in Panel C. Table reports estimates of βIV from Eq. 2. 1st and 2nd population tercile areindicators for whether a city’s 1970 population is less than 2,000 and 10,000, respectively. Panel B andPanel C estimates based on regressions that include decade interval years only (1972, 1982, 1992). Baselinemeans are the average of the dependent variable among treated cities as of 1970. Dollars in USD 2012values. “Controls” include all controls listed in Tab. 2, column(5). “Upstream & Within 50km Populationx Year FE” are controls, individually, for total population within a 50 mi radius interacted with yearFE and population upstream interacted with year FE. “State Debt Rule” defined in footnote of Tab. 2.Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

52

Table 6: Decomposing Own, Spillover, and Competition Effects

Ln(Median House Price)

(1) (2) (3)

Treated x Post δ1: 0.106∗ 0.135∗∗ 0.009(0.059) (0.066) (0.054)

Upstream Treated Pop x Post δ2: 0.039 0.037(Spillover & General Equilibrium) (0.026) (0.026)

Within 50mi Treated Pop x Post δ2: -0.015∗∗∗ -0.012∗∗(General Equilibrium) (0.006) (0.005)

Local Effects by Base Population Tercile

Treated x Post (1st tercile) 0.162∗∗∗(0.016)

Treated x Post (2nd tercile) 0.095∗∗∗(0.014)

Controls Y Y YUpstream & Within 50mi Population x Year FE Y Y YSpillover (δ2 - δ3) 0.054∗ 0.048∗

(0.028) (0.027)

F-statistic 16.47 14.09 5.33Observations 8247 8247 8247Note: Dependent variable is ln(house median house price). Table reportsestimates of δ1, δ2, and δ2 from Eq. 6. 1st tercile is 1970 population lessthan 2,000. 2nd tercile is 1970 population less than 10,000. “Controls”include all controls listed in Tab. 2, column(5). All regressions additionallyinclude controls for time trends in the local and upstream labor marketconcentration: N50mi × τt and NUS × τt. Standard errors clustered by city.∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 7: Benefit-to-Cost Calculation of CWA Mandate

Net Benefits ($bn)Local $307.3 $307.3

+ Spillover $27.2 $27.2- GE $127.8 $127.8

Net Benefits $206.4 $206.4

Cost Estimates ($bn)Federal & State $98Industrial $366Federal Grants $205.4

× MCPF 1.4 1.4

Total Cost $503.2 $287.5

Ratio 0.41 0.72Note: Net benefits calculated using estimates from Tab. 6 column (2) com-bined with municipality-specific median housing values and housing unitcounts as of 1970 from the census. Federal & State costs calculated as thetotal US expenditures spent on surface water pollution abatement in Tab.1 of Keiser & Shapiro (2019) less local costs reported by Anderson (2010)Tab. 6, converted to 2012 US dollars. Industrial costs sourced from Fig.1 of Keiser et al. (2019). Federal Grants cost sourced from EPA GrantsInformation and Control System (GICS) database, outlays from 1972-1992adjusted to 2012 USD. Marginal Cost of Public Funds (MCPF) value of 1.4is the mean value reported in Dahlby (2008), Tab. 5.3.

53

Appendix A Figures and Tables

Figure A1: Population Distribution of Local Government Sample. Source: US Census of Governments. Figure showsthe cumulative distribution function of the sample local government population size, and the corresponding share of theUS population as of 1970 and 2012. Includes 3,334 governments categorized as municipalities and townships by the Censusof Governments, which operate a wastewater treatment plant. See Section 3 and Appendix Section B for further detailson sample selection.

Figure A2: Wastewater Expenditures per capita by Size of Downstream Population Using State Compliant Plant ShareVariation. Source: USGS, Census of Governments, author’s calculations. Figure plots δt×50 + δt and δt from equation:yit =

∑tδt×50(I50 × Ss ×Dt) +

∑tδt(Ss ×Dt) + (Di ×Dt) + γt + νi + εit where the dependent variable is wastewater

expenditures per capita for city i in year t; I50 is an indicator for a city having downstream population in the bottom50th percentile, Ss is state share of compliant plants as of 1972, Dt is an indicator for year t, and Di is downstreampopulation. Bands show 95% confidence intervals. The reference year is t=1972 and the reference city is one in a lowcompliant state. Black triangles show the estimated difference in expenditures in year t relative to 1972, relative to highdownstream-high compliant state cities, and all low compliant state cities, for cities with downstream population size inthe bottom 50th percentile. Gray estimates show the difference in expenditures in year t relative to 1972 for cities withdownstream population size in the top 50th percentile, relative to low downstream-high compliant state cities, and all lowcompliant state cities. Robust standard errors are clustered at the city level. The gray bars show that cities in the top50th percentile of downstream population within high compliant states have significantly less wastewater expenditures after1972 relative to all other cities. This is due to the fact that downstream population is less predictive of ex ante compliancein low compliant states. In other words, all cities in low compliant states, regardless of their downstream population size,were more likely to be noncompliant. Consequently, their expenditures are not significantly different from low downstreampopulation cities in high compliant states.

