THE EFFECTS OF COMMUNITY FACTORS ON SCHOOL PARTICIPATION IN TURKEY:
A MULTILEVEL ANALYSIS
By
Sedat Gumus

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
K-12 Educational Administration
2012

ABSTRACT
THE EFFECTS OF COMMUNITY FACTORS ON SCHOOL PARTICIPATION IN TURKEY:
A MULTILEVEL ANALYSIS
By
Sedat Gumus
By using a recent, large-scale, and nationally representative data set, this study aims to
explore the factors associated with school participation at both the primary and secondary levels
in Turkey, with specific attention to the community-level factors. The school participation of
children at both levels has been a prominent problem in Turkey, similar to many other
developing countries. Therefore, numerous studies have been conducted to determine the factors
associated with the school participation of boys and girls so far. Existing studies in Turkey,
however, have extensively focused on the association between household-level factors and
school participation, ignoring the role of the broader environment in which children live. This
study, therefore, makes an important contribution to the current school participation literature in
Turkey by taking socio-economic context variables into account with the multilevel modeling
method. The findings highlight the importance of community/context factors in explaining
school participation in Turkey. The results of the study can help policy makers develop a
systematic understanding of the relationship between socio-economic context and school
participation, and make more appropriate decisions for improving school participation across the
country.
With respect to the household level factors, the results are in line with the previous
literature on the relationship between household-level factors and the school participation of
children in Turkey. I find that the mother’s ability to speak Turkish and the household head’s
education are positively associated with school participation, while being female, being older,

mothers’ traditional gender role attitudes, household poverty, and residing in a large household
are negatively associated with school participation. The results of the multilevel analyses, which
are the key contribution of this study, show that the school participation of children in Turkey
significantly varies between communities, but only for children aged 14–17. Specific
community-level variables such as average adult education and the average gender role attitudes
in the community, which I use as a proxy for social context, are found to be significantly
associated with school participation, while economic variables, such as community poverty and
urbanization, are not significantly associated with school participation. These results also mostly
align with the results of existing studies in other developing countries and confirm the
importance of social context in which children live for their educational involvement.

Copyright by
SEDAT GUMUS
2012

ACKNOWLEDGMENTS

It would not have been possible to complete this demanding work without the assistance,
guidance, and support of many great people. I feel most fortunate to have had Professor Amita
Chudgar as my graduate advisor. Her decision to accept working with me was truly the best
thing that happened to me as a graduate student. Her “Education, Development, and Social
Change,” a course I was fortunate enough to take at the beginning of my studies, served as a
milestone in my graduate career. Since then, she has not only been an excellent teacher, but also
a great mentor with endless patience and kindness. She also devoted significant effort and time to
this study by reading numerous drafts, giving insightful feedback, and challenging me for my
best work. Words would only prove futile if I tried to explain the extent of my gratitude to her.
I would also like to express my deepest gratitude to my committee members, Professors
David Arsen, BetsAnn Smith, and Reitumetse O. Mabokela for their countless inputs to this
work and for their unbounded guidance for my scholarly development. I am thankful to several
other past and present faculty members in my department for their generous help, support, and
encouragement during my study. To name a few: Professors Susan Printy, Christopher Dunbar,
Donald Peurach, Melinda Mangin, Muhammad Khalifa, and our department chair Professor
Marilyn Amey. I cannot fully express my appreciation to Professor Nancy Colflesh. Had she not
invited me to numerous school visits and professional development activities, it would have been
impossible for me to learn the American education system and meet many school administrators
and teachers all across Michigan. Thank you, Nancy!
Before I turn to my friends and family, I would like to thank the Turkish Ministry of
National Education for their generous support for my graduate education in the U.S. I consider

v

myself to be extremely fortunate for the opportunity to serve my country as a researcher and
educator. Michigan State University also supported my education with multiple funds. I was able
to receive a Research Development Fellowship (2010) and Summer Research Fellowship (2011)
from the College of Education, along with travel assistance to many major academic conferences.
I also want to extend my appreciation to the Hacettepe University Institute of Population Studies,
which provided the data I used for this work. Finally, I extend my thanks to Professor Doug
Campbell for editing many of my papers, including my dissertation.
I was greatly lucky to meet so many wonderful people during my years in East Lansing.
Many thanks to the great people in my cohort; Anthony Berthiaume, Ashley Johnson, Celena
Mills, Cristi Dikeou, Nicole Campbell, Nimo Abdi, Ronetta Wards, Yulonda Walker, and
William Patterson, for making our courses so fun and productive. At the College of Education, I
also would like to thank my great friends Seung-Hwan Ham, Yisu Zhou, Wang Jun Kim, Dwi
Yuliantoro, Irfan Prasetya, Dara Kim, Nabil Abdullah, Juanita Bautista, Amy Bucknell, and
Seher Ahmad, with whom I took classes, wrote papers, and talked about my research.
Finally, and most importantly, I am heartily grateful to my family. I would like to start
with my wife Emine. I do not know where to start, because I owe her so much. I truly appreciate
the fact that she has postponed many of her wishes so that she could support me to pursue my
dreams. More importantly, she became my best friend and helped me grow with her love,
patience, understanding, and so many other traits that will forever enlighten my way. I will not
ever be able to pay back everything she has done for me, but I will do my best to make her happy
and follow her own dreams without having to postpone them ever again.
I am also so thankful to my two siblings, my elder sister and my brother, for being with
my parents when I was away for so many years. This long process could not be this comfortable

vi

without their support and prayers. And, I cannot fully express my appreciation to my mother. Her
unconditional love has been the greatest motivation for me. Mom, I know that these five years
might have seemed like a lifetime to you, but I would like you to know that I appreciate your
sacrifice from the bottom of my heart. My father, a serious man with the warmest heart,
contributed the most to who I am today. I admire his endless effort, which enabled him to
overcome all the hardships in his life without losing his optimism a single day. I could not have
been able to leave everything behind and come to the U.S. without his support and
encouragement. Thanks Dad! Just knowing that you are always there for me is the greatest
feeling in my life. Last, but not the least, thank you, Yigit, my son, for the joy and happiness you
have brought to my life. Wait, Daddy is coming home to spend more time with you!
I dedicate this dissertation to the two women of my life, my mother and my wife, who
make this world a better place to live not only for myself but also for whoever has been fortunate
enough to meet them.

vii

TABLE OF CONTENTS

LIST OF TABLES ......................................................................................................... x
LIST OF FIGURES ...................................................................................................... xi
CHAPTER 1
INTRODUCTION.......................................................................................................... 1
1.1. Purpose of the Study .............................................................................................. 1
1.2. Background of Turkey ........................................................................................... 6
1.2.1. Regional Characteristics of Turkey ........................................................... 7
1.2.2. Socio-economic and Cultural Differences between Regions................... 11
1.2.3. The Educational System and Educational Attainment in Turkey ............ 12
1.2.4. Regional Differences in Educational Attainment .................................... 14
CHAPTER 2
LITERATURE REVIEW ........................................................................................... 18
2.1. Theoretical Background ....................................................................................... 18
2.1.1. Socio-cultural Considerations .................................................................. 18
2.1.2. Economic Considerations …..… ............................................................. 23
2.2. Empirical Evidence from Developing Countries ................................................. 25
2.2.1. Economic Community Factors ................................................................ 27
2.2.2. Socio-cultural Community Factors .......................................................... 31
2.3. Previous Studies in Turkey .................................................................................. 33
2.3.1. Data, Variables, and Methods Used ......................................................... 34
2.3.2. Core Findings ........................................................................................... 34
2.4. Summary and Research Questions....................................................................... 38
CHAPTER 3
DATA AND METHODOLOGY ................................................................................ 42
3.1. Data … ............................................................................................................... 42
3.1.1. Data Source .............................................................................................. 42
3.1.2. Sample Selection Decisions and Representativeness .............................. 43
3.1.3. Definition of ―Community‖ ..................................................................... 45
3.2. Variables ............................................................................................................. 45
3.2.1. Dependent Variable ................................................................................. 45
3.2.2. Independent Variables ............................................................................. 46
3.3. Correlations among Variables............................................................................... 50
3.4. Empirical Strategy .............................................................................................. 50
3.4.1. Binary Logit Model ................................................................................ 51
3.4.2. Hierarchical Linear Modeling .................................................................. 52
CHAPTER 4
DESCRIPTIVE AND PRELIMINARY ANALYSES .............................................. 56
viii

4.1. Descriptive Statistics ............................................................................................. 55
4.2. Out-of-school Children by Different Groups ..................................................... 58
4.3. Characteristics of Out-of-school Children ........................................................... 63
4.4. Results of Binary Logit Analysis ......................................................................... 65
4.4.1. Employing the Binary Logit Model for Different Age Groups ............... 67
4.4.2. Employing the Binary Logit Model for Different Regions ..................... 70
CHAPTER 5
MULTILEVEL ANALYSES ...................................................................................... 76
5.1. Results of Unconditional and Random Coefficient Models ................................ 77
5.2. Analysis of Final HLM Model ........................................................................... 80
CHAPTER 6
DISCUSSION AND CONCLUSIONS ...................................................................... 87
6.1. Summary of Findings ........................................................................................... 87
6.2. Policy Implications for Turkey ........................................................................... 92
6.3. Broader Implications beyond Turkey .................................................................. 94
6.4. Suggestions for Future Research ......................................................................... 95
APPENDIX

............................................................................................................ 100

BIBLIOGRAPHY ...................................................................................................... 105

ix

LIST OF TABLES

Table 1.1 Net enrollment rates and sex ratios ................................................................ 14
Table 4.1 List of variables and descriptive statistics .................................................... 57
Table 4.2 School participation by level-1 and level-2 variables, mean and S.D ........... 58
Table 4.3 Percentage of out-of-school children by different groups ............................ 60
Table 4.4 Regional differences in school participation by type of residence ............... 62
Table 4.5 Selected important attributes of out-of-school children (%) ........................ 64
Table 4.6 Results of binary logit analysis for whole sample ........................................ 66
Table 4.7 Binary logit analysis by age groups .............................................................. 68
Table 4.8 Binary logit analysis by regions..................................................................... 71
Table 5.1 The results of unconditional models .............................................................. 78
Table 5.2 The results of random coefficient models...................................................... 79
Table 5.3 The results of HLM analysis for whole sample and aged 14–17................... 81
Table 5.4 The results of HLM analysis for two regions (East and Central) .................. 84
Table A1 Previous studies on educational attainment issue in Turkey ....................... 100
Table A2 School participation by age in original data set ........................................... 102
Table A3 Age of children in the final data set ............................................................. 102
Table A4 Omnibus Tests of Model Coefficients ......................................................... 103
Table A5 Out-of-school girls and boys by age groups ................................................ 103
Table A6 Out-of-school children by mother’s ability to speak Turkish (Age 8-13) ... 103
Table A7 Pearson’s Correlation between level-2 variables ......................................... 103
Table A8 Kendall Tau’s nonparametric correlation .................................................... 103

x

LIST OF FIGURES

Figure 1.1 Geographical regions of Turkey .................................................................... 8
Figure 2.1 Conceptual framework ................................................................................. 40

xi

CHAPTER 1
INTRODUCTION

1.1.Purpose of the Study
This study’s purpose is to explore the factors associated with school participation at both
the primary and secondary levels in Turkey, with specific attention to community-level factors,
which have received limited attention so far in the Turkish literature. Using recent, large-scale,
and nationally representative data, this study makes an important contribution to the current
education policy conversations in Turkey.
Turkey has been decisively working to be a participant in democratic western
modernization since the Republic of Turkey was established in 1923. The start of membership
negotiations with the European Union (EU) in 2005 can be seen as one of the biggest indicators
of Turkey’s effort to achieve this aim after the establishment of the republic. In addition, this
event was an important milestone in Turkey’s long-running relationship with the EU and it
implies that Turkey has made major strides in many different economic and social areas in recent
years. However, it should be noted that Turkey still experiences significant issues in its education
sector, especially regarding access to education. To illustrate, Turkey has a very high level of
adult illiteracy and a low level of primary and secondary school participation, even compared to
the least developed European countries in its region (UNESCO, 2011). At this point, it can be
safely claimed that Turkey’s continued growth and development require a more highly educated
population, because education is considered an essential element of socio-economic development
(Hanushek & Woessmann, 2008; Lockheed, Bloch, Hamilton, & Vespoor, 1990; Psacharopoulos

1

& Patrinos, 2002).
There is a robust literature which empirically demonstrates the significant returns to
education at both individual and societal levels. In terms of economic returns, the literature
suggests a strong positive relationship between the level of education which individuals attain
and their individual earnings, as well as the economic growth of their countries (Cohen & Soto,
2007; De la Fuente & Doménech, 2006; Hanushek & Kimko, 2000; Krueger & Lindahl, 2001;
Psacharopoulos, 1994; Psacharopoulos & Patrinos, 2002; Tansel, 2004). It is also argued that
there is a considerable social return to education because of its many positive impacts on social
life, such as reducing crime rates, improving health, and promoting civic participation (Dee,
2004; Wolfe & Haveman, 2002). Therefore, the strong evidence on the positive returns to
education at both individual and societal levels has motivated many governments to invest more
in education (Hanushek, 2003). While most developed countries are currently discussing how to
raise the quality of education, providing basic education to all, regardless of geographic location
or socio-economic background, is still a main target in many developing countries (Tansel,
2004).
The international community also views basic education as a necessary prerequisite for
children in developing countries. In this respect, the international initiative ―Education for All‖
(EFA) was launched in 1990 in Jomtien, Thailand, with the aim of helping developing countries
provide at least a basic education to all members of their society. In accordance with this aim, six
education goals were generated by a diverse group of national governments, non-governmental
organizations, and international agencies. Two of these six EFA goals are committed to
increasing the school participation of children all over the world by ensuring universal primary
education and eliminating gender disparity at both primary and secondary education levels.

2

1

These goals are:
Goal 2) to ―ensure that by 2015 all children, particularly girls in difficult circumstances
and belonging to ethnic minorities, have access to and complete free and compulsory primary
education of good quality‖ and,
Goal 5) to ―eliminate gender disparities in primary and secondary education by 2005 and
achieve gender equality in education by 2015 with a focus on ensuring girls’ full and equal
access to and achievement in basic education of good quality.‖
Although both national governments and many international agencies have been working
diligently to increase access to education, millions of children still do not attend primary or
secondary school in developing countries. Even at the primary level, there was a total of 72
million out-of-school children in 2007 around the world. Turkey has also struggled with
providing basic education to all, with approximately 640,000 out-of-school children at primary
school level in 2007 (UNESCO, 2010). According to trend analysis, although the number of outof-school children will significantly decline in Turkey, there will still be around 340,000 primary
school-aged children not in school in 2015 (UNESCO, 2011). Thus, Turkey is considered to be
one of the few higher-income countries which may not achieve universal primary education by
2015. At the secondary level, school participation is even a bigger problem in Turkey because
around one-third of all secondary school aged children are not enrolled in school (UNESCO,
2010; MoNE, 2010).
It is clear that the school participation of children at both primary and secondary
education levels is still a prominent problem in Turkey and many other developing countries. In
this context, as a first step, determining the factors that are associated with the school
1

http://web.worldbank.org/WBSITE/EXTERNAL/TOPICS/EXTEDUCATION/0,,contentMDK:
20374062~menuPK:540090~pagePK:148956~piPK:216618~theSitePK:282386,00.html
3

participation of children plays a key role in providing formal education opportunities to every
child in every society. Hence, numerous studies have been conducted to determine the factors
associated with the school participation of boys and girls in developing countries. Among the
three key factors of school, family, and community, which might affect children’s educational
attainment, existing studies extensively focus on the importance of family and school
characteristics (Bramley & Karley, 2007; Buchmann & Hannum, 2001; Govinda &
Bandyopadhyay, 2010; Nechyba, McEwan, & Older-Aguilar, 1999). Many studies from
developing countries, for example, have investigated the association between various householdlevel factors, such as household wealth, household size, and parental education, and the school
participation of children in these households (Anh, Knodel, Lam, & Friedman, 1998; Chudgar,
2009; Connelly & Zheng, 2003;Huisman & Smits, 2009a; Knodel, Havanon, & Sittitrai, 1990;
Pong, 1997; Sudha, 1997). Similarly, there are several studies that have investigated the factors
associated with educational attainment in Turkey, with specific interest in the effects of family
characteristics (Bakis, Levent, Insel, & Polat, 2009; Dayioglu, Kirdar, & Tansel, 2009; Duman,
2010; Goksel, 2008; Hisarciklilar, 2002; Koc, 2008; Smits & Gunduz-Hosgor, 2006; Tansel,
2002; Tomul, 2008).
The effects of community factors on school participation, however, have not yet been
substantially explored (Buchmann & Hannum, 2001; Chudgar, 2006b; Chudgar & Shafiq, 2010;
Duncan, 1994), either in Turkey or in other developing countries, although there is a growing
interest in this subject. Several recent studies suggest that there is a significant relationship
between the community where children live and their educational outcomes, even after
controlling for household-level characteristics (Adams, 2006). Studies in developing countries
(e.g., China, India, and Mexico) show that several community-level factors, such as average

4

income, children’s overall school enrollment, and average educational attainment of adults, are
strongly associated with schooling desire or the actual school participation of children (Adams &
Hannum, 2005; Binder, 1999; Chudgar, 2008; Chudgar & Shafiq, 2010; Connelly & Zheng,
2003). There is not, however, any study that exclusively investigates the effects of community
factors on school participation of children in Turkey, to my knowledge. Only Tansel (2002) has
used, in addition to household-level variables, a few community-level explanatory variables such
as the development of streets, distance to regional metro centers, and local employment
composition, to predict the impact of context on school participation in Turkey. However, these
variables are mostly provincial-level indicators (except development of streets) and may not truly
represent the effect of the community.
In sum, the school participation of children at both the primary and secondary levels is
still an undeniable problem in Turkey. Similar to the current situation in Turkey, millions of
school-aged children do not participate in formal education around the world. There have been
many studies investigating the factors that contribute to this situation in different developing
countries, with special attention to the household-level factors. The existing literature, however,
suggests that the effects of community factors on school participation in developing countries are
also likely important, but not well researched and understood. In Turkey, there is also an
important gap in the literature on the relationship between the community in which children live
and their school participation. My aim, therefore, is to analyze the association between
community-level factors and the school participation of children in Turkey by using both sociocultural (e.g., level of adult education in the community) and economic (e.g., level of poverty in
the community) community-level variables. My specific research questions are: 1) What is the
current pattern of school participation at the primary and secondary levels in Turkey? 2) Are

5

community level factors significantly associated with school participation in Turkey, after
controlling for individual- and household-level factors?
Answering these questions can make an important contribution to the school participation
literature in Turkey by depicting the current situation and by revealing the role of communitylevel factors, which has been ignored so far. This study can also help policy makers develop a
systematic understanding of the relationship between social context and school participation, and
make more appropriate decisions for improving school participation across the country.
The following section provides detailed information about Turkey, with a specific focus
on regional differences, the Turkish education system, and the current school participation
problem in Turkey, to give the reader an understanding of the background context.
1.2.Background of Turkey
The Turkish Republic was founded in 1923, following the fall of the Ottoman Empire.
After the establishment of the republic, some of the long-held cultural, social, and religious
traditions of the empire were prohibited. Meanwhile, numerous economic, legal, and social
reforms were implemented in the first two decades of the new republic, with the aim of
modernizing the nation through the adoption of secularism and democracy. Since education was
seen as a key tool in this modernization process, substantial changes in the educational sector
were also introduced. These changes included making primary education compulsory and
adopting Roman alphabets to replace Arabic scripts (Hacettepe University Institute of Population
Studies, 2009; Kaya, 2004). Since the beginning of the republic, there have also been continued
efforts to increase access to education among different social groups in the society. Although
significant improvements have been made, a low level of educational attainment is still an
important problem in Turkey. In addition, people who live in less developed parts of the country
6

suffer more from this problem because of both low socio-economic development and traditional
cultural norms.
Turkey is geographically positioned between Asia and Europe. Its geographic position
also reflects the culture and values of the people who live in the country. Hence, diverse cultural,
educational, social, and economic situations can be observed in different parts of Turkey. The
western portion of the country, which is surrounded by European countries, mostly adopted
western values, while much of the population in the east, where Turkey has its borders with
Middle Eastern countries such as Iran, Iraq, and Syria, still generally holds to traditional beliefs
and attitudes. The same difference can also be observed between people who live in metropolitan
cities and people who live in rural settlements. It should also be noted that patriarchal ideology is
still dominant in social life, especially in less developed parts of the country, although gender
equality is protected by law (Hacettepe University Institute of Population Studies, 2004; Kaya,
2004; Smits & Gunduz-Hosgor, 2006).
1.2.1. Regional Characteristics of Turkey
Turkey has seven geographic regions, which were formed based on their climate, location,
agricultural products, and so on, but they do not represent an administrative division. These
regions were named Marmara, Aegean, Mediterranean, Black Sea, Central Anatolia, Eastern
2

Anatolia, and Southeastern Anatolia according to their geographic location. Turkey also has 81
administrative divisions (provinces), which are further divided into districts, subdivisions, and
villages (Hacettepe University Institute of Population Studies, 2009).

