International Journal of Educational Development 81 (2021) 102348

Contents lists available at ScienceDirect

International Journal of Educational Development

journal homepage: www.elsevier.com/locate/ijedudev

How far the apple falls from the tree: Intergenerational transmission of
educational attainment in Indonesia
Syed Hassan Raza a, *, Ugur Aytun b
a
School of Economics, Quaid-i-Azam University, Islamabad, Pakistan
b
Department of Economics, Kutayha Dumlupinar University, Turkey

A R T I C L E I N F O A B S T R A C T

JEL classification: We use five waves of the Indonesian Family Life Survey (IFLS) to create unique father-son matched data of the
I21 male population (aged 25–53). Using this data, we investigate the evolution of intergenerational transmission of
J62 educational attainment in Indonesia over time, by ethnicities (six major ethnic groups), and by regions (twelve
J28
provinces). To estimate the persistence in education (measured as schooling years of the father as a predictor of
Keywords: schooling years of the son), we use coefficient of correlation and regression coefficient methods. Our results from
Intergenerational education persistence
the regression coefficient suggest that persistence in Indonesia is declined over time. Whereas, the results from
Correlation decomposition
Indonesia
the coefficient of correlation reveals no secular trend in education persistence over time. Thus, to understand the
reason behind the secular trend in the correlation coefficient we decompose the correlation following Checchi
and García-Peñalosa (2008). The decomposed intergenerational correlation shows that although the persistence
is declined at the lower-end of fathers’ education distribution, it is increased at the top-end of the distribution.
Therefore, we argue that the coefficient of correlation should be used with caution while analyzing the education
policy particularly the persistence in education. Thereafter, we document the stationary distribution and find the
conclusive evidence of non-convergence in educational attainment over time among both, ethnicities and
regions.

1. Introduction circumstances on outcomes, but those outcomes that are sensitive to the
effort. Hence the concept of equality of opportunities is supported by the
Over the last few years, the concept of equal opportunity has become majority of the policymakers, think tanks, and the general public1 .
a core subject of every in-depth theoretical and empirical development One of the key apparatuses to analyze the inequality of opportunities
in economic literature. Unlike the traditional theories that stresses on or the welfare of the people in an economy is through intergenerational
the outcome of inequality as a change in consumption, wealth, and in­ persistence in economic status. Higher persistence in economic out­
come, recent literature focuses on equality of opportunity through the comes across generations will aggravate the inequality problem.
prism of circumstances and effort on individual outcomes (Roemer, Persistence may differ across the ethnic groups, genders, and regions.
1998). Factors for which individuals cannot be held responsible are Such a persistence implies the disparity of opportunities among different
defined as circumstances such as economic background, birthplace, and groups. Thus, policymakers need to quantify the transmission of eco­
ethnic group. Whereas, factors which are variables and can be controlled nomic status from one generation to another2 . Education is a prime
by individual exertion are categorized as individual efforts. Years of determinant of economic success over the longer time horizon and is a
schooling and labor supply decisions are two examples. Roemer (1998) pivotal mechanism to social mobility. It is a well-documented fact that
defined that the objective of policy equalizing opportunities is to elim­ parental schooling significantly affects children’s schooling (Hertz et al.,
inate the biased opportunities which came into being as an impact of 2007). But children’s educational attainment is also affected by other

* Corresponding author.
E-mail addresses: [email protected] (S.H. Raza), [email protected] (U. Aytun).
1
“In policy and political discourse, “equality of opportunity” is the new motherhood and apple pie. In its strongest form, the position is that equality of outcomes
should be irrelevant to policy; what matters is equality of opportunity” (Kanbur and Wagstaff, 2014)
2
Some of the recent studies presents the international evidence on the transmission mechanism of educational attainment, implications of persistence and
methodological development in this regard, see, see Azam and Bhatt (2015); Black et al. (2011), and Hertz et al (2007).

https://doi.org/10.1016/j.ijedudev.2021.102348
Received 4 February 2020; Received in revised form 23 November 2020; Accepted 31 December 2020
Available online 17 January 2021
0738-0593/© 2021 Elsevier Ltd. All rights reserved.
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

