A reader who finishes this article should be able to state four things without hesitation and one thing with deliberate humility. The four: the built environment changed materially over three decades, with curb ramps, accessible transit fleets, and accessible new construction arriving as a continuous, law-driven accumulation; telecommunications relay service grew from a statutory command into a national infrastructure, later extended to internet-based communications; long-term services for people with disabilities shifted measurably from institutions toward home and community settings, with the spending data to prove it; and all of this can be documented with named sources and dated periods. The fifth, the one that requires humility: whether the statute increased or reduced employment among people with disabilities is genuinely unresolved, because respected economists have published respected findings on both sides, and the identification problems that separate them have never been fully settled. That asymmetry, access settled and employment open, is the whole of this article. Every section below exists to earn it.

The one test this article applies

This is the hardest evidence article in its cluster, because its subject is the one where the public argument most outruns the research. The method it uses is the one the series applies throughout: assess a statute against its own aims, and name precisely which of those aims the evidence can and cannot speak to. The Americans with Disabilities Act, Public Law 101-336, stated its purposes plainly, among them assuring equality of opportunity, full participation, independent living, and economic self-sufficiency. Those are different kinds of promises. Some of them concern the physical and institutional world, and the world left tracks. Some concern labor markets, where cause and effect are entangled with everything else happening in an economy, and the tracks are faint, disputed, and sometimes drawn by the survey instrument itself rather than by the world it measures. The honest response is to grade each aim separately, and to refuse the two temptations that surround this literature: the temptation to cite one employment paper as though it settled the question, and the temptation to dismiss the uncomfortable papers as though they were never written. The provisions evaluated here are the employment title, the public accommodations title, and the telecommunications title, described in the guide to the Americans with Disabilities Act, and the integration mandate of the public-services title, whose litigation history the series traces separately.

The article organizes the record into four layers, and each layer gets its own evidentiary rating in the artifact table near the end. The first layer is the built environment: curb ramps, transit vehicles and stations, new construction and altered facilities. This is the least contested ground, because the changes are physical, cumulative, and anchored to dated regulations and dated transit-agency reports. The second layer is telecommunications, where the relay obligation created a national infrastructure that had not existed before, with the statute’s text, the implementing agency’s orders, and dated extensions to internet-based relay marking each step. The third layer is institutional care, where the integration mandate and the litigation and Medicaid policy changes that followed it drove a sustained, measured shift of long-term services from institutions toward home and community settings. The fourth layer is employment, where the literature is genuinely divided, and where the article gives the early findings, the qualifying findings, and the measurement confounds equal space and equal seriousness, because fairness to the reader is the entire competitive advantage of this piece.

A note on dates governs everything that follows. The reference date for this article is September 1, 2015. The record described is the record available through that date. Findings, filings, and agency releases through December 2014, and a small number of anniversary announcements from the summer of 2015, form the at-reference record. Anything after that date is identified with its date and kept outside the at-reference record. The wall matters most in two places. The first is the employment literature, where later research exists but does not enter as fact here. The second is terminology: a phrase that has become the standard label for the spillover pattern of curb-cut use was coined in a magazine article in the winter of 2017, two years after the reference date, and this article therefore never uses it as a term of art. It describes the documented phenomenon instead, with the pre-2015 documentation attached. The finding is older than the label, and the label’s absence costs nothing.

One more piece of framing belongs up front, because it shapes how the credit is assigned. Not all of the accessibility change of the last four decades belongs to this statute. Section 504 of the Rehabilitation Act of 1973, a funding-condition statute, had been requiring accessibility as a condition of federal funds since the 1970s, and part of the accessible built environment that the record shows was built under that earlier regime’s pressure. The article assigns credit accordingly, describing where the 1990 statute extended the obligation beyond federally funded programs and where it simply continued work that was already underway. A law that widened a mandate deserves credit for the widening, not for the whole of the movement, and the built-environment section keeps the two apart.

The four-layer evidence table at the end of the article carries the verdict in compact form: for each domain, what changed, the period, the principal sources, and a rating of settled, measurable, or contested. Settled means the direction of change is not seriously disputed and multiple sources document it. Measurable means the change is documented with numbers but the attribution to the statute, or the completeness of the change, requires qualification. Contested means respected researchers disagree about the direction of the effect itself, and no honest summary can pick a winner. The built environment and telecommunications earn settled. Institutional care earns measurable. Employment earns contested, and the article says so in the open, because a reader who leaves believing the jobs question is answered has been misled.

The four evidence layers map onto the statute’s titles in a way that keeps the measurement honest. Title I, the employment title, is the source of the contested layer: its prohibitions on employment discrimination and its reasonable accommodation requirement are the provisions whose labor market effects the economists debated. Title II, covering state and local government programs and services, is the source of two settled layers: its integration regulation produced the Olmstead mandate behind the institutional shift, and its transportation and facility obligations contributed to the transit and curb ramp changes. Title III, covering private places of public accommodation, is the other source of the built environment layer, through the new construction and alteration requirements that reshaped commercial buildings. Title IV, the telecommunications title, stands alone as the source of the relay infrastructure. Keeping those mappings explicit prevents the common error of treating the statute as a single treatment with a single effect, when it was four distinct interventions whose outcomes must be measured separately.

The One Test for this article is deliberately demanding. It does not ask the reader to recite the statute’s titles or its passage history. It asks for something harder: a measurable statement about the built environment, about telecommunications, and about institutional care, followed by an honest confrontation with the employment question, including the names on both sides of it. A reader who can say that curb ramps accumulated under dated alteration rules, that relay service became nationally uniform in July 1993, that community-based services crossed half of Medicaid long-term spending in 2013, and that Acemoglu and Angrist found declines while Kruse and Schur showed the estimates turn on definitions, has passed. A reader who can only say the law was important has not, because importance is not a finding.

The brief for this article calls it the hardest evidence piece in its batch, and the reason is the empirical-framing risk that runs through every paragraph. Most impact articles report findings that point in one direction, and the writer’s job is accuracy. This article reports a literature that points in two directions on its most contested question, and the writer’s job is fairness under conditions where fairness itself will be read as taking a side. Readers who believe the statute harmed employment will read the qualification sections as hostile. Readers who believe the statute helped will read the early-findings sections as hostile. The article’s defense is structural rather than tonal: equal length for both sides, named authors with publications and periods for every finding, the identification problems stated explicitly rather than implied, and the uncomfortable results given the same prominence as the comfortable ones. The competitive advantage the brief names, reporting the difficult finding fairly, is not a rhetorical posture. It is a set of measurable choices about space, attribution, and order, and the article makes them the same way in every section.

There is a reason the four layers get different ratings, and the reason is worth stating as a general principle about evidence. Physical changes leave the kind of evidence that accumulates without statistical argument. A ramp either exists at an intersection or it does not. A bus either has a lift or it does not. A spending share either crosses fifty percent or it does not. Incentive changes in a labor market leave a different kind of evidence, one that has to be extracted by comparing a treated group with a comparison group, and the extraction is only as good as the comparison. The built-environment and telecommunications layers are rated settled because their comparisons are clean: the world before the mandate and the world after it, documented in regulations and agency reports. The employment layer is rated contested because its comparisons are entangled with everything else that moves in an economy, and the entanglement has survived three rounds of increasingly careful research. The ratings are not grades on the statute. They are grades on what can be known, and the article treats the distinction as load-bearing.

A final piece of framework concerns attribution, the question of how much of each measured change the statute caused. The article’s discipline is to distinguish three statements that public argument constantly blends: that a change happened, that the change happened after the statute, and that the statute caused the change. The first two are often documented. The third requires a comparison, and the quality of the comparison determines how much causation can be claimed. In the built-environment layer, the comparison is the regulatory trigger: ramps appear where resurfacing occurs because the 2013 technical assistance says they must, which is about as close to a documented causal chain as observational evidence gets. In the telecommunications layer, the comparison is the statutory deadline and the agency orders that met it. In the institutional-care layer, the attribution is shared, and the article says so, because litigation, state policy, and demographics all pushed the same direction. In the employment layer, the comparison is the whole dispute. The article never upgrades a post hoc sequence into a causal claim without naming the comparison that licenses the upgrade, and where no comparison licenses it, the article reports the change and leaves the causation open.

The built environment: the least contested layer

The physical world is where the statute’s effects are easiest to see and hardest to argue with, because concrete does not revise itself in later editions. Over roughly three decades, curb ramps spread across American streets, transit fleets converted to near-universal accessibility, and new construction and altered facilities were built to accessibility standards as a matter of routine. The mechanism was not a single mandate but a stack of them: design standards for new construction, alteration rules that pulled existing facilities into compliance over time, and a Title II regulation that treated routine street work as a trigger for ramp installation. The result was continuous accumulation rather than a one-time conversion, which is why the evidence takes the form of dated benchmarks rather than a single before-and-after snapshot.

The design standards arrived early. The 1991 ADA Accessibility Guidelines set the technical requirements for accessible new construction and alterations, and they were superseded in time by the 2010 ADA Standards for Accessible Design, issued as regulation at 28 CFR Part 36. The core rule was structural: facilities built or altered after 1990 had to be accessible. That rule changed the building stock the way sediment changes a riverbed, gradually and then comprehensively, because every new building and every qualifying renovation added accessible square footage to the total. The article does not state a precise share of American buildings that are accessible, because no verified figure exists, and inventing one would be worse than admitting the gap. What the record supports is the mechanism and its direction: the stock of accessible construction grew continuously from 1990 onward by operation of law, and the alteration requirements meant that existing facilities moved in the same direction whenever they were substantially renovated.

The regulatory stack deserves a closer look, because its design explains the continuity of the change. The 1991 guidelines were the first comprehensive technical standard, and the 2010 standards that superseded them updated the technical requirements without interrupting the obligation. That continuity matters: a regulated community that faces one stable, updated standard over two decades plans differently from one that faces a new rule every few years, and the accessibility of the American building stock reflects the steadiness of the requirement as much as its stringency. The regulation at 28 CFR Part 36 carried the 2010 standards into enforceable law, and the alteration provisions carried them backward into existing buildings. The combination is what produced the sediment effect described earlier: new construction adding accessible floor area every year, renovations pulling older floor area into compliance on their own cycle, and the two streams never pausing.

The compounding logic extends to alterations, which the statute and its regulations treated as triggers for accessibility upgrades in existing facilities. An older building that underwent substantial alteration after 1990 had to bring the altered portions into compliance, which meant that renovation cycles, like paving cycles, carried the obligation into the pre-1990 stock gradually. The 2013 joint technical assistance on street resurfacing is the clearest illustration of the principle applied to the public right-of-way: routine maintenance, the kind of work cities do every year without thinking of it as construction, counted as an alteration and triggered ramp installation at every intersection in the resurfaced area. The mechanism converted the ordinary maintenance budgets of thousands of local governments into an accessibility program, year after year, without any of those governments voting to create one.

Why is the built environment evidence the least contested?

Because the changes are physical, cumulative, and dated: ramps, lifts, and accessible construction accumulate under dated regulations, and transit agencies publish dated fleet figures, so the record consists of verifiable benchmarks rather than modeled estimates that depend on survey definitions.

