The evolution of returns to education in Argentina
PRELIMINARY AND INCOMPLETE.
Florencia López Bóo ∗
Department of Economics, University of Oxford
Manor Road, Oxford, OX1 3UQ, UK.
August 2007
Abstract Returns to schooling in urban Argentina increased from 1992 to
2003, a period of economic reforms and macroeconomic volatility. In this
paper we provide the most consistent estimates of returns to education so far
by gender and by occupation. We also investigate earnings profiles over time.
This paper contributes to the existing literature by employing a variety of
methodologies in order to estimate these returns; by using macroeconomic
variables to explore shifts in earnings and by exploring sectoral earnings
differentials with instrumental variables for occupational choice. Increasing
convexity in the returns to education reflects increasing or stable earnings
for college graduates up until 2001, combined with decreasing earnings for
the less educated from 1995 onwards. Moreover, from 2001 to 2003, wages
for the less educated were falling at a faster pace than college graduate’s
wages. This result is robust to endogeneity, selection, specification and to
changes in the functional form. Higher returns to education had offset the
effect of shocks only for college completers. Men have higher returns than
women. Trade explains falling earnings for secondary completers and capital
accumulation has a differentiated effect by skills. However, after controlling
for all macro variables we still found a highly significant downward trend for
the earnings of primary completers.
JEL Classification Codes: I21, J31
Keywords: returns to schooling, earnings profiles, occupations, macroeconomic shocks,
policy swings, Argentina
∗
email: [email protected]. The author wishes to thank Francis Teal for very
helpful discussions. The paper also benefited from comments received from the University of Oxford’s
CSAE workshop, the Department of Economics Seminar at the University of Namur, the OECD Development Center and the World Bank-Latin American-Poverty and Gender group seminar participants.
All the errors however remain my sole responsibility.
1
Introduction
During the past three decades many Latin American countries underwent several macroeconomic shocks and policy swings, ranging from severe inflationary experiences, opening
of the economy and privatization to periods of real exchange rate under and over–
valuation.
In that context, how the returns to education have fluctuated over these swings
remains of central policy concern. Particularly interesting is the exploration of how
the dispersion of returns across education levels behaves over time. As rates of return
to education determine the incentives to invest in education, they are crucial for understanding how individuals change fundamental decisions on schooling, occupational
choice, fertility and training with shifts in expected returns.
Several OECD countries have experienced an increasing dispersion of wages during
the last two decades, in particular after important structural reforms and, mainly, technological change, globalization and the change in world demand. By far the biggest rise
in wage dispersion took place in the UK and the US (Layard and Nickell 2000). In particular, a large increase in the wage differentials by educational level is observed in these
countries (Bound and Johnson 1992, Katz and Murphy 1995, Machin 1996, Schmitt
1995). In the Latin-American context, the case of Argentina is particularly interesting since the increase in wage inequality was larger than in any other Latin American
country over the nineties. Moreover, both the speed and depth of the economic reforms
were also greater than in any other Latin-American country over the same period. 1
Especially, the large variation in macroeconomic aggregates provides the researcher with
exogenous shocks which are sharp enough to be natural experiments. Argentina experienced three important changes in her economic structure during the last 15 years: (i)
economic reforms (including trade opening)which lead to a higher rate of capital accumulation in the early 1990s, (iii) massive fluctuations in capital flows in 1995 and (iv) a
40% devaluation and default of the debt in 2001.
Studies for other Latin American countries have dealt with very smooth time series,
so that which shock was causing shifts in wage profiles was difficult to identify. Here,
among other things we will take advantage of the variation in our data to overcome this
methodological problem.
This paper will focus on the evolution of the rates of return to different levels of education and will also look at earnings differentials by occupation. In particular, we will
investigate the effects of a set of macroeconomic variables over the returns to human
capital during the period 1992–2003 in Argentina’s main urban centers. If individuals are imperfect substitutes in production or changes in labour supply behaviour are
different for different types of individuals, we assume that external shocks and pol1
In 1991 the Argentine economy was transformed through the establishment of a currency board
arrangement as part of a sweeping set of reforms that altered the monetary system, improved fiscal and
tax policies, liberalized trade and reformed the public sector including a rapid privatization programme
and changes in the social security system in the early nineties. By the end of 1991, nominal tariffs had
been lowered to an average level of 12% and all import licenses had been eliminated. The physical capital
stock (excluding the public sector) grew by 20% between 1992 and 1999 (FIEL 2002). The success of
these reforms brought inflation down from 1,343% in 1990 to 17.5% in 1992 and to one-digit rates from
1993 until 2002. It also translated into GDP growth rates of 10.6% in 1991, 9.6% in 1992 and always
higher– than–5% rates up until 1997 (with the exception of the 1995 crisis).
2
icy changes will impact differently on individuals wages. In essence, we will measure
private economics returns to education by using Mincer’s semi-logarithmic regressions
(Mincer 1974). However, as is well known, the causality link from education to earnings is problematic. Biases due to measurement error in reported schooling, omitted
variables, selection into employment or into certain occupations and the distinction between homogeneous vs heterogeneous returns are some of the issues we will need to
take into account. Moreover, while the standard theory of investment in human capital
put forward a concave education-earnings profile, empirical evidence from various countries has challenged the prevailing view (Behrman and Wolfe 1984, Lachler 1999, Blom,
Holm, Lauritz, and Verner 2001, Patrinos and Sakellariou 2005, Söderbom, Teal F.,
and Kahyarara 2006). This finding raises some policy concerns and deserves further
investigation.
A number of studies have measured the private rates of return to education in Argentina (Kugler and Psacharopoulos 1989, Pessino 1993, Pessino 1996, Gasparini, Marchionni, and Sosa Escudero 2001, Galiani and Sanguinetti 2003, Patrinos, Fiszbein, and
Giovagnoli 2005). However, most of them are dated, constrained by data or have not
addressed diverse methodological problems. Kugler and Psacharopoulos estimate the
private rate of return to another year of schooling in a post-hyperinflation year (1989)
at 10.3 percent in urban Argentina. They also note that returns to schooling were higher
for workers in the private sector: 9.6 versus 7.0 percent in the public sector in 1985;
and in 1989, 11.1 versus 8.9 percent. Pessino finds that, after an inflationary shock, the
returns to schooling in Buenos Aires increased from 10 percent in 1986 to 12.5 percent
in 1989. Then they dropped to 9 percent in 1990 and increased again to 10 percent
by 1993. She argues that hyperinflation was the main cause in the shifting of wage
profiles as after 1990, with the beginning of a period of very low inflation, the profiles
return to 1986 levels. She shows that, after 1990, the rate of return increases continually, especially regarding the ”college premium”. 2 Gasparini et al found increasing
returns to education on the Greater Buenos Aires from 1986 to 1998, while Galiani and
Sanguinetti; and Gasparini and Acosta find increasing returns to college graduates in
the manufacturing sector in Greater Buenos Aires, the main urban agglomerate, over
the nineties. 3 4
There are three consistent results from previous studies from Argentina:(i) returns
to education increase with the level of education (challenging the dominant theoretical
view of concave earnings function), (ii) men have higher returns to schooling compared
to females for every level of education and (iii) the overall rate of return to an additional
year of schooling is higher than the average for middle income countries (Psacharopoulos
and Patrinos 2004). Previous studies mostly compute linear returns to schooling, for
2
She estimates Mincerian wage equations on years of education, and on a set of dummies for levels
of education by occupational categories and sectors on a sample of working males aged 25–54 in the
Greater Buenos Aires for the 1986–1993 period.
3
Gasparini et al do address selectivity issues. Following Bourguignon et al. (1999) they assume that
labor market participation choices are made within the household in a sequential fashion. Spouses take
the heads labor market status into consideration to decide whether to enter the labor market or not. In
turn, other members of the family consider both the head and the spouse labor market status before
deciding whether to participate or not.
4
Galiani et al analyze the evolution of skill premiums by educational attainment levels for three skill
groups: unskilled, semi-skilled and skilled workers. They exclude self-employees, owner-managers and
unpaid workers from their analysis; and estimate time series derived from the coefficients of a set of
Mincer wage equations by gender.
3
males, wage-employees in the Greater Buenos Aires; and hence, are not able to respond
to the central question of this study: what happened to the full earnings-education
profile, for different occupations in all regions of Argentina over this period dominated
by shocks and policy swings?
On the effects of crises on the rates of return to education, the literature is scarce.
Only Pessino had studied the particular effects of the hyperinflation period on rates of
return. Most of the studies from Argentina concentrate on the effects of crisis on welfare,
employment, coping mechanisms and use of public services. 5 Here we will then try to
answer the question of whether these shocks had a differential impact by skill level.
The closest work to ours is Fizbein et al (2006). They estimate Mincerian equations
on years of education and on dummies for different categorical levels of education by
gender on the full labor force for Argentina from 1992 to 2002 and correct for the selectivity problem in the participation of women in the labor market. 6 They find that
returns to education increase with the level of education, however, they do not attempt
to correct for the selection in the participation decision for males nor endogeneity in
education, neither explore regional effects or different functional forms in the earningseducation profile.
In summary, none of these studies addresses various methodological problems at the
same time. As their estimates could be biased, they raise some uncertainty about their
findings.
Using household surveys, the objective of this paper is to estimate Mincerian returns
and wage premiums, by gender and by occupations in a more consistent manner.7 We will
check the robustness of the estimates to changes in the functional form of the earnings
function (parametrically and semi-parametrically) and we will attempt to correct for
endogeneity and selectivity problems. In addition to the before and after comparisons
common in the literature (Pessino 1993, 1996), we use 11 years of data to determine
what period-to-period movement occurred in the economy in stable periods. 8
Briefly, four main methods of estimation are used to take into consideration selection,
endogenous education and endogenous sorting between occupations: (i) standard ordinary least squares, (ii) Heckman maximum likelihood procedure which deals with the
sample selectivity issues which arise because earnings are only observed for individuals who participate in the labor force and who may form a non-random sub-sample
5
On the particular effect of crises on welfare, another study by Fizbein and Giovagnoli presents the
initial findings of a household survey dealing with the effects of the 2001 Argentine economic crisis on
welfare. The results obtained identify the limitations of the different coping mechanisms and reveal
serious effects on welfare. The evidence they present there suggests that the effects on the use of health
services have been more marked than those on the use of education services (Fiszbein, Giovagnoli, and
Aduriz 2003). McKenzie finds that the crisis had a large aggregate effect, with 78% of households
surveyed experiencing real income declines in 2002 and 63% suffering a real income fall of 20% or more.
In spite of consumer price inflation of 41%, he finds that the distribution of nominal incomes remained
remarkably constant, resulting in dramatic declines in real wages (McKenzie 2004). He also finds that
job exits rose and that existing workers were not able to increase labor hours worked to counter the
effects of existing wages, despite many workers saying that they would like to work more hours.
6
They also introduce a quantile analysis finding that returns increased during the last decade and
that men in higher quantiles have higher returns to schooling compared to those in the lower quantiles,
while for women returns are highest at the lowest quantiles.
7
Differentials in earnings by occupations have not been widely explored besides the report by the
World Bank on informality in Latin America (World Bank, 2007).
8
This is important as some groups exhibit higher variability over time in their labor market behavior
(i.e informal workers, self employed).
4
of the population, (iii) household fixed effects estimation to control for unobserved
family-specific heterogeneity. For this, estimates are based on spouse pairs, sibling
pairs, grandmother/father-granson/daughter and parent-child pairs, and (iv) we exploit
potential instruments for occupation choice to give a consistent estimate of the occupation dummies due to the fact that workers may endogenously sort between occupations
based on observable and unobservable skill differences (this point is still in progress) In
all methods, we add dummies for different occupations and we allow for the possibility
that parameters are different for the two genders. At the same time, our main variable of
interest (education) will be expressed as either: levels of education, years of education,
years of education square, years of education cube or semi parametrically in the spirit of
partial linear regression models. In order to test credentialism effects dummy variable
for each year of education completed are also included in some regressions.
We depart from studies as Fizbein et al, Galiani et al and Gasparini et al, as we do
take into account the formal vs. informal, wage–employee vs. self–employed rates of
returns, when they only keep wage–employees workers in their samples.
Another novelty here is that we will contrast returns to education trends with wage
trends, both of which are affected by macroeconomic variables, albeit differently. In some
sense we will interpret change in returns as the result of differential rates of change in
wages by skill. This idea serves as the framework to analyze the real effects of shocks
and policy swings on earnings by skills.
Finally, we also examine which (and to some extent how) our main macroeconomic
variables (GDP, unemployment, trade and capital accumulation) may have affected the
time trends in wages at different educational levels.
The rest of the paper is organized as follows: section I briefly discusses the data
and shows some summary statistics for our sample. Section II gives some background
information on education, occupations and earnings in Argentina over the nineties and
the crisis years, briefly discussing the main macroeconomic shocks and policy swings
experienced by Argentina. Section III outlines the empirical framework. Section IV
shows estimates of the earnings profiles and the calculation of rates of return to education in Argentina for a 11 year period. Section V shows additional results in which
we estimate partial linear regression models and different polynomials to check the robustness to changes in the functional form. Education is now treated as endogenous
and we control for selectivity in the decision of participate or not in the labour marked.
We also instrument for the occupational variables by the predicted probabilities from
a multinomial logit. In Section VI, we look at plausible correlations of wages trends
with macroeconomic trends, by regressing earnings on a set of macroeconomic variables.
In addition, we test whether other micro variables (changes in labor supply behavior,
public versus private jobs, occupational effects and firm size effects) might have been
factors behind the increasing convexity of the earnings-education profile. Section VII
concludes.
1
Data and some descriptive statistics
Earnings equations on the following pages are estimated using all available individual
and household level observations from the 1992–2003 May rounds of the Argentine Permanent Household Survey (EPH hereafter). The survey is conducted twice per year
5
(May and October) in urban areas by the Instituto Nacional de Estadisticas y Censos
(INDEC). 9
All urban areas with more than 100,000 inhabitants (according to the 1991 Population and Housing National Census) and all province’s capitals are currently covered
by the survey. 71% of Argentina’s urban population lives in the 31 centers covered
by the survey and Argentinas urban areas represent 87.1% of the country (one of the
largest shares in the world). Therefore, the EPH sample represents approximately,
62% of the total urban population (INDEC 2000), (W.Bank 2000a), (W.Bank 2000b)
(W.Bank 2000c). 10 The EPH is a stratified random sampling survey and the structure
of the survey is a rotating panel (i.e. 25% of the sample is replaced in each round).
