ArticlePDF Available

Abstract and Figures

Most youths in developing countries leave school with only a general academic education level, slowing down their transition to the labour market. We analyse whether work experience during school can ease youth transition to first job in Benin. Using data from the 2014 School-to-Work Transition Survey (SWTS), we estimate a three-equation model to control for endogenous treatment assignment and sample selection and a hazard frailty model. We find that working while studying eases transition from school to first job. However, these findings were significant only for men and youth who left school with at least a secondary education.
This content is subject to copyright. Terms and conditions apply.
RESEARCH ARTICLE
Working while studying: Employment premium or
penalty for youth?
Sènakpon Fidèle Ange Dedehouanou
1
| Luca Tiberti
2
|
Gbodja Hilaire Houeninvo
1
| Djohodo Inès Monwanou
3
1
Department of Economics, Université
d'Abomey Calavi (UAC), Abomey Calavi, Benin
2
Partnership for Economic Policy (PEP),
Université Laval, Québec, Canada
3
Université Nationale d'Agriculture, Kétou,
Benin
Correspondence
Sènakpon Fidèle Ange Dedehouanou,
Department of Economics, Université
d'Abomey Calavi (UAC), Qtier Tchinvié no.
237 Rue 2532 Porto-Novo, Benin.
Email: dsenakpon@gmail.com;
ange.dedehouanou@uac.bj
Funding information
Government of Canada through the
International Development Research Center
(IDRC); Department for International
Development (DFID) of the United Kingdom
(or UK Aid)
Abstract
Most youths in developing countries leave school with only
a general academic education level, slowing down their
transition to the labour market. We analyse whether work
experience during school can ease youth transition to first
job in Benin. Using data from the 2014 School-to-Work
Transition Survey (SWTS), we estimate a three-equation
model to control for endogenous treatment assignment and
sample selection and a hazard frailty model. We find that
working while studying eases transition from school to first
job. However, these findings were significant only for men
and youth who left school with at least a secondary
education.
KEYWORDS
Benin, hazard frailty model, school-to-work transition,
simultaneous equation modelling, sub-Saharan Africa, working
while studying, youth unemployment
JEL CLASSIFICATION
I21, J20, J64
1|INTRODUCTION
In reporting its 2014 and 2015 surveys, Afrobarometer (2015) noted that unemployment was the problem mostly
commonly cited by residents of 36 sub-Saharan countries, which together represent more than three-quarters of
Africa's population. According to the International Labor Office (ILO, 2012), young people are almost three times more
likely to be unemployed than are adults. For students, of particular concern is first entry into the labour market after
leaving school. In fact, African youth experience long periods of transition from school to first job, ranging from
between less than a year to 7 years (Garcia & Fares, 2008; ILO, 2015) and even more than 12years in Togo
Received: 15 July 2020 Revised: 18 October 2021 Accepted: 26 November 2021
DOI: 10.1002/jid.3605
J. Int. Dev. 2021;127. wileyonlinelibrary.com/journal/jid © 2021 John Wiley & Sons, Ltd. 1
(Manacorda et al., 2017). This paper uses data on youth from Benin to study whether working experience (mostly at
the secondary-education level) before leaving school can ease the transition from school to first job. The duration of
youth unemployment is long in Benin: 42.7% of the unemployed have spent over a year without work (INSAE, 2012).
Statistics from the School-to-Work Transition Survey (SWTS) show that only 11.2% of 15- to 29-year-olds have com-
pleted the transition to work (INSAE, 2016) and that their average unemployment spell is 58.5 months.
Most students in developing countries leave school with a general academic education level that is insufficient
to provide the skills requested by the labour market, thus limiting their job opportunities (Garcia & Fares, 2008). This
may explain why governments in many African countries, including Benin, attempt to increase employment opportu-
nities for youth through programmes and policies. The government of Benin has tried to reduce youth unemploy-
ment since 2007 through the National Agency for Promotion of Employment (ANPE) and the National Fund
Enterprise Promotion and Youth Employment (FNPEEJ). Yet, the majority of these limited interventions come post-
schooling, and their impact is not yet clearly known. The SWTS reveals, in fact, that impediments to youth employ-
ment in Benin include a scarcity of vocational and technical education, minimal professional experience, and a lack of
job search assistance (INSAE-BIT, 2013).
The motivation for this study is twofold. First, from an empirical point of view, little is known about that impact
that working while studying (hereafter: work/study) has on the school-to-work transition for youth in Africa
generallyor in Benin in particular. We know of only two published studies, both using 20122013 SWTS data, that
include Benin. Based on 28 countries, Nilsson (2015) provided only descriptive evidence regarding the relationship
between work/study and time to first job following formal schooling. Manacorda et al. (2017), who estimated a haz-
ard model on data from 23 countries, provided empirical evidence of the effect of work/study on the probability of
transition to first job and on the duration of the transition period. Yet neither study used suitable approaches to
address the endogeneity of the variable work/study.In order to understand more about the work/study combina-
tion and transition to work, we examined the 20142015 Benin SWTS data set, adopting empirical methods to deal
with endogeneity issues and to investigate the heterogeneous effects of work/study.
Second, from a policy perspective, understanding whether the work/study combination helps youth enter the
labour market could be useful for policy implementation in Benin as well as in other African countries. As mentioned,
public money is invested in dealing with barriers to youth employment, though these post-schooling interventions
may be limited in their ability to reduce transition time from school to first job. In fact, that transition could be facili-
tated if youth acquired work experience before leaving school. Such experience would allow them to become famil-
iar with the workplace environment, acquire work habits and attitudes, build a professional network, and receive
information related to the labour market.
The estimation of the causal effect of work experiences during schooling on the transition to post-education
work can be complicated by the endogeneity of working while studying decisions. Moreover, the transition spell
between school leaving and the start of post-education job is observed only for school leavers, hence leading to
potential self- selection issues. Using a nonlinear three simultaneous equations econometric model, we controlled
for these endogenous-treatment and sample-selection issues. Specifically, we addressed all these issues simulta-
neously, by implementing a Full Information Maximum Likelihood (FIML) estimator method combined with a control
function approach for endogenous regressors. We corrected for endogeneity of working while studying by
instrumenting it with the intensity of the exposure to educational reforms undertaken by the government of Benin
in primary and secondary schools. Using school reforms as a source of identification for education-related variables
is common in the literature (e.g., Ashraf et al., 2020; Duflo, 2001). To correct for potential self-selection into school
leaving we controlled for marriage before leaving school and the percentage of primary-school-age children in Benin
who were not enrolled in primary or secondary school by the time the youth was in school. We complemented with
the estimation of a hazard model with a frailtyterm which accounts for the unobserved sources of heterogeneity.
We found that work experience while studying (during the schooling year, or during breaks or summer holidays)
is negatively related to the transition-to-work period. Estimates remained robust with either external instruments
alone or with constructed instruments as proposed in Lewbel (2012, 2018a). A number of significant heterogeneous
2DEDEHOUANOU ET AL.
impacts are also found and are related to the gender and the education level of youth. Work/study eased the transi-
tion from school to first job for men and for youth who left school with at least a secondary education.
Most of the literature on the effects of work experience during schooling focuses on developed countries and
looks mainly at long-term post-schooling effects such as wages later in life (Light, 2001). Little attention has been
paid to immediate post-schooling effects on, for example, employment or the duration of unemployment. In general,
the empirical evidence is mixed regarding the impact of in-school work experience on later labour-market outcomes,
whether by education level (Molitor & Leigh, 2005) or by type of schooling (Parent, 2006).
Work experience was found to increase the probability of finding work after graduation for students at a Finnish
university (Häkkinen, 2006), though the study's significant effects seemed to disappear when the author accounted
for the endogeneity of the work/studydecision. A 2016 randomized study of the effect of students' work experi-
ence on future employment in Belgium found no evidence that employers' initial recruitment decisions were affected
by students' work/study experience (Baert et al., 2016).
The type of prior work experience has also received attention in the literature. Using data from a representative
survey of Swiss university graduates, Geel and Backes-Gellner (2012) found that work experience during school led
to shorter job searches after graduation if that prior work experience was related to the field of study.
Robinson (1999) analysed the effects of part-time student work in Australia and showed that students who held
part-time jobs during secondary school experienced shorter periods of unemployment after leaving school; such
part-time jobs may also have helped youth transition to later full-time employment. Robinson's conclusions were
similar to those of Anlezark and Lim (2011), who found that working for 5 h per week during studies had a positive
impact on full-time post-schooling employment in Australia.
Studies on the nexus between in-school work experiences and transition to work are scarce in developing coun-
tries (Nilsson, 2019). Poor quality of labour data and underdeveloped labour-market information systems in many
developing countries have impeded analyses of youth unemployment. Household surveys do not always contain
information on working youth and, therefore, are not ideal for analysing their transition to the labour market.
The SWTS, carried out in more than 30 developing countries since 2012, provides an opportunity to study youth
unemployment, though little is known about the effect of work/study on transitions to first job. In addition to the
works by Nilsson (2015) and Manacorda et al. (2017) cited earlier, other studies using the SWTS have found that lon-
ger post-schooling unemployment lowers the likelihood of getting a job for youth, suggesting that efforts to reduce
this transition-to-work period could be helpful (Atanasovska et al., 2016; Petreski et al., 2017).
2|CONCEPTUAL FRAMEWORK
Studies that explain the mechanisms through which work/study experience may influence the post-schooling
labour-market success of youth have largely relied upon the standard human capital theory, the social network or
social capital theory, or the signalling or screening theory (Geel & Backes-Gellner, 2012). The overall effect is theo-
retically ambiguous, however, and may ultimately depend upon the type of work performed during studies as well as
on local cultural or institutional barriers.