factors like economic resources, ethnic groups, and regions (Azam and stationary distribution for 12 provinces of Indonesia. We institute, small
Bhatt, 2015; Huang, 2013; Björklund and Salvanes, 2011). Transmission persistence exists in higher asset holdings regions of Indonesia such as
of education from one generation to another, or from parents to chil­ Jakarta, where persistence remains lower than 0.40 overtime. Further­
dren, is considered as a focal indicator of education mobility. Therefore, more, we document in regions with less asset holdings, long term
higher persistence of educational attainment, or lower intergenerational probability suggests that sons are more likely to obtain a university
mobility of education, is regarded as an obstacle to opportunities in the degree in comparison to higher asset holding regions such as Jakarta and
labor market (Black et al., 2011). South Sumatra.
In this paper, we thoroughly study the intergenerational education In last few years, intergenerational mobility has become an active
persistence in Indonesia and track changes in intergenerational persis­ area of research due to the economic implication of persistence between
tence in education across different birth cohorts from as early as 1940 parent’s and child’s education. The literature on intergenerational and
using two widely used measures of educational persistence: a correlation social mobility focuses primarily on income, education and occupational
coefficient (ρ) and regression coefficient (β). It is noteworthy that many choices. Majority of the empirical evidence pertains to North American
studies used regression coefficients argued that a “causal relationship” or Northern European data, but interest in this issue has increased lately
suffers from two major drawbacks. Firstly, they do not account for among development economists as well (cf. World Bank, 2005). Hertz
average schooling across generations. Secondly and more importantly, et al. (2007) argued that it is not feasible, given the existing data in
they do not allow us to analyze the sources of intergenerational mobility developing economies, to estimate the intergenerational persistence of
with respect to subgroups of population. Checchi et al. (2013) argued permanent income. Thus, they suggested using education as an alter­
that the beta coefficient might decrease in a given cohort but this finding native to study the transmission over generations. One of the key ben­
cannot explain that this stems from either compulsory education or efits of using education is that it represents more or less permanent
upward mobility of educated fathers dominating uneducated fathers. characteristics in adulthood. Thus, one of the greatest attempts in this
Since our aim in this study is to analyze the effect of different educa­ regard is attributed to (Hertz et al., 2007) who analyzed 42 countries
tional attainment categories of the father on son’s educational out­ with over 50 years of data. They found large regional differences in
comes, we adopted this methodology based on the correlation educational persistence such as Nordic countries shows the least edu­
coefficient which allows us to decompose the demographic sources of cation persistence and Latin America displays the highest education
mobility. Nonetheless, we put beta coefficients in nation-wide and persistence among those 42 countries. They conclude the average cor­
cohort-wise results for basic comparison between causal and non-causal relation between parents’ and child’s education remains 0.42 in
approach but the focus of our analysis remains on correlation coefficient 1920–1980. Likewise, Daude (2011) used the cross-country dataset for
and decomposition exercise which is non-causal in nature. 18 Latin American countries and ascertained no significant change in
Unlike previous studies on this topic, which relied on coresidence to persistence measure.
identify father-son pairs, our first contribution is to show bias in Indonesia serves as an excellent case study to explore intergenera­
persistence when limiting our data to co-residence parents. We observe tional mobility for various reasons. Firstly, Indonesian education system
the trend in correlation estimates was not clear, however, over the often categorized as high volume, low-quality enterprise which could
period of time, starting from 1941− 1945 it was 0.49 but reduced to 0.44 not achieve country’s ambition to be internationally competitive
in 1986− 1990. This unclarity in the trend of the coefficient of correla­ (Rosser, 2018). Secondly, It has been well-known for being afflicted by
tion do not exist in regression coefficients which shows clear decline in serious ethnic, regional, and other forms of communal conflicts. One of
persistence over 50 years. Thus, in order to understand this discrepancy the key examples is brutal ethno-riots after the collapse of the Suharto
in the trend, we further focus our analysis on correlation decomposition. regime. Though the ethnic hard lining is tamed over the period of time
Through correlation decomposition we found in some of the education through the policy measures taken by the Government of Indonesia
groups, persistence increased remarkably but in other groups it (Aspinall and Klinken, 2011), still ethnic identity remains an important
reasonably curtailed. For example, we observed persistence between cause of social exclusion in Indonesia. This implies that ethnic identity
unschooled fathers and sons declined but for the university education may affect the labor market outcome as well and as we have discussed
group it increased substantially over time. earlier that education is a prime determinant of long terms success in to
Thereafter, we analyzed the six major ethnicities of Indonesia and the labor market.
found higher persistence in bigger ethnic groups in comparison to rest. For example, Bennington and Habir (2003) investigated the human
In particular, we found higher and significant increase in educational resource management in Indonesia and found the existence of ethnic
persistence for Javanese, Sundanese, and Other Sumatrans. But, in up­ discrimination in the labor market. There are at least 200 ethnic groups,
ward mobility Index we observe an increase in all the ethnic groups but only four are considered to be significant numerically: Javanese (45
suggesting an increase in probability that son will attain more education %), Sundanese (14 %), Madurese (7.5 %) and Coastal Malays (7.5 %)
than his father. It is important to note that our findings on ethnicities are (Levinson, 1998). Chinese represents less than 5% of the population but
only suggestive since IFLS is not ethnicity representative data of reportedly controls bigger share of economic activity (Tan et al., 2008).
Indonesia. Later on, we obtained the stationary distribution for long run This is regardless of the fact that right after the independence in 1945,
probability to attain any particular education level. We observe the the Government of Indonesia provided everyone with constitutional
existence of long run differential in probabilities such as for Minang rights as mentioned in the constitution of Indonesia (Article, 28D)
probability of university education is 0.65 but for Batak it is 0.52 which “everyone is equal before the law, have right to work and equal opportunities
shows the existence of inequality among various ethnic groups. This in government.”
could be due to the fact we observe higher upward mobility for Batak in Moreover, Indonesia has achieved remarkable growth over the last
comparison to Minang. Even from our descriptive statistics, we observe few decades, but this growth greatly varies across regions. Balisacan
higher educational attainment in Batak for the most recent cohort while et al. (2003) argued that despite of sustained growth over two decades
comparing it to Minang. This could be due to many reasons but one of and rise in provincial GDP, the growth-poverty nexus appears strong at
the most important reasons is that Batak are treated as native de­ aggregate level. However, the picture is entirely different at regional
scendants of Indonesia and they have certain advantages regardless of level with the existing regional disparities. They documented the critical
they are less in number and this encourages Batak people to attain factors that are responsible for such disparity and they include infra­
higher societal positions via education and jobs (Prasiwi and Susandari, structure, technology, and human capital. Hill (2008) found that the
2015; Ramdani et al., 2015). growth performance of the resource-rich provinces of Indonesia varied
Similarly, we add regional dimension to our analysis and we considerably over the period of 30 years. Likewise, thirdly, Inequality of
captured the persistence estimates, conditional probabilities and opportunities in different provinces of Indonesia is yet another

2
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

important dimension for research. Similarly, India is also a diverse 2. Analytical framework
country from an ethnic and regional point of view, thus (Azam and
Bhatt, 2015) used the Indian Human Development Survey to document Intergenerational transmission of education is captured by the
the persistence over time from 1940 to 1985 and they further docu­ following regression:
mented the persistence across ethnicities and regions. Their findings S
Si,c P
= α + ρSi,c + εi,c (1)
ascertained no decline in persistence estimates, however, conditional
probabilities show a shift towards the attainment of higher education.
But based on conditional probabilities they do not find evidence of SSi represents the years of schooling of male individual i belonging to
convergence among various ethnic groups. In comparison to their paper, cohort c, whereas, SPi is the years of schooling individual i’ s parent.
instead of relying on conditional probabilities to observe convergence, Years of schooling for parents and son are both normalized to have a
we rely on stationary distribution for regions and ethnicities. standard deviation of 1. We construct matched father-son samples and
Education policy in Indonesia is mainly controlled by the central run the estimation by the cohort of the son separately in each sample.
government under the ministry of education and culture. The Regional We consider 10 cohort groups: 1941–1990 for our nation wide analysis
Autonomy Law (22 and 25) in 1999, the 2003 Education Law, and but for ethnicity and regions we consider 3 cohort groups: 1946− 1990.
subsequent regulations (e-g Government Regulation 38 of 2007) estab­ For brevity, we focus our discussions on the results from the father-son
lishes the responsibilities of various levels of government in the provi­ sample.3
sion of education services. The central government keeps its role in The coefficient ρ is the parameter of interest and the ordinary least
formulating education policies, establishing curriculum frameworks, square estimate of ρ is interpreted as a measure of intergenerational
and setting national education standards. The district governments are persistence of educational attainment. Moreover, changes in ρ not only
given responsibility for all primary and secondary education organiza­ captures father-son education transmission but a different phenomenon
tions, appointment and placement of teachers, the authority to establish such as the secular rise in schooling and changes in compulsory edu­
new schools and maintain registration of existing schools. cation. Thus, changes in ρ may reflect changes in the education policy
Twelve years of education is compulsory for everyone; however, among other things, therefore, it is referred to as an absolute measure of
private education institutions operate parallel to public institutions. In intergenerational transmission of education. The correlation coefficient
the early’ 70 s, the Government of Indonesia started a massive school can be re-written as:
expansion program throughout the country due to which enrollment ∫
( ) ( ) ⃒ ( )
rate share of a private school at the primary level and secondary level is ρ= (SiS − E SiS )(SiP − E SiP ) P(SiS ⃒SiP ) P SiP . (2)
⏟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏟⏟̅̅̅̅̅⏞⏞̅̅̅̅̅⏟⏟̅̅̅⏞⏞̅̅̅⏟
over 10 % and 18 % since 2000, respectively. The objective of this policy A B C

