Gunadi, Christian; Ryu, Hanbyul

Working Paper
Does the Rise of Robotic Technology Make People
Healthier?

GLO Discussion Paper, No. 600

Provided in Cooperation with:

Global Labor Organization (GLO)

Suggested Citation: Gunadi, Christian; Ryu, Hanbyul (2020) : Does the Rise of Robotic
Technology Make People Healthier?, GLO Discussion Paper, No. 600, Global Labor
Organization (GLO), Essen

This Version is available at:

http://hdl.handle.net/10419/222315

Standard-Nutzungsbedingungen: Terms of use:

Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your
Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.

Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial
Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them
machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise
use the documents in public.
Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen
(insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open
gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you
genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated
licence.
 Does the Rise of Robotic Technology Make People
Healthier?

Christian Gunadi∗ Hanbyul Ryu†

July 9, 2020

Abstract
Technological advancements bring changes to our life, altering our behaviors
as well as our role in the economy. In this paper, we examine the potential effect
of the rise of robotic technology on health. The results of the analysis suggest
that higher penetration of industrial robots in the local labor market is posi-
tively related to the health of the low-skilled population. A ten percent increase
in robots per 1,000 workers is associated with an approximately 10% reduction
in the fraction of low-skilled individuals reporting poor health. Further analysis
suggests that reallocation of tasks and reduction in unhealthy behavior partly
explain this finding.

JEL Classification: I10, I13, J24, O33

Keywords: Automation, Robots, Health

∗
[email protected]. Department of Economics, UC Riverside, CA 92521, USA
†
[email protected]. Department of New Southern Policy, Korea Institute for International Eco-
nomic Policy, Sejong-si 30147, South Korea

1
1 Introduction

The use of industrial robots has been rising rapidly in the United States. Between

2005 and 2017, the number of robots per 1,000 workers increased by about 70% (Fig-

ure 1). At the same time, there are questions about how the rapid implementation of

robots affects society. Most of the recent studies have been focused on the labor mar-

ket effects of robot adoption in the local economy (Acemoglu and Restrepo, 2020a;

Graetz and Michaels, 2018; Acemoglu and Restrepo, 2020b; Giuntella and Wang,

2019). These studies generally found that the rise of robotic technology harms the

labor market prospect of low-skilled workers. Relatively few studies, however, ex-

amine how exposure to robots affect the other aspects of society. An exception is a

work by Anelli et al. (2019), which found that a more intense robot penetration in

the local labor market led to a decline in new marriages and an increase in both

divorce and cohabitation, partly by rising economic uncertainty and lowering the

relative marriage-market value of men.

In this paper, we examine the potential effect of higher robot exposure on the

health of the low-skilled population. We postulate that higher penetration of robots

in a local labor market improves the health of low-skilled individuals in the local-

ity through two channels. First, robots mainly replace the physically demanding

tasks usually done by low-skilled workers, nudging these workers toward occupa-

tions with lower intensity of physical tasks, improving their health. Second, the

worsening labor market conditions will lead to a reduction in unhealthy behaviors,

such as smoking. Indeed, the literature has documented evidence that mortality

rate is pro-cyclical (Ruhm, 2000; Neumayer, 2004; Gerdtham and Ruhm, 2006),

2
partly because of reductions in job-related stress and individuals adopting healthier

lifestyle during economic downturns (Ruhm and Black, 2002; Ruhm, 2005).

We begin our analysis by examining the relationship between the rise of robotic

technology and health. We found evidence that higher exposure to robots is pos-

itively related to the health of the low-skilled population. A ten percent increase

in robots per 1,000 workers is associated with 0.5, 1.3, and 0.6 percentage points

decline in the fraction of low-skilled population reporting poor health, work disabil-

ity, and ever quit a job because of health reasons. Evaluated at the mean, these

estimates correspond to an approximately 10% decrease in each of the outcomes.

Examining the mechanisms behind these findings, we found evidence of the real-

location of tasks. A ten percent increase in robots per 1,000 workers is associated

with approximately 2% reduction in the fraction of low-skilled workers employed

in occupations classified as physically demanding.1 Similar findings are found for

employment in risky jobs. The results of the analysis suggest that a ten percent

increase in robots per 1,000 workers is associated with approximately 9% and 4%

declines in the fraction of low-skilled workers employed in high fatality rate occu-

pations and high injury rate industries, respectively. We fail to find evidence that

the fraction of the low-skilled population identified as a current or everyday smoker

is affected by robot exposure. However, there is evidence that an increase in robot

exposure is associated with a lower number of cigarettes per day consumed by ev-

eryday smokers. This finding suggests that the effect of robot exposure on health

that is coming through changing smoking behavior, if there is any, is likely to be

the result of a reduction in smoking intensity.

1
As described later, we use the Department of Labor O*NET data to classify occupations with
high physical tasks requirement.

3
 This paper is related to a growing literature examining the impacts of the indus-

trial robot. Most of these studies have been focused on the labor market effects of

robot exposure. Examining the impacts of robots across U.S. commuting zones, Ace-

moglu and Restrepo (2020a) found strong negative effects of robots on employment

and wages, especially among low skilled workers. Graetz and Michaels (2018) found

that increased robot use is associated with higher labor productivity. However, they

also found evidence that low-skilled workers lose out from the adoption of industrial

robots.2 Analyzing the effect of robots across cities in China, Giuntella and Wang

(2019) found a large negative impact of robot exposure on employment and wages

of Chinese workers, especially those who are low-skilled. Relatively few studies,

however, examine how robots affect the other socio-economic outcomes. An excep-

tion is the work by Anelli et al. (2019), which found that higher adoption of robots

in the local labor market affects the family formation, decreasing new marriages

and increasing both divorce and cohabitation. We contribute to this literature by

examining the potential role of robotic technology in improving the health of the

population, especially those who are low-skilled.

This paper is also related to studies that examine the relationship between eco-

nomic conditions and health. The seminal work by Ruhm (2000) found that mortal-

ity rate in the United States is procyclical.3 Subsequent studies have found that this

relationship hold in other countries (Neumayer, 2004; Granados, 2005; Gerdtham

and Ruhm, 2006; Lin, 2009). The reason why this is the case, however, is still

inconclusive. Ruhm and Black (2002) and Ruhm (2005) argue that reduction in
2
This is unlike the effect of Information and Communication Technology (ICT), which mainly
adversely affecting workers in the middle of skill distribution (Autor et al., 2003; Goos et al., 2014;
Michaels et al., 2014).
3
It is worth noting that a follow-up study by Ruhm (2015) suggests that mortality has shifted
from strongly procyclical to weakly related to economic conditions in recent years.

4
unhealthy behaviors, such as drinking and smoking, partly explain why mortality

decrease during economic downturns. A more recent study by Stevens et al. (2015)

argues that lower quality of health care during economic expansion may explain the

observed negative relationship between improvement in economic conditions and

health. We contribute to this literature by documenting that worsening economic

conditions due to the rise of robotic technology is associated with an improvement

in health status among the low-skilled population and by offering potential mecha-

nisms driving this relationship.

The rest of the article is constructed as follows. The next section describes the

data used in the main analysis. Section 3 describes the empirical methodology.

