You are on page 1of 34

747036

research-article2017
EUSXXX10.1177/0013124517747036Education and Urban SocietyAtaç

Empirical Article
Education and Urban Society
1­–34
Modeling Educational © The Author(s) 2017
Reprints and permissions:
Inequalities: Class, sagepub.com/journalsPermissions.nav
DOI: 10.1177/0013124517747036
https://doi.org/10.1177/0013124517747036
Academic Achievement, journals.sagepub.com/home/eus

and Regional Differences


in Turkey

Ela Ataç1

Abstract
As it has been realized that education is a key to a long-term economic growth
and to reducing social and economic disadvantages, educational inequality and
its reflections in the geography have become some of the major issues in many
countries. Turkey is in many ways a good example to analyze the relations
between class, education, and regional inequalities where education is strongly
a class-related issue and there has also been a strong dimension of “geography” as
far as the educational provision and performance are considered. The purpose
of the article is to contribute to two debates on the relation of education and
inequality in Turkey. One is a specific and practical way of understanding about
the effect of socioeconomic backgrounds of the students on their educational
achievement. The other is an understanding on causal relations based on
socioeconomic variables and geographical variations and how these lead to or
indeed are partly caused by regional inequalities in Turkey. Using the datasets
of PISA (Program for International Student Assessment) database, the datasets
of National University Entrance Examination and Census, the article finds that
for Turkish students where (the region and the place of residence) and with
whom (socioeconomic qualifications of parents) they live are the powerful
indicators of academic achievement.

1TED University, Ankara, Turkey

Corresponding Author:
Ela Ataç, Assistant Professor, Department of City and Regional Planning, TED University
Faculty of Architecture, 06420 Ankara, Turkey.
Email: ela.atac@tedu.edu.tr
2 Education and Urban Society 00(0)

Keywords
educational inequality, geography of education, Turkey

Introduction: How Education Matters


Educational inequality and its reflections in the geography have become
some of the major political and policy issues in many developed and develop-
ing countries. This is mainly because of the fact that education still lies at the
bottom of many social, political, and economic issues. As it has been realized
that education is a key to a long-term economic growth and to reducing social
and economic disadvantages (Barro, 2013; Butler & Hamnett, 2007), the aca-
demic debates have mainly evolved around the issues such as educational
variations between social, economic, and ethnic groups; the role of education
in class positions; the long-term effects of education policies on the econ-
omy; and the geography of education (Goldthorpe,1980; Maloutas, 2007;
Moore, 2004; Raftery & Hout, 1993). Turkey is in many ways a good exam-
ple to analyze the relations between class, education, and regional inequali-
ties in this sense. This is mainly because of the fact that education is strongly
class related in Turkey and there has also been a strong dimension of “geog-
raphy” as far as the educational provision and performance are considered
(State Planning Organization-T.R. Prime Ministy, 2010). The regional varia-
tions in education levels, performances, or provisions are mainly character-
ized by sharp inequalities in Turkey where western regions representing a
socioeconomic trend akin to the developed countries, the east represents its
inverse, and the central areas indicate an in-between case. The case of regional
inequality in Turkey is well known with its characteristic east and west divide
underlined in the article is not a novel finding though. However, what the
article aims to underline is the causal relationships through geographic varia-
tions based on the uneven spatiality of education-related variables. In other
words, the article aims to make case for inequality and geography in urban
and regional studies and to show how geography matters as far as educa-
tional and socioeconomic indicators are considered in Turkey. The purpose of
the article is, therefore, to contribute to two debates on education and inequal-
ity in Turkey. One is a specific and practical way of understanding the effect
of socioeconomic backgrounds of the students on their educational achieve-
ment. The other is more complicated—namely, an understanding on the
mechanisms and causal relations based on socioeconomic variables and geo-
graphical variations, and how these lead to or indeed are partly caused by
regional inequalities in the level of districts in Turkey. The evidence which
can contribute to these two debates results from the integration of three
Ataç 3

previously unconnected datasets: PISA (Program for International Student


Assessment) dataset of 2012 at country level, University Entrance
Examination (UEE) dataset of 2011 at district level, and TURKSTAT (Turkish
Statistical Institute) census dataset of 2011 at district level.
The article consists of two main phases. In the first phase, the relation of
academic achievement and socioeconomic background in Turkey is examined
via ordinal logistic regression (OLR) method based on PISA dataset. In the
second phase, by using geographically weighted regression (GWR) at dis-
trict level, the variations in the districts as well as the causal relationships
that could account for the regional educational differentials in Turkey are
analyzed. The article is, thereby, an attempt to read the new-term regional
differences and inequalities in Turkey by using global, local (spatial), and
causal statistical models. On this ground, building models to explain how
academic achievement is a class-related issue in Turkey (a class-focused
approach), and how it is related with the geographical inequalities in the
country (a geographical-based approach) is another aim of the article.
The rest of the article is, therefore, organized as follows: The theoretical
background is presented in the “The Background: Different Aspects of
Educational Inequality—At Home and Abroad” section, showing different
aspects of educational inequalities both at home and abroad. The datasets and
the methodology of OLR for the first phase of analyses and GWR for the
second part of the analyses are then reviewed in the “Modeling Educational
Inequalities in Turkey: The Data and the Method” section. This section is fol-
lowed by the interpretations of the results. The analysis results of OLR and
the class-effect on academic achievement in Turkey are evaluated in “The
Class-Effect in Academic Achievement in Turkey (OLR Analysis)” section,
whereas in the section after that the analysis of GWR is carried out. The find-
ings of GWR analysis are presented under the heading of “Geographical
Variations in Academic Achievement in Turkey.” The article is concluded in
the “Conclusion” section.

The Background: Different Aspects of Educational


Inequality—At Home and Abroad
The debates about education, class, and inequality have focused around sev-
eral different issues in the literature, including the level of educational attain-
ment or achievement and their variations between different social, income,
ethnic groups, or even genders (Bobbitt-Zeher, 2007; Butler & Hamnett,
2007; Moore, 2004); the relationship between class or socioeconomic groups
and educational opportunity or academic success (Arnot & Barton, 1992;
Blackledge & Hunt, 1985; Goldthorpe, 1980; Lynch & O’riordan, 1998; Pink
4 Education and Urban Society 00(0)

& Noblit, 1995; Raftery & Hout, 1993); the role of education in social exclu-
sion, integration, or mobility (Goldthorpe,1980; Raftery & Hout, 1993); and
its effect on long-term economic growth with the need of highly educated and
skilled workforce in the new term of knowledge-based societies (Barro,
2013; Nelson & Phelps, 1966; Social Exclusion Unit [SEU], 2001).
Among these areas of interest, class and education relation is the well-
pronounced one. This is because of the fact that educational opportunity and
achievement have for long been recognized to be a class-related issue (Butler
& Hamnett, 2007). Despite national variations,1 there is a large consensus in
the literature that children from lower class backgrounds perform less well on
average than those from higher class backgrounds (Butler & Hamnett, 2007).
In the recent literature, the students of education sociology have, thus, returned
to Bourdieu’s habitus and Bernstein’s writings to account for the intricacies of
educational strategies within contexts of inequality in terms of class, ethnicity,
and gender (Maloutas, 2007) referring to the structuralist theories of education
and class. Structuralism is the well-known theoretical model to explain class
and inequalities in education. In a structuralist view, the role of education in
reproducing class inequality is a structural inevitability. However, Bourdieu
and Passeron (1977) and Bernstein (1977) exemplify an explanatory tradi-
tion, which emphasizes the role of culture in structural determinations. The
common assumption of these views is that people are somehow pushed into
certain educational positions instead of choosing them in a free and meaning-
ful sense (Lynch & O’riordan, 1998). SEU (2001) suggests concrete examples
of this with an empirical study on the relation of education and class.
Accordingly, underachieving educational potential has, therefore, in many
ways been one of the central symbols of social and economic exclusion, which
has huge costs for the individuals, groups, and the areas that experience it. In
their study to understand the relation between educational qualifications, fam-
ily background, and social exclusion, they found that a teenager from a
deprived neighborhood is 5 times more likely to go to a failing school and less
likely to achieve good qualifications, and it is very hard for the disadvantaged
children to get rid of those class-based systems of education.
Education has also been accepted as both the cause and the result of social
and economic inequalities (Eckland & Alexander, 1980). It has been increas-
ingly recognized in the literature that education plays an important role in
class positioning in the society, specifically in upward mobility. The stratifi-
cation theorists who depicted the patterns of social mobility such as
Goldthorpe (1980), and Raftery and Hout (1993) assert that in a class-strati-
fied society, it is the equality of opportunity in education that rule social
mobility the most. Today, in Western societies, working class is shrinking and
middle class is expanding and internally diversified. The changing geometry
Ataç 5

