Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
1
LONG-RUN EFFECTS OF PUBLIC EXPENDITURE IN EDUCATION ON
POVERTY AND HEALTH*
Marisa Hidalgo-Hidalgo and Iñigo Iturbe-Ormaetxe**
April 2013
Abstract: Household characteristics may have long-run effect on individual outcomes
when adult. For instance, individuals who lived when young in households experiencing
financial problems may be more likely to be poor when adults. Public intervention in
education is one of the most important means by which governments try to soften these
effects and to promote equality of opportunity. The objective of this paper is to check
whether public expenditure in education may have an effect in reducing the probability
of being poor when adult, and to what extent. Our main finding is that, public
expenditure in primary education has a strong effect on raising individuals above the
poverty line, on reducing the probability of suffering health problems when adults and
on increasing school attendance beyond compulsory education.
Keywords: Public expenditure in education, poverty rate, intergenerational
transmission of poverty
JEL Classification: H52, I21, I23, J24, J31
* We wish to thank M. Dolores Collado and Laura Crespo for helpful comments. Financial
support from Spanish Ministry of Education and Science and FEDER funds (SEJ 2007-62656,
ECO2012-34928 and ECO2011-22919), and Instituto Valenciano de Investigaciones (IVIE) is
gratefully acknowledged.
** M. Hidalgo-Hidalgo: Universidad Pablo de Olavide de Sevilla, [email protected] I. Iturbe-
Ormaetxe: Universidad de Alicante, [email protected]
2
1. Introduction
Living in a poor household when young may have negative long-run effects on
individual welfare. When adults, these individuals are more likely to be poor, they are
also more prone to suffer health problems and less likely to stay at school after
compulsory education. These long-run effects reflect the degree of intergenerational
mobility in a society. In countries where social mobility is low, being poor when young
is a good predictor of the probability of being poor when adult, or of the probability of
suffering health problems. Public intervention in education is one of the most important
means by which governments try to soften these long-run effects of poverty and
promote equality of opportunity.
Our objective is to study whether the long-run negative effects of poverty are
mitigated by public expenditure in education, and to what extent.
With this intention, we merge data from the 2005 cross-section of the EU-SILC
(including the module on “Intergenerational transmission of poverty”) with data on
public expenditure in education from the UNESCO database.
We estimate the effect of public expenditure in education during childhood in
our variables of interest. Our main finding is that public expenditure in primary
education has a strong effect on raising individuals above the poverty line, in particular
for those individuals who have more than just compulsory education. The reason for
this can be that spending resources in primary education increases schooling beyond
compulsory education and, therefore, helps to lift individuals above the poverty
threshold when adults. We also find that public expenditure in primary education helps
to reduce the probability of suffering health problems when adults. Surprisingly, now
this effect is stronger for individuals with at most compulsory education. Regarding the
effect on the probability of attending post-compulsory education, we also some a
positive effect of our variable of interest. This seems to be consistent with the effect on
the poverty rate.
The paper is organized as follows. Section 2 reviews the literature. Section 3
describes the data used in the paper. Section 4 presents the empirical model. We discuss
our empirical results in Section 5. Finally, Section 6 concludes.
3
2. Literature Review
This paper is related to several strands of the literature. First, the literature on
intergenerational income mobility. Second, the literature that studies the effect of
schooling on several life outcomes, like health and poverty status.
The first strand studies the extent of intergenerational income mobility and
estimates the relationship between parents' economic status and a child's economic
status in adulthood. There have been some important contributions in terms of
measurement of correlations and the forces driving the relationship (see Black and
Devereux, 2011). There are two plausible mechanisms underlying the intergenerational
transmission of poverty or health. On the one hand there may be genetic differences in
ability that are transmitted from parent to child and that lead to intergenerational
persistence in poverty. On the other hand children of wealthy parents earn higher
incomes in part because they invest more in human capital and have more education.
