Robots and Non-Participation in the United States: Where Have All the Workers Gone?

Abstract

The rapid advances in automation technologies are disrupting labor markets at an unprecedented speed, contributing to the secular decline in US labor force participation and raising questions about where workers shift to when they leave the labor force. This article investigates the margins of adjustment of workers after being displaced by the introduction of industrial robots. Exploiting exogenous variation in the adoption of robots across local labor markets over time, the authors show that almost 8% of non-participants respond by enrolling in college, approximately 10% claim disability benefits, and nearly 40% retire early. The remaining non-participants rely on the income of their household members or live off their savings. These margins differ with the sociodemographic characteristics of individuals. Results also show that robots have worsened non-participants’ health, including self-reported health problems and hospitalizations related to severe mental disorders and substance abuse.

Keywords

industrial robots labor force participation margins of adjustment health

Advances in labor-replacing technologies are poised to shape the future of labor markets, fueling concerns that the automation of labor through robots and artificial intelligence is going to displace millions of workers in the years to come (Brynjolfsson and McAfee 2014; Susskind 2020). The displacement of workers is an issue with profound implications since information technologies are affecting labor markets at an unprecedented speed, forcing many to leave the labor force and seek alternative sources of income. Despite growing interest in the impact of new technologies on the labor market, little is known about how displaced workers adapt to automation. This article investigates the impact of the introduction of industrial robots on the margins of adjustment of non-participants in the United States between 1993 and 2014. We provide novel evidence about the relative contribution of the different margins of adjustment on the stock of non-participants, including education, disability insurance take-up, and early retirement, among others, as well as the demographic characteristics of non-participants. Our results highlight the need to design policies that facilitate the reallocation of the workforce to new jobs during the automation transition.

Industrial robots have been among the leading automation technologies of the past decades. The International Federation of Robotics (IFR) defines them as fully autonomous machines that can be programmed to perform various manual tasks without the intervention of a human worker. The astonishing advances in robotics in recent decades, along with a decline in their quality-adjusted prices, have led to the widespread adoption of robots in many countries and affected the demand for labor (Graetz and Michaels 2018). In the United States, the stock of industrial robots has increased from less than half a robot to more than two robots per thousand workers (approximately 180,000 units) between 1993 and 2014, displacing thousands of workers from their jobs (Acemoglu and Restrepo 2020).

Although technological advancements are generating millions of new job opportunities (Autor, Chin, Salomons, and Seegmiller 2024), these jobs frequently require collaboration between humans and machines, potentially leading to challenges for displaced workers who lack the required skills (Restrepo 2015). Displaced workers are therefore likely to become unemployed or drop out of the labor force altogether (Grigoli, Koczan, and Topalova 2020; Jaimovich, Saporta-Eksten, Siu, and Yedid-Levi 2021), unless they are endowed with easily redeployable human capital (Goldin and Katz 2010).

To identify the margins of adjustment of non-participants, we use individual-level information about the labor force status and detailed sociodemographic characteristics of the US population from the Census and the American Community Survey (ACS), including college enrollment, disability take-up, early retirement, income sources, and recent migration history. We match these data with industry-level data from the IFR about the adoption of industrial robots. We follow Acemoglu and Restrepo (2020) in constructing a plausibly exogenous measure of robot exposure at the local labor market level using a shift-share approach that interacts baseline industry employment shares within local labor markets, proxied by commuting zones (CZs; Tolbert and Sizer 1996), with the adoption of robots in the United States. Identification builds on the assumption that advances in robotics vary by industry and expose CZs differently based on their industrial composition of employment.

Literature

This article contributes to the growing literature that studies the disruptive labor market impacts of automation. Although to date there is no general consensus on its aggregate implications for labor markets, the literature agrees that technological progress is rapidly changing the demand for skills and the nature of work (Katz and Murphy 1992; Acemoglu 1999; Goos and Manning 2007; Goos, Manning, and Salomons 2009; Autor and Dorn 2013).

We focus on the stream of the literature that identifies a mismatch between the new skill requirements of jobs and the skill endowment of workers (Goldin and Katz 2010; Acemoglu and Restrepo 2018). We build on these results and investigate where workers shift to when they have been directly or indirectly displaced by the introduction of industrial robots. In particular, we build on the pioneering work of Acemoglu and Restrepo (2020), who showed that industrial robots have reduced aggregate employment and wages in the United States. We complement their work by providing a comprehensive assessment of the margins of adjustment of workers who leave the labor force because of robots, analyzing systematic differences in the adjustment across demographic groups.

The literature has identified school enrollment (Hickman and Olney 2011; Greenland and Lopresti 2016; Hunt 2017; Foote and Grosz 2020; Weinstein 2022; Di Giacomo and Lerch 2023; Ferriere, Navarro, and Reyes-Heroles 2023), disability take-up (Black, Daniel, and Sanders 2002; Autor, Dorn, and Hanson 2013; Maestas, Mullen, and Strand 2015), and early retirement (Bartel and Sicherman 1993; Peracchi and Welch 1994; Ahituv and Zeira 2011; Burlon and Vilalta-Bufi 2016) to be among the margins of adjustment of non-participants. Evidence is lacking, however, about whether these margins of adjustment are also relevant for individuals who leave the labor force when exposed to robots. We address this question and quantify the relative contribution of each margin of adjustment for the increase in non-participation caused by local robot exposure.

The article further contributes to the literature analyzing the effect of labor market shocks on migration flows (Blanchard and Katz 1992; Monras 2018; Foote, Grosz, and Stevens 2019; Yagan 2019; Howard 2020; Faber, Sarto, and Tabellini 2022) and health outcomes (Schaller and Stevens 2015; Frank, Glied, Marple, and Shields 2019; Lang, McManus, and Schaur 2019; Adda and Fawaz 2020; De Vries, Gentile, Miroudot, and Wacker 2020; Gihleb, Giuntella, Stella, and Wang 2022) to understand to what extent they are contributing to the increase in labor force non-participation in the United States.

The adverse effects of robots on employment are less visible in Europe. In these countries, the displacement of workers from low-skill (Graetz and Michaels 2018), routine task-intensive (De Vries et al. 2020), and manufacturing jobs (Dauth, Findeisen, Suedekum, and Woessner 2021) has often been compensated by the employment growth in other occupations and industries. However, it is important to account for the fact that even in European countries a fraction of the displaced workers may drop from the labor force because of a skill mismatch with the newly created jobs (Grigoli et al. 2020), suggesting that the findings of this article are also of interest for Europe.

Data

This section describes the main data sources along with a set of descriptive statistics (see Table 1).

Table 1.

