unemployment duration, technology and skills€¦ · unemployment duration, technology and skills...
TRANSCRIPT
Unemployment duration, technology and skills
Ana Bárbara Leal Rendeiro da Piedade
Thesis to obtain the Master of Science Degree in
Industrial Engineering and Management
Supervisor: Prof. Hugo Miguel Fragoso De Castro Silva
Examination Committee
Chairperson: Prof. Rui Miguel Loureiro Nobre BaptistaSupervisor: Prof. Hugo Miguel Fragoso De Castro Silva
Member of the Committee: Prof. António Sérgio Constantino Folgado Ribeiro
November 2018
ii
Dedicated to Eduardo
iii
iv
Acknowledgments
I would first like to thank my thesis adviser Professor Hugo Castro Silva for all the support received
throughout this project, his help and guidance was crucial to the making of this dissertation. The door
to Prof. Hugo was always open whenever I ran into a trouble spot or had a question about my research,
and he consistently allowed this paper to be my own work. Above all, I am grateful for his time and
for always steering me in the right direction. I would also like the thank Professor Francisco Lima, who
helped me in the project stage of this dissertation.
Finally, I must express my very profound gratitude to my parents and to my dear friends for providing
me with unfailing support and continuous encouragement throughout my years of study and through
the process of researching and writing this thesis. This accomplishment would not have been possible
without them. Thank you.
v
vi
Resumo
A crescente mudanca tecnologica tem efeitos potencialmente negativos no mercado de trabalho. Com
a tecnologia cada vez mais presente nas nossas vidas, assim como no local de trabalho, a procura
relativa de qualificacoes esta a alterar-se. Existe uma maior procura de trabalhadores com capacidade
de aprendizagem rapida e altamente qualificados em tempos de grande mudanca tecnologica. Deste
modo, trabalhadores pouco qualificados nao tem acesso as melhores oportunidades de emprego e,
consequentemente, a taxa e a duracao do desemprego para este grupo de trabalhadores aumenta.
Existe tambem forte evidencia de que o risco de separacao do trabalho e maior para pessoas nao qual-
ificadas em tempos de mudanca tecnologica. A aquisicao de capital humano, especialmente capital
especıfico as tecnologias modernas, torna-se indispensavel para operar no novo mercado de trabalho.
O objectivo desta tese e estudar a duracao do desemprego (em particular do desemprego tecnologico),
e analisar como a duracao e afectada pela complementaridade entre a tecnologia e capital humano.
Isso e conseguido atraves da modelacao do modelo Cox proportional hazards, usando o Inquerito ao
Emprego relativo a populacao Portuguesa, realizado pelo Instituto Nacional de Estatıstica. Os resulta-
dos indicam que trabalhadores mais velhos tem mais dificuldade em serem re-empregardos e portanto,
sofrem perıodos mais longos de desemprego. A probabilidade desses trabalhadores saırem do de-
semprego decresce com o aumento a idade. Os resultados indicam tambem que os homens tem uma
ligeira vantagem em relacao as mulheres, assim como indıviduos casados tambem apresentam um
maior risco de saıda do desemprego. O nıvel de capital humano e escolaridade esta muito relacionado
com a duracao do desemprego. Indivıduos com nıveis de escolaridade mais elevados apresentam uma
maior probabilidade de serem re-empregados. Estes indivıduos trabalham maioritariamente em empre-
gos mais tecnologicos e que requerem um maior nıvel de conhecimento. Sendo assim indıviduos que
trabalharam anteriormente em industrias mais intensas em tecnologia ou conhecimento apresentam
maior risco de saıda. Estes indivıduos passam em media mais tempo empregados e quando desem-
pregados, experienciam perıodos de desemprego de curta duracao.
Palavras-chave: Duracao do Desemprego; Tecnologia; Qualificacoes; Capital Humano;
Modelos de Duracao.
vii
viii
Abstract
The growing emergence of technological change in the labor market has potentially negative effects. As
technology becomes more present in our daily lives, as well as in our work place, the relative demand
for skills is shifting. The demand for fast-learning and skilled workers increase in times characterized by
technological change. Thus, unskilled people are left with few job opportunities, and might experience
higher unemployment rates, as well as longer spells of unemployment. There is also strong evidence
that unskilled individuals experience higher hazards of job separation during periods of intense techno-
logical change. Therefore, acquiring human capital, and especially technology-specific human capital, is
paramount in the new labor market. This dissertation aims to study the duration of unemployment (and,
in particular, of technological unemployment), and how is it affected by the complementarities between
technology and human capital. This is achieved by modeling unemployment duration using the Cox
proportional hazards model. We use a Portuguese data-set surveyed every quarter called Labor Force
Survey (Inquerito ao Emprego). The survey is conducted by the Instituto Nacional de Estatıstica. The
results indicate that older individuals have lower hazards of re-employment, i.e. the hazard decreases
with age, prolonging the duration of the unemployment spell. Men have slightly more chances of getting
out of unemployment. Married individuals also show an advantage. Education represents an important
factor in the chances of re-employment, as people with higher levels of education spend longer periods
of time employed, and have shorter unemployment spells. Highly educated workers will mainly perform
more knowledgeable tasks and work in high-tech manufacturing and knowledge intensive services, thus
obtaining the same results — lower unemployment durations.
Keywords: Unemployment Duration; Technology; Skills; Human Capital; Duration Models.
ix
x
Contents
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Resumo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvii
1 Introduction 1
2 Background 5
2.1 Unemployment Duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.1 Technology and Skills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 Personal Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.3 Unemployment Insurance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.4 Duration Dependence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 Human Capital . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Technological Unemployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.1 Early Stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3.2 Technological Anxiety . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Evolution of Unemployment in Portugal . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Data for the Study of Unemployment Duration 21
3.1 The Labor Force Survey (LFS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.1 Survey Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1.2 Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.3 Data Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.4 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Main Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Sample Construction and Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3.2 Characterization of the Portuguese Population . . . . . . . . . . . . . . . . . . . . 29
xi
4 Methodology 35
4.1 Survival Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.1.1 The Survival and Hazard Functions . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.1.2 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5 Results 45
5.1 Baseline Hazard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.2 Explanatory Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6 Conclusions 53
Bibliography 55
Appendix A 61
xii
List of Tables
2.1 Employment and unemployment by gender . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Unemployment duration by gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3 Employment and unemployment by age group . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Employment and unemployment by education level . . . . . . . . . . . . . . . . . . . . . . 19
2.5 Unemployed people by previous sector activity . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Employment and unemployment by region . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1 Examples of rectifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 Occupational status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Education of the unemployed population by age . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5 Technology/knowledge intensity of previous firm of the unemployed population by age . . 31
3.6 Education of the unemployed population by gender . . . . . . . . . . . . . . . . . . . . . . 31
3.7 Technology/knowledge intensity of previous employing firm of the unemployed population
by gender . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.8 Technology/knowledge intensity of previous firm of the unemployed population by education 32
3.9 Education by unemployment duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.10 Technology/knowledge intensity of previous employing firm of the unemployed population
by unemployment duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.1 Estimation results for the Cox model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
A.1 Manufacturing categories by technology intensity. . . . . . . . . . . . . . . . . . . . . . . . 61
xiii
xiv
List of Figures
1.1 Supply and demand of low-/high-skilled labor, and shifts in their wage in a time of techno-
logical change. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1 Unemployed population regarding duration of job-search . . . . . . . . . . . . . . . . . . . 6
2.2 Human capital concepts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Population of Portugal with higher education. . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Unemployment rates by education level, from 1998 to 2016. . . . . . . . . . . . . . . . . . 14
2.5 Unemployment rate in Portugal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.1 Distribution of the population. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Rotation of the LFS samples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.1 Kaplan-Meier survival estimate for age in groups . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Kaplan-Meier survival estimate for males and females . . . . . . . . . . . . . . . . . . . . 39
4.3 Kaplan-Meier survival estimate for recipients and non-recipients of unemployment insurance 40
4.4 Kaplan-Meier survival estimate for the different levels of education . . . . . . . . . . . . . 41
4.5 Kaplan-Meier survival estimate for the different levels of technology/knowledge intensity
in firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.6 Kaplan-Meier survival estimate for Non-knowledge-based job and Knowledge based job . 43
5.1 Cox proportional hazard regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
xv
xvi
Glossary
INE Instituto Nacional de Estatıstica
KIS Knowledge-Intensive Services
LFS Labor Force Survey
LKIS Less Knowledge-Intensive Services
NUTS Nomenclature of Territorial Units for Statistics
OECD Organization for Economic Co-operation and
Development
OTJ On-The-Job
RBTC Routine-Biased Technological Change
SBTC Skill-Biased Technological Change
xvii
xviii
Chapter 1
Introduction
This chapter provides information to contextualize the problem to be described, as well as a brief state-
ment of the research question. It also includes the research methodology we adopt and the main goals
and objectives.
We are living in an era highly focused on technology leading to an increasing technological change.
The relationship between technology and employment has long been debated. Though, the general
consensus in the literature is that innovation has a positive relationship with employment, the discussion
is far from over (Vivarelli, 2014).
Technological change can be brought by two alternatives: process innovation and product innovation.
Process innovation refers to the implementation of new and improved production or delivery methods.
Introducing better and cheaper methods, as in manufacturing industries. Repetitive jobs can easily be
automated, which might lead to the displacement of workers. The direct output for process innovation is
job destruction where workers are replaced by automation. The impact of such innovation can, however,
be counterbalanced by various market compensation mechanisms such as new machinery, lower prices,
new investments, and lower wages (Vivarelli, 2015). The introduction of new products through product
innovation promotes the creation of entirely new fields of labor (Marx, 1867). Therefore, the approach of
technological innovation can have a job-creating effect (Vivarelli, 2014).
Technological change favors the employment and wages of skilled workers, replacing the tasks per-
formed by the unskilled, aggravating inequalities (Acemoglu, 2002). Figure 1.1 demonstrates how the
supply and demand changes in an environment of technological change regarding the wages and quan-
tity of low-skilled and high-skilled labor.1
Where S0 is the original supply curve for labor and D0 is the original demand curve for labor. The
original point of equilibrium occurs at E0 at the price W0 and quantity Q0 . As the substitute for low-skill
labor becomes available, demand for low-skill labor will contract. As the technology complement for
1This picture and discussion is adapted from https://opentextbc.ca/principlesofeconomics/chapter/
4-1-demand-and-supply-at-work-in-labor-markets.
1
Wag
es
Wag
es
Quantity of Low-skill Labor Quantity of High-skill Labor
S0 S0
D0
D1 D0
D1
E1
E0
E0
E1
W0
W0W1
W1
Q1 Q0 Q0 Q1
(a) Technological change and low-skill labor (b) Technological change and high-skill labor
Figure 1.1: Technology and wages: supply and demand curves (a) demand for low-skill decreases whentechnology can do the same job (b) new technologies increase the demand for high-skill jobs.
high-skill labor becomes cheaper, demand for high-skill labor will expand. Therefore, low-skill workers
will have lower wages and the quantity hired will also be lower, whereas high-skill workers will have an
increase in the wage, and the quantity hired will be higher.
As seen, technological change can affect the general demand for skills. Therefore, workers displaced
by such circumstances can stumble upon difficulties in making use of their skills. This phenomenon
might translate into long-term unemployment, where such individuals not only lack the necessary skills
in a more skill-demanding labor market, but also cannot accumulate such skills while unemployed. This
leads to increased labor market segmentation, where the best job positions are only accessible to high-
skilled workers, whereas the low-skilled are excluded from high-tech jobs and experience intermittent
short-term low-pay periods of employment with longer periods of unemployment. One way to avoid
the negative consequences of the recent changes in skill requirements is acquiring skills and knowl-
edge through human capital investments (Castro Silva and Lima, 2017). At the same time, technology-
intensive firms invest in firm-specific capital in the form of on-the-job training. However, firms tend to
invest more in young and fast learning workers (Caselli, 1999).
Griliches (1969) pointed out that the complementarity between capital and skilled workers was
stronger than with unskilled people, hence, raising the productivity of skilled people relatively more,
especially in technology-intensive environments. The complementarity theory backs up the skill-biased
technological change hypothesis, where technological change favors high-skilled workers (Autor et al.,
1998). The number of middle-skilled workers started to decrease, increasing the gap between skilled
and unskilled workers (Goos and Manning, 2007), leading to a more polarized labor market. Particular
industries suffered a tremendous shift from unskilled to skilled labor. Many technology-intensive firms
2
laid-off the older workers, keeping the youngest. Younger people have more active time to recoup the
investments made by the firm, since on-the-job training takes great resources from firms.
In this thesis we study the link between technological change, human capital and the flows between
unemployment and employment from an unemployment duration perspective. We also assess the im-
pact of other variables such as gender, marital status and unemployment insurance. Furthermore, we
also contribute with an analysis of the possible ways out of unemployment, by looking at the technology
intensity of the new job, among those who managed to be re-employed. Thus, we attempt to provide
information so people can optimize their possibilities of staying employed, allowing also to lower the
unemployment rate. This study is also not only relevant not only in the context of both policy making
and labor market segmentation, but it can also contributes to educate and provide decision makers with
guidance for programs that incentivize skill acquisition, innovation, and technological employment cre-
ation.
Using data from the Portuguese quarterly employment survey Inquerito ao Emprego or Labor Force
Survey, conducted by the Instituto Nacional de Estatıstica (INE), we estimate unemployment duration
models: the probability at which an individual is re-employed, and how it is that affected by variables
such as time in unemployment, human capital and technological intensity.
The remainder of this document is structured as follows. Chapter 2 presents the literature review
regarding unemployment duration containing a discussion of the main concepts, as well as an explana-
tion of the terms human capital and technological change, as well as the history and different positions
concerning technological change. Chapter 3 presents the Labor Force Survey and characterizes the
data for the study of unemployment duration. Chapter 4 presents the methodology for fitting estimation
models, as well as the hypotheses tested. In Chapter 5 we present and discuss our econometric results,
backed up by the literature review. Finally, Chapter 6 summarizes the document, presenting the main
conclusions.
3
4
Chapter 2
Background
In this chapter we provide a discussion of extant literature on unemployment duration. We begin by intro-
ducing the main theory behind the unemployment duration subject, considering the effects of technology
and skills. We then present a more detailed concept of human capital. The last section comprises the
main concepts related with technological unemployment, as well as an historical review.