The curb ramp story deserves to be told in full, because it contains both the oldest accessibility work in the article and the most misattributed. The modern American curb ramp did not begin with the 1990 statute. In Berkeley, California, wheelchair activists including Ed Roberts, the group known as the Rolling Quads, and Hale Zukas persuaded the city’s public works department to install cement curb ramps at intersections along Telegraph Avenue in 1969 and 1970, and when official action lagged, activists poured their own concrete. The city’s Resolution No. 45,605-N.S., adopted February 13, 1973, mapped a 125-cut wheelchair route through the city. The pre-2015 documentation for this history includes Edward Steinfeld’s curb-ramp history, published through the University at Buffalo in 2007, and Bess Williamson’s account, “The People’s Sidewalks,” published in 2014. The standard Berkeley narrative, repeated in multiple sources including Williamson’s 2014 account, holds that the ramps built for wheelchair users were soon used heavily by parents with strollers, delivery workers, travelers with luggage, and cyclists. The article presents that as the historical origin story it is, not as a named social-science concept, and it does not quantify the spillover, because no verified national study of ramp beneficiaries exists. The point of the story is narrower and better supported: the pattern in which an accommodation built for disabled users gets taken up by everyone else was visible in the streets of Berkeley decades before anyone gave it a label, and the label’s 2017 coinage changes nothing about the underlying documentation.

The Berkeley history repays attention for a different reason. It is the article’s answer to the question of where American accessibility came from, and the answer is not Washington. Ed Roberts, paralyzed by polio, arrived at the University of California, Berkeley, in the 1960s and organized with other wheelchair users, the Rolling Quads, into a political force. Hale Zukas worked the municipal side. Their target was the curb, the six-inch concrete barrier that made every street crossing a negotiation. The Telegraph Avenue ramps of 1969 and 1970 were the first cuts, and the activists’ willingness to pour their own concrete when the city moved slowly is part of the documented record, not legend. The 1973 resolution mapping the 125-cut route turned a protest tactic into municipal infrastructure planning. The spillover pattern that later got its label, the strollers and delivery workers and travelers with luggage and cyclists of Williamson’s 2014 account, was visible to anyone watching those Berkeley intersections in the early 1970s. The 1990 statute nationalized the Berkeley insight. It did not invent it, and the article’s dating keeps the invention where it belongs.

What the 1990 statute added to the Berkeley story was scale and legal compulsion. The Title II regulations required curb ramps wherever new roads and sidewalks were constructed or existing ones were altered, and the joint technical assistance issued by the Department of Justice and the Department of Transportation in 2013 confirmed that street resurfacing counts as an alteration that triggers curb-ramp installation at all intersections in the resurfaced area. That ruling matters because it turned every repaving cycle into an accessibility project: cities do not repave on a disability schedule, but the regulation made the repaving schedule serve accessibility anyway. The result is that curb-ramp coverage expanded continuously after 1990 with each construction season, in every jurisdiction that maintained its streets. The article does not offer a national count of intersections with curb ramps or a percentage of compliant corners, because no citable figure was verified, and the New York survey sometimes cited in this connection postdates the reference date. The verified claim is the mechanism: the regulatory trigger, the 2013 confirmation, and the continuous accumulation that followed.

Transit accessibility supplies the hard numbers that the ramp story lacks, and they are striking. The American Public Transportation Association’s 25th-anniversary release of July 26, 2015, and the Federal Transit Administration’s announcement of July 23, 2015, both pre-reference-date, reported the same trajectory. The accessible share of the nation’s bus fleet rose from 51 percent in 1994 to 99.8 percent in 2015, achieved through lifts and ramps on the vehicles themselves. Heavy rail and subway vehicles went from 83 percent to 100 percent accessible over the same span. The light rail and streetcar fleet went from 41 percent to 88 percent. An earlier segment of the same series, drawn from the association’s 2009 Public Transportation Fact Book, shows buses moving from 60 percent accessible in 1995 to 98 percent in 2008. Those figures describe vehicles, not stations, and the article keeps that distinction explicit: a lift-equipped bus is a different achievement from an accessible platform, and the station record is less completely documented. But the vehicle trajectory is one of the clearest measured accessibility gains in the entire record, and it is dated, sourced, and large.

The bus figures repay the same close attention the ramp story received. The 51 percent accessible figure of 1994 and the 99.8 percent of 2015, with the intermediate marker of 60 percent in 1995 to 98 percent in 2008, describe a fleet replacement cycle harnessed to a legal requirement: as agencies retired old buses and bought new ones, the new ones arrived with lifts and ramps, and the old ones left. The mechanism is worth naming because it recurs across the article. The statute rarely ordered an immediate retrofit of everything. It attached accessibility to the normal replacement cycle, whether of buses, buildings, or street surfaces, and let the cycle do the work over decades. Transit fleets make unusually clean measurement subjects. Buses and rail cars are countable, centrally procured, and replaced on predictable cycles, so an accessibility mandate expressed as a vehicle standard produces a time series that can be checked against agency reports. The association and the administration figures used here come from the industry and the regulator respectively, and they agree, which is a further check.

The alteration rules deserve emphasis because they are the least understood part of the built-environment story and the part that most distinguishes the statute from a mere building code. A building code typically governs new construction. The accessibility regime reached backward into the existing stock through the alteration requirements: when a facility was altered, the altered portions had to meet the standards, and in many cases the path of travel to the altered area had to be made accessible as well. That design choice meant that the statute did not wait for the building stock to turn over. It rode the normal cycle of renovation, and every renovation cycle pulled more of the existing stock into compliance. Combined with the new-construction standards, the alteration rules explain why the change feels both gradual and, over three decades, near-total in the categories the law reached. The path-of-travel concept is what gave the rule its reach: renovating a single floor could require accessible entrances, corridors, and restrooms serving that floor, so the compliance obligation propagated outward from each renovation like ripples.

The credit question returns here, and it deserves a direct answer. The statute did not invent American accessibility. Section 504’s funding condition had been pushing federally assisted programs toward accessibility for more than fifteen years before 1990, and the early Berkeley work predates even that. What the 1990 law did was extend the obligation to the private sector and to state and local government as such, rather than as recipients of federal money, and to back the standards with a private right of action. The new construction of the 1990s and 2000s, the alteration cycles, the transit conversions, and the ramp installations triggered by resurfacing all sit on the statute’s side of that line. The university buildings, hospitals, and transit systems that had already been working under the earlier funding condition sit partly on the other. The article does not pretend the line can be drawn facility by facility. It states the direction of the credit and leaves the precise division where the record leaves it, which is qualitative.

What accessibility meant operationally for transit riders is worth spelling out, because the percentages become concrete when attached to the experience. An accessible bus, in the Federal Transit Administration’s 2015 accounting, is a bus usable by people with disabilities through features like lifts and ramps: the vehicle kneels or deploys a ramp, a wheelchair user boards and secures, and the bus continues its route. The 99.8 percent figure means that experience was available on essentially every bus in the national fleet by 2015, up from barely half in 1994. For heavy rail and subway vehicles, accessibility means level boarding or on-board accommodations that let wheelchair users ride the same cars as everyone else, and the move from 83 percent to 100 percent meant the last inaccessible cars left the fleet. The light rail and streetcar figure, 41 percent to 88 percent, reflects the harder problem of older streetcar systems with historic vehicles and constrained rights of way. The station side of the ledger, platforms, elevators, and accessible paths through stations, is the documented gap in the record: the article reports the vehicle figures because the agencies published them, and it does not extrapolate from vehicles to stations, because the verification found no citable station series and the article does not manufacture one.

The three cycles together describe a coherent theory of how the statute remade the physical world, and the theory is worth naming because it explains the otherwise puzzling fact that a 1990 law was producing new accessibility in 2015. The construction cycle added accessible floor area with every new building. The renovation cycle pulled existing buildings toward the standards with every alteration. The replacement cycle converted transit fleets and street surfaces on their own schedules, with the alteration and resurfacing triggers ensuring that accessibility rode along. None of the three cycles required a separate appropriation, a new program, or a fresh political fight. Each one harnessed a process that was already happening, building, renovating, replacing, and redirected a share of it toward accessibility by legal rule. That design is why the change was continuous rather than episodic, why it survived changes of administration without new legislation, and why the benchmarks dated 1994, 2008, 2013, and 2015 all point the same direction.

Telecommunications: a relay infrastructure built by statute

Before the statute, a deaf or speech-disabled American who needed to make a telephone call faced a fragmented landscape: some states had relay programs, many did not, and there was no national system. The statute’s telecommunications title changed that by creating a federal obligation and assigning it to an agency with the power to enforce it. Title IV added Section 225 to the Communications Act of 1934, codified at 47 U.S.C. section 225, and directed the Federal Communications Commission to ensure that telecommunications relay services were available, in the statute’s words, to the extent possible and in the most efficient manner to individuals with hearing or speech disabilities. The obligation fell on common carriers providing voice telephone service, which had to make relay service available throughout their service areas beginning July 26, 1993.

The implementation timeline shows a regulator moving faster than the statute required and then being pulled forward by technology. The Commission first ordered all carriers to provide relay service nationwide on July 26, 1991, two years before the statutory deadline, and relay service became available on a uniform nationwide basis in July 1993, as the Commission’s Second Report and Order recorded and the Federal Register confirmed in September 2000. That sequence matters for attribution: the national infrastructure that resulted was not an accident of the market and not a patchwork of state experiments that happened to converge. It was the product of a statutory command, a federal agency’s implementing orders, and a hard deadline, and the uniformity of the resulting system, the same service available across the country rather than fifty different arrangements, is the signature of federal action.

Why did relay service need a federal mandate?

Because telephone networks cross state lines and no single state’s program could compel nationwide carriers: only a federal obligation on common carriers, enforced by the Federal Communications Commission, could produce the uniform relay system that became available across the country in July 1993.

The Commission’s 1991 order deserves emphasis because it shows the regulator outpacing the statute. Ordered to have relay available by July 1993, the Commission required nationwide provision by July 1991, and the uniform national availability recorded in July 1993 reflected two years of operating experience, not a scramble to meet a deadline. The sequence matters for anyone tempted to attribute the relay system to market forces or state experimentation. The verified baseline is that no national relay infrastructure existed before the mandate: the system the statute created was national and uniform where nothing national and uniform had existed. The uniformity of the post-1993 system, the same functional service available to a caller in any state, is the observable signature of federal compulsion, and it is why the attribution in this layer is cleaner than in any other part of the article.

The statute’s relay system did not freeze in 1993. It followed communications technology, and the dated extensions show a regime adapting rather than aging. Abbreviated dialing for relay access, the three-digit 711 code, was required by October 1, 2001, under the Commission’s Second Report and Order of 2000, which made the service as easy to reach as dialing three digits rather than a full access number. Video relay service, in which a deaf caller communicates through a sign-language interpreter over video, was recognized by the Commission as a form of relay service in the Improved TRS Order of March 6, 2000, designated FCC 00-56. Internet-protocol relay, which carries relay conversations over the internet rather than the telephone network, was recognized as a form of relay service in the Commission’s IP Relay Declaratory Ruling of April 22, 2002, published at 17 FCC Rcd 7779. Each of those dates sits before the article’s reference date, and each represents the same pattern: the statutory framework absorbing a new communications medium rather than being overtaken by it.

Each technological extension tells its own story about usability. The 711 dialing code, required by October 1, 2001, addressed the discovery problem: a relay system that requires users to remember and dial a ten-digit access number is a system that many potential users will never reach, and collapsing access to three digits put relay on the same footing as other abbreviated services. Video relay, recognized in the March 6, 2000 order, addressed the modality problem: for sign-language users, typing conversations through a text relay is a poor substitute for visual communication, and video relay restored the caller’s own language to the conversation. Internet-protocol relay, recognized in the April 22, 2002 ruling, addressed the network problem: as voice communication migrated to the internet, a relay system confined to the telephone network would have stranded its users on a shrinking platform, and the ruling followed the users onto the new network. The three extensions share a logic. The statute promised functional equivalence in telecommunications, and each extension redefined what equivalence required as the underlying technology moved.