2

http://www.worldturkey.com/lang/eng/regions.php
7

Figure 1.1: Geographical regions of Turkey

3

In addition to the geographic breakdown, Turkey is also divided into five main regions
(West, South, East, North, and Central) in the literature based on socioeconomic differences and
the demographic structure of the country. Social surveys generally use these five main regions
for sampling purposes (Hacettepe University Institute of Population Studies, 2009).
The West of Turkey consists of the country’s two well-developed geographic regions,
Marmara and Aegean, and includes the country’s largest city, Istanbul. While the Aegean coast
holds a significant role for producing and exporting important crops, the northwestern part of the
region is known as the center of the financial activities in the country. As a result of the region’s
importance for the country’s economic development, most of the infrastructure of roads and
schools was built earlier in this region compared to other regions. The region, however, shows an
important diversity regarding the demographic structure and socio-economic status of its
inhabitants, especially in metropolitan cities, since many people from less developed regions of
the country have migrated to this region for socio-economic reasons during the last several

3

For interpretation of the references to color in this and all other figures, the reader is referred to
the electronic version of this dissertation.
8

decades. In terms of school participation, there are relatively higher participation rates for both
boys and girls in the West, partly because of the better infrastructure, the higher need for
educated workforce, and the weaker impact of traditional social norms in the region (GündüzHosgör & Smits, 2007; Hacettepe University Institute of Population Studies, 2009; Smits &
Gunduz-Hosgor, 2006).
The South also has some fertile agricultural areas and manufacturing sector similar to
those of the West. In addition, one of the biggest economic sectors in the area is the tourism
industry along the coastline, including Antalya, which provides a significant proportion of the
country’s tourism revenue. The South is also mostly urbanized, like the West, but inner parts of
the region are not as developed as the coastline. Because of booming industrial activities and
tourism sector, this region has also witnessed increased in-migration from the relatively less
developed regions, especially during the last two decades. As in the West, a higher level of
school participation for both boys and girls is also expected in this region, since the region
experiences relatively high level of economic development and urbanization (Gündüz-Hosgör &
Smits, 2007; Hacettepe University Institute of Population Studies, 2009; Smits & GunduzHosgor, 2006).
Central Turkey has witnessed recent development in several industrial sectors, such as
furniture and marble, although industry is not well developed compared to the West. Agriculture
is also relatively less developed in this region, mostly because of the unfavorable climate and
geographic restrictions. Ankara, the capital city of Turkey, makes the biggest contribution to the
region, as a city where many governmental activities take place. Ankara also plays a significant
role in the social development of the region, as it has some of the best public and private
universities of Turkey. Although the region, especially the countryside, might be categorized as

9

underdeveloped compared to the West and the South, the average socio-economic development
and educational infrastructures are better than in the East and the North, mostly because of the
capital city of Ankara. In general, a moderate level of school participation is expected in this
region (Gündüz-Hosgör & Smits, 2007; Hacettepe University Institute of Population Studies,
2009; Smits & Gunduz-Hosgor, 2006).
The North region is known for its important agricultural products, such as hazelnut and
tea. The fertile areas of the region, however, are mostly located along the coastline and isolated
from the rest of the country because of high mountains and intensive forests. While industrial
activities are growing in some western provinces of this region, industry is not well developed
and traditional methods are used for agriculture due to the geographic restrictions in the
northeastern part. Since males generally tend to out-migrate or not work on lands in the region,
women are mostly charged with agricultural work. The North region, therefore, has the highest
rate of female employment among all regions of Turkey. Relatively higher school participation is
expected for girls in the region because of women’s more independent position (Gündüz-Hosgör
& Smits, 2007; Hacettepe University Institute of Population Studies, 2009; Smits & GunduzHosgor, 2006).
The East of Turkey geographically consists of the Eastern and Southeastern Anatolia
regions. The overall socio-economic development of the region is considerably low compared to
the rest of the country. Furthermore, there are many hardships in the region related to the climate
and geographical circumstances. Therefore, agricultural activities are not common and the
mechanization in agriculture is very low, although most of the people are living in rural areas in
the region. However, some agricultural development has appeared recently with the ―Southeast
Anatolia Project,‖ especially in the southern part. Industrial activities are also very limited and

10

do not have a prominent impact on the economy. In terms of social life, tribal structure and
traditional patriarchal norms still somewhat maintain their influence in the region. It is also
noteworthy that the region has experienced high levels of internal and external migration from
rural areas to the larger cities, mostly because of socio-economic restrictions and terrorist
activities during the last two decades. In the East region, the school participation rate is very low,
especially for girls, partly because of the current socio-economic condition and the patriarchal
cultural norms and beliefs in the region (Gündüz-Hosgör & Smits, 2007; Hacettepe University
Institute of Population Studies, 2009; Smits & Gunduz-Hosgor, 2006).
1.2.2. Socio-economic and Cultural Differences between Regions
There are extensive differences among the regions of Turkey in terms of many socioeconomic indicators (Gündüz-Hosgör & Smits, 2007). In general, lesser development levels have
been found in existing empirical studies regarding educational, health, and economic matters in
moving from western to eastern provinces (Gedik, Sahin, & Suer, 2002; Gezici & Hewings,
2007; Gezici & Keskin, 2005; Gündüz-Hosgör & Smits, 2007; Tomul, 2007; Unal, 2008). To
illustrate, most of the eastern provinces are significantly disadvantaged in terms of literacy rates
and schooling ratios compared to the western provinces. In addition, higher fertility rates and
larger households have been observed in eastern provinces (Gedik, et al., 2002). In terms of the
relationship between location and economic development, it can be argued that the coastal
provinces of the West and the South are the most prosperous provinces of the country. In
contrast to the provinces on the western and southern coasts, many provinces on the north coast
can be defined as underdeveloped, as can almost all the provinces of the East and some of the
provinces of the Central region (Gezici & Hewings, 2007; Gezici & Keskin, 2005).
The regions of Turkey also show great differences in terms of traditional gender role

11

attitudes and the status of women. For example, 41 percent of the women have five or more
children in the countryside of the East, while only 12 percent of the women in the countryside of
the West experience the same situation (Gündüz-Hosgör & Smits, 2007). In terms of
employment, the 2008 Turkey Demographic Health Survey (TDHS) main report indicated that
30.7 percent of all women (age 15–49countrywide) reported being currently employed at the
time of the survey. While the North had the highest employment ratio, 53.0 percent, the East had
the lowest ratio of 19.9 percent, and the other three regions had employment ratios around 30
percent (Hacettepe University Institute of Population Studies, 2009). Furthermore, women in the
East region may experience additional disadvantages because a high number of women in the
region are not able to speak the official language of the country, Turkish, and some women even
4

do not have a civil marriage. It was also found that the percentages of women who accept
traditional gender roles are much higher in the East region compared to the West region
(Gündüz-Hosgör & Smits, 2007).
1.2.3. The Educational System and Educational Attainment in Turkey
The Turkish education system includes the pre-primary, primary, secondary, and higher
education levels. Primary education included five years of elementary (for children aged 6–11)
and three years of middle (aged 12–14) schooling before 1997, and just the first five years were
compulsory for every child in the country at that time. In 1997, however, the elementary and
middle schools were combined and compulsory education was extended from five to eight years
for all children whose ages fell between 6 and 14. Starting in 1997/1998, middle schools stopped
accepting new students for education and primary schools became eight years. With a recent

4

Those who do not have a civil marriage generally have religious marriage, which is not
officially accepted in Turkey.
12

educational reform in May 2012, primary schools were again divided into elementary and middle
schools. Each of these levels includes four years of education. Secondary education included at
least three years of high school before 2005. Beginning with the 2005/2006 school year, the
minimum length of secondary education was gradually extended to four years. Today, secondary
education is considered to last for a period of four years regardless of where in the country you
live (MoNE, 2010). The recent education reform that is mentioned above also made the four
years of secondary education compulsory for all boys and girls, making the total of compulsory
education 12 years in Turkey.
As previously mentioned, eight years of primary education has been compulsory for more
than a decade, and both primary and secondary schooling are free in public schools in Turkey.
However, there are still thousands of out-of-school children at both levels, mostly because of
high regional disparities and gender inequalities. According to the most recent EFA report,
Turkey is one of the few higher-income countries that may not achieve universal primary
education by 2015. In Turkey, approximately 640,000 primary school-aged children were out of
school in 2007. The profile of out-of-school children in Turkey also seems quite problematic,
because more than half of these children are unlikely to enter school later. Furthermore,
secondary school participation is fairly low in certain regions and among girls (UNESCO, 2010).
School enrollment patterns in Turkey can be seen in Table 1.1, which shows the net
5

enrollment rates and sex ratios since 2000 (MoNE, 2010). Although there have been significant
increases in enrollment rates and sex ratios at both primary and secondary levels during the last
couple of years, the un-enrollment problem has not been fully eliminated. From 2000 to 2009,
the net enrollment has increased from 95.28 percent to 98.17 percent at the primary level and
5

It is obtained by dividing the female gross schooling ratio by the male gross schooling ratio
multiplied by 100.
13

from 43.95 percent to 64.95 percent at the secondary level. Similarly, sex ratios have increased
from 89.64 percent to 98.91 percent at the primary level and from 74.41 percent to 88.59 percent
at the secondary level.
Table 1.1: Net enrollment rates and sex ratios
Net Enrollment Rates
Primary
Secondary
Year
*
**
education
education
95.28
43.95
2000/'01
92.40
48.11
2001/'02
90.98
50.57
2002/'03
90.21
53.37
2003/'04
89.66
54.87
2004/'05
89.77
56.63
2005/'06
90.13
56.51
2006/'07
2007/'08
97.37
58.56
2008/'09
96.49
58.52
2009/'10
98.17
64.95
Source: Ministry of National Education, (2010)
*
Grade 1–8

Sex Ratios
Primary
Secondary
*
**
education
education
89.64
74.41
90.71
75.87
91.10
72.32
91.86
78.01
92.33
78.72
93.33
78.76
94.11
79.65
96.39
85.81
97.91
88.99
98.91
88.59

**

Grade 9–11 until 2008–2009 and grade 9–12 after that (Only students who started high school
after 2005 had to study four years)
1.2.4.

Regional Differences in Educational Attainment

In addition to the high number of out-of-school children in Turkey, there are also
significant inequalities regarding average educational attainment between different regions. In
general, average years of schooling (AYS) for both males and females are higher, and the gap
between the two genders is lower in the West and the South. In contrast, AYS for both males and
females in the East of Turkey are relatively lower, and the gap between the two genders is higher
(Tomul, 2007). More specifically, while 21 percent of people (age 17 to 22) have less than four
years of education in the eastern part of the country, this is the case for only 2to 7 percent of the
same age group in other regions (UNESCO, 2010). Female education also shows great
disparities across the regions of the country. Overall, one in every five women has no education
14

or has not completed the first level of primary school in Turkey. However, more than half of
women who live in the East have no education or have not completed first level primary school,
while just around one in every seven women experience the same situation in the West
(Hacettepe University Institute of Population Studies, 2004). It has also been found that the
educational level of married men is fairly higher than the educational level of their wives in all
regions of the country. As expected, the difference is the lowest in the West and the highest in
the East (Gündüz-Hosgör & Smits, 2007).
The current situation in the school participation of boys and girls in the country also
reflects the situation regarding regional differences in average educational attainment. At all
levels, there is higher school participation for both boys and girls in the West, and lower
participation in the East and several provinces in Central Turkey. According to statistics from the
Ministry of National Education (MoNE), the general net enrollment ratio was 98 percent for
primary schooling and 65 percent for secondary schooling at the beginning of the 2009/2010
educational year. However, there are still several cities in the Central and the East that have
around 90 percent of primary enrollment, such as Yozgat (90), Cankiri (89), Hakkari (90), Bitlis
(94), etc. The situation is much worse in terms of the secondary education enrollment ratios
because there are several cities, all of them located in the East, that have even lower than a 40
percent enrollment ratio, such as Agri (27), Van (33), Mus (28), Bitlis (34), Sanliurfa (32),
Sirnak (34), and Siirt (38).
There is also no doubt that gender disparity in school participation is a distinctive reality
and shows significant differences between different regions of the country. Turkey has the lowest
gender parity index (ratio of girls to boys in school) at the secondary level, and the second lowest
gender parity index at the primary level, among 19 countries in Central and Eastern Europe

15

(UNESCO, 2011). In Turkey, around 60 percent of out-of-school children at the primary school
level are girls, compared a world average of 54 percent. Between the ages of 8 and 12, 7 percent
of girls do not enroll in school, while this is the case for only 2 percent of boys. Furthermore, this
gap becomes much wider at the secondary level. Particular regions also show greater gender
disparities in school participation. While girls and boys enjoy equal rates of enrollment at the
primary level and are beginning to reach that parity at the secondary level in the western half of
the country, the eastern part of Turkey has experienced high gender inequality in school
participation, especially at the secondary level. In eastern Turkey, for example, the percentage of
girls’ enrollment rates to boys’ is highest at age 9 (85 percent), while it drops below 40 percent
by age 15. This situation is even worse in rural areas of the region, since less than 20 percent of
15 years old girls are enrolled in school (UNESCO, 2010).
In terms of the actual enrollment rates, according to MoNE, the overall sex ratios for
student enrollment were .98 for primary schooling and .88 for secondary schooling in the
2009/2010 educational year for the country. According to these results, it seems that gender
disparity was almost eliminated at the primary school level for most parts of the country, except
several eastern provinces. At the secondary level, however, big disparities between girls and
boys still exist, with the greater inequality in the East. For example, sex ratios of secondary
school enrollment rates in several eastern provinces were .57 (Mus), .56 (Bitlis), .62 (Sanliurfa),
and .57 (Siirt) at the beginning of the 2009/2010 educational year. When considering the fact that
the general enrollment rates were also very low in most of the same provinces, the actual
enrollment rates of girls were significantly low in these provinces (MoNE, 2010).
In sum, Turkey has great socio-cultural and economic diversities between and within its
regions. In addition, the educational attainment of adults and school participation of children

16

show great differences between Turkish regions. While there are generally higher rates of school
participation for both girls and boys in western provinces, eastern provinces experience lower
school participation in general and higher gaps between girls’ and boys’ schooling. Because of
the regional diversity that exists within the country, Turkey serves as a very suitable place to
investigate the effect of community-level socio-cultural and economic factors on the school
participation of children.
In the remaining chapters I now turn to systematically investigating the importance of
community factors for school participation in Turkey. In the next chapter, I provide a review of
related theoretical and empirical literature in other developing countries. I also summarize the
findings of existing studies that investigate the relationship between household factors and
school participation in Turkey. In the third chapter, I set up the methodological foundation for
my statistical analyses by introducing my data set, variables, and statistical models. While the
fourth chapter includes the results of descriptive and preliminary analyses, the fifth chapter
provides the results of more advanced statistical analyses that answer the primary research
question of this study. In the last chapter, I summarize my findings and state my suggestions for
policymakers and for future research.

17

CHAPTER 2
LITERATURE REVIEW

This chapter provides a systematic review of the literature that informs this study. The
review of the literature is divided into three main areas. I first discuss theoretical literature that
highlights the importance of community characteristics in shaping individual beliefs and
behaviors, including schooling behaviors. Next, I present the existing empirical evidence on the
relationship between community-level socio-cultural and economic factors and educational
attainment in developing countries. Finally, I review the empirical literature on the determinants
of educational attainment in Turkey and the emphasis on community factors in the Turkish
literature.
2.1. Theoretical Background
2.1.1. Socio-cultural Considerations
Sociologists often claim that actions of individuals are not independent from the social
context in which they live. Based on this claim, it is also argued that children’s development and
their participation in formal education are also influenced by broader community they are part of.
In this section, therefore, I review the theoretical considerations regarding to the relationship
between socio-cultural context and individual behaviors, with specific attention to children’s
development and their educational involvement.
a) Social capital and individual behavior
Social action is generally described and explained in two different ways. While
economists often view action as guided independently by self-interest, sociologists suggest that it
18

is shaped and constrained by the social context in which individuals live. According to most
sociologists, social action is controlled by the norms and rules of society (Coleman, 1988). It is
argued that these norms and rules can shape individuals’ beliefs and behaviors through the
process of social influence. In this process, individuals feel obligated to behave in accordance
with the norms of their community because they wish to either avoid sanctions or to get others’
acceptance (Kravdal, 2002). In addition to social influence, social context can also affect
individuals’ behaviors more directly through transmission of knowledge and attitudes among
members of the society (Kravdal, 2002; Lee & Croninger, 1996). As a result of this transmission
process, which is generally called social learning, individuals may completely change or redesign
their beliefs and behaviors. In sum, from a sociological perspective, it is argued that the social
context can affect individuals’ behaviors through their social relationships with others with
whom they live.
The relationship one has with other individuals or institutions in society, in a broad view,
was conceptualized as social capital by Coleman (1988). Coleman argued that, similar to both
physical capital and human capital, social capital can also be thought of as a resource for
individuals’ behaviors. For example, obtaining information through social relationships, as one
of the most important forms of social capital, can play a very important role in individuals’
behaviors, because information generally affects behaviors. From this perspective, some people
are more advantaged than others since both the quality and quantity of information that can be
obtained through social relationships strongly depend on the social context in which individuals
live. Social capital, therefore, may also be defined as the benefits one can gain because of his/her
connections with others in a specific system (Kelly, 2002). Based on these definitions, it can be
argued that one who lives in an affluent community probably has more social capital than others

19

who live in poor or less educated communities.
Social capital also refers to the shared values that allow individuals to live together
(Fukuyama, 1999). This type of social capital is formed extensively by the norms and sanctions
in society. From this perspective, social capital not only encourages individuals to behave in
certain ways, but also limits them (Coleman, 1988). For example, norms in a community that
provides high respect and more opportunities for educated women can encourage women to get
more education and can result in high educational attainment among women in that community.
On the other hand, norms in another community can limit women’s involvement in education or
other social activities. Furthermore, it is more likely that a person who lives in the former context
will develop positive attitudes toward women’s education compared to a person who lives in the
latter context. It therefore can be argued that the social capital that individuals possess can
influence both their personal beliefs and behaviors. In this respect, social norms and social
relations among individuals in the community are often mentioned as significant contributors to
educational exclusion (Govinda & Bandyopadhyay, 2010).
b) Social context and children’s development
It is a common argument that individuals’ beliefs and behaviors are influenced by others,
as human development is a complex process that is involved in and affected by multiple
ecological systems. For example, events happen in school and at home, and interactions between
these two settings can affect a child’s development (Bronfenbrenner, 1979). In a similar vein, it
is also argued that the development of children can be affected by the community in which they
live. There is a growing literature about the effects of community-level socio-economic and
structural differences on children’s development. This literature indicates that the impact of
communities on children’s development is separate from the impact of household characteristics

20

(Adams, 2006). This impact can be derived from the peer group of children as well as the adults
with whom children communicate or whom they observe in their community. At this point, three
types of theories — collective socialization models, institutional models, and epidemic
models — suggested by Jencks and Mayer (1990) may help to explain how children’s
development can be influenced by the social context in which they live.
Collective socialization models emphasize the role modeling of adults and argue that
affluent adults in the community can help children to behave in accordance with social norms
and public order (Jencks & Mayer, 1990). Adults in a community, as role models, may either
negatively or positively influence children’s behaviors and development in that community. For
example, seeing adults who are educated and successful may encourage children to think that it
is possible to be successful by getting more education, even if they do not see similar role models
among their immediate family members. In contrast, if children do not know anybody who
graduated from high school in their community, they probably think that attending high school is
not necessary or possible for someone from their community. In addition, adults can serve as
resources for children in terms of providing information about schools and job opportunities.
Hence, children’s chances of getting more information about these matters increase with the
level of education that adults have attained in their community (Wilson, 1987).
While institutional models are also interested in the effects of adults on the development
of children, their specific focus is the role of exogenous adults who work in community
institutions, such as schools and the police force. Similar to the collective socialization models,
these models also point out the importance of the relationship between adults and children in the
same community. Specifically, it is suggested that children in affluent communities benefit more
from exogenous adults compared to children from poor communities. For example, it is assumed

