Content uploaded by Antonio Martins-Neto
Author content
All content in this area was uploaded by Antonio Martins-Neto on Apr 19, 2022
Content may be subject to copyright.
#2022-014
Routine‐biasedtechnologicalchangeandemployeeoutcomes
aftermasslayoffs:EvidencefromBrazil
AntonioMartins‐Neto,XavierCireraandAlexCoad
Published19April2022
MaastrichtEconomicandsocialResearchinstituteonInnovationandTechnology(UNU‐MERIT)
email:info@merit.unu.edu|website:http://www.merit.unu.edu
Boschstraat24,6211AXMaastricht,TheNetherlands
Tel:(31)(43)3884400
UNU-MERIT Working Papers
ISSN 1871-9872
Maastricht Economic and social Research Institute on Innovation and Technology
UNU-MERIT
UNU-MERIT Working Papers intend to disseminate preliminary results of research carried
out at UNU-MERIT to stimulate discussion on the issues raised.
Routine-biased technological change and employee
outcomes after mass layoffs: evidence from Brazil
Antonio Martins-Neto∗1,2, Xavier Cirera3, and Alex Coad4
1UNU-MERIT, The Netherlands
2Maastricht University, The Netherlands
3Finance, Competitiveness and Innovation Global Practice, World Bank
4Waseda Business School, Waseda University
Abstract
We investigate the impact of “routinization” on the labor outcomes of displaced work-
ers. We use a rich Brazilian panel dataset and an occupation-task mapping to examine
the effect of job displacement in different groups, classified according to their tasks. Our
main result is that following a layoff, workers previously employed in routine-intensive
occupations suffer a more significant decline in wages and more extended periods of
unemployment. As expected, job displacement has a negative and lasting impact on
wages. Still, workers in routine-intensive occupations are more impacted than those
in non-routine occupations in terms of wages (an increase of one point in the routine-
intensity index results in a further decline of 2 percent in workers’ relative wages) and
employment. Furthermore, our results indicate that workers in routine-intensive oc-
cupations are more likely to change occupations after the shock, and those who do
not switch occupational fields suffer a more significant decline in wages. Lastly, even
though the loss of employer-specific wage premiums explains 13 percent of displaced
workers’ drop in wages, it does not explain routine-intensive workers’ more substantial
losses.
JEL: J24, J63, O54
Keywords: Routine intensity; Job displacement; Mass layoffs; Occupational mobility;
Brazil
∗Corresponding author: martinsneto@merit.unu.edu
1
1 Introduction
Creative destruction has been referred to as the engine of modern economic growth (Aghion
et al.,2021;Aghion and Howitt,1992;Romer,1990;Schumpeter,1942) and a key driver
of productivity differences across countries (Comin and Hobijn,2010;Comin and Mestieri,
2018;Easterly and Levine,2001). Central to the process of creative destruction is techno-
logical change and how resources are reallocated to firms that are able to disrupt markets
with new technologies. The effects of this process of technological change in labor markets,
however, are not homogeneous. In recent decades, a significant amount of evidence has doc-
umented the increasing polarization and inequality in the labor markets, especially in devel-
oped economies, with the share of high-skill, high-wage, and low-skill, low-wage occupations
growing relative to those in the middle of the distribution. This “hollowing out” of the mid-
dle of the wage distribution has been commonly associated with automation and changes
in the task requirements in production. The routine-biased technological change (RBTC)
hypothesis argues that computers and robots have diminished the demand for routine, repet-
itive tasks in production, which more commonly concentrates among middle-earning workers
(Acemoglu and Autor,2011;Autor et al.,2003;Goos et al.,2009).
The phenomenon of job polarization and its association with technology adoption has
been largely tested and confirmed in the context of advanced economies (see, for instance
Acemoglu and Autor,2011;Autor et al.,2003;de Vries et al.,2020;Dustmann et al.,2009;
Fonseca et al.,2018;Goos et al.,2009;Michaels et al.,2014;Spitz-Oener,2006). In contrast,
the picture is less evident in developing economies, where indications of job polarization are
considerably weaker. Maloney and Molina (2019) and Das and Hilgenstock (2018) find little
evidence of labor market polarization or increased inequality in developing countries, either
in absolute levels of employment or share of the workforce. Gasparini et al. (2021) find similar
results for Latin America’s six largest economies, showing no evidence for polarization in the
labor market (see Martins-Neto et al. (2021) for a literature review of job polarization in
developing economies).
2
Figure 1: Evolution of routine intensity in Brazil
Source: Own elaboration. The routine-intensity (RTI) index is based on Goos et al. (2014). The
following occupations are dropped: legislators and senior officials (ISCO 11); teaching professionals and
teaching associate professionals (ISCO 23 and 33); skilled agricultural and fishery workers (ISCO 61);
and agricultural, fishery and related laborers (ISCO 92).
Despite this lack of apparent “hollowing out” in the middle of the distribution consistent
with job polarization, the empirical literature suggests a decline in routine intensity across
countries and that technological progress has diminished the demand for routine-intensive
occupations in developing countries – which is a crucial precondition for polarization. For
instance, Gasparini et al. (2021) shows a decline in job growth in routine-intensive occupa-
tions in Latin America’s largest economies, and Reijnders and de Vries (2018) document an
increasing share of non-routine occupations in developing countries’ labor forces. In Brazil,
Firpo et al. (2021) shows that despite the lack of job polarization, the routine intensity of
occupations declined considerably. Figure 1 highlights this result, displaying the decline of
routine tasks in Brazil from 2006 to 2018.
With or without the polarization in labor outcomes, a critical question for developing
countries is the implications of this decline in routine-intensive occupations in labor out-
comes, where the extent of job insecurity and informality is larger, and wages are critical for
3
the income distribution. It remains unclear whether workers employed in routine-intensive
occupations are already facing the adverse effects of this process, which groups of workers
are experiencing the adverse effects more strongly, and how large this effect is. So far, most
of the studies in developing economies have focused primarily on aggregate outcomes such
as changes in occupational employment and the extent of job polarization, thus failing to
observe workers’ transitions across occupations and the effects on individuals’ wages and
unemployment duration. This paper attempts to fill this gap in the literature.
One challenge when measuring the impact of exogenous changes in the demand for rou-
tine tasks on labor outcomes is the fact that it is difficult to disentangle the effects of “rou-
tinization” from endogenous decisions and responses from workers. Therefore, we employ an
event-study approach (Blien et al.,2021;Couch and Placzek,2010;Jacobson et al.,1993;Ra-
poso et al.,2019), treating mass layoffs as an external shock. Specifically, to better identify
the role of “routinization” on employment outcomes, we use this exogenous sizable negative
shock - mass layoffs - and explore how re-employment probabilities and wage dynamics vary
by the level of occupations’ routine intensity.
Our study contributes to the labor economics literature on outcomes across displaced
workers. Following the seminal work of Jacobson et al. (1993), studies have found that
workers face a significant decline in salaries after displacements, with sustained effects rang-
ing from 3% to 25% depending on the region and methodology (see, for instance Couch and
Placzek,2010;Eliason and Storrie,2006;Hijzen et al.,2010;Huttunen et al.,2006;Ichino
et al.,2017;Kaplan et al.,2005;Menezes-Filho,2004;Raposo et al.,2019). Using Brazil’s
matched employer-employee data set (RAIS) in the mid-1990s, Menezes-Filho (2004)’s pi-
oneering paper finds that high-tenure workers suffer a long-term loss in monthly wages of
about 20% per year. Saltiel (2018) examined the Brazilian labor market outcomes over 2002-
2012, with similar results to Menezes-Filho (2004): affected workers suffer annual earnings
losses exceeding 15-20%, and the effect persists through the medium term. We contribute
to this literature by examining a number of heterogeneities across workers’ groups, while
4
also exploring a larger and more recent sample. For instance, Menezes-Filho (2004) focuses
on male workers in the state of S˜ao Paulo and Saltiel (2018) includes only displaced work-
ers that do not face longer periods of unemployment, thus underestimating the impacts of
displacement. In addition, following (see, for instance, Bertheau et al.,2022;Fallick et al.,
2021;Lachowska et al.,2020), we estimate the loss of employer-specific wage premiums.
We also contribute to the recent literature in labor economics that has looked into the
effects of technological change and occupational differences across displaced workers. Bessen
et al. (2019) explore the direct impact of technology adoption at the firm level on workers’
probability of separation from their current jobs and their future labor prospects. They find
that automation at the firms increases workers’ separation risk and that displaced workers are
more likely to work fewer days in the years to come. However, differently from our analysis,
the authors do not explore differences between workers previously employed in different
occupations. Goos et al. (2021) examined survey data of workers previously employed in a
large Belgian establishment in the automotive sector. After the plant closed and in line with
the RBTC hypothesis, workers in routine-intensive occupations were less likely to find a job
1,5 years after the event. Additionally, for those workers who could find a job, the non-routine
content of job tasks was higher, wages were lower, and permanent jobs were less frequent.
In line with our results, they find a more significant impact on wages and employment for
displaced workers previously employed in routine-intensive occupations (compared to their
non-routine-intensive counterparts). However, the authors concentrate the analysis on a
case study of one firm, which raises questions about possible selection bias, endogeneity, and
generalizability of the results. In turn, our work is closest to a recent analysis of German
displaced workers. Using data from 1980 to 2010, Blien et al. (2021) test whether workers in
routine intensive occupations are disproportionately affected by job separation. They find
evidence that workers in routine occupations undergo more considerable and more persistent
wage losses and that the difference compared to non-routine workers has increased over time.
An important contribution of this study is to examine the impact of RBTC in the context
5
of a middle-income country such as Brazil. Brazil presents an interesting case for comparison
for various reasons besides the previously-mentioned weaker evidence of job polarization.
First, labor market institutions have exacerbated market frictions and mismatches in the
labor market (Ulyssea,2010). Second, minimum wage policies have helped decrease wage
inequality significantly in the last decade, and the wage gap between low and high skilled
workers narrowed significantly (Alvarez et al.,2018;Firpo et al.,2021). Third, productivity
growth has remained stagnant, suggesting a lack of significant technological change. Fourth,
in many developed economies, participation in GVCs spurred “routinization”. However,
Brazil has remained relatively isolated from global offshoring, with low participation in
GVCs and services trade due to restrictive trade policies and lack of skills. This combination
of labor institutions and the lack of internationalization of Brazilian companies makes the
country a very interesting case study to explore the impact of “routinization.”
An additional important contribution of the paper is the heterogeneity analysis, including
the differences between female and male workers, long- and short-tenured individuals, and
firms’ size. These dimensions seem to play a critical role in explaining the adverse effects of
displacement. Also, we explore some possible mechanisms explaining the larger decline in
wages, especially the roles of demand, job switchers, and firms’ heterogeneity.
