High School Dropouts and Transitory Labor Market Shocks - The Case of The Spanish Housing Boom

HighSchool Dropouts and Transitory Labor
Market Shocks: The Case of the Spanish Housing

Boom
Ainhoa Aparicio
No. 158
October 2010
www.carloalberto.org/working_papers
© 2010 by Ainhoa Aparicio. Any opinions expressed here are those of the authors and not those of the
Collegio Carlo Alberto.
High-School Dropouts and Transitory Labor Market
Shocks: The Case of the Spanish Housing Boom
Ainhoa Aparicio∗†
Collegio Carlo Alberto and IZA

August 17, 2010
Abstract
This paper addresses the implications of transitory changes in labor market con-
ditions for low versus high educated workers on the decision to acquire education.
To identify this effect, I use the improvement in the labor market prospects of low
educated workers motivated by the increases in employment and wages in the con-
struction sector during the recent housing boom. The estimation strategy is based
on the fact that changes in the labor market driven by the construction sector affect
only men. Increases in construction activity are found to increase men’s propensity
to drop out of high-school, relative to women. According to this finding, policies
promoting education should strengthen when in the presence of transitory shocks
in the labor market that make dropping out more attractive.
JEL codes: J24, J22, I20, L74
Keywords: High-school dropout, housing boom, Spain
∗
Collegio Carlo Alberto. Address: Via Real Collegio 30, Moncalieri (TO) 10024, Italy.
Email: [email protected]
†
I would like to thank Libertad González for her invaluable advice and support. This paper has
benefited from comments by José Ignacio Garcı́a Pérez, Kurt Schmidheiny, Núria Rodrguez Planas,
Thijs Van Rens and Julián Messina.
1
1 Introduction
The 2006 European Commission Progress Report on Education Policies states that ”the
high number of [high-school dropouts] is an obstacle to securing access to the knowledge-
based society and greater social cohesion in the European Union”1 . The European Union
established an objective for 2010 of a 10% rate of high-school dropouts, as defined by
persons aged 18 to 24 who have attained at most lower secondary education and who
declare not having received any education or training in the previous four weeks. This
objective was still unfulfilled in 2009, when the proportion of early school leavers in
the EU-27 was more than 15%. In this paper, I show that changes in labor market
conditions in recent years have prevented the achievement of a lower high-school dropout
rate, specially for males.
The desire to enter the labor market is one important motivation for dropping out of
high-school. The recent housing boom has altered the labor market opportunities faced
by youngsters and is therefore likely to be responsible for changes in high-school dropout
trends.
Having a high-school diploma has been shown to reduce unemployment and increase
wages. These differences are unrelated to individual characteristics as argued by Stern,
Paik, Catterali, and Nakata (1989). I show that, by reducing the negative consequences of
not obtaining a high-school diploma, the recent housing boom has induced more young-
sters to drop out of the educational system before getting a high-school diploma. From a
policy perspective, this finding implies that creating the right incentives to get educated
even in the presence of transitory labor market shocks can help to avoid high unemploy-
ment rates at the end of the economic upturn.
The construction sector presents two features that make it appropriate to estimate
the effect of improvements in employment prospects for low educated individuals on the
decision to drop out of school. First, it employs mainly men and second, the proportion
1
Along this paper I will use the notion of early school leavers as defined by the European Union, but
I will refer to early school leavers under the more common denomination of high-school dropouts.
2
of low educated workers is higher than in the rest of the economy. This implies that
the upturn in construction activity will predominantly affect low educated men, and this
will allow us to estimate the impact of better employment opportunities for low educated
workers by comparing male to female high-school dropout probabilities. Additionally, the
magnitude of the increase in construction activity and the differences in the incidence of
the housing boom across regions provides the necessary variation to correctly identify the
effect.
The implications of the housing boom for high-school dropouts are tested using Spanish
data. The case of Spain is of special interest because of the situation of its educational
system as well as the magnitude of the recent housing boom. Spain is the country with
the highest incidence of high-school dropouts in the OECD (see Figure 1). This incidence
has been rising in recent years, reaching 30% in 2009. Additionally, while in the US house
prices increased by 104% between 1997 and 2007, prices in Spain boosted by more than
190% in that same period (see Figure 2). Regarding the intensity of construction activity
in the period, the ratio of construction value added over GDP and the number of new
dwellings experienced a boost of 66% and 90%, respectively (see Figures 3 and 4).
In 2005, more than 29% of high-school students abandoned their studies to enter the
workforce in Spain. Out of all individuals that left high-school to work, more than 23%
of them worked in the construction sector. Disaggregating by gender, 32.61% of all males
that leave high-school to work, do it in the construction sector, while the corresponding
figure for females is negligible2 .
This paper provides new insights to the study of the long-run effects of expansions. The
empirical evidence in this field supports the theory that agents tend to devote less effort to
productivity enhancing activities during expansions because production is relatively more
profitable. See Bean (1990), Gali and Hammour (1992), Saint-Paul (1993), and Malley
and Muscatelli (1999). This paper identifies formal education as one type of productivity
2
These data are obtained from the 2005 Spanish Education and Labor Transitions Survey (Encuesta
de Transición Educativo-Formativa e Inserción Laboral (2005)).
3
enhancing activity that significantly decreases in booms.
Some earlier papers have devoted their attention to the role of labor market conditions
in explaining student drop out decisions. The vast majority of previous studies have
addressed this question by including macroeconomic indicators in the equation modeling
the probability of dropping out.
For the Spanish case, the early paper by Peraita and Pastor (2000) explores the deter-
minants of primary school dropouts focusing on the role played by family background and
economic conditions. They find that the unemployment rate negatively affects the prob-
ability of dropping out of primary school. Also using Spanish data, the influence of labor
market conditions on the demand for post-compulsory education (ages 16 to 18) is studied
by Petrongolo and San Segundo (2002). They find that the youth unemployment rate has
a positive influence on the probability of staying-on while the general unemployment rate
has a negative impact on that probability.
The previous two papers estimate the effect of labor market conditions on dropout
behavior using cross-sectional data. Closer to this paper in terms of methodology is the
study by Rees and Mocan (1997). They use a panel of districts in New York State and
conclude that there is a negative relationship between the overall unemployment rate and
the proportion of high-school students who drop out of school in a given year. Controlling
for unobserved district characteristics is essential to reach this conclusion.
Although many papers argue that labor market conditions are important in education
decisions, this is not a closed question. Using US Census data, Waren and Lee (2003) find
no effect of labor market conditions for individuals aged 16 to 19 years old on high-school
leaving. This is coherent with the conclusion reached by Schady (2004) for the impact
of the 1988-1992 macroeconomic crisis on attendance rates of children in Peru. In con-
trast, Schady’s paper also finds a high and significant impact of the crisis on educational
attainment.
The recent Spanish housing boom was caused by the confluence of financial as well as
demand factors (see Gonzalez and Ortega (2009)) and had a great influence on local labor
4
markets. It improved the labor market prospects of high-school dropouts with respect to
those of more educated workers. Theoretically, the gains from leaving the school system
are represented by the discounted sum of the interactions of probability of employment
and expected wage in each period if one is low educated, minus the discounted sum of the
interactions of probability of employment and expected wage in each period if one is highly
educated3 . In the last decade, both employment opportunities and wages have increased
sharply in the construction sector. This raises the gains from leaving the school system
for men. Figure 5 shows the huge increase in employment in the construction sector.
Figure 6 makes apparent that construction wages have risen more than the average wage
in the economy. Given that construction employs low educated workers intensively as
compared with the overall economy, agents in labor markets where the housing boom was
more pronounced are expected to leave school earlier4 .
The previous hypothesis is tested using individual data from the Spanish Labor Force
Survey in combination with data from the Spanish Regional Accounts and the Spanish
Ministry of Housing. The sample is composed by individuals aged 18 to 24, out of which
34% are high-school dropouts. Men represent 61% of total high-school dropouts. There
is notable variation across regions within Spain. The maximum is found in some Mediter-
ranean regions like Alicante and Almerı́a, where the ratios reach more than 42%, and
the minimum in the Northern regions, i.e., Navarra and Paı́s Vasco, with ratios under
20%. Regarding the evolution over time, the high-school dropout rate has continuously
decreased since 1995 when the high-school dropout rate was around 40%, until it reached
a minimum of 28% in 2000. Since then, it has remained more or less stable until 2008,
when it exceeded 30%.
The empirical correlation between the proportion of high-school dropouts and the
intensity of the housing boom as measured by the ratio between construction and total
3
This argument is made assuming a binary decision of whether to complete high-school. This is a
plausible assumption in our case because Spanish high-school students rarely work.
4
According to the data obtained from the Spanish Labor Force Survey, the average share of high-school
dropouts working in construction for the period 1995-2007 is 73.4%. In contrast, the average share of
high-school dropouts working in the overall economy for that same period is only 52.6%.
5
value added in each province is positive for both males and females. This is represented in
Figure 7. The line that approximates the relation between high-school dropouts and con-
struction activity is however steeper for males (0.55 versus 0.25 of females) and, moreover,
the slope is not significantly different from zero for females. This illustrates the findings
in the main econometric specification, where a differential positive effect of construction
activity on high-school dropouts is found for males relative to females, while the average
effect on all individuals is positive but insignificant.
The causal relationship between high-school dropouts and construction activity in a
regional labor market is estimated by using the fact that changes in construction activity
change employment prospects of men while it leaves those of women practically unaffected.
The evolution of male and female employment in construction is displayed in Figure 8.
Women have always represented a negligible fraction of construction workers.
The results show that variations in construction activity as measured by changes in
construction over total value added in a region make men more likely to drop out of high-
school, while women remain unaffected. This effect is consistent with the estimations using
number of new dwellings as an alternative measure of construction activity. In addition to
this contemporaneous effect, I also find a significant positive effect of construction activity
at the age of 17, i.e., at the time when the decision to drop out of high-school is more
likely to be taken.
I conclude that changes in labor market prospects for low relative to high educated
workers, even if they are transitory in nature, can have significant effects on schooling
decisions.
The remainder of the paper proceeds as follows. Section 2 presents the methodology
used to estimate the effect of the changes in labor market conditions as a result of the
housing boom on high-school dropout rates. Section 3 describes the databases, the vari-
ables and the sample used in the analysis. Section 4 discusses the empirical results and
includes some robustness checks and extensions. Finally, section 5 concludes.
6
2 Methodology
The aim of the empirical exercise is to test whether changes in labor market conditions
for low educated workers change individual decisions to drop out of high-school.
The construction sector is characterized by employing mostly men. In fact, the average
proportion of Spanish women employed in construction from 1980 to 2008 was only 5%. I
take advantage of this feature of the construction sector and identify the effect of variations
in labor market prospects on high-school dropout by exploring the differential effect of
variations in construction activity on males with respect to females. Thus, the explanatory
variable of interest is defined as the interaction between construction activity and a male
dummy.
The probability of high-school dropout is assumed to have a logistic form and is then
estimated by means of the following equation:
exp(βZijt )
yijt =
1 + exp(βZijt )
with
βZijt = β 0 + β 1 Cjt · maleijt + β 2 Cjt + β 3 maleijt + β 4 Xijt + β 5 Wjt + β 6 η j + β 7 ξ t + εijt
where y is an indicator variable equal to one if individual i living in province j at year