Descriptive Statistics: Margins of Adjustment of Non-participants

	US robot exposure1993–2014
	All	Q1	Q2	Q3	Q4
	(1)	(2)	(3)	(4)	(5)
Men
Non-participation rate (1990)	11.1 (2.81)	12.3 (4.03)	11.8 (2.78)	10.1 (2.24)	11.5 (2.50)
Students	0.93 (0.41)	0.96 (0.51)	1.07 (0.44)	0.91 (0.36)	0.79 (0.35)
Disability beneficiaries	1.35 (0.67)	1.53 (0.91)	1.36 (0.69)	1.20 (0.58)	1.52 (0.63)
Early retirees	3.55 (1.57)	3.89 (1.37)	3.66 (1.21)	3.06 (0.90)	4.09 (1.04)
Household income	5.34 (1.09)	5.56 (1.32)	5.63 (1.21)	5.11 (0.91)	5.24 (1.01)
Private income	4.11 (0.97)	4.56 (1.45)	4.47 (0.96)	3.83 (0.74)	3.91 (0.82)
Δ Non-participation rate (1990–2014)	4.37 (2.11)	3.83 (2.76)	3.78 (2.02)	4.36 (1.91)	5.37 (1.87)
Δ Non-participation rate (1970–1990)	3.15 (2.01)	2.74 (2.65)	3.08 (1.87)	2.71 (1.49)	4.14 (2.28)
Women
Non-participation rate (1990)	30.4 (4.81)	31.5 (7.41)	31.3 (4.69)	29.0 (4.29)	30.09 (3.81)
Δ Non-participation rate (1990–2014)	−2.76 (2.91)	−3.54 (3.11)	−2.95 (3.16)	−2.11 (2.59)	−3.24 (2.80)

Notes: Averages of the main variables used in the analysis, weighted by commuting zone (CZ) population in 1990. Column (1) reports averages over all 722 CZs in the sample. Columns (2) to (5) split the sample into four quartiles, accounting for a labor market’s exposure to robots between 1993 and 2014. Since we use data on labor market outcomes from the 1990 Census, following Acemoglu and Restrepo (2020), we rescale the 1990–2000 period to a 7-year equivalent change to achieve comparability across sample periods. Labor force participation indicators are expressed in terms of the working-age population of reference in the CZ. Standard deviations are in parentheses.

Industrial Robots

We obtain data on the adoption of robots from the International Federation of Robotics (IFR). The IFR is a survey that collects data about shipments and operational stocks of industrial robots by country, industry, and year ranging back to 1993 for 50 countries. The IFR defines an industrial robot as an “automatically controlled, reprogrammable, multipurpose manipulator, programmable in three or more axes, which can be either fixed in place or mobile for use in industrial automation applications” (IFR 2018: 29). That is, industrial robots are machines that can be programmed to autonomously perform several manual tasks (e.g., assembly, material handling, packing, and welding) without the intervention of a human worker. They are often designed as robotic arms and do not include conveyor belts, cranes, or elevators, since these machines do not meet the above requirements.

The IFR breaks down the stock of operational robots according to the International Standard Industrial Classification (ISIC), Fourth Revision, and provides consistent data for six broad industries outside the manufacturing sector (agriculture, forestry, and fishing; construction; education, research, and development; mining; utilities; and other non-manufacturing branches, such as services) and 13 industries within the manufacturing sector (automotive; basic metals; electronics; food and beverages; industrial machinery; metal products; minerals; paper and printing; plastics and chemicals; textiles; wood and furniture; other transport equipment [e.g., airplanes, locomotives, and ships]; and other manufacturing branches). These data have been widely used in the literature; however, they also include some limitations, as presented in Online Appendix A2. (Hereafter, numbering for Online Appendix material is prefaced with an “A.”)

The US stock of industrial robots has increased by approximately 1.5 robots per thousand workers—a fivefold increase, or roughly 180,000 units, compared to its 1993 level. According to the IFR (2018), this number is expected to grow even more in the future. Industrial robots are adopted primarily in a subset of manufacturing industries, such as the automotive, electronic, plastic and chemical, and metal production industries (see Table A1 for details). This characteristic exposes the Midwest of the country significantly more to their adoption, especially the local labor markets of the Rust Belt (including the states of Indiana, Michigan, and Ohio), given their specialization in industrial manufacturing industries (see Figure 1).

Figure 1.

US Robot Exposure at the CZ Level

Margins of Adjustment

To measure long-term changes in local labor market outcomes contemporaneous to the introduction of industrial robots, we obtain data from the Integrated Public Use Microdata Series (IPUMS) for the decennial Census for 1970, 1980, 1990, and 2000, and the ACS for 2007 and 2014 (Ruggles et al. 2019). These data sets are from repeated cross-sectional surveys that include between 1% and 5% of the US population and provide a comprehensive set of information at the individual and household level, including the employment status, sociodemographic characteristics, income sources, and the place of residence of households and their members.

An individual is considered to be out of the labor force if she is not employed and is not looking for work at the time the survey is conducted. Individuals are also asked about recent migration, school attendance, receipt of Social Security income, and their household role. We use these data alongside their labor force status to determine the adjustments among non-participants. We focus on the non-institutionalized population between 25 and 64 years of age. These individuals are above the usual full-time school age and below the full retirement age.

We aggregate individual-level data at the labor market level using 722 Commuting Zones (CZs) that cover the entire US mainland and act as proxies of local labor markers (Tolbert and Sizer 1996). This aggregation allows us to build a measure of the labor force non-participation rate at the local labor market level:

N P_{c, t} = \frac{L_{c, t}}{N_{c, t}}

(1)

where $L_{c, t}$ is the number of non-participants and $N_{c, t}$ is the working-age population in CZ c in year t. We decompose the non-participation rate into population subgroups to identify the margins of adjustment of workers, when they are out of the labor force:

N P_{c, t} = \sum_{m \in M} {NP}_{c, t}^{m} = \frac{1}{N_{c, t}} \sum_{m \in M} L_{c, t}^{m}

(2)

where ${NP}_{c, t}^{m}$ is the share of the population that is non-participant with margin of adjustment m (school enrollment, disability take-up, early retirement, household income, private income, and idle status).

Table 1 shows that while the US has been experiencing a reduction in female labor force non-participation over the last decades (–2.8%), labor force non-participation of men increased from 11.1% in 1990 to 15.5% in 2014. When decomposing non-participants by their margin of adjustment, we find that approximately 8% of the non-participants are enrolled in college (0.93/11.1), one out of 10 receives disability benefits from Social Security Disability Insurance (SSDI), and almost one-third receives some form of retirement income (pension plan income or Social Security early retirement benefits). We also find that approximately half of the non-participants rely on the income of their household members and more than one-third rely on private savings from previously earned income. Note that, although college enrollment, disability take-up, and early retirement are mostly mutually exclusive, this is not the case for non-participants who rely on their household income or who rely on private savings. For instance, a 62-year-old displaced worker who leaves the labor force to retire early is unlikely to enroll in college. It might be the case, however, that a college student in his late twenties lives with his parents and relies on their income. Therefore, the shares of non-participants in the individual margins of adjustment do not necessarily add up to total non-participation.

Columns (2) to (5) split the sample into four quartiles based on a CZ’s exposure to industrial robots and provide means of the main variables used in the analysis for the respective quartile. Results show that the increase in labor force non-participation of men is largest in CZs with heavy utilization of industrial robots (5.37 percentage points), whereas we observe no particular correlation between the decline in non-participation of women and robot exposure. Note, too, that male non-participation was already on an increasing trend since the 1970s in more exposed areas and, as illustrated in Table A2, these CZs are also highly exposed to import competition from China during our sample period. We account for this threat to identification in developing our empirical strategy.

As outlined, a potential margin of adjustment against robot exposure is disability enrollment. We are unable to determine, however, whether an increase in disability take-up is associated with a deterioration in the health condition of displaced workers or if individuals are misusing the SSDI as a sort of permanent unemployment insurance (Ford 2015). We investigate this question by supplementing labor market data with indicators of mental and physical health conditions at the CZ level from the Behavioral Risk Factor Surveillance System (BRFSS) of the Centers for Disease Control and Prevention (CDC) and the National Inpatient Sample (NIS) of the Healthcare Cost and Utilization Project (HCUP). We provide additional details on the data in Appendix A2.

