But an EIG reader asked on X for a more-detailed breakdown of how specific subsectors within the manufacturing sector are doing, and because EIG is very much in the business of fan service, we are happy to oblige.

The short version of what we found is that nearly every manufacturing subsector has lost jobs since the post-COVID peak roughly three years ago.3 But we also discovered a few intriguing exceptions, and each of them tells a useful story about recent trends in the U.S. economy.

Start with Figure 1, which shows net job gains or losses in each subsector since the recent peak in 2023 through the end of 2025, the last available date for which data on subsectors is available.4 To get a sense of the most recent momentum for job growth in each subsector, the chart also shows how they did in 2025 alone.

Four manufacturing subsectors have enjoyed positive net job growth since the most recent peak for overall manufacturing employment. Two of them suffered reversals last year:

• Food, beverage, and tobacco product manufacturing added an impressive 34,800 jobs on net since the peak. But that headline figure obscures a reversal last year, possibly reflecting Chinese retaliatory tariffs against American food products and immigration enforcement pressures on industries that rely heavily on migrant labor.

• Petroleum and coal products manufacturing have eked out a tiny net job gain since the 2023 peak, then gave back ground in 2025, though the 2026 oil price spike may have since staved off that slide.

Jobs in two other subsectors — electrical components and transportation equipment — not only have grown since the peak but also have continued their growth throughout 2025. The rest of this piece will focus on them.

Figure 2 graphs these two subsectors against all other manufacturing subsectors. After jobs all across manufacturing bounced back following the pandemic, only these two sectors have enjoyed consistent and (for now) sustained growth since 2023.

As of the end of last year, electrical components manufacturing employed 9.1 percent more workers than in 2019, while transportation equipment employed 5.8 percent more.

Jobs in the rest of the American manufacturing sector, in contrast, were 2.9 percent below their pre-COVID employment level. The recovery and brief expansion of 2022 and 2023, when the sector climbed above pre-pandemic employment, proved short-lived.

What accounts for the resilience of these two subsectors? We can zoom in even closer to look at the specific industries within each subsector.5 We show the results in Figure 3.

Job growth in the transportation subsector is almost entirely driven by the boom in aerospace.6 In fact, aerospace leads all manufacturing industries in job growth since 2023. Last year it also posted the largest nominal trade surplus of any manufacturing industry, at $116.7 billion.

Driving the growth has been strong commercial and passenger airline demand, including a record-high order backlog. Boeing, the world’s leading airplane manufacturer, returned to profitability and high delivery volumes in 2025 after being plagued by safety scandals, worker strikes, and financial losses the prior year. Demand has also rippled downstream: Plane engine orders have surged, likely pushing engine makers such as GE Aerospace and Evendale to expand their workforces.

Aerospace’s continued growth reflects the United States maintaining its historical edge in high-tech, high value-added manufacturing.

The employment surge in the electrical components subsector tells a different story. It owes its strength to two industries: electrical equipment manufacturing; and batteries, wires, cables, and miscellaneous electrical components.7 8 These are among the industries that bore the brunt of the China Shock and suffered acute decline in the 2000s.9 Green energy and electric vehicles likely contributed to their initial recovery, raising demand for batteries and electrical infrastructure. But their continued growth even after the cancellation of EV subsidies in 2025 points to a different, potent, and entirely unsurprising driver: Artificial Intelligence.

AI’s manufacturing footprint runs the length of the supply chain. Dedicated data centers more than doubled their real consumption of electrical equipment and components from 2019 to 2024.10 Software firms building and deploying AI show an even larger rise, likely propelled by onsite infrastructure upgrades.11 Further downstream, data center buildup has lifted demand from engineering designers and construction contractors, while surging energy requirements have pushed governments to expand electricity generation and transmission.12

In Figure 4 below, we summarize the sources of demand (from other industries and sectors) for these two industries:

Rising AI-related demand has driven up prices, giving manufacturers a strong incentive to expand. The six industries and sectors in the figure above increased their real consumption of electrical components by 15.4 percent from 2019 to 2024, while nominal consumption rose 51.9 percent. That 36.5 percentage point gap far exceeds the roughly 20 percent broad inflation over the same period, pointing to price appreciation specific to these products.

Protectionists may interpret the new electrical manufacturing jobs as a tariff success, given the duties on copper wires and electrical grid equipment that remain in place. They would be mistaken. Imports of electrical equipment and components did not decline in 2025 relative to 2024, and metals tariffs have raised input costs for the very manufacturers supposedly being protected. Their growth owes to AI-driven demand, not trade policy.

The AI supply chain’s impact on manufacturing extends beyond electrical components. Architectural metals and cooling equipment rank among the fastest-growing manufacturing industries since 2023 and sustained gains through 2025, even as their broader subsectors contracted.13 14 15 The top five industries and sectors expanding real purchases of these products are all related to information technology, led by software companies and data centers, which collectively raised their demand for pipes and cooling systems by 8.9 percent in real terms from 2019 to 2024.

AI-led job creation in manufacturing complicates the popular narrative that data centers don’t create many jobs. A Corning-NVIDIA partnership is opening three fiber optic cable factories. Electrical and cooling equipment-makers are benefitting from Virginia’s data center buildup. GE Vernova received more electrical equipment orders from data centers in the first quarter of 2026 than in all of 2025. The spotlight falls on AI labs, chip designers, and computing hubs, but the AI ecosystem reaches the factory floor. Demand for energy infrastructure and climate control apparatuses represents an opportunity for significant industrial growth.

The post-COVID manufacturing boom is fading, and current policy is not arresting the decline. The rapid erosion of job gains in chips and autos is particularly concerning. Tariffs have proven insufficient to reverse the broad-based retreat in manufacturing employment, and the remaining steel and aluminum tariffs continue to impose a cost burden on manufacturers. Even in the industries that are expanding — aerospace and electrical components — the binding constraint is supply struggling to keep up with demand, whether measured by airplane order backlogs, defense production delays, or surging components prices.

A durable industrial renaissance will require cutting input and energy costs and lowering barriers to capital formation, whether the goal is sustaining leadership in aerospace, building out the AI-related manufacturing ecosystem, or stemming job losses elsewhere.

See our GitHub with replication code here.

In May 2026, Dario Amodei, the chief executive of Anthropic, offered a compact way to think about artificial intelligence and work: “If you automate 90 percent of the job, then everyone does the 10 percent of the job.”1

Consider a lawyer who no longer needs to spend hours reviewing routine documents, leaving more time to understand a client’s objectives, test a strategy, and exercise judgment. The service improves and, because the job is done more efficiently, its price falls and demand for it might even grow. Amodei’s point was that the remaining 10 percent of work could well expand to fill the day while workers become far more productive.2 The remark drew attention partly because it sounded more optimistic than Amodei’s earlier warnings about white-collar job loss.3

Of course, something else might happen. Lawyers who have 90 percent of their jobs automated might simply end up with less legal work to do, and thus fewer of them needed to do it.

Amodei’s comment raises real economic questions: Under what conditions does the last ten per cent become a better and more productive job? When does the job merely become smaller and performed by fewer people?

Task-level analysis remains the right place to start for understanding AI exposure, or which of the tasks that make up a job might be automated. Trouble begins when a job itself is treated as a checklist and the exposed tasks are crossed out one by one, appearing to simply shrink the amount of work needed to do the job. But once some tasks disappear, the rest of the job may evolve and change character entirely, with the remaining tasks expanding or new tasks emerging to replace the old.

Tasks differ widely. They may be substitutes, complements, bottlenecks, filters for expertise, objects of worker motivation, or parts of an organisational bundle that firms redraw after automation. How a job changes after some of its tasks are automated and new ones emerge depends largely on the nature of those tasks and, especially, how the tasks relate to each other.

Tasks are a useful starting point

The task-based approach used by economists Daron Acemoglu and Pascual Restrepo in their 2019 paper starts with a useful observation: Technology rarely replaces a job title in a single stroke. Instead, it changes who, or what, performs particular activities within production.4 Their framework separates a displacement effect (capital takes over tasks previously assigned to labour) from a reinstatement effect (innovation generates new tasks in which people have a comparative advantage). Productivity gains may also increase demand for the resulting goods and services.

This approach is more informative than asking whether an entire occupation is “safe” or “at risk.” An accountant, software engineer, teacher, or nurse performs a bundle of activities. A technology may be well-suited to some activities and poorly-suited to others. Studies of large language models have measured occupational exposure by assessing the tasks for which these systems could reduce completion time.5 The resulting measures give us a useful map of technical applicability.

But the map still leaves a difficult step. Once we know where a technology might apply, we need to ask what happens to the human work that remains and its effects on wages and employment. The answer will reflect displacement, reinstatement, and demand. It will also depend on the job’s internal structure.

What happens to expertise?

Economists David Autor and Neil Thompson focus on the type of task that disappears.6 Was it among the more expert parts of the occupation, or among the less expert parts? (Autor and Thompson define expert tasks as “problem-solving and decision-making tasks requiring specialized knowledge and training.”) The same activity can be an expert task in one setting and a routine task in another. Once that activity is automated, a worker who remains in the occupation no longer needs the expertise associated with it.

Suppose software removes routine bookkeeping from an accounting clerk’s day. The remaining activities may involve exceptions, problem-solving, and communication with clients. The occupation has become more demanding. Workers with the relevant expertise may become more valuable, while fewer people qualify for the role. Wages can rise as employment falls.

In another occupation, technology might remove the most demanding diagnostic or administrative activity. The remaining role becomes accessible to a broader pool of workers. Employment may rise as wages fall because expertise is less scarce.

For practical analysis, the warning is clear. Wages and employment can move in opposite directions. A task labelled “routine” in one setting may play a different role elsewhere. Autor and Thompson show how automation can replace experts in one context and complement expertise in another. Their framework asks how automation changes the expertise threshold for the remaining tasks.

When the remaining work matters more

Another mechanism appears when the quality of one task affects the value of the others. Michael Kremer’s O-ring model captured this idea: Production can depend on several essential activities, and a weak link reduces the value of the whole effort.7 In my joint work with Avi Goldfarb, we apply that reasoning to automation when a worker has a fixed amount of time to allocate across essential tasks.8 A manager who spends less time scheduling can devote more attention to a difficult negotiation. A radiologist relieved of routine triage can spend more time on integrative diagnosis. A lawyer freed from document review can devote greater care to strategy.

Here, automation has a second effect. A machine takes over one activity, freeing up time for the remaining activities. If the quality of human work depends on attention, the quality of those residual tasks rises. We call this the focus mechanism. At the level of an occupation, it can look like augmentation even though it begins with the automation of a particular task.

This is one way to read Amodei’s final ten per cent. The tasks left to people may be the bottlenecks that determine service quality. As auxiliary tasks disappear, those bottlenecks can become more valuable.

Demand matters too. Lower costs may attract more customers, while a better service may draw in people who previously went without it. The history of bank tellers after the spread of automatic teller machines is a useful recollection that work reorganisation and demand expansion often arrive together.9

Three cautions follow.

First, average exposure can obscure the job’s architecture. Two occupations with the same average exposure may look quite different once we see whether the exposure is concentrated in a few tasks or spread across the whole bundle.

Second, adoption can be lumpy. Automating one task changes the return to automating another, so firms may adopt a package of tools rather than move through a tidy sequence.

Third, better automation can increase the value of retained labour, even as a human bottleneck remains necessary. If full automation becomes credible, the wage effect is less predictable; the worker’s contribution may become more valuable even as the firm’s outside option improves.

Jobs can be redrawn

Occupations themselves are moving targets. When AI removes part of a job, firms can redesign the human work that survives. Related work on endogenous task bundling studies this organisational margin directly.10 Consider a radiologist whose pre-AI role combines image classification with physician consultation, while quality assurance and model governance are handled by a separate technical role. If AI begins to read images, the old radiologist’s role changes. The hospital has to decide how to recombine the surviving activities.

The hospital might create a broad clinical role that bundles consultation with model governance. Or it might split the work, with a senior clinician focusing on consultation and a governance specialist monitoring the model. The broad role rewards breadth. The specialist arrangement rewards comparative advantage. A skill that carried little weight in the original role can become central after automation, while a skill that anchored the old wage may disappear from the schedule.

For measurement, this matters greatly. A study that holds occupational boundaries fixed can estimate the effect of automation within the old bundle and still miss much of the effect on workers. Where occupations are split into specialist roles, a fixed-bundle estimate may understate the return to comparative advantage. Where tasks are recombined into broader roles, the bias can run in the other direction. Changes in occupational classification belong in the analysis because they record part of the economic response.

What payroll records leave out

Jobs can change even when wage records barely move. Consider a software engineer who uses AI to reduce time spent on documentation, support tickets, and routine debugging, and then devotes additional time to refining a tool they find intrinsically rewarding. Payroll may show little change. The job itself has changed substantially.

Work on enjoyable tasks and contracting shows why this can matter.11 Workers sometimes expand their effort on tasks they value. Two jobs can then have the same hourly wage, paid hours, and wage bill while differing in total time worked and task mix. In some settings, AI reduces recorded hours. In others, it intensifies work, extends the working day or moves effort towards activities beyond formal requirements. Empirical work on AI exposure and the length of the working day points to the importance of this intensive margin.12

Enjoyment is one reason for extra effort. Promotion incentives, deadlines, professional norms, and managerial pressure can produce it as well. The practical point is that productivity gains do not automatically translate into shorter hours or better jobs. Payroll records cannot reveal the full quality of the work experience.

Turning exposure into a forecast

Exposure measures still have a clear use: They tell us where to look. Turning exposure into a forecast requires a second layer of analysis. For any occupation, decision-makers should ask at least six questions.

1. Which tasks are exposed? Are they peripheral activities, expert activities, or essential bottlenecks?

2. What expertise remains? Does automation raise or lower the threshold for entering the occupation?

3. Does freed attention improve quality? Can workers use the time freed up by AI to perform residual tasks more effectively?

4. Will firms redraw the job? Are surviving tasks likely to be rebundled into broad roles or split among specialists?

5. How elastic is demand? Will lower costs and higher quality expand the market enough to support additional employment?

6. What do the available data miss? Are actual hours, work intensity, worker well-being, and changing job boundaries visible?

The answers are likely to vary across occupations. In professional services, medicine, and management, judgment can remain a bottleneck, leaving room for focus effects. In clerical work, displacement may be more direct, although much still depends on whether AI removes expert or inexpert tasks. In software and finance, job boundaries may move quickly as firms separate oversight, client interaction, and technical execution. In delegated creative and professional work, payroll may remain stable even as the texture and intensity of work change.

For researchers, this points towards a fairly concrete data agenda. Occupation-level exposure scores should be paired with task-level time use before and after adoption, quality measures for the remaining tasks, evidence of lumpy implementation, vacancy data showing new specialist roles, and worker-reported measures of intensity and well-being.

No single model will fit every setting. The task is to work out which mechanism matters in each case.

Conclusion

What remains after automation can change in several ways. The final ten per cent may become more valuable, more demanding, easier to enter, harder to perform, or newly divided among specialists. AI may displace tasks and reinstate labour elsewhere. It may remove routine work and increase the value of expertise. It may free scarce attention for human bottlenecks, prompt firms to redraw occupational boundaries, or change the intensity of work in ways that payroll does not capture.

For policymakers and business leaders, a useful discipline is to keep asking questions after an exposure estimate is released. What remains? How do the remaining tasks fit together? Who can perform them? How will firms organise the work? Will demand expand? An exposure measure is a starting point for these questions, not the answer.

While America’s well-off places come in many forms and have many birthdates, when a neighborhood was built turns out to be a pretty good predictor of the level of economic well-being enjoyed by its residents. Generally speaking, our best-off neighborhoods are those with the shortest histories.

Youthful exuberance

Using measures of educational attainment, workforce participation, poverty, incomes, housing vacancies, and employment and business trends, EIG’s Distressed Communities Index (DCI) allows us to compare economic well-being across nearly every zip code in the United States.1 In the DCI’s methodology, a neighborhood’s distress score is equivalent to a percentile based on its average rank across the seven metrics. Each quintile is then assigned a label from “prosperous,” for the top 20 percent, to “distressed,” for the bottom 20 percent.

Plotting distress scores against community ages reveals a distinct relationship between neighborhood newness and prosperity. As a rule, economic well-being runs higher in newer places and lower in older ones.

To approximate a neighborhood’s age, I used the American Community Survey to obtain the year its median residence was built — meaning an equal number of residences in the area are older and an equal number are newer. The line in the graph below shows the average distress scores for zip codes according to those median years, grouped by decade.2

The average zip code built out in the 2010s posts a distress score of only 16.7, firmly in the prosperous quintile. It stands more than 35 points ahead of the average zip code built out in the 1980s, and it enjoys a nearly 45-point lead over the oldest neighborhoods.3

Of course, new places aren’t the only ones that are well-off. There are plenty of prosperous zip codes that were built out in earlier decades. The median home was constructed in the 1970s or 1980s in 51 percent of all zip codes nationwide and in 42 percent of zip codes that score as prosperous on the DCI. Zip codes that combine both old and new housing stock can also register mid-century medians.

But the relationship between the age of homes in a neighborhood and the economic well-being of its residents is strong. The scatterplot below shows the year in which the median home was built for the average zip code in each percentile on the DCI.4 Newer places are more likely to be thriving communities in which the population enjoys high levels of well-being.

Heritage assets

The age of the housing stock offers a proxy for where the country channels its growth and investment. True, Agglomerations readers may be more likely than most to live in older neighborhoods that combine high-levels of well-being with a historical housing stock. But such neighborhoods are the exception. Nationally, for every one prosperous “historic” neighborhood — which I’ll define as a zip code in which the median residence was built in the 1940s or earlier — there are three categorized as distressed.

Stately Massachusetts manages to have more prosperous historic zip codes than distressed ones, although there are more housing units in distressed and historic parts of Fall River, New Bedford, and Springfield than there are in heritage corners of tony Brookline, Marblehead, or Newton. In California, the ratio between prosperous and distressed historic zip codes is balanced, with every Berkeley matched by a Stockton, and every Presidio by a Skid Row.