As for the information of the survey (i.e. employment search, total hours worked and
almost all work-related questions), the period of reference is normally the week before
to the survey, which takes place either in March/April (for the May wave) or in August
(for the October wave). Nevertheless, there are some questions with special reference
periods such as total labor and non-labor income referring to a calendar month previous
to the interview, and income received from dividends, interest and utilities, referring to
12 months previous to the survey. As the year 2003 only has comparable data available
for the May wave, we have decided to take only May waves for all calculations. 11 By
doing that, only 13 out of 31 jurisdictions were left with the same conglomerates surveyed for all Mays between 1992 and 2003. In Appendix B we present the definition
of the variables used in the analysis. Table B2 gives means and standard deviations of
all variables in the survey. Basic statistics of the main variables used in this paper are
presented here below.
The table below shows how the (log of) hourly wage rates, on average, went up until
the first crisis hit the economy in 1995. Then wages started to go down, while GDP was
still going up. Real wages were higher in 2001 than in 1992 as the result of increasing
earnings during the first three years of the Convertibility plan and decreasing earnings
thereafter (see Table B2 in the Appendix of this chapter).
The upward trend in GDP ended in 1999, when the longest recession of the Argentine
economy started. GDP finally starts to recover sharply in 2003 catching up with the
1996 level (in Argentine pesos, AR$). 12
Inequality of wages (as measured by the standard error here) has also gone up over the
period as illustrated by the increasing standard deviation in column (4). The proportion
of self–employed, wage–employees and managers on the labor force has stayed almost
unchanged, as well as the concentration of the labor force in the Greater Buenos Aires
area, except for the a slight increase from 1993 onwards. Informality has increased all
over the period under study. In spite of the various shocks, the Argentine labor force
9
In 2003, a major methodological change was implemented by the INDEC, including modifications
to questionnaires and the frequency of survey visits. So far only a reduced version of the dataset of
the new EPH Continua (EPHC) is available to the public. The number of observations changed from
around 90,000 in the late 1990s and 60,000 in the last EPHs to approximately 50,000 per quarter in the
new EPHC. However, we only use the ’old’ EPH surveys, being May 2003 both the last EPH surveyed
by INDEC and the last observation we use here.
10
Approximately 40,000 households are surveyed in 28 cities. Between 800 and 1,500 households are
surveyed in each one of the 27 interior centers and 4,500 in the Greater Buenos Aires (the main urban
center which represents 12 million people, one third of the country’s population).
11
Given that surveys cover only urban areas, most statistics are not significantly affected by seasonality
issues.
12
See figures for GDP in US dollar terms in Table B.1.1, Appendix B.
6
continued to increase its quantity of education and the number of females in it. The
latter is one of the main features of the Argentinean labor markets over the nineties and
a concern for selectivity issues. 13 For instance, increasing activity rates were chiefly
pushed up by this phenomenon in the first half of the decade. The unemployment rate
underwent a sharp rise after the first international economic shock in 1995, reaching a
rate of 18.4% and remaining relatively high. 14 We explore these issues in more detail
in the next section.
13
We control for this by estimating returns by gender
In fact, Argentina is an exception in Latin America as the variance of unemployment explained by
the output gap is only 0.1%, while for other countries is close to 1 (IADB, 2005)
14
7
8
250,31
243,19
256,63
277,44
288,12
278,37
276,17
264,00
235,24
256,02
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
8.8
-10.9
-4.4
-0.8
-3.4
3.9
8.1
5.5
-2.8
5.8
5.7
17.8
21,5
16,4
15.4
14.5
13.2
16.1
17.1
18.4
10.7
9.9
0.20
(0.40)
(0.78)
(0.41)
(0.81)
0.72
0.21
(0.40)
(0.79)
0.89
0.20
(0.40)
(0.78)
1.07
0.20
(0.40)
1.10
0.20
(0.76)
(0.40)
(0.77)
1.09
0.20
(0.40)
(0.76)
1.15
0.20
( 0.41)
(0.74)
1.16
0.21
(0.40)
1.17
0.20
(0.73)
(0.41)
1.19
0.22
(0.71)
(0.42)
(0.70)
1.20
0.23
(0.41)
(0.43)
0.75
(0.43)
0.75
(0.43)
0.75
( 0.44)
0.75
(0.43)
0.75
(0.43)
0.75
(0.43)
0.75
(0.44)
0.74
(0.44)
0.75
(0.44)
0.74
(0.45)
0.72
(0.44)
(0.50)
0.48
(0.50)
0.51
(0.50)
0.52
(0.50)
0.53
(0.50)
0.54
(0.50)
0.55
(0.50)
0.56
(0.49)
0.58
(0.49)
0.60
(0.49)
0.65
(0.49)
0.69
(0.49)
Table 1: Summary statistics.
Occupations
self
wage formal wage
(5)
(6)
(7)
0.22
0.74
0.65
1.07
(0.69)
Wages
ln(w)
(4)
1.03
(0.17)
0.03
(0.17)
0.03
(0.18)
0.03
( 0.19)
0.04
(0.18)
0.03
(0.19)
0.04
(0.18)
0.03
(0.18)
0.03
(0.19)
0.04
(0.18)
0.03
(0.20)
0.04
(0.19)
(0.46)
0.30
(0.47)
0.33
(0.47)
0.33
(0.47)
0.33
(0.47)
0.34
(0.47)
0.32
(0.47)
0.34
(0.48)
0.35
(0.48)
0.36
(0.47)
0.32
(0.46)
0.31
(0.46)
(0.47)
0.34
(0.48)
0.37
(0.48)
0.36
(0.48)
0.35
(0.47)
0.34
(0.48)
0.35
(0.48)
0.36
(0.48)
0.36
(0.48)
0.35
(0.47)
0.34
(0.47)
0.34
(0.48)
(0.32)
0.11
(0.32)
0.12
(0.31)
0.10
(0.30)
0.10
(0.30)
0.10
(0.30)
0.10
(0.30)
0.10
(0.37)
0.16
(0.36)
0.15
(0.26)
0.08
(0.26)
0.08
(0.28)
Weighted sample averages
Tenure
manager ten1 ten2 ten3
(8)
(9)
(10) (11)
0.04
0.30 0.34 0.08
( 0.50)
0.57
(0.49)
0.59
( 0.49)
0.59
( 0.49)
0.60
(0.49)
0.60
(0.49)
0.61
( 0.49)
0.62
(0.48)
0.62
(0.49)
0.62
(0.48)
0.63
(0.49)
0.62
(0.48)
Other
male
(12)
0.63
1797
3192
3740
3724
3906
3942
3898
3648
3555
3961
4072
(129.80)
59.16
(138.04)
64.38
(137.93)
64.00
(142.68)
66.26
(145.29)
70.56
(148.14)
72.19
(144.57)
68.20
(149.82)
73.12
(143.69)
68.83
(136.47)
63.85
(137.68)
63.56
(139.01)
(3.74)
10.84
(3.78)
10.86
(3.75)
10.67
(3.77)
10.53
(3.79)
10.44
(3.84)
10.35
(3.84)
10.24
(3.88)
10.13
(3.90)
10.11
(3.78)
9.92
(3.78)
9.88
(3.81)
worker’s characteristics
GBA firmsize edu
(13)
(14)
(15)
2971 61.48
9.74
column 13, figures are given in absolute numbers. See Appendix B for further definitions.
rate as measured in the May wave. Tenure between 1 and 5 years is ten1, ten2 is tenure between 5 and 20 years and ten3 is more than 20 years of tenure. In
characteristics for the main occupation. Average age is 38 for every year, potential experience (age-schooling-6) is 21 for every year. ”Un” is the Unemployment
Columns (4) to (15): Source: Own calculations based on EPH, May waves, 14–65 year–old with positive earnings. Wages, occupational category and other
Columns (1) and (2): Source: Official figures, Dirección Nacional de Cuentas Nacionales (Ministery of Economy). Constant prices 1993 in Argentine pesos
(AR$). g stands for growth rate of real per capita GDP.
236,51
1993
Macro variables
year GDP
g
Un
(1)
(2)
(3)
1992 222.31 9.6
6.9
2
Education, occupations, wages and shocks: 1992-2003
Argentina has one of the most developed education systems in the Americas. Indicators
show that despite the recent economic crisis, school enrollment rates are high and drop
out levels are low (Parandekar, Espana, and Savanti 2003). 15 Average years of schooling
of the population are 8.5, significantly higher than the Latin American average of 5.9
years. 16
Argentina also compares well with East and Central Europe and East Asia, where
average educational attainment is 8.4 years and 7.6 years, respectively (Barro and Lee
2000). In fact, 16 per cent of the active population had higher education in 2003 (17 per
cent in 2002), while that figure was only 11 per cent in May 1992. We see in Figure 1 and
Table 1 that years of education for all (active, employed and unemployed population)
had increased in about a year, over 11 years (10 % increase overall, or about 1% per
year).
9
years of education completed
9.5
10
10.5
11
Figure 1: Average years of education:1990-2003
1990
1995
2000
2005
year
EMPLOYED
ACTIVE
UNEMPLOYED
Source: own calculations, EPH, Permanent Household Surveys Argentina. May waves. 14-65 year old
population.
15
Enrollment of 6-14 year-old children is about 100% all over the period under our study.
Argentinas case is also particularly interesting because of the following peculiarities of its schooling
system: (i) it is mandatory to attend until high school, (ii) the government provides education without
tuition at all levels, including University, (iii) there is no kind of restriction to the admission to a public
school (the only restriction for universities is to have high school level completed), and (iv) private
education is available at each level.
16
9
The rise in average years of education of the population in a context of rising unemployment and falling wages has to be understood in the context of secondary schooling–
12 years of education–becoming a necessity to access jobs and adequate wages (i.e workers with incomplete secondary schooling will not earn significantly more than those with
complete primary education, see Table B2). This effect acts together with the lower
opportunity cost of schooling (i.e. less foregone income given the high unemployment
rate, in particular for the semi-skilled).
In table B3a and B3b we show the percentage of the labor force in each education
level by gender and by occupation, respectively. We observe the gap between female
and male college completers in the labor force increasing steadily. In 1992: 7% of males
in the labor market completed a superior degree while 17% of the females did so. In
2003 these figures were 11% and 23%, respectively. This may raise the issue of selectivity, as only the more educated females have been self-selecting into labor market’s
participation. In Table B3b we observe the biggest increases in investment in education
among managers and wage–employees: 18% of managers, 7% of self–employed and 12%
of employees completed a superior degree in 1992. In 2003, those figures were: 27%,
13% and 17%, respectively.
On the other hand, Table B4 shows the trend of wages and hours worked by level of
education. The more notable point in this table is the loss in wages of secondary completers, their 2002 (2003) wages were 80% (65%) of their 1992 wages; while the wages
of primary completers were 84% (74%) of their wages 10 (11) years earlier. College
completers lost only 5% (20%) over this 10 (11) years. 17
When looking at wages trends by occupations, we remark from Table B5 that the
self-employed have significantly lost compared to other groups. In 2002, the real hourly
wages of managers were 87% of their 1992 wages. That figures was 96% for wage earners
and only 84% for self–employed. In 1992 the average total earnings of the latter group
were higher than the earnings of salaried workers (mainly due to a higher number of
hours worked). The relative loss for the self-employed has occurred in terms of both
hourly wages and hours of work. While earnings significantly increased on the 1990s for
the self-employed professionals, labor income substantially dropped for self–employed
workers of low education levels.
We will then study these trends in four well differentiated sub-periods which encompassed one policy swing and two different shocks:
1992-1995 (”Trade opening and economic reforms policy swing”): 18 where stabilization, together with massive privatization, opening of the economy, and free–market
oriented structural reforms were implemented. Low inflation, high growth rates, massive
entrances of foreign capital and low real exchange rate ensued. 19
The second sub–period is 1995–1999. In 1995 (”1st. external/supply shock”), the Mexican crisis affects Argentina‘s credibility to back the exchange rate and a massive flow
17
I am purposely reporting figure for 2002 and 2003 because the change in wages between these two
years is quite dramatic
18
We will use import penetration ratios and capital accumulation to identify these changes. Even if
it the trend starts in the early nineties, both import penetration and capital accumulation kept going
up until 1999.
19
This sub-period actually starts in 1991, with the implementation of the Convertibility Plan at the
end of the hyperinflation of 1989/1990. Unfortunately, the available data for 1990 and 1991 is not
comparable with ours.
10
of capital out of the country resulted, together with high increases in interest rates, and
decreases in liquidity. As a result, aggregate demand decreased by 2.8%, unemployment
increased by 8.3 percentage points, but altogether the 1995 recession was short and positive growth (on average 5.8% per year) continued until 1998. 20
The third sub–period starts in 1999, when the Russian crisis strongly affects the Brazilian economy (Argentinas main commercial partner) and halted positive growth in Argentina. This sub-period concludes with the 2001-2002 collapse (”2nd. external/supply
shock”), in which a 40 % devaluation of the peso took place, together with default of
the country’s external debt and a 11 % decrease in GDP (measured in Argentine pesos,
AR$) in 2002. The last year in our sample, 2003, shows the beginning of the recovery
(8.8 % increase in real GDP) and is our fourth sub-period. However, we will not focus
too much on this as it is too soon to analyze the consequences of the post-devaluation
period.
