An economic-history study of Rhine tolling counts 12 toll stations between Mainz and Cologne in 1250. Then imperial authority broke down during the Great Interregnum. The number of Rhine toll stations at least doubled within four years, and many of the new toll collectors lacked legal authority. A coalition led by cities and merchant interests formed the Rhine League in 1254. The league forced several lords to abandon unjust tolls at roughly ten sites and destroyed at least one toll castle.

The tolls did not impoverish every place that collected them. They financed rulers and fortifications, and some nearby towns gained business. But each toll holder could benefit while imposing part of the cost on every merchant and customer farther along the river. The burden accumulated across the route. Centuries later, the problem remained: a Prussian report counted 29 toll stations above the Dutch border in 1801, and traders sometimes used roads running parallel to the Rhine simply to avoid the charges. Historical estimates suggest that the tolls may have reduced river traffic substantially. Robert Mark Spaulding documents this later history.

The eventual response was a common regime for the whole river. Reformers first consolidated the patchwork of charges, then established the Central Commission for the Navigation of the Rhine in 1815. The Mannheim Act of 1868 barred charges imposed merely for navigating the Rhine and supported common rules for using the river. The Rhine remained governed, but it increasingly operated as one transportation network. The commission’s history tells that institutional story.

Surprisingly, this history offers a useful way to think about artificial intelligence. Most conversations about AI, at least in the present political environment, begin inside a data center. We talk about chips, models, electricity, and the enormous amounts of money being invested in all three. Yet an AI model needs data, and those data have to get somewhere — in the physical sense of traveling from one place to another.

The data may be generated on a phone, in a factory, at a hospital, or through millions of customer interactions. They may be stored in one place, cleaned in another, processed in a third, and used later to improve a model. Before any of that can happen, someone has to determine which data may be used, where they may go, how they must be protected, and what records must be kept.

In other words, AI has a transportation sector.

In a recent working paper with Bentley Coffey and Pietro Peretto, I use this idea to examine a question that deserves more attention: What happens when innovation in computing moves faster than innovation in the technical and legal systems that deliver data to it?

The paper’s most interesting implication is a conditional case for federal preemption. When state rules function like border checkpoints in a national data network, they can slow today’s AI services and the learning that produces tomorrow’s.

The factory gate is not the finish line

This paper grew out of another project in which Bentley, Pietro, and I study transportation, innovation, and economic growth.

The basic point of that paper is intuitive. Imagine a factory that becomes more efficient every year. Its workers produce more with less, and the cost of making each item keeps falling. But the cost of shipping each item to the customer never improves.

Eventually, transportation becomes the largest part of the delivered price. Further gains inside the factory matter less because the product still has to leave the factory. Sustained growth requires innovation in making things and innovation in moving them. Better production methods become much more valuable when containers, railroads, trucks, warehouses, and routing systems improve alongside them.

The same logic applies to AI. Chips, model design, training, and the cost of running models are the factory. Data collection, cleaning, transfer, storage, permission checks, monitoring, and compliance form the transportation network.

Suppose computing becomes cheaper every year while the cost of moving and governing the data needed for each request stays fixed. The delivered cost of AI then approaches a floor. Better models are still useful, but each additional improvement applies to a shrinking share of the total cost. The return to further model and compute innovation weakens.

Growth resumes when firms can also improve data logistics. Compression, caching, edge computing, interoperability, privacy-preserving technologies, better orchestration, and automated compliance can all make useful data easier to move and govern. Progress on the two sides reinforces itself: lower data-transport costs make better models more valuable, while better models increase the payoff to improving data systems.

Data barriers do damage twice

Data differ from ordinary freight in one crucial respect. A truck carrying a load of lumber cannot deliver that same lumber simultaneously to ten construction sites. Data can be reused.

A transaction record might help provide a service today, detect fraud tomorrow, test a model for bias, monitor its safety, and support a new product later. One use does not necessarily exhaust the data for other uses.

That makes barriers to data movement unusually consequential. A freight restriction mostly raises the cost of the current shipment. A data restriction can raise the cost of the current AI service and reduce the information available for future learning. It charges a toll today and makes tomorrow’s system less capable.

Our model captures both effects. Rules that fragment data across jurisdictions raise the immediate cost of delivering an AI service. They also shrink the stock of data that firms can lawfully pool, transfer, and reuse. Investment in better models and computing falls because the data needed to make that investment valuable become harder to reach.

Existing evidence points in the same direction. Research on international mobile-app markets finds that AI can greatly expand an app’s reach abroad, but that the gain is substantially smaller in countries with tighter restrictions on cross-border data flows. The relevant study by Ruiqi Sun and Daniel Trefler examines international rather than interstate commerce, but the mechanism is closely related.

Our paper develops the theory and leaves the size of these effects to future empirical work. The mechanism yields clear predictions. AI-intensive services that depend heavily on combining data across jurisdictions should be especially sensitive to regulatory fragmentation across subnational borders (e.g., U.S. states or Canadian provinces). Firms should alter where they process data—on a device, close to the user, or in a centralized data center—when policy changes the relative cost of moving it. Smaller firms may be particularly likely to avoid fragmented markets because they cannot spread the cost of dozens of compliance systems across a large customer base.

Fifty laboratories—or fifty tollbooths?

There are good reasons to let states regulate. Local preferences differ. States can experiment with different approaches. They can develop expertise and reveal which rules work before Congress imposes a national policy.

But a state can be a laboratory or a tollbooth, depending on the market being regulated.

Consider an AI service offered nationwide. If every state adopts different consent language, audit duties, impact assessments, record-retention requirements, and rules for reusing data, the company must build separate systems for separate jurisdictions. It may withdraw from some states, divide its data and models into smaller pools, or apply the strictest state’s rule nationwide. That last response creates a peculiar form of regulation: one state effectively sets national policy without a national vote. This used to be called the California effect, although I’m not so sure anyone uses that term these days because of the number of people and companies choosing to leave the regulatory environment that has been gradually built in California.

But our model doesn’t require a California effect, where all states adopt the most stringent state’s policy, in order to inhibit long-run growth in a networked market. Fifty slightly incompatible permission slips may be enough.

The Rhine toll collectors faced the same incentives. Each ruler could gain from one more charge even as the accumulation of charges reduced the value of the whole route. Better ships could not remove that institutional bottleneck. Common rules and compatible infrastructure made the river more useful because merchants could cross a boundary without renegotiating passage.

AI and cloud computing also operate through networks. A state rule governing the movement, processing, or reuse of data can impose costs well beyond that state’s borders. The case for federal action rests on that spillover: the costs of jurisdictional fragmentation cross state lines.

The conditional case for preemption

The model supports federal preemption under identifiable conditions. Federal intervention becomes more attractive when state rules primarily create cross-border data-transport costs, when their requirements are mutually incompatible, and when the lost market integration and learning exceed the benefits of local variation.

This logic can reach some state regulation of data centers. A data center is a node in a network that serves users and moves information across state and national boundaries. State rules that dictate how that node may route, process, monitor, or reuse data can fragment the wider network even when the physical facility sits in one place.

Other data-center issues may be genuinely local. Land use, water consumption, noise, local grid reliability, and the allocation of infrastructure costs could create effects concentrated near the facility. I’m not sure the case has been made for some of these local concerns, but if there were evidence there, states and localities have a much stronger claim to govern those potential harms. A blanket federal exemption for data centers would sweep away too much local knowledge without the justification of crossing boundaries. Like we’re seeing in California, if a state or localities goes too far in trying to address these issues, people and companies can still vote with their feet.

The more defensible policy would combine a federal floor for privacy, security, and accountability with preemption of incompatible state requirements that obstruct interstate data movement or force firms to operate separate technical systems in each jurisdiction. States could also rely on interoperability standards or mutual recognition, allowing different state regimes to coexist when compliance with one accepted standard satisfies another.

Trust belongs in this calculation. Clear privacy protections, security standards, auditability, and liability rules can encourage people and businesses to share data lawfully. A well-designed safeguard may increase the useful data available to AI systems even while imposing some compliance cost. The goal is an open network with enough trust to sustain participation.

That is why the case for preemption is conditional. State variation that produces useful experimentation or addresses local harms has real value. State variation that acts mainly as a series of border checkpoints has a national cost.

The network is part of the technology

AI policy tends to treat the model as the technology and regulation as something applied after the fact. The transportation perspective gives a better picture. The surrounding network (i.e., the technical architecture, data governance, and the rules that connect users to computing) is part of the productive system.

Faster chips cannot realize their full value if useful data cannot reach them. More capable models cannot fully serve a national market if every state boundary adds another toll and another permission system.

Federal preemption should remain a careful instrument. Yet when state rules divide an interstate AI network into incompatible pieces, the growth case for a national standard is real. The economic payoff from AI will depend on smarter models and on whether we build rules that let useful data move, earn trust, and generate the next round of innovation.

A free society cannot live indefinitely on a choice between loneliness and bureaucracy.

Churches, charities, neighborhoods, clubs, and local governments once did more to provide solidarity, moral formation, and practical help close to home. As those institutions weakened over the course of the 20th century, Washington increasingly became the default social problem-solver.

Just think how often you have heard someone reflexively say, “We need a regulation for that!” And Washington “solves problems” in its native language: spending, mandates, prohibitions, guidance, enforcement, and compliance.

That shift has a well-documented economic price. RegData counts about 400,000 restrictive terms in the Code of Federal Regulations in 1970 and more than 1 million today. Research also connects regulatory accumulation to slower growth, fewer new firms, and greater inequality. Those costs then fuel populist demands for still more government action, to ironically address those problems that the growth of the regulatory state is creating in the first place. Leading me to conclude:

When the pews empty, the rulebooks thicken.

And I think the public is often sold a proverbial “bill of goods” on this front. That is, regulations are often discussed as if they will definitely solve the problem that they are supposed to solve, but the reality is that we don’t know! Regulations aren’t systematically re-examined for efficacy. So we get more and more rules, and we don’t actually know if those are delivering their purported benefits.

The answer is not only systematic review of old rules but also a better process surrounding how new regulations are added to the federal code. Both of these would create more room for families, congregations, charities, markets, and local governments to solve problems before they become federal cases.

The Miami Herald edition is here. A complete ungated copy is available through ArcaMax.

My concern is the accumulation of prescriptive rules—including a statutory two-person crew requirement for Class I freight trains—on top of the rules already there.

Americans don’t experience rules one at a time; we experience the combined system. Unlike taxes, regulatory costs rarely appear on balance sheets or receipts, but some can be measured. They’re embedded in operating expenses, delayed or lost innovation and investment, reduced productivity, and higher prices.

The freight estimates are not small and are especially relevant in an era when costs and inflation are on everyone’s minds:

In recent research on the freight transportation sector, my colleagues and I estimated that a 5% increase in federal rules raises unit costs and prices by roughly 0.8% to 2.3% and reduces quantities shipped by about 1.4% to 4.1%—the regulatory equivalent of paying more to move less.

That research can be found here, by the way. Going forward,

The central question is how to achieve more safety per dollar of compliance, not how to micromanage operating practices.

I make the case for performance-based rules here, in the Washington Examiner.

The idea is that alongside traditional legal language, we also publish more computer-friendly and AI-friendly versions. We propose that:

“Structured data and machine-readable rules are made available alongside legal text.”

In practice, this could help both humans and software systems interpret regulations more consistently, thereby reducing ambiguity, improving compliance tools, and making it easier to spot conflicts or inefficiencies in complex rule systems.

But we also argue there’s an important boundary: computation can help implement and analyze rules, but it should not be treated as the source of their legitimacy or meaning.

“The ultimate decisions … cannot be made legitimate by computer analysis alone.”

In other words, making regulations more “computable” can improve transparency and administration, but deciding what the rules should be remains a fundamentally human and political responsibility.

The entire post is here.

<iframe width=”560” height=”315” src=/__u/thirdorder.substack.com/&#8221;%3C/p>

title=”Patrick A. McLaughlin - SAMCEDA&#39;s 2026 &quot;THRED Talks&quot; Series” frameborder=”0” allow=”accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share” referrerpolicy=”strict-origin-when-cross-origin” allowfullscreen></iframe>

The video is the talk as delivered. The essay below develops the same argument in a little more detail.