Human capital, in the view of Becker (1964), is valued in the market as a set of acquired experience, skills, atti-
tudes, or knowledge that may later increase workers' productivity. Firms would willingly hire educated youth that
had acquired labour market experience during their studies because they would be more useful than would their
counterparts without prior work experience. The 2012 SWTS, which collected information on entrepreneurs and fac-
tors that influenced hiring, suggested that this could be true in Benin as well. According to SWTS data, although
employers indicated that training received by job seekers was important in the recruitment process, work experience
was the main factor in hiring decisions (INSAE-BIT, 2013).
From the perspective of social network or social capital theory, investment in social networks and personal
relationshipssuch as those acquired through work/study experiencesmay influence labour-market outcomes
DEDEHOUANOU ET AL.3
positively (Mouw, 2003; Seibert et al., 2001). In general practice, a job vacancy is announced first to people working
within a company. In some cases, job openings may be kept from the general public in order to benefit trainees' or
employees' relatives. In both scenarios, previously established social or personal relationships, such as those formed
during work/study experiences, may increase the chances of finding a job because labour market information may
be shared through those networks. The social network mechanism appears to function in Benin as well: SWTS statis-
tics revealed that 51.4% of young employees got their jobs through a friend or a family member and, further, that
open positions were usually advertised first to parents or friends (INSAE-BIT, 2013).
Previous studies have also explained the positive impact of prior work/study experience on labour-market out-
comes as the result of student ability. Work/study may be a signal of unobserved ability for employers who, given
the uncertainty in the labour market, may seek to avoid unnecessary investment in screening. This signalling theory
(Spence, 1973) likely also applies to young individuals in Benin who have had the opportunity to gain work experi-
ence while studying. All three mechanisms are probably stronger in the case of individuals whose work/study experi-
ences have included apprenticeships because apprenticeships strengthen human capital, social networks, and ability
signalling, all of which are more directly related to the needs of the labour market.
It must be noted that prior work/study may have a negative effect on labour-market outcomes as well. Consid-
ering the theory of the allocation of time, a trade-off in the use of time for work vs. study is likely to occur
(Becker, 1965; Buscha et al., 2012). Allocating more time for employment may thus compromise learning and aca-
demic performance and crowd out the positive effect of human capital acquired from work/study. This mechanism
might apply only in the case of working activities during the schooling year, while no negative effects on learning is
expected when students work during breaks or summer holidays.
Reservation wages may also be a mechanism by which work/study experiences influence post-schooling labour-
market success. Such experiences may make the wage expectations of young workers more accurate because they
are based on the characteristics of the local labour market. As such, work/study experience could reduce reservation
wages and have a positive effect on labour-market outcomes after school. If youth put too much weight on gaining
work experience while studying, the reservation wage would increase. In that case, works/study experience would
have a negative effect on later labour-market outcomes by increasing the reservation wage and likely delaying the
school-to-work transition. Thus, the effect of in-school work experience on reservation wages is unclear. In Benin,
the salary expectations of young people are around three times higher than the amount of the Interprofessional
Guaranteed Minimum Wage, according to statistics from the household national survey Enquête Modulaire
Intégrée sur les Conditions de Vie des ménages(EMICoV) for 2014. The reservation wages of youth with in-school
work experience may, therefore, be even higher and that may extend the duration of the transition. The absence of
an unemployment benefit in Benin and the difficulty of finding a job in the formal sector because of competition
from the very dynamic informal sector could, however, decrease reservation wages.
3|EMPIRICAL METHODOLOGY
3.1 |Threats to identification
Two threats to identification must be addressed when estimating the effect of work/study on the transition of youth
from school to first job. First, work/study (our treatment) is likely to be endogenous. Unobserved individual charac-
teristics and/or family background might influence both the likelihood that youth will acquire work/study experience
and their degree of labour-market success after study (Hotz et al., 2002). For example, because of greater ability or
initial skills, more able or motivated youth may be pushed to start working earlier during study. As a result of that
ability, they may also have an easier transition to a first job after leaving school (Geel & Backes-Gellner, 2012).
Second, the transition from school to first job is observed only for youth who left school. A second threat to
identification, then, is the non-random nature of the choice to leave school. The school-leaving decision may, indeed,
4DEDEHOUANOU ET AL.
have been the result of unobserved motivations and preferences that may also have affected labour-market out-
comes (Mussida et al., 2016) or of parents' investment in their children's schooling. Less-motivated students may
have left school earlier, for example, but may also have performed less well in the labour market.
3.2 |Econometric modelling
We measured the transition-to-work period from school to first job as the time span (in months) between the time
respondents left school and when they got a job (i.e., when they left the transition period). Unlike previous studies,
we dealt with both potential endogeneity in our treatment condition (work/study) and sample-selection bias
(because we observed the transition-to-work period only for those who left school). To account for the endogeneity
of work/study (SW) and sample selection from leaving school (LS), we modelled the duration of the transition (T)
within the potential outcome framework and jointly estimated the following multi-equation model:
Ti¼α1WSiþβXiþu1i > 0 outcome equation ð1Þ
LSi¼1, if α2WSiþφZ1iþu2i >0
0, otherwise
selection equation ð2Þ
WSi¼1, if γZ2iþu3i >0
0, otherwise
endogenous treatment equation ð3Þ
where Xis the vector of control variables (reported in Table A1), Z1 and Z2 are the selection and the instruments
variables, respectively (which are discussed later in this section), and α1,α2,β,γ, and φare the parameters to be esti-
mated. The unobserved errors terms are normal with a mean of zero and had the following correlation structure:
corr u1,u2
ðÞ¼ρ12, corr u1,u3
ðÞ¼ρ13, corr u2,u3
ðÞ¼ρ23:
Equations 1 and 3 constitute the main part of the multi-equation model.
1
The model allows for the correlation
between the potential outcomesthe duration of the transitionand unobserved factors affecting the treatment.
The treatment variable work/studyis endogenous if the estimated correlation ρ13 0.
Equation 2 adjusts for the non-random sample selection resulting from school-leaving, given that the duration of
the transition is not observed for youth who were still in school at the time of the survey. Outcome Tis observed if the
selection variable LS is equal to one. Equations 1 and 2 thus form a block of the Heckman selection model
(Heckman, 1976; Lewis, 1974). The selection of being out of school is non-random if the estimated correlation ρ12 0.
Given that the duration of the transition (T) is left censored at zero, we estimated an interval-regression model
incorporating endogenous treatment assignment and non-random sample selection, as presented earlier. The esti-
mated parameter α1is the effect of work/study on the duration of the transition. In the counterfactual modelling
framework (Heckman & Navarro-Lozano, 2004; Imbens & Wooldridge, 2009; Rubin, 1974), α1is also interpreted as
the average treatment effect (ATE) of the treatment variable work/study (WS).
3.3 |Identification strategy
The identification of work/study effect on the duration of the transition relies on two educational reforms under-
taken by the government of Benin in primary and secondary schools (Houedenou, 2016).
2
Using school reforms as
source of identification for education-related variables is common in the literature (e.g., Ashraf et al., 2020;
Duflo, 2001). The first reform, in 2006, made the access to all public primary schools free of charge. The second
DEDEHOUANOU ET AL.5
reform introduced gratuity of education for girls in secondary Cycle I education level, starting in 2010 for girls in the
sixth grade. In 2011, it was extended to girls in the fifth grade, and in 2012 to all girls in secondary Cycle I. Crossing
the years of the reforms, the legal ages of primary and secondary Cycle I education (611 and 1215, respectively),
and the age of the individuals, we could define our instrument as the intensity of the exposure to the reform for each
individual. Youth with a diploma of primary education were potentially exposed to the primary school reform only.
Female students who attended secondary school or higher were potentially exposed to the second reform. Our
instrument is a discrete variable taking value zero if a person was not exposed to any reform at all, one if she/he was
exposed just for 1 year, and so on.
Upon controlling for schooling attainment, we argue that the exposure to the reforms (free education) is likely to
be related to work/study but unlikely to directly affect the duration of the transition (once youth leave the school).
Indeed, school reform is expected to directly affect individuals' human capital while they are at school, including
working while studying decisions and experience, and only indirectly labour market outcomes. Hence, the reform
would affect the transition length only indirectlythat is, through the investment in human capital. Free education
may relax students' non-affordability and then decrease their likelihood of looking for a job during studies to pay
education fees. Nevertheless, students may look for low-paid or unpaid work experiences during studies like intern-
ships or apprenticeships because of free education. These experiences can be particularly beneficial in reducing
unemployment spells once an individual entered the labour market.
Additionally, in the duration of the transition equation, we control for the reasons of leaving school and the level
of education attained by the youth when she left the school. These variables would reasonably capturetogether
with parents' occupationunobserved factors (such as parental engagement during youth's schooling career) which
would positively affect educational attainment and, then, the duration of the transition.
Previous studies in medium and highly developed countries used, as instruments of work/study experiences,
local employment to proxy prevailing labour-market conditions (Häkkinen, 2006; Parikh & Sadoulet, 2005). We did
not have large time-series data on local employment at our disposal. More fundamentally and in contrast to devel-
oped countries, however, the likelihood that youth in Benin would enter the labour market while still studying was
driven more by family conditions than it was by conditions in the labour market. As explained in the descriptive sta-
tistics section, the motivations of youth for work/study were chiefly the desire to earn money or help their families.
Following Lewbel's recommendations (Lewbel, 2018b), we also instrumented WS through constructed instruments
as a robustness check. In particular, as proposed in Lewbel (2012 and 2018a), and relying on the heteroscedasticity of
the error term of the endogenous variable, we constructed instrumental variables as the difference between Xand their
sample average value, then multiplied by u3. As shown in Lewbel (2012), the structural equation can be identified
only under certain hypotheses. In addition to the usual assumptions necessary for valid instruments (under-
identification, weak identification, and overidentifying restrictions), the Lewbel approach also requires that u3be het-
eroscedastic. The validity of this condition was checked through the Breusch-Pagan test (Table A2).
