What the Clean Water Act Promised

The Clean Water Act promised something no American pollution law had promised before: not a reduction, not a compromise, but an end. When Congress rewrote federal water law in 1972, it wrote two goals into the statute’s opening section and dated them both. The interim aim called for water quality sufficient to protect and propagate fish, shellfish, and wildlife and to provide for recreation in and on the water, the standard everyone soon shortened to fishable and swimmable, to be achieved by July 1, 1983, wherever attainable. The national aim was blunter and carried no qualifier at all: the discharge of pollutants into the navigable waters of the United States would be eliminated by 1985. A law that set out to end pollution on a deadline is the law this article measures, and the only fair way to measure it is against the yardstick it chose for itself.

Read plainly, the two dates did different work. The 1983 target was operational: it named a condition, fishable and swimmable, that administrators, states, and courts could recognize in the field, and it softened the demand with the phrase “wherever attainable,” an acknowledgment that some waters might not get there. The 1985 target was directional: zero discharge stated where the country was supposed to be headed even if few in the Senate chamber believed every pipe would run clean within thirteen years. Together they expressed a theory of regulation with no precedent in federal water law. Earlier statutes had asked polluters to show restraint and asked states to set standards at their own pace. The 1972 law told polluters to stop and set the clock running.

The law under examination here is the Federal Water Pollution Control Act Amendments of 1972, Public Law 92-500, enacted in October 1972 over President Richard Nixon’s veto, together with its principal later amendments, including the Clean Water Act of 1977, Public Law 95-217, and the Water Quality Act of 1987, Public Law 100-4. The federal money at the center of the story flowed through the construction grants program that paid the bulk of the cost of building and upgrading municipal sewage treatment plants, later converted by the 1987 amendments into the state revolving funds that finance treatment infrastructure through low-interest loans. Those two channels, first the grants and then the revolving funds, constitute the principal federal spending the record can tie to results. The evaluation that follows takes the statute’s two great levers as given, the permit system that capped what factories and sewage plants could release and the subsidized treatment plants that made compliance possible, and asks only what they achieved in the rivers, lakes, and estuaries of the country.

Nixon’s veto belongs in the story because it fixed the terms of every later argument about cost. The president rejected the bill as too expensive, warning that its spending commitments would burden the federal budget for years. Congress overrode him the same month with majorities large enough to suggest the country had already decided the price was worth paying. That exchange set the pattern that has held ever since: the costs of the law arrive as appropriations, contracts, and sewer bills, itemized and dated, while the benefits arrive as oxygen in a river, fish in an estuary, and children swimming without getting sick, none of which comes with a receipt. Any serious accounting of the statute has to reckon with that asymmetry, and most attempts handle it badly.

The political moment matters because it explains the audacity of the dates. The first Earth Day had drawn millions into the streets in 1970. The Environmental Protection Agency was barely two years old. Senator Edmund Muskie of Maine, chairing the subcommittee that wrote the bill, had watched the paper mill towns of his home state live with rivers that stank and ran brown, and he carried that memory into the drafting room. The Clean Air Act of 1970 had already shown that Congress could write technology-forcing law, statutes that set a health-based destination and ordered industry to invent its way there. The water bill applied the same logic to rivers. Its deadlines were not predictions. They were instruments, designed to force action by making delay visible.

It is worth pausing on what the 1972 law replaced, because the contrast clarifies what “worked” has to mean. The Federal Water Pollution Control Act of 1948 and the Water Quality Act of 1965 had built a system in which states set water quality standards and the federal government could convene enforcement conferences when pollution crossed state lines. In practice the standards came slowly, the conferences rarely produced abatement, and proving that a particular discharger had caused a particular harm downstream defeated most enforcement before it began. The 1972 rewrite discarded that architecture and substituted permits, deadlines, and federal money. Whatever the monitoring record shows after 1972 has to be read against a prior regime that had produced, by wide contemporary agreement, very little.

The honest verdict, stated up front so the evidence can be read against it, has three parts. First, the measured improvement is real, large, and causally connected to the law’s spending: rivers that could not support fish in 1972 can support them now, and the gains concentrate where the federal money went. Second, the dated goals were not met in full: no one who has examined the record claims that discharges ended in 1985 or that every waterway turned fishable and swimmable by the summer of 1983. Third, the standard test for whether a regulation was “worth it,” the comparison of monetized benefits to costs, comes out against the statute on the conventional numbers and means less than it appears to mean, because the test measures what is easy to count and misses most of what the law was for. The rest of this section lays out the evidence for each part in turn: the baseline the law inherited, the improvement the monitoring record shows, and the 2019 economic assessment that gave the question its sharpest form.

Judging the law by its own aims is the demanding choice, and it is the right one. A defender of the statute might prefer to grade it against what came before, in which case it passes easily, or against the plausible alternatives available in 1972, in which case it passes by a distance. But Congress did not promise “better than 1971.” It promised fishable and swimmable by 1983 and zero discharge by 1985, and it wrote those promises into section 101(a) of the act where no later Congress could pretend they were never made. The gap between the promise and the outcome is part of the story. So is the distance the country traveled toward a destination it never reached.

The qualifier attached to the 1983 goal, “wherever attainable,” did quiet work that is often missed. It converted what looked like an absolute command into a rebuttable presumption: waters were to be made fishable and swimmable unless a state could demonstrate, through a formal attainability analysis, that the goal could not be reached for specified reasons. In practice this meant the 1983 deadline functioned less as a guillotine than as a ratchet, pulling standards upward wherever the demonstration of unattainability failed. The zero-discharge goal of 1985 carried no such safety valve in its phrasing, and its legislative history shows Congress understood the difference. Committee reports described the elimination of discharges as a national goal to be pursued through increasingly stringent technology requirements, with the understanding that each round of permits would tighten the screws further. The dates were therefore doing double duty: announcing a moral destination and creating a legal mechanism, the recurring permit cycle, that would keep moving toward it long after the calendar dates had passed.

The technology-forcing precedent came from the air side. The Clean Air Act of 1970 had set national ambient air quality standards at levels requisite to protect public health and ordered states to achieve them within fixed deadlines, on the explicit theory that regulation could drag technology forward rather than waiting for technology to make regulation convenient. The water bill’s drafters borrowed the logic openly: set the destination in the statute, assign the Environmental Protection Agency the job of writing the technology requirements, and let the deadlines do the forcing. What the water law added was money. The air statute largely ordered private parties to spend their own funds; the water statute paired its orders to cities with federal grants covering most of the construction cost, on the theory that municipalities could not finance the required plants from local taxes alone. That pairing, command plus cash, is what distinguishes the American water program from its contemporaries, and it is what makes the grant-focused evidence of this section the right lens for judging it.

The majorities that overrode Nixon’s veto were large in both chambers, a measure of how completely the burning rivers had rearranged the politics of pollution. A president who had created the EPA through Reorganization Plan No. 3 of 1970 and signed the 1970 Clean Air Act found himself outflanked on the environment by a Democratic Congress, chose to fight on cost rather than principle, and lost. The law that emerged was thus enacted over a presidential veto at spending levels that reflected a rare moment of national willingness to pay first and count later. That willingness is the reason the costs were counted so carefully ever after, and the reason the benefits, which arrived without invoices, have been argued about ever since.

What American Rivers Looked Like Before 1972

On June 22, 1969, the Cuyahoga River in Cleveland, Ohio, caught fire. Oil-soaked debris floating on the surface ignited, from a cause that was never firmly established, and burned long enough to damage two railroad bridges before firefighters put it out. The blaze itself was minor, and the city had seen worse: the river had burned in 1952 and on other occasions, each fire fed by the same cocktail of industrial oil and untreated waste. What made 1969 different was the telling. Time magazine’s account that summer described the Cuyahoga as “chocolate-brown, oily, bubbling with subsurface gases,” and the photograph the magazine ran was reportedly taken during the earlier 1952 fire. The details of the coverage matter less than what they reveal. A river that burns repeatedly, in a major American city, in the middle of the twentieth century, is not an accident. It is a system functioning as designed, carrying away whatever industry and the city poured into it.

The Cuyahoga was not the only river that burned. The Rouge River near Detroit caught fire in October 1969 as well, when oil-soaked debris on its surface ignited. Industrial waterways in other cities suffered similar fires or the constant threat of them through the 1960s, each one fed by the same combination of petroleum waste and industrial solvents that made the Cuyahoga flammable. The plural matters because it defeats the comforting thought that Cleveland was an outlier. A country whose rivers caught fire in multiple states, in multiple years, had a national condition rather than a local embarrassment, and the condition had a single cause: untreated industrial waste discharged as a matter of routine. The fires were the most photogenic symptom. The everyday symptom was the oxygen sag, the fish kills, and the stench, present in hundreds of rivers that never made the news.

The Cuyahoga became the emblem, but the condition was general. Across the industrial Midwest and Northeast, rivers served as open sewers and waste conveyors. Steel mills, paper mills, chemical plants, and meatpacking houses discharged process waste with little or no treatment. Cities discharged raw or barely treated sewage, and combined sewer systems sent a mixture of stormwater and human waste straight into rivers every time it rained hard. The pollution was not subtle. Rivers ran with floating debris and oil slicks, stank in the summer heat, and in the worst stretches could not support fish at all. People who lived near them organized their lives around the smell. On the Androscoggin River in Maine, paper mill effluent fouled the water so thoroughly that riverside towns endured a stench of rotten eggs through the warm months, fed by paper mill effluent that depleted oxygen and killed fish by the millions. Muskie knew that river. His determination to write a federal law owed something to having grown up downwind of it.

The chemistry of the damage was straightforward, and understanding it makes the later recovery legible. Organic waste, whether human sewage or industrial effluent, feeds bacteria in the water. The bacteria consume dissolved oxygen as they break the waste down. When the waste load is heavy enough, oxygen levels sag below what fish need to survive, and the river goes biologically quiet: fish kills, vanished insect life, a bottom fauna of only the hardiest worms. Engineers called the pattern the oxygen sag, and it appeared below nearly every major municipal outfall in the country. In the Delaware estuary near Philadelphia, oxygen levels in the 1960s fell to near zero across long reaches in summer, creating a chemical block that migratory fish could not swim through. In the central basin of Lake Erie, oxygen depletion in the deep water each summer suffocated bottom life across wide areas. Fish kills that would be treated as emergencies now were routine events then, cleared away and forgotten.

Lake Erie supplied the era’s most quoted verdict on the state of American waters. By the late 1960s, popular accounts were pronouncing the lake dead or dying, choked by algal blooms fed by phosphorus from sewage and from phosphate detergents, its beaches fouled, its once-great fishery collapsed in the public imagination. The pronouncements were exaggerated, commercial fishing continued, but the exaggeration took hold because the visible evidence supported it: green mats of algae, decaying windrows of it on beaches, and water that looked nothing like the lake of memory. The 1972 agreement between the United States and Canada on Great Lakes water quality, signed the same year as the domestic statute, treated phosphorus as the enemy and set the two countries to reducing it.

The Potomac gave the capital its own humiliation. President Lyndon Johnson called the river a national disgrace in 1965, and the description stuck because anyone could verify it by standing on a bridge. The river that flowed past the monuments carried the untreated and poorly treated waste of the metropolitan region, and its condition embarrassed a government that claimed to lead the world. The Potomac’s later recovery would become one of the statute’s showcase stories, which is worth remembering when reading the baseline: the river Johnson condemned is the same river whose rebound the law’s defenders cite, and both halves of that story are true.

None of this happened for lack of law. The Federal Water Pollution Control Act of 1948 had declared a federal interest in clean water, and the Water Quality Act of 1965 had required states to adopt water quality standards for interstate waters. The machinery, however, defeated itself. Standards arrived years late. Enforcement ran through conferences that produced plans more often than abatement. And the legal theory required the government to trace pollution in a river back to a particular discharger and prove the connection, a burden that dissolved in every river with more than one pipe. By 1972 the consensus in Washington, shared by the Nixon administration’s own environmental officials, was that the existing system had failed and that only a federal permit regime with fixed deadlines and federal construction money would move the needle. That consensus is what made the audacious dates of section 101(a) politically possible.

One caution about the baseline shapes everything that follows. Systematic national monitoring of water quality barely existed before the 1970s. The country did not know, in any statistically defensible sense, what share of its rivers were fishable in 1969, because no one was measuring fishability on a national grid. The baseline is therefore reconstructed from case studies, state records, industrial discharge estimates, and contemporary journalism, all of which agree on the direction and the severity but none of which yields a tidy national percentage. That evidentiary thinness cuts two ways. It means the horrors of the baseline cannot be reduced to a single number, and it means the improvements after 1972 cannot be measured against a precise starting line either. The monitoring record that does exist, built largely with the law’s own money and requirements, begins in earnest only after the law it is asked to evaluate. Any honest account of whether the statute worked has to carry that circularity in plain view.