Therefore it is important to understand the mechanisms underlying the observed
intergenerational correlations. If they are in fact due to differential human capital
investment, this suggests a role for public provision or financing of education to
equalize opportunities. While most of the theoretical work on the intergenerational
transmission of economic status considers only parental investments in children,
governments also invest in children's human capital. Mayer and Lopoo (2008) provide
an empirical contribution that takes into account government expenditure. They assess
the relationship between government spending and intergenerational economic mobility
using data from the Panel Study of Income Dynamics and data on state spending from
the U.S. Census of Governments. They find greater intergenerational mobility in high-
spending states compared to low-spending states. They also find that spending on
elementary and secondary schooling has the largest impact on low-income children's
future income. Our paper contributes to this line of research, although there are some
important differences between Mayer and Lopoo (2008) and this paper. First, they focus
on intergenerational income transmission rather than on poverty or on health outcomes
as we do. Second, they fail to consider the endogeneity of schooling decisions to public
expenditure on education. Third, we focus on a group of European countries using data
from the EU-SILC.
4
Our identification strategy to assess the impact of government educational
spending on individual’s particular outcomes (poverty and health status) consists of
exploiting country and time differences in expenditure. Several other papers had a
similar approach. For example, state spending on health care has been associated with a
decline in infant mortality (Mayer and Sarin, 2005) and state per pupil spending on
elementary and secondary schooling is associated with higher wages once children are
adults (Grogger, 1996). However, other studies find that expenditure on schools has
little effect on test scores (e.g., Hanushek, 1996, 2001), while others find that spending
increases test scores (e.g., Hedges et al., 1992). Mayer (2002) finds that greater
spending on elementary and secondary schools increases low-income but not high-
income children's educational attainment but that greater spending on college financial
aid increases schooling for high-income but not low-income children. This result
implies that spending on elementary and secondary schooling but not spending on post-
secondary schooling promotes intergenerational mobility. In addition, this evidence
suggests the need for a more thorough investigation of the impact of different levels of
education spending on intergenerational poverty and health transmission.
Within this literature on the relationship between education spending and
educational outcomes, there are some works on the impact of expenditure on
attendance. For example, Meghir and Palme (2004) evaluate the impact of a school
reform, which took place in the 1950s in Sweden, on educational attainment and
earnings. This reform consisted of increasing compulsory schooling, among other
aspects. Thus the reform can alternatively be viewed as an increase in the per capita
public expenditure on education. They find that the reform increased both the
educational attainment and the earnings of those whose fathers had just compulsory
education.
Finally, the impact on education on health has been the object of some academic
interest. There is a growing body of work that suggests that improvements in education
can improve health. See Lochner (2011) for an interesting review of recent
developments on these “nonproduction” benefits of education. A substantial body of
recent research has tried to disentangle the causal component of the correlation and to
understand the mechanisms driving the causal link (see Cutler and Lleras-Muney (2011)
for an example of the latter and a discussion). Longitudinal studies collecting very rich
information on childhood environment and childhood outcomes have been used to
5
identify plausibly causal links between childhood environment, childhood endowments
and health and years of schooling on the one hand, and between all of these variables
and adult health on the other hand (Case, Fertig, and Paxson (2005), Currie and Hyson
(1999), and Conti, Heckman & Urzua (2010)). This paper belongs also to this body of
research. In particular we study the impact of parental poverty on individuals’ education
attainment and both variables (together with very rich information regarding child
environment) on adult health status.
3.- Data and descriptive statistics
Estimating whether government expenditure increases intergenerational mobility
requires individual-level data on adult’s income together with information on the
characteristics of the households where that adult grew up. It also requires a source of
variation in government expenditure. In this study we merge data drawn from the 2005
cross section of the EU-SILC database with data from the UNESCO database for
Education. We build a database comprising 17 European countries.1 These are the
countries in the EU-SILC database for which we have enough data on public
expenditure in education.
The 2005 cross section of the EU-SILC database includes a special module on
inter-generational transmission of poverty.2 This module has information on parental
background and childhood circumstances. This information includes, in particular,
family composition, year of birth of parents, activity and occupation of parents. One of
the questions (PM100) provides retrospective information about the economic situation
when the individual was a teenager. They are asked how frequent financial problems in
the household were when they were young teenagers. It is a categorical variable that
takes values from 1 (most of the time) to 5 (never).3 We summarize the information of
PM100 by constructing a binary variable that takes value 1 when PM100 is either 1 or
2. We call this variable “poor_past”. It tells us whether an individual experienced
1 The list of countries is: Austria, Belgium, Cyprus, Denmark, Finland, France, Greece, Hungary, Ireland,
Italy, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden and United Kingdom. 2 For an overview of EU-SILC, see Wolff, Montaigne, and Rojas González (2010). To access further
information about EU’s regulations concerning the SILC, data documentation provided by Eurostat, and
SILC variable lists, we recommend the EU-SILC web portal provided by the GESIS research institute at
http://www.gesis.org/ 3 Frequencies are: (1) Most of the time, 7.3%; (2) often, 10.9%; (3) occasionally, 21.0%; (4) rarely,
20.9%; never, 33.6%. There is a 6.4% of missing values. Total number of observations is 52,972.