To advance some of the main findings, we observe a significant and long-lasting negative
impact of job displacement on workers’ wages and employment. Based on Jacobson et al.
(1993)’s methodology, the results show a large and statistically significant wage loss associ-
ated with job displacement. Workers in the treated arm see their relative monthly earnings
decline over 20% in the year following the layoff and up to 5% five years after the event. The
shock also affects workers’ relative employment, as displaced individuals work over 15% less
in t+ 1 and 3% less five years after the layoff. In addition, we find that the loss of employer-
specific wage premiums explains 13% of the decline in wages for the treated group. Following
the initial results, we test for differences between routine and non-routine workers. First, we
find strong evidence that workers in routine-intensive occupations are more impacted than
6
those in non-routine occupations. An increase of one point in the routine-intensity index
results in a further decline of 2% in workers’ relative wages and an increase of 1% in the
chance of unemployment. Second, we explore the heterogeneity in our results and find a
more significant decline in wages for male, less educated, and long-tenured individuals in
routine intensive occupations. In addition, our findings suggest that the negative impact
is larger in sectors with a larger decline in the demand for routine tasks. Third, we show
that workers in routine-intensive occupations are more likely to change occupations after
the shock. However, those unable to switch fields experience a more significant decline in
wages. Lastly, we find that the loss of employer-specific wage premiums does not explain
routine-intensive workers’ more substantial reduction in wages.
The paper is organized as follows. The following section describes the data sources and
the definition of involuntary displacement events. Section 3 describes the empirical strategy.
Section 4 estimates the impact of job displacement in Brazil and examines the heterogeneity
across occupational groups, especially routine-intensive occupations. Section 5 examines the
heterogeneity across occupational groups and investigates the importance of the demand for
routine occupations in explaining labor outcomes from displacement; including differences
across sectors, between occupational switchers and non-switchers, and the role of firms’
heterogeneity. The last section concludes.
2 Data and sample construction
2.1 Data
To estimate the impact of displacement on wages in Brazil, we use the RAIS database
(Rela¸c˜ao Anual de Informa¸c˜oes Sociais) from 2006 to 2018. This is an administrative
database from the Brazilian Ministry of Economy considered a high-quality census of the
Brazilian formal labor market. The census includes all establishments nationwide with at
least one registered worker — even though we carry our analysis at the establishment level, we
7
refer to firms and establishments interchangeably. The data includes over 30 million employ-
ees per year, matched with firm information, including location and industry, and workers’
gender, age, education, employment status, wages, type of contract, tenure, and hiring date.
RAIS reports compensation as the monthly average wage received by each worker (including
regular salary payments, holiday bonuses, performance-based and commission bonuses, tips,
and profit-sharing agreements).
We restrict our analysis to employees in private establishments, and focus on workers
displaced in 2009-2013 due to establishments’ closure or mass layoffs (see definition below).
We observe workers’ outputs three years before displacement and five years following the
event. This period includes both a moment of fast national economic growth (2009-2012)
and a period of economic stagnation with recessions in 2015 and 2016 when GDP dropped by
3.5% and 3.2%, respectively. Thus, for workers displaced in 2009, the booming labor market
should have facilitated their reinsertion. In contrast, for workers displaced in 2013, the entire
period following the shock is a period of wage stagnation and increased unemployment.
The database includes information on each worker’s occupation, coded according to the
Brazilian Code of Occupations (CBO). To measure the task content of occupations, we follow
Goos et al. (2014), who mapped the routine intensity index (RTI) to ISCO-88 occupations.
This RTI measure is based on Autor et al. (2003) and combines five task measures from the
US Dictionary of Occupational Titles (DOT) to produce three aggregate measures: Manual,
Routine, and Abstract task measures.1The Routine Task Intensity (RTI) index takes the
difference between the log of Routine tasks and the sum of the log of Abstract and the log
of Manual tasks.2We map these occupations to the Brazilian Code of Occupations. Table 1
1Specifically, Manual tasks relate to the occupation’s demand for “eye-hand-foot coordination” (EYE-
HAND), and Abstract tasks refer to the simple average of occupations’ managerial and interactive tasks
(DCP) and mathematical and formal reasoning requirements (GED-MATH). In contrast, the Routine task
measure is a simple average of the following variables: “set limits, tolerances and standards” (STS), which
measures an occupation’s demand for routine cognitive tasks; and “finger dexterity” (FINGDEX), which
measures an occupation’s use of routine motor tasks.
2Even though the task content of occupations may differ across countries (see, for instance, Lewandowski
et al. (2019)), we assume that the ranking of occupations in terms of routine intensity may not vary signifi-
cantly across countries.
8
describes the occupations ranked by the level of routine tasks; RTI is highest at 2.24 for
office clerks (41) and lowest at -1.52 for managers of small enterprises (13).
Table 1: Routine-intensity by occupation
Occupation RTI Index
Managers of small enterprises -1,52
Drivers and mobile plant operators -1,50
Life science and health professionals -1,00
Physical, mathematical and engineering professionals -0,82
Corporate managers -0,75
Other professionals -0,73
Personal and protective service workers -0,60
Other associate professionals -0,44
Physical, mathematical and engineering associate professionals -0,40
Life science and health associate professionals -0,33
Extraction and building trades workers -0,19
Sales and service elementary occupations 0,03
Models, salespersons and demonstrators 0,05
Stationary plant and related operators 0,32
Laborers in mining, construction, manufacturing and transport 0,45
Metal, machinery and related trade work 0,46
Machine operators and assemblers 0,49
Other craft and related trade workers 1,24
Customer service clerks 1,41
Precision, handicraft, craft printing and related trade workers 1,59
Office clerks 2,24
Source: Own elaboration. The following occupations are dropped: legislators and senior officials (ISCO
11); teaching professionals and teaching associate professionals (ISCO 23 and 33); skilled agricultural and
fishery workers (ISCO 61); and agricultural, fishery and related laborers (ISCO 92). The routine-intensity
(RTI) index is based on Goos et al. (2014).
2.2 Sample construction and matching
Our identification strategy rests in examining the impacts of a sudden exogenous shock on
workers’ career prospects. Specifically, we look at individuals displaced due to establish-
ments’ closures or mass layoffs. Yet, the RAIS database does not carry information on the
year an establishment closes. Instead, as is commonly done in the literature, we use the
establishment’s unique identifier and define exits when the employer identifier ceases to exist
(see, for instance Schwerdt et al.,2010). For example, we assign an establishment in 2010
as closed if it appears in our database in the years preceding 2010 and disappears afterward.
9
Furthermore, we define mass layoffs when 30% or more workers are laid off between t−1 and
t. We impose an additional restriction to avoid capturing seasonal changes in employment
and exclude cases in which employment fluctuated by 20% in the two years before the mass
layoff, or the firm size went above 150% compared to the year of the layoff. To put it sim-
ply, we exclude cases in which the trend was already perceived in the years before or when
employment recovers in the years following. In addition, some of these events might not be
actual closures. Establishments can change their identifier in time or spin-off into different
companies. We impose an additional restriction to capture these cases and exclude cases in
which more than 50% of the employees continue under a new employer identifier.3
As it is commonly done in the mass layoff literature, we restrict our sample to full-time
prime-age workers, focusing on individuals older than 25 or younger than 50 years in the first
year of analysis (for instance, for workers displaced in 2010, the first year of study is 2007).
The reason to restrict workers’ age is that younger workers can be working as apprentices
or interns, while older workers can opt to leave the market and retire. We also restrict
establishments’ size, focusing on establishments with at least 30 employees in the first year
before the event. We also limited our sample to one observation per worker-year by choosing
the highest-paying in any given year. We also excluded observations where the data were
miscoded or missing. Furthermore, we impose that displaced individuals work in the same
company for at least three years before the layoff. By setting this restriction, we focus on
workers in stable positions, who would have likely continued had the closure not occurred.
To estimate the effects of displacement, we include a control group with workers who
continue to work at firms that had no mass layoff during the period of analysis. For this
control group, we also impose at least three years of tenure before the “potential displace-
ment”. In our analysis, displacement affects workers at different times (i.e. there is variation
in treatment timing, Roth et al.,2022), and therefore to ensure the validity of the difference-
3One downside in using RAIS database is that it only covers formal workers. In this scenario, if a worker
becomes unemployed or moves to the informal sector, which comprises about 40% of the Brazilian labor
market, we will not be able to track her. Therefore, transitions from the formal to the informal are not
captured in our analysis, being thus treated as movements to unemployment.
10
in-difference setup, we guarantee that the control group includes only workers that will never
be part of a mass layoff in subsequent years (de Chaisemartin and D’Haultfœuille,2020),
hence avoiding the problem of “forbidden” comparisons (Roth et al.,2022). However, other
than that, we do not include any additional restriction in the years following the “potential
displacement”. In other words, we aim to compare long-tenured workers with a control group
of individuals that are as similar as possible in all domains, except for the displacement.
To identify a set of control workers, we implement a two-stage matching procedure in
t−2. First, we perform exact matching on workers’ occupations (2-digits), gender, and on
Brazil’s 27 states. In the second step, we implement the coarsened exact matching (CEM)
algorithm (Iacus et al.,2012). Then, we apply the CEM algorithm on a series of covariates
at both worker-level (wage, wage growth, age, tenure, and education) and establishment-
level (number of workers, average salary, and sector (2-digits)). By including workers’ wage
growth, we ensure that workers display similar trends in salaries before the shock, a key
identification restriction of the difference-in-difference estimator.
This matching procedure yields a sample of about 135 thousand treated workers and 135
thousand workers in the control group. Table 2 shows the descriptive statistics of various
workers’ and establishments’ characteristics for workers in the treatment and control arms
two years before the layoff. The last column shows the difference between the means. Workers
in the control group earn slightly more than treated individuals, although the difference is
not statistically significant. Displaced workers have similar age and tenure as the control
group and work in larger firms. In contrast, firms’ average wage is not statistically different
between displaced and control workers. As expected, most individuals in our sample are
placed in the Southeast of Brazil. This is the most populous region in the country and
includes the state of S˜ao Paulo, the wealthiest state in Brazil. In addition, about one-third
is employed in the manufacturing sector, and about one-third of workers in the sample are
female. Less than 10% of our sample has a college degree. In contrast, 49% has only a
high-school diploma (Figure A1 shows the histogram of the routine-intensity index for the
11
matched sample).