of occupational positions has upgraded the requirements of social mobility


and has increased the hopes and fears of success and failure (Maloutas, 2007).
As high-level education is considered a safe way for the social mobility pros-
pects of the young generation or in other words as a path to university, a good
job, and a more affluent life, middle classes in many developed and develop-
ing countries are increasingly concerned to ensure that their children get the
opportunities necessary to ensure their social reproduction (Butler & Hamnett,
2007). Education has thus become a privileged investment area for the fami-
lies which pays back in increased social mobility and a substantial compo-
nent in the formation of broader trends of social inequality (Maloutas, 2007).
Another focus on education is based on its relation with long-term eco-
nomic growth. Since the late 1980s, macroeconomists has focused on long-
term issues of economic growth. Yet, the effect of human capital in general
and education in particular on long-term economic growth is now more popu-
lar than it has been in 30 years, and it has thus become one of the key issues
of macroeconomy. The literature on the relation between long-term economic
growth and education draws upon the very large number of empirical studies.
Barro (2013) is one of the researchers who found empirically that education
has a significant explanatory power for economic growth in their study where
they analyze the relation between economic growth and primary school
attainment and attainment of females for the time period of 1960 and 2000
with comparable examinations of different countries. It is stated in the study
that the strong relation between education and economic growth arises
through two channels: The first one is about the fact that the more human
capital facilities such as health or education in a country, the higher its absorp-
tion of superior technologies from leading countries. This channel is mainly
concerned with the schooling at the secondary and high levels as well.
Second, a country that has a high ratio of human capital which tends to be
more difficulty to adjust than physical capital grows rapidly as well. These
are the reasons why the process of education has been considered as an
investment area for the policy makers or a safe path of economic growth for
the countries in a long term.
The relation between economic growth and education has also created a
recent awareness that education and its equal provision is a key factor in the
face of a knowledge-based society. The key issue for politicians in devel-
oped nations is the awareness that unless the issue of educational perfor-
mance which becomes also an urban issue affecting social justice and
economic competitiveness is addressed, the nations will face the danger that
those of their populations who do not have good educational skills will expe-
rience labor market exclusion and will at best only be able to aspire to a
minimum wage standard of living (Butler & Hamnett, 2007). An empirical
6 Education and Urban Society 00(0)

model put forward by Nelson and Phelps (1966) shows that education is
important as an explanation of why countries might fail to use the best-
practice technology. The countries where efficient economies cannot per-
form are also those where education facilities are inadequate and inefficiently
produced or served. This basically means that there is also a strong link
between scientific advance, innovation, or technology and the way in which
education has facilitated the development of knowledge in a country (Stevens
& Weale, 2003).
All these international academic debates on education and inequality have
also been some major sources of social, political, and academic controversies
in Turkey where there are significant variations in class, income groups, and
socioeconomic status and where “economic growth” has been very much at
the center of the recent public and political issues in the last decades. To sim-
plify the condition of Turkey in social and economic aspects, the world devel-
opment indicators of World Bank (2013) by countries are summarized in
Table 1 in a comparable way.
As can also be seen from these figures in the table, Turkey is in many ways
an interesting case where economic growth has an indirect relation with the
indicators of population, technology, social life, and economy. According to
World Bank, Turkey is an upper-middle-income country located in the region
of Europe and Central Asia. As can also be seen from the growth rates of
Turkey in the table, economic growth lies at the heart of its macroeconomic
strategies as well. Although Turkey’s performance about economic growth
has been impressive since the early 2000s, it is the year of 2012 when the
growth has slowed, per capita income has stagnated, and unemployment—
particularly youth unemployment—is rising. Yet, according to the world
development indicators of World Bank given in the table, although Turkey is
still among the 20 biggest economies of the world in terms of current GDP
(since the late 1990s), the indicators other than the economic ones (such as
GDP and annual growth) represent dramatically lower figures compared with
the other countries. It is especially the indicators about women, education,
and the use of technology where Turkey ranked at the bottom of the list. In
specific, the indicators where Turkey takes place at the bottom of the list are
high technology, women in primary education, women in labor force, out-of-
school children and number of students per teacher which means that Turkey
performs very poorly in terms of education, innovation, and women issues.
Without question, these three problem areas are the key issues for Turkey
which also pave the way of all kinds of inequalities.
As far as education is concerned in Turkey, it can be stated that in the last
few years, several developments have been achieved specifically in primary
education. Compulsory education in Turkey has been extended from 5 to 12
Ataç 7

Table 1.  Turkey’s Ranking by World Development Indicators.

Economic indicators Population indicators

2013 GDP (million US$) Annual growth Fertility Infant mortality

US 14,296,446 China 7.68 Korea 1.19 Japan 2.90


China 7,892,710 India 6.90 Spain 1.27 Korea 3.70
Japan 4,055,838 Indonesia 5.58 Poland 1.29 Italy 3.70
Germany 3,153,493 Turkey 4.19 Germany 1.39 Germany 3.90
France 2,346,558 Brasil 3.02 Italy 1.39 Netherlands 4.00
UK 2,302,811 Korea 2.90 Japan 1.40 Australia 4.10
Brasil 2,008,235 Australia 2.44 Switzerland 1.52 Switzerland 4.20
Italy 1,731,484 US 2.22 China 1.56 Spain 4.30
Russia 1,682,153 UK 2.16 Canada 1.61 France 4.40
India 1,594,667 Canada 2.00 Netherlands 1.68 UK 4.60
Canada 1,456,386 Switzerland 1.77 Russia 1.70 Poland 5.20
Australia 1,208,811 Japan 1.60 Brasil 1.80 Canada 5.20
Spain 1,119,003 Mexico 1.39 UK 1.83 US 6.90
Korea 1,059,861 Russia 1.34 US 1.86 Russia 10.40
Mexico 996,807 Poland 1.26 Australia 1.86 China 12.30
Indonesia 805,659 France 0.66 France 1.99 Mexico 14.50
Turkey 763,572 Germany 0.30 Turkey 2.09 Turkey 15.40
Netherlands 729,373 Netherlands −0.50 Mexico 2.27 Brasil 16.00
Switzerland 554,725 Spain −1.67 India 2.47 Indonesia 29.30
Poland 444,198 Italy −1.75 Indonesia 2.48 India 52.10

Technological indicators Unemployment indicators

Youth unemployment
Use of Internet High technology Unemployment (%) (%)

Netherlands 93.96 Korea 27.10 Korea 3.10 India -


UK 89.84 China 26.97 India 3.60 China -
Japan 89.70 Switzerland 26.51 Japan 4 Japan 6.5
Switzerland 86.34 France 25.90 Switzerland 4.40 Germany 7.80
Canada 85.80 UK 21.86 China 4.60 Switzerland 8.50
Korea 84.77 Netherlands 20.41 Mexico 4.90 Mexico 9.20
US 84.20 US 17.82 Germany 5.30 Korea 9.30
Germany 84.17 Japan 16.80 Russia 5.50 Netherlands 11.00
Australia 83.00 Germany 16.02 Australia 5.70 Australia 12.20
France 81.92 Mexico 15.92 Indonesia 6.30 Canada 13.70
Spain 71.64 Canada 14.09 Brasil 6.50 Russia 13.80
Russia 67.97 Australia 12.91 Netherlands 6.70 Brasil 15.00
Poland 62.85 Russia 10.01 Canada 7.10 US 15.50
Italy 58.46 Brasil 9.63 US 7.40 Turkey 16.90
Brasil 51.04 India 8.07 UK 7.50 UK 20.70
Turkey 46.25 Poland 7.81 Turkey 8.70 France 23.90
China 45.80 Spain 7.67 Poland 10.40 Poland 27.30
Mexico 43.46 Italy 7.24 France 10.40 Indonesia 31.30
India 15.10 Indonesia 7.05 Italy 12.20 Italy 40.00
Indonesia 14.94 Turkey 1.88 Spain 26.30 Spain 55.50

(continued)
8 Education and Urban Society 00(0)

Table 1. (continued)

Indicators about women population Education indicators

Women in primary Women in labor force Out-of-school children Number of students per
education (%) (%) (%) teacher

Korea - Russia 48.85 Canada - Japan -


China - Canada 47.21 Switzerland - Spain -
Switzerland - France 47.05 China - Russia -
UK - Switzerland 46.15 Brasil - Mexico -
Germany - Netherlands 46.10 Japan 0.1 Australia -
Canada - UK 45.91 Germany 0.15 Canada -
Brasil - Germany 45.86 UK 0.17 Switzerland -
Japan 100.00 US 45.79 Korea 0.48 Poland 9.53
Spain 99.74 Spain 45.55 Spain 0.49 Italy 11.38
France 99.64 Australia 45.38 Italy 0.54 Germany 12.56
Italy 99.22 Poland 45.12 France 0.66 France 12.94
Netherlands 99.15 China 43.86 Netherlands 0.94 Netherlands 14.11
India 98.57 Brasil 43.70 Russia 2.49 US 14.70
Russia 98.12 Japan 42.6 Mexico 2.51 China 15.14
Mexico 97.97 Italy 41.84 Australia 2.55 Indonesia 15.38
Australia 97.64 Korea 41.60 Poland 2.90 UK 15.87
Poland 97.17 Indonesia 37.97 India 5.02 Korea 15.94
Indonesia 93.77 Mexico 37.17 Indonesia 6.71 Brasil 17.26
US 92.63 Turkey 30.66 Turkey 6.82 Turkey 20.13
Turkey 92.60 India 24.17 US 7.22 India 30.78