t is a high-school dropout and zero otherwise, C measures the intensity of construction
activity, male is a dummy taking value 1 if the individual is a man and 0 if a woman, X
is a vector of individual and family characteristics including age, an indicator for being an
immigrant, a dummy for high-school graduated father, a binary variable for high-school
graduated mother, an indicator for university graduated father, a dummy for university
graduated mother, a binary variable for father working, an indicator for mother working,
a dummy for absent mother, a binary variable for absent father and an indicator for
living without any of the parents, W represents three regional variables, namely, total
value added, youth unemployment rate, and overall unemployment rate, η is a vector of
7
province dummies, and ξ includes year dummies5 . Finally, ε is the residual.
The high-school dropout rate in a province is likely to exhibit time dependence. To
account for Bertrand, Duflo, and Mullainathan (2004)’s critique, standard errors are clus-
tered at the province level6 .
Taking equation 1 as a base, I estimate the contemporaneous effect of construction
activity on high-school dropouts. For those individuals that dropped out of high-school
at an earlier time, the coefficient β 1 is reflecting the influence of construction activity on
the probability of reinsertion to the education system. However, many individuals may
have actually taken their decisions to drop out of high-school at the average age at which
students are in high-school. To include this possibility, I estimate the differential effect
of construction activity when the individual was 17 years old for males with respect to
females. Therefore, the key explanatory variable in this specification is the interaction
of construction activity at the age of 17 and a male dummy. The estimated equation
remains as follows:
exp(βZijt )
yijt =
1 + exp(βZijt )
with
17 17
βZijt = β 0 +β 1 Cjt ·maleijt +β 2 Cjt +β 3 Cjt +β 4 maleijt +β 5 Xijt +β 6 Wjt17 +β 7 η j +β 8 ξ t +εijt
where C 17 reflects the construction activity measure when the individual i was 17 years
old, W 17 contains total value added in the province at the age of 17, youth unemployment
rate, and overall unemployment rate. The rest of variables are as defined above7 .
5
The information on parents’ education and employment status is only available for cohabiting fathers
and mothers. Hence, the dummies for absent father and absent mother serve to account for the missing
observations.
6
The paper by Bertrand, Duflo, and Mullainathan (2004) calls our attention to the fact that esti-
mated standard errors are inconsistent when outcomes are serially correlated and the estimation is done
employing difference-in-differences techniques.
7
Initially, I planned to include contemporaneous total value added in the equation but it has proven
to be highly collinear with total value added at 17 and hence, I have excluded it. This should not be
a concern because the high correlation existing between total value added and total value added at 17
assures that I am controlling adequately for the impact of economic activity on high-school dropout.
Moreover, results are statistically invariant to the inclusion of contemporaneous total value added.
8
As in the equation for the contemporaneous effect, standard errors are clustered at
the province level.
3 Data and descriptive statistics
3.1 Data sources and sample definition
The data used in the estimations contains individual data from the Spanish Labor Force
Survey as well as province level data from the Spanish Regional Accounts and the Spanish
Ministry of Housing.
The Spanish Labor Force Survey has been collected quarterly since 1976. In practice,
more than 60000 households containing around 180000 individuals are surveyed in each
wave. The information contained in the survey includes individual characteristics such as
age, gender and level of education, details about working status and family characteristics.
The population of interest is formed by working age individuals, i.e., individuals aged 16
to 65 years old.
The sample selected for the analysis is composed by individuals aged 18 to 24 years
old. These individuals constitute the set of potential high-school dropouts as defined by
the European Union. Only the waves corresponding to the fourth quarter of each year are
included. I have chosen the fourth quarter because the decision to drop out of high-school
will be observed at the beginning of the academic year which starts in September and then
new dropouts will only be observable in the fourth quarter of the year. The time period
covered comprises the years from 1995 to 2007. This is determined by the availability of
the data on construction.
The Spanish Regional Accounting data follows the methodology established by the
1995 European System of National and Regional Accounts. The data contains informa-
tion on GDP decomposed by economic sectors defined as agriculture, energy, industry,
construction and services. The information is geographically disaggregated up to the
province level. The data has yearly frequency and is available for the period including
9
the years 1995 to 2007.
The Spanish Ministry of Housing collects administrative data on dwelling acquisition,
sole prices and dwelling construction. In the analysis, the number of new dwellings will be
used8 . This data is obtained from a registry containing architect’s construction permis-
sions. Hence, the reported information is very accurate and comprises the whole universe
of dwellings in Spain. The data is aggregated at the province and year level and covers
the period 1991 to 2008.
3.2 Variable definition and descriptive statistics
The variable high-school dropout is defined according to the European Union criteria and,
consequently, it is a binary indicator equal to one if an individual fulfills the following two
conditions: (i) her level of education is inferior to high-school graduate and (ii) declared
not to have followed any official education in the preceding four weeks, and zero otherwise.
The explanatory variable of interest is intensity of construction activity. This is mea-
sured as construction over total value added in the province of residence during the year
when the individual is surveyed. This measure is interacted with a male dummy in order
to address the differential effect of construction activity on males compared to females.
When exploring the effect of construction activity at the age of 17 on high-school
dropouts, the value assigned to each individual corresponds to the measure of construction
activity at the time she was 17 years old. This restricts the time period included in the
sample because the information on construction activity at 17 is not available for all
individuals in the years 1995 to 2001. To avoid sample selectivity issues, only the years
for which this information is available for all individuals (2002 to 2007) are included in
the analysis.
The descriptive statistics for the variables included in the main regression are displayed
8
The number of new dwellings is a good proxy of expectations about employment in the construction
sector. In fact, Figure 4 shows that the number of new dwellings measure captures better the end of the
housing boom than the number of constructed dwellings measure.
10
in Table 1. The final sample is composed by 230635 individuals. One third of them are
high-school dropouts and around one half of the sampled individuals are male. They are
21 years old on average.
Regarding family characteristics, 11% of sampled individuals have a cohabiting fa-
ther with high-school education and the proportion of individuals who have a cohabiting
mother with high-school education is only one percentage point lower. Only 9% of sam-
pled individuals have a university graduated cohabiting father and the proportion is 2
percentage points lower for university graduated cohabiting mothers. Additionally, 59%
of individuals have a cohabiting father who is working while only 30% of them have a
cohabiting mother who is working. Finally, 22% of sampled individuals do not live with
their fathers. The corresponding proportion for individuals not living with their mothers
is reduced to 17% and individuals not living with any of their parents are only 5%.
Average construction over total value added assigned to individuals in the sample is
around 10% and the average number of constructed dwellings per individual aged 18 to
24 is 0.034.
The reported figures are similar to the ones for the sample included in the regression
estimating the effect of construction activity at age 17. These are displayed in Table 2.
The number of individuals is reduced to 86953. The only remarkable differences in terms
of descriptive statistics are that the high-school dropout rate is slightly lower, cohabit-
ing parents are more educated on average, the proportion of sampled individuals with
working cohabiting father is hardly reduced, the proportion of individuals with working
cohabiting mother is a little higher, and the proportion of absent fathers and mothers as
well as emancipated youngsters is higher as compared to the descriptive statistics for the
sample included when estimating the contemporaneous effect. These differences are intu-
itive because the period is restricted to recent years and educational attainment, women
participation in the labor market and youngsters emancipation have been continuously
increasing in time.
Table 3 contains the descriptive statistics for the sample included in the regressions
11
where construction activity at 17 is measured using number of new dwellings at 17. The
period included is shorter and more recent on average than the period in the contempo-
raneous effect regressions but longer and less recent on average than the period in the
regression where construction activity is measured using construction value added at 17.
Therefore, the number of observations, 143440, as well as the mean values for high-school
dropout rate, parents’ education attainment, mother’s working status and emancipated
are in between those in the two tables described above.
4 Empirical results
The objective of the empirical exercise is to test whether changes in labor market condi-