The infrastructure nobody notices

When a bridge works, nobody notices it.

People cross it on the way to work, school, a doctor’s appointment, or dinner. Trucks carry goods over it. Emergency vehicles use it. Thousands of plans depend on it, but almost nobody pauses to say, “Thank you, bridge.”

We notice the bridge when it stops working.

A closure creates detours. A weight restriction disrupts deliveries. Deferred maintenance becomes an emergency repair, usually at a much higher cost than ordinary upkeep would have required.

Regulations operate in much the same way—as infrastructure that people rarely notice.

They are part of the infrastructure through which economic and social activity moves. They establish rights, responsibilities, standards, permissions, and prohibitions. They can protect health and safety, make transactions more predictable, and create the conditions under which markets function.

Regulation is, in that sense, part of the operating system of the economy.

Calling regulation infrastructure does not mean that every regulation is good, any more than calling roads infrastructure means every road was well designed. It means that regulations form a durable system whose performance depends on its design, condition, and maintenance.

And like roads and bridges, regulatory infrastructure deteriorates when governments concentrate on construction and neglect upkeep.

Government is organized for construction

Most governments have a process for creating new regulations.

There may be a proposal, public notice, stakeholder consultation, a hearing, legal review, and sometimes an economic analysis. That machinery may work imperfectly, but it exists.

Far fewer governments have an equally serious process for rules that are already on the books.

Once adopted, a regulation can remain in force long after the technology, industry, risk, or administrative practice that gave rise to it has changed. The rule persists because persistence is the default.

That does not mean old regulations are necessarily bad. Some old rules remain essential. Others may be outdated, duplicative, contradictory, unnecessarily prescriptive, or simply ineffective.

The deeper problem is that, without a maintenance process, governments often do not know which category a rule belongs in.

Rulemaking is treated as a series of individual construction projects. But the regulatory code is a portfolio of accumulated infrastructure.

Every new rule tends to have a sponsor. The accumulated system rarely has an owner.

Good regulatory management is not anti-regulation. It is anti-neglect.

An inventory is not a verdict

Years ago, I created RegData to quantify regulation at the federal, state, and local levels.

At the federal level, the number of regulatory restrictions grew from roughly 400,000 in 1970 to around 1.1 million in the figures I used for the talk. California’s regulatory code contained approximately 420,000 restrictions. Idaho’s contained about 31,000.

Those numbers do not prove that California’s rules are bad or Idaho’s are good. A restriction count is not a benefit-cost analysis, a measure of enforcement quality, or a ranking of regulatory performance.

It is an inventory.

No serious transportation department would say, “Counting bridges tells us nothing about their structural condition, so there is no reason to count them.”

The inventory is what makes the next questions possible:

Which bridges are in good condition? Which need repair? Which are heavily used? Which are redundant? Which were designed for traffic patterns that no longer exist?

The same logic applies to regulations.

How many requirements exist? Which agency owns each one? When was it last reviewed? Does it still address the problem for which it was created? Does it conflict with another rule? Could its objective be achieved more simply?

Counting is not the end of regulatory management. But it is difficult to imagine competent management beginning without it.

And the size comparison raises an unavoidable question: if one jurisdiction’s regulatory infrastructure is many times larger than another’s, does it have a maintenance system that is correspondingly more serious?

Having the most bridges would not, by itself, prove that a state had the best transportation system. Having the most regulations does not prove that a jurisdiction has the best regulatory system.

Condition matters more than volume.

Rusty bolts

Some of the clearest evidence of regulatory neglect comes from provisions that seem almost comically outdated.

San Mateo County, for example, has a dedicated licensing chapter for fortune telling, covering activities such as clairvoyance, palmistry, astrology, spiritism, and fortune telling. The material I reviewed for the talk set the licensing fee at $2,400 per year, payable quarterly in advance.

I am not here to defend fortune tellers. But the provision raises reasonable maintenance questions.

What problem is the license intended to solve? If the objective is preventing fraud, why is a special occupational license preferable to generally applicable anti-fraud laws? Why was that particular fee selected? What evidence would tell us whether the rule works?

California’s regulations also retain real-estate rescission language involving the telegraph. Another filing rule still contemplates paper copies prepared for placement in a manila folder and attached with an Acco fastener—a piece of office hardware that much of the modern economy has managed to leave behind.

Are telegraphs responsible for California’s housing shortage? No.

Are Acco fasteners the principal obstacle to innovation? Of course not.

But the smallness of these examples is part of the point. Each provision is too minor to provoke a major reform campaign. Each can remain untouched for decades because the cost of fixing it appears larger than the immediate benefit.

A single rusted bolt does not prove that a bridge is failing. But enough rusted bolts, obsolete fittings, and layers of improvised patchwork tell us something important about the maintenance culture.

The provisions are not necessarily the disease. They are diagnostic evidence.

The Third Order problem

Regulators generally make decisions one rule at a time.

Businesses do not comply with rules one at a time.

A business complies with Rule A, Rule B, Rule C, and hundreds or thousands of other requirements simultaneously. Those requirements may come from different agencies and levels of government. They may use different definitions, deadlines, applications, reporting systems, and enforcement practices.

The first-order question surrounding a regulation is usually: What problem is this rule intended to solve?

The second-order question is: How will regulated parties respond? Will they alter a product, hire compliance personnel, change a production process, delay a project, or avoid a market?

The third-order question is: What happens when those responses interact with the rest of the regulatory system?

A rule can have a sensible purpose in isolation and still contribute to a dysfunctional portfolio. Requirements can overlap or contradict one another. Sequential approvals can turn several modest delays into a multiyear process. Fixed compliance costs can be manageable for established firms and prohibitive for potential entrants.

The burden of the system can therefore be greater than the sum of its individual rules.

This is one reason regulatory accumulation is easy to miss. Each rule has a rationale. Each burden may appear modest when viewed alone. By the time the cumulative effects emerge—in the form of fewer startups, slower projects, higher prices, or investment diverted toward compliance—no single regulation appears responsible.

Government manages the flow of new rules.

The public experiences the stock.

What regulatory maintenance looks like

Treating regulation as infrastructure points toward a practical management system. It requires at least three disciplines.

1. Build an inventory

A jurisdiction should know what it owns.

That means more than producing a page count. Governments should identify regulatory requirements, permits, forms, responsible agencies, affected activities, and the date on which each provision was last reviewed or amended.

An inventory does not prejudge whether a rule should stay or go. It creates the information necessary to make that determination.

2. Inspect and repair the existing stock

Old regulations should be reviewed systematically rather than only when a controversy erupts.

The relevant question is not simply whether a rule is old. It is whether the rule still works.

A review might conclude that a regulation should be preserved, strengthened, clarified, made technologically neutral, simplified, consolidated with another provision, or retired altogether.

Physical infrastructure inspection does not begin with a presumption that every old bridge must be demolished. Regulatory inspection should not begin with a presumption that every old rule must be repealed.

But neither should age confer permanent immunity from scrutiny.

3. Establish budget discipline

Without a constraint on regulatory growth, the stock will usually continue to expand.

A regulatory budget can take several forms. The basic idea is to require policymakers to recognize that regulatory capacity is finite and that new obligations should prompt examination of existing ones.

When an agency proposes a new requirement, it might be required to identify outdated requirements that can be simplified, consolidated, or removed. The point is not to claim that every new rule is bad. It is to make maintenance part of the rulemaking process.

British Columbia provides the most important example. The province created an inventory, established a goal of reducing regulatory requirements by roughly one-third, ultimately reduced the stock by more than one-third, and then used regulatory-budget discipline to keep it from simply growing back.

Research I coauthored on the British Columbia experiment estimated that the reform increased the province’s economic growth rate by about one percentage point.

The precise numerical target is less important than the institutional lesson: maintenance happened because someone was made responsible for it.

Without a mechanism that assigns that responsibility, the unglamorous work of regulatory upkeep will almost always lose out to more visible priorities.

Maintenance is not demolition

The familiar political framing pits regulation against deregulation.

The infrastructure framing suggests a more useful distinction: management versus neglect.

Maintaining a bridge may mean replacing it. It may also mean strengthening it, repairing its surface, changing its load rating, redesigning an interchange, or adding new safety features.

Regulatory maintenance should allow for the same range of outcomes.

Some rules should be removed. Some should be simplified. Some should be combined. Some should be updated for modern technology. Some may need to be strengthened because they no longer provide the protection the public expects.

The goal is not the smallest possible regulatory code.

The goal is a regulatory system that delivers public value without imposing unnecessary cost, delay, complexity, or uncertainty.

Indeed, regular review can strengthen the legitimacy of regulation. People are more likely to trust a system that can explain why its rules exist, whether they work, and how they have been updated in response to changing circumstances.

A government should be able to answer four basic questions about its regulatory infrastructure:

What do we own?

What condition is it in?

Which parts remain essential?

What is our maintenance plan?

Failure is usually quiet

A neglected bridge can eventually fail in a dramatic and visible way.

Regulatory infrastructure more often fails quietly.

The startup that is never formed does not issue a press release. The housing project abandoned during predevelopment never appears in the housing stock. The investment redirected from innovation to compliance is difficult to observe. The price increase caused by an accumulation of small requirements arrives without an itemized regulatory receipt.

That invisibility is one reason neglect can persist for so long.

But invisible costs are still costs.

A jurisdiction that continually adds to its regulatory infrastructure without inventorying, inspecting, and maintaining what it has already built should expect deterioration. It may not arrive as a sudden collapse. It will show up as fewer entrants, slower approvals, delayed investment, higher prices, and slower growth.

Regulations are infrastructure.

And infrastructure fails when it is only built and never maintained.

Thanks to SAMCEDA for inviting me to develop and present this argument.

Further reading

Regulatory Reform 101 for County and Municipal Policymakers

Explore RegData and QuantGov

Regulation and Economic Growth: Evidence from British Columbia’s Experiment in Regulatory Budgeting

The key question is not whether Congress can add another requirement to railroads’ compliance stack. It is whether the policies it adopts will move investment toward the things that actually reduce risk in a modern freight network. Set aside the politics for the moment. The first question is whether the RSA would actually improve safety.

The RSA would tighten inspection requirements, broaden hazardous-material oversight, increase penalties, expand defect-detection requirements, and impose more prescriptive operating rules, including crew-size mandates. Some of those provisions address real failure modes. But the bill too often assumes that safety is produced mainly by visible labor and process inputs rather than by information, capital investment, and operational learning.

The strongest case in the bill is for provisions tied to identifiable failure modes, such as defect detection and hazardous-material information; the weakest is for blanket input mandates, such as minimum crew sizes rules and mandates around the number and timing of human-performed (and only human-performed) inspections. The mistake is to assume that safety is produced mainly by crew counts and mandated procedures when the more important margin is information, capital investment, maintenance targeting, and operational learning.

This seems obvious, but we should not assume that regulatory actions lead to the desired outcomes. In this case, let’s not assume that rail safety is produced mainly by labor inputs and procedural commands, but instead look at historical data. Start with the minimum crew size requirement contained in the bill, mandating that at least two people be in the crew of virtually every Class I freight train. This cannot be justified by data, as the Federal Railroad Administration itself admitted in 2016 when it could not “provide reliable or conclusive statistical data to suggest whether one-person crew operations are generally safer or less safe than multiple-person crew operations.” So it is not a given that more people on the train equals more safety, and the picture gets worse for safety when you consider the tradeoffs that railroads will face, as I explain below.

Improvements in safety are increasingly coming from earlier detection of equipment failure, better maintenance targeting, stronger track and equipment, and control systems that allow problems to be identified before they become accidents. On the other hand, the leading cause of accidents and incidents in rail is human error, and that has been the case for well over a decade.

Put differently, safety is increasingly being produced by information, capital, and operational learning, not by simply putting more people in the train crew or mandating a minimum amount of specific human activities.