Table 2: Comparison of treated and control groups after matching
Control Treated
Mean Standard Deviation Mean Standard Deviation Difference
Wage 1478 1498.46 1487 1508.68 9.442
Wage Growth .11 0.26 .1 0.26 -0.005***
Worker’s age 35 6.34 35 6.34 -0.002
Gender .32 0.47 .32 0.47 —
Illiterate or primary school .026 0.16 .026 0.16 0.000
Primary school graduate .16 0.37 .16 0.37 0.000
Middle school graduate .24 0.43 .24 0.43 -0.000
High-school graduate .49 0.50 .49 0.50 -0.000
College degree .081 0.27 .081 0.27 0.000
Tenure 63 43.24 63 43.25 -0.282**
Size (30-49) .15 0.35 .12 0.33 -0.025***
Size (50 - 99) .17 0.38 .17 0.37 -0.008*
Size (100-499) .38 0.49 .4 0.49 0.011
Size (500+) .29 0.46 .32 0.46 0.022***
Firm’s average wage 1533 1201.26 1548 1225.56 15.328
Agriculture and Extractive .025 0.16 .025 0.15 -0.001
Manufacturing .36 0.48 .36 0.48 -0.000
Services .61 0.49 .61 0.49 0.001
North .02 0.14 .02 0.14 —
Northeast .12 0.32 .12 0.32 —
Southeast .71 0.46 .71 0.46 —
South .12 0.33 .12 0.33 —
Central-West .036 0.19 .036 0.19 —
Observations 135.566 — 135.566 — —
Table shows averages for baseline. The last column is the coefficient of a simple regression of treatment
status on the variable, with robust standard errors. The groups are perfectly matched for gender,
occupation, and state. Stars indicate whether this difference is significant. * p <0.10, ** p <0.05, *** p
<0.01.
3 Empirical strategy
We are interested in exploring how workers in different occupational groups respond to a
sudden shock in their careers. In doing so, we follow extensive literature and employ an
event-study approach (Blien et al.,2021;Couch and Placzek,2010;Jacobson et al.,1993;
Raposo et al.,2019). Mass layoffs are taken as external shocks to estimate the effect of an
involuntary job loss on earnings and employment prospects. In essence, we aim to compare
12
the wage and employment changes of treated individuals over the medium-run with the wage
changes that would have occurred if they had not lost their jobs. Given that we aren’t able to
observe the latter, we build a control group. In doing so, a robust methodology is the use of
matching techniques in combination with difference-in-differences (DiD) methods (see Blien
et al.,2021;Cunningham,2021;Heckman et al.,1997). Following the matching procedure
described in the previous section, we follow Jacobson et al. (1993) and estimate:
yit =α0+
5
X
k=−3,k̸=−2
[νk
t+νk
tTiβk] + λi+θt+δs+σj+ϵijst (1)
where yit is the outcome of interest (relative monthly salary or employment). Relative wages
are measured compared to worker’s compensation in t−2, while employment is a dummy
equal to one is the worker has any positive labor earnings in a given year. Wages are taken
as zero whenever the individuals are unemployed. Tiis a treatment indicator that is equal to
one if the worker faced a layoff and zero otherwise, and νtrepresents time-to-event dummies,
from 3 years before the event to five years after it (t-2 is the baseline). The coefficients
βkare our outcome of interest and measure the differences in relative earnings or relative
employment for displaced and non-displaced workers from three years before the shock to
five years after. λiand θtrepresent individual and time fixed effects and capture permanent
unobserved individual characteristics and general patterns in the economy, respectively. In
contrast, σjand δsrepresent common region and sector effects. To estimate the difference
between routine and non-routine occupations, we follow Blien et al. (2021) and modify
Equation 1 such that:
yit =α0+
5
X
k=−3,k̸=−2
[νk
t+νk
tTiβk+νk
tRT Iiαk]+
5
X
k=−3,k̸=−2
νk
tTiRT Iiρk+λi+θt+δs+σj+ϵijst (2)
where RT Iiis the routine intensity index described in Table 1. In addition to worker fixed
effects, time-to-event dummies, and regular year, region, and sector dummies, Equation 2
13
also includes interactions between time-to-event dummies and routine intensity (αkνk
tRT Ii)
and the triple interaction between routine intensity, treatment, and time to event dummies
(P5
k=−3νk
tTiRT Iiρk). The interaction between time-to-event dummies and routine intensity
captures common trends in the occupational groups irrespective of treatment, while the
triple interaction term measures the additional effect in a specific year due to an increase in
the routine intensity index. The latter is our main outcome of interest.
4 The adverse effects of job displacement
We start by first exploring the impact of job displacement on wages and employment in
Brazil. Figure 2 plots the coefficients from Equation 1 and shows as expected that displaced
individuals face a substantial decline in relative wages in t+ 1 compared to the control
group, which is only partially recovered in the following years. For instance, in t+5, treated
individuals earn over 5% less than the control group. The identifying restriction rests on
whether displaced and non-displaced workers have parallel trends in the outcome variables
before the event. In the years before the displacement, the coefficients were not statistically
different from zero, which implies that the earnings profiles of workers were the same up
to the shock. However, following the shock, treated workers earn substantially less (about
20%) than two years before the event. Our results are larger than those in Saltiel (2018) but
much smaller than those from Menezes-Filho (2004), who found salary losses of up to 30%.
The differences with our results are likely related to differences in our sample. For instance,
Saltiel (2018) focuses on displaced workers that find a job in the year of displacement, thus
resulting in estimates that are biased towards smaller negative effect sizes. On the other
hand, in addition to focusing exclusively on the state of S˜ao Paulo, Menezes-Filho (2004)
does not perform a matching between control and treated workers, thus likely resulting in
larger negative results.
14
Figure 2: Effect of displacement on relative wages
The figure shows the estimates of time-to-event dummies interacted with a displacement indicator from
a regression including individual, region, sector, time-to-event dummies, and year fixed effects. The
dependent variables is relative wages. Relative wages are measured dividing worker’s monthly average
wage by the worker’s average wage in year t−2. Year t−2 is the base year. Vertical bars show estimated
95% confidence interval based on standard errors clustered at individual level.
Figure 3 presents the effect on the other outcome of interest, workers’ employment. In
years preceding the shock, given that workers were employed full-time, the coefficients are
equal to zero. However, following the displacement, treated workers are 18% less likely to
be in formal jobs than the control group. In the following years, the impact on employment
declines to 5%, with the impact lasting over the medium-run. For instance, in year t+5,
displaced individuals are 3.4% less likely to be in formal employment than the control group
(Table A1 in the Appendix presents the coefficients for each year).
15
Figure 3: Effect of displacement on employment
The figure shows the estimates of time-to-event dummies interacted with a displacement indicator from
a regression including individual, region, sector, time-to-event dummies, and year fixed effects. The
dependent variables is employment. Employment is a dummy equal to one is the worker has any positive
labor earnings in a given year. Year t−2 is the base year. Vertical bars show estimated 95% confidence
interval based on standard errors clustered at individual level.
We further explore the heterogeneity of our results and group workers into different
categories to examine some of the drivers of the adverse effects of displacement. Table 3
shows the baseline estimates of the averages of the estimates over the 6 years from the shock
(from tto t+ 5) of time-to-event dummies interacted with a displacement indicator from a
regression including individual, region, sector, time-to-event dummies, and year fixed effects.
16
Table 3: Effect of displacement on relative wages and employment by group
Relative wages Relative employment
Mean effect Stand. Errors Mean effect Stand. Errors Observations
Age
40 years or younger -0.0726*** (0.00232) -0.0524*** (0.00114) 1.734.309
41 years or older -0.0688*** (0.00314) -0.0662*** (0.00184) 702.449
Tenure
72 months or less -0.0646*** (0.00242) -0.0489*** (0.00124) 1.547.145
73 months or more -0.0845*** (0.00300) -0.0694*** (0.00155) 889.612
Education
Without high-school -0.0536*** (0.00269) -0.0557*** (0.00149) 1.027.583
High-school -0.0819*** (0.00276) -0.0550*** (0.00138) 1.201.266
College graduate -0.101*** (0.00757) -0.0689*** (0.00339) 207.909
Gender
Female -0.0886*** (0.00347) -0.0756*** (0.00190) 779.831
Male -0.0635*** (0.00224) -0.0473*** (0.00110) 1.656.927
Firm size
100 or less employees -0.104*** (0.00368) -0.0799*** (0.00185) 735.912
101 or more employees -0.0579*** (0.00219) -0.0476*** (0.00114) 1.700.846
The table shows averages of the estimates over the 6 years from the shock (from tto t+5) of time-to-event
dummies interacted with a displacement indicator from a regression including individual, region, sector,
time-to-event dummies, and year fixed effects. In other words, the table shows the “average over years”
obtained from a single dummy variable for the entire period t:t+5. The dependent variables are relative
wages and employment. Relative wages is measured dividing worker’s monthly average wage by the
worker’s average wage in year t−2. Employment is a dummy equal to one is the worker has any positive
labor earnings in a given year. Year t−2 is the base year. Standard errors clustered at individual level
are reported in parenthesis. ***, ** and * respectively indicate 0.01, 0.05 and 0.1 levels of significance.
Several interesting facts emerge. First, in terms of employment, and consistent with the
literature (Deelen et al.,2018), the adverse effects are more significant for older workers. 41
years or older workers face a decline in employment about 1 percentage points larger than
younger individuals. In contrast, older workers are less affected in terms of relative wages.
In addition, similar to Saltiel (2018), the impact is more significant for long-tenured workers,
reflecting the importance of breaking employer-employee matching and the destruction of
firm-specific human capital for explaining sustained wage losses. Workers with over 72
months of experience in the same firm see their wages declining on average 8% relative to
two years before the displacement, while short-tenured individuals see a decline of 6%. Long
tenured workers are also more impacted by a decrease in relative employment (6.9%) than
short-tenure workers (4.8%). We also find that male workers are less impacted than female
individuals in terms of relative wages and employment. The results are different to those
17
observed in Carneiro and Portugal (2006), who use Portuguese matched employer-employee
database and find that the effects of displacement are larger for men (12%) than women
(9%). In addition, we find that high-educated workers are more significantly affected both
in terms of wages and employment. Furthermore, workers in smaller companies see a more
substantial decline in wages and employment than those in larger companies.
5 Routine intensity and the cost of displacement
5.1 The role of tasks
Mass layoffs are a natural experiment to explore the role of routine intensity in the impact
on workers, given that they represent an exogenous and often unexpected shock to workers.
If wages solely reflect workers’ observable characteristics, pre- and post- displacement wages
should show minor variation. However, some factors external to the worker can affect the
labor outcomes of the displaced. Critical among these factors are those that affect routine
intensity of those tasks. A first element that suggests that the type of tasks carried out by the
worker matters is technology. Technological progress does not equally affect all occupations.