A model of relative wages determination
We will argue that changes in relative demand of labor might have affected relative
wages. Relative wages in the economy are derived from changes in both derived demand
for labor and supply for each type of skill. As Pessino (1996) we assume imperfect
substitution among labor types, an that increases in demand or supply will change the
relative wages. 21 Following Welch and Freeman (Welch 2003, Freeman 1979) and
assuming a CES production function, we have that:
c = (1/σ)(db − sb)
W
(1)
Where W is the ratio of wages for more skilled or more educated labor with respect
to less educated or less skilled, d and s are relative demand and supply respectively for
skilled labor, σ is the elasticity of substitution for these two types of labor and a hat
over a variable indicates rate of change. So, if there is an increase in demand for highly
educated individuals relative to supply, we should expect an increase in their relative
wage as long as the elasticity of substitution is not infinity. This relative wage increase
will be manifested essentially through an increase in the rate of return when one makes
the assumption that high skilled people corresponds to highly educated and viceversa.
We argue that there may be two channels via which we expect our macroeconomic
variables (shocks or policy swings) to affect W .
Firstly, if elasticities of substitution among age groups, experience vs. unexperienced
groups, educated vs. uneducated are non infinite, human capital cannot be treated as
a homogeneous input with a single rental price. In this sense, capital deepening, trade
and skill biased technological change might change our σ parameter. Given how the
composition of output was affected over this period, we can also expect a change in
labor demand that was not uniform across sectors of the economy. 22
20
A set of support measures from the IMF together with different mechanisms of government support
to commercial banks helped to stopped the 1995 financial crisis and help to restore positive expectations.
Because of that, the convertibility regime remained in place.
21
There is also imperfect substitution between similar educated workers in different age groups. Manacorda et al (2006) show that there was a des-acceleration of the growth rate of educational attainment
across cohorts (i.e change in relative supply of highly educated workers across age group lower than the
relative supply change).
22
Of course, other type of shocks, such as real exchange rate shocks will affect the composition of
output between different sectors of the economy and this raises the issue of what type of shock caused
what.
11
The second channel through which I expect the wages to change differently hinges
on the differentiated capacity individuals have to smooth consumption (given their differentiated access either to capital markets or to household insurance) and therefore on
how they might change labour supply and occupation decisions. If an individual belongs
to a wealthy household, consumption can be smoothed over through transfers between
family members. If an individual does not belong to a wealthy household (or if the
wealth that used to exist has been depleted), the whole labor supply decision of the
family will change more dramatically (I use household size as a proxy for labour supply
behaviour and a determinant of the reservation wage). I expect women of households
of this type to increase their labor supply more than men, unless their earnings has
been so depressed that the wealth and substitution effect cancel each other. Decisions
of the young and the old will also be altered. In the case of the young, where more
fundamental investments decisions are being taken, the whole human capital accumulation decision may change labor supply. If they belong to a wealthy household, they will
probably increase their investment in schooling (if real wages are depressed); although
the expected rate of return to education will also affect their schooling decisions. With
respect to the old, I expect to find increases in labor supply for the less wealthy and
those who do not have children, or whose children are unable to support them. 23
Another aspect (that could affect both elasticities and somehow with smoothing) is
the sorting across firms, occupations, sectors and quality-schools. Fafchamps et al show
how much of the total returns to education in 11 African countries is due to sorting across
firms and across occupations within firms (Fafchamps, Soderbom, and Benhassine 2006).
Also see Soderbom et al for an investigation of the relationship between earnings and
firms size (Söderbom, Teal, and Wambugu 2005).
It has also been well documented the existence of inter-industry wage premium
(Dickens, Katz, and Lang 1986, Krueger and Summers 1989). In another paper, Katz
argues that much of the shift in relative demand in the US can be accounted for by
observed shifts in the industrial and occupational composition of employment toward
relatively skill-intensive sectors, the majority reflecting shifts in relative labor demand
occurring within detailed sectors. These shifts are likely to reflect skill-biased technological changes (Katz and Murphy 1995). Another source of heterogeneity we will not
be able to address here is the type of school these individuals have attended. Andres
shows that workers educated in private schools have higher returns than those in public
schools and the quality of schooling significantly affects returns (Andres 2003).
Is in this theoretical framework, together with the preliminary empirical evidence
just provided, in which we want to answer the following questions: What has been
the implications of increasing supply of education for the returns to education in this
economy? How do returns to education change in a middle income economy, dominated
by wage employment ? Also, how have different types of shocks changed the returns to
education across occupations? Now we turn to our baseline model of earnings.
23
Actually, unemployment rates rose between 1992 and 1998 partly due to the fact that spouses and
youngsters decided to start seeking for a job (’added– worker effect’). This fact suggests that part of the
causality could have been from inequality to unemployment: the drop in wages of low-income household
heads triggered a jump of their relatives from home to the labor market, a fact that could have fed the
increase in unemployment.
12
3
3.1
Earnings model and Identification strategy
Basic equations
One of the aims of this paper is to consistently estimate the earnings-education profile,
and to investigate if there is any evidence of changes in that profile that has been
affected by shocks (or policies) over time. For that purpose the baseline model is based
on Mincer equations (Mincer 1974). We expect the trend of our beta coefficients for
the educational levels and the time variable to reflect how aggregate changes did impact
differently on wages of different individuals. Despite the increasing evidence of convexity
in the earnings-education relationship, Mincerian returns remain popular and have been
widely estimated. We follow Pessino (1993, 1996), Gasparini (2001) and Fizbein (2006)
and start by estimating the by-levels function specified here below: 24
lnWit
exp2it
+ β3 P ricit + β4 Seciit + β5 Seccit
100
+β6 Supiit + β7 Supcit + regiondummies + µit
= α + β1 expit + β2
(2)
For the estimation of this model, the dependent variable is the log of hourly real
wages (at 1998 prices). The independent variables are exp which is the potential expe2
rience measured as: age – number of years in education – 6, exp
100 which is the square of
potential experience by 100 (to facilitate interpretation), P ricit , Seciit , Seccit , Supiit
and Supcit refer to dummy variables for: primary complete, secondary incomplete, secondary complete and Superior (including college) incomplete and complete, respectively
(P riiit is primary incomplete or no education and is the regressor). Region dummies
for the 13 regions used in this study are also included but not reported. Finally µit is
the error term and i and t denote individual and time respectively.
We use other specifications and include a quadratic function of age (instead of potential
experience) in the same equation we will add tenure dummies, a dummy for self employment (inteacted with education to capture the heterogeneity of this group) and a
dummy for informality.
We therefore did not include in equation (2) variables that may be channels through
which education affects earnings, e.g. firm size or sector. Later, in the specification
shown in Table 13 to 15 we add dummies for male, firm size, a dummy for whether
the person works in the public sector, number of persons in the household, sector and
occupation fixed effects because we want to see how much is captured by those variables.
Finally, we also estimate equation (2) including time dummies (in order to capture
the individual–invariant time effects) and interacting our education variable with time
(in order to capture the educational–and–time–variant effect), as in the specification
shown below.
lnWit
= α + βt t + βe expit + βe2
exp2it
+ Σg (t ∗ DSigt ∗ βgt )
100
+regiondummies + µit
(3)
24
For Argentina, Gasparini also includes a male variable, a quadratic function of age and a dummy
for youngsters less than 18 year old. He performs this regression on the sub–samples of heads, spouses
and others members of the household between 14 and 65 year–old.
13
where t is time, DSigt is a dummy variable that indicates a schooling group g in period
t and βgt is a schooling effect in period t. 25 In the next sub-section we explain how we
will measure Mincerian returns.
3.2
Rates of Return
Rates of Return will be defined here as the Mincerian returns, or wage premium (Mincer,
1974). In the case of the log-lin specification in (2), our returns per additional year of
education will be defined as follows:
c3 ) − 1]/SP ric
RoR(P ric)t = [exp(β
(4)
c4 ) − 1]/SSeci − SP rimc
RoR(Seci)t = [exp(β
(5)
c5 ) − 1]/SSecc − SSeci
RoR(Secc)t = [exp(β
(6)
c6 ) − 1]/SSupi − SSecc
RoR(Supi)t = [exp(β
(7)
c7 ) − 1]/SSupc − SSupi
RoR(Supc)t = [exp(β
(8)
where SP rimc , SSeci , SSecc , SSupi and SSupc are the total number of years of schooling
for each successive level of education. In the Argentine case these are (not considering
drop outs case): 7, 10, 12, 15 and 17, respectively. For instance RoR(P rimc)t would
be the return to Primary complete over Primary incomplete or no education (which are
basically drop outs). We report these returns by levels of educations and by occupation
in the next section for different specifications of earnings models.
3.3
Semi-parametric estimations, polynomials and the Dummy Variable Approach
In order to grant more freedom to the relation between the variables and to allow
for non-linearities in the education-earnings profile, we will now use the continuous
variable in education (S=years of schooling completed) and will estimate this function
semi-parametrically. We will use the partial linear regression model first suggested by
Yatchew (1997). This estimation combines parametric with non-parametric techniques,
by implementing the difference-based algorithm for estimating partial linear regression
models.
As particular cases, we will polynomials (square and cubic) and the Dummy Variable
Approach in the education earnings profiles. The basic polynomial equation (for the
pooled data) is:
lnWit
= α + βt t + βe expit + βe2
exp2it
+ f (Sit ) + regiondummies + µit
100
Here we have introduced the non-linearity via f (Sit ). The function f is a smooth,
single valued function with a bounded first derivative. In this model the parametric
(Xβ) and non-parametric f (Sit ) parts are additively separable. We also estimate an
equation with one dummy variable for each year of education completed (S=1, 2 ...19).
This (dummy variable) approach, in contrast, gives a function that ”jumps” each time
one moves from one level of education to another.
25
We did not interact experience and experience squared as our results from (2) (see Table 7) showed
there was no significant change over time in those two variables’ point estimates. Therefore, we take
these variables at their average level in this equation.
14
3.4
Endogeneous education, selection and occupational choice: strategies (in progress)
In terms of the specifications we chose here, it is widely recognized that using OLS
to estimate returns to education from cross sections is problematic. Education can be
correlated with the earnings residual due to unobserved ability. And it could be that
unobserved ability is correlated with the returns, we therefore allow for this endogeneity
in section 5. As is of common knowledge, the OLS estimator will give biased estimates
of the returns to education if education is ”endogenous” and also if we have sampleselection issues.
The common concern is that education could be correlated with unobserved labor
market ability, and that the returns would be upward biased. 26 Potential instruments
to correct for endogeneity has to be variables that are correlated with education and
uncorrelated with the earnings residual. Family background variables have been used
as instruments for education in many previous studies, primarily on the grounds that
such variables should have no causal effect on earnings. Social and natural experiments
are also useful and many studies using institutional variations in schooling due to such
factors as proximity to schools, minimum school-leaving age etc. have been used to
instrument for schooling. Card (1995, 1999 and 2001) provides a summary of some
of the recent studies that use this approach and include Angrist and Krueger, 1991,
(Butcher and Case 1994) Card, 1995a, and Harmon and Walker, 1995, among others. 27
In all waves in our data there is information on parent’s education for the sub-sample
of youth still living with their parents at the moment of the survey. The percentage
of less–than–30 living with their parents in our sample is about 75% when taking all
individuals, but it is 47% when taking our sample of interest-all those employed between
14 and 65 year–old. Due to this severe selectivity, we have decided to discard this
strategy. An alternative to the IV technique is to either use repeated observations
on the same individual over time (THE USE OF THE PANEL COMPONENT OF
THE EPH TO ESTIMATE RETURNS TO EDUCATION IS STILL IN PROGRESS)
or observations from different individuals within the same family to difference out the
variables generating correlation in the residuals in a fixed effects approach (Ashenfelter,
O. and Zimmerman, D. J.,1997). Arguably, a good part of the unobserved heterogeneity
is common to family members. Consequently, differences in unobserved ability and their
impact in determining education should be lower within rather than between families.
Earnings functions can be estimated on twin-samples, siblings, father-son or motherdaughter pairs using a fixed effects or first-differencing approach. By introducing subsamples of households with at least two individuals of a given gender in employment (and
more stringently households with brothers/sisters, father-son or mother-daughter pairs
in employment) the fixed effects method effectively controls for all household variables
that are common across these individuals within a given household. 28 //
Finally, within a basic Harris-Todaro framework I allow for both skills and unob26
A common finding in the empirical literature is that estimated returns rise as a result of treating
education as an endogenous variable (Card 2001).
27
Butcher and Case analyze in depth the effect of sibling composition in educational achievement.
28
As a robustness test, we took characteristics of the house at the household level as wealth proxies
which are not supposed to have causal effects on earnings either (this could be argued further). Variables
like house type, number of rooms in the house (excluding kitchen and bathroom), number of nonshared rooms in the house and the sort of tenancy agreement (whether owned or rented) were used as
instruments (we are not reporting these results, but they are available under request).
15
served high ability types. The model should predict for the relative rates of formal
employment, informal employment and unemployment for each skill group. To test the
model - i.e., to understand the interaction of human capital and wage-setting institutions
in producing earnings differences - requires earnings data on individuals across skill in
both the formal and informal sectors (as I have). I will model occupation sorting using
a latent variable approach. Here the latent variable is the propensity to sort into a given
sector (i.e the instrument variable) . Only the occupation outcome and not the underlying propensity are actually observed. The key assumptions that underlie the model
are the determinants of the propensity to sort into sector. We assume that both age
and education determine this sorting decision. Education by increasing human capital
makes a job in the formal sector more likely, it is assumed, because such human capital is
more valuable in the formal sector. Age matters as it is related to general labor market
experience which, if valued differently across sectors, may lead to sectoral reallocations
over time. While both education and age are observable this model assumes that there
are characteristics of the individual/ job which can be observed by the individual but
not by me. The most commonly used approach to dealing with selection in a model of
this kind is that of Lee (1983) who first proposed a generalisation of the two-step selection bias correction method introduced by Heckman (1979). If there is sorting along
the lines hypothesised here then in principle all the results of the OLS earnings function
will be biased. In particular any finding of sectoral differences in earnings between the
formal and informal sector need not imply that a Harris-Todaro model is relevant for
understanding these differences, they may arise from a Roy sorting model.