Source. http://databank.worldbank.org/data/

years, and the net enrollment ratio has nearly reached a 100% (SPO, 2010).
The Turkish higher education system has also been expanding rapidly in the
last decade. The estimated gross and net enrollment rates in 2004 were 30%
and 17%, respectively, up from 18% and 9% in 1994 (Hatakenaka, 2006).
Moreover in Turkey, there is a strong sense among the public that a high-level
education degree is a safe path to a more affluent life, a good job, or high
income. However, despite the significant processes in educational policies,
the issues directly related with education at any level such as family back-
ground (i.e., size, income, and education level) or place of residence remain
critical in determining access to education, the quality of education, and even
academic achievement in the country (SPO, 2010). What is more, in Turkey,
there has also been a strong geographical dimension to the unequal distribu-
tions of education, income, welfare, or opportunity.
As far as inequalities are concerned, there is a long-standing literature on
the unequal geographical distribution of welfare in Turkey in many aspects.
It is for the reason that geography is increasingly becoming the basis for
accessing many forms of welfare and opportunities in the country. In the
recent years, therefore, a large body of works on educational inequalities
Ataç 9

based on class and geography in Turkey has appeared in the literature in the
same manner. Most of the studies deal with the effect of social and economic
backgrounds of children on school enrollment (Smiths & Hoşgör, 2006;
Tomul, 2011), the link between income inequality and educational variables
(Duman, 2008; Duygan & Guner, 2006), or the unequal geography of educa-
tion variables through the country (SPO, 2010; Tomul, 2007). However, to
quantitatively identify the relationships between class-academic achieve-
ment and geography that the article also aims to is seldom attempted. Among
these a few attempts, there is the work of Tomul (2007) where he examines
the relation of schooling years and gender in the geography of Turkey for the
years of 1990 and 2000. He finds with this study that school years of both
male and female increase in all regions in 2000, and this increase occurs
more in the regions in which years of schooling are low. In 1990 and 2000,
male years of schooling are higher than those of female in all regions, but the
gap between them is still higher in the regions that take place in the eastern
and southeastern parts of Turkey. An examination of the net enrollment ratio
distribution by province made by SPO (2010) shows in the same manner that
regional differences persist and that the ratio of children left out primary
education is higher in the central Anatolian and eastern regions. Among the
voluminous literature on the effect of socioeconomic inequalities in Turkey
on education, Duygan and Guner (2006) make use of modeling the relation
between education policies and fertility to show the impact on persistence of
inequality where Tansel and Bircan (2010) compute the relation between
male wage inequality which is rather high in Turkey and education returns.
Tansel and Bircan find that education contributed to higher wage inequality
through both within and between dimensions; the within-groups inequality
increased and between-groups inequality decreased over the period of 1994-
2002. However, there is some research which shows regional inequalities,
and its relation with social and economic development may be context-
dependent. Ferreira, Gignoux, and Aran (2011) find in their article where
they measure inequality of opportunity using socioeconomic variables at
regional level that rural versus urban birth and parental education appear to
be the main correlates of economic advantage in Turkey. The language spo-
ken at home and the number of siblings are also important in that sense.
Interestingly, the broad geographical division of east–west and center in
which a woman was born appears less important. As wealth distributions,
consumption, and education levels do differ substantially across these
regions, this finding suggests that such differences are due to heterogeneity
in the composition of the population across regions, in terms of the other
circumstances, rather than to any intrinsic regional effects.
10 Education and Urban Society 00(0)

Nevertheless, almost none of the studies on educational inequalities in


Turkey explicitly examine the effect of socioeconomic variables on educa-
tional achievement considering regional inequalities or look deeper into
effect of both class positions and the geography on education in Turkey.
Recalling the international literature on stimulating role of education and
its equal provision lying at the heart of innovation, long-term economic
growth, and class positioning, it is reasonable to state for Turkey that the
relations between educational achievement-opportunities-socioeconomic
variables and regional inequalities should be analyzed by causalities to
achieve a long-term and sustainable economic growth and social develop-
ment. This is exactly what the article aims to. On this ground, the article, in
the next section, introduces the modeling methods of educational inequali-
ties in Turkey.

Modeling Educational Inequalities in Turkey: The


Data and the Method
As has been explained before, the study comprises of two phases of analysis:
(a) analyzing the effect of family background, home assets, and urban effect
as a sign of class-effect, and (b) analyzing the effect of socioeconomic envi-
ronment (education level, fertility, migration, demographic composition, etc.)
on academic achievement in Turkey. To analyze the “class-effect” in
academic achievement, PISA dataset of 2012 at country level is used. PISA
is a program directed by Organisation for Economic Co-Operation and
Development (OECD), which aims to evaluate education systems worldwide
by assessing the competencies of 15-year-olds in reading, mathematics, and
science. The information is collected through background questionnaires that
provide context which can help analysts interpret the results (OECD, 2012).
There are three kinds of surveys in PISA dataset that provide information
about the students’ backgrounds, schools, and their learning experiences.
These are student questionnaires, parent questionnaires, and school question-
naires. As parent questionnaires are not available for Turkey, student ques-
tionnaires are used to find out family backgrounds and learning environments
of the students in the article.
As PISA dataset does not include geographically based data at micro
level, to analyze geographical variations and relations between academic
achievement and the socioeconomic variables in the second phase of the
analyses, UEE dataset of 2011 and census data of 2011 at district levels are
used. Census dataset is provided by TURKSTAT, which includes education
levels, age groups, and migrant population at district level. The dataset of
UEE is provided by Directorate of National Education,
Ataç 11

and it basically comprises the scores of the students in the main fields of
Turkish-Mathematics, Mathematics-Science, and Equal-weighted. The data
also include the information of the students who achieved the required level
at UEE to attend at any undergraduate program of universities in Turkey. In
the article, these data are used as the main indicator of academic achieve-
ment. Basically, by “academic achievement” or “achievement at UEE,” it is
referred to the share of the students who achieve or exceed the required level
of attainment in an undergraduate program at any university in Turkey in the
total number of students who take UEE. This may not be an ideal indicator
to define “success” for obvious reasons though. It may, for instance, fail to
differentiate the students who are very bright, gifted, or have many other
abilities but failed at UEE. However, despite some caveats on this dataset of
UEE, it is observed by initial analyses that the dataset has strong ability to
separate the districts where the students are performing well or poorly at
UEE. On this basis, as far as district-level analyses are concerned, dataset
gives reliable results.
The article first analyzes how family background, home assets, and type
of settlement are effective in academic achievement and then it examines the
relationship between achievement at UEE and the socioeconomic character-
istics of the districts in Turkey. Following a global regression analysis (OLR
model) to examine the relations between family background and academic
performance at PISA, spatial variations in the relationships are examined
with a local regression analysis (GWR) at UEE dataset to reveal geographical
variations. The details about these two models are given below.

First Phase: OLR Model


Logistic regression is a classical regression model where the independent
variable is used to predict the dependent variable. It is called binary logistic
regression when the dependent variable has two categories. If the dependent
variable has more than two categories, it becomes a multinomial logistic
regression. It is an OLR if the category of the dependent variable is ranked.
As PISA scores are defined in a range that gives ordinal score levels from A
to F, OLR is used in the study.
In the scope of the article where it is aimed to find out the causal rela-
tionships between PISA score levels and the variables related to family
backgrounds, home assets of the students, and urban effect, ordinal regres-
sion is used for a causal analysis which identifies the strength of the effect
that the independent variables have on a dependent variable. The question
that it is posed for this phase is therefore that “What is the strength of the
12 Education and Urban Society 00(0)

relationship between PISA score levels and socioeconomic backgrounds of


the students in Turkey?”
OLR is a well-used model because it has ability to overcome many restric-
tive assumptions of ordinary least squares (OLS). In ordinary regression, a
linear relationship between the dependent (factor variable which must be cat-
egorical) and independent variables (categorical or continuous) is a must-test
whereas in OLR this relationship may be linear or nonlinear. In logistic
regression, the distribution may also be not normally distributed whereas nor-
mal, Poisson or binominal distributions work well in OLR. However, there
are also some norms in logistic regression. First, the dependent variable used
in OLR should be dichotomous in nature for binary regression in terms of
data level. For the binary logistic regression, dependent variables are in two
categories (assumed as 0 and 1), and for multinomial logistic regression, the
dependent variable has more than two categories. One should also know
about OLR that although it does not assume a linear relationship like basic
regression models, there should be a linear relationship between the odds
ratio and the independent variable. As it uses the maximum likelihood
method, a large dataset is required for OLR (Cabrera, 1994; Lall, Campbell,
Walters, & Morgan, 2002; Peng, Lee, & Ingersoll, 2002).
OLR predicts the cumulative probability of the dependent variable order,
and it gives maximum likelihood estimation used to predict the logit or odds
ratio for the dependent variable. In SPSS, the “parameter estimate” is the b
coefficient (beta value) used to predict or calculate the log odds (logit) of the
dependent variable. Z is the logit for a dependent variable, and the logistic
prediction equation where b0 is constant and k is independent (X) variables is
(Statistics Solutions, 2016)

 prob ( event ) 
z=ln ( odds ( event ) ) =ln 
 prob ( nonevent ) 
  (1)
prob ( event )
=ln = b0 + b1X1 + b2 X 2 +  + bk X k .
1 − prob ( event ) 