The picture is very different across much of the rest of the country.

Consider the city of Chicago. Of its historic zip codes, one prosperous spot lies at the heart of downtown and three “comfortable” zips (second quintile on the DCI) line Milwaukee Avenue as it extends northwest out of the city. Covering Wicker Park, Logan Square, and adjacent neighborhoods, these zip codes encompass some of the city’s trendiest enclaves. Yet they are far outnumbered by the 16 deeply distressed zip codes arrayed across the city’s South and West Sides with housing of similar vintages.

Or consider Ohio, where more than half of the state’s nearly one hundred historic zip codes are outright distressed. Only eight are prosperous, and only two of those are urban, covering the Hyde Park and Oakley neighborhoods in Cincinnati. Pennsylvania has 14 distressed historic zip codes for every one prosperous one, and Missouri 18. Florida’s only heritage zip code, just north of downtown Jacksonville, is distressed.

The determining factor, of course, is investment. Some historic neighborhoods have retained their value and received ample investment over the years in upgrades and rehabilitations. But this feat does not come easily for the typical neighborhood as its housing stock fades into obsolescence.

Building the American dream

The other way investment flows into a community is new builds. By definition, an old housing stock is synonymous with a lack of new housing construction. Where new housing gets built follows a distress-prosperity gradient, as people seek out amenities, opportunities, and high levels of well-being. But it also reinforces the gradient, by channeling new investment, new residents, and new incomes into already-growing and well-off places.

Consider two emblematic cities that have approximately the same population (~275k people) but couldn’t be more different: Buffalo, New York, and Gilbert, Arizona. The former is one of the most distressed cities on the DCI and has experienced long-term population decline, while the latter is one of the best-off boomtowns. Only 4 percent of all housing units in Buffalo have been built since 2000, and the median home in 15 of its 17 zip codes was constructed before 1940. In Gilbert, by contrast, 56 percent of all homes have been constructed since 2000.

In some ways, the comparison is unfair. Gilbert is part of the country almost completely unencumbered by history, while Buffalo suffers from the weight of it — saddled with legacies of segregation and deindustrialization. But it also points to a key fact: Unencumbrance makes it easier for new Sun Belt cities to thrive, just as the burdens of the past — which include zoning codes and other administrative accretion — make it harder for their Rust Belt cousins to renew. Housing is not the root cause of Buffalo’s travails, but speak with Rust Belt civic leaders, and a lack of the type of housing people want usually comes up quickly as a key barrier to their regeneration. A quality housing/quality workforce chicken-and-egg problem is one of several obstacles standing between them and renewal.

Economically distressed neighborhoods have less housing overall, and the housing they do have skews older.5 The below graph shows the count of all housing units constructed in each decade across all prosperous and distressed zip codes in the country (about 5,000 zip codes each). Very few housing units have been constructed in distressed communities in the past 15 years, and one-fifth of distressed community housing stock predates 1940. Top quintile communities are skewed in the other direction, with a preponderance of housing units built in more recent decades, and 36 percent since the turn of the millennium.

Where we build thus reinforces geographic inequality. Forty-one percent of all housing units built since 2020 have risen in prosperous communities. Only 8 percent have risen in distressed ones. These figures likely would have been even more skewed away from distressed communities absent the federal Opportunity Zones tax incentive.

The phenomenon does not only operate between cities or metro areas, but within them, too. Across the greater Dallas-Forth Worth metro area, 1.16 million new housing units have been constructed since the turn of the millennium. Of those, 677,000 have risen in prosperous zip codes, compared to just 72,500 in distressed zip codes. Suburban Dallas is home to five of the 10 fastest growing cities in the country, all of which are considered prosperous on the DCI.

Even in the country’s most vibrant growth pole, neighborhood age and well-being are closely linked. The median residence was built in 1974 in the typical distressed zip code in Dallas-Fort Worth, compared to 2001 in the typical prosperous one. Growth has occurred in distressed zip codes — the average one has welcomed nearly 2,100 new housing units since 2000 — but at about a quarter the magnitude of well-off parts of the metro area.

Rehabbing Pleasantville

Housing growth is part of the recipe that sustains high levels of well-being in a community.

Americans have always had a predilection towards the new, and new housing is no exception. New builds command a premium in the market. But when that preference for newness grafts onto a map of already-steep geographic inequality, and widespread variation in where new building is even possible, it becomes a dynamic that reinforces social and economic divides. Where we build impacts where we grow. Where we build affects where opportunity rises. The fact that building is so hard in many corners of the country is part of what contributes to the lack of economic opportunity in them.

Consider the country’s quintessential mid-century city, Los Angeles. Only five zip codes have had at least one-third of their housing stock constructed since 2000. Three cover downtown, representing the sorts of densely urban distressed zip codes that have absorbed growth in recent years. The fourth is Koreatown, an at-risk zip code (fourth quintile) that has densified its commercial corridor. And finally there’s Playa Vista, a master-planned district that arose on a massive old industrial site and was reborn as one of the most prosperous mixed-use corners of the city.

Such second leases on life are rare, even at much smaller scales than Playa Vista. But they do exist: in Highland Park, Denver, where single family neighborhoods abut riverine mixed-use districts; in Over the Rhine, Cincinnati, where reinvestment has revived one of the nation’s finest concentrations of Italianate architecture; in South Bend, Indiana, where new builds along the St. Joseph River are mirrored by new student housing adjacent to the University of Notre Dame; and in Washington, DC, where more housing units have risen along the Southwest Waterfront in the years since 2000 than were constructed in the 1960s.

But for every Georgetown or Beacon Hill, in which prosperity has endured across generations, the fact remains that the highest levels of well-being offered by the U.S. economy are much more typically found in wholly new places, like Playa Vista, Gilbert, or Dallas’s northern exurbs. That our best-off communities tend to have the shortest histories is in some ways a triumph of the suburbs, a uniquely American social innovation. But it also reflects a failure of more mature communities to retain the capacity to grow. Millions of Americans are mired in low-investment, low-opportunity neighborhoods, and many aging neighborhoods find themselves battling deteriorating social and economic conditions, because they are structurally and administratively unable to renew.

The tight connection between place and opportunity that pervades our society at the nation’s semiquincentennial surely would have shocked the nation’s founders. But patterns of growth and investment are not inevitable. Public policy choices, not just private economic calculations, help explain why the U.S. economy is much better at building prosperous communities from scratch than it is at sustaining prosperity in legacy neighborhoods. As the nation turns 250 years old, there’s much it can learn from its youngest places — and the older ones that have managed to rejuvenate.

***

The Distressed Communities Index (DCI) is EIG’s flagship data tool for exploring geographic inequality in the United States. To learn more about the DCI, explore the interactive map, or download the data for your own use, visit: eig.org/dci.

Artificial Intelligence appears destined to reshape the labor market and how firms produce goods and services. Workers, firms, and policymakers will need timely information about how the economy is changing, answering questions such as:

• How many firms use AI? How many workers do?

• How do firms that use AI differ from those that do not?

• What industries are expanding and what industries are contracting?

• What skills and occupations are in high demand, and which are increasingly less valuable?

• What types of workers are switching jobs and which are losing jobs and stuck in unemployment or falling out of the labor force?

• What types of new businesses are being created and how fast are they growing?

High-quality data enables displaced workers to find a new career, students to choose the right skills for the new labor market they will face, and firms to choose which investments to make and which operations to expand or shrink.

The good news is that there is already widespread agreement that something needs to be done.1 Not only that, but much of the statistical infrastructure needed to measure AI’s impact on the economy — surveys that can be expanded, administrative data to combine, human capital within the federal statistical system — already exists.

Policymakers simply need to make some targeted investments and scale what is already working.

Table 1 below summarizes my recommendations. For the full discussion of which measures of AI’s impact are currently available; a summary of findings thus far; and a detailed description of what can be done to improve measurement of AI, see the full report here.

To generate the economic data that policymakers and the public rely on, the government’s statistical agencies use various methods to navigate a delicate balance between precision and the protection of privacy. As EIG research director Nathan Goldschlag explains in a new Agglomerations post, the Commerce Department has now banned the use of one such method, making the production of quality data more difficult.

On June 4, Commerce banned a practice called “noise infusion,” a vital tool that statistical agencies use to provide granularity without violating the privacy of the businesses and individuals from whom it collects information. Noise infusion, Goldschlag explains, is the practice of slightly altering data in order to preserve both the general accuracy of the data and the identities and proprietary information of private-sector actors.

In the example he offers, noise infusion can be used to “fuzz” a business’s employment numbers to preclude their rivals from gaining competitive intel. He presents the hypothetical of a county with a small number of breweries. If the owner of one brewery is able to infer the number of employees its rivals have by a process of elimination, that owner might be able to operationalize an advantage. Noise infusion changes the specifics ever-so-slightly in order to give a useful picture for researchers and policymakers without disclosing too much on any given business.

The motivation behind the ban is likely that Commerce wants to improve the precision of the data that statistical agencies generate. Removing noise infusion, however, will have a different effect. Because the agencies are legally bound to confidentiality, they will have to resort to broader aggregations, such as at the state level, rather than at the county or municipal level. Businesses themselves may also end up withholding a lot more information in their survey responses on the front end. Perhaps worst of all, the statistical agencies will be forced to stop publishing certain data series altogether.

Noise infusion is a vital way to balance between granularity and privacy. Commerce’s blunt ban — “Noise infusion shall not be used for any statistical product” — undermines the ability of the agencies to provide researchers and policymakers with the best data they can. Policymakers should take note and read Nathan’s post.

Policy

John Lettieri, President and CEO of the Economic Innovation Group, was quoted in the New York Times responding to the passage of the 21st Century ROAD to Housing Act, arguing that the bill is a welcome step towards addressing the nation’s entrenched housing shortage — but far from the final step:

“I think the American people are running low on patience broadly on this issue, and that has been true for years...”

Mr. Lettieri said the bill was not “the magic fix-it-all button” on housing “that people would love to see pushed here.” But he said the measure was still a rare and significant achievement that would “show demonstrable progress.”

EIG also issued a statement praising the bill, encouraging Congress to continue playing a central role in helping communities build at the scale the moment demands:

EIG remains committed to advancing research and policy ideas, such as Right to Build Zones, that make housing abundance a national priority and expand access to opportunity. The 21st Century ROAD Act is an important first step, but there is still work to be done.

Tina Lee, Jess Remington, and Adam Ozimek recapped a convening about Right to Build Zones hosted by EIG in March of this year. “The goal of the convening,” they write, “was to strengthen the policy’s structural design with input from leading experts, practitioners and industry.” The guests included several members of EIG’s housing advisory board: Scott Alter, Alex Armlovich, Bobby Fijan, Arpit Gupta, Paul Williams, Emily Hamilton, Michael Novogradac, and Mike Kingsella.

Around the Horn

• John Lettieri was also quoted by Bruce Golding in The Independent about the 21st Century ROAD to Housing Act.

• The Economic Innovation Group is proud to be a partner of this year’s Abundance Conference.

• In Newsweek, Billal Rahman, highlights a new report that examines earnings linked to high-skill immigration using EIG data.

• In The Hill, AEI’s Brent Orell cites work from EIG research director Nathan Goldschlag and former research associate Sarah Eckhardt on AI’s labor-market impact.

Research & Analysis

AI versus the China Shock
Adam Ozimek, Jason Harrison, and Nathan Goldschlag ask: Will AI do to laptop workers what the China Shock did to so many factory workers? “Hoping to avoid the same mistakes of the past,” they write, “researchers and policymakers will be tempted to say yes. But after a closer look at the details of the China Shock, combined with our understanding of workers exposed to AI, we arrive at a different conclusion. The real lessons of the China Shock emerge not from its similarities with the AI Shock, but from its differences. And these lessons should mitigate concerns about a massive upcoming disruption, not exacerbate them.”

AI Shocked workers are likely to be better educated, have higher incomes, and live in more places with high levels of human capital — all of which contribute to far greater resilience than their China Shocked predecessors.

The New Bazaar

• AI Shock vs China Shock: Adam Ozimek and host Cardiff Garcia chat about Adam’s new analysis (see above) contrasting the characteristics of China Shocked with those of workers likely to be affected by the AI Shock. They also discuss how to think about automation, the collection of tasks that make up a job, the resilience of “messy” jobs, and much more.

• A City from Scratch: California Forever has a bold and ambitious plan: to erect an entire new city in California, from scratch, about a one-hour drive north of San Francisco. It has bought about 70,000 acres of land in Solano County, and it was the city to be two things: 1) an urbanist, walkable city of about 400,000-500,000 people, zoned for density instead of suburban sprawl, and located in a place where you would not expect to find it; and 2) a tech manufacturing center, with new factories making actual physical stuff, and a big shipbuilding operation. Cardiff is rooting for it to succeed, but he also had questions. Jan Sramek, Founder and CEO of California Forever, was exactly the guest to answer them.

• Debt and Deficits: Time to Worry?: Martha Gimbel, executive director of The Budget Lab at Yale, returns to the New Bazaar to chat with Cardiff about all things debt and deficits. Martha argues that the fiscal choices of the past decade have already led to higher costs on mortgages, car loans, and small-business debt. She and Cardiff discuss what might happen next given the frightening projections for future borrowing.

EIG Videos of the Month

Ben Glasner comprehensively summarizes the new retirement policy from the Trump administration to expand access to retirement accounts:

Also, did you know that the New Bazaar now also does video episodes? Here is Cardiff Garcia’s chat with Martha Gimbel, available on EIG’s YouTube page, where you’ll find all full episodes of The New Bazaar in addition to shorter clips:

The policy bans the use of something called noise infusion, a necessary tool used by statistical agencies, in particular the Census Bureau, to publish granular data about individuals and businesses without disclosing their identities, thereby protecting their privacy.

The likely reason for the policy change is that the Trump administration wants to improve the precision of certain data series. But the likely result of the policy will be that the government publishes less data, weakening the public’s ability to understand what is happening in their local communities and the American economy.

This consequence would be particularly damaging because individuals, businesses, and policymakers rely on these data to make decisions. Workers use these data to decide whether or where to move, which skills to acquire, and which jobs to seek. Businesses use these data to decide whether and what type of investments to make. And policymakers use these data to design policies and evaluate their effects on constituents. Without the data, they are all flying blind.

I agree with the goal of making data more precise. This just isn’t the way to do it.

Disclosure Avoidance Primer

Disclosure avoidance refers to the set of methods used by the federal stats agencies to protect the privacy of individuals and businesses that contribute to statistical estimates. The agencies are required by law to protect the data of respondents to their surveys.1 In practice, this means they must ensure that it is impossible to back out the information of a specific individual or business from any statistical product.

An example is the County Business Patterns (CBP) data, which provides statistics on business activity by industry and geography. Consider the following three scenarios:

• There is only one brewery in a small county. If the CBP published the exact count of brewery employees in that county, it would be disclosing the information of one business (how many workers it employs), a clear violation of the law.2

• There are two breweries in a small county, and the CBP again publishes the exact count of brewery employees. If I own one of those breweries, I could learn how many employees my competitor has, again violating the law.

• There are more than two breweries in a small county, but the CBP chooses not to publish the total number of brewery employees out of concern that it might compromise the privacy of the businesses. If I’m a prospective brewery owner, I may deem the project too risky to pursue without information about the market I’m entering.

These hypotheticals illustrate a core tradeoff of disclosure avoidance: Granular data is often more useful than aggregate data, but it also risks disclosing private information.

The stats agencies have several tools to manage this tradeoff, including coarse aggregations (releasing less data), cell suppression (removing certain statistics from the data that is released), and noise infusion (fuzzing the data).

It is the third option, noise infusion, that is being banned by the Commerce Department. When I say that noise infusion means “fuzzing the data,” I’m not using a technical term. What I mean is that noise infusion is a method of slightly altering granular data in such a way that the privacy and identity of individual people and businesses are protected in the statistics derived from them. The data are not so much altered, however, that they become unhelpful.

Another example helps to explain how it works. The CBP uses a type of noise infusion known as multiplicative noise, in which the employment of each business in a county is multiplied by a random number, increasing or decreasing it by a small amount. Go back to our hypothetical county, and this time there are 20 total breweries. Noise infusion allows the CBP to publish the total number of brewery workers in that county because it is not based on each brewery’s exact employment. But the data on total brewery employment in the county will be close enough to its actual value for it to be useful to anyone who needs this information to make decisions.

For any statistic that covers a large number of people or businesses, the noise cancels out and the statistic ends up very close to the actual value. For a specific industry in a small county with only five or ten businesses, the statistical and actual value may differ. But the differences tend to be small, and even in those cases it is nonetheless better to have some data rather than none. In the absence of noise infusion, for example, the stats agencies might be forced to only give the count of brewery employees at the state level. This information is far less useful to local policymakers and businesses — or to a prospective employer looking to build a new brewery in a specific county with unique business conditions that may differ from other parts of the state.

The practice of noise infusion became contentious around the 2020 decennial census because of the Census Bureau’s introduction of a new disclosure method called differential privacy, which is another type of noise infusion.

Differential privacy is similar to multiplicative noise infusion in that it adds noise to protect the privacy of people and businesses. But differential privacy is fundamentally different in that it gives the Census Bureau a better understanding (through complicated math) of how changes in the amount of noise added will affect the risk of disclosing too much private information.

The worry on the part of data users was that the Census Bureau would use — or misuse, in their view — differential privacy to err too far in the direction of protecting privacy, thus compromising the precision of the data.

Importantly, as I’ve already explained, differential privacy is not the only type of noise infusion used by the stats agencies. The County Business Patterns, Business Dynamics Statistics, and Quarterly Workforce Indicators have been using the simpler multiplicative noise infusion for decades. But the new policy from the Commerce Department would ban multiplicative noise infusion in addition to differential privacy.