3.5
The effects of macroeconomic variables on wages by skills
Macro shocks can affect cohorts differently depending on the rate of substitution between experienced and unexperienced workers, educated and uneducated ones, different
expectations about their level of future wealth, and the time in their life cycle at which
the different shocks occurred. The main assumption in the literature is that groups of
workers categorized under some definition are different factors in the production function. Standard human capital models of the age-education-earnings profile of a cohort
which posit that earnings rise with age or experience solely as result of individual investment behavior are incomplete. For instance, difference in the activities of different
type of workers and in the demand for those activities decisively influences the shape
of that profile. Crises certainly may have affected the way wages were bargained in the
informal sector and also how devalation and inflation in 2001 may have played a role in
increasing the wage flexibility of the self employed and the informal wage–employees.
One particularly interesting period in the data is 1992-1998. In this period GDP
growth was benefiting far more the college completers. That leaves room to further
explore the argument of increasing demand for skills. There is widespread agreement
on the fact that in developed countries there has been a shift in demand away from
unskilled labor in favor of skilled workers during the last two decades. Two competing explanations have been proposed to explain this shift in the relative demand
for skilled labor: the impact of trade on low-wage (developing) countries, and skillbiased technological change (Berman, Bound, and Griliches 1994, Berman, Bound, and
Machin 1998, Machin 1996, Wood 1995, Galiani and Sanguinetti 2003). We will be
exploring both hypotheses in section 6.
16
4
Estimates
In order to predict the effects of macroeconomic variables in the structure of wages we
will proceed in the following manner: first we will present the estimated wage equations
by level of education and then by gender for different occupations together with the
calculation of returns. Second, we will see if the effects of shocks or economic reforms
show a consistent pattern; that is whether effects encountered in the early nineties
(opening of the economy and structural reforms), 1995 (financial crisis) or 2001–2002
(collapsing GDP, devaluation) are really ”outliers” from otherwise smooth series on the
rate of return to different human capital arguments.
4.1
Wage profiles
Estimates of the earnings equation specified in (2), by year, are reported in Table 7. We
take year 1992 and primary incomplete workers as the base year/category; and estimate
the same regression with year fixed effects. From those equations we predicted the
average real earnings, by education level. These are showed in Figure 2. 29
29
Results of the specification using age instead of experience and adding dummies for male and tenure
as well as the one including a dummy for self-employed interacted with education to equation (2) are
available under request.
17
.5
predicted log real hourly wage
1
1.5
2
Figure 2: Predicted average real wages, by level of education: 1992-2003
1990
1995
2000
2005
year
PRIM
COLL
SEC
Source: own calculations, EPH, Permanent Household Surveys Argentina. Dependent variable is the
log of (positive)hourly wages at 1998 pesos prices on employed people for the cities of: La Plata,
Córdoba, Paraná, Comodoro Rivadavia, Neuquen, Jujuy, Rı́o Gallegos, Salta, San Luis, Santa Rosa,
Tierra del Fuego, Capital and Conurbano Bonaerense. May waves based on Table 16
18
We can see how wages for every education level show a positive trend up until 1994.
For college graduates, wages fall in 1996, but recover quickly in 1997 up until 1999,
when they stagnate until 2001. Only in 2002, predicted wages fell by 18% with respect
to 2001, and kept falling until 2003, summing up to a total fall of 11% over the period
under study (1992 is the base year). For those workers with up to secondary school,
wages are decreasing very slowly between 1995 and 2001. Their wages fell by 21% in
2002, adding up to a total fall of 45% for the whole period. For primary complete, post1995, there is only an increase in 2000 (from 2.52 to 2.69 Argentine pesos per hour).
The crisis hit them harshly (- 23% change) and they experiment a total fall of 41 %
over the period. For primary incomplete (not shown in the graph) there is only a slight
increase in 1998 (from 2.45 Argentine pesos per hour to 2.47). The first thing we notice
particularly in 1995 (see the ineteaction of the college completer dummy with thime vis
a vis other levels of education in table 16) is that the more the stock of general human
capital, the less is the impact of the shocks in real wages. This may be related to the
fact that education may served as a tool to secure a job over a crisis. 31 Figure 3 (based
on regressions in table B6) shows predicted earnings by occupations. Wages increase for
all occupations until 1994. In 1995, self-employed see their wages decrease much faster
than wage-employees and managers. In 1997 and 1998, in spite of the recent crisis,
wages increase for managers and self–employed, while earnings of the wage–employee
stagnate and those of the informal waged workers kept falling. Overall, wages for all
occupations have a decreasing trend from 1996 onwards, except for the last year in our
sample, 2003.
30
30
The upward trend in wages started in 1991. The implicit increase in wages brought about by the low
real exchange rate in 1991 was aggravated when considering the relative price of capital goods versus
labor, since wages continued to be taxed at high rate, while investments in physical capital become
cheaper through basically zero tariffs to the imports of intermediate capital goods. When one considers
labor in the aggregate, this increase in its relative price, should imply a decrease in labor relative to
physical capital demand. The relative price of labor increased by 40% between 1990 and 1993 according
to the Economic Programming Secretary (Secretaria de Programacion Economica).
31
Probit results are available under request. Also, see Gould, Moav et al (2001) for a model that
endogenously generates the patterns of wage inequality (within and between groups) and educational
attainments seen throughout the last few decades. Their model is based on the disproportionate effects
of technological changes on the depreciation of general versus technology-specific skills, and the resulting
precautionary factor in the demand for general education which guards against the higher depreciation
risk of technology-specific skills. They assume that individuals, given their level of ability, choose to
invest in general skills through education or in technology-specific skills through on-the-job training.
Since the return to ability is higher as an educated worker, higher ability individuals choose to invest in
general education and workers with lower ability choose to invest in technology-specific skills. However,
changes in technology render technology-specific skills obsolete. Consequently, less educated workers,
who are relatively more invested in technology-specific skills, will suffer higher rates of human capital
deterioration due to technological improvements. Therefore, an increase in the rate of technological
progress will increase the education premium. (Gould and Weinberg 2001)
19
.5
.6
predicted log real hourly wage
1
predicted log real hourly wage
.8
1
1.2
1.4
1.5
1.6
Figure 3: Predicted average real wages, by occupation: 1992-2003
1990
1995
2000
2005
1990
1995
Year
2000
2005
Year
selfemp
wagemp
selfemp
wagempinf
manag
manag
wagemp
Source: own calculations, EPH, Permanent Household Surveys Argentina. May wave
Another remarkable point is that the largest gap in earnings across occupations occurs between self-employed secondary and self-employed college graduates (see Figure
5 in Appendix). This is consistent with the fact that the self employed are a very heterogenous group that can range from street vendors who use small amounts of physical
capital to an entrepreneur producing goods with varying amounts of equipment. The
gap between categories only starts increasing in 1995, because of the hourly wages of
the self-employed and the informal workers falling faster than those of the wage employees. From 2001, however, we observe a very similar rate of change in wages across
occupations.
Time trends by levels of education In Table 16 (eq. (4) time dummies estimates
show the effects of shocks not explained by any of the explanatory variables included.
When we look at the time trend (column (1) in Table 2) we see the common downward
trend in wages already discussed in section 4.1. The trend by education level (columns
(3), (5) and (7)) shows secondary completers as the big losers of the period, mainly
due to the losses they suffer over the last part of the decade. 32 There is also a fairly
significant negative trend for secondary incomplete workers over time (see Table 16).
32
Wages for those with primary complete fell by 40%, for the secondary completers by 45% and for
college completers by 11% for the 11-year period (1992 is the base year).
20
Table 2: Time trends: overall and by education level (base: 1992, Primary incomplete)
year time trend pric (%) time-pric secc (%) time-secc supc(%) time-supc
(1)
(2)
(3)
(4)
(5)
(6)
(7)
1992
100
0
100
0
100
0
100
1993
109.1
-5
104.1
-7.2
101.9
-3.1
106
119.1
-5.6
113.5
-5.3
113.8
5.4
124.5
1994
1995
112
-1.8
110.2
-1.7
110.3
23.9
135.9
1996
109.8
-4.9
104.9
-2.3
107.5
20.7
130.5
1997
101.9
-0.3
101.6
1.4
103.3
27.2
129.1
1998
105
-4.4
100.6
-5.5
99.5
26.3
131.3
1999
105.9
-9.5
96.4
-6.6
99.3
21.1
127
97.4
-0.6
96.8
0.3
97.7
26.7
124.1
2000
2001
99
-7.6
91.4
-6.3
92.7
25.1
124.1
2002
71.9
-1.3
70.6
1
72.9
29.4
101.3
2003
69.2
-10.1
59.1
-14.5
54.7
19.6
88.8
Note: All figures in this table are derived from Table 16 in the Appendix. The ”time trend” index in
column (1) is derived from the dummy time trend in the data. The percentage changes in columns (2),
(4) and (6) are the changes in the returns to a given level of education with respect to a 1992 worker
with primary incomplete. These percentage changes are taken from the (highly significant) coefficients
of the interaction of level of education and time, and then multiplied by 100. The indices in columns
(3), (5) and (7) come from adding the percentage change in wages for a given level of education from
the general ”time trend” in column (1). For instance, column (3)=(1)+(2), column (5)=(1)+(4) and
column (7)= (1)+(6).
If we only take the period 1992-2002, wages for primary completers decreased by
29.4% 33 for secondary completers wages fell by 27.1% and for college completers wages
increased by 1.3%. After controlling for experience, region fixed effects and education,
college completers lost 12.5% of their wages in only one year (2002/2003). When taking
the period 1992-1998, primary ccompleters and secondary completers remained almost
unchanged with respect to 1992, while those that finished a college degree gained 31.3%.
Finally, if we look at the first four years in our sample we observe that, while primary school completers wages increased by 10.2%, those finishing secondary have seen
an increase of 10.3% and those finishing college have seen an increase in wages of 35.9%.
Overall, these figures suggest that after each supply crisis, all educational categories
have lost about the same percentage. However, the type of crises matters, as the loss
was about 5% for all in 1995 ; and about 20% in 2002 (or 30% from 1999-2002).
33
Even if the individual-invariant time effect from 1992 to 1994 ”helped” them not to loose even more.
21
Other controls The results for the specification with controls are presented in table
13 to 15. When controls are introduced for firm size, a dummy for whether the person
works in the public sector, number of persons in the household, sector and occupation
fixed effects we can observe a significant decrease in the incomplete and complete college
education dummies coefficient in all tables. This might be explained by the fact that
higher returns to those workers (in the non-controlled equation) were premiums to the
size of the firm or the particular sector (or occupation) where the individual works.
The male dummy shows that males are receiving, on average, 10% higher earnings than
females on average. With respect to firm size, for each additional person working in the
worker’s firm, an individual is receiving a 3 to 4% premium on his wage, on average.
Sector and occupation fixed effects dummies are very significant for the pooled equations.
When adding the controls into the square, the cubic and the dummy specification, results
are similar.
These results suggest that the size of our estimates of the returns should be taken as noisy
estimates of the actual rates of return. However, we still have increasing convexity over
time (except in table 15, when occupations FE are included. This needs to be explored
further). Household size is very significant and negative and particularly negative and
big over crises (see the estimates for 1995 and from 2000 onwards on Table 13). We
explore later what macroeconomic variables might have played a role in the observed
increasing gap between secondary and college completers. We come back to this point
in Section 6.
22
Table 3: Rates of Return by level over time, full sample(%)
Year
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
Primary vs none
1.8
0.8
1.0
1.0
2.1
3.2
1.5
0.1
1.2
1.7
2.5
0.4
Secondary vs Primary
12.3
8.9
10.2
11.0
14.2
17.3
13.5
11.6
12.7
14.9
16.9
11.0
College vs Secondary
28.1
30.4
35.6
44.2
49.8
58.1
53.5
45.9
51.1
57.4
58.6
56.7
Notes: this table reports the wage returns to primary versus none education, secondary versus primary
and college versus secondary workers. The series are obtained from year specific regressions of log
wages on a constant, potential experience, its square and 5 educational dummies for 6 levels of
education. The series in the table are the exponential of the coefficient on the given educational
dummy minus 1, divided by the number of years of education needed to complete that given level.
Source: own calculations on EPH, Permanent Household Surveys Argentina. May waves.
4.2
Rates of Return
Overall Rates of Return Overall, the returns to primary schooling remained almost
unchanged and small during this decade. Both wage–employees and self–employed get,
on average, 1.5% higher earnings per additional year of education. This result corresponds well to an economy with universal primary education. The (already low) returns
to incomplete secondary schooling also remained unchanged. The returns to complete
secondary education increased towards 1997 and 2002, but the overall change was small.
34 The returns to university education (complete or incomplete) increased dramatically over the period. Only the return to college complete versus secondary complete,
increased from 28.12% in 1992 to 57% in 2003.
In order to interpret this table better, we plot both the differences between the college
and secondary premiums and the difference between secondary and primary premiums
here below. This graph summarizes the evidence presented in this section and provides
validation for the statement that increasing convexity in returns was a novel feature of
the Argentine economy over the nineties.
34
Using a dynamic cohort analysis for Buenos Aires for the period 1980-1999, Margot shows that
workers with secondary incomplete experience rather stable returns, which are on average 12 percent
although decreasing in recent years, reaching 10 percent in 1999. Workers with secondary complete
experience slightly higher returns– at 13 percent on average for the whole period – but very stable over
time and reaching 11 percent in 1999. Workers with complete higher education seem to be experiencing
increasing returns, especially in recent years, reaching 23 percent in 1999 (Margot 2001).These figures
correspond approximately to the ones we obtain in our paper for secondary (complete and incomplete).
23
.1
.15
.2
.25
.3
.35
Figure 4: Evolution of wage premium: 1992-2003
1990
1995
2000
2005
year (4 digits)
diffCOLLSEC
diffSECPRIM
Source: own calculations, EPH, Permanent Household Surveys Argentina. 14–65 year-old employed
d ) − 1]/5 − [exp(βSeccomp
d ) − 1]/5, while
and positive wages, May waves. diffCOLLSEC=[exp(βSupcom
d ) − 1]/5 − [exp(βP rimcomp
d ) − 1]/7
diffSECPRIM=[exp(βSeccomp
Rates of Return by Gender Results in Tables 19 (Men) and 20 (Women) demonstrate that rates of returns to education for men are much higher than for women, at
every level. Returns to primary completion have gone down for women over time, and
stayed flat with some ups and downs for men. Among secondary completers: men experiment a decreasing trend until 2001. After the crisis the returns to them has started
to raise. For women, those returns stayed flat. Returns for both categories rose steadily
among college graduates. They have increased as well for men with superior incomplete
education, but not for women. For college graduates, there is a peak in 2003 for men
and a peak in 2001 for women.