Exponential of the parameter estimate (beta value) gives the odds ratio of
the dependent variable. One can find the probability of the dependent vari-
able from this odds ratio as well. When the exponential beta value is greater
than one, the probability of higher category increases, and if the probability
of exponential beta is less than one, then the probability of higher category
decreases. Exponential beta value is interpreted with the reference category
where the probability of the dependent variable will increase or decrease. In
Ataç 13

continuous variables, it is interpreted with one unit increase in the indepen-


dent variable corresponding to the increase or decrease of the units of the
dependent variable (Lall et al., 2002; Peng et al., 2002).
For the first phase of analyses, dependent and independent variables are
defined in a categorical system illustrated in Table 2. As it is the “achieve-
ment” or “success” that the article aims to understand and explain, dependent
or the factor variable is defined as PISA score levels (from A to F and a Z
category), whereas education level of father, education level of mother, occu-
pation group of father, time of computer use, home possessions, and home
educational resources are constructed as the independent variables or covari-
ates. For the PISA score levels, mathematics scores are used and six catego-
ries from A (the highest score) to F (the lowest score) and a subcategory, Z,
are defined. In this grouping, the plausible values found in the dataset of
PISA 2012 are used, and the levels are defined using the cut points given by
PISA (PISA scale in OECD, 2012) as well.
For education levels of the parents, seven categories are defined in accor-
dance with the International Standard Classification of Education (ISCED)
levels: no education, primary education, lower secondary education, upper
secondary education, postsecondary nontertiary education, bachelor’s or
equivalent level, and master’s, doctoral, or equivalent level. In PISA surveys,
students are also asked about the occupations of their fathers, and the occupa-
tions are coded in reference to International Standard Classification of
Occupations (ISCO08) by International Labor Organization in 2008. These
categories are collapsed into four main groups in the study: managers and
professionals (white-collar classes); technicians, associate professionals,
clerical employees, and armed forces; service sector employees; and workers
(Table 2).
Time of computer use represents the average time (in minutes) spent by a
single student in a day. Within the PISA survey, students are asked about
their home possessions and educational resources. For home possessions,
they are able to make multiple selections for the items such as own room,
desk, study place, computer, textbooks, and so on. For home educational
resources, students are able to make multiple selections among the items like
laptop, desktop, tablet computer, Internet connection, cell phone, USB stick,
and so on. In what follows, information and communications technologies
(ICTs) availability at home are also asked to students in a same manner to
understand the relations they built with communication techniques and
innovative systems. For understanding the “urban effect” at student achieve-
ment in PISA, student dataset is joined with the school dataset and this vari-
able is incorporated to the analysis in four categories: metropolis, city, town,
and small town.
14 Education and Urban Society 00(0)

Table 2.  Variables Used in OLR Analysis.


Dependent variable Independent
(the factor) variables(covariates) Categories Abbreviations

PISA score levels in Education level of Master’s, doctoral, or [edu_father=1]


math: father/mother equivalent level [edu_mother=1]
Level A (the (categorical) Bachelor’s or equivalent level [edu_father=2]
highest) [edu_mother=2]
Level B Postsecondary nontertiary [edu_father=3]
Level C education [edu_mother=3]
Level D Upper secondary education [edu_father=4]
Level E
Family background

[edu_mother=4]
Level F (the lowest)
Lower secondary education [edu_father=5]
Level Z (sublevel)
[edu_mother=5]
Primary education [edu_father=6]
[edu_mother=6]
No education [edu_father=7]
[edu_mother=7]
Occupation status of Managers and professionals [occup_father=1]
father (categorical) Technicians, associate [occup_father=2]
professionals, clerical
employees, and armed forces
Service sector employees [occup_father=3]
Workers [occup_father=4]
Time of computer Minutes per day TIMEINT
use (continuous)
Home possessions Possession index (high scores HOMEPOS
(continuous) represent high number of
Home assets

items)
Home educational Educational resource index HEDRES
resources (high scores represent high
(continuous) number of items)
ICT availability at Availability index (high scores ICTHOME
home represent high number of
items)
Urban effect

Type of settlement Metropolis [metropolis = 1]


(categorical) City [city = 2]
Town [town = 3]
Small town [small town = 4]

Note. OLR = ordinal logistic regression; PISA = Program for International Student
Assessment; ICT = information and communications technology.

The results of this model are presented in the section “The Class-Effect in
Academic Achievement in Turkey.”

Second Phase: GWR Model


A local analysis using statistical technique of GWR is undertaken in the study
as the second phase of analyses. GWR is a simple model that extends the
Ataç 15

traditional regression by allowing local rather than global parameters to be


estimated and measuring spatial nonstationarity (Fotheringham, Charlton, &
Brunsdon, 2001). The OLR model, for instance, improperly measures the
spatial interactions, and one or more relevant variables are either omitted
from the model or are represented by an incorrect functional form
(Fotheringham, Charlton, & Brunsdon, 1997, in Huang & Leung, 2002;
Fortheringham, 1997). In regression models where the cases are geographical
locations, regression coefficients may not remain fixed over space either.
GWR is, thereby, introduced as a technique for exploring these phenomena
(Brunsdon, Fotheringham, & Charlton, 1998; Fortheringham, 1997).
The logic behind local and global regression models and the examples are
introduced by Fotheringham, Brunsdon, and Charlton (2002); Fotheringham,
Charlton, and Brunsdon (1996); Brunsdon et al. (1998); and Fotheringham
et al. (1997). The principle of GWR model is simply based on the first law of
geography introduced by Tobler (1970, in Miller, 2004) as “everything is
related to everything else, but near things are more related than distant
things.” In the calibration of GWR, one parameter is estimated for the rela-
tionship between each independent variable and the dependent variable, and
this relationship is assumed to be constant across the study region. Instead of
producing a single global average parameter estimate for each relationship,
GWR produces a set of local parameter estimates that can be mapped as well
(Fotheringham et al., 2001).
In the analysis of the “geography effect” in academic achievement in Turkey,
the relationships between the academic achievement and socioeconomic factors
are generally assumed to be stationary in space. When a single regression equa-
tion has been employed in the study, it produces a “global” relationship that
might not be valid over the entire study area. In fact, it is reasonable to assume
that the relationships between academic achievement and socioeconomic fac-
tors at district level are different in Turkey. In other words, parameters of the
regression models are different in different districts, and every district in Turkey
has its unique local regression parameters representing the relationships. GWR
is, on this basis, an exploratory technique mainly intended to indicate where
nonstationarity is taking place on the map varying from their global values. With
reference to Bivand, Pebesma & Gomez-Rubio (2015), its basis is the concern
that the fitted coefficient values of a global model may not represent detailed
local variations in the data adequately. Yet, it differs in not searching for local
variation in data space but by moving a weighted window over the data,
­estimating one set of coefficient values at every chosen fit point. GWR is, there-
fore, a relatively technique that extends an OLR model of Equation 2 by a­ llowing
local variations in rates of changes. The coefficients in the extended model are
­specific to location i rather than assumed to be constant (Huang & Leung, 2002).
16 Education and Urban Society 00(0)

An OLR model can be expressed as

P
Yi = a0 + ∑a x
k =1
k ik + θ1 , (2)

where i = 1, , n and Y is the dependent variable represented as a linear com-


bination of independent variables Xk, k = 1, 2; P; a0; a1 , , a p are parameters
and θ1 is independent normally distributed error term with zero mean and
constant variance.
It is important to note that the parameters in Equation 2 are assumed to be
the same across the study area in the OLR model. However, this assumption
is not always valid because different locations may have different parameters
(Huang & Leung, 2002). GWR makes the weighting system dependent on the
location in geographical space and, therefore, allows local rather than global
parameters to be estimated. The parameter estimates become specific to loca-
tion i and the GWR model can be expressed as

P
Yi = ai 0 + ∑a
k =1
ik xik + θ1 , (3)

where i = 1, 2, , n and aik is the value of the kth parameter at location i
(Bivand, 2015; Fotheringham et al., 1997).
Spatial variation of academic achievement in Turkey may exist as there
are many variations of socioeconomic parameters in the districts of Turkey.
For example, the difference in fertility rates in the west and in the east may
affect the level of academic achievement differently in different districts of
the country. In the article, the technique of GWR is employed to explore the
relationships between the achievement in UEE and various factors over the
country. There are, without question, many aspects such as social, economic,
geographical, financial, or even historical factors related to the academic
achievement in Turkey. Considering the data availability at district level, the
main factors affecting the academic achievement in UEE are selected via
stepwise regression model (Huang & Leung, 2002). The determinant factors
of academic achievement in Turkey, therefore, include the share of female
illiterates (FEMILLIT) and the share of university graduates (UNI) as the
strong indicators of educational development in the related district, the share
of young population (YOUNG) and the child–woman ratio which represents
fertility rates (CWR), and the share of migrated population (MIGRANT) as
an indicator of the level of migration in the related district. The results of this
Ataç 17

model are presented in the section “Geographical Variations in Academic


Achievement in Turkey.”