Why I’m Worried

Noise infusion is a vital tool in an agency’s toolkit to release more data than it otherwise would. An excerpt from Abowd, Stephens, and Vilhuber (2005) makes this point nicely for the Quarterly Workforce Indicators (QWI) data:

“Because of the fine detail offered by the published statistics and the confidential nature of the micro-data used to compile the statistics, confidentiality protection is a critical and integral part of the design of the QWI system. Only the application of state-of-the-art protection methods allows the Census Bureau to publish these statistics.”

Noise infusion makes possible the granular tabulations of QWI data, which are used to inform the public about employment, hires, separations, and earnings by industry, geography, and worker characteristics including age, sex, race, and education.

If the Commerce Department policy stands and noise infusion is banned, which disclosure avoidance tools could the agencies use?

The policy states: “Coarsening shall be the preferred category of Disclosure Avoidance methods for all statistical products.”

And: “Suppression shall be permitted as a last resort.”

As a reminder, coarsening means releasing less granular data, and suppression means no longer publishing certain statistics altogether.

And that is why you should be worried. Either the agencies decide it’s actually okay to publish data for the county with only a handful of breweries, or, more likely, they decide that there will simply be no county-level brewery data at all. The data might get rolled up to the state level, which is far less detailed and helpful.

Now what?

Remember again the fundamental tradeoff.

On the one hand, the statistical agencies must adhere to strict legal requirements that ensure the privacy of individuals and businesses. Those privacy guarantees, importantly, encourage survey responses by giving respondents confidence that their data won’t be made public.

On the other hand, the statistical agencies are charged with producing quality data about the nation’s people and economy. Excessive focus on disclosure risks can degrade the quality of data, reducing its value to the individuals, businesses, and policymakers that rely on them.

Outside of clearcut cases like our small county with just one or two breweries, the statistical agencies exercise significant discretion in terms of how strict the disclosure-avoidance protocols should be. In the brewery example, how many breweries does a county need for the data to be considered safe? If one brewery was much bigger than the others, could the smaller ones back out the data of the bigger one?

This ambiguity leaves a lot of room for choice. The agencies can choose to publish as much data as possible given legal constraints, or they can avoid the risk and design rules that reduce the amount of data released. The agencies often choose the latter.

The 2022 proposal to change the Current Population Survey Public Use microdata files is an instructive example. In early 2022, the Census Bureau announced a plan to more than double the population threshold for suppression and round wages. By December of 2022, the Bureau walked back those changes due to pushback from data users, finding alternative approaches less damaging to the usability of the data.

Given this tendency towards limiting data release, which I myself observed over years working within the statistical system,3 it is entirely reasonable for policymakers to rebalance the scales and place greater emphasis on producing more and higher-quality data.

But if the administration’s goal is to pursue that rebalancing, banning a specific disclosure avoidance method is not the way to do it.

Without changing the laws governing privacy protections at the statistical agencies, or promoting a culture and leadership more focused on the quality and volume of data they produce, removing noise infusion as a disclosure avoidance technology will lead to less data overall.

Worse yet, the prospect of less data comes at exactly the wrong moment. There is growing bipartisan agreement that we need significant investments in our statistical system to improve our measurement of AI’s impact on the economy.

I hope to be proven wrong, and that the rollout turns out to be more sensible than I’ve imagined it here. But this new policy has me worried.

The China Shock refers to the widespread job loss that certain workers and communities suffered because of the rapid rise of trade with China in the 2000s.

The AI Shock refers to the same thing happening to white collar workers from the rise of Artificial Intelligence.

For now, the AI Shock is merely a potential shock. But a number of economists, journalists, and other experts have drawn worrying parallels between these two shocks.

We understand why. Across a large body of literature, researchers have found that the China Shock had big and enduring impacts on the local economies and workers who were most exposed to it.

Even more concerning, the extent of the China Shock’s damage was a surprise to economists and other experts. It had long been known that expanding trade will produce winners and losers, but economists tended to think that jobless workers would adjust by moving to the better job opportunities created in a growing economy. Outmigration and flexible wages would restore balance to local labor markets, while many factories would be repurposed to make different products or provide services.

For too many communities, however, especially certain manufacturing towns in the Midwest and South, those adjustments never came. Jobs were lost and not replaced. Jobless people who stayed in their towns simply stopped looking for work. Health, crime, and family formation deteriorated.1

Will AI do to laptop workers what the China Shock did to so many factory workers? Hoping to avoid the same mistakes of the past, researchers and policymakers will be tempted to say yes.

But after a closer look at the details of the China Shock, combined with our understanding of workers exposed to AI, we arrive at a different conclusion.

The real lessons of the China Shock emerge not from its similarities with the AI Shock, but from its differences. And these lessons should mitigate concerns about a massive upcoming disruption, not exacerbate them.

But to understand how we discovered these lessons, we first need to explain how we compared the experiences of China Shocked workers to the potential outcomes for AI Shocked workers. We begin with our measures for each set of workers.

China Shock Exposure

The workers most exposed to the China Shock were highly concentrated in a select few manufacturing industries. In fact, the vast majority of American workers had no exposure at all.

Here is how we know.

A paper published in 2021 by David Autor and his co-authors, building upon earlier work, produced a measure of trade exposure by industry — specifically, the increase in Chinese imports as a share of each industry’s output.

We largely follow this approach. But to arrive at a measure of China Shock exposure for workers, we did two things. First, to measure industry exposure, we chose the time period of 1991 to 2007, a window that captures both the 1990s rise in Chinese import competition and the acceleration after China’s 2001 WTO accession. Second, we adapted the Autor measure — again, the rise of Chinese imports within each industry — by assigning exposure to individual workers based on the industries they worked in.

Our findings:

• For 82 percent of American workers — including everyone outside of manufacturing — there is no measured exposure. They worked in industries in which there was no rise in Chinese imports as a share of their industries’ output between 1991 and 2007.

• Another 14 percent of American workers faced positive but still modest exposure. The rise in Chinese imports was less than 15 percentage points as a share of their industries’ output.

• In contrast, roughly 5 percent of workers were in industries where the growth in Chinese imports was quite large — 15 percentage points or more — relative to output in their respective industries. For example, Chinese imports climbed by 15 points in electrical machinery and tires, 54 points in computers and related equipment, 70 points in toys and sporting goods, and 77 points in footwear. All are manufacturing industries.

It is this last category — the 5 percent most exposed workers — that we use as our comparison group in the rest of this analysis.

In other words, when we compare workers exposed to the China Shock to workers exposed to the AI Shock, we are specifically referring to the 5 percent of American workers who were most exposed to Chinese import competition between 1991 and 2007.

This group numbers about 6 million workers. They all worked in manufacturing, representing about a third of all workers in manufacturing industries. We explore their other characteristics in the sections to come.2

AI Shock Exposure

There is no directly comparable dollar based exposure measure for AI.

Instead, we use the approach in a 2024 paper by Tyna Eloundou and co-authors to identifying the set of workers, based on their occupations, that are most exposed to AI.

Eloundou and co-authors begin by measuring whether the use of AI, potentially with additional specialized software, could reduce the time to complete a task by at least half. To do this they prompt both expert human reviewers and an early version of OpenAI’s ChatGPT-4 to classify tasks based on this time reduction rubric, finding a high degree of agreement between them. For our analysis, we will refer to such tasks as those that can be “done by AI.”

Eloundou and co-authors then aggregate from the task level to the occupation level, computing the share of tasks that can be done by AI.

What is useful about this measure is that it focuses on what could be automated, not what has been automated today. Because the AI shock is mostly yet to come, a forward looking exposure measure is more appropriate.

Another complicating factor is that it is not yet possible to know how much task exposure will lead to actual job loss. For the China Shock, empirical estimates show that the 5 percent most exposed workers were at serious risk of job loss. For the AI shock, however, there is no evidence about the share of a job’s tasks being done by AI that would constitute such risk.

Even if a very high share of an occupation’s tasks can be automated by AI, the occupation itself may survive as a new and different combination of tasks.3 As there is no obvious cutoff for which workers will face displacement (job loss) pressures, we consider multiple cutoffs for the pool of workers that are AI exposed. Table 1 below shows the percentage of the workforce employed in occupations at each respective amount of task exposure to AI.

Our narrowest definition of high exposure assumes that the AI shock only affects the workers at the most extreme end, where AI can replace 90 percent of their tasks. These workers represent less than 1 percent of the workforce.

Our broadest definition assumes the AI shock disrupts a much wider swathe of jobs, including everyone where at least half of their tasks could be done by AI, representing closer to one out of every four workers.

But as our analysis will illustrate, the results of our comparison between China Shocked workers and AI Shocked workers are consistent regardless of where we draw the line.

AI Shocked Workers: More educated, better paid

With highly exposed groups for our two shocks defined, we can explore how they are similar and how they differ. Two differences are especially notable.

1. Compared to China Shocked workers, AI shocked workers are far more educated.

Figure 2 shows that the majority of China Shocked workers have an education level of a high school degree or less, while only 6 percent have an advanced degree.

Using the broadest sample of AI Shocked workers — at least half of their tasks are exposed to AI — the majority of them have a bachelor’s degree or higher. Workers in this group are more likely to have an advanced degree (18 percent) than just a high school diploma or less (17 percent).

And the more concentrated the AI shock, the more educated the group becomes. Workers in the most exposed AI Shock group, representing less than 1 percent of the workforce, are three times more likely to have an advanced degree than to have a high school diploma or less.

In short, regardless of how broad or narrow we define the AI shock, it will affect a group of workers who are substantially more educated than workers who were most exposed to the China Shock. As AI exposure rises, so does educational attainment.

One of the reasons for the disparity is simply that average education rates increase over time, with the China Shock reflecting an early 1990s workforce. Between the early 1990s and early 2020s, the share of workers with a high school degree or less fell from about 48 percent to just over 30 percent. But even if we redo the analysis comparing shocked workers in the same period, the educational gaps persist.4

2. AI Shocked workers also have higher earnings than China Shocked workers.

Regardless of which AI Shock exposure group we use, its share of the top third of all earners is much higher than the China Shock share, while the reverse is true for the respective shares of the bottom third of earners, as shown in Figure 3.

Only about a third of China Shocked workers were in the top tercile (top 33 percent) of overall earners during the time of that shock, whereas 43 percent of the broadest group of AI exposed workers made the top tercile. Meanwhile, when looking at the most exposed group of AI Shocked workers, nearly two thirds of them were in the top tercile.

In contrast to the heavy concentration of AI Shocked workers among the top tercile of earners, China Shocked workers were fairly equally distributed across the wage spectrum, with around a third of workers in each tercile.

Not all workers are equally vulnerable

Why does it matter that AI shocked workers are more educated and have higher earnings? The reason is that education and earnings greatly affect the ability of workers to adapt to job loss.

One piece of evidence supporting this view is that the least educated are more likely to be unemployed. Census data going back over 80 years shows that unemployment for the least educated third of workers is always above, and usually at least double, that of the most educated third.

Another piece of evidence is that even among those workers who lost jobs from the China Shock, not all of them suffered equally.

Difficult adjustments and longlasting effects were concentrated among workers with lower initial wages.5 Low wage manufacturing workers in highly exposed industries lost about 1.2 years of initial earnings over the following 16 years relative to comparable workers in less exposed industries. But middle wage workers lost much less, and the estimated effect for the highest paid third of workers was essentially zero.

High wage workers were able to leave exposed firms and industries without large earnings losses. Lower wage workers, on the other hand, were more likely to remain at their initial employer until the firm underwent mass layoffs, pushing those workers into worse jobs or forcing them to exit the labor force entirely.

The proof that more educated and higher paid workers adapt better to shocks is not just limited to the China Shock research. A similar pattern occurred during the Great Recession. Workers with more education were less likely to lose their jobs initially, and when they got a new one it was less likely to be part time or to pay lower earnings. Looking at the longrun effects of local labor market shocks from the Great Recession, Danny Yagan found that the workers with the highest initial earnings were least likely to be jobless in 2015, long after the recession had officially ended.

The relationship between education and adaptability to local shocks has also been found in the decline of coal mining.

One reason that better educated, higher paid workers are more able to adapt is they are more willing to move. Another reason is that their skills are more transferrable.

Whatever the cause, the evidence on a wide range of economic shocks illustrates that when job loss is concentrated among lower wage and less educated workers, it is likely to generate different adjustment problems than a shock concentrated among workers with higher pay and more college degrees. We consider this an important reason to worry less about the AI shock mirroring the China Shock.

Where the shocks are

The disruptiveness of an economic shock is determined not just by which people get hit but also which places. Research has shown that one reason the China Shock was so disruptive was that it was geographically concentrated.

Local labor markets adapt to small losses of employment all the time. Even outside of recessions, roughly 30 million jobs across the United States are lost every year due to layoffs or business closures. But the national unemployment rate stays low as workers reallocate from shrinking sectors or firms to growing ones. This activity is simply part of the everyday churn of the economy.

When job losses are large and concentrated in a few specific places, in contrast, it becomes much more difficult for those local economies to adapt. Rather than Schumpterian creative destruction, the job loss is large relative to the local economy. The reallocation will take longer, and the struggle to adapt also can generate demand shortfalls as local spending pulls back and spills over into other sectors. Nationally set monetary policy will do little to stabilize local economies suffering from demand problems that arise from concentrated shocks.

If the few million jobs lost from the China Shock had been scattered evenly throughout the country, they would have represented just a small fraction of normal economic churn, and thus workers would have bounced back more easily. Instead, the shock was heavily concentrated in manufacturing towns in the Rust Belt and South.

The good news is, the evidence we have suggests that the AI shock will be far less geographically concentrated. Figures 5 and 6 below show the respective place level exposures to the China Shock and the AI shock across Commuting Zones (CZs). China Shock exposure is concentrated in a much narrower set of manufacturing heavy labor markets. AI exposure, in contrast, is more evenly dispersed throughout the country.

The difference is also obvious when looking at relative exposure in the hardest hit places, as shown in Figure 7. The 10 Commuting Zones most exposed to the China Shock average about 5.8 times the exposure of the average one; the next 40 about 3.2 times the exposure of the average one; and the next 50 about 2.1 times.

AI exposure is much flatter, with the top 10 CZs only about 1.2 times the average CZ.

Which types of places

How a place adapts to a shock also depends on the characteristics of the place.

Research shows that local economies whose workers have higher average levels of education adapt much more easily to economic shocks than places with lesser educated workers. The reasons for their resilience go beyond the fact that higher educated workers themselves are more resilient. Places with more human capital tend to have faster rising populations, which helps to boost dynamism and the growth of underlying demand growth. A more educated population will also be more innovative and have more entrepreneurs who identify next best uses for displaced capital and labor.

When we compare China Shocked places to AI shocked places, it is clear that education levels are a major differentiator. Figure 8 below looks again at exposure ranked Commuting Zones, but it compares the average share of workers with a high school degree or less against the average share of those workers for all Commuting Zones in the relevant year.6 (We use 1990 education data for the China Shock and 2024 education data for the AI Shock.)

The most China Shocked places had populations with lower than average education levels at the time, while the opposite is true for the most AI exposed.

Looking at the share of the population with bachelor’s degrees or higher, we see a complementary pattern. China Shocked places are slightly below average, while AI shocked places are way above average.

Missing Something?

We have presented the evidence for why the China Shock was far worse than the AI Shock is likely to be. What might the opposite case look like? We’re happy to play devil’s advocate, both because it’s a good intellectual habit and also because the case ends up looking so much weaker than ours.

One possibility is that the AI shock materializes much faster than the China Shock. If AI agents prove to be as capable as their creators suggest, displacing a worker whose primary tasks occur on a laptop could be as simple as installing and initializing the AI agent.

A second possibility is that the AI Shock ends up displacing a much bigger share of workers than we expect. In our most aggressive AI Shock scenario, 27 percent of workers would be exposed, 5.7 times larger than the 5 percent exposed to the China Shock. If most or all of these exposed workers lose their jobs — especially if, coinciding with the first possibility, they lose their jobs quickly — an utterly massive share of the labor force will find itself suddenly unemployed. The job losses and subsequent demand destruction could be enormous.

That’s the worst case scenario: AI hits faster and hits bigger than even pessimistic forecasts. We find this outcome unlikely, for two reasons.

First, rolling out AI at such scale requires significant computing hardware, which takes time to build. Data center construction is a real constraint and is already causing policy blowback. The prices of information processing equipment have also surged, rising at the fastest pace since 1959. The current AI buildout is already facing big obstacles. Imagine how much bigger they would be for the kind of buildout required to replace millions of workers.

Second, if the AI Shock were to produce vast job losses, the magnitude of the effect itself makes economic policy more powerful in offsetting the effects on demand. Recall that monetary policy, for instance, was of limited use for helping local economies destroyed by the concentrated China Shock. If the AI Shock proves to be bigger and more widespread, monetary policy regains in potency.

What is more, significant losses of jobs because of AI would be accompanied by rapid economic growth. Both economic growth and AI augmentation are helpful for smoothing the economic adjustment in a variety of ways, including boosting the demand for other types of workers.

Nevertheless, we admit that an AI Shock large enough to upend the labor market overnight would be unprecedented, which means that reasoning about it involves uncertainty in both positive and negative directions — about adjustment costs but also benefits, layoffs but also progress, unemployment but also innovation. The economy would be entering uncharted territory.

In such a world, comparisons to the China Shock, or to any other great historical shock, would therefore be pointless.

A tale of two shocks

It is true that the impacts of the China Shock were more negative than many economists expected. But it is also the case that those negative effects are no longer a mystery. Thanks to a growing body of research, quite a bit is known about which types of people and places were more affected than others.

In sum, the China Shock was disruptive because it 1) disproportionately hurt workers with less education, 2) was geographically concentrated, and 3) inflicted its worst damage on places with lower levels of human capital.

It is still early days for AI and so a large dose of humility is appropriate. But so far as we can tell, the AI Shock is aimed at the kinds of people and places that actually weathered the China Shock just fine.