21

that police often behave differently in rich and poor communities, and this may negatively affect
the criminal records of children who live in poor communities. Likewise, it is well known that
affluent communities often get better teachers, principals, social workers, etc. who can positively
influence children’s development by supporting their educational attainment and achievement
(Jencks & Mayer, 1990).
Lastly, epidemic models focus on how peers influence the behaviors of each other (Crane,
1991; Gephart, 1997; Jencks & Mayer, 1990). These models assume that children’s behaviors are
not independent from how their peers behave in the same community. For example, living in a
community where every child is involved in some kind of illegal activity may make it hard for
children to avoid similar activities. Similarly, peer groups can influence the educational
attainment of children, independently from family characteristics. If children live in a community
where everyone goes to school, they may also feel compelled to go to school, even if it is not
theirs or their families’ intentional selection. Based on these considerations, it is argued that even
children from very similar families can behave distinctly different in poor or affluent
communities because of the general behaviors of children in the community (Jencks & Mayer,
1990).
In general, theoretically suggested close association between the social context and
individuals’ behaviors implies that the schooling decisions of children or their families are
affected by the social context in which they live. In this regard, social context can influence
children’s school participation in variety of ways (Garner & Raudenbush, 1991). Parents, for
example, often benefit from the experience of other parents in their community when they make
important decisions about the education of their own children (Lee & Croninger, 1996).
Children’s willingness to attend school can also be affected by the level of education that adults

22

attain in their community, as illustrated before. Similarly, children can be influenced by the
school participation pattern of their peers in their community (Dreze & Kingdon, 2001).
2.1.2. Economic Considerations
There are also theories that focus on the effects of community-level economic resources
and incentives on individuals’ behaviors and children’s development. In terms of educational
attainment, classic economic models assume that parents

6

make a rational decision by

considering both the costs of schooling and its future return when they decide whether to send
their children to school or not (Basu & Van, 1998; Blundell, Dearden, Meghir, & Sianesi, 1999;
Smits, 2007). From this perspective, children’s school participation may also be affected by the
economic characteristics of their community, as these characteristics can impact both the costs of
schooling and its expected returns. Costs of schooling can affect families’ schooling decisions in
two different forms, direct costs and opportunity costs. Direct costs include costs of school
materials, such as books, uniforms, supplementary tools, school fees, travel costs, etc. As
previously mentioned, both primary and secondary schools are free of charge, and school books
are distributed by government for free in Turkey. However, families may still have to deal with
some direct costs, such as unofficial school fees, uniforms, special educational materials, etc.,
and this can partly affect the schooling decision of families, especially poor ones.
More important than the direct costs of schooling, families also face important
opportunity costs as long as they send their children to school. In this regard, the community in
which families live is a very important factor in their schooling decisions for their children, as
communities differ in terms of the child labor market. To illustrate, a greater number of
6

It is known that schooling decisions of young children are generally made by their parents in
developing countries (Huisman & Smits, 2009a).
23

employment opportunities for young children in a community can increase the opportunity costs
of schooling, which may negatively influence the schooling decisions of children or their
families (Binder, 1999). If parents see that there are existing child-labor opportunities in their
community and the returns to education are relatively low, they may prefer having their children
work and contribute to the family budget, rather than sending them to school (Webbink, Smits, &
Jong, 2008). Therefore, poor families in rural areas, where children can be employed easily in
agricultural activities, may be less willing to send their children to school (Chudgar, 2006a).
Families in more modern areas, however, may not find many employment opportunities for their
children, even if they wish, as mechanization reduces the need for unskilled labor (Webbink, et
al., 2008).
Future earning is another important factor that can also affect children’s school
participation. According to human capital theory, education is a crucial investment that can
increase the income prospects of individuals (Becker, 1962). From this perspective, it can be
argued that parents compare the prospective returns to education with its current costs, including
both the direct and the opportunity costs, when they decide whether to send their children to
school or not, and they make a favorable decision if the prospective returns exceed its immediate
costs (Blundell, et al., 1999; Tansel, 2002). In this regard, the structure of the adult labor market
in the community may play a key role in families’ schooling decisions. If parents observe that
educated people earn more in their community, they will be more willing to educate their
children (Chudgar & Shafiq, 2010; Smits, 2007). However, if education does not substantially
increase job opportunities and economic welfare in their society, parents may not seek higher
levels of education for their children. Thus, in less industrialized societies, parents might be less
willing to send their children to school.

24

Similarly, it is argued that in less developed societies where capital can be still directly
transferred, via the family farm, etc., parents may invest less in their children’s education than
families in more developed societies where education is an important determinant of future
welfare (Smits, 2007). In addition, living in a less developed community can be an important
drawback for children’s development regardless of their family backgrounds. For example, poor
communities most probably have lower quality institutional infrastructures (schools, hospitals,
family support services, etc.) than affluent communities. These low-quality institutional
infrastructures in poor communities may not provide enough support for the development of
children in those communities. In addition, children who live in these communities may not have
direct access to or information about high quality education and better employment opportunities
(Galster, Marcotte, Mandell, Wolman, & Augustine, 2007). In this context, it can be argued that
families’ schooling decisions for their children can be influenced by general economic conditions
and employment opportunities in their communities (Gephart, 1997).
2.2.Empirical Evidence from Developing Countries
Place of residence (e.g., urban vs. rural) has been traditionally used in the literature in
order to analyze and predict the effect of social context on educational attainment, since urban or
rural residency is often highly related to communities’ socio-economic condition in developing
countries. In existing studies, place of residence has often been found among the biggest
predictors of children’s school participation in developing countries. King and Lillard (1983)
indicated that the probability of attaining higher levels of education was much lower for children
who lived in rural areas than it was for children who lived in urban and semi-urban areas in the
Philippines. Similarly, it was found that girls who lived in the rural areas of Pakistan had much
less chance to attend or complete primary school compared to their male peers as well as to girls
25

in urban areas (Lloyd, Mete, & Grant, 2007). A study in five Arab countries, Morocco, Algeria,
Tunisia, Egypt, and Syria, also showed that living in the countryside implied less schooling for
both boys and girls, with a higher impact on girls, compared to living in cities (Smits, 2007).
Huisman and Smits (2009a) also found in their study of 30 developing countries that living in a
rural area was negatively associated with girls’ school enrollment, although it was not
significantly related to boys’ schooling, after other factors were controlled for. In the case of
Turkey, the extant literature indicates that living in rural areas is among the most prominent
disadvantages in terms of the educational attainment of both boys and girls, with the greater
impact on girls (Duman, 2010; Hisarciklilar, McKay, & Wright, 2010; Smits & Gunduz-Hosgor,
2006; Tansel, 2002).
However, more recent studies, which particularly focus on the effects of social context on
educational attainment, have used more detailed variables than type of residence to analyze the
effects of both economic and socio-cultural community factors on educational attainment. At this
point, economic indicators such as the average income of community members, the available
labor market, the level of industrialization in the community, and socio-cultural indicators such
as average educational attainment, adult literacy ratios, and level of women’s empowerment in
the community, have been mentioned in the literature as important community-level factors
affecting educational attainment in developing countries. Existing studies generally use
nationally representative individual-level data sets and employ various statistical methods such
as ordinary least squares (OLS), ordered probit and logit models, hierarchical linear modeling
(HLM), etc., in order to investigate the relationship between community level variables and
educational attainment.

26

Given the context above, in the next section I review the recent literature on the effects of
both economic and socio-cultural community factors on educational attainment in developing
countries.
2.2.1. Economic Community Factors
The strong relationship between household-level poverty and lower educational
attainment is well documented in the literature. There is extensive research, both in Turkey and
in many other developing countries, which uniformly suggests that children from poor families
have significantly fewer chances to attain more education compared to children from more
affluent families (Anh, et al., 1998; Bakis, et al., 2009; Brown & Park, 2002; Filmer, 2000;
Hannum, 2003; Huisman & Smits, 2009a; Smits & Gunduz-Hosgor, 2006; Tansel, 2002; Tomul,
2008; UNESCO, 2010). Furthermore, it is argued that the concentration of poverty in a certain
community may negatively affect the educational attainment chances of children in that
community (Garner & Raudenbush, 1991) because of the lack of economic and social resources.
Although the relationship between the general economic condition of a community and the
school participation of children in that community has not yet been explored substantially,
several recent studies in developing countries encourage researchers to look at this relationship
more extensively. Community level per capita income, for example, was found to be
significantly associated with school enrollment in China, after controlling for other factors such
as gender, age, and family resources (Adams & Hannum, 2005). It was also found that districtlevel economic conditions, which included average per capita expenditure, average household
expenditure, and average per child expenditure on education, positively associated with girls’
school participation in India (Chudgar, 2006b). These findings have been also confirmed by
several similar studies conducted in developing countries (Binder, 1999; Brown & Park, 2002;

27

Connelly & Zheng, 2003; Hannum, 2003). Based on this evidence, it can be argued that children
from wealthier communities may have more chances to enroll in school compared to children
from poorer communities, even if they are coming from very similar families and have very
similar demographic characteristics. This is generally explained by the effects of communities on
returns to education via school quality or/and labor market structure (Brown & Park, 2002).
In the literature, the availability or the quality of physical infrastructures in the
community is also mentioned as an important determinant of educational attainment in
developing countries. According to Dreze and Kingdom (2001), for example, availability of a
post office and piped water in the community positively affects the school participation of
children. In addition, the degree of modernization of communities or regions, which is closely
related to the availability and the quality of physical infrastructures, was found to be a very
significant determinant for school participation of both boys and girls in many developing
countries (Buchmann & Brakewood, 2000; Dreze & Kingdon, 2001; Huisman & Smits, 2009a;
Smits, 2007). In order to explain the positive effect of modernization on children’s school
participation, it is argued that transportation to school would be easier and pressure on parents to
educate their children would be higher in more developed areas (Huisman & Smits, 2009a). It
should also be considered that these well-developed areas probably have better teachers and
school facilities, which may also influence the school participation of children.
In Turkey, Tansel (2002) investigated whether development, or modernization, of the
streets on which families live affects their children’s school participation or not. In this study,
urban streets were categorized as developed, undeveloped, or a squatter settlement. This
classification was done according to interviewers’ own observations of several factors. A
developed street, for example, had residences with higher rents, enjoyed a better transportation

28

system, and was relatively closer to the city center compared to an undeveloped street. A squatter
settlement was defined as a settlement that was built on public or other individuals’ land without
a construction permit. Results of the study showed that living on an undeveloped street or in a
squatter settlement negatively and strongly affected the children’s chance of middle school
attainment after controlling for individual- and household-level characteristics, while it did not
significantly influence primary schooling. In order to account for this result, the author noted that
generally the poorest migrants who did not have enough education to acquire well-paid jobs
lived on undeveloped streets and in squatter settlements. It was also discussed that middle and
high schools might be neither available nor close enough for children to access safely and easily
in these settlements.
The structure of the labor market in the community is also considered an important factor
affecting children’s and their families’ desire for education, and therefore children’s actual
school participation. This effect could either be negative because of child-labor activities, or it
could be positive because of well-paid job opportunities for more educated people (Buchmann &
Brakewood, 2000). Child labor is often found to be a significant factor in terms of limiting the
school participation of children in developing countries (Canagarajah & Coulombe, 1997;
Psacharopoulos, 1997). For example, Chudgar (2006b) found that the prevalence of child labor
in Indian districts is negatively associated with the school participation of both boys and girls in
those districts. It is further argued that the negative effects of child labor on school participation
might be higher in less developed communities because child-labor activities are more common
in these areas (Webbink, et al., 2008). Smits (2007), however, found that dropping out of school
is more common among older boys in the more developed regions of five Arab countries. The
author accounted for this finding by arguing that one of the possible reasons for this situation

29

could be the greater number of opportunities to earn money for these children in more developed
areas.
Beyond the effect of child labor, the labor market for adults can also affect children’s
school participation because of its impact on future earning opportunities. Increases in future
earning opportunities may motivate parents to invest more in their children’s education because
of their wish for their children’s future well-being and/or their expectations for future economic
assistance. For example, it was found that there was a significant positive relationship between
children’s school participation in Kenya and their parents’ anticipation of future economic
assistance (Buchmann, 2000). At this point, it may be assumed that the school participation of
children will be higher in more industrialized communities, where educated people could secure
higher earnings. There is not, however, enough empirical evidence to confirm this assumption.
Data from 30 developing countries showed that the proportion of the work force with a
white-collar job in the district was significantly associated with higher school-enrollment rates. It
was, however, found that this variable was no longer significant in multivariate analyses where
other factors were controlled for (Huisman & Smits, 2009b). Similarly, Smits (2007) found that
the prevalence of white-collar jobs in the region unexpectedly did not have any positive impact
on school participation in five Arab countries. Tansel (2002) suggested that higher rates of
industrial employment at the provincial level may result in higher educational attainment in
Turkey, although its effect was only significant for a few cases. There are also studies in
developing countries that suggest that the structure of the labor market can contribute to the
gender gap in school participation. Buchmann (2000, p. 1371) concluded her study of child
schooling in Kenya by noting that ―if parents perceive limited returns to girls' education due to
gender discrimination in the labor market, girls' school enrollment suffers.‖

30

2.2.2. Socio-cultural Community Factors
In addition to economic factors, several studies in developing countries suggest that
community-level socio-cultural factors may also influence children’s school participation, but
there have not been enough studies to substantially explore this relationship too. For example,
the literature is lacking in terms of exploring the effects of the average educational attainment of
adults and peers in the community on children’s school participation, although research from
both Turkey and other developing countries uniformly illustrates the strong positive relationship
between parents’ educational attainment and children’s school participation (Anh, et al., 1998;
Buchmann & Brakewood, 2000; Chudgar, 2009; Connelly & Zheng, 2003; Dreze & Kingdon,
2001; Handa, 1996; Huisman & Smits, 2009a; Sathar & Lloyd, 1994; Smits & Gunduz-Hosgor,
2006; Tansel, 2002; Tomul, 2008).
There are, however, several recent studies that have included adults’ average educational
attainment and the school participation of peer groups in the community as predictors of the
children’s school participation in developing countries. In China, Connely and Zheng (2003)
showed that the proportion of the neighborhood that was in school was positively associated with
children’s school enrollment. In their study, they found that living in a village that had a high
level of school enrollment significantly increased the schooling chance of both boys and girls,
with more effect on girls. Community-level adult literacy was also found to be positively
associated with higher school participation for children (Buchmann & Brakewood, 2000;
Chudgar, 2009). Similarly, the proportion of literate women and the percentage of the women
with higher education in the district were found to be positive indicators for school participation
for both boys and girls, with more impact on girls, in India (Chudgar, 2009).
Empowerment of women at the community level is also mentioned in the literature as an

31

important determinant of children’s welfare, including their school participation, in developing
countries (Webbink, et al., 2008). Chudgar (2006b) suggested that as an indicator of the overall
status of women within a community, the degree of women’s media exposure and women’s
autonomy in family decision making are positively associated with school participation of both
boys and girls, with higher impact on girls. The percentage of women of age 40–69 in the
district, as an indicator of women’s life conditions in society, was also found to be significantly
associated with girls’ school enrollment in a study of 30 developing countries (Huisman &
Smits, 2009a). In addition, the presence of women’s associations in villages was found to be
positively related to the current school enrollment and grade attainment of girls in rural India
(Dreze & Kingdon, 2001).
In this context, it can be argued that community-level socio-cultural factors may be more
important for girls’ schooling than for boys’ schooling in developing countries. Because
women’s roles are often specified as wife and mother in many societies, including Turkey, it is
believed that formal education is not necessary for women to perform their social roles (Bakis, et
al., 2009; Csapo, 1981).As another cultural obstacle for girls’ schooling in developing countries,
it has been argued that parents tend to attach more importance to the education of their sons than
their daughters in countries where sons are considered to be responsible for supporting their
parents in their old age (Huisman & Smits, 2009a; Schultz, 2002). Similarly, the marriage
tradition in society may impact parents’ decisions about investing in their daughters’ education.
Because girls tend to marry into the families of their husbands in many developing countries,
parents may be concerned that the returns to investment for the education of their daughters go to
the husband’s family. So, they may favor the education of their sons over the education of their
daughters (Smits & Gunduz-Hosgor, 2006). From a different perspective, Chudgar (2006b)

32

mentioned marriageability as a key indicator for families’ decisions about the education of their
daughters. She emphasized that in societies where women are expected to ―marry up,‖ getting
more education may not raise girls’ chances of marriageability. Due to the fact that it is very
important for women to be married in many societies, girls may have much fewer schooling
opportunities compared to boys if they are expected to marry someone who has more education.
In sum, previous studies regarding the effect of community factors on the school
participation of children suggest that there are significant relationships between community-level
economic factors, such as average per capita income and job market, and the school participation
of children in several developing countries. Similarly, existing studies have also found strong
associations between socio-cultural community factors, such as adult education, cultural norms,
and women’s empowerment, and children’s school participation in developing countries. The
research on the relationship between community-level factors and school participation, however,
is relatively new and the findings are not verified in different countries and with different data
sets.
2.3. Previous Studies in Turkey
In this section, I review the existing quantitative studies that concern the low levels of
educational attainment in Turkey. I introduce the variables and methods used in these studies and
summarize their core findings. Since existing studies mostly focus on the importance of
household characteristics for educational attainment, this review informed my study by exposing
the common household-level factors that have been found to be significant predictors of
educational attainment in Turkey. I therefore controlled for these household-level factors in my
study while investigating the relationship between community-level factors and the school
participation of children.
33

2.3.1. Data, Variables, and Methods Used
Although factors related to educational attainment have been extensively studied in many
developing countries during the last couple of decades, there were not many studies of this issue
until recently in Turkey. During the last 10 years, however, several studies that focus on the
educational attainment problem in Turkey have been conducted. These studies aimed to
determine the factors that are associated with educational attainment by using different
perspectives. While some studies primarily focused on the gender disparity in educational
attainment and factors that may cause it, others focused on the relationship between family
background characteristics such as parental education, sibling size, ethnicity, etc., and
educational attainment. Data used in these studies mostly came from the State Institute of
Statistics (SIS) of Turkey, while some studies used data from the Turkey Demographic and
Health Survey (TDHS) conducted by Hacettepe University’s Institute of Population Studies in
coordination with Macro International. Different statistical methods, such as OLS and probit and
logit models, were used in these studies depending on their primary interest and dependent
variables. While some studies used children’s school enrollment status (enrolled or not) at the
time of the surveys as a dependent variable, others used the final grade attainment of children to
determine their level of educational attainment. Detailed information about data, variables, and
methods used in these studies can be seen in Table A1 in the Appendix.
2.3.2. Core Findings
a) Household wealth
Household wealth is known to be a significant indicator of children’s school participation
in developing countries. It has often been found that there is a strong negative association
between household poverty and children’s school participation in many developing countries
34

(Filmer, 2000). Related studies in Turkey also confirm these findings. In general, it was found
that children in the wealthiest 20 percent of households had five times more chance to attend
higher education than their counterparts in the poorest 20 percent in Turkey (UNESCO, 2010).
Tansel (2002) also showed that per adult expenditures in the family were significantly associated
with children’s schooling at all levels (elementary, middle, and secondary). Studies that used
different measures of household wealth, such as wealth index, household income, etc., also found
a strong relationship between household wealth and educational attainment of both genders in
Turkey (Kirdar, 2009; Smits & Gunduz-Hosgor, 2006; Tomul, 2008). As discussed in earlier
sections, the negative influence of household poverty on children’s school participation can be
explained by its aggravating effect on the impacts of the direct and the opportunity costs of
schooling.
b) Parental education
It is a well-established argument that parental education is one of the most powerful
determinants of the educational participation of children in many developing countries
(UNESCO, 2010). In the case of Turkey, almost all of the existing studies indicate a strong
association between parental education and children’s schooling (Bakis, et al., 2009; Goksel,
2008; Hisarciklilar, 2002; Koc, 2008; Smits & Gunduz-Hosgor, 2006; Tansel, 2002; Tomul,
2008), in line with international research (Anh, et al., 1998; Chudgar, 2009; Connelly & Zheng,
2003; Dreze & Kingdon, 2001; Huisman & Smits, 2009a; King & Lillard, 1983). In general,
parental education may impact children’s education via different channels. For example, parents
who attain a higher level of education probably have more positive attitudes toward schooling. In
addition, these parents can be good role models for their own children in terms of educational
attainment. Parental education is also often highly correlated with other factors (e.g., household