2.1 Unemployment Duration
The unemployment duration study can provide answers to some vital questions about unemployment:
1) ”How long do workers stay unemployed?”, 2) ”How does the duration of unemployment vary?” or
3) ”What are the destination states of the unemployed?”. The answers to these questions are vital to
analyze the behavior and performance of labor markets. It is not enough to study the labor market only
by analyzing static variables, as the employment and unemployment rates. To really understand labor
markets dynamics one must analyze the flows in and out of unemployment. Unemployment duration
refers to the amount of time that an individual stays unemployed — the period of time which occurs
when an individual is looking for the first job or is between jobs. Unemployment duration is one of the
factors for high unemployment rates. If unemployed people could find and accept new jobs quicker, the
unemployment rate would be lower. The length of joblessness is affected by many variables. Variables
which can be related with the labor market itself or with personal characteristics of the candidate. In a
lesser extent can be related to the availability of job positions. Factors like age (Haile, 2004; Wolff, 2005),
gender (Hernæs and Strøm, 1996), work experience and labor market history (Haile, 2004) and level of
college education (Nickell, 1979; Ashenfelter and Ham, 1979; Lancaster and Nickell, 1980; Kiefer, 1985)
have great impact in the employability of an individual. This factor can distinguish people experiencing
long periods of unemployment from those who experience short-term unemployment.
Figure 2.1 shows the Portuguese population searching for a job for less than a year (short-term
unemployment) and for a year or more (long-term unemployment). It is noticeable that until 2008 the
average population searching for a job remained the same, after which a new peak was reached by in
5
2011. This can be related with the financial crisis in Portugal, where many firms were forced to make
cuts in resources, namely in labor.
0
100
200
300
400
500
6001970
1975
1980
1985
1990
1995
2000
2005
2010
2015
2020
Less than 1 year 1 year or more
Figure 2.1: Number of unemployed residents in Portugal searching for a job.Note: Data from 1974 to 2016 presented in thousands. Dashed lines correspond to breaks in the series. (Source:
INE, PORDATA)
Skills and the level of human capital can influence the unemployment duration. In technology-
intensive industries, educated displaced workers tend to have higher post-displacement employment
rates and a better chance at being re-employed at a full-time job (Farber, 2003). Low-skilled people,
on the other hand, have a lower complementarity with capital and are less productive when working in
advanced technological environments.
2.1.1 Technology and Skills
Authors such as Nickell (1979), Ashenfelter and Ham (1979), Lancaster and Nickell (1980), and Kiefer
(1985) studied the implication of human capital investments and education applied to unemployment
duration, by using the years of schooling as a descriptive variable. Typically, highly educated individuals
have higher wages and spend more hours at work during their lives (Ashenfelter and Ham, 1979). A
person with more human capital has a smaller probability of being redundant in the workplace and to be
displaced (Nickell, 1979). Well-educated individuals have higher levels of firm-specific human capital,
since education and firm-specific training are complements (Kiefer, 1985). They are also more likely to
receive on-the-job training and to stay in the firm for a longer period of time (Mincer, 1991). Therefore,
there is a relationship between the probability of an individual entering unemployment and the level of
schooling (Azariadis, 1976). Nickell (1979) and Kiefer (1985) suggest that there is a negative relation
between education and unemployment duration. Schooling up to 12 years can reduce the expected
length of unemployment by more than 4%, and qualifications above that level can reduce up to 12%
6
(Nickell, 1979). Job opportunities rise with the amount of years spent on education (Kettunen, 1997).
On the other hand, Ashenfelter and Ham (1979) and Ciuca and Matei (2011) found that people with
higher education are not advantaged on the labor market, hence finding no evidence of the impact on
unemployment duration but concluded that work experience can reduce the duration of unemployment
spells.
Stigler (1962) also claims that it is difficult for the individual to collect information on possible em-
ployers, firms and wage offers. Studying the demand of labor takes time and resources. Mincer (1991)
provides evidence to explain why educated workers have lower unemployment incidence and experience
shorter durations of unemployment, by focusing on the search behavior. The evidences are: 1) costs of
on-the-job search compared to costs of searching while unemployed are lower for educated individuals;
2) educated workers are more efficient in the job search activity; and 3) both firms and workers tend
to search more intensively to fill more skilled job positions. The length of job-search affects both the
matching function and the wage function (Blanchard and Diamond, 1994).
Griliches (1969) provides evidence through various papers indicating that capital and skilled labor
are relatively more complementary than are capital and unskilled labor. This hypothesis is called capital-
skill complementarity. By this, Griliches (1969) meant that, although capital can be complementary to
all levels of skills, the level of complementarity tend to be higher for skilled labor. Such complementari-
ties were studied in the manufacturing industry in the United States, where companies that used more
capital per worker, hired more educated workers and paid them higher wages (Goldin and Katz, 1996).
As a result of the rapid growth in the demand for more-skilled workers, wage inequalities started to be
an issue (Autor et al., 1998; Bound and Johnson, 1992; Murphy and Katz, 1992). While the share of
workers with higher qualifications was rising, the wages of unskilled workers were decreasing (Addison
and Teixeira, 2001). Low-skilled people have higher hazards of job separation.
The literature suggests different reasons for the decline in unskilled labor, being one of them the skill-
biased (or unskilled labor saving) technological change (SBTC) hypothesis (Berman et al., 1998; Katz
and Autor, 1999; Acemoglu and Autor, 2010). SBTC is a shift in the production that changes the relative
demand for skills (Castro Silva and Lima, 2017), favoring skilled over unskilled workers (Violante, 2000),
where capital reallocates from slow to fast learning workers (Caselli, 1999). The increase in the share of
well-educated workers in employment came to prove that skilled workers are in a better position in the
labor market (Addison and Teixeira, 2001).
In 1990, the demand for medium-skilled workers decreased when compared with the demand for
high and low-skilled people. With this phenomenon the polarization of labor markets rose all over the
world: in the United States (Autor et al., 2006; Autor and Dorn, 2013), in Europe (Spitz-Oener, 2006;
Goos et al., 2009) and in the United Kingdom (Goos and Manning, 2007). The SBTC is insufficient in
supporting the recent situation facing job polarization. Goos et al. (2014) explained the job polarization
7
effect through routine-biased technological change (RBTC) and task offshoring. RBTC means that tech-
nological change is biased against labor in routine tasks. RBTC and task offshoring combined provoke a
decrease in the demand for middle-skilled workers (Autor et al., 2003, 2006; Goos and Manning, 2007;
Autor and Dorn, 2013).
Autor et al. (2003, 2006) analyzed job polarization by studying the effects of computers in the relative
demand for skills. Their model predicts that labor intensive firms performing routine tasks, will invest
more in computer capital, as their prices decreases. Therefore, low-skilled labor is substituted by com-
puter capital. Autor et al. (2003, 2006) suggest assigning more abstract tasks to high-skilled workers,
routine tasks to middle-skilled and manual tasks to the low-skilled. The assumption of SBTC is that new
technologies like computers affect jobs and skill requirements in jobs.
In times characterized by technological change, the average unemployment duration will probably
rise (Wolff, 2005). Low-skilled and less educated workers will face difficulties in finding jobs in techno-
logical sectors, hence spending more time unemployed and searching for new jobs.
2.1.2 Personal Characteristics
Other variables such as gender and marital status may affect the unemployment duration. Nickell (1979)
found that unmarried men have longer expected duration of unemployment when compared to married
men. The expected duration of unemployment decreases with the number of dependent children (Nick-
ell, 1979). Haile (2004) concludes that married people have 24% lower hazard of re-employment when
compared to single people.
Regarding gender Ciuca and Matei (2011) found no relevant difference between men and women,
just a lightly higher hazard for men. On the other hand, Hernæs and Strøm (1996) finds that the exit
probability out of unemployment is higher for women than men, and attributes that to the lower reserva-
tion wage of women. In this case, women would accept jobs with lower wages than men.
There is also a considerable distinction in results regarding developed and developing countries.
Tansel and Tasci (2010) compared the unemployment duration for men and women in Turkey and con-
cluded that the hazard was substantially lower for women. Bowers and Harkess (1979) studied the
impact of age and gender in the British labor market, and found a rise in the rate of entry of men to
the unemployment register and a fall in the expected duration of an entry. For women there was a rise
in expected duration but no fall in entry rates. Comparing men and women, the expected duration of
women entrants declined against men, as did rates of entry to the register. Labor market prospects as
measured by expected duration moved in favor of younger workers independently of gender.
Ollikainen (2003) finds that in Finland women aged 16 to 19 years old experience shorter durations
8
compared to men when exiting to employment, but longer durations when exiting to economic inactivity.
Women have higher hazard of re-employment between 16 and 29 and men have higher hazards when
aged between 20 and 39 years.
Age is another important factor regarding displacement of workers. Friedberg (2003) analyzed how
older workers are affected by technological change. Older people will consider whether to upgrade a
skill, since they have fewer years left in the labor market in comparison with younger people. With the
increase of technology intensity, it is less likely for a person to find a new job after being displaced. Such
situation is aggravated when the people in question are low-skilled and older. It is more difficult for them
to find new post-displacement employment. Therefore, the length of unemployment will be higher for
older workers and workers with low levels of education, as the average weeks of unemployment rise
proportionally with age (Wolff, 2005). Ciuca and Matei (2011) concluded that people between the ages
of 36 and 55 years have higher probabilities of remaining unemployed, attributing this fact to the adapt-
ability to changes in the labor market. Haile (2004) finds that older workers have 59% lower hazard of
re-employment when compared to younger workers.
Friedberg (2003) analyzes the relationship between computer usage and retirement, estimating that
computer use lowers the likelihood of retirement. This means that skilled workers stay in the labor market
for longer periods, filling vacancies that could be occupied by younger unemployed individuals. Older
and poorly educated workers remain unemployed for longer periods of time, times which are charac-
terized by low human capital investments, therefore facing long-term unemployment. Also, older people
may decline jobs for other reasons, such as mobility as a result of family and other responsibilities (Haile,
2004).
Haile (2004) found that previous jobs and labor market history had importance on the re-employment
hazard. Workers who had unskilled manual jobs experience longer periods of unemployment compared
with those who had high-technological jobs or managerial positions. Those who worked in small and
medium sized firms have 28% higher hazards, thus experiencing shorter periods of unemployment,
when compared with those working in large firms.
2.1.3 Unemployment Insurance
Authors such as Burda and Sachs (1988); Katz and Meyer (1990); Meyer (1990, 1995); Bover et al.
(2002) studied the effect of unemployment insurance on the duration of unemployment. Unemployment
insurance is a form of compensation for unemployed people. They receive a support income every
month. Mortensen (1970) was the first to include unemployment benefits into the job search analysis.
He concludes that individuals would accept a job offer if the benefit of it was larger than their reservation
wage. The income from the unemployment insurance system sets the price at which an unemployed
individual is willing to work. The higher the unemployment benefit, the weaker the incentive to accept a
9
job offer is (Hernæs and Strøm, 1996).
Katz and Meyer (1990) provide evidence that countries with generous unemployment benefits have
higher unemployment rates, as well as, longer periods of unemployment spells. They find that an in-
crease of one week of benefits, increases the average length of unemployment spells from 0.16 to 0.20
weeks. Burda and Sachs (1988) also finds a correlation between a measure of the generosity of un-
employment insurance benefits and the ratio of long-term unemployment. The impact of benefit levels
on the conditional probability of getting a job in any given moment is significant for the first 20 weeks
(Nickell, 1979). Hunt (1995) studied the effects of unemployment benefits in Germany and found that for
recipients of unemployment insurance between 44 and 48 years there was a great increase in unem-
ployment duration when compared to younger people, while the effect for people between 49 and 57 was
smaller. Bover et al. (2002) concludes that the hazard rate for recipients is double the rate of workers
without benefits, when the largest effects occur for a three-month period of unemployment. Therefore,
unemployment insurance reduces the hazard of leaving unemployment.
The rates of re-employment and job search increase in times where the benefits are likely to pre-
scribe (Katz and Meyer, 1990). This does suggest that many unemployed people are comfortable with
the income from the unemployment insurance, and therefore may function as a large disincentive to
work. Unemployment insurance assists efficient job search, by giving recipients more time to search for
offers and analyze them. However, unemployment insurance has the disadvantage of possibly prolong-
ing the unemployment spell.
2.1.4 Duration Dependence
The duration of unemployment itself can also be a decisive factor. There is a negative relationship be-
tween unemployment duration and the likelihood of finding a job (negative duration dependence), i.e.
long-term unemployment negatively affects the re-employment chances (Steiner, 1990).
The probability of an unemployed individual finding a job declines steadily after the first six months of
a spell (Nickell, 1979). Long periods of unemployment are usually times characterized by low training,
or even loss of human capital ( through obsolescence) and may be seen by employers as a signal of
reduced productivity. On the other hand, positive duration dependence periods may occur as the result
of the long-term unemployed being less selective when it comes to accepting jobs (Hernæs and Strøm,
1996).
Van Den Berg and van Ours (1994) studied the duration dependence in France, the Netherlands and
the United Kingdom. They found that in France there was no duration dependence during the first year,
while for the Dutch there was a non-monotonous (inverse-U shaped) duration dependence over the first
10
three quarters. In the UK male individuals experienced a decline in the exit rate over duration. Van Den
Berg and Van Ours (1998) studied the French focusing on younger individuals. They found negative
duration dependence for young women and no significant duration dependence for young men during
the first year.
2.2 Human Capital
The term capital comprises any activity or action that yields income. In the same line of reasoning, any
type of schooling and training can be perceived as an investment in capital. Capital which can later yield
income when employed in the labor market. The following section will present human capital comprising
the subjects in Figure 2.2.
Humancapital
Innateability Schooling Training
Generalcapital
Firm-specificcapital
Jobinstructiontraining
Jobrotation Coaching Apprentice-ship
Figure 2.2: How human capital can be acquired.
In the 18th century, the definition of capital was extended to include the concept of human capital
(Smith, 1776). Smith (1776) included the inhabitants’ useful knowledge and characteristics, as abilities
and qualities (either innate or acquired) since it provides and increases wealth for both the individual
and society in general (Laroche et al., 2017). Human capital is a measure of the economic value of an
employee’s skill set. It can be acquired through a number of different sources1: innate ability, schooling
and, training. Innate ability refers to the amount and diversity of skills from innate differences. Different
people have different characteristics, and those characteristics can have effects on the productivity and
efficiency of a worker. Schooling constitutes the most common source of acquiring human capital. Train-
1See Acemoglu and Autor (2009) for an excellent discussion of this, and other Labor Economics topics.