was to reduce the education differential among the various regions and Consequently, as shown in term (A) of the above equation, the cor­
provide an equal opportunity of education to everyone in Indonesia. In relation coefficient can change over time because of dispersion of son’s
this regard, Hertz and Jayasundera (2007) investigated the policy and parent’s education (standardized) around their respective means, or
outcome of school expansion programs on intergenerational mobility in because of educational attainment of sons condition on the parents ed­
Indonesia. They documented that higher mobility is attributed to the ucation (conditional probability) as shown in term (B) and term (C),
exposure to new schools under this policy by younger cohorts whose which is unconditional probability of parent’s education. In order to
fathers are educated. However, they used only one wave of Indonesian investigate the stability of coefficient of correlation, we further
Family Life Survey (IFLS) and their research design and objectives are decompose the correlation coefficient by using the empirical analogue of
different from this study. Their study focuses on the impacts of school Eq. (2).
expansion program and this paper focuses on decomposition and long ∑∑ ∑∑
( ) ( ) ⃒
run evolution of persistence. ρ=
̂ (SiS − E SSi )(SiP − E SiP ) P(SiS ⃒SiP ) P(SiP ) = rS,P (3)
⏟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏟⏟̅̅̅̅̅⏞⏞̅̅̅̅̅⏟⏟̅̅⏞⏞̅̅⏟
Even though it has been argued here earlier, but understanding the S S SP
A B C
SS S P

issue of intergeneration mobility from the mentioned perspective is

important in Indonesia and this has received relatively less attention where SS , SP ∈ {0, 1, 2, ⋯, 16} representing a maximum of 16 years of
mainly due to the availability of suitable data. In the absence of longi­ schooling, and thus ̂ρ for each cohort is the sum of 289 elements. In the
tudinal panel data, the majority of the studies focused on those house­ empirical literature, it has been argued that change in term A could be
hold members who coreside with their parents. This restriction attributed to convergence towards higher level of education (Azam and
significantly reduces the observations and generates a serious sample Bhatt, 2015), thus, since the term B shows the changes in sons’ educa­
selection issue, because the distribution of both generations’ education tion conditional on the education of their fathers, it can be tracked by
is different i.e. who lives with parents might attain higher education in policymakers. For example, decreasing of the term signals that policies
contrast to those who are living outside. Alternatively, it can be argued enforcing schooling improve equality of opportunities. Moreover, with
as well, in order to acquire higher education sons do relocate to other the passage of time, the country develops and if education is a normal
place like big cities or abroad, thus, they obtain more education than good, one can expect an increase in the educational level of parents
parents. across generations. Term C represents the changes, which are indepen­
We discuss this issue in the data section after limiting our sample to dent of fathers’ education. It is related with the development and
coresiding household members, which amounts to only 35 % of the institutional infrastructure of a country (Checci, Firorio & Leonardi,
observations in the data. The sample-based on coresidence restriction 2013). Now we focus on conditional probability (term B) as shown in Eq.
does not allow us to perform the cohort wise analysis and long-term (3) to obtain the decomposed conditional probability.
trend in intergeneration transmission because such as condition will With the conditional probabilities, Pr(SSi |SPi ), we construct two
only be satisfied for younger adults (Azam and Bhatt, 2015). We focus indices for upward and downward mobility respectively. The reason
our analysis on sons due to early marriages and relocation of daughters these indices rely on the conditional probabilities (term B’s) is that they
with lower tracking rate. Additionally, men are considered breadwin­ purely capture the probability that a son obtains the educational level
ners for the family due to their stereotypical roles in a traditional society above (or below) his father. For t, j ∈ {0, 1….,E}, The upward mobility
and developing country like Indonesia, thus, analyzing the inequality of
opportunities for breadwinners is even desirable from research
perspective. Although, we performed the analysis for daughters as well
and we observed the similar results, thus, in the interest of brevity we 3
We obtain similar results from the mother-son sample, which we can pro­
present the results for sons’ sample. vide upon request.

3
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Table 1 who live in the same household. Thus, they captured the bias in inter­
Descriptive statistics nation wide. generational elasticities and it was 12%–39% as identified by the parent-
Cohorts Father’s Years of Schooling Son’s Years of Schooling child pairs only on coresidence information. When we restrict the data to
coresident pairs, we observed that 65 % of the observations dropped in
1941− 1945 5.48 7.17
1946− 1950 6.06 7.70 both father-son and mother-son samples. We have performed the same
1951− 1955 6.39 8.03 analysis on the mother-son pairs as well but in the interest of brevity in
1956− 1960 6.52 8.10 this paper we focus on father-son pairs. This could be the reason that our
1961− 1965 6.76 8.72 estimates also differ in both full and coresident samples as in Table 1 we
1966− 1970 6.96 9.44
1971− 1975 7.26 9.71
observe that our intergenerational elasticity estimates are under­
1976− 1980 7.52 10.34 estimated when imposing the coresident condition. Therefore, it is the
1981− 1985 7.55 10.58 advantage in IFLS that it enables us to make father-son and mother-son
1986− 1990 7.57 10.70 pairs without being contingent on children residing with parents. Head
Source: Author’s calculation from IFLS (Notes: We calculated cohort wise of the household or any other householder over 18 years of age is asked
average years of schooling for both fathers’ and sons’ distribution by using all to provide the information on non-co-residing house members deceased
five wave of Indonesian Family Life Survey). or alive. Information on education, sex, birth year is collected for three
types of non-co-residing household members which include parents,
index is defined as siblings and children.
∑ ⃒ Unlike Azam and Bhatt (2015), we are able to track non-coresiding
pr(SiS = j⃒SiP = t) (4) fathers, children and siblings as well over five waves which cover over
21 years of data, and all this makes our sample bigger and interesting. By
j>t