6
difficulties when teenager or not. Four of the countries in our sample (Austria, France,
Greece, and Portugal) do not report information on economic circumstances when
young (variable PM100) and so are not included in the graphical below, although we
will use them in Section 4.
Since we want to use the intergenerational module, we have to exclude from the
2005 cross section all individuals who are not in the age range (25-65) or are not the
selected respondent. Since we want to assess the long-run effect of household
characteristics, we also exclude those individuals who lived in a collective house or an
institution when young.
In Figure 1 below we represent for each country the fraction of individuals for
whom the variable poor_past is 1 in blue (PM100 is either 1 or 2). We also represent in
red the fraction of individuals who were most of the time in difficult circumstances
(PM100 is 1). As can be observed there is a lot of variation across countries, mainly
with regard to the fraction of individuals who had financial problems often or most of
the time. This goes from a maximum of 30.2% in Italy to a minimum of 7.4% in
Norway. The overall mean of poor_past is 19.3%.
0.1
.2.3
IT HU UK IE CY ES FI NL LU BE SE DK NO
Figure 1: Financial problems when young by country
Often or most of the time
Most of the time
7
As we said above, to the EU-SILC sample we merge data on public expenditure
in education from the UNESCO Database for Education. The UNESCO Database for
Education contains country data on public expenditure in education per student as a %
of per capita GDP at three levels (primary, secondary, tertiary).4 We want to construct a
variable that imputes to each individual a measure of the public resources used to
finance his/her education. We propose a simple approach which consists of assigning to
each individual a number (called “exp_p”) that corresponds to average public spending
in primary education during the years this individual age was between 6 and 11. This
could be considered as a rough measure of school quality. The idea is that this
individual went through primary education in that age bracket. As an example, consider
an individual who was born in Spain in 1970. This individual went to primary education
between 1976 and 1981. We compute average expenditure in primary education in the
years from 1976 to 1981, and assign this number to the variable exp_p for this
individual. In a similar way we can impute expenditure in secondary and tertiary
education. However, not all individual go to secondary or tertiary education. This is
why we focus only on expenditure in primary education, since this is the only
expenditure that all individuals within a country enjoy.
To illustrate the data we use, in Figure 2 below we show the ratio of per capita
expenditure on primary and secondary education to GDP per capita for the 17 countries
in our sample.
4 See http://www.uis.unesco.org/Education/Pages/default.aspx
8
As can be seen in Figure 2, there is variation both across countries and through
time. If we focus on primary education, Greece and Sweden are the lowest and highest
spending countries in this period, respectively. We also see that per capita expenditure
in primary education was below per capita expenditure in secondary education for most
countries (with the exceptions of Denmark, Hungary, Norway and Sweden). Regarding
per capita expenditure in secondary education over GDP per capita, Greece is again the
lowest spending country, while Belgium is now the highest spending country.
We also assume that educational attainment does not change after the age of 25
and we restrict our sample to include individuals born between 1960 and 1980. We also
exclude those individuals who were not born in the country. Our final sample consists
of 66,377 individuals from 17 countries.
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
02
04
06
0
1970 1980 1990 2000 1970 1980 1990 2000 1970 1980 1990 2000
1970 1980 1990 2000 1970 1980 1990 2000
AT BE CY DK ES
FI FR GR HU IE
IT LU NL NO PT
SE UK
Primary education
Secondary education
Source: UNESCO database.
Figure 2: Expenditure in education as % of per capita GDP
9
Our objective is to study whether public expenditure in primary education has an
effect on adult circumstances. To do this we focus on three endogenous variables that
are potentially interesting. The first one is the individual’s current poverty status. This is
the information contained in the variable HX080, which is an indicator of whether the
individual lives in a family with income below the poverty threshold. The poverty line
corresponds to 60% of equivalized household disposable income. We define a dummy
variable called “poor” which is 1 whenever HX080 is 1. The mean value of poor is
12.05%. We represent in Figure 3 the percentage of individuals below the poverty line
in each country. The maximum value corresponds to Spain (17.6%) and the minimum to
Denmark (5.8%).