The statutory language repays quotation, because it shows Congress assigning a problem to an agency rather than pretending to solve the engineering itself. Title IV directed the Federal Communications Commission to ensure that relay services were available to the extent possible and in the most efficient manner. That phrasing is a delegation: the statute set the goal and the deadline, July 26, 1993, for carrier provision throughout service areas, and left the Commission to work out what relay service meant in practice as technology changed. The delegation is why the system could absorb video and internet-protocol relay a decade later without new legislation. A statute that had specified the technology of 1990 would have been obsolete by 2000. A statute that specified the function and empowered the regulator stayed current, and the dated orders of 2000, 2001, and 2002 are the evidence that the design worked as intended.

The telecommunications layer earns its settled rating for a specific reason. Unlike employment, where the outcome of interest is entangled with the whole economy, the relay story has a clean chain from statute to agency order to dated nationwide availability, and the infrastructure did not exist before in anything like its post-statute form. A skeptic can dispute the cost of the system, the details of its funding, or the quality of particular services, and the article takes no position on those questions because the verification record does not resolve them. What the record does resolve is existence and scale: a national relay infrastructure, created by a statutory command and extended by dated agency orders through 2002, where none had existed before. That is the kind of finding this article calls settled, and the rating means only that, not that every question about the system is answered.

Institutional care: from institutions toward community

The third layer concerns where people with significant disabilities live and receive long-term services, and it is the layer where a single court decision and a financing system interacted to produce a measurable national shift. The statute’s public-services title contains an integration regulation, 28 C.F.R. section 35.130(d), which requires public entities to administer services in the most integrated setting appropriate to the needs of the individual. For the first nine years of the statute’s life, that regulation’s full implications for residential care were contested. The Supreme Court resolved the contest in Olmstead v. L.C., 527 U.S. 581, decided June 22, 1999, holding that the title prohibits the unjustified segregation of individuals with disabilities and requiring public entities to provide community-based services when the services are appropriate, when the affected person does not oppose them, and when the placement can be reasonably accommodated given the public entity’s resources and the needs of others. The decision’s background and its place in the statute’s litigation history belong to the series’ account of ADA Supreme Court cases, and this article takes the holding as its starting point rather than re-litigating it.

The holding did not by itself move anyone out of an institution. What moved the system was the interaction between the legal rule and the financing: Medicaid pays for the overwhelming share of publicly funded long-term services and supports, and the program’s structure determines whether community services are a real alternative or a paper right. For the program’s architecture, the series carries a separate explainer of the Medicaid statute’s structure. The relevant fact here is that Medicaid’s home and community-based services, the waivers and state-plan options that fund care outside institutions, expanded enormously in the years after the 1999 decision, driven by litigation enforcing the integration mandate and by state policy choices responding to it.

Did the 1999 integration decision end institutionalization?

No. It required community placement where appropriate and unopposed, and the financing followed: home and community-based services rose from 18 percent of Medicaid long-term spending in 1995 to 51 percent in 2013, a shift of the system rather than its abolition.

The numbers are the strongest in this layer of the article. Home and community-based services rose from 18 percent of Medicaid long-term services and supports spending in 1995 to 51 percent in 2013, and fiscal year 2013 was the first year in which community-based services constituted a majority of Medicaid long-term spending, with total Medicaid long-term spending reaching 146 billion dollars that year. Those figures come from the Kaiser Family Foundation and University of California, San Francisco report “Medicaid Home and Community-Based Services Programs: 2013 Data Update,” by Terence Ng, Charlene Harrington, MaryBeth Musumeci, and Priya Ubri, published through Brandeis and Kaiser in 2016 reporting the 2013 data, and from the Centers for Medicare and Medicaid Services’ 2013 Medicaid long-term services expenditure report, released through the agency’s announcement channel in July 2015, both inside or reporting inside the reference window. A Congressional Research Service report from 2014, number R43804, described the same trajectory in plainer terms, saying the community-based share had more than doubled from 18 percent in 1995 to about half in 2012. A 2015 Health Affairs policy brief, “Rebalancing Medicaid Long-Term Services And Supports,” treated the shift as the central fact of the field. The article presents the spending shift as the measured outcome, not as proof that the statute alone caused it, because state policy, litigation, and demographic pressure all contributed, but the direction, the magnitude, and the timing, accelerating after the 1999 decision, are all documented.

The data sources for the spending shift complement each other, and the article names all four because each contributes something the others lack. The Kaiser Family Foundation and University of California, San Francisco report provides the long series, 1995 to 2013, with the program-level detail that shows which waiver authorities drove the growth. The Centers for Medicare and Medicaid Services expenditure report for 2013 provides the official federal accounting, with the 146-billion-dollar total that anchors the percentages. The Congressional Research Service report from 2014 provides the plain-language summary, the more-than-doubling from 1995 to about half in 2012, in the form legislators actually read. The 2015 Health Affairs policy brief provides the field’s own characterization of the shift as rebalancing, the term of art for the movement of long-term services from institutions toward community settings. Four sources, four institutional perspectives, one directional finding: the convergence is what makes the measurement credible.

The three conditions of the Olmstead holding deserve to be unpacked, because each one is doing analytical work. The appropriateness condition means the decision does not order community placement for people whose needs cannot be met there; it is an anti-segregation rule, not an anti-institution rule, and it leaves room for institutional care where the individual’s needs require it. The non-opposition condition means the affected person, or their guardian, must not oppose the placement; the decision protects autonomy in both directions, against unwanted institutionalization and against unwanted discharge. The reasonable-accommodation condition means the public entity’s resources and the needs of others count; a state is not required to dismantle its service system overnight or to deny services to one population to serve another. Together the three conditions describe a qualified right rather than an absolute one, and the litigation that followed spent years working out what each qualification meant in practice for waiting lists, funding formulas, and service definitions.

The regulation behind the decision is older than the decision itself. Section 35.130(d) of Title 28 of the Code of Federal Regulations required from the statute’s early years that public entities administer services in the most integrated setting appropriate, and the 1999 decision gave that regulatory phrase its authoritative judicial meaning. The sequence matters because it shows the integration mandate was not invented in 1999. It was present in the regulatory structure from the beginning, contested for nine years, and then construed by the Court. The years between the regulation’s issuance and the decision help explain why the spending shift measured from 1995 already shows movement before 1999: the legal standard was operating, if less forcefully, before the Court clarified it, and Medicaid waiver policy was moving independently in the same direction.

The transmission from legal rule to spending shift ran through two channels, and the article distinguishes them. The first was litigation: Olmstead enforcement suits against states with large institutional populations and long waiting lists for community services, which produced settlements and court orders requiring states to expand community capacity. The second was policy: state Medicaid agencies, reading the direction of the law, expanded home and community-based waiver programs and state-plan options, sometimes under litigation pressure and sometimes ahead of it. The two channels reinforced each other, because each settlement set a precedent that other states’ policymakers could read, and each voluntary expansion reduced the litigation risk for the states that undertook it. The article does not quantify the relative contribution of litigation and voluntary policy, because no verified decomposition exists. It describes the mechanism qualitatively and lets the spending data carry the measurement.

The rating for this layer is measurable rather than settled, and the distinction is deliberate. Settled would imply that the change is complete and its attribution is clean. Neither is true. Institutions continued to operate, the community-based majority was a spending majority rather than a population majority, and the causes included state waiver policy and budget politics alongside the integration mandate. What the record supports without qualification is the measurement: the share of public long-term spending going to home and community settings roughly tripled over eighteen years, crossing the majority line in 2013. That is a large, dated, sourced change in the direction the statute’s integration mandate pointed, and it is more than enough to count as an outcome, even if the mandate was only one of its causes.

The 2013 milestone deserves a moment of its own, because crossing a majority line is the kind of event that changes how a field talks about itself. When community-based services became the majority of Medicaid long-term spending, the institutional model stopped being the default setting of public long-term care and became the minority alternative. The field’s term for the shift, rebalancing, captures exactly that: not the abolition of institutions, which continued to serve people whose needs required them, but the reweighting of the system toward the setting the integration mandate preferred. A reader encountering the 51 percent figure should understand what it marked: the year the financing of American long-term care tipped, after eighteen years of movement, toward the community side of the ledger, in the direction the 1999 decision had pointed and at a pace the litigation and waiver expansions had set.

Two dating cautions apply to this layer. First, the spending series runs through 2013, the most recent verified data year, and this article does not project it forward. A later projection that the home and community share would reach 63 percent was identified during verification as a forecast rather than a measured fact, and it is excluded on that ground. Second, the integration mandate’s later legal history is date-walled: a 2025 Department of Justice opinion taking a different view of the mandate postdates the reference date by a decade and reflects a later administration’s position, and it has no place in an article measuring what the statute changed through 2015. The rating for this layer is measurable with shared attribution: the spending shift is documented, the legal pressure is documented, and the precise division of credit between the statute, Medicaid policy, and state decisions cannot be isolated from the available data.

Employment: the early findings

The employment title took effect on July 26, 1992, for employers with twenty-five or more employees, and its coverage expanded on July 26, 1994, to employers with fifteen or more employees, according to the Equal Employment Opportunity Commission’s official questions-and-answers publication on the statute. The Commission’s 1994 press notice estimated that the expansion brought the covered employer count to roughly 666,000 businesses and 86 million workers, up from 264,000 businesses. The public-services title’s employment provisions for state and local governments took effect earlier, on January 26, 1992. Those dates matter because the entire employment literature is organized around them: every paper in the debate defines a before and an after, and the credibility of each paper depends in part on whether its after actually starts when the law started.

The first wave of findings was unfavorable, and the article reports it at full length because the brief for this piece names fair reporting of the uncomfortable finding as its entire competitive advantage. Daron Acemoglu and Joshua D. Angrist published “Consequences of Employment Protection? The Case of the Americans with Disabilities Act” in the Journal of Political Economy, volume 109, number 5, in October 2001, pages 915 to 957, working from a working paper circulated as National Bureau of Economic Research Working Paper 6670 in July 1998. Using March Current Population Survey data from 1988 through 1997, they found a sharp drop in the employment of disabled workers after the statute went into effect, for men of all working ages and for women under forty. The effects were larger in medium-size firms, which the statute covered while exempting small firms, and in states with more disability-related discrimination charges, a pattern that points toward the statute rather than a background trend, since a background trend would not have respected the law’s coverage boundaries. The decline reflected reduced hiring rather than increased separations, which weighs against the interpretation that the law raised firing costs and for the interpretation that it raised the expected cost of taking on a disabled worker. The authors examined whether the rise in disability benefit receipt could explain the decline and concluded it did not appear to explain it alone, leaving the statute as a likely cause. They found little evidence of an effect on workers without disabilities. The mechanism they proposed was the cost of accommodation and the cost of potential termination disputes, both of which the statute imposed on employers and both of which could rationally make employers more cautious about hiring.

Did respected economists find that the statute reduced employment?

Yes. Acemoglu and Angrist, in the Journal of Political Economy in October 2001, and DeLeire, in the Journal of Human Resources in autumn 2000, both found post-statute employment declines among disabled workers, using different surveys and methods within a year of each other.