tracking the siblings and children we matched the information with the
and the downward mobility index is defined as
father dead or alive. This enables us to track almost all the members
∑ ⃒
pr(SiS = j⃒SiP = t). (5) associated with house whether living in or out of the house and make
j<t father-son pairs. This certainly increases the predictive power of our
estimates. Moreover, when it comes to years of schooling, it is not
The upward mobility index basically represents the cumulative sum
readily available, the common estimation strategy is to replace the level
of conditional probability that a son attains higher education in com­
of attained education by the number of regular years needed to obtain
parison to his parents. Whereas, the downward mobility index is the
(Black and Devereux, 2010). Fortunately, IFLS allows us to impute years
cumulative sum of a son attaining less education than his parents.
of schooling based on two questions (AR16 and AR17) which clearly
Lastly, we built 5-by5 transition matrices based on regrouped years
asks the respondents about highest level of schooling attended and
of schooling into five different categories of education attained such as
whether they have graduated from that level or not, if not then how
no schooling, primary, middle, senior high and university. To keep the
many years they spent in that particular level. In addition to this, IFLS
analysis precise, we focused on educational attainment categories here
further provides us information on ethnic affiliation (DL01) and resi­
instead of years of schooling. We compute the long-run stationary dis­
dence status (SC01) due to which we are able to capture the ethnic and
tribution over educational outcomes implied by these transition
regional dimension concerning persistence in education. Although, IFLS
matrices to assess the possibility of long-run convergence.
data is not designed to be a representative of ethnicities but still it helps
to provide some important insights about various demographics and
3. Data
existing inequality between different ethnicities present in IFLS. Thus,
we put our estimates concerning major ethnicities in Indonesia in Ap­
We used all five waves of the Indonesia Family Life Survey (IFLS)
pendix A and this objective can be investigated further in the future with
from 1993 to 2014. The IFLS is an ongoing longitudinal survey of
help of ethnicities representative data.
Indonesia, which is representative of over 80 % percent of the Indone­
For region-wise estimates, IFLS is a representative of the 13 original
sian population and contains 30,000 individuals living in 13 provinces.
IFLS provinces in 1993, collectively (the population of which constituted
It contains a wealth of information on education, ethnicity, geographic
of 83 % of Indonesia in 1993). The cross-sectional weights constructed in
location, consumption, labor supply, income, migration, and health of
each wave help bring it back to be representative- of the original
the household. Although, IFLS is a rich dataset in terms of covariates
provinces for that particular year. The end result is a weight that makes
which could certainly strengthen the analysis but our father-son
the IFLS of each round representative of 13 original provinces and one
matching is unique in a way that we created it. For example, we used
needs to exclude the households located outside the original provinces
different data files related to the available information about parents,
and combine the post-1993 split provinces. We have done this exercise
children, and siblings to construct this matching but not all the infor­
which makes our key variables close to SUSENAS but for later rounds
mation about the covariates are consistently available in all the files.
regardless of adjusting for over and under-sampling, the representation
Thus, adding covariates would reduce our sample size for this unique
issue remains which makes it problematic to use IFLS for province-level
father-son matching. This is the reason many studies including (Aydemir
comparisons. We have tried to address this issue by separately using the
and Yazici, 2019; Azam and Bhatt, 2015; Checci, Firorio & Leonardi,
first wave of IFLS which is the provincial representative of Indonesia in
2013) model this relationship without any covariates. Our sample in­
1993 as a robustness check for the entire analysis using all five waves.
cludes those who were between the ages of 25 and 53 during the five
We outline our results for provinces from wave 1 in Appendix-B
waves of IFLS, therefore, our sample was born between 1941 and 1990.
Tables B1 and B2.
It is important to note data limitation occurs when surveys are limited to
We have presented our descriptive statistics related tables in
co-residing households. Fortunately, IFLS allows us to track the
Tables and Figures section. In Table 1 we have descriptive statistics
non-coresiding household member, therefore, we do not suffer with
showing nationwide average years of schooling over ten successive co­
potential bias in the estimates.
horts. Here we see a clear increase in average years of schooling for all
Francesconi and Nicoletti (2006) studied 11 waves of the British
sons and fathers over time. Average years of schooling among Indone­
Household Panel Survey (BHPS) and investigated occupational prestige.
sians above the age of 25 doubled since 1980 (EDUCATION SYSTEM
They used coresident data and argued the existing bias in persistence
PROFILES, 2019) we observe the similar trend in our data as well. We
estimates for occupational prestige if one rely upon coresident parents
have used broad categorization for cohorts in ethnic and regional
because such a condition is more likely to be satisfied for young adults
dimension because in older cohorts for some of the ethnic groups we

4
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Table 2 created a downward bias in intergenerational elasticity estimates.

Intergenerational persistence in education attainment, full sample versus cor­ The value of the regression coefficient is 0.47 and the correlation
esidence sample. coefficient is 0.53 for the coresident sample which implies that regres­
Father-Son Pairs Full Coresident sion coefficients are roughly 6% lower in the coresident sample.
Sample Sample Whereas, the value of regression coefficient is 0.52 and correlation co­
Father’s years of schooling (̂
ρ) 0.49** 0.53**(0.007) efficient is 0.49 in the full sample which means correlation coefficients
(0.004) are 3% lower in the full sample, which is also consistent with previous
Father’s years of schooling (̂
β) 0.52** 0.47**(0.006) work by Azam and Bhatt (2015). The differences we observe are sta­
(0.004)
tistically significant as suggested by joint equality test. It is also evident
R 2 0.222 0.262
that standard deviation in son’s years of schooling is higher in full
observation 41,275 14,518
S.D in Son’s Years of Schooling (σs ) 4.007 3.647
sample, whereas in co-resident sample it significantly low, whereas the
S.D in Father’s Years of Schooling (σf ) 3.767 4.125
opposite is true for standard deviation in father’s years of schooling.
σs/ 1.063 0.884
Ratios of standard deviations are much smaller in co-resident sample,
σf which causes the estimated correlation to be much closer in both
χ2 = 229.37 with p value = 0.000 for equality of (̂ρ )
in both samples. samples.
Moreover, we found those sons who coreside with their parents have
χ2 = 416.61 with p value = 0.000 for equality of (̂β)
in both samples significantly higher persistence in comparison to entire sample. As dis­
cussed earlier when limiting data to coresidents we are more likely to
Robust standard errors in parentheses. analyze the young adults, therefore, it can be argued over time for
** p < 0.01, * p < 0.05, + p < 0.1. attainment of higher education children moves out or relocate to other
cities or countries. Thus, in our case we infer, when we limit the data to
have low observations. Thus, using fifteen years cohorts we are able to coresidents we are more likely to analyze the young children where
address this issue. education attainment is similar to their parents. Hence, we observe
higher persistence in the coresident sample as we document 4% over­
4. Empirical results estimation in correlation estimates. We have also performed the joint
equality test and it suggests that estimates from both samples are not the
This section presents our empirical results. We first present our same. Therefore, it is right to argue about the existing bias in estimates
findings for pooled sample and then we present cohort wise intergen­ from coresident sample.
eration persistence in educational attainment, at aggregate level, by
major ethnicities and by major provinces in Indonesia.
4.2. Nationwide cohort analysis of intergenerational education mobility in
Indonesia
4.1. Intergenerational educational persistence in Indonesia
In Table 3, we further analyze the pooled sample with 10 distinct
We outlined our pooled results in Table 2, where in two different cohorts to capture the change in persistence over time. Interestingly,
columns we show our estimates for full sample and coresident sample. starting from 1941–1990, except for three cohorts between 1951–1965
As argued earlier, one of the major benefits of using IFLS is that it en­ we observe a visible overall decline in intergenerational elasticity esti­
ables us not to be contingent on parents residing with child. So in order mates but an unclear trend in the coefficient of correlation estimates
to illustrate the consequence, we exploited this data restriction to co- especially in the middle cohorts starting from 1960 to 1980. On the
resident households and found that only 35 % of the observations whole, we document persistence estimates decline in the context of the
were present in coresident sample in contrast to full sample. In addition correlation coefficient from 0.49 to 0.44 which means 5% decline in 50
to this, we compare two measures of the persistence of intergenerational years, and for intergenerational elasticity estimates, we observe 19 %
education attainment: the regression coefficient ̂
β and the coefficient of decline. But we do not observe any secular trend in the coefficient of
correlation ̂
ρ . The discrepancy between beta and correlation coefficients correlation since time variation is not clear. A joint equality test of co­
is evident in Table 2 for our nationwide results. Moreover, in the same efficients rejects the null hypothesis that the coefficients are constant
table, we observe the differences in the estimates when imposing the across cohorts. This implies, while there is no secular trend in the
coresident condition which validates the finding of Francesconi and intergenerational transmission of education from father to son over
Nicoletti (2006) as we see that the restricting data to co-residents time, there are significant swings. As we show next in the decomposition