Empirical Strategy

We estimate the effect of robots on the margins of adjustment of non-participants using a stacked first-difference specification with 722 CZs and three time periods (1993–2000, 2000–2007, 2007–2014).¹ The key estimating equation is given by:

Δ {NP}_{c, (t_{0}, t_{1})}^{m} = β^{m} US robot exposur e_{c, (t_{0}, t_{1})} + X'_{c, (t_{0}, t_{1})} θ + δ_{(t_{0}, t_{1})} * μ_{d (c)} + ε_{c, (t_{0}, t_{1})}

(3)

where ${NP}_{c, t}^{m}$ is the share of the population that is non-participant with margin of adjustment m. To account for potential sources of bias that might confound the estimates of the labor market effect of robots, Equation (3) also includes a set of covariates, such as pre-existing trends in US labor force participation and manufacturing employment, a measure of the China trade shock (Autor et al. 2013), and a vector of start-of-sample-period regional characteristics and economic variables (gender, race, ethnicity, education, age, state of birth, population, industry, and occupation composition). We keep CZ characteristics constant at their 1990 levels to avoid contamination by endogenous adjustments in the structure of local labor markets in response to robot adoption. Moreover, we include year fixed effects $δ_{(t_{0}, t_{1})}$ and US Census division fixed effects $μ_{d (c)}$ . We provide additional details about covariates in Table A2.

We build a measure of robot exposure at the CZ level using a shift-share approach. We follow Acemoglu and Restrepo (2020) and match industry-level data from the IFR with employment counts from the Census:

US robot exposur e_{c, t} = \sum_{j \in J} ℓ_{c, j}^{90} [\frac{R_{j, t}^{US} - R_{j, t - 1}^{US}}{L_{j, 90}^{US}} - g_{j, (t, t - 1)}^{US} \frac{R_{j, t - 1}^{US}}{L_{j, 90}^{US}}]

(4)

The term in brackets (shift-component) is a measure of industrial robot density, computed as the US-wide change in the stock of robots in industry j ∈ J, relative to its workforce in 1990, and adjusted for the adoption of robots that is driven by overall industry output growth, $g_{j, (t, t - 1)}^{US} = Δ \ln (Y_{j, t}^{US})$ , where $Y_{j, t}^{US}$ is output in industry j in year t in the United States. The industry-level shock is apportioned across local labor markets using CZs’ industry employment shares, $ℓ_{c, j}^{90} = \frac{L_{c, j}^{90}}{L_{c}^{90}}$ . The baseline employment shares are kept constant to avoid endogeneity and serial correlation concerns across periods of our stacked first-difference specification.

Identification builds on the assumption that advances in robotics vary by industry and expose CZs differently depending on the industrial composition of employment. The adoption of robots, however, is likely to be correlated also with local labor market conditions that affect the demand for labor and, therefore, labor force participation rates. For instance, positive demand shocks might induce US firms to raise both capital and employment, biasing upward the estimates of the effect of robots on labor demand.

To address the endogeneity concern and identify robot adoption that is driven by the supply channel, we instrument the shift-component of Equation (4) using contemporaneous changes in the stock of robots in seven European countries with a comparable adoption of robots as the United States:

EU 7 robot exposur e_{c, t} = \sum_{j \in J} ℓ_{c, j}^{70} \frac{1}{7} \sum_{i \in EU 7} [\frac{R_{j, t}^{i} - R_{j, t - 1}^{i}}{L_{j, 90}^{i}} - g_{j, (t, t - 1)}^{i} \frac{R_{j, t - 1}^{i}}{L_{j, 90}^{i}}]

(5)

where $R_{j, t}^{i}$ is the stock of robots in country $i \in EU 7$ at time t in industry j. The EU7 countries include Denmark, Finland, France, Italy, Spain, Sweden, and the United Kingdom. The share-component uses (plausibly exogenous) employment shares from 1970 to focus on the industrial composition of employment that precedes the introduction of industrial robots, which started in the 1980s (Acemoglu and Restrepo 2020).

Our instrumental variables (IV) strategy aims to identify the labor market effects of exogenous improvements in robotics available to US firms. The strategy relies on the assumption that the adoption of robots in European countries is positively related to the adoption of robots in the United States but is unrelated to domestic labor market conditions. The first condition can be easily verified (see first-stage results in Table 2), while we discuss possible threats to identification related to the exclusion restriction in Appendix A4. Reassuringly, the empirical results are robust and survive a variety of tests, including checks for international product market competition from Europe, industry trends (Goldsmith-Pinkham, Sorkin, and Swift 2020), and labor market pre-trends.

Table 2.

Robots and Non-participation

	Working-age population		Men	Women
	(1)	(2)	(3)	(4)
Panel A: OLS results
US robot exposure	0.195** (0.076)	0.093*** (0.033)	0.179*** (0.035)	0.008 (0.035)
Panel B: IV results
US robot exposure	0.224** (0.093)	0.192*** (0.063)	0.345*** (0.073)	0.046 (0.061)
Panel C: First-stage results
EU7 robot exposure	0.773*** (0.055)	0.738*** (0.050)	0.738*** (0.050)	0.738*** (0.050)
Observations	2,166	2,166	2,166	2,166
Covariates
Division FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Pre-trends		Yes	Yes	Yes
Chinese imports		Yes	Yes	Yes
Demographics		Yes	Yes	Yes
Industries		Yes	Yes	Yes
Occupations		Yes	Yes	Yes

Notes: Instrumental variable (IV) estimates of the effect of US robot exposure on the non-participation rate at the CZ level. Changes are expressed in percentage points of the working-age population and are multiplied by 100. Independent variables are standardized to have mean zero and standard deviation of 1. There are three time periods and 722 CZs. Column (1) includes year dummies, nine census divisions, and their interactions. Column (2) also includes changes in the non-participation rate and in the manufacturing employment rate between 1970 and 1990. Column (3) includes a measure of exposure to Chinese imports. Column (4) also controls for demographic (share of individuals aged between 25 and 34 years, 35 and 44 years, 45 and 54 years, the share of white men, Blacks, Hispanics, women, individuals born in the same US state as their current residency, and individuals with less than a college degree, and logarithmic population), industry (shares of employment in the construction, manufacturing, mining, research, service, and utilities sectors) and occupation (share of offshorable and routine task-intensive occupations) characteristics of CZ in 1990. Columns (5) and (6) report estimates of the effect for men and women separately. Standard errors are robust against heteroskedasticity and allow for clustering at the state level. Regressions are weighted by CZ population in 1990. EU7 countries include Denmark, Finland, France, Italy, Spain, Sweden, and the United Kingdom; FE, fixed effects; OLS, ordinary least squares.Coefficients with ***, **, and * are significant at the 1%, 5%, and 10% confidence level, respectively.

Results

Margins of Adjustment of Non-participants

We start the analysis by presenting the impact of robot exposure on US labor force non-participation. Table 2 reports ordinary least squares (OLS) and IV estimates, as well as first-stage results. The coefficients are standardized and represent the estimated labor market effect of a one standard deviation increase in robot exposure. Regressions are weighted by the 1990 CZ population and standard errors allow for arbitrary clustering at the state level.

Column (1) presents a baseline specification that accounts only for division-specific business cycles. The following columns include additional covariates that might confound the labor market effect of robots. The absolute size of OLS estimates is smaller than that of IV estimates across all specifications since US robot adoption is likely to be correlated with omitted demand shocks that bias the estimates toward zero.

In line with the findings of Acemoglu and Restrepo (2020), our results show that robots have an adverse effect on US labor markets. They suggest that each robot reduces employment by approximately six workers in a local labor market. We find that, on average, a one standard deviation increase in robot exposure decreases labor force participation by 0.192 percentage points, which translates into a decrease in the local labor force of approximately four workers for each additional robot.

