De-industrialization and inequality in the United States

Last year, I met a writer and political theorist named Benjamin Studebaker. He was an heir to the Studebakers, a family who made their fortune in South Bend, Indiana during the early 1900's. I'd read articles he'd written about his family and his hometown, and out of curiosity, I met with him privately to learn more.

Studebaker began in the wagon business, transitioning to automobiles during the 1910's as the private car market took off. The company limped along through the Great Depression, but flourished during World War II under the Lend Lease Act where their trucks became a staple on Allied front lines. At its peak, Studebaker employed over 500,000 South Bend residents.

Pictured: The Studebaker Proving Grounds

The company saw a slow decline during the postwar period as GM, Ford, and Chrysler pushed smaller competitors out of the automotive industry. Within a decade, the company which employed nearly all of the adult male population in South Bend vanished.

Today, many lower and middle income residents are service workers at the University of Notre Dame, or work in retail or food service. White-collar residents are often expats from Chicago and Indianapolis, commuting in and out of the city to save on rent.

Pictured: Abandoned Studebaker Trucks

The words my friend used to describe his hometown included "bombed out" and "rotten". No significant wealth is produced in the area, and it's difficult for the native residents to find stable income or housing.

This is an unfortunate reality in much of Middle America; the opportunities may have left decades ago, but the families remain, while socioeconomic conditions around them decay.

Pictured: Abandoned house in South Bend set to be demolished. Part of a wider campaign to remove "blight" from the city to raise property values and make the city more attractive to tourists and homeowners.

It's worth noting that manufacturing hasn't "disappeared", as much as the job market within manufacturing has shrunk. The United States remains the world's fourth largest steel producer, the second largest automaker, and the single largest aerospace exporter. From this perspective, American manufacturing is still quite powerful.

However, the change which has really wounded the U.S. is the availability of work within the manufacturing sector. This chart represents the total percentage of the U.S. workforce employed in manufacturing. Since the mid-1960's (around the same time as the Studebaker closures), the percentage of Americans employed in manufacturing has dropped at a near-constant rate.

During my analysis, I tried looking for structural breaks in all of my available datasets, including this one. I was unable to pick up on any significant breaks in the rate of change, meaning that this downward trend never saw a significant period of recovery between the mid-1960's through today.

We can observe this general decline in the average inflation-adjusted wages of manufacturing workers, as well. Wages shoot up drastically and peak just before 1980, before experiencing a sharp decline the following year.

To get a sense for how the wage trends truly "behave", I used the Ruptures library in Python. This package detects likely points in the timeline in which the underlying data generating process changed due to an external event. Indeed, the first regime begins in 1971 and ends in 1987, which bookends the parametric "bump" in relative wages. At first glance, it may seem that a spike in wages contradicts the relative decline in manufacturing jobs. However, as an industry shrinks, the lower-skilled, lower-paid workers are typically pushed out first, increasing the representation of higher-paid workers in the average.

The second regime shows a flattening out of the curve, with wages falling far below the peak, stabilizing over time. Rather than a sign of life, the periodic bumps may reflect system-wide layoffs of manufacturing workers, in the same manner as the post-1970 period. Indeed, some of the most significant losses in terms of job counts occurred between the year 2000 through the 2010's, with over 5 million jobs lost. The final regime was strongly influenced by the Covid 19 pandemic, grouping the rise and fall into a single trend.

The constant decline in manufacturing jobs paints a negative picture of both the peaks and the valleys in the data. Peaks indicate that entry-level, low-skilled become less common, while valleys indicate that the wages tend to stabilize at lower levels for the remaining workers.

We can further contextualize these trends by visualizing wage shifts for all private sector workers (marked in red). This includes growing sectors, such as work in service jobs and tech. At a glance, it appears that the two trend lines move alongside one another, reflecting shocks in the job market which affected the entire workforce. However, the combined workforce also includes sectors which are growing in numbers, so occasionally, an increase in the line does indicate job recovery alongside improving wages.

Like the manufacturing dataset, the unusual wage spikes in 2008 and 2020 in both trends was due in part to low-paid workers exiting the workforce, while relatively high-paying jobs remained. This explains why wages increased, despite a visible drop in manufacturing employment in the same year.