Section 4 documents the main finding. Section 5 explores the potential mechanisms

explaining the main finding. Section 6 concludes.

2 Data

2.1 IFR Robot Data

We obtain the statistics on the operational stock of robots from the International

Federation of Robotics (IFR). The statistics come primarily from the information

provided by nearly all industrial robot suppliers to the IFR Statistical Department.

The IFR data has information on the operational stock of “industrial robots” in

more than 50 countries from 1993 to 2017, defined as “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 appli-

cations.”

5
 There are a few limitations of using IFR data. First, the statistics on the oper-

ational stock of robots are only available at the national level across the years. To

obtain a measure of robot exposure at the local level, similar to recent studies (Ace-

moglu and Restrepo, 2020a; Graetz and Michaels, 2018; Giuntella and Wang, 2019),

we use the variation in the initial distribution of industrial employment in U.S. lo-

calities and the difference in robot use across industries over time. The intuition is

that cities that are historically more dependent on robot-intensive industries will

have a higher number of robots per worker compared to other areas. Second, the

IFR industrial classification is coarse, and it is only available since 2004, limiting

our analysis period to 2004 onwards.4 Additionally, not all robots are classified into

one of IFR industry classification. For those that are unclassified, we allocate it to

each industry in the same proportion as the classified robot data.

Using the information available from the IFR data, we constructed the robot

exposure measure at the local level as follows:

J
X Rjt
Robotsmt = πmj,1960 × (1)
j=1
Lj,1960

where πmj,1960 is the share of industry j employment in MSA m in 1960. We use

the share of industry in 1960 to focus on the city’s specialization in industries that

predates the rise of robots in the early 1990s. Rjt is the total stock of robot employed

in industry j at time t. Lj,1960 is the number of workers employed in industry j

in 1960. It follows that the robot exposure measure, Robotsmt , predicts that cities
4
We use the broad IFR industry classification in creating the robot exposure measure:
food/beverages and tobacco products, textiles, wood products, paper products, plastic and chemical
products, glass/ceramics and other mineral products, metal, electronics, automotive, other transport
equipment, other manufacturing branches, agriculture, mining, utilities, construction, education,
and all other non-manufacturing branches.

6
that are more dependent on robot-intensive industries in 1960, partly because these

cities have comparative advantages (i.e., resources, location) to specialize in those

industries, will have a higher number of robots per worker today.

On average, there are 3.32 robots per 1,000 workers across the cities in our sam-

ple (Table 1). Many of the cities with the highest predicted robot exposure are lo-

cated in the Midwest (Appendix Table 1). This is unsurprising since the automotive

industry, which is the top robot-intensive industry (Appendix Table 2), is mainly

concentrated in this region.

2.2 Health Status Data

The measures of health used in our analysis are obtained from the Current Popula-

tion Survey (CPS) available on IPUMS (Flood et al., 2020). Administered monthly

to over 65,000 households in the United States, CPS provides information on edu-

cation, labor force status, and other aspects of the U.S. population. Over time, the

CPS has added supplemental information on special topics such as health status

and tobacco use in some months. The health status information, in particular, is

available starting from 1996 in March CPS (CPS-ASEC). Throughout the analysis,

we focus on the sample of individuals between the ages of 25 and 60 to avoid poten-

tial bias associated with changes in perceived/actual health after retirement (Coe

and Zamarro, 2011; Mazzonna and Peracchi, 2012).

The health status in March CPS indicates an individual’s health on a five-point

scale (Excellent, Very Good, Good, Fair, or Poor). Specifically, the question is worded

as follows: ”Would you say your health in general is excellent, very good, good, fair,

or poor?” We use this information to construct our main outcome: the share of the

7
population in a city reporting poor health. In addition to health status, the March

CPS also asks additional questions with regards to work disability and whether an

individual ever quit a job because of health reasons. We use this information to

construct additional health outcomes in the analysis. In an average city, the frac-

tion of low-skilled population with no high school diploma reporting poor health is

higher than their high-skilled counterparts: five percent of the low-skilled popula-

tion reports that they are in poor health, while only 2 percent of the high-skilled

population with at least a high school diploma reports that they are in poor health

(Table 1).5 Similar patterns between low- and high-skilled populations are observed

for the fraction of population reporting work disability or ever quit a job because of

health reasons.

3 Empirical Methodology

To examine the effect of robot exposure on health, we estimate the following empir-

ical specifications:

yct = δc + δt + β1 ln(Robotsc,t−2 ) + Xct0 β2 + εct (2)

where yct is the outcome for Metropolitan Statistical Area (MSA) c at time t. As

mentioned in the previous section, we consider three health outcomes: the share of

population reporting poor health, the share of population reporting work disability,

and the share of population reporting ever quit a job because of health reasons. Our

main coefficient of interest is β1 , which corresponds to unit increase in y following

5
Throughout the paper, we define low-skilled as individuals with no high school diploma, while
high-skilled is defined as those with at least a high school diploma.

8
an increase of one in the natural log of robots per 1,000 workers. We lagged the

effects since it should take some time for individuals to adjust to an increase in robot

exposure. Xct is a vector of city-level control variables which include the population

share of blacks, the population share of females, and the unemployment rate. δc

and δt are MSA and year fixed effects, respectively. All regressions are weighted by

the MSA population in 2000. Unless otherwise specified, our period of the analysis

is 2006 to 2017. This is because the earliest robot by industry data is only available

starting from 2004, and the latest IFR data that we can obtain is 2017. We include

all MSA in IPUMS 5% 1960 Census that can be consistently identified in March

CPS between 2004 and 2017 in our analysis.

Since we use predicted rather than actual robot exposure, there are fewer con-

cerns that local unobserved factors will bias our estimates. However, to further

address the endogeneity concerns, we use the variation in the robot use across in-

dustries in the European countries as an instrument, similar to Acemoglu and Re-

strepo (2020a) and Giuntella and Wang (2019). The main idea is that factors that

contribute to the rise of robots in these other economies are unlikely to be corre-

lated with unobserved factors affecting health in U.S. localities. Specifically, the

instrument is constructed as follows:

J EU
X Rjt
Robotsmt = πmj,1960 × (3)
j=1
Lj,1960

where the definition of the variables is the same as before except for Rjt
EU
, which is

now defined as the total operational stock of industrial robots in European coun-

tries.6
6
We use the sum of operational stock of robot in the United Kingdom, Finland, Denmark, France,
Norway, Spain, and Sweden to construct the instrument.

9
 To be valid, this instrument must fulfill two conditions. First, the instrument

must be strongly correlated with the endogenous variable. The first-stage anal-

ysis results suggest that this is indeed the case (Appendix Table 3). The robust

F-statistics are around 27, well above the Staiger and Stock (1994) rule of thumb of

10. The interpretation of the estimate is that a one percent increase in predicted

robot exposure constructed using the variation in the robot use across industries

in the European countries is associated with a roughly 0.5% rise in the predicted

robot exposure in the U.S. cities. Second, the instrument must not be correlated

with unobserved local factors affecting the health of individuals in U.S. localities.

Although this condition is essentially untestable, we will provide supporting evi-

dence that this condition is fulfilled in the next section.