The old cliché is that history does not repeat itself but does often rhyme. If the AI Shock turns out to be as bad as the China Shock despite such different underlying characteristics, it would turn the cliché on its head: an instance in which history did repeat itself but skipped the rhyming. It’s not impossible, but we find this outcome unlikely.

Appendix

In the preceding analysis, we compare the education levels of the workers most exposed to the China Shock to those who are most exposed to the AI shock. One complicating factor is that these two shocks are occurring at different points in time, and education levels among workers overall climbed.

Absolute increases in education over time are relevant for resilience, but we can also compare relative education rates by comparing each group of exposed workers to the average education rates of the overall workforce from the same time period. In other words, we can compare the China Shocked workers to the average education of workers between 1991 and 1993. And we can compare the education of AI Shocked workers to average education levels of all workers between 2021 and 2022.

The results are directionally the same: Workers exposed to the China Shock are still less educated, while the Al exposed groups are more educated.

Specifically, compared to the overall 1991 to 1993 workforce, workers exposed to the China Shock are 10.4 percentage points more likely to be high school graduates or less and 5.9 percentage points less likely to have a bachelor’s degree or higher.

In contrast, the most AI Shock exposed groups are substantially more educated than the 2021 to 2022 workforce. Depending on the exposure group, they are 13.7 to 24.1 percentage points less likely to have a high school diploma or less, and 13.1 to 29.9 percentage points more likely to have bachelor’s degrees or higher.

Taking the z scores for each group shows consistent results, with China Shock exposed workers below their period average and every exposure level of AI shock above its period average.

While overall national jobs numbers continue to inch upward, more granular data at the state and sectoral levels reveal a startling trend across the country: healthcare jobs in many states are booming, but most other sectors are stagnant.

“And no state appears as reliant on the healthcare and social assistance sector to support its labor market,” write EIG senior fellow Kenan Fikri and research associate Thomas Cronin, “as the nation’s most populous and the world’s fourth largest economy: California.”

New research from Fikri and Cronin finds that since March 2022 California has increased employment in these fields by a massive 25.3 percent, but has seen negative private-sector jobs growth otherwise.

California stands out from the pack, but it is not the only state where healthcare is masking labor market weakness overall. Georgia, New York, Oregon, New Jersey, Maryland, and Colorado all have experienced healthcare job growth exceeding 17 percent while employment in all other sectors in these states has expanded by 3 percent or less.

Nationally, jobs in healthcare and social assistance increased by a strong 16.3 percent, compared to just 2.9 percent in the non-health segments of the economy.

Amplifying EIG’s publishing of this research at Agglomerations, the Wall Street Journal printed a story in its Thursday May 14 edition anchored by these stunning findings. “EIG analyzed a three-year span of the most recent Labor Department data to cover a mixed period for the labor market, which saw a dramatic cooling off after a postpandemic hiring boom,” Harriet Torry and Josh Ulick conveyed to Journal readers.

While healthcare and social assistance jobs are not a bad thing to add to a state’s employment mix, the wages associated with them tend to be lower than in fields in which these states have suffered losses. Moreover, California, Oregon, and other states may see the healthcare sector struggle going forward with the dissipation of federal Medicaid spending.

“If this federal support starts to dry up,” Fikri told the Journal, “we’re going to soon discover whether the healthcare boom has legs sufficient to stand independently.”

Policy

• EIG submitted a comment letter arguing that the Department of Labor’s proposal to raise prevailing wages for H-1Bs and green card applicants is based on flawed premises. EIG found that the proposal fails to account for differences in worker education, experience, and qualifications when estimating prevailing wages. The letter also argued that the proposal would continue to advantage large outsourcing firms and fail to select for top talent. EIG urges the Department of Labor to adopt the alternative proposal, known as Experience Benchmarking, to correct these flaws. EIG also noted, however, that fixing flaws within the H-1B program is a suboptimal solution overall and recommends market-based immigration reforms, such as selecting H-1Bs with the highest salary offers.

• EIG issued a statement in response to the House passage of the 21st Century ROAD to Housing Act. John Lettieri, President and CEO of the Economic Innovation Group. “America cannot lower housing costs without building substantially more homes of all types,” said John Lettieri, EIG President and CEO. “This legislation takes meaningful steps toward that goal, and Congress should move quickly to get it across the finish line.”

Research & Analysis

Smells like Teen Unemployment
The overall labor market appears reasonably healthy, but real weakness has seeped in and virtually all of it is concentrated in the young. Employment has fallen not just for recent college grads but also for noncollege young adults and for teenagers, making it unlikely that the cause of these problems is AI replacing worker tasks. Something else is going on. But what?

How Students and Recent Grads are Responding to AI
Sarah Eckhardt and Nathan Goldschlag find that undergraduates “are flocking towards the most-AI-exposed degrees, with enrollment in those degrees up 8 percent last year compared to 2017. This trend holds despite a notable decline in Computer Science degrees, one of the most-AI-exposed degrees, but whose decline is more than offset by increases in other exposed degrees like Engineering.” One reason why is that the jobs these students are likely to be offered still pay very high wages.

The Problem of Partial Noncompete Bans
Partial noncompete bans can be net positives if they are carefully written and apply to a wide swath of workers, but too often the exemptions — some of them quite absurd — lead to mass confusion, chicanery on the parts of employers, and unenforceability. The evidence is clear, write Thomas Cronin and Ben Glasner: “Banning noncompetes for certain job types or income levels helps curb some abuses, but it leaves intact many of the mechanisms by which noncompete agreements hurt most workers and stifle innovation and entrepreneurship. The best way to achieve a stronger, fairer, and more dynamic economy is for states to ban noncompetes entirely.”

Around the Horn

• For the National Bureau of Economic Research, Nathan Goldschlag and coauthors have issued a working paper examining the relationship between patents and economic growth.

• Kyla Scanlon compiles ways to respond to the rise of AI, including “Be AI native, not AI proof” citing the post on students and young graduates from Sarah Eckhardt and Nathan Goldschlag.

• In City Journal, Jordan McGillis explains the fraught labor dynamics hindering autonomous trucking in California.

• In National Review, McGillis argues against a federal subsidy for an Alaskan pipeline.

• In The Dispatch, McGillis applies EIG’s Distressed Communities Index to family housing decisions.

• And also in National Review, McGillis has published a longform reported essay on the U.S. Geological Survey’s Earth Mapping Resources Initiative.

• In Newsweek, Billal Rahman quotes Sam Peak on increased scrutiny of the H-1B program.

• In an essay for the Washington Post, the University of Notre Dame’s Alexander Kustov highlights EIG’s survey work showing public support for high-skill immigration.

• Also in the Washington Post, reporters Shira Ovide and Rachel Lerman drew input from Nathan Goldschlag in their coverage of AI’s effects on the labor market.

• Coverage of the Trump administration’s new retirement plan continues, with reporters citing Ben Glasner and Sarah Eckhardt’s pivotal research on the issue in USA Today and Forbes.

EIG Chart of the Month

Via Kenan Fikri’s post on the dominance of health care jobs in the California labor market:

There is another data point making me skeptical that AI is behind the labor market woes of young people: The unemployment rate for teenagers (ages 15–18) has risen more than for any other category of young workers. From its post-pandemic low, teenage unemployment has climbed 4 percentage points, compared to the 1.5 percentage point increase for recent college graduates.

Is it plausible that the cause of these trends is primarily AI? In other words, how likely is it that AI is the main cause of employment struggles not just for recent college graduates, but also noncollege youths and even teenagers?

What do teenagers do?

When looking at the kinds of entry-level, white-collar “laptop” jobs that recent college graduates tend to do, it is natural to wonder if AI is playing a role in the lack of hiring. The same is not true when looking at the kinds of work that teenagers do.

The industries that hire teenagers are simply not the ones that AI is likely to disrupt by displacing a large share of workers. Of the teens who were employed from 2022 through 2024, for example, a combined 75 percent of them worked in Accommodation and Food Services; Retail Trade; Arts, Entertainment, and Recreation; and Health Care and Social Assistance. ChatGPT can do a lot of things, but it can’t help if you want fries with that.

Measures of AI exposure at the occupation level tell a similar story. According to three separate measures of occupational AI exposure from influential research papers, only a small share of teens — between 3.8 and 8 percent, depending on the measure — are in the most AI-exposed occupations.

To whatever extent the struggles of teenage workers have the same cause as the employment struggles of other young adults, that cause is almost certainly AI.

Side note: Are the kids alright though?

Closer scrutiny of the teen labor market reveals an interesting wrinkle in the story. By one key metric, conditions are simply back to what they were before COVID.

The employment rate for teenagers, which is the share of all teens who are employed, started surging in 2021 as the pandemic recovery took hold and climbed far above its pre-pandemic high.1 It finally peaked in early 2023 before subsequently falling to where it is now, roughly consistent with its level from before the pandemic.

Put another way, the labor market for teens has indeed weakened, but only as a reversion to a historically normal level, and a pretty strong one at that.

Contrast the teen story with employment for young adults. The employment rate for both college and noncollege young adults (ages 22–25) barely recovered to its pre-pandemic level before stagnating and then falling again in recent years.

Why teen employment rose to greater heights during the pandemic than young-adult employment is itself an interesting question. Part of the story likely has to do with post-pandemic policies, which by design made it easier to not work for people who benefited from them — a group that largely excluded teens. Stimulus checks went to their parents. Teens mostly never qualified for expanded unemployment insurance. And because most teens live with their parents, they get nothing from an eviction moratorium. Nor do people who haven’t been to college yet benefit from pauses in student loan payments.

Still, the relative resilience of the teen labor market is consistent with the fact that it has deteriorated in recent years, just as the labor markets for recent college graduates and noncollege young adults have deteriorated. Something is damaging the labor market for young people. If, as I have already argued, that something is unlikely to be AI, then what is it?

Low Hire, Low Fire

I have to admit that current labor market conditions are kind of weird. This is not a normal business cycle.

In addition to the struggles of young workers, another sign of weakness for the overall labor market is the slowdown in real and nominal wage growth. The quits rate, which tends to climb during strong labor markets as workers leave their jobs for better opportunities, is down from its post-pandemic highs. And interest rates are probably still above neutral, meaning that the Fed still has some pressure on the brakes of the economy as it tries to fight inflation.

On the other hand, the prime employment rate, which is the share of all adults aged 25–54 who are employed, has moved down only slightly from its peak. Total unemployment has inched up, but only a bit. Taken by themselves, and in the context of previous historical periods, these two measures would suggest a healthy environment for jobs.

One way to summarize this bizarre labor market is that real weakness has crept into it, but it seems concentrated almost entirely on the youth. Just why the youth are bearing the brunt is unclear, an ongoing economic mystery.

For teenagers, one possible contributing factor is that because their employment rate climbed so high after the pandemic, it thus had the farthest to fall. But that point doesn’t resolve why youth of all kinds, including groups that didn’t see a post-pandemic surge, have also experienced these deteriorating conditions. Teenagers are hardly the only canary in this coalmine.

Time to Zoom Out

Somewhere between the economic measures showing weakness and those showing strength is the middling reality of the low hire and low fire economy, in which employers are not laying off workers but neither are they hiring many of them. The lack of visibility into the economic future has left businesses paralyzed by a generalized uncertainty.

There are plenty of good reasons for it: tariffs that move up and down violently, wars abroad and the resulting shocks to energy prices, interest rates that might go up or down, high levels of deportations, and — yes, it’s fair to include — the unknowable effects of AI on the economy and society.

But while AI is a possible cause of uncertainty, one of several, it’s important to recognize that employers not hiring because they are frozen by AI uncertainty is not the same thing as AI replacing the tasks of young workers. Those are two different stories.

Economists and researchers and journalists are mostly focused on the story of task displacement. They are constantly trying to assess how real it is, whether it has started, how far it will go, which jobs are vulnerable to it, and so on.

I think it’s time to zoom out and instead look at AI uncertainty and also the many other types of possible macro uncertainty the economy now confronts. Given the mystery surrounding youth labor-market weakness, it would be helpful to widen our scrutiny beyond the question of which tasks performed by recent college grads might someday be taken by AI, and instead investigate the strange macroeconomy we are now living through.

See our GitHub with replication code here.

Unless, that is, you’re a babysitter.

Washington, DC bans noncompete agreements for workers earning below roughly $160,000 a year. But the ban expressly excludes “casual babysitters,” meaning that a teenager in high school who occasionally watches their neighbor’s kids can be legally barred from babysitting for anyone else.

While defenders of noncompetes argue that they act as protection for trade secrets, client relationships, and training investments, none of this applies to casual babysitters. They are both highly replaceable and unlikely to know anything damaging to their employers.

The absurdity of DC’s babysitter exception is emblematic of what noncompete policy often looks like when states opt for a partial ban: divisions by income, occupation, or industry that tend to be arbitrary, confusing, and difficult to enforce.

Truly Un-exception-al

Businesses take advantage of this confusion by applying noncompete agreements far more broadly than is necessary to protect their proprietary information. News stories and multiple studies and surveys have uncovered countless examples of employment restrictions on workers whose job mobility poses no special threat to their employers: sandwich makers, house cleaners, temporary warehouse workers, and more.1

By one 2014 estimate, 38 percent of workers have worked under a noncompete at some point in their lives. This despite the fact that other enforcement mechanisms, such as nondisclosure agreements, already exist to protect business interests and are much less harmful to employees and the economy as a whole.2 3

Some states are beginning to recognize the problems with partial noncompete bans. Washington State, for example, first adopted a noncompete ban only for workers with annual earnings below $100,000 in 2020. But just six years later, it passed a new law replacing that partial ban with a full retrospective noncompete ban for all workers,4 extending protections to the remaining 15 percent of the state’s workforce and 43 percent of the state’s total earnings.5 (Retrospective noncompete bans are those that invalidate existing bans in addition to outlawing future ones.)

Washington now joins California, Minnesota, Oklahoma, and North Dakota in fully banning noncompetes.

Unfortunately, workers in the other 45 states and DC are still reliant on incomplete and uneven protections. Large parts of the workforce remain vulnerable to contracts that suppress mobility and discourage entrepreneurship.

The story of Epic, a megalithic electronic health records (EHR) company, offers a particularly stark example of the rationale in favor of universal bans, with no carveouts, on noncompete agreements. Based in Wisconsin, which only limits how long noncompetes can last, Epic requires its employees to sign a noncompete that prevents them from working for any of the roughly 4,500 companies across the United States on its “blacklist” that either compete with Epic or are connected to its customers and partners. This noncompete effectively forces workers leaving the company to exit the healthcare industry entirely. Former Epic employees can’t even create their own healthcare start-up or business for the duration of their noncompete clause.

Leaving workers with such narrow options effectively negates their power to negotiate with their employer. Noncompetes thus can suppress wages, restrict the flow of talent and ideas, and limit new business formation.

Job-Type Noncompete Bans

By far the most common type of partial noncompete ban is based on job type. These laws selectively protect workers in specific industries or with particular job titles or legal statuses — such as physicians, healthcare workers, or employees paid by the hour — on the theory that noncompetes for these types of workers are especially likely to harm them or the wider economy. All other workers are still subject to noncompetes from their employers.

Noncompete bans that focus on specific job types are often so comically narrow that they may as well not exist at all. Kentucky and Iowa, for example, ban noncompete agreements only for nurses, home health aides, and other health staff who contract with healthcare employment agencies instead of working for hospitals directly. Connecticut has a noncompete ban for workers specifically in non-cable broadcasting companies. And in Vermont, noncompetes are banned only between barbering and cosmetology students and their universities. Predictably, less than 0.1 percent of each state’s workforce is covered by each of these provisions.6

Even some of the broadest noncompete bans by job type end up covering very few workers. Nevada’s noncompete ban for all hourly workers, for example, covers only about 10 percent of its workforce.7

Job-type noncompete bans are also the easiest bans for firms to circumvent. A mere title or classification change can be the difference between whether or not a noncompete is enforceable for a worker doing otherwise similar tasks. Under Nevada’s noncompete ban for hourly workers, for example, a retail worker paid hourly may be protected, while a salaried worker doing nearly identical work is not. These laws simply fail to be a meaningful impediment to firms imposing noncompete agreements on their employees.

And in edge cases where the legal enforceability of a worker’s noncompete agreement is unclear, many workers will assume their noncompetes are binding. Studies show that even clearly unenforceable noncompete agreements significantly reduce workers’ job mobility behavior and new employers’ willingness to hire them.

Finally, noncompete bans by job type can distort the labor market by artificially making some jobs more appealing to workers than others. When only some jobs carry noncompetes, workers may avoid otherwise better matches in favor of less restrictive roles, leading talent to flow toward jobs with fewer contractual barriers rather than where it would be most productive. This misallocation will ultimately reduce both job match quality and overall economic efficiency.

Income-Based Bans

Income-based noncompete bans forbid noncompetes for all workers earning below a certain income level. One upside of these bans is that they are much simpler, making it apparent when a noncompete is enforceable and when it is not. Another is that they tend to cover much wider swaths of the workforce, depending on the level of the income cutoff.

While most job-type bans apply to less than 5 percent of their state’s workforce, New Hampshire’s noncompete ban for workers earning less than 200 percent of the state’s minimum wage — about $29,000 per year in 2024 — covers more than 25 percent of the state’s workers. And the income-based noncompete thresholds are typically much higher in the other states that have them. Oregon, for example, bans noncompete agreements for those earning below about $116,000 per year in 2024 dollars, protecting about 85 percent of its workforce.8

Higher earners, however, are by definition left exposed by income-based noncompete bans. These workers are typically the most specialized and therefore the most likely to engage in the kinds of productive activities, like starting businesses and transferring skills across firms, that are restricted by noncompetes.

Narrow specialization also makes it substantially more difficult to find a new job that doesn’t violate the terms of their noncompete without leaving the entire industry, so the harm of a noncompete agreement to high-income workers themselves can be especially pronounced.