For example, a well known fact is that automation diminishes the demand for routine tasks
(Autor and Dorn,2013), so that workers in routine-intensive occupations suddenly find
themselves in a less favorable market, making it a challenge to recover from losing their jobs.
Arnoud (2018) shows that even under low technology adoption, the threat of automation
can lower wage growth of occupations more susceptible to automation. Second, structural
transformation and the decreasing share of manufacturing in the economy can also reduce
the demand for routine intensive occupations and worsen the outcomes of displaced workers
in those sectors (B´ar´any and Siegel,2018).
18
Figure 4: Effect of displacement on relative wages by occupational group
The figure shows the estimates of time-to-event dummies interacted with a displacement indicator from
a regression including individual, region, sector, time-to-event dummies, and year fixed effects. The
dependent variables is relative wages. Relative wages is measured dividing worker’s monthly average
wage by the worker’s average wage in year t−2. Year t−2 is the base year. Low-routine are workers
in the first quartile of the routine-intensity index, while high-routine indicates workers in the fourth
quartile. Vertical bars show estimated 95% confidence interval based on standard errors clustered at
individual level.
Figure 4 provides a first glance at the impact of job displacement on workers’ wages
for different occupational groups. Using the routine-intensity index described in Table 1,
we group workers into low-routine occupations (first quartile) and high-routine occupations
(fourth quartile). The figure compares both groups and shows that workers in high-routine
occupations are substantially more harmed than those at the bottom of the routine distri-
bution. The decline in wages in t+ 1 is over 5% larger for workers in the fourth quartile,
with the effect persisting over the medium run.
Table A2 shows the results of a similar exercise than in Table 3, and reinforces the fact
that workers in high-routine occupations are more impacted in terms of wages. In addition,
Figure 5 shows that workers in the fourth quartile are 2% more likely to face unemployment
in the years following the shock.
19
Figure 5: Effect of displacement on employment by occupational group
The figure shows the estimates of time-to-event dummies interacted with a displacement indicator from
a regression including individual, region, sector, time-to-event dummies, and year fixed effects. The
dependent variables is employment. Employment is a dummy equal to one is the worker has any positive
labor earnings in a given year. Year t−2 is the base year. Low-routine are workers in the first quartile
of the routine-intensity index, while high-routine indicates workers in the fourth quartile. Vertical bars
show estimated 95% confidence interval based on standard errors clustered at individual level.
These initial results, however, do not account for some critical differences between occu-
pational groups; especially the fact that trends in wages might differ between occupational
groups. A more formal estimate is presented in Figure 6, which offers the estimates from
Equation 2 and presents the coefficient of the triple interaction term (ρk) taking relative
wages as the dependent variable. Table A5 in the Appendix show the coefficients for both
wages and employment. The interpretation of these estimates is by how many percentage
points the earnings loss in a specific year is magnified due to an increase in 1 point in the
routine intensity index, which in turn varies from -1.52 to 2.24.
20
Figure 6: Routine task intensity and the effect of displacement on relative wages
The figure shows the estimates of the triple interactions between time-to-event dummies interacted with
a displacement indicator and a routine intensity measure from a regression including individual, region,
sector, time-to-event dummies, time-to-event dummies interacted with the routine intensity measure,
and year fixed effects. The dependent variables is relative wages. Relative wages is measured dividing
worker’s monthly average wage by the worker’s average wage in year t−2. Year t−2 is therefore the
base year. Vertical bars show estimated 95% confidence intervals based on standard errors clustered at
individual level.
The results suggest that an increase in 1 point in the RTI results in a further decline of
about 2% on relative wages across the years and up to five years following the shock. For
instance, a worker previously employed in metal and machinery (RTI equals 0.46) would face
a decline 2% lower than a worker once hired as precision, handicraft, and craft printing (RTI
equals 1.59). In addition, Figure 7 shows that workers in routine-intensive occupations are
also more likely to face more extended periods of unemployment – a 1 point increase in the
RTI increases the chance of unemployment by 1%. Our findings are similar to those in Blien
et al. (2021) and Goos et al. (2021), who also find a negative impact of being previously
employed in routine-intensive occupations. In addition, the results are somewhat consistent
with Firpo et al. (2021), who find some evidence of earnings polarization in Brazil.4
4Regarding the more significant adverse effect of displacement for workers in high routine occupations,
Table A3 suggests that these workers are less likely to be part of a mass layoff (compared to a firm clo-
sure). Workers experiencing a mass layoff are significantly less routine intensive on average than workers
experiencing a firm closure (RTIs of 0.17 vs 0.31 respectively, p-value <10%).
21
Figure 7: Routine task intensity and the effect of displacement on employment
The figure shows the estimates of the triple interactions between time-to-event dummies interacted with
a displacement indicator and a routine intensity measure from a regression including individual, region,
sector, time-to-event dummies, time-to-event dummies interacted with the routine intensity measure,
and year fixed effects. The dependent variables is employment. Employment is a dummy equal to one is
the worker has any positive labor earnings in a given year. Year t−2 is therefore the base year. Vertical
bars show estimated 95% confidence intervals based on standard errors clustered at individual level.
Keeping our focus on job displacement and routine intensity, we investigate the robustness
of our findings according to heterogeneity in individuals’ characteristics.5Table 4 examines
the heterogeneity of these findings, as in Table 3, while estimating Equation 2. Some in-
teresting findings emerge in the analysis of relative wages. First, as in the previous results,
the effect of routine-intensity is larger for long-tenured individuals. Second, we observe that
the impact is more significant for male and less-educated individuals. Lastly, the impact
is larger for workers previously employed in larger establishments. As for employment, we
observe similar results. Older and long-tenure individuals face more extended periods of
unemployment. In addition, female and college graduate workers don’t show statistically
significant results. In the following section, we try to account for these significant impacts,
5We cannot rule out that our estimates for RTI and displacement outcomes may not correspond to causal
estimates of RTI on displacement outcomes, because of potential correlations between RTI and workers’
characteristics (such as gender, education, or other variables that remain unobserved). While a detailed
analysis of workers’ characteristics is beyond the scope of the current paper, further work on different
samples would be welcome.
22
considering differences across sectors, the role of job switchers, and firm heterogeneity.
Table 4: Effect of routine intensity on relative wages and employment by group
Relative wages Relative employment
Mean effect Stand. Errors Mean effect Stand. Errors Observations
Age
40 years or younger -0.0141*** (0.00223) -0.00651*** (0.00105) 1.687.041
41 years or older -0.0253*** (0.00307) -0.0128*** (0.00173) 677.402
Tenure
72 months or less -0.0126*** (0.00236) -0.00521*** (0.00115) 1.504.215
73 months or more -0.0243*** (0.00289) -0.0113*** (0.00142) 860.227
Education
Without high-school -0.0284*** (0.00286) -0.0153*** (0.00151) 976.112
High-school -0.0119*** (0.00257) -0.00586*** (0.00123) 1.193.220
College graduate -0.00236 (0.00658) 0.00187 (0.00293) 195.111
Gender
Female -0.00973*** (0.00349) 0.00186 (0.00175) 758.996
Male -0.0187*** (0.00217) -0.00930*** (0.00104) 1.605.447
Firm size
100 or less employees -0.00307 (0.00356) 0.000296 (0.00168) 706.671
101 or more employees -0.0212*** (0.00213) -0.00941*** (0.00106) 1.657.772
The table shows averages of the estimates over the 6 years from the shock (from tto t+ 5) of the
triple interactions between time-to-event dummies interacted with a displacement indicator and a routine
intensity measure from a regression including individual, region, sector, time-to-event dummies, time-
to-event dummies interacted with the routine intensity measure, and year fixed effects. In other words,
the table shows the “average over years” obtained from a single dummy variable for the entire period
t:t+5. The dependent variables are relative wages and employment. Relative wages is measured dividing
workers monthly average wage by average wage in year t−2. Employment is a dummy equal to one is
the worker has any positive labor earnings in a given year. Standard errors clustered at individual level
are reported in parenthesis. ***, ** and * respectively indicate 0.01, 0.05 and 0.1 levels of significance.
Our main finding suggests that workers in routine-intensive occupations face a more
considerable decline in wages and employment following a mass layoff. In other words there
is evidence of non-routinization affecting workers outcomes in Brazil, given that the falling
demand for routine workers has impacted their ability to find similar, good-paying jobs. As
a result, a critical question is to understand what are the main the mechanisms that could
explain these effects on workers.
5.2 The role of decreasing demand
In exploring the possible mechanisms, we first look at differences in the demand for routine
occupations across sectors and test whether workers initially employed in industries with
23
falling demand for routine tasks are more considerably affected. Second, we test for the
importance of job switchers in explaining our results. For example, following a displacement
and given the lower demand for these occupations, workers may move to occupationally
distant jobs, which is usually associated with lower re-employment wages (Huckfeldt,2018;
Lysho,2020). Therefore, we examine first whether workers in routine occupations are more
likely to switch fields, and then we test for different impacts among switchers and non-
switchers. Lastly, differences in firm characteristics can help shed some light on our results.
In particular, there are significant differences in firms’ wage premiums in Brazil (Alvarez
et al.,2018). In this context, we examine whether workers in routine occupations are more
likely to move to low-paying firms upon re-employment. Given the decline in the demand for
such occupations, displaced individuals could face more difficulties finding a better-paying
firm, thus, ending with a lower wage.
Figure 1 shows a constant decline in routine intensity in Brazil from 2006 to 2018. Yet, the
aggregate measure hides significant heterogeneity across sectors. The decrease in the demand
for routine occupations combines within-industry and between-industry changes. On the one
hand, as firms adopt more sophisticated and automated technologies, a given industry will
use less routine employment to produce similar output levels. On the other hand, routine
intensity differs across sectors, such that sectoral employment shifts also explain aggregate
occupational share changes (Goos et al.,2014).
24
Figure 8: Routine Intensity and Change in Industries’ Employment Share
Circles represent 87 sectors, weighted by total employment in 2006. The x-axis is the RTI index,
calculated as the weighted occupational index. The y-axis is the change in the share of employment in
each sector from 2006 to 2018, measured in percentage points. The coefficient in the linear regression is
-0.00117, with standard error equal to 0.0012.
As a first exercise, we test for the association between initial routine intensity across
sectors and the change in employment share from 2006 to 2018. Figure 8 shows a negative
correlation, albeit weak, between RTI and employment change across industries in Brazil,
hence suggesting that most of the changes in the index might have occurred within sectors.