The Hudson River supplied a different kind of baseline evidence, the kind discovered rather than seen. General Electric discharged polychlorinated biphenyls, PCBs, from its capacitor plant at Fort Edward starting in 1947 and at Hudson Falls starting in 1952, continuing through 1977; researchers detected the contamination in fish in 1969, the state issued fish-consumption advisories in 1975, and direct discharges were estimated as high as roughly 1.3 million pounds. Nothing burned and nothing stank; the river looked fine while its fish carried poison. The Hudson case previewed the limits of the 1972 law’s design: a statute built to attack visible organic waste through treatment plants would struggle with persistent toxics that accumulated silently, a struggle that belongs to the second half of this article but whose origins predate the law.

The failure of the pre-1972 regime deserves a closer look, because it explains why Congress chose permits and deadlines over gentler tools. The 1965 Water Quality Act required states to adopt water quality standards for interstate waters and authorized the federal government to call enforcement conferences when those standards were violated. The conferences were elaborate, quasi-judicial proceedings, and they almost never produced cleanup. Between 1965 and 1972 the conference mechanism ground through a handful of cases while thousands of dischargers continued as before, because the law still required proof connecting specific pollution to specific sources in specific rivers, a burden that dissolved in every waterway with more than one pipe. The lesson Congress drew was institutional rather than moral: the problem was not only that polluters were stubborn but that the legal theory made enforcement structurally impossible. The 1972 rewrite changed the theory. A discharger would need a permit, the permit would set numeric limits, and violation of the limits would be violation of the law, with no proof of downstream harm required.

The politics that made the rewrite possible had been building since the middle of the 1960s. The Santa Barbara oil spill of 1969 blackened California beaches on the evening news. The first Earth Day in April 1970 brought millions into the streets. A series of less telegenic calamities, fish kills, beach closures, drinking water scares, accumulated into a public mood that treated pollution as a national emergency rather than a local nuisance. Muskie, with national ambitions and the subcommittee gavel, had both the motive and the vehicle. The country that passed the 1972 law had spent five years watching its waters fail on television, and it voted accordingly.

The Record of Improvement After 1972

Where did the improvement show up most clearly?

Dissolved oxygen rose in many rivers and the share of monitored waters meeting fishable standards increased from 1972 onward, with the largest gains downstream of municipal plants that received federal construction grants. National totals remain estimates because monitoring coverage was uneven across states and decades.

The most direct evidence of change sits at the end of the pipes the grants rebuilt. In 1972 most American cities treated their sewage partially or not at all; primary treatment, which settles out solids but leaves dissolved organic matter largely untouched, was common, and raw discharge was not rare. The construction grants program paid up to 75 percent of the cost of building secondary treatment plants under Public Law 92-500, which use biological processes to break down organic waste before it reaches the river; Congress reduced the federal share to 55 percent in the 1981 amendments (Public Law 97-117), effective for fiscal years beginning on or after October 1, 1984. Between 1968 and 1996 the population served by at least secondary treatment rose from 85.9 million to 164.8 million, while the share receiving raw or less-than-secondary treatment fell from about 39 percent to about 9 percent; municipal plants cut their organic-material discharges 43 percent from 1972 to 1996 despite influent loads 35 percent higher. This was the statute’s central physical bet: that the worst pollution came from identifiable pipes, that those pipes belonged mostly to cities and factories, and that money plus permits could fix them. On the evidence of the rivers, the bet paid.

Boston Harbor is the story the law’s defenders tell first, and it earns the telling because a federal judge made it happen under the statute’s authority. By the early 1980s the harbor was among the foulest in the nation, receiving enormous volumes of barely treated sewage and sludge each day. In 1985 the state created the Massachusetts Water Resources Authority, and in 1986 the federal judge overseeing the case set the construction schedule for the Deer Island treatment plant, one of the largest public works projects in American history. Within years of the new plant coming online, the harbor’s condition transformed: sludge discharge ended in December 1991, upgraded primary treatment came online in January 1995, the first secondary-treatment battery followed in July 1997, and the final phase was completed in 2000. The harbor did not clean itself. A judge reading the 1972 law held a state to compliance, the money flowed, and the water changed.

The Potomac traced a longer arc from the same starting point. The river Johnson had condemned received massive treatment upgrades at the region’s plants, most notably the Blue Plains facility serving Washington, opened as a primary-treatment plant in the 1930s, expanded with secondary units in 1959, and rebuilt with advanced treatment in the 1970s to about 300 million gallons a day by 1983, and the combined effect of permits and plant construction gradually lifted the oxygen levels that had made long reaches of the river nearly lifeless. By the 1990s the Potomac supported a recovered fishery, striped bass and shad ran the river again, and the bald eagle, absent as a breeder for generations, returned to nest along its shores. The recovery was not a return to wilderness; the Potomac remained an urban river carrying the waste of millions. It became instead a working river that could sustain fish and recreation, which is precisely what the 1983 goal had asked of it.

The Delaware told a similar story in the language of chemistry. The oxygen block that had sealed the estuary near Philadelphia in the 1960s broke apart as treatment plants up and down the river upgraded under permit pressure and grant financing. Average oxygen levels near Philadelphia rose from about 2 parts per million in 1968 to 3.5 in 1981 and about 5 by 1987, and striped bass and American shad returned in large numbers in the 1990s, after some $1.5 billion in new wastewater plants along the river. The commission’s long monitoring record, one of the best in the country, shows the recovery as a steady climb across the 1970s and 1980s rather than a sudden break, which is what one would expect from a program that rebuilt plants city by city over many years.

On the West Coast, the Willamette River in Oregon became the textbook case. Through the 1960s the Willamette ran so foul below Portland’s outfalls that state officials posted warnings and the river’s summer oxygen levels collapsed. Oregon had begun its own cleanup program before 1972, and the federal law accelerated and financed what the state had started; by 1973 and 1974, United States Geological Survey summer studies found mean daily dissolved oxygen above 5 milligrams per liter throughout the river, and salmon and steelhead moved through reaches that had been written off. The Willamette mattered to the national argument because it showed the mechanism working at full strength: permits set the limits, grants built the plants, monitoring confirmed the result, and a river came back.

Lake Erie’s reversal may be the largest single achievement in the record. The 1972 agreement with Canada committed both countries to cutting phosphorus loads, and the domestic program attacked the same pollutant from the municipal side: treatment plants added phosphorus removal, and states began restricting phosphate detergents. Phosphorus loadings to the lake fell from about 29,000 to about 11,000 metric tons per year, roughly a two-thirds cut achieved by the late 1980s, and spring isothermal total phosphorus in the central basin dropped from about 20 micrograms per liter in the 1970s to the 10-microgram target by 1987. Algal blooms diminished through the late 1970s and 1980s, oxygen conditions in the central basin improved, and the walleye fishery that had been given up for dead rebounded so strongly that the lake became one of the premier walleye fisheries on the continent. The “dead lake” of the 1960s pronouncements became, by the 1980s, a lake that supported a thriving fishery and a recreation economy. Few environmental reversals anywhere have been as complete or as well documented. Beyond the showcase rivers, the national monitoring record tells a consistent story, though it requires careful reading. Beginning in the 1970s, the Environmental Protection Agency compiled the National Water Quality Inventory, the biennial report to Congress required by section 305(b) of the act, drawing on state assessments of whether waters supported their designated uses. The inventories recorded the assessed share cycle by cycle: in 1996, states assessed 693,905 river and stream miles and found 36 percent impaired; in 2000, 699,946 miles with 39 percent impaired; in 2002, 695,540 miles with 45 percent impaired; in 2004, 16 percent of the nation’s 3.5 million stream miles with 44 percent impaired. Because each cycle assessed different waters against standards that changed over time, the series is a set of snapshots rather than a trend line. The U.S. Geological Survey’s analyses of long-term monitoring stations told the recovery story in the language of chemistry: declining concentrations of fecal coliform bacteria and biochemical oxygen demand at many sites, rising dissolved oxygen, the signature of organic waste retreating from the nation’s rivers, while nutrients showed mixed trends. The modern causal analysis sharpened the point: the share of waters meeting fishable standards rose 12 percentage points between 1972 and 2001. Two independent lines of evidence, the states’ use assessments and the survey’s chemistry, pointed the same direction.

The geography of the improvement is as telling as its scale. The gains were not spread evenly like rain; they concentrated downstream of the municipal treatment plants that received federal construction money, fading with distance as dilution and natural processes took over. This gradient is exactly what the statute’s theory predicted. The law attacked point sources, the pipes, and the money rebuilt the biggest pipes first, so the rivers that ran past upgraded cities improved first and most. Industrial discharges, covered by the same permit system, added their own reductions as technology standards forced entire categories of plants to install treatment. The pattern refutes the lazy alternative explanation that the rivers cleaned themselves through deindustrialization or luck: the timing tracks the grants, the location tracks the plants, and the chemistry tracks the pollutants the permits targeted.

That said, the national numbers deserve the skepticism that all administrative data deserves, and the honest case for the statute is stronger for admitting the weaknesses. The monitoring network expanded enormously after 1972, which means the set of waters being assessed in 1995 was not the set assessed in 1975; comparing the two is comparing different rivers, not the same river twice. States have always monitored troubled waters more intensively than healthy ones, because monitoring budgets chase problems, so the assessed share of waters has always looked worse than the unassessed share would. And the standards themselves tightened over time: a river judged fishable under 1975 criteria might fail under 1995 criteria that measured more pollutants at lower detection limits. Each of these biases works against showing improvement, which makes the recorded improvement more credible rather than less, but they also mean no single national percentage should be quoted as though it were a census. The record shows a large, real, geographically patterned recovery. It does not show a number with two decimal places.

There is a second caveat that matters more, because it defines the boundary of this section. The improvements documented above concern the pollutants the 1972 law was designed to reach: organic waste, suspended solids, bacteria, phosphorus, the conventional pollutants of municipal sewage and industrial process water. These were the pollutants killing fish outright and closing beaches, and on them the record is unambiguous. What the monitoring record of the 1970s and 1980s could not yet fully show, and what the law’s permit system reached only partially, was the universe of toxic pollutants, the metals, solvents, and synthetic chemicals that persist at low concentrations and harm through accumulation rather than oxygen depletion. The statute addressed toxics through later provisions and later rules, with less money and less success. That unfinished business belongs to the second half of this article. It does not erase the oxygen that returned to the rivers, but it limits what the oxygen proves.

One more piece of the Layer 1 record deserves emphasis because it answers the question the statute’s skeptics ask first: whether the improvement would have happened anyway. The sharpest test comes from the timing of the grants themselves. Federal construction money did not arrive everywhere at once; it moved through applications, approvals, and construction schedules that stretched across two decades, so different cities upgraded their plants in different years. Rivers downstream of early-upgraded plants improved early; rivers downstream of late-upgraded plants improved late. That staggered pattern is difficult to square with any explanation that does not run through the statute’s money. General economic trends do not stagger themselves to match EPA grant disbursement schedules. The improvement has the law’s fingerprints on it, in the order the cities were rebuilt.

So far this section has told the municipal story, because the grants are where the money went and the clearest evidence sits. But the permit system reached far beyond city sewage plants, and the industrial half of the record deserves its own accounting. The 1972 law required the Environmental Protection Agency to write technology-based effluent limits for categories of industrial dischargers, numeric caps derived from what the best available treatment technology could achieve, applied uniformly to every plant in the category regardless of the condition of the receiving river. Over the following decade the agency issued guidelines across 56 industrial categories, from pulp and paper to iron and steel to organic chemicals, each one forcing the laggards in the category up toward the performance of the leaders. The reductions were enormous in the aggregate: industry’s oxygen-consuming organic discharges fell 93 percent between 1973 and 1987.

The pulp and paper industry illustrates the mechanism at its most muscular. Paper mills had been among the most visible polluters of American rivers, discharging fiber, organic solids, and chemicals that depressed oxygen for miles downstream; the Androscoggin’s stench was largely their work. The effluent guidelines for the industry, issued in the 1970s and tightened thereafter, required mills to install secondary treatment comparable to what cities were building, and the industry’s discharges fell sharply across the decade. On the toxic side, plants using the best available technology later achieved a 96 percent reduction in chloroform releases to water under the 1998 Cluster Rule. The iron and steel industry, concentrated along the rivers of the Great Lakes and the Monongahela Valley, underwent a similar transformation under its own set of guidelines, as permit limits forced treatment across the industry. These were not voluntary improvements. They were permit limits backed by the act’s enforcement provisions, and the compliance data show an industry-by-industry march toward cleaner effluent.