4 Results

4.1 Main Findings

Before reporting the results from our main empirical specifications, we present the

visual evidence on the relationship between robot exposure and health in Figures 2

and 3. We separate the analysis by two skill groups: low-skilled is defined as indi-

viduals with no high school diploma, while high-skilled is defined as those with at

least a high school diploma. This is based on our hypothesis that the rise of robotic

technology mainly substitutes for physically demanding tasks that were usually

done by the low-skilled workers, nudging these workers towards occupations that

are less physically demanding. Therefore, we should see that the effect of robots on

health to be concentrated among the low-skilled population. Consistent with this

10
hypothesis, we see that cities that had a high growth of robots per 1,000 workers

between 2005 and 2017 experienced a decline in the share of low-skilled population

reporting poor health (Figure 2a). The slope of the fitted line implies that a one per-

cent increase in robot exposure is associated with a 1.13% decline in the fraction of

low-skilled population reporting poor health. Although it is imprecisely estimated,

we also see there is a negative relationship between the growth of robots per 1,000

workers with other measures of health outcome such as the share of low-skilled

population reporting work disability or ever quit a job because of health reasons

(Figures 2b and 2c).

On the other hand, there is not much evidence that the health outcomes for high-

skilled individuals are affected by the rise of robotic technology (Figure 3). The

slope of the fitted line suggests that there is a positive relationship between the

growth of robot exposure and the fraction of the high-skilled population reporting

poor health, but this estimate is small in magnitude and not statistically significant

(Figure 2a). Qualitatively similar findings are found for the share of high-skilled

population reporting work disability and ever quit a job because of health reasons

(Figures 2b and 2c).

We report the results from our main empirical specifications in Table 2. Similar

to visual evidence, the effects of robot exposure are mainly concentrated on the low-

skilled population. A ten percent increase in robots per 1,000 workers is associated

with about 0.3 percentage point decrease in the share of low-skilled population re-

porting poor health (Column 2 of Panel A). The results from the IV model suggest

that this estimate is an overestimation of the true effect (∼ 0.5p.p.). Evaluated at

the sample mean, this estimate corresponds to about a 10% decline. Qualitatively

11
similar findings are found for the other health outcomes: a one percent increase in

robot exposure is associated with approximately 1.3 and 0.6 percentage points de-

cline in the share of low-skilled population reporting work disability and ever quit

a job for health reasons, respectively.

On the contrary, the estimates on the high-skilled population are smaller in

magnitude and not statistically different from zero. At a 90% significance level,

evaluated at the sample mean, for a 10% increase in robots per 1,000 workers, we

can rule out an effect size larger than a 4% decline in the fraction of high-skilled

population reporting poor health. The results from the IV model suggest a larger

magnitude of the effect, but it is not statistically significant. Qualitatively similar

results are obtained for the other health outcomes.

In sum, the results of the analysis in this section document evidence of a negative

relationship between the rise of robotic technology and the fraction of low-skilled

population reporting poor health. However, the mechanism driving this finding is

still unclear. The next sections are devoted to checking the robustness of this finding

as well as exploring the potential mechanisms explaining this result .

4.2 Robustness Checks

In the main empirical specifications, we choose to measure the two-year lagged ef-

fects of robot exposure, mainly because it should take some time for individuals to

adjust in response to the rise of robotic technology in their locality. However, this

choice may seem arbitrary. Therefore, we check the robustness of our findings when

one- or three-year lagged robot exposures are used in the analysis (Appendix Table

4). Although some of the estimates become imprecisely estimated, the results of this

12
exercise are largely in line with the findings from the main empirical specifications.

Another concern is that our findings may be driven by an outlier city with high

growth of robot exposure experiencing a large decline in population reporting poor

health. To check for this, we conducted a leave-one-city-out analysis, excluding one

city in the sample one by one and re-estimating the effect. The results of this exer-

cise are reported in Figure 4. For the fraction of the low-skilled population reporting

poor health, the range of the estimates is quite narrow. Most of the estimates lie

between -0.030 and -0.038 (Figure 4a). In Figure 4b, we also report the uncertainty

around the estimates. There is no evidence that the main findings are driven by a

specific city. Similar findings are also found for other health outcome measures.

Finally, Goldsmith-Pinkham et al. (2018) argues that the empirical specifica-

tions in which the variable of interest is constructed using a shift-share approach

(equation 1) are similar to difference-in-differences methodology. In other words,

the rise of robotic technology in the 1990s can be thought of as a ‘policy’ shock, and

the industry shares serve as a proxy for the exposure to the shock. Cities that rely

more on industries that are experiencing a higher rate of automation because of

robotic technology will be more exposed to the shock. In this case, the assumption

for the estimates to be valid is that cities that were experiencing high growth of

robot per 1,000 workers in 2005-2017 period would have a similar change in the

fraction of low-skilled population reporting poor health as cities with low growth of

robot exposure in the absence of the rise of robotic technology in the 1990s. It is not

possible to test for this assumption directly, but we can provide supporting evidence

that this assumption is met by checking the pre-1990 trends. In Figure 5, we graph

the relationship between the 2005-2017 growth of robots per 1,000 workers and the

13
1980-1990 growth of low-skilled population reporting work disability. The slope of

the fitted lines suggests that the pre-1990 trend of work disability rate between

cities that had a high growth of robot exposure in the 2005-2017 period and those

with low growth is similar. Unfortunately, information on poor health status and on

whether an individual ever quit a job for health reasons in 1980 and 1990 IPUMS

5% Census is not available, limiting the analysis only on the work disability rate.

Nonetheless, the finding in Figure 4 further supports the validity of the estimates

obtained in the main empirical analysis.

5 Potential Mechanisms

5.1 Reallocation of Tasks

The analysis in the previous section documents evidence of a negative relation-

ship between robot exposure and the share of low-skilled population reporting poor

health outcomes. However, it is unclear how the rise of robotic technology may

affect health. We hypothesize that robots mainly substitutes for the physically de-

manding tasks usually done by low-skilled workers, nudging these workers towards

occupation with less physical tasks. In this subsection, we examine whether there

is evidence to support this hypothesis from the data.

To examine the potential reallocation of tasks in response to the growth in robots

per 1,000 workers, we obtain information on the importance of physical abilities in

a given occupation from the U.S. Department of Labor O*NET dataset. O*NET

ratings reflect experts’ evaluation of how important an ability is in the occupa-

tion. Within physical ability group, O*NET measures the importance of the fol-

14
lowing abilities in a standardized scale ranging from 0 (min) to 100 (max): dynamic

flexibility, dynamic strength, explosive strength, extent flexibility, gross body co-

ordination, gross body equilibrium, stamina, static strength, and trunk strength.7

Unfortunately, not all occupations in O*NET occupation codes can be crosswalked

to IPUMS consistent occupation codes (OCC1990). However, we manage to assign

physical abilities score to 315 out of 341 occupations listed in IPUMS OCC1990. We

then took an average of the O*NET physical abilities rating, classifying occupations

with a score above the median as physically demanding occupations.8

In Table 3, we report the effect of robot exposure on the fraction of low-skilled

workers whose occupations can be assigned physical abilities score employed in

physically demanding occupations. There is evidence that the rise of robotic technol-

ogy is associated with a lower share of low-skilled workers employed in physically

demanding jobs: a ten percent increase in robots per 1,000 workers is associated

with about 0.7 percentage points decrease in the fraction of low-skilled workers em-

ployed in physically demanding occupations. Evaluated at the sample mean, this

estimate corresponds to a 1.2% decline. Although some of the estimates are impre-

cisely estimated, the results from the IV model suggest a larger magnitude of 2%

reduction.