This has a clear policy implication. Regulation is not just a rule. It is also a capital-allocation decision—there are second- and third-order effects to think about. A mandate to spend more on a staffing input, or on a rigid inspection process, is also a decision not to spend on something else. In rail, that opportunity cost is rarely trivial. It may mean fewer detectors, slower equipment renewal, delayed infrastructure investment, or slower adoption of predictive maintenance and other technologies that address risk more directly. Once that tradeoff is recognized, the usual Washington debate starts to look badly framed.

Safety and productivity rise together

The empirical evidence points in the same direction. In recent work with Bentley Coffey and Pietro Peretto, we examine how federal regulatory accumulation has affected freight transportation over the last half century. The accumulation point matters: firms respond not just to one new rule, but to the stock of restrictions they face and the way that stock shapes productivity, investment, and output over time. Train operations are a complicated system, and they are regulated by a complicated system of rules. It’s never one rule by itself. Using 50 years of industry-specific regulatory data matched to public freight data, we find that higher regulatory burdens are associated with lower labor productivity in all four major freight modes. In rail, the effects are broader: higher regulatory burdens are associated with lower labor, fuel, and capital productivity.

That last margin is the key one here. The technologies and operating improvements that make railroads safer—detectors, track renewal, equipment upgrades, predictive maintenance, and control systems—are capital-intensive. If regulation lowers capital productivity or diverts capital into lower-value uses, it slows the very process through which safety has historically improved. In our simulations, a 5 percent increase in rail regulatory restrictions raises rail unit costs by 2.3 percent and reduces rail output by 4.1 percent in the first year. That is not just a cost story. It is a safety story, because the margin being squeezed is capital formation. Because productivity growth also slows, the losses persist and compound. The system gets smaller, and a smaller, less productive system has less capacity to finance the fixed investments that improve safety.

That is exactly what we should expect once transportation is treated as part of the growth process rather than as a background condition. When freight moves more cheaply and reliably, markets widen, the return to innovation rises, and firms have stronger incentives to improve both production and distribution. Rail safety belongs in that same story because many of the same investments reduce cost and reduce risk at the same time.

History points the same way. In earlier work with Jerry Ellig on railroad safety after the Staggers Rail Act of 1980, we found that partial economic deregulation, not additional safety regulation, best explains the major improvement in railroad safety. Railroads with better incentives and stronger finances could invest more in track, equipment, maintenance, and new technologies. From 1978 to 2013, total accidents on the major Class I railroads and their predecessors fell from more than 11,000 to 1,867 even as revenue ton-miles roughly doubled. In rail, safety and productivity are complements.

What a serious safety agenda would look like

A serious agenda would do three things: target demonstrated failure modes, rely on performance standards where possible, and preserve flexibility in how railroads combine labor, capital, and technology. Provisions aimed at identifiable failure modes deserve serious attention. Measures that improve defect detection, for example, have a much stronger logic than measures that simply prescribe visible inputs. The same is true of better hazardous-material information and emergency response. If a proposal begins with a real mechanism of harm, it at least starts on solid ground.

But much of the current agenda is built around visible inputs rather than system performance. The blanket two-person crew requirement is the clearest example. The empirical basis for it has never been strong. More important, it asks the wrong question. Railroads do not make staffing decisions in isolation. They make them alongside decisions about route characteristics, control systems, monitoring technology, rolling stock, and operating procedures. Freezing one input by law is a poor substitute for understanding how the system actually produces safety.

The same concern applies to inspection mandates that prescribe process rather than outcomes. Of course inspections matter. But the relevant questions are which inspections, where, how often, and at what opportunity cost. In a technologically changing industry, Congress is often tempted to regulate what it can easily see and count. That is politically understandable. It is also often counterproductive. A process mandate can satisfy the demand for action while slowing the adoption of technologies that do the job better.

A more serious agenda would start from a different premise. Safety policy should target demonstrated failure modes, require accountability for results, and preserve room for firms to combine labor, capital, and technology in the least-cost way. That means more emphasis on performance standards and less on design standards and rigid mandates. It means a higher burden of proof for mandates with obvious costs and uncertain benefits. And it means remembering that the cost of moving freight does not stay in the freight sector. It works its way through supply chains and into the prices of ordinary goods, with the burden falling hardest on households least able to absorb it.

The larger point, though, is broader than any single bill. Rail safety is not mainly a staffing problem. It is about investment across all inputs of production. If Congress wants safer railroads, it should ask which policies increase the sector’s capacity to invest in detection, maintenance, infrastructure, and operational learning. If it keeps regulating the most visible inputs instead, it will keep regulating the wrong margin.

Below is some of what I wrote in my column about the policy influence of the RegData project. Do give it a read, and while you’re there, maybe subscribe too:

The most important milestone for any measurement project is adoption—when institutions start treating it as a legitimate input, not a curiosity.

Here are three concrete examples that matter because they sit at different choke points of policy-making: macro policy discussion, congressional analysis, and state-level reform.

1. The Federal Reserve: measurement becomes a macro input. In a January 14, 2026, speech on regulation and the supply side, Federal Reserve Governor Stephen Miran pointed to “progress in measuring regulation” using advances in natural language processing and highlighted QuantGov as a concrete example—moving beyond page counts toward counting restrictions and quantifying regulations by industry. That is a notable shift: it signals that regulatory accumulation is increasingly viewed as something that can be discussed with evidence, not just asserted.

2. CBO: regulatory measurement enters scorekeeping. In materials discussing its work, the Congressional Budget Office has noted that it “relied in part on RegData—a database linking federal regulations to industries” in its analysis of H.R. 1. That single sentence matters because CBO is the institution Congress leans on for neutral analysis; if regulation can be measured well enough to appear in that analytical machinery, it is less likely to remain an unscored, untracked policy lever. See the CBO blog post and the related CBO presentation.

3. Texas, Virginia, Idaho, Ohio, and more states: measurement as a reform baseline. In April 2025, when Texas created the Texas Regulatory Efficiency Office, Governor Greg Abbott cited one of my studies that, using RegData, showed that Texas was the fifth most regulated state in the country, with 274,469 regulatory restrictions in its administrative code (See reporting and the underlying policy brief). Virginia, taking inspiration from the RegData approach, has spent the past four years inventorying its regulatory restrictions and tracking progress toward its goal of cutting red tape by 25 percent (it actually topped that—cutting about 33 percent by the end of Governor Glenn Youngkin’s term). Several other states, including Idaho and Ohio, have overtly followed the RegData methodology for similar red tape reduction. This is what progress looks like: reform debates anchored to quantitative baselines that can be checked after the bill-signing ceremony. The business adage, “What gets measured, gets managed,” is also true in government.

But if you read Justice Gorsuch’s concurrence (scroll down about 28 pages), it’s hard to miss that he’s doing something else: he’s trying to reframe a familiar fight about the administrative state as a problem of institutional design. The Wall Street Journal editorial board noticed that too, describing the concurrence as an effort to push Congress back toward its constitutional role rather than a simple “power grab” by courts.

I want to take Gorsuch’s framing seriously, because it aligns with something regulatory economists have been saying for decades, and because it points toward an underappreciated bottleneck in the nondelegation revival project:

Congress can’t reclaim power if it doesn’t know how much power it has actually ceded.

That sounds abstract. It’s not. It’s a data problem. And it’s similar to the problem that motivated the creation of RegData in the first place (namely, how can we know the economic effects of the growing administrative state if we can’t quantify it?), and it happens to be the subject of the Nondelegation Project.

Sidenote: I’m speaking at a webinar on Thusrday, February 26, at 2:00 PM ET, about the Nondelegation Project, which (among other things) uses AI to produce a comprehensive inventory of Congress’s statutory delegations and tells you whether the language in the statute is vague, or specific. I’ll be joined on the panel by Pacific Legal Foundation’s legal scholar Luke Wake, former director of Virginia’s Office of Regulatory Management Reeve Bull, Vulcan Technologies (an AI RegTech startup) founder and CEO Tanner Jones, and moderated by PLF’s Mitchell Scacchi. Register for the webinar here.

Back to Justice Gorsuch…

“Not anti-administrative state. It is pro-Congress.”

Gorsuch’s most quoted line will probably be this one:

“The major questions doctrine is not ‘anti-administrative state.’ It is pro-Congress.”

That’s not a rhetorical flourish. It’s a claim about incentives and institutional drift.

Gorsuch leans explicitly into what economists call the principal–agent problem: principals sometimes give their agents broad delegations because they can’t anticipate every eventuality, but that same looseness creates the risk that delegees will “exploit” ambiguity to expand their own discretion. Courts, in Gorsuch’s telling, sometimes have to develop default rules to protect the principal, because inevitably situations will arise that were not anticipated in the original “contract,” or in this case, the statutory delegation from Congress to an agency. Or maybe in the original, original contract — the Constitution.

This is, basically, principal–agent theory in judicial prose.

And it fits cleanly with how a systems-oriented approach to the regulatory process: the outcomes we get aren’t just a function of “good intentions” or “expertise”; they’re a function of process design and whether that design aligns incentives between Congress (principal), regulators (agents), and the public.

So far, so familiar.

The part that isn’t familiar, at least in the public debate, is what Gorsuch says near the end. He’s not just defending a doctrine. He’s defending the legislative process itself as the stabilizing mechanism:

laws that survive the legislative process “tend to endure,” letting people plan “when the rules” don’t “shift from day to day”… and the legislative process is a “bulwark of liberty.”

Economists would agree that policymakers should prioritize policies that let people plan and let businesses make long term investment decisions. If that is the goal, and durable, democratically accountable lawmaking is the mechanism that gets us there, then the relevant question isn’t merely whether courts should enforce clearer statements.

The relevant question is whether Congress has the information infrastructure to legislate clearly and to oversee what happens after it delegates.

Right now, it clearly does not.

The missing balance sheet of delegated power

Congress has a Congressional Budget Office. It has budget baselines, scoring conventions, reconciliation instructions, and a whole ecosystem built around a single idea:

if you don’t measure something, you can’t manage it.

Now you could argue that even if you can measure something, politics still prevents good management. That’s where the structure of the system comes into play—checks and balances, separation of powers, etc. But as far as even simple measurement goes, for the regulatory state (arguably at least half of modern governance), we mostly operate without an equivalent ledger.

That’s one reason debates about “the administrative state” tend to be so airy. People argue in abstractions because they don’t have a shared set of facts about:

• how many binding constraints exist in the regulatory code,

• which agencies are producing them,

• which industries they target,

• what their impact is, and

• which statutory delegations are doing the most “work” in the background.

This is the gap RegData and its umbrella project, QuantGov, were designed to address.

A brief refresher on RegData as regulatory accounting

A decade ago, regulation was often “measured” by counting pages in the Federal Register. QuantGov’s “about” page captures why that’s noisy—pages include non-regulatory material, and even deregulation can require pages.

RegData instead counts regulatory restrictions directly in the Code of Federal Regulations (CFR) using terms that usually signal binding constraints—“shall,” “must,” “may not,” “prohibited,” “required,” and similar phrases. And it also (probabilistically) tells you what parts of the economy (e.g., industries such as manufacturing or home construction) are affected by those regulatory restrictions.

That yields a running tally of the “stock” of constraints—something closer to regulatory accounting than regulatory rhetoric.

A few highlights—sorry if you already know these facts, but not all my readers have seen this before:

• The CFR had about 400,000 restrictions in 1970, but has grown to more than 1 million restrictions.

• This accumulation of regulations over time has been shown to slow annual economic growth by nearly one percentage point.

• And yes: it is genuinely hard for humans to even read the modern regulatory corpus. It would take over three years just to read the CFR.

Here’s the connection to Learning Resources:

Gorsuch is basically saying, “If an agent is claiming extraordinary power, courts should hesitate unless Congress spoke clearly.”

But clarity doesn’t come from vibes. It comes from institutional capacity—and in this space, capacity begins with measurement.

Which brings me to the next step.

From counting restrictions to tracing authority

RegData answers: How many constraints exist, which agencies are they coming from, and what parts of the economy are affected?

But nondelegation (and major questions) ultimately ask a different question:

Where did the authority come from?