11
ing is usually acquired after schooling and is associated to a set of skills needed in a particular industry
for a particular job. Firms have interest in investing in their workers, sometimes paying for such training,
though informal on-the-job training also occurs. Through training a person can achieve higher levels of
knowledge and productivity.
As far back as Adam Smith’s time, economists had observed that the efficiency of production was
not only dependent on equipment or land, but also on peoples’ abilities. Until then, labor was treated as
an undifferentiated mass of workers, aggregating skilled and unskilled workers. At that time, the view
on training as human capital investment, was pessimistic. Pigou (1920) believed that there would be
an under-supply of trained workers, since companies would not want to train their employees, for the
possibility of them being taken by rivals. Although, the human capital concept was broadly forgotten until
the early 1960’s, where the concept is renewed by Becker (1964). Becker (1964) first contribution was to
make a distinction between specific and general human capital. Specific capital is knowledge acquired,
linked to a certain activity performed during a specific job or task. Companies are willing to pay for this
type of training since it is not (fully) transferable. By contrast, as Pigou (1920) defended, companies are
often reluctant to stump up for general human capital. This investment represents knowledge transverse
to every company. Workers with general training can make use of that training in other companies, for
instance higher-paying companies.
Firm-specific training can be divided into formal training and informal or on-the-job training. This
constitutes an important source of increased productivity and consequently higher wages, as workers
gain more experience at work (Becker, 1964). Formal training is often performed away from the job or
through computer-based programs, whereas on-the-job training occurs at the work place while doing
daily tasks. The purpose of this training is to provide the worker with task-specific knowledge and skills
directly related to job requirements. There are various methods to achieve this purpose: job instruction
training, job rotation, coaching and apprenticeship. Job instruction training consists of four steps (prepa-
ration, present, try out and follow up) and is used when (1) there is a need to teach manual skills or
procedures, (2) train new employees or apprentices at the work place or (3) prepare competent workers
who are able to perform specific job tasks at any level. This technique achieves greater productivity
and fast results. Job rotation consists of a systematic movement of employees from a job to another
within the organization, with the desire of achieving various different human resources objectives. A well
performed job rotation program can decrease training costs while increasing the impact and efficiency
of training. Coaching is a one-to-one guidance and instruction that helps to quickly identify weak areas.
It also offers the benefit of improving knowledge, skills and work performance. This method is directed
at workers with performance weaknesses, but also as of a motivational tool for those with good perfor-
mance. Apprenticeship is one of the oldest forms of training and was the major approach to learning a
craft. Is designed to provide planned and practical instruction over a significant time span where a new
generation of practitioners learn a skill. The objective of this training is to make the trainers all-round
craftsmen.
12
Firms decide how much to invest in training by calculating the net present value of the costs and ben-
efits of such decision. The total investments in on-the-job training was almost as large as the investment
in education (Mincer, 1958). Becker (1964) also suggested that job changes were more frequent within
unskilled workers opposed to skilled. Becker (1964) noticed that the most common and observable
source of human capital was through schooling and that people were acquiring general human capital
at their own expense. People had to take on debts to pay for education before entering the labor market.
Something that economists avoided to stress was the fact that people were investing in themselves and
that those investments were large (Schultz, 1961). Becker (1964) assumed that people would carefully
calculate how much to invest in such capital and compare it to expected future earnings from different
career paths. This helped him realize why younger generations would spend more money and time in
schooling than older ones. Young people had more years ahead of them, over which they could amortize
such investments (Schultz, 1961).
The spread of education can be explained by technological change. Advances in technology made
skills profitable, hence raising the demand for education (Becker, 1964). The general analysis of in-
vestment in human capital made by Becker (1964) came to conclude that the unemployment rate tends
to be inversely related to the skill level. People with higher levels of education have higher chances of
staying employed (Kodde, 1988). We are converging to a highly technological world, where the number
of skilled people is rising every year. In Portugal the number of highly-educated people rose from about
560 thousand in 2000 to 1.6 millions in 2016. (Figure 2.3).
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1996
1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018
Figure 2.3: Number of residents in Portugal with a higher level of education.Note: Data from 1985 to 2016 presented in millions. Dashed lines correspond to breaks in the series. (Source:
INE, PORDATA)
13
Nowadays, the labor market is in constant change. About half of new employer-employee matches
end in a period of a year, and one-fifth of all workers have been in their firm for less than a year (Farber,
1999). With the evolution of technology in past years, older crafts and routinized tasks susceptible to be
automated are disappearing from the labor market. The advances made in the robotics field are further
accelerating the pace at which these individuals are being replaced (Frey and Osborne, 2017). Each
year, 10% of the jobs are destroyed (Davis and Haltiwanger, 1999). Such circumstance only makes
finding new jobs and specializing in new skills more complicated. As there are indicators that skilled and
better educated workers have an advantage to better adjust to the implementation of new technologies
in general (Bartel and Lichtenberg, 1987), i.e. education enhances adaptability to change (Riddell and
Song, 2011), and leads to a more efficient decision making (Schultz, 1975). Workers who stay with their
firms for longer periods of time tend to accumulate more human capital, enhancing their productivity
(Addison and Teixeira, 2001). In Figure 2.4 it is possible to observe unemployment in Portugal by level
of education. It is clear that the rate of unemployment for people with low levels of education is substan-
tially higher.
-
10
20
30
40
50
60
70
80
90
100
Basic High-school College
Figure 2.4: Unemployed population of Portugal divided by education level: basic, high school and college.Note: Data from 1998 to 2016 presented as a percentage of the unemployed population. (Source: INE, PORDATA)
2.3 Technological Unemployment
There are different types of unemployment with distinct causes: frictional unemployment, structural un-
employment, cyclical unemployment and technological unemployment. The fear for the latter type
of unemployment always arisen in ages strongly characterized by a profound technological change (Vi-
varelli, 2014). “Technological unemployment” was a term originally popularized by Keynes (1930),—
”This means unemployment due to our discovery of means of economizing the use of labor outrunning
14
the pace at which we can find new uses for labor”. This phenomenon takes place when new machinery
and other forms of technology are more efficient than human labor or at least make human labor more
efficient. Hence, firms will increase capital and decrease labor, leading to a rise in unemployment.
However, this issue has been talked about since at least Ancient times. These worries and anxi-
eties over technological unemployment are not new to the modern era. Technological advances often
appeared to take away jobs, but in the long run they led to the creation of more, albeit possibly different
jobs. In order to understand if this time is different, it is imperative to comprehend the evolution and
history throughout the years (Mokyr et al., 2015).
2.3.1 Early Stages
The Industrial Revolution happened in the 1770’s in Britain. The population was hit by the angst of
job-loss and the intensified growth of mass unemployment due to the impact of this devastating action.
In sequence of this technological change, movements against this evolution of technology started to
appear. The Luddite Movement appeared in the 19th century and was a famous anti-technology group
of textile workers, that destroyed weaving machinery in form of protest (Vivarelli, 2014). This group of
workers feared that their craft’s skills would become obsolete.
Keynes (1930) prophesied an alarming future for economics: assuming that no significant wars or in-
crease in population occurred, the economic problem could be solved, at least within a period of hundred
years. This meant that the economic problem was not a permanent problem of the human race. The
economic problem that Keynes (1930) refers to is the economic recession from the Great Depression.
The biggest cause for the event was the Wall Street crash. Although it happened in the United States
it did not take long to spread worldwide. The stock market crash in 1929 wiped out a generous amount
of nominal wealth, both corporate and private. Before the crash, the unemployment rate in the United
States was 3.2%, and by 1933 it had risen to 24.9%.
In an attempt to understand the Great Depression, Keynes (1930) developed what went on to be
called Keynesian theory. This theory supported the idea that, in the short run, the way out of recession
was through aggregate demand, the total expenditure in the economy. He proposed an increase in
government expenditures and lower taxes in difficult times, in order to stimulate demand and encourage
people to spend their money. The Keynesian theory considered personal savings a drag on the economy.
Despite all the stagnation and poor economic performance around the world, including in the United
Kingdom, Keynes (1930) remained optimistic. His idea rested on the power of technology as a driver
of growth, rather than a cause of unemployment. He forecasted that technology would complement,
empower and raise the worker of tomorrow (Wilson, 1999). He also predicted that the economy would
be so productive that people would not have to work more than fifteen hours in a week. Yet almost a
15
century later this has not been the case. The average workweek, in 1930 was 47 hours, compared to the
39 hours in 1970. It almost seemed that Keynes (1930) prediction were right. However, since 1970, the
working week remained unchanged, at 40 hours (Wilson, 1999). Technology has had ambiguous effects
on the labor market, ranging from fierce opposition and fears of automated job-loss to an uncertainty as
to the extension of modern technologies taking away jobs.
2.3.2 Technological Anxiety
Technological change has created anxiety throughout history. This anxiety can manifest itself through
two related concerns. First, the most common and discussed concern is that the rapid technological
change will cause a replacement of labor for machines, leading to a widespread increase in technologi-
cal unemployment. The second concern is related with the moral implications of this progress for human
society and prosperity. The Industrial Revolution was considered a symbol of dehumanization, where the
most human activity — work — was transformed into something entirely inhuman (Mokyr et al., 2015).
The debate about technological unemployment has long been a matter of discussion. During the
Industrial Revolution, economists were divided over whether technological progress would lead to a
lower rate of employment. The participants of this debate can be divided into two distinct groups, the
optimists and the pessimists. The optimists defend that the long-term effects of automation on the
labor market and productivity are clearly beneficial (Wilson, 1999). A factory that saves money on labor
through automation will either (1) generate an increase in demand by lowering prices, that may lead to
the creation of new jobs, or (2) generate more profit and consequently pay higher wages, leading to an
increase in consumption or investment, and thus more employment. The optimistic point of view also
agrees that technology will continue to grow and expedite, and that innovation might cause disruption in
jobs in the short to medium-term. On the other hand, pessimists believed that machines could remove
humans from the labor force more permanently, defending that new technologies would lead to a lasting
and significant decline in the number of workers employed, causing long-term unemployment.
2.4 Evolution of Unemployment in Portugal
This section seeks to characterize the evolution of unemployment in Portugal comparing statistics from
2006 and 2016. We compare two points in time with 10 years apart to see how the values differ. The
information available in Figure 3.1 can be used as input for computing equations 4.1 and 4.2. In 2016,
of the total population (10.3 million people), only 5.2 million were in the labor force. Of this latter group,
4.6 million individuals were employed and 0.6 million were unemployed.
Figure 2.5 shows that Portugal had a period of fluctuation with more or less constant values of unem-
ployment, closely matching the economic cycle. However, after the year 2008 we can observe a general
trend of quickly increasing unemployment until 2013. The increasing levels of unemployment are mainly
16
0
2
4
6
8
10
12
14
16
18
1980
1985
1990
1995
2000
2005
2010
2015
2020
Figure 2.5: Unemployment rate in Portugal.Note: Data from 1983 to 2017 presented as percentage. Gray bars correspond to periods of economic contraction
and dashed lines to breaks in the series. (Source: INE, PORDATA)
due to the international economic crisis that began in late 2008, the effects of which were, and continue
to be, strongly felt in Portugal — in 2013, the unemployment rate was 16.2%, a record high. In Figure
2.5 the periods of economic contraction are identified by gray bars. It is clear that during these periods
there is an increase in the unemployment rate.
To further evaluate the state of unemployment in Portugal we proceed to study the unemployment
rates by gender, age, education level, region and sector activity. In Chapter 2 we analyzed unemploy-
ment duration and concluded that there are several factors affecting it. The factors that we are going to
analyze can play an important role in the unemployment duration of an individual.
The study by gender, in Table 2.1, allows us to conclude that, although the rate of unemployment has
increased for both men and women, the rate of unemployment of women suffered a lower variation. The
variation of unemployment between 2006 and 2016 for men was 4 percentage points (p.p. henceforth),
whereas for women it was 1.6 p.p.
In Table 2.2 we show the average unemployment duration by gender. Unemployment duration for
both men and women is dominated by three different intervals: between one to six months; between 12
and 24 months; and 25 months and over. Observing the row 25 months and over for men and women we
conclude that long-term unemployment is increasing. In 2016, close to half of the unemployed men and
women are searching for a job for more than 25 months. Although, there are studies (Tansel and Tasci,
2010, for example) referring that women suffer longer periods of unemployment, by the data provided
we observe that the values are similar. This shift in the duration of unemployment can be also related to
17
Table 2.1: Rates of employment and unemployment by gender.Note: Data presented in thousands (Source: INE, PORDATA)
Gender State 2006 % 2016 % Variation (p.p.)
Men Employed 2 779.9 93.0 2 361.4 89.0Unemployed 208.7 7.0 291 11.0 4.0
Total 2 988.6 100.0 2 652.4 100.0
Women Employed 2 362.9 90.4 2 243.8 88.8Unemployed 249.8 9.6 282 11.2 1.6
Total 2 612.7 100.0 2 525.8 100.0
Table 2.2: Unemployment duration by gender.Note: Data presented in thousands (Source: INE, PORDATA)
Gender Unemploymentduration 2006 % 2016 % Variation (p.p.)
Men less than 1 month 12.4 6.0 10.4 3.6 -2.4from 1 to 6 months 54.1 26.0 69.0 23.7 -2.3from 7 to 11 months 23.4 11.2 25.1 8.6 -2.6from 12 to 24 months 44.8 21.5 50.3 17.3 -4.225 months and over 73.5 31.3 136.2 46.8 15.5
Total 208.7 100.0 291.0 100.0
Women less than 1 month 14.4 5.8 12.0 4.3 -1.5from 1 to 6 months 85.4 34.2 74.7 26.5 -7.7from 7 to 11 months 30.7 12.4 26.2 9.2 -3.2from 12 to 24 months 44.8 18.0 44.8 15.9 -2.125 months and over 73.5 29.6 124.3 44.1 14.5
Total 249.8 100.0 282.0 100.0
the technology intensity felt in recent years.