Rates of Return by Occupation Results in charts B1, B2 and B4 in Appendix
B show how the returns to wage–employees college graduates rise steadily, while the
returns to self–employed college graduates react more to changes in the macroeconomic
conditions as already observed in the last section with their wages. For managers, the
gap in the returns to different educational categories did not grow. 35 36
In the same charts, we see that for a wage–employee completion of secondary school
meant, on average, a return between 10% and 15% per additional year, while for a self–
employed it meant a return between 7% and 20% premium for additional year. Clearly,
complete university education has a higher rate of return (between 30% and 60% for
wage–employees and between 30% and 70% for the self–employed).
35
These charts are the result of calculating equations (5), (6) and (9) after OLS estimations.
Here, there is a clear endogeneity problem as we are assuming that one chooses the level of education
once the occupation has been already chosen. We will deal with this in an extension of this chapter
36
24
5
Robustness tests: functional form, endogeneity and selectivity
5.1
Semi-parametric estimations, polynomials and the Dummy Variable Approach
We report in Tables 17 and 18 the estimates and the changes in the interaction of S 2
and time of the polynomial versions (square and cubic) by occupations and for the
pooled data. 37 Both the square and the cubic function are mostly well accepted by
the data. However the cubic shows higher levels of significance. Results for the full
equation (including controls) and by year are not reported, but available upon request.
In that table the signs of the coefficients display as expected for the pooled data (positive
for the linear, negative for the square and positive for the cubic). When splitting the
sample by occupations, changes in the squared education coefficients turned out to be
non significant for the managers sample (except for 1993 and 1996) and the same is true
for changes in the cubic coefficient. The self–employed display significant coefficients (at
the 99% level) for the linear, the square and cubic terms. For the wage-employees, the
cubic is still significant at 99% level, while the square is only significant at 90% level,
and the linear term is non-significant.
Changes in the cubic term estimates are also significant for most of the years. In
particular for the wage employees and the pooled data (exceptions are 1993, 1996, 1999
and 2002, and 1999 and 2002 for the wage employees)
Changes in the square term (for a specification including only the linear and the
square term as showed in Table 17) are very significant for every year for the pooled
data and the wage employee series. Self–employed do not have significant coefficients
neither from 1996 until 1999 nor in 2002/2003.
Results in this section, by means of our cubic and square coefficients and the dummy
variable approach (not reported but available under request), seem to confirm that there
has been a significant shift in that profile and in the next sub-section we show that shift
was robust to endogeneity.
5.2
Endogenous education
Here below we present results from our household fixed effects on the sub-samples of
workers within a household who are related in any way (e.g. father-daughter, motherson, brother-sister or grandomother/son-mother/father or grandaughter/son) . Results
below are for the ’all type of relations’ equation (see estimates in table 10, for all type
of relations. 38 .
37
It has to be noticed that the question referring to completed years of education was not asked in the
questionnaires prior to 1995. Therefore, for 1992-1994, a categorical variable showing levels of education
was used to impute values to our continuous schooling variable, S.
38
The divisions are based on the notion that any relation may not have the same genetic ties as
blood relations. Although our sample of ”All family” relations in household fixed effects estimates
excludes non-blood relations such as parents-in-law and any servants residing in the household, it includes
grandchildren. To increase the robustness of the estimates, we divide individuals into tighter groupings:
sibling pairs (brothers and sisters) and parent-child pairs. The father-son pairs are, more specifically,
male children of the household head. The mother-daughter pairs are the female children of spouses of
the household head
25
Table 4: Rates of Return by level over time, HH fixed effects, all types of family relations.
Pooled data(%)
Year Primary vs none Secondary vs Primary College vs Secondary
1992
2.9
15.4
34.5
1995
1.5
11.8
38.7
1996
1.6
13.4
40.3
1997
2.6
15.2
46.4
1998
2.5
14.5
48.2
1999
0.8
11.2
36.7
2000
0.9
11.1
36.2
2001
2.0
12.9
45.1
2002
1.9
12.8
40.2
2003
0.9
11.0
33.3
Notes: this table reports the wage returns to primary versus none education, secondary versus primary
and college versus secondary workers. The series are obtained from year specific regressions of log
wages on a constant, potential experience, its square and 5 educational dummies for 6 levels of
education, plus household fixed effects on the sub-samples of wage-workers within a household who are
related in any way (e.g. father-daughter, mother-son, brother-sister). The series in the table are the
exponential of the coefficient on the given educational dummy minus 1, divided by the number of years
of education needed to complete that given level. Coefficients are presented in TAble 9
Source: own calculations, EPH, Permanent Household Surveys Argentina. May waves.
FE estimates are lower than the OLS estimates, a finding consistent with previous
literature. Some part of the attenuation could be due to measurement error. However,
the FE findings confirm that the convexity of the education-earnings profile in previous
section is not an artifact of heterogeneity. We have to add two qualifications here. First,
that the the gap between secondary and primary returns, instead of increasing slightly
now stay rather flat, which supports even more the argument of increasing convexity.
Second, the gap in returns between college completers and secondary completers shows
the same increasing trend stopping in 1998 as before, but two years in the sample showed
movements in opposite directions. The year 2000, when the difference stayed the same
as in 1999 for the fixed effects estimation of the returns (as opposed to an increasing
trend in the previous estimation); and 2003, where the gap increases in the fixed effects
estimation and decreases in the non-fixed effects estimation. Moreover, the size of the
gaps are bigger in the household fixed effect estimation, even if the coefficients are
smaller (due to OLS being upward biased). This evidence supports the idea that the
shift in convexity, seems to be robust to endogeneity, at least for the first six years of
the series. This is suggestive evidence of the increasing convexity not being explained
by any sort of shifts in the distribution of ability of this population, but rather by the
actual change in premiums to schooling.
5.3
Selectivity
Estimates in Table 8 (women’s participation) and 9 (participation in the labor force)
correct for selection bias by using the Heckman maximum likelihood procedure and incorporate LAMBDA into earnings functions estimates. The selectivity-corrected earnings
functions reported in Table 8 and 9 include the standard variables -education, experience
and its square and the regional dummies. Household demographic variables (hhead=1
26
if household head, married=1 if married or living together, numchildHH= number of
child in a household) are used as exclusion restrictions; these variables are believed to
determine participation in work but do not directly affect labor marker earnings. All
are individually statistically significant. Importantly, the returns to college graduates
remains significantly increasing relative to those with secondary education or primary
education even after controlling for selection bias. However, the returns to college completers (and consequently the gap between educational categories) fell for every year
after the Heckman correction. This is consistent with the fact that college education
has been the single main important factor to escape unemployment over this period
(probits and occupational choice multinomial logits are available under request)
6
The effects of macroeconomic variables on wages by skills
In Table 5 we see the regressions of earnings by levels of education over GDP, unemployment (lagged by one period) and a time trend. In table 6 we add to the same
regressions: import penetration ratios (measured as imports over valued added by sector
and averaged by year=MVA) and capital accumulation (measured as gross investment
in machinery and equipment over valued added by sector and averaged by year). 39
Table 5: Education-level earnings equations, GDP and Unemployment, by levels of
education: Pooled data 1992-2003
Primary
Secondary
College
(1)
(2)
(3)
exp
.029
.044
.042
(.002)∗∗∗
exp2 /100
male
self
ln-GDP
ln-U (lagged)
year trend
Obs.
R2
(.002)∗∗∗
(.003)∗∗∗
-.037
-.062
-.066
(.003)∗∗∗
(.005)∗∗∗
(.007)∗∗∗
.065
.135
.245
(.011)∗∗∗
(.012)∗∗∗
(.015)∗∗∗
-.078
-.141
-.018
(.014)∗∗∗
(.020)∗∗∗
(.023)
1.114
1.360
1.407
(.098)∗∗∗
(.103)∗∗∗
(.124)∗∗∗
-.132
-.221
-.216
(.033)∗∗∗
(.039)∗∗∗
(.047)∗∗∗
-.042
-.034
-.021
(.003)∗∗∗
(.003)∗∗∗
(.004)∗∗∗
38032
.133
30989
.225
20998
.203
Primary referes to all those who have completed at most Primary level education (7 years). Secondary
refers to those completing Secondary education (12 years) and College are all the college completers
(between 15 and 18 years of education). See notes Table 7.
39
Data was only available for 1992-1999 for those variables, which explains the shorter time span for
table 6. We also tried other macroeconomics variables, namely employment, sub-employment, exchange
rate and the growth rate of all of them. The only significant results were for the ones left in the tables
here reported.
27
Table 6: Education-level earnings equations and macro variables, by levels of education:
Pooled data 1993-1999
Primary
Secondary
College
(1)
(2)
(3)
exp
.032
.046
.043
(.002)∗∗∗
exp2 /100
male
self
ln-GDP
ln-U (lagged)
ln-Investment in Machinery and Equip
ln-MVA
year trend
Obs.
R2
(.002)∗∗∗
(.003)∗∗∗
-.044
-.073
-.068
(.004)∗∗∗
(.006)∗∗∗
(.009)∗∗∗
.068
.158
.226
(.013)∗∗∗
(.015)∗∗∗
(.019)∗∗∗
-.039
-.139
.026
(.016)∗∗
(.023)∗∗∗
(.026)
.272
1.174
1.247
(.445)
(.547)∗∗
(.742)∗
-.116
-.241
-.244
(.067)∗
(.081)∗∗∗
(.102)∗∗
.083
.051
.083
(.015)∗∗∗
(.019)∗∗∗
(.025)∗∗∗
.046
-.531
-.240
(.166)
(.208)∗∗
(.260)
-.056
.011
-.013
(.014)∗∗∗
(.017)
(.021)
28243
.099
21668
.18
14266
.176
Primary refers to all those who have completed at most Primary level education (7 years). Secondary
refers to those completing Secondary education (12 years) and College are all the college completers
(between 15 and 18 years of education). We estimate this only for 1992-1999 due to the fact that data
for trade and machinery and equipment is only available for that period. MVA stands for import over
value added. See notes Table 5.
28
We see the somehow ”regressive” effect of GDP. Unemployment is not significant
and the time trend is negative for all levels of education; and highly significant for
primary and secondary. When we add the controls for import penetration and capital
accumulation in column (4), there is still a highly significant downward trend in the
wages of primary completers. GDP is only important for secondary completers, and
unemployment gains significance for primary and secondary completers. Trade seems to
have highly significant and negative effects on secondary completers, and negative (non
significant) for college completers. Capital accumulation is highly significant and positive for all levels of education. The capital investment coefficient for college completers
is 3 percentage points higher than the estimate for primary completers (and statistically
different) and 5 percentage points higher than for secondary completers (and also statistically different). This is preliminary evidence of the effect of some macroeconomic
variables in wages by skills.
29
7
Conclusions
There are four main empirical findings in this paper. First, we observe increasing returns to education with falling average wages for the whole period. Returns to education
reflect the always increasing wages for college graduates up until 2001, combined with
always decreasing wages for the less educated from 1995 onwards. Moreover, from 2001
to 2003 wages for the less educated were falling at a faster pace than college graduate’s
wages. This result is robust to endogeneity, selectivity and to changes in the functional
form (including parametric and semi-parametric techniques). In fact, the household
fixed effects estimation gives an even higher rate of increase in convexity as the gap
between primary and secondary returns becomes flatter and the gap between secondary
and university returns stays equally steep. When testing other specifications (such as
public vis a vis private jobs, labour supply behavior, firm size, occupation and sector
fixed effects) the convexity remained.
Second, given a supply or external shock, the higher the stock of human capital, the
less the impact on the wages levels. As seen in Table 16, we can argue that education
increased the ability of individuals to deal with disequilibria and hence better process
information on how to adjust quickly (Schultz 1975, Gould and Weinberg 2001).
Third, the type of shock matters as external or supply shocks do not seem to have
affected the increasing trend in the returns. They rather had an equal impact across
skill levels, but had non–neutral effects across occupations, clearly affecting more the
self-employed and the informal workers. In contrast, the opening of the economy and the
reforms had non–neutral effects across skills. When looking at the returns to different
occupations, after controlling by education and potential experience, we see that some
occupations receive higher returns and less variance (i.e formal wage employees) than
others. Over the full period (and in particular over crises), the self–employed and the
informal reacted differently to shocks and lost more than any other occupational category. One explanation might be that the self–employed and informal workers could/had
to adjust via wages reductions while other categories have gone into unemployment.
Finally, when analyzing in a very preliminary way the effects of macroeconomic variables on wages we find that: i) capital accumulation is positively correlated with the
increasing trend in returns to schooling for college graduates; ii) trade seems to have a
very significant negative impact on Secondary completers, but it is not significant for
other educational groups; iii) GDP seems to have a regressive effect on wages (benefiting
significantly more the college completers); iv) (lagged) unemployment does not have any
effect on wages (unless we control for trade and capital accumulation) and v) after controlling for all these macroeconomic variables, there is still a highly significant downward
trend in wages for Primary and Secondary completers. These results are consistent with
various ”stories” on how wages are set at different levels of disaggregation. Skill biased
technological change, trade and different abilities to smooth consumption might be part
of the explanation as to why a shock in output or any change in the macro time series
will make the wages of the unskilled to fall faster, or the wages of the skilled to increase
faster.
To finalize, the contribution to the existing literature has focused on new methods
and new sample utilised, and also on a further look at the effects of macroeconomic variables on the returns to education. So far, there were no studies to our knowledge, trying
to assess the impact of relaxing the ”constant-education-slope” assumption commonly
maintained in the literature, beyond some by-levels regressions. In this paper that as30
sumption is explored further; and overall, our results cast doubt on the interpretation of
education in the constrained to have a homogenous impact in earnings equation common
in the literature. Moreover, we run several robustness tests that confirm our final results.