Another way to examine the reallocation of tasks in response to the rise in robotic

technology is to analyze whether the fraction of low-skilled workers employed in

risky jobs is affected by an increase in robot exposure. To do this, we first obtained

the number of fatalities associated with each occupation from 2000 Census of Fatal

Occupational Injuries (CFOI) and crosswalked it at the two-digit level to IPUMS

7
The description for each ability is reported in Appendix Table 5.
8
Appendix Table 6 reports the 10 most/least physically demanding occupations based on O*NET
ratings in the sample.

15
consistent occupation codes. Then, we divide the number of fatalities with the

number of workers employed in each occupation obtained from 2000 CPS ASEC

to acquire the fatality rate corresponding to each occupation. Finally, we classify

individuals working in occupations with a fatality rate above the median as being

employed in high fatality risk jobs. It should be noted that this is not the only way

to classify an individual as working in a risky job. For example, a person could

be employed in an industry with a high injury rate. To explore this possibility, we

obtained the injury rate associated with each industry from 2000 Survey of Occu-

pational Injuries and Illnesses (SOII) and crosswalked it at the two-digit level to

IPUMS consistent industry code (IND1990). Once again, we classify individuals

working in industries with injury rate above the median as being employed in a

high injury risk industry.9

The results of this exercise are reported in Table 4. Focusing on the IV esti-

mates, there is evidence that the rise in robotic technology is negatively related

with the share of low-skilled workers employed in risky jobs: a ten percent increase

in robots per 1,000 workers is associated with approximately 0.3 and 0.2 percentage

points decrease in the fraction of low-skilled workers employed in high fatality rate

occupations and high injury rate, respectively. The magnitude of the effects is eco-

nomically meaningful. Evaluated at the sample mean, these estimates correspond

to about 9% and 4% reduction in the fraction of low-skilled workers employed in

high fatality rate occupations and high injury rate industries.

9
The fatality rate across occupations as well as the injury rate across industries are reported in
Appendix Tables 7 and 8.

16
5.2 Changes in Unhealthy Behavior

Reallocation of tasks is not the sole mechanism through which the rise of robotic

technology may affect health. Recent studies have documented evidence that mor-

tality rate is pro-cyclical (Ruhm, 2000; Neumayer, 2004; Gerdtham and Ruhm,

2006), partly because of the reduction in unhealthy behaviors during recessions,

such as smoking (Ruhm, 2005).10 In this subsection, we examine whether there is

evidence that a change in smoking behavior is one mechanism through which robot

exposure affects health.

For this analysis, we obtain the CPS supplement on tobacco use from IPUMS.

CPS tobacco use survey is not conducted every year; between 2004 and 2017, we

have data for 2006, 2007, 2010, 2011, 2014, and 2015. In some survey years, CPS

collected tobacco use information twice. In this case, we use both surveys and divide

the survey weights by two to make the sample representative of the U.S. popula-

tion. Similar to Ruhm (2005), we define current smokers as individuals who stated

that they have smoked 100 or more cigarettes in their lifetime and who currently

report smoking some days or every day. To construct additional measures of smok-

ing behavior, we also use the information on whether an individual is an everyday

smoker, and the average number of cigarettes currently smoked daily if the indi-

vidual is an everyday smoker. Throughout this analysis, we focused on low-skilled

population, whose labor market prospects are adversely affected by robot exposure

(Acemoglu and Restrepo, 2020a; Graetz and Michaels, 2018; Giuntella and Wang,

2019) and whom the health outcomes are found to be improved by the rise of robotic
10
Ruhm (2005) argues that reduction in job-related stress and increases in non-market leisure
time incentivize individuals to adopt healthier lifestyle during recessions.

17
technology (Table 2).

The results of this exercise are reported in Table 5. We found no evidence that

the rise in robotic technology has a statistically significant effect on the fraction

of the low-skilled population identified as a current or everyday smoker. However,

the IV estimates suggest that a ten percent increase in robots per 1,000 workers is

associated with one fewer cigarettes per day among everyday smokers. Evaluated at

the sample mean, these estimates correspond to about a 10% reduction in cigarettes

per day among everyday smokers. These findings suggest that the effect of robot

exposure on health that is coming through changing smoking behavior, if there is

any, is likely to be the result of a reduction in the intensity of smoking.

6 Conclusion

The use of industrial robots has increased substantially in the United States. As

such, there are interests in understanding more of how the rise of robotic technology

will affect our behavior and our role in the economy. In this paper, we attempt to

quantify the effect of robots on health. We hypothesize that higher penetration of

industrial robots in a local economy will improve the health of low-skilled workers

in the locality by nudging these workers toward occupations with lower intensity of

physical tasks and by reducing unhealthy behaviors such as smoking.

We have reached a few main findings. First, we document evidence that higher

penetration of industrial robots in the local labor market is positively related to the

health status of low-skilled individuals. A ten percent increase in robots per 1,000

workers is associated with 0.5, 1.3, and 0.6 percentage points decline in the share

of low-skilled population reporting poor health, work disability, and ever quit a job

18
because of health reasons. Evaluated at the sample mean, these estimates corre-

spond to an approximately 10% decrease in each of the outcomes. Second, we found

that this effect is partly explained by the reallocation of tasks and a reduction in

unhealthy behaviors. A ten percent increase in robots per 1,000 workers is associ-

ated with 2, 9, and 4 percent decline in the share of low-skilled workers employed in

physically demanding occupations, occupations with high fatality rate, and indus-

tries with high injury rates respectively. We fail to find evidence that the fraction

of the low-skilled population identified as a current or everyday smoker is affected

by robot exposure. However, there is evidence that an increase in robot exposure

is associated with a lower number of cigarettes per day consumed by an everyday

smoker. This finding suggests that the effect of robot exposure on health that is

coming through changing smoking behavior, if there is any, is likely to be the result

of a reduction in smoking intensity.