tions that improve labor market perspectives for low educated workers make more likely
that individuals dropout of high-school.
The identification strategy makes use of the fact that men experience better employ-
ment opportunities as a result of an increase in construction activity while women do
not.
In practice, I focus on the sign of the coefficient associated to the interaction of the
measure of construction activity and a male dummy in the equation for the probability
of being a high-school dropout. A positive and significant value of this coefficient is
interpreted as transitory improvements in labor market prospects for low educated workers
inducing a rise in the probability of high-school dropout.
4.1 Contemporaneous effect
In this section, I discuss the estimated differential effect of changes in regional construction
activity over time on high-school dropout probability for males versus females.
Table 4 displays the results for the estimation in which the probability of dropping
out of high-school is assumed to have a logistic functional form.
The impact of construction activity on male versus female high-school dropout prob-
12
ability is positive and significant. In contrast, the average effect of construction activity
is positive but statistically indistinguishable from zero. This corroborates the hypothesis
that labor market shocks that improve employment prospects for low educated workers
significantly increases the likelihood that individuals dropout of high-school. The esti-
mated coefficient associated to the variable construction activity interacted with male in
the logistic equation is 2.899. This is interpreted as a one percentage point increase in
construction activity as measured by the share of total value added that is produced by
the construction sector inducing an increment in the probability of high-school dropout
for the average individual of 0.006 (1.74%).
One could argue that the intensity of construction activity may affect the high-school
dropout rate not through labor market conditions but through affordability of housing.
The latter channel could operate such that when construction activity increases, housing
prices decrease and more individuals dropout of high-school to start working and buy a
house. However, in our data, construction activity is positively correlated with housing
prices. Hence, if the previous critique was applicable, more intense construction activity
together with higher prices would induce less dropouts. However, the estimation results
show that men dropout more in the presence of higher housing prices. This is counterin-
tuitive given that it is male working status what usually determines household formation.
Hence, our identification strategy is assuring that the estimated effect is not a consequence
of changes in affordability of housing.
Another source of concern could be that the recent housing boom is contemporaneous
to an important economic upturn and, therefore, the estimated effect could be a con-
sequence of men being differently affected by the economic cycle. The results reached
by Yamashita (2008) when exploring the effects of the Great Depression on educational
attainment suggest that the average effect of the economic cycle is negligible. However,
I still address the possibility that the general level of economic activity has an effect on
high-school dropouts by including regional total value added as a control in the estimated
equations. Any persisting bias is expected to be negative because women’s labor supply
13
is more affected by the economic cycle9 . This indicates that, if the economic cycle was
influencing our results, the actual effect would be higher in magnitude than the estimated
one.
Additionally, I discuss the potential existence of reverse causality, i.e., variations in
high-school dropouts causing changes in construction activity. Our results could be ex-
plained by reverse causality if increments in the high-school dropout rate in the province
reduce construction wages and therefore increase construction activity by increasing its
profitability. However, as it was previously noted, construction wages rose more than
wages in other sectors during the period of study and therefore if the labor supply of
high-school dropouts has some influence on construction activity through wages, our co-
efficient of interest would be downward biased. Additionally, the literature on the causes
of the housing boom (see Glaeser, Gyourko, and Saiz (2008) for a recent example) indi-
cates that labor supply does not play a role in explaining the expansion of the construction
sector in recent years and then, I expect no impact of reverse causality in the estimations.
In the empirical analysis I define the relevant labor market for an individual at the
province level. Given the low geographical mobility of the Spanish population, this is
considered a reasonable assumption10 . However, if some individuals dropout of high-school
to become construction workers in a different province where the demand of construction
workers is higher, the actual impact of the housing boom on high-school dropouts is more
pronounced than the estimated by defining the labor market at the province level.
Regarding the rest of coefficients in the logistic regression, males are more likely to
drop out of high-school. The influence of age could be represented using an inverted U-
shape with a maximum in 21 years old. Immigrants are significantly more likely to drop
out. Individuals are less likely to drop out if their parents are more educated and if their
parents are working. This is coherent with the findings of Maani and Kalb (2007) that
9
In unreported regressions, I have included the interaction of total value added and a male dummy as
an additional control variable. The associated coefficient is negative and insignificant. This constitutes
additional evidence that the effect of the economic cycle, if there is any, is stronger on women.
10
The low mobility of the sampled individuals is illustrated by the fact that more than 87.6% of sampled
individuals live in their province of birth.
14
economic resources play a significant role in school leaving decisions. Finally, those living
without their fathers are less likely to drop out of high-school while living without their
mothers or on one’s own are associated to a higher probability of being a high-school
dropout. The positive coefficient associated to the youth unemployment rate and the
negative coefficient for the general unemployment rate are coherent with the findings of
Petrongolo and San Segundo (2002). However, in our regressions the coefficient for the
overall unemployment rate is not significant. This is due to the correlation between the
unemployment rate and total value added.
4.2 Effect of construction activity at the age of 17
The estimation results for the impact of construction activity at age 17 on the probability
of high-school dropout are displayed in Table 5. The differential impact of construction
activity at 17 on males with respect to females is positive and significant. The point
estimate is 1.904. This can be interpreted as a one percentage point increase in construc-
tion over total value added inducing an increase of 0.003 (1.17%) in the probability of
dropping out of high-school for the average individual. This effect is consistent with the
contemporaneous one in the sign, but it is smaller in magnitude. Also in consonance with
the contemporaneous effect results, the average effect of construction activity at 17 on
both males and females is statistically indistinguishable from zero while total value added
at 17 has a positive and significant impact. In contrast, contemporaneous construction
activity is positive and significant. This differs from the result obtained for the coefficient
associated to this variable in the regression for the contemporaneous effect. The reason
is that it contains the average and the differential contemporaneous effect and the latter
is significant.
The estimated coefficients for the rest of controls are extremely similar to the ones
obtained in the contemporaneous regression.
15
4.3 Construction activity measured by the number of new dwellings
In order to test the consistency of the results under different measures of construction ac-
tivity, the analysis is repeated substituting construction over total value added by number
of new dwellings over population aged 18-24 in the province. The estimation results are
displayed in Table 6 for the contemporaneous effect. The point estimate for the differential
effect of construction activity as measured by number of new dwellings over population
aged 18 to 24 on males versus females is 3.875. This is equivalent to a marginal effect
of 0.77 when the covariates are evaluated at their mean values. This implies that one
extra new dwelling per one hundred individuals aged 18-24 increases the probability of
high-school dropout by 0.02 (7.03%) on average. Similarly to the results when construc-
tion activity is measured by value added, the average effect of construction activity as
measured by number of new dwellings is negligible. Total value added is also positive
but the estimate becomes insignificant. The rest of coefficients have extremely similar
magnitudes and significance levels to the ones reported for the contemporaneous effect of
construction activity as measured by construction value added.
The results for the estimation of the effect of construction activity at 17 as measured by
number of new dwellings can be found in Table 7. The point estimate for the differential
effect of construction activity at 17 on males versus females is 2.126 which is equivalent
to a marginal effect evaluated at the mean values of the covariates of 0.38. Intuitively,
this means that one extra new dwelling at the age of 17 per one hundred individuals
aged 18-24 increases the probability of high-school dropout by 0.004 (1.29%). The rest of
coefficients are very similar to the ones displayed in Table 5.
In general, the estimations of the contemporaneous and the construction activity at
age 17 effects are robust to the use of different measures of construction activity.
4.4 Robustness checks and additional specifications
Effects of changes in other economic sectors