Table 3
Cohort wise mobility estimates.
Variables 1941− 1945 1946− 1950 1951− 1955 1956− 1960 1961− 1965 1966− 1970 1971− 1975 1976− 1980 1981− 1985 1986− 1990

ρ
̂ 0.495** 0.418** 0.417** 0.408** 0.425** 0.434** 0.455** 0.524** 0.494** 0.443**
(0.023) (0.021) (0.017) (0.015) (0.012) (0.010) (0.010) (0.011) (0.012) (0.014)
̂
β 0.601** 0.577** 0.560** 0.556** 0.564** 0.527** 0.496** 0.510** 0.456** 0.413**
(0.028) (0.029) (0.023) (0.020) (0.016) (0.012) (0.011) (0.011) (0.011) (0.013)
σs 4.253 4.085 3.938 4.008 4.090 3.840 3.621 3.633 3.428 3.646
σf 3.498 2.956 2.927 2.944 3.086 3.170 3.315 3.7367 3.712 3.917
R2 0.235 0.184 0.175 0.174 0.190 0.206 0.219 0.274 0.240 0.194
Observations 1562 1855 2740 3714 4787 5755 5980 4936 5406 4507

χ 2 = 82.21 with p value = 0.000 for equality of.

(̂
ρ ) in all cohorts.
χ 2 = 115.61 with p value = 0.000 for equality of.
(̂
β) in both samples.
Robust standard errors in parentheses.
** p < 0.01, * p < 0.05, + p < 0.1.

5
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Table 4
Decomposed correlation.
Decomposed ρ 1941–1945 1946− 1950 1951− 1955 1956− 1960 1961− 1965 1966− 1970 1971− 1975 1976− 1980 1981− 1985 1986− 1990

Father: No Schooling
Son: No Schooling 0.208 0.139 0.101 0.089 0.076 0.047 0.024 0.025 0.026 0.034
Son: Primary 0.100 0.095 0.093 0.080 0.070 0.089 0.083 0.112 0.099 0.091
Son: Middle − 0.004 0.000 0.000 0.000 0.003 0.004 0.007 0.015 0.019 0.015
Son: Senior High − 0.016 − 0.015 − 0.020 − 0.009 − 0.012 − 0.013 − 0.014 − 0.009 − 0.008 − 0.011
Son: University − 0.017 − 0.012 − 0.007 − 0.013 − 0.005 − 0.004 − 0.004 − 0.007 − 0.011 − 0.010
Percentage of Group 55.0 % 49.5 % 40.1 % 36.3 % 31.1 % 28.3 % 21.1 % 26.1 % 25.4 % 27.0 %
Contribution to
Correlation
Father: Primary
Son: No Schooling − 0.004 0.001 0.004 0.006 0.011 0.009 0.007 0.005 0.005 0.009
Son: Primary − 0.023 0.011 0.036 0.045 0.069 0.083 0.103 0.095 0.086 0.078
Son: Middle 0.002 0.000 0.000 0.000 0.004 0.011 0.018 0.023 0.034 0.033
Son: Senior High 0.012 − 0.005 − 0.019 − 0.024 − 0.034 − 0.040 − 0.044 − 0.032 − 0.034 − 0.030
Son: University 0.016 − 0.006 − 0.018 − 0.019 − 0.025 − 0.026 − 0.027 − 0.026 − 0.027 − 0.034
Percentage of Group 0.6 % 0.2 % 0.7 % 2.0 % 5.9 % 8.5 % 12.5 % 12.5 % 13.0 % 12.7 %
Contribution to
Correlation
Father: Middle
Son: No Schooling − 0.006 − 0.005 − 0.002 ——— − 0.001 − 0.005 − 0.002 − 0.001 0.000 − 0.001
Son: Primary − 0.008 − 0.009 − 0.018 − 0.012 − 0.013 − 0.009 − 0.009 − 0.007 − 0.006 − 0.005
Son: Middle 0.002 0.000 − 0.001 0.000 − 0.002 − 0.007 − 0.006 − 0.004 − 0.004 − 0.003
Son: Senior High 0.033 0.031 0.034 0.023 0.021 0.021 0.018 0.009 0.010 0.007
Son: University 0.039 0.039 0.047 0.040 0.032 0.025 0.022 0.020 0.017 0.012
Percentage of Group 12.2 % 13.4 % 14.4 % 12.6 % 8.7 % 5.8 % 5.1 % 3.3 % 3.5 % 2.3 %
Contribution to
Correlation
Father: Senior High
Son: No Schooling ——— ——— ——— − 0.002 − 0.002 − 0.004 − 0.003 − 0.001 − 0.003 − 0.008
Son: Primary − 0.008 − 0.007 − 0.011 − 0.011 − 0.009 − 0.010 − 0.017 − 0.009 − 0.012 − 0.016
Son: Middle 0.002 0.000 0.000 0.000 − 0.003 − 0.005 − 0.006 − 0.007 − 0.008 − 0.006
Son: Senior High 0.040 0.041 0.044 0.042 0.041 0.045 0.044 0.032 0.030 0.026
Son: University 0.089 0.071 0.093 0.095 0.093 0.098 0.098 0.113 0.106 0.096
Percentage of Group 24.9 % 25.1 % 30.3 % 30.6 % 28.2 % 28.6 % 25.5 % 24.6 % 23.0 % 20.9 %
Contribution to
Correlation
Father: University
Son: No Schooling ——— ——— ——— − 0.002 ——— ——— ——— − 0.001 − 0.003 − 0.007
Son: Primary − 0.003 − 0.001 − 0.002 − 0.003 − 0.002 − 0.002 − 0.002 − 0.006 − 0.005 − 0.012
Son: Middle 0.001 0.004 0.007 0.003 − 0.004 − 0.004 − 0.001 − 0.001 − 0.003 − 0.005
Son: Senior High 0.010 0.006 0.011 0.015 0.014 0.012 0.021 0.013 0.015 0.012
Son: University 0.028 0.040 0.044 0.062 0.103 0.119 0.145 0.170 0.169 0.176
Percentage of Group 7.3 % 11.7 % 14.4 % 18.5 % 26.1 % 28.8 % 35.8 % 33.6 % 35.2 % 37.2 %
Contribution to
Correlation
Correlation 0.493 0.418 0.416 0.405 0.425 0.434 0.455 0.521 0.492 0.441
coefficient

Notes: Years of schooling is grouped to refer to attended stages of schooling. No schooling: 0 years; Primary: 1–6; Middle: 7–9 years; Senior High: 10–12; and University:
13 years or more.