35

wealth and household size) that may affect school participation of children (Koc, 2008).
c) Household size
Household size or number of siblings is another important factor affecting children’s
educational attainment in developing countries (Anh, et al., 1998; Huisman & Smits, 2009a;
Knodel,et al., 1990; Pong, 1997; Sudha, 1997). In Turkey, belonging to a large family was also
found to be a negative indicator of both boys’ and girls’ schooling (Hisarciklilar, 2002).
Similarly, Koc (2008) found that the number of siblings was negatively associated with
educational attainment for both genders, with more impact on girls. Being a member of a larger
family was also found to be among the significant factors that contributed educational attainment
inequality among girls (Duman, 2010). Bakis et al. (2009) also found that the number of siblings
was negatively associated with educational attainment, but it was only significant for girls at the
secondary level. The negative relationship between family size and the school participation of
children can also be explained by both the direct and the opportunity costs of schooling. It is
obvious that the direct costs of schooling would be more challenging for families with more
children. In addition, some of the children, generally older siblings, may have to work either at
home or outside to support their family and their siblings’ schooling.
d) Mother’s ability to speak Turkish
Speaking a minority language is known as one of the most significant obstacles for
children’s schooling in many developing countries, along with poverty and parental education
(UNESCO, 2010).In Turkey, it was found that girls who have a non-Turkish mother tongue are
very disadvantaged in terms of school participation. The current EFA report (UNESCO, 2010)
refers to these girls as one of the most marginalized groups in the world regarding educational
attainment. Smits and Gunduz-Hosgor (2006) also indicated that having a mother who could not
36

speak Turkish was a significant obstacle to schooling in Turkey, especially for girls. This may be
explained in several different ways. First, women who do not know the country’s official
language probably become more dependent on their husbands, have to accept more traditional
gender role attitudes, have relatively lower chance to get information about schooling, and lose
their bargaining power for their daughters’ education (Gündüz-Hosgör & Smits, 2007; Smits &
Gunduz-Hosgor, 2006). Second, most of the children whose mother cannot speak Turkish start
learning Turkish when they begin first grade and they are challenged with many hardships
because there are no special programs in schools for children who speak Turkish as a second
language. Hence, these challenges may force students to drop out of school earlier than they
desire. In fact, dropout rates were found to be much higher for ethnic non-Turkish children
compared to ethnic Turkish children, even at very early grades (Kirdar, 2009).
e) Gendered cultural values
Traditional cultural beliefs and attitudes may also be negatively associated with
children’s school participation. Traditional cultural values may limit children’s formal schooling,
especially in rural and more culturally traditional areas (UNESCO, 2010). This limitation could
be even larger for girls, since traditional gender roles and expectations in many societies mostly
assume that girls stay within the home and are responsible for things such as cooking, child
rearing, and other household chores (Bakis, et. al, 2009; Smits & Gunduz-Hosgor, 2006). At this
point, Smits and Gunduz-Hosgor (2006) found that when a mother agrees with the following
proposition, ―It is better for a male than for a female child to have education,‖ her daughter’s
secondary school enrollment chances decrease significantly in Turkey, after controlling for other
household-level variables such as parental education, family income, household size, etc.

37

2.4. Summary and Research Questions
Low participation in formal schooling is still a significant problem for many developing
countries, even at the primary school level. Turkey is also among the countries that struggle with
low school participation at both the primary and secondary levels. In order to ameliorate this
situation, it is vital to understand the factors that are related to school participation. Hence, an
extensive number of studies have been conducted in order to determine these factors by generally
focusing on household level variables in both Turkey and other developing countries. Most of
these studies have used nationally representative individual-level data sets and have employed
statistical analyses such as ordinary least squares (OLS), ordered probit and logit models, etc.
Results of existing studies suggest that several household-level factors, including level of
parental education, household income, family size, gendered cultural values, etc., are strongly
related to the school participation of children in Turkey, similar to many other developing
countries.
There is, however, not enough research that explores how the factors outside of the
household influence family schooling decisions in developing countries (Chudgar, 2006b). This
is also the same for research in Turkey. Several social and economic theories, however, suggest
that individuals’ behaviors and children’s development can be significantly influenced by the
social context in which they live. In addition, there are empirical studies from developing
countries that confirm these theories by showing significant relationship between communitylevel factors and children’s school participation, after controlling for individual- and householdlevel variables. In sum, both the theoretical and the empirical literatures suggest that children’s
school participation can be affected by community-level economic and socio-cultural factors,
such as availability and quality of infrastructures, general economic conditions, norms and values,

38

and level of adults’ educational attainments.
In this context, it is important to explore how different community-level factors are
associated with the school participation of children in Turkey, and how these associations vary
by region, in order to inform policy aimed at providing formal education to every child in every
part of the country. Exposing this relationship is important for helping policymakers develop a
systematic understanding of the relationship between the context in which children live and their
school participation in Turkey. In addition, this study can also inform educational policy in other
developing countries that suffer from the same problem. Turkey serves as a very suitable place
for such a study since school participation shortfalls are mostly concentrated in particular areas
that are often found to be the least developed parts of the country in terms of many socioeconomic variables.
In summary, my aim is to analyze the effects of community-level factors on the school
participation of children in Turkey by using the latest available nationally representative
individual-level data set. In addition to commonly used household-level variables (parental
education, household size, household poverty, etc.), community-level variables (average adult
education, level of poverty, etc.) will also be included in this study as independent variables.
Beyond investigating the relationship of community-level factors to school participation, I will
also descriptively depict the current picture of school participation in Turkey. In my descriptive
analyses, I will explore the percentage of in-school and out-of-school children by examining
children’s demographic characteristics, household factors, and their type of residence. I will also
present the most prominent characteristics of out-of-school children in my descriptive analyses.

39

Factors Affecting School Participation in Turkey
(Conceptual Framework)

7

Figure 2.1: Conceptual framework
Demographic Factors
 Gender
(Male +)
 Age ()
Household Factors





Parental education (+)
Household poverty ()
Household size ()
Mother’s ability to speak
Turkish (+)
 Mother’s traditional
gender role attitudes ()

Community Factors

Primary and
secondary
school
participation

 Level of adult education (+)
 Mothers’ traditional gender
role attitudes ()
 Level of poverty ()
 Type of residence (Urban )

Research Questions
1) What is the current pattern of school participation at the primary and secondary levels in
Turkey?


What is the percentage of out-of-school children by demographic characteristics (age
and gender), household-level factors (parental education, household wealth,
household size, etc.), and place of residence (urban/rural and regions)?

7

The (+/) signs assigned to variables are guided by existing findings in the literature and by
informed expectations based on these existing findings.
40



What are the most prominent characteristics of out-of-school children by region?



What is the relative importance of individual-level factors (demographic
characteristics of children and household factors) in school participation in terms of
sign and magnitude by region?

2) Are community-level factors significantly associated with school participation in Turkey,
after controlling for individual- and household-level factors?


What is the relative importance of economic and socio-cultural community factors for
school participation of children in Turkey?

41

CHAPTER 3
DATA AND METHODOLOGY

This chapter presents the data and methodology used in this study. It includes detailed
information on the variables and the model specifications that were used for the statistical
analyses in this study.
3.1.Data
3.1.1. Data Source
The study’s data come from the 2008 Turkey Demographic and Health Survey (TDHS2008), which was conducted by Hacettepe University Institute of Population Studies. The survey
is the ninth national demographic survey series and it is comparable to the worldwide standards
of the Demographic and Health Survey (MEASURE/DHS+) project. The survey contains a
variety of questions about households, including household assets, household size, demographic
characteristics and educational attainment of members, etc. (Hacettepe University Institute of
Population Studies, 2009). The data collected via the previous national demographic surveys
have been used for a variety of studies, especially in the health and education fields. Specifically,
TDHS-1993 and TDHS-1998 were used in several existing studies (e.g., Smits & GunduzHosgor, 2006; Kirdar, 2009; Dayıoglu, Kirdar, & Tansel, 2009) that aimed to determine factors
associated with school participation in Turkey. However, TDHS-2008 is the latest survey in this
series, and to my knowledge it has not been used in any previous study with a related aim.
Academic staff of the Hacettepe University Institute of Population Studies trained,
supervised, and coordinated the field staff, which included 19 teams with 8 people on each team.
Data collection was undertaken by these teams between the first week of October, 2008, and the
42

first week of the December, 2008. TDHS-2008 sample was selected by using a weighted,
multistage, stratified cluster sampling method in order to provide accurate estimates for
important demographic characteristics for various domains, such as Turkey as a whole, five main
regions, urban and rural areas, etc. The Address Based Population Registration System (ABPRS2007), a newly developed system that links each citizen to a specific address, was used to acquire
information about all settlements of Turkey. Based on the settlement lists created using ABPRS2007, a systematic random sample was selected with probability proportional to size (PPS).
Although 13,521 households were initially selected for the survey, only 11,911 of them were
considered as available for data collection, because others were not occupied at the time. The
survey was successfully administrated to 10,525 households in all 81 of Turkey’s provinces, and
it included information about more than 40,000 individuals. Thus, the overall response rate was
88.4 percent, with a higher response rate for rural areas (95 percent) and a lower response rate
for urban areas (86.1 percent). As part of TDHS-2008, 8,003 ever married women (15–49 years)
from 10,525 households were further selected for the women’s survey, which included more
detailed information about their children and themselves, such as family planning, fertility, and
infant and child mortality, etc. The total number of women who were successfully surveyed was
7,405, with a 92.5 percent response rate (Hacettepe University Institute of Population Studies,
2009).
3.1.2. Sample Selection Decisions and Representativeness
As seen in Table A2 in the Appendix, TDHS-2008 data shows that percentages of out-ofschool children are relatively high for children age 6 (30.6 percent) and 7 (5 percent). It can be
interpreted that some of these children may turn six after the beginning of the school year, so
they could not enroll (Hacettepe University Institute of Population Studies, 2009). Also, late
43

enrollment, which is very common especially in the countryside of Turkey, may be another
reason for these high un-enrollment ratios. Since most of these children will eventually attend
school, they should not be considered as out-of-school children in statistical analyses. Thus, I
kept only children older than 8 years old for my further analyses, similar to several previous
studies (e.g., Kirdar, 2009; Smits & Gunduz-Hosgor, 2006). I chose 17 as an upper age boundary
because it is the age at which most children complete their secondary-level education in Turkey.
In my final sample, I only included children of household heads and children whose mothers
participated in the women’s survey, because I intended to use some information from the
women’s survey, such as mothers’ Turkish speaking ability and their gender role attitudes, as
independent variables. While individual-level variables were created from this new sample,
community-level variables were created from the original data set and were incorporated into the
new sample in order to create a more comprehensive representation of the community.
It should also be noted here that my final sample does not represent the children whose
mothers were older than 50 at the time of data collection, because only 15–49-year-old women
were included in the TDHS-2008 women’s survey. However, the chance of having a mother
older than 50 is not very high for 8–17 year old children in Turkey. For example, a 50-year-old
woman should have given birth when she was 33 years old to have children in this age group,
and this age goes up for women older than 50. In Turkey, around 7 percent of all births were
given by women 35 or older and only around 1.5 percent of all children born belong to women
40 or older during the 1990s (Kirdar, 2009). Based on this information, the total number of 8–17year-old children who are missing in the final sample is probably under 5 percent. In addition,
TDHS-2008 data is weighted, and the data set includes a weight variable. Thus, I applied the
sample weights to ensure representative estimates.

44

3.1.3. Definition of “Community”
In the TDHS-2008 data, individuals are grouped in households, and households are
grouped in clusters. Small living areas that include approximately 100 households are defined as
clusters. A total of 634 clusters were selected for data collection. While 400 of the selected
clusters are located in urban areas, which have a population of at least 10,000 people, 234
clusters are located in rural areas, where the population of residents is less than 10,000. Twentyfive households from each cluster in a standard urban settlement and 15 households from each
cluster in a standard rural settlement were selected for the survey in order to have an accurate
representation of the clusters within each region (Hacettepe University Institute of Population
Studies, 2009). In the Turkish context, people who live in the same cluster probably have very
similar socio-economic characteristics, use the same external resources, and socially interact
with each other. I therefore defined each cluster as a community for this study because
communities are often described as small living areas where residents have various social
interactions and share common external institutions such as schools and hospitals (Binder, 1999).
Therefore, this data set matches the study’s objectives well, as this study aimed to explore the
current pattern of school participation, and to analyze the relationship between community
characteristics and children’s school participation in Turkey.
3.2.Variables
3.2.1. Dependent Variable
School participation: The dependent variable for this study was school participation,
which is a dichotomous variable indicating whether the child attended school during the school
year when the survey was conducted or not (1 = attended, 0 = not attended).
45

3.2.2. Independent Variables
Within the multilevel structure of my data, I considered both children’s demographic
characteristics (age and gender) and household-level factors (parental education, household
wealth, household size, etc.) as one level and community characteristics as another level in order
to make the analyses more manageable. When data has a multilevel structure, as in this study,
elementary units (individuals in this study) are referred to as level-1units, and the clusters
(communities in this study) are referred to as level-2 units (Skrondal & Rabe-Hesketh, 2004). In
other words, level-1 units are grouped within level-2 units (Smith, 2011), as individuals are
grouped in communities in the case of the present study. Thus, there are two groups of
independent variables in this study: those that correspond to students’ demographic and
household characteristics (individual level) and those that correspond to their community
(community level).
a) Individual-level variables
At the individual level, in addition to children’s demographic characteristics (age and
gender), household-level factors, which are often found to be significantly associated with
children’s school participation in Turkey, are included in the study.
Demographic characteristics of children: These variables include the child’s gender and
age. While gender is a dichotomous variable [female (1) and male (0)], age is a continuous
variable, ranging from 8 to 17. In Turkey, it is known that there is a big gender disparity in
school participation, especially at the secondary level, so I expected to find that the school
participation of female students is significantly lower. In terms of age, I expected to find a
negative relationship between age and school participation since many children may drop out
when they get older due to various socio-economic reasons, such as marriage, work, cultural
46

restrictions, etc.
Household poverty: TheTDHS-2008 data set does not include information about
households’ income or expenditures; rather, it includes wealth quintiles (poorest, poorer, middle,
richer, and richest) as indicators of families’ economic condition. These wealth quintiles were
generated based on the wealth index, which was constructed by using data on possession of
consumer goods, dwelling and household characteristics, and assets. This wealth index is known
as a good indicator of household wealth, because it is consistent with household income and
expenditures (Rutstein, 1999, as cited in Hacettepe University Institute of Population Studies,
2009). As mentioned in the literature review chapter, empirical studies have shown that
household poverty is one of the most significant limitations on school participation of children in
many developing countries, including Turkey. Thus, in order to control for household poverty, I
created a binary variable, with (1) for a child who lives in a poor household (one of the lowest
two wealth quintiles; poorest and poorer) and (0) otherwise. Based on the findings of previous
literature, I expected to find a negative relationship between household poverty and the school
participation of children.
Household heads’ education: The international literature uniformly highlights the
importance of parental education for a child’s school participation. It is also well documented in
the literature that the mother’s and father’s levels of education are generally highly correlated.
Therefore, I included a variable that indicated the household head’s level of education in order to
control for the effect of parental education. According to the data, around 65 percent of
household heads in Turkey have lower than second-level primary school (equivalent to middle
school) education. Thus, I created two categories to indicate the level of education that
household heads attained: (0) lower than second-level primary and (1) second-level primary or

47

higher. I expected to find that the household head’s level of education was positively associated
with children’s schooling, in line with previous literature.
Mothers’ Turkish speaking ability: The mother’s ability to speak Turkish was also
included as an independent variable in this study since previous studies have indicated that it is
significantly related to school participation in Turkey, even after controlling for mothers’ level of
education. This variable was a dichotomous variable, (1) if the mother could speak Turkish and
(0) otherwise. As Smits and Gunduz-Hosgor (2006) found, I expected to find a positive
relationship between mothers' ability to speak Turkish and the school participation of their
children, especially for younger children.
Household size: Household size (numbers of household members) is often found to be
significantly associated with school participation in both Turkey and other developing countries.
Thus, I also included it as an independent variable in this study. This variable was a continuous
variable ranging from 2 to 17. I expected to find a negative relationship between household size
and school participation, as previous research suggests.
Mothers’ traditional gender role attitudes: In order to control for gendered cultural
values, mothers’ responses to a survey question were included as an independent variable in this
study. This question was: ―It is better to educate a son than a daughter.‖ This variable was
dichotomous, coded (1) when women agreed with this proposition, and (0) when they disagreed.
It was expected that this variable would be negatively associated with school participation, since
this variable represents more traditional gender role attitudes at the household level.
b) Community-level variables
Level of adult education in the community: This variable was created and included in this
study in order to explore the relationship between community-level adult education and

48

children’s school participation, net of demographic and household level variables. This variable
was defined as the proportion of adults who completed at least second-level primary school in
the community. In order to construct this variable, the number of adults (18 years old and older)
who completed at least second-level primary school in each community was divided by the total
number of adults in the same community. Since this variable represented the availability of
educated adult role models in the community, I expected to find a positive relationship between
this variable and the school participation of children.
Level of poverty in the community: This variable was constructed and included in this
study in order to investigate the relationship between school participation of a child and the
general economic condition of her/his community. In the TDHS-2008 data set, the economic
condition of each household is qualified with one of five wealth quintiles (poorest, poorer,
middle, richer, and richest), as mentioned above where the household poverty variable is defined.
This variable represents the proportion of households that were in one of the lowest two wealth
quintiles (poorer or poorest) for each community. This variable could be viewed as an indicator
of the quality of the general infrastructure and the schools in the community, the prevalence of
child labor, the family’s ability to borrow money from neighbors, etc. I therefore expected to find
a negative relationship between this variable and the school participation of children.
Gender role attitudes in the community: Acceptance of traditional gender roles in a
community was measured by women’s response to an interview question (―It is better to educate
a son than a daughter‖), which was also used at the individual level. The proportion of women
who agreed with this statement was constructed for each community. Based on the previous
literature and the theoretical framework discussed in the previous chapter, I expected to find a
negative relationship between this variable and school participation.

49

Urbanization: In the TDHS-2008, settlements with a population of at least 10,000 were
defined as urban, while settlements with less than 10,000 were defined as rural. As is often found
in both Turkey and other developing countries, I expected to find a positive relationship between
living in an urban settlement and school participation, because urban settlements in developing
countries generally have better community infrastructures, school facilities, future job
opportunities, etc.
3.3.Correlations among Variables
Inter-correlations between independent variables at both levels were estimated before
conducting statistical analyses in order to avoid multicollinearity between variables. I conducted
both ―Pearson’s correlation‖ and ―Kendall’s tau b‖ tests, since many of my variables were binary.
At level-1, inter-correlations between independent variables were not very strong (ranging from
0 to 0.47), indicating a limited possible multicollinearity problem. At the second level, however,
stronger correlations were found between independent variables. Specifically, a very high
correlation (0.82) between average adult education and community poverty was determined (see
Table A7 and Table A8 for the correlation between all level-2 variables). Thus, I estimated two
different models, each of which included either average adult education or community poverty at
the second level, when I included level-2 variables in my analyses, and I report the results of
both models.
3.4. Empirical Strategy
Empirical analyses in this study included two parts. I start by providing detailed
descriptive analyses, followed by more extensive modeling and statistical analyses of factors
associated with children’s school participation in Turkey.