Finally, concerning the estimation of selection Equation 2, it should be noted that, as shown in Wool-
dridge (2010, chapter 15), when (1), (2), and (3) are jointly estimated through full information maximum likelihood, it
is not necessary to include exclusion variables. Yet, estimates are more reliable if one or more covariates in (2) are
not included in (1) and (3). Hence, we included in (2) two different covariates: one indicating whether the youth was
married before leaving school and a macroeconomic-level variable indicating the percentage of primary-school-age
children in Benin who were not enrolled in primary or secondary school by the time the youth was in school. Reason-
ably, both covariates are good determinants of leaving school.
3.4 |Survival modelling as alternative estimation method
For robustness check, we also consider the transition-to-work period from school to first job as a survival outcome.
We estimate a hazard model with a frailtyterm which accounts for the unobserved sources of heterogeneity.
3
6DEDEHOUANOU ET AL.
Survival models are generally used to describe and explain the occurrence and the duration of an event (Cleves
et al., 2016). The duration T of the transition from school to the first job is assumed to be a random variable whose
cumulative distribution function represents the probability that there is an exit from the transition before or at time
t:
Ft,θðÞ¼PTtðÞ,8t0ð4Þ
The probability that the duration of the transition exceeds tis defined as the survival function:
St,θðÞ¼1Ft,θðÞ¼PT>tðÞ ð5Þ
and the instantaneous probability of the transition exit at time t, conditional upon that exit has not yet occurred, is
defined as the hazard or riskfunction:
ht,θðÞ¼lim
Δ!
PtTtþΔjTtðÞ
Δ¼ft,θðÞ
St,θðÞ ð6Þ
ft,θðÞbeing the density function and θa vector of parameters to be estimated. Here, tis the number of months spent
in transition after leaving school until the first job. We estimate the following discrete time proportional hazard
model with the frailtyterm θ
i
:
ht,X,θðÞ¼θ1λ0tðÞExp αWSiþβX1iþδX2it
ðÞ ð7Þ
θ
i
represents the individual-specific random effect that accounts for unobserved sources of heterogeneity and is
assumed to take a multiplicative form. λ
0
(t) is a baseline hazard that summarises the duration dependence in the haz-
ard common to each i. It is the instantaneous risk of exiting the transition when all covariates are zero. It is expressed
either as a logarithmic or a polynomial function of the survival time tper individual-month.
4
WS is work/study, the
variable of interest. X1 and X2 are, respectively, the vectors of fixed and time-varying explanatory variables, and α,β,
and δare coefficient and vectors of coefficients to be estimated. The hazard model (7) is estimated by maximum like-
lihood using a gamma distribution for the unobserved individual heterogeneity (Jenkins, 1995, 1997; Meyer, 1990).
4|DATA AND DESCRIPTIVE STATISTICS
4.1 |Data source
We used data from the School-to-Work Transition Surveys (SWTS) for Benin, carried out between December 2014
and January 2015 by the Institut National de la Statistique et de l'Analyse Economique (INSAE) in collaboration with
the International Labor Organization (ILO) and the MasterCard Foundation in a project entitled Work4Youth.The
20142015 SWTS is a nationally representative sample of individuals 1529 years old. The survey used a six-section
questionnaire to collect rich, detailed information about young individuals, including personal and household demo-
graphic characteristics, formal education/training, employment history, and aspirations.
4.2 |Data summary and definition of variables
The Benin SWTS includes information for 4306 individuals aged 1529 who were interviewed. We removed 1370
of these individuals who had never been in school. Our main equation was run on a sample of 1162 youth who were
DEDEHOUANOU ET AL.7
no longer in school at the time of the survey because the duration of the school-to-work transition was observed for
these individuals only. We accounted for sample-selection issues by additionally considering 1771 youth who were
still in school at the time of the survey.
5
The variables used in this study are defined in Table A1.
The main outcome variable is the transition from school to first job, expressed as the transition period and
defined as the number of months the youth spent in transition between leaving school and first job. The first job is
either salary work or self-employment (we excluded unpaid family workthat is, work for the benefit of the family).
6
Each individual was observed over a defined time interval T; the lower limit is the month and year of leaving school,
and the upper limit corresponds to the month and year in which the respondent started her or his first job or the
month and year of the survey, in cases in which the youth had not left the transition period at the time of the survey.
Work/study is the treatmentvariable of interest. To define this variable, we used the following survey ques-
tion: Have you ever worked while studying (outside apprenticeship)?Answers were either (a) no,(b) yes, during
the school year,(c) yes, outside the school year (summer break, holiday),or (d) yes, during and outside the school
year.The variable work/studywas thus defined as a dummy variable with a value of 1 if the youth was involved
in remunerated jobs while in school and 0 if not. Of the 1162 individuals aged 1529, 17.38% had worked while
studying.
Other variables (mostly time-invariant) that were included in the econometric analysis are defined in Table A1. A
few remarks on the explanatory variables that may help strengthen the identification strategy are worth making here.
For example, the parental occupation is important to capture the income or wealth level of the household, as well as
social networks which significantly ease the transition to work in the context of Benin (as shown by the statistics
reported earlier in Section 2). Also, the current residence of the youth (urban/rural and geopolitical department) at
the time of the survey may have changed from her or his residence at the point at which the transition began. We
thus additionally controlled for whether the youth had always lived in the same community (not moved) because
residence-related variables could reflect social-mobility potentially linked to the transition-to-work period. Other var-
iables were intended to serve as proxies for fixed, unobserved individual characteristics that might also have
explained work/study behaviour (Geel & Backes-Gellner, 2012; Wenz & Yu, 2010). One of these was information
concerning life goals. This variable captured unobserved individual motivations or aspirations that may have
influenced both the decision to choose work/study earlier and post-schooling labour-market behaviour. Finally, we
introduced a categorical variable to identify the reason why the individual stopped studying; this variable stood as a
proxy for specific individual shocks that may have affected school-leaving decisions.
Three macroeconomic variables were included in the econometric analysis. The first was gross domestic product
(GDP) per capita in constant prices. This variable took into account macroeconomic conditions in the country that
may have influenced labour-market behaviour or created financial constraints. The second was the youth unemploy-
ment rate, which was taken to reflect variations in labour-market conditions over time (changes in labour regulations,
for example). Both variables are included in the hazard model as time-variant variables. However, in the multi-
equation model (1)(3), they were averaged for each individual over the transition period. These two variables cap-
tured economic shocks that could have influenced individuals' decisions. The third variable was the percentage of
primary-school-age children who were not enrolled in primary or secondary school; this was used as a selection vari-
able in the multi-equation model and was averaged for each individual over the school-attendance period. These
macroeconomic variables stemmed from the World Bank's annual World Development Indicators database.
4.3 |Descriptive statistics
Table 1 summarizes the transition profile of the 1162 youth who had left school at the time of the survey. The tran-
sition was observed between January 1993 and December 2014. For those who had exited the transition period,
the median age upon entering the transition period was approximately 22; it was 15 for youth who were still in the
transition at the time of the survey. The median age for individuals who had left the transition was 25, and the
8DEDEHOUANOU ET AL.
median transition-to-work period was 1.75 years. This figure is close to those found for (transited) youth in franco-
phone Africa: on average 1 year in Côte d'Ivoire, one-and-a-half years in Burkina Faso, and nearly 5 years in
Cameroon (Garcia & Fares, 2008). The median (unfinished) transition-to-work period for individuals who had not yet
left the transition from school to first job was more than 4 years.
The exit from the transition was also gender-sensitive. Men were more likely (42.01%) to exit the transition
period than were women (38.33%). The fact of being a man may offer more opportunities for work/study, which
allowed men to exit the transition earlier. Cultural and sociological constraints often limit African women's participa-
tion in the labour market, and this is especially true in Benin. The cumulative distribution function of the duration of
the transition period by gender shows that men had a higher probability of exiting earlier (Figure 1). This remained
true through the 150th month, at which point the probability for both sexes was about equal.
Table 2 shows the distribution of youth who worked while studying. We report statistics for those who had
already left school and those still in school at the survey time. A large proportion of youth in our sample were full-
time students. A small percentage of those with work/study experience worked only during the school year. Part-
time work has been reported to have a negative effect on students' academic performance when it is done for long
hours during schooling days (Anlezark & Lim, 2011; Jewell, 2014). Youth in our sample seemed, in general, more
likely to work part-time during summer breaks and holidays, suggesting a reduced impact on academic performance.
The data in Table 2 also indicate no clear differences in whether or not respondents were still in transition or in
the type of work performed by those with work/study experience. Those who performed some work during summer
or holiday breaks alone were relatively better represented among those who had left the transition period and, espe-
cially, among those who had transitioned to salary work. Those with combined work experience during and outside
of school were more prevalent among those who had transitioned to self-employment.
Data from the 2014 SWTS for Benin do not report the characteristics of the work performed by youth while
studying. As is common in Benin, however, that work was likely to be casual or undertaken in small businesses
owned by their families. The motivations of youth to undertake work/study experiences, as recorded in the 2012
SWTS for Benin, were mostly to earn moneyor help familyand less to acquire work experience or consolidate
a resumeor establish contacts for possible future employmentin other words, for financial reasons more than
out of career aspirations, probably due to their living conditions during the study. There can be no doubt, nonethe-
less, that they may still acquire worthwhile skills, of which they may have been unaware, such as management or
other abilities beneficial for attracting future employment.