The 1977 amendments, Public Law 95-217, extended the permit system’s reach from the conventional pollutants to the toxic ones. The amendments grew out of a 1976 consent decree in litigation brought by the Natural Resources Defense Council, which had sued the Environmental Protection Agency for failing to regulate toxic discharges; the decree committed the agency to controlling 65 toxic pollutants across 21 industrial categories. Section 307 of the act gave the agency authority to set toxic effluent standards, and the pretreatment program required industrial dischargers to treat their waste before sending it to municipal plants, so that toxics would neither pass through city systems into rivers nor concentrate in sewage sludge. The toxics program never achieved the comprehensive success of the conventional-pollutant program; the chemistry was harder, the monitoring was thinner, and many toxic discharges continued under limits that reflected feasibility more than safety. But the direction was the same. The full story of toxics, including what the program missed, belongs to the second half of this article.

One institutional footnote clarifies what the numbers above do and do not prove. The 1987 amendments, the Water Quality Act, are sometimes described as the moment Congress discovered nonpoint source pollution, the runoff from farms, streets, and construction sites that no pipe collects. Section 319 of the amended act created a grant program for state nonpoint source management, funded at the low hundreds of millions of dollars a year (the fiscal year 2012 appropriation was $164.5 million), a fraction of the point-source spending it accompanied. The nonpoint story, the pollutants the permit system could not reach and the waters that stayed impaired because of them, is the second half’s subject. It is noted here only to fix the boundary: everything documented in this section concerns the point sources the 1972 law aimed at, and the law aimed at them because they were the sources killing rivers outright.

Enforcement gave the permits their bite, and the compliance record shows a system that largely worked as designed. The act gave the Environmental Protection Agency and the states authority to fine violators, to seek court orders, and in extreme cases to pursue criminal penalties; citizens could also sue dischargers directly, a provision that turned every riverside community into a potential enforcer. Municipal and industrial dischargers filed regular discharge monitoring reports, creating a paper trail that made violations visible in a way the old conference system never had. Compliance was imperfect: smaller municipalities struggled, and some industries treated penalties as a cost of doing business. The mechanism mattered because it solved the precise problem that had defeated the 1965 regime. No one had to prove that a particular pipe had harmed a particular fish. The permit said what the pipe could release, the monitoring report said what it released, and the difference was the violation.

The regulatory meaning of “fishable” deserves a final note, because the word does quiet analytical work throughout this section. A waterway counted as fishable when it met the state water quality standards for the propagation of fish and wildlife, standards that specified numeric criteria for dissolved oxygen, temperature, acidity, and other parameters. Meeting the standard did not mean the water was pristine, and it did not mean every species had returned; it meant the chemistry could support a functioning aquatic community, as verified by monitoring. This is a modest definition, and its modesty is a strength for evaluation purposes: when the record says a river became fishable, it reports a measured chemical fact rather than an aesthetic judgment. The gains documented above are therefore gains in the statute’s own currency, measured against the interim goal Congress set, and they show that goal being met in waterway after waterway even as it was missed as a national deadline.

The 2019 Reckoning: Gains, Costs, and the Measurement Gap

The improvement documented above is visible, large, and geographically patterned, but visibility is not causality and pattern is not proof. A skeptic could always argue that the rivers recovered for other reasons: that factories closed, that cities shrank, that the economy simply produced less waste. Answering that skeptic required something the earlier literature never quite assembled, a national dataset long enough to span the grant era and a research design capable of isolating the law’s contribution from everything else that changed after 1972. That answer arrived in 2019, when David Keiser and Joseph Shapiro published their assessment in the Quarterly Journal of Economics. It is the canonical modern treatment of the question this article asks, and it deserves to be presented whole, because its two halves are routinely quoted separately by people making opposite arguments.

The 2019 Study Documented Gains and a Cost Shortfall

A 2019 Quarterly Journal of Economics study by David Keiser and Joseph Shapiro found large water quality gains attributable to federal grants, but conventionally measured benefits fell short of costs. The authors argue standard measures omit nonuse and health values, so the shortfall reflects measurement limits rather than a verdict that the statute failed.

The paper’s first achievement was archival. Keiser and Shapiro assembled water quality readings from the Environmental Protection Agency’s STORET database and from U.S. Geological Survey monitoring, building one of the largest environmental datasets ever compiled: about 50 million individual readings from more than 240,000 monitoring sites, spanning 1962 to 2001. They tracked the pollutants the statute targeted, dissolved oxygen and its deficit, biochemical oxygen demand, fecal coliform bacteria, total suspended solids, and they tracked the regulatory bottom line as well, whether a waterway met the standards that made it legally fishable. No earlier analysis had watched American rivers this closely for this long. The dataset let them ask not only whether rivers got cleaner but when, where, and in what sequence, which is what separates a story from a test.

The data did not assemble itself, and its provenance matters for trusting it. The STORET database grew out of the monitoring requirements the statute itself imposed: permit holders had to sample their effluent, states had to assess their waters for the biennial inventory, and the resulting measurements flowed into federal repositories over decades. The Geological Survey’s National Water-Quality Assessment program, launched in 1991, added systematic long-term stations designed for trend detection rather than compliance. This means the dataset that judges the law is partly the law’s own creation, the circularity noted in the baseline section, but it also means the measurements were taken for regulatory purposes by people with no stake in flattering the statute. A monitoring record built to enforce permits is a credible witness when it reports that the permits worked.

The test itself exploited the way the grant money actually moved. Federal construction grants did not arrive everywhere at once. They moved through applications, engineering reviews, funding waves, and construction schedules that stretched across decades, so a treatment plant in one city might complete its grant-funded upgrade in 1976 while a comparable plant in another city waited until 1984. Keiser and Shapiro used that staggered timing the way a laboratory uses a control group: they compared water quality downstream of plants just before and after their upgrades against plants still waiting their turn, watching whether the rivers diverged exactly when the money arrived. The design also let them map how far the improvement traveled, with detectable effects extending on the order of twenty-five miles downstream of upgraded plants before dilution erased the signal. If the rivers had been cleaning themselves for unrelated reasons, there would be no reason for the cleanup to start the year the local plant finished construction. Again and again, it did.

What they found on the quality side was unambiguous, and it is worth stating at length because this half of the paper gets less attention than it deserves. Grant-funded upgrades caused large, statistically robust reductions in pollution downstream. Dissolved oxygen rose. Biochemical oxygen demand, the standard measure of the organic waste that suffocates rivers, fell. Fecal coliform concentrations, the marker of sewage contamination that closes beaches and shellfish beds, dropped sharply. Total suspended solids declined. And the regulatory bottom line moved with the chemistry: waters downstream of upgraded plants became significantly more likely to meet the standards that define fishable, the very goal Congress had dated to 1983. The effects were not small adjustments at the margin. They were the kind of changes that turn a dead reach into a living one, and they showed up across regions, across decades, and across the different pollutants the grants were meant to address. The gains, moreover, endured. The paper found that the improvements downstream of grant-funded plants persisted for about 30 years after the upgrades were completed, not as a fading echo but as a sustained new condition. That persistence matters because it answers the suspicion that treatment plants work well when new and degrade when old; across the study window, the rivers stayed cleaner. It also matters for the economic half of the analysis, because a benefit that lasts for decades is worth far more than a benefit that lasts for years, and any fair ledger has to count the full stream. Taken together, the quality findings amount to the strongest causal evidence ever assembled that the statute’s central spending program did what Congress paid it to do. The money went to treatment plants, the plants discharged cleaner effluent, and the rivers downstream got measurably, durably cleaner. On the question “did the law clean the water,” this half of the paper answers yes, with the kind of evidence economists trust most.

The paper’s second half asked the harder question, and it answered it with equal candor. Having established that the grants caused real improvement, Keiser and Shapiro asked what that improvement was worth in dollars and how the worth compared to what the grants cost. Valuing cleaner rivers is not straightforward, because no one buys “river cleanliness” in a market with a posted price. The authors used the standard economist’s workaround: housing prices. Homes near cleaner rivers sell for more than otherwise identical homes near dirtier ones, and that premium reveals, indirectly, what households are willing to pay for water quality. By measuring how home values changed when nearby rivers cleaned up after grant-funded upgrades, the paper estimated the dollar value households placed on the improvement, aggregated those values across the affected populations, and set the total against the outlays for the grants themselves.

The result is the finding that launched a hundred arguments. Conventionally measured benefits, the housing-price values, fell well short of the costs of the grants. The ledger, drawn the standard way, shows the country spending more to clean its rivers than the nearby households’ revealed willingness to pay for the cleanup. Keiser and Shapiro reported this plainly rather than burying it, and the plainness is to their credit: a paper that had found enormous gains and then hidden an unfavorable benefit-cost comparison would deserve no one’s trust. The shortfall is real within the method’s terms. Households near the cleaned rivers, judging by what they paid for homes, valued the cleanup at less than it cost to produce. Anyone who cites the paper for its gains and ignores this half is misrepresenting it, and anyone who cites it for the shortfall and ignores the gains is doing the same. The paper contains both findings on purpose, and they have to be carried together.

What the authors did next is what makes the paper the canonical treatment rather than merely a provocative one. They argued, at length and carefully, that the shortfall measures the incompleteness of the benefit ledger rather than the failure of the statute. The housing-price method captures only the values that get capitalized into nearby real estate: the pleasure of living near a clean river, the recreation access, the view. It captures nothing of what economists call nonuse values, the worth people place on knowing that rivers run clean even when they never visit them. A household in Arizona that never sees the Cuyahoga may still value the fact that American rivers no longer catch fire; a parent may value the bequest of fishable rivers to children not yet born; an angler may value the option of fishing a recovered river someday. These are real components of the value of clean water, documented in decades of stated-preference research, and the hedonic method sees none of them, because none of them moves a home price.

The ledger is incomplete in a second way, and here the comparison with the statute’s sister law becomes instructive. The benefits of the Clean Air Act, as conventionally tallied, are dominated by reductions in mortality from fine particulate exposure, a pathway where epidemiology supplies dose-response relationships and where the value of mortality risk reduction is among the best-studied numbers in economics. The air law’s benefits therefore arrive at the benefit-cost table nearly complete, counted in lives extended. The water law targeted a different universe of pollutants, organic waste, bacteria, suspended solids, the conventional pollutants that kill fish by depleting oxygen rather than killing people through chronic exposure. For these pollutants the health science offers far thinner dose-response evidence: cleaner rivers surely mean fewer gastrointestinal illnesses among swimmers and safer shellfish, but the epidemiology needed to convert those avoided illnesses into dollars at national scale barely exists. The contrast is stark: analyses of the Clean Air Act can tally mortality reductions from fine particles with established science, while no comparable dose-response ledger exists for the conventional pollutants the water law targeted, so the two statutes’ benefits reach the benefit-cost table with very different degrees of completeness. A verdict that compares their ratios without adjusting for that difference is not comparing the statutes. It is comparing the sciences.

The measurement gap: the Clean Water Act is the clearest case in American regulation of a statute whose benefits are systematically harder to measure than its costs, and any benefit-cost verdict on it is really a verdict on the measurement method rather than on the statute.

That sentence states the paper’s deepest implication in its strongest form, and it is worth distinguishing the implication from several things the paper did not say. Keiser and Shapiro did not conclude that the statute failed, and they did not write a brief for repeal or retrenchment. They did not claim the grants were wasted money; they claimed the standard tools undercount the returns. They did not argue that benefit-cost analysis should be abandoned; they argued that its results for this statute should be read with the missing categories in view. And they did not pretend the missing categories could be filled with precision: nonuse values and health pathways are omitted precisely because they resist the kind of measurement the housing-price method performs, and acknowledging the omission is not the same as repairing it. The paper’s honesty runs in both directions. It refuses to inflate the measured benefits, and it refuses to mistake the measured benefits for the whole.

The cost side of the paper’s comparison deserves the same scrutiny the authors gave the benefit side, because “costs” in the public argument about the statute usually means something broader than what the paper tallied. Keiser and Shapiro compared the outlays for the construction grants against the housing-price benefits, which is the right comparison for judging the grant program specifically. But the full cost of the statute includes the industrial compliance spending forced by the effluent guidelines, the municipal operating costs borne by sewer ratepayers after the federal construction money was spent, and the administrative costs of the permit system itself. None of those appear in the paper’s ratio, and a critic who wants to argue the statute cost more than the paper says has legitimate material to work with. The honest response is quantitative modesty: the paper measured the benefit-cost arithmetic of the grants, the federal spending channel, not of the entire regulatory enterprise, and broader tallies would need to count both more costs and, as the authors argue, more benefits.

