Research Highlights Article
February 26, 2026
IT investment and the earnings distribution
The geographic dimension of wage polarization.
Source: graphy111
Economists have studied the relationship between technological change and labor market inequality for decades, but most of that research has examined trends at the national level.
In a paper in the American Economic Journal: Macroeconomics, authors Jan Eeckhout, Christoph Hedtrich, Roberto Pinheiro take a different approach by examining how information technology adoption and job displacement vary across metropolitan areas with different housing costs. Their findings reveal that technological change does not affect all places equally and that local housing prices play an important role in determining which jobs disappear and where.
Since the 1970s, inequality has risen substantially faster in larger cities than in smaller ones. At the same time, this geographic pattern has coincided with well-documented labor market trends driven by technological change and trade.
"If technology really is behind this change in the aggregate, we had a hunch that there must be something about it. Maybe it also explains the differences across cities," Hedtrich told the AEA in an interview.
To study the issue at the right level of granularity, the researchers obtained access to establishment-level data on information technology expenditures from the Ci Technology Database, which covered over 200,000 sites across the United States in 2015.
Their empirical analysis established two central facts. First, firms in expensive cities invest more heavily in information technology per worker. Second, routine cognitive occupations, such as bookkeepers, administrative assistants, and data entry clerks, have declined substantially faster in expensive cities.
Routine cognitive jobs and the cost of living
The chart below shows the changing relationship between rental prices and the share of routine cognitive jobs in metropolitan areas. In 1990 the relationship was positive, but by 2015 it turned negative.
According to the authors, the mechanism underlying these patterns reflects basic economic logic about relative prices. Because information technology can be purchased at similar prices everywhere, while labor costs vary substantially with local housing prices, firms in expensive locations face stronger incentives to substitute capital for labor. This substitution effect concentrates in occupations where technology can most easily replace human tasks.
"In a place like New York City, it's really expensive to hire people because they have to live close by. But if you can outsource some of the work to a small machine that fits on a desktop, you don't have to compensate for rent," Hedtrich said.
The challenge in quantifying this mechanism is that housing prices themselves are equilibrium outcomes determined by productivity, amenities, and other factors that are highly correlated with wages and occupations. So the researchers developed an equilibrium model where workers choose locations and occupations based on wages, housing costs, and their preferences, while firms choose optimal combinations of differently skilled workers and capital given local productivity and input prices.
The model—along with previous evidence based on national time-series— allowed the researchers to estimate occupation-specific elasticities of substitution between information technology and labor. They found that IT complements labor in nonroutine cognitive occupations—such as scientists and engineers—with an elasticity of substitution of 0.87, meaning that better technology complements these workers. In contrast, IT substitutes for labor in routine cognitive occupations with an elasticity of 2.5, making displacement more likely.
If the mechanism we describe is relevant, we should pay attention to the restructuring of work in expensive versus cheap cities. The adoption of so-called AI and its effects should be particularly pronounced in expensive cities.
Christoph Hedtrich
Using the model, the researchers conducted a counterfactual exercise where they simulated the labor market allocation if IT prices had remained at their 1990 levels rather than falling by 65 percent. The results indicate that falling IT prices can explain approximately 30 percent of the relative decline in routine cognitive employment between expensive and inexpensive cities. The model also predicts that the wage premium for nonroutine cognitive work relative to routine cognitive work is higher in more expensive locations, consistent with patterns observed in the data between 1990 and 2015.
Overall, the authors’ findings show that job polarization and rising inequality reflect not just technological forces but the interaction between technology, geography, and housing markets.
Looking ahead, Hedtrich suggests that this framework offers guidance for thinking about artificial intelligence and other emerging technologies: "If the mechanism we describe is relevant, we should pay attention to the restructuring of work in expensive versus cheap cities. The adoption of so-called AI and its effects should be particularly pronounced in expensive cities."
♦
“IT and Urban Polarization” appears in the January 2026 issue of the American Economic Journal: Macroeconomics.