High earners furthermore account for a disproportionately large share of total income, which means that income-based noncompete bans protect a much smaller share of total earnings than of workers. New Hampshire’s ban on noncompetes for workers earning below twice the minimum wage, for example, covers 25 percent of its workforce but only 6 percent of the state’s total income. So failing to ban noncompetes for these workers leaves a great deal of economic activity vulnerable to restriction.

Income-based noncompete bans are an effective way to protect lower-wage workers from abusive business practices, but states still sacrifice a great deal of productive activity by allowing noncompetes for high earners.

Just Go All the Way

We recognize that political reality differs from state to state. In some places, given the absence of federal action, a full ban may not yet be possible. Advocates of a universal ban on noncompetes sometimes need to make concessions to protect at least some workers. They might acknowledge that a partial noncompete ban has risks but nonetheless consider it better than no ban at all.

Consider a few recent examples of states moving towards a partial ban:

• Utah passed out of committee a noncompete ban for non-exempt workers, students, contractors, and employees earning below $155,000 per year;

• Virginia’s state legislature approved new protections for licensed medical practitioners and discharged employees;

• New York state is debating a noncompete ban for workers earning less than $500,000 per year.

Partial bans, especially when applied clearly and to a wide swath of the labor force, can still be large net positives for workers and the economy, reducing barriers to mobility and improving labor market dynamism. They are especially worthwhile if they prove to be an incremental step on the way towards a complete ban, as was the case for Washington.

But we also encourage advocates of partial bans to remember the evidence: Banning noncompetes for certain job types or income levels helps curb some abuses, but it leaves intact many of the mechanisms by which noncompete agreements hurt most workers and stifle innovation and entrepreneurship.

The best way to achieve a stronger, fairer, and more dynamic economy is for states to ban noncompetes entirely.

See our GitHub with replication code here.

Not so. Healthcare job growth varies dramatically by state, as does the extent to which healthcare contributes to total job growth. And no state appears as reliant on the healthcare and social assistance sector to support its labor market as the nation’s most populous and the world’s fourth largest economy: California.

Another take on a “health-full” labor market

The total number of jobs in the United States increased by 4.7 percent from March 2022 to March 2026, but the differences across sectors were massive.1 Jobs in healthcare and social assistance increased by a whopping 16.3 percent, compared to just 2.9 percent in the non-health segments of the economy.2 Artificial intelligence (AI) may dominate financial markets, but healthcare still reigns supreme in the labor market.

Scan the figures in the charts below, and it becomes apparent just how geographically disparate job growth has been over the past four years even in this quintessentially local sector.

California outpaces every other state with its 25.3 percent job growth in healthcare and social assistance from March 2022 to 2026 — and that growth wasn’t being powered by a quickly rising population. The number of California residents increased by less than 1 percent between 2022 and 2025.3

But what really distinguishes California is how it combines such massive growth in healthcare with such little growth in other parts of the economy.

California eked out total job growth of a mere 3.4 percent over the past four years, the bulk of which was racked up at the beginning of the period (see the top figure). Take out healthcare and social assistance, and employment across all other sectors in the state declined by 0.3 percent. Over the past four years, employment in the non-care portions of the economy shrunk in only three other states, led by DOGE-walloped DC.

The growth gap between the health and non-health parts of the economy (the length of the bars in the above graph) is greater in California than in any other state.

Healthcare now accounts for 17.4 percent of all jobs in California, up three percentage points in four years. That is twice the rate at which healthcare is expanding its dominance over the labor market nationwide.

All subsectors go

Growth was solid across all four main subsectors of the California care economy. Employment in hospitals increased 5.2 percent, in ambulatory medical care (think doctors’ and dentists’ offices) by 15.7 percent, in nursing home facilities by 17.2 percent, and in social assistance by 27.2 percent over the past three years.4

Even more granularly, 36 of the 39 narrow industries that make up the care sector added jobs over the most recent three-year window.5 In absolute terms, services for the elderly and disabled topped the list, adding a whopping 211,000 jobs (31.5 percent increase) and accounting for nearly half of all job growth in the state’s care sector.6 Mental health practitioners led the way in percentage terms, with the ranks of California therapists nearly doubling (92.7 percent growth, or 27,212 new jobs).7

These numbers are huge but not anomalous. Nearly every corner of the care economy is growing, and no single corner can explain the larger sector’s might.

Nevertheless, it is striking how exceptionally low-paid these new jobs in services for the elderly and disabled are, coming in at $487 a week on average (or just over $25,000 per year) in the state.8 This is the lowest paid sub-sector of the entire care economy and one of the lowest paid in the entire economy, too.

Stepping back, it is also notable just how starkly the robust growth across basically every corner of the care sector compares to weak growth across the rest of California’s economy.

Trading down?

Lay out all the pieces and it becomes clear that, as the California economy evolves, jobs in high-wage sectors are being replaced by jobs in low-wage sectors. Jobs are shifting from manufacturing, information (tech), and finance and professional services into much lower-paying care roles. States such as Idaho are adding jobs in both high- and low-wage sectors. California is undergoing a great labor market transformation towards lower-paid work.

Consider this rough estimate: Multiply the average weekly wage in each sub-sector in Q3 2022 by the number of jobs gained or lost over the subsequent three years (essentially an expanded version of the above chart with the most granular subsectors available), and California workers shed around $1.26 billion in inflation-adjusted earnings a week through this sectoral trading down. That equates to $65.4 billion annually, and this amount is actually an understatement of the composition change’s effect because the total number of private sector jobs grew slightly (by 28,900, as shown in the rightmost column above) over this time period.

Diagnostic check

Could these findings be figments of the data — statistical or administrative blips that don’t reflect economic realities on the ground?

Our analysis was conducted using two Bureau of Labor Statistics’ (BLS) datasets, State and Area Employment, Hours, and Earnings (SAE); and QCEW. The latter, which is derived from state unemployment records, feeds into the former, which adds survey responses from BLS’s Current Employment Statistics series to offer even more timely estimates. QCEW is derived from high-quality administrative data and forms the benchmark for numerous other labor market series. California’s sheer size translates into high-quality data estimates across both sources.

Nor do there appear to be any California-specific biases in the largest subsectors driving the trends, such as Services for the Elderly and Persons with Disabilities (or NAICS 62412 in the classification system used by statistical agencies; see the footnotes). Concerns about how home health aides are scoped and categorized at the establishment level — is each home they work in considered an establishment, or only the main employer or agency home office? — do not apply to this exploration of jobs. What is more, any anomaly in one corner of the Health Care and Social Assistance sector (NAICS 62) is unlikely to explain weaknesses in other sectors. It’s highly unlikely that a care worker would be misclassified into manufacturing, retail, or tech, for example. Across the board, the state’s relative estimates appear in line with others; they simply add up to a story that is unique to the Golden State.

California has the highest foreign-born share of the population of any state. Perhaps the spike in care jobs somehow extends from President Trump’s immigration crackdown, as once-informal jobs are formalized or native-born workers begin to occupy the positions vacated by departing migrants?

Here too, the evidence doesn’t match up. The largest increase in care economy jobs appears in the first quarter of 2024, well before the election. The rate of increase in elderly services held steady around 3 percent per quarter from Q3 2024 through Q3 2025. And while it’s true that immigrants are overrepresented in the home health care industry (NAICS 621) and undocumented immigrants occupy a range of healthcare support roles, we would expect to find evidence to support the formalization hypothesis in construction and other industries, too, where the immigrant share of the workforce is even higher. The fact still stands that few sectors are expanding in the Golden State beyond healthcare.

Prognosis

What does the present composition of job growth mean for the future of the California economy? It’s hard to tell, but the rate at which the Golden State is shedding jobs in its highest-value sectors like tech and professional services while adding them in decidedly duller corners with more dismal wages is cause for concern. With a gubernatorial election this year, a billionaire tax possibly on the ballot, and soon-to-be-former Governor Newsom preparing for a potential presidential run, closer examination of the California model is also of national interest.

California’s population is not especially old or feeble. Only 16.5 percent of its population is 65 or older, making it the sixth-youngest state on this measure. Californians also report fewer chronic health conditions than almost anywhere else. So why is its healthcare sector so mighty?

One plausible partial explanation is that state policies are driving the outcome — in other words, that government subsidies are allowing more people to access more healthcare and social support, stoking demand.

And indeed, many of those new therapy jobs are probably supported by the state’s Children and Youth Behavioral Health Initiative, a $4.6 billion dollar program launched in 2021 to connect youth with mental health services. Enrollment in Medi-Cal, the state’s medicaid program, has increased significantly over the past decade as the state has continuously expanded coverage, including in recent years to undocumented populations. As a result, only 5.9 percent of Californians went without health insurance in 2024, a record low. The extensive margin (new enrollment) isn’t the only one at work; the intensive margin (use of healthcare services) is too. The state legislature has noted that Medi-Cal spending per enrollee is increasing even faster than the number of enrolled. Both greater utilization and greater coverage are driving demand and creating jobs in the process.

So how should we think about the care economy in the broader context of the California labor market? Is it an economic engine, a useful jobs sponge? Or does it represent an unhealthy dependence? Is the care economy picking up the slack left by weakness in other sectors or crowding out growth?

And given how intimately involved the public sector is in healthcare finance, we have to ask where the public support is coming from and what tradeoffs are involved with raising resources from one corner of the economy to spend them in another.

And then, finally, we have to ask how the political economy changes when healthcare spending effectively becomes a jobs program.

We might start to get some answers to these questions over the coming months, as the Medicaid- and healthcare-related portions of the One Big Beautiful Bill Act work their way through state coffers. Medi-Cal’s expansion was not only fueled by the state’s swelling tax receipts after the pandemic. It was also made possible by federal largesse and a tax on private plans that will be subject to heightened federal scrutiny going forward.

States that expanded Medicaid under the Affordable Care Act now have a nearly 3 percentage point greater share of their workforce in healthcare jobs than those that didn’t on average. A dozen expansion states have triggers that could discontinue or dial back their support if federal funding dries up — which it is poised to do. We may soon get a better sense of the extent to which the healthcare jobs juggernaut has been fueled by Washington, and whether it has legs strong enough to stand independently.

Where does that leave us? Healthcare has driven the plot in the national jobs story for several years running. The subplot of weak growth in other good jobs sectors may carry the narrative from here.

They say the future happens first in California. What we don’t know is how healthy it will look.

Students consider current labor market conditions when choosing a degree, so it isn’t too surprising that AI could shape students’ decisions today.1

But the study does not ask students exactly how AI has shaped their choices, only whether it has. Are students shying away from fields that have more exposure to AI, perhaps worried that AI will shrink the number of jobs available to them? Or are students shifting towards those fields, preparing for a future in which they will have to be comfortable using AI?

To find out, we can check enrollment for groups of degrees based on the AI exposure of the jobs that students with those degrees are likely to take, as shown in Figure 1.

As is clear, undergraduates are flocking towards the most-AI-exposed degrees, with enrollment in those degrees up 8 percent last year compared to 2017. This trend holds despite a notable decline in Computer Science degrees, one of the most-AI-exposed degrees, but whose decline is more than offset by increases in other exposed degrees like Engineering.

(A quick primer on our methodology before we move on: To estimate AI exposure, the general approach taken by researchers is to look at the tasks that workers perform in a given occupation and count how many of those tasks AI is likely to be good at. The more tasks associated with a given occupation that AI can perform, the higher the AI exposure of a worker in that occupation. AI exposure measures have important limitations, most notably that exposure can mean either augmenting tasks or automating tasks. We estimate the AI exposure of degrees chosen by college students by using the typical occupations that graduates of each degree end up choosing.2 See the Appendix for more details.)

Why are students still pursuing AI-exposed degrees? Part of the answer is shown in Figure 2: Wages are highest for those with the most-AI-exposed degrees.

In addition to our earlier work showing that AI-exposed occupations have higher incomes, this finding is also consistent with research from Morgan R. Frank, Alireza Javadian Sabet, Lisa Simon, Sarah H. Bana, and Renzhe Yu. They used data from LinkedIn profiles, combined with degree-level AI-exposure inferred from 3 million higher education course syllabi, and found that after the launch of ChatGPT, highly-exposed students enjoyed higher salaries and found jobs more quickly.

What About Recent Graduates?

While students today have the chance to switch degrees, those who graduated before the rise of generative AI are stuck with the degrees they have. It is theoretically plausible that young graduates with AI-exposed degrees face lower labor demand in their field and are forced to find work elsewhere. Think software developer turned retail manager. But is it in the data?

We can use American Community Survey (ACS) data to see the degrees and occupations of young graduates. The table below shows, for example, the top ten occupations of young graduates with a Computer and Information Sciences degree and how they have changed — or not changed — over time. The overwhelming majority of Computer Science graduates work in Computer and Mathematics occupations, the share having fallen less than a percentage point from the pre-COVID average to the years 2023 and 2024.

It doesn’t seem like young graduates with Computer Science degrees are upending their careers to find work. What about those with other AI-exposed degrees?

We first looked at the top three most-common occupations for each degree during the years 2015–2019. For Computer Science degree holders, this would be Computer & Mathematical, Office & Administrative Support, and Management occupations from the table above. We can then determine how many young graduates with each degree in subsequent years have also ended up in those same three occupations. The results are in Figure 3.

Across all three groups — Computer Science graduates; all graduates with the most-AI-exposed degrees including Computer Science; and all graduates with any degree — the share of graduates going to the most-common occupations has changed little over time. This result is consistent with a more-complex but generalized measure of how occupation choices change over time, as shown in the Appendix.

Taking Stock

There are several possible explanations for why we don’t see students avoiding AI-exposed degrees or young graduates working in occupations outside their field.

First, how AI is impacting the job market today is not at all clear. As Jed Kolko recently put it, this research is still in its early innings. If students are looking at the labor market to decide what to study, it’s not clear how AI would reorder their choice.

Second, the content of degrees or occupations themselves may change. If the curricula of AI-exposed degrees change to focus on skills complementary or resilient to AI, students may not feel the need to avoid those degrees. Similarly, the mix of tasks workers do in highly-AI-exposed occupations may be changing in ways that leave plenty of work for newly minted college graduates.

Third, as we noted above, the AI-exposure measures themselves have a lot of limitations. Knowing that a job contains tasks that an AI might be good at doesn’t necessarily tell you what will happen to employment in that job. This could help explain why Computer Science had decreased enrollment while engineering, which is also classified as highly AI-exposed, had increased enrollment.

Finally, there might be some behavioral frictions that slow the reallocation of students and graduates to occupations outside their field. Rather than looking far and wide for work, young highly-AI-exposed graduates might hold out for work in their field, waiting in unemployment longer.

Perhaps next year’s enrollment data will show movement away from AI-exposed degrees, but we doubt it. Not all “exposure” is the same. Occupation and sector-specific bottlenecks, diffusion dynamics, task reallocation, and output demand effects will jointly determine labor demand — and, in turn, the degrees students choose.

See our github with replication code here.

APPENDIX

Classifying degrees by AI exposure

We classify groupings of degrees by AI exposure using two pieces of information: (1) occupational-level AI-exposure measures and (2) the cross-sectional mapping of degrees to occupations. We utilize the Eloundou et al. (2024) GPT-4 beta scores, the most commonly used measure in the literature, as our occupational measure of AI exposure. This approach essentially walks occupational exposure back onto degrees based on how often individuals with each degree are observed in each occupation before the introduction of generative AI.

We start by using a sample of college graduate workers between the ages of 22 and 27, inclusive, in the 1-year ACS for years 2009 to 2019. Using these years avoids contamination with AI-induced changes in degree-to-occupation flows and changes during COVID. In each year, each individual in the sample is observed to be in one of 39 groupings of degrees in the ACS’s classification scheme and the Census 2010 4-digit occupation code. We compute the weighted mean of occupational AI-exposure scores for each degree-occupation pair. We then use this weighted mean to classify the 39 degrees into 5 quintiles based on AI exposure.

The AI-exposure scores are crosswalked to 2018 census occupation codes following our methodology in “AI and Jobs”, and to Census 2010 codes based on a crosswalk constructed using ACS samples with codes available under both vintages (occ2010 and occ for post-2018 samples). In some analyses, like Table 1 and Figure 2, we collapse occupation detail to 24 major SOC occupation groups. The degree-level exposure measures are shown below.

Measuring Enrollment

The ACS data is not well-suited to measuring how responsive college enrollment counts are to the introduction of generative AI. The ACS only reports college majors for those who have received their diploma. Students who graduated in 2023 and before were unable to adapt their degree choice in response to developments in AI that occurred in that year or after. Students graduating in 2024 would have needed to select a major by their sophomore or junior years (2023 or 2022), and so may have already been anticipating AI disruption when making this selection, but many would have been locked in to a major before ChatGPT’s release in November 2022. The 2025 cohort of college graduates with a bachelor’s degree are the most likely to have been capable of taking AI into account in their degree selection, but the 1-year ACS runs only through 2024.

To address these issues, we rely on the National Student Clearinghouse Research Center, which reports the number of enrollees — graduate and undergraduate — by declared major. This provides a more timely estimate of student majors. The clearinghouse received enrollment data submissions from 97 percent of all Title IV, degree-granting institutions in the U.S. in the fall of 2024. Coverage rates vary by year, but are handled with a weighing system described on their methodology pages here. Degree majors follow a different classification as the ACS, though most majors match up one-to-one.

Additional Figures

The figure below shows the change in degree-to-occupation shares for young college graduates with a Computer and Information Sciences degree. The figure shows the L1 distance, which is computed as the sum of absolute differences between shares in a given year and the base year. It answers the question: “How different is the occupation mix of students with a given degree now than it was in the past?”

If students were forced to look outside their field for a job, this should show up as an increase, even if temporary, at the end of each of these lines — the occupation mix would be different from prior years. Across different baselines, changes in degree-to-occupation shares are relatively stable over the last few years.