54

Figure A3: Plant Service Population vs Local Government Population. Source: US Census of Governments; EPACWNS. Figure shows relationship between local government population as reported by Census of Governments, and servicepopulation of its corresponding plant as reported by EPA. Includes 3,226 governments with population less than 30,000.Regression coefficient (r) estimated from Census population =r Plant population + e. Each dot represents approximately60 cities.

55

Table A1: Specification Sensitivity of CWA Mandate Effect

Panel A: Wastewater Expenditures per capita (N = 14, 866) (1) (2) (3) (4)

Primary’72xPost 48.946∗∗ 99.263∗∗∗ 128.011∗∗∗ 140.412∗∗∗(23.193) (35.069) (49.539) (46.999)

Panel B: Dissolved Oxygen (N = 13, 976) (1) (2) (3) (4)

Primary’72xPost -1.223∗∗∗ 0.710∗∗ 0.553∗∗∗ 1.375∗∗∗(0.335) (0.339) (0.060) (0.337)

Panel C: Ln(Population) (N = 8, 264) (1) (2) (3) (4)

Primary’72xPost -1.584∗∗∗ 0.193∗∗∗ 0.084∗∗∗ 0.128∗(0.386) (0.070) (0.017) (0.069)

Panel D: Ln(Med. House Price) (N = 8, 264) (1) (2) (3) (4)

Primary’72xPost -0.284∗∗∗ -0.213∗∗∗ 0.274∗∗∗ 0.015(0.100) (0.056) (0.019) (0.052)

YearFE Y Y Y YCityFE Y Y YRegion & Income Trend Y YBaseline Controls YNote: Dependent variables are wastewater expenditures per capita, five-year average dissolved oxygen in mg/l, ln(population)and ln(median house price) in the Panels A, B, C, and D, respectively. Table reports estimates of βIV from Eq. 2.Ln(Population) and Ln(Med. house price) estimates include decade interval years only (1972, 1982, 1992). Region & In-come trends include (1) region-specific indicators based on BEA US regions: New England, Mideast, Great Lakes, Plains,Southeast, Southwest, Rocky Mountain, and Far West each interacted with a linear time trend; and (2) county-level averageincome per capita, averaged from 1967 and 1972, interacted with a linear time trend. Baseline controls include time trends ofseveral pre-CWA characteristics (averaged from 1967-1972): share of employment in water-polluting industries, growth from1956 to 1970 in manufacturing employment; and annual federal, state, local intergovernmental grants; and distance from citycentroid to coastline (ie, all controls in Tab. 2, column (4)). Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗p < 0.01

56

Table A2: Effect of CWA Mandate on Local Government Budgets with State x Year FE

PANEL A: Expenditures Per CapitaTotal Wastewater Other

Total Capital Operating Total PublicSafety PublicWorks GenAdmin Welfare Rec

Primary’72xPost 598.041∗ 81.054 53.487 25.174 318.448 -10.810 295.661 -30.659 73.292 -9.037(357.307) (74.949) (66.719) (25.378) (268.918) (51.357) (254.071) (23.847) (50.094) (20.536)

PANEL B: Revenues Per CapitaTotalRevenues UserFees Taxes Debt

Total FedGrants Own Total Wastewater Total Property Sales&License LongTerm ShortTerm

Primary’72xPost 553.241∗ -10.676 472.877∗ -3.051 25.835 -37.354 -81.396 44.023 1902.494 -519.559∗∗(294.804) (42.501) (256.795) (99.442) (26.920) (77.230) (63.640) (42.387) (1608.830) (223.448)

Controls Y Y Y Y Y Y Y Y Y YState x Year FE Y Y Y Y Y Y Y Y Y YF-statistic 10.36 10.36 10.36 10.36 10.36 10.36 10.36 10.36 10.36 10.36Clusters 2975 2975 2975 2975 2975 2975 2975 2975 2975 2975Obs. 14866 14866 14866 14866 14866 14866 14866 14866 14866 14866Note: Dependent variables are in 2012 dollars per capita. Table reports estimates of βIV from Eq. 2. “Controls” include all controls listed in Tab. 2, column(5),as well as interaction of state x year fixed effects. Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table A3: Effect of CWA Mandate on Local Government Budgets with State Debt Rules x Year FE

PANEL A: Expenditures Per CapitaTotal Wastewater Other

Total Capital Operating Total PublicSafety PublicWorks GenAdmin Welfare Rec

Primary’72xPost 504.439∗ 166.393∗∗∗ 113.274∗∗ 54.372∗∗∗ 354.722∗ 52.008 171.544 67.567 54.854∗ 8.750(264.373) (62.760) (56.406) (20.220) (196.435) (32.421) (166.942) (41.193) (33.020) (15.473)

PANEL B: Revenues Per CapitaTotalRevenues UserFees Taxes Debt

Total FedGrants Own Total Wastewater Total Property Sales&License LongTerm ShortTerm