Our second variable of interest is a variable describing individual’s health status.
In the EU-SILC there are several variables that provide information on health status.
From these variables we choose one (PH030) which describes whether individuals are
limited in daily activities because of health problems. This is a better measure of health
than PH010 (general health status). The reason is that several authors consider it as a
0.0
5.1
.15
.2
ES PT GR IT HU UK IE FI AT FR CY BE NO LU NL SE DK
Figure 3: Poverty rate by country
10
quasi-objective indicator.5 Possible answers to this question are 1 (strong limitation), 2
(limitation), and 3 (no limitation). We build a binary variable called “limited” that takes
value 1 when PH030 is either 1 or 2. In Figure 4 we represent the percentages of the
variable limited for the countries in our sample. The maximum value is in Finland
(27.6%) and the minimum in Greece (6.8%). The overall mean is 11.87%.
Our third and last endogenous variable measures whether the individual has
more than just compulsory education. We use the variable PE040 that describes the
highest ISCED education level attained by the individual. We construct a dummy
variable (“post-compulsory education”) that takes value 1 when PE040 is 3, 4 or 5. That
is, our variable corresponds to having more than lower secondary education. In Figure 5
we represent the fraction of the population in each country that has more than
5 There is an intense debate on the reliability of work-disability as true measure of health status in the
literature (see Benitez-Silva et al. (2004)). However work-disability has been found to be powerful
predictors for a range of outcomes and behaviors. Examples of such phenomena are labor supply
decisions (Stern (1989); Dwyer and Mitchell (1999)) and individuals’ decisions to apply for, and the
government’s decision to award, disability insurance benefits from the Social Security Administration
(Benitez-Silva et al. (1999)).
0.1
.2.3
FI HU IE ES LU BE SE AT PT NO UK NL FR CY DK IT GR
Figure 4: Limited individuals by country
11
compulsory education. The maximum is in Norway (96%) and the minimum in Portugal
(31.2%). The overall mean is 74.2%.
Our next objective is to study the correlation between our measure of the
economic situation when the individual was a teenager (poor_past) and each one of our
three endogenous variables of interest (poor, limited, noncomp_educ). The simplest way
of doing this is to compute probabilities for these three variables, conditional on the two
possible values of the variable poor_past. When we do this, we find a strong correlation
between this variable and each one of our three endogenous variables. This is what we
show in Table 1 below.
Table 1: Long-run effects of poverty
Poor Limited
Post-comp.
ed.
4.164 4.421 31.169
poor_past=0 10,50 11,14 78,98
1.718 1.583 5.595
poor_past=1 18,08 16,65 59,11
5.882 6.003 36.764
all 11,96 12,21 75,14
0.2
.4.6
.81
NO SE UK FI AT BE DK FR NL CY HU LU GR IE IT ES PT
Figure 5: Fraction with post-compulsory education
12
To read Table 1, let us focus on the first column (the one labeled “Poor”). The
probability of being below the poverty line today, conditional on not having
experienced financial problems when teenager, is 10.50%. But for those who
experienced those problems it is 18.08%. We also find striking differences with regard
to the probability of having limitations because of health problems (11.14% vs. 16.65%)
and in the probability of attending post-compulsory education (78.98% vs. 59.11%). We
illustrate this correlation in a series of figures in which we represent the probability of
being poor today (Figure 6), having limitations (Figure 7), and attending post-
compulsory education (Figure 8), depending on whether the household where the
individual was living had a good, bad, or very bad financial situation.
In Figure 6 we observe that the relation between being poor in the past and today
is weaker in those countries where the poverty rate is low, as in the Scandinavian
countries.
0.1
.2.3
IE ES IT BE CY HU LU UK FI NL NO SE DK
Figure 6: Poverty rate by financial situation when teenager
Good Bad
Very bad
13
Comparing figures 6 and 7, we see that the effect of past poverty on the
probability of experiencing limitations because of health problems today happens for all
countries. Moreover, this effect seems to be independent of the proportion of
individuals with limitations in the country.
Finally, in Figure 8 we observe that the effect of living in a poor household on
the probability of having more than compulsory education is larger in countries with a
low percentage of individuals with more than compulsory education. For countries
where almost the whole population has more than compulsory education (Norway,
Sweden, UK, and Finland) there are almost no differences.