exercise of the correlation coefficient, the lack of secular trend in the fathers with all five education attainment categories of sons.
transmission masks differences in what is driving the correlation in older The last row of Table 4, titled “Correlation coefficient,” corresponds
∑ ∑
versus younger cohorts. Therefore, following section is related with to SS rS,P , which is the correlation coefficient estimated by cohort.
SP
counteraction forces of correlation coefficient which beta coefficient
Every row in this table corresponds to rS,P as stated in Eq. (3) it is a
cannot deepen. To do this, we decompose the correlation coefficient and
obtain conditional correlations which shows extent to which a child’s component of the correlation coefficient linking sons’ education while
education is affected by father’s education as given. We also estimated holding to parent’s educational category constant and sum of all the
this coefficient for regions and obtained upward and downward mobility rows gives us a decomposed estimation correlation which is equal to our
indices. persistence estimates in Table 2.4 For every category of father’s educa­
tion we obtained the group contribution of that category to the total
correlation. It is interesting to see group contribution of fathers with no
4.3. Correlation decomposition schooling declined over the period of time. As we observe in the first
cohort unschooled fathers were contributing 55 % to the total correla­
We decompose the correlation to investigate the liable factors un­ tion which reduces to 27 % in the last cohort. In fact, to shed light on the
derlying the trend of the estimates of persistence, according to Eq. (3). In study of Akresh et al. (2018) further, we can see from this decomposition
Table 4, we present the results. For this exercise, we group years of
schooling into five distinct education attainment categories: no
schooling, primary, middle, senior high and university education. In the 4
In Table 2 we used years of schooling to capture the educational persistence
decomposition table, we put the correlation between the educational but in Table 3 we used 5 education categories to obtain the decomposed
attainment of sons’ condition on their father’s education group. For persistence estimates, thus we observe small differences in the estimates
instance first group in the table shows the correlation of unschooled because of this difference.

6
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

educational attainment clearly improves for younger cohorts, since we

more masses in university education and fewer masses in primary or
middle. This can be explained as an implication emerging from educa­
tion policies in Indonesia as we document the evidence that long-run
probability for the attainment of university education increased and a
clear decrease in no-schooling, primary and middle stages of education.

4.4. Cohort analysis over regions

For region-wise analysis, first, we plot various characteristics and

indicators in order to understand the existing regional differences in
Indonesia. Indonesia has 34 different provinces of Indonesia and 8 of
which are created in 1999. Thus, to avoid the split off issue we rely on 12
provinces consistently covered in IFLS out of 13. In Fig. 4 we plot the
percentage of the population in IFLS residing in the concerned prov­
inces. We found an extremely skewed concentration of population in
Fig. 1. Transmission Rate (father to son).
some of the regions such as 19 % of the observations belongs to West
Java, 12 % from Central Java and 9% from Jakarta. Moreover, in Fig. 5
exercise that the 1973 school construction program had a strong nega­ we plot region-wise total asset holdings (in USD) using all five waves of
tive intergenerational effect on 1966–1970 and 1971–1975 cohorts with IFLS. We define asset as sum of reported value of house, other physical
uneducated fathers. The probability of having an uneducated son in asset and financial assets such as receivables, bonds etc. There we
these periods declined significantly compared to previous periods. observed highest asset holdings are in Jakarta followed by Bali and West
Contrary, the probability of having a primary degree of a son with an Nusa Tenggara stands last among the concerned regions. We computed
uneducated father increased in these cohorts. In later cohorts, the the average asset holdings to better understand about wealthier and
probability of having middle or senior high level with same level father least wealthy regions. This enables us to observe how the educational
slightly declined. This may be explained that the construction program persistence evolves in these regions. Thus, to maintain the brevity and to
has lagged effects on post-primary education. But on the other hand, we better visualize the result for stationary distribution we kept our analysis
see this decline is compensated by the group of fathers with university limited to Jakarta, two medium asset holding regions like South Sumatra
education which was 7% in the first cohort and increased to 37 % in the and Yogyakarta, and the region with the least asset holdings which is
last cohort. Thus, the steady trend in correlation coefficient we estimate West Nusa Tenggara.
by cohort in Table 3 reflects two countervailing forces. While the major In addition to this, we outline the computed years of schooling for
contributor to intergenerational persistence in educational outcomes concerned regions of Indonesia in Table 5. There we observe 8.97
comes from unschooled fathers in older cohorts, in younger cohorts the average years of schooling for fathers in Jakarta which is the highest in
persistent is mainly driven by fathers with university education. among all the regions in the most recent cohort. Whereas, in Bali we
To make it clearer, we plot in Fig. 1, the computed the percentage of observed 7.38 years of schooling which is least among all the regions in
group contribution to the coefficient of correlation from the same edu­ the most recent cohort. It is surprising to see in a region like Yogyakarta,
cation category of father-son pairs. Hence we call it transmission rate, by where average asset holding is lesser than Jakarta and Bali, we observe
transmission we mean the flow of same education attainment from fa­ remarkable progression in years of schooling for sons. In fact, highest
thers to sons. In Fig. 1, it is evident that the fathers with a university average years of schooling are 12.26 years in the most recent cohort.
education were transmitting less than 8% of their education to their sons We estimate the educational persistence for each region using 3 co­
but this increases to 37 % in the last cohort. However, another major horts with fifteen years gap starting from 1946 and ending in 1990. As
change occurred for unschooled father’s education transmission to their we have mentioned earlier in the data section that only the first wave of
sons where we see the transmission rate decline from 54 % to 27 % over IFLS is actually representative of the provincial population originally
the all ten cohorts. The third major change in transmission rate is covered in IFLS. For the later-rounds representation remains a conten­
evident from fathers with primary education which started with less tious issue so we have performed our analysis by using all five waves and
than 1% in the oldest cohort to 12 % in the youngest cohort. only first wave separately as a robustness check. In Table B1 we present
We extended the analysis further by obtaining the term B in Eq. (3) our estimates for provinces by using first wave only and while doing this
which is the conditional probabilities of son achieving various levels of exercise we keep in our mind that we will have a lower number of ob­
educations. Checchi and García-Peñalosa (2008, 2013) discussed that servations, so, some of our estimates may be imprecisely estimated
term B in Eq. (3) is the correct measure for examining the transmission of relative to our main results. Table B2 shows the direction of estimates
education. To summarize their story, a system will achieve the equality over regions in different cohorts by using all five waves and first wave
of opportunities if the probability of attaining a particular degree for the only. It was really interesting to see for the majority of provinces (except
son was independent of the father’s educational achievement. In the last three provinces where we observe the different direction of estimates
panel of Fig. 2, we see a major decline in the probability of a son across cohorts) we observe the similar direction of estimates across all
achieving no education given father is unschooled. This proves the three cohorts.
universalization of education towards a higher level as it can also be Table 6 shows the results for this objective and it is evident that
seen in the first panel as well. Whereas the probability of a son achieving persistence estimates vary substantially among all the regions for all the
university education increases given father is having a university cohorts; although, just like our estimates in ethnicity analysis we observe
education. somewhat unclear trends in some of the regions. But in some regions like
In Fig. 3, we plot the stationary distribution by cohorts. First, we North Sumatra, Jakarta, West Nusa Tenggara, and Yogyakarta we
construct the transition matrices and later we solve them in order to document a significant increase in persistence over time. It is interesting
obtain stationary distribution. For each cohort, we have a 5⨯5 matrix to document an increase in persistence in the highest asset holding and
corresponding to 5 stages of education attainment for son and father. lowest asset holdings regions of the country. This raises important
Here it is evident that we do not observe the long-run convergence but questions about existing inequality in these regions or why these two
probability suggests the substantial decline in the first three education regions behave similarly when it comes to persistence in education. In the
stages over the period of time. Here we observe the distribution of rest of the regions, we do not observe a clear trend in persistence