Thomas DeLeire reached a similar conclusion by a different route. His paper, “The Wage and Employment Effects of the Americans with Disabilities Act,” appeared in the Journal of Human Resources, volume 35, number 4, in the autumn of 2000, pages 693 to 715, using the Survey of Income and Program Participation rather than the Current Population Survey, and studying disabled men. He found that employment among men with disabilities was on average 7.2 percentage points lower over the post-statute period than before the act was passed, with the decline visible as early as 1990 and continuing through early 1995. He found no change in the wages of disabled men associated with the statute’s passage. The two papers together, one in the profession’s leading general-interest journal and one in its leading labor journal, using different surveys and arriving at the same directional result within a year of each other, constituted the state of knowledge at the turn of the century: the statute appeared to have reduced the employment of the people it was meant to help.

One timing qualification attaches to DeLeire’s result and must be stated plainly, because it limits what the paper can prove about causation. DeLeire’s post-statute period begins after January 1991, which is before the employment title took effect on July 26, 1992. Part of the employment decline his paper measures therefore predates the enforceability of the employment mandate. A decline that begins before a law takes effect cannot have been caused by the law’s enforcement in the period before it took effect, though it could reflect employer anticipation of the law, or it could reflect something else entirely. The article flags the timing because the distinction between a law’s passage, its effective date, and its enforcement bite is exactly the kind of detail that separates careful reading of this literature from careless citation of it. The Acemoglu and Angrist paper does not share this particular problem to the same degree, since its data run from 1988 through 1997 and its identification leans on the firm-size and charge-rate patterns, but the general caution applies: in this literature, when the clock starts determines what the estimate means.

The Acemoglu and Angrist research design deserves a detailed walk-through, because its strengths are what make the later qualifications meaningful rather than dismissive. The March Current Population Survey supplement is the federal government’s principal annual source on employment and disability, and the 1988 to 1997 window gave the authors four pre-statute years and five post-statute years. The firm-size pattern is the design’s sharpest feature. The statute exempted small employers, first those under twenty-five workers and then those under fifteen, so if the law caused the employment decline, the decline should be larger where the law applied and smaller where it did not. That is what the authors found: medium-size firms showed larger effects than the small firms the law left alone. A background trend, a recession, or a change in survey methods would not naturally respect the statute’s coverage thresholds, so the pattern points toward the law rather than toward the background. The charge-rate pattern works the same way: states with more disability-related discrimination charges, where the law bit harder, showed larger employment effects.

The hiring-versus-separations distinction is the paper’s most underappreciated finding. An employment decline can come from fewer hires or more fires, and the two have different implications for what the law did. If the statute raised firing costs, through the difficulty of terminating a worker who might sue, employers would retain reluctant hires and the decline would show up as reduced separations. If the statute raised the expected cost of employing a disabled worker, through accommodation costs and litigation risk, employers would become more cautious about hiring and the decline would show up as reduced hiring. The authors found the latter: the decline reflected reduced hiring, not increased separations. That finding weighs against the firing-costs story and for the accommodation-cost story, and it shaped everything written afterward, because the Jolls and Prescott qualification about one-time accommodation costs is a direct descendant of this distinction. The hiring margin matters for how the finding should be read. A decline driven by reduced hiring suggests a quieter mechanism: disabled job seekers facing longer searches and fewer offers, with the harm distributed across people the data never identifies as having been turned away. It is also the pattern most vulnerable to the confounds, because hiring is the margin most sensitive to the business cycle and to the availability of benefit income as an alternative to job search.

The disability-transfer test is the paper’s honesty about its own limits. If the employment decline simply reflected disabled workers moving onto benefit rolls, the statute’s employment provisions would be a sideshow and the benefits system the main event. The authors tested that hypothesis and found that the rise in disability transfers did not appear to explain the employment decline alone, leaving the statute as a likely cause. The word alone is doing the work in that sentence: the authors did not claim benefits were irrelevant, only that benefits could not account for the full decline. The Kruse and Schur confound papers would later argue the benefits story was bigger than Acemoglu and Angrist allowed, and the Zebley history below gives the benefits expansion its full due. But the 2001 paper’s treatment was careful on its own terms, and a reader who dismisses it as ideologically motivated has not read it.

DeLeire’s paper supplies the independent replication that makes the early finding hard to wave away. Different survey, the Survey of Income and Program Participation rather than the Current Population Survey; different population slice, disabled men; same directional result within a year of the other paper’s journal publication. The 7.2-percentage-point figure is the paper’s headline: employment among men with disabilities averaged that much lower over the post-statute period than before passage, with the decline visible as early as 1990 and continuing through early 1995. The wage finding is the dog that did not bark: the statute left the wages of disabled men unchanged, which means the adjustment, whatever its cause, happened on the employment margin rather than the wage margin. In a textbook labor market adjustment, an increase in the cost of employing a group would show up partly in lower wages for that group as employers passed the cost through. That wages did not move suggests the adjustment happened on the quantity margin rather than the price margin: fewer hires rather than lower pay.

The timing caveat on DeLeire’s paper needs its full statement, because it is the kind of detail that separates serious engagement from citation mining. The paper’s post-statute period begins after January 1991. The employment title took effect on July 26, 1992. Eighteen months of the measured decline therefore predate the law’s enforceability. Three interpretations compete for those eighteen months. The first is anticipation: employers who knew the law was coming adjusted their hiring before it took effect, which would mean the law caused the early decline through expectations rather than enforcement. The second is coincidence: the early-1990s recession, which began in 1990, hit disabled workers first and hardest, and the decline’s early start reflects the cycle rather than the statute. The third is passage effects: the law was enacted in 1990, and whatever signal enactment sent, separate from enforceability, could have moved behavior. The paper itself cannot distinguish among the three, and the article does not choose for it. The point of the caveat is narrower: a causal claim about a law’s enforcement cannot rest on a decline that began before enforcement existed, and any citation of DeLeire that omits the January 1991 start date is an incomplete citation.

The nondisabled finding in the Acemoglu and Angrist paper is a quiet but important part of the result. The authors found little evidence that the statute affected the employment of workers without disabilities, which is what the design predicts if the mechanism runs through accommodation costs and disability-specific liability: those costs attach to disabled workers specifically, so a statute-caused employment effect should appear in the disabled series and not in the nondisabled series. A finding of parallel declines in both series would have pointed to a common shock, a recession or a survey change, affecting everyone. The divergence between the series is therefore doing diagnostic work, the same kind of work as the firm-size and charge-rate patterns. It does not prove the statute caused the disabled-worker decline, because a shock correlated with disability status could be responsible, but it rules out the simplest alternative explanations, the ones that would move both series together.

The early findings also need their scope stated accurately. They concern employment levels among people who report disabilities in household surveys, not the statute’s other aims, and they say nothing about the built environment, telecommunications, or institutional care, where the evidence runs the other way. A critic who cites the Acemoglu and Angrist paper as proof that the statute failed wholesale is misusing it, because the paper is a paper about employment and only about employment. A supporter who dismisses the paper because its conclusion is unwelcome is misusing the research enterprise, because the paper was published in a leading journal, used standard methods, and survived the ordinary scrutiny of the field. The defensible position, and the one this article holds throughout, is that the early findings are real findings that must be confronted, and that confronting them means examining the later work that qualifies them rather than pretending they were never published.

Employment: the state-law qualification

The first major qualification of the early findings came from exploiting a feature of American federalism that the early papers had treated as background noise. The statute’s employment-discrimination provisions were an innovation in some states and not in others, because a number of states already had disability-discrimination laws resembling the federal statute’s before 1990. That variation creates a natural comparison: if the federal statute caused the employment decline, the decline should appear where the federal provisions were genuinely new and should be muted where state law had already imposed similar obligations. Christine Jolls and J.J. Prescott pursued exactly that comparison in “Disaggregating Employment Protection: The Case of Disability Discrimination,” circulated as National Bureau of Economic Research Working Paper 10740 in September 2004 and as Harvard Law and Economics Discussion Paper number 496 in February 2005. The working paper number deserves care, because some references miscite it, and the verified number is 10740. Jolls had laid the groundwork in a solo paper the same year, “Identifying the Effects of the Americans with Disabilities Act Using State-Law Variation: Preliminary Evidence on Educational Participation Effects,” published in the American Economic Review, volume 94, number 2, in May 2004, pages 447 to 453, in the Papers and Proceedings issue.

Their finding materially qualifies the early results without erasing them. Relative employment of disabled workers declined significantly just after enactment in the states where the federal provisions were a substantial innovation, but held stable in the states with pre-existing regimes resembling the statute. The innovation-related declines were concentrated in the initial years after enactment, which is to say they were largely transitory rather than permanent. And the mechanism traced to the reasonable-accommodation requirement, with its one-time costs of physical and procedural adjustment, rather than to the employment-protection rules governing termination, which would have implied a persistent drag on hiring. The National Bureau of Economic Research’s own digest of the paper, published in November 2004, summarized the bottom line in terms this article quotes directly because the source is authoritative and the sentence is careful: “Apart from a shortterm effect of the ADA’s requirement of special accommodations, the ADA was not causally linked to declining disabled employment over much of the 1990s.” That is a very different statement from the early papers’ suggestion of a persistent statute-caused decline, and it comes from the same kind of careful empirical work.

The research design behind the state-law qualification is worth understanding in detail, because it is the cleanest comparison the literature produced. American federalism meant the 1990 statute landed on fifty different legal landscapes. In some states, disability-discrimination law already resembled the federal provisions: employers faced similar accommodation duties and similar liability before 1990, so the federal law changed the forum and the enforcement machinery more than the substantive obligation. In other states, the federal provisions were a genuine innovation, imposing duties that had not existed before. Jolls and Prescott’s insight was to treat that variation as a natural experiment. If the federal statute caused the employment decline, the decline should appear in the innovation states and be muted in the already-regulated states. If the decline reflected a national trend, a recession, or a survey artifact, it should appear in both groups alike. The design isolates the statute’s contribution the way the firm-size pattern in the early paper isolated it, but at the state level rather than the employer level. By contrasting states where the federal statute was a genuine innovation with states where it duplicated existing law, the design separates the statute’s distinctive contribution from the national trends, the business cycle, and the benefits expansion, all of which affected both groups of states alike.

The transitory finding is the result that most changes the policy reading. The innovation-state declines were concentrated in the initial years after enactment and then faded. A permanent employment-protection effect, the kind where fear of termination lawsuits depresses hiring indefinitely, would not fade; it would persist as long as the law persisted. A one-time adjustment cost, the kind where firms pay to install accommodations, revise procedures, and learn the new rules, would spike and then decay as the adjustment completed. The data matched the second pattern. That match is what moved the debate from a story about the law’s permanent incentive effects to a story about its transition costs, and the policy implications of the two stories differ completely. Transition costs argue for patience and perhaps for adjustment assistance. Permanent incentive effects would argue for redesigning the law. The paper’s contribution was to show the evidence fit the first story better than the second.

The mechanism finding refines the story further. Within the employment provisions, two kinds of cost could depress hiring: the cost of reasonable accommodation, which is largely incurred once, when a workspace is modified or a procedure established; and the cost of employment protection, the expected cost of disputes over termination, which recurs with every hiring decision. Jolls and Prescott traced the immediate post-enactment effect to the accommodation requirement rather than the termination rules. Accommodation costs are largely one-time expenditures incurred when a workplace is brought into compliance: physical modifications, equipment purchases, revised procedures. An employer who has paid them has no continuing reason to avoid disabled applicants, which is why a decline driven by those costs should fade as compliance is completed. Employment protection, the expected cost of disputes over termination, is different in kind: it attaches to every hiring decision as a recurring risk premium, because any disabled hire is a potential future dispute. The early papers’ mechanism story leaned toward the recurring cost, which would depress hiring permanently. Jolls and Prescott’s evidence pointed to the one-time cost, which depresses hiring only until the adjustment is complete. The difference between a recurring tax on hiring and a one-time transition expense is the difference between a structural flaw and a startup cost, and the paper’s contribution was to move the empirical weight from the first interpretation to the second.