Primary’72xPost 892.322∗∗∗ 191.950∗∗∗ 630.292∗∗ -33.178 54.316∗∗ 323.044∗ 269.689 53.628∗ 1892.010 -159.298(317.435) (52.968) (286.294) (83.663) (23.138) (194.497) (189.888) (32.422) (1378.153) (103.647)

Controls Y Y Y Y Y Y Y Y Y YState Debt Rule x Year FE Y Y Y Y Y Y Y Y Y YF-statistic 11.19 11.19 11.19 11.19 11.19 11.19 11.19 11.19 11.19 11.19Clusters 2975 2975 2975 2975 2975 2975 2975 2975 2975 2975Obs. 14866 14866 14866 14866 14866 14866 14866 14866 14866 14866Note: Dependent variables are in 2012 dollars per capita. Table reports estimates of βIV from Eq. 2. “Controls” include all controls listed in Tab. 2, column(5),as well as interaction of state debt rules x year fixed effects. “State Debt Rule” defined in footnote of Tab. 2. Standard errors clustered by city. ∗ p < 0.10, ∗∗p < 0.05, ∗∗∗ p < 0.01

Table A4: Effect of CWA Mandate on Local Government Budgets with LCV Score x Year FE

PANEL A: Expenditures Per CapitaTotal Wastewater Other

Total Capital Operating Total PublicSafety PublicWorks GenAdmin Welfare Rec

Primary’72xPost 358.906 167.726∗∗∗ 125.494∗∗ 41.899∗∗ 203.682 46.077∗∗ 75.965 13.187 61.763∗ 6.689(273.097) (60.953) (54.230) (20.614) (201.777) (22.379) (181.373) (32.818) (34.346) (13.928)

PANEL B: Revenues Per CapitaTotalRevenues UserFees Taxes Debt

Total FedGrants Own Total Wastewater Total Property Sales&License LongTerm ShortTerm

Primary’72xPost 773.731∗∗∗ 181.802∗∗∗ 464.616∗ -48.076 49.831∗∗ 144.901 99.484 45.497 1569.240 -46.707(267.549) (49.385) (237.447) (85.577) (21.794) (140.428) (135.599) (30.292) (1153.702) (94.757)

Controls Y Y Y Y Y Y Y Y Y YLCV Score x Year FE Y Y Y Y Y Y Y Y Y YF-statistic 14.71 14.71 14.71 14.71 14.71 14.71 14.71 14.71 14.71 14.71Clusters 2975 2975 2975 2975 2975 2975 2975 2182975 297583 2975Observations 14866 14866 14866 14866 14866 14866 14866 14866 14866 14866Note: Dependent variables are in 2012 dollars per capita. Table reports estimates of βIV from Eq. 2. “Controls” include all controls listed in Tab. 2, column(5) aswell as 1971-72 state League of Conservation Voters (LCV) Score (http : //scorecard.lcv.org/scorecard/archive). Standard errors clustered by city. ∗ p < 0.10,∗∗ p < 0.05, ∗∗∗ p < 0.01

57

Table A5: Effect of CWA Mandate on Local Government Budgets, Excluding Coastal Cities

PANEL A: Expenditures Per CapitaTotal Wastewater Other

Total Capital Operating Total PublicSafety PublicWorks GenAdmin Welfare Rec

Primary’72xPost 346.254 86.647 56.197 31.132∗ 307.917∗ 27.728 172.420 11.022 77.819∗∗ 18.927(256.332) (58.612) (56.119) (18.189) (184.760) (18.768) (160.812) (32.712) (33.884) (13.923)

PANEL B: Revenues Per CapitaTotalRevenues UserFees Taxes Debt

Total FedGrants Own Total Wastewater Total Property Sales&License LongTerm ShortTerm

Primary’72xPost 716.497∗∗∗ 189.352∗∗∗ 505.120∗∗ -0.151 29.278 250.417 169.121 81.352∗∗∗ 795.145 -53.786(274.404) (54.073) (250.239) (84.777) (20.556) (161.882) (157.468) (27.828) (1100.366) (86.162)

Controls Y Y Y Y Y Y Y Y Y YF-statistic 13.78 13.78 13.78 13.78 13.78 13.78 13.78 13.78 13.78 13.78Clusters 2553 2553 2553 2553 2553 2553 2553 2553 2553 2553Obs. 12564 12564 12564 12564 12564 12564 12564 12564 12564 12564Note: Dependent variables are in 2012 dollars per capita. Excludes cities located within 50 km of a coastline. Table reports estimates of βIV from Eq. 2.“Controls” include all controls listed in Table 2, column(5). Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table A6: Effect of CWA Mandate on Local Government Budgets, Excluding Ohio and MississippiWatershed

PANEL A: Expenditures Per CapitaTotal Wastewater Other

Total Capital Operating Total PublicSafety PublicWorks GenAdmin Welfare Rec

Primary’72xPost 270.005 194.951∗∗∗ 142.354∗∗ 52.917∗∗ 103.536 48.247 -104.630 52.170 72.718∗∗ 35.031∗(310.458) (71.087) (62.610) (21.008) (240.029) (48.417) (213.302) (43.830) (36.755) (18.270)