We divided the analysis of unemployment by age (Table 2.3) into four age groups: from 15 to 24,
from 25 to 34, from 35 to 44 and from 45 and more. The unemployment rate in the age group from 15
to 24 rose from 17.9% to 28%, with a variation of 10.1 p.p. These figures suggest that graduates and
younger individuals may find it harder to find jobs than before. The recent European recession greatly
impacted youth, only aggravating the situation.
In Table 2.4 we present unemployment by level of education. It is possible to observe that the results
remain stable between 2006 and 2016. However, it is also visible that we are growing towards a more
skilled population — the number of people with basic education is falling and the values for individuals in
higher education almost doubled in ten years. Although in absolute terms the numbers are very distinct
for tertiary education in absolute terms, the rate of unemployment suffered a small variation of 1.1 p.p.
The data available on unemployment by sector or occupation is limited. However, this analysis is
crucial in our study since there are sectors and activities more vulnerable to technological change and
18
Table 2.3: Rates of employment and unemployment by age group.Note: Data presented in thousands (Source: INE, PORDATA)
Age State 2006 % 2016 % Variation (p.p.)
from 15 to 24 Employed 442.6 82.1 262.4 72.0Unemployed 96.2 17.9 101.8 28.0 10.1
Total 538.8 100.0 364.2 100.0
from 25 to 34 Employed 1 337.3 90.3 923.1 87.5Unemployed 143.6 9.7 131.7 12.5 2.8
Total 1 480.9 100.0 1 054.8 100.0
from 35 to 44 Employed 1 325.5 93.2 1 308.1 91.5Unemployed 96.2 6.8 121.2 8.5 1.7
Total 1 421.7 100.0 1 429.3 100.0
45 and over Employed 2 037.5 94.3 2 111.6 90.6Unemployed 122.6 5.7 218.3 9.4 3.7
Total 2 160.1 100.0 2 329.9 100.0
Table 2.4: Rates of employment and unemployment by education level.Note: Data presented in thousands (Source: INE, PORDATA)
Education State 2006 % 2016 % Variation (p.p.)
Basic Employed 3 628.8 91.7 2 227.4 88.2Unemployed 327.4 8.3 299.0 11.8 3.5
Total 3 956.2 100.0 2 526.4 100.0
High school Employed 788.1 91.3 1 182.1 87.8Unemployed 74.7 8.7 165.0 12.2 3.5
Total 862.8 100.0 1 347.1 100.0
College Employed 725.9 92.8 1 195.8 91.7Unemployed 56.5 7.2 109.0 8.3 1.1
Total 782.4 100.0 1 304.8 100.0
to the relative changes in demand for skills. In Table 2.5 we analyze the previous employment sector
of unemployed for a duration of less than 8 months. The results allow us to conclude that the number
of unemployed people in the agriculture and industry sectors remained almost unchanged on absolute
terms, though the variation between 2006 and 2016 was greater for the industry sector. For the service
sector we observe a large increase in the amount of people unemployed, as well as an increase in the
respective unemployment rate, suggesting a shift from the industry sector to the service sector.
In Table 2.6 we present unemployment rates by regions, according to NUTS II. Between 2006 and
2016 the unemployment rate rose all over Portugal, although some regions suffered higher increases.
Azores and Madeira had the largest variations of all of the Portuguese regions. Portugal has a higher
population density in North, Center and Lisbon. In more populated regions, typically near cities, there
are more job offers than in less populated areas. This does not mean that Lisbon has lower unemploy-
19
Table 2.5: Unemployed people by previous sector activity.Note: Data presented in thousands (Source: INE, PORDATA)
Sectoral activity 2006 % 2016 % Variation (p.p.)
Agriculture 9.9 2.8 11.8 2.5 -0.8Industry including Construction 155.2 44.2 147.4 31.1 -13.1Services 186.2 53.0 314.9 66.4 13.4Total 351.3 100.0 474.1 100.0
ment rates as seen below.
Table 2.6: Rates of employment and unemployment by region according to level NUTS II.Note: Data presented in thousands (Source: INE, PORDATA)
Region 2006 (%) 2016 (%) Variation
Portugal 7.7 11.1 3.4North 8.9 12 3.1Center 5.5 8.4 2.9Lisbon 8.5 11.9 3.4Alentejo 9.2 12.1 2.9Algarve 5.5 9.2 3.7Azores 3.8 11.1 7.3Madeira 5.4 12.9 7.5
20
Chapter 3
Data for the Study of Unemployment
Duration
The data for our study originates from the Portuguese Labor Force Study (LFS), obtained from the
National Institute of Statistics (INE) in Portugal. The data available for econometric analysis is limited to
the time span between the year 2011 and 2013 because more recent data was not made available to
the author at the time of this study. This chapter is organized as follows: Section 3.1 describes the LFS,
Section 3.2 presents the definition of the main variables used in the study of unemployment duration,
and Section 3.3 introduces the description our study.
3.1 The Labor Force Survey (LFS)
The LFS is a household survey, carried out every three months by sampling and conducted by INE, and
has a quasi-longitudinal capacity (Addison and Portugal, 2001). INE provides the quarterly and annual
results of the survey covering the entire national territory, with the objective of characterizing the work
in Portugal, particularly the transitions between employment and unemployment. The official statistics
regarding the condition of work consider many characteristics of the Portuguese population. Some of
the collected data in the survey is related with the sector of economic activity, education and professional
qualifications, job search and career path, as well as, personal characteristics.
This survey collects a generous amount of individual information allowing to cross variables, which
help increase our understanding of the national reality in terms of employment (Correia and Lima, 2006).
The LFS is a major source of information on the personal characteristics of the working-age population,
including age, sex, marital status, level of education, and family structure.
Employment estimates can be analyzed at different levels of detail: by demographic characteris-
tics, industry and occupation, job tenure, and usual and actual hours worked. The survey includes
questions allowing analyses of many topical issues, such as involuntary part-time employment, multiple
21
job-holding, and absence from work. Unemployment estimates are presented by demographic group,
duration of unemployment and activity before looking for work. Information on industry and occupation,
and reason for leaving the last job is also available for individuals currently unemployed or not in the
labor market but with recent labor market involvement.
The LFS is to divides the working-age population (age ≥ 15) into three mutually exclusive groups
of people — employed, unemployed and inactive or not in the labor force — and to give explanatory
data on members of each of these categories. In Figure 3.1 it is possible to observe how the population
is divided, as well as the data regarding the Portuguese population in 2016. Since the survey is per-
formed regularly, it is not only possible to have information on the structure of the labor market and its
functioning, but also to conduct an analysis from quarter to quarter.
Totalpopulation(10.3)
Notinlabourforce(3.7)
Under15(1.4)
Inlabourforce(5.2)
Employed(4.6)
Unemployed(0.6)
Figure 3.1: Total population of Portugal according to labor force state.Note: Data presented in millions (Source: INE, PORDATA 2016)
The unemployment rate is one of the most relevant indicators of the economic performance. The
labor force consists of those who want to work, whether employed or unemployed (Equation 3.1). Those
who are 15 years of age or older and are not institutionalized but officially neither employed nor unem-
ployed, are not considered in the labor force. The unemployment rate is calculated by the ratio between
the unemployed and employed population within the labor force (Equation 3.2).
Labor force = employment+ unemployment∗ (3.1)
The unemployment∗ in Equation 3.1 refers only to the number of unemployed individuals, excluding
the inactive.
Unemployment rate (%) =unemployed
labor force× 100 (3.2)
22
3.1.1 Survey Methodology
The population characterized in the LFS are individuals residing in the Portuguese territory (in family
housing, considering main residence). The sample used is probabilistic and stratified at level II of the
Nomenclature of Territorial Units for Statistical Purposes (NUTS II), and follows a quality standard to
ensure that the information reflects the behavior of the population and is characterized by a low margin
of error.
The sampling unit is the household, selected from a sampling basis, named Mother-Sample built by
INE based on the General Population and Housing Census of 2001 (Census 2001), to conduct house-
hold surveys. Currently, about 22,500 accommodation units are surveyed quarterly. The LFS sample is
evenly distributed for the weeks of each quarter. Each accommodation is associated with a pre-defined
week, called reference week. The interviews are made in the week following the reference week, or at
most two weeks after.
The sample consists of six sub-samples, also known as rotations. In each new quarter, a rotation
stops being surveyed, and is replaced by a new one. For example, in the schematic represented in
Figure 3.2, we can see that in the third quarter of 2016 (3Q 2016), the set of accommodations A is no
longer surveyed, and the set G starts the inquisitive survey.
Once selected to be part of the sample, an accommodation is visited during six consecutive quarters.
This repetitive pattern allows us to calculate the evolution indicators, as in two consecutive quarters, five
of the six rotations are common. Thus, we can record transitions out of unemployment for up to five
quarters. The transition rate is obtained by identifying the unemployed population contained in the sur-
vey, and who got re-employed in a given quarter (Addison and Portugal, 2001). Figure 3.2 shows that
between the second (2Q) and third (3Q) quarters of 2016, the sets of accommodation B, C, D, E and F
are surveyed in both quarters.
By analyzing the flows in and out of unemployment we are able to define specific characteristics of
the labor market. The LFS is flexible to the extent that it allows to work with the information obtained on
relevant labor market aspects regarding all market sectors of the economy in a consistent method.
23
Quarter
2015
1Q 2Q 3Q 4Q
2016
1Q 2Q 3Q 4Q
6 5 4 3 2 1
6 5 4 3 2 1
6 5 4 3 2 1
A
B
C
Seto
fHouseholds
2017
1Q 2Q 3Q 4Q
D
E
F
G
6 5 4 3 2 1
6 5 4 3 2 1
6 5 4 3 2 1
6 5 4 3 2 1
Figure 3.2: Rotation of the Labor Force Survey samples.(Source: INE)
3.1.2 Data Collection
The collection of data for the LFS is carried out each quarter during the week following the LFS reference
week. The information is collected through a direct interview to all of the members in the household.
When one of the members of the household cannot respond, the information is obtained through another
member of the household able to do so — these kinds of responses are called proxy. The interviewer
first obtains socio-demographic information for each household member and then obtains labor force
information for all members aged 15 and over. In subsequent monthly interviews the interviewer confirms
the socio-demographic information collected in the first month and collects the labor force information
for the current quarter.
3.1.3 Data Accuracy
The LFS samples only a part of the population. Therefore, the information collected in the sample,
should be transposed to the entire population for a more general analysis. For that we use sampling
weights associated to each individual’s unique identification. This is accomplished via an estimation
process: each individual in the sample represents a subset of individuals from the entire population,
with similar characteristics.
There is an error associated with this process, as different samples result in different values for the
estimates. Thus, a dispersion in the obtained values is expected. This dispersion constitutes the sam-
pling error. Therefore, it is possible to measure the variability resulting from the use of a sample that is
intended to reflect the entire population. The relative measure used is the coefficient of variation. It con-
24
sists of dividing the standard deviation by the estimate value. The calculation of coefficients of variation
allows establishing a quality criteria of the statistical information for the revelation of the estimates.
3.1.4 Limitations
Official statistics have limitations and can, many times, understate the true unemployment rate. Apart
from the typical adversities associated with surveys, such as inaccuracy in responses, there are some
other issues. As Figure 3.1 reflects, the population is categorized regarding their labor market status in
three categories. However, between categories there are ”gray areas”, as underemployed people are
not counted: some part-time workers are considered as fully employed, when in reality they would like
to work full-time, and also those who are working at jobs below their skill levels and/or pay grades. In
Portugal, there is a sizable amount of part-time workers (9.1%, according to OECD data from 2016),
which should not be undervalued. However, this can work both ways: for part-time workers desiring to
work full-time and the inverse situation, where full-time workers would prefer to work part-time.
Another issue is the inclusion or exclusion of the unemployed in the labor force. Since the unemploy-
ment rate is measured as a percentage of the labor force, and to be officially unemployed one needs to
be actively searching for a job, a person who does not fulfill this latter requirement will not be counted.
This means that the so-called discouraged workers do not count towards the unemployment rate be-
cause they are officially classified as not in labor force. Studies show that people become discouraged
and abandon job searching after repeated unsuccessful attempts. However, they might accept a job if
it came along. This category of workers constitutes hidden unemployment (McConnell et al., 2017). In
Portugal, in 2000 the number of discouraged workers was 12,000 people, whereas in 2013 the number
increased to 118,000 people. These figures on discouraged workers indicate that the measured unem-
ployment rate may underestimate the true magnitude of unemployment.
Finally, there are other limitations regarding the LFS. For instance, the cost of the overall sample size
and the level of detail achieved when analyzing the results.
3.2 Main Definitions
The notion of employment and unemployment comes from the theory of supply of labor as a production
factor. The supply of labor is the total number of hours workers are willing to work at a given wage rate.
In this section we present the general concepts and definitions of employment and unemployment, as
well as the main variables used in our analysis.
Employment: The employed individuals are those who, during the surveyed reference week:
1. performed any work at all at a job or business, i.e., paid work in the dynamics of employer-
employee, or self-employment, including also unpaid family work.
25
2. had a job but were not present due to factors such as illness, personal or family responsibilities,
vacation or other reasons.
Unemployment: Unemployment corresponds to an underutilized supply of labor; the functional defi-
nition of unemployment is based on the job-search activity and the availability to accept a job. Therefore,
the unemployed individuals are those who during the surveyed reference week:
1. were on temporary layoff during the reference week with an expectation of recall and were available
for work, or
2. were without work, but had looked for a job in the past four weeks, and were available for work, or
3. had a new job to start within four weeks from the reference week and were available for work.
The available individuals are those who: (1) could have worked in the reference week if an appropri-
ate job had been presented; (2) the reason of not accepting a job was a temporary matter; (3) already
had a job starting in the immediate future. Note that full-time students looking for full-time jobs are not
considered available for work during the reference week.
The variables used for the analysis of unemployment duration includes age, gender, educational
level, unemployment duration, class of worker, and industry level of technology intensity. Next, we intro-
duce the definitions of the terms and variables used in the LFS.
Age: The information about age is collected for every household member in the survey, although the
information about labor market activity is only collected for people aged 15 and over. Age is divided in
categories: 1) from 15 to 24, 2) from 25 to 34, 3) from 35 to 44, 4) from 45 to 54, 5) from 55 to 64 and
6) more than 64. We choose to keep individuals until the age of 64, because it was the mean age of
retirement in 2013.
Education: Corresponds to the highest level of schooling completed. This variable is divided into
the following categories: 1) basic, 2) high-school, and 3) college.
