SERIES

IZA DP No. 10077

PAPER

In God We Learn? Religions’ Universal Messages,

Context-Specific Effects, and Minority Status

Pierre-Guillaume Méon
DISCUSSION

Ilan Tojerow

July 2016

Forschungsinstitut
zur Zukunft der Arbeit
Institute for the Study
of Labor
 In God We Learn? Religions’ Universal
Messages, Context-Specific Effects,
and Minority Status

Pierre-Guillaume Méon
Université libre de Bruxelles (CEB)

Ilan Tojerow
Université libre de Bruxelles (CEB and DULBEA)
and IZA

Discussion Paper No. 10077

July 2016

IZA

P.O. Box 7240

53072 Bonn
Germany

Phone: +49-228-3894-0
Fax: +49-228-3894-180
E-mail: iza@iza.org

Any opinions expressed here are those of the author(s) and not those of IZA. Research published in
this series may include views on policy, but the institute itself takes no institutional policy positions.
The IZA research network is committed to the IZA Guiding Principles of Research Integrity.

The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center
and a place of communication between science, politics and business. IZA is an independent nonprofit
organization supported by Deutsche Post Foundation. The center is associated with the University of
Bonn and offers a stimulating research environment through its international network, workshops and
conferences, data service, project support, research visits and doctoral program. IZA engages in (i)
original and internationally competitive research in all fields of labor economics, (ii) development of
policy concepts, and (iii) dissemination of research results and concepts to the interested public.

IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion.
Citation of such a paper should account for its provisional character. A revised version may be
available directly from the author.
IZA Discussion Paper No. 10077
July 2016

ABSTRACT

In God We Learn? Religions’ Universal Messages,

Context-Specific Effects, and Minority Status*
We study the relationship between major religious denominations and individuals’ levels of
education, using the World Values Survey. In a first step, running country-by-country
regressions, we report first-time evidence that no single denomination has a universal effect
on education. Each denomination has a positive and statistically significant effect in some
countries, a negative and statistically significant effect in others, and a statistically
insignificant effect elsewhere. In a second step, we relate the sign of the impact of a
denomination in a country to whether the denomination is a minority in that country. We find
that denominations that are a minority in a country are more likely to be associated with a
higher level of education, and less likely to be associated with a lower level of education in
that country. In both steps, the findings are independent from the specification of the
regressions used in the first stage to determine the sign of the impact of denominations on
educational outcomes. The finding of the second step is moreover robust to defining minority
denominations using various thresholds. It is robust to controlling for whether the
denomination is a state religion, for the country’s level of democracy, per capita GDP, or level
of education, to introducing denomination- and country- fixed effects, and to controlling for
the identity of the largest other denomination in the country.

JEL Classification: I2, O5, Z1

Keywords: religion, education, minority

Corresponding author:

Ilan Tojerow
Université libre de Bruxelles
CP 114/02
50, Av. F.D. Roosevelt
1050 Brussels
Belgium
E-mail: itojerow@ulb.ac.be

*
We thank François Facchini, Victor Ginsburgh, Erich Gundlach, Petros Sekeris, participants of the
European Public Choice Society conference in Cambridge and at the European Workshop in Political
Macroeconomics, in Mainz, for helpful comments and suggestions, as well as participants in seminars
at Tulane University, the University of Hamburg, the University of Portsmouth, the University of Liège,
and Université libre de Bruxelles. The authors claim the sole responsibility for remaining errors and
approximations.
1. Introduction

Nearly one century after the publication of Max Weber’s (1904) classic and after
spending decades in the purgatory of economics, religion has once again resurfaced as an
explanation of economic performance. Barro and McCleary (2003, 2006) have thus
econometrically established a link between religiosity and growth in contemporary
economies. Becker and Wössmann (2009) moreover report a cross-country association
between per capita GDP and the share of Protestants in 1900. Various channels have been
investigated to account for the correlation between specific religious denominations and
economic performance, be it the role of the state, as argued by Kuran (1996, 1997, 2004) or
Platteau (2008), the impact of religion on values and attitudes towards economic activities,
which is Weber’s (1904) classic argument and has for instance been illustrated by Guiso et
al. (2003) or Hillman (2007).

One implication of this impact of religion on attitudes concerns education, which

directly promotes development. While the impact of education on development is
uncontroversial, as Hannushek and Woessmann’s (2008) survey shows, some denominations
put more emphasis on education than others. Some denominations even promote literacy as a
religious duty. For instance, Botticini and Eckstein (2005, 2007) recall that male Jews must be
able to read the Torah in the Synagogue and teach it to their sons. Similarly, the principle of
the priesthood of all believers implies that each Protestant should be able to read the Holy
Scriptures.

All the arguments relating specific denominations to education share a common

feature: They assume that denominations carry universal messages resulting in a universal
incentive to acquire more, or less, education. They therefore imply that the impact of each
denomination should be the same across countries. Yet, religion is also a social activity that
could be considered as a club good (Iannaccone, 1992), hence we may assume that the same
denomination may prompt its members to adopt different behaviors in different social
contexts. A denomination may be dominant in one country and persecuted in others. It may be
a majority in some countries and a minority in others. As a result, one may expect the impact
of religious denominations to be context-dependent.

Unfortunately, that contention has received little attention so far. Most existing studies
of the impact of religion on education are either country-specific or based on cross-country
regressions. This is the case of Chiswick (1988) or Glaeser and Sacerdote (2008) for the

1
United States, Brown and Taylor (2007) for the United Kingdom, or Blunch (2007) for
Ghana. By construction, they cannot therefore compare the effect of religious denominations
across countries. Some studies take a more macroeconomic standpoint, and are based on
cross-country comparisons. This is what Guiso et al. (2003), Schaltegger and Torgler (2010),
or Arruñada (2010) do, although they focus on attitudes rather than on education. However, in
such regressions one runs the risk of attributing to religion differences that are related to other
country-specific traits, such as geography or institutional quality. Most of all, those
regressions force the effect of religious denominations to be the same across countries.

In this paper, we specifically let the impact of religious denominations differ across
countries. To do so, we follow Guiso et al. (2003) and use the World Values Survey, which
provides individual data on educational attainment, religious denomination, and religiosity in
a large sample of countries. Unlike Guiso et al. (2003), we take advantage of the two-level
structure of the World Values Survey to separately estimate the impact of religious
denominations in all countries.

Our first key finding is that no single religious denomination has a universal impact on
education. Each denomination has at the same time a positive and statistically significant
effect in some countries, a negative and statistically significant effect in other countries, and a
statistically insignificant effect elsewhere. In other words, we find the effect of religious
denominations to be country-specific. The finding is robust to controlling for individuals’
level of religiosity or not, and to focusing on individuals born in their country of residence. It
holds equally for individuals below and over forty years old, and for male and female
respondents.

In a second step, we investigate what determines the effect of a religious denomination

in a country estimated in the first step. We more specifically test the role of the
denomination’s minority status. We expect that being a minority religion is a key determinant
of the impact of a denomination on educational outcomes, although the sign of the effect on
education of being a minority is ambiguous a priori. One the one hand, if minority
denominations are discriminated against by teachers (Coate and Loury, 1993, Hannah and
Linden, 2012), internalize negative stereotypes (Hoff and Pandey, 2006), have a limited
access to social capital (Coleman, 1988), or oppose the values of the majority to avoid a loss
in self-image (Akerlof and Kranton, 2000, 2002), then they should be more likely to result in
lower educational outcomes. On the other hand, if investing in education is a way to cope
with discrimination (Brenner and Kiefer, 1981), or if members of minority religions

2
compensate for their lack of connections with the majority by stronger ties within the
community (Coleman, 1988), benefit from positive stereotypes (Shih et al., 2002), or develop
an identity that emphasizes academic effort (Akerlof and Kranton, 2002), then minority
denominations may be associated with better educational outcomes.

To determine which effect prevails, we perform a cross-country study of the

determinants of the marginal effect on education of religious denominations being
significantly negative, significantly positive, or insignificant. In a series of multinomial logit
regressions, we regress the impact of a religious denomination in a particular country to a
dummy variable capturing the denomination’s minority status in the country. We find that the
likelihood of the impact on education of a denomination in a country being significantly
negative is lower, and the likelihood of the impact being significantly positive larger, if the
denomination is a minority. The finding is robust to a series of alternative definitions of
minority denominations. It does not depend on the specification of the regressions used in the
first stage to determine the sign of the impact of denominations on educational outcomes. It is
robust to controlling for whether the denomination is a State religion, for the country’s level
of democracy, per capita GDP, or level of education. It resists introducing denomination- and
country- fixed effects, and to controlling for the identity of the largest other denomination in
the country.

Those results directly contribute to our understanding of the impact of religious

denominations on education, by emphasizing that their impact is context-dependent and
highlighting that it depends on being a minority. They more broadly contribute to our
knowledge of the impact of culture, surveyed in Alesina and Giuliano (2015), and identity,
surveyed by Kranton (2016). We show that the same cultural trait or identity characteristic,
religious denomination, can have opposite effects across countries depending on the context.

To achieve those results, the rest of this paper is organized as follows. In the next
section, we survey the literature on the impact of religious denominations on education. In the
following section, we perform a country-by-country study of the impact of denominations on
education. We observe wild differences across countries in the effect of religious
denominations on education that are hidden by cross-country regressions. In section 4, we
investigate the determinants of the sign and statistical significance of the marginal impact of
religious denominations. The last section concludes.

3
2. The Impact of Religious Denominations on Education

In this section, we discuss the possible relationship with education of the main religious
denominations in our study. We first, provide arguments suggesting a uniform impact of those
denominations across countries, then argue that their impact is likely context-dependent.

2.1. The Implication of Religious Denominations for Education

Some religious denominations have been interpreted as conveying universal norms and
incentives to acquire more or less education. We review those denominations here, focusing
in turn on Judaism, Islam, Protestantism, Buddhism, and Hinduism.

Judaism has often been assumed to emphasize education. An alternative explanation of

the impact of Judaism on education, for instance put forward by Brenner and Kiefer (1981), is
that Jews have historically been discriminated against in most parts of the world, and faced a
risk of confiscation of their physical capital. As a result, they had an incentive to invest in
education, which is a portable form of capital. This may have tilted the trade-off between the
quantity and the quality of children towards quality, as argued by Chiswick (1988). Botticini
and Eckstein (2005, 2007) stress the obligation for male Jews not only to be able to read the
Torah in the Synagogue but more to the point to teach it to their sons. The stance of Judaism
on general education is, however, more mixed. Hollander et al. (2003) report a theological
debate on the issue. While some interpretations of the Torah recommend a balance between
religious studies and worldly occupations, Hollander et al. (2003) claim that specific
interpretations of the Bible encourage religious versus general studies. They recall that
interpretations known as Midrashim claim that Jews should engage only in studies of the
Torah, and not secular studies. Hollander et al. (2003), however, admit that mainstream
Jewish thought holds a more balanced view, at least for the broad population, as opposed to
selected scholars.

Bobrick (2001) argues that Islam is based on a more oral tradition than Judeo-Christian
religions, resulting in lower incentives concerning literacy for Muslims. The word “Quran”
means “recitation” in Arabic. Another feature of Islam that may negatively affect educational
outcomes is the observance of Ramadan. Oosterbeek and van der Klaauw (2013) compare the
results of Muslim and non-Muslim students in microeconomics tests at the University of
Amsterdam, during years when Ramadam falls during the course and at another time. They

4
find that one additional week of Ramadan exposure reduces the final grade of Muslim
students for the microeconomics course by almost 10% of a standard deviation.