50

In the first part, I illustrated the school participation ratios of children by different groups,
such as gender, age, household head’s education, household wealth, household size, mother’s
Turkish speaking ability, type of residence (urban/rural), and regions. I then explored the
relationship between my individual-level variables (demographic characteristics of children and
household level factors) and school participation by using the binary logit model. I also
conducted the same binary logit model for different regions in order to see the regional
differences in this relationship.
In the second part, I employed multilevel analyses by using hierarchical linear modeling
(HLM)to investigate the significance of the community-level variables, which I primarily
focused on in this study.Below, I explain the binary logit model and HLM approaches, and I
discuss the models’ specifications and their applications in this study.
3.4.1. Binary Logit Model
The binary logit modeling approach is known as a more appropriate regression method to
use when the dependent variable is binary, compared to the linear probability modeling (LPM)
approach, because of several issues. First of all, since the dependent variable can only get values
0 and 1, error terms cannot be normally distributed, which is an important assumption of LPM.
In addition, when we use a LPM with a binary outcome, it is possible to get predictions either
less than zero or greater than one, which do not make sense because probabilities must be
between zero and one. More importantly, it is unrealistic to assume a linear relationship between
probability and all possible values of independent variables. Using LPM with binary dependent
variables thus has some significant drawbacks, but these drawbacks can be overcome by using a
binary logit model (Long, 1997; Wooldridge, 2003), where instead of directly modeling the 0-1
outcome, we model the chance of being 1, which can take any value from 0 to 1 and is mapped

51

into all possible values via the logit link.
In this context, I used a binary logit model to investigate the relationship between my
outcome variable (school participation) and my level-1 variables (age, gender, household head’s
education, household poverty, mother’s gender role attitudes, household size, and mother’s
Turkish speaking ability),since my dependent variable, school participation, was binary:

Logit (Pij) = Log [ Pij / (1-Pij)] = pij(1)

Full logit model:

Logit (Pij) = β0j +β1jFEMALE +β2jAGE +β3jHH_ED + β4jHH_POVERTY
+β5j HH_SIZE+β6j GEN_RL +β7jTURKISH + rij(2)

In this model, [Pij / (1-Pij)] is the odds of being in school compared to being out of school
for a child i in community j (Tranmer & Elliot, 2008). The results of the estimated effects of
independent variables were interpreted in terms of the odds ratio. While an effect on the odds
ratio greater than one represents a positive relationship between school participation and the
given variable, an effect on the odds ratio that is smaller than one indicates a negative
relationship.
3.4.2. Hierarchical Linear Modeling
When I took community-level variables into account, I used HLM to address several
issues that were derived from the multilevel nature of the data. First, individuals who belonged to
the same community may have had more commonalities. Thus, this situation can result in

52

dependencies among observations and can violate the common classical statistics assumption
that observations are independently distributed (De Leeuw & Meijer, 2008). In addition,
variation of observations may substantially change across different communities, so withingroup variation should be taken into account. HLM therefore was an appropriate method to use
in this study, since it allows intercepts and slopes to vary by groups (Raudenbush & Bryk, 2002;
Skrondal & Rabe-Hesketh, 2004). In this method, I first used an unconditional model (a model
with a random intercept only) in order to investigate if there was significant variation in the
intercept across communities:

Level 1: Yij= β0j+ rij
Level-2: β0j = γ00 +u0j

(3)

Where
Yij

is the dependent variable for each individual in a community,

β0j

is the mean outcome for community j,

rij

is a level-1 random error that is assumed to be normally distributed with a mean of zero
and constant variance,σ2,

γ00

is the average of community means or a grand mean and a random error,

u0j

is the random effect associated with community j, and it is assumed that u0j is normally
distributed with a mean of zero and a variance, τ00.

I also calculated the intra-class correlation (ICC),

which is the proportion of level-2

variance to the total variance. Thus, denotes the importance of level-2 variables in explaining
the variance in the outcome variable (Raudenbush & Bryk, 2002):

53

⁄

⁄

Finally, I used two-level HLM to investigate the relationship between school
participation and community-level factors, after controlling for both children’s demographic
characteristics (age and gender) and household-level factors (household head’s education,
household wealth, household size, etc.).
At level-1, for child i in community j,

pij= β0j +β1jFEMALE +β2jAGE +β3jHH_ED + β4jHH_POVERTY
+β5j HH_SIZE+β6j GEN_RL +β7jTURKISH + rij(5)

In this equation, for a child i in community j, pij is a function of child’s gender
(FEMALE), age (AGE), household head’s education (HH_ED), household’s poverty
(HH_POVERTY), household’s size (HH_SIZE), mother’s gender role attitudes (GEN_RL), and
mother’s Turkish speaking ability (TURKISH). β0j is an intercept for community j and rij is
random error, while β1j,β2j,…., β7j are coefficients for each level-1 variable, which varies across
different communities, and the differences are random and controlled by one dispersion
parameter.
At level-2,

β0j = γ00 +γ01CADULT_ED + γ02CGEN_RL + γ03URBAN + u0j(6)

54

In this equation, intercept β0j is a function of community-level variables; adult education
(CADULT_ED), gender role attitudes (CGEN_RL), and urbanization (URBAN). While γ00 is a
constant, which represents the grand mean, and u0j is a random effect that captures the variations
caused by the unobserved community factors, γ01, γ02, γ03 are the coefficients associated with each
level-2 variables. In the application of HLM, I allowed the remaining level-1 coefficients to stay
fixed at level-2. As mentioned in the previous section, I also estimated the same model with level
of poverty (CPOVERTY) at the second level instead of adult education (CADULT_ED),
because of the high correlations between these two variables.

55

CHAPTER 4
DESCRIPTIVE AND PRELIMINARY ANALYSES

In this section, I intend to answer my first research question (What is the current pattern
of school participation at the primary and secondary levels in Turkey?) by descriptively
depicting the school participation pattern in Turkey. The results of descriptive analyses include
the means and standard deviations for all of the variables used in this study, and the percentage
of out-of-school and in-school children by different groups, including all level-1 variables, type
of residence (urban/rural), and regions. Chi-square tests were applied to investigate the
significance of the relationship between school participation and different groups for each
independent variable. I also identified the most prominent characteristics of out-of school
children in Turkey by presenting the percentages of the out-of-school children who shared the
same characteristics. Finally, before I started multilevel statistical analyses, I employed binary
logistic regression analyses for more advanced investigation of the relationship between
individual-level independent variables and school participation. All statistical analyses presented
in this chapter were conducted in SPSS version 17.
4.1. Descriptive Statistics
This study used multilevel data from the 2008 round of the Turkey Demographic and
Health Survey. Descriptive statistics for all dependent and independent variables used in this
study are summarized in Table 4.1.In addition, Table 4.2presents the means and standard
deviations for each independent variable by the school participation status of children. Recall
that sample weight has been applied to data before computing the descriptive statistics.

56

Table 4.1: List of variables and descriptive statistics
Min
Max
Dependent Variable
School Participation*
Level-1 Independent Variables
Household Head’s Education*
Mother Speak Turkish*
Age of Child
Gender*
Household Poverty*
Number of Household Members
Mother’s Gender Role Attitudes*
Level-2 Independent Variables
Average Adult Education
Community Poverty
Average Gender Role Attitudes
Type of Residence (Urban/Rural)*
Valid N (listwise) = 6,640
* Variables are binary

Mean

Std. Deviation

N

0

1

.88

.33

6,667

0
0
8
0
0
2
0

1
1
17
1
1
17
1

.34
.94
12.37
.50
0.43
5.66
.14

.48
.24
2.86
.50
0.50
2.07
.35

6,675
6,675
6,675
6,675
6,675
6,675
6,647

0
0
0
0

0.94
1
0.5
1

0.39
0.44
0.12
0.65

0.21
0.37
0.12
0.48

613
613
613
613

57

Table 4.2: School participation by level-1 and level-2 variables, mean and S.D
Out-of-school
In-school
Total
Level-1 Variables (Individual and family
variables)
Number of Household Members
6.64
5.52
5.66
(2.44)
(1.97)
(2.07)
Age of Child
15.30
11.95
12.37
(1.73)
(2.74)
(2.86)
Household Poverty
0.66
0.40
0.43
(0.47)
(0.49)
(0.50)
Household Head’s Education
0.15
0.37
0.34
(0.36)
(0.48)
(0.48)
Gender (female=1)
0.59
0.48
0.50
(0.49)
(0.50)
(0.50)
Mother Speak Turkish
0.86
0.95
0.94
(0.35)
(0.22)
(0.24)
Mother’s Gender Role Attitudes*
0.23
0.13
0.14
(0.42)
(0.34)
(0.35)
Level-2 Variables (Community
variables)
Average Adult Education
0.27
0.39
0.37
(0.15)
(0.19)
(0.19)
Community Poverty
0.59
0.40
0.43
(0.36)
(0.36)
(0.37)
Average Gender Role Attitudes*
0.18
0.13
0.13
(0.12)
(012)
(0.12)
Urban Residence
0.61
0.75
0.74
(0.49)
(0.43)
(0.44)
* Higher means represents more traditional gender role attitudes.
4.2. Out-of-school Children by Different Groups
Table 4.3 presents the percentage of in-school and out-of-school children by different
groups, including all level-1 variables, type of residence (urban/rural), and regions. The results of
chi-square tests indicate that for all of these variables, there were significant differences between
groups in terms of the school participation of children. As seen in Table 4.3, children from larger

58

households, children from poorer households, older children, females, children from households
where the head had lower than middle school education, children from the East region, children
whose mothers could not speak Turkish, children whose mothers had more traditional gender
role attitudes, and children from rural areas were significantly disadvantaged compared to their
counterparts in terms of school participation.
According to the results of the descriptive analyses, it is clear that school participation at
both the primary and secondary levels is still a serious problem in Turkey, with a much more
dramatic form at the secondary level as the rate of school participation is much lower for
children age 14 years old and older. Just by looking at the descriptive statistics, it can also be
argued that household poverty and household head’s education stand out as major factors for
children’s school participation, although all independent variables seem to be systematically
related to school participation. To illustrate, almost one-fourth of children in the poorest
households were out of school compared to only 4 percent in the richest households. In terms of
the household head’s education, the percentage of out-of-school children (16 percent) in
households where the household head had lower than middle school education was almost three
times the percentage of out-of-school children (5.4 percent) in households where the household
head had at least a middle school education.

59

Table 4.3: Percentage of out-of-school children by different groups
Out-of-school (%)
In-school (%)
Number of Household Member*
Less than Six
7.9
92.1
Six and More
18.8
81.2
Age of Children*
8-10 years
1.0
99.0
11-13 years
3.7
96.3
14-17 years
29.0
71.0
Gender*
Female
14.7
85.3
Male
10.1
89.9
Wealth Quintile*
Poorest
23.1
76.9
Poorer
14.2
85.8
Middle
12.0
88.0
Richer
6.2
93.8
Richest
4.1
95.9
Household Head’s Education*
Lower than Middle School
16
84
Middle School or Higher
5.4
94.6
Mother Speak Turkish*
Yes
11.3
88.7
No
29.3
70.7
Mothers’ Gender Role Attitudes*
Negative
19.9
80.1
Positive
11.1
88.9
Regions*
West
11.5
88.5
South
11.1
88.9
Central
9.6
90.4
North
8.9
91.1
East
17.4
82.6
Type of Residence*
Urban
10.3
89.7
Rural
18.3
81.7
Total
12.4
87.6
* Chi-Square tests indicate significant differences between groups (p≤ 0.01).
60

Table 4.3 also shows the relationship between the school participation of children and
their place of residence, regarding both types of residence (urban/rural) and region. In terms of
the type of residence, the ratio of out-of-school children was around 10% in urban areas, while it
was significantly higher (18.3%) in rural settlements. Therefore, it can be interpreted that in
Turkey the chance of being in school is higher for the children who live in urban areas compared
to children who live in rural areas. Based on the numbers presented in Table 4.3, it can also be
argued that children in the East region were significantly disadvantaged in terms of school
participation. To illustrate, the percentage of out-of-school children in the East region (17.4%)
was almost two times of the percentage of out-of-school children in the North region (8.9%).
This result was somewhat expected as the East is the least developed region of Turkey in terms
of many socio-economic variables. It was unexpected, however, to find that the country’s two
most developed regions, the West and the South, also experienced relatively lower school
participation. One possible reason for this pattern could be the growing trend of immigration
from less developed parts of the country to the urban centers of these two regions over the years.
In order to check the appropriateness of this interpretation, I descriptively analyzed how type of
residence (urban/rural) interacted with regions in regard to children’s school participation status.

61

Table 4.4: Regional differences in school participation by type of residence
Region
Type of Residence
Out-of-school (%)
In-school (%)
West
Rural
10.9
89.1
Urban
11.6
88.4
South
Rural

13.2

86.8

Urban

10.4

89.6

Rural

17.2

82.8

Urban

6.7

93.3

Rural
Urban

17.5
3.6

82.5
96.4

Rural
Urban

25.1
12.2

74.9
87.8

Central

North

East

According to Table 4.4, the West is the only region where the proportion of out-of-school
children in urban areas was higher than the proportion of out-of-school children in rural areas.
Interestingly, the percentage of out-of-school children in urban areas of the West was almost the
same as the percentage of out-of-school children in urban areas of the East. Similar to the West,
the South also had a much higher out-of-school children percentage in urban areas compared to
the North and the Central. Hence, it seems that the school participation problems in the West and
the South are quite different than the rest of the country. As mentioned above, children of
immigrant families, who often live in the suburbs of metropolitan cities, could possibly be a
reason for this pattern. In contrast, out-of-school children in the North and the Central are mostly
concentrated in rural areas. This could be explained by socio-economic underdevelopment,
geographic hardships, agricultural employment, and unavailability of schools, especially at the
62

secondary level, in the rural areas of these regions. As expected, the highest percentages of outof-school children in both urban and rural areas were found in the East. Specifically, children
living in the rural parts of the East region seem to be the most marginalized groups in terms of
their access to education, because around a quarter of them do not attend school.
4.3. Characteristics of Out-of-school Children
In order to identify better the out-of-school children in Turkey, the common
characteristics of out-of-school children should be identified. Understanding the common
characteristics of these children may help to determine the areas that should be primarily focused
on in order to raise the overall school participation in the country. For example, although
focusing on children whose mothers cannot speak Turkish could be a helpful strategy to increase
school participation in certain areas or for certain groups, it cannot be an effective way to raise
the overall school participation across the country since children whose mothers cannot speak
Turkish are mostly concentrated in certain areas, and their numbers are relatively small (14.2
percent) among all out-of-school children. In addition, based on the numbers presented in Table
4.3, it can be argued that the policies intended to increase school participation in Turkey should
pay special attention to the children who live in the East region. Although this interpretation is
reasonable, it should also be noted that around one-third of all out-of-school children in Turkey
live in the West region. Thus, there is also an immediate need for specific policies that should
focus on the school participation problem in the West region. I therefore present the most
prominent characteristics of out-of-school children in a percentage form in Table 4.5. In order to
identify how characteristics of out-of-children differ by regions, I also present the same
information separately for each region.

63

Table 4.5: Selected important attributes of out-of-school children (%)
Characteristics
Total
West South Central
Age (14–17)
88.1
92.4 89.6
93.4
HH Head’s ed. lower than middle
84.9
80.5 88.8
78.4
school
In poor households*
66.0
42.0 79.2
54.8

North
91.9
83.1

East
80.5
91.1

81.7

89.0

Household size (Six and more)

62.4

47.4

61.3

38.5

60.2

89.2

Live in urban settlements

61.1

85.9

68.5

50.6

24.7

42.1

Female

58.7

53.1

63.1

59.5

44.7

64.2

Live in the West

34.7

-

-

-

-

-

Live in the East

34.0

-

-

-

-

-

Mother accepts traditional gender role

22.6

22.7

27.7

22.6

17.6

21.4

Mother cannot speak Turkish

14.2

0

4.6

0

0

40.2

* Include the lowest and the second lowest poverty quintiles.

According to Table 4.5, 88.1 percent of all out-of-school children in Turkey are between
14 and 17 years old, pointing to the dramatic school participation problem at the secondary
school level. Also, most (84.9 percent) of the out-of-school children live in households where the
household head has lower than a middle school education. Similarly, a significant number of
children who are out of school are from poor (65.9 percent) and large (six and more members)
(62.4 percent) households. Although descriptive analyses show that the percentage of out-ofschool children is significantly higher in rural areas compared to urban areas, indeed more than
half of the out-of-school children live in urban areas of the country. This is mostly because of the
high population of urban areas in Turkey. Hence, the school participation problem in urban areas
should not be ignored. It is also important to highlight here that the West and the East together
include more than two-thirds of all out-of-school children in Turkey.
When the characteristics of out-of-school children in different regions were investigated,
several important differences emerged. First of all, children aged 14–17 years constitute around
90 percent of all out-of-school children in four of the five regions. This means that school
64

participation is not a big issue for younger children in these regions. However, almost 20 percent
of all out-of-school children were younger than 14 years old in the East region. Given the fact
that the East had the highest proportion of out-of-school children, school participation for
younger children is still an important problem for this region. Another notable finding is that a
very high proportion (40.2 percent) of out-of-school children in the East region had a mother
who could not speak Turkish. In line with expectations, the East region also had the highest
proportion of female out-of-school children among all regions. In terms of the type of residence,
it was found that the significant proportion of the out-of-school children lived in urban areas in
the West (85.9 percent) and the South (68.5 percent) regions. This could be explained by the
high level of urbanization in these two regions. In contrast, only one fourth of out-of-school
children lived in urban settlements in the North region. It is important to note here that because
of the geographic constrains and pervasive agricultural works, a significant number of people
live in rural areas in this region. These reasons may also contribute to the high number of
children who cannot attend school in rural areas in this region, as discussed later in this chapter.
4.4. Results of Binary Logit Analysis
In this section, I introduce the results from the binary logistic regression analyses, where
the dependent variable ―school participation‖ is a dichotomous variable having the value 1 if the
child is in school and 0 otherwise. These analyses allowed me to conduct a more advanced
investigation of the relationship between the individual-level independent variables and school
participation. Since this study primarily focused on the role of community-level variables,
individual-level independent variables became secondary in the HLM analyses. I therefore
intended first to find out which of the individual-level variables were significantly associated
with the school participation of children before I started taking community-level variables into
65

account by conducting HLM analyses. I also conducted binary logistic regression analyses
separately for different age groups and for different regions in order to see how the effects of the
individual-level independent variables on school participation differed for children in different
age groups and in different regions. The results of the binary logistic regression analyses are
reported in the form of odds ratios for the accessibility and the convenience of the interpretations.
Table 4.6: Results of binary logit analysis for whole sample
Exp (B)
(n= 6,640)
Gender (female)
.61**
Age

.53**

Mother speaks Turkish

1.75**

Household head’s education

2.20**

Mother accepts traditional gender role

.76**

Household poverty

.41**

Number of household members

.88**

** p≤ 0.01
Before starting interpretations of the results shown in Table 4.6, it is important to note
that the model chi-square for the binary logistic model is significant (p≤ 0.01), indicating that the
overall model significantly improved over the null model (see Table A4 in Appendix).
According to the results of the analysis, there are statistically significant (p≤ 0.01) relationships
between all of the individual-level independent variables and the school participation of children,
as seen in Table 4.6.As explained in the methods section, an odds ratio greater than one indicates
a positive relationship between school participation and the given variable, while an odds ratio
that is smaller than one is a sign of a negative relationship. Based on the odds ratios presented in
Table 4.6, it can be stated that the mother’s ability to speak Turkish and the household head’s
education were positively associated with school participation, while being female, being older,
66

the mother’s traditional gender role attitudes, household poverty, and residing in a large
household were negatively associated with school participation. All of these associations are in
line with most of the previous Turkish and international research and with my expectations in
terms of both their significance and the sign of the association.
As mentioned earlier, one of the most important advantages of reporting the odds ratio is
its convenience of interpretation. The odds ratio enables us to interpret the likelihood of having
the value of 0 or 1 for each individual case regarding different independent variables. In the case
of my results, for example, the odds of being in school for a girl is 0.39 (1-0.61) times (or 39
percent) less than the odds of being in school for a boy, holding all other factors constant.
Likewise, the odds of being in school decrease by 0.47 times with each one-year increase in the
age of a child, while having one more household member decreases the odds of being in school
by 0.12 times. It was also found that the mother’s acceptance of traditional gender roles
decreased the child’s school participation chance by 0.24 times. According to the odds ratio for
the mothers’ ability to speak Turkish, it can be stated that a child whose mother could speak
Turkish was 75 percent (1.75-1) more likely to be in school compared to a child whose mother
could not speak Turkish. Similarly, a child’s chance of being in school was 120 percent higher in
a household where the household head had at least a middle school education. Finally, children
who lived in poor households (households in the lowest two wealth quintiles) had 59 percent less
chance to be in school compared to their counterparts.
4.4.1. Employing the Binary Logit Model for Different Age Groups
I also analyzed the data separately by different age groups in order to see if the
relationship between each individual-level independent variable and the school participation
status of children differed by age group. I first divided the sample into three age groups, 8–10
67

years, 11–13 years, and 14–17 years, with each group roughly consisting of one-third of my final
sample. It was assumed that these age groups represented the children who were supposed to be
in the first level of primary school (1–5 grades), second level of primary school (6–8 grades), and
secondary school, respectively. I analyzed the same binary logit model separately for these three
age groups. Since the results of the analyses for the first two groups looked very similar and
there were not many out-of-school children in the first group (see Table 4.3), I merged these two
groups and created a new age group (8–13 years), then employed the binary logit model. Table
4.7 presents the results of the binary logistic regression analyses by two age groups (8–13 years
and 14–17 years) in terms of the odds ratios.
Table 4.7: Binary logit analysis by age groups
8–13 years
Exp (B)
(n=4,204)
Gender (female)
.26**