19
 7 Tables and Figures

Table 1: Summary Statistics

Mean SD Min. Max

Robots per 1,000 Workers 3.32 6.17 0.10 67.82

Fraction Black 0.13 0.10 0.00 0.56
Fraction Female 0.51 0.02 0.39 0.64
Unemployment Rate 0.07 0.04 0.00 0.29

Low-skilled
20

Fraction Reporting Poor Health 0.05 0.04 0.00 0.32

Fraction Reporting Work Disability 0.14 0.07 0.00 0.71
Fraction Reporting Ever Quit for Health Reasons 0.05 0.04 0.00 0.42

High-skilled
Fraction Reporting Poor Health 0.02 0.02 0.00 0.33
Fraction Reporting Work Disability 0.06 0.04 0.00 0.31
Fraction Reporting Ever Quit Job for Health Reasons 0.03 0.03 0.00 0.25
Notes: Estimates are based on International Federation of Robotics (IFR) data and Annual Social and
Economic Supplement (ASEC) of the Current Population Survey obtained from IPUMS.
 Table 2: The Effect of Robot Exposure on the Share of Population Reporting Poor Health Outcomes
Poor Health Work Disability Ever Quit Job Because
of Health Reasons
(1) (2) (3) (4) (5) (6)
Panel A (Low-skilled)
OLS: ln (Robot Exposure t-2) -0.029* -0.032** -0.048* -0.048* -0.024* -0.023
(0.016) (0.015) (0.026) (0.026) (0.014) (0.014)

2SLS: ln (Robot Exposure t-2) -0.047* -0.056** -0.132** -0.134** -0.062* -0.059*
(0.027) (0.027) (0.065) (0.066) (0.032) (0.033)

Mean of Dep. Var. 0.05 0.05 0.14 0.14 0.05 0.05

Panel B (High-skilled)
OLS: ln (Robot Exposure t-2) 0.005 0.003 0.002 0.002 0.010 0.009
(0.007) (0.007) (0.013) (0.013) (0.009) (0.009)
21

2SLS: ln (Robot Exposure t-2) 0.025 0.022 -0.023 -0.023 -0.011 -0.012
(0.016) (0.016) (0.021) (0.021) (0.015) (0.015)

Mean of Dep. Var. 0.02 0.02 0.06 0.06 0.03 0.03

Controls:
MSA and Year Fixed Effects Yes Yes Yes Yes Yes Yes
MSA Characteristics No Yes No Yes No Yes
Observations 1584 1584 1584 1584 1584 1584

Notes: Notes: The estimates show the effect of robot exposure on the share of population reporting poor health. Low-skilled
is defined as individuals without a high school diploma. High-skilled is defined as individuals with at least a high school
diploma. Control for MSA characteristics include population share of blacks, population share of female, and unemploy-
ment rate. The instrument in Panel B is constructed based on the number of operational robots in European countries. All
regressions are weighted by MSA population in 2000. Standard errors clustered at the MSA level are reported in paren-
theses. * p < .1, ** p < .05, *** p < .01
 Table 3: The Effect of Robot Exposure on
the Fraction of Low-skilled Workers Employed
in Physically Demanding Occupations
(1) (2)
OLS: ln (Robot Exposure t-2) -0.073** -0.066*
(0.035) (0.034)

2SLS: ln (Robot Exposure t-2) -0.115* -0.102

(0.067) (0.068)

Mean of the Outcome Variable 0.58 0.58

Controls:
MSA and Year Fixed Effects Yes Yes
MSA Characteristics No Yes
Observations 1584 1584

Notes: The estimates show the effect of robot exposure on the

share of low-skilled workers reporting working in physically
demanding occupations. Low-skilled is defined as individuals
without a high school diploma. Control for MSA characteristics
include population share of female, population share of blacks,
and unemployment rate. The instrument in Panel B is con-
structed based on the number of operational robots in European
countries. All regressions are weighted by MSA population in
2000. Standard errors clustered at the MSA level are reported
in parentheses. * p < .1, ** p < .05, *** p < .01

22
 Table 4: The Effect of Robot Exposure on the Fraction of
Low-skilled Workers Employed in Risky Jobs
High Fatality Rate High Injury Rate
Occupations Industries
(1) (2) (3) (4)
OLS: ln (Robot Exposure t-2) -0.190*** -0.178** -0.032 -0.026
(0.072) (0.073) (0.081) (0.082)

2SLS: ln (Robot Exposure t-2) -0.296** -0.277** -0.217* -0.232*

(0.124) (0.129) (0.130) (0.133)

Mean of the Outcome Variable 0.32 0.32 0.54 0.54

Controls:
23

MSA and Year Fixed Effects Yes Yes Yes Yes

MSA Characteristics No Yes No Yes
Observations 1584 1584 1584 1584

Notes: The estimates show the effect of robot exposure on the share of low-skilled workers
employed in risky jobs. High fatality rate occupations are defined as occupations with fa-
tality rate above median. High injury rate industries are defined as industries with injury
rate above median. Low-skilled is defined as individuals without a high school diploma.
Control for MSA characteristics include population share of blacks, population share of
female, and unemployment rate. The instrument in Panel B is constructed based on the
number of operational robots in European countries. All regressions are weighted by MSA
population. Standard errors clustered at the MSA level are reported in parentheses. *
p < .1, ** p < .05, *** p < .01
 Table 5: Exposure to Robot and Smoking Behavior Among Low-skilled Population
Fraction of Fraction of Cigs/Day
Current Smoker Everyday Smoker (Everyday Smoker)
(1) (2) (3) (4) (5) (6)

OLS: ln (Robot Exposure t-2) 0.056 0.052 0.051 0.047 0.892 0.873
(0.056) (0.056) (0.042) (0.041) (1.659) (1.614)

2SLS: ln (Robot Exposure t-2) -0.007 -0.014 0.039 0.033 -9.561* -9.581*
(0.078) (0.075) (0.078) (0.077) (5.722) (5.659)
24

Mean of Dep. Var. 0.24 0.24 0.19 0.19 10.43 10.43

Controls:
MSA and Year Fixed Effects Yes Yes Yes Yes Yes Yes
MSA Characteristics No Yes No Yes No Yes
Observations 840 840 840 840 840 840

Notes: The estimates show the effect of robot exposure on the smoking behavior among low-skilled population.
Low-skilled is defined as individuals without a high school diploma. Control for MSA characteristics include
population share of female, population share of blacks, and unemployment rate. The instrument in Panel B is
constructed based on the number of operational robots in European countries. All regressions are weighted by
MSA population in 2000. Standard errors clustered at the MSA level are reported in parentheses. * p < .1, **
p < .05, *** p < .01
 Figure 1: Rise of Robots in the United States

1.8
1.6
Robots per 1,000 Workers

1.4
1.2
25

1
.8

2005 2010 2015 2020

Year

Notes: The estimates are based on IPUMS Annual Social and Economic Supplement of the Current Population Survey (CPS ASEC) and
International Federation of Robotics (IFR) data.
 Figure 2: Robot Exposure and Health Outcomes (Low-skilled)

Growth in the Share of Low−skilled Population Reporting Work Disability

Growth in the Share of Low−skilled Population Reporting Poor Health

2
Coeff=−1.126, p−value=0.009

1
1

0
0

−1
−1

Coeff=−0.393, p−value=0.174
−2

−2
.4 .6 .8 1 1.2 .4 .6 .8 1 1.2
2005−2017 Growth in Robots per 1,000 Workers 2005−2017 Growth in Robots per 1,000 Workers

(a) Poor Health (b) Work Disability

Growth in the Share of Low−skilled Population Reporting Ever Quit for Health Reasons
26

2
1
0
−1
−2 Coeff=−0.357. p−value=0.326

.4 .6 .8 1 1.2
2005−2017 Growth in Robots per 1,000 Workers

(c) Ever Quit for Health Reasons

Notes: Growth rates are calculated by taking first difference of natural log. The analysis uses 93 MSA in which the growth rates between 2005
and 2017 can be calculated. Size of the circle represents the weight assigned to that particular observation. Each observation is weighted
by the commuting zone population in 2000. The growth in health outcomes are based on 2005 and 2017 IPUMS CPS-ASEC data. Robot
exposure measure is constructed based on IPUMS 5% 1980 Census and 2005-2017 International Federation of Robotics (IFR) data.
 Figure 3: Robot Exposure and Health Outcomes (High-skilled)