The state-law variation result does not refute the early findings so much as re-describe them. It says, in effect, that the early papers were measuring something real but misattributing its duration and its mechanism: a genuine short-run adjustment cost, concentrated where the law was genuinely new, mistaken for a permanent employment-protection effect. Refutation would mean the early papers made an error: bad data, a coding mistake, a misread table. Nothing in Jolls and Prescott suggests that. Re-description means the early papers measured something real and characterized it incompletely: a genuine short-run, accommodation-driven, innovation-concentrated decline, mistaken for a persistent employment-protection effect. The early estimates are not wrong within their design. They are incomplete without the state variation, the way a photograph is incomplete without its caption. Equal length for both sides is not a courtesy to the authors. It is the only way to show the reader how empirical knowledge actually advances: not by one paper demolishing another, but by each paper narrowing what the previous one could claim.

The solo Jolls paper from the same year extends the state-variation logic to a different margin. “Identifying the Effects of the Americans with Disabilities Act Using State-Law Variation: Preliminary Evidence on Educational Participation Effects,” published in the American Economic Review’s Papers and Proceedings issue of May 2004, applied the same innovation-versus-pre-existing-law comparison to educational participation rather than employment. The article notes the paper’s existence and its method, the state-law comparison, without claiming more for its results than the verification supports, because the verified findings in the record concern the employment estimates. The point of noting it is the research program it reveals: Jolls was not writing a single contrarian paper but developing a general method, using state-law variation as an identification strategy, and applying it across outcomes. A reader who encounters the working paper and the journal article together should understand them as two applications of one idea, not as two independent replications.

Employment: the measurement confounds

The second major challenge to the early findings attacks not the mechanism but the measurement itself, and it is the most technically demanding part of the literature. Douglas Kruse and Lisa Schur, in “Employment of People with Disabilities Following the ADA,” published in Industrial Relations, volume 42, number 1, in January 2003, pages 31 to 66, and in “Does the Definition Affect the Outcome? Employment of People with Disabilities Under Alternative Disability Definitions,” published as a chapter in the 2003 Upjohn Institute volume “Why the Decline in Employment of People with Disabilities: A Policy Puzzle,” edited by David Stapleton and Richard Burkhauser, pages 279 to 300, argued that the employment decline the early papers measured may be substantially an artifact of how disability is measured in surveys. Their argument has three strands, and each deserves to be stated separately because they are logically independent.

The first strand concerns who reports a disability. The standard survey measure is self-reported work limitation, and that measure is endogenous to the very law being studied. The statute may have reduced the stigma of identifying as disabled, leading more people to report disabilities in surveys, and the newly reporting group would disproportionately include people with milder limitations or weaker labor-force attachment, pulling the measured employment rate of the disabled population down even if no individual’s employment changed. Alternatively, improved workplace accessibility might have allowed some people to stop reporting work limitations they previously reported, which would have the opposite compositional effect. Either way, the population being measured changed at the same time as the law took effect, which means the before-and-after comparison is not comparing like with like. Kruse and Schur tested this directly by constructing fourteen alternative disability measures, and their result is the paper’s headline: with the standard measures they reproduced the post-statute decline found by the earlier studies, but with measures more closely tied to the statute’s own definition of disability, employment was found to increase. That single contrast, same data, different definition, different sign, is the strongest evidence in the literature that the definition does the work.

Could survey measurement alone explain the employment finding?

Possibly in large part. Kruse and Schur’s fourteen alternative disability measures show the estimated effect changing sign with the definition: standard self-reported work-limitation measures reproduce the decline, while measures closer to the statute’s own definition show employment increasing after enactment.

The endogeneity of self-reported disability is worth illustrating concretely, because it is the subtlest of the confounds. Suppose the statute succeeds at its cultural goal and reduces the stigma of identifying as disabled. More survey respondents then answer yes to the disability question, including some whose limitations are mild and whose employment prospects were already strong, and some whose limitations are severe and whose employment prospects are weak. The measured employment rate of the self-reported disabled group falls, not because anyone lost a job, but because the group’s composition changed. Alternatively, suppose improved workplace accessibility lets some workers manage their limitations without reporting them. They exit the measured group, and if they were employed at above-average rates, the measured employment rate of those who remain falls for the same compositional reason. In both scenarios the survey shows a decline and the statute caused it, but the decline reflects measurement rather than labor market harm. Kruse and Schur’s contribution was to show that this is not a theoretical worry: changing the disability definition changes the measured answer, including reversing its sign.

The second strand concerns the concurrent expansion of federal disability benefits. The early 1990s saw substantial growth in disability benefit receipt, and benefit receipt mechanically reduces measured employment, because beneficiaries face strong incentives not to work. The benefits story has its own documented history. In Sullivan v. Zebley, 493 U.S. 521, decided February 20, 1990, by a seven-to-two vote, the Supreme Court required the Social Security Administration to abandon its listings-only method for children’s Supplemental Security Income disability claims and provide an individualized functional assessment comparable to the adult standard. The agency issued revised childhood-disability rules on February 11, 1991, published at 56 Federal Register 5534, and finalized them after September 9, 1993. The aftermath was large: a 1994 General Accounting Office report, HEHS-94-225, titled “Social Security: Rapid Rise in Children on SSI Disability Rolls Follows New Regulations,” found that by March 1, 1994, the agency had notified roughly 452,500 class members of their readjudication rights, about 321,600 responded, and about 287,900 cases were readjudicated. The Congressional Record for 1995, volume 141, recorded the children’s SSI program growing from 1.2 billion dollars to 4.5 billion dollars, attributing the growth to the early-1990s recession, a 1989 agency outreach effort, a mental-impairment listings change, and the Zebley decision, then estimated to add roughly one million children, while noting the accounting office’s finding of no solid evidence that parents were coaching children to qualify. The history of Supplemental Security Income in this series supplies the program context. The article keeps the causal chain tight here, as the verification requires: the Zebley expansion is chiefly a children’s-benefits story, and its relevance to the adult employment estimates runs through the broader early-1990s growth in disability benefit receipt and through Kruse and Schur’s argument that benefit growth confounds the employment estimates. The benefits confound is cited to Kruse and Schur; the program-growth facts are cited to the accounting office and the Congressional Record. The timing sharpens the confound: the Zebley policy change, with revised rules in 1991 and final rules after September 1993, unfolded across exactly the years in which the early employment studies measured their post-statute periods, which is why economists treat the benefits expansion as a serious rival explanation rather than a footnote.

The third strand is the business cycle. Disabled workers are more vulnerable to economic downturns than nondisabled workers, and the early 1990s contained a recession. A decline in disabled employment measured across the statute’s enactment period therefore mixes the law’s effect with the cycle’s effect, and the early papers’ controls for the cycle may not have captured the differential impact fully. Kruse and Schur named differential business-cycle effects as one of their three confounds alongside the reporting-composition and benefit-growth mechanisms. Correcting for it requires modeling how disability interacts with the cycle, which the early studies did not attempt in full. The point is not that the cycle explains everything. It is that a portion of every measured change belongs to it, and no study reviewed here claims to have isolated that portion precisely.

Taken together, the confound papers do not prove the statute helped employment. They prove something narrower and, for honest assessment, more important: that the early finding of harm rests on measurement choices that are contestable, on a benefit expansion happening at the same time, and on a recession that hit disabled workers hardest. The decline the early papers found is concentrated in the years immediately following the statute even on the standard measures, and it reverses sign under alternative definitions. A reader who wants to cite the early papers as proof of harm must explain why the standard survey definition is the right one, why the benefit expansion does not confound the estimate, and why the state-law variation shows only transitory effects. Those are answerable questions, but they are questions, not settled matters, and the literature has not settled them.

The coaching non-finding in the Zebley debate is worth pausing over, because it illustrates the article’s standard for what counts as evidence. During the controversy over the children’s program’s growth, a charge circulated that parents were coaching children to appear disabled and qualify for benefits. The General Accounting Office investigated and found no solid evidence supporting the charge, and the 1995 Congressional Record debate recorded that finding. The episode matters for two reasons. First, it shows the verification working as it should: a widely repeated claim was checked against the auditing agency’s findings and did not survive, so the article reports the program’s growth without the coaching story. Second, it is a small-scale model of the whole employment debate. A striking claim about the disability system circulates, an authoritative source examines it, and the claim either survives with documentation or fails for lack of it. The article applies that standard to every number it reports, which is why the benefits confound is sourced to Kruse and Schur’s measurement argument and the program facts to the accounting office, and why no figure appears without its source attached.

Employment: what the literature honestly supports

With the three bodies of work laid out at equal length, the summary can be brief, because the structure has done the work. Respected economists found that employment among disabled workers fell after the statute took effect: Acemoglu and Angrist in the Journal of Political Economy in October 2001, DeLeire in the Journal of Human Resources in autumn 2000. Respected economists found the effect concentrated where the law was new and largely transitory, tracing to one-time accommodation costs rather than persistent employment protection: Jolls and Prescott in the 2004 working paper, Jolls solo in the American Economic Review in May 2004. Respected economists found the measured decline sensitive to the disability definition, confounded by benefit growth, and exposed to business-cycle effects: Kruse and Schur in Industrial Relations in January 2003 and in the 2003 Upjohn volume. No consensus exists. The article states that flatly, because anything else would be a misrepresentation, and it extends no finding to any current proposal, because the literature it surveys ends at the reference date and speaks to the statute as enacted and amended through 2008, not to anything proposed after.

The complication the brief names deserves its explicit paragraph, because both misuses of this literature are live in public argument. Critics cite the early findings as proof that the statute harmed its intended beneficiaries, as though a 2001 estimate using 1988 to 1997 survey data were a verdict on the law for all time. Supporters dismiss the early findings entirely, as though publication in leading journals, replication across two surveys, and survival of ordinary peer scrutiny counted for nothing. Both positions are indefensible on the record. The defensible position is the one this article has built toward across four sections: the identification problems are severe, the later work materially qualifies the early results without erasing them, and the question remains open. A reader who can state that sentence, with the authors and the years attached, has gotten everything this literature can honestly give.

The coverage change mid-period belongs in this section rather than in the measurement discussion, because it affects what the later estimates measure. The ADA Amendments Act of 2008 expanded the statute’s definition of disability, broadening the protected class for the years after its passage, and the coverage expansion is treated in the series’ guide to the ADA Amendments Act of 2008. The article claims no verified employment effect from the 2008 change, because no verified post-2008 employment study sits in the record, and the date-wall rule forbids manufacturing one. The amendment matters here only as a boundary marker: the employment literature surveyed above studies the statute largely as it operated before the coverage expansion, and estimates from that period do not automatically describe the statute as amended. The survey-based disability definitions the researchers used never mapped neatly onto the legal definition in any year, and the 2008 widening increased the distance between the measured group and the protected group.