Technology/knowledge intensity of previous employment sector: Classification of firms regard-
ing the manufacturing technology or services knowledge intensive, according to the OECD’s definition,
using Eurostat’s version with NACE Revision 2 codes at the 2-digit level1. This variable is divided into
the following categories: 1) high-technology (including medium-high-technology) manufacturing, 2) low-
technology (including medium-low-technology) manufacturing, 3) knowledge-intensive services (KIS),
and 4) less knowledge-intensive services (LKIS).
Knowledge intensity of previous job: Classification of knowledge intensity of previous occupation.
Distinguishes between Knowledge-based and Non-knowledge-based jobs. Knowledge-based occupa-
tions include ISCO-08 groups 1 (managers), 2 (professionals) and 7 (craft and related trades workers).1Table A.1 presented in Appendix A presents details on Eurostat’s classification.
26
We chose these three groups as knowledge-based, once managers have to have a general knowledge
about a business, professionals are usually educated individuals that practice a specified professional
activity, and craft and related trades workers are specialized in particular professional that requires
knowledge to execute it.
Reason for unemployment: The reason for unemployment is divided in three categories — 1) Fired
collectively or individually, 2) Temporary job, and 3) Other. The category other, includes any other reason
such as illness or incapacity, study or training or even early retirement.
Residency location: Location of residency locations in Portugal according to NUTS II. Portugal is
divided in seven subdivisions: 1) North, 2) Algarve, 3) Center, 4) Lisbon, 5) Alentejo, 6) Madeira and 7)
Azores.
Unemployment duration: The duration of unemployment is measured by the number of quarters
during which an individual has been on temporary layoff or without job and searching for one (it is re-
quired to search for a job at least once every four weeks). The unemployment spell is interrupted by any
period of work or exit from the labor force.
3.3 Sample Construction and Characterization
Between 2011 and 2013 a total of 134,956 individuals responded to the LFS, recording 479,326 ob-
servations. However, our analysis considers an exclusive sample. As stated in Subsection 3.1.4 there
is always some inaccuracy associated with responses in a survey. We encountered contradictory re-
sponses especially regarding dates. For example, an individual stating that he lost his job after the date
of the interview. We eliminated observations with incomplete or contradictory information, but in order
to keep the greatest number of observations we rectified some that were possible (see Table 3.1 for
examples of rectifications).
Table 3.1: Examples of rectifications
ID Unemployment date Interview date Occupational status Rectification
1 2011q1 2010q4 Unemployed No rectification — delete2 2012q2 Employed2 2011q4 2012q3 Unemployed Unemployment date — 2012q23 2011q2 2012q2 Unemployed3 2011q1 2012q3 Unemployed Unemployment date — 2011q23 2011q2 2012q4 Unemployed
We limited our sample to include only unemployed individuals between the ages of 15 and 64, and de-
cided to work with all unemployed individuals, including those who enter the study already unemployed.
After eliminating and rectifying contradictory answers and implementing the restrictions presented, the
27
sample was reduced to 25,336 observations from 11,806 individuals.
This section provides an overview of the sub-sample constructed by characterizing the covariates
that according to the literature may affect unemployment duration. Those covariates and their respec-
tive descriptions are presented in Section 3.2. In the last subsection we characterize the Portuguese
population using the last quarter of the original sample.
3.3.1 Summary Statistics
Table 3.2 presents the descriptive statistics at entry of the main covariates used. We use only the last
observations of each individuals to avoid misinterpretation, since we have more than one observation
for the same individual. The sample is composed by 25,336 observations corresponding to 11,806 in-
dividuals, for which the average unemployment duration is 4.98 quarters with a maximum recorded of
17 quarters. The unemployment duration is calculated with the information about when the individual
became unemployed (for those who are already unemployed at the date of the interview). The mean
age for the individuals in our sample is 38 years. Analyzing the age by groups, we find a greater number
of observations between the ages 25 and 44, representing close to half of the population.
Regarding gender, 55% of our sample are men. Over 40% of the sampled individuals are married
and, 89% were born in Portugal. About 41% of the unemployed were fired from their previous job, and
39% of the unemployed in our sample receive unemployment insurance. The residency location is im-
portant for our study regarding the availability of more technological firms and job positions. There is
more people living in North (26%) and Lisbon (19%) than other areas of Portugal. However, there is
more people applying for the same job in more populated areas, thus increasing the competition be-
tween candidates.
By analyzing education we conclude our sample is fairly uneducated in line with the Portuguese pop-
ulation: 66% of people have basic education, while 22% hold a high school diploma, and only 12% have
superior education. Consequently, more people work in non-knowledge-based jobs. More than 80%
of the unemployed previously worked in services compared to 18% that worked in the manufacturing
industry. Portugal has a higher share of service focused companies as opposed to companies in the
manufacturing sector, but it might be possible that manufacturing companies have greater necessity for
workers (keeping workers for longer periods of time). Only 5% worked in high- and medium-high-tech
firms. However, it is to note that the number of people working in less knowledge-intensive services is
similar to the number in knowledge-intensive services.
Finally, about 29% of our sample experienced failure (re-employment) until the end of the observation
time, corresponding to 3,395 individuals. In the following study we will analyze the differences between
those who were re-employed and those who remain unemployed.
28
Table 3.2: Descriptive statistics
Mean Std. Dev.
Unemployment duration (quarters) 4.98 3.29Short-term unemployment 0.54 0.49Long-term unemployment 0.46 0.49
Age (years) 37.99 12.46Age: 15-24 0.17 0.37Age: 25-34 0.25 0.25Age: 35-44 0.24 0.43Age: 45-54 0.22 0.41Age: 55-64 0.12 0.33
Female 0.45 0.50Married 0.43 0.50Born in Portugal 0.89 0.32Unemployment insurance 0.39 0.49Residency location
North 0.26 0.44Algarve 0.14 0.34Center 0.12 0.33Lisbon 0.19 0.39Alentejo 0.12 0.32Madeira 0.08 0.28Azores 0.10 0.30
Reason for separationCollective/individual dismissal 0.43 0.49Temporary job 0.36 0.48Other 0.21 0.40
EducationBasic 0.66 0.47High-school 0.22 0.41College 0.12 0.33
Technology/knowledge intensity of previous sectorLow-tech manufacturing 0.13 0.34High-tech manufacturing 0.05 0.22Less knowledge-intensive services 0.42 0.49Knowledge-intensive services 0.40 0.49
Knowledge intensity of previous job positionKnowledge-based 0.40 0.49Non-knowledge-based 0.60 0.49
Number of observations 25,336Number of individuals 11,806Number of failures 3,395Proportion of failures (%) 28.76
Note: Statistics computed using only the last observation of each individual.
3.3.2 Characterization of the Portuguese Population
In this section, we characterize the last quarter of available data, the fourth quarter (October 1st to De-
cember 31st) of 2013, considering only individuals between the ages of 15 and 64. The last surveyed
quarter counts with a total has a 26,857 observations. Table 3.3 presents the occupational status of
2,943 individuals.
29
Table 3.3: Occupational status
Frequency Percentage
Employed 15,515 57.77Unemployed (looking for first job) 316 1.18Unemployed (looking for new job) 2,627 9.78Student (15 years of age and more) 3,010 11.21Domestic 1,332 4.96Retired 1,892 7.04Another Inactive 2,165 8.06
Total 26,857 100.00
We will perform an analysis of the unemployed population, using the covariates and pairing them
with the technological intensity of previous sector and education. The results regarding employment will
be obtained from the data of previous employment experiences of the unemployed population.
Age
In Table 3.4 we present the education level of the unemployed population by age groups. We conclude
that higher the age, the lower is the number of people with tertiaty education. Meaning that older
people have less educational qualifications. Consequently, we observe a higher percentage of people
with tertiaty education in the age group of 25 to 34 (26.9%) and between the ages 45 and 54 (17.2%). It
should be noted that in the age group between 15 and 24 the number of individuals with post-secondary,
non-tertiary education is higher than the individuals with only below upper secondary education. This
trend results in younger generations with higher educational qualifications.
Table 3.4: Education of the unemployed population by age
Education 15-24 % 25-34 % 35-44 % 45-54 % 55-64 % >64 % Total
Basic 217 40.3 305 45.4 426 60.1 498 78.4 312 81.9 7 87.5 1,765High-school 250 46.5 186 27.7 161 22.7 104 16.4 48 12.6 1 12.5 750College 71 13.2 181 26.9 122 17.2 33 5.2 21 5.5 0 0.0 428
Total 538 100 672 100 709 100 635 100 381 100 8 100 2,943
From the analysis of technology intensity in Table 3.5 we verify a greater number of observations be-
tween the ages of 35 and 44 with 26.5% of the observations, and we also have a considerable amount
of missing data regarding technology intensity (from individuals that did not respond in the LFS). In gen-
eral, we have fewer people working in manufacturing, i.e. technological jobs, with more people working
in less knowledge-intensive services with 63.4% of the total observations. We have a low percentage of
people working in high-technological jobs, however we record a considerable number of people working
in those jobs between the ages of 25 and 34. Though it is a small increment we find a growing number
of people working in high-technological jobs as the age decreases, without considering individuals in the
age group of 15-24 that unassumingly are still studying for higher levels of education. Those who are
30
not studying and are between 15 and 24 years have lower qualifications for high-technological jobs.
Table 3.5: Technology/knowledge intensity of previous firm of the unemployed population by age
Sector 15-24 % 25-34 % 35-44 % 45-54 % 55-64 % >64 % Total
HT 6 2.0 15 2.5 14 2.1 13 2.1 4 1.2 1 14.3 53LT 27 9.2 54 9.0 75 11.2 113 18.6 60 17.2 1 14.3 330KIS 57 19.3 173 28.6 164 24.5 90 14.8 58 16.7 1 14.3 543LKIS 205 69.5 362 59.9 417 62.2 391 64.5 226 64.9 4 57.1 1,605
Total 295 100 604 100 670 100 607 100 348 100 7 100 2,531
Gender
In Table 3.6 we can see a similar number of observations for male and female, the percentage of females
with college and high-school education is superior to the proportion of males in those educational levels.
We have 22% fewer females with basic education that are distributed in higher educational levels. This
does not mean that there are more women working in high-technological jobs as suggested in Table 3.7.
Quite the opposite, we observe a higher number of men working in technological jobs in general, while
women work more in services. 28% of the observations of women are working in knowledge-intensive
services, against 15.5% of the observations for men.
Table 3.6: Education of the unemployed population by gender
Education Male % Female % Total
Basic 1,042 69.2 723 50.3 1,765High-school 335 22.2 415 28.9 750College 130 8.6 298 20.8 428
Total 1,507 100 1,436 100 2,943
Table 3.7: Technology/knowledge intensity of previous employing firm of the unemployed population by gender
Sector Male % Female % Total
HT 34 2.6 19 1.6 53LT 187 14.1 143 11.8 330KIS 205 15.5 338 28.0 543LKIS 897 67.8 708 58.6 1,605
Total 1 323 100 1,208 100 2,531
Education
We compared directly technology intensity to college education level of the unemployed. The data sug-
gests that that is no correlation between the level of education and people working in high-technological
31
firms. In fact, more than half of the people working in those firms have just the minimum level of edu-
cation. These results are contradicting literature, though, we have to bear in mind that we only have 53
observations that were previously in high-tech firms. Most of the individuals that possess college edu-
cation are working in knowledge-intensive services, while people basic education work predominantly
in less knowledge-intensive services. Individuals with high-school education are more distributed be-
tween the different categories of technology intensity of firms, with more than double of people working
in knowledge-intensive services when compared to individuals with basic education. We also see a de-
crease of individuals working in less knowledge-intensive services as the level of education increases,
and the opposite occurs for the number of individuals working in knowledge-intensive services.
Table 3.8: Technology/knowledge intensity of previous firm of the unemployed population by education
Sector Basic % High-school % College % Total
HT 31 2.0 15 2.5 7 2.0 53LT 252 16.0 62 10.1 16 4.6 330KIS 191 12.1 153 25.0 199 57.4 543LKIS 1,099 69.9 381 62.4 125 36.0 1,605
Total 1,573 100 611 100 347 100 2,531
Unemployment duration
From Table 3.9 we notice that there is a distinction between level of education regarding the number of
people unemployed. These figures suggest that each additional level up of college education doubles
the chances of being re-employed: 62.15% of the unemployed have only basic education, 24.15% have
high-school education, and 13.70% have college education. Meaning that people with college education
have higher probabilities of being employed, and re-employed if unemployed.
Table 3.9: Education by unemployment duration
Education 0 1 2 3 4 5 6 7 8 Total
Basic 443 320 300 225 125 70 36 24 30 1,573High-school 228 129 111 59 35 26 9 6 8 611College 158 78 47 24 14 12 7 4 3 347
Total 829 527 458 308 174 108 52 34 41 2,531
32
Table 3.10: Technology/knowledge intensity of previous employing firm of the unemployed population byunemployment duration
Sector 0 1 2 3 4 5 6 7 8 Total
HT 15 14 10 3 4 1 4 1 1 53LT 74 65 72 43 27 24 9 7 9 330KIS 222 133 68 53 24 21 9 6 7 543LKIS 518 315 308 209 119 62 30 20 24 1,605
Total 829 527 458 308 174 108 52 34 41 2,531
33
34
Chapter 4
Methodology
From the literature review in Chapter 2, we concluded that technology intensity and skills or human
capital can much influence the duration of unemployment as well as the job-to-job flow. This study
focuses on the duration of unemployment in Portugal. We now present the methodology used to analyze
individual’s unemployment time controlling for different variables.
4.1 Survival Models
In this section we discuss the econometric models used for the estimation of unemployment duration.
Though duration models originate from the health sciences (where they are usually called survival mod-
els), in more recent times many researchers have applied them to the study of economic phenomena,
of which Lancaster (1979) and Nickell (1979) are two seminal contributions in the field of unemploy-
ment duration. The approach to model the re-employment probabilities is based on the hazard function
(Kiefer, 1988).
For the purpose of our work, we will estimate unemployment duration models to ascertain how it is
influenced by an individual’s human capital and the technological intensity of the previous employment
experience, while controlling for several other variables. This allows us to compare how the role of hu-
man capital varies with different levels of technological intensity. Our approach will be to estimate the
probability of re-employment, accounting for time in unemployment, through maximum likelihood.