A key feature of Islam is that it offers religious education including basic reading of the
Koran in madrasas and maktabs (Borooah and Iyer, 2005, and Chaudhury and Rubin, 2011).
The impact of madrasas on non-religious education is ambiguous. On the one hand, religious
schools devote time and effort to teaching religious subjects at the expense of secular
subjects. Chaudhary and Rubin (2011) argue that, as a result, a greater prevalence of Muslim
religious schools results in a wider Muslim-Hindu literacy gap in India. On the other hand, it
can be argued that Muslim schools can provide a complementary educational system. In
Bangladesh, state-registered and -financed madrasas called aliyah teach Bengali, English,
mathematics and sciences, and their curricula are nationally defined by a national board that
also runs national examinations (Asadullah and Chaudhury, 2010). A similar system was
implemented in Indonesia (Newhouse and Beegle, 2006) and India (Borooah and Iyer, 2005).
Whether chartered madrasas can provide a quality of education comparable to that of secular
schools is a point of empirical debate. In some Indian states, madrasas are viewed as
complements to the formal education sector (Borooah and Iyer, 2005). In Indonesia,
Newhouse and Beegle (2006) could find no difference between religious and secular private
schools in the public examination records of graduates of secondary junior schools. In
Bangladesh, Asadullah et al. (2007) also find no difference between religious and secular
schools at the secondary level, but observe a learning deficit among graduates of primary
madrasas. Finally, Sander (2010) observes that Muslims in the US exhibit higher education
than Protestants and Catholics.

Protestantism is usually believed to encourage literacy, because each believer should

be able to have an unmediated access to the Holy Scriptures, in line with the principle of the
priesthood of all believers. Becker and Woesmann (2009) recall that Martin Luther explicitly
advocated education. He not only made it an obligation for rulers to build schools, but also
urged parents to keep their children to school (Luther, 1524, 1530). Literacy, and education in
general, are therefore hardwired in Protestant ethics, because they are a religious duty.
Woodberry (2011) moreover argues that Protestant missionaries exported literacy by starting
schools and pioneering phonetic alphabets and forms of writing accessible to a larger
population.

Interestingly, although his main argument focused on work ethics, Weber (1904) cited
his student Offenbacher’s (1900) study on secondary school choices of Catholics, Protestants

5
and Jews in the first chapter of his book, pointing to an over-representation of Protestants in
institutions that prepared for technical and commercial occupations (Realgymnasium and
Realschulen), while Catholics preferred a more general type of education (Gymnasium).
Becker and Woesmann (2009) even argue that the impact of Protestantism on literacy
accounts for most of the higher affluence of Protestants in XIXth century Prussia.

Indeed, greater literacy rates among Protestants than Catholics have been repeatedly
observed in various countries and times, such as the US in mid-XIXth century (Go and
Lindert, 2010), the first half of the XXth century (Goldin and Katz, 2000), late XIXth century’s
Ireland (Cipolla, 1969), or Finland (Markussen, 1990). More recently, Glaeser and
Glendon (1998) found a stronger connection between religiosity and education among
Calvinist Protestants than among Catholics, on contemporary US data. Blunch (2007) found,
in his study of educational attainment in Ghana, that Protestant breeds of Christianity are
associated with higher levels of education than Catholicism.

One should note, however, that Protestantism is heterogeneous, and that some
Protestant denominations hold conservative views on the teaching of scientific disciplines,
perceived as hostile to their faith and to the conviction that the Bible is inerrant, as Darnell
and Sherkat (1997) or Beyerlein (2004) argue. Accordingly, Darnell and Sherkat (1997),
Lehrer (1999), or Beyerlein (2004) have observed lower educational levels among
conservative Protestant denominations than among other Protestant and non-Protestant
denominations, using various surveys in the US.1

Roman Catholicism is the implicit reference group in Max Weber’s (1904) book, and
probably in most of the works devoted to the educational advantage of Protestants.2 It is
therefore implicitly assumed to be less conducive to education. This implicit belief may
contrast with the network of Catholic schools and higher education institutions around the
world. Morey and Piderit (2010), for instance, count no less than 220 Catholic colleges and
universities only in the United States. The Catholic Church also had developed an intellectual
tradition on education and the role of reason, going back to the Early Church Fathers and the
Middle Ages. However, while that tradition recognized that faith and reason were compatible,

1
Conservative Protestants are defined in those studies as three denominations: Fundamentalists,
Pentecostal/Charismatic, and Evangelical. Christians. Beyerlein (2004) observes that the result that conservative
Protestants are less educated is driven by Fundamentalist and Pentecostal Protestants, while Evangelicals exhibit
above-standard educational attainments.
2
Weber’s (1904) comments of Offenbacher’s (1900) figures on the secondary school choices of Catholics,
Protestants and Jews, for instance essentially focus on the difference between Protestants and Catholics.

6
it considered that secular matters were subordinate to the Church’s religious teaching. A
specificity of the Catholic Church vis-à-vis Protestant denominations, and most other
denominations, is that it has a central authority that can establish an official doctrine. It only
started establishing its doctrine on education in the XXth century, essentially in two
documents (McClelland, 1996). The first is the encyclical Divini illius magistri published by
Pope Pius XI in 1929. Although it acknowledged the role that the State could play in
education, the encyclical recalled that secular education was subordinate to religious
education and forbade Catholic children to attend laic schools. The second document is the
Second Vatican Council’s Declaration on Christian Education, entitled Gravissimum
educationis and promulgated in 1965 by Pope Paul VI. It took a more moderate stance,
emphasizing the positive roles of education and science, and featuring no explicit ban on
secular or laic education. The overall position of the Roman Catholic Church on education
therefore evolved over the course of the century, and its overall impact is ambiguous,
especially as the official doctrine may be applied and interpreted in different ways in different
countries or by different institutions.

The impact of other religions has, to our knowledge, received less attention, at least in
the economics literature. One may contend that Buddhism has a positive impact on
education, because of the importance it gives to universal access to the teaching of the
Buddha. Ling (1984) argues that Buddhism is essentially a matter of teaching. The Buddha
himself is portrayed as a teacher with an aim to address everyone without discrimination. As a
result, in some traditions, Buddhism stresses the need to educate the largest number of people
to read and write. In both Burma and Thailand, monasteries were thus instrumental in
spreading literacy, Ling (1984) argues, resulting in high literacy rates.3 Secular governments
even leveraged on those monasteries to spread education. Using contemporary data,
Sander (2010) reports evidence that Buddhists having lived in the US at least since age 16
exhibit higher educational achievements than Protestants and Catholics. By contrast,
Ling (1984) stresses that Buddhism’s emphasis on universal access to the teaching of the
Buddha stands in stark contrast with Hinduism, which sees parts of the population as unfit to
the teaching of Brahmans in Hinduism.

3
Ling (1984) recalls that in the first decade of the XXth century, the rate of literacy for all Burmese males was 49
percent, while only 12 percent for Madras.

7
2.2. Country-Specific Impacts of Denominations on Education: the Role of Minority Status

The views surveyed above imply that the impact of a religious denomination on
education is independent from the context where it occurs. However, religion is intrinsically a
social activity. Its impact on individual believers is therefore likely affected by their
environment, and the behavior of others, be it members of the same denomination or of other
groups in society. In other words, it is bound to be context-dependent. If the impact of a
denomination is context-dependent, then one should isolate the dimensions of the context that
determine its impact. In this section, we argue that there are strong reasons to expect that
being a minority religion is a key determinant of the impact of a denomination on educational
outcomes, although the impact may a priori be ambiguous.

At first pass, it is tempting to expect minority religions to lead to lower educational

outcomes, because their members have more limited opportunities. The impact of minority
denominations may, however, also be due to the attitude of their very members. First, in his
paper introducing the notion of social capital, Coleman (1988) emphasizes the role of social
capital for the accumulation of human capital. He defines social capital as the social structures
that facilitate certain actions. Children from families that are integrated in a network may thus
benefit from their families’ network when acquiring human capital. Conversely, children who
lack such a network would perform less well. In line with that contention, Coleman (1988)
observes that children from families who have moved more frequently tend to drop-out of
school more often. By definition, the networks of families belonging to minority
denominations are more limited, and one may contend that their children have fewer
opportunities to benefit from interactions with the rest of the population than children from
the majority, and acquire less human capital.

Minority religions may also directly suffer from discrimination in education. Hannah
and Linden (2012) document that graders tend to give lower grades to papers that are
randomly attributed to lower caste students. The same mechanism may apply to students
whose denomination is stigmatized.

Part of the effect of being a minority religion may also be driven by the reaction of the
discriminated-against students. Investing less in the acquisition of human capital may be a
rational response to discrimination, as Coate and Loury (1993) argue, if discriminated-against
students face a larger cost of acquiring education than their non-discriminated counterparts.
Behavioral mechanisms may also be a work. The Pygmalion effect initially documented by

8
Rosenthal and Jacobson (1968) prompts pupils to conform to their teachers’ expectations.
Accordingly, pupils belonging to minority religions may suffer from their teachers’ lower
expectations, and end up acquiring less human capital. In a similar way, students may
erroneously internalize a negative stereotype of inferiority. Hoff and Pandey (2006) thus
observed that simply revealing that pupils were members of a disadvantaged caste reduced
their performance in solving mazes on their own, and reduced their expectations of their own
performance. By the same token, membership in a minority religion may lower the
performance of pupils and students.

Akerlof and Kranton’s (2000) theory of identity and its (2002) application to education
provide a similar rationale. Akerlof and Kranton’s (2002) model assumes that students not
only maximize the pecuniary payoffs of education but also get utility from identifying with a
group. To do so, they must share the attributes of the group and conform to its norms in terms
of effort. Students will identify with the group that provides them the largest utility. Students
from the leading group are those that best conform to the school’s ideal. They therefore exert
effort to reach the school’s ideal and achieve academic success. However, identifying with the
leading group is costly in terms of self-image if one does not share its attributes, for instance
in terms of looks or social network. Students who do not have those attributes may therefore
be better off identifying with another group that rejects the school’s ideal, emphasizes low
effort, and obtains lower academic success. Belonging to a minority religion may precisely be
the type of attributes that prevents conforming to the ideal of the leading group, who typically
belongs to the majority. Minority students may therefore choose to identify with a group that
opposes the school’s ideal, with negative consequences on their academic achievements.

While the mechanisms discussed so far suggest that members of minority religions
should acquire less education, a closer look at the same arguments can also result in the
opposite contention. First, Coleman’s (1988) concept of social capital is not only quantitative.
Families belonging to minority denominations may compensate their relative lack of
connections with the rest of society by stronger relationships and a larger commitment to
education within the family. Coleman’s (1988) thus recalls the case of Asian immigrant
families who purchased two copies of textbooks needed by the child, so that the mother would
be able to help her child. In addition, the minority group may collectively compensate for its
minority status by developing stronger community ties, resulting in more social capital, not
less, and better educational outcomes. He thus relates the better performance of Catholic
schools in the US to the role of the adult community around those schools.

9
 Disproportionately investing in education may also be a way to compensate for
discrimination. In extreme cases, a minority facing persecution or the need to seek refuge
would have an incentive to invest in education, because human capital, unlike physical
capital, cannot be seized and is easily transferred across countries. Brenner and Kiefer (1981)
use that line of reasoning to explain the finding that the level of education of Palestinians
living in Arab countries after 1948 increased, and resembled that of Jews living in the US.