Growth in the Share of High−skilled Population Reporting Work Disability

Growth in the Share of High−skilled Population Reporting Poor Health

2
1

1
0

0
−1

−1
−2

Coeff=0.234, p−value=0.623
Coeff=−0.119, p−value=0.652
−3

−2
.4 .6 .8 1 1.2 .4 .6 .8 1 1.2
2005−2017 Growth in Robots per 1,000 Workers 2005−2017 Growth in Robots per 1,000 Workers

(a) Poor Health (b) Work Disability

Growth in the Share of High−skilled Population Reporting Ever Quit for Health Reasons
27

2
1
0
−1
−2 Coeff=0.077, p−value=0.810

.4 .6 .8 1 1.2
2005−2017 Growth in Robots per 1,000 Workers

(c) Ever Quit for Health Reasons

Notes: Growth rates are calculated by taking first difference of natural log. The analysis uses 77 MSA in which the growth rates between 2005
and 2017 can be calculated. Size of the circle represents the weight assigned to that particular observation. Each observation is weighted
by the commuting zone population in 2000. The growth in health outcomes are based on 2005 and 2017 IPUMS CPS-ASEC data. Robot
exposure measure is constructed based on IPUMS 5% 1980 Census and 2005-2017 International Federation of Robotics (IFR) data.
 Figure 4: Robustness Check (Leave-one-out Test)
80

0
60
Frequency

−.02
40

Effects
20

−.04
0

−.038 −.036 −.034 −.032 −.03

−.06
Effects

0 50 100 150
Excluded MSA ID
(a) Distribution of the Estimates (Poor
Health) (b) Leave-one-out Test (Poor Health)

0
60

−.02
Frequency

−.04
40

Effects
−.06
20

−.08
−.1
0

−.065 −.06 −.055 −.05 −.045

Effects 0 50 100 150
Excluded MSA ID

(c) Distribution of the Estimates (Work Dis.) (d) Leave-one-out Test (Work Dis.)
80

0
−.01
60
Frequency

−.02
Effects
40

−.03
20

−.04
−.05
0

−.03 −.028 −.026 −.024 −.022 −.02

Effects 0 50 100 150
Excluded MSA ID

(e) Distribution of the Estimates (Quit) (f) Leave-one-out Test (Quit)

Notes: Subfigures on the left show the distribution of the estimates from the leave-one-out
exercise. Subfigures on the right show the estimate of the effect when MSA ID in the cor-
responding x-axis is excluded from the regression. The blue line represents the coefficient
estimates, while the green dash lines represent the 90% confidence interval constructed
based on standard errors clustered at the MSA. All regressions are weighted by MSA pop-
ulation in 2000 and include controls for MSA and year fixed effects.

28
 Figure 5: Robustness Check (Checking Pre-trends)

.4

.4
1980−1990 Growth in Work Disability Rate

1980−1990 Growth in Work Disability Rate

Coeff=−0.025, p−value=0.598 Coeff=0.044, p−value=0.445
.2

.2
0

0
−.2

−.2
−.4

−.4
.4 .6 .8 1 1.2 1.4 0 .2 .4 .6 .8 1
29

2005−2017 Growth in Robots per 1,000 Workers (Constructed Using Robot Stock in U.S.) 2005−2017 Growth in Robots per 1,000 Workers (Constructed Using Robot Stock in European Countries)

(a) Constructed Using Robot Stock in the U.S. (b) Constructed Using Robot Stock in European Countries

Notes: Growth rates are calculated by taking first difference of natural log. The estimates for work disability rate among low-skilled workers
are calculated based on 1980 and 1990 IPUMS 5% Census data. Robot exposure measure is constructed based on IPUMS 5% 1980 Census and
2005-2017 International Federation of Robotics (IFR) data. Size of the circle represents the weight assigned to that particular observation.
Each observation is weighted by the MSA population.
References

Acemoglu, D. and Restrepo, P. (2020a). Robots and jobs: Evidence from us labor

markets. Journal of Political Economy, 128(6):2188–2244.

Acemoglu, D. and Restrepo, P. (2020b). Unpacking skill bias: Automation and new

tasks. Technical report, National Bureau of Economic Research.

Anelli, M., Giuntella, O., and Stella, L. (2019). Robots, labor markets, and family

behavior.

Autor, D. H., Levy, F., and Murnane, R. J. (2003). The skill content of recent tech-

nological change: An empirical exploration. The Quarterly journal of economics,

118(4):1279–1333.

Coe, N. B. and Zamarro, G. (2011). Retirement effects on health in europe. Journal

of health economics, 30(1):77–86.

Flood, S., King, M., Rodgers, R., Ruggles, S., and Warren, R. J. (2020). Integrated

public use microdata series, current population survey: Version 7.0. Minneapolis:

University of Minnesota.

Gerdtham, U.-G. and Ruhm, C. J. (2006). Deaths rise in good economic times: evi-

dence from the oecd. Economics & Human Biology, 4(3):298–316.

Giuntella, O. and Wang, T. (2019). Is an army of robots marching on chinese jobs?

Goldsmith-Pinkham, P., Sorkin, I., and Swift, H. (2018). Bartik instruments: What,

when, why, and how. Technical report, National Bureau of Economic Research.

30
Goos, M., Manning, A., and Salomons, A. (2014). Explaining job polarization:

Routine-biased technological change and offshoring. American economic review,

104(8):2509–26.

Graetz, G. and Michaels, G. (2018). Robots at work. Review of Economics and

Statistics, 100(5):753–768.

Granados, J. A. T. (2005). Recessions and mortality in spain, 1980–1997. European

Journal of Population/Revue européenne de Démographie, 21(4):393–422.

Lin, S.-J. (2009). Economic fluctuations and health outcome: a panel analysis of

asia-pacific countries. Applied Economics, 41(4):519–530.

Mazzonna, F. and Peracchi, F. (2012). Ageing, cognitive abilities and retirement.

European Economic Review, 56(4):691–710.

Michaels, G., Natraj, A., and Van Reenen, J. (2014). Has ict polarized skill demand?

evidence from eleven countries over twenty-five years. Review of Economics and

Statistics, 96(1):60–77.

Neumayer, E. (2004). Recessions lower (some) mortality rates:: evidence from ger-

many. Social science & medicine, 58(6):1037–1047.

Ruhm, C. J. (2000). Are recessions good for your health? The Quarterly journal of

economics, 115(2):617–650.

Ruhm, C. J. (2005). Healthy living in hard times. Journal of health economics,

24(2):341–363.

Ruhm, C. J. (2015). Recessions, healthy no more? Journal of health economics,

42:17–28.

31
Ruhm, C. J. and Black, W. E. (2002). Does drinking really decrease in bad times?

Journal of health economics, 21(4):659–678.

Staiger, D. O. and Stock, J. H. (1994). Instrumental variables regression with weak

instruments. National Bureau of Economic Research.