If you can’t trace a regulation back to the statute that purportedly authorizes it, you can’t even begin to talk seriously about whether an agency has exceeded a delegation—or whether Congress delegated something too broadly in the first place.

That’s the core motivation behind the Nondelegation Project: an AI-enhanced database that links each CFR part to the U.S. Code sections it cites as statutory authority, and then classifies those delegations as general (broad/vague) or specific (more constrained).

AI can take a 190,000-page, 245-volume CFR and make it navigable as an interactive, searchable authority map. Such an exercise yields some descriptive facts that, to me, are another “wow” moment:

• The database captures 56,000+ delegations, and about 37% are coded as general—broad, vague, open-ended grants.

• The agencies with the highest number of general delegations include FERC and EPA—but the agencies with the highest percentage of general delegations include DOJ and NASA.

• Using RegData’s restrictions metric, EPA is (unsurprisingly) the most pervasive user of vague statutory delegations in the dataset. EPA also happens to be the agency with the most restrictions in the CFR.

However, this is where an intriguing (and somewhat counterintuitive) point emerges:

The nondelegation problem isn’t just about “the big villain agencies.”

The early data suggest that broad delegations are not confined to the usual caricatures. Some agencies that are not central to the political imagination of “the regulatory state” still operate under a high share of general authority language.

That matters for doctrine and for reform along the nondelegation lines.

Because if you’re trying to “revive” nondelegation, or even just apply major questions more rigorously, you need to know where general delegations are concentrated and what regulatory output they enable. You need to know which statutes are authorizing the most regulations, and whether they are vague, and how vague they are.

A doctrine without an inventory is like auditing without books.

Major questions as an information-forcing rule

One of the more productive ways to read Gorsuch’s Learning Resources concurrence is as a defense of an information-forcing default rule.

In private law, default rules often exist because contracting parties can’t (or won’t) specify everything ex ante. Courts sometimes force clarity when the consequences are large, precisely to avoid opportunism and misunderstanding.

Gorsuch is telling a similar story in public law: “highly resourceful” executive actors (i.e., agencies) have incentives to exploit ambiguity; courts respond by demanding clearer authorization for extraordinary claims of power.

What’s missing in the discourse is this: information-forcing rules only work if someone has the infrastructure to produce and verify the information.

That’s why I think the most important connection between Learning Resources and nondelegation isn’t ideological. It’s operational.

If courts are going to insist on clarity in delegations (which, thankfully, seems to be the trend), we should expect:

1. More litigation focused on identifying the “true” statutory basis for major regulatory programs.

2. More strategic drafting by agencies to shore up authority citations.

3. More pressure on Congress to write statutes that actually specify the scope of delegated authority in a way that can be monitored.

All three of these require what the regulatory system rarely provides: a usable, public-facing map from statute → delegation → regulation → binding constraints.

That’s what RegData + the Nondelegation Project are trying to approximate.

A counterintuitive implication: this isn’t necessarily deregulatory

There’s a cartoon version of major questions and nondelegation that treats them as shorthand for “less regulation.”

That’s not obviously right.

Gorsuch’s own language is closer to: make Congress own the big decisions.

And that kind of “owning” can cut both ways:

• If Congress wants a sweeping policy, it can enact it clearly and explicitly—making it more durable and less vulnerable to whiplash.

• If Congress can’t agree, the doctrine blocks executive improvisation that substitutes administrative creativity for legislative consensus.

So the deeper stakes aren’t “more regulation vs. less regulation.” The stakes are:

policy by durable legislation vs. policy by delegated improvisation.

That’s why the WSJ editorial board’s framing—that Gorsuch is trying to revive Congress—lands in an important way even if you don’t share their ideology.

If you care about legitimacy, predictability, and accountability, the “revive Congress” project has to include the informational plumbing.

Where this points: an actual delegation ledger

I’m not naïve about Congress. Legislating is hard, slow, and often ugly.

But there are tangible, boring, technocratic reforms that would make “pro-Congress” doctrines less of a slogan and more of a workable governance model:

1. A standing, public inventory of delegations.
   Treat statutory delegations the way we treat fiscal outlays: catalog them, tag them, track them, and update them. The Nondelegation Project is a prototype for what this could look like at scale.

2. A regulatory “stock” metric in routine oversight.
   Congress can’t do oversight if it doesn’t know the stock and flow of constraints. Counting restrictions isn’t perfect, but it’s comprehensive, transparent, and replicable. It captures far more of the regulatory universe than cost-benefit analysis does in practice, without any of the possibility of gamesmanship (deliberately bending the analysis to match a predetermined conclusion).

3. Institutionalized lookback and subtraction.
   Regulatory accumulation is the default setting. Absent deliberate management, the stock grows, with all sorts of negative consequences. I’ve been making this point for years.

4. Use AI for oversight, not just for rulemaking.
   AI will increasingly help identify deregulatory targets, including rules that are “authorized but not required,” or that rely on vague statutes likely to be challenged in a post-Chevron world. This is already happening.

None of this resolves the deepest normative disagreements about the administrative state.

But it does something more basic: it gives Congress, courts, and the public a way to argue from shared facts rather than from caricatures.

Closing thought

Gorsuch’s concurrence is a reminder that constitutional structure is not a museum piece. It’s an operating system. And operating systems fail when they lack feedback loops.

If major questions is going to be “pro-Congress,” then Congress needs the tools that principals use everywhere else: measurement, auditing, and the ability to see what its agents are doing.

RegData started as a way to quantify what was previously unquantified. The Nondelegation Project is an attempt to connect those quantities back to their sources of authority.

In other words: it’s an attempt to build the ledger that a “revive Congress” project presupposes.

The OECD’s December 2025 Economic Outlook is explicit that the investment shortfall is not purely cyclical. Business investment has been held back by uncertainty and demand, but also by structural factors; and in the OECD’s own list of pro‑investment reforms, cutting regulatory costs sits alongside competition reforms, easing permitting and licensing rules (which are, of course, simply a subcategory of regulatory costs…), and the usual macro foundations. The longer chapter on reigniting investment pushes the same theme: policy uncertainty is not just about trade or fiscal policy; it also reflects regulatory policy, raising risk premia and firms’ hurdle rates.

This mechanism closely parallels the argument in my 2020 publication (with Bentley Coffey and Pietro Peretto), “The Cumulative Cost of Regulations.” We show that a growing regulatory stock distorts business investment, thereby slowing TFP growth and ultimately overall economic growth. It is notable and appropriate that the OECD now explicitly situates regulatory burden within its macro productivity framework rather than treating it as an ancillary institutional issue.

One of OECD’s important contributions in this line of work is measurement, as they explain in this working paper. It introduces a task‑based measure of regulatory compliance costs, in a similar but more extensive manner as followed by Trebbi and Zhang (2022). This measure estimates the share of labor, and in the U.S. case the share of the wage bill, devoted to regulation‑related tasks such as reporting, inspections, audits, and compliance activities. From the OECD study, for the United States, the OECD estimates that roughly 4.2 percent of the wage bill is absorbed by regulatory tasks, amounting to more than $500 billion annually.

Two features of this approach are worth emphasizing. First, it treats compliance as a factor‑input wedge, measuring real resource diversion rather than statutory text. Second, it is conceptually distinct from text‑based regulatory stock measures, including State RegData. That distinction is a feature, not a flaw. If the two approaches align, the result is powerful cross‑validation. If they do not, the divergence is informative about what different measures of regulation are actually capturing.

The OECD’s state‑level results make this comparison feasible. Regulatory compliance costs vary substantially across U.S. states, with Idaho at the low end and states such as New Jersey near the high end. Moreover, Idaho stands out as exhibiting a decline in compliance costs after 2019, following the state’s adoption of its Red Tape Reduction Act and Licensing Freedom Act and other regulatory reforms led by Governor Brad Little. The OECD treats this as a validation exercise for its measure, tying a discrete institutional reform to a measurable reduction in compliance intensity.

The OECD also links higher compliance costs to weaker outcomes. Using within‑state variation over time, it finds that increases in regulatory compliance costs are associated with lower labor productivity and a smaller employment share in young firms. These magnitudes are modest but economically meaningful, and they point squarely at investment and dynamism as the transmission mechanisms.

All of this raises an obvious next question: how does the OECD’s task‑based compliance measure relate to State RegData’s text‑based restriction counts? On its face, the OECD index is not derived from State RegData. It is built from occupational task descriptors and employment and wage data. That independence makes the comparison more valuable, not less.

I put together a quick analysis of the OECD data compared to my State RegData numbers from 2023, and the initial results look promising. The nearby figure shows the scatterplot of the two datasets for the US states available in both. The correlation coefficient is 0.32 – not bad for a single year of data. I hope to be able to look at first-differenced data soon and see how changes in regulation, as measured by RegData, correlate with changes in regulatory compliance costs, as measured by OECD.

Figure 1: Correlation of compliance costs and regulatory restrictions

Figure 1. OECD compliance costs and State RegData restrictions across U.S. states (2023). OECD compliance cost index: share of the wage bill devoted to regulation-related tasks, from Figure 2.2 in Andrews, Turban, and Tyros (2026), OECD Economics Department Working Papers No. 1856. State RegData restrictions: counts of restrictive terms in state administrative codes (snapshot dates in 2023 from State RegData). The fitted line is an OLS regression of the OECD index on ln(restrictions); N=48 (states with 2023 State RegData snapshots available; excludes DC, WV, and AR).

A natural research agenda follows. First, the OECD compliance index should be systematically compared with State RegData restriction density across states and over time. Second, Idaho’s post‑2019 reform provides a clean case study for examining whether changes in compliance costs are mirrored by changes in regulatory stock measures, and where they are not. Third, business investment outcomes should be brought explicitly into the analysis, exploiting state‑level variation to test whether and to what degree compliance costs, regulatory stock, or both explain investment behavior. Finally, both measures should be used jointly to assess whether regulatory burden operates through legal accumulation, administrative practice, or enforcement intensity.

Regulatory burden is now a part of the macroeconomic discussion of productivity and investment—something that is long overdue, as I recently wrote at Barron’s. The U.S. federal structure offers an opportunity to push this work further, using state‑level variation to clarify measurement, validate competing indicators, and better identify the channels through which regulation affects growth.

My argument is pretty simple: legal/regulatory changes (especially zoning and permitting) are necessary, but they may not be sufficient. Reforms underdeliver when they stop at the “we passed a bill” stage—because housing affordability is the end result of a long production chain, and that chain only moves when someone is accountable for making it move.

What the housing supply agenda gets right

For a long time, housing policy talk leaned heavily on demand-side moves—subsidies, credits, and other ways to help buyers or renters pay more. Or it devolved into blame games.

The more recent shift toward supply is a welcome change. Across states and cities, we’ve seen reforms that loosen zoning, simplify permitting, and legalize more housing types—often in places that used to reserve large portions of land for only one kind of housing.

That’s the right direction. When supply can’t expand fast enough and a new generation wants housing, prices rise.

Why results are mixed so far

Many of these reforms are still not clearly delivering broad-based price relief—at least not yet. That doesn’t mean the reforms are wrong. It means we should be honest about what it takes to translate “legalized” into “built,” and “built” into “cheaper.”

A few reasons the affordability payoff can take longer than advocates (or critics) expect:

• Timing and implementation delays. Many reforms have been tied up in litigation, phased in slowly, or require agencies and local governments to rewrite rules before anything changes on the ground.

• Housing supply is a pipeline, not a switch. Permits need to turn into projects; projects need to become completed homes; completed homes need to be numerous enough to influence the broader market.

• Short-run economic headwinds can swamp early gains. Interest rates, construction costs, and demand pressures can overpower near-term effects even when permitting activity improves.

In other words: some places are successfully changing what’s allowed and what gets permitted (including more “missing middle” housing). That’s meaningful progress. But it’s not automatic, it’s not always fast, and the positive effects (towards more affordable housing) can be masked by other economic factors.

The other half of the affordability equation: the cost to build

Even if zoning and permitting are fixed, housing can remain expensive if it’s still costly to produce. As I wrote, “Legalizing more housing is only one part of a much larger system that determines housing costs.” The production cost isn’t driven only by land and labor. It’s also shaped by layers of regulation that sit “on top” of zoning: building codes, design requirements, energy rules, and a long list of mandates.