Stereotypes and identities may also encourage the acquisition of human capital. Indeed,
all stereotypes are not necessarily negative. If a minority group is for instance perceived as
“good at math” or “hard-working”, then teachers may raise their expectations, resulting in a
positive Pygmalion effect for the minority group. Shih et al. (2002) report experimental
evidence that subtly activated positive stereotypes can enhance academic performance of the
target group.4

Akerlof and Kranton’s (2000, 2002) model of identity performance also produces mixed
predictions. Not being able to identify with the leading group may enhance academic
performance for students who identify with a group that lacks the attributes of the leading
group but values academic performance. The 2002 model applied to academic performance
considers three groups of students, referred to as jocks, nerds and burnouts, and two types of
skills, for simplicity looks and academic ability. Jocks are the leading group, insofar as they
get the largest utility from identifying with the group, followed by nerds, and burnouts. The
key group attribute is looks for jocks and academic ability for nerds, while burnouts have no
specific attribute. The model implies that a student whose looks do not conform to those of
jocks will opt to identify with one of the two other groups. Only if her ability is too small will
she identify with burnouts. With sufficient ability, she will identify with nerds. As that group
emphasizes academic performance, students identifying with it will have an incentive to
increase academic effort to conform to the group’s ideal, thereby improving their academic
performance. The model therefore implies that if students from a religious minority are
prevented from identifying with the leading crowd, they may have an incentive to increase
academic effort. Their minority status would thus result in the acquisition of more human
capital, not less.

4
Positive stereotypes are, however, no unmitigated blessing. Shih et al. (2002) report that positive stereotypes
can indeed reduce academic performance if they are blatantly imposed on their targets. See Czopp et al. (2015)
for a survey of the effects of positive stereotypes.

10
 In a similar way, Iannaccone (1992) and Berman (2000) emphasize that religion is a
club good, insofar as the benefit from religious participation depends not only on individuals’
own inputs but also on the inputs of others. The impact of a given denomination on behavior
in general therefore depends on the size of the religious community or the need to distinguish
its members from the rest of society. Berman (2000) accordingly argues that the lengthening
of religious studies (yeshiva) among ultra-orthodox Jews in Israel is a rational reaction to the
difficulty to signal commitment to the ultra-orthodox community in a predominantly Jewish
society. Berman (2000) remarks that ultra-orthodox Jews with the same geographic origin as
those who live in Israel but who live in countries that are not predominantly Jewish, for
instance in Central and Eastern Europe, Canada or the US, stop attending yeshiva much
earlier than in Israel. Berman’s (2000) interpretation of this finding is that sending a signal
that one belongs to the ultra-orthodox community is much easier in predominantly non Jewish
societies than in a society that is predominantly Jewish. The impact of belonging to the ultra-
orthodox denomination therefore has a clear context-dependent effect on religious education.
It is weaker in countries where Jews are a minority. As attending yeshiva comes at the cost of
secular education, one should expect ultra-orthodox Jews to have an incentive to acquire more
secular education in countries where Judaism is a minority religion. The same line of
reasoning may apply to other denominations elsewhere. What it implies is that not being a
minority may adversely affect the acquisition of secular human capital.

Chaney (2011) suggests a final mechanism associating a minority religion with a

greater emphasis on education, based on the reaction of religious elites, at least in the case of
Islam. He argues that fostering logical methods poses a dilemma to Islamic elites: on the one
hand, the art of discussion, ilm al-kalam, can contribute to attract converts. On the other hand,
it may also lead some Muslims to question the authority of religious elites. Choosing between
the two therefore depends on the context. Where Islam is a minority, or still new, it is more
important to attract converts. Conversely, where Islam is an established majority, it is more
important to secure the position of the religious elite. He argues that this mechanism sheds
light on the behavior of the Ottoman Empire in Anatolia. When the Ottoman Empire
conquered the initially Christian region in the XIVth and XVth centuries, they first encouraged
ilm al-kalam, leading to a period of scientific achievements. Once the region had become
predominantly Muslim, religious elites supported a move away from rational sciences,
eventually leading to a scientific decline. The upshot of this argument is that the stance of the

11
Muslim elites on education in the Ottoman Empire depended on whether Islam was a minority
or not.

The notion that the impact on education of a denomination depends on its minority
status, though appealing, has received little direct empirical attention, but indirect evidence
suggests that minority religions tend to perform better than other denominations in terms of
academic achievements. For instance, Sander (2010) reports that Jewish, Muslim, and
Buddhist residents in the US have higher educational achievements than Protestant and
Catholics, while their religions only represent 1.4, 0.5, and 0.5 percent of the population.

Chaudhary and Rubin (2011) report similar evidence for Muslims in India. They note
that Muslims living in districts with more Muslims have lower literacy rates. They for
instance note that in Bengal 21.1 percent of Hindu males are able to read and write but only
10.9 percent of Muslim males. In Madras, where the share of Muslims is much lower, literacy
rates of Hindus and Muslims are comparable. Borooah and Iyer (2005) made similar findings.

3. Global versus Country-Specific Effects: an Estimation

In this section, we investigate the impact of specific religious denominations on

education. After presenting the data, we apply the same method as standard cross-country
studies, and simply regress individuals’ levels of education on a dummy variable capturing
their confession, thereby forcing the impact of each denomination to be the same across
countries. These regressions only serve as a benchmark against which to assess our key
contribution. The key contribution of this section appears in the following sub-section, where
we allow the impact of each denomination to be country-specific, and report major
differences across countries.

3.1. Data

Our key data was retrieved from the World Values Survey. The survey has been carried
in a large number of countries since 1981. It results in a two-level dataset, where the country
of individual respondents can be identified.

The World Values Survey covers a lot of issues, and more specifically contains
questions about education, religion, and religiosity. Respondents are asked to indicate their
level of education. It results in a variable that can assume three values: lower education,
middle education, and higher education. An individual is considered to have a lower

12
education if he/she inadequately completed elementary education, completed elementary
education, or has not completed secondary school. The individual falls in the middle category
if he/she has completed secondary school or has some university-preparatory type or
secondary intermediate general qualification. The individual will be considered as having
received higher education if he/she has a university degree or at least a higher education with
a lower-level tertiary certificate.

The World Values Survey also contains information about respondents’ religious
denominations. As there are many denominations across the world, which would result in
many denominations appearing only in one country, we pooled denominations together
following Guiso et al.’s (2003) classification: Catholic, Christian-Protestant, Christian-
Orthodox, Buddhist, Muslim, Jewish, and Hindu. Other less frequent religions were pooled
under the label “other religions”, and we kept a separate category for respondents reporting no
religious affiliation.

*** Insert Table 1 around here ***

Table 1 presents statistics of our final sample of 147,049 respondents, distributed across
77 countries and 3 waves of the World Values Survey. 5 The top of the table reports the
worldwide distribution of religious denominations. It appears that more than 40% of the
individuals in our sample are affiliated with one of the Christian denominations, i.e.
Catholicism, Protestantism and Christian Orthodoxy. More than 50% go to religious services
once a year or less, while 30% go to services at least once a week. The rest of the table
provides for the distribution across education levels and for demographics.

3.2. Global Estimates

As a first step, we relate at the global level individual educational levels to religious
denominations using a cross-country regression similar to those of Guiso et al. (2003),
Schaltegger and Torgler (2010), or Arruñada (2010). We therefore estimate the following
ordered logit model for each denomination:

5
See Table A1 in the Appendix for details on the sample composition broken down by country.

13
Prob(Educationi = y) = f(Denominationki, Ci, Countryi, Yeari) (1)

Where:
- Educationi is the ordered variable measuring respondents i’s level of education;
- y ϵ (lower education, middle education, and higher education);
- Denominationki is a dummy variable capturing whether respondent i’s religious
denomination is denomination k, where k  (Catholic, Christian-Protestant, Christian-
Orthodox, Buddhist, Muslim, Jewish, and Hindu, Other religions, and no religious
affiliation);
- Ci is a vector of control variables controlling for respondent i’s demographics and
religiosity: Age, gender, marital status, income (by decile), and social class. We
include a dummy variable set to one if respondent i declares to be a religious person,
and three dummy variables controlling for respondent i’s church attendance: one
capturing whether the respondent goes to church once a week, another if he/she goes
to church once a month, and a third if he/she goes to church one a year;
- Countryi is a fixed country effect;
- Yeari is a dummy variable controlling for the survey wave in which the respondent filled in
the questionnaire.

We control for religiosity to distinguish its impact from that of denominations. By

doing so, we distinguish the intensity of religious practices, which has been found to correlate
with education for instance by Mocan and Pogorelova (2014), from the affiliation to a given
denomination.

Country effects are included to control for systematic differences across countries. As
a result, the observed effects of religious denominations are at work within countries, and can
for instance not be attributed to differences in education policies.

As the dependent variable is an ordered discrete variable, we estimate the relationship

between education and religious denomination using an ordered logit model. Each religious
denomination is considered in a specific regression. This means that the impact that each
denomination is assessed by comparing it to a reference group consisting of all other
respondents. However, it is clear from expression (1) that the coefficient of each
denomination is by construction the same across countries. The result of estimating
expression (1) should therefore be interpreted as measuring the average impact on education
at the global level of belonging to a given denomination.

14
 The result of those regressions are reported in Table 2, where each column reports the
results for one denomination. The coefficient of interest is the coefficient of the denomination
dummy, which captures a different denomination in each column. It for instance captures
whether respondent i is a Catholic in column 1.1 and whether he/she is a Protestant in column
1.2. The sign of the coefficient of the dummy variable captures whether at the global level
belonging to a denomination increases or decreases the probability to have acquired a higher
level of education.

*** Insert Table 2 around here ***

Most religious denomination dummies exhibit an insignificant coefficient. That is the

case of the Catholic, Protestant, Hindu, Buddhist denominations, and of denominations
pooled under the label other denominations. Those denominations therefore appear unrelated
to respondents’ levels of education worldwide.

However, we observe that the Jewish and Christian Orthodox dummies exhibit positive
coefficients that are significant at the five and one-percent levels respectively. Jewish and
Christian Orthodox respondents accordingly have a higher level of education on average at
the global level. Conversely, the Muslim dummy bears a negative coefficient that is negative
at the one-percent level, suggesting that being Muslim is on average associated with a lower
level of education. Table 2 also shows that being a non-religious person bears a positive sign
significant at the one-percent level, meaning that non-religious persons have on average a
higher level of education.

3.3. The Heterogeneity of the Impact of Religious Denominations on Education

Table 2 seems to deliver a clear message: Most religious denominations seem unrelated
to the level of education with one negative exception, Islam, and two positive exceptions,
Judaism and Christian Orthodoxy. These findings at the global level however rest on the
assumption that the marginal impact of each denomination is the same across countries. This
assumption makes sense if one believes that religious denominations convey universal
messages and prescriptions. Now, our specifications so far force the marginal effect of
religious denominations to be the same across countries. As a consequence, the estimates of
Table 2 may hide important cross-country heterogeneities. In fact, they simply prevent us
from testing the notion that religious denominations have universal effects.

15
 To test that notion, one has to allow the marginal effect of each denomination to differ
across countries. With this end in view, we therefore performed another set of regressions.
We still regressed the level of education achieved by individual respondents on their religious
denomination and a series of control variables, but instead of pooling all countries together,
we ran the following regression for each denomination k and each country j:

Prob(Educationi = y)jk = f(Denominationki, Ci, Yeari) (2)

Where variables are defined as in Expression (1). The key difference between the two
specifications is that Expression (2) includes no country fixed effect, because it is estimated
separately for each country.

As we considered one denomination at a time and up to 72 countries, we ran 502

regressions and obtained as many country-denomination-specific marginal effects, Effectjk.
Reporting all those regressions would be tedious and impossible to interpret. What really
matters instead, is the distribution of coefficients across countries for each denomination. We
therefore report coefficients according to their sign and significance.6 Those coefficients must
be interpreted as capturing the country average effect on education of a given denomination,
as opposed to the global average effect estimated in previous sub-section and in the rest of the
literature.