I was able to find significant co-integration in the first five years of the data, meaning that the wages for both manufacturing and the entire private sector moved in unison. However, past the first five years, there appeareded to be no co evidence in favor of co-integration. This suggests that following the first 5 years during the overlap (between the first set of break points), wages for manufacturing experienced a decoupling with the rest of the economy, and stagnated while wages elsewhere dropped, and then shot up. While visually unintitive, it makes sense, as manufacturing workers would come to represent a smaller and smaller piece of the American wages, and would thus have a smaller overall impact on the broader trends. We can observe this decoupling more broadly by looking at the location of each break point. The wages for all workers experienced ruptures that seemingly did not appear within the manufacturing wages (e.g., in 1999 and 1980), demonstrating that the two trends are often moving independently.

Finally, we can observe trends in unemployment at the state level. This map represents the raw counts of workers employed in manufacturing at a given point in time, with color intensity representing the relative "health" of the market during that year.

1990: The data available ranges from 1990 to 2024. Within the dataset, employment sits at its healthiest level. Recall from the previous chart that the representation of manufacturing workers in the 1970's is nearly twice that of 1990.

1995: As the years pass, the balance shifts, with some states experiencing and increase, and some a decrease in employment.

2000: However, by 2000, it appears that the job market in manufacturing is reasonably stable. The hardest-hit state appears to be New York, which saw approximately 250,000 workers leave the between 1990 and 2000.

2005: The most dramatic shifts were yet to appear. Between 2000 and 2005, all states in the Rust Belt region experienced a significant decline in employment. in manufacturing workers.

2010: By 2010, employment reaches its lowest point for the entirety of the dataset. Several factors played into this rapid decline. Firstly, manufacturing in the United States became far more productive, slashing the number of workers required to keep any given operation functional. Additionally, intense global competition with countries such as China and India moved much of the world's manufacturing overseas.

2015: During my conversations with Studebaker, I'd asked him if simply re-shoring the lost factories through policy would help reverse this decline. His response was simply, "No". Let's say we bought out a car manufacturing plant from overseas, and rebuilt it in South Bend. No matter how good our intentions were, a modern plant requires far fewer skilled laborers, and wouldn't be able to offer the few workers it needed competitive pay. He described the contemporary factory model as a set of "lego kits". Rather than requiring a fixed headquarters and any kind of centralized operations, modern factories are built to be taken down, and quickly reconstructed elsewhere in the world, preferably in areas where workers make fewer demands about their salaries. The distribution of factories around the world is, from a profit standpoint, the path of least resistance.

2020: Through the 2010's and up until the 2020's, manufacturing saw limited recovery, although the raw counts never returned to those of the 1990's. Nearly half of the gains nationwide happened within the "rust belt" states of Michigan, Indiana, Ohio, and Wisconsin. However, some of this recovery occurred elsewhere in the southern U.S., indicating a kind of "domestic outsourcing", where opportunities shift around to more convenient locations within the United States.

By 2024, it appears that these shocks are here to stay. While new opportunities may have cropped up elsewhere in the country, the towns which once relied on manufacturing remain, without the promise of stable employment they once offered. What these regions continue to offer is potentially intellectual work; many of the nation's top research universities remain in this region, attracting significant talent from around the U.S. and abroad. However, as Studebaker noted in our conversations, most of the work offered by universities is not intellectual work, but low-skill service work, and in limited supply. The high-value opportunities are also in jeoporady, as growing tuition costs and slashing of funding for public schools makes accessing and contributing to these instutions prohibitively expensive.

Conclusion

What my investigation clarified to me is that prospects of "reviving" industry are somewhat misguided. Firstly, the "de-industrialization" of the United States does not inherently mean that manufacturing itself has left the country. In fact, the shrinking number of jobs is a testament to the increased efficiency of American industry, which is still competitive on a global scale. However, my main takeaway from this analysis was that the stable, high-paying work offered by industry in the middle 20th century, which supported so many families, has disappeared in many regions, and has not been meaningfully replaced. During my analyses, I found that the down trend in employment in the industrial sector was co-integrated with inequality (measured by the Gini coefficient) at a significance level of 0.13. While this high cutoff does not present a cut-and-dry relationship, it does suggest that the kinds of work we have available today in many regions are more precarious than work offered in the past, and may be one of many factors driving socioeconomic inequality. As we enter a period where intellectual labor is able to be automated, it may be fruitful to look into the past at previous times, where technological advancement overhauled the kinds of work we consider valuable.