To have a better grasp of these dynamics, we decompose the difference from 2006 to 2018 in
the RTI index into changes within and between industry groups:
∆RT I =X
i
∆RT IiSi0+X
i
RT Ii0∆Si+X
i
∆RT Ii∆Si(3)
where iindexes industries. RT Iiaccounts for the routine-intensity (measured as the occu-
pational weighted index) of industry iand ∆RT Iiaccounts for the change in RTI of unit
i.Siis the share of industry iin total employment, and ∆Siis the change in the share in
total employment of industry iover the period. The first term in the RHS is the contri-
bution of RTI growth in each industry (within industry), assuming that employment shares
remain unchanged. The second term in the RHS is related to changes in employment shares
25
(between industry), while the RTI index in each sector is kept constant. Finally, the third
term is a dynamic term, giving the contribution to the total RTI index due to a rise in the
employment share in sectors whose RTI has increased in the period.
Table 5: RTI decomposition, 2006-2018
Mean
RTI 2006 RTI 2018 Total Change
0.402 0.283 -0.12
Decomposition (Raw)
Within Between Dynamic
-0.112 -0.012 0.005
Decomposition (Percentage)
Within Between Dynamic
0.94 0.10 -0.04
All proportions and means are weighted by occupational employment in 2006 or 2018. ∆ is the change
in the average proportion or mean from 2006 to 2018.
Table 5 presents the results of the shift-share decomposition. From 2006 to 2018, the
RTI index in Brazil dropped 0.12 points (30%), explained mainly by changes within sectors.
Specifically, 94% of the decline is explained by within-sector variations, while between sector
changes explain only 10%. Figure 9 presents the change in routine intensity across sector
over the period. (Table A6 in the Appendix shows the within-sector change in RTI for the 87
sectors over the period, ranked according to the size of the decline). Although most research
relates the decline in routine tasks to automation in manufacturing, the services industry
presents a more significant decrease in routine intensity in Brazil. In particular, the decline
in routine intensity in services was twice as large as that for manufacturing. Hence, we
document a significant decrease in routine occupations not related to reallocation of workers
but to within sector changes, and that go beyond manufacturing.
26
Figure 9: Change in Sectors’ Routine Intensity
The y-axis is the mean change in RTI for each sector from 2006 to 2018.
Our next step lies in using these differences across sectors. First, we divide our sample
into workers initially employed in manufacturing and non-manufacturing and re-estimate
Equation 2. In addition, we look at within-industry changes in the RTI index and split our
sample into individuals initially employed in sectors above the median or below or equal the
median. Table 6 presents the results of both exercises, suggesting that workers previously
employed in manufacturing face a larger decline on wages and more extended periods of
unemployment. Furthermore, when focusing on the decline in RTI across industries, the
impact is more considerable for sectors with a more significant decline in the demand for
routine tasks. Therefore, demand seems to be playing a sizable role in explaining differences
across occupational groups in Brazil.
27
Table 6: Effect of displacement on wages by sector group
Relative wages Relative employment
Mean effect Standard Errors Mean effect Standard Errors Observations
Sector
Manufacturing -0.0204*** (0.00361) -0.00954*** (0.00176) 864.288
Non-Manufacturing -0.0140*** (0.00215) -0.00609*** (0.00105) 1.500.156
Sector
Below or equal the median -0.0127*** (0.00359) -0.00260 (0.00167) 677.844
Above the median -0.0196*** (0.00211) -0.00995*** (0.00106) 1.686.600
The table shows the baseline estimates of averages of the triple interactions between time-to-event dum-
mies interacted with a displacement indicator and a routine intensity measure from a regression including
individual, region, sector, time-to-event dummies, time-to-event dummies interacted with the routine in-
tensity measure, and year fixed effects. The dependent variables are relative wages and employment.
Relative wages is measured dividing workers monthly average wage by average wage in year t−2. Em-
ployment is a dummy equal to one is the worker has any positive labor earnings in a given year. Year t−2
is the base year. Columns (1) and (2) split the 87 sectors between non-manufacturing and manufacturing.
Columns (3) and (4) split the sample into those with a decline in the RTI index below the median and
above the median. The median is equal to -.055. Standard errors clustered at the individual level are
reported in parenthesis. ***, ** and * respectively indicate 0.01, 0.05 and 0.1 levels of significance.
5.3 Job switchers
Within a more competitive labor market, with few opportunities, workers in routine occu-
pations could be more inclined (or forced) to change fields in search for better, high-paying
jobs. In this section, we estimate whether workers in routine-intensive occupations are more
likely to move to different professions following a layoff and the impacts of these transitions.
In doing so, we create a dummy equal to 1 if individuals switch occupations (2-digits) and
zero otherwise. In our sample, switching to a different occupation is observed for 49.743
workers (about 20% of individuals) (see Table A4 for a descriptive analysis of switchers and
non-switchers). Columns (1)-(2) in Table 7 show baseline estimates that include a set of
Mincerian workers’ characteristics and year, sector, and region effects. The columns differ
in methodology, with column (1) using a ordinary least squares model (OLS), while column
(2) employs a probit model. In addition, columns (3)-(4) show similar regressions using
a different definition of job switchers, now including only individuals that change broader
occupations (1-digit).
28
Table 7: Routine-intensity and the probability of switching occupations
Dependent variable: dummy indicator of switching occupations
(1) (2) (3) (4)
OLS Probit OLS Probit
RTI 0.0135∗∗∗ 0.0684∗∗∗ 0.0116∗∗∗ 0.0626∗∗∗
(0.000976) (0.00412) (0.000938) (0.00426)
Treated 0.0872∗∗∗ 0.342∗∗∗ 0.0723∗∗∗ 0.309∗∗∗
(0.00149) (0.00607) (0.00142) (0.00626)
Treated ×RTI 0.0158∗∗∗ 0.0249∗∗∗ 0.0130∗∗∗ 0.0216∗∗∗
(0.00136) (0.00522) (0.00130) (0.00537)
Female -0.0200∗∗∗ -0.0767∗∗∗ -0.0231∗∗∗ -0.0948∗∗∗
(0.00199) (0.00738) (0.00190) (0.00764)
Worker’s age 0.000166 -0.0315∗∗∗ -0.000406 -0.0352∗∗∗
(0.00114) (0.00414) (0.00108) (0.00421)
Squared age -0.0000685∗∗∗ 0.000150∗∗∗ -0.0000534∗∗∗ 0.000208∗∗∗
(0.0000152) (0.0000563) (0.0000143) (0.0000572)
Tenure -0.000337∗∗∗ -0.00152∗∗∗ -0.000287∗∗∗ -0.00140∗∗∗
(0.0000174) (0.0000799) (0.0000166) (0.0000826)
Primary school graduate 0.0169∗∗∗ 0.0192 0.0150∗∗∗ 0.0206
(0.00540) (0.0234) (0.00495) (0.0247)
Middle school graduate 0.0293∗∗∗ 0.0743∗∗∗ 0.0275∗∗∗ 0.0806∗∗∗
(0.00539) (0.0231) (0.00495) (0.0243)
High-school graduate 0.0281∗∗∗ 0.0718∗∗∗ 0.0293∗∗∗ 0.0893∗∗∗
(0.00536) (0.0229) (0.00494) (0.0242)
College degree 0.0515∗∗∗ 0.165∗∗∗ 0.0547∗∗∗ 0.195∗∗∗
(0.00611) (0.0252) (0.00569) (0.0264)
Log(firm size) -0.00372∗∗∗ -0.0208∗∗∗ -0.00261∗∗∗ -0.0173∗∗∗
(0.000670) (0.00258) (0.000638) (0.00268)
Year Yes Yes Yes Yes
Region Yes Yes Yes Yes
Sector Yes Yes Yes Yes
Observations 262716 262714 262716 262711
The table shows the baseline estimates of switching occupations and a routine intensity measure from a
regression including individual, region, sector, and year. In columns (1) and (2), the dependent variable
is a dummy variable equals to one if individuals switch occupations (2-digits) and zero otherwise. In
columns (3) and (4) we use a broader definition of occupation and define workers’ occupations at 1-digit
level. Therefore, the dependent variable is equal to 1 if workers transition across occupations at 1-digit
and zero otherwise. Robust standard errors are reported in parenthesis. ***, ** and * respectively
indicate 0.01, 0.05 and 0.1 levels of significance.
The main finding is that the routine-intensity index correlates significantly with the
probability of moving to different occupations, even if not in the treatment group. The
29
second line in Table 7 shows that treated workers (those part of a mass layoff) are also more
likely to switch occupations, consistent with other findings in the literature (Nedelkoska et al.,
2015). We also find that treated individuals previously employed in routine occupations are
more likely to transition to different professions. In addition, we observe that male and more
educated workers are more likely to switch jobs, whereas long-tenured individuals are less
likely to follow this path. Overall, the results suggest that the falling demand for routine
tasks not only affects workers’ employment outcomes but also increases the likelihood to
change professions.
We further test whether displacement affects job switchers differently compared to non-
switchers. In particular, we separate our sample between job switchers and those workers
that remained in the same occupation (2-digits) and re-estimate Equation 2. Column (1) in
Table 8 presents the results for the group of workers that switch occupations and column (2)
for the workers that have remained in the same occupational group. Interestingly, workers
initially in routine-intensive occupations and moving to different occupations are significantly
less affected than workers who do not switch occupations. This aligns nicely with economic
intuition. Falling demand for routine tasks requires the reallocation of workers from declining
occupations to other more promising – workers who comply with such inter-occupational
selection dynamics should be rewarded compared to workers who “stubbornly” linger in their
original declining occupation. Similarly, workers switching to other occupations presumably
select from amid a broader opportunity set than workers searching only within their current
occupation. Hence switching workers would be associated with a better-matching labor
market opportunity if their search space is wider.
30
Table 8: Effect of displacement on wages for switchers and non-switchers
Dependent variable: relative wages
(1) (2)
Switchers Non-switchers
Mean effect -0.0698∗∗∗ -0.0865∗∗∗
(0.00534) (0.00208)
RTI -0.00769 -0.0242∗∗∗
(0.00512) (0.00195)
Individual Yes Yes
Year Yes Yes
Region Yes Yes
Observations 446886 1917558
R-squared 0.374 0.374
The table shows averages of the estimates over the 5 years from the shock (from tto t+ 4) of the
triple interactions between time-to-event dummies interacted with a displacement indicator and a routine
intensity measure from a regression including individual, region, sector, time-to-event dummies, time-
to-event dummies interacted with the routine intensity measure, and year fixed effects. The dependent
variable is relative wages. Relative wages is measured dividing workers monthly average wage by average
wage in year t−2. Year t−2 is the base year. Job switchers are defined as workers that change occupations
between the year before the shock and the first year of re-employment. Occupations are defined at the
2-digit level. Column (1) restrict the sample to switchers, while column (2) focus on non-switchers.