The hedonic method itself carries known limitations that cut in both directions, and the paper’s readers should understand them. Housing prices capitalize only the values buyers perceive, and only to the extent buyers are informed; a household that does not know the river got cleaner cannot pay for the knowledge, and a household that expects the cleanup to reverse will discount it. Sorting complicates the picture further: cleaner rivers attract higher-income households whose willingness to pay differs from the population average, so the measured premium reflects the preferences of movers rather than of everyone. These are standard caveats in environmental economics, and they mean the housing-price estimate is best read as a lower bound on use values rather than a complete accounting of them. The authors’ argument does not depend on the method being perfect. It depends on the method being systematically incomplete in one direction, capturing a slice of the value while the costs are captured whole, and on that point the logic is difficult to escape.

A fuller ledger, if one could be drawn, would add the categories the method misses. Commercial and recreational fishing gains that accrue to visitors rather than nearby homeowners. Avoided treatment costs for downstream drinking water systems drawing from cleaner rivers. The existence and bequest values discussed above, which stated-preference surveys have repeatedly found to be large relative to use values for iconic waters. And the health pathways, the gastrointestinal illnesses not contracted by swimmers, the developmental harms not suffered by children exposed to cleaner water, whose absence from the ledger reflects the thinness of the epidemiology rather than the thinness of the effect. None of these can be filled in with the confidence of the housing-price estimates, which is why the authors present them as an argument rather than a number. But the direction of the bias is not in doubt, and its size is plausibly large enough to change the sign of the verdict. That possibility is what elevates the paper from a cost study to a caution about cost studies.

The paper landed in a literature that had been circling the question for years without quite answering it. Earlier benefit-cost treatments of the water law had generally found costs exceeding measured benefits, using cruder data and simpler methods, and defenders of the statute had generally answered by questioning the methods rather than producing better numbers. Keiser and Shapiro changed the terms by producing the better numbers themselves and then showing, from inside the analysis, why even good numbers undercount. The paper has been widely cited since, in economics and in regulatory policy debates, often by writers who quote only one of its halves. That selective quotation is the fate of any honest two-sided finding, and it is worth resisting here: the paper’s contribution is not the gains alone or the shortfall alone but the demonstration that both can be true at once, and that their coexistence indicts the measurement rather than the law.

Where the Evidence Leaves the Question

Set against the promise Congress wrote in 1972, the record of this section supports the three-part verdict stated at the outset. The rivers got markedly cleaner, and the cleaning traces to the statute’s money with a clarity rare in regulatory history: the grants rebuilt the plants, the plants discharged cleaner effluent, and the monitoring record, crowned by the 2019 causal analysis, shows the rivers responding where and when the money arrived. Against the dated goals the verdict is equally clear and less flattering. Discharges did not end in 1985. Not every waterway was fishable and swimmable by the summer of 1983. A law that promised an end delivered a large, documented, incomplete improvement, and the distance between the promise and the delivery is not a footnote. It is the central fact about the statute’s ambitions, and any account that rounds it off is flattering the law at the expense of the truth.

The third part of the verdict is the one that should change how the question is asked. The conventional benefit-cost test comes out against the statute, and the 2019 paper shows why that result should be handled with tongs: the test counts the costs completely and the benefits partially, because costs arrive as appropriations and benefits arrive as oxygen, fish, and the quiet knowledge that rivers no longer burn. None of that excuses weak analysis or justifies any price. It means the number at the bottom of the ledger is a property of the yardstick, and a yardstick that cannot see nonuse values or health pathways will always undervalue a law whose benefits live mostly in those categories. The statute’s defenders should stop citing the benefit-cost shortfall as though it were a clerical error, and its critics should stop citing it as though it were a conviction. It is neither. It is a measurement with known blind spots, and the blind spots are the story.

One discipline the evidence imposes is worth stating explicitly, because it guards against both romantic and cynical readings. The question is not whether American rivers in 2012 were pristine; they were not. The question is what they would have looked like without the statute, and the staggered-grant evidence gives the closest thing to an answer the historical record permits: rivers downstream of cities that upgraded early improved early, rivers downstream of cities that upgraded late improved late, and the difference tracks the money rather than the business cycle. The counterfactual is not a return to wilderness. It is the baseline extended forward, more people, more waste, the same pipes, rivers asked to absorb two more decades of growth without the plants the grants built. Against that counterfactual the statute’s achievement is not subtle. It is the difference between rivers that recovered and rivers that would have gotten worse.

The record assembled here covers what the grants and the permits cleaned up: the organic waste, the sewage bacteria, the oxygen-sucking effluent that made rivers burn and stink and die. It does not cover the waters that stayed impaired after the plants were rebuilt, the pollutants the permit system was never designed to reach, or the sources no permit ever touched. Those unfinished portions of the ledger, the share of waters still missing their goals decades after the deadlines passed and the reasons the statute’s absolute aims remain unmet, belong to the second half of this article, which takes up the question where the measurements of success run out.

The improvement story is real, and the first half of this article gave it its due: oxygen deficits shrank downstream of the treatment plants the federal grants built, the share of monitored waters able to support fish rose over the decades, and the best modern research attributes a meaningful part of that change to the statute rather than to coincidence. That is the first half of the ledger. This section takes up the rest of it: the waters that never made the turn, the sources the statute was never written to reach, the deadlines Congress printed in the text and then watched pass, and the reason the whole record resists a single verdict.

The Unfinished Half: Attainment Against Standards

The most quoted figures in debates about the Clean Water Act are the attainment shares: the percentages of assessed rivers, lakes and estuaries that fail to support one or more of the uses their states designated for them. Those figures are worth examining closely, because they are routinely misread in exactly the way the verification notes for this article warn against. They measure waters against standards at a single point in time. They do not measure change.

Why can the impaired share not measure improvement?

The share measures assessed waters against state-designated uses at one point in time, not whether conditions are improving. A river counts as impaired when monitoring or modeling shows it fails any designated use, such as fishing or swimming. It is a snapshot against standards, so a falling or rising share cannot be read as a trend.

The source most writers reach for is the Environmental Protection Agency’s National Water Quality Inventory, the Report to Congress compiled from state assessments under section 305(b) of the act. In the 2004 reporting cycle, the most fully documented national snapshot available in the article’s period, states assessed 563,955 miles of rivers and streams, which is 16 percent of the nation’s estimated 3.5 million stream miles. Of those assessed miles, 44 percent were reported as impaired, meaning not clean enough to support at least one designated use, while 56 percent fully supported all assessed uses, with 3 percent of the total rated as threatened by a deteriorating trend. For lakes, ponds and reservoirs, states assessed 39 percent of the national acreage and found 64 percent of the assessed acres impaired. For estuaries, states assessed 29 percent of the national square mileage and found 30 percent impaired. When the inventory broke the numbers down by individual use rather than by overall status, 36 percent of the stream miles assessed for fish, shellfish and wildlife protection and propagation were impaired for that use, and 28 percent of the waters assessed for recreation were impaired for that use.

Those numbers deserve two qualifications before they can be used honestly. The first is that the assessed fraction is not a random sample of the nation’s waters. States direct their limited monitoring budgets toward waters they have reason to worry about, toward high priority uses, and toward waters with known discharges, so the assessed set tilts toward trouble. The inventory itself cautions readers on this point. The second qualification is the one that undoes most public argument: the report states explicitly that the 2004 cycle should not be compared with earlier reports. Assessment methods changed, the portion of waters assessed varied from cycle to cycle, six states did not submit waterbody-level data electronically in 2004 so the cycle covered roughly 130,000 fewer stream miles than the 2002 cycle, and states redefined some of their designated uses over time. A writer who lines up the 1998, 2002 and 2004 impaired percentages and draws a trend line is manufacturing a finding the underlying data do not support. The attainment share answers one question, whether the assessed waters meet the standards states set for them, and it is silent on the question most readers actually want answered, whether conditions are getting better.

EPA’s answer to the sampling problem was a probability-based survey, the Wadeable Streams Assessment, which sampled about 1,392 stream sites at random in 2000 to 2004 and rated 28 percent of the nation’s stream miles in good condition, 25 percent fair, and 42 percent poor. A random sample can support statements about the whole resource rather than about the waters states chose to examine, and it is the right tool for the question the inventory cannot answer, but it cannot backfill history: the randomized record starts decades after the act, so the cleanest before-and-after comparison the statute will ever get remains the targeted monitoring record with all its flaws. The measurement gap is not only about benefits. It is about baselines too.

What the shares do establish, read properly, is scale. Even allowing for the sampling tilt, the 2004 snapshot says that among the waters states chose to look at, impairment was the common condition for rivers and the dominant condition for lakes. The inventory’s own analysis of causes named pathogens, mercury, nutrients and organic enrichment with low dissolved oxygen as the leading causes of impairment, and its analysis of sources named agriculture as the leading source for rivers and streams, atmospheric deposition as the leading source for lakes and estuaries, hydrologic modification close behind, and a large category of unknown or unspecified sources. Municipal discharges, the target the statute was built to control, no longer led the list. That shift in the source ranking is one of the most consequential facts in the whole evidence record, because it describes a problem the statute’s machinery was only partly designed to solve.

The companion mechanism to the attainment reports is the impaired waters list under section 303(d). When a state finds that technology based discharge limits are not enough to bring a water body up to its standards, it lists the water and eventually develops a total maximum daily load, a pollution budget that allocates allowable loading among sources. By the 2000s, tens of thousands of water body segments sat on state 303(d) lists awaiting or covered by such budgets. The lists are useful as an index of where the permit system hit its limits, and they are frequently misread as an index of failure. A water body can sit on the 303(d) list for years while its conditions improve, because the listing reflects whether standards are met, not whether loading is falling. Again the measure is against standards, not change.

The 303(d) program’s own history reinforces the point about pace. Through the 1980s, EPA and the states produced few TMDLs, and a wave of citizen suits in the 1990s forced the agencies onto court-ordered schedules: by 2001 about 1,500 TMDLs had been established against more than 20,000 impaired waterbodies on the 1998 state lists, and by 2008 more than 34,000 TMDLs had been developed since 1996, toward an EPA estimate that as many as 40,000 might ultimately be required. But a TMDL is a calculation, not a cleanup. The statute sets no deadline for achieving the load reductions a TMDL assigns to nonpoint sources, and the allocations follow the statute’s two tracks: the point-source side gets written into enforceable permit limits while the nonpoint side waits on voluntary action. The list grows faster than the load falls, which is one more reason the lists mislead when they are read as a scoreboard of success or failure.

Why does the statute leave agricultural runoff outside the permit system?

Yes. The statute regulates discharges from discrete conveyances such as pipes, called point sources, and expressly leaves most agricultural stormwater and irrigation return flows outside the discharge permit system. Congress added section 319 in 1987 to address runoff through state plans and federal grants, a voluntary funding channel rather than a permit requirement.

This is the structural heart of the unfinished half, and it should be stated as a matter of statutory architecture rather than of neglect. The 1972 act defines its regulatory reach around the point source, a discernible, confined and discrete conveyance from which pollutants are discharged. The permit system built on that definition, the National Pollutant Discharge Elimination System under section 402, obliges municipalities and industries to meet technology based limits and to face enforcement for violating them. Agricultural stormwater discharges and return flows from irrigated agriculture are expressly excluded from that system by section 402(l)(1). The exclusion was not an oversight that slipped past the drafters. It reflected a deliberate 1972 bargain about federal reach, about the administrative impossibility of permitting millions of fields, and about leaving land use decisions with the states. Whatever one thinks of the bargain, it is part of the statute’s design, and describing the nonpoint gap as a failure of will misdescribes what Congress actually wrote.

What Congress did write for runoff came fifteen years later. The Water Quality Act of 1987 added section 319, which directs states to assess their nonpoint source problems, to adopt management programs describing the best management practices they will promote, and to implement those programs with the help of federal grants. The grants fund terracing, buffer strips, animal waste management, streambank stabilization and the other field level practices that reduce runoff loading, and they fund the planning and monitoring that tells a state where its runoff problems are worst. What section 319 does not do is regulate. It sets no discharge limits for farms, it issues no permits to fields, and it carries no enforcement tool against a landowner whose runoff keeps a stream impaired. The federal role is to pay for and plan around a source category the federal permit system does not touch.

The total maximum daily load program illustrates the same boundary from a different angle. A TMDL assigns allowable pollutant loads to point sources as wasteload allocations and to nonpoint sources as load allocations, and on paper it reaches the whole watershed. In practice the point source side of a TMDL is enforceable through permit limits and the nonpoint side is not, because the statute gives the administering agencies no lever to compel nonpoint reductions. States may use their own authority, or the leverage of grant conditions, or simply persuasion, and the results vary enormously from state to state. Where state programs are strong and well funded, nonpoint loading has fallen in specific watersheds. Where they are not, the TMDL is a budget with no collector.