Models for the analysis of survival data which have three main characteristics: (1) the dependent
variable is the duration (survival time) until the occurrence of a well-defined event, (2) observations are
censored, in the sense that for some units the event of interest has not occurred at the time the data
are analyzed, and (3) there are explanatory variables whose effect on the duration we wish to assess or
control for.
35
4.1.1 The Survival and Hazard Functions
Survival analysis uses two quantitative terms; the survival function (S(t)) and the hazard function (h(t)).
We use continuous-time proportional hazard models. This choice is justified given the ease of imple-
mentation and interpretation of continuous models. Let T be a non-negative continuous random variable
representing the duration of a specific state, i.e. the time elapsed until the occurrence of the event of
interest (spell length), with a cumulative distribution function defined as follows:
F (t) = Pr(T < t) (4.1)
For the study of unemployment duration, the hazard function allows to estimate the instantaneous
probability of an individual getting re-employed at an instant t, given that he remained unemployed until t
(Portugal and Addison, 2008). Assuming the probability density function f(t) and cumulative distribution
function F (t) of T , the hazard function is given by Equation 4.2.
h(t) = lim∆t→ 0
P (t ≤ T < t+4t | T > t)
4t=
f(t)
1− F (t)=f(t)
S(t)(4.2)
Where S(t) is the survivor function. Note that h(t) gives the conditional density of T given T > t. The
hazard function can be modeled using non-parametric, semi-parametric and parametric approaches,
introduced in the following subsections. F (t) is also called failure function in the survival analysis. The
survivor function is the probability of an individual surviving, in this case staying unemployed, during t,
the elapsed time since the entry state at the beginning of the study.
If the hazard rate is not constant over time, there is duration dependence. Positive duration depen-
dence, (h(t)/∆t > 0), means that the hazard increases with the passage of time: more time in unem-
ployment would increase the chances of re-employment. Negative duration dependence, (h(t)/∆t < 0),
means that the probability of exiting unemployment decreases with time spent in unemployment.
Censoring
Censoring is present when for some subjects the event of interest is not observed during the observation
time. There are two types of censoring: left and right-censoring. Left censoring occurs when the
individual in study achieves the event before it is observed, i.e. the event occurs before the start of
the study. Right censoring means that the duration of the study is not long enough for the event to occur.
This can happen for two reasons: (1) the subject experiences the event after the end of the study or (2)
the subject has withdrawn from the study.
4.1.2 Models
Non-parametric methods, impose no restrictions on the shape of the hazard function and can be used
as a first step to decide between different probability functions (Steiner, 1990; Mussida, 2007). However,
their non-parametric nature does not allow us to understand the effects of the covariates in our analysis.
36
Non-parametric graphical methods are useful to show the data on duration in a preliminary analysis
(Kiefer, 1988). Among the various non-parametric estimation methods that consider censoring we use
the Kaplan-Meier estimator of the survival function for right censored data for a preliminary test of our
hypotheses.
The semi-parametric estimation Cox (1972) proportional hazards model is a popular method for
survival analysis because of its simplicity and straightforward interpretation (Arellano, 2008). The model
is mentioned as semi-parametric since no particular form of the distribution is assumed for the survival
time; it is based on the assumption of proportional hazards, explained later on. Cox (1972) proposed the
using the partial likelihood method, allowing the inference of the coefficients without having to specify
any functional form for the baseline hazard function. Partial likelihood depends only on the ranking of
events (death times), since this determines the risk set at each death time, in contrast with maximum
likelihood that focuses on individuals (spells) (Jenkins, 2005). Consequently, the inference about the
effect of the explanatory variables on the hazard function depends only on the rank order of the survival
times.
For an individual with vector of covariates z = (z1, ..., zp), at a time t, the proportional hazard model
defines the hazard function as:
λ(t; z) = exp(β′z) λ0(t) (4.3)
h(t, z) factors into a function of t and a function of z, in case that two different individuals have
the probability of being re-employed proportional to t. β is a p × 1 vector of regression coefficients
representing the effect of the covariates on survival and λ0(t) represents the base-line hazard function
for an individual for whom is associated a vector z = 0. The baseline hazard does not contain any of the
covariates while the exponential contains the accounts for the covariates, but not for the time (Peters,
2016). As a result, this model can only handle variables that do not change over time. This model is a
proportional hazards model since the hazard function corresponding to two individuals with covariates
z1 and z2 are proportional.
λ(t; z1)
λ(t; z2)= exp{β′(z1 − z2)} (4.4)
This formulation indicates that the covariates have a multiplicative effect on the hazard function, i.e.
the hazard ratio is constant over time. The definition of hazard ratio comes from the division of the pre-
vious equation by λ0(t).
On the other hand, parametric models fit a specified distribution in terms of unknown parameters
to calculate the survival function. They allow for the estimation of the parameters of the covariates but
require the researcher to make assumptions on the shape of the baseline hazard. However, the choice
of the baseline hazard function has to be made under proper conditions, or else the estimates results
may be unreliable and unstable.
37
Though our choice falls on the semi-parametric Cox model, for robustness sake, we also estimated
our models using the Weibull distribution. According to Kiefer (1988) the most appropriate specification
for this kind of studies is the Weibull model. The Weibull distribution is a two-parameter hazard function.
However, the assumption that the baseline hazard of the unemployment duration is not constant is theory
driven. Lancaster (1979) provides evidence that the probability of leaving unemployment decreases
with the length of the spell, and it does not decrease monotonically. Katz and Meyer (1990) show that
the exit rate to employment increases as the unemployment insurance benefits gets closer to expire.
This means that the search activity of the unemployed increases in times like that. We find that Cox
proportional models works best since it does not rely on the restrictions of the baseline hazard.
4.2 Hypotheses
After presenting the discussion of literature in Chapter 2 and presenting the main variables to be con-
trolled for, we now present the hypotheses that we will test in this dissertation. The hypotheses presented
are theory driven and constructed according to the extant literature.
The literature points to a negative relationship between age and the probability of re-employment
(Ciuca and Matei, 2011; Friedberg, 2003; Haile, 2004). Older people stay in unemployment for longer
periods of time, while younger people have higher probabilities of re-employment. The average num-
ber of weeks of unemployment rises proportionally with age (Wolff, 2005). In Figure 4.1 we observe
that older age groups have higher hazards, i.e. the older the individual, the lower the hazard of re-
employment. This leads to our first hypothesis:
1. Older individuals have lower hazards of re-employment.
Regarding gender, there is no general consensus on the role of gender with regards to unemploy-
ment duration. Ciuca and Matei (2011) find higher hazard for men, though the difference is small when
compared to women, while Hernæs and Strøm (1996) find that the probability of re-employment is
higher for women than men. Tansel and Tasci (2010) compared the unemployment duration for men
and women in Turkey, and concluded that the hazard was substantially lower for women. Bowers and
Harkess (1979) finds that in the British labor market, men have shorter expected unemployment dura-
tions than women. A study from Finland by Ollikainen (2003) states that women have a higher risk of
re-employment than men when exiting to employment. Our own non-parametric analysis in Figure ??
shows no relevant difference between men and women, however we adopt the hypotheses of Ciuca and
Matei (2011) study in Romania, for the reason that Portugal and Romania are somewhat similar in terms
of economic development.
2. Men have higher hazards of re-employment.
38
0.00
0.25
0.50
0.75
1.00
Surv
ival
pro
babi
lity,
S(t)
0 5 10 15 20Time in unemployment
15-24 25-34 35-4445-54 55-64
Figure 4.1: Kaplan-Meier survival estimate for age categorized in groups
0.00
0.25
0.50
0.75
1.00
Surv
ival
pro
babi
lity,
S(t)
0 5 10 15 20Time in unemployment
Male Female
Figure 4.2: Kaplan-Meier survival estimate for males and females
There is a wide evidence that unemployment insurance prolongs unemployment duration (Burda and
Sachs, 1988; Katz and Meyer, 1990; Meyer, 1990, 1995; Bover et al., 2002). Katz and Meyer (1990)
39
and Burda and Sachs (1988) find that the more generous the unemployment benefits is, the higher un-
employment rates are and the longer are the spells of unemployment. Bover et al. (2002) points to a
two-fold increase in the hazard rate for recipients of unemployment insurance. Figure 4.3 shows a pos-
itive relationship between survival in unemployment and insurance recipients. In line with the literature,
we advance our third hypothesis:
3. Unemployment insurance recipients have lower hazards of re-employment.
0.00
0.25
0.50
0.75
1.00
Surv
ival
pro
babi
lity,
S(t)
0 5 10 15 20Time in unemployment
No unemployment insurance Unemployment insurance
Figure 4.3: Kaplan-Meier survival estimate for recipients and non-recipients of unemployment insurance
One of the objectives of this dissertation is to estimate the impact of education on the duration of
unemployment spells. Figure 4.4 shows the survival rates for the three levels of education using the
non-parametric estimator Kaplan-Meier. The estimates indicate a positive correlation between the level
of education or the years spent in education and the hazard of re-employment (Nickell, 1979; Kiefer,
1985). Highly educated people receive more job offers and opportunities, hence having more chances
when it comes to being re-employed or staying employed. 12 years of schooling can reduce the ex-
pected duration of unemployment by more than 4%, and qualifications above that level can reduce up to
12% (Nickell, 1979). Our forth hypothesis is thus:
4. More educated individuals have higher hazards of re-employment.
The following two hypotheses are related to previous job experiences and aim to assess how technol-
40
0.00
0.25
0.50
0.75
1.00
Surv
ival
pro
babi
lity,
S(t)
0 5 10 15 20Time in unemployment
Basic High-school College
Figure 4.4: Kaplan-Meier survival estimate for the different levels of education
ogy and knowledge intensity affects the length of unemployment and the relative demand for skills. We
classify firms as Low-technology manufacturing, Less-knowledge intensive services, Knowledge inten-
sive services and High-technology manufacturing. Technology or knowledge intensity of previous em-
ployment is related to education. Highly educated workers will predominantly work in High-technological
jobs and in Knowledge-intensive services, therefore those two categories will have advantage over Low-
technological job and Less-knowledge intensive services in terms of employment. This is in line with
the capital-skill complementarity hypothesis. Capital and skilled labor are relatively more complemen-
tary than are capital and unskilled labor (Griliches, 1969). Technological change demands for more
educated and skilled workers. Beyond that, experience in these more demanding sectors result in the
acquisition of skills that are both more valuable in the labor market, and are in higher demand, which
would suggest that individuals that have such skills will experience shorter unemployment periods:
5. Individuals that worked in technology/knowledge intensive firms have higher hazards of
re-employment.
Literature regarding the impact on unemployment of previous jobs and labor market history is scarce.
This dissertation aims to provide more information and context to allow for more conclusions. Fol-
lowing the same line of reasoning of the previous two hypotheses it is expected that highly educated
workers would perform Knowledge-based tasks and less educated workers the Non-knowledge-based
tasks. People working in Knowledge-based jobs will also acquire more skills (by training) and thus
41
0.00
0.25
0.50
0.75
1.00
Surv
ival
pro
babi
lity,
S(t)
0 5 10 15 20Time in unemployment
LT LKIS KIS HT
Figure 4.5: Kaplan-Meier survival estimate for the different levels of technology/knowledge intensity in firms
achieving greater labor market experience required in more technological jobs. Workers who performed
unskilled manual jobs experience longer periods of unemployment compared with those who had high-
technological jobs or managerial positions (Haile, 2004). However, the Kaplan-Meier survival estimates
shown in Figure 4.6 reveal no sizable difference between Knowledge-based and Non-knowledge-based
jobs. These results are counter intuitive. By estimating and controlling for other covariates we aim to
study the real impact of this variable in the duration of unemployment, and test our last hypothesis:
6. Individuals that performed knowledge-based jobs have higher hazards of re-employment.
42
0.00
0.25
0.50
0.75
1.00
Surv
ival
pro
babi
lity,
S(t)
0 5 10 15 20Time in unemployment
Not knowledge based Knowledge based
Figure 4.6: Kaplan-Meier survival estimate for Non-knowledge-based job and Knowledge based job
43
44
Chapter 5
Results
In this section we present the results of the unemployment duration analysis and discuss the main find-
ings in the light of the literature. To validate our hypotheses, we estimated various models including
different combinations of the different variables of interest, already presented in previous sections. All
model in Table 5.1 control for age, gender, marital status, country of origin, unemployment insurance,
reason for job separation and education, as well as year, quarter and regional effects. We control for
year and quarter in every model to consider hidden exterior information common to every quarter, such
as macroeconomics tendencies, as GDP and changes in unemployment. The coefficients are presented
in the form of hazard ratios (exponential of the coefficients). Hazard ratios enable us to interpret the re-
sults in a simpler and more direct way when compared to the coefficients. From the hazard ratios we
can identify if a variable provides either an advantage or a disadvantage to individuals. If the hazard
ratio is lower than 1 means that the subject has lower risk of getting re-employed and experience longer
unemployment duration. On the other hand, a hazard ratio higher than 1 means increased risk of re-
employment and shorter unemployment duration. A hazard ratio of 0.5 is interpreted as a decrease of
50% in the exit rate from unemployment, when compared to the base level of the covariate.
Every model has 24,221 observations in total, fewer than the number presented in the summary
statistics, since not every individual provided information about the reason of unemployment. We record
a number of 3,326 exits to employment. Our study portrays the Portuguese population. After including
the sampling weights in our models, the population for which we are controlling for is composed by
7,080,283 individuals. This study only considers exits to employment, excluding exits to inactivity. Model
1 shows the impact of education along with the other covariates. Since education is a categorical
variable we can compare the unemployed individuals with only basic education and individuals with
higher levels of education. Models 2 and 3 include all variables in Model 1 plus knowledge intensity of
previous job (in Model 2) or technology/knowledge intensity of previous sector (in Model 3). By including
education and knowledge intensity of previous job or technology/knowledge intensity of previous sector
in the same model we can see how they influence each other. Model 4 includes all the covariates
mentioned above.