The Spanish housing boom was contemporaneous to an important economic upturn.
16
Production was increased and employment was generated also in the rest of economic
sectors. Hence, other sectors may have played a role in explaining education decisions.
This is partly taken into account in the previous estimations by including total value
added, youth unemployment rate and unemployment rate as controls in the specifications.
However, as all other sectors affect differently males and low educated workers, if some
other sector’s share in total value added is correlated with the construction share that
other sector could be behind our results.
The empirical exercise is repeated for each of the other economic sectors including each
sector’ share in total value added instead of the construction share. Given the correlations
between sector shares in total value added displayed in Table 8, other sector’s level of
activity could explain the results found for the construction sector in the following cases:
(i) agriculture activity has a positive impact on high-school dropouts for males versus
females, (ii) industry activity has a negative effect on male with respect to female dropout,
(iii) energy activity increases the likelihood of dropout for males versus females and, (iv)
services activity has a positive and strong impact on male relative to female high-school
dropouts.
Figure 9 represents, for each economic sector, the share of male workers as well as the
proportions of male and female workers that are high-school dropouts. The agriculture,
energy and industry sectors are characterized by the prevalence of male workers while
the service sector employs the majority of working women. Regarding male high-school
dropouts, agriculture and industry have relatively high incidence of male high-school
dropouts while energy and services have relatively low male high-school dropout rates.
Female dropouts rates are high in agriculture, medium in services and low in energy and
construction.
Given the employment composition described above, other sectors’ effect in the prob-
ability of high-school dropout are supportive of our main hypothesis and, therefore, con-
sistent with the results found for the construction sector in the following cases: (i) the
agriculture and industry sector have a positive impact on high-school dropouts for males
17
versus females (an increase in agriculture or industry activity constitutes and improve-
ment in labor conditions for low educated men), (ii) the energy sector have a negative
impact on high-school dropouts for males versus females (an increment in energy activity
enhances employment opportunities for highly educated men), or (iii) the service sector
has a negative impact on high-school dropouts for males versus females (services growth
is associated with better labor market prospects for highly educated men and better
employment opportunities for all low and highly educated women).
Table 9 displays the results of the estimations using other sectors’ level of activity and
their interaction with male as alternative explanatory variables. I find that the coefficients
for agriculture, industry and energy are statistically insignificant while the service sector
has a negative influence on high-school dropouts. Hence, the alternative explanation in
which other sectors are responsible for the effect found for construction does not apply
in this context. However, the results found for the service sector support the hypothesis
that improvements in labor market perspectives for low educated workers induce more
dropouts while increases in expected gains of education cause less dropouts.
Moreover, the regression simultaneously including estimates for the differential effect
of construction and services activities on males versus females gives almost indistinguish-
able estimates from the ones found when studying the construction and services effects
separately. This can be explained by the low correlation between the construction and
service share in value added. It constitutes additional evidence that changes in the service
sector are unlikely to have any impact in our results for construction but have their own
effect instead.
Sample composed by non-migrants
One could argue that the estimated effect is a consequence of changes in the compo-
sition of the pool of individuals in each province. This would happen if individuals with
high propensity to drop out of high-school move to provinces with an increasing level of
construction activity. To address this, I have repeated the analysis over individuals that
live in the same province where they were born. The results, which are displayed in Table
18
10, are almost invariant with respect to those obtained using the full sample.
Sample composed by different age groups
The sample used in the previous regressions is composed by individuals aged 18 to
24 years old. This has been chosen in order to be coherent with the European Union
definition of high-school dropouts. Additionally, 18 years old is the minimum age to have
finished high-school. However, the group of individuals aged 18 to 24 years old is relatively
broad and the effects of labor market conditions may be different for younger relative to
older individuals. The effect of construction activity on males versus females is estimated
on four different subsamples composed by individuals aged (i) 18 to 20, (ii) 21 to 24, (iii)
18 to 22 and (iv) 23 to 24. As it can be seen in Table 11, all the estimated coefficients are
significant and consistent with the existence of a positive impact of better employment
prospects for low educated individuals on high-school dropouts. The magnitudes of the
effects are slightly larger for older samples, indicating that those are the ones more affected
by labor market conditions when taking their decisions on education.
Effects of construction activity at the age of 18
In our previous analysis I have taken into account the possibility that individuals take
their decisions to drop out of high-school influenced by labor market conditions at 17,
the average age in high-school. However, grade retention is severe in Spain. This means
that individuals could have taken their decision to drop out of high-school influenced by
labor market conditions later, at 18. This is also the age at which they are legally allowed
to start working in construction11 . To account for this possibility, the equation for the
probability of high-school dropout is estimated including the interaction of construction
activity at the age of 18 and male as key explanatory variable.
The results displayed in Table 12 are consistent with the findings for the impact of
construction activity at age 17. However, they are slightly smaller in magnitude12 .
11
The Spanish legislation states that workers under 18 are not allowed to perform certain tasks. Those
tasks comprise almost all the tasks that are required in building. However, data shows that more than
20% of total number of 16-17 aged individuals that work do it in the construction sector. This figure
almost doubles the average ratio of construction over total number of workers (11.5%).
12
A similar exercise is performed including construction activity at the age of 16 interacted with male
19
Placebo test
Gender is assigned randomly to each individual by nature in a way such that the
population is expected to be composed 50% by males. In this section, I randomly assign a
fictitious gender to sampled individuals and repeat the analysis by assuming that fictitious
males are affected by the housing boom while fictitious females do not. This exercise is
repeated several times and the coefficients are close to zero and mainly insignificant. This
shows that it would have been extremely difficult to have found a positive relationship if
males were not differently affected by construction with respect to females.
5 Conclusion
The European Commission in its 2010 Report on Education Policies states that ”targeted
measures for preventing [high-school dropout] should be further mainstreamed”. To guar-
antee the efficiency of education policies to reduce the incidence of high-school dropout
is important to understand the determinants of the decision to drop out.
This paper shows that labor market shocks that improve prospects for low educated
relative to high educated workers increase the probability of dropping out from high-
school.
The effect is identified by means of the increase in employment opportunities for
low educated male workers provoked by the housing boom in Spain. The housing boom
constituted a worldwide phenomenon caused by financial and demand factors that affected
local labor markets. The Spanish housing boom is a great opportunity to identify labor
market changes due to the big magnitude of the boom and its heterogenous incidence
across regions.
The estimated effect is such that a one percentage point increase in the construc-
tion share in total value added increases the probability of high-school dropout for males
as key explanatory variable. The point estimate for the differential effect of construction activity on
males is positive. However, the effect is statistically insignificant due to the reduction in the sample
which covers only the years 2003-2007.
20
by 1.74% while it leaves the probability of high-school dropout for females unaffected.
Additionally, a one percentage point increase in construction activity at the age of 17,
i.e., when the individual was more likely to be in high-school, increases the probability
of high-school dropout for males from 18 to 24 years old by 1.17% while it leaves the
probability of high-school dropout for females unchanged. These findings are consistent
with the estimations obtained using number of new dwellings as an alternative measure
of construction activity. I have ruled out the possibility that the estimated results are the
reflection of changes in other economic sectors. Additional results for the impact of ser-
vices activity on male versus female probability of high-school dropout provide additional
support to my hypothesis.
These conclusions suggest that individual schooling decisions largely respond to labor
market conditions even when the changes in those conditions are transitory. Hence, policy
designers should strengthen their efforts to incentivate individuals to get more education
in the presence of booms in economic sectors that use low educated workers intensively.
21
References
Bean, C. R. (1990): “Endogenous growth and the procyclical behaviour of productivity,”