0.1
.2.3
.4
FI LU HU NO BE CY SE ES IE UK NL DK IT
Figure 7: Limited by financial situation when teenager
Good Bad
Very bad
14
4. Empirical model
Since we intend to identify the causal relationship that public expenditure on
education has on our endogenous variables of interest, we need to control as accurately
as possible for additional factors affecting these dependent variables. Our set of
additional explanatory variables includes gender, age, age squared, together with a set
of household characteristics (family composition, education and occupation of parents,
etc). These household characteristics are included in the special module on inter-
generational transmission of poverty of the 2005 cross section of EU-SILC. The
inclusion of these additional controls helps us to reduce selection bias. Now the
important issue is that, when we include all these controls, we cannot identify the effect
of the variable poor_past. The reason is simply that this variable is very well explained
by the set of household characteristics we are using. However this is not a problem,
since our interest is not to identify the effect of the variable poor_past, but the effect of
the variable exp_p. In fact, there are two approaches that yield very similar estimates for
the effect of exp_p. In the first one, we regress our variable of interest with respect to
poor_past, exp_p, some individual characteristics (age, gender) and a set of country
dummies. In the second approach, we replace the variable poor_past with household
characteristics from the module. In the sequel, we will focus on this second approach.
0.2
.4.6
.81
NO SE UK FI BE DK CY HU NL LU IE IT ES
Figure 8: Post-compulsory education by financial situation when teenager
Good Bad
Very bad
15
Another reason for not including the variable poor_past in our analysis is that this
allows us to use data of the four countries that lack information of that variable (Austria,
France, Greece, and Portugal). In Table 2 we present the main descriptive statistics.
Insert Table 2 here
We propose to estimate a separate Probit model for each one of our three
endogenous variables of interest:
��� = �� + ���_�� + ���� + ��� + ��� , (1)
where ��� is our endogenous (binary) variable for individual i who lives in country c.
Variable �_�� is our measure of the size of the educational budget invested in a
particular individual, while ��� denotes family background variables to control for
differences in household characteristics. The term �� contains a set of dummy variables
to control for country-specific characteristics.
We assume that the error term ��� is not correlated with �_��. In addition, the
error term (conditional on the rest of explanatory variables) follows a normal
distribution. However, we check that a linear probability model gives almost identical
results for the estimation of the coefficient of interest (the coefficient of �_��).
We estimate each model for all individuals in the sample and for individuals
with different levels of education. In particular, we divide individuals into two groups
according to whether they have at most compulsory education or they have more than
just compulsory education. In addition, and because of the existing evidence on gender
differences in education, health, and poverty, we estimate gender-specific equations.
Finally, since we combine individual-level data with group-level data in our variable of
interest (exp_p), errors are clustered at the country and year of birth level.6
The crucial issue for identification is the assumption regarding exogeneity of
public expenditure. Variation in this variable arises because of differences in
expenditure across countries at the same point in time and differences in country
expenditure over time. Either difference could be partly endogenous with respect to
poverty rate, because given the same tax rate countries with a low proportion of poor
households raise more revenue than countries with a high proportion of poor
6 See Moulton (1986) for the importance of controlling for cluster effects.
16
households. In addition, some forms of expenditure are entitlements so that countries
with many poor households will have to spend more than countries with few poor
households. Other invariant country characteristics such as the political and social
climate could also be related to both country expenditure and children’s eventual
income. To partially address this issue, we add country fixed effects to the model to
control invariant factors within countries.
One nice feature of our data on public expenditure is that it is a percentage of per
capita GDP, which eliminates economic cycle issues.
5. Results
In Tables 3, 4, and 5 we compute the marginal effects corresponding to some of
the explanatory variables on our three endogenous variables poor, limited, and
noncomp_educ, respectively. For the sake of brevity we do not present the marginal
effects corresponding to the country dummies. We will focus first on Table 3, where the
endogenous variable is poor. We present the results of five different models. In Model 1
we include all individuals. In models 2 and 3 we separate individuals by gender. In
Model 4 we restrict our sample to individuals with at most compulsory education.
Finally, in Model 5 we only consider individuals with more than just compulsory
education. We do the same in Table 4, where the dependent variable is limited.
In Table 3 we see that the effect of being raised in a family with only a mother
on the probability of being poor strong, particularly for individuals who have at most
compulsory education. Another variable that has a positive effect on that probability is
having a father that is unemployed. However, both variables turn not significant when
we focus on individuals with post-compulsory education. On the other hand, it is clear
that parental education reduces very much the probability of being poor.