The Class-Effect in Academic Achievement in


Turkey (OLR Analysis)
With class-effect, it is referred here to the socioeconomic background of the
students which is defined in the article with the indicators of home assets,
family background, and the size of settlement where the students live in.
OLR results for the effect of home assets, family background, and the size of
settlement on academic achievement are presented in Table 3. The table
shows specifically about the relationship between the explanatory variables
and the outcomes. The coefficients are used to predict the probability of the
first order. The basic question that OLR analysis seeks to answer is the
strength of the relationship between ranked PISA score levels and socioeco-
nomic backgrounds of the students. To interpret the relations between PISA
levels and the variables, odds ratios (probability of the dependent variable)
need to be evaluated. To do so, the variables which have the probability value
lower than .05 are eliminated first. In this regard, it is only the statistically
significant categories which are evaluated (bold values in the table) in the
analysis. Interpretations of this table are based on the reference groups shown
in the table.
According to Table 3, from the most to the least important variables on
educational achievement are family background, home assets, and urban
effect, respectively. There are three subvariables under family background.
These are occupation of father, education level of father, and education level
of mother. The table shows, first and foremost, that for the variable of occu-
pation of father, workers (occup_father=4) is the reference group. Other sig-
nificant categories are, therefore, evaluated based on this variable. Based on
the variables of occupation of father in the table, it can be said that the stu-
dents whose fathers are included in the groups of technicians and profession-
als (occup_father=2) and managers (occup_father=1) are far more successful
at PISA test than that of found in the occupation group of workers. The ratio
of this difference is 1.24 for technicians and professionals, whereas it is 1.37
for the white-collar managers, which simply means that as far as the occupa-
tion of father is concerned for academic achievement in Turkey, being in
high-status occupation group matters.
This table also depicts how children of high-educated parents are more
successful than the ones who have low-educated parents. For the variable of
education of father, the reference group is the fathers with no education (edu_
father=7). It is shown in the table that the students whose fathers have master’s,
18 Education and Urban Society 00(0)

Table 3.  OLR Results for Home Assets, Family Background, and Urban Effect on
Academic Achievement.
Standardized Factor
Odds factor change in Confidence
Determinants Coefficienta SD Significance ratiob changec percentd intervals

Home assets
TIMEINT −.002 .000 .000 1.00 1.00 36.73 [−2.22, −1.70]
HOMEPOS .327 .049 .000 1.39 1.02 51.00 [0.23, 0.43]
HEDRES .133 .040 .001 1.14 1.01 42.02 [0.05, 0.21]
CTHOME .036 .032 .259 1.04 1.00 38.15 [−0.02, 0.10]
Family background
[occup_father=1] .314 .100 .002 1.37 1.03 50.37 [0.11, 0.51]
[occup_father=2] .217 .096 .025 1.24 1.02 45.68 [0.02, 0.40]
[occup_father=3] .097 .061 .111 1.10 1.01 40.54 [−0.02, 0.21]
[occup_father=4] 0a Reference group
[edu_father=1] .931 .156 .000 2.54 1.16 93.33 [0.60, 1.20]
[edu_father=2] .372 .142 .009 1.45 1.05 53.36 [0.09, 0.65]
[edu_father=3] .579 .139 .000 1.78 1.08 65.66 [0.30, 0.85]
[edu_father=4] −.084 .214 .693 0.92 0.98 33.81 [−0.50, 0.33]
[edu_father=5] .082 .123 .504 1.09 1.01 39.95 [−0.15, 0.32]
[edu_father=6] .074 .117 .525 1.08 1.01 39.63 [−0.15, 0.30]
[edu_father=7] 0a Reference group
[edu_mother=1] .874 .162 .000 2.40 1.15 88.19 [0.55, 1.19]
[edu_mother=2] .473 .143 .001 1.60 1.07 59.01 [0.19, 0.75]
[edu_mother=3] .766 .124 .000 2.15 1.10 79.11 [0.52, 1.09]
[edu_mother=4] −.393 .275 .153 0.67 0.90 24.83 [0.93, 0.14]
[edu_mother=5] .135 .102 .186 1.14 1.01 42.11 [–0.65, 0.33]
[edu_mother=6] .460 .082 .000 1.58 1.04 58.29 [0.29, 0.62]
[edu_mother=7] 0a Reference group
Urban effect
[metropolis = 1] −2.072 .567 .000 1.13 0.31 4.63 [0.03, 0.25]
[city = 2] −1.123 .277 .001 0.66 0.73 11.97 [0.12, 0,66]
[town = 3] −.420 .334 .000 0.33 0.87 24.17 [0.23, 1.20]
[small town = 4] 0a Reference group

aParameter estimate = b value.


bOdds ratio = Exp(b).
cStandardized factor change = Exp(b × SD).
dFactor change in percent = Exp(b – 1) × 100.

doctoral, or equivalent degree are 2.54 times more successful than those
whose fathers have no education. As to the reference group of “no education,”
this ratio is 1.45 for the students whose fathers have bachelor or equivalent
level and 1.78 for those whose fathers have upper secondary education.
Taking these figures into account, it can be concluded that education level of
the father has a great impact on academic achievement of the students and
Ataç 19

this relation is more apparent for higher education. As far as the education
level of mother is concerned, the table shows similar results with the educa-
tion level of father in general. However, the details give interesting findings
for the effect of occupation of mother on academic achievement of the stu-
dents. For this variable, the reference group is the mothers with no education
(edu_mother=7). As the table indicates, there are four levels of education,
which are considerably effective in academic achievement as reference to the
level of “no education.” First, it is seen from the table that even a small
increase in education level of the mother matters in high success of the students
in Turkey. In concrete terms, even the students whose mothers have primary
school education (edu_mother=5) are 1.14 times more successful than the
ones who have uneducated mothers. It is important to remind here that the
same group for the fathers gives statistically insignificant ratios, which means
that primary school education matters in academic achievement only for the
mothers of the students. This refers to the necessity of making a distinction
between education levels of the mothers and the fathers that have a positive
effect on academic achievement of the students. One can infer from the OLR
results that for the fathers it is the highest education level, which takes a
greater role in the high success of the students, whereas even a primary school
education level of the mothers can enhance the level of the academic achieve-
ment of the students. This basically means that it is specifically low level of
education of the mothers, which has a pronounced negative effect on aca-
demic achievement of the children and vice versa. Smiths and Hoşgör (2006)
explain this in another way. According to their work on analyzing the effect
of family background on school enrollment of the children in Turkey, moth-
ers who completed primary education do their best to realize the same for
their daughters, probably because they are aware of the importance of female
education. On this ground, it can be claimed that the strong effect of educa-
tion level of mother on academic achievement is both the reason and result of
the high correlation between socioeconomic background of the children and
their educational achievement at school.
Home assets as a strong indicator to understand the economic and social
background of a student matters in educational achievement as well. As can
be seen from the table, it can be inferred that home assets are as effective as
living in the metropolis or a high-status occupation of a father for educational
achievement of a student. OLR results show in detail that it is, respectively,
home possessions (nearly the same ratio with the highest status occupation
group of a father, managers), home educational resources (nearly the same
ratio with living in a metropolis), and time of computer use. But what needs
to be paid more attention in this relation is that economic background of the
children is as effective as the social background of their families in Turkey.
20 Education and Urban Society 00(0)

OLR results also make it clear that there is an “urban effect” on academic
achievement for Turkey. The parameters of urban effect given in Table 3
show that type of settlement that the students are located in is one of the most
important variables on educational achievement. Students who live in
metropolises are more successful than the ones who live in small cities,
towns, or small towns. This finding will be clear in the second phase of analy-
ses (GWR analysis with UEE results) where academic achievement and
socioeconomic parameters are examined in a geographical-based analysis.
This finding is the result of the residential and regional inequalities in Turkey.

Geographical Variations in Academic Achievement


in Turkey (GWR Analysis)
GWR analysis is run for the districts of all cities in Turkey (Figure 1) to
understand spatial variations as well as the causal relations between academic
achievement and socioeconomic parameters such as education level, migra-
tion, fertility, and young population.
Before analyzing spatial variations in the determinants of academic
achievement in Turkey, the regression model (OLS) representing the average
relationships of 968 districts in Turkey between the academic achievement
(success in UEE) and socioeconomic parameters is obtained. The intercept
via OLS model of the variables is obtained as follows:

Y( s ) = b1 + b2 ⋅ FEMILLIT + b3 ⋅ UNI + b4 ⋅ YOUNG


(4)
+ b5 ⋅ CWR + b6 ⋅ MIGRANT.