PANEL B: Revenues Per CapitaTotalRevenues UserFees Taxes Debt

Total FedGrants Own Total Wastewater Total Property Sales&License LongTerm ShortTerm

Primary’72xPost 629.428∗∗ 159.209∗∗∗ 474.635∗ 77.479 55.203∗∗ 295.589∗ 137.857 158.170∗∗∗ 1644.020 -136.693(297.573) (54.267) (263.963) (81.943) (23.287) (167.622) (154.649) (46.832) (1122.285) (118.719)

Controls Y Y Y Y Y Y Y Y Y YF-statistic 11.54 11.54 11.54 11.54 11.54 11.54 11.54 11.54 11.54 11.54Clusters 2183 2183 2183 2183 2183 2183 2183 2183 2183 2183Obs. 11084 11084 11084 11084 11084 11084 11084 11084 11084 11084Note: Dependent variables are in 2012 dollars per capita. Excludes cities located within HUC 5 and 7 (the Ohio River and Mississippi watersheds). Table reportsestimates of βIV from Eq. 2. “Controls” include all controls listed in Tab. 2, column (5). Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table A7: Effect of CWA Mandate on Local Government Budgets Using Full Sample

PANEL A: Expenditures Per CapitaTotal Wastewater Other

Total Capital Operating Total PublicSafety PublicWorks GenAdmin Welfare Rec

Primary’72xPost 1030.885 113.473 52.433 59.284∗∗ 445.543∗ 79.295∗∗∗ 290.658 15.047 23.711 36.832∗(636.863) (81.764) (74.568) (28.474) (240.805) (26.544) (219.634) (29.283) (37.866) (18.907)

PANEL B: Revenues Per CapitaTotalRevenues UserFees Taxes Debt

Total FedGrants Own Total Wastewater Total Property Sales&License LongTerm ShortTerm

Primary’72xPost 1219.948∗∗∗ 188.476∗∗∗ 787.733∗∗∗ 15.673 55.339∗∗ 250.768∗ 231.516∗ 19.706 1661.706 -24.438(333.335) (62.562) (286.091) (105.571) (26.348) (131.098) (126.629) (33.241) (1424.464) (106.490)

Controls Y Y Y Y Y Y Y Y Y YF-statistic 17.98 17.98 17.98 17.98 17.98 17.98 17.98 17.98 17.98 17.98Clusters 8957 8957 8957 8957 8957 8957 8957 8957 8957 8957Observations 40563 40563 40563 40563 40563 40563 40563 40563 40563 40563Note: Sample includes population of municipal wastewater treatment plants with primary or secondary treatment listed in the CWNS survey, including thosethat were built after 1972. Dependent variables are in 2012 dollars per capita. Table reports estimates of βIV from Eq. 2. “Controls” include all controls listedin Table 2, column(5). Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

58

Table A8: Determinants of Ex Ante CWA Mandate Compliance Status Using Full Sample

Cross Section (1972) Panel

(1) (2) (3) (4) (5) (6) (7)

Downstream Population -0.026∗∗∗ -0.026∗∗∗(0.006) (0.006)

Dowstream Population x StateShare’72 x Post -0.520∗∗∗ -0.380∗∗ -0.273 -0.277 0.033(0.174) (0.178) (0.195) (0.197) (0.235)

Dowstream Population x Post -0.002 -0.002 -0.009 -0.009 -0.034∗∗(0.009) (0.009) (0.010) (0.011) (0.013)

StateShare’72 x Post -1.190∗∗∗ -1.306∗∗∗ -1.558∗∗∗ -1.570∗∗∗ -1.370(0.179) (0.184) (0.256) (0.319) (2.025)

River FE Y YBaseline Controls YCity & YearFE Y Y Y Y YRiverPopulation x YearFE Y Y Y Y YBaseline Controls x YearFE Y Y Y YIncome Trend Y Y Y YRegion Trend Y Y Y YWatershed x Year FE Y YState Debt Rule x Year FE YState x Year FE YChange in P(noncompliance) given:

1 SD increase DSpop -2.94% -2.89% -2.25% -1.72% -2.02% -2.03% -3.60%1 SD increase StateShare -5.13% -5.06% -5.49% -5.54% 0.10%