7
Further descriptive statistics on sociodemographic variables are presented in Table 3. Significant differences
were observed only in some cases. Young people who worked during their studies seemed to have, on average, a
TABLE 1 Transition from school to first job: A summary
Sample of youth that already left school (1162)
Those who exited from the
transition
Those still in the
transition
% of youth 40.19 59.81
Median transition-to-work period (years/months) 1.75/21 4.42/53
Median age of entering in the transition (years) 22.08 15.25
Median age of exiting from the transition (years) 25 -
% that exited into self-employment 23.84 -
% that exited into salary work 16.35 -
% that had immediate transitions (transition length of 0)
among those who transited
21.84 -
Source: Calculations based on 2014 SWTS data.
DEDEHOUANOU ET AL.9
briefer transition period from school to first job compared to those who focused only on their studies. Table 3 also
shows that, on average, individuals who worked during schooling were those who left school with at least a second-
ary education, suggesting that time spent on work while studying may not impede school performance, as discussed
above.
Young men and those whose parents worked in agricultural-related activities seemed more inclined to work
while studying. They engaged in activities often reserved for men and which were likely to be performed at specific
times of the year. Women, in contrast, were often confined to housework, a phenomenon more common in agricul-
tural households and especially in rural areas where a male workforce was more often required.
Individuals with no work/study experience were more likely to receive general academic training. They were also
more likely to leave school for economic reasons or with the aspiration of having a good family life, suggesting that
they may have left school early in order to work full time rather than combine work and study. This was more likely
for those for whom school was unaffordable or who lived in poorer households; these respondents were also more
likely to need money to help their families, as shown by the figures on life goals in Table 3.
5|RESULTS AND DISCUSSION
5.1 |Estimation results from the duration of transition model
Table 4 reports results related to the duration of the transition from school-to-work period (full results are available
in Table A3). Specification A is the standard interval regressionwithout accounting for sample-selection bias and
endogeneity. Specification B corrects for potential non-random selection bias. Specification C (our main specifica-
tion) adds the correction of potential endogeneity in the treatment variable to A and B through external IV.
Specifications D and E conduct sensitivity analysis to check whether the relationship between work/study and
the duration of the transition period are robust to changes of specifications and samples. Specification D restricts
FIGURE 1 Cumulative distribution function of the duration of the transition period by gender [Colour figure can
be viewed at wileyonlinelibrary.com]
10 DEDEHOUANOU ET AL.
the sample to those whose work-search period was less than the 99th percentile (207 months).
8
While, in the con-
text of Benin, it is quite common for people to search quite a long time for a job, those showing an excessively high
job-search period may actually also have fewer incentives to look for work or may stop their searches temporarily or
permanently. Finally, specification E uses constructed IVs and can be seen as a robustness check of specification C.
9
In the multi-equation modelling framework (specifications C, D, and E), pairwise correlations between the error
term of the endogenous treatment equation (Equation 3) and of the duration of transition equation (Equation 1) are
all significant and take the expected sign, indicating the potential existence of treatment endogeneity biases. A sim-
ple joint significance test on the instruments shows that our instruments are jointly statistically significantly corre-
lated to the likelihood of work/study (Equation 3). In addition, as shown in Table A2, we performed additional tests
on the Lewbel's constructed instruments (Lewbel, 2012, 2018a) using a two-stage least square (2SLS) regression
method with a continuously updated GMM estimator (Baum et al., 2007).
10
Our set of constructed instruments
strongly passed the standard test of weakness of instruments, the Hansen J overidentification test and the Breusch-
Pagan test for homoscedasticity of the error term of the endogenous variable, as proposed in Lewbel (2012, 2018a).
The various estimation results shown in Table 4 all indicate that work/study reduces the length of the school-
to-first-job transition period. When potential selection and endogeneity biases are not taken into account (A), work/
study experiences reduce the transition period by 12 months. Accounting for selection bias (B), the impact is slightly
higher (13 months). When we corrected for treatment endogeneity (C), the effect was much larger (roughly
39 months, with respect to the average unemployment period of about 59 months in the overall sample). Ignoring
sample-selection and, in particular, endogeneity issues would underestimate (in absolute terms) the estimated coeffi-
cient of work/study, creating a downward bias. This may mean that people with unobserved academic skills were
TABLE 2 Distribution (%) of youth who ever worked while studying
Sample of youth that already left school
Sample of youth
still in school at the
time of survey
Total
(1162)
Those still
in
transition
(695)
Those not
in
transition
(467)
Those
exited into
salary work
(190)
Those exited
into self-
employment
(277) Total (1771)
(a) Worked during the
school year
3.44 3.31 3.64 3.68 3.61 2.15
(b) Worked outside the
school year (summer
break and holiday)
6.97 6.04 8.35 11.05 6.50 9.15
(c) Worked during and
outside the school year
6.97 5.61 8.99 7.37 10.11 6.38
(d) Worked in any of the
categories above (a
+b+c)
17.38 14,96 20,99 22,11 20,22 17,68
Among (d)
(e) Had additionally an
experience of
internships or
apprenticeships during
study
16.34 14.42 18.37 26.19 12.50 13.38
(f) Not worked 82.62 85.04 79.01 77.89 79.78 82.33
Source: Calculations based on 2014 SWTS data.
DEDEHOUANOU ET AL.11
TABLE 3 Descriptive Statistics
Total
sample
(1162)
Sample of youth who
worked while studying
Sample of youth who did
not work while studying Mean
(202) (960)
ttestVariables Mean Mean Mean
Time-invariant variables
Duration of the
transition (months)
58.52 (52.63) 42.32 (43.38) 62.22 (53.93) ***
Head (of household)
or spouse
0.44 0.41 0.45
Gender: Male 0.48 0.59 0.45 ***
Married before 0.13 0.09 0.14
Have children 0.44 0.38 0.46
Live always area 0.88 0.88 0.88
Secondary educ 0.47 0.57 0.44 **
Domain study 0.85 0.74 0.87 **
Parental education
No schooling 0.46 0.52 0.44
Primary education 0.27 0.17 0.30 ***
At least secondary
education
0.26 0.30 0.25
Milieu: Urban 0.70 0.67 0.71
Age at school-leaving 16.11 (4.82) 17.02 (5.03) 15.91 (4.76)
Reasons to stop study
Drop out 0.29 0.33 0.28
Work/married/
parents/
distance/others
0.20 0.18 0.21
Economic 0.31 0.19 0.34 ***
Graduated 0.19 0.28 0.16 *
Life goal
Professional 0.20 0.29 0.18
Social 0.04 0.05 0.03
Money 0.35 0.41 0.34
Family 0.40 0.26 0.44 ***
Profession of parents
Agricultural 0.25 0.38 0.22 **
Elementary 0.23 0.17 0.24
Other 0.51 0.43 0.52
Macroeconomic variables
Youth unemployment
rate
1.95 (0.24) 1.95 (0.27) 1.95 (0.24)
GDP per capita 350,580 (16,378) 351,514 (17,669) 350,367 (16,098)
12 DEDEHOUANOU ET AL.
less likely to work while studying, but those same skills would have helped them in reducing the duration of the tran-
sition period.
When we restricted our analyses to those whose work-search period was less than the 99th percentile
(207 months), the effect, though reduced (37 months), was still strongly significant (as expected because we
TABLE 3 (Continued)
Total
sample
(1162)
Sample of youth who
worked while studying
Sample of youth who did
not work while studying Mean
(202) (960)
ttestVariables Mean Mean Mean
Children out of school 29.73 (11.93) 28.76 (11.19) 29.95 (12.07)
Note: Standard deviation in brackets for continuous variables.
*p< 0.10. **p< 0.05. ***p< 0.01. Source: Calculations based on 2014 SWTS data.
TABLE 4 Estimation results of the duration of the school-to-first-job transition period
Outcome:
Duration of
transition
(A) Interval
regression:
Eq1
(B) Interval
regression
with
sample
selection:
Eq1
(C) Interval
regression with
endogenous
treatment and
sample selection
(with external IV):
Eq1
(D) Interval regression
with endogenous
treatment, sample
selection (with external
IV and excluding
potential outliers): Eq1
(E) Interval
regression with
endogenous
treatment, sample
selection (with
constructed IV):
Eq1
Work/study 11.642** 12.593*** 38.549** 37.395** 37.484**
corr(e.Eq2,
e.Eq1)
0.511*** 0.560*** 0.617*** 0.555***
corr(e.Eq3,
e.Eq1)
0.361* 0.367* 0.346*
corr(e.Eq2,
e.Eq3)
0.390 0.381 0.420**
Joint
significance
test of all
instruments
(chi2)
14.01 14.27 9.67
Prob > chi2 0.0295 0.0268 0.0463
Observation 1162 2910 2909 2896 2910
Uncensored 1056 1056 1056 1043 1056
Left-censored 106 106 106 106 106
Right-
censored
000 0 0
Selected 1162 1162 1149 1162
Nonselected 1748 1747 1747 1748
Note: All regressions also control for urban/rural and department residency. Eq1 identifies Equation 1 presented in
Section 3.2. Full results, including parameters of Equations 2 and 3, are shown in Table A4.
*p< 0.10. **p< 0.05. ***p< 0.01. Source: Calculations based on 2014 SWTS data.
DEDEHOUANOU ET AL.13
excluded those with extremely long search periods). There was practically no difference when we used constructed
instruments in the endogenous treatment equation. Because our identification came from different sources (external
and constructed IVs) and specifications C to E yield very close estimates, we had sufficient confidence that our esti-
mated effects were reliable. Finally, the coefficients of the other explanatory variables took the expected sign or
were not statistically significant.
5.2 |The hazard of exiting the transition from school to the first job
Table 5 shows the estimation results of the discrete-time proportional hazards model that incorporates unobserved
heterogeneity. We report as well estimations results of the hazard model without unobserved heterogeneity. Both
specifications include a quadratic formulation of the baseline hazard function.