European Economic Review, 34(2-3), 355–363.
Bertrand, M., E. Duflo, and S. Mullainathan (2004): “How Much Should

We Trust Differences-in-Differences Estimates?,” The Quarterly Journal of Economics,
119(1), 249–275.
Gali, J., and J. Hammour (1992): “Long Run Effects of Business Cycles,” (92-26).
Glaeser, E. L., J. Gyourko, and A. Saiz (2008): “Housing supply and housing
bubbles,” Journal of Urban Economics, 64(2), 198–217.
Gonzalez, L., and F. Ortega (2009): “Immigration and Housing Booms: Evidence
from Spain,” CReAM Discussion Paper Series 0919, Centre for Research and Analysis
of Migration (CReAM), Department of Economics, University College London.
Maani, S. A., and G. Kalb (2007): “Academic performance, childhood economic

resources, and the choice to leave school at age 16,” Economics of Education Review,
26(3), 361–374.
Malley, J., and V. A. Muscatelli (1999): “Business cycles and productivity growth:
Are temporary downturns productive or wasteful?,” Research in Economics, 53(4), 337–
364.
Peraita, C., and M. Pastor (2000): “The Primary School Dropout in Spain: The
Influence of Family Background and Labor Market Conditions,” Education Economics,
8(2), 157–168.
Petrongolo, B., and M. J. San Segundo (2002): “Staying-on at school at 16: the
impact of labor market conditions in Spain,” Economics of Education Review, 21(4),
353–365.
22
Rees, D. I., and H. N. Mocan (1997): “Labor market conditions and the high school
dropout rate: Evidence from New York State,” Economics of Education Review, 16(2),
103–109.
Saint-Paul, G. (1993): “Productivity growth and the structure of the business cycle,”
European Economic Review, 37(4), 861–883.
Schady, N. R. (2004): “Do Macroeconomic Crises Always Slow Human Capital Accu-
mulation?,” World Bank Economic Review, 18(2), 131–154.
Stern, D., I.-W. Paik, J. S. Catterali, and Y.-F. Nakata (1989): “Labor mar-
ket experience of teenagers with and without high school diplomas,” Economics of
Education Review, 8(3), 233–246.
Waren, J. R., and J. C. Lee (2003): “The Impact of Adolescent Employment on High
School Dropout: Differences by individual and labor-market characteristics,” Social
Science Research, 32, 98–128.
Yamashita, T. (2008): “The effects of the Great Depression on educational attainment,”

Discussion paper, Reed College.
23
Figures
Figure 1: High-school dropouts across countries
Data source: Eurostat. Information available at http://epp.eurostat.ec.europa.eu/tgm/table.do?

tab=table&init=1&language=en&pcode=tsisc060. The primary source is the EU Labor Force Survey.
The y-axis refers to the variation rate in the proportion of high-school dropouts from 2000 to 2008 and
the x-axis represents the value of the high-school dropout rate in 2008. The vertical line depicts the EU
objective established for 2010.
24
Figure 2: House prices over time
Data source: Spanish Ministry of Housing. Each data point corresponds to average house prices in
the second quarter of the year for the period 1995 to 2009.
25
Figure 3: Ratio of construction value added and GDP over time
Data source: Spanish National Accountings. This information can be found at the National Statistics
Institute webpage at the address http://www.ine.es/jaxiBD/tabla.do?per=03&type=db&divi=CNTR&idtab=1.
Each data point corresponds to the second quarter of each year for the period 1995 to 2009.
26
Figure 4: Number of constructed dwellings over time
Data source: Spanish Ministry of Housing. The data points are the result of adding up non-
subsidized and subsidized dwellings. Information on non-subsidized dwellings is accessible online from
http://www.mviv.es/es/index.php?option=com content&task=view&id=379&Itemid=434 and data on
subsidized dwellings can be found at http://www.mviv.es/es/index.php?option=com content&task=view&id=318
&Itemid=431. This data covers the period 1991-2008.
27
Figure 5: Number of workers in the construction sector over time
Data source: Spanish Labor Force Survey. Each data point corresponds to the second quarter of each
year for the period 1980-2007.
28
Figure 6: Average wages over time
Data source: Labor Costs Quarterly Survey (Encuesta Trimestral de Coste Laboral). Each data
point corresponds to the second quarter of each year for the period 2000 to 2008.
29
Figure 7: Correlation between dropout rate and construction
activity averaged by province
Males Females
.15
.2
.15
.1
.1
.05
.05
0
.06 .08 .1 .12 .14 .06 .08 .1 .12 .14

construction over total value added construction over total value added
Dropout rate Fitted values Dropout rate Fitted values
The estimated slope for men is 0.55 and is significant at the 10% level while the estimated slope for
women is 0.25 and is statistically indistinguishable from zero. The data on dropout rates is computed
using the Spanish Labor Force Survey. The data on construction over total value added is obtained from
the Spanish Regional Accounts. Averages by province are computed over the period 1995 to 2007.
30
Figure 8: Number of male and female construction workers over
time
Data source: Spanish Labor Force Survey. Each data point corresponds to the second quarter of each
year for the period 1980 to 2007.
31
Figure 9: Gender and high-school dropout rates across sectors
32
33
Data source: Spanish Labor Force Survey. The data corresponds to the second quarter of each year
for the period 1995-2007. The share of dropouts is computed as the proportion of working age individuals
that do not have a high-school degree and declare not to have received any education or training in the
previous four weeks.
34
Figure 10: Placebo test
Each bar corresponds to the point estimate of the coefficient associated with the interaction of
construction value added and male where male is a random variable equal to one with probability 0.5
and equal to 0 with probability 1. Therefore, the variable takes different values in each of the 100 draws.
In each regression, the dependent variable is equal to one if the individual is a high-school dropout
and 0 otherwise. Construction value added is measured relative to total value added in the province.
All estimations include the same controls, namely, construction value added, total value added, a male
dummy, age binary indicators, an immigrant dummy, parents characteristics including father and mother
high-school graduate dummies, father and mother university graduate indicators, father and mother
working dummies, binary variables equal to one if the parent is present in the household as well as an
emancipated dummy, youth and total unemployment rates, year dummies as well as province binary
indicators. The individual data is obtained from the Spanish Labor Force Survey. The information on
value added is part of the Spanish National Accounting data and can be found in the webpage of the
Instituto Nacional de Estadı́stica. The sample contains individuals aged 18 to 24 years old surveyed in
the fourth quarter of each year. The time period comprises the years from 1995 to 2007. Standard errors
are clustered at the province level.
35
Tables
Table 1: Descriptive statistics for the contemporaneous effect

estimations
Mean Sd Min Max

high-school dropout 0.33 0.47 0 1
construction over total VA by male 0.048 0.051 0 0.202
construction over total VA 0.095 0.027 0.04 0.202
new dwellings by male 0.017 0.023 0 0.125
new dwellings 0.034 0.022 0.002 0.152
total value added 1.95e+07 2.82e+07 570919 1.67e+08
youth unemployment rate 0.15 0.059 0.028 0.363
unemployment rate 0.066 0.028 0.015 0.164
male 0.511 0.5 0 1
age 21.054 1.983 18 24
immigrant 0.028 0.165 0 1
father high-school grad 0.111 0.314 0 1
mother high-school grad 0.096 0.294 0 1
father university grad 0.092 0.289 0 1
mother university grad 0.071 0.257 0 1
father working 0.588 0.492 0 1
mother working 0.296 0.457 0 1
absent father 0.217 0.412 0 1
absent mother 0.168 0.373 0 1
on one’s own 0.053 0.224 0 1
province 26.341 14.328 1 52
year 2000.283 3.655 1995 2007
weights 22210.2 14599.69 1038 414671
The number of included observations is 230635. The individual data is obtained from the Spanish
Labor Force Survey. The information on value added is collected from the Spanish Regional Accounts and
the data on number of new dwellings can be found in the webpage of the Spanish Ministry of Housing.
36
Table 2: Descriptive statistics for the effect of construction ac-
tivity at age 17 estimations. Construction activity measured by
value added
Mean Sd Min Max

construction over
total VA at 17 by male 0.047 0.049 0 0.189
construction over total VA at 17 0.093 0.023 0.04 0.189
total VA at 17 1.85e+07 2.65e+07 570919 1.56e+08
male 0.509 0.5 0 1
age 21.103 1.994 18 24
immigrant 0.045 0.208 0 1
father absent 0.31 0.463 0 1
mother absent 0.191 0.393 0 1
on one’s own 0.069 0.253 0 1
province 26.277 14.308 1 52
year 2004.31 1.719 2002 2007
weights 24197.04 16812.11 1038 414671
Labor Force Survey. The information on value added is collected from the Spanish Regional Accounts.
37
Table 3: Descriptive statistics for the effect of construction ac-
tivity at age 17 estimations. Construction activity measured by
number of new dwellings
Mean Sd Min Max