Kleibergen-Paap F-statistic 24.048 23.27 17.98 11.973 5.845Stock-Yogo critical value 22.30 22.30 22.30 22.30 22.30Observations 5282 5282 40563 40563 40563 40563 40563Note: Sample includes population of municipal wastewater treatment plants with primary or secondary treatment listed in the CWNS survey,including those that were built after 1972. The dependent variable in (1) and (2) is an indicator for primary treatment as of 1972. Thedependent variable in (3)-(7) is an indicator for primary treatment as of 1972 interacted with a post CWA indicator. The “River” fixed effectis denoted by the terminating point of the city’s nearest river. The Baseline Controls in specifications (1) and (2) include: distance to theriver terminating point, pre-CWA (averaged from 1967-1972) intergovernmental grant receipt, growth from 1956 to 1970 in manufacturingemployment, share of pre-CWA employment in water pollution industries, and distance to ocean. River Population is size of total river networkpopulation as of 1970. Baseline Controls in columns (3)-(7) include time trends of several pre-CWA characteristics (averaged from 1967-1972):share of employment in water-polluting industries, growth from 1956 to 1970 in manufacturing employment; and annual federal, state, localintergovernmental grants; and distance from city centroid to coastline. Income trend is county-level average income per capita, averaged from1967 and 1972, interacted with a linear time trend. Region trend consists of 8 indicators based on BEA US regions: New England, Mideast,Great Lakes, Plains, Southeast, Southwest, Rocky Mountain, and Far West each interacted with a linear time trend. Watershed includes one of18 hydrologic units across the coterminous US. State Debt Rule takes one of three values for none, statutory, or constitutional state requirementsas of 1970 that local governments balance their budget at the end of each fiscal year (source: Bohn & Inman 1996). Downstream populationis normalized by its standard deviation. Stock-Yogo critical values based on the weak identification values for 10% maximal IV size. Marginaleffects evaluated at the baseline mean of noncompliance (90%). Standard errors clustered by city. ∗ (p<0.10), ∗∗ (p<0.05), ∗∗∗ (p<0.01).

59

Table A9: Local Government Growth Robustness Checks

(1) (2) (3) (4) (5)

Panel A: Dissolved Oxygen (mg/l)

Primary’72xPost 1.830∗∗∗ 1.075∗∗ 0.442 0.740∗ 1.860∗∗∗(0.547) (0.442) (0.447) (0.438) (0.514)

Observations 13976 13976 11811 10354 37747F-statistic 10.814 11.974 11.291 8.902 16.154

Panel B: Ln(Median House Price)

Primary’72xPost 0.003 0.135∗∗ 0.060 0.109 0.260∗∗∗(0.089) (0.066) (0.064) (0.079) (0.080)

Observations 8264 8264 7043 6120 23601F-statistic 9.518 14.627 13.499 10.946 17.331

Panel C: Ln(Population)

Primary’72xPost -0.061 0.233∗∗∗ 0.207∗∗ 0.249∗∗ 0.306∗∗∗(0.105) (0.088) (0.087) (0.105) (0.112)

Observations 8264 8264 7043 6120 23601F-statistic 9.518 14.627 13.499 10.946 17.331

Controls Y Y Y Y YState x Year FE YLCV Score x Year FE YExcl. coastal cities YExcl.HUC5,7 YFull Sample Y

Note: Dependent variables are: five-year lagged average dissolved oxygen in mg/l, ln(median house price), andln(population) for panels A, B, and C, respectively. Housing price and population regressions include decade intervalyears only (1972, 1982, 1992). Table reports estimates of βIV from Eq. 2. “Controls” include all controls listed in Tab. 2,column(5). “LCV” is state League of Conservation Voters Score as of 1971-1972. “Excl. coastal cities” excludes all citieswithin 50 km of a coastline. “Excl.HUC5,7” excludes cities in the Ohio and Mississippi watersheds. “Full Sample” includesall municipal wastewater treatment plants with primary or secondary treatment listed in the CWNS survey, includingthose that were built after 1972. Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

60

Table A10: Effect of CWA Mandate on Skill Composition

OLS Two Stage Least Squares

(1) (2) (3) (4) (5)

Panel A: Ln(Population with a College Degree) (N=7,602 )

Primary’72xPost -0.033 0.152 0.144 0.033 -0.070(0.021) (0.121) (0.122) (0.115) (0.127)

x (1st pop tercile) 0.546∗∗∗ 0.517∗∗∗(0.136) (0.130)

x (2nd pop tercile) 0.137∗∗∗ 0.133∗∗∗(0.034) (0.033)

First Stage F-statistic 14.213 13.974 5.425 4.343Baseline mean: 2,553

Panel B: Share of Population with College Degree (N=7,602 )

Primary’72xPost -0.004∗ 0.020 0.017 0.008 -0.002(0.002) (0.013) (0.013) (0.012) (0.013)

x (1st pop tercile) 0.002 -0.000(0.012) (0.011)

x (2nd pop tercile) -0.005 -0.005∗(0.003) (0.003)

First Stage F-statistic 14.213 13.974 5.425 4.343Baseline mean: 10.7%

Controls Y Y Y Y YUpstream & Within 50mi Population x Year FE Y Y YState Debt Rule x YearFE YClusters 2949 2949 2949 2949 2949

Note: The dependent variable is ln(population with a college degree) in Panel A and share of population with a collegedegree in Panel B. Sample includes panel of cities with annual observations for educational attainment in each decade. 1stand 2nd population tercile are indicators for 1970 population less than 2,000 and 10,000, respectively. Regressions includedecade interval years only (1972, 1982, 1992). Table reports estimates of βIV from Eq. 2. “Upstream & Within 50kmPopulation x Year FE” are controls, individually, for total population within a 50 mi radius interacted with year FE andpopulation upstream interacted with year FE. “State Debt Rule” defined in footnote of Tab. 2. “Controls” include allcontrols listed in Tab. 2, column(5). Standard errors clustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