Some of these rules provide benefits people clearly want and are willing to pay for. Others add costs with fuzzier returns.

The big takeaway on building cost is this: even if a housing unit can be more easily built from a zoning and permitting perspective, actual construction costs may mean the new build isn’t economically viable. And if it’s not viable, it won’t get built—meaning the legal reform doesn’t translate into real supply.

Regulatory streamlining with a center of gravity

One of the examples I discuss in the op-ed is Virginia, which pursued a broader regulatory reform effort with oversight from a dedicated central office (its Office of Regulatory Management). The Virginia approach is to inventory existing rules, eliminate unnecessary requirements, and justify what remains. Within that broader effort, Virginia streamlined parts of its residential building code, and state officials estimated the change reduced the cost of constructing a new home by $24,000.

I also point to other state efforts with similar “system-wide” ambitions—initiatives that force agencies to re-justify rules, set reduction targets, or require staged cuts with legislative oversight.

I used to stand by the axiom, “What gets measured, gets managed.” But these days, having studied broad reform efforts across numerous states and jurisdictions, I’ve refined my take. Measurement alone doesn’t appear sufficient if no one is handed the job of management. Looking across states at where red tape reduction has been very successful vs. moderately or not successful, I’ve identified one common thread: where reform has a real management structure, it’s more likely to translate into real change. Virginia’s Office of Regulatory Management is a great example of creating a management job in charge of effecting the reform.

Do permitting and zoning reforms require similar oversight offices? Is it sufficient to merely create the option for multi-family housing to be built where single-family was previously the only type allowed? Or do policymakers need to actively check in on whether those multi-family permits are actually being issued? Based on what I’ve seen for broad reg reform efforts, you need to actively check in on things and probably need to actively manage those people who are responsible for permitting. The reform has effectively changed their job, and someone needs to make sure they’re doing the new job correctly.

Success depends on oversight, execution, and accountability—someone waking up every day responsible for results. When responsibility is spread across lots of actors, it becomes easy for everyone to say the right things while nothing moves quickly. When there’s clear ownership, reforms have a better chance of producing measurable outcomes.

Two takeaways I hope policymakers (and advocates) absorb

1. Don’t treat zoning reform and regulatory streamlining as competing agendas. If you want affordability, you’ll probably need both: make it easier to build and make it less expensive to build.

2. Shift focus from announcements to operations. Passing a reform is the start of the work, not the finish line. The real question is: who is responsible for ensuring the reform translates into permits, projects, completions, and (eventually) prices?

Concluding thoughts

If a policymaker were to ask me to evaluate whether a housing reform agenda is likely to deliver, I’d look for a few concrete signals that go beyond supply legalization (i.e., zoning/permitting reforms):

• Is there a named office / leader responsible for implementation and outcomes?

• Are there metrics that track the full pipeline (not just “units legalized”)?

• Are policymakers pairing supply legalization with cost-focused reforms that make projects pencil out?

• Is there a plan for follow-through over multiple years, not just one legislative session?

Housing reform is overdue in much of the country. But if we want it to make a real dent in prices, we have to treat it like a management challenge that requires sustained execution.

Introduction

Regulations are created to solve real problems, but over time they tend to accumulate, often without systematic review. I study the economics of regulation, which I’ll explain momentarily, and my talk today summarizes evidence on how regulatory accumulation affects economic outcomes and how jurisdictions can modernize their regulatory systems without sacrificing legitimate public goals such as safety, health, or environmental protection.

The evidence supports three core findings:

· Regulatory accumulation is the default, absent deliberate management.

· Accumulated regulations have measurable economic costs, including slower growth, higher prices, reduced business formation, and increased inequality.

· Targeted regulatory management can reverse these effects, often at very high returns on investment.

I. A Framework for Understanding Regulation

What do I mean when I say, “the economics of regulation?” I generally think about it in three stages.

Stage 1: How Regulations Are Made

Nearly all U.S. jurisdictions follow a formalized rulemaking process. Examples of key institutional features may include:

· Notice-and-comment requirements

· Required economic analysis (e.g., cost–benefit analysis, paperwork burden estimates)

· Requirements—or lack thereof—for periodic review of existing rules

These features determine how easily regulations enter the system and how rarely they exit. Basically, how the sausage is supposed to get made, if there is a formalized process…

Stage 2: Regulatory Outputs

Beyond the formalized process, regulators produce outputs, including:

· The quantity of rules on the books

· The rate at which new rules are added

· The design of rules (e.g., flexibility, exemptions for small businesses, performance-based vs. prescriptive standards)

Basically, how the sausage actually gets made (as opposed to what the process claims it does). Over the past decade, much of my work has focused on this stage—developing quantitative measures of regulatory output that can be compared across jurisdictions and over time.

Stage 3: Economic Outcomes

Regulatory outputs can be linked to measurable outcomes, which is what economists like me need to do in order to make claims about the relative merits of different policies. Examples of economic outcomes include:

Business outcomes:

· New business formation

· Employment

· R&D spending and patenting

Household outcomes:

· Consumer prices

· Income levels

· Poverty rates

· Income inequality

The central question is, for someone studying the economics of regulation, is not whether regulations exists, but how regulations affects people’s economic lives.

II. What the Data Show

1. Regulatory accumulation is the norm

The buildup of regulations over time seems to be the default setting in modern democracies, at all levels of governance that I’ve studied so far. Using data from the RegData project:

· Federal regulatory restrictions increased from about 400,000 in 1970 to about 1.1 million today.

· States vary widely: Idaho has about 27,000 restrictions, California about 400,000, and the median state about 125,000.

· State regulatory codes grow about 1–2% per year on average, absent intervention.

In short: without active management, regulations steadily pile up.

2. Accumulation has economic consequences

The findings are consistent across multiple studies:

· Economic growth: Regulatory accumulation is associated with roughly 1 percentage point lower annual growth on average. Given typical growth rates of 2–3%, this is substantial.

· Business formation: New firm creation declines as regulatory burdens rise.

· Prices: Higher regulatory burdens correlate with higher consumer prices.

· Poverty and inequality: States with greater regulatory accumulation tend to have higher poverty rates, and evidence suggests regulation contributes to income inequality, particularly through restrictions such as occupational licensing that limit entry and upward mobility.

These estimates are typically produced while controlling for taxes, demographics, and other policies. Regulation is not the only driver—but it is a meaningful one.

III. Addressing a Common Objection: “But Regulations Are Meant to Help”

A common concern is that regulations were enacted to address real harms—fire safety, health risks, environmental damage—and therefore reducing regulation must worsen outcomes. That framing misses the point.

The issue is not regulation versus no regulation, but whether existing rules still deliver benefits commensurate with their costs.

A concrete example from housing policy illustrates the challenge:

· Many building codes require multi-stairwell apartment designs based on fire standards developed a century ago.

· Modern safety technologies (sprinklers, alarms, materials) substantially reduce the original risk.

· Allowing single-stair designs can lower construction costs, increase density, and reduce housing prices—without sacrificing safety.

The economic harm arises not from safety regulation per se, but from outdated rules that are never revisited. Most jurisdictions lack systematic processes to ask whether a rule is still working as intended and whether benefits still justify costs.

IV. Evidence That Reform Works

Jurisdictions that actively manage regulation see measurable gains, including:

· British Columbia reduced its regulatory stock by roughly one-third and experienced improvements in growth and entrepreneurship.

· Virginia’s Office of Regulatory Management (ORM) estimated about $1.4 billion in savings from systematic review and modernization.

As I’ve written here in the past, successful reforms share three features:

· A quantitative baseline

· A clear target and tracking mechanism

· Accountability—someone responsible for results

These are management reforms, not ideological exercises.

V. Housing Affordability and ROI

Housing illustrates why this matters now:

· Virginia identified building code provisions that raised costs without commensurate safety benefits.

· Estimated impact: about $24,000 reduction in home construction costs per unit.

· The reform effort required roughly three years and four FTEs, yielding an estimated ROI over 1,000% (when looking at total estimated annual cost savings of $1.4 billion).

New AI-based tools further reduce the cost of identifying red tape and conducting economic analysis, making these reforms feasible even for smaller jurisdictions.

VI. Implications for States, Counties, and Cities

Any jurisdiction can adopt a pragmatic regulatory management strategy:

· Use data to establish a baseline

· Review regulations systematically

· Focus on outdated, duplicative, ineffective, or unnecessarily costly rules

The empirical evidence predicts:

· Faster economic growth

· More startups

· More affordable housing

· Greater economic mobility

Please don’t fall for the false dichotomy of regulation vs no regulation. This is not about abandoning protections, if they actually work. It is about making the regulatory system work better.

So why is ATI—more specifically, the American Association of Railroads’ waiver request to expand its use—still waiting for an answer from the FRA?

A recent Washington Examiner report by Christian Datoc explored this question, noting that the FRA is taking longer than expected to decide whether to expand automated inspection under a waiver filed this spring. That delay has raised eyebrows not because ATI is controversial, but because the technical case for it has been consistent for years and across administrations.

Datoc summarizes the situation this way:

“The Federal Railroad Administration is taking its time deciding a potential expansion of automated railroad track inspection technology… The decision itself involves a waiver… that would provide rail companies some exemptions from adhering to 1971’s regulations dictating the method and frequency at which railroads must be visually inspected for defects.”

In my own recent DC Journal op-ed, I described the broader pattern: regulators, unions, and other stakeholders sometimes align—intentionally or not—to slow the adoption of safety-enhancing technologies. ATI has become the clearest example, and it reminds me of Bruce Yandle’s famous Bootleggers and Baptists explanation for why we sometimes observe unlikely partners coalescing around certain regulatory policies.

But the current delay raises a genuine question: What is driving the holdup now? Several explanations deserve consideration.

Possibility 1: Union Pressure Still Matters

Some labor unions remain skeptical of expanded ATI use. They argue that ATI expansion could reduce inspection jobs and might not improve safety.

Datoc reports:

“Union members specifically argue that the expansion of the technology would not demonstrably improve rail safety and could potentially jeopardize inspection jobs.”

While in reality ATI tends to shift work toward repairs—because it finds more defects than manual inspection—it’s reasonable to assume unions still worry about long-term automation trends. Union opposition played a key role in slowing ATI waivers under the Biden administration, and it may be part of the timeline today as well.

Regulatory agencies rarely move quickly when influential stakeholders object.

Possibility 2: Federal Agencies Are Still Recovering From Shutdown Disruptions

Another possibility is simply bandwidth. DOT has been juggling a number of operational priorities, especially ongoing air traffic controller staffing shortages and post-shutdown disruptions.

ATI waiver decisions require careful technical review and legal vetting. If DOT staff have been focused on stabilizing aviation operations or responding to urgent system issues, it wouldn’t be surprising if lower-profile decisions are temporarily paused or slowed.

Possibility 3: The FRA Is Being Extra Cautious After Recent Litigation

ATI waivers have been the subject of major litigation. Most notably, the D.C. Circuit Court of Appeals ruled in 2023 that FRA’s rejection of BNSF Railway’s ATI expansion request was “arbitrary and capricious” and called the pilot program “an unqualified success.”

When an agency has been reversed in court, it often proceeds more cautiously the next time a similar issue comes up. The FRA may simply be building a more robust administrative record before issuing a decision—especially if it anticipates challenges from either carriers or unions.

A Broader Pattern: When Regulations Accumulate, Innovation Slows

Regardless of the reason for the current delay, ATI illustrates a larger systemic problem: regulations written for a very different era create friction and uncertainty for new technologies.

1970s visual-inspection rules force railroads into a waiver-based system. Waivers invite stakeholder disputes. Disputes invite litigation. Litigation encourages procedural caution. And all of this produces delay—even when the evidence for a new technology is positive, as FRA has repeatedly acknowledged.

This is the dynamic of regulatory accumulation: the interaction of many old rules and risk-averse processes that make it difficult for safer technologies to scale.