11
Table 5 also displays the estimated
gamma variance, which is the proportion of the random term variance in the total variance from the estimation of
the hazard model incorporating unobserved individual heterogeneity. The value of the gamma variance is 3.214, and
the likelihood ratio test does not reject its significance at 5% level, suggesting the presence of individual unobserved
heterogeneity indeed. This type of heterogeneity generally appears in non-experimental data. Nicoletti and
Rondinelli (2010) show that ignoring that heterogeneity in the duration model could result in biased estimated coeffi-
cients of the explanatory variables.
The signs of the significant estimated coefficients of the explanatory variables are consistent throughout the
estimations' results. In Table 5, we report both the estimated hazard coefficients and the hazard rate.
12
Accounting
for the unobserved individual heterogeneity increases the estimated hazard coefficients in absolute value. The com-
parison between the two specifications in Table 5 suggests that omitting unobserved heterogeneity would bias
downward the estimated hazard coefficient of Work/study.
There is a significant and positive relationship between Work/study and the hazard of exiting the transition
from school to the first job. The instantaneous probability of exiting the transition after leaving school increases
with the experience of work while studying. The value of the hazard rate of work/study indicates that the
expected hazard rate is approximately two times higher for youth who have worked during studies compared
to their counterpart.
5.3 |Heterogeneous effects in the hazard of exiting the transition
We estimated additional regressions to explore whether the above-estimated impact of work/study differed among
youth groups. The regression results of the hazard model with unobserved heterogeneity, presented in Table 6 show
significant heterogeneous impacts of work/study by gender and education level.
There is a significant and positive relationship between work/study and the hazard of exiting the transition from
school to the first job for men. The estimated coefficient of work/study is not significant for women. The hazard rate
of exiting the transition is approximately three times higher for youth men who have worked during studies than
their male counterparts. Youth women who have worked during studies have a hazard rate not statistically different
from their female counterparts. The results suggest either that the types of economic activities performed by men
may have been more favourable to later transition to work or that the local environment viewed men who had
acquired skills in work/study experiences more positively than if women did so. Relatedly, the post-schooling life
goals of women in Benin may differ from those of men. Women, for example, are more likely to look for maternity
and household chores than are men, and local market conditions are more discriminatory against women such that
any acquired skill is less valuable in the labour market.
Regarding the level of education, Table 6 shows that the hazard of quitting the transition increases with work/
study as the youth left school with at least a secondary education level. Previous estimation results above show that
14 DEDEHOUANOU ET AL.
leaving school with at least a secondary education level reduces the transition spell (Table A3). Still, those with that
education level have the riskto exit the transition at a rate 0.59 time less than that of youth who have left the
school with an elementary education level (Table 5). Hence, this suggests that work/study experience, coupled with
TABLE 5 Estimates of hazard of exiting the transition from school to the first job
Model without unobserved
heterogeneity
Model with unobserved
heterogeneity
Hazard
coeff
Hazard
rate
Hazard
coeff
Hazard
rate
Work/study 0.2875* 1.3331* 0.5907** 1.8053**
Age leave school in month (t) 0.0336*** 1.0341*** 0.0578*** 1.0595***
t(spell month identifier, by subject) 0.0289*** 0.9715*** 0.0299*** 0.9706***
tsquared 0.0001*** 1.0001*** 0.0001* 1.0001*
Head or spouse (of household) 0.2393 0.7872 0.0240 1.0243
Gender: Male 0.0907 1.0949 0.0989 0.9059
Have children 0.5575*** 0.5727*** 1.5860*** 0.2047***
Live always area 0.1729 0.8412 0.1036 1.1091
Educ secondary 0.3920*** 0.6757*** 0.5278** 0.5899**
Domain study 0.2054 1.2280 0.2368 0.7891
Life goal (Social) (reference is: Professional) 0.4438 1.5586 0.1419 0.8677
Life goal (Money) 0.2110 1.2349 0.1433 1.1540
Life goal (Family) 0.1734 1.1893 0.0515 0.9498
Father has primary education (reference is: No
education)
0.3214* 0.7251* 0.4753* 0.6217*
Father has at least secondary education 0.2385 0.7878 0.6044* 0.5464*
Mother has primary education (reference is: Education) 0.0081 0.9919 0.0416 0.9593
Mother has at least secondary education 0.2188 0.8035 0.3939 1.4828
Elementary profession of parents (reference is:
Agricultural)
0.3738* 0.6881* 1.0481*** 0.3506***
Other profession of parents 0.0737 0.9290 0.2061 0.8138
Milieu: Urban 0.0094 0.9907 0.4780** 0.6200**
Stop study (Work/married/parents/distance/others) 0.0923 1.0967 0.0066 0.9934
Stop study (Economic) 0.0764 0.9264 0.3840 0.6811
Stop study (graduated) (reference is: Drop out) 0.4677** 1.5963** 0.0544 0.9471
Youth unemployment rate 0.8327*** 2.2995*** 1.3606*** 3.8986***
GDP per capita 0.0000** 1.0000** 0.0000 1.0000
Constant 10.4999*** 0.0000*** 21.7325*** 0.0000***
Variance gamma 3.336 3.214
LR test of gamma var =0 (chibar2) 69.042 48.9434
Prob chibar2 0.000 1.3e-12
Number of unit (individuals) 1162 1162 1162 1162
Observations 65,209 65,209 65,209 65,209
*p< 0.10. **p< 0.05. ***p< 0.01. Source: Authors' calculations using data from SWTS (2014).
DEDEHOUANOU ET AL.15
at least some secondary education, was likely to be more beneficial for early entry into the labour market. Also,
work/study experiences undertaken at a higher education level may be more qualifying and may more closely match
labour demands.
TABLE 6 Hazards estimations of exiting from the transition from school to the first job: heterogenous impacts
By sex (hazard rate) By level of education (hazard rate)
Female Male
At least
secondary Elementary
Working while studying 0.7600 2.9600** 3.0013** 1.0345
Age leave school in month (t) 1.0582*** 1.0598*** 1.0590*** 1.0613***
t(spell month identifier, by subject) 0.9847** 0.9592*** 0.9637*** 0.9735***
tsquared 1.0000 1.0001*** 1.0002* 1.0000
Head or spouse (of household) 1.1167 1.1097 1.1809 1.0484
Have children 0.0919*** 0.4243** 0.2704*** 0.1646***
Live always area 1.2658 0.9534 1.6726 0.7349
Educ secondary 0.5906 0.6371
Domain study 0.7329 0.7826 0.5014 0.8672
Life goal (Social) (reference is: Professional) 0.8104 1.1044 1.0121 0.4509
Life goal (Money) 0.7743 1.3919 1.3906 0.9070
Life goal (Family) 0.8739 0.7992 1.0307 0.7150
Father has primary education 0.8463 0.4394** 0.5820 0.8766
Father has primary education (reference is: No
education)
0.4308* 0.6707 0.6291 0.5804
Father has at least secondary education 0.9782 1.0316 1.1593 0.7176
Mother has primary education (reference is: Education) 2.0630 1.3410 1.5760 0.4306
Mother has at least secondary education 0.4947 0.2444*** 0.3503* 0.3576**
Elementary profession of parents (reference is:
Agricultural)
1.2122 0.6016 1.1575 0.5603
Other profession of parents 0.9982 0.4134** 0.4682* 0.8613
Milieu: Urban 1.2751 0.8085 3.1781** 0.4806*
Stop study (Work/maried/parents/distance/others) 1.2092 0.3421** 2.8308** 0.2504***
Stop study (Economic) 0.9816 0.8964 1.5979 0.7121
unemployment_youth 3.9102*** 4.0078*** 4.0478*** 4.1533***
GDP per capita 1.0000 1.0000 1.0000 1.0000
Gender: Male 1.1127 0.8356
Constant 0.0000*** 0.0000*** 0.0000*** 0.0000***
Variance gamma 2.969 3.149 3.929 2.652
LR test of gamma var =0 (chibar2) 4.417 46.775 28.1647 17.8749
Prob chibar2 0.01779 4.0e12 5.6e-08 0.000012
Number of unit (individuals) 586 576 554 608
Observations 35,911 29,298 19,460 45,749
*p< 0.10. **p< 0.05. ***p< 0.01. Source: Authors' calculations using data from SWTS (2014).
16 DEDEHOUANOU ET AL.
6|CONCLUSIONS AND POLICY IMPLICATIONS
Long school-to-work transition spells represent a big socio-economic issue for African countries that serious and
effective actions should be taken urgently. Several programmes or strategies have been implemented in Benin over
the last decade to deal with youth unemployment. One government strategy is the youth volunteer programme that
provides young people seeking first jobs the chance to learn in public and private businesses after graduation. In this
study, we explored the potential of an alternative, complementary, approach that may smooth the transition of youth
to the labour market.
We analysed the effect of work experience while studying on the ability of youth to transition from school to
first job in Benin. Our analyses focused on adolescents, and most of the in-school work experiences we examined
were jobs performed during summer breaks or holidays. Among various possible (a priori) undetermined effects on
employment, work/study may help youth acquire work experience before they leave school, allowing them to
become familiar with the barriers or impediments to employment that most post-graduation interventions and poli-
cies are already trying to address.
Our multi-equation modellingwhich corrected for treatment endogeneity and sample-selection biasesand the
estimation of the hazard of exiting from the transition add to the scarce literature on the effects of work/studyon
the transition-to-work period. Our results show that work/study is negatively related to the transition-to-work
period and increased the hazard of quitting the transition. Significant heterogeneous impacts were also found: work/
study eased the transition from school to first job for men and for youth who left school with at least a secondary
education level. Unfortunately, however, our data did not allow us to determine the kind of work youth performed
during their studies and, therefore, possible sources of differences between boys and girls.
The results here provide useful information for the implementation of effective employment policies that can
accelerate the transition of young people to their first job at the end of their studies. The results draw attention to
the importance of temporary job experiences for youth during summer breaks or holidays and of expanded school
programmes that include apprenticeships.