61

Appendix B Data Sampling Restrictions

First, I restrict my analysis to the roughly 8,318 facilities (out of 20,361 total municipal plants) that

were listed as operational wastewater treatment plants as of 1972. Thus, my analysis does not include

cities that built a wastewater treatment plant after the CWA came into effect. This restriction

increases the likelihood that compliant and noncompliant cities shared important ex ante unobservable

characteristics that determine economic growth such as willingness of their taxpayer base to invest in

long-lasting infrastructure projects. I further exclude plants that ceased operation over time by

including only facilities that appear in each decade and in at least half of the 13 surveys between 1972

and 2003. This restriction further drops 22% of the facilities that appeared in the 1972 survey, leaving

6,440 plants. To reduce potential instances of measurement error or misreporting, I exclude wastewater

treatment plants that did not meet all of the following criteria: maintains facility type “wastewater

treatment plant” as opposed to sewer system, septic, or other (excludes 12.6% of facilities); reports

having wastewater treatment plant technology and is recorded as a wastewater treatment facility

(excludes 2.3% of facilities); continues to have a plant in a given year if it had a plant in the prior year

(excludes 3.7% of facilities), and does not downgrade technology type from secondary to primary

(excludes 19% of facilities). These additional sample restrictions eliminate approximately 2,462 plants.

These sample restrictions serve to reduce measurement error of treatment plant technology and

help to ensure that variation across my treatment and control cities is driven primarily by differences

in the CWA technology standard, as opposed to cyclical infrastructure degradation, or structural

municipal decline. Appendix Tab. A11 shows descriptive statistics comparing my restricted sample to

the full population of municipal treatment plants. By utilizing only continuously operating plants, the

population size of cities in my analysis is larger, on average, than the mean plant-operating

municipality. My sample of cities also has larger budgets, a more educated/higher income population,

and they are located slightly closer to coastlines compared to the population of municipalities with

plants. As a robustness check, I re-estimate my main empirical specifications without imposing either

the 1972 criteria or the misreporting exclusions. I discuss these results in Section 5.3.

62

Table A11: Pre-Clean Water Act Descriptive Statistics

Sample Population P-value fordifference in means

City CharacteristicsPopulation 35,075.293 8,983.219 0.000Median House Price ($) 97,469.633 83,975.008 0.000Share of population with a college degree 0.111 0.096 0.000Dissolved oxygen (mg/l) 7.782 7.912 0.001League of Conservation Voters score 45.106 40.750 0.502

County-level labor marketCounty income per capita ($) 23,345.873 21,959.055 0.000County employment share in manufacturing 0.368 0.355 0.001County employment share in water-polluting manufacturing 0.148 0.138 0.001Manufacturing employment growth (1956-1970) -0.341 -0.249 0.000

Expenditures per capitaTotal expenditures ($) 1,069.634 939.643 0.000

Wastewater 78.823 55.897 0.000Total other 654.168 575.120 0.000

Public works 132.903 104.675 0.000Public safety 383.904 361.362 0.016General & admin. 64.846 60.927 0.024Health & welfare 29.262 17.201 0.000Recreation 43.254 30.954 0.000

Revenues per capitaTotal revenues pc ($) 1,046.335 906.366 0.000

Intergovernment revenues 180.897 158.043 0.001Revenues from own sources 865.476 748.372 0.000

Total taxes 417.781 322.316 0.000Property taxes 332.554 253.272 0.000Sales & License taxes 85.219 69.116 0.000

Total user fees 108.877 73.691 0.000Wastewater user fees 31.001 20.218 0.000

Long-term debt outstanding 1,387.677 1,479.486 0.597Short-term debt outstanding 94.886 53.042 0.000

GeographyRiver Population as of 1970 (th.) 11,677.515 13,272.658 0.000Distance to waterbody (km) 46.543 60.722 0.054Distance to river mouth (’000 km) 1,147,195.000 1,316,238.500 0.000Distance to navigable river (km) 203.900 219.154 0.002Distance to Great Lake (km) 717.817 801.116 0.000Distance to Ocean (km) 491.426 587.582 0.000

Number of Cities 2,965 6,150Panel Frequency 5.3 4.6Observations 14,863 26,216Note: All variables measured as means in 1967 and 1972. P-value denotes significance of difference in means.Dollars in USD 2012 values.

63

Appendix C Downstream & Upstream Population Calculations

C.1 Downstream Calculation

I construct the downstream population component of the instrument using digital spatial maps

sourced from the National Hydrography Dataset Plus of the US Geological Survey (USGS). These

maps contain hydrologic information for over 2.6 million stream segments averaging 1 kilometer in

length. Every river segment possesses three identifying attributes that allow me to trace out all

possible linkages in the US river system: a segment identification code, the code of the immediate

upstream river segment, and the code of the immediate downstream river segment. In addition, all

segments include an identifier for the terminal point of its river network (i.e., the river “mouth”). The

combination of network linkages across segments and terminal point identifiers allows me to identify

upstream versus downstream relationships across cities located on the same major river (e.g., the

Mississippi) as well as across cities on differing tributaries sharing the same major river basin (e.g., the

Illinois and Ohio rivers, which both feed into the Mississippi).