As I noted in my DC Journal piece:

“New, safer tools are often treated as presumptively risky until proven safe in every imagined scenario.”

ATI isn’t the only example of this dynamic. FRA has oscillated on crew-size mandates, for instance—withdrawn in 2019 because accident data didn’t support it, then revived in 2024 without new evidence—leading to ongoing legal challenges.

A More Predictable Path Forward

Rather than rely on one-off waivers that can be delayed for reasons ranging from political pressure to staff bandwidth constraints, we’d be better off with:

• Performance-based regulation.
If ATI meets or exceeds measurable safety benchmarks, then automated inspection should “count” for regulatory compliance.

• Transparent cost-benefit analysis before imposing or reversing major rules.
That applies to visual-inspection rules and crew-size mandates alike.

• Scalable regulatory sandboxes.
If a technology performs well in a structured pilot, it should automatically advance toward full approval—without needing to restart the process with every new administration.

These approaches would speed the adoption of safety-enhancing tools and reduce the burden on agencies that are already stretched thin.

A Moment for Clarity, Not Criticism

The central question isn’t why an administration is moving slowly, or which stakeholder bloc is influencing the timeline. It’s whether safety policy is being anchored to evidence.

ATI has a strong record. It has bipartisan technical support. And it has the potential to prevent derailments—something everyone agrees is a priority.

Whatever is driving the current delay—union concerns, agency workload, procedural caution, or some mix of all three—greater transparency would help. And clearer, performance-based rules would make future decisions far less dependent on politics, bandwidth, or administrative turnover.

[…] the Nondelegation Project [is] a database linking every part of the Code of Federal Regulations (CFR) to its underlying statutory authority. With this project, the power of AI has transformed an otherwise incomprehensible, more than 190,000-page behemoth spanning 245 volumes into an interactive website accessible to anyone.

The idea is to use AI to understand how many regulations depend upon vague or even overtly unrelated statutes as their statutory authority. The conclusion:

[about] 37% of the over 56,000 delegations captured in the database are general (broad, vague, open-ended) grants of authority to regulatory agencies.

Read the blog post for more, and if you want to know how we used AI, the testing protocol, etc., check out the working paper itself. And the database’s website itself is pretty user-friendly - check it out and if you have any thoughts or suggestions for the next iteration, leave a comment!

As usual, Wayne Crews was quick to post his analysis, which you can read for yourself here. I want to go a bit deeper, at least in some dimensions.

My first question of the agenda is: Can we call it a deregulatory agenda? To answer that, we need to know how much deregulation can we expect over the next 12 months. Although the rhetoric has largely been that the administration will deregulate whenever and wherever possible, the truth is in the details of what rules actually get published. So the purpose of my analysis today is simply to figure out the percentage of rulemakings in the agenda that appear to be deregulatory in nature. The short answer: more than 60 percent of the rulemakings are deregulatory in nature.

The Big Deregulation Has Begun

The Spring 2025 Unified Agenda lists 3,816 rulemakings — up from 3,331 in the Fall 2024 Agenda under Biden. On raw counts alone, this looks like an expansion, not a rollback. However, in the esoteric world of administrative law (I’ve got to be the first person to ever write that phrase), it requires a “rulemaking” to add to or subtract from the regulatory code. Thus the raw count hides the real story.

By my estimation, about 2,404 of these 3,816 rulemakings (roughly 63 percent) are deregulatory in intent. Let me explain how I arrived at that number. That total count of 2,404 deregulatory actions includes:

• 1,001 explicitly identified as deregulatory under Executive Order 14192,

• 701 additional rulemakings with deregulatory titles (terms like rescind, repeal, revoke, remove, delete, abolish, sunset, obsolete), and

• 702 additional rulemakings with abstracts describing the removal or rescission of existing regulations.

By those metrics, the early data point to an unprecedented wave of deregulatory activity.

The Next Frontier: AI-Assisted Deregulation

I suspect that the next unified agenda will have an even higher percentage of deregulatory rulemakings, as well as a higher total volume of rulemakings. There are several reasons that this will likely be the case:

• Traditionally, the spring agenda in a new administration is relatively small compared to the first fall agenda. That’s because it takes considerable time for an administration to get appointees selected, confirmed, and situated, and then those appointees must get their respective agencies to move in the right direction too.

• AI will likely play a greater role in helping the administration identify deregulatory targets. AI will augment the deregulatory efforts by doing background legal and economic research to identify, for example, regulations that are authorized but not required by statute, regulations that depend upon vague statutes that could arguably be illegal in a post-Chevron world, or regulations that overlap or conflict with other regulations. See my Nondelegation Project over at Pacific Legal Foundation for an example of how AI can be used to identify potentially illegal regulations.

In traditional notice-and-comment rulemaking, humans select targets, draft rules, and respond to public comments. AI can now play a support role at every stage — from target identification and proposal drafting to automated response analysis during the comment phase. The more the administration uses AI, the more we can expect it to achieve (just like in the private sector!).

Expect a Deregulatory Tsunami in the Fall 2025 Agenda

Agencies plan to update their regulatory (or deregulatory) agendas for the Fall 2025 Unified Agenda. We may not actually see that published until early 2026, but when it comes out, I expect this initial deregulatory wave—large as it is—look relatively small compared to what the Fall Agenda contains.

But even if only the current plans hold, this could become the largest single-year reduction in the federal regulatory code in modern history.

If you’re reading this, you’re likely well aware that such regulatory accumulation is not benign — creating undesirable effects on economic growth, new business formation, consumer prices, and even income inequality.

What can policymakers do about it? I often talk about “regulatory reform,” as if everyone knows what that means. But I shouldn’t make that assumption. Here’s what I mean by regulatory reform (and I think these apply to any jurisdiction — national, subnational, local — and even to corporations and other organizations that have internal regulations):

Create a mechanism that forces red tape reduction. This requires a transparent and conspicuous commitment by the reforming government. The mechanism can take the shape of a targeted reduction (e.g., “cut regulation by 25%”), regulatory sunsetting, or a 1-in, X-out pay-as-you-go scheme (e.g., Trump I’s 1-in, 2-out or Trump II’s 1-in, 10-out executive orders). For a concrete model, see my testimony on Virginia’s approach to reducing regulatory burden, and the growth results from British Columbia’s regulatory budgeting experiment.

Limit the creation of new regulations to those that make sense. One approach for this is REINS-style legislation, where the legislature actively votes up-or-down on any new regulation that exceeds some pre-determined significance threshold (e.g., any regulation expected to cost more than $1 million or $10 million per year).

Allocate some resources to the production of data and analysis. Call me biased because of my economics background, but I think that cost-benefit analyses can be used to inform regulatory design itself as well as legislators’ votes (if a REINS-style law exists) and to guide red tape reduction efforts. Independence is key here — the people producing the analysis shouldn’t have incentive to cook the books one way or another. At some point, we should use AI to produce these analyses, so long as we have verified that the AI produces good analyses. For the state of play (and how often analysis falls short), see my explainer Regulation 360: Why We Must Reform Our Regulatory System (and How to Do It).

Set up an office to oversee everything. In 2001, when British Columbia set out to cut red tape by 25%, it created a new office called the Ministry of Deregulation (which was subsequently renamed). Virginia set up an Office of Regulatory Management. In short, it needs to be someone’s job to make sure that the regulatory reform doesn’t lose steam when the political winds shift. I lay out this institutional design in Virginia’s Approach to Reducing Regulatory Burden.

And by the way, I really think some state needs to create a position called the Minister of Deregulation. I’d be interested. I can’t imagine a better job title.

Lawmakers have returned to DC and face fiscal deadlines, but an equally urgent task is tackling the staggering stock of outdated regulations. As I note in a recent op-ed posted at National Review Online:

Millions of Americans voted for an America that’s freer to build homes, roads and other infrastructure, for lower prices, and for better jobs. They know regulations are slowing growth, increasing costs, and sapping productivity and wages. Soon, the political calendar flips to next year’s midterm elections. Will they view the administration’s early work [on deregulation] as proof that red tape will be held in check and that an economy of more abundance is on the way? These things take time.

That’s the core demand: clearing the path to abundance by identifying and cutting red tape. And while the deregulatory efforts of the Trump administration are underway, Congress needs to play a role here too.

The logic is straightforward: we budget what we value. If we want growth, we must budget the regulatory state with the same seriousness we bring to fiscal policy. (OK - maybe we aren’t adequately serious about budgeting fiscal policy, but we at least try…). It can’t be left up to the White House alone.

As I recently told the Joint Economic Committee, each year, the unchecked accumulation of rules and regulations robs the average American of an estimated $18,000 and the U.S. of roughly (and conservatively) 0.8 percent in GDP growth. Businesses pay rising compliance costs — estimates range into the trillions annually — and one recent study suggests about 1.3 percent of all labor costs go toward complying with regulations rather than production. The dynamic toll this takes matters more: untold amounts of compounding growth lost to delays and uncertainty.

How should this sort of effect be considered by Congress in its budgetary process? This isn’t about one rule — it’s about an economy gummed up by millions of “shalls” and “musts.” At some point, the negative effects of regulatory accumulation on growth, and the corollary positive effects of regulatory reform, need to make it into Congressional planning and budgeting.

You don’t overcome this macroeconomic headwind without measuring it. The government budgets federal dollars; it barely counts federal mandates. Only a thin slice of rules carries a proper impact analysis, and even these are inconsistent across agencies.

Originally published in National Review on September 12, 2025.

Below, I’ve adapted my recent response to a “question for the record” from Rep. Scott Fitzgerald, who chairs that Subcommittee.

Question from Rep. Fitzgerald: In your testimony, you discussed the costs that regulations impose upon the economy and how the regulatory impact analyses performed by agencies in the rulemaking process often underestimate these costs. Do you think that agencies are doing a thorough enough job to ascertain the true cost of new regulations? If not, what can be done to ensure that agencies provide the public with a comprehensive understanding of the cost of regulations during the rulemaking process?

My response: Regulatory impact analyses (RIAs) and regulatory flexibility analyses (RFAs) have been a feature of the federal regulatory process for several decades. In theory, analyses such as these help regulatory agencies ensure that new regulations solve a significant problem at a reasonable cost. To know whether a regulation solves a significant problem at a reasonable cost, the regulatory agency needs to know whether a significant problem exists, the root cause of the problem, alternative solutions that address the root cause, the effectiveness of each alternative in solving the problem, the benefits to society of each alternative, and the costs to society of each alternative. This is the information that RIAs and RFAs could provide.

Unfortunately, the vast majority of new regulations avoid analysis.[1] Historically, RIAs are produced for less than one percent of all new rules.[2] Similarly, initial RFAs are often not produced when the promulgating agencies certify (often incorrectly) that a new regulation would not affect small entities.[3]

Another misconception is that the analyses accompanying proposed regulations are rational and dispassionate tools that ensure new regulations deliver more benefits than costs and avoid disproportionately impacting small businesses when possible. In reality, RIAs are often of such low quality that their usefulness in informing decisions on what and how to regulate is dubious.[4] Furthermore, because RIAs are typically produced by the same agency that is promulgating a new regulation, the analysis often resembles advocacy for a decision that was made before the analysis was even begun—an attempt to persuade the reader that the new regulation will lead to substantial net benefits, rather than a sober assessment and comparison of the net benefits of a set of different regulatory options, including the option of not regulating.

These flaws surrounding RIAs and RFAs stem from two features of the regulatory process. First, to ensure that RIAs and RFAs are truly dispassionate analyses, we need to reconsider who produces them. While there is value in attaching an economist to the team that is crafting a new regulation within an agency, the risk of such setups is that the economist could be influenced to try to use the analysis as an opportunity to demonstrate that the agency is making the best possible choice. In other words, when an analysis is produced by the agency that is promulgating the new regulation, the result can be mere advocacy, dressed up as scientific analysis.[5]

Better oversight would help minimize this problem. Bills such as the Prove It Act of 2025 (H.R. 1163) would empower a separate set of analysts to assess the quality of some of the analyses related to new rulemakings and give them some authority to force an agency to improve the quality of its analysis if it is insufficient.