*** Insert Figure 1 around here ***

Figure 1 summarizes the distribution of the coefficients of religious denominations

resulting from the estimation of a parsimonious model, where the level of education is
explained by the respondent’s religious denomination, and individual demographic
characteristics, namely age, gender, marital status, income, social class, and religiosity when
significance is set at the ten-percent threshold. 7 A simple glance at the figure reveals that each
religious denomination has a significantly negative effect in some countries, and an
insignificant or significantly positive effect in others.

6
As the estimated model is a logit model, we cannot interpret their magnitude.
7
Table A2 in the appendix exhaustively reports the specific sign and significance of the coefficient of each
denomination in each country.

16
 To be more specific, Figure 1 shows that Catholicism has a significantly negative
impact on education in 11.1 percent of the countries in our sample where there are Catholics,
and a significantly positive impact in 29.2 percent of those countries. Being Protestant has a
significantly negative impact in 20.3 percent of countries, and a significantly positive impact
in 29.0 percent of countries. This finding is at odds with the view that Protestant ethics should
give an incentive to acquire more education everywhere. Being a Christian Orthodox results
in a lower level of education in 20.8 percent of countries and a higher level of education in
22.6 percent of countries. Buddhism can also have a negative effect, in 25.6 percent of
countries, and a positive effect, in 30.2 percent of countries. Being Muslim reduces the level
of education in 39 percent of countries, but increases it in 10.2 percent of countries. Being
Jewish has a negative effect on the level of education in 13 percent of countries, while it has a
positive impact in 26.1 percent of countries. Again, this is at odds with the view that Judaism
provides a universal incentive to acquire education. Being Hindu is associated with a lower
level of education in 22.2 percent of countries, but has a positive effect in 29.6 percent of
countries. The Other denominations category also splits evenly between a negative and a
positive impact, which is somewhat unsurprising, as this category is by definition
heterogeneous. Finally, even the impact of reporting no religious denomination at all varies
across countries. It is negative in 5.6 percent of countries, insignificant in 62.5 percent, and
positive in 31.9 percent of countries.

*** Insert Table 3 around here ***

Table 3 complements Figure 1 by considering the three standard levels of statistical

significance. The top panel of the table considers statistical significance at the one-percent
level, the middle panel statistical significance at the five-percent level, and the bottom panel
statistical significance at the ten-percent level. The first three columns report the share of
coefficients corresponding to each sign, and the fourth column the number of regressions run
for each denomination. This number varies across denominations, because a denomination
has to be present in a country for the regression to be performed in that country. For instance,
the impact of being Hindu could not be assessed in Algeria, simply because there was no
respondent reporting being Hindu in that country.

A cursory look at Table 3 reveals that no cell in the table features a zero. This
confirms the finding of Figure 1 that each religious denomination can have a significantly

17
negative, significantly positive, or insignificant effect depending on the country under
consideration. Moreover, the finding holds regardless of the level of significance. It even
holds in the top panel of Table 3, where the level of significance is set at the one-percent
level, and the number of insignificant coefficients is therefore mechanically larger than in the
other two panels.

3.4. Robustness Checks

In this sub-section, we check the robustness of our findings along two dimensions. The
first dimension is the robustness of the finding that the effect of religious denominations is
heterogeneous across countries. The second dimension is that the sign and significance of the
effect of a denomination in a country must be robust too. Table 4 therefore reports not only
the distribution of coefficients for various alternative specifications but also the percentage of
coefficients that are categorized in the same way as with the baseline specification. To
facilitate comparisons, the first column of the table reports the distribution of coefficients
obtained with the baseline specification. All the results reported in the table use the ten-
percent level of significance.

One may argue that, by controlling for religiosity, the baseline specification may strip
denominations of their effect, because it operates through religiosity. To make sure that it is
not the case, Table 4 reports the results of estimating the baseline specification, but without
controlling for religiosity. The outcome of this robustness check is reported in the first
column of the table. Again, each denomination can have either a significantly negative
impact, an insignificant impact, or a significantly positive impact in at least some countries of
the sample. Moreover, for all religious denominations, the share of coefficients that are
categorized in the same way exceeds 90 percent of countries.

One may be concerned that income is endogenous to education, and that it correlates
with both the level of education and the religious denomination of an individual. To make
sure that our results are not driven by the inclusion of income in the set of independent
variables, we estimate another specification that does not control for income. The second
column of Table 4 reports the distribution of coefficients when income is not controlled for in
Equation 2. It confirms our key result that no denomination has the same marginal effect
across countries. Furthermore, the sign and significance of coefficients is identical to those
obtained with the benchmark specification in 91 percent of countries or more for all
denominations.

18
 Because the role of religion may have changed over time, we distinguish individuals
by age. Hence, we run all the country-denomination regressions separately for respondents
below and above the age of 40. The outcomes of the two series of regressions are reported in
Columns 4.4 and 4.5 of Table 4. The distribution of coefficients in the two columns is similar.
In addition, in both columns the distribution of coefficients remains similar to the benchmark
distribution.

Because some religions ascribe different roles to men and women, our results may in
fact capture a gender gap. To check whether it is the case, we consider two subsamples made
up of only female or male respondents in Columns 4.6 and 4.7 of Table 4. Again, the
distribution of coefficients is similar across the two columns, and more importantly identical
to the benchmark categorization in 70.5 to 92.3 percent of countries.

Because in some countries, followers of a particular denomination may have migrated

in the country, our results may capture the effect of migration rather than denominations. The
third wave of the World Values Survey allows distinguishing respondents who are not native
of their country of residence. We therefore dropped non-natives from the sample to run the
benchmark specification only on natives. Although the size of the sample substantially
shrinks and the period of observation is different, Column 4.8 of Table 4 reports that a
majority of coefficients are identical to those obtained using the benchmark specification.

Finally, higher education may result in, or be the outcome of, religious conversions.
To make sure that our results are not driven by the impact of education on the choice of a
religious denomination, we focus on the subset of respondents who were raised religiously.
The information is only available in the third wave of the World Values Survey, which is why
we run our regressions on that specific wave. The results are reported in the last column of
Table 4. They are very close to those of the benchmark. Coefficients fall in the same category
as for the benchmark regression in at least 94 percent of countries for each denomination.

The main message from the above results is that no religious denomination has a
homogeneous effect across the world. In other words, the effect of religious denominations is
country-specific. Admittedly, some denominations, e.g. Islam, exhibit a more frequent
negative effect, and others, like Judaism, exhibit a more frequent positive effect. However,
Islam is associated with more education in 13.6 percent of countries in our sample, and
Judaism is associated with less education in 10.9 percent of countries.

19
 If the impact of religious denominations is not universal but country-specific, the key
question then becomes to determine what may drive a religious denomination to have a
negative effect in one country, a positive effect in another, and be plainly insignificant
elsewhere. This is the question that we address in the next section.

4. Explaining Heterogeneity: the Role of being a Minority

4.1. Baseline Results

To explain why a religious denomination may have a positive effect on education in

some countries and a negative or insignificant effect in others, we use the outcome of
previous section as our dependent variable. Specifically, we use as observation the outcome
of each regression reported in Table 3 and create a categorical variable capturing the sign of
the marginal effect of denomination j in country i. Each observation is therefore a
denomination-country pair. The categorical variable can take three modalities: negative and
significant at the ten-percent level, insignificant at the ten-percent level, or positive and
significant at the ten-percent level. We then estimate the following multinomial model:

Prob(Effectji = x) = f(Minorityji, Cji) (3)

where :
- Effectji is the sign of the coefficient of denomination j in country i in Table 3;
- x  (significantly negative, insignificant, significantly positive);
- Minorityji is a dummy variable signaling whether denomination j is a minority
religion in country i;
- Cji is a vector of control variables.

Our key explanatory variable of interest is denomination j’s minority status in country
i, because we expect members of minority religions to face incentives that differ from those of
members of majority religions. Table 5 below reports the number of countries in our sample
where each denomination is present, and the number and shares of countries where each
denomination is a minority. Here, we define a denomination as a minority if its share of
followers in a country is smaller or equal to five-percent according to the World Values
Survey. It shows that, while the frequency varies across denominations, each denomination is
a minority in some countries. Judaism stands out, as it is a minority in every country in our
sample.

20
 *** Insert Table 5 around here ***

We control for a series of country characteristics that can be subsumed into two
groups. The first group of variables describes a country’s type of government and government
policy. Kuran (1996, 1997, 2004) and Platteau (2008) emphasize the relationship between
Islam and the State. Specifically, we define a dummy variable indicating whether
denomination j is a State religion in country i. We used the list of State religion provided by
Barro and McCleary (2005), and set the dummy variable to one when the State religion of
Country i, if any, corresponded to denomination j. We also control for a country’s degree of
democracy, as measured by the PolityIV index (Marshall et al., 2011). We expect religious
denominations to matter less in more democratic countries, because democracy allows for a
better representation of all the components of society. The first group of variables also
contains a measure of the official hostility of the government towards religion, retrieved from
the Religion and State Project carried out by the Association of Religion Data. That variable
increases with hostility. The effect of that variable is a priori ambiguous. On the one hand,
one may expect that more secular governments foster education for all citizens, regardless of
their denomination. In that case, hostility of the government towards religion should reduce
the influence of denominations on education. On the other hand, if the government’s hostility
leads to bar education to religious pupils, then the variable should correlate positively with the
probability that a denomination has a negative effect on education.

The second set of country characteristics captures a country’s level of development.

We thus control for per capita GDP at purchasing power parity retrieved from the Penn World
Tables dataset. One may expect that more developed countries have a better education
infrastructure, and that religious denominations therefore matter less in those countries. Also,
religiosity has been found to decrease with income (see e.g. McLeary and Barro, 2006)
although the effect may not be causal, as Becker and Woessmann (2013) argue. Finally,
Bettendorf and Dijkgraaf’s (2010) finding that the impact of religiosity on individual income
is conditional on the income level of individuals’ country of residence may also apply to the
impact of religious denominations on education. We also control for the average level of
education in the country. The idea here is that religious denominations could simply not
matter if the entire population was educated. This intuition is in line with the finding of
Borooah and Iyer (2005) that differences in the school enrolment of boys among Muslim and

21
Hindus are essentially observable in the poorer regions of India, where school enrolments are
low. We therefore expect the role of religious denominations to be more limited in countries
where the average level of education is higher. Specifically, the likelihood of religious
denominations having a significant positive or negative effect should be lower in countries
with a higher level of education. We therefore control for the average number of years of
secondary education in the population, as measured by Barro and Lee (2013).

Finally, we control for the number of observations used in regressions in each country.
In our sample that number ranges from 282 in the Dominican Republic to 6745 in South
Africa. The smaller the number of observations, the larger standard errors, and the smaller t-
statistics. As a result, the likelihood of observing an insignificant coefficient is larger the
smaller the number of observations. We therefore control for that variable in all regressions to
avoid our results being biased by a statistical artifact.

We estimate Equation 3 as a multinomial logit model, because the dependent variable

is an unordered trichotomous variable. In all regressions, the reference category is the
category corresponding to denomination j having no significant effect on education in country
i.

As there is no standard threshold to define a minority religion, we first estimate

Equation 3 using alternative definitions of the minority dummy variable. Specifically, we
create a dummy variable taking the value one in denomination-country pair ji, if the share of
respondents in the World Values Survey reporting denomination j as his/her religion in
country i is lower than a given threshold, and consider in turn thresholds ranging from one to
fifty percent. Table 6 reports the results of estimating Model 3 using a multinomial logit
regression with those alternative dummy variables. The bottom row of the table reports the
number of minority denominations in the sample obtained with each definition, which
unsurprisingly increases as the threshold to define a minority religion increases.