Interestingly, we find that the marginal effect of exp_p has always a negative
sign. This means that indeed educational expenditure in primary education reduces the
probability of being poor when adult, although this effect is not always statistically
significant. To give an idea of the size of the effect of exp_p on poor, consider the
model in which we pool all individuals (Column 1). In this case we obtain a marginal
effect of -.00216. This means that increasing exp_p by one percentage point (recall it is
17
measured as % of per capita GDP) reduces the probability of being poor by 0.216
percentage points. This is a sizable effect since the mean of poor is 12.05%. Moreover,
increasing exp_p by one standard deviation reduces the probability of being poor by
0.216*8.555 = 1.8 percentage points. This represents a 15% of the mean value of the
variable poor.
In Table 4 we find some differences with respect to Table 3. Although in all
cases the sign of the marginal effect of exp_p is negative, the only case in which we find
a statistically significant effect is when we consider only individuals with at most
compulsory education. In fact, for this group the marginal effect is very large.
Now we focus on Table 5, where the endogenous variable is post-compulsory
attendance. Again we observe that the being raised in a family with only a mother
reduces the probability of attending post-compulsory education, and mostly for women.
On the other hand, it is clear that parental education increases very much the probability
of attending post-compulsory education. We find that the marginal effect of exp_p has
always a positive sign. This means that indeed educational expenditure increases the
probability of attending post-compulsory education, although this effect is only
statistically significant in the first model in which we pool all individuals. The
magnitude of the effect is similar when we divide the sample by gender, although we
lose significance.
6. Conclusions
TBW.
18
References
Benítez-Silva H., Buchinsky M., Chan H-M., Rust J., Sheidvasser S. (1999): “An
empirical analysis of the social security disability application, appeal and award
process,” Labour Economics 6, 147–178.
Benítez-Silva H., Buchinsky M., Chan H-M., Cheidvasser S., Rust J. (2004): “How
large is the bias in self-reported disability?,” Journal of Applied Econometrics 19,
649–670.
Black, S. and Devereux, P. (2011): “Recent Developments in Intergenerational
Mobility.” In: Ashenfelter, O., Card, D. (Eds.), Handbook of Labor Economics, Vol.
4, Part B. Elsevier, Amsterdam, 1487–1541.
Case, A. Fertig, A. and Paxson, C. (2005): “The lasting impact of chilhood health and
circumstance,” Journal of Health Economics 24, 365-389.
Conti. G, Heckman, J. and Urzua, S. (2010): “The Education-Health Gradient,”
American Economic Review 100, 2, 234-238.
Cutler, D.M., Lleras-Muney, A. (2006): “Education and Health: Evaluating Theories
and Evidence”. NBER Working Paper 12352.
Currie, J., and Hyson, R. (1999): “Is the impact of health shocks cushioned by socio-
economic status? The case of low birthweight,” American Economic Review Papers
and Proceedings 89, 2, 245-250.
Dwyer D.S. and O.S. Mitchell (1999): “Health problems as determinants of retirement:
are self-rated measures endogenous?” Journal of Health Economics 18(2): 173–193.
Grogger, J. (1996): “School expenditures and post-schooling earnings: evidence from
the high school and beyond,” Review of Economics and Statistics 78, 4, 628–637.
Hanushek, E. (1996): “School resources and student performance.” In: Burtless, G.
(ed.), Does Money Matter? The Effect of School Resources on Student
Achievement and Adult Success. Brookings Institution Press, Washington, D.C.
19
Hanushek, E. (2001): “Spending on schools.” In: Moe, T. (Ed.), A Primer on
American Education. Hoover Institution Press, Palo Alto, CA, pp. 69–88.
Hedges, L.V., Laine, R., Greenwald, R. (1992): “Does money matter? Meta-analysis of
studies of the effects of differential school inputs on student outcomes.” Educational
Researcher 23, 3, 5–14.
Lochner, L. (2011): “Non-Production Benefits of Education: Crime, Health, and Good
Citizenship.” In: Hanushek, E., S. Machin, and L. Woessmann (eds.), Handbook of the
Economics of Education, Vol. 4, Ch. 2, Amsterdam: Elsevier Science.