Standard errors clustered at the individual level are reported in parenthesis. ***, ** and * respectively
indicate 0.01, 0.05 and 0.1 levels of significance.
5.4 Firms fixed effects
A third and related mechanism to explain worse employment outcomes for routine occupa-
tions is that workers in high routine intensive occupations have to move to low-paying and
worse companies to find a job in the same field. For instance, less productive firms tend to
have lower adoption of more sophisticated technologies, thus continuing to demand labor to
perform routine tasks and being the main option for dismissed high-routine workers.
To test this hypothesis, we estimate whether workers in routine-intensive occupations
move to low-paying companies more frequently. First, we estimate firms’ paying heterogene-
ity and decompose earnings using an AKM decomposition. Specifically, to calculate firms’
fixed effects, we regress the log of monthly wages on a set of individual, firm, and year fixed
effects. To ease the computational burden, we estimate this specification for two different
periods - 2007 to 2012 and 2013 to 2018, and limit the sample to the largest connected set
31
within each of these samples.6
With firms fixed effects in hand, we estimate Equation 1 using the fixed effects as the
outcome variable and compare with the impacts on workers’ logarithm of monthly wages.
Figure 10 shows the results of this exercise, indicating that the loss of employer-specific wage
premium responds to about 13% of the adverse effect on wages. Our results are closer in
magnitude to those in Lachowska et al. (2020), who finds that employer-specific premiums
explain 17% of wage losses in the state of Washington, but significantly smaller than those
observed in Germany (Fackler et al.,2021). The small effect on firm wage premium losses is
likely related to a weakening pass-through from firm characteristics to wages in Brazil. For
instance, Alvarez et al. (2018) shows the decline in firm productivity pay premium explained
about 40% of the decrease in earnings inequality in Brazil between 1996 and 2012. As a
result, workers are increasingly more likely to move to firms with equal paying premiums.
Figure 10: Job displacement and the loss of employer-specific wage premium
The figure shows the estimates of time-to-event dummies interacted with a displacement indicator from
a regression including individual, region, sector, time-to-event dummies, and year fixed effects. The
dependent variables are logarithm of monthly wages and firms’ fixed effects. Firms’ fixed effects are
identified using a AKM model. Year t−2 is the base year. Low-routine are workers in the first quartile
of the routine-intensity index, while high-routine indicates workers in the fourth quartile. Vertical bars
show estimated 95% confidence interval based on standard errors clustered at individual level.
Following this analysis, we estimate Equation 2 to test whether the RTI is associated with
6We assume that establishments’ wage premium is set at the firm level. Therefore, we estimate estab-
lishments fixed effects at the company level.
32
movements to low-paying firms. On the one hand, Figure 11 confirms that workers previously
employed in routine-intensive occupations face a more significant decline in wages, even when
excluding those workers that are not employed (using the logarithm of wages exclude those
workers with missing information on wages). On the other hand, Figure 11 show that the
routine-intensity index is not statistically associated with a decline in firm’s fixed effects,
thus suggesting that workers previously employed in routine-intensive occupations were not
more likely to transition to low-paying firms.
Figure 11: Effect of displacement on employment by occupational group
The figure shows the estimates of the triple interactions between time-to-event dummies interacted with
a displacement indicator and a routine intensity measure from a regression including individual, region,
sector, time-to-event dummies, time-to-event dummies interacted with the routine intensity measure,
and year fixed effects. The dependent variables are logarithm of monthly wages and firms’ fixed effects.
Firms’ fixed effects are identified using a AKM model. Year t−2 is therefore the base year. Vertical
bars show estimated 95% confidence intervals based on standard errors clustered at individual level.
6 Conclusion
Job polarization and the decrease in the demand for routine occupations have been widely
documented in the US and European countries. The main culprit of these changes affecting
workers unequally is technological changes. Much less is known about the importance of this
phenomenon in developing countries where the rate of diffusion and adoption of advanced
technologies is much slower. This paper offers a detailed analysis of the employment dynam-
33
ics associated with routine workers in Brazil, using mass layoffs as a natural experiment to
identify employment effects.
Job displacement has a significant impact on workers’ careers. Wages are significantly
depressed in the short-run and only recover partially in the medium run. Displaced workers
are also more likely to face extended periods of unemployment. The results show a large and
statistically significant wage loss associated with job displacement. Workers displaced see
wage declines of up to 5% even five years after the displacement event. In addition, consistent
with most findings in the literature, we find that female, less-educated, long-tenured, and
older workers are more significantly affected by displacement.
But while all workers experience significant declines in wages and employment opportu-
nities following a mass layoff event, those in routine intensive occupations fare much worse.
While we cannot measure technological progress directly, the results show that job dis-
placement’s adverse outcomes are worse in sectors where the demand for routine jobs has
decreased over time. Moreover, not only do workers in routine occupations and sectors with
larger demand decline experience more wage losses, but they are also more likely to have to
switch occupations. However, we do not find evidence of a necessary move towards “worse”
firms.
Some policy prescriptions may tentatively be offered. Workers in routine-intensive oc-
cupations appear to be especially vulnerable after a mass layoff, presumably related to the
difficulty of finding a new job requiring similar skills. Specifically, we find that the effect is
only significant for less-skilled individuals. Public policies need to seek to train and qualify
displaced workers, assisting the development of new skills to reduce the harmful impacts of
displacement.
More research is needed on the heterogeneity of the results. Especially a more nuanced
view of these groups and the tasks they perform and the differences between workers that
move to different sectors and occupations. In addition, the design of policy interventions
requires a deeper understanding of the skills, especially soft skills, that facilitate job transi-
34
tion. This is critical to understanding the role of skill mismatch and designing appropriate
policies for reinserting routine-intensive workers into the labor market. Finally, as the pace
of technological change and automation shows no sign of slowing down, policy interventions
for training and reskilling displaced workers can only be expected to grow in importance.
Finally, more granular evidence is needed linking directly events of technology upgrad-
ing with changes in the skill composition at the level of the firm over a period of time.
Rather than inferring technological trends based on the changes in occupation skills within
sectors, more granular data is needed to identify better how specific technologies affect the
employment outcomes of different types of workers.
Acknowledgments We benefited from comments from Jaime Arellano-Bover, Nanditha
Mathew, Pierre Mohnen, Gaurav Nayyar, and Tania Treibich. We also thank seminar atten-
dees at the World Bank’s FIE Knowledge Sharing Seminar, the UNU-MERIT seminar, and
the 2022 Workshop on Technology, Employment and Industrial Dynamics in Pisa. Antonio
Martins-Neto acknowledges support by the Comprehensive Innovation for Sustainable De-
velopment (CI4SD) Research Programme of the United Nations University - Maastricht Eco-
nomic and Social Research Institute on Innovation and Technology (UNU-MERIT), Maas-
tricht, the Netherlands.
35
References
Acemoglu, D. and D. Autor (2011). Skills, Tasks and Technologies: Implications for Employ-
ment and Earnings. In O. Ashenfelter and D. Card (Eds.), Handbook of Labor Economics,
Volume 4 of Handbook of Labor Economics, Chapter 12, pp. 1043–1171. Elsevier.
Aghion, P., C. Antonin, and S. Bunel (2021). The Power of Creative Destruction. Harvard
University Press.
Aghion, P. and P. Howitt (1992). A model of growth through creative destruction. Econo-
metrica 60 (2), 323–351.
Alvarez, J., F. Benguria, N. Engbom, and C. Moser (2018). Firms and the decline in earnings
inequality in brazil. American Economic Journal: Macroeconomics 10 (1), 149–89.
Arnoud, A. (2018). Automation threat and wage bargaining. Unpublished Manuscript,Yale
University.
Autor, D. H. and D. Dorn (2013). The growth of low-skill service jobs and the polarization
of the us labor market. American Economic Review 103 (5), 1553–97.
Autor, D. H., F. Levy, and R. J. Murnane (2003). The Skill Content of Recent Technological
Change: An Empirical Exploration*. The Quarterly Journal of Economics 118 (4), 1279–
1333.
Bertheau, A., E. M. Acabbi, C. Barcelo, A. Gulyas, S. Lombardi, and R. Saggio (2022).
The unequal cost of job loss across countries. Working Paper 29727, National Bureau of
Economic Research.
Bessen, J. E., M. Goos, A. Salomons, and W. Van den Berge (2019). Automatic reaction-
what happens to workers at firms that automate? Boston Univ. School of Law, Law and
Economics Research Paper .
Blien, U., W. Dauth, and D. H. Roth (2021). Occupational routine intensity and the costs
of job loss: evidence from mass layoffs. Labour Economics 68, 101953.
B´ar´any, Z. L. and C. Siegel (2018). Job polarization and structural change. American
Economic Journal: Macroeconomics 10 (1), 57–89.
Carneiro, A. and P. Portugal (2006). Earnings Losses of Displaced Workers: Evidence from
a Matched Employer-Employee Data Set. IZA Discussion Papers 2289, Institute of Labor
Economics (IZA).
Comin, D. and B. Hobijn (2010). An exploration of technology diffusion. American Economic
Review 100 (5), 2031–59.
Comin, D. and M. Mestieri (2018). If technology has arrived everywhere, why has income
diverged? American Economic Journal: Macroeconomics 10 (3), 137–78.
36
Couch, K. A. and D. W. Placzek (2010). Earnings losses of displaced workers revisited.
American Economic Review 100 (1), 572–89.
Cunningham, S. (2021). Causal Inference: The Mixtape. Yale University Press.
Das, M. and B. Hilgenstock (2018). The Exposure to Routinization: Labor Market Im-
plications for Developed and Developing Economies. IMF Working Papers 2018/135,
International Monetary Fund.
de Chaisemartin, C. and X. D’Haultfœuille (2020). Two-way fixed effects estimators with
heterogeneous treatment effects. American Economic Review 110(9), 2964–96.
de Vries, G. J., E. Gentile, S. Miroudot, and K. M. Wacker (2020). The rise of robots and
the fall of routine jobs. Labour Economics 66, 101885.
Deelen, A., M. de Graaf-Zijl, and W. van den Berge (2018). Labour market effects of job
displacement for prime-age and older workers. IZA Journal of Labor Economics 7 (1),
1–30.
Dustmann, C., J. Ludsteck, and U. Sch¨onberg (2009). Revisiting the German Wage Struc-
ture*. The Quarterly Journal of Economics 124 (2), 843–881.