In addition to Acemoglu and Restrepo (2020), we account for the effect of robots during and after the Great Recession as well. The results are relatively similar with and without accounting for the 2007–2014 period (see Table A14), suggesting that the increase in non-participants accounts for more than half of the employment effects of robots.

Table 2, columns (3) and (4) analyze the impact of robot exposure on non-participation by gender. Results show that the effect is fully driven by men. Even though Acemoglu and Restrepo (2020) and Lerch (2025) found that robot exposure decreases the employment rate of both men and women, only men respond by leaving the labor force. The magnitude of this result suggests that the introduction of robots between 1993 and 2014 has had an impact on the labor force non-participation rate of men that is equivalent to approximately 18% of its secular increase.² The following analysis focuses on the characteristics of displaced male workers after they have left the labor force.

Table 3 decomposes the effect of robots on non-participation of men from Table 2 into the main margins of adjustment of non-participants. In particular, we focus on college enrollment, disability insurance take-up, early retirement, household income, private savings, and idle individuals. We focus on these margins as they have been identified as being the most relevant for non-participants (Deshpande and Lockwood 2022) and are detected by Census and ACS data. The table reports the estimated coefficients and standard errors using Equation (3), as well as the share of each margin of adjustment on the effect on non-participation of men from Table 2, panel B, column (3) (0.345).

Table 3.

Margins of Adjustment of Non-Participants

	Main margins of adjustment			Alternative sources of income
	School enrollment	Disability take-up	Early retirement	Private income	Household income
	(1)	(2)	(3)	(5)	(6)
US robot exposure	0.026*** (0.010)	0.036** (0.015)	0.135*** (0.019)	0.102 (0.063)	0.173*** (0.030)
Relative contribution to effect on non-participation (%)	7.5	10.4	39.1	29.6	50.1
Observations	2,166	2,166	2,166	2,166	2,166
Covariates	Yes	Yes	Yes	Yes	Yes

Notes: The coefficient and the relative contribution of each margin of adjustment to the increase in non-participation due to the introduction of robots. Specifically, “Relative contribution to effect on non-participation” reports for each column the ratio between the coefficient from a regression of the change in NP^m on robot exposure (and covariates) and the total impact on non-participation of men from the instrumental variables (IV) regression in Table 2, panel B, column (3). Standard errors are robust against heteroskedasticity and allow for clustering at the state level. Regressions are weighted by commuting zone (CZ) population in 1990.Coefficients with ***, **, and * are significant at the 1%, 5%, and 10% confidence level, respectively.

As we anticipated in the Data section, some of the margins of adjustment might not be mutually exclusive. For instance, non-participants who rely on their household income or who rely on private savings might also be enrolled in college, be disability income beneficiaries, or be early retirees. Therefore, the estimated effect on the margins of adjustment of non-participants does not necessarily add up to the overall effect on non-participation from Table 2.

Table 3, column (1), shows that 7.5% of the non-participants are enrolled in college. These individuals are either delaying their labor market entry or they are temporarily leaving the labor force to invest in additional human capital to increase their labor market competitiveness, either through a GED, college, or a graduate degree.³ Table A3 shows that the effect is mostly driven by individuals below the age of 35, implying that the effect is predominantly confined to individuals in their early prime age. This finding is consistent with the result of Di Giacomo and Lerch (2023), who illustrated the surge in enrollment in US community colleges resulting from robot adoption. Similarly, it echoes the findings of Dauth et al. (2021), indicating that in Germany, robots have bolstered the ratio of college graduates while reducing the proportion of workers who have completed an apprenticeship.⁴

Column (2) shows that 10.4% of non-participants receive disability income from SSDI. This result is in line with previous literature that shows that labor demand shocks and poor labor market conditions are significantly affecting disability take-up in the United States (e.g., Black et al. 2002; Autor et al. 2013; Maestas et al. 2015). Two channels may fuel the rise in disability take-up of prime-age workers. First, SSDI acts as a sort of permanent unemployment insurance against the risks of automation and worse future job prospects (Ford 2015). Second, adverse shocks affect the health condition of workers who, as a consequence, become eligible for disability benefits (Rege, Telle, and Votruba 2009; Sullivan and Von Wachter 2009; Schaller and Stevens 2015; Adda and Fawaz 2020).

In line with this framework, the fraction of disability claims related to difficult-to-verify impairments has risen substantially over the past four decades (Autor and Duggan 2003; Liebman 2015), suggesting that SSDI could be misused as a sort of permanent unemployment insurance against adverse shocks (Ford 2015; Deshpande and Lockwood 2022). Other empirical work argues that poor labor market conditions could also directly affect workers’ health (Hollingsworth, Ruhm, and Simon 2017), and that poor health is one of the main reasons individuals do not join the labor force (Parsons 1980a, 1980b; Krueger 2017). Evidence shows that the high job insecurity during periods of poor labor market conditions and the exposure to trade shocks during the 2000s led to a substantial increase in mental health problems among working-age Americans (Lang et al. 2019; Adda and Fawaz 2020; Pierce and Schott 2020), with the strongest deterioration in health conditions among white males (Case and Deaton 2015, 2017). While we cannot exclude that non-participants are misusing SSDI as insurance against adverse labor market shocks (Ford 2015; Deshpande and Lockwood 2022), in the upcoming section on health outcomes we provide suggestive evidence that the increase in disability take-up might be associated with a worsening of health outcomes among non-participants.

Table 3, column (3) shows that 39.1% of non-participants retire early. This result confirms previous findings that highlighted that workers older than 55 are eligible for some form of retirement income and retire early when facing labor-displacing technological progress (Bartel and Sicherman 1993; Peracchi and Welch 1994; Ahituv and Zeira 2011; Burlon and Vilalta-Bufí 2016). Adding up the estimates from columns (2) and (3), we find that approximately half of the non-participants leave the labor force permanently, as only a few disability beneficiaries and retirees rejoin the labor force in the future (Parsons 1980a, 1980b; Krueger 2017).

Our results suggest that nearly 6 out of 10 non-participants are either enrolled in school, receiving disability insurance, or retiring early. This finding, however, raises the question of where the remaining 40% of individuals, who are not participating in the labor force due to robots, shift to when they leave the labor force. We address this question in the subsequent columns of Table 3. Columns (4) and (5) show the results on the share of non-participants who rely on income from their household members (including partners, parents, adult children, or siblings) as well as for the share of non-participants who rely on the income they earned in the past year. We find that half of the non-participants live on their household members’ income and almost one-third have earned some income in the previous year. Accounting only for the margins of adjustment of non-participants who are not enrolled in school, who do not receive any disability benefits, and who are not retired, we instead find that 25.1% of them rely on the income of other household members, and 14.3% live off their savings. These non-participants are more likely to (re)join the labor force in future years, as are non-participants who are enrolled in school, compared to disability insurance beneficiaries and early retirees. Finally, we find that less than 5% of non-participants are idle and do not fall into any of these categories.⁵

Demographics of Non-participants

To understand the underlying causes that drive individuals out of the labor force and into a particular margin of adjustment, we must account for their sociodemographic characteristics. For this purpose, we break down the estimated effect of robots on non-participation of men by race and ethnicity (whites and racial/ethnic minorities), age groups (25–34, 35–44, 45–54, and 55–64 years), and education levels (with and without a college degree). This breakdown is achieved by using a decomposition exercise analogous to Equation (2), but for demographic groups $g \in G$ instead of margins of adjustment $m \in M$ :

{NP}_{c, t} = \sum_{g \in G} {NP}_{c, t}^{g} = \frac{1}{N_{c, t}} \sum_{g \in G} L_{c, t}^{g}

(6)

where ${NP}_{c, t}^{g}$ is the share of the population that is non-participant with demographic characteristics g. Whites include individuals who, when asked about their race and ethnicity in the Census, report to be “White” and of non-“Hispanic, Latino, or Spanish” origin. Racial and ethnic minorities include all demographic groups who are either not white (e.g., Blacks, Asians, American Indian, and Alaska natives), or are of Hispanic, Latino, or Spanish origin.