The policy implications regarding change or reorientation in existing strategies for dealing with youth unemploy-
ment in Benin are clear. Existing programmes/projects address youth unemployment post-schooling, giving youth
training and skills that are valued by potential employers only after graduation. Job policy interventions need to be
reoriented or extended in order to promote or encourage the engagement of young people in well designed in-
school work experiences. In order to extend the benefits of such programmes to women, additional research is
needed into the type of in-school work boys and girls perform, and interventions must be designed to reduce labour-
market constraints against women (during and after school). Otherwise, interventions may serve only to increase the
gender employment gap.
ACKNOWLEDGEMENTS
The authors are grateful to Abdelkrim Araar and Jorge Davalos for technical support and guidance, to two anony-
mous referees for excellent comments, and to Sessinou Erick A. Dedehouanou, Urielle Judith Tossou, and Mahouli
Mireille-Marie Mintogbe for their feedback on an earlier draft of this paper, as well as to participants to 2017
(Nairobi) and 2018 (Bangalore) PEP Annual Conferences, to 2018 PEGNet Conference in Cotonou, and to 2019
UNU WIDER Conference in Bangkok for their valuable comments and suggestions.
This research was carried out with financial and scientific support from the Partnership for Economic Policy
(PEP) and with funding from the Department for International Development (DFID) of the United Kingdom (or UK
Aid) and the Government of Canada through the International Development Research Center (IDRC).
DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable
request.
DEDEHOUANOU ET AL.17
ORCID
Sènakpon Fidèle Ange Dedehouanou https://orcid.org/0000-0003-2511-5341
ENDNOTES
1
Both equations form what is called, in the impact-evaluation literature, the endogenous treatment-regression model or
the endogenous dummy-variable model (Cameron & Trivedi, 2005; Wooldridge, 2010).
2
Government decree of 21 November 2015.
3
The frailtyterm is commonly interpreted as the impact of (unobserved) omitted variables on the hazard rate.
4
Note that the estimation of the discrete time duration model requires reorganizing the cross-section database into unbal-
anced panel data, using information on the month-year of leaving school (the starting point of the transition) and the
duration of the transition. The panel data make it possible to exploit time varying variables corresponding to the different
months-years in which the youth was at riskof exiting from the transition.
5
Three observations were removed because of missing values and inconsistencies in the data.
6
Studies mostly define the transition period as time elapsed after leaving schooleither upon graduation or upon early exit
without completionuntil the first moment of employment in any job or the first regular job (Fares et al., 2005). The ILO
SWTS applies the definition of the school-to-work transition as the passage of a young person (aged 15 to 29) from the
end of schooling to the first regular or satisfactory job(Elder, 2009). Unfortunately, we were not able to use the defini-
tion of ILO SWT because too few observations (73) exited with a first regular or satisfactory job, compared to our more
flexible definition of exiting as either a salary work or self-employed (467). Yet it is worth noting that all of the youth in
our sample who had exited from the transition, using our definition, did not report any other job until 2014, the time of
the survey. The time elapsed from the first job until 2014 was more than 1 year for about 91% of respondents and more
than 2 years for 78% of them. Hence, we can assume that our definition of first job refers to a fairly stable job.
7
Most of the youth from the SWTS database aspire to succeed professionally.
8
The 99th was an arbitrarily choice of threshold for possible outliers. A lower threshold does not qualitatively change our
results.
9
To check whether youth with zero transition length drove the results, we ran the main specification on a sample that
excludes those with immediate transition and reporting, as a reason for leaving school, failure in exams or no interest for
schooling(i.e., those who are more likely to drop out of schools while they work during schooling. The results, reported
in Table A4, hold, although the magnitude decreases to around 32.
10
To the best of our knowledge, however, there are no suitable tests for IV validity that control for possible selection bias.
11
We got similar estimation results for the specification without unobserved heterogeneity and including a logarithmic for-
mulation for the duration dependence. Convergence issues arose when estimating the model with unobserved heteroge-
neity incorporating the logarithmic form of the baseline hazard.
12
Note that only the signs matter for the estimated hazard coefficients while the hazard rate is interpreted, with respect to
the unity, as the percentage change in the hazard for a one-unit change in the covariates.
REFERENCES
Afrobarometer. (2015). Par Où Commencer? Concilier Les Objectifs de Développement Durable et Les Priorités Populaires
(Dépêche No. 67-17).
Anlezark, A., & Lim, P. (2011). Does combining school and work affect school and post-schooling outcomes? Research report.
National Centre for Vocational Education Research.
Ashraf, N., Bau, N., Nunn, N., & Voena, A. (2020). Bride price and female education. Journal of Political Economy,128(2),
591641. https://doi.org/10.1086/704572
Atanasovska, V., Angjelkovska, T., & Davalos, J. (2016). Unemployment period and vertical skills mismatches: The case of
Macedonia's youth. PEP Working Paper 2016-18.
Baert, S., Rotsaert, O., Verhaest, D., & Omey, E. (2016). Student employment and later labor market success: No evidence
for higher employment chances. Kyklos,69(3), 401425. https://doi.org/10.1111/kykl.12115
Baum, C. F., Schaffer, M. E., & Stillman, S. (2007). Enhanced Routines for Instrumental Variables/Generalized Method of
Moments Estimation and Testing. The Stata Journal,7(4), 465506. https://doi.org/10.1177/1536867X0800700402
Becker, G. S. (1964). Human capital: A theoretical and empirical analysis, with special reference to education. New York:
National Bureau of Economic Research, The University of Chicago Press.
18 DEDEHOUANOU ET AL.
Becker, G. S. (1965). A Theory of the Allocation of Time. Economic Journal,75(299), 493517. https://doi.org/10.2307/
2228949
Buscha, F., Maurel, A., Page, L., & Speckesser, S. (2012). The Effect of Employment while in High School on Educational
Attainment: A Conditional Difference-in-Differences Approach. Oxford Bulletin of Economics and Statistics,74(3),
380396. https://doi.org/10.1111/j.1468-0084.2011.00650.x
Cameron, A. C., & Trivedi, P. K. (2005). Microeconometrics: Methods and applications. Cambridge University Press. https://
doi.org/10.1017/CBO9780511811241
Cleves, M. A., Gould, W. W., & Gutierrez, R. G. (2016). An introduction to survival analysis using Stata (Revised Third ed.).
College Station, Texas: Stata Press.
Duflo, E. (2001). Schooling and labor market consequences of school construction in Indonesia: Evidence from an unusual
policy experiment. American Economic Review,91(4), 795813. https://doi.org/10.1257/aer.91.4.795
Elder, S. (2009). ILO school-to-work transition survey: A methodological guide. International Labor Office.
Fares, J. L., Guarcello, M., Manacorda, F., Rosati, C., Lyon, S., & Valdivia, C. (2005). School-to-work transition in sub-Saharan
Africa: An overview. Understanding Children's Work Working Paper. https://doi.org/10.2139/ssrn.1780265
Garcia, M., & Fares, J. (Eds.). (2008). Youth in Africa's labor market. The World Bank. World Bank Publications.
Geel, R., & Backes-Gellner, U. (2012). Earning while learning: When and how student employment is beneficial. Labor,26(3),
313340. https://doi.org/10.1111/j.1467-9914.2012.00548.x
Häkkinen, I. (2006). Working while enrolled in a university: Does it pay? Labor Economics,13, 167189. https://doi.org/10.
1016/j.labeco.2004.10.003
Heckman, J. (1976). The common structure of statistical models of truncation, sample selection and limited dependent vari-
ables and a simple estimator for such models. Annals of Economic and Social Measurement,5, 475492.
Heckman, J., & Navarro-Lozano, S. (2004). Using matching, instrumental variables, and control functions to estimate eco-
nomic choice models. Review of Economics and Statistics,86,3057. https://doi.org/10.1162/
003465304323023660
Hotz, V. J., Xu, L. C., Tienda, M., & Ahituv, A. (2002). Are there returns to the wages of young men from working while in
school? Review of Economics and Statistics,84, 221236. https://doi.org/10.1162/003465302317411497
Houedenou, F. (2016). Phénomène de déperdition scolaire féminine: analyse et perspectives d'action pour le développement
des compétences au Bénin. J. Rech. Sci. Univ. Lomé (Togo), Série B,18(2), 5767.
ILO. (2012, 25 July). The youth employment crisis: A call for action. Proceedings of the 101st International Labor Confer-
ence. International Labor Office. http://www.ilo.org/ilc/ILCSessions/101stSession/texts-adopted/WCMS_185950/
langen/index.htm
ILO. (2015). Global employment trends for youth 2015: Scaling up investments in decent jobs for youth. International Labor
Office.
Imbens, G. W., & Wooldridge, J. M. (2009). Recent developments in the econometrics of program evaluation. Journal of Eco-
nomic Literature,47,586. https://doi.org/10.1257/jel.47.1.5
INSAE. (2012). Rapport sur Enquête Modulaire Intégrée sur Les Conditions de Vie des Ménages 2ème Edition (Emicov 2011).
Institut National de la Statistique et de l'Analyse
Economique.
INSAE. (2016). Transition de L'école Vers La Vie Active des Jeunes Femmes et Hommes Au Bénin (Rapport Final, 2014). Institut
National De La Statistique et de l'Analyse
Economique.
INSAE and BIT. (2013). Transition de L'école Vers La Vie Active des Jeunes Femmes et Hommes au Bénin (Work4Youth Série De
Publication No. 7). Institut National de la Statistique et de l'Analyse
Economique (INSAE); Bureau International du Tra-
vail (BIT).
Jenkins, S. P. (1995). Easy estimation methods for discrete-time duration models. Oxford Bulletin of Economics and Statistics,
57, 129138. https://doi.org/10.1111/j.1468-0084.1995.tb00031.x
Jenkins, S. P. (1997). Discrete time proportional hazards regression. Stata Technical Bulletin,39,2232.