Second, there simply are not enough resources dedicated to the production of RIAs. More manpower would help increase the percentage of regulations that are analyzed and could also help increase the quality of RIAs. But as noted above, it may not be ideal to simply allocate more manpower to the agency promulgating a new regulation. An alternative approach could involve setting up a new office designed explicitly to produce RIAs. The Office of Information and Regulatory Affairs (OIRA) within the Office of Management and Budget (OMB) examines agency-produced RIAs, but it is relatively understaffed and does not produce RIAs itself. If another office, housed, for example, within the Council of Economic Advisers, were charged with streamlined production of independent RIAs, there would finally be an element of “competition” inside the executive branch. For years, regulatory economists have complained that there is a monopoly of sorts on the production of RIAs—only the promulgating agency produces the RIAs.[6] Creating a separate group of economists who are independent of the regulators could break this stranglehold on the production of estimates of benefits and costs of new rules. Furthermore, it would have the added benefit of independence. Finally, if analyses are produced independently and relatively early in the rulemaking process, staff in OIRA, relevant congressional committees, and the Small Business Administration would be able to use the analyses in their oversight capacities.

In summary, the concerns about the quality of regulatory analyses are very valid. Independent researchers have shown that the quality of analyses produced by regulatory agencies is often insufficient. Proposals such as the Prove It Act would help increase the quality of analyses by empowering another body with a greater oversight role. To further address the quality issue, as well as to expand the number of proposed rules that are accompanied by high quality analyses, policymakers should consider creating an office, possibly within the Council of Economic Advisers, that is charged with the independent production of RIAs.

[1] Patrick A. McLaughlin and Casey B. Mulligan, “Three Misconceptions about Federal Regulation,” Journal of Benefit-Cost Analysis 13, no. 3 (2022): 287–309, https://doi.org/10.1017/bca.2022.13.

[2] McLaughlin and Mulligan.

[3] Keith W Holman, “The Regulatory Flexibility Act at 25: Is the Law Achieving Its Goal?” Fordham Urban Law Journal 33, no. 4 (2006): 101–18.

[4] Jerry Ellig and Patrick A. McLaughlin, “The Quality and Use of Regulatory Analysis in 2008,” Risk Analysis 32, no. 5 (May 2012): 855–80, https://doi.org/10.1111/j.1539-6924.2011.01715.x.

[5] Christopher Carrigan and Stuart Shapiro, “What’s Wrong with the Back of the Envelope? A Call for Simple (and Timely) Benefit–Cost Analysis,” Regulation & Governance 11, no. 2 (June 2017): 203–12, https://doi.org/10.1111/rego.12120.

[6] Strictly speaking, anyone could produce an RIA. But only the promulgating agency can produce an RIA before announcing a proposed rulemaking. All others must await the publication of the proposed rule before beginning any analysis. Even worse, if they want their analysis to be considered by the promulgating agency, they must then produce any analysis quickly enough to submit it to the agency during the comment period—typically 45 to 60 days.

Regulation typically increases the production costs of goods, and these costs are passed on to the consumer in the form of higher prices. A study published in 2017 (ungated working paper version here) combined data from the Bureau of Labor Statistics, the Bureau of Economic Analysis, and the RegData database to study the relationship between prices and consumer choices.[1] It found that a 10 percent increase in total regulation leads to a nearly 1 percent increase in consumer prices. Furthermore, they found that the effects of these price increases are regressive: The poorest income groups experience the highest proportional increases in the prices they pay. This is consistent with spending patterns broken down by income level. Low-income households tend to spend a greater portion of their incomes on necessities such as utilities, food, and healthcare; unfortunately, these goods also tend to be more regulated than other consumer and household goods.

It is perhaps not surprising, then, that regulatory accumulation also has a positive statistical relationship with poverty rates; as regulation grows, poverty rates also tend to rise (ungated working paper version here).[2] Regulatory accumulation can also contribute to income inequality as wage growth shifts from low-income workers to compliance-related workers such as managers, lawyers, and accountants.[3] Indeed, a recent study in the European Journal of Political Economy found that a higher incidence of federal regulations on a state’s economy causes greater income inequality.[4]

Dustin Chambers and Colin O’Reilly think most of these effects stem from limits to competition.[5] Here’s the abstract to their 2021 article, simply and brilliantly entitled, “The Economic Theory of Regulation and Inequality”:

Stigler (Bell J Econ Manag Sci 2:3–21, 1971) proposed that regulation benefits politically influential interest groups rather than advancing the public interest. The Stiglarian perspective predicts that regulation raises barriers to entry that limit competition and creates economic rents for incumbents. Apart from the direct economic harm of such policies, regulation generates additional consequences. One hypothesized consequence ushered by anticompetitive rules is the widening of income disparities. This article therefore surveys the growing empirical literature that studies whether regulation ultimately exacerbates income inequality. Beginning with the literature on entry and start-up regulation, we find that these rules, as predicted by Stigler, limit entry and dampen entrepreneurship. Moreover, recent studies also indicate that these regulations are associated with higher income inequality. We also review the literature on occupational licensure. Consistent with Stigler (Bell J Econ Manag Sci 2:3–21, 1971), the literature chronicles widespread use of barriers to entry in labor markets, which have documented regressive effects on the distribution of income. Finally, we review research on financial regulation, in which studies have shown that some financial regulations are associated with less entrepreneurship and higher income inequality. Taken together, the recent empirical literature buttresses and extends the implications in Stigler (Bell J Econ Manag Sci 2:3–21, 1971). Regulation tends to benefit incumbents by limiting entry of economic participants, be it firms or workers, and exacerbates income inequality.

That tracks for me, although it doesn’t fully capture some of the second order and third order effects listed above. For example: relatively less competition in a given industry can lead to higher prices - that makes sense. But the fact that regulatory impact tends to be concentrated on those goods that are also necessities (like electricity or phone service) would probably be true even if those industries had perfectly competitive markets.

In summary - I agree with Chambers and O’Reilly that the regulatory accumulation serves to restrict competition for firms as well as in labor markets. Deregulation can therefore help unwind a lot of the regressive effects of regulatory accumulation, by spurring more competition, leading to lower prices and better labor opportunities.

But I also think some of the regressive effect on households derives from the unhappy coincidence that the most regulated goods tend to also be goods that every household has to have. Less regulated goods seem to be more optional (like Netflix and Spotify subscriptions, or the fancier trim on a new automobile). Thus the distribution of spending in household budgets would seem to require some regressivity, even if the prices for necessities like electricity, etc., were to be significantly lowered by increased competition. Deregulation will help alleviate this effect, but as long as electricity production and distribution is relatively more regulated than some of the optional goods that higher income households purchase more than lower income households, I don’t see this effect going away entirely.

[1] Dustin Chambers, Courtney A. Collins, and Alan Krause, “How Do Federal Regulations Affect Consumer Prices? An Analysis of the Regressive Effects of Regulation,” Public Choice 180 (2017): 1–34.

[2] Dustin Chambers, Patrick A. McLaughlin, and Laura Stanley. “Regulation and Poverty: An Empirical Examination of the Relationship Between the Incidence of Federal Regulation and the Occurrence of Poverty Across the US States.” Public Choice 180, no. 1–2 (2019): 131–144.

[3] James B. Bailey, Diana W. Thomas, and Joseph R. Anderson, “Regressive Effects of Regulation on Wages,” Public Choice (2018): 1–13; Dustin Chambers, Patrick A. McLaughlin, and Laura Stanley, “Barriers to Prosperity: The Harmful Impact of Entry Regulations on Income Inequality,” Public Choice 180, no. 1–2 (2019): 165–190; and Sean Mulholland, “Stratification by Regulation: Are Bootleggers and Baptists Skill-Biased?” Public Choice (2018): 1–26.

[4] Sanchari Choudhury, “The Causal Effect of Regulation on Income Inequality Across the US States,” European Journal of Political Economy 80 (2023)

[5] Dustin Chambers and Colin O’Reilly, “The Economic Theory of Regulation and Inequality.” Public Choice (2022): 63–78.

Regulatory accumulation, as I've learned through years of measurement, seems to be the norm. So let me show you what I mean. When I began working in the regulation space, I was trained as an economist. I like numbers. I like data, and I need data to make some kind of diagnosis about whether regulation and the build-up of regulation is indeed problematic. So from the very beginning, I set out to study and produce data on regulation. Let me just show you what I mean by the accumulation of regulation. Here is a video I made about 11 years ago.

That's me when I was a much younger man. What I was trying to do in that video is to visualize regulatory accumulation. These books that I’m stacking are all books of federal regulations in the US. And I started with the regulations that were on the books in 1950. It was about 6 or 7 books, so a small little chunk on this table there is regulation that was in effect in 1950. And then I kept adding and adding and adding books to the pile as time went on.

So you can see how many books there were in 1960, 1970, all the way up to 2013 when I made this video. And by the way, a funny thing happened while we were making the video. This wall of books started to fall on me and the cameraman had to come out and help me hold the books up. And so while there are a lot of companies who will say that they feel crushed by regulation, I'm probably the only person in the world who can say that I was literally almost crushed by regulation. But I survived, but I think the lesson is that it is a very tangible thing. It's a very easy thing to observe in almost every jurisdiction I've looked at, that regulations tend to pile up. They tend to accumulate. And that has consequences: there is no such thing as a free lunch.

One of the things I want to understand as an economist is how expensive is this lunch that we're building up over time? What is the cost of regulatory accumulation? So, I started to generate data. The whole project, which I’ve named RegData, is available online: data for the United States at the federal level, at the state level, data for Canada at the federal level, at the provincial level, data for Australia, some data for the UK. And obviously I can't read all these books, so I had to use computer programs to go through the text of these regulations and measure different dimensions that I could then use in economic modelling to understand, again, the cost of this lunch. So that was a step towards developing a science around regulation. As physicist and Nobel laureate Heike Kamerlingh Onnes once said, through measurement to knowledge. That was the whole purpose of this endeavor. And the knowledge that I came to early on in the process was simple enough: regulation piles up. Regulatory accumulation happens.

This is one slice of the data. If we look at federal regulations in the US over time, starting in 1970, we look at words within the regulations that create prohibitions, statements that mean “you can't do this,” or obligations, or statements that mean, “you have to do this.” And we're trying to find all those words that appear in the text, these hundreds of thousands of pages of text that are in effect in a given year. You can see on the horizontal axis, these are the different years that we have data for. In 1970, there were about 400,000 of these prohibitions and obligations, what we call regulatory restrictions, on the books. And today there's about 1,100,000 regulatory restrictions. And I think the bigger thing than the sheer volume is just the consistency of the growth. You see more or less a very, very consistent slope over time.

And if you think about, this seems inevitable in most modern democracies. As Congress creates more and more regulatory agencies over time (we have roughly 450 in Washington now), those agencies hire people. These people believe that their job is to make more and more regulations, and they do their job. They're human beings. They respond to incentives. And this is the result. You create more agencies, more agencies employ more people, and they make more regulations. That is what the data tell us.

What can we learn from this? What insights can I offer you as an economist, now that I've used the data in my research? The first thing I started to look at is whether we can answer the obvious question of whether the accumulation of regulations affects overall economic growth. I'm an economist, so of course that's the first thing I look at. And in order to understand this question, I needed a model of economic growth. Economists tend to do their analysis with models, and so I worked with a couple of other professors to develop an endogenous growth model. And one feature of an endogenous growth model is that economic growth doesn't happen by chance.

Instead, the agents within the model, individuals and firms, are the drivers of economic growth. They do this by making conscious decisions to invest in research and development, to invest in opening new shops, to invest in buying new machinery for factories, to open new shops. All of these these investment decisions are the ultimate drivers of economic growth in the long run. What my co-authors and I wanted to understand is, “How does the picture I just showed you affect those investment decisions? How does regulation and the accumulation of regulation change what firms do? And then, ultimately, how do these changes in business decisions change the economic growth path of an economy?” The results (journal version here) were pretty dramatic.