The coefficient of determination for this OLS model is R2 = .820, which


means that the OLS model explains a substantial amount of (82%) the varia-
tion in percent census undercount. The estimated OLS regression coefficients
are as follows: b1 = 25.6, b2 = −1.4, b3 = 2.0, b4 = −1.3, b5 = −0.4, and b6 =
0.5. The variables are significant at .001 level. The new OLS equation with
the coefficients is, thereby, given below:

Y( s ) = 25.6 + ( −1.4 ) ⋅ FEMILLIT + 2.0 ⋅ UNI


(5)
+ ( −1.3) ⋅ YOUNG + ( −0.4 ⋅ CWR + 0.5 ⋅ MIGRANT ) .
Based on these coefficients of global regression analysis, it can be claimed
that university graduation in a district (which can be interpreted as a highly
educated environment) has probably the highest explanatory power to
describe academic achievement in Turkey. This is followed, respectively, by
Ataç 21

Figure 1.  Province and district borders of Turkey.

female illiteracy, young population, child–woman ratio, and migrants. Among


all the variables, it is the female illiteracy, young population, and child–
woman ratio, which have negative relation with academic achievement. On
the contrary, university graduation and migrated population are positively
related with the student achievement in UEE. By looking at this equation, one
can also claim that academic achievement is very much dependent on educa-
tion levels, such as university graduation or female literacy, rather than other
factors. However, it is important to bear in mind that these relations may vary
in the geography as far as the local regression model (GWR) is concerned.
To analyze the spatial or local variation of relationships between academic
achievement and the other variables, GWR model in Equation 3 is adopted.
GWR makes the weighting system dependent on the location in geographical
space and, therefore, allows local rather than global parameters to be esti-
mated (Huang & Leung, 2002). Before employing GWR model to the data-
set, spatial autocorrelation (Moran’s I) tool is run for “regression residuals”
to ensure that the residuals are spatially random, which means that the regres-
sion model is well specified for the predictions without missing any explana-
tory variable. The results of spatial autocorrelation of the regression residuals
of the model are given in Figure 2 (on the left). As shown in the figure,
Moran’s index value is 0.01, where Z score is 0.71 and, thereby, the signifi-
cance level of the pattern is defined as neither clustered nor dispersed. This
can also be seen from the cold-to-hot rendered map of standardized residuals
given in Figure 2 (on the right). Standardized residuals have a mean of 0 and
22 Education and Urban Society 00(0)

Figure 2.  Spatial autocorrelation (Moran’s I) results of GWR model.


Note. GWR = geographically weighted regression.

a standard deviation of 1. As far as the standard deviation scores between


−0.5 and 0.5 represent the largest distribution in the map, this model can be
defined as well specified for further predictions.
For the employed GWR model, no data found larger than 30 which may
cause unreliable predictions. In testing the explanatory power of the variables
used in the analysis, local R2 values are also used. Local R2 values range
between .0 and 1.0 indicating how well the local regression model fits
observed y values. As very low and high values indicate the local model is
performing poor, mapping the values can help one to find out the clues about
the variables that may be missing from the regression model or that may not
be useful to explain the dependent variable. The variables used in GWR
model are the ones which give reliable results according to the R2 values.
To better understand the regional variations, the coefficient raster sur-
faces (spatial distributions of parameter estimates) of the explanatory vari-
ables are examined. When GWR model is used to model the explanatory
variables, it is possible to predict the values or understand the factors that
contribute to dependent variable outcomes. It is also possible with this
model to examine the relationships between academic achievement and a
variety of socioeconomic variables of the country. On this basis, examining
the coefficient surfaces given by GWR model helps to see where and how
much variation is presented in the Turkish geography.
As has been explained before, GWR techniques allow mapping statis-
tics, which denote local relationships in the study area and it creates coef-
ficient raster surfaces for the model intercept and each explanatory variable.
Figure 3 depicts the percentage maps of the related variables to see how
percentage values are varying in the country without a causal or relational
Ataç 23

Figure 3.  Percentage maps of academic achievement and the socioeconomic


parameters.

concern. Figure 4, on the contrary, shows the maps of the intercept terms
and spatial distributions of the parameter estimates calculated via GWR
method to understand the local relations between academic achievement
and the socioeconomic variables in Turkey. Based on these maps, it is
apparent that there are significant variations in the relations of academic
achievement and the explanatory variables and “geography matters” for
every map produced.
The coefficient surfaces of model intercept are given in the first map of
Figure 4. Huang and Leung (2002) define intercept terms as the “basic level
of the independent variable” in GWR method. In this case, this variable is
academic achievement (achievement at UEE), and the map of intercepts indi-
cates the fundamental level of academic achievement excluding the effects of
all factors on education across the country. The intercept map of Turkey
shows, first and foremost, that the geography of academic achievement in
Turkey represent a clear spatial variation with higher parameters in the
24 Education and Urban Society 00(0)

Figure 4.  Spatial distribution of intercept and socioeconomic parameters in GWR.


Note. GWR = geographically weighted regression.

eastern and central regions, and lower ones in the western and southwestern
regions of the country. In other words, from east to west, the success in UEE
is getting relatively higher in Turkey, which represents a greater inequality in
education that can be visibly seen in the geography. This map can be seen as
the corrected version of the percentage map of UEE success given in Figure
3. As can be seen from the map, in terms of exam success, there is a greater
gap between the western and the eastern regions of the country. Two of these
sets of maps, therefore, show how academic achievement is inequally distrib-
uted throughout the country.
To understand the effective reasons behind this inequal distribution in the
country, the GWR maps of socioeconomic variables can be analyzed. The
GWR map of female illiterates is given in the second map of Figure 4. The
OLS results show how highly significant the relationship between the suc-
cess in UEE and female illiteracy is (see Equation 5). But the GWR map of
Ataç 25

female illiterates makes it clear that this relationship is also highly visible in
the geography. According to the map, the inverse relationship between female
illiteracy and UEE success is stronger mainly in the eastern, southeastern,
and southwestern parts of the country where the female illiteracy rates are
relatively higher (see second map in Figure 3). In other words, female illit-
eracy is more important in these regions than the other ones. It can also be
seen from this map that from east to west, this relation is getting weaker for
Turkey. For the western cities like Istanbul, Bursa, Ankara, and Izmir, female
illiteracy seems to be inefficient in explaining UEE results. This means that
although in many developed countries, illiteracy becomes one of the ineffi-
cient variables to define any social, economic, or sociological fact, it is still
one of the main determinant features in understanding regional inequalities in
Turkey where this rate is still high in the eastern regions of the country.
As far as education level is concerned, there are greater gaps between the
regions in Turkey, which lays the grounds for the inequal distribution of
academic achievement through the country (see second and third maps in
Figure 3). The third map of Figure 4 indicates the GWR map of university
graduates in Turkey. It is found in the GWR analysis of university graduates
that the parameter estimates of university graduation are positive for all
groups. This means that in every district of Turkey, high education or a
highly educated environment is the strongest indicator of academic achieve-
ment of students. Recalling Equation 5, the parameter estimate of university
graduation in OLS model is found 2.0, which means that by not only global
measure but also local or spatial analysis, two variables are found strongly
related in Turkey. The spatial distribution of GWR parameters in university
graduation shows that this direct relation between university graduation and
academic achievement is stronger in the western (mainly in Aegean Region),
northwestern (around Antalya, Burdur, Isparta, Denizli, and Muğla), and
southern regions (like Artvin, Rize, and Sinop) of the country. This means
that in these regions, although university graduation rates are not relatively
higher compared with the other regions of the country (see the percentage
map of university graduation in Figure 3), a highly educated environment is
rather effective in academic achievement. Referring to the previous map on
GWR parameters of female illiterates, especially for the northwestern part
of the country that represents the highest parameters in both maps, it can be
stated that in this area the share of educated population has the most impor-
tant effect in academic achievement. In other words, among all the regions,
this part of the country known with its tourism potential in Turkey is the one
where academic achievement means a highly educated environment such as
highly educated families or parents. This finding proves the strong relation
between high socioeconomic status of the parents and the academic
26 Education and Urban Society 00(0)

achievement of a student found in the OLR results (see the findings of Phase
1), and these maps show that it can be clearly seen in the southwestern part
of the country.
The spatial pattern of the young population parameters in GWR is given
in the fourth map of Figure 4. It is interesting to note that the GWR map of
the young population is quite similar to its percentage map in terms of the
main divisions between eastern, central, and western regions. This is mainly
because of the fact that the divisions are very distinct as far as the young
population or the fertility rates are considered in Turkey (see percentage
maps of young population and CWR in Figure 3). As shown in its percentage
map in Figure 3, it is mainly the eastern region of the country, which hosts the
greater share of young population. However, when one looks at the GWR
map of young population which depicts local similarities, it becomes appar-
ent that the strength and the form of this relation are quite different in the
regions. For instance, according to the categories defined in GWR map of
young population, it is found that the parameters were negative in the first
and second area groups located in the eastern and central regions. It simply
means that in the regions where there is a great amount of young population,
it is hard for them to be involved in the successful group of UEE. In other
words, one can claim that the greater the number of the youth, the lower the
chance of being successful at UEE in Turkey, and its highest effect can be
observed from the eastern regions of the country. On the contrary, the GWR
parameters of young population have a positive effect on academic achieve-
ment in the west. Because of the fact that these areas have relatively smaller
number of young population compared with the eastern and central regions,
the bigger the number of youth, the more their change to be successful at
UEE. Looking at the maps, therefore, one can claim that academic achieve-
ment is highly related with youth population, even if it is in different ways.
This relation is highly and negatively effective in the eastern and central
regions, whereas it is positively effective in the west.
Spatial distribution of the parameters of child–woman ratio which can be
seen as the strong indicator of fertility rate is presented in the fifth map of
Figure 4 and its percentage map is given in the fifth map of Figure 3. GWR
map of CWR shows a pattern, which is quietly different from that of young
population, and this indicates that the effect of the number of children and the
number of young population behaves different on academic achievement in
Turkey. This is mainly because of the fact that in developed western regions
where there is a variety of social, cultural, and economic opportunities and
high-performing schools, greater numbers of young population may become
an advantage in the larger cities. But this is exactly not the case for children
population. High number of children in a region means high fertility rates,
Ataç 27