We also attempt to endogeneize education and correct for selectivity. Furthermore, we
consider the action of the individual-invariant time effect multiplied by the education
estimate at given points in time to analyze the effects of aggregate shocks. Previous
studies missed the fact that important increases in premiums to education may come
hand in hand with other ”time effects” by educational levels. Besides, we are not aware
of any study looking at the macro variables behind these trends with our methodology.
One of the implications of this paper is that, differences in the returns to different occupations may be part of the explanation as to why in a country with high-unemployment
we do not observe workers switching from wage employment to self employment over
this period. Increasing informality and an increasing premium to formality has been
one of the most clear outcomes of successive crises in the Argentine labor markets.
Appendix
31
32
9205
.322
9656
.317
1.187
(.029)∗∗∗
1.226
(.029)∗∗∗
.871
(.029)∗∗∗
.886
(.029)∗∗∗
.634
(.024)∗∗∗
.692
(.024)∗∗∗
.412
(.023)∗∗∗
.443
(.023)∗∗∗
.206
(.021)∗∗∗
.239
(.021)∗∗∗
-.058
(.003)∗∗∗
-.056
(.003)∗∗∗
(.002)∗∗∗
(.002)∗∗∗
9846
.345
(.028)∗∗∗
1.261
(.028)∗∗∗
.863
(.024)∗∗∗
.632
(.022)∗∗∗
.386
(.021)∗∗∗
.181
(.003)∗∗∗
-.060
(.002)∗∗∗
8926
.316
(.032)∗∗∗
1.267
(.032)∗∗∗
.863
(.027)∗∗∗
.659
(.026)∗∗∗
.412
(.025)∗∗∗
.219
(.004)∗∗∗
-.059
(.002)∗∗∗
8832
.333
(.032)∗∗∗
1.293
(.033)∗∗∗
.890
(.028)∗∗∗
.653
(.027)∗∗∗
.344
(.025)∗∗∗
.149
(.004)∗∗∗
-.061
(.002)∗∗∗
8688
.352
(.032)∗∗∗
1.367
(.033)∗∗∗
.946
(.028)∗∗∗
.693
(.027)∗∗∗
.403
(.025)∗∗∗
.215
(.004)∗∗∗
-.057
(.002)∗∗∗
8303
.359
(.032)∗∗∗
1.420
(.031)∗∗∗
.925
(.029)∗∗∗
.650
(.027)∗∗∗
.419
(.026)∗∗∗
.229
(.004)∗∗∗
-.052
(.002)∗∗∗
7530
.337
(.036)∗∗∗
1.312
(.035)∗∗∗
.830
(.031)∗∗∗
.612
(.030)∗∗∗
.322
(.029)∗∗∗
.135
(.004)∗∗∗
-.060
(.002)∗∗∗
6905
.334
(.039)∗∗∗
1.343
(.039)∗∗∗
.898
(.035)∗∗∗
.647
(.034)∗∗∗
.350
(.032)∗∗∗
.171
(.004)∗∗∗
-.061
(.002)∗∗∗
Table 7: Earnings equations: Men, Argentina 1992-2003. Naive estimates
y94
y95
y96
y97
y98
y99
y2000
(3)
(4)
(5)
(6)
(7)
(8)
(9)
.043
.042
.043
.043
.041
.042
.044
6761
.35
(.041)∗∗∗
1.442
(.040)∗∗∗
.941
(.036)∗∗∗
.659
(.035)∗∗∗
.382
(.034)∗∗∗
.207
(.004)∗∗∗
-.044
(.002)∗∗∗
y2001
(10)
.037
5842
.344
(.044)∗∗∗
1.448
(.044)∗∗∗
1.005
(.040)∗∗∗
.704
(.040)∗∗∗
.388
(.038)∗∗∗
.241
(.005)∗∗∗
-.044
(.003)∗∗∗
y2002
(11)
.037
4092
.372
(.051)∗∗∗
1.481
(.051)∗∗∗
1.072
(.046)∗∗∗
.720
(.046)∗∗∗
.478
(.043)∗∗∗
.273
(.006)∗∗∗
-.050
(.003)∗∗∗
y2003
(12)
.042
Note 2: Controls for municipalities fixed effects present (see Note 1 in Figure 2 for details on the sample)
Note 1: Estimates receive one star if they are significant at the 90% significance level, two stars at the 95% significance level and three stars at the 99% level.
Obs.
R2
Supc
Supi
Secc
Seci
Primc
exp2 /100
exp
y93
(2)
.042
y92
(1)
.041
33
.236
(.023)∗∗∗
.194
(.023)∗∗∗
-.038
(.026)∗∗
-.135
(.026)∗∗∗
.047
(.022)∗∗∗
.027
(.021)∗
-.141
(.022)∗∗∗
-.219
(.020)
(.022)∗∗∗
-.071
-.096
(.020)∗∗∗
-.090
(.003)∗∗∗
-.091
(.003)∗∗∗
.050
(.001)∗∗∗
.049
(.001)∗∗∗
-.079
(.004)∗∗∗
-.072
(.004)∗∗∗
-.298
(.013)∗∗∗
-.279
(.013)∗∗∗
-.501
(.013)∗∗∗
-.517
(.014)∗∗∗
.949
(.036)∗∗∗
.990
(.036)∗∗∗
.648
(.039)∗∗∗
.763
(.041)∗∗∗
.435
(.033)∗∗∗
.545
(.033)∗∗∗
.133
(.034)∗∗∗
.289
(.035)∗∗∗
.045
(.030)
.141
(.030)∗∗∗
-.056
(.006)∗∗∗
-.058
(.003)∗∗∗
(.006)∗∗∗
(.003)∗∗∗
(.024)∗∗∗
0.275
(.025)∗∗∗
.006
(.021)∗∗∗
.058
(.022)∗∗∗
-.134
(.021)
-.079
(.003)∗∗∗
-.093
(.001)∗∗∗
.052
(.005)∗∗∗
-.069
(.013)∗∗∗
-.279
(.014)∗∗∗
-.515
(.037)∗∗∗
1.039
(.039)∗∗∗
.718
(.033)∗∗∗
.483
(.034)∗∗∗
.190
(.031)
.041
(.006)∗∗∗
-.058
(.003)∗∗∗
(.025)∗∗∗
.349
(.027)∗∗∗
.079
(.023)∗∗∗
.212
(.023)∗∗
-.054
(.022)
-.046
(.003)∗∗∗
-.097
(.001)∗∗∗
.054
(.004)∗∗∗
-.050
(.013)∗∗∗
-.305
(.014)∗∗∗
-.477
(.041)∗∗∗
1.031
(.042)∗∗∗
.659
(.037)∗∗∗
.417
(.036)∗∗
.089
(.034)
-.004
(.006)∗∗∗
-.078
(.003)∗∗∗
(.024)∗∗∗
.440
(.026)∗∗∗
.167
(.024)∗∗∗
.152
(.024)∗∗∗
.128
(.023)
.018
(.003)∗∗∗
-.079
(.001)∗∗∗
.048
(.004)∗∗∗
-.039
(.013)∗∗∗
-.296
(.014)∗∗∗
-.445
(.044)∗∗∗
1.112
(.043)∗∗∗
.740
(.038)∗∗∗
.461
(.038)∗∗∗
.128
(.034)
.018
(.006)∗∗∗
-.079
(.003)∗∗∗
(.025)∗∗∗
.488
(.026)∗∗∗
.205
(.024)∗∗∗
.180
(.024)∗∗∗
-.049
(.023)∗∗∗
.025
(.003)∗∗∗
-.089
(.001)∗∗∗
.051
(.005)∗∗∗
-.039
(.013)∗∗∗
-.288
(.014)∗∗∗
-.446
(.048)∗∗∗
1.159
(.045)∗∗∗
.777
(.040)∗∗∗
.548
(.040)∗∗∗
.238
(.036)∗∗∗
.102
(.006)∗∗∗
-.065
(.003)∗∗∗
(.025)∗∗∗
.522
(.026)∗∗∗
.333
(.024)∗∗∗
.157
(.024)∗∗∗
.002
(.023)∗∗∗
.031
(.003)∗∗∗
-.089
(.001)∗∗∗
.052
(.005)∗∗∗
-.049
(.013)∗∗∗
-.30
(.014)∗∗∗
-.415
(.050)∗∗∗
1.263
(.047)∗∗∗
.846
(.041)∗∗∗
.523
(.040)∗∗∗
.210
(.037)∗∗∗
.096
(.006)∗∗∗
-.059
(.003)∗∗∗
(.026)∗∗∗
.554
(.027)∗∗∗
.33
(.025)∗∗∗
.246
(.025)∗∗∗
.031
(.024)
.045
(.003)∗∗∗
-.080
(.001)∗∗∗
.049
(.005)∗∗∗
-.045
(.014)∗∗∗
-.318
(.014)∗∗∗
-.403
(.055)∗∗∗
1.188
(.050)∗∗∗
.771
(.045)∗∗∗
.489
(.043)∗∗∗
.112
(.039)
.003
(.006)∗∗∗
-.067
(.003)∗∗∗
(.026)∗∗∗
.534
(.028)∗∗∗
.321
(.025)∗∗∗
.222
(.025)∗∗∗
.013
(.024)∗∗∗
.030
(.003)∗∗∗
-.082
(.001)∗∗∗
.050
(.005)∗∗∗
-.033
(.014)∗∗∗
-.300
(.014)∗∗∗
-.402
(.054)∗∗∗
1.195
(.050)∗∗∗
.808
(.045)∗∗∗
.548
(.043)∗∗∗
.174
(.040)∗∗∗
.105
(.007)∗∗∗
-.060
(.003)∗∗∗
(.027)∗∗∗
.626
(.028)∗∗∗
.404
(.026)∗∗∗
.285
(.027)∗∗∗
.064
(.025)∗∗
.091
(.003)∗∗∗
-.078
(.001)∗∗∗
.049
(.005)∗∗∗
-.028
(.014)∗∗∗
-.300
(.014)∗∗∗
-.387
(.062)∗∗∗
1.287
(.057)∗∗∗
.831
(.051)∗∗∗
.603
(.048)∗∗∗
.148
(.045)∗∗
.101
(.007)∗∗∗
-.056
(.003)∗∗∗
Table 8: Earnings equations: Women (corrected for selectivity), Argentina 1992-2003
y93
y94
y95
y96
y97
y98
y99
y2000
y2001
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
.036
.037
.046
.048
.043
.042
.044
.042
.041
(.028)∗∗∗
.654
(.030)∗∗∗
.408
(.027)∗∗∗
.325
(.028)∗∗∗
.062
(.027)
.086
(.003)∗∗∗
-.082
(.002)∗∗∗
.051
(.005)∗∗∗
-.031
(.014)∗∗∗
-.319
(.015)∗∗∗
-.357
(.069)∗∗∗
1.211
(.062)∗∗∗
.808
(.056)∗∗∗
.545
(.051)∗∗∗
.194
(.048)
.067
(.007)∗∗∗
-.065
(.004)∗∗∗
y2002
(11)
.045
(.032)∗∗∗
.592
(.034)∗∗∗
.414
(.031)∗∗∗
.310
(.031)∗∗∗
113
(.030)
.068
(.004)∗∗∗
-.085
(.002)∗∗∗
.051
(.006)∗∗∗
-.034
(.016)∗∗∗
-.299
(.016)∗∗∗
-.307
(.075)∗∗∗
1.186
(.068)∗∗∗
.697
(.061)∗∗∗
.481
(.056)∗∗∗
.173
(.053)
.061
(.009)∗∗∗
-.071
(.004)∗∗∗
y2003
(12)
.047
Note 2: Controls for municipalities fixed effects present (see Note 1 in Figure 2 for details on the sample)
Note 1: Estimates receive one star if they are significant at the 90% significance level, two stars at the 95% significance level and three stars at the 99% level.
W aldChi2
Supc
Supi
Secc
Seci
Primc
exp2 /100
exp
numofchildsHH
married
hhhead
Supc
Supi
Secc
Seci
Primc
exp2 /100
exp
y92
(1)
.036
34
29714
14795
5166.853
.000
30726
15584
5298.462
.000
1.022
(.023)∗∗∗
1.067
(.022)∗∗∗
.780
(.024)∗∗∗
.857
(.024)∗∗∗
.531
(.020)∗∗∗
.615
(.020)∗∗∗
.337
(.019)∗∗∗
.420
(.019)∗∗∗
.150
(.018)∗∗∗
.214
(.017)∗∗∗
-.023
(.003)∗∗∗
-.028
(.002)∗∗∗
(.003)∗∗∗
(.002)∗∗∗
30469
15852
5856.232
.000
(.022)∗∗∗
1.069
(.023)∗∗∗
.804
(.020)∗∗∗
.549
(.019)∗∗∗
.347
(.018)∗∗∗
.137
(.003)∗∗∗
-.023
(.002)∗∗∗
30666
14331
4531.72
.000
(.025)∗∗∗
1.058
(.026)∗∗∗
.774
(.022)∗∗∗
.544
(.022)∗∗∗
.353
(.020)∗∗∗
.145
(.004)∗∗∗
-.018
(.002)∗∗∗
31602
14265
4798.12
.000
(.025)∗∗∗
1.055
(.026)∗∗∗
.787
(.023)∗∗∗
.518
(.022)∗∗∗
.296
(.021)∗∗∗
.094
(.004)∗∗∗
-.020
(.002)∗∗∗
30035
14215
4964.005
.000
(.026)∗∗∗
1.148
(.027)∗∗∗
.850
(.023)∗∗∗
.594
(.023)∗∗∗
.373
(.021)∗∗∗
.174
(.004)∗∗∗
-.020
(.002)∗∗∗
28862
13818
5361.33
.000
(.026)∗∗∗
1.199
(.026)∗∗∗
.840
(.024)∗∗∗
.554
(.023)∗∗∗
.378
(.022)∗∗∗
.176
(.004)∗∗∗
-.011
(.002)∗∗∗
26555
12634
4137.764
.000
(.029)∗∗∗
1.064
(.029)∗∗∗
.732
(.026)∗∗∗
.473
(.026)∗∗∗
.273
(.024)∗∗∗
.077
(.004)∗∗
-.009
(.002)∗∗∗
25003
11708
3749.2
.000
(.031)∗∗∗
1.091
(.031)∗∗∗
.792
(.028)∗∗∗
.538
(.028)∗∗∗
.324
(.026)∗∗∗
.141
(.005)
-.005
(.002)∗∗∗
24620
11558
3731.711
.000
(.033)∗∗∗
1.142
(.033)∗∗∗
.801
(.030)∗∗∗
.543
(.030)∗∗∗
.338
(.028)∗∗∗
.151
(.005)∗∗∗
.017
(.002)∗
24620
10122
3053.819
.000
(.037)∗∗∗
1.109
(.036)∗∗∗
.842
(.033)∗∗∗
.549
(.033)∗∗∗
.353
(.031)∗∗∗
.165
(.005)∗
.010
(.003)∗∗∗
y2002
(11)
.008
See Notes in Table 7. Heckman maximun likelihood estimations to correct for selection. Exclusion restrictions are all significant, not reported due to lack of space.