The chart below shows the percentage change in enrollment between 2024 and 2025 by degree and AI exposure. Bubble size is weighted by 2025 enrollment. We can see the 8 percent decline in enrollment in Computer and Information Sciences, which is among the most-exposed degrees. Engineering and business degrees, which are also highly exposed, saw rising enrollment. These two different trends net out to the slight increase in enrollment for all highly exposed degrees shown above. The linear trend is statistically insignificant at the 5% level (-17.17 [8.97]).

References

Blume-Kohout, M. E., & Clack, J. W. (2013). Are graduate students rational? Evidence from the market for biomedical scientists. PLoS One, 8(12), e82759.

Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2024). GPTs are GPTs: Labor market impact potential of LLMs. Science, 384(6702), 1306-1308.

Frank, M. R., Sabet, A. J., Simon, L., Bana, S. H., & Yu, R. (2026). AI-exposed jobs deteriorated before ChatGPT. arXiv preprint arXiv:2601.02554.

Long, M. C., Goldhaber, D., & Huntington-Klein, N. (2015). Do completed college majors respond to changes in wages?. Economics of Education Review, 49, 1-14.

Ryoo, J., & Rosen, S. (2004). The engineering labor market. Journal of political economy, 112(S1), S110-S140.

On the final day of April, President Trump signed an executive order to address one of the biggest wealth-building gaps in America’s economy: that too many workers, including 42 percent of full-time private-sector workers, lack access to retirement plans.

With the finalization of a policy hinted at in February’s State of the Union address, the Trump administration has created a new path to retirement for these workers.

As we explain in our detailed Q&A about the new policy, the government is creating the infrastructure and parameters for financial institutions to offer workers a retirement product similar to the Thrift Savings Plan (TSP), which is the retirement vehicle currently available to federal employees and members of the military. Moreover, these new individual retirement accounts will offer a federally-supported $1,000 annual match to low-income workers, consistent with the Saver’s Match legislation passed by Congress in 2022.

These new products will be listed on TrumpIRA.gov, along with an explanation for how eligible workers can claim the Saver’s Match. The new web site will be launched by the Treasury Department by January 1st, 2027, according to the executive order.

This policy builds upon years of EIG research, led by senior economist Ben Glasner. As Glasner wrote in 2023, “Retirement accounts are the largest source of aggregate fungible wealth for American households and are an important tool to build a nest egg for the future. Unfortunately, access to employer-provided retirement plans remains deeply divided across earning levels and regions of the country.”

With the launch of this new policy, EIG research estimates that 54 million more workers will now have access to a retirement plan. Of those 54 million, 11.5 million will be eligible for the full Saver’s Match, with another 14.6 million eligible for a partial Saver’s Match.

The new retirement initiative is a clear victory for economic inclusion, and it also creates momentum for even more progress. The next step is for Congress to pass the Retirement Savings for Americans Act (RSAA), a bipartisan piece of legislation that was built on the recommendations of a 2021 EIG white paper co-authored by economist Teresa Ghilarducci and Kevin Hassett, the current Director of the National Economic Council. The RSAA, most recently reintroduced by Representatives Lloyd Smucker and Terri Sewell and Senators John Hickenlooper and Thom Tillis in April of 2025, would automate enrollment, set a default contribution rate, and increase the match.

For media coverage of the new Trump administration retirement plan announcement citing EIG’s research and work behind the scenes, check out Axios, Semafor, CNBC, Barron’s, and USA Today.

Policy

• The EIG Opportunity Zones Coalition sent a letter to congressional leaders urging them to remove a provision in the Senate’s 21st Century ROAD to Housing Act that would restrict built-to-rent (BTR) development by requiring investors to sell properties within seven years. The coalition warns this would effectively eliminate Opportunity Zone investment in BTR housing, since OZ investments must be held for 10 years to receive the full tax benefits. The provision could potentially reduce new home construction by an estimated 72,000 units annually.

  ---

Research & Analysis

Manufacturing job gains are fading, but growth persists in dynamic places
The post-pandemic recovery of manufacturing jobs has been strikingly uneven across states and metropolitan areas. “Over four-fifths of this employment growth has occurred in just 354 out of 3,143 counties,” writes Jason He in his detailed analysis of the manufacturing sector. The two states that really stand out are Florida, because of its high manufacturing dynamism (elevated startup and exit rates), and Texas, which added the most manufacturing jobs of any state and accounted for over a third of the country’s manufacturing employment growth since 2019, writes He. National manufacturing employment has slipped since hitting a post-pandemic high in the third quarter of 2023.

H-1B Workers are Critical for AI Dominance
H-1B workers now represent an estimated 4.3 percent of the nation’s total AI workforce, and each year the share of H-1B visa holders who work in AI keeps climbing. Furthermore, write Sarah Eckhardt and Jason He, “Nearly four out of five new H-1B holders working in AI also completed their education in American universities, compared to 52 percent of H-1Bs overall. Losing many of these workers to China would undermine the domestic AI industry and could threaten American national security.” The United States should embrace

The Problem of Population Loss
Do residents of stagnant local communities recognize population loss as a problem? Sarah Eckhardt and Adam Ozimek answer: “The data show that local perceptions match demographic reality. Eighteen percent of respondents living in counties that lost population between 2010 and 2020 identified population loss as being a major problem for their community, versus 10 percent of those living in growing places. Including those who see it as a minor problem, 54 percent of respondents in declining counties view population loss as an issue, versus 34 percent in growing counties.” In addition to their direct contributions to the AI sector, these workers generate substantial fiscal benefits for the government. The United States should reform its high-skilled immigration system to make it easier for these workers to come, work, and stay here.

Around the Horn

• At the Center for Economic Policy Research, Nathan Goldschlag and coauthors explain why universities are losing top AI researchers to the private sector.

• In the Wall Street Journal, Jordan McGillis argues that female labor force attachment is a key driver of household income dispersion.

• In the New York Times, Jessica Grose cites Adam Ozimek and Nathan Goldschlag’s Agglomerations analysis on what’s happening in the young-adult job market. Rogé Karma cites the same analysis in The Atlantic.

• In Business Insider, James Rodriguez features Jess Remington’s research on the bifurcation of the housing market.

• In Newsweek, Billal Rahman quotes Sam Peak on the effects of the Trump administration’s $100,000 H-1B fee.

• In the Minnesota Star Tribune, Emma Nelson draws upon EIG’s Great Transfer-mation to show the rising share of Americans’ incomes deriving from federal social programs.

  ---

EIG Chart of the Month

From the post by Sarah Eckhardt and Adam Ozimek showing that residents of stagnant communities recognize population loss and a problem for them.

EIG Video of the Month

Jason Harrison and Ben Glasner riff on Jess Remington’s post in March, How the Housing Market Split in Two, which showed the differing experiences between new and existing homeowners:

The issue is certainly not that politicians neglect retirees as a group, or that policymakers are unwilling to use public money to help them. The United States already heavily subsidizes retirement savings. In 2019, federal income and payroll tax expenditures related to retirement savings totaled $276 billion, according to the most recent report that offers this breakdown from the Congressional Budget Office.1

The real problem is that the existing retirement system is badly designed to benefit the workers and retirees who most need help.

Retirement tax breaks from the government, primarily through 401(k) and similar plans, disproportionately benefit the highest-earning households. The primary reason for this disparity is not that lower-income workers choose to save a smaller share of their income, but rather that so many of them lack access to an employer-sponsored retirement savings vehicle in the first place.

The problem is obvious when simply viewing the key stats, starting with the outcomes…2

• Households in the top 20 percent by income received more than 60 percent of the benefits from “exclusions for pensions and retirement savings accounts” (the CBO category that represents the foregone taxes collected because of retirement program tax breaks).

• The bottom 40 percent of households by income together received less than 5 percent of the benefits.

• Within the bottom 20 percent of households by income, four out of five of them received none of these benefits at all.

… and then looking at the data on access to retirement accounts:

• Roughly 54 million workers, or about 47 percent of all full-time and part-time private-sector workers between the ages of 18 and 65, are not offered any retirement plan at work.3

• Among full-time private-sector workers, 42 percent lack access to an employer-sponsored plan. The situation is even worse for part-time workers: 79 percent have no access to an employer-sponsored plan.

• Nearly 80 percent of workers in the lowest earnings decile lack access, compared with just 18 percent in the highest decile.

• The income disparity is accompanied by racial and ethnic, generational, and rural vs urban disparities. Black workers, young workers, workers in small firms, and those in rural communities, for example, are all significantly less likely to have access as well.

For more on the methodology behind these estimates, including data on how many workers lack access to an employer match, see our earlier detailed analysis.

A New Approach

Closing the access gap and reducing the inequality in retirement savings will be impossible without first building the infrastructure that makes it easier for workers to save.

The federal government has started taking the necessary first steps, but there is confusion about what exactly has been announced, the likely effects of these new policies, and what policymakers should do next.

Here we attempt to answer some of the most common questions.

What just happened?

On Thursday, April 30th, the White House issued an executive order to expand retirement plan access for private-sector workers without employer-sponsored coverage.

Specifically, the government is creating the infrastructure and parameters for financial institutions to offer these workers a retirement product similar to the Thrift Savings Plan (TSP), which is the retirement vehicle currently available to federal employees and members of the military. The TSP offers low-fee index funds that invest in equities, bonds, and Treasuries.

In addition, the federal government will match half of each eligible worker’s contributions to this plan up to a limit of $1,000 (or 50 percent of the first $2,000 contributed by the worker).

How will workers know which plan they should choose?

These new products will be listed on TrumpIRA.gov, along with an explanation for how eligible workers can claim the Saver’s Match. The listed plans will be vetted by the Treasury Department to ensure their offerings align with those offered by the TSP. The plans will also be required to have no minimum contribution or balance.

In what ways will this new retirement policy help workers?

The first way is simply that more workers will have access to a tax-advantaged retirement plan. Retirement plans are incredibly effective wealth-building tools,4 but not every employer offers them, leaving many workers — especially low-income workers — excluded from this path to financial security in old age.

The second way this policy will help workers is through the match, which will boost each worker’s annual retirement contribution.

Where is the money for the match going to come from?

We need to take a step back to explain the answer.

The Saver’s Credit, which has been available to taxpayers since 2002, is a nonrefundable tax credit of up to half an individual’s contributions to a 401(k) or other retirement vehicle. It is capped at $1,000 per year. That it is nonrefundable means that many low-income taxpayers cannot get the full credit because their overall tax liability is too low (less than $1,000).

In 2022, Congress passed the Secure 2.0 Act, which replaces the Saver’s Credit with something called the Saver’s Match starting in 2027. The Saver’s Match is effectively a fully refundable tax credit of up to $1,000 (matching up to half the eligible taxpayer’s first $2,000 contributed to a retirement plan) to be deposited directly into an individual’s retirement account.

The Trump administration is making it possible for eligible workers who currently lack an employer-sponsored retirement plan to also have access to the Saver’s Match through their new retirement account. Thus, according to the administration, this matching benefit for its new plan has already been passed by Congress under the Secure 2.0 Act, and no new legislation is needed.

So is every worker who gets access to this new retirement plan eligible for the $1,000 match?

No, because the Saver’s Match is targeted by income. The only workers who are eligible to receive the full $1,000 match are married couples filing jointly who make at or below $41,000; heads of household who make at or below $30,750; and single filers making at or below $20,500.

If a worker makes more than that, given their filing status, they aren’t necessarily out of luck. They may fall into phase-out ranges. The Saver’s Match phases out linearly over modified adjusted gross income ranges of $41,000 to $71,000 for joint filers; $30,750 to $53,250 for heads of household; and $20,500 to $35,500 for single and other filers.

When does the new retirement policy take effect?

The program will launch in 2027. According to the executive order, the Treasury Department must also launch TrumpIRA.gov by January 1, 2027.

Can’t workers who don’t have access to an employer-sponsored retirement vehicle just set up their own retirement accounts, and access the Saver’s Match that way?

Yes. But what we know about low-income workers is that they overwhelmingly tend not to because of the administrative burden.5 Setting up such a plan on your own isn’t easy. The Trump administration is proposing to take on the administrative burden so that workers don’t have to. The new accounts will be offered by private providers and overseen by the Treasury Department.

And workers will then be automatically enrolled in those accounts?

No. Auto enrollment would require new legislation from Congress. So would increasing the matching amount beyond $1,000 and adding a default contribution rate.

How many workers will end up benefitting from the Trump administration’s new retirement plan?

We’ll give you the topline numbers first:

• 54 million total workers will become eligible for a new retirement plan.6

• Of those 54 million, 11.5 million will be eligible for the full Saver’s Match.

• Of those 54 million, another 14.6 million will be eligible for a partial Saver’s Match.

Here’s how the numbers break down by single, married filing jointly, and head of household:

And here’s how the math works. (If these details don’t interest you, feel free to skip ahead to the next question.)

As already noted, roughly 54 million workers currently lack access to an employer-sponsored retirement plan. Conceivably, all of them will be eligible for the new retirement plans created by the Trump administration.

Not all of them, however, will be eligible for the Saver’s Match of up to $1,000. This is where the calculations become a little complicated.

We can start by looking at how many workers will be eligible for the full Saver’s Match regardless of whether or not they already have access to a retirement plan. Using the Saver’s Match income thresholds and data from the Survey of Income and Program Participation (SIPP), we estimate that 15.1 million workers would qualify for the full 50 percent match.7 This includes full-time, part-time, self-employed, and government workers.

But only 3.5 million of these workers currently have a qualifying retirement account. What this means is that because of the Trump administration’s plan — here comes the answer — 11.5 million workers will have new retirement accounts through which they can receive the full Saver’s Match, or roughly 8 percent of workers.

Having done a similar calculation for workers with incomes in the phase-out range, we estimate that another 14.6 million workers will have new accounts through which they can receive a partial Saver’s Match.8

Why is President Trump doing this?

The president noted in the State of the Union speech in February that half of working Americans lack access to a retirement plan with matching employer contributions, pulling directly from work we did here at EIG.

What effects can we expect from the Trump administration’s new policy?

It’s impossible to know exactly how many workers will actually start saving more for retirement once they have access to these new accounts.

For the roughly 26 million workers who will gain new access to either the full or partial Saver’s Match, however, it is very likely to induce more savings. Survey evidence shows that interest in participation rises sharply once workers understand that a match is available. We also know from the federal Thrift Savings Plan that introducing a match increased employee participation by roughly 22 percentage points.

You mentioned that to automatically enroll these workers, increase the match, or set a default contribution rate, Congress would have to pass new legislation. Is someone writing that legislation?

It has already been written! The Retirement Savings for Americans Act (RSAA) is a bipartisan piece of legislation that was built on the policy recommendations of a 2021 EIG white paper by economists Teresa Ghilarducci and Kevin Hassett, who is currently President Trump’s Director of the National Economic Council. The RSAA was most recently reintroduced by Representatives Lloyd Smucker and Terri Sewell and Senators John Hickenlooper and Thom Tillis in April of 2025.

Like President Trump’s current plan, the RSAA would give workers without employer-sponsored retirement plans access to a new retirement vehicle similar to the federal Thrift Savings Plan, with similar investment options and employee ownership of the plan. But the RSAA would also go much further and offer to workers:

• Automatic enrollment in their new plan

• A default contribution rate set at 3 percent of a worker’s income

• Matching contributions of up to 5 percent

In short, the RSAA would extend to low-income and moderate-income families the same opportunity to build wealth that higher earners have long enjoyed.

Can you quantify the likely benefits for workers if the RSAA is passed?

Differences in wealth accumulation across households are driven not only by income but by unequal access to employer-sponsored retirement plans and participation in asset markets like equities and housing. Low-income households participate in these wealth-building systems at far lower rates, and that gap compounds over time.

What makes the case for a policy like the Retirement Savings for Americans Act more than a matter of fairness is the scale of the projected economic benefits.

Economists Pavel Brendler and Moritz Kuhn have estimated that every dollar of public spending on RSAA would result in roughly $2.40 in worker contributions and investment returns for workers and their families.

Younger workers would eventually enjoy the biggest gains, as they benefit most from time and compounding. According to Brendler and Kuhn, enacting RSAA today would amount to roughly $157 billion in additional accumulated retirement wealth, measured in 2024 dollars, for the nearly 5 million American households currently aged 25–29 by the end of their 40-year working life.

Simulation work from RAND shows that the RSAA would “enable the lowest earners (those who consistently earn in the bottom 10 percent of the earnings distribution) to save approximately $126,000 over a 40-year working career.” A worker at the median of the earnings distribution could approach $585,000 in savings.

If a lot of workers end up with tax-advantaged retirement accounts, and many of them are also getting a 5 percent match via public funds, won’t the policy have a big fiscal cost? Has anyone modeled that?

Yes — and in fact the fiscal effects over the long run are positive, not negative. Modeling from RAND suggests that implementing RSAA could generate more than two trillion dollars in net federal and state savings over forty years, largely because higher retirement assets reduce future spending on asset-tested programs such as Supplemental Security Income and Medicaid.

Okay, so what happens now?

It will be up to Congress whether to enact the remaining core elements of RSAA, such as auto enrollment and expanding the match. We hope it does. Policies that are fiscally responsible, address a big lingering inequality, and offer a boost to those who need it most are rare. With bipartisan support already on the table and key figures in the administration enthusiastic about the proposal, the moment for legislative action on retirement is now.

But the consequences of population loss can unfold gradually, making them difficult to recognize as they occur. Unlike discrete shocks — such as the COVID pandemic — which generate immediate momentum for action, the effects of population decline can be subtle and diffuse enough to lack the kind of urgency that motivates policy responses.

This raises a question: do affected residents themselves recognize population loss as a problem? Survey data suggests that they do.

The demographic problem is real

Population loss is undoubtedly a widespread issue. While demographic change has profound effects on a local economy and community, it’s unclear whether individuals actually perceive population loss in the first place.

We can test this question using a nationally representative survey of 2,005 respondents conducted by EIG and Echelon in 2022.1 In that survey, individuals were asked: “How much of a problem are each of the following in your local area these days, if at all?” One of the options was “population loss.” Importantly, we know the zip code where each respondent lives, which we then match to county-level economic and demographic data.2

The data show that local perceptions match demographic reality. Eighteen percent of respondents living in counties that lost population between 2010 and 2020 identified population loss as being a major problem for their community, versus 10 percent of those living in growing places. Including those who see it as a minor problem, 54 percent of respondents in declining counties view population loss as an issue, versus 34 percent in growing counties.