new dwellings at 17 by male 0.014 0.019 0 0.152
new dwellings at 17 0.028 0.019 0.002 0.152
total value added at 17 1.53e+07 2.40e+07 0 1.56e+08
male 0.509 0.5 0 1
age 21.103 1.987 18 24
immigrant 0.035 0.183 0 1
absent father 0.254 0.435 0 1
absent mother 0.177 0.382 0 1
on one’s own 0.06 0.237 0 1
province 26.332 14.324 1 52
year 2002.59 2.57 1999 2007
weights 23186.12 15571.18 1038 414671
Labor Force Survey. The information on number of new dwellings can be found in the webpage of the
Spanish Ministry of Housing. The missing values in the variable total value added at 17 are dummied
out.
38
Table 4: Contemporaneous effect. Construction activity mea-
sured by value added
basic unemployment parents year province
(1) (2) (3) (4) (5)
construction VA by male 2.151 2.038 3.060 3.028 2.899
(0.473)∗∗∗ (0.468)∗∗∗ (0.549)∗∗∗ (0.542)∗∗∗ (0.563)∗∗∗
contruction VA -2.581 0.866 0.919 2.771 1.186
(1.524)∗ (1.455) (1.312) (1.903) (0.955)
total value added -4.22e-09 -2.45e-09 -7.78e-10 -1.47e-10 1.90e-09
(9.82e-10)∗∗∗ (1.12e-09)∗∗ (8.89e-10) (1.20e-09) (1.42e-09)
male 0.434 0.455 0.496 0.5 0.521
(0.052)∗∗∗ (0.051)∗∗∗ (0.059)∗∗∗ (0.058)∗∗∗ (0.061)∗∗∗
19 years old 0.221 0.223 0.218 0.22 0.227
(0.015)∗∗∗ (0.014)∗∗∗ (0.015)∗∗∗ (0.015)∗∗∗ (0.015)∗∗∗
20 years old 0.333 0.339 0.303 0.306 0.32
(0.02)∗∗∗ (0.02)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗
21 years old 0.393 0.401 0.325 0.329 0.347
(0.02)∗∗∗ (0.02)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗
22 years old 0.399 0.411 0.289 0.296 0.316
(0.021)∗∗∗ (0.021)∗∗∗ (0.024)∗∗∗ (0.024)∗∗∗ (0.025)∗∗∗
23 years old 0.435 0.449 0.257 0.264 0.288
(0.023)∗∗∗ (0.023)∗∗∗ (0.025)∗∗∗ (0.024)∗∗∗ (0.026)∗∗∗
24 years old 0.47 0.489 0.217 0.226 0.253
(0.026)∗∗∗ (0.027)∗∗∗ (0.029)∗∗∗ (0.029)∗∗∗ (0.031)∗∗∗
immigrant 0.47 0.524 0.14 0.128 0.127
(0.089)∗∗∗ (0.09)∗∗∗ (0.089) (0.092) (0.087)
father high-school grad -1.218 -1.203 -1.171
(0.037)∗∗∗ (0.036)∗∗∗ (0.037)∗∗∗
mother high-school grad -1.007 -.986 -.948
(0.037)∗∗∗ (0.036)∗∗∗ (0.036)∗∗∗
father university grad -2.055 -2.059 -2.050
(0.06)∗∗∗ (0.061)∗∗∗ (0.062)∗∗∗
mother university grad -1.899 -1.895 -1.871
(0.069)∗∗∗ (0.07)∗∗∗ (0.071)∗∗∗
father working -.356 -.367 -.377
(0.018)∗∗∗ (0.018)∗∗∗ (0.016)∗∗∗
mother working -.041 -.053 -.056
(0.021)∗∗ (0.022)∗∗ (0.02)∗∗∗
absent father -.432 -.458 -.434
(0.036)∗∗∗ (0.029)∗∗∗ (0.026)∗∗∗
absent mother 0.127 0.143 0.129
(0.034)∗∗∗ (0.032)∗∗∗ (0.032)∗∗∗
emancipated 1.156 1.163 1.116
(0.05)∗∗∗ (0.05)∗∗∗ (0.05)∗∗∗
(0.835)∗∗∗ (0.73)∗∗∗ (0.753) (0.273)∗∗
unemployment rate 3.411 2.435 2.464 -1.544
(1.958)∗ (1.670) (1.818) (1.034)
The dependent variable is equal to one if the individual is a high-school dropout and 0 otherwise.
Construction value added is measured relative to total value added in the province. The coefficients are
marked with * if the level of significance is between 5% and 10%, ** if the level of significance is between
1% and 5% and *** if the level of significance is less than 1%. The basic regression includes construction
value added interacted by male, construction value added, total value added, a male dummy, age binary
indicators and an immigrant dummy. The second column adds parents characteristics to the basic
specification including father and mother high-school graduate dummies, father and mother university
graduate indicators, father and mother working dummies, binary variables equal to one if the parent is
present in the household as well as an emancipated dummy. The third column includes, in addition to
the controls in column 2, youth and total unemployment rates. The fourth column adds year dummies.
Finally, the fifth column includes all previously mentioned controls plus province binary indicators. The
individual data is obtained from the Spanish Labor Force Survey. The information on value added is
39
part of the Spanish National Accounting data and can be found in the webpage of the Instituto Nacional
de Estadı́stica. The sample contains individuals aged 18 to 24 years old surveyed in the fourth quarter
of each year. The time period comprises the years from 1995 to 2007. Standard errors are clustered at
the province level. The number of included observations is 230635.
40
Table 5: Effect of construction activity at age 17. Construction
activity measured by value added
(1) (2) (3) (4) (5)
construction VA at 17 by male 2.428 2.336 2.094 2.088 1.904
(0.888)∗∗∗ (0.874)∗∗∗ (1.003)∗∗ (1.009)∗∗ (0.995)∗
construction VA at 17 -4.515 -3.448 -3.622 -3.904 -.704
(2.082)∗∗ (2.142) (1.906)∗ (2.000)∗ (1.605)
tva17 -9.50e-10 -2.25e-10 1.34e-09 9.18e-10 5.29e-09
(1.33e-09) (1.12e-09) (9.70e-10) (1.01e-09) (1.45e-09)∗∗∗
contruction VA 8.075 7.403 7.707 6.787 2.571
(1.792)∗∗∗ (1.546)∗∗∗ (1.393)∗∗∗ (1.655)∗∗∗ (1.256)∗∗
male 0.473 0.489 0.642 0.643 0.67
(0.094)∗∗∗ (0.091)∗∗∗ (0.105)∗∗∗ (0.106)∗∗∗ (0.106)∗∗∗
19 years old 0.249 0.257 0.252 0.251 0.282
(0.027)∗∗∗ (0.027)∗∗∗ (0.028)∗∗∗ (0.028)∗∗∗ (0.027)∗∗∗
20 years old 0.345 0.36 0.33 0.326 0.397
(0.031)∗∗∗ (0.034)∗∗∗ (0.032)∗∗∗ (0.033)∗∗∗ (0.03)∗∗∗
21 years old 0.38 0.405 0.313 0.307 0.401
(0.046)∗∗∗ (0.051)∗∗∗ (0.045)∗∗∗ (0.046)∗∗∗ (0.038)∗∗∗
22 years old 0.355 0.385 0.248 0.241 0.358
(0.058)∗∗∗ (0.063)∗∗∗ (0.058)∗∗∗ (0.06)∗∗∗ (0.051)∗∗∗
23 years old 0.321 0.355 0.149 0.14 0.278
(0.056)∗∗∗ (0.06)∗∗∗ (0.053)∗∗∗ (0.057)∗∗ (0.055)∗∗∗
24 years old 0.297 0.337 0.057 0.048 0.196
(0.061)∗∗∗ (0.067)∗∗∗ (0.058) (0.061) (0.051)∗∗∗
immigrant 0.671 0.715 0.332 0.325 0.301
(0.08)∗∗∗ (0.082)∗∗∗ (0.078)∗∗∗ (0.079)∗∗∗ (0.079)∗∗∗
(0.039)∗∗∗ (0.039)∗∗∗ (0.037)∗∗∗
mother high-school grad -.975 -.982 -.949
(0.043)∗∗∗ (0.043)∗∗∗ (0.044)∗∗∗
(0.082)∗∗∗ (0.082)∗∗∗ (0.086)∗∗∗
(0.074)∗∗∗ (0.074)∗∗∗ (0.075)∗∗∗
(0.03)∗∗∗ (0.03)∗∗∗ (0.029)∗∗∗
(0.027)∗∗∗ (0.028)∗∗∗ (0.023)∗∗∗
absent father -.327 -.361 -.349
(0.04)∗∗∗ (0.039)∗∗∗ (0.038)∗∗∗
absent mother -.011 0.007 0.009
(0.042) (0.042) (0.043)
emancipated 0.923 0.925 0.908
(0.054)∗∗∗ (0.054)∗∗∗ (0.057)∗∗∗
(0.767)∗∗ (0.686)∗∗∗ (0.666)∗∗ (0.411)∗
(2.511)∗ (2.227) (2.100)∗ (1.489)
marked with * if the level of significance is between 5% and 10%, ** if the level of significance is between 1%
and 5% and *** if the level of significance is less than 1%. The basic regression includes construction value
added at 17 interacted by male, construction value added at 17, total value added at 17, construction
value added, a male dummy, age binary indicators and an immigrant dummy. The second column
adds parents characteristics to the basic specification including father and mother high-school graduate
dummies, father and mother university graduate indicators, father and mother working dummies, binary
variables equal to one if the parent is present in the household as well as an emancipated dummy. The
third column includes, in addition to the controls in column 2, youth and total unemployment rates. The
fourth column adds year dummies. Finally, the fifth column includes all previously mentioned controls