N (total)
N (uncensored)
W aldχ2 statistic
p-value (Wald)
Supc
Supi
Secc
Seci
Primc
exp2 /100
exp
Table 9: Selection equations (dependent vble=1 if hourly wages are positive), Argentina 1992-2003
y92
y93
y94
y95
y96
y97
y98
y99
y2000
y2001
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
.026
.023
.024
.020
.021
.023
.019
.015
.014
.004
15940
7309
2526.412
.000
(.041)∗∗∗
1.129
(.041)∗∗∗
.829
(.037)∗∗∗
.538
(.037)∗∗∗
.386
(.035)∗∗∗
.152
(.006)∗∗
.015
(.003)∗∗
y2003
(12)
.007
35
2912
14690
.254
3001
15482
.276
1.077
(.025)∗∗∗
1.015
(.025)∗∗∗
.813
(.026)∗∗∗
.788
(.027)∗∗∗
.586
(.022)∗∗∗
.600
(.022)∗∗∗
.355
(.021)∗∗∗
.393
(.021)∗∗∗
.191
(.019)∗∗∗
.213
(.019)∗∗∗
-.046
(.003)∗∗∗
-.052
(.003)∗∗∗
3175
15720
.316
(.024)∗∗∗
1.121
(.025)∗∗∗
.836
(.021)∗∗∗
.584
(.021)∗∗∗
.356
(.019)∗∗∗
.162
(.003)∗∗∗
-.055
(.001)∗∗∗
6573
14271
.238
(.034)∗∗∗
.982
(.035)∗∗∗
.670
(.030)∗∗∗
.482
(.029)∗∗∗
.263
(.027)∗∗∗
.122
(.004)∗∗∗
-.056
(.002)∗∗∗
6689
14229
.259
(.035)∗∗∗
1.042
(.037)∗∗∗
.734
(.032)∗∗∗
.541
(.031)∗∗∗
.284
(.028)∗∗∗
.122
(.004)∗∗∗
-.054
(.002)∗∗∗
5887
14128
.294
(.033)∗∗∗
1.098
(.035)∗∗∗
.782
(.030)∗∗∗
.578
(.030)∗∗∗
.324
(.027)∗∗∗
.181
(.004)∗∗∗
-.053
(.002)∗∗∗
3969
13702
.301
(.031)∗∗∗
1.151
(.032)∗∗∗
.779
(.029)∗∗∗
.526
(.027)∗∗∗
.328
(.025)∗∗∗
.172
(.004)∗∗∗
-.051
(.002)∗∗∗
5565
12528
.269
(.039)∗∗∗
.992
(.040)∗∗∗
.669
(.036)∗∗∗
.454
(.034)∗∗∗
.213
(.032)∗∗
.064
(.004)∗∗∗
-.057
(.002)∗∗∗
5765
11613
.263
(.043)∗∗∗
.980
(.044)∗∗∗
.637
(.039)∗∗∗
.468
(.038)∗∗∗
.203
(.035)∗∗∗
.093
(.005)∗∗∗
-.060
(.002)∗∗∗
5761
11461
.279
(.045)∗∗∗
1.091
(.046)∗∗∗
.717
(.041)∗∗∗
.500
(.040)∗∗∗
.281
(.037)∗∗∗
.136
(.005)∗∗∗
-.044
(.002)∗∗∗
5493
10064
.26
(.053)∗∗∗
1.053
(.055)∗∗∗
.736
(.049)∗∗∗
.495
(.047)∗∗∗
.263
(.043)∗∗∗
.135
(.006)∗∗∗
-.044
(.003)∗∗∗
3732
7261
.247
(.057)∗∗∗
.940
(.058)∗∗∗
.636
(.052)∗∗∗
.473
(.050)∗∗∗
.262
(.047)∗∗
.099
(.006)∗∗∗
-.049
(.003)∗∗∗
(.001)∗∗∗
(.001)∗∗∗
See Notes in Table 7.
No. groups
Obs.
R2 within
Supc
Supi
Secc
Seci
Primc
exp2 /100
exp
y2003
(12)
.040
Table 10: Fixed Effects estimates of Earnings Equations, Argentina 1992-2003. All type of family relations
y92
y93
y94
y95
y96
y97
y98
y99
y2000
y2001
y2002
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
.037
.035
.040
.039
.040
.040
.039
.040
.042
.036
.035
36
See Notes Table 7.
Obs.
R2
HHsize
log (firmsize)
male
Supc
Supi
Secc
Seci
Primc
exp2 /100
exp
12803
.331
14015
.335
-.007
(.003)∗∗∗
-.010
(.003)∗∗∗
.018
(.003)∗∗∗
.008
(.004)∗∗
.135
(.012)∗∗∗
.111
(.013)∗∗∗
.999
(.025)∗∗∗
1.035
(.026)∗∗∗
.717
(.025)∗∗∗
.768
(.026)∗∗∗
.514
(.021)∗∗∗
.584
(.021)∗∗∗
.306
(.020)∗∗∗
.355
(.020)∗∗∗
.142
(.018)∗∗∗
.192
(.018)∗∗∗
-.045
(.003)∗∗∗
-.047
(.001)∗∗∗
(.003)∗∗∗
(.001)∗∗∗
14565
.375
(.003)∗∗∗
-.011
(.003)∗∗∗
.033
(.012)∗∗∗
.147
(.024)∗∗∗
1.063
(.024)∗∗∗
.748
(.020)∗∗∗
.525
(.019)∗∗∗
.321
(.018)∗∗∗
.143
(.003)∗∗∗
-.048
(.001)∗∗∗
-.053
13494
.363
(.003)∗∗∗
-.018
(.003)∗∗∗
.032
(.012)∗∗∗
.130
(.026)∗∗∗
1.021
(.026)∗∗∗
.669
(.022)∗∗∗
.487
(.022)∗∗∗
.277
(.020)∗∗∗
.133
(.003)∗∗∗
13393
.369
(.003)∗∗∗
-.010
(.003)∗∗∗
.029
(.013)∗∗∗
.128
(.026)∗∗∗
1.044
(.027)∗∗∗
.716
(.023)∗∗∗
.492
(.022)∗∗∗
.246
(.021)∗∗∗
.089
(.003)∗∗∗
-.052
(.001)∗∗∗
13404
.389
(.003)∗∗∗
-.012
(.003)∗∗∗
.038
(.013)∗∗∗
.142
(.026)∗∗∗
1.118
(.027)∗∗∗
.757
(.023)∗∗∗
.570
(.022)∗∗∗
.322
(.020)∗∗∗
.161
(.003)∗∗∗
-.046
(.001)∗∗∗
12752
.399
(.003)∗∗∗
-.012
(.003)∗∗∗
.036
(.013)∗∗∗
.142
(.028)∗∗∗
1.143
(.027)∗∗∗
.748
(.024)∗∗∗
.513
(.023)∗∗∗
.312
(.021)∗∗∗
.157
(.003)∗∗∗
-.040
(.001)∗∗∗
11764
.402
(.003)∗∗∗
-.016
(.003)∗∗∗
.042
(.014)∗∗∗
.124
(.029)∗∗∗
1.056
(.029)∗∗∗
.669
(.026)∗∗∗
.477
(.025)∗∗∗
.226
(.023)∗∗∗
.077
(.003)∗∗∗
-.049
(.002)∗∗∗
1992-2003. Controlling for firm size, labor supply
y95
y96
y97
y98
y99
(4)
(5)
(6)
(7)
(8)
.036
.036
.036
.033
.035
(.001)∗∗∗
Table 11: Earnings equations, Argentina
y92
y93
y94
(1)
(2)
(3)
.034
.034
.035
10789
.402
(.003)∗∗∗
-.020
(.004)∗∗∗
.050
(.014)∗∗∗
.137
(.031)∗∗∗
1.079
(.031)∗∗∗
.732
(.028)∗∗∗
.551
(.027)∗∗∗
.290
(.025)∗∗∗
.158
(.003)∗∗∗
-.044
(.002)∗∗∗
10602
.413
(.003)∗∗∗
-.024
(.004)∗∗∗
.042
(.015)∗∗∗
.153
(.032)∗∗∗
1.117
(.032)∗∗∗
.736
(.029)∗∗∗
.543
(.028)∗∗∗
.281
(.026)∗∗∗
.151
(.003)∗∗∗
-.028
(.002)∗∗∗
behavior and sector
y2000
y2001
(9)
(10)
.034
.027
9364
.417
(.004)∗∗∗
-.029
(.004)∗∗∗
.060
(.016)∗∗∗
.143
(.035)∗∗∗
1.089
(.035)∗∗∗
.747
(.031)∗∗∗
.554
(.031)∗∗∗
.299
(.029)∗∗∗
.151
(.004)∗∗∗
-.037
(.002)∗∗∗
6818
.428
(.004)∗∗∗
-.017
(.005)∗∗∗
.050
(.017)∗∗∗
.130
(.039)∗∗∗
1.116
(.039)∗∗∗
.756
(.035)∗∗∗
.534
(.035)∗∗∗
.332
(.033)∗∗∗
.177
(.004)∗∗∗
-.039
(.002)∗∗∗
fixed effects
y2002
y2003
(11)
(12)
.031
.033
Table 12: Time trend and time-varying returns to schooling: Argentina 1992-2003
Pooled data
.085
(.020)∗∗∗
.188
(.020)∗∗∗
.138∗∗∗
(.021)
.114
(.022)∗∗∗
.050∗∗
(.022)
.047
(.022)∗∗
.061
(.024)∗∗
.031
(.024)
-.012
(.025)
-.239
(.027)∗∗∗
-.367
(.030)∗∗∗
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
Primc
.209
(.017)∗∗∗
.421∗∗∗
(.018)
.639
(.018)∗∗∗
.870∗∗∗
(.022)
1.110
(.022)∗∗∗
Seci
Secc
Supi
Supc
Pric*1993
-.042
(.024)∗
-.054
(.024)∗∗
-.037
(.025)
-.075
(.025)∗∗∗
-.038
(.025)
-.037
(.026)
-.109
(.027)∗∗∗
-.063
(.028)∗∗
-.076
(.029)∗∗∗
-.061
(.031)∗∗
-.052
(.035)
Pric*1994
Pric*1995
Pric*1996
Pric*1997
Pric*1998
Pric*1999
Pric*2000
Pric*2001
Pric*2002
Pric*2003
Seci*1993
-.074
(.025)∗∗∗
-.075
(.025)∗∗∗
-.101
(.026)∗∗∗
-.107
(.026)∗∗∗
-.075
(.026)∗∗∗
-.087
(.026)∗∗∗
-.157
(.028)∗∗∗
-.141
(.029)∗∗∗
-.162
(.030)∗∗∗
-.142
(.031)∗∗∗
-.124
(.035)∗∗∗
Seci*1994
Seci*1995
Seci*1996
Seci*1997
Seci*1998
Seci*1999
Seci*2000
Seci*2001
Seci*2002
Seci*2003
Secc*1993
-.063
continued on next page
37
(pooled data)
(.025)∗∗
-.051
(.025)∗∗
-.078
(.026)∗∗∗
-.044
(.026)∗
-.024
(.027)
-.070
(.027)∗∗∗
-.078
(.028)∗∗∗
-.064
(.029)∗∗
-.069
(.030)∗∗
-.056
(.031)∗
-.105
(.035)∗∗∗
Secc*1994
Secc*1995
Secc*1996
Secc*1997
Secc*1998
Secc*1999
Secc*2000
Secc*2001
Secc*2002
Secc*2003
Supi*1993
-.049
(.030)
-.024
(.030)
-.027
(.030)
.014
(.030)
.036
(.030)
.029
(.030)
-.012
(.031)
.004
(.034)
.009
(.032)
.054
(.028)
.005
(.029)
Supi*1994
Supi*1995
Supi*1996
Supi*1997
Supi*1998
Supi*1999
Supi*2000
Supi*2001
Supi*2002
Supi*2003
Supc*1993
-.024
(.028)
.045
(.028)
.24
(.031)∗∗∗
.207
(.031)∗∗∗
.250
(.031)∗∗∗
.286
(.031)∗∗∗
.236
(.033)∗∗∗
.219
(.034)∗∗∗
.271
(.034)∗∗∗
.269
(.035)∗∗∗
.266
(.041)∗∗∗
Supc*1994
Supc*1995
Supc*1996
Supc*1997
Supc*1998
Supc*1999
Supc*2000
Supc*2001
Supc*2002
Supc*2003
Obs.
R2
168965
0.37
38
Table 13: Changes in the square coefficients of earnings equations with time FE : 19922003
self
wage
manager
pool
(1)
(2)
(3)
(4)
S2 /100*1993
.601
.199
-1.204
.282
(.172)∗∗∗
S2 /100*1994
S2 /100*1995
S2 /100*1996
S2 /100*1997
S2 /100*1998
S2 /100*1999
S2 /100*2000
S2 /100*2001
S2 /100*2002
S2 /100*2003
Obs.