Table 4, panel A, illustrates the relative contribution of each group to the aggregate effect of robots on non-participation. Results show that the increase in non-participation is fueled in similar proportions by whites and racial/ethnic minorities and that it is caused predominantly by less-educated workers. This result is in line with previous literature, which argues that workers with a college degree are often employed in occupations that require the use of communication and interpersonal skills that are more difficult to automate (Acemoglu and Autor 2011). However, this effect is also mechanically small due to the relatively low share of the population with a college degree (below 32% between 1990 and 2014). The same reasoning applies to racial and ethnic minorities, who make up less than 36% of the population.

Table 4.

Non-participation by Demographics

	College degree		Race and ethnicity		Age
	Yes	No	White	Non-white	25–34	35–44	45–54	55–64
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Panel A: Breakdown of non-participation rate by demographics
US robot exposure	0.033*** (0.009)	0.312*** (0.068)	0.179*** (0.021)	0.167** (0.076)	0.048 (0.036)	0.071*** (0.022)	0.075*** (0.020)	0.151*** (0.020)
Panel B: Non-participation rate within demographic groups
US robot exposure	0.124*** (0.034)	0.436*** (0.100)	0.247*** (0.038)	0.631*** (0.126)	0.289*** (0.089)	0.227*** (0.080)	0.216*** (0.080)	0.475*** (0.078)
Observations	2,166	2,166	2,166	2,166	2,166	2,166	2,166	2,166
Covariates	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Notes: Instrumental variable (IV) point estimates of the effect of US robot exposure on non-participation by race/ethnicity, age, and education (ΔNP^g_c,t). Panel A decomposes the effect estimated in Table 2 by demographic group, and panel B analyzes the effect within each group (Δ ${NP}_{c, t}^{φ}$ ). Regressions include the full battery of controls from our preferred specification and are weighted by commuting zone (CZ) population in 1990. Standard errors are robust against heteroskedasticity and allow for clustering at the state level.Coefficients with ***, **, and * are significant at the 1%, 5%, and 10% confidence level, respectively.

Finally, the effect of robots on labor force dropouts increases with age (though not monotonically among each group). Individuals between 55 and 64 years account for 40% of the overall increase in non-participation. Again, this result is likely to be subject to underlying composition effects unrelated to robot adoption, such as population aging, and differences in the size of the population attributable to older individuals having lower labor force participation rates.

To account for these effects, we further analyze changes in the non-participation rate within population subgroups. This analysis is achieved by reconstructing our dependent variables, the change in non-participants of demographic group g, $L_{c, t}^{g}$ , divided by the population count of demographic group g, $N_{c, t}^{g}$ . As a result, Equation (7) provides a measure of the non-participation rate within demographic groups, ${NP}_{c, t}^{φ}$ :

{NP}_{c, t}^{φ} = \frac{L_{c, t}^{g}}{N_{c, t}^{g}}

(7)

The impact of robot exposure on demographic-specific non-participation rates obtained using the regression analysis presented in Equation (3) is illustrated in Table 4, panel B. When accounting for the relative population size of demographic groups, the adverse effect of robots on the non-participation rate of racial and ethnic minorities stands out. The increase is more than twice that of whites. This result follows from the fact these workers are often over-represented in blue-collar jobs with a high workload of manual tasks, which are more susceptible to automation through industrial robots (Lerch 2025).⁶

Migration

The literature on regional shocks has identified workers’ geographic mobility as one of the key channels through which labor markets adjust to local economic shocks (Blanchard and Katz 1992; Howard 2020). A recent study by Faber et al. (2022) has shown that the introduction of robots has a negative effect on the population by reducing the inflow of prospective migrants in the local labor market. This effect is likely to affect the composition of the local labor force. In particular, the reduced inflow of workers could mechanically increase the share of non-participants.

Table 5 confirms these results, showing that robots reduce the inflow of migrants from different US states. While they do not affect the outflow of migrants, they also do not affect the inflow of migrants from other CZs in the same state. Moreover, column (5) shows the impact of robots on non-participation among individuals who live in their state of birth.⁷ The estimate is not economically nor statistically different from our preferred specification’s estimate.

Table 5.

Migration Flows and Place of Birth

	Outflows		Inflows		Non-participation
	Same state	Different state	Same state	Different state	Same state
	(1)	(2)	(3)	(4)	(5)
US robot exposure	−0.009 (0.103)	0.073 (0.123)	−0.041 (0.079)	−0.241*** (0.048)	0.299*** (0.049)
Observations	2,166	2,166	2,166	2,166	2,166
Covariates	Yes	Yes	Yes	Yes	Yes

Notes: Instrumental variable (IV) estimates of the effect of US robot exposure on migration flows across commuting zones (CZs), both within and outside of the state, as well as non-participation rates by state of birth. We compute outflows (inflows) in columns (1) and (2) (columns (3) and (4)) as the share of individuals who migrated away from (to) a CZ. The denominator equals the population in their CZ of residence before (after) moving. Note the information about individuals’ migration status changes over time. In particular, the Census asks whether a person changed his residence in the previous 5 years, while the ACS asks whether a person changed his residence in the previous year. We follow Molloy, Smith, and Wozniak (2011) in building measures of 5-year migration flows from the American Community Survey (ACS) by using four times the annual migration flow of a CZ. Column (5) is computed as the number of non-participants in a CZ born in the same state as their current residency divided by the population of the CZ born in the same state as their current residency. All regressions include the full battery of controls from our preferred specification.Coefficients with ***, **, and * are significant at the 1%, 5%, and 10% confidence level, respectively.

These results suggest that the effect of robot exposure on non-participation is more likely to reflect individuals’ change in employment status rather than changes in the composition of the labor force through migratory flows. We address this point further in the next section.

Change of Employment Status

Here, we shed light on whether the introduction of robots has affected the flow of non-participants in and out of employment. The results are summarized in Table 6. Column (1) shows the effect of robots on non-participation of individuals who moved from employment to non-participation. These individuals have been (directly or indirectly) displaced from the labor market and, as a consequence, left the labor force. Column (2) shows the impact on non-participation of individuals who were already non-participants and those who moved from unemployment to non-participation. These non-participants were unable or unwilling to seek employment when exposed to robots in their CZ of residence.

Table 6.

Non-participants Previously Employed

	Worked past year
	Yes	No
	(1)	(2)
US robot exposure	0.242*** (0.065)	0.103*** (0.032)
Observations	2,166	2,166
Covariates	Yes	Yes

Notes: Instrumental variable (IV) point estimates of the effect of US robot exposure on non-participation by employment status in the previous year. Regressions include the full battery of controls from our preferred specification and are weighted by commuting zone (CZ) population in 1990. Standard errors are robust against heteroskedasticity and allow for clustering at the state level. Coefficients with ***, **, and * are significant at the 1%, 5%, and 10% confidence level, respectively.