*** Insert Table 6 around here ***

The top panel of Table 6 reports the coefficients measuring the impact of the
explanatory variable on the probability that denomination j has a positive effect significant at
the ten-percent level in country i. Conversely, the bottom panel of Table 6 reports the
coefficients measuring the impact of the explanatory variable on the probability that

22
denomination j has a negative effect significant at the ten-percent level in country i. The
unreported modality pertains to the coefficient not being significant at the ten-percent level.

The first column of Table 6 reports the result of estimating Equation 3 when a
denomination is assumed to be a minority in a country if less than one percent of respondents
claim to belong to that denomination. We observe that that variable exhibits a positive
coefficient in the top panel and negative coefficient in the bottom panel of the column and
that both are significant at the ten-percent level. This implies that if being a minority religion
decreases the probability for religion j to be negatively associated with a lower level of
education in country i, and that it increases the probability that the religion be positively
associated with a higher level of education in the country.

Columns 6.2 to 6.7 of Table 6 perform the same regression with increasing levels of
the threshold used to define a minority religion: 5%, 10%, 20%, 30%, 40%, and 40%
respectively. They exhibit the same pattern as the first column. Specifically, the coefficient of
the minority dummy is positive and statistically significant at the ten-percent level or beyond
in the top panel of each column, and negative and statistically significant at the ten-percent
level or beyond in the bottom panel of each column. There are only to exceptions. In Column
4, where the minority threshold is set to 20 percent, the coefficient of the minority dummy is
insignificant at standard levels of confidence in the top panel of the column. It is, however,
significantly negative at the ten-percent level in the bottom panel. Conversely, when the
threshold is set to 30 percent, in Column 5, the coefficient fails to be statistically significant in
the bottom panel of the column, but is positive and significant at the five-percent level in the
top panel.

In the last column of Table 6, we experiment with a continuous way to capture a

denomination’s minority status in a country. We compute the share in the country of all the
other denominations, which is also the complement to one of the share of the denomination of
interest. That share increases if the denomination of interest becomes a smaller minority.
Again, we observe that that measure of minority status positively correlates with the
probability of denomination j to have a positive effect on education in country j, and
correlates negatively with the probability of the denomination having a negative effect on
education in the country.

Table 6 therefore sketches a consistent pattern of the effect of minority religions on the
educational achievement of their members a country. Minority religions are less likely to

23
reduce and more likely to increase the educational achievement of their members with respect
to their fellow citizens. The finding is robust to various definitions of minority denominations.

Table 6, however, allows comparing alternative definitions of the minority religion

dummy variable. Specifically, we observe that the results that are the most statistically
significant are the results obtained when the threshold for a denomination to be considered as
a minority is set to five percent. In that particular case, the coefficient of the dummy variable
is significant at the five-percent level in the top panel and significant at the one-percent-level
in the bottom panel of Table 6. This suggests that the five-percent threshold best identifies
minority religions. We therefore use it as our benchmark in the rest of this paper.

4.2. Robustness Checks

The dependent variable in Equation 3 is a categorical variable defined on the outcomes

of a series of regressions run in different countries with the same specification of Equation 2.
However, the result may be specific to that specification. In Section 2, we experimented with
a series of alternative specifications of Equation 2, and showed that the finding that no
denomination had the same effect across countries was robust to using alternative
specifications. However, one may still be concerned that the results of Table 6 are driven by
the particular specification of Equation 2 used to define the dependent variable.

To make sure that this is not the case, we therefore estimated Equation 3 using
definitions of the dependent variable obtained in turn from all the specifications of Equation 2
estimated in Section 2. The results of those estimations are reported in Table 7.

Our benchmark specification of Equation 2 controls for an individual’s level of

religiosity, which may interact with education. We therefore estimated Equation 2 without
controlling for religiosity, use that specification to define the dependent variable in Equation
3. Column 7.1 of Table 7 reports the outcome of that estimation. The minority dummy still
exhibits a positive sign, significant at the ten-percent level, in the top panel of the column, and
a negative sign, significant at the one-percent level in the bottom panel, in line with our
benchmark estimates. Column 7.2 reports the outcome of the same exercise, but drops income
in the estimation of Equation 2 instead of religiosity. Here, the coefficient is statistically
insignificant in the top panel of the column, but still negative and significant at the one-
percent level in the bottom panel.

24
 *** Insert Table 7 around here ***

Columns 7.3 and 7.4 distinguish respondents by age, as one may argue that the role of
religion may have changed over time. We therefore estimated Equation 3 separately for
respondent below and above 40 years. Yet, the results of the two columns are similar. In both
columns, the minority variable is statistically significant at the one-percent level in the top
panel of the table, and insignificant in the bottom panel.

Finally, Columns 7.5 and 7.6 provide separate results for male and female
respondents. Again, and possibly surprisingly, we find similar results for the two genders,
suggesting that the effect of being a minority religion does not operate through a differential
effect for men and women. More precisely, we find that the minority variable is statistically
significant at the one-percent level in the top panel of the table, and insignificant in the
bottom panel for both genders.

One may be concerned that the impact of being a minority denomination may be driven
by the behavior of immigrants, who often belong to a denomination that differs from the
denomination of the host country, which may confound or results. We therefore estimated
Equation 2 on the subset of survey respondents who were born in their country of residence.8
Column 7.7 reports those results. It confirms the positive and statistically significant effect of
being a minority denomination in the top panel of Table 7. The impact, however, turns out
statistically insignificant in the bottom panel.
So far, we have relied on bivariate regressions. To make sure that our results are not
driven by an omitted variable bias in Equation 3, we control for a series of factors. As there is
no specification to guide us in the choice of the list of control variables to use when
estimating Equation 3, we started by including each variable in turn, before including them all
in the same regression. The outcomes of those regressions are reported in Table 8.

*** Insert Table 8 around here ***

In the first column, we control for the dummy variable indicating whether
denomination j is a State religion in country i. In the second column of Table 8, we control

8
Note that the information was only available in the third wave of the World Values Survey, while the main
specification aggregates information from several waves.

25
Country i’s level of democracy as measured by the PolityIV index. In the third column, we
control for the country’s PPP GDP per capita. In Column 8.4, we control for the country’s
years of schooling. In Column 8.5, we control for the number of observations in country i. In
Column 8.6, we control for no specific variable, but include denomination fixed effects to
control for any effect of any denomination that would be constant across countries. Column
8.7 controls for all variables at the same time, and Column 8.8 adds denomination fixed
effects to the list of control variables. Column 8.9 replaces the control variables that are
country-specific by country fixed effects, while controlling for denomination fixed effects at
the same time.

The impact of a religious denomination may depend on the denomination with which
it competes. For instance, being Christian orthodox in a country where the main other
denomination is Catholicism may result in very different incentives than being Christian
orthodox where the main other denomination is Islam. This source of heterogeneity is not
controlled by country- or denomination-fixed effects, because it is specific to denomination-
country pairs. We therefore controlled for denomination j’s main alternative in country i.
Specifically, for each denomination-country pair (j, i), we defined eight dummy variables
coding the largest denomination in country i beside denomination j. For instance, the first
dummy variable is set to one for Country i and Denomination j if Catholicism is the main
denomination beside Denomination j, and set to zero elsewhere. We defined such a dummy
variable for each denomination in our sample. Column 8.11 reports the result of a regression
controlling for those dummy variables, together with country- and denomination- fixed
effects. As an ultimate robustness check, Column 8.12 complements Column 8.11 by
controlling for the other country-denomination pair dummy, namely State religion.

The results reported in Table 8 are homogeneous. In the nine columns of the table, the
coefficient of the minority dummy variable is positive and significant at the five-percent level
or beyond, signaling that all specifications imply that members of minority religions tend to
accumulate more education than their fellow citizens of another denomination.

Similarly, the bottom panel of Table 8 implies that when a religion is a minority in a
country, the members of the religion are less likely to have acquired less education than their
fellow citizens. The coefficient of the minority variable in the bottom panel of Table 8 is in
general negative and statistically significant well beyond the ten-percent level. In Columns
8.7 and 8.8, where the all set of control variables are controlled for, the coefficient turns out
insignificant. However, one should remark that the number of observations in the regression

26
is also limited by data availability. When those variables are replaced by fixed-country
effects, which allows controlling for all country-specific omitted variables without reducing
the size of the sample, the coefficient of the minority dummy is negative and significant at the
one-percent level.

A by-product of Table 8 is to allow assessing the impact of control variables. Unlike

the minority dummy variable, other control variables are in general statistically insignificant.
One exception is the variable that measures the average number of years of schooling in the
population. That variable is statistically insignificant in the top panel of Table 8 but bears a
negative and statistically significant at the ten- or five-percent level, depending on the
specification. This suggests that any denomination is less likely to have a negative impact on
its members’ educational achievements in a country if the average level of education in that
country is larger, which is in line with our expectations.

The dummies capturing the main alternative denomination are nearly always statistically
insignificant. Moreover none is significant in more than one specification, which is why we
do not report their coefficients. The regressions reported in Table 8, however, show that
controlling for the main alternative denomination does not affect our main conclusions. We
observe that that the minority religion dummy variable exhibits a positive coefficient in the
top panel and negative coefficient in the bottom panel of the column and that at least one of
them is statistically significant.

5. Concluding Remarks

We have used the World Values Surveys to identify the impact of major religious
denominations on individuals’ levels of education in a large sample of countries. Two main
results emerge from our results.

First, no denomination has a uniform effect on the level of education of its members
across countries. In other words, we find that each denomination correlates positively with
education in some countries, negatively in other countries, and does not correlate with
education in others. In a nutshell, no denomination has a universal effect on education.

The second main result is that the propensity of a denomination to have a positive,
negative or insignificant effect on education is not randomly distributed across countries. We
provide robust evidence that minority religions are more likely to have a positive effect, and
less likely to have a negative effect on the level of education of its members. This finding is in

27
line with the theories that emphasize the role of religions as a club good. The finding may also
be in line with theories that suggest that members of minority religions must invest in
education to compensate for their minority status.

One may argue that our results capture the fact that different blends of the same religion
with different implications for education are subsumed under the same name. For instance,
Calvinists and Lutherans are pooled together under the label Protestantism, or Islam pools
together Shia and Sunni Muslims. While we cannot fully dismiss that point, it only applies to
a subset of religious denominations. It for instance does not apply to Catholicism, which is
more centralized and therefore more uniform. Yet, our findings for Catholicism are in line
with those obtained for other denominations. This is, however, food for further research.

28
References

Akerlof, George A. and Kranton, Rachel E. 2000. “Economics and identity,” Quarterly
Journal of Economics, 115(3): 715–753.

Akerlof, George A., and Rachel E. Kranton. 2002. “Identity and schooling: Some lessons for
the economics of education,” Journal of Economic Literature 40(4): 1167–1201.

Alesina, A. and Paola Giuliano. 2015. “Culture and institutions,” Journal of Economic
Literature, 53(4): 898-944.

Arruñada, B. 2010. “Protestants and Catholics: Similar Work Ethic, Different Social Ethic,”
Economic Journal 120: 890–918.

Asadullah, Mohammad Niaz, and Nazmul Chaudhury. 2010. “Religious Schools, Social
Values, and Economic Attitudes: Evidence from Bangladesh,” World Development
38(2): 205–217.