What would it take to settle the question, and why has nothing settled it yet? The question is worth asking because it shows what the literature lacks. A settling study would need to handle all four difficulties simultaneously: a comparison group untouched by the statute, a disability definition tied to the law rather than to self-reported work limitation, a control for the concurrent benefit expansion that separates the two treatments, and a business-cycle adjustment specific to disabled workers. No published study in the at-reference record does all four. The early papers handled the comparison through firm size and charge rates but used the standard definition and the standard cycle controls. The state-law paper improved the comparison but kept the survey definitions. The measurement papers attacked the definition and named the confounds but did not supply the full alternative estimate. Each paper advanced one margin while leaving the others, which is why the literature reads as a sequence of partial corrections rather than a convergence. A reader who understands that structure understands why the article rates the layer contested: not because the researchers are confused, but because the problem is genuinely hard.

The identification problem at the heart of the employment literature deserves a general statement, because it explains why competent researchers reached incompatible conclusions from honest work. The problem is that the statute arrived in a world where many things were changing at once, and the researcher’s task is to isolate the statute’s contribution from everything else. The early 1990s brought a recession followed by a long expansion, and disabled workers experience the business cycle differently from nondisabled workers: they are often hired later in expansions and dismissed earlier in contractions, which means that a comparison spanning those years confounds the statute’s effect with the cycle’s. The same years brought the benefits expansion documented above, which changed the financial calculus of work for some people with disabilities independently of any employer obligation. Several states already had disability discrimination laws, which means the federal statute’s arrival was not a uniform treatment. And the surveys used to measure outcomes rely on self-reported disability, which means the measured group itself shifts in response to the policy being evaluated. A before-and-after comparison absorbs all of those movements into a single number and labels it the statute’s effect, which is why the early findings, though real as measurements, are fragile as causal claims.

The no-extension rule stated earlier bears repeating in its strongest form, because it is the article’s principal discipline against misuse. Nothing in the employment literature surveyed here speaks to any proposal made after the reference date. The estimates describe the statute as enacted in 1990 and amended in 2008, operating in the labor markets of the 1990s and early 2000s, measured with the surveys and definitions of that era. A proposal to change accommodation rules, benefit structures, or enforcement mechanisms would operate in a different economy, under a different coverage definition, with different measurement tools, and the old estimates would not transfer. The article states this boundary explicitly because the temptation to transfer is the most common misuse of empirical research in policy argument: a number estimated for one intervention in one period gets wielded as evidence about a different intervention in a different period, and the original authors would not recognize the use. The literature ends where it ends. The article ends it there.

Litigation: volume, geography, and concentration

The statute is enforced through three channels: private lawsuits, agency charges, and the Justice Department’s own compliance programs. The record for each channel is partial, and the article reports each with its source and period, noting the concentration patterns that the bare totals conceal. The recurring error in public discussion is to cite a filing count without noting that a small number of filers account for a large share of it, and the recurring error in the other direction is to treat the litigation record as though it described the statute’s effects rather than its enforcement. Lawsuits measure disputes, not outcomes, and the article keeps that distinction in view.

The employment-enforcement channel runs through the Equal Employment Opportunity Commission. In fiscal year 2014, the Commission reported 25,369 disability-basis charges, representing 28.6 percent of all charges, the fourth-largest basis after retaliation at 37,955, race at 31,073, and sex at 26,027. The figures come from the Commission’s fiscal year 2014 enforcement and litigation data release, issued under Chair Jenny R. Yang, and the release notes that a single charge can allege multiple bases, so the categories overlap. The article does not disaggregate the disability figure into statute-specific components, because the Commission’s disability basis covers both the 1990 statute and Section 501 of the Rehabilitation Act, and no verified disaggregation exists. The charge count measures complaints filed with the agency, not lawsuits, not findings of violation, and not the employment effects discussed in the preceding sections, and it should not be read as any of those things.

The charge process deserves a brief explanation, because the 25,369 figure is often misread. A charge is a complaint filed with the Commission alleging discrimination, and the Commission’s published basis categories describe what the complaint alleged, not what the Commission found. The release notes that a single charge can allege multiple bases, which means the 25,369 disability charges, the 37,955 retaliation charges, the 31,073 race charges, and the 26,027 sex charges overlap: one worker’s complaint can appear in several categories. The disability basis covers both the 1990 statute and the Rehabilitation Act’s federal-employment provisions, and the article does not disaggregate the two, because no verified disaggregation exists and inventing one would violate the article’s own rules. The charge count is best read as a measure of complaint volume, a signal of where workers perceive problems, rather than as a measure of violations found or of the statute’s effects.

The public-accommodations litigation channel is where the volume sits. Seyfarth Shaw, a law firm whose ADA Title III practice has tracked federal filings from the PACER system since 2013, reported 2,722 Title III filings in 2013 and 4,436 in 2014, an increase of 63 percent in a single year. California led with 1,866 filings in 2014, followed by Florida with 1,553 and New York with 212. The article uses only the 2013 and 2014 figures, because the 2015 full-year total postdates the reference date and is date-walled out, and because the firm’s tracking began in 2013, so no comparable series exists for earlier years from this source. The employment-title federal lawsuits, by the same firm’s comparison, held steady across 2012 to 2014 at well under half the Title III volume. That comparison matters because public discussion routinely conflates the two titles: the litigation surge that people argue about is overwhelmingly a public-accommodations phenomenon, driven by the title that covers restaurants, stores, hotels, and theaters, not the title that covers hiring and firing.

Why do a few plaintiffs file so many of the suits?

Because the economics reward it in some states: an NBC Bay Area investigation aired in March 2014 found that just 30 plaintiffs filed over half of California’s ADA suits since 2005, in a state whose own law allows up to 4,000 dollars in statutory damages per violation.

The concentration finding comes from investigative reporting rather than an academic study, and the article attributes it accordingly. NBC Bay Area aired an investigation in March 2014, reported on Lexology on March 5, 2014, by reporters Vicky Nguyen, Jeremy Carroll, and Kevin Nious, examining more than 10,000 disability-access lawsuits filed since 2005 in the five states with the largest disabled populations. California accounted for 7,188 of those suits, more than the other four states combined, and more than half of the California suits were filed by just 30 plaintiffs. The same reporting supplied the economic context: California’s own disability-access law allows statutory damages of up to 4,000 dollars per violation, creating a damages incentive that the federal statute’s public-accommodations title, which provides no money damages to private plaintiffs, does not create. The article reports the concentration factually and notes the incentive structure factually, without characterizing the filers’ motives, because the record documents the pattern and the incentive, not the intent. A later industry analysis of filing concentration, covering 2009 through 2023, is methodologically stronger but postdates the reference date and is date-walled out; the article cites only the 2014 figures.

The Justice Department’s own enforcement channel is documented through its Project Civic Access program, a Title II initiative that negotiates settlement agreements with cities and counties to bring local-government facilities, polling places, courthouses, and similar sites into compliance. The department reported 218 Project Civic Access settlement agreements as of July 2015, with 15 new agreements that year, in its “Twenty-Five Years of Progress” anniversary publication of July 2015. The program began with the City of Toledo settlement of August 23, 1999. The article reports only that figure for department enforcement, because no verified aggregate count of the department’s other disability enforcement actions was found, and the verification record explicitly excludes invented totals. Project Civic Access is a compliance program for local governments, not private-plaintiff litigation, and the article does not merge the two.

The three enforcement channels need to be kept conceptually separate, because they measure different things and the public argument routinely blends them. Private lawsuits under the public-accommodations title measure disputes between individuals and businesses over physical and policy barriers: a restaurant without an accessible entrance, a hotel that refuses a service animal, a theater with no accessible seating. Agency charges under the employment title measure complaints filed with the Equal Employment Opportunity Commission, most of which never become lawsuits and many of which the agency closes without a finding. Department compliance agreements measure negotiated settlements with local governments, which produce accessibility improvements without any plaintiff at all. A single filing count that merges the three would be meaningless, and the article reports each channel with its own source, its own period, and its own unit of measurement.

The Title I versus Title III divergence is the litigation section’s most important structural fact. Employment-title federal lawsuits held steady from 2012 to 2014 at well under half the volume of public-accommodations filings, by the Seyfarth comparison. That divergence tells a story about where the statute generates disputes. The employment title’s procedures route complaints through the Commission first, which filters and sometimes resolves them before litigation. The public-accommodations title allows private plaintiffs to file directly in federal court, which lowers the procedural barrier to bringing cases relative to the employment title’s agency-first procedure. The two titles also differ in what compliance requires: an employer’s accommodation duty is individualized and fact-specific, which makes cases expensive to bring, while a missing curb ramp or an inaccessible restroom is observable and repeatable across defendants, which makes cases cheap to replicate. The filing counts reflect those structural differences, not just the underlying rate of violations.

The damages incentive explains the geography. The federal public-accommodations title provides injunctive relief and attorney’s fees but no money damages to private plaintiffs, which means a federal-only case offers a plaintiff little direct financial recovery. California’s parallel state law changes the calculus by allowing statutory damages of up to 4,000 dollars per violation, which turns each noncompliant facility into a potential damages claim and makes serial filing economically rational. Florida’s volume has its own drivers in the state’s large hospitality and retail sectors and its own state-law framework. The article does not adjudicate which state’s regime is better policy. It reports the incentive structure that the 2014 investigation documented and lets the reader see why the national filing map looks the way it does. A filing count without the incentive context invites the inference that California businesses are uniquely noncompliant. A filing count with the incentive context invites the more careful inference that California’s legal regime rewards filing.

Project Civic Access illustrates the department’s distinct enforcement philosophy. Rather than waiting for private plaintiffs, the department selected cities and counties and negotiated comprehensive settlement agreements covering the full range of local-government facilities and programs: city halls, courthouses, polling places, parks, libraries, and emergency services. The 218 agreements as of July 2015 represent 218 local governments that surveyed their facilities, identified barriers, and committed to remediation schedules under federal oversight. The program began with Toledo in August 1999 and grew steadily, with 15 new agreements in the anniversary year. The compliance-program model differs from litigation in its unit of change: a private lawsuit typically fixes one defendant’s one facility, while a Civic Access agreement commits an entire municipal government to a multi-year remediation plan. The article reports the 218 figure as the verified department-enforcement number and states plainly that no verified aggregate exists for the department’s other disability enforcement work, because the gap in the record is itself information.

The disputes-versus-outcomes distinction needs one more explicit statement, because it governs how every number in this section should be read. A filing count measures how often people went to court or to the agency. It does not measure how often violations occurred, because most violations never produce a filing and some filings allege violations that did not occur. It does not measure compliance rates, because the denominator, the number of covered facilities or employment decisions, is not in the filing data. And it does not measure the statute’s effects on the world, because litigation is the enforcement channel, not the outcome channel: the ramps, the relay system, and the spending shift are the outcomes, and they are documented elsewhere in this article. The filing counts belong in an impact article for a different reason. They show where the statute generated friction, which titles produced disputes, how concentrated the filers were, and what incentives shaped the geography. Read that way, the 4,436 Title III filings of 2014, the 25,369 agency charges, and the 218 compliance agreements are not a scorecard. They are a map of the enforcement terrain, and the article presents them as a map.