We find that approximately 70% of the increase in non-participants observed in Table 2 (0.242/0.345) consists of individuals who were employed in the previous year. This result shows that robots have displaced from the labor market a significant fraction of workers. The remaining 30% consists of non-participants who were not employed in the past year.

Health Outcomes

Next, we analyze how robot exposure affects the health outcomes of non-participants. Table 7 reports estimates of the labor market effect of robot exposure on self-reported health conditions. We find that robot exposure increases the share of non-participants who report suffering from persistent physical health problems. Although the estimated impact of robots on the share of individuals who report fair or poor health and persistent mental problems is positive, these effects are not statistically significant.⁸

Table 7.

Self-perceived Health

	No good health	Physical problems	Mental problems
	(1)	(2)	(3)
US robot exposure	0.083 (0.081)	0.155*** (0.046)	0.049 (0.054)
Observations	2,166	2,166	2,166
Covariates	Yes	Yes	Yes

Notes: Instrumental variable (IV) estimates of the effect of US robot exposure on the change in self-reported health. Changes are expressed in percentage points of non-employed individuals and are multiplied by 100. Column (1) reports individuals that report fair or poor health. Columns (2) and (3) report individuals who suffered from physical or mental problems in more than 14 days of the previous 30 days. All regressions include the full battery of controls from our preferred specification. Coefficients with ***, **, and * are significant at the 1%, 5%, and 10% confidence level, respectively.

Additionally, Table A4 shows that exposure to robots increases regular drinking behavior and decreases physical activity, raising the share of non-employed individuals who suffer from obesity, in particular among whites. Moreover, Table A5 shows that robot exposure has no adverse effects on the self-perceived health conditions of employed workers, suggesting that the impact of robots on individuals’ health depends on their employment status.⁹

The adoption of robots might affect workers’ health conditions in two (not mutually exclusive) ways. First, workers who already suffer from health problems are less productive and are, therefore, more likely to be displaced by robots. These workers mechanically increase the share of non-employed individuals who suffer from health problems (Frank et al. 2019). Second, the health condition of displaced workers deteriorates after a job loss (Sullivan and Von Wachter 2009). This mechanism is likely to be reinforced by the fact that for many US workers, losing their jobs is associated with losing health insurance (Schaller and Stevens 2015). Indeed, Table A4 shows that robot exposure decreases the share of non-employed whites with health care coverage. This, in turn, increases the share of individuals who cannot afford a doctor visit for cost reasons and who had no medical checkup in the previous two years. The inability to be treated by a doctor in case of need may substantially worsen individuals’ health conditions which, if neglected for too long, may become severe and eventually lead to a disability. If this is the case, we should also observe a rise in hospitalization rates with acute health problems in areas that are more exposed to robots.

Table 8 summarizes the results on the effect of robots on the share of hospital admissions with a length of stay of more than seven days (they are most likely to be related to a severe health condition) and a disability-related diagnosis.¹⁰ We use shares in terms of hospital admissions (rather than in per capita terms) because the sample of US community hospitals is changing from year to year, making a comparison of per capita values across years uninformative.¹¹

Table 8.

Hospitalizations

	All admissions	All disability-related admissions	Arthritis or rheumatology	Back or spine problems	Circulatory system diseases	Respiratory system diseases	Mental disorders	Diabetes
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
US robot exposure	1.023** (0.455)	1.145** (0.478)	0.440* (0.231)	−0.325 (0.495)	0.544 (0.411)	0.484* (0.235)	1.877*** (0.609)	0.282 (0.215)
Observations	2,166	2,166	2,166	2,166	2,166	2,166	2,166	2,166
Covariates	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Notes: Instrumental variable (IV) estimates of the effect of US robot exposure on the share of hospital admissions with a length of stay of more than seven days and a diagnosis related to a cause of disability. Changes are expressed in percentage points of commuting zone (CZ) hospital admissions multiplied by 100. Column (1) reports all admissions with a length of stay of more than seven days. Column (2) reports all disability-related admissions with a length of stay of more than seven days. Columns (3) to (8) report admissions diagnosed with arthritis or rheumatology, back or spine problems, circulatory system diseases, respiratory system diseases, mental disorders, and diabetes, with a length of stay of more than seven days. All regressions include the full battery of controls from our preferred specification.Coefficients with ***, **, and * are significant at the 1%, 5%, and 10% confidence level, respectively.

Results show that robots increase the relative share of admissions with an acute health condition and, in particular, that they increase the relative share of admissions diagnosed with mental disorders.¹² Robot exposure also increases the relative share of severe hospital admissions related to substance abuse, such as alcohol, drug (in particular opioid), and tobacco abuse (see Table A9). These diagnoses are often related to mental disorders.¹³

These findings provide suggestive evidence that the increasing disability take-up in CZs that are more exposed to robots might be associated with a deterioration in individuals’ health condition.

Conclusions

This article investigates the margins of adjustment of US workers after they were displaced by industrial robots between 1993 and 2014. By exploiting plausibly exogenous variation in robot exposure across local labor markets over time, we find that robots have an adverse effect on the labor force participation of men, but not of women. The impact is concentrated among workers without a college degree, it increases with age, and is strongest among racial and ethnic minorities.

Overall, each additional robot drives four workers out of the labor market. The margins of adjustment for these individuals include college enrollment (7.5%), disability take-up (10.4%), and early retirement (39.1%). Moreover, non-participants (in particular those who do not fall in any of these categories) often rely on income of their household members or live off their savings. The rising enrollment in SSDI coincides with robots negatively affecting workers’ self-reported health, and an increasing share of hospital admissions diagnosed with acute mental disorders.

The rapid progress in automation technologies is likely to intensify the mismatch between the skill requirements of jobs and the skills of workers. As a result, workers may increasingly drop out of the labor force to seek alternative sources of income, unless they are endowed with redeployable human capital. Our findings highlight the need for policymakers to design policies that facilitate the transition of the workforce to new jobs, and to improve the interplay between workers and machines through education and on-the-job training.

Supplemental Material

sj-pdf-1-ilr-10.1177_00197939251373051 – Supplemental material for Robots and Non-participation in the United States: Where Have All the Workers Gone?

Supplemental material, sj-pdf-1-ilr-10.1177_00197939251373051 for Robots and Non-participation in the United States: Where Have All the Workers Gone? by Giuseppe Di Giacomo and Benjamin Lerch in ILR Review

Footnotes

Acknowledgements

We are extremely grateful to Giovanni Pica for his guidance in this project. We thank Daron Acemoglu, David Autor, David Dorn, Marius Faber, Lorenz Kueng, Fabrizio Mazzonna, Matt Notowidigdo, Luigi Pistaferri, Pascual Restrepo, Isaac Sorkin, and the participants of the USI Brown Bag Seminars; the 2021 Young Swiss Economists Meeting; the 3rd QMUL Economics and Finance Workshop; as well as the faculty of the Department of Economics at Università della Svizzera italiana and Stanford University for valuable feedback. We also thank Pascual Restrepo for sharing their data.

Giuseppe Di Giacomo gratefully acknowledges financial support from the Swiss National Science Foundation grant No. 100018_197802.

The views expressed in this article are those of the authors and do not necessarily reflect the official position of the Swiss Federal Finance Administration, the Federal Department of Finance, or the Federal Council.

ORCID iDs

Giuseppe Di Giacomo

Benjamin Lerch

An Online Appendix is available at .

For general questions as well as for information regarding the data and/or computer programs, please contact the authors at benjamin.lerch@efv.admin.ch; giuseppe.digiacomo@usi.ch.