The more a county’s population shrank over the decade, the more likely its residents were to report population loss as a problem. In the top 10 percent of the fastest growing counties, 7 percent of respondents report that population loss is a major problem for their communities. In the fastest shrinking counties, that number rises to 23 percent.

Regression analysis demonstrates population loss is a non-linear problem. Among growing places, 1 percentage point slower population growth is associated with a 0.2 percentage point increase in the probability that survey respondents identify population loss as a major problem. In shrinking counties, the effect is 5 times larger.

Nobody’s business

Residents of shrinking communities don’t just recognize population loss in the abstract.

The survey data also shows that individuals identify specific economic symptoms that typically accompany demographic decline.

One of the most important consequences of population loss is lack of dynamism. A growing body of scholarship shows that demographic decline is linked to lower rates of business formation.3 Fewer residents doesn’t only mean fewer entrepreneurs, but also a smaller labor force. Both issues make it less likely that a community will be able to generate startups.

The survey data indicates that residents do perceive this connection between population loss and less dynamism playing out in real life. Respondents were asked how the number of new businesses has changed over the past 10 years, as well as how the availability of restaurants, stores, and entertainment options has evolved. In both cases, individuals living in areas with slower population growth are more likely to report that the number of businesses declined over the period.

What’s more, another survey question illustrates that the lack of new businesses rises to the level of concerning in demographically struggling places. Individuals reported whether a lack of new businesses or investment is a problem in their community, and regression analysis shows this is statistically significant related to population change. A 1 percentage point decrease in population growth is associated with a 0.79 percentage point increase in concern about insufficient investment and a 0.49 percentage point increase in concern about business decline.

In short, individuals in shrinking places do understand they have fewer new businesses and that this is a problem.

Wider problems

Unsurprisingly, the lack of investment in shrinking places is not an isolated problem. Another unavoidable result of demographic decline and lack of investment is that the tax base erodes. This in turn causes other deteriorations in quality of life. Population loss can also lead to local vacancies as some neighborhoods empty out and businesses shutter without replacement.

Our survey results are consistent with all of these issues. Regression analysis shows individuals in shrinking places are more likely to report that vacancy, the availability of good jobs, and poor public service are a problem in their community, and the effect is statistically significant. School quality is directionally worse, though not statistically significant.

Interestingly, people in declining places are very likely to report that both there were not enough workers (77 percent of respondents), and not enough good jobs (73 percent of respondents).

People living in declining places are less likely to support immigration than those living in growing ones.

Encouraging immigration to distressed regions is one of the main policy levers available to address population loss and its associated economic consequences.4 Yet residents of the places most in need of immigration are among those least likely to support it.

Residents of shrinking communities are less likely to view immigration as a net positive for the local or national economy, and are less supportive of temporary seasonal migration and immigration designed to address labor shortages. They are less likely to support high-skilled immigration even if it is framed as supporting entrepreneurship.

Support for immigration remains lower among those living in shrinking counties when controlling for urban versus rural status, political affiliation, gender, and educational attainment, suggesting that other factors play a role.

The contradiction between recognizing population loss as a local problem and being more likely to oppose a viable remedy poses a real challenge for policymakers.5 Well-designed place-based programs are unlikely to succeed without a corresponding effort to build public understanding of immigration’s role in reversing economic decline.

Still, there is reason for optimism. Support for high-skilled immigration remains strong: 61 percent of residents in shrinking communities support it, compared with 71 percent in growing ones. Policymakers should address the sources of opposition in declining areas, but this relatively high amount of baseline support suggests that efforts to expand high-skilled immigration in these regions may face only limited resistance.

See our github with replication code here.

Last August, the Chinese State Council announced its most comprehensive AI policy initiative to date. One directive of this “AI+” initiative is to expand China’s AI talent pool by encouraging enterprises to appeal to skilled workers using equity, stock options, and other incentives. The Chinese State Council has also announced a new visa for foreign STEM graduates.

Meanwhile, support for the American AI workforce is conspicuously absent from the White House’s recent AI Legislative Framework. And recent proposed and enacted changes to the H-1B visa program (for skilled immigrants) and OPT (Optional Practical Training, which eases a college student’s path to employment after graduation) either make it harder for AI experts to remain in the United States or fail to help their prospects.

The booming AI sector in the United States has greatly benefited from immigrant founders and engineers.1 Without sustaining and enhancing America’s AI talent advantage, we risk ceding ground to China in a competition with significant national security implications.

By the numbers

Estimates vary widely, but there are between about 50,000 and 200,000 AI jobs in the United States.2 Workers employed in these jobs are in high demand, with postings accelerating by the day.3

And as the sector becomes more and more important, the share of H-1B workers who work in AI has also climbed.

Using the most recently available FOIA data on H-1B approvals, we estimate that nearly a thousand workers who received H-1Bs in 2024 work in AI-related occupations,4 representing 1.12 percent of all approvals that year.

That figure may not sound huge, but this share far exceeds AI’s presence in the broader American workforce, in which AI jobs account for only 0.06 percent of all jobs.5 (The AI share of H-1B approvals also does not include university researchers in AI, for which an uncapped, or theoretically unlimited, number of H-1Bs can be issued.)

These annual flows of H-1B workers into AI are starting to add up. Using the share of approved H-1B Labor Condition Applications (LCAs) — a prerequisite for filing an H-1B petition — we estimate that H-1B workers now represent 4.3 percent of the nation’s total AI workforce.6

Nearly four out of five new H-1B holders working in AI also completed their education in American universities, compared to 52 percent of H-1Bs overall. Losing many of these workers to China would undermine the domestic AI industry and could threaten American national security.

In the small but rapidly growing AI labor market, the addition of 1,000 high-skilled workers each year would have an outsized impact on American competitiveness. With reforms to the H-1B visa selection process, we could attract even more AI experts.

How to Triple the Number of AI Workers on H-1Bs

Had H-1Bs been selected by EIG’s proposed wage-ranking system in 2024, the number of AI workers admitted would have more than tripled, rising to 3,330.7 A shift of that magnitude would amount to a meaningful expansion of the American AI talent base.

H-1B workers in AI are already well compensated, earning a mean wage of $150,000 — 37 percent above the already high average wage of H-1B holders broadly. A wage-ranked selection system would boost their mean wage even higher to $169,000.

Beyond their direct contributions to AI development, each worker generates substantial fiscal returns: the average federal fiscal impact of current H-1B AI workers is $38,000 per worker, rising to $44,000 under a wage-ranking system.8

Another way to clearly see the superiority of the wage-ranking model is to simulate what would have happened if it had already been adopted in the past. Under wage-ranking, the share of new H-1B visa recipients in AI would have been 3.9 percent in 2024 rather than 1.1 percent.9

The current H-1B lottery-based system favors large tech firms and relies on easily-manipulated occupational classifications.10 By reflecting actual market demand signaled through wages, a ranking system would satisfy frontier AI startups’ hunger for talent, generate positive fiscal impacts for the American people, and boost American innovation into the future.

Data through 2025 reveals that manufacturing jobs growth over the past six years has been highly concentrated. Over four-fifths of this employment growth has occurred in just 354 out of 3,143 counties. These high-growth counties accounted for around a quarter of 2019 U.S. employment in the sector. Areas that have sustained such growth into 2025 tend to have high dynamism and low pre-existing manufacturing dependence. Many legacy industrial cities continued to struggle.

Where are the new manufacturing jobs?

In the map below, pink-shaded counties, representing 62.3 percent of manufacturing employment in 2019, saw zero or negative growth between 2019 and the first three quarters of 2025. National recovery in manufacturing after 2020 was therefore driven by blue-shaded counties with positive net job growth. The high-growth counties with over 350 net new jobs accounted for 26.5 percent of national manufacturing employment pre-COVID but have contributed 83 percent of growth since then.

Aside from Florida, where most counties experienced gains in manufacturing employment, and the New England states, where nearly all counties lost manufacturing jobs, most states exhibit a common pattern: job losses in rural counties and large metropolitan cores, and gains in small cities and suburban areas.

Between 2019 and the first three quarters of 2025, 95.6 percent of net manufacturing jobs growth at the state level was concentrated in just twelve states, which can be grouped into three regions: the Southeast, Great Plains, and Mountain West.

The southeastern states of Florida, Georgia, Alabama, and South Carolina together account for 35.8 percent of state-level manufacturing jobs growth. “Superstar” manufacturing counties in this region are anchored by existing heavy-industrial clusters such as Huntsville–Decatur, Alabama; Spartanburg, South Carolina; and Savannah, Georgia, with growth driven primarily by expansions in the aerospace and automotive industries.

Florida stands out for having the largest share of counties with significant manufacturing increases. The state’s strength stems partly from its comparatively high dynamism in the sector, such as robust firm entry and job churn. In 2019, the manufacturing startup rate in Florida was 6.9 percent, compared with 4.7 percent nationally. Subsequently, Florida’s startup rate in manufacturing increased steadily each year, reaching 8.1 percent in 2023. By contrast, the national startup rate in the sector rose briefly to 5.3 percent in 2022 before falling back to 4.8 percent in 2023.

Texas added the most manufacturing jobs of any state, accounting for over a third of the country’s manufacturing employment growth since 2019. Manufacturing expansion in Texas spans multiple industries and is concentrated around the four largest metropolitan areas: Dallas–Fort Worth, Houston, San Antonio, and Austin.

The three states in the Lower Missouri River Basin, Kansas, Missouri, and Nebraska, also saw notable manufacturing job gains. Growth is concentrated in the suburbs of large metropolitan areas such as Kansas City, as well as in small and mid-size urban centers like Garden City, Kansas, and Columbia, Missouri. Alongside their traditional strengths in food processing and aerospace manufacturing, these states have also seen substantial job gains in chemicals, pharmaceuticals, machinery, and electric vehicle components.

Finally, four Mountain West states — Arizona, Nevada, Utah, and Idaho — recorded both large absolute gains in manufacturing employment and percentage increases exceeding 8 percent. Semiconductor manufacturing in Arizona’s Phoenix–Tucson corridor and Boise, Idaho (home to Micron), benefited from CHIPS and Science Act funding alongside rising global demand. Nevada increased manufacturing employment by 13.4 percent between 2019 and 2025, partly driven by expansion of the existing Tesla Gigafactory in the Reno metro area. Smaller urban centers such as Idaho Falls, Idaho, and Cedar City, Utah, experienced growth in chemical manufacturing, while suburbs of Salt Lake City saw a pronounced expansion in pharmaceutical production.

Has manufacturing growth held up?

Comparing manufacturing employment levels between the 2019 annual average and the 2025 Q1–Q3 average masks key dynamics in the intervening quarters. After the 2020 recession, manufacturers benefited from rebounding demand, monetary easing, federal investment, and the resumption of global trade. However, as the economy began to cool in 2022, manufacturing startup and job-turnover rates began declining, indicating that the brief upswing in dynamism has largely eroded. Supply chains have also faced disruptions from tariffs and armed conflicts.

Nationally, manufacturing employment has slipped from its post-pandemic peak of 12.9 million in the third quarter of 2023 to 12.6 million in the third quarter of 2025.

The map above shows which core-based statistical areas (CBSAs)1 are continuing to grow their manufacturing employment and which are past their post-2019 peak levels. CBSAs that sustained manufacturing growth into 2025 expanded employment by 12.9 percent, outstripping total workforce expansion by 3.1 percentage points and raising manufacturing’s share of total employment by 0.2 percentage points. Yet these growth centers account for only 13 percent of U.S. manufacturing employment as of 2019, demonstrating a high degree of concentration.

Among states that added substantial manufacturing jobs, sustained growth was concentrated in specific metro areas. Of the 67 Texas CBSAs that increased manufacturing employment, 23 continued to expand into 2025, accounting for 45.8 percent of the state’s manufacturing workforce in 2019. An overwhelming 95 percent of Nevada’s pre-pandemic manufacturing employment was in CBSAs that maintained growth through 2025. In contrast, Arizona’s CBSAs in the Phoenix–Tucson corridor experienced rapid semiconductor and EV-driven expansion but reached peak manufacturing employment in 2023. Only 0.5 percent of the state’s 2019 manufacturing jobs were in areas that maintained their post-pandemic growth trajectory.

The time series of manufacturing employment in CBSAs that are still growing and have not yet reached their employment peak reveals a critical turning point in the post-pandemic recovery. Employment in these growing CBSAs continued to rise through mid-2022 despite reversals in a few sparsely populated areas.

Beginning in the third quarter of 2022, however, the series experienced a sharp decline as increasing numbers of CBSAs reached their peak employment levels and exited the sustained growth category. This inflection point coincides closely with the Federal Reserve raising interest rates throughout 2022 to combat inflation. The tightening of monetary conditions, together with a slowdown in the broader economy, likely curtailed post-pandemic workforce expansion across many manufacturing businesses and contributed to the widespread peaking of employment levels observed thereafter.

The seeds of manufacturing divergence were planted before the COVID recession. In 2019, a quarter of manufacturing firms in metropolitan areas2 with employment growth into 2025 were younger than five years old, compared to 22 percent for metro areas that grew but reversed course and 20 percent for those that never recovered. This higher concentration of young firms suggests stronger entrepreneurial capacity in the years leading up to 2019, which likely boosted startup formation after the pandemic shock. Startup rates confirm this pattern: sustained-growth areas consistently had a larger share of new manufacturing firms between 2019 and 2023.

Metro areas that sustained growth into 2025 had the highest manufacturing establishment entry rates from 2019 to 2023 compared to the other two categories. While entry rates declined after 2022 across most metro areas, corresponding to employment peaking nationwide, the decline was least pronounced in metropolitan statistical areas (MSAs) with sustained growth, falling by only 0.5 percentage points while areas that reversed course or declined saw entry rates fall by more than twice as much.

Those with consistent growth into 2025 also show higher establishment exit rates, indicating elevated churn and dynamism. However, exit rates vary less across metro area categories than entry rates, suggesting ease of entry for new manufacturers likely contributed more to divergent growth paths than factory closures.

Our examination of post-pandemic manufacturing employment trajectories reveals high geographic concentration and widespread reversal of gains. Existing factories hire more workers during periods of high demand, monetary expansion, and government investment, but new manufacturers and plants appear to be crucial for enduring employment gains.

An American manufacturing renaissance requires understanding the policy conditions that enabled higher dynamism in areas with sustained manufacturing growth. We already know some levers that would work: removing regulatory hurdles for opening new factories, making it cheaper and easier to build workforce housing, and raising labor mobility through right-to-work laws and noncompete bans.

Policymakers must take dynamism seriously as a prerequisite for resilient manufacturing ecosystems. And we have shown the places that could offer a few lessons.

H-1B immigration has long been proven to boost innovation, entrepreneurship, and America’s macroeconomic outcomes. New research from EIG’s Adam Ozimek and Sarah Eckhardt shows that it also provides a shot in the arm for America’s fiscal woes.

In a report we published on March 17, Adam and Sarah find that the average H-1B household contributes $30,050 net each year to the federal budget. That’s 2.6 times the $11,530 contribution of a typical U.S. household.

Adam and Sarah furthermore break down the fiscal benefits for all 50 states and the District of Columbia. In 49 states, H-1B households are net fiscal positives, contributing an average of $5,040. These benefits are evident even in low-income states like Mississippi and West Virginia, which draw net fiscal impacts from H-1B households of $4,600 and $5,780 respectively.

As EIG has argued previously, the H-1B program leaves much room for improvement. This report models the fiscal impact at the federal level and the state level of implementing two high-leverage reforms: authorizing H-1B holders’ spouses to work in the United States, and changing the H-1B selection process from a lottery to a wage-rank system. Combining our two reform proposals would more than double the positive fiscal impacts, to $65,020 at the federal level and $10,650 at the state level.

Check out the full report, a handy one-pager, and an accompanying Agglomerations post here.

Policy

• Washington State became the fifth state to impose a full ban on noncompete contracts, which will take effect next year. The bill passed with bipartisan support and follows a 2020 reform that banned noncompetes for middle and low wage workers. EIG released a statement in response, quoting CEO John Lettieri: “Washington State’s action marks a significant step forward in restoring dynamism and opportunity in the labor market. Noncompete agreements prevent workers from taking better jobs, limit their earning potential, and impede the formation of new businesses. By moving to curb their use, Washington is helping to unlock opportunities for workers while strengthening competition and innovation across its economy.”

• Following the launch of the new EIG housing vertical in February, we are delighted to announce the members of our new Housing Advisory Council:

  
  Scott J. Alter, Co-Founder and Principal, Standard Communities
  Alex Armlovich, Housing Program Officer, Coefficient Giving
  Alfonso Costa, Jr, COO, Falcone Group
  Arpit Gupta, Associate Professor, NYU Stern
  Bobby Fijan, Co-Founder, Building Housing for American Families
  Colin Higgins, Executive Director, National Housing Crisis Task Force
  Edward Glaeser, Professor of Economics, Harvard University
  Emily Hamilton, Senior Research Fellow, Mercatus Center, GMU
  John Zeanah, Chief of Development and Infrastructure, City of Memphis
  Michael Novogradac, Managing Partner, Novogradac & Company LLP
  Mike Kingsella, CEO, Up for Growth
  Paul Williams, Founder and Executive Director, Center for Public Enterprise
  Sean Rust, VP of Development, NexMetro Communities
  Stephanie Kestelman, Director of Housing, Arnold Ventures

  ---

Research & Analysis

AI and Young-adult Jobs: The Real Mystery
To understand what’s happening in the labor market for young adults, it’s necessary to look at the employment rate for college and non-college workers. Both are lagging behind the employment rate for adults older than 25. Something mysterious is happening, but as Adam Ozimek and Nathan Goldschlag write, this does not seem like a story about AI displacing entry-level college graduates.