45
Table 5.1: Estimation results for the Cox model
Model 1 Model 2 Model 3 Model 4
Age: 25-34 0.909 0.896* 0.885* 0.879*(0.060) (0.059) (0.058) (0.058)
Age: 35-44 0.733*** 0.718*** 0.705*** 0.698***(0.052) (0.051) (0.050) (0.050)
Age: 45-54 0.630*** 0.616*** 0.606*** 0.599***(0.049) (0.048) (0.047) (0.047)
Age: 55-64 0.410*** 0.399*** 0.397*** 0.391***(0.041) (0.040) (0.040) (0.039)
Female 0.953 0.987 0.981 1.002(0.043) (0.046) (0.044) (0.047)
Married 1.219*** 1.215*** 1.222*** 1.219***(0.063) (0.063) (0.063) (0.063)
Born in Portugal 1.171** 1.172** 1.169** 1.171**(0.086) (0.086) (0.086) (0.086)
Unemployment insurance 0.907* 0.904* 0.907* 0.906*(0.047) (0.047) (0.047) (0.047)
Reason for separation: Temporary job 1.363*** 1.371*** 1.328*** 1.335***(0.067) (0.072) (0.070) (0.070)
Reason for separation: Other 1.291** 1.137** 1.139** 1.136**(0.071) (0.067) (0.067) (0.067)
Education: High-school 1.092* 1.086 1.079 1.074(0.057) (0.057) (0.057) (0.057)
Education: College 1.291*** 1.225*** 1.212*** 1.172**(0.086) (0.087) (0.082) (0.084)
Previous job: Knowledge-based 1.124** 1.088*(0.054) (0.056)
Previous sector: Less-knowledge intensive service 1.068 1.098(0.076) (0.080)
Previous sector: Knowledge intensive service 1.298*** 1.298***(0.094) (0.094)
Previous sector: High-tech manufacturing 1.188* 1.219**(0.118) (0.0123)
Observations 24,221 24,221 24,221 24,221Failures 3,326 3,326 3,326 3,326
Hazard ratios, and standard errors (in brackets). All models are estimated with sampling weights and control foryear, quarter and regional effects. The base level of each categorical variable is omitted (age: 15-24; reason forseparation: fired collectively or individually; education: basic; previous sector: low-tech manufacturing; previousjob: non-knowledge-based) — * significant at 10%; ** significant at 5%; *** significant at 1%
5.1 Baseline Hazard
The construction of the baseline hazard form Model 4 (Figure 5.1) allows us to see the duration de-
pendence in unemployment. By looking at the baseline hazards for exiting unemployment we see that
there is negative duration dependence (the likelihood of staying unemployed decreases with time) up to
a peak at the at the analysis time between 10-13 quarters. However, after reaching the peak, the hazard
rates appear to decline sharply. It was not expected to have the highest hazard in the interval between
10 quarters (2.5 years) and 13 quarters (3.25 years). The baseline hazard function follows and inverted
U-shape — increasing hazard in the beginning as individuals use those initial periods for job search-
46
ing or recipient of unemployment insurance fall out of the benefit program, followed up by deceasing
chances of exit that translate into long-term unemployment. A typical discovery when negative duration
dependence is observed is that it is not possible to distinguish whether longer duration spells result in
lower exit rates, or whether there is unobserved heterogeneity leading to low exit rates, remaining in
unemployment for longer. The information contemplated in inquiries, such as the LFS, is often times
insufficient to clearly understand the true effects of unemployment duration.
.000
4.0
006
.000
8.0
01.0
012
.001
4Sm
ooth
ed h
azar
d fu
nctio
n
0 5 10 15Time in unemployment
Figure 5.1: Cox proportional hazard regression
However, the shape of the baseline hazard is very important. Decreasing unemployment exit rates
by duration (holding other characteristics constant) indicate that unemployment has a scarring effect.
Scarring is the negative long-term effect that unemployment has on future labor market possibilities.
Machin and Manning (1999) say that the presence of unemployment benefits that decline with duration
and active labor market policies targeted at long-term unemployed may lead to rising exit rates with
duration.
5.2 Explanatory Variables
The age group estimations show that the older the individual is, the worse are their prospects of exiting
unemployment and thus, the longer the unemployment spell will be. The estimates indicate that the age
group more sensitive to exits to active labor market is the age group between the ages of 15-24. The
results show that the hazard decreases with age: the difference between the age group 15-24 and 25-34
47
groups is not significant but the age group 35-44 has a 30.2% smaller hazard of re-employment (results
from Model 4). The findings regarding the elderly are noteworthy. The oldest age group results show
that people with more than 55 years are less likely to exit unemployment and to get re-employed. The
most probable exit for the latter group would be exiting the labor force. The results, and therefore con-
clusions, from the remaining models are very similar. With the exception that the hazard ratios increase
slightly in every age group by removing previous job and previous sector from the model. These results
support Hypothesis 1, which stated that older people have lower hazards of exiting unemployment.
The variable gender is not significant in any of the models, not allowing us to take a position and be
certain regarding the results obtained. According to the estimation results in Models 1, 2 and 3, men
have a slight advantage when compared to women, meaning that the latter group has smaller hazards
of re-employment. However, the percentage is minimal, and the estimate is not significant. The trend
towards equality between genders means that the Portuguese population is well balanced regarding the
number of men and women working in technological and knowledge-based jobs. The results from the
complete model (Model 4) are not conclusive, and so Hypothesis 2 is not fully validated.
Being married can attenuate the negative psychological effects of unemployment and might increase
the incentives for re-employment. The difference between married and unmarried individuals can be
related to the need to provide for other, meaning that married people will experience an urgency to find a
job, and will accept a job offer more quickly. The results show that married individuals have higher haz-
ards of re-employment when compared to non-married individuals. The covariate married is significant
across all models and vary marginally between models. Married individuals have 21.9% more hazards
of re-employment than unmarried individuals.
Similarly to the previous analyzed covariate, the estimates of the hazard ratios for individuals born
in Portugal remains fairly unchanged across the different models. Born in Portugal is significant at 5%
in all models, and individuals that are born in Portugal have on average 17.1% more chances of being
re-employed than people that are born outside of Portugal. These results show that companies might
prefer to hire people from their own country for reasons such as language.
The effects of the covariate unemployment insurance on the unemployment duration were discussed
in Chapter 2. The position of the various authors is unanimous regarding the effects of unemployment
benefits. As anticipated, unemployment insurance lowers the chances of an individual getting out of un-
employment. This covariate is significant at 10% for all models and holds approximately the same value.
In Portugal, the unemployment insurance can go from 150 days (5 months) to as far as 540 days (18
months), depending on the age of the individual. The data presented show that people entitled to unem-
ployment insurance, will most likely wait until the end of the benefit to accept a job offer. This trend leads
to an increase in the duration of the unemployment spell. Recipients of unemployment benefits have less
9.3%-9.6% chances of getting re-employed, according to our estimates. Our results support the hypoth-
48
esis that individuals with unemployment insurance have lower hazards or re-employment (Hypothesis 3).
We also control for the covariate reason (for unemployment) in every model, where individuals are
compared to those who were fired collectively or individually from their previous job. The values of the
estimates for individuals with other reasons for unemployment show an advantage when compared to
individuals that were fired from their previous job. The estimates are significant at the 1% level for peo-
ple displaced by temporary job and at the 5% level for people with other reasons. The values of the
estimates do not vary much between the different models. Individuals previously working in temporary
job have 33.5% more chances of leaving unemployment (Model 4). This advantage may be because
they already knew that that job would end in a near future, giving them an advantage in the job seeking
activity, and an anticipation factor by searching for jobs earlier while employed. Individuals displaced by
other reasons have 13.6% more hazard of re-employment. This latter group of individuals are usually
displaced in their own terms, i.e. the decision to leave the job is of their own choice. This fact also
provided a timing advantage, over fired individuals. Amongst the people with other reasons for being
unemployed, there are individuals that quit their current job to search for better ones or that already have
better job offers.
Education is one of the most discussed factors for unemployment duration. The extant literature
points to a positive relationship between the level of education and the hazard of re-employment. Hy-
pothesis 4 is supported by the results of our estimates. We can see that the significance of this covariate,
as well as the values of the hazard ratios, decrease by adding more covariates to the model. In Model
1, where we control for education alone, high-school is significant at 10% but is not significant for the
remaining models, suggesting that there is no considerable difference between high-school and basic
education in terms of re-employment once we account for previous job experience we do find a signifi-
cance difference to college education in all models. The decrease in the significance might result from
the correlation with the variables previous job and previous sector, not allowing us to distinguish the
true covariate that is affecting the exit of unemployment and the duration of unemployment. This is the
reason why, in Model 1 we control solely for education. People with high-school education have 9.2%
higher hazards of exiting unemployment, and people with higher education have an advantage of more
29.1%. This means that having university qualifications does increase the exit rate and shortens the
duration of unemployment spells. Hence, validating Hypothesis 4 of our analysis.
The covariates previous job and previous sector carry information about past employment experi-
ences. We now examine whether previous unemployment affects the exit rate in the current spell. This
phenomenon may occur not only because of differentiation against people with unemployment histories,
but also due to the deterioration of human capital and work habits, resulting in lower exit rates to employ-
ment. The results from the previous job estimation is significant at the 5% level in Model 2 and at 10%
in Model 4. In Model 4 the significance decreases further with the addition of previous sector, where
the following sequence occurs: highly educated people perform knowledge-based tasks in high-tech
49
manufacturing or in knowledge intensive services. The covariates college education, previous job and
previous sector are highly correlated. Thus, there is an effect from not accumulating human capital on
the job (while being unemployed), as well as a separate negative impact from being unemployed.
According to Model 4, people performing knowledge-based tasks have 8.8% more chance of getting
re-employed. Our estimate shows that the difference between individuals in low-tech manufacturing and
less-knowledge intensive services is not significant with this combination of variables. A greater and
significant difference in hazard ratio can be observed in the group of people performing knowledge in-
tensive services, with 29.8% more hazard of re-employment. High-tech manufacturing is significant at
the 10% level and provides 21.9% more chances in the exit of unemployment. The increase in the values
of the hazard ratios of less-knowledge intensive services and high-tech manufacturing (by adding the
covariate previous job), from Model 3 to Model 4 may be related with the collinearity between previous
job and previous sector. These results validate Hypotheses 5 and 6.
In general, our estimates confirm Hypothesis 1, which emphasizes the importance of personal char-
acteristics such as age. Confirming the conclusions of Friedberg (2003), Wolff (2005), and Ciuca and
Matei (2011) that the average number of weeks in unemployment rise proportionally with age. The
results of the estimates regarding gender were not significant in the combination of variables in our
study for the unemployment duration. As discussed in the literature, there is not a definite conclusion
on whether men or women have an advantage before the other, in exiting unemployment. In our esti-
mations, men appear to have more hazard of exit from unemployment. Hypothesis 2 is not supported
by our estimates as we find no significant difference. We confirm Hypothesis 3 validating also the find-
ings of Burda and Sachs (1988), Meyer (1990), Meyer (1995), Katz and Meyer (1990) and Bover et al.
(2002). As Mortensen (1970) concluded, unemployment insurance lowers the hazard of re-employment
and individuals only accept a new job if the benefit of it is larger than their reservation wage.
Hypotheses 4, 5 and 6 represent the core research of our analysis and were supported by our
estimates. The impact of human capital and skills on the average unemployment duration is pretty
straightforward. Nickell (1979), Ashenfelter and Ham (1979), Lancaster and Nickell (1980), and Kiefer
(1985) analyzed the impact of human capital in unemployment duration and reached the conclusion that
human capital enhances the chance of an individuals’ re-employment. Higher-educated workers have
higher wages, are more prone to receive other types of complementary training provided by the com-
pany they are working for, and have lower hazards of job separation. Nickell (1979) and Kiefer (1985)
indicate a negative relationship between the level of education and unemployment duration. Griliches
(1969) presents the hypothesis capital-skill complementarity, by saying that capital and skilled labor are
more complementary than capital and unskilled labor. In times characterized by technological change,
the average unemployment duration will rise (Wolff, 2005), since companies will demand more skilled
and educated workers. Technological change lowers the chances of people with less education and
less labor market experience to find new jobs (especially when the previous experience was in low-tech
50
manufacturing or less-knowledge intensive services), as the demand for high qualified and skilled indi-
viduals rises. By analyzing the covariates previous job and previous sector we conclude that evaluating
previous labor market experiences can complement the analysis and enable us to take some interesting
conclusions. We conclude that high technological sectors and more knowledge-based tasks are more
demanded. Hence, unemployed individuals that previously worked in those specific jobs and sectors
are much more likely to be re-employed, especially in times characterized by technological change.
51
52
Chapter 6
Conclusions
The present work analyzed the determinants of unemployment duration in Portugal between the years
2011 and 2013, by using a data-set surveyed every quarter — Inquerito ao Emprego. The aim of this
study was to, first, find which determinants affected unemployment duration and, second, to estimate the
impact of each covariate, by focusing on technology and skills. We analyzed nine determinants (age,
gender, marital status, country of origin, unemployment insurance, reason for separation, education,
knowledge intensity of previous job and technology/knowledge intensity of previous sector) of unem-
ployment duration.
The results of the preceding analysis show that there is no simple explanation for unemployment
duration, and that it cannot be explained solely by traditional supply-demand arguments. An individual
re-employment probability is affected by many variables, variables which can be unobserved. Among
the many determinants, we have personal characteristics, previous labor market experiences, economic
trends. There is also a strong state dependence in the unemployment process. The results of the es-
timates were in line with the extant literature. In Hypothesis 1 and 2 we tested the effect of age and
gender, respectively. We found a negative relationship between age and unemployment duration, i.e.
the probability of re-employment decreases with age, and found that the difference between sexes is not
significant in Portugal, but men have slightly higher hazards than women. Our estimates also confirm
Hypothesis 3. Unemployment insurance recipients spend in average more time unemployed.
Regarding Hypotheses 4, 5 and 6, we find that education, knowledge intensity of previous job and
technology/knowledge intensity of previous firm are related. We find education to be one of the most de-
cisive determinants for the duration of unemployment. People with more years of schooling and higher
levels of education have higher hazard of re-employment. The difference between people with basic
and high-school education is not so significant. By analyzing previous labor market experiences (knowl-
edge intensity of previous job and technology/knowledge intensity of previous firm) we can understand
the impact of technological change. In times of great technological change, the relative demand for
skills change, people working in high-tech and knowledge-based positions are better in terms of em-
53
ployment opportunities. These individuals stay employed for longer periods of time, and unemployed
for shorter periods of time. While unemployment individuals working in low-tech manufacturing and less
knowledge-intensive services have higher rates of displacement due to the technological progress and
automation of manual tasks. These individuals are characterized by low levels of human capital, hence
prolonging their unemployment duration by lower complementarity than capital and skilled and more ed-
ucated labor. The results also point to a negative duration dependence for exit of unemployment, after
a maximum peak of 10-13 quarters, where more time spent in unemployment causes a decrease in the
probability of re-employment, as suggested by the ”scarring” theory of unemployment.