Asadullah, Mohammad Niaz, Nazmul Chaudhury, and Amit Dar. 2007. “Student achievement
conditioned upon school selection: Religious and secular secondary school quality in
Bangladesh,” Economics of Education Review 26(6): 648–659

Barro, R. and Jong-Wha Lee. 2013. “A new data set of educational attainment in the world,
1950–2010,” Journal of Development Economics 104: 184–198.

Barro, Robert J, and Rachel M McCleary. 2006. “Religion and economy,” The Journal of
Economic Perspectives 20(2): 49–72.

Barro, Robert J, and Rachel M McCleary. 2003. “Religion and economic growth across
countries,” American Sociological Review 68: 760–781.

Barro, Robert J, and Rachel M. McCleary. 2005. “Which countries have state religions?”
Quarterly Journal of Economics 120 (4): 1331–1370.

Becker, Sascha O., and Ludger Woessmann. 2009. “Was Weber Wrong? A Human Capital
Theory of Protestant Economic History,” Quarterly Journal of Economics 124(2): 531–
596.

Berman, Eli. 2000. “Sect, Subsidy, and Sacrifice: An Economist’s View of Ultra-Orthodox
Jews,” Quarterly Journal of Economics 115(3): 905–953.

Bettendorf, L.; Dijkgraaf, E. 2010. “Religion and Income: Heterogeneity between Countries,”
Journal of Economic Behavior and Organization 74 (1-2): 12–29.

29
Beyerlein, Kraig. 2004. “Specifying the Impact of Conservative Protestantism on Educational
Attainment,” Journal for the Scientific Study of Religion 43(4): 505–518.

Blunch, Niels-Hugo. 2007. “Human Capital and Religion in Ghana,” Cambridge Working
Papers in Economics 0770.

Bobrick, Benson. 2001. Wide as the waters: The story of the English Bible and the revolution
it inspired. Simon & Schuster.

Borooah, Vani K. and Sriya Iyer. 2005. “Vidya, Veda, and Varna: The influence of religion
and caste on education in rural India,” Journal of Development Studies 41(8): 1369–
1404.

Botticini, Maristella, and Zvi Eckstein. 2005. “Jewish Occupational Selection: Education,
Restrictions, or Minorities?” Journal of Economic History 65(4), 922–948.

Botticini, Maristella, and Zvi Eckstein. 2007. “From farmers to merchants, conversions and
diaspora: Human capital and Jewish history,” Journal of the European Economic
Association 5(5):885–926.

Brenner Reuven, and Nicholas M. Kiefer, 1981. “The Economics of Diaspora: Discrimination
and Occupational Structure,” Economic Development and Cultural Change 29(3): 517–
534.

Brown, Sarah, and Karl Taylor. 2007. “Religion and education: evidence from the National
Child Development Study,” Journal of Economic Behavior & Organization 63(3):439–
460.

Chaney, Eric, 2011. “Islam and Human Capital Formation: Evidence from Pre-Modern
Science,” in Rachel M. McCleary (ed.), Oxford Handbook of the Economics of Religion,
Oxford University Press, 81-91.

Chaudhary, Latika, and Jared Rubin. 2011. “Reading, Writing, and Religion: Institutions and
Human Capital Formation,” Journal of Comparative Economics 39(1): 17-33.

Chiswick, Barry R. 1988. “Differences in Education and Earnings Across Racial and Ethnic
Groups: Tastes, Discrimination, and Investments in Child Quality,” Quarterly Journal of
Economics 103(3): 571-597.

Cipolla, Carlo M. 1969. Literacy and Development in the West (Harmondsworth, UK:
Penguin.

30
Coate, Steven, and Glenn Loury. 1993. “Will Affirmative-Action Policies Eliminate Negative
Stereotypes?” American Economic Review 83(5): 1220–1240.

Coleman, James S. 1988. “Social capital in the creation of human capital,” American Journal
of Sociology 94, S95–S120.

Czopp, Alexander M., Aaron C. Kay, and Sapna Cheryan. 2015. “Positive stereotypes are
pervasive and powerful,” Perspectives on Psychological Science 10(4): 451–463.

Darnell, Alfred and Darren E. Sherkat. 1997. “The Impact of Protestant Fundamentalism on
Educational Attainment,” American Sociological Review 62(2): 306–315.

Glaeser, E.L. and Glendon, S. 1998. “Incentives, predestination and free will,” Economic
Inquiry 36(3): 429–43.

Glaeser, Edward L., and Bruce I. Sacerdote. 2008. “Education and Religion,” Journal of
Human Capital 2(2): 188–215.

Go, Sun, and Peter Lindert. 2010. “The Uneven Rise of American Public Schools to 1850,”
Journal of Economic History 70(1): 1–26.

Goldin, Claudia, and Lawrence Katz. 2000. “Education and Income in the Early 20th Century:
Evidence from the Prairies,” Journal of Economic History 60(3): 782–818.

Guiso, Luigi, Paola Sapienza, and Luigi Zingales. 2003. “People’s opium? Religion and
economic attitudes,” Journal of Monetary Economics 50(1):225–282.

Hanna, Rema N., and Leigh L. Linden. 2012. “Discrimination in grading,” American
Economic Journal: Economic Policy 4(4), 146–168.

Hanushek, Eric A, and Ludger Woessmann. 2008. “The role of cognitive skills in economic
development,” Journal of economic literature 46(3): 607–668.

Hillman, Arye L. 2007. “Economic and security consequences of supreme values,” Public
Choice 131(3): 259–280.

Hoff, Karla, and Priyanka Pandey. 2006. “Discrimination, social identity, and durable
inequalities,” American economic review 96(2), 206–211.

Hollander, Gideon, Kahana, Nava, and Tikva Lecker. 2003. “Religious and secular human
capital: An economic model,” Journal of Socio-Economics 32(5): 489–498.

Iannaccone, Laurence R., 1992. “Sacrifice and Stigma: Reducing Free-Riding in Cults,
Communes, and Other Collectives,” Journal of Political Economy 100(2): 271–291.
31
Kranton, Rachel E. 2016. “Identity Economics 2016: Where Do Social Distinctions and
Norms Come From?” American Economic Review 106(5): 405–409.

Kuran, Timur. 1996. “The discontents of Islamic economic morality,” American Economic
Review Papers and Proceedings 86(2): 438–442.

Kuran, Timur. 1997. “Islam and underdevelopment: an old puzzle revisited,” Journal of
Institutional and Theoretical Economics 153(1): 41–71.

Kuran, Timur. 2004. “Why the Middle East is economically underdeveloped: historical
mechanisms of institutional stagnation,” Journal of Economic Perspectives 18(3): 71–90.

Lehrer, Evelyn L. 1999. “Religion as a Determinant of Educational Attainment: An Economic

Perspective,” Social Science Research 28(4): 358–379.

Ling, Trevor. 1984. “Buddhism and education in Burma and Thailand,” Religion 14(1): 53-
65.

Luther, M., “An die Ratsherren aller Städte deutschen Landes, dass sie christliche Schulen
aufrichten und halten sollen (To the Councilmen of All Cities in Germany That They
Establish and Maintain Christian Schools),” in Dr. Martin Luthers Werke: Kritische
Gesamtausgabe, Vol. 15 (Weimar, Germany: Verlag Hermann Böhlhaus Nachfolger
[1524] 1899).

Luther, M., “Eine Predigt, daß man Kinder zur Schule halten solle (A Sermon on Keeping
Children in School),” in Dr. Martin Luthers Werke: Kritische Gesamtausgabe, Vol.
30, Part 2 (Weimar, Germany: Verlag Hermann Böhlhaus Nachfolger [1530] 1909).

Markussen, Ingrid. 1990. “The development of writing ability in the Nordic countries in the
eighteenth and nineteenth centuries,” Scandinavian Journal of History 15(1-2): 37–63.

Marshall, M.G., K. Jaggers, and T.R. Gurr. 2011. “Polity IV Project: Political regime
characteristics, and transitions, 1800-2010: Dataset users’ manual,” University of
Maryland, Center for International Development and Conflict Management.

McClelland, V. Alan. “Wholeness, faith and the distinctiveness of the Catholic school,” in
McLaughlin, B.Terence, Joseph O'Keefe, and Bernadette O'Keeffe. The contemporary
Catholic school: Context, identity, and diversity. Psychology Press, 1996, 155-161.

Mocan, Naci, and Luiza Pogorelova. 2014. “Compulsory schooling laws and formation of
beliefs: education, religion and superstition,” NBER Working Paper 20557.

32
Morey, Melanie M. and Piderit, John J. 2010. Catholic higher education: A culture in crisis.
Oxford University Press, 2010.

Newhouse, David, and Beegle, Kathleen. (2006). “The effect of school type on academic
achievement: Evidence from Indonesia,” Journal of Human Resources 41(3): 529–557.

Offenbacher, Martin.1900 “Konfessionelle und soziale Schichtung. Eine Studie über die
wirtschaftliche Lage der Katholiken und Protestanten in Baden,” Volkswirtschaftliche
Abhandlungen der Badischen Hochschulen, Bd. IV (5).

Oosterbeek, Hessel, and Bas van der Klaauw. 2013. “Ramadan, Fasting and Educational
Outcomes,” Economics of Education Review 34: 219-226.

Paul VI, 1965. Gravissimum Educationis [Declaration on Christian Education], accessed 13

July 2016.
http://www.vatican.va/archive/hist_councils/ii_vatican_council/documents/vat-
ii_decl_19651028_gravissimum-educationis_en.html

Pius XI, 1929. Divini Illius Magistri [Encyclical on Christian education], accessed 13 July
2016.
http://w2.vatican.va/content/pius-xi/en/encyclicals/documents/hf_p-
xi_enc_31121929_divini-illius-magistri.html

Platteau, Jean-Philippe. 2008. “Religion, politics, and development: Lessons from the lands of
Islam,” Journal of Economic Behavior & Organization 68(2): 329–351.

Rosenthal, Robert, and Lenore Jacobson. 1968. “Pygmalion in the classroom,” Urban Review
3(1), 16–20.

Sander, William. 2010. “Religious Background and Educational Attainment: The Effects of
Buddhism, Islam, and Judaism,” Economics of Education Review 29(3): 489–493.

Schaltegger, C, and B. Torgler. 2010. “Work ethic, Protestantism, and human capital,”
Economics Letters (107): 2, 99–101.

Shih, Margaret, Ambady, Nalimi, Richeson, Jennifer A., Fujita, Kentaro, and Gray, Heather
M. (2002). “Stereotype performance boosts: the impact of self-relevance and the
manner of stereotype activation,” Journal of personality and social psychology, 83(3),
638–647.

33
Weber, Max, “Die protestantische Ethik und der ‘Geist’ des Kapitalismus”, Archiv für
Sozialwissenschaft und Sozialpolitik 20 (1904/1905), 1–54 and 21: 1–110; reprinted in
Gesammelte Aufsätze zur Religionssoziologie, pp. 17–206, 1920; English translation as
The Protestant Ethic and the Spirit of Capitalism, translated by Talcott Parsons (London,
UK: Routledge Classics, [1930] 2001).

Woodberry, Robert D. 2011. “Religion and the Spread of Human Capital and Political
Institutions: Christian Missions as a Quasi-Natural Experiment”, in Rachel M. McCleary
(ed.), Oxford Handbook of the Economics of Religion, Oxford University Press, 111-131.