I assign each city centroid to its closest stream segment using GIS software. My criteria for

matching cities to a stream segment is to select the six closest stream segments to a city centroid and

assign the city to the stream segment with the lowest branching level. This approach accounts for the

tendency of cities to divert wastewater effluent into the main river segment closest to their city as

opposed to a small tributary.

I then calculate each city’s cumulative downstream population through a recursive “search tree”

algorithm as follows: I first find the terminal point, or the mouth, of each river network and assign

this segment a downstream population of xi = 0 and a current population of xj equal to the

population of a city at that mouth, if one exists. Moving upstream along stream segments, indexed by

j for current and i for the relative downstream segment, I sum the population xi of any cities located

along those segments until a branching occurs. The branch point is again treated as a temporary

“river mouth” with a downstream population of∑i

0 xi, and the process repeats itself until the source

(j = N) of the river is reached, with a total downstream population of xj +∑N

0 xi.

C.2 Upstream Calculation

I use digital maps on river networks from the National Hydrography Dataset of the USGS to observe,

for every stream segment i in the contiguous US, which stream segment is immediately upstream and

immediately downstream of that segment i. This process requires that I reverse the recursive approach

outlined in Appendix C.1 as follows: I first find the start point, or headwaters, of each river and assign

this segment an upstream population of ui = 0 and a current population of uj equal to the population

of any city at that segment, if one exists. Moving downstream along connecting stream segments,

64

indexed by j for current and i for the relative upstream segment, I sum the populations ui of any

cities located upstream of segment j until a branching occurs. Unlike the downstream calculation

approach, here I must account for the multitude of instances where several headwaters enter the same

branch point as I follow a network downstream. Thus, at a given branch point, I temporarily “hold”

the upstream population value for one stream network and “wait” for the recursive process from other

streams above the branch point to arrive at that same branch point. Once all distinct stream networks

arrive at their common branch point, I aggregate the upstream populations of each stream network so

that a given branch point has an upstream population of∑i

N ui and the process repeats itself until the

mouth of the river (j = 0) is reached with a total upstream population of uj +∑i

N ui.

65

Appendix D Dissolved Oxygen & Water Quality

I focus on dissolved oxygen as my preferred measure of water quality for two main reasons. First,

dissolved oxygen plays a crucial role in water ecosystems: insufficient levels of dissolved oxygen can

cause fish, amphibians, and plant life to die off. Because the primary goal of the CWA was to restore

and maintain the biological integrity of US surface waters and to make all water “fishable and

swimmable,” dissolved oxygen provides an holistic measure of the effectiveness of the CWA technology

mandate in meeting the CWA goals. Second, dissolved oxygen is directly impacted by municipal

sewerage. Secondary treatment can increase dissolved oxygen levels by removing harmful bacteria

from the wastewater effluent, including fecal coliforms, and nutrients such as nitrogen and

phosphorous (Minnesota Pollution Control Agency 2009). These pollutants are potentially hazardous

to human health and can induce eutrophication, thereby reducing the clarity and aesthetic value of

surface waters.

In summary, high levels of dissolved oxygen correlate with water quality attributes that are

likely to be valued by individuals, such as visual aesthetics and the opportunities for fishing and

swimming recreation. However, dissolved oxygen may not provide the most salient metric for water

quality. Visual clarity of water, for example, can be high even if the water quality is inhospitable to

aquatic life and dissolved oxygen levels are low. In absence of large fish kills or algal blooms, variation

in dissolved oxygen, nitrogen or phosphorous may be unobservable to the eye (Leggett & Bockstael

2000). Turbidity provides a closer measure of water clarity, however turbidity is not closely related to

overall ecosystem health. To the extent that the dissolved oxygen improvements from secondary

treatment are largely undetected by local residents, my estimates on the local value of water quality

from wastewater treatment infrastructure will be attenuated toward a null effect.

Appendix E Comparison of Water Quality Results

Keiser & Shapiro (2018) (K&S) focus on municipal wastewater treatment plants as the relevant

treatment unit and employ a triple difference estimation strategy to show that water quality

downstream of grant-receiving plants improved significantly more than that of plants that did not

receive federal grants. Their study isolates changes to water quality within 25 miles downstream of a

wastewater treatment plant after the plant receives an infrastructure grant. In contrast, my paper

considers changes to average surface water quality within 25 miles of the city center for cities under

pressure to comply with the CWA mandate. Both ex ante compliant and noncompliant cities could

receive EPA infrastructure grants. A second major difference is that K&S focus on dissolved oxygen

deficit (among others) as their measure of water quality, whereas I focus on dissolved oxygen in its

compound form. Because healthy levels of dissolved oxygen can differ across water bodies depending

66

on the ambient temperature, salinity, and depth, researchers sometimes consider dissolved oxygen

saturation (or dissolved oxygen deficit) as a standardized measure. I focus on the compound form to

reduce potential mis-measurement from converting dissolved oxygen to dissolved oxygen saturation,

and then aggregating up water quality readings to a city-level average. Because my empirical

approach relies on within-city variation over time, any cross sectional differences across geographic

locations with respect to their water chemistry is unlikely to bias my results.