Jewell, S. (2014). The impact of working while studying on educational and labor market outcomes. Business and Economics
Journal,5, 110.
Lewbel, A. (2012). Using heteroscedasticity to identify and estimate mismeasured and endogenous regressor models. Journal
of Business and Economic Statistics,30,6780. https://doi.org/10.1080/07350015.2012.643126
Lewbel, A. (2018a). Identification and estimation using heteroscedasticity without instruments: The binary endogenous
regressor case. Economics Letters,165,1012. https://doi.org/10.1016/j.econlet.2018.01.003
Lewbel, A. (2018b). The identification zoo-meanings of identification in econometrics. The Journal of Economic Literature.,
57(4), 835903. https://doi.org/10.1257/jel.20181361
Lewis, H. G. (1974). Comments on selectivity biases in wage comparison. Journal of Political Economy,82, 11451155.
https://doi.org/10.1086/260268
Light, A. (2001). In-school work experience and the returns to schooling. Journal of Labor Economics,19,6593. https://doi.
org/10.1086/209980
DEDEHOUANOU ET AL.19
Manacorda, M., Rosati, F. C., Ranzani, M., & Dachille, G. (2017). Pathways from school to work in the developing world. IZA
Journal of Labor and Development,6(1), 1. https://doi.org/10.1186/s40175-016-0067-5
Meyer, B. D. (1990). Unemployment insurance and unemployment spells. Econometrica,58(4), 757782. https://doi.org/10.
2307/2938349
Molitor, C. J., & Leigh, D. E. (2005). In-school work experience and the returns to two-year and four-year colleges. Economics
of Education Review,24, 459468. https://doi.org/10.1016/j.econedurev.2004.09.003
Mouw, T. (2003). Social capital and finding a job: Do contacts matter? American Sociological Review,68(6), 868898. https://
doi.org/10.2307/1519749
Mussida, C., Sciulli, D., & Signorelli, M. (2016). Early school leaving and work outcomes in developing countries. Quaderni del
Dipartimento di Economia, Finanza e Statistica 26/2016. University of Perugia, Department of Economics.
Nicoletti, C., & Rondinelli, C. (2010). The (mis)specification of discrete duration models with unobserved heterogeneity: A
Monte Carlo study. Journal of Econometrics,159(1), 113. https://doi.org/10.1016/j.jeconom.2010.04.003
Nilsson, B. (2015). Does the work-study combination among youth improve the transition path? (Technical Brief No. No.2).
Genève 22. www.ilo.org/w4y
Nilsson, B. (2019). The school-to-work transition in developing countries. The Journal of Development Studies,55(5),
745764. https://doi.org/10.1080/00220388.2018.1475649
Parent, D. (2006). Work while in high school in Canada: Its labor market and educational attainment effects. Canadian Jour-
nal of Economics,39, 11251150. https://doi.org/10.1111/j.1540-5982.2006.00384.x
Parikh, A., & Sadoulet, E. (2005). The effect of parents' occupation on child labor and school attendance in Brazil. Working
paper No. 1000, UC Berkeley.
Petreski, M., Mojsoska-Blazevski, N., & Bergolo, M. (2017). Labor-market scars when youth unemployment is extremely
high: Evidence from Macedonia. Eastern European Economics,55(2), 168196. https://doi.org/10.1080/00128775.
2016.1261631
Robinson, L. (1999). The effects of part-time work on school students. LSAY Research Reports. Longitudinal Surveys of
Australian Youth Research Report No. 9.
Rubin, D. B. (1974). Estimating causal effects of treatments in randomized and nonrandomized studies. Journal of Educational
Psychology,66, 688701. https://doi.org/10.1037/h0037350
Seibert, S. E., Kraimer, M. L., & Liden, R. C. (2001). A social capital theory of career success. The Academy of Management
Journal,44(2), 219237.
Spence, M. (1973). Job market signaling. Quarterly Journal of Economics,87(3), 355374. https://doi.org/10.2307/1882010
Wenz, M., & Yu, W. C. (2010). Term time employment and the academic performance of undergraduates. Journal of Educa-
tion Finance,35(4), 358373. https://doi.org/10.1353/jef.0.0023
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data (2nd ed.). MIT Press.
How to cite this article: Dedehouanou, S. F. A., Tiberti, L., Houeninvo, G. H., & Monwanou, D. I. (2021).
Working while studying: Employment premium or penalty for youth? Journal of International Development,
127. https://doi.org/10.1002/jid.3605
20 DEDEHOUANOU ET AL.
APPENDIX A. ADDITIONAL TABLES
TABLE A1 Description of variables
Variables Definition of variables
Working while studying Has worked while studying =1; 0 otherwise
Head or spouse (of household) Is the head of the household or the spouse of the head =1; 0 otherwise
Gender: Male Is a male =1; 0 otherwise
Married before Is married for the first time before leaving school =1; 0 otherwise
Have children Have one or more living children =1; 0 otherwise
Live always area Has always lived in that commune (not moved) =1; 0 otherwise
Educ secondary Has attained at least the secondary education level =1; 0 otherwise
Age leave School Age (in years) when left school
Domain study Has being student in a general programme =1; 0 otherwise
Father has no schooling Father has no schooling education level =1; 0 otherwise
Father has primary education Father had attained the primary education level =1; 0 otherwise
Father has at least secondary
education
Father had attained at least the secondary education level =1; 0 otherwise
Mother has no schooling Mother has no schooling education level =1; 0 otherwise
Mother has primary education Mother had attained the primary education level =1; 0 otherwise
Mother has at least secondary
education
Mother had attained at least the secondary education level =1; 0 otherwise
Agricultural profession of parents
(reference)
Agriculture and qualified agricultural workers =1; 0 otherwise
Elementary profession of parents Elementary profession =1; 0 otherwise
Other profession of parents Others professions =1; 0 otherwise
Milieu: Urban Resides in an urban area =1; 0 otherwise
Stop study (drop out) (reference) Has interrupted study because of: not pass exam/no interest for school =1; 0
otherwise
Stop study (Work/married/
parents/distance/others)
Has interrupted study because of: work/married/parents/distance/others =1; 0
otherwise
Stop study (Economic) Has interrupted study for economic reason =1; 0 otherwise
Stop study (graduated) Has interrupted study because for graduation =1; 0 otherwise
Life_goal His most important objective in life
Life goal (Professional) (reference) Succeeding professionally =1; 0 otherwise
Life goal (Social) Contributing to society =1; 0 otherwise
Life goal (Money) Earn lots of money =1; 0 otherwise
Life goal (Family) Have a good family life =1; 0 otherwise
Youth unemployment rate (during
transition)
Youth unemployment rate at the national level (from World Development
Indicators database). Averaged on the transition period.
GDP per capita (during transition) Gross domestic product (GDP) per capita in constant prices at the national level
(from World Development Indicators database). Averaged on the transition
period.
Children out of school (during
schooling)
The percentage of primary-school-age children who are not enrolled in primary or
secondary school (from World Development Indicators database). Averaged on
the schooling period.
DEDEHOUANOU ET AL.21
TABLE A2 Weak identification and overidentification tests
Constructed instruments
Weak identification test (CraggDonald Wald Fstatistic) 26.228
Hansen Jstatistic (overidentification test of all instruments) 5.402
p0.141
BreuschPagan/CookWeisberg test for Ho: Constant variance (chi2) 245.79
Prob > chi2 0.000
Observations 1162
Note: Survey weights included. Parameters for all the other variables are not reported.
Source: Calculations based on 2014 SWTS data.
TABLE A3 Estimation results of the duration of the school to first job transition period (showing full results of
Table 4)
Interval regression
(without
corrections) (A)
Interval regression with
correction of sample
selection (B)
Interval regression with Correction of
sample selection and endogenous
treatment (with external IV) (C)
Duration of
transition: Eq1
Duration of
transition:
Eq1
Leave
school:
Eq2
Duration of
transition:
Eq1
Leave
school:
Eq2
Work/
study:
Eq3
Work/study 11.642** 12.593*** 0.218 38.549** 0.891*
Head or spouse (of
household)
4.484 3.743 3.515 0.005
Gender: Male 4.242 4.099 0.107 4.156 0.094 0.122
Have children 26.309*** 22.931*** 23.240***
Live always area 4.329 4.905 0.341* 3.067 0.282 0.403***
Age leave School in
year
6.844*** 6.681*** 0.910*** 6.672*** 0.873***
Educ secondary 9.540*** 9.245*** 9.474***
Domain study 0.330 1.932 0.381*** 5.518 0.469*** 0.637***
Life goal (Social)
(reference is:
Professional)