34
Tables
Table 1: Descriptive Statistics:
Education Level, Religious Affiliation, Religiosity and Demographics

Religious Denomination:
Catholic 0.277
Christian, Protestant 0.158
Christian, Orthodox 0.003
Buddhist 0.204
Muslim 0.132
Jewish 0.024
Hindu 0.028
Other affiliations 0.042
No religious affiliation 0.133

Religiosity:
Religious Person (yes) 0.699

Goes to Religious Services at least once a Week 0.329

Goes to Religious Services once a Month 0.118
Goes to Religious Services once a year / only on Special Holidays 0.247
Goes to Religious Services less than once a year 0.306

Education:
Lower (elementary education and below) 0.358
Middle (intermediate & full secondary) 0.421
Upper (lower & upper-level tertiary certificate) 0.221

Demographic Characteristics
Age (years) 41.217
(15.971)
Female (yes) 0.508
Married (yes) 0.591
Social Class 3.356
Income (decile) 4.546
(2.383)

Number of Observations 147,049

Notes: Weighted Sample, 77 countries, 3 waves (1994-1999, 1999-2004, 2005-2007).

35
 Table 2: Education Level and Religious Affiliation, by Religion (Ordered Logit)
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9
No
Catholic Protestant Jewish Muslim Orthodox Hindu Buddhist Others
Religion
Denominat -
0.020 -0.030 0.335** 0.166*** 0.026 -0.103 0.063 0.190***
ion 0.548***
(0.020) (0.022) (0.111) (0.031) (0.027) (0.063) (0.053) (0.035) (0.024)

Religiosity yes yes Yes yes yes yes Yes yes Yes
Controls yes yes Yes yes yes yes Yes yes Yes
Country
yes yes Yes yes yes yes Yes yes Yes
FE
Years FE yes yes Yes yes yes yes Yes yes Yes
Pseudo R2 0.171 0.171 0.171 0.172 0.171 0.171 0.171 0.171 0.171
Number of 14704 14704
147049 147049 147049 147049 147049 147049 147049
Obs. 9 9
* ** ***
Absolute standard errors are reported between brackets. p < 0.05, p < 0.01, p < 0.001.

36
 Table 3: The distribution of estimates across countries
Religious Affiliation Significant and Not Significant and Number
negative at significant at positive at of
(Row %) (Row %) (Row %) countries
3.1. 3.2 3.3 3.4
1% level
Catholic 2.8 81.9 15.3 72
Christian, Protestant 11.6 72.5 15.9 69
Christian, Orthodox 13.2 69.8 17.0 53
Buddhist 23.3 58.1 18.6 43
Muslim 28.8 64.4 6.8 59
Jewish 10.9 73.9 15.2 46
Hindu 18.5 55.6 25.9 27
Other affiliations 9.8 83.6 6.6 61
No religious affiliation 1.4 86.1 12.5 72
5% level
Catholic 8.3 68.1 23.6 72
Christian, Protestant 17.4 56.5 26.1 69
Christian, Orthodox 18.9 58.5 22.6 53
Buddhist 23.3 51.2 25.6 43
Muslim 39.0 52.5 8.5 59
Jewish 13.0 65.2 21.7 46
Hindu 18.5 51.9 29.6 27
Other affiliations 13.1 75.4 11.5 61
No religious affiliation 2.8 75.0 22.2 72
10% level
Catholic 11.1 59.7 29.2 72
Christian, Protestant 20.3 50.7 29.0 69
Christian, Orthodox 20.8 56.6 22.6 53
Buddhist 25.6 44.2 30.2 43
Muslim 39.0 50.8 10.2 59
Jewish 13.0 60.9 26.1 46
Hindu 22.2 48.1 29.6 27
Other affiliations 13.1 73.8 13.1 61
No religious affiliation 5.6 62.5 31.9 72

37
 Table 4: Distribution of Estimates across Countries by Specification
Benchmark No No Young Old Men Women Native Raised
Religiosity Income (<41) (>40) only religiously
(wave 3) (wave 3)
4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9
Catholic
% Significant and negative 11.1 16.7 8.3 9.7 15.3 12.5 8.3 17.5 13.6
% Significant and positive 29.2 27.8 29.2 23.6 22.2 27.8 22.2 12.5 9.1
% Not significant 59.7 55.6 62.5 66.7 62.5 59.7 69.4 70 77.3
Identical to benchmark at: 91.0% 92.3% 70.5% 75.6% 74.4% 75.6% 87.2% 93.6%
Protestant
% Significant and negative 20.3 20.3 20.3 15.9 17.4 17.4 17.4 17.5 15.9
% Significant and positive 27.5 27.5 27.5 21.7 30.4 18.8 34.8 15 13.6
% Not significant 52.2 52.2 52.2 62.3 52.2 63.8 47.8 67.5 70.5
Identical to benchmark at: 97.4% 94.9% 76.9% 74.4% 80.8% 75.6% 91.0% 92.3%
Orthodox
% Significant and negative 20.8 20.8 22.6 22.6 22.6 18.9 9.4 17.2 18.8
% Significant and positive 22.6 22.6 22.6 28.3 28.3 28.3 32.1 34.5 28.1
% Not significant 56.6 56.6 54.7 49.1 49.1 52.8 58.5 48.3 53.1
Identical to benchmark at: 100.0% 98.7% 82.1% 78.2% 82.1% 80.8% 92.3% 100.0%
Buddhist
% Significant and negative 25.6 25.6 27.9 23.3 18.6 25.6 25.6 8 11.1
% Significant and positive 30.2 27.9 27.9 32.6 51.2 48.8 39.5 68 66.7
% Not significant 44.2 46.5 44.2 44.2 30.2 25.6 34.9 24 22.2
Identical to benchmark at: 98.7% 94.9% 89.7% 80.8% 85.9% 82.1% 94.9% 98.7%
Muslim
% Significant and negative 38.3 38.3 40 38.3 41.7 35 48.3 27.6 35.5
% Significant and positive 11.7 15 11.7 16.7 23.3 15 20 34.5 16.1
% Not significant 50 46.7 48.3 45 35 50 31.7 37.9 48.4

38
Identical to benchmark at: 97.4% 91.0% 74.4% 73.1% 84.6% 70.5% 87.2% 94.9%
Jewish
% Significant and negative 13 10.9 15.2 17.4 21.7 10.9 26.1 10 8.8
% Significant and positive 26.1 26.1 28.3 45.7 34.8 45.7 39.1 56.7 61.8
% Not significant 60.9 63 56.5 37 43.5 43.5 34.8 33.3 29.4
Identical to benchmark at: 98.7% 94.9% 75.6% 80.8% 78.2% 79.5% 94.9% 98.7%
Hindu
% Significant and negative 22.2 22.2 14.8 22.2 11.1 14.8 22.2 13.3 12.5
% Significant and positive 29.6 33.3 29.6 40.7 59.3 48.1 48.1 60 68.8
% Not significant 48.1 44.4 55.6 37 29.6 37 29.6 26.7 18.8
Identical to benchmark at: 98.7% 96.2% 88.5% 89.7% 92.3% 88.5% 96.2% 97.4%
Other Affiliations
% Significant and negative 13.1 14.8 19.7 16.4 21.3 8.2 19.7 0 3
% Significant and positive 13.1 14.8 13.1 21.3 18 21.3 16.4 23.3 27.3
% Not significant 73.8 70.5 67.2 62.3 60.7 70.5 63.9 76.7 69.7
Identical to benchmark at: 97.4% 92.3% 73.1% 80.8% 79.5% 79.5% 93.6% 98.7%
No Religious Affiliation
% Significant and negative 5.4 8.1 4.1 9.5 5.4 5.4 10.8 0 2.3
% Significant and positive 33.8 32.4 36.5 10.8 35.1 27 25.7 20 22.7
% Not significant 60.8 59.5 59.5 79.7 59.5 67.6 63.5 80 75
Identical to benchmark at: 88.5% 93.6% 66.7% 73.1% 78.2% 71.8% 89.7% 94.9%

39
 Table 5: Minority religions in our sample
Number of countries Number of countries Percentage of countries
where the denomination where the denomination where the denomination
is present is a minority is a minority
Catholic 72 26 36.11
Protestant 69 32 46.38
Muslim 60 31 51.67
Orthodox 53 34 64.15
Jewish 46 46 100
Buddhist 43 36 83.72
Hindu 27 24 88.89
Notes: A denomination is considered as a minority in a country if its share of followers in the
country is smaller or equal to five-percent according to the World Values Survey

40
 Table 6: Impact of being a minority religion: alternative definitions of religious minorities
(6.1) (6.2) (6.3) (6.4) (6.5) (6.6) (6.7) (6.8)
Definition of the minority dummy <1% <5% <10% <20% <30% <40% <50% Share others
Modality of the dependent variable
Positive and significant at 10%
Minority religion 0.467* 0.574** 0.488* 0.348 0.737** 0.745* 0.728* 1.104**
(0.256) (0.282) (0.295) (0.309) (0.364) (0.395) (0.442) (0.560)
Negative and significant at 10%
Minority religion -0.529* -0.753*** -0.717*** -0.521* -0.369 -0.587* -0.619* -0.939**
(0.275) (0.270) (0.272) (0.284) (0.301) (0.307) (0.331) (0.422)
Observations 370 370 370 370 370 370 370 370
Number of minority denominations 179 229 250 272 292 301 315 -
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Table 7: Impact of being a minority religion: alternative specifications of first-stage regressions

(7.1) (7.2) (7.3) (7.4) (7.5) (7.6) (7.7)
Specification No religiosity No income <40 >40 Men Women Only natives
Modality of the dep. var.
Positive and significant at 10%
Minority 0.476* 0.418 1.035*** 1.723*** 1.354*** 1.361*** 2.665***
(0.280) (0.277) (0.278) (0.298) (0.287) (0.278) (0.504)
Negative and significant at 10%
Minority -0.875*** -0.796*** 0.0764 0.0623 -0.172 0.277 0.0400
(0.269) (0.270) (0.273) (0.270) (0.278) (0.272) (0.397)
Observations 370 370 370 370 370 370 208
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

41
 Table 8: Impact of being a minority religion with control variables
(8.1) (8.2) (8.3) (8.4) (8.5) (8.6) (8.7) (8.8) (8.9) (8.10) (8.11) (8.12)
Modality of the dep. var.
Positive and significant at 10%
Minority religion (<5%) 0.584** 0.635** 0.592** 0.993*** 0.598** 0.694** 1.202*** 1.443*** 1.040*** 1.072** 1.106** 1.320**
(0.296) (0.296) (0.285) (0.324) (0.283) (0.317) (0.364) (0.404) (0.400) (0.418) (0.491) (0.515)
State religion 0.0800 0.217 0.275 0.240 2.965*
(0.714) (0.868) (0.875) (1.047) (1.539)
Democracy -0.000818 0.0142 0.0109
(0.0262) (0.0376) (0.0380)
GDP -7.36e-06 -1.57e-05 -1.38e-05
(1.02e-05) (1.66e-05) (1.68e-05)
Years of schooling -0.0305 -0.00152 0.000887
(0.0547) (0.0830) (0.0846)
Number of obs. in country i -0.000107 -0.000109 -0.000109
(0.000111) (0.000121) (0.000123)

Negative and significant at 10%

Minority religion (<5%) -0.625** -0.566** -0.682** -0.531* -0.768*** -1.048*** -0.277 -0.480 -1.344*** -1.176*** -1.471*** -1.457***
(0.285) (0.285) (0.277) (0.308) (0.271) (0.322) (0.343) (0.397) (0.423) (0.443) (0.525) (0.525)
State religion 0.753 0.749 0.757 1.081 0.0658
(0.506) (0.588) (0.642) (0.842) (1.140)
Democracy -0.0351 0.0134 0.0259
(0.0254) (0.0379) (0.0405)
GDP -1.74e-05 6.28e-06 4.60e-06
(1.19e-05) (1.92e-05) (1.95e-05)
Years of schooling -0.143** -0.171* -0.169*
(0.0585) (0.0908) (0.0930)
Number of obs. in country i 6.25e-05 2.52e-05 2.56e-05
(0.000106) (0.000118) (0.000121)
Denomination fixed effects no no no no no yes no yes yes yes yes yes
Country fixed effects no no no no no no no no yes yes yes yes
Main alternative den. Dum. no no no no no no no no no no yes yes
Observations 370 335 359 301 370 370 286 286 370 370 370 370
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