R2
(.081)∗∗
(.510)∗∗
(.075)∗∗∗
.431
.273
-.687
.287
(.171)∗∗
(.076)∗∗∗
(.488)
(.071)∗∗∗
.265
.224
-.273
.207
(.176)
(.080)∗∗∗
(.517)
(.074)∗∗∗
.269
.281
-1.141
.239
(.175)
(.079)∗∗∗
(.590)∗
(.074)∗∗∗
.196
.344
-.757
.252
(.188)
(.079)∗∗∗
(.556)
(.076)∗∗∗
.431
.366
-.745
.349
(.183)∗∗
(.078)∗∗∗
(.491)
(.074)∗∗∗
.703
.293
-.657
.359
(.179)∗∗∗
(.081)∗∗∗
(.488)
(.074)∗∗∗
.304
.416
-.436
.347
(.201)
(.085)∗∗∗
(.508)
(.080)∗∗∗
.643
.497
-.108
.482
(.202)∗∗∗
(.083)∗∗∗
(.558)
(.079)∗∗∗
.288
.453
-.679
.373
(.209)
(.084)∗∗∗
(.541)
(.081)∗∗∗
.099
.585
-.975
.372
(.254)
(.098)∗∗∗
(.687)
(.102)∗∗∗
34229
.267
128860
.36
5875
.313
168965
.334
See notes Table 7. Potential experience Potential experience square, and interactions of time and years
of education and time of years of education squared, plus regions controls are included but not
reported. This specification includes only a linear term and a square term in years of education.
Estimates receive one star if they are significant at the 90% significance level, two stars at the 95%
significance level and three stars at the 99% level.
39
40
30666
31602
.118
(.013)∗∗∗
.096
(.013)∗∗∗
-.068
(.016)∗∗∗
-.080
(.015)∗∗∗
1.027
(.025)∗∗∗
1.037
(.025)∗∗∗
.754
(.026)∗∗∗
.750
(.026)∗∗∗
.490
(.023)∗∗∗
.522
(.022)∗∗∗
.278
(.022)∗∗∗
.337
(.022)∗∗∗
.084
(.021)∗∗∗
.135
(.020)∗∗∗
-.022
(.004)∗∗∗
-.021
(.002)∗∗∗
(.004)∗∗∗
(.002)∗∗∗
30035
(.013)∗∗∗
.138
(.016)∗∗∗
-.103
(.025)∗∗∗
1.108
(.026)∗∗∗
.803
(.023)∗∗∗
.555
(.022)∗∗∗
.346
(.021)∗∗∗
.158
(.004)∗∗∗
-.023
(.002)∗∗∗
28862
(.013)∗∗∗
.167
(.016)
-.016
(.026)∗∗∗
1.144
(.026)∗∗∗
.783
(.024)∗∗∗
.504
(.023)∗∗∗
.345
(.021)∗∗∗
.153
(.004)∗∗∗
-.016
(.002)∗∗∗
26555
(.013)∗∗∗
.201
(.016)∗∗∗
-.050
(.028)∗∗∗
1.018
(.028)∗∗∗
.673
(.026)∗∗∗
.429
(.025)∗∗∗
.239
(.023)∗∗∗
.062
(.004)∗∗∗
-.018
(.002)∗∗∗
25003
(.014)∗∗∗
.192
(.017)∗∗∗
-.082
(.030)∗∗∗
1.040
(.031)∗∗∗
.733
(.028)∗∗∗
.490
(.027)∗∗∗
.288
(.025)∗∗∗
.128
(.005)∗∗∗
-.016
(.002)∗∗∗
24620
(.014)∗∗∗
.218
(.017)∗∗∗
-.067
(.032)∗∗∗
1.088
(.032)∗∗∗
.740
(.029)∗∗∗
.496
(.029)∗∗∗
.297
(.027)∗∗∗
.134
(.005)
.005
(.002)∗∗∗
24620
(.016)∗∗∗
.277
(.019)∗∗∗
-.076
(.035)∗∗∗
1.047
(.035)∗∗∗
.758
(.031)∗∗∗
.477
(.031)∗∗∗
.301
(.029)∗∗∗
.138
(.006)
-.008
(.003)∗∗∗
15940
(.017)∗∗∗
.270
(.020)∗∗∗
-.094
(.039)∗∗∗
1.039
(.039)∗∗∗
.728
(.036)∗∗∗
.457
(.035)∗∗∗
.331
(.033)∗∗∗
.134
(.006)
-.002
(.003)∗∗∗
y2003
(9)
.014
Note 2: Controls for municipalities fixed effects present (see Note 1 in Figure 2 for details on the sample)
Note 1: Estimates receive one star if they are significant at the 90% significance level, two stars at the 95% significance level and three stars at the 99% level.
Obs.
R2
formal
self
Supc
Supi
Secc
Seci
Primc
exp2 /100
exp
Table 14: Pooled earnings equations with occupation dummies (corrected for selectivity), Argentina 1992-2003
y95
y96
y97
y98
y99
y2000
y2001
y2002
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
.022
.022
.025
.021
.019
.020
.010
.016
References
Andres, L. (2003): “Private Education versus Public Education: The effects of the
quality of the alternative schooling in Argentina,” Discussion paper, mimeo, University of Chicago.
Barro, R. J., and J. W. Lee (2000): “International Data on Educational Attainment
Updates and Implications,” Discussion paper, National Bureau of Economic Research,
Inc, NBER Working Papers: 7911 2000.
Behrman, J., and B. Wolfe (1984): “The Socioeconomic Impact of Schooling in a
Developing Country,” The Review of Economics and Statistics, 66(2), 296–303.
Berman, E., J. Bound, and Z. Griliches (1994): “Changes in the Demand for
Skilled Labor within U.S. Manufacturing: Evidence from the Annual Survey of Manufactures,” Quarterly Journal of Economics, 109(2), 367–97, Boston U and NBER; U
MI and NBER; Harvard U and NBER.
Berman, E., J. Bound, and S. Machin (1998): “Implications of Skill-Biased Technological Change: International Evidence,” Quarterly Journal of Economics, 113(4),
1245–79, Boston U and NBER; U MI and NBER; U College London and CEPR,
London School of Econ.
Blom, A., N. Holm, Lauritz, and D. Verner (2001): “Education, Earnings, and
Inequality in Brazil, 1982-98: Implications for Education Policy,” Discussion paper,
Unlisted; Unlisted; Unlisted, The World Bank, Policy Research Working Paper Series:
2686 2001.
Bound, J., and G. Johnson (1992): “Changes in the Structure of Wages in the 1980’s:
An Evaluation of Alternative Explanations,” American Economic Review, 82(3), 371–
92.
Butcher, K. F., and A. Case (1994): “The Effect of Sibling Sex Composition on
Women’s Education and Earnings,” Quarterly Journal of Economics, 109(3), 531–63,
VA Polytechnic Institute State U; Princeton U.
Card, D. (2001): “Estimating the Return to Schooling: Progress on Some Persistent
Econometric Problems,” Econometrica, 69(5), 1127–60, U CA, Berkeley.
Dickens, W. T., L. F. Katz, and K. Lang (1986): “Are Efficiency Wages Efficient?,”
Discussion paper, National Bureau of Economic Research, Inc, NBER Working Papers: 1935 1986.
Fafchamps, M., M. Soderbom, and N. Benhassine (2006): “Job sorting in African
Labor Markets,” Discussion paper, CSAE Working Paper series WPS/2006-02.
FIEL (2002): Productividad, competitividad y empresas. Los engranajes del
crecimiento. Buenos Aires.
Fiszbein, A., P. I. Giovagnoli, and I. Aduriz (2003): “The Argentine Crisis and Its
Impact on Household Welfare,” CEPAL Review, 0(79), 143–58, World Bank; World
Bank; Instituto Brasileiro de Opiniao Publica e Estatistica.
41
Freeman, R. B. (1979): “The Effect of Demographic Factors on Age-Earnings Profiles,” Journal of Human Resources, 14(3), 289–318.
Galiani, S., and P. Sanguinetti (2003): “The impact of trade liberalization on
wage inequality: evidence from Argentina,” Journal of Development Economics, 72,
497–513.
Gasparini, L., M. Marchionni, and W. Sosa Escudero (2001): “Characterization of inequality changes through microeconometric decompositions: The case of
Greater Buenos Aires,” in The microeconomics of income distribution dynamics: A
comparative analysis of selected developing countries., ed. by W. B. IADB. Washington DC.
Gould, E., M. O., and B. Weinberg (2001): “Precautionary Demand for Education,
Inequality and Technological Progress,” Journal of Economic Growth, 6, 285–315.
INDEC (2000): “Incidencia de la pobreza y de la indigencia en el aglomerado del Gran
Buenos Aires,” Discussion paper, Ministerio de Economı́a y Producciòn, Secretarı́a
de polı́tica econòmica, Instituto Nacional de Estadı́stica y Censos.
Katz, L. F., and K. M. Murphy (1995): “Changes in Relative Wages, 1963-1987:
Supply and Demand Factors,” in Labor economics. Volume 1 Labor supply and
labor demand.Elgar Reference Collection., pp. 384–427. shenfelter,-Orley-C.; Hallock,Kevin-F., eds., Harvard U and NBER; U Chicago and NBER.
Krueger, A. B., and L. H. Summers (1989): “Efficiency Wages and the Wage Structure,” Discussion paper, National Bureau of Economic Research, Inc, NBER Working
Papers: 1952 1989 with the RePEc Project: http://repec.org.
Kugler, B., and G. Psacharopoulos (1989): “Earnings and Education in Argentina: An Analysis of the 1985 Buenos Aires Household Survey,” Economics of
Education Review, 8(4), 353–65, World Bank.
Lachler, U. (1999): “Education and Earnings Inequality in Mexico,” Discussion paper, Unlisted, The World Bank, Policy Research Working Paper Series: 1949 1999
http://econ.worldbank.org/docs/635.pdf Information provided in collaboration with
the RePEc Project: http://repec.org.
Layard, R., and S. Nickell (2000): “Unemployment in Britain,” in Cheltenham,
U.K, vol. 122 of Junankar,-P.-N., ed. The economics of unemployment. Volume 3.
International Library of Critical Writings in Economics, pp. 375–423. Elgar; distributed by American International Distribution Corporation, and Northampton,
Mass.
Machin, S. (1996): “Changes in the Relative Demand for Skills,” in Market failures,
their symptoms and policy responses, pp. 129–46. Cambridge; New York and Melbourne: Cambridge University Press, Booth,-Alison-L.; Snower,-Dennis-J., eds. Acquiring skills, U College London and Centre for Econ Performance, LSE.
Margot, D. (2001): “Rendimiento a la educacion en Argentina: un analisis de cohortes,” Discussion paper, Documento de trabajo numero 33, Universidad Nacional
de la Plata.
42
McKenzie, D. J. (2004): “Aggregate Shocks and Urban Labor Market Responses:
Evidence from Argentina’s Financial Crisis,” Economic Development and Cultural
Change, 52(4), 719–58, Stanford U.
Mincer, J. (1974): “Schooling, experience and earnings,” Human behavior and social institutions. National Bureau of Economic Research, distributed by Columbia
University Press, New York.
Parandekar, S., S. Espana, and M. P. Savanti (2003): “The impact of the crisis on the Argentine educational process,” World Bank Working Paper Series 03/03
Background paper number 7.
Patrinos, H. A., A. Fiszbein, and P. I. Giovagnoli (2005): “Estimating the returns
to education in Argentina: 1992-2002,” Discussion paper, The World Bank, Policy
Research Working Paper Series: 3715 2005.
Patrinos, H. A., and C. Sakellariou (2005): “Economics volatility and returns to
education in Venezuela: 1992-2002,” Applied Economics (forthcoming).
Pessino, C. (1993): “From aggregate shocks to labour markets adjustments: shifting
of wages profiles under hyperinflation in Argentina,” Discussion paper, Documento
de Trabajo CEMA numero 95.
(1996): “Returns to Education in Greater Buenos Aires 1986-1993: From
Hyperinflation to Stabilization and Beyond,” Cuadernos de Economia (Pontifical
Catholic University of Chile), 33(99), 205–26, CEMA, Instituto Universitario, Buenos
Aires.
Psacharopoulos, G., and H. A. Patrinos (2004): “Human Capital and Rates of
Return,” Hellenic Parliament; World Bank Johnes,-Geraint; Johnes,-Jill, eds. International handbook on the economics of education. Cheltenham, U.K. and Northampton,
Mass.: Elgar, 2004; 1-57.
Schmitt, J. (1995): “The Changing Structure of Male Earnings in Britain, 1974-1988,”
London School of Econ Freeman,-Richard-B.; Katz,-Lawrence-F., eds. Differences and
changes in wage structures. National Bureau of Economic Research Comparative Labor Markets Series. Chicago and London: University of Chicago Press, 1995; 177-204.
Schultz, T. W. (1975): “The Value of the Ability to Deal with Disequilibria,” Journal
of Economic Literature, 13(3), 827–46.
Söderbom, M., F. Teal, and A. Wambugu (2005): “Unobserved heterogeneity and
the relation between earnings and firm size: evidence from two developing countries,”
Economics Letters, 87(2), 153–159.
Söderbom, M., W. A. Teal F., and G. Kahyarara (2006): “The Dynamics of
Returns to Education in Kenyan and Tanzanian Manufacturing,” Oxford Bulletin of
Economics and Statistics, 68(3), 261.
W.Bank (2000a): Implementation completion report.41950-AR. Washington DC.
(2000b): Poverty and Policy in Latin America and Caribbean, Latin America
and Caribbean Regional Study.
43
(2000c): Poverty Report for Latin America and Caribbean Region.
Welch, F. (2003): “Education in Production,” in Cheltenham, U.K, vol. 159 of
Hanushek,-Eric-A., ed. The economics of schooling and school quality. Volume 1.
Labor markets, distribution and growth. Elgar Reference Collection. International
Library of Critical Writings in Economics, pp. 29–53. Elgar; distributed by American
International Distribution Corporation, and Northampton, Mass., Unlisted.
Wood, A. (1995): “How Trade Hurt Unskilled Workers,” Journal of Economic
Perspectives, 9(3), 57–80, Institute of Development Studies, U Sussex.
44