14–17 years
Exp (B)
(n=2,436)
.72**

Age

.60**

.66**

Mother speaks Turkish

2.13**

1.57**

Household head’s education

.86**

2.54**

Mother accepts traditional gender role

.72**

.78**

Household poverty

.30**

.43**

Number of household members

.82**

.89**

** p≤ 0.01
According to the results presented in Table 4.7, gender was much more strongly
associated with the school participation of children in the first age group (8–13 years) compared
to the school participation of children in the second age group (14–17 years). The odds ratio of
.26 implies that a girl is 74 percent less likely to be in school compared to a boy in the first age
group, holding all other factors constant. It should be noted here that the proportion of out-ofschool children was very low (2.4 percent) for 8–13-year-old children, as most of these children
68

were in their compulsory primary school education. Thus, it can be argued that most out-ofschool children in this age group probably belonged to marginal groups, such as the poorest
households, ethnic minorities, and highly conservative families. It is a well-established argument
in the literature that girls living in these kinds of marginal groups are generally more
disadvantaged in terms of school participation. Thus, this could be one possible reason for the
strong association of gender with school participation for younger children. Descriptive analysis,
presented in Table A5, also supports this argument by showing that a significant proportion of all
out-of-school children (76.3 percent) in the first age group were female, while 56.4 percent of all
out-of-school children were female in the second age group.
As seen in Table 4.7, there was also a much stronger association between the mother’s
ability to speak Turkish and school participation for children in the first age group, implying that
a significant number of the out-of-school children in this age group belonged to ethnic
minorities. For the first age group, descriptive analyses show that 9.5 percent of children whose
mothers could not speak Turkish were not in school, while only 1.9 percent of children whose
mothers spoke Turkish were not in school (see Table A6 in Appendix). Household poverty,
household size, age, and mother’s traditional gender role attitudes were also more strongly
associated with the school participation of children in the first age group compared to the second
age group, but the differences between magnitudes were not very high.
As expected, the household head’s education was found to be positively and very
strongly associated with the school participation of children between 14–17 years old, which
included the majority of out-of-school children in Turkey. Surprisingly, however, the household
head’s education was found to be negatively associated with the school participation of children
in the first age group. Because the proportion of out-of-school children in this age group was

69

very low, a few unusual cases from marginal groups might explain this unexpected result. For
example, conservative/religious parents may intentionally refuse to send their daughter to mixedgender schools, especially at the second part of primary school (6–8 grades), even if they are
well educated. This could be one possible explanation given the fact that most out-of-school
children were girls at this level. In addition, children who have special educational needs may
represent an important proportion of out-of-school children in this age group. It can be argued
that more educated parents have a better sense of their children’s special needs and are able to
provide these needs outside of the formal education system. It is also known that some welleducated parents, even some public figures, choose to educate their children themselves rather
than sending them to school, although home schooling is not legal in Turkey. In sum, these few
unusual cases may help to explain the negative impact of the household head’s education on a
child’s schooling in the first age groups, since the proportion of out-of-school children was very
low and all other important factors were controlled for.
4.4.2. Employing the Binary Logit Model for Different Regions
Turkey shows great disparities among its regions in terms of many economic, cultural,
and social areas, as illustrated in Chapter 1. In general, the West and the South are known as the
most advanced and urbanized regions of the country. According to descriptive analyses
presented at the beginning of this section, the West and the South also have similar patterns in
terms of the school participation of children. The percentages of out-of-school children are very
similar in these two regions, and their urban settlements share the significant portion of these
children, unlike the other regions. Compared to the West and the South, however, the North and
the Central regions enjoy a lower percentage of out-of-school children in general, especially in
urban areas. In contrast, rural areas of both the North and the Central regions contain
70

considerably higher percentages of out-of-school children compared to the West and the South.
As expected, since it is known as the least developed region of the country in terms of many
socio-economic variables, descriptive analyses indicated that the East had the highest percentage
of out-of-school children in both the urban and the rural areas among all regions of Turkey.
Given both great socio-economic and cultural disparities among regions of Turkey and
the differences found in the patterns of school participation in these regions, I decided to analyze
the data separately for different regions in order to see if the relationship between each
individual-level independent variable and the school participation status of children differed by
region. Therefore, I applied my binary logit model separately to all five regions. The results of
these analyses are presented in Table 4.8.
Table 4.8: Binary logit analysis by regions
West
Exp (B)
(n=1,278)
Gender (female)
.89**

South
Exp (B)
(n=886)
.53**

Central
Exp (B)
(n=1,131)
.54**

North
Exp (B)
(n=627)
1.60**

East
Exp (B)
(n=2,718)
.36**

.54**

.49**

.50**

.46**

.55**

0

1.93**

0

0

1.85**

Household head’s education

1.82**

2.79**

2.66**

1.47**

2.53**

Mother accepts traditional gender role

.85**

.95**

.42**

.87**

.80**

Household poverty

.43**

.37**

.35**

.09**

.45**

Number of household members

.90**

.80**

.94**

.80**

.86**

Age
Mother speaks Turkish

** p≤ 0.01
According to the results of my analyses, several important differences among regions
emerged. For example, being female was found to be a significant advantage for children in the
North, where females were 1.6 times more likely to be in school, in contrast to all other regions.
This finding could be explained by the more independent position of women in the North, as
discussed in Chapter 1. In all other regions, however, being female had a significantly negative
71

impact on school participation, with the highest impact in the East and the lowest impact in the
West, as expected. In the East region, being female is associated with a 64-percent decline in the
probability of school participation compared to an 11-percent decline in the West region.
According to the results, the mother’s ability to speak Turkish is associated with around a 90percent increase in the probability of enrollment in the East and the South regions. The mother’s
ability to speak Turkish, however, was not found to be an important factor for the school
participation of children in three of the five regions, the West, the North, and the Central. These
regions do not historically include a high population of ethnic minorities, but it is known that
many people with different ethnic backgrounds have migrated to these regions, especially to the
big cities. However, it seems that these immigrants mostly learn Turkish to maintain their lives,
as there are just a handful mothers (0.3 percent in the West, 0.3 percent in the North, and 0.1
percent in the Central) who could not speak Turkish in these regions, according to the data.
Another eye-catching finding is that the impact of poverty on school participation was
lowest in the East and highest in the North. While being poor is associated with a 55-percent
decline in the probability of school participation in the East region, it is associated with a 91percent decline in the probability of school participation in the North region. Although the East is
economically the least developed part of the country, it seems that social factors such as
household head’s education and gender role attitudes are more important factors for schooling in
this region. Hence, economic incentives may not solve the issue entirely in this region. Rather,
some social interventions such as educating parents about schooling, using media and other
resources to change gender perceptions, and establishing specific educational settings, such as
girls’ schools, in the region should be considered. This result could also be explained by the lack
of variation in household wealth in the East (72.3 percent of the population in the East lives in

72

one of the two lowest wealth quintiles).
The high impact of household poverty on school participation in the North may be
explained in two different ways. First, this region is known for its important agricultural products,
such as hazelnut and tea. Some families from less developed parts of Turkey temporarily migrate
to this region in certain seasons to work in agriculture, since traditional agricultural methods,
which require more human power, are generally used in the region because of the geographic
difficulties and the types of products. As discussed before, employment in agriculture is often
mentioned as an important drawback for the school participation of poor children in the literature.
Therefore, it can be argued that children from poor families in the region may easily be
employed in agriculture and give up schooling, especially in rural areas. The second reason for
the strong impact of poverty on school participation in the region could be the geographic
structure of the region itself. Because of the very mountainous territory and extensive forests,
houses are built far away from each other, and even villages are spread out in wide areas in the
region. Thus, accessing school can be very difficult and expensive, especially at the secondary
level. This may put an additional burden on children from poor families and make it hard for
them to continue their education when considering the fact that poor households are generally
located in rural areas.
Finally, the results indicated that a mother’s gender role attitude was a very important
determinant of schooling for children in the Central region because a child whose mother agreed
with the proposition ―it is better to educate a son rather that daughter‖ had 58 percent less chance
to be in school compared to one whose mother disagreed. Traditional gender role attitudes are
often mentioned as an important drawback for children in the East region, but it seems that this is
also an important issue in the Central region. This result is not very surprising because several

73

highly conservative cities are located in this region. Thus, this region also deserves attention
regarding traditional gender role attitudes and their impact on school participation.
In sum, the results of analyses reported in this chapter make clear that school
participation at both the primary and secondary levels is still a serious problem in Turkey.
Descriptive statistics revealed clear and systematic associations between all individual-level
independent variables and the school participation of children. The results of the binary logistic
regression analyses also confirm these associations by indicating statistically significant
relationships between all independent variables and school participation. According to the results,
the mother’s ability to speak Turkish and the household head’s education were positively
associated with school participation, while being female, being older, the mother’s traditional
gender role attitudes, household poverty, and residing in a large household were negatively
associated with school participation.
The results of the analyses also indicated that there were some important differences in
the relationship between independent variables and the school participation of children in
different age groups. For example, the effects of being female and the mother’s ability to speak
Turkish are much stronger for younger children. In addition, household head’s education was
unexpectedly found to be negatively associated with school participation for younger children,
while it has a strong positive relationship with school participation of older children. There are
also several notable regional differences in terms of the relationship between several
demographic and household factors and school participation. The positive relationship between
being female and school participation and the extreme impact of poverty on school participation
in the North region stand out. In addition, the Central region differs from the other four regions
in terms of the strong negative relationship between mothers’ acceptance of traditional gender

74

roles and the school participation of children, while the East region differs from other regions
with the highest negative impact of being female on school participation.
After introducing the associations between individual-level independent variables and the
school participation of children in this chapter by using various descriptive and binary logistic
regression analyses, the next chapter builds on the results of this chapter by taking communitylevel independent variables into account and applying multilevel modeling approach. Similar to
analyses conducted in this chapter, analyses in the next chapter are also conducted for different
age groups and regions.

75

CHAPTER 5
MULTILEVEL ANALYSES

In the first part of my analyses, the relationship between various demographic and
household characteristics of children and their school participation was explored. Beyond the
systematic and clear association between different independent variables and the school
participation of children that was shown in the descriptive analyses, a statistically significant
relationship between each independent variable and school participation was also found by using
a more advanced statistical method, binary logistic regression. In this section, I introduce the
results of HLM analyses in order to answer my primary research question: Are community-level
factors significantly associated with school participation in Turkey, after controlling for
individual- and household-level factors?
I first employed an unconditional model (a model with a random intercept only) in order
to investigate if there was significant variation in the intercept across communities, as an
indicator of the differences in school participation among the communities. In order to determine
the amount of variation at the individual and community levels, intra-class correlation (ICC) was
calculated based on the results of the unconditional model [See equation (3)]. ICC in this case
provides a measure of the proportion of variance in the outcome at the community level [See
equation (4) for the measurement of ICC].
After confirming that there were significant cross-community variations in the outcomes
in the unconditional model, I next employed the random coefficient model [See equation (5)], to
investigate if the variation at the second level still existed after introducing the level-1
independent variables in the model. I checked for significance in the variance at level-2

76

(community level) and calculated ICCs again for the models where there was significant
variance.
If, after accounting for individual demographic and home background variables, I
observed that community-level variations were still significant, then as the next step I employed
the two-level HLM model in order to investigate which specific community variables were
significantly associated with the school participation of children, controlling for demographic
characteristics and household-level factors. For these models, the specific community variables I
systematically included were average adult education, average gender role attitudes, type of
residence (urban/rural), and community poverty.
The above analysis was conducted for the whole sample, and separately for different age
groups and regions, similar to my binary logistic regression analyses. All of the analyses
presented in this section were conducted via statistical software of HLM 6.08 (Raudenbush, Bryk,
& Congdon, 2004).
5.1. Results of Unconditional and Random Coefficient Models
Results of the unconditional model for the whole sample showed that there were
statistically significant differences across communities in Turkey in terms of children’s school
participation (Variance Component (VC): 0.499, p≤ 0.01), as seen in Table 5.1. Based on this
result, Intra-class Correlation (ICC) was calculated as 0.132. This means that 13.2 percent of the
total variance in school participation was between communities, while 86.8 percent was within
communities. This result confirmed that HLM was an appropriate method for this study.
However, the unconditional models for different age groups showed different results. The results
of the unconditional models indicated that there was not a statistically significant variation
between communities in terms of the school participation of younger children, aged 8–13 years.
77

This result is not surprising as there were not many out-of-school children in this age group;
schooling is compulsory at the primary level in Turkey. In the logistic regression analyses,
however, gender and several household-level factors, such as household size, mother’s ability to
speak Turkish, and household poverty, were found to be strongly associated with the school
participation of these children. Thus, it can be argued that variation in the school participation of
children aged 8–13 mostly occurred at the household level rather than the community level.
Result of the unconditional model for the second age group (14–17 years) indicated that
there were statistically significant (VC: 0.867, p≤ 0.01) differences between communities in
terms of children’s school participation in this age group. According to this result, almost 21
percent of the total variance (ICC: 0.209) in school participation was between communities for
this age group. For the whole sample and the second age group (14–17 years), which showed
significant variation between their communities, random coefficient models were also employed
in order to investigate if there still were significant variations among communities after
introducing the individual-level variables into the models. A significant variation among
communities was maintained in both groups, as seen in Table 5.1 and Table 5.2.
Table 5.1: The results of unconditional models
VC
Whole Sample
0.499**
Age 8–13
0.791
Age 14–17
0.867**
West
0.466
South
0.643**

ICC***
0.132
0.209
0.163

Central
1.043**
North
1.627**
East
1.265**
* p≤ 0.05, ** p≤ 0.01
*** ICCs represent the total variance in school participation
only for the significant VCs).

78

0.241
0.331
0.278
between communities (calculated

Table 5.2: The results of random coefficient models
VC
Whole Sample
0.404**
Age 8–13
Age 14–17
0.378**
West
South
0.159

ICC***
0.109
0.103
-

Central
0.470*
0.125
North
1.692
East
0.744**
0.184
* p≤ 0.05, ** p≤ 0.01
*** ICCs represent the variance in school participation between communities after accounting
for the individual level variables (calculated only for the significant VCs).
I also conducted HLM analyses separately in each region for children aged 14–17, as
previous analyses showed that there was not significant variation between communities in terms
of the school participation of younger children. Results of the unconditional models for different
regions indicated that there were statistically significant variations between communities in
terms of children’s school participation in all regions, except the West region. However,
significant variations between communities remained in only two regions, the East region and
the Central region, after taking individual-level independent variables into account. Thus, it
appears that communities matter less in terms of children’s school participation within specific
regions, and it is likely that the national results were driven by a few regions where there were
cross-community variations in the likelihood of the school participation after accounting for the
home background.
One reason for this could be the similarities among communities within the same region
in terms of the level of urbanization, geographic structure, socio-economic and cultural
differences, etc. Thus, it may be assumed that individual-level differences were more influential
than communities in terms of the school participation of children at the regional level. In addition,
sample size could be an issue when analyzing the data separately for each region because sample
79

sizes decreased significantly when I separated the whole sample into different regions and
dropped the children aged 8–13.
5.2. Analysis of Final HLM Model
Based on the results of the unconditional and the random coefficient models, I employed
a final two-level HLM model to four groups(whole sample, children aged 14–17, and two
regions, Central and East) that continued to show significant variations among their communities
in terms of children’s school participation, after taking individual-level variables into the account.
For each group, two separate models were run. As explained in the methods section, because of
the high correlation between poverty and adult education at the community level, I included
adult education in the first model and poverty in the second model. Results of the two different
models for the whole sample and for the children aged 14–17 are presented in Table 5.3.

80

Table 5.3: The results of HLM analysis for whole sample and aged 14–17
Variables
Odds Ratios (Model 1) Odds Ratios (Model 2)
Whole
Whole
Sample Aged 14–17
Sample
Aged 14–17
Level-1 Variables
Gender (female)
0.57**
0.68**
0.58**
0.69**
Age

0.52**

0.64**

0.53**

0.64**

Mother speaks Turkish

1.53**

1.41*

1.58**

1.47*

Household head’s education

1.54*

1.77**

1.84**

2.14**

Mother’s traditional gender role 0.95

1.00

0.97

1.03

Household poverty

0.59**

0.67**

0.56**

0.64**

Number of household members

0.90**

0.91**

0.89**

0.90**

Level-2 Variables
Average Adult Education

15.17**

16.51**

Average Gender Role Attitudes

0.13**

0.12**

0.06**

0.05**

Urban Residence

0.91

1.02

1.04

1.16

-

-

0.59

0.57

Community Poverty
* p≤0.05, ** p≤0.01

-

-

HLM analyses of the whole sample and the children aged 14–17 showed similar results
with the logistic regression analyses in terms of the significance and the magnitude of the odds
ratios for level-1 variables, with a few exceptions. One exception was that the odds ratios for
household head’s education decreased significantly in HLM analyses, compared to the logistic
regression analyses for both groups. Introducing community-level adult education into the
models may partly explain this result, because the magnitude of odds ratios for household head’s
education increased when community-level poverty was introduced to the models instead of
community-level adult education. Compared to the results of binary logistic regression analyses,
mother’s ability to speak Turkish also became less significant for the children aged 14–17 in
both HLM models. The biggest difference between the results of the logistic regression analyses
and the HLM analyses at level-1 was, however, the mother’s acceptance of traditional gender
role attitudes. This variable became insignificant both for the whole sample and for the children
81

aged 14–17 in the HLM results. Other than these specific differences, the HLM results at level-1
were mostly aligned with the logistic regression analyses presented in the previous chapter in
terms of the association between level-1 variables and the school participation of children.
In terms of the level-2 variables, it was found that there was a very strong and
significantly positive relationship between average adult education in the community and
children’s school participation, after controlling for the individual- and household-level variables.
As shown in Table 5.3, a child who lived in a community where all adults had at least a middle
school education was 15 times more likely to be in school compared to a child who lived in a
community that did not include any adult with a middle school or higher level of education. As
seen in Table 5.3, the importance of the community-level adult education became much stronger
for the children aged 14–17. Similarly, the average gender role attitude in the community was
significantly associated with the school participation of children who lived in that community.
This means that a child from a community that held more traditional gender role attitudes
suffered more in terms of his/her school participation compared to a child who came from a very
similar family but lived in a less traditional community. The importance of the community-level
gender role attitudes for children’s school participation also increased by the children’s age, but
the magnitude of the increase was relatively small.
Results of the HLM analyses also indicated that living in an urban or rural residence did
not significantly influence the school participation of children after controlling for various
individual-level and community-level variables. In the descriptive analyses, it was found that
children who lived in rural areas were significantly disadvantaged in terms of their access to
schooling, compared to children who lived in urban areas. Given this fact, the results of the HLM
analyses implied that differences between rural and urban areas in terms of the school

82

participation of children were not merely caused by the location itself, but may mostly have been
derived from the household-level and community-level disparities between these locations. This
result and interpretation, however, should be read with more caution since this variable is binary
and is not similar to other level-2 variables in terms of the variation it contains.
Finally, my second models for both groups showed similar results with the first model in
terms of both level-1 and level-2 variables. However, the new variable — community level
poverty — that was added into the model instead of the community-level adult education did not
seem to be an important factor for children’s school participation, net of other community-level
and individual-level variables. To the extent that this variable is seen as a proxy for a
community’s economic status, it can be argued that community-level economic factors such as
the quality of the infrastructure in the community, the economic conditions of neighbors, the
availability or ease of access to school, etc., were not important determinants of school
participation in Turkey. Rather, families’ own economic condition was an important factor that
may influence school participation of children regardless of the economic condition of the
community in which they live. This result also aligns with the previous finding that living in an
urban or rural area is not significantly related to school participation.
As mentioned earlier, the final HLM model was also employed for the East and the
Central regions. The results of HLM analyses for these regions are illustrated in Table 5.4. The
results were mostly in line with the HLM results for the whole sample and for children aged 14–
17, with a few differences. The child’s gender, for example, was more strongly associated with
school participation in both regions compared to the whole sample and to children aged 14–17,
with a stronger association in the East region. This finding is aligned with the results of the
logistic regression analyses presented in the previous chapter. In addition, for the Central region