This is a picture of two very different economies. Because we had a model, we were able to simulate a world in which there was less regulation compared to the real world in which there was regulatory accumulation. The solid line at the bottom is the actual GDP per capita in America, in the US over time. The dotted line there is the simulated economy.

In this simulated economy, we held regulation constant at the level that we observed in 1980. The experiment here is, if regulation hadn't grown beyond the level it was at in 1980, what would economic growth have been? And that's what this dotted line tells you. In this simulation, if regulations had not accumulated, the economy would have been about 25% bigger by 2012. That's a big number. Break it down into a per capita impact. That's $13,000 per capita in the US, and the total number is about $4,000,000,000,000 ($4 trillion in 2012 dollars). $4 trillion in lost economic growth. To put it another way, that $4 trillion of lost economic growth is equivalent to the fourth largest potential economy in the world in 2012, between Japan and Germany.

The growth effect is clearly massive, especially if you let it accumulate over time because of the compounding of lost growth on lost growth on lost growth. You get to a point where perhaps the disease for the patient becomes worse than the treatment that the regulation provides. If you had asked me to give my prognosis on regulatory policy and the economy about 8 or 9 years ago, when I was just finishing this research, I would have been very pessimistic. I would not have seen any real future for escaping this trap of regulatory accumulation. But then I learned about some innovative approaches to regulatory reform that have been very successful. The first place that I heard about doing this was the Canadian province of British Columbia. As you probably know, British Columbia is the westernmost province of Canada, on the Pacific Ocean, and its largest city is Vancouver.

British Columbia experienced a really bad economy in the nineties. In fact, the people of British Columbia referred to the 1990s as the “dismal decade” because the economic growth was so consistently low over that whole 10-year period. So in 2001, a new government was elected. One of the campaign promises of that government was to increase economic growth, and one of the ways they wanted to do that was to reduce regulation. They recognized that regulation had been piling up for too long and that it was getting in the way of business. So they promised to cut regulation by a third. They got elected. They got into office and realized they didn't really know how to deliver on that promise.

What does it mean to cut regulation by a third? They tried to figure out, well, how do we measure regulation in the first place? Let's set a baseline, and then let's track progress and see if we can actually cut regulation by a third. And that's exactly what they did. They used an army of interns to go through all their regulations. It wasn't as big as the pile of books I showed you earlier, but it was maybe half as big. And in their case, they developed a metric that's very similar to the one I ended up using. These were developed independently, but they counted up the regulatory requirements that are in the text of the regulations, and they came up with a number. It was about 400,000 that were in force in the year 2001. And then they set a target.

That was the next phase of this reform. Their target, as they had promised in their campaign, was to reduce regulation by a third. They also set up a very robust system for updating these figures and tracking progress. And that's an important element of the reform here, which I'll talk about later, that subsequent reformers have noticed and emulated. If you just tell regulators to go and cut regulation, you might get a result, but you probably won't. You have to actively supervise their efforts. And that's because, in a previous life, I worked in a regulatory agency in Washington. And if you work in an agency, you are usually rewarded for making new rules.

That's just the incentive of government employees and most regulatory agencies. For every new regulation I worked on, I got a plaque to hang on my wall. If I worked on a regulation that didn't get finalized and didn't become a rule, then of course there was no plaque. So it's just a response to incentives. So in this case, British Columbia wanted to change the incentives of the rank-and-file employees in these agencies, and that required a lot of effort. It took a lot of active intervention. They went so far as to send people from the regulatory agencies into the companies that they were regulating, to spend a day or two per week in those companies just to talk to them, to understand the pain points that the regulations were creating, to sort of humanize the companies that they're working with as opposed to working against. This whole system was developed on the fly and it was very effective.

They ended up achieving their goal and more in British Columbia within 3 years. By 2004, they had reduced regulations not just by a third, but by 40%.

So I did a study on this. The figure above is from my paper. The bottom line is British Columbia. This is the average level of regulation for all the different industries in the economy for British Columbia over time.

There are some years where you don't have complete data, but it's very clear that British Columbia, according to my computer programs, has reduced regulation substantially. And then the other line is the province of Ontario. Because we have this experiment, economists will call it a natural experiment, where there's one province that goes through this reform and other provinces that don't, we can compare the economic outcomes between the provinces. In contrast to the study I described earlier, we don't need an elaborate model now. I don't have to use an economic model to simulate what happens if regulation doesn't grow. I can look at British Columbia and say what happens if regulation doesn't grow and compare it to other provinces. The result was significant, and that is part of the optimistic part of my message here. British Columbia, by reducing regulation by 40% in three years, caused (I use that word very deliberately) an increase of about 1.1 percentage points to economic growth. That is a massive effect in terms of growth.

With that effect in mind, here's a way of visualizing it. This is a way of looking at the growth path of British Columbia before and after the reform. So all these grey lines are different provinces in Canada. The blue line is British Columbia. The red line is Canada as a whole, the whole country. And this is growth, real growth per capita, indexed to 1981.

And you can see that British Columbia is generally at the bottom. This is in the years of the 1980s to the 1990s. This is the dismal decade that I was talking about earlier. So the growth was bad and now after the reform this is indexed back to the year 2002. How did British Columbia rank first, second or third in terms of growth rate among the provinces in all the subsequent years? It really did. It's very obvious in the data. And when you control for relevant factors, the numbers just jump out. British Columbia experienced a growth boom as a result of this reform.

Now others have taken note. It took a little while for the word to get out in America about what British Columbia had done, but eventually the word got out. I have personally worked with several policymakers, particularly at the state level in the US, to understand what British Columbia has done and to actually implement some similar policies. We are now seeing a wave of regulatory reform spreading across America. I've worked in 10 different states that have implemented some form of regulatory reduction efforts that are at least partly modelled on what British Columbia has done. And generally speaking, this is what the states are doing. They start by just quantifying what you've got. Obtain a baseline and go from there.

And to extend the medical analogy, the first thing you do when you go to the doctor is you step on a scale to see how much you weigh, you strap something on your arm to take your blood pressure, and you get some data. The doctor gets some data, and they want to track those data over time. Because if you suddenly lose weight or suddenly gain weight, that could be an indication of a problem. That's what we're doing in America now. We're starting to quantify, we're getting that first data point of how regulated that state is right now, and then we can start to learn if we're going up or down and what that means. And the other thing I want to mention about this first step in the states is that they're not just counting how much. They're also trying to understand what kind of regulation we have. Specifically, one important detail is that their states tend to look at how much of our regulations come from federal mandates from Washington? How much of our regulation comes from the state legislature? So state legislatures could, with that information, potentially pass a law saying you have to have a certain amount of regulation in a certain area.

How much of that is mandated? And then the third category is how much of a regulation is discretionary? I think the equivalent in European countries would probably be how much is the nation adding over and above what the EU regulations would require—“gold plating,” as they call it. Maybe I'm getting the analogy a little bit wrong, but the states are trying to understand how much leeway they have to cut back on regulations with this last group of this last category that they put them in. And then the rest pretty much follow what British Columbia has done, which is set a target. And those targets vary from state to state. Some states want to cut by a third, some states want to cut by 25%. Some states are just implementing some sort of ongoing one-in, one-out or one-in, two-out regulatory budget. But all of them are understanding the need for hard numbers, for a quantitative baseline, and then tracking progress relative to the vaseline. They're also increasingly finding cheaper ways of doing all this, with AI playing bigger roles over time.

If you're using people to do all of this, it can be a little more expensive operationally. But the state of Ohio, for example, recently contracted with the company Deloitte to use AI and LLMs to go through their regulations and create an account, a baseline, and then track progress with computer programs. And of course I've shown you data that shows it can be done. I've done it myself. And so, increasingly, software is a solution to that, which makes it cheaper to implement. And then finally, we've seen more and more states create new roles. The state of Virginia, for example, has created a new office called the Office of Regulatory Management (ORM). It's staffed by 3 people, 3 FTEs, one of which is sort of procedural, and the other two doing the actual work.

And with those 2 FTEs, Virginia’s ORM is able to monitor all the different regulatory agencies in the state. They interface with the heads of those agencies and ask the heads what you've done to achieve your goals. They have a goal in Virginia to reduce regulations by 25% before the governor's term is up. And as I said, it's about changing the nature of the job of these people who work in the agencies to make them believe that their job is not just to be a rule-maker, but to be a rule-manager, to manage a portfolio of regulations and to optimize that portfolio, not just to add more and more on top of it. Other states have done the same thing. They have different titles for the people who take on these oversight roles within the regulatory agencies, or they keep it in the governor's office. But it's an important thing. I've seen some states that have not been very intentional about creating oversight positions, and the results are not as good in those states that do.

Overall, the results are very good. I mentioned that I measured regulation at the state level. I want to show you a couple of pictures here to perhaps really share my optimism. This is how regulated different states are. This is in the year 2023. A darker state is more regulated. A brighter state is less regulated.

For the most part, the most regulated states are not really surprising. California, New York, Illinois, New Jersey, Texas. Wll, I suppose Texas might be a bit surprising. It's the fifth most regulated state, and it really should be doing a lot better. But on the other hand, Idaho is the least regulated state, followed by South Dakota and North Dakota.

What's more interesting and more optimistic is not to look at a cross-section, but instead look at trends over time. The figure below shows two groups over time: a typical state (status quo state), and a reform state. Typical states are not actively reforming their process and just letting regulations pile up. So that's the 41 other states that haven't made any changes and what their average regulatory growth path looks like. It's just regulatory accumulation, and you already know the growth consequences of that.

The bottom group is the states that have adopted a British Columbia-style regulatory budget process, and you can see that they're starting to make progress. They are reversing the trend. The outcome of regulatory policy doesn’t have to be, at least in the states, a proliferation of regulation. Some states have shown that if you deal with this consciously, you can reverse it and escape the trap of regulatory accumulation and its consequences.

One more dose of optimism. Below is the average GDP growth of the 2 groups of states. The group on the left is the reform group, the states that have changed regulations, changed their regulatory process. The group on the right is the status quo group, the states that have allowed regulatory accumulation to continue. It's not a huge difference. It's about two-tenths of a percentage point. But over time, I expect that difference to grow as the reform states cut more red tape, cut more regulations.

Also, over time, the difference between the two will widen, and you'll see as changes set in for the the overall performance of the economy, the gap between the two groups will, I predict, get bigger. Another reason to be optimistic, at least if you're in one of the reform-minded states. Or countries.

[Adapted from a speech given at the Frankfurt School for German and Global Regulation (FCCR)]

But if you’re in a state (or city, or country) where those reforms are simply infeasible, here are some second-best options. Call it the top five things your jurisdiction can do if a regulatory budget or regulatory sunsetting is not in the cards:

1. Start the process of re-examining old regulations in an area that generally has bipartisan appeal – occupational licensing, for example, or criminal justice. Virginia did this back in 2018/2019. A bipartisan bill, signed by Democratic governor Ralph Northam, required VA’s occupational licensing office and VA’s criminal justice office to review their regulations and identify 25% to cut.

2. Set up a process where old rules are at least re-examined by agencies, for the purpose of at least seeing what is working and what is not. President Obama did this, to some positive effect. Socrates supposedly said, “The unexamined life is not worth living” which, by the way, I don’t ascribe to at all. But I will say that the unexamined regulation may not be worth keeping.

3. Create a better system for examining new regulations. Full blown regulatory impact analysis, including a benefit-cost analysis, is nice if you can do it. But at least get an alternatives analysis. For an excellent guide on setting up a streamlined, low-cost process in a state, see Virginia’s Office of Regulatory Management’s guide here. It makes analysis simple enough that non-economists (or one-armed economists) could easily do it.

4. Permitting transparency. Virginia is the leader in this category too. Mostly by virtue of creating a website that let’s you track the status of a permit application, Virginia Department of Environmental Quality cut permit processing time by 72%. Virginia has since expanded the permit transparency website to all agencies.

5. Change the way that regulators deal with businesses. British Columbia had regulators start spending the occasional working day physically in the offices of small businesses, to understand their pain points better. This kind of approach amounts to trying to change the culture at agencies. Instead of us vs them, it’s us and them. Instead of rule-makers, maybe they can be rule managers.