41
plus province binary indicators. The individual data is obtained from the Spanish Labor Force Survey.
The information on value added is part of the Spanish National Accounting data and can be found in the
webpage of the Instituto Nacional de Estadı́stica. The sample contains individuals aged 18 to 24 years
old surveyed in the fourth quarter of each year. The time period comprises the years from 2002 to 2007.
Standard errors are clustered at the province level. The number of included observations is 86953.
42
sured by number of new dwellings
(1) (2) (3) (4) (5)
new dwellings by male 2.674 2.473 3.942 3.893 3.875
(0.64)∗∗∗ (0.574)∗∗∗ (0.801)∗∗∗ (0.8)∗∗∗ (0.854)∗∗∗
new dwellings -4.275 1.933 1.693 2.810 -.926
(1.743)∗∗ (1.340) (1.263) (1.433)∗∗ (0.692)
(1.03e-09)∗∗∗ (9.86e-10)∗∗ (7.26e-10) (8.22e-10) (1.44e-09)
male 0.549 0.565 0.655 0.657 0.666
(0.028)∗∗∗ (0.027)∗∗∗ (0.032)∗∗∗ (0.032)∗∗∗ (0.034)∗∗∗
19 years old 0.221 0.223 0.217 0.219 0.227
(0.015)∗∗∗ (0.014)∗∗∗ (0.015)∗∗∗ (0.015)∗∗∗ (0.015)∗∗∗
20 years old 0.333 0.338 0.302 0.304 0.32
(0.02)∗∗∗ (0.02)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗
21 years old 0.394 0.401 0.323 0.328 0.347
(0.02)∗∗∗ (0.02)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗
22 years old 0.401 0.41 0.287 0.294 0.316
(0.021)∗∗∗ (0.021)∗∗∗ (0.023)∗∗∗ (0.023)∗∗∗ (0.025)∗∗∗
23 years old 0.437 0.447 0.255 0.262 0.288
(0.022)∗∗∗ (0.023)∗∗∗ (0.025)∗∗∗ (0.025)∗∗∗ (0.026)∗∗∗
24 years old 0.472 0.487 0.215 0.223 0.252
(0.026)∗∗∗ (0.027)∗∗∗ (0.029)∗∗∗ (0.029)∗∗∗ (0.031)∗∗∗
immigrant 0.485 0.52 0.14 0.123 0.128
(0.093)∗∗∗ (0.09)∗∗∗ (0.09) (0.091) (0.087)
(0.038)∗∗∗ (0.037)∗∗∗ (0.037)∗∗∗
mother high-school grad -1.004 -.990 -.949
(0.036)∗∗∗ (0.036)∗∗∗ (0.036)∗∗∗
(0.061)∗∗∗ (0.061)∗∗∗ (0.062)∗∗∗
(0.07)∗∗∗ (0.07)∗∗∗ (0.071)∗∗∗
(0.018)∗∗∗ (0.017)∗∗∗ (0.016)∗∗∗
(0.021)∗∗ (0.022)∗∗ (0.02)∗∗∗
absent father -.421 -.460 -.435
(0.031)∗∗∗ (0.031)∗∗∗ (0.026)∗∗∗
absent mother 0.118 0.141 0.13
(0.034)∗∗∗ (0.033)∗∗∗ (0.032)∗∗∗
emancipated 1.153 1.159 1.116
(0.05)∗∗∗ (0.05)∗∗∗ (0.05)∗∗∗
(0.791)∗∗∗ (0.695)∗∗∗ (0.765) (0.281)∗∗
(2.136)∗∗ (1.841)∗ (1.923)∗∗ (1.102)∗
Number of new dwellings is measured relative to population aged 18 to 24 in the province. The coefficients
are marked with * if the level of significance is between 5% and 10%, ** if the level of significance is
between 1% and 5% and *** if the level of significance is less than 1%. The basic regression includes
number of new dwellings interacted by male, number of new dwellings, total value added, a male dummy,
age binary indicators and an immigrant dummy. The second column adds parents characteristics to
the basic specification including father and mother high-school graduate dummies, father and mother
university graduate indicators, father and mother working dummies, binary variables equal to one if the
parent is present in the household as well as an emancipated dummy. The third column includes, in
addition to the controls in column 2, youth and total unemployment rates. The fourth column adds
year dummies. Finally, the fifth column includes all previously mentioned controls plus province binary
indicators. The individual data is obtained from the Spanish Labor Force Survey. The information on
43
number of new dwellings can be found in the webpage of the Spanish Ministry of Housing. The sample
contains individuals aged 18 to 24 years old surveyed in the fourth quarter of each year. The time period
comprises the years from 1995 to 2007. Standard errors are clustered at the province level. The number
of included observations is 230635.
44
activity measured by number of new dwellings
(1) (2) (3) (4) (5)
new dwellings at 17 by male 1.806 1.804 2.337 2.269 2.126
(0.761)∗∗ (0.73)∗∗ (0.843)∗∗∗ (0.854)∗∗∗ (0.901)∗∗
new dwellings at 17 0.479 2.464 3.163 1.848 0.55
(1.878) (1.534) (1.263)∗∗ (1.401) (1.400)
tva17 -2.55e-09 -8.32e-10 9.85e-10 6.03e-10 2.80e-09
(1.13e-09)∗∗ (1.04e-09) (8.24e-10) (8.28e-10) (7.04e-10)∗∗∗
new dwellings 2.038 4.055 3.853 3.687 1.036
(1.325) (1.499)∗∗∗ (1.523)∗∗ (1.497)∗∗ (0.724)
male 0.639 0.648 0.773 0.775 0.789
(0.033)∗∗∗ (0.032)∗∗∗ (0.035)∗∗∗ (0.035)∗∗∗ (0.037)∗∗∗
19 years old 0.257 0.268 0.265 0.26 0.261
(0.022)∗∗∗ (0.021)∗∗∗ (0.021)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗
20 years old 0.374 0.398 0.368 0.358 0.365
(0.029)∗∗∗ (0.028)∗∗∗ (0.026)∗∗∗ (0.027)∗∗∗ (0.024)∗∗∗
21 years old 0.41 0.444 0.366 0.351 0.362
(0.035)∗∗∗ (0.033)∗∗∗ (0.028)∗∗∗ (0.029)∗∗∗ (0.027)∗∗∗
22 years old 0.384 0.439 0.322 0.298 0.306
(0.04)∗∗∗ (0.04)∗∗∗ (0.037)∗∗∗ (0.038)∗∗∗ (0.036)∗∗∗
23 years old 0.35 0.422 0.241 0.208 0.218
(0.041)∗∗∗ (0.042)∗∗∗ (0.038)∗∗∗ (0.039)∗∗∗ (0.037)∗∗∗
24 years old 0.347 0.438 0.188 0.149 0.16
(0.044)∗∗∗ (0.048)∗∗∗ (0.044)∗∗∗ (0.043)∗∗∗ (0.041)∗∗∗
immigrant 0.598 0.631 0.229 0.215 0.223
(0.088)∗∗∗ (0.086)∗∗∗ (0.087)∗∗∗ (0.087)∗∗ (0.086)∗∗∗
(0.036)∗∗∗ (0.036)∗∗∗ (0.035)∗∗∗
mother high-school grad -.987 -.996 -.956
(0.039)∗∗∗ (0.04)∗∗∗ (0.039)∗∗∗
(0.07)∗∗∗ (0.07)∗∗∗ (0.072)∗∗∗
(0.066)∗∗∗ (0.066)∗∗∗ (0.067)∗∗∗
(0.02)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗
(0.024)∗∗∗ (0.024)∗∗∗ (0.021)∗∗∗
absent father -.346 -.401 -.382
(0.034)∗∗∗ (0.035)∗∗∗ (0.031)∗∗∗
absent mother 0.019 0.055 0.061
(0.037) (0.036) (0.035)∗
independent 1.079 1.086 1.053
(0.054)∗∗∗ (0.054)∗∗∗ (0.055)∗∗∗
(0.827) (0.761) (0.768) (0.329)
(2.467)∗∗∗ (2.187)∗∗∗ (2.218)∗∗∗ (0.992)∗
Number of new dwellings is measured relative to population aged 18 to 24 in the province. The coefficients
are marked with * if the level of significance is between 5% and 10%, ** if the level of significance is between
1% and 5% and *** if the level of significance is less than 1%. The basic regression includes number of
new dwellings at 17 interacted by male, number of new dwellings at 17, total value added at 17, number
of new dwellings, a male dummy, age binary indicators and an immigrant dummy. The second column
45
The information on number of new dwellings can be found in the webpage of the Spanish Ministry of
Housing. The sample contains individuals aged 18 to 24 years old surveyed in the fourth quarter of
each year. The time period comprises the years from 1999 to 2007. Standard errors are clustered at the
province level. The number of included observations is 143440.
46
Table 8: Correlations between value added shares of economic
sectors
Agriculture Industry Energy Construction Services
Agriculture 1
Industry -0.174 1
Energy 0.011 -0.162 1
Construction 0.103 -0.439 0.054 1
Services -0.484 -0.649 -0.233 0.009 1
Data source: Spanish Regional Accounts. The correlation is computed for the sample included in the
regression for the contemporaneous effect.
47
Table 9: Contemporaneous effect. Other sectors activity mea-
sured by value added
agriculture industry energy services
(1) (2) (3) (4)
Agriculture VA by male 0.457
(0.44)
Agriculture VA -.777
(0.58)
Industry VA by male 0.158