42
Figures

Fig. 1: The distribution of estimates across countries controlling for religiosity (10% significance)

73.8
70
60.9 62.5
59.7
60 56.6
50.7 50.8
48.1
50 44.2
39
40
30.2 31.9
29.2 29 29.6
30 25.6 26.1
22.6 22.2
20.3 20.8
20
13 13.1 13.1
11.1 10.2
10 5.6

Negative Insignificant Positive

43
Appendix A: Additional Tables

Table A1: Sample Composition

Europe Africa Asia / Oceania Americas

Country Obs. Country Obs. Country Obs. Country Obs.
#
Albania * 1,811 Algeria* 378 Armenia# 1,622 Argentina# * 1,929
$ $ #$
Andorra 878 Burkina Faso 1,087 Australia 2,780 Brazil# $ 2,360
# $ # $
Belarus 1,409 Egypt* 5,628 Azerbaijan 1,545 Canada* 2,919
Bosnia and Ethiopia$ 1,337 Bangladesh# * 2,680 Chile# * $ 2,580
Herzegovina# * 1,847 Ghana $
1,379 China* $
1,091 Colombia #
2,866
Bulgaria# $ 1,518 Mali$ 894 Hong Kong$ 1,043 Dominican
Cyprus$ 1,012 Morocco* 568 India# * $ 4,711 Republic# 282
# # $ #
Czech Republic 783 Nigeria * 3,402 Indonesia* 2,274 El Salvador 926
# $ $
Estonia 802 Rwanda 1,331 Iran* 4,112 Mexico # * 2,629
South Africa#
Finland# $ 1,554 *$ 6,745 Iraq* $ 4,227 Peru # * $ 3,509
#$ $
Georgia 3,223 Tanzania* 975 Japan* 1,687 Puerto Rico# * 1,483
# $
Germany 3,000 Uganda* 520 Jordan* 1,095 Trinidad and
Italy$ 602 Zambia$ 1,023 Kyrgyzstan* 921 Tobago $ 921
New Zealand United
Latvia# 906 Zimbabwe* 748 #$
1,463 States # * $ 3,054
#
Lithuania 816 Pakistan * 1,168 Uruguay # $ 1,202
# #
Macedonia * 1,329 Philippines* 1,082 Venezuela * 1,866
Moldova# * $ 2,634 Saudi Arabia* 1,198
Norway # $ 1,766 South Korea$ 1,191
Poland# $ 1,777 Taiwan# $ 1,175
Romania# $ 2,581 Thailand $ 1,480
Russian
Federation# 1,441 Vietnam * $ 2,215
$
Serbia 974
Serbia and
Montenegro# * 2,770
Slovakia# 859
$
Slovenia 764
Spain# * $ 2,644
Sweden# * $ 2,306
Switzerland# $ 1,837
# $
Turkey * 5,360
Ukraine# $ 2,545
Total 51,748 Total 26,015 Total 40,760 Total 28,526
Note: Sample Wave: # 1994-1999, * 1999-2004, $ 2005-2007.

44
 Table A2: Country−denomination−specific Coefficients: Sign and Statistical Significance at 10%
Country / No
Catholic Protestant Orthodox Buddhist Hindu Jewish Muslim Others
Denomination Relig.
Albania Not Sign. Not Sign. + Not Pres. Not Pres. Not Sign. − + −
Algeria Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. +
Andorra − Not Sign. + Not Pres. Not Sign. Not Pres. Not Sign. + Not Sign.
Azerbaijan Not Sign. Not Sign. Not Sign. Not Pres. Not Pres. Not Sign. − Not Pres. +
Argentina Not Sign. − + + + Not Sign. Not Pres. Not Sign. Not Sign.
Australia − Not Sign. Not Sign. Not Sign. + + Not Sign. Not Sign. +
Bangladesh Not Sign. Not Sign. − Not Sign. Not Sign. + Not Sign. Not Pres. Not Sign.
Armenia Not Sign. + Not Sign. Not Pres. Not Pres. Not Pres. Not Pres. Not Sign. Not Sign.
Bosnia and
Not Sign. Not Sign. Not Sign. Not Pres. Not Pres. Not Sign. Not Sign. Not Pres. Not Sign.
Herzegovina
Brazil Not Sign. − − Not Sign. Not Pres. + Not Pres. + Not Sign.
Bulgaria Not Sign. Not Sign. + − Not Sign. + − + Not Sign.
Belarus − Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. Not Sign. Not Pres. Not Sign.
Canada − + Not Sign. + Not Sign. Not Sign. Not Sign. Not Sign. Not Sign.
Chile Not Sign. − Not Pres. Not Pres. − Not Sign. Not Pres. Not Sign. +
China + Not Sign. − Not Sign. Not Pres. Not Pres. Not Sign. Not Pres. Not Sign.
Taiwan + Not Sign. Not Pres. Not Sign. Not Pres. Not Pres. + Not Sign. +
Colombia Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. + Not Pres. Not Sign. Not Sign.
Cyprus Not Sign. + + Not Pres. Not Pres. + − Not Sign. +
Czech republic Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Sign.
Dominican Republic Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Sign. Not Sign.
El Salvador + − Not Pres. Not Sign. Not Pres. Not Pres. Not Pres. Not Pres. Not Sign.
Ethiopia + + − − Not Pres. Not Sign. − Not Sign. Not Sign.
Estonia Not Sign. Not Sign. Not Sign. + Not Pres. Not Pres. Not Sign. Not Pres. −
Finland Not Sign. − Not Sign. Not Pres. Not Pres. Not Sign. Not Sign. + +
Georgia Not Sign. Not Sign. Not Sign. − Not Sign. − Not Sign. Not Sign. Not Sign.

45
Germany − Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. Not Sign. Not Sign. +
Ghana + Not Sign. Not Sign. − Not Pres. Not Pres. − Not Sign. Not Sign.
Hong kong Not Sign. + Not Pres. − + Not Pres. Not Sign. − +
India + + + − Not Sign. Not Sign. − Not Sign. Not Sign.
Indonesia + + Not Pres. Not Pres. Not Pres. Not Pres. − Not Sign. Not Sign.
Iran Not Sign. Not Pres. + Not Pres. Not Pres. Not Pres. Not Sign. Not Sign. Not Sign.
Iraq + Not Pres. Not Sign. Not Pres. Not Pres. Not Pres. − + +
Italy Not Sign. Not Pres. Not Pres. Not Sign. − Not Pres. Not Pres. Not Sign. +
Japan + + Not Sign. Not Sign. Not Pres. Not Sign. Not Pres. Not Sign. Not Sign.
Jordan + Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. − − Not Pres.
South Korea + Not Sign. + Not Sign. + − + + Not Sign.
Kyrgyzstan + + Not Sign. Not Sign. + Not Sign. − Not Sign. +
Latvia Not Sign. Not Sign. Not Sign. − Not Pres. Not Sign. + Not Pres. Not Sign.
Lithuania Not Sign. Not Sign. Not Sign. Not Pres. Not Sign. + Not Sign. Not Pres. Not Sign.
Mali + + − Not Pres. Not Sign. Not Sign. − Not Sign. Not Pres.
Mexico Not Sign. − Not Pres. + + Not Sign. Not Sign. Not Sign. +
Moldova Not Sign. Not Sign. − Not Pres. Not Pres. Not Sign. Not Sign. Not Sign. +
Morocco Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. + Not Pres. Not Pres.
New Zealand Not Sign. − Not Pres. Not Sign. Not Sign. + Not Sign. Not Sign. +
Nigeria + + Not Pres. Not Pres. Not Pres. Not Sign. − Not Sign. Not Sign.
Norway + − Not Sign. Not Sign. Not Pres. − Not Sign. Not Sign. +
Pakistan Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. − Not Pres. +
Peru + − Not Pres. + Not Sign. Not Sign. Not Pres. Not Sign. Not Sign.
Philippines Not Sign. + Not Pres. Not Pres. Not Pres. Not Pres. + Not Sign. −
Poland Not Sign. Not Sign. Not Sign. − Not Pres. − Not Pres. Not Sign. +
Puerto Ricco Not Sign. − Not Pres. Not Sign. Not Pres. Not Sign. Not Pres. Not Sign. Not Sign.
Romania Not Sign. Not Sign. Not Sign. + Not Pres. Not Sign. − Not Pres. Not Sign.

46
Russian federation + + Not Sign. + Not Pres. + Not Sign. Not Pres. Not Sign.
Rwanda − Not Sign. − + Not Pres. − Not Sign. − Not Sign.
Saudi Arabia Not Pres. Not Pres. Not Pres. Not Pres. Not Sign. Not Pres. Not Sign. Not Sign. −
Slovakia Not Sign. Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. Not Pres. Not Sign. Not Sign.
Vietnam − Not Sign. − Not Sign. Not Pres. − Not Pres. − +
Slovenia Not Sign. Not Sign. − + Not Pres. Not Pres. Not Sign. Not Sign. +
South Africa Not Sign. Not Sign. − Not Sign. − + Not Sign. Not Sign. Not Sign.
Zimbabwe Not Sign. Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. Not Sign. Not Sign. Not Sign.
Spain Not Sign. Not Sign. Not Pres. Not Sign. Not Pres. Not Pres. − Not Sign. Not Sign.
Sweden + − Not Sign. − Not Pres. Not Sign. Not Sign. Not Sign. +
Switzerland − Not Sign. + + + Not Sign. Not Sign. Not Sign. +
Thailand Not Pres. − Not Pres. − Not Pres. Not Pres. + − Not Sign.
Trinidad and Tobago + Not Sign. Not Sign. Not Sign. − Not Pres. Not Sign. − Not Sign.
Turkey Not Sign. Not Sign. Not Sign. Not Pres. Not Pres. Not Sign. Not Sign. Not Sign. +
Uganda Not Sign. Not Sign. Not Pres. Not Pres. Not Pres. Not Pres. Not Sign. Not Sign. Not Sign.
Ukraine Not Sign. + Not Sign. + − Not Sign. Not Sign. Not Pres. +
Macedonia Not Sign. + + Not Pres. Not Pres. Not Sign. − Not Sign. +
Egypt Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. Not Pres. − + Not Pres.
Tanzania + + Not Sign. Not Pres. Not Pres. Not Sign. − Not Sign. Not Sign.
United states Not Sign. + Not Sign. Not Sign. Not Sign. + + − Not Sign.
Burkina Faso + + − Not Pres. − + − − Not Sign.
Uruguay Not Sign. − Not Pres. + Not Pres. Not Sign. Not Pres. Not Sign. Not Sign.
Venezuela Not Sign. − Not Sign. + + Not Pres. Not Pres. Not Sign. Not Sign.
Serbia and Montenegro Not Sign. Not Sign. Not Sign. − Not Pres. Not Sign. − Not Sign. Not Sign.
Zambia Not Sign. + + Not Sign. Not Sign. Not Pres. − Not Sign. Not Sign.
Serbia Not Sign. Not Sign. + Not Pres. Not Pres. Not Sign. − Not Pres. Not Sign.

47
Number of
Countries/Regressions 72 69 53 43 60 46 27 61 74
(by denomination):

48