83

the number of household members was not significant in both models, and the household head’s
education was not significant in the first model. More interestingly, household poverty became
non-significant for both regions in both models. Accounting for community-level variables in the
framework of HLM could have caused these results. In addition, the significant decrease in the
sample sizes and lesser variation between households in terms of poverty levels in the same
region could be the other possible explanations for these results.
Table 5.4: The results of HLM analysis for two regions (East and Central)
Variables
Odds Ratios (Model-1) Odds Ratios (Model-2)
East
Central
East
Central
Level-1 Variables
Gender (female)
0.33**
0.46**
0.34**
0.47**
Age
0.54**
0.49**
0.55**
0.50**
Mother speaks Turkish***
1.42*
1.44*
Household head’s education
Mother accepts traditional gender role
Household poverty
Number of household members

1.91**
0.89
0.89
0.91*

1.94
0.57
0.68
0.98

2.06**
0.88
0.93
0.91**

2.33*
0.57
0.69
0.97

Level-2 Variables
Average Adult Education
Average Gender Role Attitudes
Urban Residence

42.38**
0.59
1.17

20.07**
0.31
1.16

0.41
1.31

0.38
1.23

Community Poverty
0.21**
0.33
* p ≤ 0.05, ** p ≤ 0.01
*** The models for the Central region could not be run with the variable of mother’s ability to
speak Turkish, because there is only a handful mothers in this situation in the Central region.
In terms of the level-2 variables, average adult education was found to be strongly
associated with the school participation in both regions, with a more dramatic form in the East
region. This result was in line with the results of HLM analyses for the whole sample and
children aged 14–17. However, it should be emphasized here that the association between
average adult education and school participation of children was much stronger in the East

84

region. Given the fact that the East region is the least developed region of the country in terms of
many socio-economic variables, children who live in this region are the most disadvantaged in
terms of access to schooling. It seems, however, that having more educated adults in a
community may eliminate the negative influence of different socio-economic factors, and may
help with increasing the school participation of children in that community.
According to the HLM results for both the East and the Central regions, the average
gender role attitude in the community was not significantly associated with the school
participation, while this variable was found to be significant for both the whole sample and for
children aged 14–17. This result seems surprising because the relationship between mother’s
acceptance of traditional gender roles and the school participation of children was strongest in
these two regions according to the results of the binary logistic regression analyses. This
surprising result may have been caused by the relatively lesser variation among communities at
the regional level in terms of this variable. It is logical to assume that this variable has more
variation between regions than within regions. In addition, the numbers of communities are
relatively low for these regions, so this could be another reason for this result.
Another interesting result of the HLM analyses for the two regions was the significant
association between community poverty and school participation in the East region, different
from the results of other HLM analyses. As mentioned earlier, there was less variation in the East
region compared to other regions in terms of household poverty, because almost three of four
households lived in poverty in this region. Thus, it was found that household poverty had the
weakest association with school participation in the East region among all regions at level-1.
However, it seems that there was significant variation in community poverty in the East region,
and living in an affluent community had a significant positive impact on students’ school

85

participation in this region. Given the fact that this region was the least developed part of the
country, it can be argued that community-level economic factors such as the quality of
infrastructure, economic conditions of neighbors, availability and ease of access to school, etc.,
may still have been important determinants of school participation of children in this region. In
addition, given the strong impact of community-level adult education on school participation in
the East region, this result may also have been affected by the high correlation between
community-level adult education and community poverty.
In sum, the results of multilevel analyses showed that there were significant differences
between communities in terms of the school participation of children in Turkey, with the
exception of children aged 8–13. According to the results, living in a community with higher
average adult education or with more positive gender role attitudes was strongly associated with
the chance of participating in formal education. However, community poverty and urban
residence were not found to be important determinants of school participation. According to the
regional analyses, there were significant variations between communities only in the Central and
the East regions, after accounting for individual-level variables. The results of HLM analyses for
these regions show that average adult education in the community was also strongly associated
with school participation in both regions. While average gender role attitudes in the community
and urban residence were not significantly associated with school participation in both regions,
community poverty was found to be significantly associated with the school participation of
children in the East region. Most importantly, taken together the findings from this chapter
underscore the importance of taking into account the child’s context in order to better explain her
likelihood of school participation at the secondary level in Turkey.

86

CHAPTER 6
DISCUSSION AND CONCLUSIONS

6.1. Summary of Findings
The primary purpose of this study was to investigate the association between socioeconomic community characteristics and school participation at both the primary and secondary
levels in Turkey. Previous studies in Turkey had paid significant attention to the relationship
between household factors and school participation, but the impact of community on the school
participation of children had not been explored. Thus, this study attempted to fill this gap in the
Turkish literature by using a large-scale, nationally representative data set, TDHS-2008. Data
was organized at two levels, individual and community, and was analyzed by using appropriate
statistical methods. In order to investigate the relationship between individual-level factors, such
as gender, household wealth, household size, etc., and the school participation of children, binary
logit models were employed. Then, the community-level factors of average adult education,
community poverty, urbanization, and average gender role attitudes were also taken into account,
and HLM analyses were conducted.
Results of the study confirm the findings of previous studies in terms of the relationship
between household-level factors and the school participation of children in Turkey. Binary
logistic regression analyses indicate that demographic characteristics of children (gender and age)
and household-level factors (education of household head, household size, the mother’s ability to
speak Turkish, mothers’ gender role attitudes, and household poverty) are significantly
associated with children’s school participation. All of the associations are in the expected
direction and align with previous studies, both in Turkey and in other developing countries, as

87

discussed in the literature review chapter. For example, girls, older children, children whose
mother cannot speak Turkish, and children who live in poor households are more likely to be out
of school compared to their counterparts.
Several important regional differences also emerged in terms of the relationship between
the household and demographic characteristics of children and their school participation. These
differences are explained by the socio-economic, demographic, and geographic disparities
between regions. For example, the North region is the only region where girls are more likely to
be in school compared to boys, net of other household characteristics. As mentioned in the first
chapter, women generally are in charge of agricultural work and enjoy relatively higher
independency in this region. Thus, the mother’s more independent position may ease the school
participation of girls in this region. In addition, males tend to out-migrate beginning at early ages
because of the employment limitations in non-agricultural sectors. This may also force boys to
drop out of school when they reach a certain age. In the North region, the negative relationship
between household poverty and school participation is the strongest among all regions of Turkey.
Geographic and agricultural structures of the region could be important factors causing this result.
Having agricultural sectors that require high levels of human work force participation may
encourage children from poor families to give up schooling and work in agriculture in the region.
In addition, because of the rough geographic structure and extensive forests, houses are generally
located far from each other and from city centers in the region. It can thus be very challenging
for children to commute to school, especially in rural areas where most poor people live.
Being female had the strongest negative relationship with school participation in the East
region. This result was expected because the East region is the least developed part of Turkey in
terms of many socio-economic variables, and its residents are known for their strong traditional

88

norms and patriarchal culture. Mothers’ acceptance of traditional gender role attitudes, however,
shows the strongest association with school participation in the Central region. Traditional
gender role attitudes are especially known as an important drawback for girls’ education in the
East region, so this result is a little bit surprising. However, the Central region also has several
very conservative cities, where girls’ school participation is traditionally low, although it is
socio-economically well developed compared to the East region. Thus, it is reasonable to have a
strong relationship between the mother’s traditional gender role attitudes and school participation
in the Central region, after controlling for other important household-level socio-economic
variables.
Finally, the mother’s ability to speak Turkish is only significant in two regions, the East
and the South, mostly because there are not many mothers who cannot speak Turkish in the other
three regions. The East region is known as the most ethnically diverse region of Turkey and
includes many people, especially women at older ages, who cannot speak Turkish. Although the
South region is less diverse in terms of ethnicity, it receives many immigrants from the East
region as do the West and some parts of the Central regions. Hence, it seems that people who
have migrated to the South region may have some adaptation problems and preserve their
language. Immigrant women in this region may also not learn Turkish if they generally live with
other immigrant families and do not feel the need to learn Turkish. According to the results,
children from these families experience significant school participation problems. The results
also suggest that perhaps those who migrate to the West and the Central regions better adapt to
their new places and learn Turkish, compared to those who migrate to the South region.
The results of the HLM analyses, in which the community-level variables are taken into
account, include the biggest contribution of this study to the field. First of all, the results show

89

that there are significant differences between communities in terms of the school participation of
children in Turkey. This is, however, only true for secondary school–aged children. The reason
for this could be the fact that primary education is compulsory and only a few children in each
community do not participate in primary education. Thus, there is not significant variation in
school participation at the primary level between communities. In addition, attending primary
school may be seen as a norm even in the less educated communities. For example, uneducated
adults in the community may not see the benefit of attending higher levels of schooling and may
not be good role models for children to continue their education, but they may at least encourage
students to attend primary education to obtain basic skills. Traditional gender role attitudes in the
community also may not be a big obstacle for younger children, because they are not yet seen as
marriageable.
In terms of the specific community-level variables, average adult education and the
average gender role attitudes in the community, which I use as a proxy for social context, were
found to be significantly associated with school participation, while economic variables,
community poverty and urbanization, were not significantly associated with school participation.
These results also mostly align with the results of previous studies dealing with the relationship
between community-level factors and school participation in other developing countries (Adams
& Hannum, 2005; Binder, 1999; Chudgar, 2006a, 2008; Chudgar & Shafiq, 2010; Connelly &
Zheng, 2003). While the strong relationship between social context variables and school
participation is expected, finding no significant relationship between community-level economic
variables and school participation is little bit surprising because urban and affluent communities
often enjoy better educational and other community infrastructures. However, it can be argued
that although these communities may offer better educational infrastructures for children, they

90

may also provide more opportunities to earn money by working in different informal job markets.
Thus, these educational opportunities may in conflict with the child labor market (Chudgar &
Shafiq, 2010). In addition, the direct costs of schooling, such as informal school fees, costs of
school costumes, and costs of transportation and meals, could be higher in affluent communities.
If there are a few poor families in an affluent community, these higher educational costs could be
an important drawback for their children’s schooling. Thus, after controlling for household
economic situations, it seems reasonable that I did not find a significant relationship between
community-level economic conditions and school participation.
The results of the study mostly confirm the sociological theories that argue that the social
context can affect individuals’ behaviors through different mechanisms. For example, finding a
significant relationship between gender role attitudes in the community and the school
participation of children supports the common argument that defines social norms as a
significant contributor to educational exclusion (Govinda & Bandyopadhyay, 2010). In addition,
sociological theories assign a very important role to adults in the community for children’s
development in that community. For example, it is argued that educated adults can help children
in their community to learn acceptable behaviors better (Jencks & Mayer, 1990), and parents
may also benefit from other adults in their community in terms of the education of their children
(Lee & Croninger, 1996). The results of this study support these arguments by indicating a
strong relationship between community-level adult education and the school participation of
children, after controlling for many important individual-level variables. However, the results of
the study cannot confirm the economic argument that schooling decisions of families are
influenced by the general economic conditions in their communities (Gephart, 1997). Instead, it
was found that the economic condition at the household level was strongly related to the school

91

participation of children, whether they live in a poor or an affluent community.
6.2. Policy Implications for Turkey
First of all, the results of this study in terms of the relationship between household factors
and the school participation of children mostly align with previous studies. My results show that,
for example, household wealth and parental education are still important determinants of school
participation in Turkey. Therefore, the efforts that have been shown by the Turkish government,
such as educating parents about the importance of schooling and providing financial support for
poor families to send their children to school, especially for girls, should be continued. Regional
analyses also show that there is more need for financial assistance in different forms, such as
providing cash and free shuttles to schools, in the North region. In addition, programs that are
initiated either by government or by non-profit organizations with the aim of supporting girls’
schooling need to be continued, especially in eastern Turkey. Changing traditional gender role
attitudes should also be targeted all over the country, with specific attention to the Central region.
The results of the study also suggest that children from different ethnic backgrounds
should be paid special attention in terms of their school participation in the East and the South
regions. The strong association between the mother’s ability to speak Turkish and the school
participation of children in these regions points to the significant role that language background
may play. Thus, the government should take the necessary steps to increase school participation
among children whose mother tongue is different than Turkish. First of all, women who cannot
speak Turkish in the East and the South regions can be identified, and special education
programs, similar to literacy programs, can be organized with the aim of teaching Turkish to
these women. In the meantime, languages other than Turkish can be used to educate parents
about the importance of schooling and inform them about educational process of their children if
92

the parents, especially mothers, do not know Turkish. In addition, educational programs can be
differentiated and adapted to the special needs of these children, especially in the early grades,
because many of these children start their education with insufficient Turkish proficiency.
Teachers should also be trained to teach Turkish, which is the official and the primary education
language in Turkey, as a second language to these children. Similar policy changes may both
encourage the families to send their children to school and minimize the drop-outs by reducing
the hardships these children may encounter during the first years of their education.
The biggest contribution of this study to the literature is from exploring the association
between community-level variables and the school participation of children in Turkey. The
strong relationship between community-level adult education and school participation suggests
that there is a need for community-level empowerment in terms of adult education. Likewise, the
significant association found between average gender role attitudes in the community and the
school participation of children confirms the need for educating adults about the importance of
schooling for young children, especially for girls. Therefore, governments should focus on adult
education at the community level as one of the methods for increasing the school participation of
children. Communities where both the average adult education and school participation of
children are low can be identified and the adults in these communities, including the ones who
do not have school-aged children, can be educated about the importance of schooling. Formal
degree programs, such as intensive primary and secondary education programs and literacy
certificate programs, targeting adults specifically, can also be organized in these communities.
Schools in these communities can be used for these programs during the after-school hours or on
weekends. In these ways, both the relationship between community and schools can be improved
and the benefits of formal education can be better realized by adults in the community. Adults’

93

involvement in formal education can also provide good examples for children in these
communities. In this process, it is also important to focus on changing the gender role attitudes in
the most traditional communities, given the importance of community-level gender role attitudes
to the school participation of children.
The results of the study also show that living in a more affluent community and in an
urban location are not significantly associated with school participation, after controlling for a
variety of individual-level factors. It can, therefore, be argued that allocating all the resources to
building better schools, roads, and other infrastructures may not be the most effective way to
ameliorate the school participation problem. The results of the study imply that community-level
social empowerment projects, such as increasing adult education, changing traditional gender
role attitudes, educating community members about the importance of education, etc., should
also be initiated in addition to traditional economic approaches. This proposition does not
suggest stopping building better schools or improving roads to schools, but emphasizes the
importance of the social context, which has not been paid enough attention by policy makers, for
children’s schooling. It is also necessary to differentiate the approaches for increasing school
participation in different regions. For example, community poverty was only significant in the
East region. This may suggest that there is still need for improving community-level
infrastructures in certain areas of this least developed region of Turkey.
6.3. Broader Implications beyond Turkey
Beyond its unique contribution to Turkish literature, this study also is an important
addition to the school participation literature in developing countries. First of all, the results of
the study confirm the findings of several previous studies from different developing countries by
indicating the significant role of social context in the school participation of children in Turkey.
94

Consistent with the results of previous studies (Buchmann & Brakewood, 2000; Chudgar, 2006b;
Chudgar, 2009; Dreze & Kingdon, 2002), the results of this study show that social context
factors, such as the level of adult education and women’s empowerment in the community, can
be considered to be crucial for school participation of children in developing countries in
addition to well-recognized household-level factors, such as parental education, household
wealth, household size, etc.
Based on the findings of this study and previous studies from other developing countries,
it can be suggested that developing countries should pay special attention to community-level
interventions specifically targeting adults as another way to increase the school participation of
young cohorts. The international community has placed great emphasis on the importance of
increasing general school participation and eliminating gender disparity at both the primary and
secondary school levels in developing countries, especially during the last two decades. Given
the high number of out-of-school children and significant gender disparities in many developing
countries, this emphasis should be continued. It is, however, also necessary to encourage
developing countries to initiate policies to increase the level of adult education and empower
women in society. Beyond their direct positive impacts on the socio-economic development of
societies, these kinds of policies can also help to increase the school participation of children and
reduce the gender inequalities in schooling. Therefore, international organizations, such as
UNESCO, OECD, and The World Bank, should increase their emphases on the importance of
adult education and women’s empowerment in developing countries.
6.4. Suggestions for Future Research
The main findings of this study point to the significant association between communitylevel factors and the school participation of children in Turkey. As a first step in terms of
95

exploring the influence of the social context on school participation in Turkey, this study makes
a significant contribution to the literature. However, there is a need for more research on the
impact of the social and economic context in which individuals live on their schooling behaviors
in Turkey. In this study, data from the TDHS 2008 were used. TDHS 2008 is the latest
implementation of the demographic and health survey series, which has been conducted every
five years since 1968. Therefore, this study can be replicated by using data from the previous
versions of the TDHS in order to see if the results are similar or how the effects of the social and
economic context on school participation have changed over time. In addition, the Institute of
Statistics of Turkey (IST) produces similar nationally represented data sets. Therefore, data from
IST may also be used to replicate this study.
This study makes clear that the social context variables of the level of adult education and
average gender role attitudes in the community are strongly associated with the school
participation of children in Turkey. However, the relationship between economic context
variables and school participation is more ambiguous. The results do not show a significant
relationship between community poverty and urbanization, which are both thought to be
indicators of the economic context, and the school participation of children. It should be noted,
however, that these variables are not direct measures of economic factors such as the quality of
community infrastructures and the prevalence of child labor. Similarly, the TDHS 2008 data
does not include variables about schools and their instructional and infrastructural qualities. Thus,
it will be useful for future researchers to use more direct measures of these factors, where
available, in order to more clearly explore the relationship between specific economic context
factors or school-level factors and the school participation of children in Turkey.
Finally, it should be noted here that qualitative approaches to school participation

96

research are often ignored in both Turkey and other developing countries. Thus, there is also a
need for more qualitative studies to reveal the reasons behind the relationships between various
factors and the school participation of children indicated by statistical analyses. Although
possible reasons are speculated on, based on existing theories and previous research, it is
necessary to confirm these speculations with qualitative studies. Although there have been
several qualitative studies that have explored the factors causing the low school participation of
girls in the East region, this study indicates more research areas for qualitative researchers. For
example, the reasons behind the strong association between household poverty and school
participation in the North region, or the positive relationship between being female and school
participation in the same region, could be uncovered by using a qualitative approach. Similarly,
the strong association between traditional gender role attitudes and school participation in the
Central region, and the significant relationship between mothers’ ability to speak Turkish and
school participation in the East and the South regions, could be important topics to focus on for
future qualitative researches.
At the community level, the causal mechanism between the level of adult education and
school participation of children can also be revealed by using qualitative research approaches.
For example, the strong relationship between adult education and school participation of children
in a community may be caused by the direct interactions between adults and children or by
historically low numbers of educational facilities in the community. Thus, qualitative studies
may help to uncover the real reasons beyond this relationship. It is also important to explore the
possible mediating factors between adult education and school participation of children. This can
be done either by using different quantitative methods (e.g., Structural Equation Modeling) or
qualitative studies. Similarly, future studies may focus on the ways in which traditional gender

97

role attitudes influence school participation of children and how this influence can be eliminated.

98

APPENDIX

99

APPENDIX

Table A1: Previous studies on educational attainment issue in Turkey
Author(s)/DateDependent Independent
Study
Variable
Variables
Tansel (2002)
Years of
Age of children, Parents’
(Determinants of
schooling education and self employment,
school attainment of attained
per adult expenditure, only
boys and girls in
mother/father, development of
Turkey: individual,
street, location of residence
household and
(urban/rural and regions),
community factors)
population density, distance to
metro centers and Istanbul,
provincial percentages of the
employed people in industry
and in the service sectors.
Hisarciklilar (2002) Final
Parental education and
(A censored
grade
occupation, family income per
regression model for attainment capita, family size, existence of
the educational
an elderly household member,
attainment of boys
existence of a female relative,
and girls in Turkey)
mother’s participation in the
labor market, location of
residence (urban/rural and
regions).
Smits and GunduzSchool
Age of children, number of
Hosgor (2006)
enrolment brothers and sisters, birth order,
(Effects of family
extended family, parental
background
education, parental occupation,
characteristics on
income level of family,
educational
mother's age at the marriage,
participation in
gender role attitudes, mother’s
Turkey)
Turkish proficiency, regions,
and urbanization.
Tomul (2008) (The
relative effects of
family socioeconomic
characteristics on
participation
in education in
Turkey)

Final
grade
attainment

Age, family size, parental
education, fathers’ self
employment, place of residence
(rural/urban and regions),
income per head.