Mayer, S. E. (2002): “The Influence of Parental Income on Children's Outcomes: A
Review.” Report to the New Zealand Ministry of Social Development.
Mayer, S. E., Lopoo, L. M. (2008): “Government spending and intergenerational
mobility,” Journal of Public Economics 92, 139–158.
Mayer, S.E., Sarin, A. (2005): “An assessment of some mechanisms linking economic
inequality and infant mortality,” Social Science and Medicine 60, 3, 439–455.
Meghir, C., Palme, M. (2005): “Educational Reform, Ability, and Family Background,”
American Economic Review 95, 1, 414-424.
Moulton, B. (1986): “Random group effects and the precision of regression estimates”
Journal of Econometrics, 32(3): 385-397.
Stern S. (1989): “Measuring the effects of disability on labor force participation”.
Journal of Human Resources 24: 361–395.
Wolff, P., Montaigne, F., and Rojas González, G. (2010): “Investing in statistics: EU-
SILC,” Chapter 2 in A. B. Atkinson and E. Marlier (eds), Income and Living
Conditions in Europe. Luxembourg: Eurostat, 38-55.
20
Table 2: Summary Statistics
Variable Mean St. Dev. Min Max N. obs.
Poor today 0.120 0.326 0 1 66363
Limited 0.119 0.323 0 1 66191
Post-compulsory education 0.742 0.438 0 1 65384
Poor when young 0.193 0.395 0 1 49188
Exp. primary educ. (% GDP per
capita) 16.91 8.551 4.358 56.05 66003
Single mother family 0.0734 0.261 0 1 64649
Father secondary education 0.437 0.496 0 1 57733
Father tertiary education 0.109 0.312 0 1 57733
Mother secondary education 0.414 0.492 0 1 60000
Mother tertiary education 0.0786 0.269 0 1 60000
Father unemployed 0.00909 0.0949 0 1 60266
Father blue collar 0.493 0.500 0 1 55779
Female 0.509 0.500 0 1 66377
Age 34.98 5.450 24 45 66377
Age squared 1254 379.1 576 2025 66377
Not citizen 0.00457 0.0674 0 1 66302
21
Table 3: Endogenous variable is poor
(1) (2) (3) (4) (5)
VARIABLES All Men Women Comp. Educ
Post-Comp.
Educ
age -0.0133** -0.0238*** -0.00299 0.00548 -0.0175***
(0.00523) (0.00603) (0.00600) (0.0135) (0.00432)
age2 0.000179** 0.000331*** 2.87e-05 -9.63e-05 0.000232***
(7.36e-05) (8.51e-05) (8.51e-05) (0.000191) (6.19e-05)
gender 0.0152*** 0.0528*** 0.00871***
(0.00286) (0.00867) (0.00260)
singlemotherfam 0.0370** 0.0381* 0.0373* 0.145*** 0.00383
(0.0147) (0.0206) (0.0195) (0.0395) (0.0124)
oldfather 0.00787 0.00459 0.0116 0.00557 0.00615
(0.00641) (0.00886) (0.00905) (0.0141) (0.00693)
oldmother 0.00627 0.0180 -0.00595 0.0222 -0.00535
(0.00975) (0.0149) (0.0114) (0.0202) (0.00997)
educfatherM -0.0527*** -0.0499*** -0.0547*** -0.0795*** -0.0202***
(0.00419) (0.00619) (0.00528) (0.0126) (0.00387)
educfatherH -0.0543*** -0.0446*** -0.0640*** -0.0889*** -0.0232***
(0.00479) (0.00652) (0.00635) (0.0306) (0.00462)
educmotherM -0.0535*** -0.0508*** -0.0562*** -0.0753*** -0.0285***
(0.00462) (0.00648) (0.00602) (0.0144) (0.00427)
educmotherH -0.0484*** -0.0482*** -0.0483*** -0.0521 -0.0257***
(0.00539) (0.00666) (0.00775) (0.0485) (0.00482)
father_manual -0.00309 -0.00967** 0.00310 -0.0239*** -0.00395
(0.00320) (0.00397) (0.00475) (0.00919) (0.00306)
fatherunemp 0.0956*** 0.112*** 0.0832** 0.183*** 0.0253
(0.0283) (0.0401) (0.0348) (0.0633) (0.0264)
exp_p -0.00216*** -0.00258*** -0.00177* -0.00292 -0.00143**
(0.000767) (0.000962) (0.000966) (0.00211) (0.000683)
noncitizen 0.0825** 0.0653* 0.103** 0.0962 0.0749**
(0.0337) (0.0357) (0.0495) (0.0778) (0.0313)
Observations 47,301 22,931 24,370 11,803 35,244
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
22
Table 4: Endogenous variable is limited
(1) (2) (3) (4) (5)
VARIABLES All Men Women Comp. Educ
Post-Comp.