Stevens, A. H., Miller, D. L., Page, M. E., and Filipski, M. (2015). The best of times,

the worst of times: understanding pro-cyclical mortality. American Economic

Journal: Economic Policy, 7(4):279–311.

32
Appendix

Appendix Table 1: Cities with Highest/Lowest Predicted Exposure to

Robots in 2017
MSA Name Predicted
Robots/1000 Workers

Panel A: 10 Cities with Highest Predicted Robot Exposure

Flint, MI 67.82
Detroit, MI 30.98
Lansing-East Lansing, MI 26.52
Saginaw-Bay City-Midland, MI 22.33
South Bend-Mishawaka, IN 21.73
Jackson, MI 19.40
Racine, WI 14.37
Fort Wayne, IN 13.99
Toledo, OH/MI 13.24
Ann Arbor, MI 12.87

Panel B: 10 Cities with Lowest Predicted Robot Exposure

Laredo, TX 0.21
Austin, TX 0.39
Eugene-Springfield, OR 0.43
Fargo-Moorhead, ND/MN 0.43
Colorado Springs, CO 0.54
Columbus, GA/AL 0.56
Washington, DC/MD/VA 0.56
Pensacola, FL 0.56
Macon-Warner Robins, GA 0.58
Charleston-North Charleston, SC 0.58
Notes: The estimates are based on International Federation of Robotics (IFR) data. The number of workers
in an industry is obtained from IPUMS 5% 1960 Census.

33
 Appendix Table 2: Highest/Least Robot-Intensive Industries in 2017
MSA Name Robots/1000 Workers

Panel A: 5 Highest Robot-Intensive Industries

Automotive 159.84
Electrical/Electronics 28.34
Plastic and Chemical Products 18.16
All Other Manufacturing Branches 11.64
Food Products and Beverage; Tobacco Products 6.56

Panel B: 5 Least Robot-Intensive Industries

All Other non-Manufacturing Branches 0.01
Construction 0.05
Mining and Quarrying 0.07
Textiles, Leather, Wearing Apparel 0.09
Wood and Wood Products Including Furnitures 0.23

Notes: The estimates are based on International Federation of Robotics (IFR) data. The
number of workers in an industry is obtained from IPUMS CPS ASEC 2017.

34
 Appendix Table 3: First Stage Estimates
ln (US Robot Exposure t-2)
(1) (2)

ln (EU Robot Exposure t-2) 0.497*** 0.493***

(0.096) (0.095)

Robust F-Stats. 26.90 26.95

Controls:
MSA and Year Fixed Effects Yes Yes
MSA Characteristics No Yes
Observations 1584 1584

Notes: The estimates are based on International Federation of Robots

(IFR) data and Annual Social Economic Supplement (ASEC) of CPS from
2004 to 2015.

35
 Appendix Table 4: Exposure to Robot and Share of Low-Skilled Population
Reporting Poor Health Measures (Robustness – Lagged Effect Choice)
Poor Health Work Disability Quit Job Because of
Health Reasons
(1) (2) (3) (4) (5) (6)
Panel A (Lagged One Year)
OLS: ln (Robot Exposure t-1) -0.030** -0.033** -0.058** -0.057** -0.014 -0.013
(0.014) (0.015) (0.025) (0.025) (0.013) (0.013)

2SLS: ln (Robot Exposure t-1) -0.056** -0.062*** -0.148** -0.150** -0.061** -0.060*
(0.024) (0.024) (0.062) (0.062) (0.030) (0.031)

Observations 1716 1716 1716 1716 1716 1716

Panel B (Lagged Three Years)

OLS: ln (Robot Exposure t-3) -0.026 -0.031* -0.047* -0.048* -0.022 -0.021
36

(0.018) (0.017) (0.024) (0.025) (0.013) (0.014)

2SLS: ln (Robot Exposure t-3) -0.046 -0.055* -0.120** -0.122** -0.049* -0.047
(0.031) (0.031) (0.060) (0.062) (0.028) (0.030)

Observations 1452 1452 1452 1452 1452 1452

Controls:
MSA and Year Fixed Effects Yes Yes Yes Yes Yes Yes
MSA Characteristics No Yes No Yes No Yes

Notes: Notes: The estimates show the effect of robot exposure on the share of population reporting poor health. Low-
skilled is defined as individuals without a high school diploma. High-skilled is defined as individuals with at least a high
school diploma. Control for MSA characteristics include population share of blacks, population share of female, and un-
employment rate. The instrument in Panel B is constructed based on the number of operational robots in European coun-
tries. All regressions are weighted by MSA population in 2000. Standard errors clustered at the MSA level are reported
in parentheses. * p < .1, ** p < .05, *** p < .01
 Appendix Table 5: O*NET Physical Abilities
Ability Description
Dynamic Flexibility The ability to quickly and repeatedly bend, stretch, twist, or reach out with your body, arms, and/or legs.
Dynamic Strength The ability to exert muscle force repeatedly or continuously over time.
Explosive Strength The ability to use short bursts of muscle force to propel oneself (as in jumping or sprinting), or to throw an object
Extent Flexibility The ability to bend, stretch, twist, or reach with your body, arms, and/or legs.
Gross Body Coordination The ability to coordinate the movement of your arms, legs, and torso together when the whole body is in motion.
Gross Body Equilibrium The ability to keep or regain your body balance or stay upright when in an unstable position.
Stamina The ability to exert yourself physically over long periods of time without getting winded or out of breath.
Static Strength The ability to exert maximum muscle force to lift, push, pull, or carry objects.
Trunk Strength The ability to use your abdominal and lower back muscles to support part of the body repeatedly or continuously
over time without ’giving out’ or fatiguing.
37
 Appendix Table 6: Most/Least Physically Demanding Occupations
ONET Score

Panel A: 10 Most Physically Demanding Occupations

Dancers 67.39
Roofers and Slaters 52.56
Plasterers 47.67
Concrete and Cement Workers 46.94
Millwrights 46.33
Recreation Workers 46.11
Structural Metal Workers 45.81
Masons, Tilers, and Carpet Installers 45.40
Helpers, Constructions 44.33
Carpenters 44.28

Panel B: 10 Least Physically Demanding Occupations

Chief Executives and Public Administrators 0.17
Technical Writers 0.33
Metallurgical and Materials Engineers, Variously Phrased 0.33
Sales Engineers 0.33
Record Clerks 0.33
Urban and Regional Planners 0.33
Financial Managers 0.33
Interviewers, Enumerators, and Surveyors 0.44
Physicists and Astronomers 0.50
Architects 0.67

Notes: The estimates are based on ONET data. ONET score shows the importance of physi-
cal abilities in a given occupation on a standardized scale ranging from 0 (min) to 100 (max).