What the measurement frame leaves out

An evidence article should state what it could not verify as plainly as what it verified. Seven categories of figures were excluded during the verification behind this article, and each exclusion was deliberate rather than an oversight. No citable national count of intersections with curb cuts was found, so this article describes the installation mechanism rather than inventing a coverage figure. No verified figure states the precise share of post-1990 buildings meeting the accessibility standards, so the article describes the structural change without a percentage it cannot source. No aggregate count of all Department of Justice ADA enforcement actions was verified beyond the 218 Project Civic Access agreements, so the article does not extrapolate from one program to the whole. The EEOC’s disability charge figures cannot be disaggregated into ADA and Rehabilitation Act components, so the article reports the combined figure with the caution attached. No study quantifying the spillover beneficiaries of curb ramps was verified, so the spillover appears as reported anecdote rather than as a number. The full-year 2015 Title III filing total postdates the reference date and is excluded on dating grounds, as are the later administrative opinions on the integration mandate. The exclusions share a principle: a number without a citable source is not evidence, and an evidence article that invents numbers where the record is silent forfeits the trust its verified numbers earned.

The exclusions also discipline how the reader should receive claims encountered elsewhere. Any account of the statute’s impact that states a national curb-cut count, a precise accessible-building share, a total Department enforcement figure, or a quantified spillover population is either citing a source this article’s verification did not find or asserting what it cannot source, and the reader knows which questions to ask of it. The same discipline applies to the employment literature: a citation of DeLeire’s 7.2 percentage point figure without its January 1991 starting line, or of the Acemoglu and Angrist finding without the later qualifications, is a partial reporting of the record. The article’s transparency about its own limits is what licenses its confidence about its verified claims, and that exchange, limits stated plainly for trust earned honestly, is the compact this article offers its reader.

The four-layer evidence table

The table below compresses the article’s verdict into the findable artifact the brief requires. Each row gives the domain, what changed, the period over which the change is documented, the principal sources, and the rating. The ratings use the three words defined in the framework section: settled, where the direction of change is not seriously disputed and multiple sources document it; measurable, where the change is documented with numbers but attribution or completeness requires qualification; contested, where respected researchers disagree about the direction of the effect itself.

Domain	What changed	Period	Principal sources	Settled, measurable, or contested
Built environment	Curb ramps spread via alteration and resurfacing triggers; transit fleets converted toward near-universal vehicle accessibility; new construction and altered facilities built to accessibility standards as a continuous accumulation	1969 to 2015 for ramps; 1990 to 2015 for construction standards; 1994 to 2015 for transit fleets	1991 ADA Accessibility Guidelines and 2010 ADA Standards, 28 CFR Part 36; DOJ and DOT Joint Technical Assistance 2013; APTA 25th-anniversary release July 26, 2015; FTA announcement July 23, 2015; Berkeley Resolution 45,605-N.S. 1973; Steinfeld 2007; Williamson 2014	Settled
Telecommunications	Statutory relay obligation created a uniform national telecommunications relay infrastructure; extended to 711 abbreviated dialing, video relay, and internet-protocol relay by dated agency orders	1991 to 2002 for implementation and extensions; obligation effective July 26, 1993	ADA Title IV, 47 U.S.C. section 225; FCC orders of July 26, 1991, March 6, 2000, and April 22, 2002; FCC Second Report and Order 2000; Federal Register September 2000	Settled
Institutional care	Long-term services shifted from institutions toward home and community settings; community-based share of Medicaid long-term spending rose from 18 percent to 51 percent, crossing the majority line	1995 to 2013 for spending data; 1999 for the integration decision	Olmstead v. L.C., 527 U.S. 581 (1999); 28 C.F.R. section 35.130(d); Kaiser Family Foundation and UCSF 2013 Data Update; CMS 2013 expenditure report July 2015; CRS R43804 (2014); Health Affairs 2015	Measurable
Employment	Early studies found post-statute employment declines; later work found effects concentrated where the law was new and transitory, and estimates sensitive to disability definition, benefit growth, and the business cycle	1988 to 2004 for the core published estimates	Acemoglu and Angrist, Journal of Political Economy, October 2001; DeLeire, Journal of Human Resources, Autumn 2000; Jolls and Prescott, NBER Working Paper 10740, September 2004; Jolls, American Economic Review, May 2004; Kruse and Schur, Industrial Relations, January 2003, and Upjohn volume 2003	Contested

Assessing a statute against its aims

The series thesis thread for this article is the method it has been practicing throughout: assess a statute against its aims, and name precisely which of those aims the evidence can and cannot speak to. The 1990 statute promised equality of opportunity, full participation, independent living, and economic self-sufficiency. The record through September 2015 speaks clearly to three of the four. Full participation in the built environment advanced through ramps, vehicles, and construction standards, documented in dated regulations and fleet reports. Full participation in communications advanced through the relay infrastructure, documented in the statute’s text and the implementing orders. Independent living advanced through the shift toward community-based services, documented in the spending data. Economic self-sufficiency, the aim most dependent on labor markets rather than on physical or institutional barriers, is the one the evidence cannot resolve, because the labor market entangles the statute’s effect with benefit policy, business cycles, survey definitions, and the staggered novelty of the law across states.

That pattern carries a general lesson about what statutes can be shown to have done. Laws that command physical or institutional changes leave the kind of evidence that accumulates: ramps get built, fleets get converted, spending shares move. Laws that operate through incentives in complex systems leave evidence that has to be extracted by statistical argument, and the extraction is only as good as the comparison the researcher can construct. The employment literature on this statute is a case study in the difference. The early papers constructed one comparison and found harm. The state-law paper constructed a better comparison and found transitory adjustment. The measurement papers questioned whether the comparison was measuring the right population at all. None of those moves is dishonest, and the sequence is how empirical knowledge is supposed to work: claim, qualification, deeper qualification. The failure mode is not in the research but in the citation of it, when a single paper is lifted out of the sequence and presented as the verdict.

The citation failure mode is the practical enemy of that skepticism, and the article has been written as an inoculation against it. The failure mode works like this: a striking single-study result gets lifted out of its literature, stripped of its qualifications, and deployed as a verdict in an argument the study never addressed. The early employment papers have suffered this treatment from critics of the statute, and the qualifying papers have suffered the mirror-image treatment from its defenders, who cite the transitory finding as though it erased the short-run decline rather than explaining it. The inoculation is the article’s standing rule, repeated in every section: no empirical claim without its authors, its publication, and its period, and no claim extended beyond the comparison its authors constructed. A reader who internalizes that rule will disagree about policy on honest grounds. The disagreements will be about what to do given the evidence, not about what the evidence says, and that is the only ground on which the argument can honestly proceed.

The article’s own verdict, then, is the namable claim stated at the outset and earned across every section: access is settled, employment is not. Three decades of evidence establish large change in physical access, in communications, and in the placement of long-term services, and leave the statute’s employment effect genuinely unresolved, which means confident claims in either direction about jobs are claims the literature does not support. A reader who can state that sentence, attach the four layers to their sources and periods, and explain why the employment question resists an answer, has everything this article was built to give. Researchers assembling a cross-statute evidence file can keep the layer ratings alongside other statutes in the legislation study notebook, where the settled, measurable, and contested labels transfer directly.

The practical value of the namable claim is that it functions as a filter for everything the reader will encounter afterward. Any confident assertion that the statute raised disabled employment, or that it lowered it, can be checked against the literature’s actual state: early findings of decline, material later qualifications, documented confounds, no consensus. A claim that survives that filter is worth engaging. A claim that does not is advocacy wearing the clothes of research, and the reader who has worked through the four layers will know the difference. That is the durable benefit of an evidence article that refuses to manufacture certainty. It does not tell the reader what to think about the statute and jobs. It tells the reader how to think about anyone who claims to know.

The statutory aim of economic self-sufficiency deserves a final reflection, because it is the aim the evidence cannot resolve and therefore the aim that tests the article’s method most severely. The statute promised disabled Americans not just access to buildings and telephones but a foothold in economic life, and the employment literature is the record of whether that promise was kept. The record’s answer is that no one knows, and the article’s answer is to say so at length rather than to fill the gap with a preferred story. There is a temptation, when evidence is inconclusive, to treat the absence of proof as proof of absence, or to treat a contested literature as a license to pick the congenial half. The article resists both. It holds the early findings and the qualifying findings in the same frame, gives each its full weight, and leaves the question where the literature left it: open. That restraint is not agnosticism for its own sake. It is the precise application of the series thesis, naming which aims the evidence can speak to and which it cannot, and refusing to let the desire for a tidy verdict override what the research actually supports.

Frequently Asked Questions

Q: Did the Americans with Disabilities Act increase employment for disabled people?

The honest answer is that no one knows, because the research does not agree. Daron Acemoglu and Joshua Angrist, in the Journal of Political Economy in October 2001, and Thomas DeLeire, in the Journal of Human Resources in autumn 2000, found employment declines among disabled workers after the statute took effect, using Current Population Survey data from 1988 to 1997 and Survey of Income and Program Participation data respectively. Later work qualified those findings substantially: Christine Jolls and J.J. Prescott, in a 2004 working paper, found the declines concentrated in states where the law was new and largely transitory, while Douglas Kruse and Lisa Schur, in 2003 publications, showed the estimated effect changing sign with the disability definition used. With respected findings on both sides and severe identification problems acknowledged by all sides, the only defensible statement is that the employment effect is unresolved.

Q: Did the Americans with Disabilities Act reduce employment instead?

Some respected studies say it did, and other respected studies say the finding does not hold up, so neither answer can be stated as fact. The reduction claim rests on Acemoglu and Angrist’s 2001 paper and DeLeire’s 2000 paper, both of which found lower employment among disabled workers after the statute’s arrival. The counter-evidence has three parts: Jolls and Prescott’s state-law comparison found the decline only where the federal provisions were an innovation and fading within a few years; Kruse and Schur showed the result depends on which survey definition of disability the researcher uses; and the early-1990s expansion of disability benefits plus a recession confound the estimates. The article gives both sides equal length because the literature itself is balanced, and it reports the identification problems explicitly rather than smoothing them over.

Q: What is the curb cut effect from the Americans with Disabilities Act?

The phrase in this question did not exist during the period this article covers. Civil rights advocate Angela Glover Blackwell coined it in a Stanford Social Innovation Review article in the winter of 2017, after the September 2015 reference date, so this article never uses it as a term of art. The documented phenomenon it names is older and well sourced. In Berkeley, California, wheelchair activists including Ed Roberts and Hale Zukas won cement curb ramps at Telegraph Avenue intersections in 1969 and 1970, and the city’s Resolution 45,605-N.S. of February 13, 1973, mapped a 125-cut wheelchair route. Multiple pre-2015 sources, including Bess Williamson’s 2014 account, record that the ramps were soon heavily used by parents with strollers, delivery workers, travelers with luggage, and cyclists. The 1990 statute then scaled the pattern nationally through alteration rules and the 2013 confirmation that street resurfacing triggers ramp installation.

Q: How did the Americans with Disabilities Act change buildings?

It made accessibility a routine requirement of construction rather than an optional feature. The 1991 ADA Accessibility Guidelines, later superseded by the 2010 ADA Standards for Accessible Design at 28 CFR Part 36, required new construction and alterations after 1990 to be accessible. The alteration rules reached backward into the existing stock: when a facility was substantially renovated, the altered portions and often the path of travel to them had to meet the standards. Over three decades, every new building and every qualifying renovation added accessible square footage by operation of law. The article states no precise share of accessible buildings because no verified figure exists. Part of the credit belongs to the earlier funding-condition statute, Section 504 of the Rehabilitation Act of 1973, which had pressed federally assisted facilities toward accessibility since the 1970s.

Q: Did the Americans with Disabilities Act end institutionalization?