1

Since we use data on labor market outcomes from the 1990 Census and health outcomes from the BRFSS and NIS that range only up to 2011, following , we rescale the 1990–2000 and 2007–2011 periods to 7-year equivalent changes to achieve comparability across periods.

2

We compute this number by de-standardizing the coefficient from column (3) and dividing it by the adjusted secular increase in labor force participation of men from (0.345/0.491 × 1/4.37 × 24/21).

3

The General Educational Diploma (GED) is a high school equivalent diploma designed for individuals who have not completed high school. The GED can be used to apply to college just like a traditional high school diploma.

4

Our results on schooling focus on attendance in degree-granting institutions. They do not include enrollment in a trade or business school, or company training, as information about enrollment in these programs is not detected by Census and ACS data. However, certificates or other non-degree credentials are potentially important for affected workers seeking to transition to new employment. This fact may be particularly true for workers over 35 years of age. Indeed, Katz, Roth, Hendra, and Schaberg (2022) exploited data from four randomized controlled trials (RCTs) on sector-focused training programs designed for low-wage workers and found a significant increase in earnings after participants complete the training. They suggested that the earnings gains result from placing participants in higher-paying jobs, rather than solely increasing employment rates. On the contrary, other papers have shown that US training programs produce mixed results, with few instances of substantial and lasting increases in earnings (Greenberg, Michalopoulos, and Robbins 2003; Card, Kluve, and Weber 2018; ).

5

Results are available on request.

6

Table A27 in applies the same decomposition exercise to each margin of adjustment across demographic groups to investigate whether significant differences exist across non-participants depending on their demographics. Results suggest that non-participants who are enrolled in college are between 25 and 34 years, white, and enrolled in graduate or professional schools, given that most of them have already achieved a bachelor’s degree. Middle-aged (35–54 years) and older workers (55–64 years), by contrast, do not invest in additional human capital by enrolling in college, but rather leave the labor force permanently by enrolling in disability insurance or retiring early. We also find that a significant fraction of non-participants across all demographic groups relies on their household income, their private savings, or, in particular for less-educated racial and ethnic minorities, are idle.

7

Unfortunately, the US Census/ACS do not provide information about the place of birth of individuals more granular than the state.

8

reports the same results for employed workers and the entire population.

9

This result is in line with , who showed that robot adoption reduces work-related injuries in the manufacturing sector.

10

shows that the estimates of the effect of robots on mental disorders do not differ significantly when lowering the length-of-stay threshold for acute health conditions to four days or increasing it up to 14 days.

11

This approach does not entirely resolve the issue of changes in the sample of hospitals, which complicates the interpretation of our results. Specifically, robot adoption could influence both the numerator and the denominator of our regression. It may increase the number of severe cases while reducing hospital admissions for non-severe cases, as individuals with less insurance coverage might limit their use of medical services. We also run a regression of the log of total hospital admissions on robot exposure and find no statistically significant results.

12

The effect on all disability-related admissions (column (2)) remains positive and statistically significant when we construct the variable as a share of all hospital admissions with a length of stay exceeding seven days.

13

Table A8 reports estimates of the effect of robots on the share of disability-related hospital admissions by diagnosis including all admissions, while reports estimates of the impact of robots on the share of diagnoses that are not directly related to a disability. We find no evidence that robots are systematically affecting the health outcomes of these individuals.

References

Acemoglu

Daron

. 1999. Changes in unemployment and wage inequality: An alternative theory and some evidence. American Economic Review 89(5):1259–78.

Acemoglu

Daron

Autor

David

. 2011. Skills, tasks and technologies: Implications for employment and earnings. In Card

David

Ashenfelter

Orley

(Eds.), Handbook of Labor Economics, Vol. 4, pp. 1043–171. Amsterdam: North Holland, Elsevier.

Acemoglu

Daron

Restrepo

Pascual

. 2018. The race between man and machine: Implications of technology for growth, factor shares, and employment. American Economic Review 108(6):1488–542.

Acemoglu

Daron

Restrepo

Pascual

. 2020. Robots and jobs: Evidence from US labor markets. Journal of Political Economy 128(6):2188–244.

Adda

Jérôme

Fawaz

Yarine

. 2020. The health toll of import competition. Economic Journal 130(630):1501–40.

Ahituv

Avner

Zeira

Joseph

. 2011. Technical progress and early retirement. Economic Journal 121(551):171–93.

Autor

David

Chin

Caroline

Salomons

Anna

Seegmiller

Bryan

. 2024. New frontiers: The origins and content of new work, 1940–2018. Quarterly Journal of Economics 139(3):1399–465.

Autor

David

Dorn

David

. 2013. The growth of low-skill service jobs and the polarization of the US labor market. American Economic Review 103(5):1553–97.

Autor

David

Dorn

David

Hanson

Gordon

. 2013. The China syndrome: Local labor market effects of import competition in the United States. American Economic Review 103(6):2121–68.

10.

Autor

David

Duggan

Mark G.

2003. The rise in the disability rolls and the decline in unemployment. Quarterly Journal of Economics 118(1):157–206.

11.

Bartel

Ann P.

Sicherman

Nachum

. 1993. Technological change and retirement decisions of older workers. Journal of Labor Economics 11(1):162–83.

12.

Black

Dan

Daniel

Kermit

Sanders

Seth

. 2002. The impact of economic conditions on participation in disability programs: Evidence from the coal boom and bust. American Economic Review 92(1):27–50.

13.

Blanchard

Olivier

Katz

Lawrence F.

1992. Regional evolutions. Brookings Papers on Economic Activity 23(1):1–76.

14.

Brynjolfsson

Erik

McAfee

Andrew

. 2014. The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies. New York: W. W. Norton & Company.

15.

Burlon

Lorenzo

Vilalta-Bufí

Montserrat

. 2016. A new look at technical progress and early retirement. IZA Journal of Labor Policy 5(1): Article 5.

16.

Card

David

Kluve

Jochen

Weber

Andrea

. 2018. What works? A meta-analysis of recent labor market program evaluations. Journal of the European Economic Association 16(3):894–931.

17.

Case

Anne

Deaton

Angus

. 2015. Rising morbidity and mortality in midlife among white non-Hispanic Americans in the 21st century. Proceedings of the National Academy of Sciences 112(49):15078–83.

18.

———. 2017. Mortality and morbidity in the 21st century. Brookings Papers on Economic Activity 48(1):397–476.

19.

Dauth

Wolfgang

Findeisen

Sebastian

Suedekum

Jens

Woessner

Nicole

. 2021. The adjustment of labor markets to robots. Journal of the European Economic Association 19(6):3104–53.

20.

Deshpande

Manasi

Lockwood

Lee

. 2022. Beyond health: Non-health risk and the value of disability insurance. Econometrica 90(4):1781–810.

21.

De Vries

Gaaitzen J.

Gentile

Elisabetta

Miroudot

Sébastien

Wacker

Konstantin M.

2020. The rise of robots and the fall of routine jobs. Labour Economics 66:101885.

22.

Di Giacomo

Giuseppe

Lerch

Benjamin

. 2023. Automation and human capital adjustment: The effect of robots on college enrollment. Journal of Human Resources. Advance online publication. https://doi.org/10.3368/jhr.1222-12684R1

23.

Faber

Marius

Sarto

Andreés P.

Tabellini

Marco

. 2022. Local shocks and internal migration: The disparate effects of robots and Chinese imports in the US. NBER Working Paper No. 30048. Cambridge, MA: National Bureau of Economic Research.

24.