which can be easily associated with the least developed regions such as low
levels of education, poor health care, or low status of women in social and
economic life in a region. As has been seen in the percentage maps in Figure
3, fertility is still high in the eastern regions of Turkey. Although fertility rates
are expected to decline, in the east, it remains considerably higher than the
rest of the country. When the GWR parameters on child–woman ratio are
analyzed, it is observed that in every region of the country, the higher the
fertility rates, the lower the chance to be successful at UEE. This finding can
also be seen from the OLS equation (see Equation 5), where the effect of
child–woman ratio on academic achievement is defined by a negative rela-
tion (–0.4). What is more important about this relation is that it is highly vis-
ible and effective in the eastern and southeastern regions of the country as
GWR map illustrates. According to the map, in the eastern and southeastern
regions of the country, the negative relation between the fertility rates and
academic achievement represents the highest scores of GWR, which means
that the higher fertility rates are directly related not only with the low levels
of education of the parents but also with educational failures of the students.
The lower scores observed in the western and southern regions and the mod-
erate levels seen in central regions prove this finding as well. From east to
west, fertility rates and their effect on academic achievement get relatively
lower. On this basis, one can claim that higher fertility rate is still one of the
strongest indicators of educational failure or educational inequalities in
Turkey even in the local level.
The sixth map of Figure 4 illustrates the spatial variation of migrated
population in GWR model. Although at district level, spatial distribution of
migrated population represents a dispersed pattern compared with the other
variables, its percentage map in Figure 3 shows that there are some main
clusters around the largest cities of the country such as Istanbul, Ankara,
Izmir, and Bursa in the west, and Antalya and Mersin in the south where the
share of the migrants is considerably higher. But as far as the effect of
migrated population on academic achievement is concerned, GWR map
depicts interesting findings. The map shows where migration has a positive
effect on student achievement in UEE and where it has no meaning at all.
Put more concretely, it is apparent from the map that it is only the central
corridor from north to south where migration has a negative effect on aca-
demic achievement. These regions are the central Anatolian cities where
out-migration rates are relatively high. On this account, it can be claimed
that out-migration has a negative effect on academic achievement in this
larger area. But although the eastern and southeastern regions are known
with the out-migration trends, it seems that they do not generally affected by
this trend in terms of educational achievement. On the contrary, even a small
28 Education and Urban Society 00(0)

number of migrated population effects education positively in this area. This


is more strongly experienced in the western regions of the country where
in-migration trends can be seen both as the cause and result of educational
achievement. It can be defined as a “reason” because it has been known that
one of the purposes of the people who move from eastern or central areas to
the more developed western regions is the high-performing schools or the
variety in educational opportunities of the west. However, it is also a “cause”
because the economic, social, cultural, and educational opportunities of the
west contribute not only the student achievement but also the quality of the
child’s experience at school. On this basis, it can be suggested that as far as
migration is concerned, it is only good for academic achievement in the
west, especially in the regions of Aegean and Marmara where the GWR
scores are the highest.

Conclusion
Many of Turkey’s most disadvantaged children grow up without the skills
needed to thrive in the new century as the findings of the article indicate.
Whether in educational achievement between income groups or socioeco-
nomic groups or across geographical locations, inequality persists and
becomes deeper in Turkey. Low levels of education among disadvantaged
children create less visible but long-term social, economic, and even cultural
problems, especially in an economy where higher skills are more valuable
than they were before. As has been explained before, in the related literature,
some researchers claim, on one hand, that educational inequality is due to
family background or social or economic class. The others, on the other hand,
argue that the state should take most of the responsibility for the inequal dis-
tribution of educational achievement. This study shows, first and foremost,
that from the most to the least important variables on educational achieve-
ment in Turkey are family background, home assets, and urban effect, respec-
tively. In detail, it has been also noticed that as far as the occupation of father
is concerned for academic achievement in Turkey, being in high-status occu-
pation groups such as managers and professionals brings more change for the
students to be included in the groups of high achievement. Besides, it is not
only the occupation of father but his level of education that has a greater
effect on academic achievement of a student. The higher the degree that a
father has, the higher his child’s academic achievement is. But the article also
shows that education level of mother is more effective than that of father on
the academic achievement of the students. The findings indicate that even a
small increase in education level of the mother can enhance the academic
achievement of a child. What the article also underlines with its findings is
Ataç 29

that economic background of the students is as effective as their social back-


ground on academic achievement. Regression models show that home pos-
sessions are effective in the same way with the occupation of father or the
birthplace on academic achievement. According to Duman (2008), if high-
quality education were freely available to all children, family background
would play a less significant role in determining the academic achievement.
If educational opportunities are limited by economic or social backgrounds,
education can in fact exacerbate the differences in initial conditions rather
than reduce them. This is exactly what is concerned for Turkish case as the
article shows.
What makes the further findings of the article interesting is to see that
besides inequalities based on social and economic backgrounds of the chil-
dren, regional educational inequalities are also stricter than it is thought to be
as far as academic achievement is concerned. As there are greater gaps
between the regions of Turkey in terms of social, economic, and cultural
aspects, the geography of academic achievement is also an important issue,
which is therefore worth to be analyzed in detail. The findings of the second
phase of the article where the effect of socioeconomic parameters on aca-
demic achievement is analyzed at district level show that there appear to be
regions in which students are disadvantaged in UEE just because of the
region or location that they live in. In a country where university graduation
has a strong effect on class positions, this issue needs more attention of the
government, educators, or the academics. As the article demonstrates, female
illiteracy, a high level of fertility rate, and low level of education have an
adverse effect on academic achievement in Turkey. Conversely, high levels
of university graduates which mean a highly educated environment in an area
appear to have a positive effect on achievement at UEE in almost every
region of the country. Among socioeconomic variables tested by GWR, it is
only the youth and migrated population that can be both advantageous and
disadvantageous depending on the geography. It is, of course, impossible to
provide a causal or a direct link between achievement at UEE and these
socioeconomic factors, but the article gives the circumstantial evidence
because the family background and the socioeconomic environments of the
students are considered.
The article thus finds out that for Turkish children, where they live is a
powerful predictor of academic achievement. According to the key findings
of the article, students from high socioeconomic backgrounds with more edu-
cated parents and living in larger cities especially in the western regions of
the country are more likely to be attended at higher education. This means
that greater numbers of students in Turkey are underachieving at UEE
because of the regional inequalities in education as well as the inequalities
30 Education and Urban Society 00(0)

based on socioeconomic differences, which are having a long-term damaging


effect on life chances. Family background has long been recognized as a key
factor influencing educational achievement in almost everywhere in the
world, and the article also proves this finding for Turkish case once more.
This is a problem per se that can be tackled by developing policies which, for
instance, introduces funding to the children from disadvantaged background
or increasing the level of attainment at school. But the study shows that the
main concern should be more on the regional inequalities in Turkey, which
are harder to tackle with.
Again with reference to the findings of the article where it is underlined
that even a small increase in the education level of a woman can enhance
her child’s academic achievement, it can be claimed that reaching low-
educated female population would be a major concern of policy makers in
Turkey. One of the ways to tackle with lower academic achievement in
Turkey may be to increase the level of schooling of female population.
However, this solution can only be effective if it is supported with the fac-
tors by which the school enrollment and completion of female population
are furthered.
It is of course right to say that a high-level education degree is not enough
to solve all the social or economic problems in a country or does not guaran-
tee a fair social order as well. But high-level education is still the safest way
to develop talent and build the skills for the youth and their economic and
social well-being, especially in a country like Turkey where even a small
enhancement in the degree of education matters in a positive way. Although
the gap between the rich and the poor has slightly tightened in Turkey over
the past 15 years, income inequality is still one of the greatest problems of
Turkey. As one of the latest newsletters of TURKSTAT (2015) describes, the
share of the top quintile by the equalized household disposable income is
45.9% while the share of the bottom quintile is 6.2%, which means that total
income received by the richest population of the country is 7.4 times greater
than that of the poorest population of the country. Income inequality has
always been a good indicator to understand the capacity of the physical and
human capital in a country (Goldthorpe, 1980). As the main concern is educa-
tion here, it can be safely argued that it is mainly the countries with higher
income inequality where socioeconomic backgrounds of the students take a
greater role in their academic success. The United Kingdom, the United
States, and Italy are some of the best examples to illustrate this according to
the PISA results (OECD, 2012). Based on the research of PISA, in countries
where income inequality is not that much higher such as Nordic countries
(Finland, Norway, and Denmark), it is observed that rather than their socio-
economic backgrounds, it is mainly the children’s own abilities and skills that
Ataç 31

define their educational achievement at school. In Asian countries such as


Korea and Japan which put equity in education at the heart of their national
development strategies, it is observed with the PISA tests that even the stu-
dents from lower socioeconomic backgrounds performed better than that of
any other countries. These countries’ ability on innovation and technology is
therefore not a coincidence and heavily based on their way of employing
equity in education.
In an era of growing inequality, decision makers who wish to gain a long-
term economic growth and build a strong human capacity should, therefore,
focus on equity as a state policy and give children equal opportunities in
education to raise a more innovative, technology-oriented, and qualified
future generation. They must lay the grounds for them to achieve higher lev-
els of education and eventually high-qualified jobs, which are seen today as
the safest way of more affluent lives. Turkish education system should also
bring up innovative children who are capable of contributing to new tech-
nologies. To do so, policies that enable everyone to benefit equally from edu-
cational opportunities to reduce the negative impact of regional inequalities
on academic achievement are needed. In these policies, special attention
should be paid to the children from lower backgrounds, rural areas, and
female population. Investing in the equal provision and high quality of edu-
cation can barely guarantee a long-term enhancement both in human capital
and economic growth.