7
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Fig. 2. Probability of son attaining various levels of education conditional on father’s education.

estimates. We have also presented our estimates from first wave by Thereafter we plot the upward and downward mobility index con­
keeping in mind that we will have a lower number of observations, so, structed from Eqs. (4) and (5) respectively. As we can see in Fig. 6, again
some of our estimates may be imprecisely estimated relative to our main observe reduced margins between the probability of different regions
results. It was really interesting to see for the majority of provinces that son will achieve higher education than his father in 1961− 1975.
(except three provinces where we observe the different directions of es­ Although, we do not observe a close convergence here but overtime for
timates across cohorts using all 5 waves and only first wave) we observe all the regions we see a positive trend in the upward mobility index
the similar direction of estimates across all three cohorts. except West Nusa Tenggara. In Fig. 7, we see an increase in downward

8
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Fig. 3. Stationary Distribution Plot over Cohort.

education stages for instance we found in all the regions chances are
higher that son will achieve university education, and the trend for no
schooling, primary and middle declines over time. Although, we found
similar variations in different regions but it cannot be argued as
convergence across regions.

5. Conclusion

In this paper, we thoroughly investigate the long-term persistence of

education attainment in Indonesia. We exploit the long panel of IFLS and
document the educational persistence and contributed to the literature
by decomposing the correlation followed by an attempt to investigate
the long-run convergence with the help of stationary distribution. Using
IFLS enabled us to track 41,275 father-son pairs which certainly in­
creases the reliability of our estimates. In addition to this, IFLS enabled
us to observe the evolution of historic differential in years of schooling
by ethnicities and by regions.
We investigate the educational persistence in Indonesia with three
Fig. 4. Region wise population percentage.
different objectives. First, we capture the degree of education persis­
Source: Author’s calculation from IFLS (Notes: We calculated percentage of tence in Indonesia without adding a time dimension. We compared two
population distribution by province using all five wave of Indonesian Family different samples and found those sons who co-reside with their parents
Life Survey). have a significantly higher coefficient of correlation in comparison to
the entire sample which implies higher persistence in the coresident
mobility for North Sumatra but again we observe the change in middle sample. Thus, we document 4% overestimation in correlation estimates
cohort and it eventually declined. Similarly, for Central Java we observe for the coresident sample of IFLS. We further analyze the pooled sample
a steep decline in downward mobility after 1961− 1975. Regardless of with 10 successive cohorts to capture the change in persistence over
differential patterns, it can be argued we observed an overall decline in time. From 1941–1990, we do not find a clear time trend in the coeffi­
downward mobility for all the regions. cient of correlation estimates. However, after decomposing the corre­
Lastly, as discussed earlier we obtained the stationary distribution lation of coefficient estimates, we uncover that the correlation at the
for four different regions. This classification is based on what we lower end of fathers’ education distribution is offset by the increase at
observed in the data so we picked one region which has the biggest asset the top end of fathers’ education distribution. More specifically, the
holdings, two mid asset holding regions and last one is the least asset correlation between unschooled fathers to unschooled sons decline from
holding region. In Fig. 8 we plot our stationary distribution results. In 0.20 to 0.03 in 50 years but on the other hand, the correlation between
Jakarta we conclude there is not much variation about the probability of the university education group of fathers and sons increased from 0.02
attaining university or higher secondary education from 1946− 1960 to 0.17. In other words, the general trend of the correlation coefficient
and 1975− 1990. However, in the middle cohort we observe a substan­ masks two opposite time trends of correlation between the least
tial increase in the chances of attainment of university education and a educated fathers and their sons and the most educated fathers and their
decline in senior higher secondary education. In Yogyakarta and West sons.
Nusa Tenggara, we observe a noteworthy increase in stationary distri­ Our results have noteworthy implications for policy in Indonesia. For
bution over time suggesting higher rate that son will achieve university instance, the decline in persistence at the lower end of fathers’ education
education in long run in comparison to the rest of the regions. shows that public education is progressively able to compensate for the
Furthermore, we observe the shift in the stationary distribution for all lack of education inputs in the family. We also document the significant

9
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Fig. 5. Region wise average asset holdings (in USD).

Source: Author’s calculation from IFLS (Notes: We calculated average asset holdings by regions using all five wave of Indonesian Family Life Survey).