Ferriere

Axelle

Navarro

Gaston

Reyes-Heroles

Ricardo

. 2023. Escaping the losses from trade: The impact of heterogeneity and skill acquisition. Manuscript. https://rreyes-heroles.com/uploads/3/6/2/8/36287730/fnrh_nov_2023.pdf

25.

Foote

Andrew

Grosz

Michel

. 2020. The effect of local labor market downturns on postsecondary enrollment and program choice. Education Finance and Policy 15(4):593–622.

26.

Foote

Andrew

Grosz

Michel

Stevens

Ann

. 2019. Locate your nearest exit: Mass layoffs and local labor market response. ILR Review 72(1):101–26.

27.

Ford

Martin

. 2015. Rise of the Robots: Technology and the Threat of a Jobless Future. New York: Basic Books.

28.

Frank

Richard G.

Glied

Sherry A.

Marple

Keith

Shields

Morgan

. 2019. Changing labor markets and mental illness: Impacts on work and disability. RDRC Paper No. NB19-05. Cambridge, MA: National Bureau of Economic Research.

29.

Gihleb

Rania

Giuntella

Osea

Stella

Luca

Wang

Tianyi

. 2022. Industrial robots, workers’ safety, and health. Labour Economics 78:102205.

30.

Goldin

Claudia

Katz

Lawrence F.

2010. The Race Between Education and Technology. Cambridge, MA: Harvard University Press.

31.

Goldsmith-Pinkham

Paul

Sorkin

Isaac

Swift

Henry

. 2020. Bartik instruments: What, when, why, and how. American Economic Review 110(8):2586–624.

32.

Goos

Maarten

Manning

Alan

. 2007. Lousy and lovely jobs: The rising polarization of work in Britain. Review of Economics and Statistics 89(1):118–33.

33.

Goos

Maarten

Manning

Alan

Salomons

Anna

. 2009. Job polarization in Europe. American Economic Review 99(2):58–63.

34.

Graetz

Georg

Michaels

Guy

. 2018. Robots at work. Review of Economics and Statistics 100(5):753–68.

35.

Greenberg

David H.

Michalopoulos

Charles

Robins

Phillip K.

2003. A meta-analysis of government-sponsored training program. Industrial and Labor Relations Review 57(1):31–53.

36.

Greenland

Andrew

Lopresti

John

. 2016. Import exposure and human capital adjustment: Evidence from the US. Journal of International Economics 100:50–60.

37.

Grigoli

Francesco

Koczan

Zsoka

Topalova

Petia

. 2020. Automation and labor force participation in advanced economies: Macro and micro evidence. European Economic Review 126:103443.

38.

Hickman

Daniel C.

Olney

William W.

2011. Globalization and investment in human capital. Industrial and Labor Relations Review 64(4):654–72.

39.

Hollingsworth

Alex

Ruhm

Christopher J.

Simon

Kosali

. 2017. Macroeconomic conditions and opioid abuse. Journal of Health Economics 56:222–33.

40.

Howard

Greg

. 2020. The migration accelerator: Labor mobility, housing, and demand. American Economic Journal: Macroeconomics 12(4):147–79.

41.

Hunt

Jennifer

. 2017. The impact of immigration on the educational attainment of natives. Journal of Human Resources 52(4):1060–118.

42.

[IFR] International Federation of Robotics. 2018. World Robotics 2018. Technical Report. IFR: Frankfurt, Germany.

43.

Jaimovich

Nir

Saporta-Eksten

Itay

Siu

Henry E.

Yedid-Levi

Yaniv

. 2021. The macroeconomics of automation: Data, theory, and policy analysis. Journal of Monetary Economics 122:1–16.

44.

Katz

Lawrence F.

Murphy

Kevin M.

1992. Changes in relative wages, 1963–1987: Supply and demand factors. Quarterly Journal of Economics 107(1):35–78.

45.

Katz

Lawrence F.

Roth

Jonathan

Hendra

Richard

Schaberg

Kelsey

. 2022. Why do sectoral employment programs work? Lessons from WorkAdvance. Journal of Labor Economics 40(S1):S249–91.

46.

Krueger

Alan B.

2017. Where have all the workers gone? An inquiry into the decline of the US labor force participation rate. Brookings Papers on Economic Activity 48(2):1–87.

47.

Lang

Matthew

McManus

T. Clay

Schaur

Georg

. 2019. The effects of import competition on health in the local economy. Health Economics 28(1):44–56.

48.

Lerch

Benjamin

. 2025. From blue- to steel-collar jobs: The decline in employment gaps? American Economic Journal: Macroeconomics 17(1):126–60.

49.

Liebman

Jeffrey B.

2015. Understanding the increase in disability insurance benefit receipt in the United States. Journal of Economic Perspectives 29(2):123–50.

50.

Maestas

Nicole

Mullen

Kathleen J.

Strand

Alexander

. 2015. Disability insurance and the Great Recession. American Economic Review 105(5):177–82.

51.

Molloy

Raven

Smith

Christopher L.

Wozniak

Abigail

. 2011. Internal migration in the United States. Journal of Economic Perspectives 25(3):173–96.

52.

Monras

Joan

. 2018. Economic shocks and internal migration. CEPR Discussion Paper No. 12977. London: Centre for Economic Policy Research.

53.

Naidu

Suresh

Sojourner

Aaron

. 2020. Employer power and employee skills: Understanding workforce training programs in the context of labor market power. Roosevelt Institute, December.

54.

Parsons

Donald O.

1980a. The decline in male labor force participation. Journal of Political Economy 88(1):117–34.

55.

Parsons

Donald O.

1980b. Racial trends in male labor force participation. American Economic Review 70(5):911–20.

56.

Peracchi

Franco

Welch

Finis

. 1994. Trends in labor force transitions of older men and women. Journal of Labor Economics 12(2):210–42.

57.

Pierce

Justin R.

Schott

Peter K.

2020. Trade liberalization and mortality: Evidence from US counties. American Economic Review: Insights 2(1):47–63.

58.

Rege

Mari

Telle

Kjetil

Votruba

Mark

. 2009. The effect of plant downsizing on disability pension utilization. Journal of the European Economic Association 7(4):754–78.

59.

Restrepo

Pascual.

2015. Skill mismatch and structural unemployment. Massachusetts Institute of Technology, Job Market Paper.

60.

Ruggles

Steven

Flood

Sarah

Goeken

Ronald

Grover

Josiah

Meyer

Erin

Pacas

Jose

Sobek

Matthew

. 2019. IPUMS USA: Version 9.0 [data set]. Minneapolis, MN. https://doi.org/10.18128/D010.V9.0

61.

Schaller

Jessamyn

Stevens

Ann Huff

. 2015. Short-run effects of job loss on health conditions, health insurance, and health care utilization. Journal of Health Economics 43:190–203.

62.

Sullivan

Daniel

Von Wachter

Till

. 2009. Job displacement and mortality: An analysis using administrative data. Quarterly Journal of Economics 124(3):1265–306.

63.

Susskind

Daniel

. 2020. A World Without Work: Technology, Automation, and How We Should Respond. London: Penguin UK.

64.

Tolbert

Charles M.

Sizer

Molly

. 1996. US commuting zones and labor market areas: A 1990 update. Technical Report No. 278812. United States Department of Agriculture, Economic Research Service.

65.

Weinstein

Russell

. 2022. Local labor markets and human capital investments. Journal of Human Resources 57(5):1498–525.

66.

Yagan

Danny

. 2019. Employment hysteresis from the Great Recession. Journal of Political Economy 127(5):2505–58.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

2.24 MB