100

Data Set
Household
Income and
Expenditure
Survey
(1994)

Method
used
Ordered
probit
model

Household
Labor Force
Survey
(1988)

Censored
ordered
probit
model

-Turkish
Fertility
Survey
(1978)

Multivariate
logistic
regression

-Turkey
Demographi
c and Health
Survey
(1998)
Household Binary
Budget
logistic
Survey
regression
(2005)

Table A1: (Cont’d)
Goksel (2008)
(Determinants of
school attainment in
Turkey
and the impact of the
extension of
compulsory
education)

Final
grade
attainment

Koc (2008)
(Sibling composition
and educational
attainment of boys
and girls)

Final
grade
attainment

Kirdar (2009)
(Explaining ethnic
disparities in school
enrollment
in Turkey)

School
enrollment
and years
of
schooling

Dayıoglu,
Kirdar, and Tansel
(2009) (Impact of
sibship size, birth
order and
sex composition on
school enrolment in
urban Turkey)
Bakis, et, al. (2009)
(Determinants of
access to education
in Turkey)

School
enrollment

Duman (2010)
(Female education
inequality in Turkey:
factors
affecting girls’
schooling decisions)

School
enrollment

School
enrollment

Children’s age, parental
education, whether mother and
father are self-employed,
presence of father, presence of
mother, total household
expenditure, place of residence
(urban/rural), number of
children and percentage of
boys or girls in the household.
Parental education, working in
agriculture (for both parents),
mother working or not, family
income per capita, number of
brothers and sisters, birth
order, place of residence
(urban/rural and regions).
Location of residence
(urban/rural and region),
mother’s age, literacy status
and Turkish language
proficiency, father’s years of
education, sibling composition,
father’s job type, and wealth.
Age of children, number of
siblings, parental education,
presence of father, wealth,
reside in city, population of
city, mother's age, mother's age
at first marriage, mother not
married, mother tongue, and
regions.
Gender of child, income per
head, education of household
head, mothers’ education,
gender of household member,
sibling size, social security
membership, agricultural
activities, regions.
Age of the child, compositionsize of the family, parental
education, self-employment,
rural-urban location,
logarithmic disposable income
per head, and regional
dummies
101

Household
Ordered
Income and
probit
Expenditure model
Survey (1994
and 2002)

Household
Labor Force
Survey
(2006)

Ordered
probit and
cencored
ordered
probit
models

Turkey
Probit
Demographic estimation
and Health
model
Survey (1993
and 1998)

Turkey
OLS and
Demographic 2SLS
and Health
Survey
(1998)

Household
Budget
Survey
(2003)

Logistic
regression

Household
Income and
Expenditure
Survey
(2006)

Censored
ordered
probit
model

Table A2: School participation by age in original data set
Not Enrolled

Enrolled

Frequency

%

Frequency

%

Total

6

210

30.6

477

69.4

687

7

36

5.0

689

95.0

725

8

12

1.6

728

98.4

740

9

6

0.9

671

99.1

677

10

11

1.6

694

98.4

705

11

18

2.5

692

97.5

710

12

16

2.3

687

97.7

703

13

62

9.0

625

91.0

687

14

116

18.1

524

81.9

640

15

185

29.4

445

70.6

630

16

208

33.7

409

66.3

617

17

233

41.8

325

58.2

558

1113

13.8

6966

86.2

8079

Total

Table A3: Age of children in the final data set
Frequency

Percent

Cumulative
Percent

8

740

11.1

11.1

9

677

10.1

21.2

10

705

10.6

31.8

11

710

10.6

42.4

12

703

10.5

53.0

13

688

10.3

63.3

14

640

9.6

72.9

15

632

9.5

82.3

16

620

9.3

91.6

17

560

8.4

100.0

6675

100.0

100.0

Age

Total

102

Table A4: Omnibus Tests of Model Coefficients

Step 1

Chi-square

df

Sig.

Step

1.414E9

7

.000

Block

1.414E9

7

.000

Model

1.414E9

7

.000

Nagelkerke R2 = 0.401/ Cox & Snell R2 = 0.213
Table A5: Out-of-school girls and boys by age groups
In school (%)

Out-of-school (%)

Male

Female

Male

Female

8–13 years

51.5

48.5

23.7

76.3

14–17 years

52.4

47.6

43.6

56.4

Table A6: Out-of-school children by mother’s ability to speak Turkish (Age 8-13)
Mother can speak Turkish
Out-of-school
In-school
No

9.5

90.5

Yes

1.9

98.1

Total

2.4

97.6

Table A7: Pearson’s Correlation between level-2 variables
ADULT_ED

Gender role

C_Poverty

ADULT_ED

1

Gender role

-.380

**

1

C_Poverty

-.815**

.354**

1

* p ≤ 0.05, ** p ≤ 0.01
Table A8: Kendall Tau’s nonparametric correlation
ADULT_ED
Urban
* p ≤ 0.05, ** p ≤ 0.01

.488**

Gender role
-.163**

103

C_Poverty
-.555**

BIBLIOGRAPHY

104

BIBLIOGRAPHY

Adams, J. (2006). Community matters in China. Research in Sociology of Education, 15, 15–41.
Adams, J., & Hannum, E. (2005). Children's social welfare in China, 1989–1997: Access to
health insurance and education. The China Quarterly, 181(1), 100–121.
Anh, T. S., Knodel, J., Lam, D., & Friedman, J. (1998). Family size and children’s education in
Vietnam. Demography, 35(1), 57–70.
Bakis, O., Levent, H., Insel, A., & Polat, S. (2009). Türkiye’de Egitime Erisimin Belirleyicileri.
Istanbul: Egitim Reformu Girisimi, Sabanci Universitesi.
Basu, K., & Van, P. H. (1998). The economics of child labor. American Economic Review, 88(3),
412–427.
Becker, G. S. (1962). Investment in human capital: a theoretical analysis. The Journal of
Political Economy, 70(5), 9–49.
Binder, M. (1999). Community effects and desired schooling of parents and children in Mexico.
Economics of Education Review, 18(3), 311–325.
Blundell, R., Dearden, L., Meghir, C., & Sianesi, B. (1999). Human capital investment: the
returns from education and training to the individual, the firm and the economy. Fiscal
Studies, 20(1), 1–23.
Bramley, G., & Karley, N. K. (2007). Homeownership, poverty and educational achievement:
School effects as neighbourhood effects. Housing Studies, 22(5), 693–721.
Bronfenbrenner, U. (1979). The Ecology of Human Development: Experiments by Nature and
Design. Cambridge, MA: Harvard University Press.
Brown, P. H., & Park, A. (2002). Education and poverty in rural China. Economics of Education
Review, 21(6), 523–541.
Buchmann, C. (2000). Family structure, parental perceptions, and child labor in Kenya: What
factors determine who is enrolled in school. Social Forces, 78(4), 1349–1378.
Buchmann, C., & Brakewood, D. (2000). Labor structures and school enrollments in developing
societies: Thailand and Kenya compared. Comparative Education Review, 44(2), 175–
204.
Buchmann, C., & Hannum, E. (2001). Education and stratification in developing countries: a
review of theories and research. Annual Review of Sociology, 27, 77–102.
105

Canagarajah, S., & Coulombe, H. (1997). Child labor and schooling in Ghana: World Bank.
Chudgar, A. (2006a). Factors Associated with Desicions to Educate Sons and Daughters in India.
Ph.D Thesis, Stanford University.
Chudgar, A. (2006b). The role of context in explaining gender differences in school participation
in rural India: An empirical analysis. Unpublished paper, Stanford University.
Chudgar, A. (2008). Looking beyond the household: The importance of community level factors
in understanding underrepresentation of girls in Indian education. In M. A. Maslak (Ed.),
The Structure and Agency of Women’s Education (pp. 201–215). Albany, NY: State
University of New York Press.
Chudgar, A. (2009). Does adult literacy have a role to play in addressing the Universal
Elementary Education challenge in India? Comparative Education Review, 53(3), 403–
433.
Chudgar, A., & Shafiq, M. N. (2010). Family, community, and educational outcomes in South
Asia. Prospects, 1–18.
Cohen, D., & Soto, M. (2007). Growth and human capital: good data, good results. Journal of
Economic Growth, 12(1), 51–76.
Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of
Sociology, 95–120.
Connelly, R., & Zheng, Z. (2003). Determinants of school enrollment and completion of 10 to 18
year olds in China. Economics of Education Review, 22(4), 379–388.
Crane, J. (1991). The epidemic theory of ghettos and neighborhood effects on dropping out and
teenage childbearing. American Journal of Sociology, 96(5), 1226–1259.
Csapo, M. (1981). Religious, social and economic factors hindering the education of girls in
Northern Nigeria. Comparative Education, 17(3), 311–319.
Dayioglu, M., Kirdar, M. G., & Tansel, A. (2009). Impact of sibship size, birth order and sex
composition on school enrollment in urban Turkey. Oxford Bulletin of Economics and
Statistics, 71(3), 399–426.
De la Fuente, A., & Doménech, R. (2006). Human capital in growth regressions: how much
difference does data quality make? Journal of the European Economic Association, 4(1),
1–36.
De Leeuw, J., & Meijer, E. (2008). Introduction to multilevel analysis. In J. De Leeuw & E.
Meijer (Eds.), Handbook of Multilevel Analysis (pp. 1–76). New York, NY: Springer
Science+Business Media, LLC.
106

Dee, T. S. (2004). Are there civic returns to education? Journal of Public Economics, 88(9–10),
1697–1720.
Dreze, J., & Kingdon, G. G. (2001). School participation in rural India. Review of Development
Economics, 5(1), 1–24.
Duman, A. (2010). Female education inequality in Turkey: factors affecting girls' schooling
decisions. International Journal of Education Economics and Development, 1(3), 243–
258.
Duncan, G. J. (1994). Families and neighbors as sources of disadvantage in the schooling
decisions of white and black adolescents. American Journal of Education, 20–53.
Fernandez Kelly, P. (2002). Commentary: Uses and misuses of social capital in studying school
attainment. In B. Fuller & E. Hannum (Eds.), Schooling and Social Capital in Diverse
Cultures (pp. 73–83). Oxford: Elsevier Science.
Filmer, D. (2000). The structure of social disparities in education: gender and wealth. World
Bank Policy Research Working Paper No. 2268.
Fukuyama, F. (1999). The Great Disruption: Human Nature and the Reconstitution of Social
Order. New York: Free Press.
Galster, G., Marcotte, D. E., Mandell, M., Wolman, H., & Augustine, N. (2007). The influence
of neighborhood poverty during childhood on fertility, education, and earnings outcomes.
Housing Studies, 22(5), 723–751.
Garner, C. L., & Raudenbush, S. W. (1991). Neighborhood effects on educational attainment: A
multilevel analysis. Sociology of Education, 64(4), 251–262.
Gedik, A., Sahin, N., & Suer, S. (2002). The Importance of the Socio-Demographic Indicators in
the Regional Disparities in Turkey, 1990-1994. Paper presented at the 42nd European
Congress of the Regional Science Association Conference, Dortmund, Germany.
Gephart, M. A. (1997). Neighborhoods and communities as contexts for development. In J.
Brooks-Gunn, G. J. Duncan & L. J. Aber (Eds.), Neighborhood Poverty: Context and
Consequences for Children (pp. 1–43).
Gezici, F., & Hewings, G. J. D. (2007). Spatial analysis of regional inequalities in Turkey.
European Planning Studies, 15(3), 383–403.
Gezici, F., & Keskin, B. (2005). Interaction between Regional Inequalities and Internal
Migration in Turkey. Paper presented at the European Regional Science Association
Conference.
Goksel, I. (2008, 11–12 September). Determinants of School Attainment in Turkey and the
107

impact of the extension of compulsory education. Paper presented at the Seminar on child
labour, education and youth employment: A challenge for growth, Madrid.
Govinda, R., & Bandyopadhyay, M. (2010). Social exclusion and school participation in India:
Expanding access with equity. Prospects, 1–18.
Gündüz-Hosgör, A., & Smits, J. (2007). The status of rural women in Turkey: What is the role of
regional differences. In V. Moghadan (Ed.), Empowering Women: Participation, Rights
and Women's Movements in the Middle East, North Africa, and South Asia (pp. 180–202):
Syracuse University Press
Hacettepe University Institute of Population Studies. (2004). Turkey Demographic and Health
Survey, 2003: Hacettepe University Institute of Population Studies, Ministry of Health
General Directorate of Mother and Child Health and Family Planning, State Planning
Organization and European Union. Ankara, Turkey.
Hacettepe University Institute of Population Studies. (2009). Turkey Demographic and Health
Survey, 2008: Hacettepe University Institute of Population Studies, Ministry of Health
General Directorate of Mother and Child Health and Family Planning, T.R. Prime
Ministry Undersecretary of State Planning Organization and TUBITAK. Ankara, Turkey.
Handa, S. (1996). Maternal education and child attainment in Jamaica: testing the bargaining
power hypothesis. Oxford Bulletin of Economics and Statistics, 58(1), 119–137.
Hannum, E. (2003). Poverty and basic education in rural China: Villages, households, and girls'
and boys' enrollment. Comparative Education Review, 141–159.
Hanushek, E. A. (2003). The Failure of Input based Schooling Policies. The Economic Journal,
113(485), F64–F98.
Hanushek, E. A., & Kimko, D. D. (2000). Schooling, labor-force quality, and the growth of
nations. American Economic Review, 90(5), 1184–1208.
Hanushek, E. A., & Woessmann, L. (2008). The role of cognitive skills in economic
development. Journal of Economic Literature, 46(3), 607–668.
Hisarciklilar, M. (2002). A Censored Regression Model For The Educational Attainment Of
Boys And Girls In Turkey. Paper presented at the ERC/METU International Conference
in Economics, Middle East Technical University, Ankara.
Hisarciklilar, M., McKay, A., & Wright, P. (2010, 3–6 January ). Gender Based Differences in
Educational Achievement in Turkey: What Has Changed Over Time? Paper presented at
the 30th Annual Conference of the MEEA, Atlanta.
Huisman, J., & Smits, J. (2009a). Effects of household-and district-level factors on primary
school enrollment in 30 developing countries. World Development, 37(1), 179–193.
108

Huisman, J., & Smits, J. (2009b). Keeping children in school: Household and district-level
determinants of school dropout in 363 districts of 30 developing countries. NICE
Working Paper 09-105.
Jencks, C., & Mayer, S. E. (1990). The social consequences of growing up in a poor
neighborhood. In L. Lynn & M. McGeary (Eds.), Inner-city Poverty in the United States,
Washington, D.C., National Academy Press.
Kaya, I. (2004). Social Theory and Later Modernities: The Turkish Experience. Liverpool:
Liverpool Univ Pr.
King, E. M., & Lillard, L. A. (1983). Determinants of Schooling Attainment and Enrollment
Rates in the Philippines. Santa Monica. CA: The Rand Corporation.
Kirdar, M. G. (2009). Explaining ethnic disparities in school enrollment in Turkey. Economic
Development and Cultural Change, 57(2), 297–333.
Knodel, J., Havanon, N., & Sittitrai, W. (1990). Family size and the education of children in the
context of rapid fertility decline. Population and Development Review, 31–62.
Koc, G. Z. (2008). Sibling Compositions and Educational Attainment of Boys and Girls. M.A.,
Istanbul Technical University.
Kravdal, Ø. (2002). Education and fertility in sub-Saharan Africa: Individual and community
effects. Demography, 39(2), 233–250.
Krueger, A. B., & Lindahl, M. (2001). Education for growth: Why and for whom. Journal of
Economic Literature, 39(4), 1101–1136.
Lee, V. E., & Croninger, R. G. (1996).Social Capital and Children’s Development: The Case of
Education. Madison, WI: Wisconsin Center of Education Research, University of
Wisconsin.
Lloyd, C., Mete, C., & Grant, M. (2007). Rural girls in Pakistan: Constraints of policy and
culture. In M. Lewis & M. E. Lockheed (Eds.), Exclusion, Gender, and Education: Case
Studies from the Developing World. Washington, DC: Center for Global Development.
Lockheed, M. E., Bloch, D., Hamilton, A. O., & Vespoor, A. M. (1990). Primary Education: A
World Bank Policy Paper. Washington, D.C.: The World Bank.
Long, J. S. (1997). Regression Models for Categorical and Limited Dependent Variables (Vol. 7):
Sage Publications, Inc.
MoNE. (2010). National Education Statistics. Retrieved from
http://sgb.meb.gov.tr/istatistik/meb_istatistikleri_orgun_egitim_2009_2010.pdf
109

Nechyba, T., McEwan, P., & Older-Aguilar, D. (1999). The impact of family and community
resources on student outcomes: An assessment of the international literature with
implications for New Zealand. Wellington, Ministry of Education. http://www.minedu.
govt. nz/index.cfm.
Pong, S. L. (1997). Sibship size and educational attainment in Peninsular Malaysia: do policies
matter? Sociological Perspectives, 40(2), 227–242.
Psacharopoulos, G. (1994). Returns to investment in education: A global update. World
Development, 22(9), 1325–1343.
Psacharopoulos, G. (1997). Child labor versus educational attainment Some evidence from Latin
America. Journal of Population Economics, 10(4), 377–386.
Psacharopoulos, G., & Patrinos, H. A. (2002). Returns to investment in education: a further
update. The World Bank Policy Research Working Paper 2881.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical Linear Models: Applications and Data
Analysis Methods (2nd ed.). Thousand Oaks, CA: Sage Publications, Inc.
Raudenbush, S. W., Bryk, A. S., & Congdon, R. (2004). HLM for Windows. (Version 6.08)
[Computer software]. Lincolnwood, IL: Scientific Software International, Inc.
Rutstein, S. (1999). Wealth versus Expenditure: Comparison between the DHS Wealth Index and
Household Expenditures in Four Departments of Guatemala.Calverton, Maryland: ORC
Macro.
Sathar, Z. A., & Lloyd, C. B. (1994). Who gets primary schooling in Pakistan: Inequalities
among and within families. The Pakistan Development Review, 33(2), 103–134.
Schultz, T. P. (2002). Why governments should invest more to educate girls. World Development,
30(2), 207–225.
Skrondal, A., & Rabe-Hesketh, S. (2004). Generalized Latent Variable Modeling: Multilevel,
Longitudinal, and Structural Equation Models. USA: CRC Press.
Smith, R. B. (2011). Multilevel Modeling of Social Problems: A Causal Perspective. New York,
NY: Springer Science+Business Media
Smits, J. (2007). Family background and context effects on educational participation in five Arab
countries. Nijmegen Center for Economics (NiCE). Working Paper, 07-106.
Smits, J., & Gunduz-Hosgor, A. (2006). Effects of family background characteristics on
educational participation in Turkey. International Journal of Educational Development,
26(5), 545–560.
110

Sudha, S. (1997). Family Size, sex composition and children's education: Ethnic differentials
over development in peninsular Malaysia. Population Studies, 51(2), 139–151.
Tansel, A. (2002). Determinants of school attainment of boys and girls in Turkey: Individual,
household and community factors. Economics of Education Review, 21(5), 455–470.
Tansel, A. (2004). Education and labor market outcomes in Turkey. World Bank Report:
Background Papers and Studies for Turkey.
Tomul, E. (2007). The change in educational inequality in Turkey: A comparison by regions.
Educational Planning, 16(3), 16–24.
Tomul, E. (2008). The relative effects of family socio-economic characteristics on participation
in education in Turkey. Eurasian Journal of Educational Research, 30, 153–168.
Tranmer, M., & Elliot, M. (2008). Binary Logistic Regression: Cathie Marsh for Census and
Survey Research, Paper.
Unal, C. (2008). Insani Gelismislik Endeksine Göre Türkiye’nin Bölgesel Farklılıkları (Regional
differences in Turkey by human development index). Cografi Bilimler Dergisi, 6(2), 89–
113.
UNESCO. (2010). Education for All Global Monitoring Report 2010: Reaching the
Marginalized. Paris: UNESCO.
UNESCO. (2011). Education for All Global Monitoring Report 2011. Regional Overview:
Central and Eastern Europe and Central Asia. Paris: UNESCO.
Webbink, E., Smits, J., & Jong, E. (2008). Household and context determinants of child labor in
156 districts of 11 developing countries. Nijmegen Center for Economics (NiCE),
Working Paper 08-114.
Wilson, W. J. (2012). The truly disadvantaged: The inner city, the underclass, and public policy.
Chicago: University of Chicago Press.
Wolfe, B. L., & Haveman, R. H. (2002). Social and Nonmarket Benefits from Education in an
Advanced Economy. Paper presented at the Federal Reserve Bank of Boston Journal
Conference Series.
Wooldridge, J. M. (2003). Introductory Econometrics, A Modern Approach (2nd ed.). Mason,
OH: South-Western, Thomson Learning.

111