54
Bibliography
Acemoglu, D. (2002). Technical Change, Inequality, and the Labor Market. Journal of Economic Litera-
ture 40(1), 7–72.
Acemoglu, D. and D. Autor (2009). Lectures in Labor Economics.
Acemoglu, D. and D. Autor (2010). Skills, Tasks and Technologies: Implications for Employment and
Earnings.
Addison, J. T. and P. Portugal (2001). Unemployment Duration: Competing and Defective Risks. Journal
of Human Resources XXXVIII(1), 156–191.
Addison, J. T. and P. Teixeira (2001). Technology, Employment and Wages. Review of Labour Economics
and Industrial Relations 15(2), 191–219.
Arellano, M. (2008). Duration Models. Technical report, MIT.
Ashenfelter, O. and J. Ham (1979). Education, Unemployment, and Earnings. Journal of Political Econ-
omy 87 (5), 99–S116.
Autor, D. and D. Dorn (2013). The Growth of Low-Skill Service Jobs and the Polarization of the US Labor
Market. The American Economic Review 103(5), 1553–1597.
Autor, D., L. F. Katz, and M. Kearney (2006). The Polarization of the U. S. Labor Market. The American
Economic Review 96(2), 189–194.
Autor, D., L. F. Katz, and A. Krueger (1998). Computing Inequality: Have Computers Changed the Labor
Market? The Quarterly Journal of Economics 113(4), 1169–1213.
Autor, D., F. Levy, and R. Murnane (2003). The Skill Content of Recent Technological Change: An
Empirical Exploration. The Quarterly Journal of Economics 118(4), 1279–1333.
Azariadis, C. (1976). On the Incidence of Unemployment. The Review of Economic Studies 43(1),
115–125.
Bartel, A. and F. Lichtenberg (1987). The Comparative Advantage of Educated Workers in Implementing
New Technology. The Review of Economics and Statistics 69(1), 1–11.
Becker, G. (1964). Human Capital: A Theoretical and Empirical Analysis with Special Reference to
Education (3rd ed.). Chicago: The National Bureau of Economic Research.
55
Berman, E., J. Bound, and S. Machin (1998). Implications of Skill-Biased Technological Change: Inter-
national Evidence. The Quarterly Journal of Economics 113(4), 1245–1279.
Blanchard, O. and P. Diamond (1994). Ranking, unemployment duration, and wages. The Review of
Economic Studies 61(3), 417–434.
Bound, J. and G. Johnson (1992). Changes in the Structure of Wages in the 1980’s: An Evaluation of
Alternative Explanations.
Bover, O., M. Arellano, and S. Bentolila (2002). Unemployment Duration, Benefit Duration and the
Business Cycle. The Economic Journal 112(479), 223–265.
Bowers, J. K. and D. Harkess (1979). Duration of Unemployment by Age and Sex. Economica 46(183),
239–260.
Burda, M. C. and J. D. Sachs (1988). Assessing High Unemployment in West Germany. The World
Economy 11(4), 445–574.
Caselli, F. (1999). Technological Revolutions. The American Economic Review 89(1), 78–102.
Castro Silva, H. and F. Lima (2017). Technology, Employment and Skills: A look into job duration.
Research Policy 46(8), 1519–1530.
Ciuca, V. and M. Matei (2011). Survival rates in unemployment. International Journal of Mathematical
Models and Methods in Applied Sciences 5(2), 362–370.
Correia, M. J. and F. Lima (2006). O Inquerito ao Emprego: o que e e para que serve? Technical report.
Cox, D. R. (1972). Regression Models and Life-Tables. Journal of the Royal Statistical Society 34(2),
527–541.
Davis, S. J. and J. Haltiwanger (1999). Gross Job Flows. In Handbook of Labor Economics (1th ed.),
Volume 3, Chapter 41, pp. 2711–2805. Elsevier.
Farber, H. (1999). Mobility and stability: The dynamics of job change in labor markets. In Handbook of
Labor Economics, Volume 3, Chapter 37, pp. 2439–2483. Elsevier.
Farber, H. (2003). Job Loss in the United States, 1981-2001.
Frey, C. B. and M. Osborne (2017). The Future of Employment: How Susceptible are Jobs to Computa-
rization? Journal of Technological Forecasting and Social Change 114(1), 254–280.
Friedberg, L. (2003). The Impact of Technological Change on Older Workers : Evidence from Data on
Computer Use. Industrial and Labor Relations Review 56(3), 511–529.
Goldin, C. and L. F. Katz (1996). The Origins of Technology-Skill Complementarity. The Quarterly
Journal of Economics 113(3), 693–732.
56
Goos, M. and A. Manning (2007). Lousy and Lovely Jobs: The Rising Polarization of Work in Britain.
Review of Economics and Statistics 89(1), 118–133.
Goos, M., A. Manning, and A. Salomons (2009). Job Polarization in Europe. The American Economic
Review 99(2), 58–63.
Goos, M., A. Manning, and A. Salomons (2014). Explaining Job Polarization: Routinize-Biased Techno-
logical Change and Offshoring. The American Economic Review 104(8), 1–20.
Griliches, Z. (1969). Capital-Skill Complementarity. The Review of Economics and Statistics 51(4),
465–468.
Haile, G. A. (2004). Re-employment Hazard of Displaced German Workers : Evidence from the GSOEP.
Technical report, Lancaster University Management School, Lancaster.
Hernæs, E. and S. Strøm (1996). Heterogeneity and unemployment duration. Labour 10(2), 269–296.
Hunt, J. (1995). The Effect of Unemployment Compensation on Unemployment Duration in Germany.
Journal of Labor Economics 13(1), 88 – 120.
Jenkins, S. P. (2005). Survival Analysis. Unpublished manuscript, Institute for Social and Economic
Research, University of Essex, Colchester, UK , 121.
Katz, L. F. and D. Autor (1999). Changes in the Wage Structure and Earnings Inequality. Handbook of
Labor Economics 3, 1463–1555.
Katz, L. F. and B. Meyer (1990). The Impact of the Potential Duration of Unemployment Benefits on the
Duration of Unemployment. The Journal of Public Economics 41, 45–72.
Kettunen, J. (1997). Education and Unemployment Duration. Economics of Education Review 16(2),
163–170.
Keynes, J. M. (1930). Essays in Persuasion. In Economic Possibilities for our Grandchildren, pp. 321–
332.
Kiefer, N. (1985). Evidence on the Role of Education in Labor Turnover. Journal of Human Resources,
University of Wisconsin Press 20(3), 445–452.
Kiefer, N. (1988). Economic Duration Data and Hazard Functions. Journal of Economic Literature 26(2),
646–679.
Kodde, D. (1988). Unemployment expectations and human capital formation. The European Economic
Review 32(8), 1645–1660.
Lancaster, T. (1979). Econometric Methods for the Duration of Unemployment. Econometrica - Journal
of the Econometric Society 47 (4), 939–956.
Lancaster, T. and S. Nickell (1980). The Analysis of Re-Employment Probabilities for the Unemployed.
57
Laroche, M., M. Merette, and G. Ruggeri (2017). On the Concept and Dimensions of Human Capital in
a Knowledge-Based Economy Context. Canadian Public Policy 25(1), 87–100.
Machin, S. and A. Manning (1999). The causes and consequences of longterm unemployment in Eu-
rope. Handbook of Labor Economics 3, 3085–3139.
Marx, K. (1867). Capital: A Critique of Political Economy. In T. Frederick Engels, Ernest Untermann,
eds. Samuel Moore, Edward Aveling (Ed.), The Process of Production of Capital (4th ed.), Volume I,
pp. 543. Chicago: Charles H. Kerr and Co.
McConnell, C. R., S. L. Brue, and D. A. Macpherson (2017). Contemporary Labor Economics (11th ed.).
McGraw-Hill Education, 2015.
Meyer, B. (1990). Unemployment Insurance and Unemployment Spells. Econometrica - Journal of the
Econometric Society 58(4), 757–782.
Meyer, B. (1995). Lessons from the U.S. Unemployment Insurance Experiments. Journal of Economic
Literature 33(1), 91–131.
Mincer, J. (1958). Investment in Human Capital and Personal Income Distribution. Journal of Political
Economy 66(4), 281–302.
Mincer, J. (1991). Education and Unemployment. National Bureau of Economic Research (w3838), 35.
Mokyr, J., C. Vickers, and N. Ziebarth (2015). The History of Technological Anxiety and the Future of
Economic Growth: Is This Time Different? Journal of Economic Perspectives 29(3), 31–50.
Mortensen, D. (1970). Job Search, the he Duration of Unemployment, and the Phillips Curve. The
American Economic Review 60(5), 847–862.
Murphy, K. and L. F. Katz (1992). Changes in Relative Wages, 1963-1987: Supply and Demand Factors.
The Quarterly Journal of Economics 107 (1), 35–78.
Mussida, C. (2007). Unemployment Duration and Competing Risks: A Regional Investigation. Ph. D.
thesis, Universita Cattolica del Sacro Cuore.
Nickell, S. (1979). Education and Lifetime Patterns of Unemployment. Journal of Political Econ-
omy 87 (5), 117–131.
Ollikainen, V. (2003). The Determinants of Unemployment Duration by Gender in Finland.
Peters, R. (2016). Analysis of Unemployment Data and Intervention Instruments. Ph. D. thesis, Faculty
of Science, Utrecht University.
Pigou, A. C. (1920). The Economics of Welfare (4th ed.). London: Macmillan and Co., 1932.
Portugal, P. and J. T. Addison (2008). Six Ways to Leave Unemployment. Technical Report 4.
58
Riddell, W. C. and X. Song (2011). The Impact of Education on Unemployment Incidence and Re-
employment Success: Evidence from the U.S. Labour Market. Institute for the Study of Labor (5572),
35.
Schultz, T. (1961). Investment in Human Capital. The American Economic Review 51(1), 1–17.
Schultz, T. (1975). The Value of the Ability to Deal with Disequilibria. Journal of Economic Litera-
ture 13(3), 827–846.
Smith, A. (1776). An Inquiry into the Nature and Causes of the Wealth of Nations, Volume 2.
Spitz-Oener, A. (2006). Technical Change, Job Tasks, and Rising Educational Demands: Looking out-
side the Wage Structure. Journal of Labor Economics 24(2), 235–270.
Steiner, V. (1990). Long-Term Unemployment, Heterogeneity, and State Dependence. Empirica - Astrian
Economic Papers 17 (1), 41–59.
Stigler, G. (1962). Information in the Labor Market. Investment in Human Beings LXX (5), 94–105.
Tansel, A. and H. M. Tasci (2010). Hazard Analysis of Unemployment Duration by Gender in a Develop-
ing Country: The Case of Turkey.
Van Den Berg, G. J. and J. C. van Ours (1994). Unemployment Dynamics and Duration Dependence in
France , the Netherlands and the United Kingdom. The Economic Journal 104(423), 432–443.
Van Den Berg, G. J. and J. C. Van Ours (1998). Duration dependence and heterogeneity in French youth
unemployment durations. Journal of Population Economics 12(2), 273–285.
Violante, G. (2000). Skill-Biased Technical Change. New Palgrave Dictionary of Economics, 1–9.
Vivarelli, M. (2014). Innovation, Employment and Skills in Advanced and Developing Countries: A Survey
of Economic Literature. Journal of Economic Issues 48(1), 123–154.
Vivarelli, M. (2015). Innovation and Employment. IZA World of Labor (154), 1–10.
Wilson, P. (1999). Did the Luddites Get it Right? Automation and the Labour Market. The Student
Economic Review 31(December 1996), 25–31.
Wolff, E. (2005). Computarization and Rising Unemployment Duration. The Eastern Economic Jour-
nal 31(4), 507–537.
59
60
Appendix A
Table A.1: Manufacturing categories by technology intensity. Source: Eurostat (see alsohttp://ec.europa.eu/eurostat/cache/metadata/Annexes/htec_esms_an3.pdf).
Manufactoring Industries Code Description
High-technology 21 Manufacture of basic pharmaceutical products and preparations26 Manufacture of computer, electronic and optical products
Medium-high-technology 20 Manufacture of chemicals and chemical products27-30 Manufacture of electrical equipment, machinery and equipment n.e.c.,
motor vehicles, trailers and semi-trailers and other transport equipment
Medium-low-technology 19 Manufacture of coke and refined petroleum products22-25 Manufacture of rubber and plastic products, other non-metallic mineral
products, basic and fabricated metals, except machinery and equipment33 Repair and installation of machinery and equipment
Low-technology 10-18 Manufacture of food products, beverages, tobacco products, textile, leatherand other products, wood and of products, paper and paper products,printing and reproduction of recorded media
31-32 Manufacture of furniture; other manufacturing
Knowledge-intensiveservices
50-51 Water transport and air transport58-63 Publishing activities; motion picture, video and television programming
production, sound recording and music publish activities; programmingand broadcasting activities; telecommunications; computer programming,consultancy and related activities; Information service activities
64-66 Financial and insurance activities69-75 Legal and accounting activities; activities of head offices, management
consultancy activities; architectural and engineering activities, technicaltesting and analysis; scientific research and development; advertisingand market research; scientific and technical activities; veterinary activities
78 Employment activities80 Security and investigation activities84-93 Public administration and defence, compulsory social security; education,
human health and social work; arts, entertainment and recreation
Less knowledge-intensive services
45-47 Wholesale and retail trade; Repair of motor vehicles and motorcycles49 Land transport and transport via pipelines52-53 Warehousing and support activities for transportation; postal and
courier activities55-56 Accommodation and food service activities68 Real estate activities77 Rental and leasing activities79 Travel agency, tour operator reservation service and related activities81 Services to buildings and landscape activities82 Office administrative and other business support activities94-96 Activities of membership organization; repair of computers and
personal and household goods; other personal service activities97-99 Activities of households as employers of domestic personnel;
undifferentiated goods- and services-producing activities of privatehouseholds for own use; activities of extraterritorial organizations
61