Easterly, W. and R. Levine (2001). It’s not factor accumulation: Stylized facts and growth
models. The World Bank Economic Review 15 (2), 177–219.
Eliason, M. and D. Storrie (2006). Lasting or latent scars? swedish evidence on the long-term
effects of job displacement. Journal of Labor Economics 24 (4), 831–856.
Fackler, D., S. Mueller, and J. Stegmaier (2021). Explaining Wage Losses After Job Displace-
ment: Employer Size and Lost Firm Wage Premiums. Journal of the European Economic
Association 19 (5), 2695–2736.
Fallick, B., J. C. Haltiwanger, E. McEntarfer, and M. Staiger (2021). Job displacement and
job mobility: The role of joblessness. Working Paper 29187, National Bureau of Economic
Research.
Firpo, S., A. Portella, F. Riva, and G. ´
Ubida (2021). The changing nature of work and
inequality in Brazil (2003-19): A descriptive analysis. WIDER Working Paper Series
wp-2021-162, World Institute for Development Economic Research (UNU-WIDER).
Fonseca, T., F. Lima, and S. C. Pereira (2018). Job polarization, technological change and
routinization: Evidence for portugal. Labour Economics 51, 317–339.
Gasparini, L., I. Brambilla, G. Falcone, C. Lombardo, and A. C´esar (2021). Routinization
and Employment: Evidence for Latin America. CEDLAS, Working Papers 0276, CEDLAS,
Universidad Nacional de La Plata.
Goos, M., A. Manning, and A. Salomons (2009). Job polarization in europe. American
Economic Review 99 (2), 58–63.
37
Goos, M., A. Manning, and A. Salomons (2014). Explaining job polarization: Routine-biased
technological change and offshoring. American Economic Review 104 (8), 2509–26.
Goos, M., E. Rademakers, and R. R¨ottger (2021). Routine-biased technical change:
Individual-level evidence from a plant closure. Research Policy 50 (7), 104002.
Heckman, J., H. Ichimura, and P. E. Todd (1997). Matching as an econometric evaluation
estimator: Evidence from evaluating a job training programme. Review of Economic
Studies 64 (4), 605–654.
Hijzen, A., R. Upward, and P. W. Wright (2010). The income losses of displaced workers.
The Journal of Human Resources 45 (1), 243–269.
Huckfeldt, C. (2018). Understanding the Scarring Effect of Recessions. 2018 Meeting Papers
1207, Society for Economic Dynamics.
Huttunen, K., J. Moen, and K. G. Salvanes (2006). How Destructive Is Creative Destruc-
tion? The Costs of Worker Displacement. IZA Discussion Papers 2316, Institute of Labor
Economics (IZA).
Iacus, S. M., G. King, and G. Porro (2012). Causal inference without balance checking:
Coarsened exact matching. Political Analysis 20 (1), 1–24.
Ichino, A., G. Schwerdt, R. Winter-Ebmer, and J. Zweim¨uller (2017). Too old to work, too
young to retire? The Journal of the Economics of Ageing 9, 14 – 29.
Jacobson, L. S., R. J. LaLonde, and D. G. Sullivan (1993). Earnings losses of displaced
workers. The American Economic Review 83 (4), 685–709.
Kaplan, D. S., R. Robertson, and G. M. Gonz´alez (2005). What Happens to Wages after
Displacement? Econom´ıa Journal 0 (Spring 20), 197–242.
Lachowska, M., A. Mas, and S. A. Woodbury (2020). Sources of displaced workers’ long-term
earnings losses. American Economic Review 110 (10), 3231–66.
Lewandowski, P., A. Park, W. Hardy, and D. Yang (2019). Technology, Skills, and Glob-
alization: Explaining International Differences in Routine and Nonroutine Work Using
Survey Data. IBS Working Papers 04/2019, Instytut Badan Strukturalnych.
Lysho, A. F. (2020). Occupational mobility, occupational distance & unemployment insur-
ance. Available at https://arnelyshol.com/files/paper˙occupationalmobility.pdf .
Maloney, W. F. and C. Molina (2019). Is Automation Labor-Displacing in the Developing
Countries, Too? Robots, Polarization, and Jobs. World Bank.
Martins-Neto, A., N. Mathew, P. Mohnen, and T. Treibich (2021). Is There Job Polarization
in Developing Economies? A Review and Outlook. CESifo Working Paper Series 9444,
CESifo.
38
Menezes-Filho, N. (2004). The Costs of Displacement in Brazil. Econometric Society 2004
Latin American Meetings 200, Econometric Society.
Michaels, G., A. Natraj, and J. V. Reenen (2014). Has ICT Polarized Skill Demand? Ev-
idence from Eleven Countries over Twenty-Five Years. The Review of Economics and
Statistics 96 (1), 60–77.
Nedelkoska, L., F. Neffke, and S. Wiederhold (2015). Skill mismatch and the costs of job
displacement. In Annual Meeting of the American Economic Association.
Raposo, P., P. Portugal, and A. Carneiro (2019). The sources of the wage losses of displaced
workers: the role of the reallocation of workers into firms, matches, and job titles. Journal
of Human Resources.
Reijnders, L. S. and G. J. de Vries (2018). Technology, offshoring and the rise of non-routine
jobs. Journal of Development Economics 135, 412–432.
Romer, P. M. (1990). Endogenous technological change. Journal of Political Economy 98 (5),
S71–S102.
Roth, J., P. H. Sant’Anna, A. Bilinski, and J. Poe (2022). What’s trending in difference-
in-differences? a synthesis of the recent econometrics literature. arXiv preprint
arXiv:2201.01194 .
Saltiel, F. (2018). The effect of involuntary displacement on brazilian workers. Working
paper, University of Maryland, College Park.
Schumpeter, J. A. (1942). Capitalism, socialism and democracy.
Schwerdt, G., A. Ichino, O. Ruf, R. Winter-Ebmer, and J. Zweim¨uller (2010). Does the
color of the collar matter? employment and earnings after plant closure. Economics
Letters 108 (2), 137 – 140.
Spitz-Oener, A. (2006). Technical change, job tasks, and rising educational demands: Look-
ing outside the wage structure. Journal of Labor Economics 24 (2), 235–270.
Ulyssea, G. (2010). Regulation of entry, labor market institutions and the informal sector.
Journal of Development Economics 91 (1), 87–99.
39
Appendix A Additional estimates
Figure A1: Histogram routine-intensity index
40
Table A1: Effect of displacement on wages and employment
Dependent variables
(1) (2)
Relative wage Relative employment
Time-to-event (t-3) 0.00533∗∗∗ -0.0000887∗∗∗
(0.000569) (0.0000261)
Time-to-event (t-2) — —
— —
Time-to-event (t-1) -0.00527∗∗∗ 0.000123∗∗∗
(0.000776) (0.0000243)
Time-to-event (t) -0.00108 0.000111∗∗∗
(0.00197) (0.0000273)
Time-to-event (t+1) -0.198∗∗∗ -0.179∗∗∗
(0.00253) (0.00134)
Time-to-event (t+2) -0.0652∗∗∗ -0.0494∗∗∗
(0.00274) (0.00134)
Time-to-event (t+3) -0.0554∗∗∗ -0.0409∗∗∗
(0.00297) (0.00145)
Time-to-event (t+4) -0.0536∗∗∗ -0.0364∗∗∗
(0.00302) (0.00155)
Time-to-event (t+5) -0.0572∗∗∗ -0.0344∗∗∗
(0.00312) (0.00163)
Individual Yes Yes
Year Yes Yes
Region Yes Yes
Sector Yes Yes
Observations 2.436.759 2.436.759
R-squared 0.379 0.420
The table shows the baseline estimates of the estimates of time-to-event dummies interacted with a
displacement indicator from a regression including individual, region, sector, time-to-event dummies,
and year fixed effects. The dependent variables are relative wages and employment. Relative wages is
measured dividing worker’s monthly average wage by the worker’s average wage in year t−2. Employment
is a dummy equal to one is the worker has any positive labor earnings in a given year. Year t−2 is
the base year. Heteroskedasticity robust standard errors clustered at individual level are reported in
parenthesis. ***, ** and * respectively indicate 0.01, 0.05 and 0.1 levels of significance.
41
Table A2: Effect of displacement on relative wages and employment by occupational group
Low-Routine High-Routine
Relative wages -0.0531*** -0.102***
(0.00353) (0.00426)
Relative employment -0.0438*** -0.0664***
(0.00184) (0.00210)
Individual Yes Yes
Year Yes Yes
Region Yes Yes
Sector Yes Yes
Clus. individual Yes Yes
Observations 659.277 537.543
The table shows the baseline estimates of the estimates of time-to-event dummies interacted with a
displacement indicator from a regression including individual, region, sector, time-to-event dummies,
and year fixed effects. The dependent variables are relative wages and employment. Relative wages is
measured dividing workers monthly average wage by average wage in year t−2. Employment is a dummy
equal to one is the worker has any positive labor earnings in a given year. Low-routine are workers in the
first quartile of the routine-intensity index, while high-routine indicates workers in the fourth quartile.
Standard errors clustered at individual level are reported in parenthesis. ***, ** and * respectively
indicate 0.01, 0.05 and 0.1 levels of significance.
42
Table A3: Comparison of workers in establishment’s closure and mass layoff
Closed Mass layoff
Mean Standard Deviation Mean Standard Deviation Difference
Routine task index .31 1.09 .17 1.08 -0.143*
Wage 1683 1666.62 1309 1324.80 -370.565***
Wage Growth .1 0.27 .1 0.25 0.001
Worker’s age 35 6.30 35 6.38 0.180
Gender .34 0.47 .31 0.46 -0.027
Illiterate or primary school .018 0.13 .033 0.18 0.009**
Primary school graduate .14 0.34 .19 0.39 0.047***
Middle school graduate .21 0.41 .26 0.44 0.052***
High-school graduate .53 0.50 .46 0.50 -0.076***
College degree .1 0.30 .065 0.25 -0.032**
Tenure 65 43.51 61 42.91 -4.627**
Firm’s size 421 505.01 771 1364.19 340.591***
Size (30-49) .13 0.34 .11 0.32 -0.017
Size (50 - 99) .18 0.38 .16 0.37 -0.015
Size (100-499) .4 0.49 .39 0.49 -0.014
Size (500+) .29 0.45 .34 0.47 0.046
Firm’s average wage 1781 1377.09 1337 1025.35 -442.953***
Agriculture and Extractive .02 0.14 .028 0.17 0.001
Manufacturing .38 0.49 .34 0.47 -0.038
Services .6 0.49 .63 0.48 0.037
North .015 0.12 .024 0.15 0.009**
Northeast .081 0.27 .15 0.36 0.062***
Southeast .78 0.42 .64 0.48 -0.132***
South .11 0.31 .14 0.34 0.031*
Central-West .021 0.14 .05 0.22 0.030***
Observations 64.433 0.00 71.133 0.00 —
Table shows averages for baseline. The last column is the coefficient of a simple regression of treatment
status on the variable, with robust standard errors. Stars indicate whether this difference is significant. * p
<0.10, ** p <0.05, *** p <0.01.