Educ
age 0.00444 0.0147*** -0.00538 0.0195** 0.00115
(0.00360) (0.00462) (0.00488) (0.00854) (0.00387)
age2 -8.35e-06 -0.000162** 0.000139** -0.000210* 2.96e-05
(5.14e-05) (6.65e-05) (7.01e-05) (0.000122) (5.53e-05)
gender 0.0109*** 0.00225 0.0153***
(0.00280) (0.00596) (0.00317)
singlemotherfam 0.0517*** 0.0360* 0.0651*** 0.0566* 0.0439***
(0.0132) (0.0191) (0.0204) (0.0320) (0.0136)
oldfather -0.00713 0.00564 -0.0184** -0.0129 -0.00798
(0.00570) (0.00906) (0.00817) (0.0116) (0.00634)
oldmother 0.00327 -0.00361 0.0105 0.0105 0.000760
(0.00946) (0.0126) (0.0127) (0.0218) (0.0117)
educfatherM -0.00720* -0.00740 -0.00712 0.0125 0.000635
(0.00425) (0.00596) (0.00563) (0.0117) (0.00451)
educfatherH -0.0136** -0.0190** -0.00806 0.0949*** -0.00565
(0.00546) (0.00785) (0.00796) (0.0361) (0.00602)
educmotherM -0.0167*** -0.0165** -0.0162** -0.00786 -0.00775
(0.00456) (0.00666) (0.00639) (0.0127) (0.00497)
educmotherH -0.0315*** -0.0298*** -0.0328*** -0.000540 -0.0198***
(0.00609) (0.00796) (0.00843) (0.0349) (0.00637)
father_manual 0.0165*** 0.0197*** 0.0135*** 0.0144** 0.0126***
(0.00293) (0.00401) (0.00430) (0.00626) (0.00340)
fatherunemp 0.0149 0.0532 -0.0166 0.0645 -0.00542
(0.0220) (0.0372) (0.0294) (0.0520) (0.0259)
exp_p -0.000744 -0.00127 -0.000295 -0.00257* -0.000306
(0.000473) (0.000816) (0.000696) (0.00140) (0.000472)
noncitizen 0.0176 0.0123 0.0212 -0.0361 0.0311
(0.0204) (0.0282) (0.0327) (0.0362) (0.0261)
Observations 47,290 22,924 24,366 11,806 35,230
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
23
Table 5: Endogenous variable is postcomp_educ
(1) (2) (3)
VARIABLES All Men Women
age -0.0102 -0.00505 -0.0176**
(0.00656) (0.00838) (0.00706)
age2 7.67e-05 3.13e-05 0.000157
(9.45e-05) (0.000121) (0.000101)
gender 0.0324***
(0.00451)
singlemotherfam -0.0773*** -0.0706** -0.0784***
(0.0212) (0.0336) (0.0247)
oldfather -0.0207** -0.0323** -0.0110
(0.00896) (0.0134) (0.0111)
oldmother -0.0175 -0.00629 -0.0263
(0.0146) (0.0223) (0.0171)
educfatherM 0.131*** 0.136*** 0.125***
(0.00626) (0.00833) (0.00777)
educfatherH 0.187*** 0.205*** 0.169***
(0.00525) (0.00703) (0.00654)
educmotherM 0.113*** 0.120*** 0.106***
(0.00641) (0.00812) (0.00859)
educmotherH 0.171*** 0.171*** 0.173***
(0.00605) (0.00906) (0.00705)
father_manual -0.0559*** -0.0440*** -0.0647***
(0.00446) (0.00614) (0.00595)
fatherunemp -0.0213 0.0448 -0.0769*
(0.0288) (0.0344) (0.0407)
exp_p 0.00148* 0.00155 0.000965
(0.000892) (0.00127) (0.000980)
noncitizen -0.0372 -0.0317 -0.0494
(0.0336) (0.0420) (0.0479)
Observations 47,062 22,805 24,180
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1