Place and Male Educational Attainment Gaps
There is significant educational attainment variability between men and women within the nation’s most distressed counties, writes Sarah Eckhardt. She investigates plausible reasons and finds: “We can see that on average, counties that have large black populations, are located in the south, are rural, or were a manufacturing hub in the ‘90s, have a more negative gap [female minus male high school education rates] than those that are not. Those that have a higher-education institution located in the county, or where a large share of households with children are run by a single parent, have a more positive gap.”

How the Housing Market Split in Two
“New homeowners are now burdened with some of the highest housing costs in decades,” writes Jess Remington, “while existing homeowners — those who have owned their homes for longer than a year — are actually spending less on housing than in the past.” The new homeowner penalty, which refers to housing costs for new homeowners above those paid by existing homeowners, hit a 34-year high in 2024. Just as worrying is that too many of the housing policies favored by policymakers would widen the gap even further.

Does a legacy in manufacturing preclude a future in it?
Kenan Fikri looks at the economic geography of manufacturing activity across the country and finds that new manufacturing is now “avoiding hubs of established prowess and instead gravitating towards places with little history in manufacturing. This tendency is not new. A fact of modern U.S. manufacturing appears to be that it trends towards deagglomeration.” Among other discoveries, Kenan also finds that the United States “has only 1.5 young manufacturing firms per 10,000 people today — a figure that has been bumping along at all-time lows for 15 years and counting.”

The New Bazaar

The Roots of our Zero-Sum Moment
Stefanie Stantcheva is an economist at Harvard and the head of the Social Economics Lab, where her team has done extraordinary work investigating how people form their opinions about economic and political topics. She speaks with Cardiff Garcia about the findings in her paper (with co-authors Sahil Chinoy, Nathan Nunn, and Sandra Sequeira), “Zero-Sum Thinking and the Roots of US Political Differences,” which was just published in the American Economic Review.

Stefanie and Cardiff discuss:

• Zero-sum similarities and differences between Democrats and Republicans

• The economic geography of zero-sum thought

• The finding that surprised her the most

• The generational gap in zero-sum thinking between young and old

• The policy implications of her research

• A preview of her upcoming work on zero-sum thinking and AI

Ideas for a Post-YIMBY Housing Future
Arpit Gupta, a finance professor at NYU, speaks with Cardiff about his latest contributions to the study of housing affordability, remote work, artificial intelligence, and finance. Arpit has offered a variety of reform proposals that both complement YIMBY ideas and also prepare for a future after a YIMBY victory. Among them are re-thinking property taxes, accelerating depreciation schedules, and making it easier for factory housing to get to market.

Around the Horn

• Using the story of the player piano, Adam Ozimek writes in The Atlantic on the durable demand for bespoke, human labor even as technological change would appear to make it obsolete.

• Emily Peck in Axios cites Jess Remington’s Agglomerations post on the housing market’s splitting in two.

• Eric Levitz in Vox cites EIG’s recent publication from Google economists Zanna Iscenko and Fabien Curto Millet on how to interpret LLMs’ impact on the economy.

• In the Washington Post, columnist George Will cites Sarah Eckhardt, Ben Glasner, and Connor O’Brien’s geographic analysis of recent electoral politics.

• Also in the Washington Post, Kevin Schaul and Shira Ovide cite EIG’s research as a counterweight to the runaway narrative that AI is harming the labor market for young college grads.

• And Jordan McGillis makes that same case in National Review, arguing that college grads are markedly better off than their non-college peers.

• The Seattle Times cites EIG’s work on noncompetes as the state of Washington prepares a noncompete ban.

  ---

EIG Chart of the Month

Via Jess Remington’s post on how the housing market split in two:

EIG Video of the Month

Ben Glasner riffs on the work of Stefanie Stantcheva, as discussed recently on The New Bazaar. See more EIG videos here.

The first post in this series documented U.S. manufacturing’s stasis ever since the Great Recession of 2008 — low rates of job creation and job losses, low rates of job turnover, and low startup rates.

This post similarly peers under the surface of the sector to examine what geography and business dynamism, together, can tell us about the health of American manufacturing today.

The data paint the picture of new manufacturing activity — in the form of new establishments and new firms — avoiding hubs of established prowess and instead gravitating towards places with little history in manufacturing.

This tendency is not new. A fact of modern U.S. manufacturing appears to be that it trends towards deagglomeration.

What do these location choices reveal about the drivers of U.S. competitiveness today? Why is more manufacturing in an area associated with less manufacturing dynamism? And why does having a legacy in manufacturing seem to make it harder to build a future in it?

The answer to these questions might hold the key to American re-industrialization.

Deagglomeration nation

The total number of manufacturing firms nationwide fell slightly from 2020 to 2023, as it has consistently since about 1998.1 This decline was not pervasive, however. It was primarily due to the slow but steady hollowing out of the nation’s existing manufacturing hubs.

Nikhil Kalathil of Carnegie Mellon and coauthors developed a framework that classifies counties based on the nature of manufacturing sector agglomeration within them. Covering 1,800 counties with sufficiently sized manufacturing bases, the authors identify four different archetypes of agglomeration:2

• High agglomeration areas where firms locate with both peers and suppliers

• Peer (horizontal) agglomeration areas where firms locate with others in similar industries and similar positions in the supply chain

• Supplier (vertical) agglomeration areas where firms locate within a particular supply chain

• Low agglomeration areas where firms locate with few peers or suppliers

High agglomeration counties remain home to the bulk of the nation’s manufacturing firms and jobs, but they lost over 1,300 companies on net between 2020 and 2023. By contrast, low agglomeration (+424), peer (+38), and supplier (+74) agglomeration counties all added to their manufacturing bases on net.

In percentage terms, high agglomeration counties shed 0.7 percent of their manufacturing firms, while firm counts grew by 1 percent in low and peer agglomeration counties.3 Firm dynamics — the constant churn of openings and closings — are gradually pulling the sector’s center of gravity out of traditional hubs.

Economic Geography 101, revisited

At first glance, this trend towards deagglomeration is surprising because fundamental precepts of economic geography around industry clustering and agglomeration — from Marshall (1890) to Krugman (1992) — were forged in a goods-producing world.

In these canonical frameworks, firms co-locate to take advantage of information flows and knowledge spillovers, tap deep pools of specialized labor, reduce transportation costs, and build relationships.

Part of the magic of agglomeration is that it should spur dynamism by its very nature. Mash up a critical mass of complementary people and firms, thinkers and doers, innovators and imitators and they will unlock progress.

If expertise is an advantage and new commercial opportunities are more perceptible with proximity, you’d expect new businesses to start where an industry already has a presence, too.

In the economists’ jargon, agglomerations emerge because they offer firms increasing returns.

That is, until they don’t, at which point they start offering diminishing returns instead.

Diminishing returns can set in with size. As agglomerations grow, so do congestion costs, which include traffic and high prices of land and labor.

Diminishing returns can also set in with time. As firms and industries mature, they tend to seek out more generic and less specialized locations, as they rely less on innovation and agglomeration-based advantages to stay competitive and more on driving down costs of production.

Agglomerations may grow less dynamic over time, too, as a sort of spatial industrial sclerosis develops and winning firms eventually grow older and bigger, workforces age, technologies get locked in, vested interests accrue, and entrepreneurial vim fades.

A map of attrition

Manufacturing’s deagglomeration should be interpreted through this framework as a symptom of a sector that has lost its dynamism. The series of state-level scatterplots below show how decline and deagglomeration have proceeded together.

Each dot represents a state, sized by manufacturing’s share of the workforce. The y-axis represents the startup rate, or the share of all manufacturing firms in the state that started in the past year. The x-axis represents the death rate, or the share of all manufacturing firms that shuttered in the past year. Values are averaged for each decade. The 45-degree line represents balance, where each dying manufacturer is replaced by one new one. States above the line enjoy more manufacturing firm births than deaths; states below, the opposite.

The steady march of states below the 45 degree line shows how manufacturing has faded. The mainly vertical progression shows that American manufacturing has adjusted to economic change on the entry margin — that is to say that dynamism has fallen because the startup rate has collapsed while the rate of failing firms has remained largely unchanged. And the startup rate’s collapse has had a profound impact on the states in which manufacturing constitutes a bigger share of the economy (larger points in the graph) by pushing them more deeply into that attrition territory.

As a result, manufacturing is deagglomerating because the only places still experiencing net entry are those with less of a manufacturing base to start. The sector’s heartlands have lost the ability to launch more firms than they lose each year.

The finding holds for counties and at the establishment level too. Looking at even more recent QCEW data for the past three years, the number of new manufacturing establishments — which includes new firms and branch plants or new outposts of existing firms — increased by 4.3 percent in counties that had no specialization in the sector, compared to around 2 percent for counties moderately specialized in it and a decline of -0.1 percent for counties significantly specialized.4^,5

We’ve become so familiar with new manufacturing establishments opening in empty fields that it’s easy to overlook the shocking revelation in these figures.

Places with large and diversified manufacturing bases appear to be less conducive to startups and less attractive to expanding firms than places that represent a blank slate.

On the one hand, this tendency might attest to the comparative strengths of new locations unsaddled by a legacy in the sector. But on the other hand, it signals that something has gone deeply wrong in our agglomerations.

But what is it? What mix of factors have conspired to send American manufacturing agglomerations-first into the dynamism doldrums?

The sector has been battered by automation, off-shoring, and foreign competition, of course. But local factors, and how places respond to economic shocks and technological change, matter too.

To see the future of manufacturing in America, we need to look beyond today’s companies and ask why new ones aren’t waiting in the wings. We need to look past surface-level decline and into the dynamics of resilience and renewal at the local level.

The stakes are high, because until we have a better understanding of what holds manufacturing back in the places that embody our national expertise in the sector, the country risks continuing down a quixotic and futile path of implementing industrial policy without the industrial base.

Searching for startups

If there’s one measure that best symbolizes renewal, it is startups. New firms with new technologies, products, or business models to replenish the stock of enterprises that inevitably thins through the course of economic churn and change.

Normalized by population, the country has only 1.5 young manufacturing firms per 10,000 people today — a figure that has been bumping along at all-time lows for 15 years and counting. Zooming out, that means for every 1 million Americans, there are only 150 manufacturing firms of any size or speciality that have launched within the past five years.

Bright spots

Those young firms, few as they may be, point to enduring advantages of making it in America, and they have helped bolster an embattled sector.

What is more, dozens of metropolitan areas are nurturing new manufacturers at much higher rates than the country overall.

These include:

• Major metropolitan engines such as Los Angeles, CA, and Miami, FL.

• Competitive manufacturing clusters such as Elkhart, IN, and Holland, MI

• Mid-sized micropolitans like Cookeville, TN, and Somerset, PA

• Emergent western production hubs such as Burley, ID, Evanston, WY, and St. George, UT

• Mid-sized advanced technology centers such as Boulder, CO, and Burlington, VT.

Those bright spots point to a real competitive advantage that explain how the country remains a technological and, yes, a manufacturing superpower. But they remain the exception. Most major metros and historic manufacturing clusters track the nation.

Limited startup activity reigns along much of the East Coast, large stretches of the Southeast, and even midwestern metropolitan areas such as Columbus, OH, Indianapolis, IN, and Pittsburgh, PA, with world-beating research universities and longstanding efforts to integrate leading edge innovation into legacy manufacturing bases. Boomtowns such as Atlanta, Dallas, Nashville, and Phoenix trail the nation on spawning new manufacturers, too.

Los Angeles is a particularly interesting case study. Its manufacturing sector remains more entrepreneurial than most other major cities. The region is garnering renewed attention as it builds on its aerospace roots to become a center of defense-related “hard-tech.” Alumni from SpaceX and other firms are launching new manufacturing startups in the classic spinout process that makes strong clusters (greased, in this case, by California’s prohibition on noncompete agreements). Such entrepreneurial ferment is a key ingredient in dynamism.

Those advanced manufacturing startups attest to the area’s strengths in risk capital, know-how, and talent. They prove that the location itself still has the power to inspire entrepreneurship.

And yet, the number of young manufacturers in metro Los Angeles fell to its lowest level in decades in 2023.

This larger sectoral trend in the region attests to the area’s weaknesses, notably high costs across the board for both firms and workers.

What happens when congestion costs overpower the forces of agglomeration? People and businesses leave. Firms fail to start. The magnetic pull and inherent dynamism of a place like Los Angeles is dampened, leaving manufacturing in the nation’s second metro area smaller and less innovative than it could be.

Virtuous restoration

Los Angeles tells the national story. Metropolitan areas performing below their potential, summing up to a nation performing below its potential, too. Agglomerations past their prime and struggling to battle decay. Agglomerations that enervate rather than invigorate dynamism.

The spread of manufacturing itself is not inherently negative. The sector is an engine of economic development and opportunity for the communities into which it enters. Manufacturers have a long history of making location decisions to avoid having to compete with other firms for labor. Corners of the United States have real comparative advantages based on the costs of land, labor, and energy. New clusters may be forming in some of these low-agglomeration areas, too.

Deagglomeration is only a problem insofar as it is a symptom of the poor health of the nation’s manufacturing heartlands. That is exactly the diagnosis presented here.

Industrial policy now captivates both parties. The federal government has pledged trillions in subsidies to big firms to make it in America. The Trump administration has tried to fundamentally reset the terms of trade with tariffs. Entrepreneurship has been almost completely neglected. The low- to no-cost work of dismantling barriers to commercializing innovations and growing new firms has been largely ignored. Fundamental questions about how to strengthen U.S. competitiveness — and the regional foundations of U.S. competitiveness — remain unanswered.

To put it plainly: The nation will fail to activate a manufacturing renaissance without revitalizing innovation and entrepreneurship within its established agglomerations.

Luckily, there’s no shortage of ways to get started. Ban noncompete agreements so that nimble new firms can spin-out from lumbering old ones. Liberalize housing construction so that more talented people can afford to live in our most productive regions. Streamline regulations so that redeveloping brownfield sites can be cost- and time-competitive with building on greenfield ones.

And the best part of all is that local, state, and national leaders can all do their bit to make progress.

Geography and dynamism, together, help explain how manufacturing in America arrived in its current state. They also show how the sector can climb out of it.

In a new EIG report, my colleagues Adam Ozimek and Sarah Eckhardt demonstrate that H-1B immigration has an immediate and positive fiscal impact. Ozimek and Eckhardt analyze the taxes that H-1B immigrant households — the most numerous in the high-skill category — pay to federal, state, and local coffers, and the benefits they draw from the public budget. They find that each H-1B household has an average positive federal fiscal impact of $30,050 per year, a figure that is 2.6 times greater than that of the average American household. Moreover, at the state-and-local level, the average H-1B impact is over $5,040 annually.

In other words, oft-maligned H-1B workers are plugging the federal deficit and replenishing strained state finances. How?

H-1Bs are remarkably net-positive for two main reasons. First and foremost, they make a lot of money. H-1Bs are very well-paid compared to the average worker because they work, almost exclusively, in high-wage, in-demand sectors like tech and medicine. With an annual income of $130,075, the median H-1B worker earns at the 91st percentile nationally.

The second reason H-1Bs contribute so robustly to government finances is demography. More than 60 percent of H-1B recipients are under the age of 35, which means that H-1Bs tend to consume very little in the way of public services.

Demography does, though, cloud the fiscal impacts of one subset of H-1B workers: those with a dependent spouse and children, a group that makes up 25 percent of H-1B households. While they are still net-positive at the federal level, H-1Bs who bring with them a spouse and children are a net fiscal negative at the state and local level due to state spending on public education.

Critically, this negative fiscal impact mirrors that of Americans of similar age, income, and fertility. Until reaching an income of $250,000, single-earner American families with children, too, are net fiscal negatives at the state level. All of these facts point to a promising reform: authorizing the spouses of H-1Bs to work too.

Reforming the H-1B program to include work authorization for all H-1B spouses — who themselves enter the country on H-4 visas and are generally barred from working — would amplify positive fiscal impacts and reduce the fiscal drag introduced by the one-quarter of H-1B households that currently have children and just one earner. Because H-1B spouses tend to be comparably skilled to the visa holders, our report finds that authorizing them to work would boost the average positive fiscal impact of an H-1B household by 50 percent, to $45,850 at the federal level and $7,670 at the state-and-local levels. Our report also finds that by authorizing work for H-1B spouses, even H-1B families with children would become net fiscal positives in most states.

Another high-leverage reform opportunity is to alter the way we select H-1Bs. The current system is a pure lottery, provided H-1B applicants meet the basic educational and occupational criteria. We propose ending the lottery and enhancing the composition of H-1B cohorts by selecting applicants based on a wage ranking system that rewards age-adjusted earnings. Wage ranking the selection process would alone boost the fiscal impacts per H-1B to $48,700 at the federal level and $7,980 at the state-and-local levels.

Together, selecting H-1B applicants based on wage ranking and authorizing their spouses to work would push their net fiscal impacts each year to $65,020 at the federal level and $10,650 at the state-and-local levels.

The H-1B program has taken heat from conservatives and progressives alike in recent years, yet this report shows that the roughly 700,000 H-1B visa holders in the country at this time are of enormous fiscal value, generating a cumulative federal net fiscal impact of around $20 billion each year. Our proposed reforms would double that figure to $40 billion and provide an ongoing revenue stream, unlike the Trump administration’s one-time $100,000 H-1B fee, which does not directly address the program’s composition problem. It’s important to note that this profound, direct, near-term fiscal impact does not even factor in the tertiary economic benefits of the program that stem from increased innovation, entrepreneurship, and productivity growth. These well-documented spillovers indirectly boost tax revenues even higher in the long term.

As the nation careens toward fiscal crisis, the H-1B program should not only be judged on innovation grounds or on cultural grounds, but on the unifying goal of getting our federal, state, and local budgets in order. Improving the program by picking the best applicants and authorizing their spouses to work is a rare, politically-palatable way to plug fiscal gaps.