No, and the article does not claim that it did. What happened was a sustained shift, not an abolition. The Supreme Court’s 1999 decision in Olmstead v. L.C., 527 U.S. 581, held that unjustified segregation violates the statute’s public-services title and required community-based services where appropriate, unopposed, and reasonably accommodated. Litigation enforcing that mandate, together with state Medicaid policy, moved the financing: home and community-based services rose from 18 percent of Medicaid long-term services and supports spending in 1995 to 51 percent in 2013, the first year community services were the majority, with total long-term spending at 146 billion dollars. Institutions continued to operate, the majority was a spending majority, and state policy shared the credit, which is why this layer earns a measurable rating rather than a settled one.

Q: How many Americans with Disabilities Act lawsuits are filed each year?

The answer depends on which channel is counted, and the channels differ enormously. For public-accommodations suits in federal court, Seyfarth Shaw’s PACER tracking reported 2,722 filings in 2013 and 4,436 in 2014, a 63 percent single-year increase, with California at 1,866 and Florida at 1,553 in 2014. Employment-title federal suits held steady from 2012 to 2014 at well under half the Title III volume. At the Equal Employment Opportunity Commission, fiscal year 2014 saw 25,369 disability-basis charges, 28.6 percent of all charges. The filing counts come with a concentration warning: a March 2014 NBC Bay Area investigation found that just 30 plaintiffs filed over half of California’s disability-access suits since 2005. The 2015 full-year filing total postdates the reference date and is excluded here.

Q: What do economists say about the Americans with Disabilities Act?

They agree on the physical and institutional changes and disagree on the jobs question, and the disagreement is the story. On employment, Acemoglu and Angrist (Journal of Political Economy, October 2001) and DeLeire (Journal of Human Resources, autumn 2000) found post-statute declines; Jolls and Prescott (NBER Working Paper 10740, September 2004) and Jolls solo (American Economic Review, May 2004) found the declines concentrated where the law was new and transitory; Kruse and Schur (Industrial Relations, January 2003, and a 2003 Upjohn volume chapter) showed the estimates depend on the disability definition and are confounded by benefit growth and the business cycle. No consensus exists. On the built environment, telecommunications, and institutional care, the economic and policy literature documents large, dated changes without serious dispute. The article’s rule is that every empirical claim carries its authors, publication, and period.

Q: Did the Americans with Disabilities Act change telephone access?

Yes, fundamentally. Title IV added Section 225 to the Communications Act of 1934, directing the Federal Communications Commission to ensure telecommunications relay services for people with hearing or speech disabilities, with common carriers required to provide relay throughout their service areas beginning July 26, 1993. The Commission had ordered nationwide relay a year earlier, on July 26, 1991, and uniform national availability arrived in July 1993. The system then absorbed new technology by dated orders: the 711 abbreviated dialing code was required by October 1, 2001; video relay service was recognized as relay in the Improved TRS Order of March 6, 2000; and internet-protocol relay was recognized in the declaratory ruling of April 22, 2002. Before the statute, relay was a state-by-state patchwork; after it, a national infrastructure existed where none had before.

Q: When did the employment provisions of the Americans with Disabilities Act take effect?

The employment title took effect on July 26, 1992, for employers with twenty-five or more employees, and coverage expanded on July 26, 1994, to employers with fifteen or more employees, according to the Equal Employment Opportunity Commission’s official questions-and-answers publication. The Commission estimated the 1994 expansion brought roughly 666,000 businesses and 86 million workers under coverage, up from 264,000 businesses. The public-services title’s employment provisions for state and local governments took effect earlier, on January 26, 1992. The dates matter because every employment study in the literature defines a before and an after around them: DeLeire’s post-period beginning in January 1991, for example, starts before the title was enforceable, a timing detail that limits the causal reading of his estimates.

Q: How did the 711 dialing code change relay access?

It collapsed the access number to three digits. Before the Federal Communications Commission’s Second Report and Order of 2000, reaching relay service meant dialing a full telephone number that varied by provider and state. The order required carriers to support 711 as the nationwide abbreviated dialing code for telecommunications relay access by October 1, 2001, so a deaf or speech-disabled caller could reach a relay operator the way any caller reaches directory assistance. The change belongs to the statute’s second decade: the relay obligation itself dates to Title IV and took effect in July 1993, but the usability of the system kept improving as the Commission extended the framework. Video relay followed in the March 2000 order and internet-protocol relay in the April 2002 ruling, each folding a new medium into the same statutory structure.

Q: What legal mechanism kept adding curb ramps after 1990?

The alteration trigger. The Title II regulations required curb ramps wherever new roads and sidewalks were constructed or existing ones were altered, and the 2013 joint technical assistance from the Department of Justice and the Department of Transportation confirmed that street resurfacing counts as an alteration. That confirmation turned every repaving cycle into a ramp project: at each resurfaced intersection, ramps had to be installed. Because cities resurface streets continuously, the regulation produced a steady accumulation of ramps year after year without any separate appropriation or program. The mechanism explains why ramp coverage grew as a function of the construction calendar rather than arriving all at once. No verified national count of compliant intersections exists, so the article documents the trigger and its continuous operation instead of inventing a total.

Q: Why do researchers disagree about who counts as disabled in survey data?

The disagreement is methodological, and Douglas Kruse and Lisa Schur made it central to the employment debate. The early studies typically defined disability through self-reported work limitations in surveys such as the Current Population Survey, but self-reports are endogenous to the policy environment: the statute may reduce stigma and increase reporting, which mechanically lowers the measured employment rate of the reporting group, or better workplace access may let some workers stop reporting limitations, changing the group’s composition. Kruse and Schur tested fourteen alternative disability measures and found that the post-statute employment decline appeared with standard measures but was concentrated in the years right after passage, while measures closer to the statute’s own protected class showed employment increasing. The lesson is that the measured answer depends on the definition, and no single survey definition is neutral.

Q: How did Medicaid spending shift toward home and community services?

Home and community-based services rose from 18 percent of Medicaid long-term services and supports spending in 1995 to 51 percent in 2013, crossing into the majority for the first time, with total long-term spending at 146 billion dollars in 2013. The figures come from the Kaiser Family Foundation and UCSF 2013 data update and the Centers for Medicare and Medicaid Services 2013 expenditure report of July 2015, with a 2014 Congressional Research Service report describing the share as more than doubling from 1995 to about half in 2012. The shift followed the Supreme Court’s 1999 Olmstead integration decision and the litigation and waiver policy it set in motion. The article attributes the direction and magnitude to the record while noting that state budget choices and demographics shared the credit, which is why the institutional-care layer is rated measurable rather than settled.

Q: What are Project Civic Access agreements?

They are settlement agreements between the Department of Justice and cities or counties to bring local-government facilities into compliance with the statute’s public-services title, covering city halls, polling places, courthouses, parks, and similar sites. The department reported 218 such agreements as of July 2015, with 15 new that year, in its 25th-anniversary publication, and the program began with the City of Toledo settlement of August 23, 1999. Project Civic Access is a compliance program negotiated with local governments, not private-plaintiff litigation, and the article keeps the two separate. The 218 figure is the only verified aggregate department-enforcement number in the record; the verification found no citable total for the department’s other disability enforcement actions, so none is stated.

Q: Why do California and Florida dominate ADA Title III filings?

In 2014, California produced 1,866 federal Title III filings and Florida 1,553, together accounting for the great majority of the 4,436 national total tracked by Seyfarth Shaw, with New York a distant third at 212. The concentration has a documented economic explanation for California: the state’s own disability-access law allows statutory damages of up to 4,000 dollars per violation, an incentive the federal Title III, which provides no money damages to private plaintiffs, does not create. A March 2014 NBC Bay Area investigation added the filer-concentration dimension, finding that just 30 plaintiffs filed over half of California’s disability-access suits since 2005. The article reports the geography, the damages incentive, and the concentration factually, without characterizing motives, and it notes that the filing surge is a Title III phenomenon, not an employment-title one.

Q: How did disability benefits growth affect the employment findings?

The growth of federal disability benefits in the early 1990s is one of the documented confounds that make the employment studies hard to interpret. The Supreme Court’s 1990 decision in Sullivan v. Zebley required individualized assessments for children’s Supplemental Security Income claims, and the General Accounting Office reported in 1994 that the children’s SSI program grew from 1.2 billion to 4.5 billion dollars as rolls expanded. Broader disability benefit receipt also grew in the period, giving some adults with disabilities income support that competed with employment. Kruse and Schur named this benefits growth as a confound because the early employment studies could not fully separate it from the statute’s effect: when two policies change simultaneously, a before-and-after comparison cannot reliably assign the employment change to one of them.

Q: What happened to bus and rail accessibility between 1994 and 2015?

The fleets converted almost completely. The American Public Transportation Association’s July 26, 2015 anniversary release and the Federal Transit Administration’s July 23, 2015 announcement reported the accessible share of the national bus fleet rising from 51 percent in 1994 to 99.8 percent in 2015, through lifts and ramps on the vehicles. Heavy rail and subway vehicles went from 83 percent to 100 percent accessible, and the light rail and streetcar fleet from 41 percent to 88 percent, over the same span. An earlier segment from the association’s 2009 fact book shows buses moving from 60 percent in 1995 to 98 percent in 2008. The figures describe vehicles rather than stations, a distinction the article keeps explicit. The trajectory is among the clearest measured accessibility gains in the record, and it is dated, sourced, and large.

Q: Why does part of the accessibility credit belong to an earlier statute?

Because Section 504 of the Rehabilitation Act of 1973 conditioned federal funding on accessibility more than fifteen years before the 1990 law, and facilities built or renovated under that funding condition were already moving toward accessibility. The 1990 statute’s distinctive contribution was extending the obligation beyond federally funded programs to private businesses and to state and local government as such, backed by a private right of action, plus the design standards, alteration rules, and transit requirements that drove the post-1990 accumulation. The article assigns the widening to the 1990 law and the earlier movement to the funding-condition regime, without pretending the line can be drawn facility by facility. Crediting the whole change to one statute would misdescribe a movement that began in Berkeley streets and federal grant conditions before the 1990 law existed.

Q: Did the Americans with Disabilities Act cover video and internet-based relay calls?

The original 1990 text addressed the telephone network, but the regulatory framework proved adaptable. The Federal Communications Commission recognized video relay service, which connects sign language users through video interpreters, as a form of telecommunications relay service in its Improved TRS Order of March 6, 2000. Internet Protocol relay, which routes relay calls over internet connections, was recognized in the Commission’s declaratory ruling of April 22, 2002. The Commission had also required 711 as the nationwide abbreviated dialing code for relay access, with compliance by October 1, 2001. Each of these actions predates this article’s reference date, and each extended the Title IV obligation into a new medium without new legislation. The telecommunications layer of the statute’s impact therefore includes both the original nationwide relay system and its migration to video and internet technologies.

Q: Why did medium-size employers show the largest employment effects in the early studies?

Acemoglu and Angrist reported that the employment decline they measured was larger in medium-size firms, and the pattern fit their proposed mechanism. Small firms were exempt from the employment title, so the statute’s obligations fell on firms above the size threshold, with medium-size employers facing accommodation and potential litigation costs that were large relative to their resources. The authors also found that the decline reflected reduced hiring rather than increased separations, which weighs against the interpretation that employers responded by firing disabled workers and in favor of the interpretation that they became more cautious about hiring. Larger effects in states with more discrimination charges pointed the same way, toward perceived legal risk shaping hiring decisions. Later work questioned whether these patterns reflected the statute or the confounds, but the firm-size pattern was part of the original finding.