(0.279)
Industry VA 1.273
(0.803)
Energy VA by male 0.672

(0.62)
Energy VA -1.539
(1.174)
Services VA by male -.710

(0.296)∗∗
Services VA -.362
(0.552)
Each sector value added is measured relative to total value added in the province. The coefficients
are marked with * if the level of significance is between 5% and 10%, ** if the level of significance is
between 1% and 5% and *** if the level of significance is less than 1%. All equations include the same
controls, namely, total value added, a male dummy, age binary indicators, an immigrant dummy, parents
characteristics including father and mother high-school graduate dummies, father and mother university
graduate indicators, father and mother working dummies, and binary variables equal to one if the parent
is present in the household as well as an emancipated dummy, youth and total unemployment rates, year
dummies as well as province binary indicators. The individual data is obtained from the Spanish Labor
Force Survey. The information on value added is part of the Spanish National Accounting data and can
be found in the webpage of the Instituto Nacional de Estadı́stica. The sample contains individuals aged
18 to 24 years old surveyed in the fourth quarter of each year. The time period comprises the years from
1995 to 2007. Standard errors are clustered at the province level. The number of included observations
is 230635.
48
sured by value added. Sample composed by non-migrants
(1) (2) (3) (4) (5)
construction VA by male 2.251 2.117 2.999 2.961 2.816
(0.522)∗∗∗ (0.517)∗∗∗ (0.606)∗∗∗ (0.596)∗∗∗ (0.611)∗∗∗
contruction VA -2.918 0.948 1.079 2.994 1.217
(1.679)∗ (1.541) (1.389) (1.963) (1.033)
(9.48e-10)∗∗∗ (1.10e-09)∗∗ (8.71e-10) (1.19e-09) (1.72e-09)
1% and 5% and *** if the level of significance is less than 1%. The basic regression includes construction
value added interacted by male, construction value added, total value added, a male dummy, age binary
indicators and an immigrant dummy. The second column adds parents characteristics to the basic
specification including father and mother high-school graduate dummies, father and mother university
present in the household as well as an emancipated dummy. The third column includes, in addition to
the controls in column 2, youth and total unemployment rates. The fourth column adds year dummies.
Finally, the fifth column includes all previously mentioned controls plus province binary indicators. The
individual data is obtained from the Spanish Labor Force Survey. The information on value added is
part of the Spanish National Accounting data and can be found in the webpage of the Instituto Nacional
de Estadı́stica. The sample contains individuals aged 18 to 24 years old surveyed in the fourth quarter
of each year who live in their province of birth. The time period comprises the years from 1995 to 2007.
Standard errors are clustered at the province level. The number of included observations is 203335.
49
sured by value added. Sample composed by different age groups
18-20 21-24 18-22 23-24
(1) (2) (3) (4)
construction VA by male 2.339 3.224 2.588 3.622
(0.554)∗∗∗ (0.736)∗∗∗ (0.673)∗∗∗ (0.628)∗∗∗
contruction VA 0.446 1.675 1.021 1.509

(1.201) (1.239) (1.067) (1.336)
Observations 95668 134967 163414 67221
1% and 5% and *** if the level of significance is less than 1%. The displayed coefficients are associated
to the variables construction value added by male and construction value added. In the first column
the sample is composed by individuals aged 18 to 20, in the second by individuals aged 21 to 24, in the
third by individuals 18 to 22 and in the fourth by individuals 23 to 24. All equations include the same
controls, namely, total value added, a male dummy, age binary indicators, an immigrant dummy, parents
characteristics including father and mother high-school graduate dummies, father and mother university
present in the household as well as an emancipated dummy, youth and total unemployment rates, year
dummies as well as province binary indicators. The individual data is obtained from the Spanish Labor
Force Survey. The information on value added is part of the Spanish National Accounting data and
can be found in the webpage of the Instituto Nacional de Estadı́stica. The sample contains individuals
surveyed in the fourth quarter of each year. The time period comprises the years from 1995 to 2007.
Standard errors are clustered at the province level.
50
activity measured by value added
(1) (2) (3) (4) (5)
construction VA at 18 by male 1.916 1.838 1.669 1.658 1.557
(0.707)∗∗∗ (0.705)∗∗∗ (0.787)∗∗ (0.791)∗∗ (0.78)∗∗
construction VA at 18 -4.693 -3.925 -3.888 -4.144 -1.662
(1.959)∗∗ (2.072)∗ (1.860)∗∗ (1.901)∗∗ (1.565)
construction VA 8.421 8.218 8.549 7.312 2.875
(1.645)∗∗∗ (1.485)∗∗∗ (1.403)∗∗∗ (1.597)∗∗∗ (1.130)∗∗
marked with * if the level of significance is between 5% and 10%, ** if the level of significance is between 1%
and 5% and *** if the level of significance is less than 1%. The basic regression includes construction value
added at 18 interacted by male, construction value added at 18, total value added at 18, construction
value added, a male dummy, age binary indicators and an immigrant dummy. The second column
The information on value added is part of the Spanish National Accounting data and can be found in the
webpage of the Instituto Nacional de Estadı́stica. The sample contains individuals aged 18 to 24 years
old surveyed in the fourth quarter of each year. The time period comprises the years from 2001 to 2007.
Standard errors are clustered at the province level. The number of included observations is 104299
51

High School Dropouts and Transitory Labor Market Shocks - The Case of The Spanish Housing Boom

Uploaded by

Copyright:

Available Formats

High School Dropouts and Transitory Labor Market Shocks - The Case of The Spanish Housing Boom

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

High School Dropouts and Transitory Labor Market Shocks - The Case of The Spanish Housing Boom

Uploaded by

Copyright:

Available Formats

High­School Dropouts and Transitory Labor

Market Shocks: The Case of the Spanish Housing

Collegio Carlo Alberto and IZA

JEL codes: J24, J22, I20, L74

Keywords: High-school dropout, housing boom, Spain

βZijt = β 0 + β 1 Cjt · maleijt + β 2 Cjt + β 3 maleijt + β 4 Xijt + β 5 Wjt + β 6 η j + β 7 ξ t + εijt

where y is an indicator variable equal to one if individual i living in province j at year

3 Data and descriptive statistics

3.1 Data sources and sample definition

3.2 Variable definition and descriptive statistics

4.1 Contemporaneous effect

4.2 Effect of construction activity at the age of 17

4.4 Robustness checks and additional specifications

Effects of changes in other economic sectors

Bean, C. R. (1990): “Endogenous growth and the procyclical behaviour of productivity,”

Bertrand, M., E. Duflo, and S. Mullainathan (2004): “How Much Should

Maani, S. A., and G. Kalb (2007): “Academic performance, childhood economic

Yamashita, T. (2008): “The effects of the Great Depression on educational attainment,”

Figure 1: High-school dropouts across countries

Data source: Eurostat. Information available at http://epp.eurostat.ec.europa.eu/tgm/table.do?

.06 .08 .1 .12 .14 .06 .08 .1 .12 .14

Dropout rate Fitted values Dropout rate Fitted values

Table 1: Descriptive statistics for the contemporaneous effect

Mean Sd Min Max

Mean Sd Min Max

Mean Sd Min Max

Industry VA by male 0.158

Energy VA by male 0.672

Services VA by male -.710

contruction VA 0.446 1.675 1.021 1.509

Observations 95668 134967 163414 67221

You might also like

HighSchool Dropouts and Transitory Labor