38
 Appendix Table 7: Fatality Rates Across Occupation
Occupation Title Fatalities per 100,000 Workers IPUMS OCC1990 Codes
Managerial and Professional Specialty
Executive, Administrative, and Managerial 2.00 occ1990>=3 & occ1990<=37
Professional Specialty 1.19 occ1990>=43 & occ1990<=200

Technical, Sales, and Administrative Support

Technicians and Related Support Occupations 4.40 occ1990>=203 & occ1990<=235
Sales Occupations 2.38 occ1990>=243 & occ1990<=290
Admin. Support Occupations, including clerical 0.54 occ1990>=303 & occ1990<=391

Service Occupations
Private Household Service Occupations 0.32 occ1990>=405 & occ1990<=408
Protective Service Occupations 10.57 occ1990>=415 & occ1990<=427
Service Occupations, except protective and household 1.16 occ1990>=434 & occ1990<=469

Farming, Forestry, and Fishing

Farming Operators and Managers 29.38 occ1990>=473 & occ1990<=476
Other Agricultural and Related Occupations 15.87 occ1990>=479 & occ1990<=489
Forestry and Logging Occupations 109.32 occ1990==496
Fishers, Hunters, and Trappers 104.79 occ1990==498

Precision Production, Craft, and Repair

Mechanics and Repairers 6.94 occ1990>=505 & occ1990<=549
Construction Trades 10.62 occ1990>=558 & occ1990<=599
Extractive Occupations 93.32 occ1990>=614 & occ1990<=617
Precision Production Occupations 3.01 occ1990>=628 & occ1990<=699

Operators, Fabricators, and Laborers

Machine Operators, Assemblers, and Inspectors 3.05 occ1990>=703 & occ1990<=799
Transportation and Material Moving Occupations 23.89 occ1990>=803 & occ1990<=859
Handlers, Equipment Cleaners, Helpers, and Laborers 11.52 occ1990>=865 & occ1990<=890

Notes: The estimates are obtained based on the data published in 2000 Census of Fatal Occupational Injuries (CFOI). CFOI statistics only report
the total number of fatalities by industries. To obtain fatality rate in an industry, we divide the total number of fatalities by the number of workers
employed in the industry obtained from 2000 CPS and multiply it with 100,000.

39
 Appendix Table 8: Injury Rates Across Industries
Industry Title Injuries/Illnesses per 100 Workers IPUMS IND1990 Codes
Agriculture, Forestry, and Fishing
Agricultural Production (Crops) 6.70 ind1990==10
Agricultural Production (Livestock) 10.40 ind1990==11
Agricultural Services 6.80 ind1990>=20 & ind1990<=30
Forestry 8.80 ind1990==31
Fishing, Hunting, and Trapping 6.70 ind1990==32

Mining
Metal Mining 4.90 ind1990==40
Coal Mining 7.50 ind1990==41
Oil and Gas Extraction 4.20 ind1990==42
Nonmetallic Minerals, except fuels 4.30 ind1990==50

Construction 8.30 ind1990==60

Manufacturing
Food and Kindred Products 12.40 ind1990>=100 & ind1990<=122
Tobacco 6.20 ind1990==130
Textile Mill Products 6.00 ind1990>=132 & ind1990<=150
Apparel and Other Textile Products 6.10 ind1990>=151 & ind1990<=152
Paper and Allied Products 6.50 ind1990>=160 & ind1990<=162
Printing and Publishing 5.10 ind1990>=171 & ind1990<=172
Chemical and Allied Products 4.20 ind1990>=180 & ind1990<=192
Petroleum and Coal Products 3.70 ind1990>=200 & ind1990<=201
Rubber and Misc. Plastics 10.70 ind1990>=210 & ind1990<=212
Leather and Leather Products 9.00 ind1990>=220 & ind1990<=222
Lumber and Wood Products 12.10 ind1990>=230 & ind1990<=241
Furniture and Fixtures 11.20 ind1990==242
Stone, Clay, and Glass Products 10.40 ind1990>=250 & ind1990<=262
Primary Metal Industries 12.60 ind1990>=270 & ind1990<=280
Fabricated Metal Products 11.90 ind1990>=281 & ind1990<=301
Industrial Machinery and Equipment 8.20 ind1990>=310 & ind1990<=332
Electronic and Other Electric Equipment 5.70 ind1990>=340 & ind1990<=350
Transportation Equipment 13.70 ind1990>=351 & ind1990<=370
Instruments and Related Products 4.50 ind1990>=371 & ind1990<=381
Misc. Manufacturing Industries 7.20 ind1990>=390 & ind1990<=392

Transportation and Public Utilities

Railroad Transportation 3.60 ind1990==400
Local and Interurban Passenger Transit 8.00 ind1990>=401 & ind1990<=402
Trucking and Warehousing 7.90 ind1990>=410 & ind1990<=412
Water Transportation 7.00 ind1990==420
Transportation by Air 13.90 ind1990==421
Transportation Services 3.20 ind1990>=422 & ind1990<=432
Communications 2.60 ind1990>=440 & ind1990<=442
Electric, Gas, and Sanitary Services 6.30 ind1990>=450 & ind1990<=472

Wholesale Trade
Durable Goods 5.10 ind1990>=500 & ind1990<=532
Non-durable Goods 6.90 ind1990>=540 & ind1990<=571

Retail Trade
Building Materials and Garden Supplies 8.20 ind1990>=580 & ind1990<=590
General Merchandise Stores 5.90 ind1990>=591 & ind1990<=600
Food Stores 8.00 ind1990>=601 & ind1990<=611
Automotive Dealers and Service Stations 5.60 ind1990>=612 & ind1990<=622
Apparel and Accessory Stores 3.70 ind1990>=623 & ind1990<=630
Furniture and Homefurnishings Stores 4.70 ind1990>=631 & ind1990<=640
Eating and Drinking Places 5.30 ind1990==641
Miscellaneous Retail 3.90 ind1990>=642 & ind1990<=691

40
 Continued – Appendix Table 8: Injury Rates Across Industries
Industry Title Injuries/Illnesses per 100 Workers IPUMS IND1990 Codes
Finance, Insurance, and Real Estate
Depository Institutions 1.40 ind1990>=700 & ind1990<=701
Non-depository Institutions 1.10 ind1990==702
Security and Commodity Brokers 0.60 ind1990==710
Insurance Carriers 1.00 ind1990==711
Real Estate 4.10 ind1990==712

Services
Business Services 3.20 ind1990>=721 & ind1990<=741
Auto Repair, Services, and Parking 5.00 ind1990>=742 & ind1990<=751
Misc. Repair Services 4.90 ind1990>=752 & ind1990<=760
Hotels and Other Lodging Places 6.90 ind1990>=762 & ind1990<=770
Personal Services 3.30 ind1990==761, (ind1990>=771 & ind1990<=791)
Motion Pictures 3.40 ind1990>=800 & ind1990<=801
Amusement and Recreation Services 6.90 ind1990>=802 & ind1990<=810
Health Services 7.40 ind1990>=812 & ind1990<=840
Legal Services 0.70 ind1990==841
Educational Services 3.20 ind1990>=842 & ind1990<=861
Social Services 6.10 ind1990>=862 & ind1990<=871
Museums, Botanical, Zoological Gardens 5.20 ind1990==872
Membership Organizations 3.00 ind1990>=873 & ind1990<=881
Engineering and Management Services 1.70 ind1990>=882 & ind1990<=893
Public Administration 3.20 ind1990>=900 & ind1990<=932

Notes: The injury rates are obtained from 2000 Survey of Occupational Injuries and Illnesses (SOII). SOII incidence rates represent
the number of injuries and illnesses per 100 full-time workers.

41