43
Table A4: Comparison of switchers and non-switchers
Non-switcher Switcher
Wage 1666 1705.67 1681 1706.56 15.590
Wage Growth .13 0.42 .14 0.44 0.011*
Worker’s age 37 6.37 35 6.01 -1.786***
Gender .32 0.47 .31 0.46 -0.017
Illiterate or primary school .02 0.14 .016 0.12 -0.004
Primary school graduate .16 0.37 .13 0.33 -0.034**
Middle school graduate .24 0.43 .23 0.42 -0.016
High-school graduate .5 0.50 .54 0.50 0.040
College degree .08 0.27 .094 0.29 0.014
Tenure 76 44.34 70 36.66 -6.248***
Firm’s size 605 1057.99 591 1122.36 -13.920
Size (30-49) 39 5.79 39 5.76 0.088
Size (50 - 99) 72 14.45 72 14.53 0.246
Size (100-499) 252 112.91 249 111.57 -3.288
Size (500+) 1554 1483.60 1681 1672.40 126.883
Firm’s average wage 1686 1318.80 1750 1420.57 64.250
Agriculture and Extractive .012 0.11 .0091 0.10 -0.003
Manufacturing .35 0.48 .42 0.49 0.068**
Services .64 0.48 .57 0.50 -0.065**
North .02 0.14 .021 0.14 0.002
Northeast .11 0.32 .11 0.31 -0.005
Southeast .71 0.46 .71 0.45 0.005
South .12 0.33 .13 0.33 0.006
Central-West .038 0.19 .03 0.17 -0.008***
Observations 213.404 — 49.743 — —
Table shows averages for baseline. The last column is the coefficient of a simple regression of treatment
status on the variable, with robust standard errors. Stars indicate whether this difference is significant. * p
<0.10, ** p <0.05, *** p <0.01.
44
Table A5: Effect of displacement on wages and employment
Dependent variable
(1) (2)
Relative wage Relative employment
Time-to-event (t-3) 0.000714 -0.0000184
(0.000542) (0.0000222)
Time-to-event (t-2) — —
— —
Time-to-event (t-1) -0.00279∗∗∗ 0.0000283
(0.000701) (0.0000196)
Time-to-event (t) -0.00269 0.0000416∗
(0.00173) (0.0000230)
Time-to-event (t+1) -0.0202∗∗∗ -0.0103∗∗∗
(0.00241) (0.00123)
Time-to-event (t+2) -0.0185∗∗∗ -0.00845∗∗∗
(0.00276) (0.00123)
Time-to-event (t+3) -0.0182∗∗∗ -0.00725∗∗∗
(0.00289) (0.00133)
Time-to-event (t+4) -0.0245∗∗∗ -0.0103∗∗∗
(0.00292) (0.00143)
Time-to-event (t+5) -0.0226∗∗∗ -0.00933∗∗∗
(0.00293) (0.00151)
Individual Yes Yes
Year Yes Yes
Region Yes Yes
Sector Yes Yes
Observations 2.364.444 2.364.444
The table shows the baseline estimates of averages of the triple interactions between time-to-event dum-
mies interacted with a displacement indicator and a routine intensity measure from a regression including
individual, region, sector, time-to-event dummies, time-to-event dummies interacted with the routine in-
tensity measure, and year fixed effects. The dependent variables are relative wages and employment.
Relative wages is measured dividing worker’s monthly average wage by the worker’s average wage in year
t−2. Employment is a dummy equal to one is the worker has any positive labor earnings in a given
year. Year t−2 is the base year. Heteroskedasticity robust standard errors clustered at the individual
are reported in parenthesis. ***, ** and * respectively indicate 0.01, 0.05 and 0.1 levels of significance.
45
Table A6: Change in RTI index by sector, 2006-2018
Sector RTI change Sector RTI change
Libraries, archives, museums and other cul-
tural activities
-0,735 Manufacture of wood and of products of wood and cork,
except furniture; manufacture of articles of straw and
plaiting materials
-0,054
Information service activities -0,502 Manufacture of basic metals -0,053
Scientific research and development -0,407 Other manufacturing -0,050
Extraction of crude petroleum and natural gas -0,370 Remediation activities and other waste management
services
-0,049
Travel agency, tour operator, reservation ser-
vice and related activities
-0,358 Food and beverage service activities -0,045
Computer programming, consultancy and re-
lated activities
-0,352 Rental and leasing activities -0,041
Manufacture of coke and refined petroleum
products
-0,308 Repair and installation of machinery and equipment -0,036
Public administration and defence; compul-
sory social security
-0,305 Manufacture of other transport equipment -0,035
Activities auxiliary to financial service and in-
surance activities
-0,289 Activities of membership organizations -0,034
Creative, arts and entertainment activities -0,283 Construction of buildings -0,034
Residential care activities -0,268 Manufacture of leather and related products -0,033
Insurance, reinsurance and pension funding,
except compulsory social security
-0,245 Manufacture of beverages -0,031
Activities of head offices; management consul-
tancy activities
-0,235 Manufacture of chemicals and chemical products -0,029
Veterinary activities -0,217 Manufacture of machinery and equipment n.e.c. -0,028
Forestry and logging -0,213 Mining of metal ores -0,027
Undifferentiated goods- and services-
producing activities of private households
for own use
-0,213 Manufacture of computer, electronic and optical prod-
ucts
-0,023
Financial service activities, except insurance
and pension funding
-0,176 Manufacture of rubber and plastics products -0,021
Education -0,167 Services to buildings and landscape activities -0,021
Air transport -0,152 Manufacture of fabricated metal products, except ma-
chinery and equipment
-0,014
Programming and broadcasting activities -0,150 Telecommunications -0,014
Legal and accounting activities -0,146 Retail trade, except of motor vehicles and motorcycles -0,012
Human health activities -0,141 Accommodation -0,011
Water collection, treatment and supply -0,129 Mining support service activities -0,009
Crop and animal production, hunting and re-
lated service activities
-0,124 Mining of coal and lignite -0,007
Electricity, gas, steam and air conditioning
supply
-0,119 Fishing and aquaculture -0,006
Publishing activities -0,115 Specialized construction activities -0,006
Manufacture of other non-metallic mineral
products
-0,113 Warehousing and support activities for transportation -0,004
Water transport -0,112 Manufacture of motor vehicles, trailers and semi-
trailers
-0,003
46
Sector RIT change Sector RIT change
Other professional, scientific and technical ac-
tivities
-0,106 Manufacture of furniture -0,001
Waste collection, treatment and disposal ac-
tivities; materials recovery
-0,106 Manufacture of electrical equipment -0,001
Manufacture of food products -0,100 Security and investigation activities 0,001
Real estate activities -0,098 Wholesale trade, except of motor vehicles and motor-
cycles
0,005
Architectural and engineering activities; tech-
nical testing and analysis
-0,096 Manufacture of textiles 0,007
Sports activities and amusement and recre-
ation activities
-0,088 Other personal service activities 0,011
Manufacture of paper and paper products -0,079 Land transport and transport via pipelines 0,023
Postal and courier activities -0,079 Manufacture of wearing apparel 0,026
Civil engineering -0,075 Wholesale and retail trade and repair of motor vehicles
and motorcycles
0,045
Employment activities -0,073 Motion picture, video and television programme pro-
duction, sound recording and music publishing activi-
ties
0,054
Office administrative, office support and other
business support activities
-0,072 Activities of households as employers of domestic per-
sonnel
0,067
Printing and reproduction of recorded media -0,071 Sewerage 0,077
Advertising and market research -0,068 Manufacture of tobacco products 0,135
Repair of computers and personal and house-
hold goods
-0,065 Social work activities without accommodation 0,177
Other mining and quarrying -0,060 Gambling and betting activities 0,221
Manufacture of pharmaceuticals, medicinal
chemical and botanical products
-0,055
47
TheUNU‐MERITWORKINGPaperSeries
2022-01 Structuraltransformationsandcumulativecausationtowardsanevolutionary
micro‐foundationoftheKaldoriangrowthmodelbyAndréLorentz,TommasoCiarli,
MariaSavonaandMarcoValente
2022-02 Estimationofaproductionfunctionwithdomesticandforeigncapitalstockby
ThomasZiesemer
2022-03 Automationandrelatedtechnologies:Amappingofthenewknowledgebaseby
EnricoSantarelli,JacopoStaccioliandMarcoVivarelli
2022-04 Theold‐agepensionhouseholdreplacementrateinBelgiumbyAlessioJ.G.Brown
andAnne‐LoreFraikin
2022-05 GlobalisationincreasedtrustinnorthernandwesternEuropebetween2002and
2018byLoesjeVerhoevenandJoRitzen
2022-06 GlobalisationandfinancialisationintheNetherlands,1995–2020byJoanMuysken
andHuubMeijers
2022-07 Importpenetrationandmanufacturingemployment:EvidencefromAfricaby
SolomonOwusu,GideonNdubuisiandEmmanuelB.Mensah
2022-08 AdvanceddigitaltechnologiesandindustrialresilienceduringtheCOVID‐19
pandemic:Afirm‐levelperspectivebyElisaCalzaAlejandroLavopaandLigiaZagato
2022-09 Thereckoningofsexualviolenceandcorruption:Agenderedstudyofsextortionin
migrationtoSouthAfricabyAshleighBickerCaarten,LoesvanHeugtenandOrtrun
Merkle
2022-10 TheproductiveroleofsocialpolicybyOmarRodríguezTorres
2022-11 SomenewviewsonproductspaceandrelateddiversificationbyÖnderNomaler
andBartVerspagen
2022-12 ThemultidimensionalimpactsoftheConditionalCashTransferprogramJuntosin
PerubyRicardoMorelandLizGirón
2022-13 Semi‐endogenousgrowthinanon‐WalrasianDSEMforBrazil:Estimationand
simulationofchangesinforeignincome,humancapital,R&D,andtermsoftrade
byThomasH.W.Ziesemer
2022-14 Routine‐biasedtechnologicalchangeandemployeeoutcomesaftermasslayoffs:
EvidencefromBrazilbyAntonioMartins‐Neto,XavierCireraandAlexCoad