The practical consequence is that the country’s dominant remaining water quality problems sit largely outside the enforceable core of the statute. The 2004 inventory’s source ranking, with agriculture first among river sources and atmospheric deposition first among lake and estuary sources, is the empirical face of that design. Nitrogen and phosphorus from fertilizer and manure feed the nutrient enrichment that impairs rivers and lakes across the farm belt and the coastal plain. Sediment from fields and construction clouds streams. Pathogens from livestock and wildlife close beaches and shellfish beds. Mercury arrives through the air from sources no water permit reaches. None of these yields to a pipe by pipe permit program, and the statute’s framers knew the permit program was pipe by pipe. The unfinished half is not mainly a story of permits poorly enforced. It is a story of a permit system that did what it was built to do while the problem moved, or always partly lived, somewhere the permits do not reach.

The exclusion itself was debated rather than smuggled: agricultural stormwater and irrigation return flows were left outside the permit system from the start, so the regulatory reach ended where the fields began. The result was a deliberate division of labor: command the pipes, subsidize the fields. Section 319’s machinery reflects that division. States file assessment reports identifying the waters impaired by nonpoint sources, then management programs describing the practices and schedules they will promote, and EPA approves the programs and awards grants that fund buffer strips, terracing, animal waste systems, streambank work and the monitoring that locates the worst problems. The section 319 appropriation ran $200.9 million in fiscal year 2010, $175.5 million in fiscal year 2011, and $164.5 million in fiscal year 2012, a fraction of what the construction grants and the later revolving fund directed at point sources, and the money flows through soil and water conservation districts and state agencies whose relationships with landowners depend on cooperation rather than compulsion.

The results vary the way voluntary programs always vary. States that paired 319 money with their own cost share funds and strong district networks have documented load reductions in specific watersheds. States that treated the program as planning money have produced plans. Neither outcome reflects anyone’s dereliction, because the statute never gave the program a tool sharper than persuasion plus payment. To call this a failure of will is to imagine that the will in question was ever armed. Congress chose to spend rather than to command on runoff, and the water quality record on runoff is the record of that choice playing out over forty years.

The Visible Wins: What Disappeared

Against the attainment statistics sits a different kind of evidence, and it is the reason ordinary Americans hold a far more favorable view of the act than the impaired percentages would suggest. The conditions that motivated the statute have largely vanished from the country’s major industrial rivers, and that disappearance is documented, dated and beyond serious dispute.

The emblem is the Cuyahoga River fire of June 22, 1969. An oil slick on the river in Cleveland caught fire and burned with enough intensity to damage railroad bridges, and the image traveled nationally because the river had burned before. The 1969 blaze was only the most famous of several fires on an industrial river so saturated with oil, solvents and floating debris that combustion was a recurring hazard rather than a freak event. Rivers in other industrial cities carried their own versions of the same condition: stretches of the Hudson, the Delaware, the Potomac and the Charles that functioned in the early 1970s as open industrial sewers, places where dissolved oxygen crashed, fish kills were routine, and the smell on a summer day announced the river before it came into view. The statute’s framers wrote with those rivers in front of them, and the permit system they built was aimed squarely at the municipal and industrial pipes that fed them.

What followed is the least contested part of the record. The construction grants program financed thousands of municipal treatment plant upgrades, the permit system imposed technology based limits on industrial dischargers, and the combined sewer overflow policy the Environmental Protection Agency issued in 1994 began the long work of separating or controlling the systems that dumped raw sewage into rivers during storms. The Beach Act of 2000, formally the Beaches Environmental Assessment and Coastal Health Act, extended federal support for state beach monitoring programs, which is why the public record on recreational water safety is far richer after 2000 than before. By the 1980s and 1990s, the spectacle conditions were gone.

The scale of the buildout behind that change is worth pausing over, because it explains both the achievement and the cost side of the later benefit cost debate. The construction grants program of the 1970s and early 1980s paid up to 75 percent of the capital cost of eligible municipal treatment works, and the grants represented about $650 billion in spending. The money rebuilt the treatment infrastructure of a generation of American cities, and the share of the sewered population receiving at least secondary treatment rose from about 61 percent in 1968 to about 87 percent in 1996. On the industrial side, EPA issued technology based effluent guidelines industry by industry, so that by the 1980s a paper mill or a steel plant faced nationally uniform limits on what its outfall could carry. The combined sewer overflow policy of 1994 then addressed the systems, concentrated in older Northeastern and Midwestern cities, that mixed sewage and stormwater in single pipes and discharged the mixture untreated during heavy rain; the policy required permittees to implement nine minimum controls and to develop long term plans for bringing the overflows under control, a program of sewer separation and storage tunnels that continued for decades; in the Washington area the 2002 long-term control plan was estimated at $1.265 billion under a 2004 consent decree, on a 20-year schedule. Each of these programs aimed at a pipe, and each of them moved a number that monitoring could see.

The Keiser and Shapiro results give that transformation a multi pollutant signature: downstream of grant receiving plants, the dissolved oxygen deficit fell, biochemical oxygen demand fell, fecal coliform counts fell, and the share of readings consistent with fishable water rose, with the effects visible across the decades the monitoring covered. The Delaware estuary’s recovery from the oxygen sag that had left its urban reach nearly lifeless is the textbook illustration of the mechanism, and the Hudson, the Potomac and the Charles each carry a version of the same story from industrial sewer to recreational river. These are the waters the permit system was built for, and they are the waters where the system’s record is strongest. No American industrial river has repeated the Cuyahoga’s 1969 performance. Fish kills on the scale of the early 1970s became newsworthy events rather than background facts of river life. The rivers that the statute’s supporters had described as dying now support recreational boating, waterfront redevelopment and, in many reaches, the return of fish populations that had been absent for a generation.

This is why the formal statistics and public perception diverge so sharply, and both can be true at once. A river can be transformed from an open sewer into a working urban waterway, a change any resident of 1972 would recognize as a triumph, and still fail to meet the designated uses its state assigned it, because the uses are demanding and the remaining impairments come from sources the permit system never touched. The person who swims in a river that no longer stinks is experiencing a genuine victory. The state assessor who lists that same river as impaired for nutrients is also recording a genuine fact. The attainment statistics measure the distance from a standard. The visible wins measure the distance from 1972. Confusing the two is the standard error of the whole debate.

One more distinction belongs here because readers merge the two constantly. Nothing in the evidence reviewed in this article describes tap water. Drinking water is governed by a separate statute, the Safe Drinking Water Act of 1974, with its own monitoring rules, its own violation records and its own enforcement history. The gains documented here are gains in rivers, lakes and estuaries. When a household’s faucet runs brown, that is a failure of a different system, and attributing it to the Clean Water Act’s record, in either direction, is a category error.

The Deadlines Written Into the Statute

The series thesis for this article asks that the statute be judged against its own stated aims, and the Clean Water Act stated two of its aims as dates. Section 101(a)(2) set an interim goal: water quality sufficient to protect and propagate fish, shellfish and wildlife, and to provide recreation in and on the water, to be achieved by July 1, 1983. Section 101(a)(1) set the national goal: the discharge of pollutants into the navigable waters to be eliminated by 1985. The language is in the statute as enacted in 1972. Neither date was met.

Did the written deadlines ever bind?

No. The interim goal of water quality supporting fishing and swimming by mid 1983, and the national goal of eliminating pollutant discharges by 1985, were both written into section 101 of the 1972 act. Monitoring showed neither had been reached by its date, and Congress never revised the dates or enforced them.

The dates made sense in the political moment that produced them, which is part of why they survived unamended. The 1972 act was written in an era of unusual legislative confidence that technology forcing worked: the 1970 Clean Air Act had set health based air standards on fixed timetables, and the water bill’s managers, above all Senator Edmund Muskie of Maine, who chaired the Subcommittee on Air and Water Pollution and managed the bill on the Senate floor, framed water cleanup as the parallel campaign. The declaration of goals in section 101, that the objective of the act is to restore and maintain the chemical, physical and biological integrity of the nation’s waters, was written in the same spirit. The dates were the urgency made concrete. When the 1977 amendments extended a series of industrial compliance deadlines that the original timetables had made unworkable, Congress adjusted the machinery but left the statement of purpose alone, and the 1987 amendments did the same. Aspirational dates are a normal legislative device, and no one who voted for them was later eager to stand up and lower them. So the 1983 and 1985 language stayed in the statute long after it had become a historical artifact, a promise the law kept making every day it went unmet.

The 1983 interim goal deserves a precise reading because its failure is measurable. Fishable and swimmable are not metaphors in the statute; they name designated uses that states assign to waters and then assess. The 2004 inventory, taken more than two decades after the deadline, found 36 percent of the stream miles assessed for fish, shellfish and wildlife protection and propagation impaired for that use, and 28 percent of the waters assessed for recreation impaired for that use. The interim goal was not close. It was not met in 1983, and on the inventory’s own terms it had not been met twenty years later.

The 1985 zero discharge goal was more ambitious still, and its nonattainment was never in doubt. Eliminating the discharge of pollutants would have meant the end of permitted industrial and municipal outfalls entirely, a transformation no one in the regulatory system treated as operational. The permit system was built to reduce discharges toward the goal, not to reach it, and the goal functioned as a statement of direction rather than as an enforceable command. No deadline enforcement action was ever brought on the theory that discharges continued after 1985, because the goal was written as a national aspiration in section 101, the declaration of goals and policy, rather than as a prohibition with a penalty attached.

Congress had opportunities to revisit the dates. The 1977 amendments, enacted as the Clean Water Act of 1977, adjusted a series of compliance deadlines for industrial dischargers, and the 1987 amendments rewrote the funding model and added the nonpoint and stormwater provisions. Neither reauthorization revised the 1983 and 1985 goal language in section 101, which stands in the statute as a permanent reminder of how far the enacted ambition ran past the achievable. There is an honest debate about how much weight to give this. Defenders of the statute argue that aspirational goals are a normal legislative device, that the dates expressed urgency rather than a forecast, and that no serious participant in 1972 believed discharges would literally reach zero within thirteen years. That defense has force, but it does not erase the fact. A statute that announces a thirteen year path to zero discharge and then presides over continued permitted discharge for the next forty years has, on the plain meaning of its own text, missed the aim it set. Any assessment that counts only the improvements and not the missed dates is grading the statute against a curve it did not choose for itself.

The Two Symmetrical Overreaches

The evidence record now contains four things that have to be held in the mind at once, and the characteristic error of writing about the Clean Water Act is to drop one of them. The first is the measured improvement: pollution loads from the sources the statute regulated fell substantially, oxygen levels recovered downstream of the grant funded plants, and the best causal research attributes a real share of that change to the statute. The second is the unmet aims: the interim and zero discharge dates passed without the goals being reached, and the attainment statistics show designated uses still unmet across large fractions of assessed waters decades later. The third is the measurement problem: the statute’s benefits are unusually hard to quantify, the monitoring record is fragmented across thousands of local collectors rather than gathered by a single standardized network, and any benefit cost verdict is therefore hostage to what the chosen method can see. The fourth is the structural gap: the dominant remaining sources, above all agricultural and urban runoff, sit outside the enforceable permit core by statutory design, not by administrative accident.

From these four facts, two symmetrical overreaches are commonly drawn, and both are selections rather than conclusions. The first overreach says the act failed: half the assessed waters still fall short of their standards, the leading modern assessment finds measured costs exceeding measured benefits, the deadlines were missed. Each premise in that sentence is true, and the conclusion does not follow, because the premises describe the distance from standards and from aspirations, not the distance from 1972. A statute can miss its own deadlines and still transform the rivers it was written to save. The second overreach says the act succeeded absolutely: rivers no longer catch fire, the open sewer conditions are gone, the public’s memory of the transformation is warm. Each premise in that sentence is also true, and that conclusion does not follow either, because the vanished spectacles were never the whole of the statute’s ambition. The act promised fishable and swimmable waters by 1983 and the elimination of discharges by 1985, and pointing at a river that no longer burns does not answer the promise.

Holding all four together produces a more demanding and more accurate verdict. The statute worked powerfully within its design envelope: the pipe by pipe permit system, backed by hundreds of billions in federal construction money, cut the loadings it was built to cut, and the rivers downstream of those pipes show it. The statute did not work outside that envelope, because outside that envelope there is mostly planning, grants and state discretion rather than regulation. The benefit cost debate is in large part an argument about how much of the value created inside the envelope our measurement tools can see. And the deadlines stand as the statute’s own confession that its authors imagined a faster and more complete transformation than any permit system could deliver. None of these four points cancels the others. The honest account of the Clean Water Act is not a grade but a map: here is what it reached, here is what it never reached, here is what we can measure, and here is what the measurements miss.