13.983 14.478* 0.170 14.434* 0.135 0.070
Life goal (Money) 0.763 1.072 0.768*** 1.076 0.730*** 0.071
Life goal (Family) 2.590 0.719 0.472*** 1.230 0.493*** 0.246**
Father has at least
secondary
education
7.665** 7.390* 0.216 7.185* 0.200 0.139
Father has at least
secondary
education
4.105 5.091 0.048 4.323 0.016 0.156
Mother has at least
secondary
education
3.636 4.195 0.070 3.181 0.084 0.094
22 DEDEHOUANOU ET AL.
TABLE A3 (Continued)
Interval regression
(without
corrections) (A)
Interval regression with
correction of sample
selection (B)
Interval regression with Correction of
sample selection and endogenous
treatment (with external IV) (C)
Duration of
transition: Eq1
Duration of
transition:
Eq1
Leave
school:
Eq2
Duration of
transition:
Eq1
Leave
school:
Eq2
Work/
study:
Eq3
Mother has at least
secondary
education
5.557 4.557 0.156 4.510 0.152 0.061
Elementary profession
of parents
10.529** 8.689* 0.028 4.981 0.061 0.625***
Other profession of
parents (reference
is: Agricultural)
2.267 1.726 0.016 0.183 0.023 0.230**
Stop study (Work/
married/parents/
distance/others)
4.694 4.074 4.032
Stop study (Economic) 0.378 0.356 0.501
Stop study (graduated)
(reference is: drop
out)
6.362 5.957 5.736
Youth unemployment
rate
33.780*** 13.481 12.117
GDP per capita (in
000's)
0.307* 0.126 0.167
IV: Exposure to
education reform
for 1 year
0.170
IV: Exposure to
education reform
for 2 years
0.038
IV: Exposure to
education reform
for 3 years
0.410
IV: Exposure to
education reform
for 4 years
0.017
IV: Exposure to
education reform
for 5 years
0.273
IV: Exposure to
education reform
for 6 years
0.660**
Married before 10.009*** 9.900*** 0.455**
Children out of school 0.631*** 0.607***
Constant 210.007*** 102.552 3.023*** 101.864 2.499** 0.289
corr(e.Eq2, e.Eq1) 0.511*** 0.560***
(Continues)
DEDEHOUANOU ET AL.23
TABLE A3 (Continued)
Interval regression
(without
corrections) (A)
Interval regression with
correction of sample
selection (B)
Interval regression with Correction of
sample selection and endogenous
treatment (with external IV) (C)
Duration of
transition: Eq1
Duration of
transition:
Eq1
Leave
school:
Eq2
Duration of
transition:
Eq1
Leave
school:
Eq2
Work/
study:
Eq3
corr(e.Eq3, e.Eq1) 0.361*
corr(e.Eq2, e.Eq3) 0.390
Observation 1162 2910 2909
Uncensored 1056 1056 1056
Left-censored 106 106 106
Right-censored 0 0 0
Selected 1162 1162
Nonselected 1748 1747
Interval regression with endogenous
treatment, sample selection (with external
IV and excluding potential outliers) (D)
Interval regression with endogenous
treatment, sample selection (with
constructed IV) (E)
Duration of
transition:
Eq1
Leave
school:
Eq2
Work/
study:
Eq3
Duration of
transition:
Eq1
Leave
school:
Eq2
Work/
study:
Eq3
Work/study 37.395** 0.880* 37.484** 0.930***
Head or spouse (of
household)
1.443 0.015 3.546 0.006
Gender: Male 2.602 0.102 0.122 4.472 0.087 0.044
Have children 22.366*** 23.467***
Live always area 2.628 0.286 0.401*** 3.276 0.275 0.328***
Age leave School in year 6.159*** 0.846*** 6.670*** 0.873***
Educ secondary 9.788*** 9.187***
Domain study 5.106 0.469*** 0.635*** 5.148 0.470*** 0.651***
Life goal (Social) (reference
is: Professional)
13.835 0.118 0.068 14.491* 0.128 0.127
Life goal (Money) 2.340 0.712*** 0.074 0.820 0.720*** 0.091
Life goal (Family) 2.437 0.482*** 0.247** 0.857 0.492*** 0.245**
Father has at least
secondary education
7.801** 0.203 0.137 7.192* 0.199 0.137
Father has at least
secondary education
4.860 0.011 0.154 4.239 0.013 0.181*
Mother has at least
secondary education
1.483 0.095 0.093 3.320 0.085 0.137
Mother has at least
secondary education
1.737 0.145 0.058 4.757 0.147 0.084
Elementary profession of
parents
4.183 0.059 0.631*** 5.172 0.063 0.649***
Other profession of parents
(reference is:
Agricultural)
0.862 0.022 0.233** 0.206 0.022 0.266***
24 DEDEHOUANOU ET AL.
TABLE A3 (Continued)
Interval regression with endogenous
treatment, sample selection (with external
IV and excluding potential outliers) (D)
Interval regression with endogenous
treatment, sample selection (with
constructed IV) (E)
Duration of
transition:
Eq1
Leave
school:
Eq2
Work/
study:
Eq3
Duration of
transition:
Eq1
Leave
school:
Eq2
Work/
study:
Eq3
Stop study (Work/married/
parents/distance/others)
5.141 4.064
Stop study (Economic) 0.610 0.592
Stop study (graduated)
(reference is: Drop out)
5.549 5.810
Youth unemployment rate 11.185 11.827
GDP per capita (in 000's) 0.214 0.167
IV: Exposure to education
reform for 1 year
0.172
IV: Exposure to education
reform for 2 years
0.036
IV: Exposure to education
reform for 3 years
0.412
IV: Exposure to education
reform for 4 years
0.020
IV: Exposure to education
reform for 5 years
0.281
IV: Exposure to education
reform for 6 years
0.668**
Constructed iv_Sex 0.544***
Constructed
iv_Live_always_area
0.554*
Constructed iv_Life_goal 0.008
Constructed
iv_Occupation_parents
0.129
Married before 9.673*** 0.455** 9.760*** 0.450**
Children out of school 0.591*** 0.607***
Constant 84.967 2.317** 0.289 101.270 2.495*** 0.339
corr(e.Eq2, e.Eq1) 0.617*** 0.555***
corr(e.Eq3, e.Eq1) 0.367* 0.346*
corr(e.Eq2, e.Eq3) 0.381 0.420**
Observation 2896 2910
Uncensored 1043 1056
Left-censored 106 106
Right-censored 0 0
Selected 1149 1162
Nonselected 1747 1748
Note: In all regressions, we control for urban/rural and department residency.
*p< 0.10. **p< 0.05. ***p< 0.01. Source: Authors' calculations using data from SWTS (2014).
DEDEHOUANOU ET AL.25
TABLE A4 Estimation results of the duration of the school to first job transition period (alternatives
specification: removing immediate transitions and with failure exams or no interest in schoolingas a reason for
dropping out)
Interval regression with endogenous treatment, sample
selection (with external IV and excluding potential
outliers)
Duration of
transition: Eq1
Leave school:
Eq2
Work/study:
Eq3
Work/study 32.232** 0.087
Head or spouse (of household) 6.944* 0.036
Gender: Male 4.577 0.015 0.035
Have children 16.610***
Live always area 7.007 0.318 0.396**
Age leave School in year 5.569***
Educ secondary 8.844**
Domain study 2.625 0.006 0.712***
Life goal (Social) (reference is: Professional) 12.825 0.263 0.090
Life goal (Money) 4.077 0.561*** 0.049
Life goal (Family) 2.223 0.559*** 0.279**
Father has at least secondary education 3.715 0.270 0.126
Father has at least secondary education 6.174 0.068 0.205*
Mother has at least secondary education 3.350 0.145 0.137
Mother has at least secondary education 4.432 0.136 0.201
Elementary profession of parents 10.294** 0.075 0.466***
Other profession of parents (reference is: Agricultural) 4.339 0.098 0.102
Stop study (Economic) 3.360
Stop study (graduated) (reference is: Work/married/
parents/distance/others)
2.958
Youth unemployment rate 17.073
GDP per capita (in 000's) 0.217
Exposure to educ reform for 1 year 0.355*
Exposure to educ reform for 2 years 0.035
Exposure to educ reform for 3 years 0.583**
Exposure to educ reform for 4 years 0.009
Exposure to educ reform for 5 years 0.234
Exposure to educ reform for 6 years 0.471
iv_Sex_
iv_Live_always_area
iv_Life_goal
iv_Occupation_parents
Married before 12.102*** 0.386
Children out of school 0.709***
Constant 207.530*** 3.360*** 0.334
26 DEDEHOUANOU ET AL.
TABLE A4 (Continued)
Interval regression with endogenous treatment, sample
selection (with external IV and excluding potential
outliers)
Duration of
transition: Eq1
Leave school:
Eq2
Work/study:
Eq3
corr(e.Eq2, e.Eq1) 0.416***
corr(e.Eq3, e.Eq1) 0.349*
corr(e.Eq2, e.Eq3) 0.044
Observation 2503
Selected 756
Nonselected 1747
Note: In all regressions, we control for urban/rural and department residency.
*p< 0.10. **p< 0.05. ***p< 0.01. Source: Authors' calculations using data from SWTS (2014).
DEDEHOUANOU ET AL.27
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Article
Youth bulges in developing countries may carry both a potential for growth via demographic dividends, and ticking political time bombs, depending on the success of authorities in providing youth with adequate opportunities as they transit into the labour markets of the twenty-first century. In this article I examine the theoretical and empirical research on school-to-work transitions (SWT) in developing countries. After a discussion of the attempts at operationalising the concept of school-to-work transitions from a statistical point of view, I review the theoretical settings suitable for analysing the SWT. Despite an extensive search and matching literature, few models seem adapted to developing countries’ labour markets, and even fewer are empirically tested. I then examine the determinants of transition lengths at the individual and macro level. Findings indicate that education is not always associated with shorter durations to first employment, and that the reasons may be higher expectations, reservation wages, or queuing. Women generally experience longer transitions in the labour market, and evidence from labour market interventions is mixed. Many factors likely to influence the school-to-work transition have not been studied from the point of view of school-to-work transitions, however, and potential directions for future research are presented.
Article
Over two dozen different terms for identification appear in the econometrics literature, including set identification, causal identification, local identification, generic identification, weak identification, identification at infinity, and many more. This survey: (i) gives a new framework unifying existing definitions of point identification; (ii) summarizes and compares the zooful of different terms associated with identification that appear in the literature; and (iii) discusses concepts closely related to identification, such as normalizations and the differences in identification between structural models and causal, reduced form models. ( JEL C01, C20, C50)
Article
We extend our 2003 paper on instrumental variables and generalized method of moments estimation, and we test and describe enhanced routines that address heteroskedasticity- and autocorrelation-consistent standard errors, weak instruments, limited-information maximum likelihood and k-class estimation, tests for endogeneity and Ramsey's regression specification-error test, and autocorrelation tests for instrumental variable estimates and panel-data instrumental variable estimates.