Declaration of Conflicting Interests


The author(s) declared no potential conflicts of interest with respect to the research,
authorship, and/or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publi-
cation of this article.

Note
1. These variations are examined by PISA (Program for International Student
Assessment) tests of Organisation for Economic Co-Operation and Development
(OECD), which help to compare educational attainment and achievement in var-
ious countries. For more, see OECD (2012).

References
Arnot, M., & Barton, L. (Eds.). (1992). Voicing concerns: Sociological perspectives
on contemporary education reforms. Oxfordshire, UK: Triangle Books.
32 Education and Urban Society 00(0)

Barro, J. R. (2013). Education and economic growth. Annals of Economics and


Finance, 14, 277-304.
Bernstein, B. (1977). Class, codes and control: Towards a theory of educational
transmission (Vol. 3). London, England: Routledge.
Bivand, R. S., Pebesma, E., & Gomez-Rubio, V. (2015). Geographically weighted
regression. Applied Spatial Data Analysis with R, New York,NY: Springer-Verlag.
Blackledge, D., & Hunt, B. (1985). Sociological interpretations of education. London,
England: Croom Helm.
Bobbitt-Zeher, D. (2007). The gender income gap and the role of education. Sociology
of Education, 80, 1-22.
Bourdieu, P., & Passeron, J.-C. (1977). Reproduction in education, society and cul-
ture. Beverly Hills, CA: SAGE.
Brunsdon, C., Fotheringham, A. S., & Charlton, M. (1998). Geographically weighted
regression: Modelling spatial non-stationarity. The Statistician, 47, 431-443.
Butler, T., & Hamnett, C. (2007). The geography of education: Introduction. Urban
Studies, 44, 1161-1174.
Cabrera, A. F. (1994). Logistic regression analysis in higher education: An applied
perspective. In J. C. Smart (Ed.), Higher education: Handbook of theory and
research (pp. 225-256). New York, NY: Agathon Press.
Duman, A. (2008). Education and income inequality in Turkey: Does schooling mat-
ter? Financial Theory and Practice, 32, 369-385.
Duygan, B., & Guner, N. (2006). Income and consumption inequality in Turkey:
What role does education play? In S. Altug & A. Filiztekin (Eds.), The Turkish
economy: The real economy, corporate governance and reform and stabiliza-
tion policy. (63-91) London, UK: Routledge Curzon Studies in Middle Eastern
Economies.
Eckland, B. K., & Alexander, K. L. (1980). The national longitudinal study of the
high school senior class of 1972. In A. C. Kerchoff (Ed.), Research in sociology
of education and socialization (Vol. 1, pp. 189-222). Greenwich, CT: JAI Press.
Ferreira, H. G., Gignoux, J., & Aran, M. (2011). Measuring inequality of opportu-
nity with imperfect data: The case of Turkey. Journal of Economic Inequality,
9, 651-680.
Fotheringham, A. S. (1997). Trends in quantitative methods I: Stressing the local.
Progress in Human Geography, 21, 88-96.
Fotheringham, A. S., Brunsdon, C., & Charlton, M. E. (2002). Geographically
weighted regression: The analysis of spatially varying relationships. Chichester,
UK: Wiley.
Fotheringham, A. S., Charlton, M., & Brunsdon, C. (1996). The geography of param-
eter space: An investigation of spatial non-stationarity. International Journal of
Geographical Information Systems, 10, 605-627.
Fotheringham, A. S., Charlton, M., & Brunsdon, C. (1997). Measuring spatial varia-
tions in relationships with geographically weighted regression. In M. M. Fischer &
A. Getis (Eds.), Recent developments in spatial analysis: Spatial statistics, behav-
ioural modelling, and computational intelligence. Berlin, Germany: Springer.
Ataç 33

Fotheringham, A. S., Charlton, M., & Brunsdon, C. (2001). Spatial variations in


school performance: A local analysis using geographically weighted regression.
Geographical and Environmental Modelling, 5, 43-66.
Goldthorpe, J. H. (1980). Social mobility and class structure in modern Britain.
Oxford, UK: Clarendon Press.
Hatakenaka, S. (2006, November). Higher education in Turkey for 21st century:
Size and composition. Retrieved from http://siteresources.worldbank.org/
EXTECAREGTOPEDUCATION/Resources/444607-1192636551820/S._
Hatakenakas_report_on_Higher_Education_in_Turkey_for_21st_Century_
Nov_2006.pdf
Huang, Y., & Leung, Y. (2002). Analyzing regional industrialization in Jiangsu
province using geographically weighted regression. Journal of Geographical
Systems, 4, 233-249.
Lall, R., Campbell, M. J., Walters, S. J., & Morgan, K. (2002). A review of ordinal
regression models applied on health-related quality of life assessments. Statistical
Methods in Medical Research, 11, 49-67.
Lynch, K., & O’riordan, C. (1998). Inequality in higher education: A study of class
barriers. British Journal of Sociology of Education, 19, 445-478.
Maloutas, T. (2007). Middle class education strategies and residential segregation in
Athens. Journal of Education Policy, 22(1), 49-68.
Miller, H. J. (2004). Tobler’s first law and spatial analysis. Annals of the Association
of American Geographers, 94, 284-289.
Moore, R. (2004). Education and society: Issues and explanations in the sociology of
education. Cambridge, UK: Polity Press.
Nelson, R. R., & Phelps, E. S. (1966). Investment in humans, technological diffusion,
and economic growth. The American Economic Review, 56, 69-75.
Organisation for Economic Co-Operation and Development. (2012). Key Findings of
PISA. Retrieved from http://www.oecd.org/pisa/aboutpisa/
Peng, C. Y., Lee, K. L., & Ingersoll, G. M. (2002). An introduction to logistic regres-
sion analysis and reporting. The Journal of Educational Research, 96, 3-14.
Pink, W. T., & Noblit, G. W. (Eds.). (1995). Continuity and contradiction: The futures
of the sociology of education. Cresskill, NJ: Hampton Press.
Raftery, A. E., & Hout, M. (1993). Maximally maintained inequality: Expansion, reform,
and opportunity in Irish education, 1921-75. Sociology of Education, 66, 41-62.
Smiths, J., & Hoşgör, A. G. (2006). Effects of family background characteristics
on educational participation in Turkey. International Journal of Educational
Development, 26, 545-560.
Social Exclusion Unit. (2001). Preventing social exclusion. London, England:
Author.
State Planning Organization–T.R. Prime Ministry. (2010). Millennium Development
Goals Report Turkey 2010 (In coordination with United Nations). Retrived from
https://issuu.com/undpturkiye/docs/turkey_2010_millenium_development_g
Statistics Solutions. (2016). A manual on dissertation statistics in SPSS. Retrieved from
http://www.statisticssolutions.com/wp-content/uploads/2013/10/SPSS-Manual.pdf
34 Education and Urban Society 00(0)

Stevens, P., & Weale, M. (2003, August). Education and economic growth. London,
England: National Institute of Economic and Social Research. Retrieved from
http://cee.lse.ac.uk/conference_papers/28_11_2003/martin_weale.pdf
Tansel, A., & Bircan, F. (2010, December). Wage inequality and returns to education
in Turkey: A quantile regression analysis (Discussion Paper No. 5417) Retrieved
from https://www.econstor.eu/bitstream/10419/51627/1/670469890.pdf.
Tobler W. (1970). A computer movie simulating urban growth in the Detroit region”.
Economic Geography, 46(2): 234-240.
Tomul, E. (2007). The change in educational inequity in Turkey: A comparison by
region educational planning. A Journal Dedicated to Planning, Change, Reform,
and the Improvement of Education, 16(3), 16-25.
Tomul, E. (2011). Measuring regional inequality of education in Turkey: An evalua-
tion by Gini index. Egitim ve Bilim, 36(160), 133-143.
Turkish Statistical Institute. (2015, September). Income and Living Conditions Survey.
Retrieved from http://www.turkstat.gov.tr/PreHaberBultenleri.do?id=18633

Author Biography
Ela Ataç received bachelor’s degree (graduated with honor) in city and regional plan-
ning from Gazi University, Faculty of Architecture, Department of City and Regional
Planning in 2005. She received master’s degree from the same department in 2008
and PhD degree from Middle East Technical University, Department of City and
Regional Planning, in 2014. From 2005 to 2015, she worked as research assistant and
instructor in Gazi University, Department of City and Regional Planning. She has
been working at TED University, Department of City and Regional Planning, as assis-
tant professor since 2016. Her main research areas are urban geography, population
geography, urban sociology, segregation, inequality, and qualitative and quantitative
methods in social sciences. She has various publications in the international and
national journals on the subjects of residential segregation, inequality, poverty, and
the use of qualitative and spatial techniques in urban studies.

You might also like