Each overreach also has a constituency, which is why both survive every round of the argument. Advocacy groups cite the impaired shares and the missed deadlines to argue for broader authority and tighter standards, because numbers that describe distance from a standard are useful for demanding movement toward it. Regulated industries and their analysts cite the benefit cost ratios to argue that the statute overspent, because numbers that describe costs against measured benefits are useful for resisting new obligations. Both selections are sincere, and both are incomplete in the same way: each treats one layer of the evidence as the whole record. The selection mechanism is worth naming because it operates on the reader as well. A person who encounters the 44 percent impaired figure without the sampling and trend qualifications will carry away a story of failure. A person who encounters the 0.25 benefit cost ratio without the authors’ caveats will carry away a story of waste. A person who encounters only the Cuyahoga’s transformation will carry away a story of triumph. The article’s job is to make all three stories harder to carry away alone, by keeping the four clauses in view at the same time and letting their tension do the work that a single verdict cannot.

The map metaphor has one more implication. A map does not tell you whether the journey was worth it; it tells you where you went. The statute’s defenders and its critics both want the evidence to settle a question of worth, and the evidence keeps declining the assignment, because worth depends on values the measurements omit and on aims the statute stated but never enforced. What the record can settle is the geography: what changed, where, by whose action, and what remains. That is less satisfying than a verdict and more useful than one.

Where the Gains Showed Up, and Where They Did Not

The national averages conceal a geography, and the geography is part of the evidence. The gains cluster where the statute’s money and permits were aimed, and the shortfalls cluster where they were not.

The clearest gains appear downstream of the municipal treatment plants that the construction grants program financed. This is not an impression but a finding. Keiser and Shapiro’s 2019 analysis matched grant dates and locations, obtained through two Freedom of Information Act requests, to monitoring stations upstream and downstream of the receiving plants, and found that pollution fell significantly downstream after grants arrived, across measures including the dissolved oxygen deficit, biochemical oxygen demand, fecal coliforms and the share of waters rated fishable. The effect is visible in the industrial river corridors of the Northeast and Midwest, where federal money rebuilt the treatment infrastructure of cities whose rivers had been the worst cases of the early 1970s. The statute’s spending channel and the statute’s permit channel pointed at the same targets, and the monitoring record shows the targets moved.

The lagging regions are, with depressing consistency, the places where the dominant sources are diffuse. Across the farm belt and the coastal plain, nitrogen and phosphorus from fertilizer and manure continue to enrich rivers and lakes, and the 2004 inventory’s ranking of agriculture as the leading source of river impairment is the national summary of thousands of local versions of that story. The Gulf of Mexico’s summer hypoxic zone, the oxygen-starved area fed by nutrient loads carried down the Mississippi River system, has been routinely mapped since 1985 and has recurred every summer since, its size varying with spring rainfall and river flow. The 2002 zone measured about 8,400 square miles, then the largest on record; the 2010 zone measured 7,722 square miles. No permit in the Clean Water Act’s system reaches the fields that generate those loads, so the dead zone persists as the purest available illustration of the structural gap: the statute’s machinery is complete and its reach stops at the field’s edge.

Urban stormwater occupies a middle position. The 1987 amendments required permits for stormwater discharges associated with industrial activity and for large and medium municipal separate storm sewer systems, with the Phase I rules issued in 1990 and the Phase II expansion to smaller systems following at the end of the decade. This brought a portion of urban runoff into the permit world, but the program’s history is one of slow implementation, general permits with modest requirements, and loadings that track development and pavement rather than treatment. A city whose sewage plant was rebuilt with federal grants can show a transformed river at the plant’s outfall while the same river stays impaired from the storm drains that empty into it a mile downstream. The two facts coexist because they describe two different regulatory regimes applied to the same water.

The arid West adds its own variant, where irrigation return flows, excluded from the permit system by the same provision that excludes agricultural stormwater, carry salts and selenium into rivers that cannot dilute them. The Great Lakes carry the toxic legacy variant: persistent contaminants like PCBs and mercury, already in sediments and cycling through food webs, that no discharge limit on a current outfall can remove, and that keep fish consumption advisories in place long after the pipes that introduced them were controlled. Each of these is a different face of the same pattern. Where the pollutant comes out of a pipe, the statute has a strong record. Where it comes off the land, out of the air or out of the sediment, the record thins, and the geography of impairment follows the geography of sources the statute was not written to command.

Two large programs illustrate the boundary at watershed scale. The Chesapeake Bay partnership, formed in 1983 by the bay states, the District of Columbia and the federal government, spent decades pursuing voluntary nutrient reduction goals for the nation’s largest estuary, and when the voluntary goals repeatedly slipped, EPA established the Chesapeake Bay total maximum daily load on December 29, 2010, the largest such pollution budget ever issued, covering nitrogen, phosphorus and sediment across a 64,000-square-mile watershed. The Bay TMDL is the most ambitious attempt to use the statute’s planning tools against a fundamentally nonpoint problem: a pollution budget of 185.9 million pounds of nitrogen, 12.5 million pounds of phosphorus, and 6.45 billion pounds of sediment per year, reductions of roughly 25, 24, and 20 percent from prior loads, it remains the largest such budget EPA had developed. The Great Lakes carry the parallel story for toxics: the 1987 agreement designated 43 Areas of Concern around the lakes where beneficial uses were impaired, 26 on the American side, 12 on the Canadian side, and 5 shared across the border, many of them by contaminated sediments and legacy pollutants. Decades of remediation work had delisted four by the time of writing, Collingwood Harbour in 1994, Severn Sound in 2002, the Oswego River in 2006, and Wheatley Harbour in 2010, with Spanish Harbour and Jackfish Bay designated as in recovery rather than delisted, because dredging contaminated sediment is slow and expensive and because the pollutants in question are already in the system rather than still arriving through outfalls.

Urban stormwater’s regulatory history shows the same pattern in miniature. The Phase I stormwater rule of November 16, 1990, brought medium and large municipal separate storm sewer systems, 11 industrial categories, and construction sites of five acres or more into the permit system; the Phase II rule of December 8, 1999, extended coverage to regulated small systems and construction sites of one to five acres. But stormwater permits are typically general permits built around best management practices rather than numeric effluent limits, their requirements phase in slowly, and the pollutant loads they govern track rooftops, parking lots and traffic rather than treatment plant performance. The result is the split screen described above: a river transformed at the sewage plant’s outfall and still impaired at the storm drain, both facts true, both produced by the statute, each by a different title of it.

The Readings People Still Argue About

Two disputes dominate the literature on whether the act worked, and both persist because each side has a genuine piece of the evidence. They deserve to be presented at full strength rather than split down the middle, because the reason each dispute survives is instructive in itself.

The Authors Read the Ratio as a Lower Bound

Conventional measurements say no, but their authors say the measurements are incomplete. Keiser and Shapiro’s 2019 analysis found measured benefits, mostly through nearby home values, equal to roughly one quarter of costs. The same authors argue the true value is higher because standard methods miss health effects, nonuse values and ecological feedbacks.

The headline finding comes from the Quarterly Journal of Economics paper the brief centers, “Consequences of the Clean Water Act and the Demand for Water Quality,” by David A. Keiser of Iowa State University and Joseph S. Shapiro of the University of California, Berkeley, published in 2019. After establishing that grants caused real pollution reductions downstream, the authors asked what those reductions were worth. Their main benefit measure works through housing markets: cleaner rivers raise the value of nearby homes, and the change in home values reveals what residents were willing to pay. On that measure, the benefits came to roughly one quarter of the costs, a ratio near 0.25 overall, rising to about 0.53 in areas where outdoor fishing or swimming is common and residents therefore value fishable water more highly. The cost side of their calculation put the annual cost of making one river mile fishable at about 1.5 million dollars. Presented alone, the finding reads as a verdict: the country spent far more cleaning its rivers than the clean rivers were worth.

The authors themselves present it as anything but a verdict, and the brief for this article insists that the caveats be given the same weight as the headline. The home value method captures only the value that nearby residents reveal through housing choices. It misses the value people place on knowing a river is clean even if they never visit it, the nonuse or existence values that economists consider real but that leave no trace in property records. It misses health pathways, because the epidemiological work connecting surface water quality to human health outcomes is thin compared with the vast literature on air pollution and health, so health benefits that may exist cannot be counted. It misses ecological feedbacks, the ways cleaner water supports fisheries, wetlands function and downstream systems in chains too complex for a housing regression to price. And it is depressed by a lack of information: as Shapiro has noted in describing the research, people cannot pay for cleaner rivers they do not know are cleaner, and water quality information reaches the public far less reliably than air quality alerts. The authors interpret their measured ratio as a lower bound on the true ratio, not as an estimate of it.

The measurement dependence becomes clearer when the water record is set beside the air record. In a companion 2019 survey of the economics of water regulation, Keiser and Shapiro reported that about two thirds of surface water quality regulations failed a formal cost benefit test in government analyses, compared with roughly one in five drinking water regulations and fewer than one in ten air pollution regulations. Shapiro has contrasted those ratios with the Environmental Protection Agency’s retrospective estimate for the Clean Air Act, which put that statute’s benefit cost ratio in the tens to one. The gap is not mainly a story about one statute being wise and the other wasteful. Air quality research faces the same evidence problem but with one decisive advantage: a national monitoring network that records the same pollutants in the same units everywhere, continuously and automatically. Water monitoring, by contrast, is fragmented across thousands of state and local collectors taking grab samples on different schedules for different purposes, which is why assembling the QJE paper’s dataset required merging EPA STORET and USGS monitoring data and filing two records requests. When measurement is standardized, benefits are easier to find. When it is not, costs, which arrive as budget line items, dominate the ledger by default.

There are also substantive reasons the ratio looks bad, and the literature names them. The statute’s uniform technology standards applied the same treatment requirements to very different rivers, which meant expensive cleanup in places where few people valued it alongside underinvestment where many did. The standards focused on end of pipe treatment rather than on cheaper ways to reduce pollution at the source, and they left the cheapest abatement opportunities, in agriculture, outside the system entirely. And surface water quality is partly substitutable in a way air quality is not: a family faced with a dirty river can drive to a cleaner lake, which reduces what they will pay to clean the dirty one, while no one can drive away from the air. Each of these points is a real criticism of the statute’s efficiency, and each of them is simultaneously a reason the measured ratio understates the value of what was achieved.

The scale of the spending behind the ratio is itself part of the story. In a companion 2019 paper, a working paper released through the National Bureau of Economic Research in July 2019, Keiser and Shapiro estimated that public and private American sources had spent nearly 5 trillion dollars in 2017 dollars since the founding of the Environmental Protection Agency to provide clean rivers, lakes and drinking water, amounting to about 0.8 percent of gross domestic product in most years, with surface water pollution control accounting for roughly 3 trillion dollars of that total since 1970. The same paper’s abstract frames the puzzle in a single sentence: despite that spending, over half of rivers and substantial shares of drinking water systems violate standards, and polls have for decades ranked water pollution as Americans’ leading environmental concern. The spending figure and the violation figure coexist because they describe different things: the money bought treatment plants and permits, while the violations increasingly come from sources the plants and permits were never built to reach. The authors’ four conclusions in that paper are worth quoting in substance: water pollution has fallen since the 1970s laws, in part because of them; the investments could have been more cost effective; most recent studies find the benefits of surface water cleanup smaller than the costs, though the studies may undercount important benefit categories; and economic research on water pollution remains thin compared with the attention given to air. The thinness of the research is itself evidence for the measurement gap: the economics profession has produced far fewer studies of water benefits than of air benefits, because the data to study are so much harder to assemble.

The caveats deserve to be spelled out rather than summarized, because the brief for this article requires them to be given the same weight as the headline. Nonuse values are the largest missing category: economists have long documented that people value the existence of clean rivers they will never visit, and the housing market method captures none of it. Health pathways are the second missing category: the epidemiological literature connecting surface water exposure to illness is sparse, partly because exposure is hard to measure and partly because drinking water treatment breaks the most direct pathway, so any health gains from cleaner rivers go uncounted by default. The information problem is third: water quality data reach the public through advisories and occasional news coverage rather than through the daily indexes and alerts that broadcast air quality, so residents may not know their river improved and cannot bid that knowledge into home prices. The 0.53 ratio in areas where outdoor fishing or swimming is common is the authors’ own evidence for this mechanism: where people pay attention to water quality, the measured benefits rise. None of this proves the true ratio exceeds one. It proves the measured ratio is a floor, and a verdict built on a floor is a verdict about the floor.