67

Appendix F Decomposition of Water Quality Effects: Spillover & Placebo Results

I report results of Eq. (6) where yijt is water quality in Tab. A12. The first two rows provide the

effects of local (δ1) and upstream (δ2) abatement from the CWA mandate on local water quality. I

measure exposure to abatement by aggregating total upstream population in the second and third

columns as well as total number of cities upstream in the fourth and final columns. The spillover

estimate is marginally insignificant when estimated using total population upstream. The city count

exposure measure provides more precision, possibly an artifact of how secondary treatment is not

easily scale-able according to population. The columns marked “SD” normalize population and city

count by their respective standard deviations in order to compare the spillover to local treatment

effects. Water quality improvements from upstream secondary treatment adoption account for

between 5 and 6% of local adoption. Thus, most of the observed water quality improvements from the

CWA mandate are a result of local adoption.42

The magnitude of my estimated upstream-to-local abatement efforts ratio (5-6%) implies that

pollution externalities from one municipality to others were a minor part of any possible market failure

corrections addressed by the CWA. This is a surprising result in light of the fact that inadequate

wastewater treatment in the early twentieth century led to several litigation disputes across cities

(McQuillin 1912). One explanation for this puzzle is that overall surface water quality in the 1960’s

and 70’s were substantially improved relative to the early decades of the twentieth century (Stets et al.

2012). Nutrient levels, industrial waste, and high levels of sediment can all exacerbate harmful effects

of municipal waste and create adverse conditions for downstream locations. In an environment with

less ambient pollution, however, surface waters are better able to self-purify, and bacteria that escapes

primary treatment may readily digest in streams before flowing far downstream (Phelps 1914).

The last two rows provide placebo tests for the underlying mechanisms driving water quality

improvements following the CWA. If parallel trends assumptions hold, then any upstream spillover

effects on local water quality following the CWA should only come from upstream abatement efforts

(as opposed to unobserved secular changes upstream). To test this assumption, I run a specification of

Eq. (6) where I identify δ2: the upstream spillover effect among only cities with no treated upstream

neighbor. Columns (2) and (4) show that the “placebo” upstream effect is statistically insignificant

and substantially smaller in magnitude than the true upstream spillover effects shown in columns (1)

and (3). As a second placebo, I test whether local water quality improvements are affected by

mandated compliance of proximate cities located on a separate river network. The fourth row, δ3

42The biochemistry involved in the dilution of oxygen-consuming matter that passes through primary treatment can differgreatly across watersheds based on temperature, volumetric flow, and exposure to nutrients, among other variables.Consequently, the water quality results shown in Tab. A12 should be interpreted strictly as a national average ratherthan a mechanical result of secondary treatment.

68

estimates changes in local water quality resulting from exposure to treated cities within a 50 mile

radius of city i. Again, the estimates are statistically insignificant and substantially smaller in

magnitude than the local effects and the true upstream spillover effects shown in columns (1) and (3).

This suggests that variation in water quality in Eq. (6) are a result of local and upstream secondary

treatment technology adoption, as opposed to unobserved regional changes in wealth or labor markets

that might lead to spurious improvements in water quality.

Table A12: Effect of CWA Mandate on Water Quality: Spillover and Placebo Estimates

Population (mn) City Count

(1) (2) (SD) (3) (4) (SD)

δ1: Treated x Post 0.856∗∗ 0.972∗∗ 0.856∗∗ 0.871∗∗ 0.962∗∗ 0.871∗∗(0.384) (0.438) (0.384) (0.386) (0.390) (0.386)

δ2: Upstream Treated Exposure x Post 0.153 0.038 0.019∗∗ 0.053∗∗(0.113) (0.028) (0.008) (0.023)

δ2: Upstream Non-treated Exposure x Post 0.060 -0.012(0.182) (0.009)

δ3: Within50mi Treated Exposure x Post 0.014 0.002(0.037) (0.004)

Controls Y Y Y Y Y YWithin 50mi Population x Year FE Y Y Y Y Y YRatio of Spillover:Local Effect 0.045 0.060

Pre-policy mean of DO2: 7.77F-statistic 14.37 10.35 14.37 14.36 13.46 14.36Observations 13946 11553 13946 13946 11553 13946

Note: Dependent variable is dissolved oxygen (mg/l). Table reports estimates of δ1, δ2, and δ2 from Eq. 6. Columns (SD)measure Upstream and Within50mi populations in terms of standard deviations. Population values in millions. Averageupstream city has a population of 22,000. “Controls” include all controls listed in Table 2, column(5) as well as time trendsin local labor market concentration: N50mi× τt where N50mi is the population within 50 miles of the city. Standard errorsclustered by city. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

69