Causal attribution is the second dispute, and it asks a narrower question: of the improvement the monitoring record shows, how much did the statute cause? The naive reading credits the act with everything that happened after 1972. The skeptical reading notes that some pollution declines were already underway before the act, driven by state programs, by the economic decline of the dirtiest industries, and by the simple fact that the worst practices of the 1960s were already drawing political fire. Keiser and Shapiro’s design addresses this directly by comparing waters downstream of plants that received grants at different times, so the comparison isolates the effect of the grant from the background trends affecting all rivers. The result, that grants caused significant reductions across multiple pollutants, survives the skepticism in its essentials while conceding its limits: it measures the effect of the grants, which is the largest and most identifiable piece of the statute, not the effect of the permit system as a whole, and it cannot separate the act’s influence from the broader shift in norms and enforcement that the act itself helped create. The careful statement is that the statute caused a substantial share of the observed improvement, not all of it, and the exact share remains contested because the counterfactual, an America that kept building treatment plants without the federal mandate, cannot be observed.

What the Record Plainly Does Not Show

After the qualifications, it is worth stating the negative findings plainly, because the article’s honesty depends on them. The statute did not bring nonpoint runoff under effective control. Section 319 grants and state management plans reduced loading in specific watersheds, but the national source rankings in the 2004 inventory, with agriculture leading for rivers and atmospheric deposition leading for lakes and estuaries, describe a problem the enforceable law does not reach, and nothing in the subsequent record through the article’s period reversed that ranking. The statute did not eliminate toxic contamination of the kind that keeps fish consumption advisories posted: mercury and PCBs appear among the leading causes of lake impairment in the inventory, and the advisories reflect pollutants already in sediments and food webs rather than pollutants still coming out of pipes. The statute did not meet its own deadlines, the 1983 interim goal or the 1985 zero discharge goal, and the goal language stands unamended as a measure of the distance between enacted ambition and operating reality.

The fish consumption advisory record sharpens the point about toxics. By the 2000s, states had advisories in effect covering 35 percent of the nation’s total lake acres and 24 percent of its total river miles in 2003 data, driven principally by mercury, PCBs, dioxins, DDT, and chlordane, with mercury behind about three-quarters of advisories at least in part. An advisory is a designated use failure: a water that cannot safely support fishing fails the fishable goal no matter how clear it looks. And advisories are the impairment most resistant to the permit system, because they reflect pollutants deposited years or decades earlier, cycling through sediments and food webs, rather than pollutants arriving through a pipe that a permit can close. The advisory map is therefore the visible-wins story in reverse: where the burning rivers showed what the statute could eliminate, the advisories show what it could not reach.

The industrial toxics story adds a final negative finding with its own data source. The Toxics Release Inventory, created under the Emergency Planning and Community Right to Know Act of 1986 with reporting that began in 1988, showed comparable on- and off-site releases down 45.6 percent, from 3.35 billion pounds in 1988 to 1.82 billion pounds in 1996, and onsite surface-water discharges down 65 percent (27 million pounds) from 1988 to 2001, which is a genuine reduction in loadings. But the inventory measures what facilities report releasing, not what reaches people, and the advisory map shows the gap between the two. Loadings fell; exposure pathways persisted.

The wetlands side of the outcomes ledger belongs to a different part of the statute and tells a different story. Wetlands are protected not by the discharge permit system but by the dredge and fill permit program of section 404, administered by the Army Corps of Engineers with environmental guidelines and veto authority held by the Environmental Protection Agency. The national wetlands inventory compiled by the Fish and Wildlife Service shows a long deceleration of loss: average annual net losses of 458,000 acres from the mid 1950s to the mid 1970s, 290,000 acres per year from the mid 1970s to the mid 1980s, and 58,500 acres per year from 1986 to 1997, followed by a reported net gain of 191,750 acres between 1998 and 2004, an average annual gain of about 32,000 acres, attributed largely to restoration, creation and the spread of freshwater ponds. The no net loss policy announced in 1989 set the political target, and the numbers show the target roughly reached in acreage terms by the early 2000s, with the important caveat that the inventory measured acres rather than wetland function or quality, and that coastal watersheds continued to lose ground, about 59,000 acres per year from 1998 to 2004 in the joint federal assessment of the Atlantic, Gulf and Great Lakes coasts. It is a genuine success of a related program, and it does not belong on the discharge ledger. Keeping the two ledgers separate is part of reading the evidence correctly.

Closing Assessment

The evidence record for the Clean Water Act, taken whole, supports a verdict with four clauses, and the article has earned each of them. First, the statute caused substantial and measurable improvement in the waters its machinery reached: industrial and municipal loadings fell, oxygen recovered downstream of the grant funded plants, and the conditions that made rivers burn disappeared from the country’s industrial corridors. Second, the statute missed the aims it set for itself: the 1983 interim goal and the 1985 zero discharge goal were not met, and large shares of assessed waters still fail to meet their designated uses decades later. Third, the leading economic assessment finds that conventionally measured benefits fall short of costs while arguing, at equal length, that the conventional measures miss most of what clean water is worth, which makes the benefit cost debate a debate about measurement rather than a verdict on the statute. Fourth, the dominant remaining problems sit outside the statute’s enforceable design by deliberate choice, which means the unfinished half is not a story of weak enforcement but of a strong permit system applied to a shrinking share of the problem.

None of the four clauses is optional. Drop the first and the account becomes a counsel of despair that no one who watched the rivers change would recognize. Drop the second and it becomes a celebration graded on a curve the statute never chose. Drop the third and the cost numbers harden into a false precision. Drop the fourth and the attainment statistics look like an enforcement failure rather than a design boundary. The measurement gap the first half of this article named is the thread that runs through all four: a statute whose costs arrive as budget line items and whose benefits arrive as unpriced changes in rivers, health and existence values will always look worse on paper than in the water. The honest reader’s conclusion is not that the act worked or failed but that it worked powerfully where it was aimed, that where it was aimed was never the whole problem, and that the whole problem was never going to be solved by the means the statute chose.

That conclusion is also a baseline for judging every claim made about the statute afterward. Any argument that the act failed because the impaired shares are high must answer the sampling and trend qualifications and the design boundary. Any argument that it succeeded because the rivers no longer burn must answer the deadlines and the advisory map. Any benefit cost verdict must state, at the same length as the ratio, what the measurement method cannot see. The evidence record does not pick a side in those arguments. It sets the terms on which they can be honestly conducted, and it disqualifies the versions that select one layer and discard the other three.

The Four-Layer Evidence Table

The table below compresses the article’s evidence into the four layers the brief specified. Each row names the outcome measured, the period the measurement covers, the principal source, the direction and rough magnitude of the finding, and whether the finding is settled or contested.

Outcome measured	Period	Principal source	Direction and rough magnitude	Settled or contested
Water quality downstream of grant-funded municipal plants: dissolved oxygen deficit, biochemical oxygen demand, fecal coliforms, share of waters rated fishable	Monitoring 1962 to 2001; analyzed in work published 2019	David A. Keiser and Joseph S. Shapiro, Quarterly Journal of Economics, 2019	Improvement; grants significantly reduced pollution downstream across multiple measures; share of waters rated fishable rose 12 percentage points, 1972 to 2001; about 1.5 million dollars in annual cost per fishable river mile (2014 dollars)	Settled on direction and attribution
Benefit-cost ratio of surface water cleanup, benefits measured mainly through nearby home values	Spending 1970 to 2014; analyzed in work published 2019	Keiser and Shapiro, Quarterly Journal of Economics, 2019; the authors’ survey of government regulatory cost-benefit analyses	Costs exceed measured benefits; ratio of measured benefits to costs near 0.25 overall, about 0.53 where outdoor fishing or swimming is common; about two thirds of surface water regulations fail formal cost-benefit tests in government analyses	Contested; the authors interpret the measured ratio as a lower bound and argue standard methods omit nonuse values, health pathways, and ecological feedbacks
Share of assessed waters not meeting designated uses	2004 reporting cycle	U.S. Environmental Protection Agency, National Water Quality Inventory: Report to Congress, 2004 cycle	44 percent of assessed river and stream miles impaired; 64 percent of assessed lake acres impaired; 30 percent of assessed estuary square miles impaired; leading sources include agriculture for rivers and atmospheric deposition for lakes and estuaries	Settled as a snapshot of assessed waters against standards; contested when read as a trend, which the report itself warns against
Disappearance of burning-river and raw-sewage conditions in major industrial rivers	From the Cuyahoga fire of June 1969 through the 1980s and 1990s	Contemporary press and municipal records; EPA historical accounts; state beach monitoring records after the Beach Act of 2000	Conditions eliminated; no repetition of the 1969-scale events; routine fish kills and open-sewer reaches ended in the major industrial corridors	Settled

A Short Study Section

A student working through this article should test it the way the article tests the statute: against its own stated aims. Start by writing down the four layers in your own words, with the source and period attached to each number you cite, and check whether any of your numbers lack a source. Then ask of each layer what it cannot show: the improvement studies cannot show what would have happened without the grants, the benefit cost ratio cannot show values the housing market does not price, the attainment shares cannot show change over time, and the visible wins cannot show progress toward the written deadlines. Finally, hold the two overreaches side by side and notice what each one drops. The exercise that matters is not deciding whether the act worked. It is being able to say, with a source for every clause, what it did, what it did not do, what it never tried to do, and why the numbers that look like answers are often measurements of something else. The companion reference for this article’s sources and figures is maintained alongside the series.

the VaultBook legislation study notebook

Frequently Asked Questions

Q: Did the Clean Water Act clean up American rivers?

Yes on the measures the monitoring record supports, and no on the act’s own final goal. The economists David Keiser and Joseph Shapiro compiled decades of monitoring data from the EPA’s STORET database and USGS monitoring records for a 2019 Quarterly Journal of Economics study and found that water pollution concentrations fell substantially across the United States in the decades after 1972, with dissolved oxygen deficits shrinking and the share of waters too polluted to be fishable declining. The gains clustered downstream of municipal sewage treatment plants, where federal construction grants paid for upgraded treatment. But the same record shows the 1985 national goal of eliminating pollutant discharges was not met, and EPA’s state-based assessments have consistently found roughly half of assessed river miles still failing to meet at least one designated use, so the cleanup was real and incomplete.

Q: How much has US water quality improved since the Clean Water Act?

The headline finding from the Keiser and Shapiro 2019 Quarterly Journal of Economics work is that pollution concentrations fell substantially, with the share of waters meeting fishable standards rising 12 percentage points from 1972 to 2001, the gains concentrated within about 25 miles downstream of grant-recipient plants and persisting for roughly 30 years. Nearby housing values rose by about one quarter of the grant cost, which was itself a measured benefit. In physical terms, EPA’s National Water Quality Inventory series reported 36 percent of surveyed river miles impaired in 1996, 45 percent impaired in 2002, and 44 percent impaired in 2004, but those are snapshots against tightening standards and expanding assessment, not a trend line. The honest summary is that monitored conventional pollutants declined markedly, especially oxygen-demanding waste downstream of upgraded plants, while the official attainment share looks flat because states assess more waters against stricter standards over time.

Q: Did the Clean Water Act meet its 1985 goal?

No. Congress declared in section 101(a)(1) of the 1972 act a national goal of eliminating the discharge of pollutants into navigable waters by 1985, and an interim goal in section 101(a)(2) of reaching water quality sufficient to protect fish, shellfish, and wildlife and to allow recreation by July 1, 1983. Neither deadline was met. Point source discharges continued under permits, which the act itself authorizes, so zero discharge was never achieved in the literal sense. The 1983 interim goal came closer: conventional pollutants such as oxygen-demanding waste and pathogens fell sharply downstream of upgraded municipal plants, and EPA’s 2002 and 2004 inventory reports showed 55 to 56 percent of assessed river miles fully supporting all designated uses. But 44 to 45 percent of assessed miles were still impaired in those reports, so the interim goal was approached rather than attained.

Q: Do the Clean Water Act’s benefits exceed its costs?

By the standard measures economists use, no, and the authors of the leading study say that verdict should be read with caution. Keiser and Shapiro’s 2019 Quarterly Journal of Economics analysis found that the measured benefits of the act’s municipal grants, estimated mainly through changes in nearby housing values, amounted to roughly one quarter of the grants’ costs, and their related 2019 survey found that most studies estimate the benefits of surface water cleanup at less than the costs. They calculated a cost of about $1.5 million per river mile per year to make a mile fishable, in 2014 dollars. But they argued the measured benefits miss nonuse or existence values, health effects, and benefits people cannot perceive because water quality information is poor. Their conclusion is two-sided: some investments may genuinely fail a benefit-cost test, and the true benefits are plausibly larger than the measured ones.

Q: Why are half of US waters still polluted after the Clean Water Act?

Because the statute was designed to attack one category of pollution and the remaining pollution sits mostly in another category it barely reaches. The act’s core prohibition covers discharges from point sources, meaning pipes and discrete conveyances, which federal construction grants and technology-based permits brought under control. EPA’s National Water Quality Inventory reports identified the leading sources of the remaining impairment as agricultural activities, hydrologic modifications such as channelization and diversions, and unknown or unspecified sources, with pathogens, habitat alteration, and oxygen depletion among the leading causes in the 2004 reporting cycle. Those are nonpoint sources, meaning diffuse runoff that the act addresses only through grants and state planning under section 319, with no permit requirement. Only about 16 to 19 percent of the nation’s river miles were assessed in the 2002 and 2004 reports, so the half-impaired figure describes the waters states chose to monitor, not every mile.

Q: Did the Clean Water Act stop rivers from catching fire?

It stopped the conditions that made river fires possible, which is the most visible of the act’s achievements. The Cuyahoga River in Ohio caught fire repeatedly through the mid-twentieth century, most famously in June 1969, because it carried heavy loads of oil, industrial waste, and debris that could ignite. No comparable river fire has occurred since the early 1970s, a change the act’s historians attribute to the combination of the permit ban on unpermitted discharges, technology-based limits on industrial dischargers, and the municipal treatment upgrades the construction grants financed. The fires were a symptom of extreme organic and oil loading, and those loadings fell sharply in the industrial corridors the act targeted. The record here is one reason public memory of the statute is so favorable even where the formal attainment statistics are not.

Q: What is the biggest source of water pollution the Clean Water Act does not reach?

Nonpoint source pollution, meaning pollution that arrives in waterways as diffuse runoff rather than through a pipe, with agricultural runoff as its largest component. The act’s prohibition in section 301 applies to the discharge of a pollutant by a person, and the point source definition excludes agricultural stormwater discharges and irrigation return flows, so fertilizer, pesticide, and sediment runoff from farm fields is not subject to the permit system at all. Instead the act addresses nonpoint pollution through section 319 state management programs and grants, which carry no enforceable discharge limits. EPA’s 2004 National Water Quality Inventory named agricultural activities among the top sources of impairment in rivers and streams, and the economists’ assessments consistently describe nonpoint runoff as the dominant remaining source, which is why the impaired share of waters has barely moved even as point source pollution fell.

Q: How do economists measure Clean Water Act benefits?

Mainly through revealed preference methods, because clean water has no market price. In the leading 2019 assessment, Keiser and Shapiro measured the benefits of municipal treatment grants by comparing housing values near river segments downstream of grant-recipient plants with values near comparable segments, using a triple-difference design to isolate the grant’s effect, and found that each grant raised nearby housing values by about $7 million on average against an average grant cost of about $31 million, in 2014 dollars. Other studies in their survey use recreation demand models, estimating how much more people would pay for trips to cleaner water, and hedonic methods that extract the value of clean water from property markets. What these methods share is that they capture only use values, the worth of water people swim in, fish in, or live near, and they miss nonuse or existence values, the worth people place on clean rivers they will never visit, which is one reason the authors argue measured benefits understate the true total.

Q: How did dissolved oxygen levels in US rivers change after the Clean Water Act?

They rose, and dissolved oxygen is the single most tracked indicator of the act’s effect. Keiser and Shapiro’s 2019 Quarterly Journal of Economics study compiled monitoring records from the EPA’s STORET database and USGS monitoring and found that dissolved oxygen deficits, the shortfall of oxygen in water caused by decomposing organic waste, fell substantially across the decades following 1972. The mechanism is direct: municipal sewage treatment plants, rebuilt or upgraded with federal construction grant money, removed the organic loads that had been stripping oxygen from rivers downstream. The gains were concentrated within about 25 miles downstream of grant-recipient plants and persisted for roughly 30 years. Low dissolved oxygen is what kills fish and produces the dead zones and foul odors of the pre-1972 era, so this single measurement captures both the ecological and the human experience of the cleanup.

Q: Which pollutants showed the biggest improvement in monitored waters after the Clean Water Act?

The conventional pollutants that municipal sewage treatment is designed to remove, led by oxygen-demanding waste and fecal bacteria. Keiser and Shapiro’s 2019 Quarterly Journal of Economics compilation of decades of monitoring data found the largest and most consistent declines in dissolved oxygen deficits and in the bacterial and organic loadings discharged by municipal plants, because the construction grant program spent hundreds of billions of dollars upgrading exactly that infrastructure. Industrial toxic discharges also fell where technology-based permit limits applied, but the grant-driven municipal story dominates the monitoring record. Nutrients such as nitrogen and phosphorus improved far less, because they arrive largely through agricultural runoff that the permit system does not reach. That split explains the shape of the modern record: the pollutants the act’s grants and permits targeted fell sharply, while the pollutants from the sources it left alone did not.

Q: What did the 2019 Keiser and Shapiro study find about Clean Water Act benefits and costs?

David Keiser of Iowa State University and Joseph Shapiro of the University of California at Berkeley published two linked assessments in 2019: “Consequences of the Clean Water Act and the Demand for Water Quality” in the Quarterly Journal of Economics, and a broader survey of water pollution economics in the Journal of Economic Perspectives. The Quarterly Journal of Economics paper assembled decades of monitoring data, obtaining the dates and locations of every municipal construction grant through Freedom of Information Act requests, and used a triple-difference design comparing waters before and after grants, upstream versus downstream, across plants. It found substantial water quality gains attributable to the grants, with the share of waters meeting fishable standards rising 12 percentage points from 1972 to 2001, at a cost of about $1.5 million per fishable river mile per year in 2014 dollars. Measured benefits, estimated through nearby housing values at about 25 percent of grant costs, fell short of costs, though the authors argue the true benefits are larger.

Q: Why do Keiser and Shapiro say their Clean Water Act benefit estimates understate the statute’s true value?

Because the housing-value method they used captures only what nearby homebuyers know and pay for. In the 2019 Quarterly Journal of Economics paper and the related Journal of Economic Perspectives survey, Keiser and Shapiro list the omissions explicitly: people have incomplete information about water pollution and its health implications, so home prices cannot reflect risks buyers do not perceive; the estimates exclude nonuse or existence values, the value people place on knowing rivers are clean even where they never visit; they abstract from general equilibrium effects across housing markets; and they exclude health pathways that drinking water and air pollution studies capture more readily. They also note that the measured benefits reflect only a 25-mile radius around affected segments, missing recreation trips that travel farther. Their caveat is symmetrical with the headline finding: costs exceeded measured benefits, and measured benefits are an incomplete measure of benefits.

Q: How did Keiser and Shapiro isolate the effect of Clean Water Act construction grants from other causes?

With a triple-difference research design built on the geography of the American river network. For their 2019 Quarterly Journal of Economics paper, Keiser and Shapiro obtained the date and location of each municipal construction grant through two Freedom of Information Act requests, then mapped upstream and downstream waters using a dataset of 70 million nodes describing the entire United States river network. They compared water quality before versus after each grant, in waters downstream of the recipient plant versus upstream of it, and across plants that received grants at different times. Because a plant’s upstream waters share its local conditions but not its grant, the upstream-downstream contrast filters out regional trends, and the timing contrast filters out national ones. The design showed gains concentrated within 25 miles downstream of recipient plants and persisting for 30 years, which is the signature of the grants rather than of unrelated improvement.

Q: Did the Clean Water Act construction grants program deliver benefits greater than its costs?

Not by the measured benefits the economists could quantify. The 2019 Quarterly Journal of Economics paper by Keiser and Shapiro found that the average grant project cost about $31 million in 2014 dollars while raising the value of housing within 25 miles of the affected river by about $7 million, a benefit-to-cost ratio of roughly 0.25 on that measure. Whether the program passes a broader benefit-cost test depends on the unmeasured categories: the authors argue the ratio could exceed their estimate once nonuse values, health effects, and the value perceived by people beyond the 25-mile radius are counted, while conceding that some investments may genuinely fail the test. The federal government spent about $650 billion on the grants in total, making this the largest measured water quality investment in American history.

Q: What share of assessed US waters still fail to meet their designated uses under the Clean Water Act?

Roughly 44 to 47 percent of assessed river and stream miles in the EPA reporting cycles of the early 2000s. The 2004 National Water Quality Inventory, which summarized state assessments of 16 percent of the nation’s 3.5 million river and stream miles, reported 44 percent impaired, meaning not clean enough to support at least one designated use such as fishing or swimming, with 56 percent fully supporting all assessed uses. The 2002 cycle reported 45 percent impaired among 19 percent of miles assessed. The 1996 report found 36 percent of surveyed miles impaired. The impairment share is not a trend line: states assess different waters in different cycles, against standards that tighten over time, and only a minority of total miles is ever assessed. But the stability of the figure across cycles is the factual basis for the summary that about half of assessed waters fall short more than fifty years after the act’s passage.

Q: How close did the country come to the Clean Water Act’s 1983 fishable-swimmable goal?

Closer than to the 1985 zero-discharge goal, but still short. Section 101(a)(2) of the 1972 act set an interim national goal of achieving, by July 1, 1983, water quality sufficient to protect and propagate fish, shellfish, and wildlife and to allow recreation in and on the water, the objective everyone calls fishable-swimmable. Keiser and Shapiro’s 2019 Quarterly Journal of Economics analysis found that the share of waters violating fishability standards fell substantially after 1972, with the gains concentrated downstream of the municipal plants the construction grants upgraded. EPA’s 2004 National Water Quality Inventory reported that 36 percent of stream miles assessed for the fish, shellfish, and wildlife use were impaired for that use, and 28 percent of waters assessed for recreation were impaired. So the country moved measurably toward the interim goal without attaining it, which fits this article’s series thesis: a statute assessed against its own stated aims, including aims written as deadlines that were never met.

Q: What share of remaining water quality impairment comes from nonpoint sources the Clean Water Act does not reach?

The largest share, though EPA’s inventories report it by source category rather than as a single number. In the 2004 National Water Quality Inventory, the top sources of impairment in rivers and streams were agricultural activities, hydrologic modifications such as diversions and channelization, and unknown or unspecified sources, with pathogens, habitat alterations, and organic enrichment or oxygen depletion as the leading causes. Keiser and Shapiro’s 2019 Journal of Economic Perspectives survey describes nonpoint runoff as the dominant remaining source and notes that it is the category the statute reaches only through section 319 grants and state planning, with no permit requirement and no enforceable limits. Because the point source program spent decades and hundreds of billions of dollars driving down pipe discharges, the pollution that remains is disproportionately the pollution the act’s central mechanism was never designed to touch.

Q: Why is public perception of the Clean Water Act more positive than attainment statistics?

Because people judge the statute by the visible disasters it ended, while the statistics judge it by designated uses it never fully attained. Rivers catching fire, raw sewage floating in harbors, and industrial corridors choked with waste were the conditions that produced the 1972 act, and those conditions ended in the early 1970s: no river has burned the way the Cuyahoga did in June 1969, and the organic loadings that created such fires fell sharply under the permit and grant programs. Decades of Gallup polling ranked water pollution as Americans’ top environmental concern, which keeps the memory of the pre-1972 baseline vivid. The attainment statistics, by contrast, measure whether assessed waters meet every designated use under standards that have tightened over time, and they count nutrient and sediment impairment from farm runoff that was never the public’s image of dirty water. The two verdicts use different yardsticks, and both are accurate on their own terms.

Q: Why are Clean Water Act outcomes harder to measure than Clean Air Act outcomes?

Because water monitoring is manual, local, and fragmented, while air monitoring is automated and standardized. Shapiro put the contrast directly in the coverage of the 2019 studies: air pollution and greenhouse gas measurements are typically automated and standard, while water pollution is more often a person going out in a boat and dipping something in the water. Keiser and Shapiro had to compile data from the EPA’s STORET database and USGS monitoring records and file two Freedom of Information Act requests just to locate every municipal grant before any analysis could begin. Air quality also rests on a denser, more uniform monitoring network and on pollutants that disperse in ways models can track, whereas water quality varies mile by mile along a river network and responds to rain events that sampling schedules miss. The result is that the Clean Air Act’s record rests on continuous national datasets, while the Clean Water Act’s record had to be reconstructed from scattered local measurements decades later.

Q: Why are the Clean Water Act’s costs easier to measure than its benefits?

Because the costs arrive as invoices and the benefits arrive as experiences nobody prices. The act’s largest costs are the construction grants, about $650 billion in federal spending on municipal sewage treatment plants, plus the compliance costs industry and municipalities report under the permit system, all of it recorded in budgets, appropriations, and capital accounts. Keiser and Shapiro’s 2019 survey in the Journal of Economic Perspectives estimated total public and private spending on surface water pollution control at roughly $3 trillion since 1970, in 2017 dollars. The benefits have no equivalent paper trail: cleaner water raises no revenue, and its value shows up only indirectly in housing prices, recreation choices, and health outcomes that must be estimated statistically. Economists call this the measurement gap, and it is this article’s namable claim: the Clean Water Act is the clearest case in American regulation of a statute whose benefits are systematically harder to measure than its costs, so any benefit-cost verdict is partly a verdict on the measurement method.