differences in attaining various levels of education for sons conditioning international evidence such as for Canada (Latif, 2019), Latin America
on father’s education especially for higher secondary or above. From Region (Neidhöfer et al., 2018), India (Azam, 2016), Australia (Rana­
conditional probabilities, we can see the policies which are able to singhe, 2015) and South Africa (Kwenda et al., 2015). Though govern­
transport those sons of less-educated fathers to advanced level of edu­ ments of these countries have managed to reduce the persistence in
cation are not able to move those sons to advanced level where fathers primary to tertiary education, but university education has higher
have higher education. For instance, if the father has senior higher persistence to parent’s education for various reasons. However, these
secondary education, the probability son will have a university educa­ studies point to economic performance and public spending in education
tion remains constant. Our findings are in line with some of the is a key to accelerate the mobility of younger cohorts to a college degree.
In this sense, income inequality, poverty, economic growth, and assor­
tative mating are determinants to explain cross-country differences of
Table 5
educational persistence at the top end of education distributions
Region wise descriptive stats.
(Neidhöfer et al., 2018).
Father-Son Sample For ethnicities, we present our estimates in Appendix A and we
Regions 1946− 1960 1961− 1975 1976− 1990 observe people belonging to Batak and Sundanese have fewer chances of
West Java obtaining a university education in comparison to the rest of the eth­
Father’s Years of Schooling 7.05 7.33 7.56
nicities. Thus, ethnic identity remains an important factor for policy­
Son’s Years of Schooling 8.14 9.27 10.15
Central Java makers to address these differences in Indonesia. Thus, it could be
Father’s Years of Schooling 6.19 6.60 7.08 argued that policy should be more focused on social inclusion consid­
Son’s Years of Schooling 7.54 8.74 10.18 ering the persistence in education in major ethnicities of Indonesia.
North Sumatra OECD (2015) recommended that there is a need to expand the provision
Father’s Years of Schooling 6.75 7.22 7.68
of education beyond basic level to disadvantaged groups in Indonesia in
Son’s Years of Schooling 8.51 9.73 10.94
Jakarta order to achieve the high-income status.
Father’s Years of Schooling 7.22 7.96 8.97 In our region-wise analysis, we found significant differences at the
Son’s Years of Schooling 9.51 10.63 11.88 regional level because in some of the regions we see higher persistence
South Sulawesi
but in some of the regions trends in persistence estimates is not clear. We
Father’s Years of Schooling 5.30 6.47 7.32
Son’s Years of Schooling 6.59 8.63 10.10
documented that Jakarta, Yogyakarta, and West Sumatra have the least
Lampung persistence in education or in other words these are most mobile regions
Father’s Years of Schooling 5.92 6.18 7.17 which imply lower inequality of opportunities in these regions. This fact
Son’s Years of Schooling 7.17 8.44 9.76 is further validated with the economic size of these three regions as
South Sumatra
Jakarta is the commercial and official capital of Indonesia and whereas
Father’s Years of Schooling 6.96 7.22 7.99
Son’s Years of Schooling 8.27 9.37 10.43 two other regions have a higher human development index in compar­
West Sumatra ison to the rest of the regions. Among these provinces, Sleman regency of
Father’s Years of Schooling 7.11 7.43 8.54 Yogyakarta has remarkable success in terms of education policies, ac­
Son’s Years of Schooling 8.43 9.84 11.66
cording to the Indonesian Local Education Governance (ILEG) index
West Nusa Tenggara
Father’s Years of Schooling 6.290 6.46 6.74
developed by Al-Samarrai (2013)5 . In contrast, South Sumatra, West
Son’s Years of Schooling 6.95 8.71 10.43 Java, and Central Java are the most immobile regions, along with low
Yogyakarta
Father’s Years of Schooling 6.60 8.14 8.34
Son’s Years of Schooling 9.47 10.97 12.26
5
Bali Al-Samarrai (2013) assigned a ILEG score to a randomly selected 50
Father’s Years of Schooling 5.94 6.51 7.38 regencis in Indonesia. Components of education governance are following: i)
Son’s Years of Schooling 7.11 9.51 10.88 education services provision standards ii) transparency and accountability iii)
East Java management control systems iv) management information systems v) efficient
Father’s Years of Schooling 6.23 6.20 7.38
resource use. Since the scope of the regencies was subject to eight regions and
Son’s Years of Schooling 7.54 8.74 9.18
three of them is matched with our study, we limited our assessments with these
Source: Author’s calculation from IFLS regions.

10
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Table 6
Mobility estimates over regions.
Father-Son Sample

Regions 1946− 1960 1961− 1975 1976− 1990

West Java
ρ
̂ 0.502** (0.021) 0.505** (0.013) 0.559** (0.015)
R2 0.240 0.265 0.312
Observations 1502 3128 2531
Central Java
ρ
̂ 0.555** (0.018) 0.494** (0.018) 0.555** (0.018)
R2 0.285 0.232 0.285
Observations 1701 1867 1701
North Sumatra
ρ
̂ 0.235** (0.042) 0.310** (0.027) 0.398** (0.033)
R2 0.056 0.096 0.131
Observations 597 1236 1260
Jakarta Fig. 6. Upward Mobility Index over Regions.
ρ
̂ 0.341** (0.033) 0.359** (0.018) 0.373** (0.023)
R2 0.115 0.179 0.198
Observations 765 1613 1129
South Sulawesi
ρ
̂ 0.326** (0.035) 0.311** (0.027) 0.401** (0.033)
R2 0.094 0.096 0.132
Observations 799 1235 1258
Lampung
ρ
̂ 0.366** (0.065) 0.358** (0.046) 0.421** (0.035)
R2 0.117 0.095 0.169
Observations 236 616 725
South Sumatra
ρ
̂ 0.478** (0.063) 0.480** (0.036) 0.597** (0.029)
R2 0.141 0.177 0.285
Observations 438 886 912
West Sumatra
ρ
̂ 0.391** (0.052) 0.339** (0.032) 0.281** (0.037)
R2 0.155 0.102 0.081
Observations 432 843 761
West Nusa Tenggara Fig. 7. Downward Mobility Index over Regions.
ρ
̂ 0.387** (0.050) 0.411** (0.027) 0.462** (0.024)
R2 0.152 0.188 0.232
growth have lower persistence estimates. Thus, in this case, we suggest
Observations 311 861 1034
Yogyakarta the redistribution policy by the government on spending related to
ρ
̂ 0.248** (0.046) 0.330** (0.020) 0.355** (0.024) human development. This will certainly facilitate more immobile re­
R2 0.061 0.192 0.217 gions to obtain higher mobility or lower persistence in education which
Observations 458 859 596 will reduce the problem arising from inequality of opportunities.
Bali Moreover, mobility index and stationary distribution results also
ρ
̂ 0.435** (0.044) 0.335** (0.038) 0.433** (0.032) suggest no convergence for the attainment of higher secondary or above
R2 0.153 0.097 0.215 education among the regions. But when plotting stationary distribution,
Observations 257 619 645 we observe more masses in to university education for least asset
East Jave holding regions like Yogyakarta and West Nusa Tenggara which suggest
0.548** (0.030) 0.512** (0.020) 0.502** (0.020)
the education-related policy is able to uplift the people towards uni­
ρ
̂
R2 0.237 0.326 0.218
versity education in these regions. Chetty et al. (2014) correlated the
Observations 1048 2019 1820
regional variation in their intergenerational mobility measure for the
Robust standard errors in parentheses. United States along with local area characteristics. They concluded
** p < 0.01, * p < 0.05, + p < 0.1. regional variation in public funding of education may be associated with
the observed regional variation in intergenerational mobility in income.
HDI scores6 . Even though some regencies in Central Java such as Similarly, it could be the topic of interest in the future to evaluate the
Kebumen, Wonogiri, Purvorejo, Wonosobo, and Sragen have a high role of public expenditures in education to understand the existing
ILEG Index score, these education policies should spread to other re­ regional differences in persistence for Indonesia.
gencies. Besides, the intergenerational effects of policies of these mu­
nicipalities may take time to realize because Education Law and Author statement
subsequent regulations allowing local governments to implement their
education policies are recent developments (2003 and later). Rana­ We hereby confirm that the corresponding author Dr. Syed Hassan
singhe (2015) documented the similar trends in persistence across Raza proposed the idea, worked on relevant literature, and wrote the
different regions of Australia as they find regions with higher economic paper. Whereas, the second author Dr. Ugur Aytun managed the entire
work related to data management such as data clearing, estimations, and
most importantly helped the corresponding author to address the re­
6
viewer’s comments.
https://www.bps.go.id/dynamictable/2020/02/18/1772/indeks-pemban­
gunan-manusia-menurut-provinsi-metode-baru-2010− 2019.html

11
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Fig. 8. Stationary Distribution over Regions.

Declaration of Competing Interest

The authors report no declarations of interest.

Appendix A

Cohort Analysis by ethnicity

We moved our intergenerational mobility analysis to ethnicity since Table A1

the development history of Indonesia is also subject to the problems Descriptive statistics on ethnic groups in Indonesia.
rooted in ethnic differences such as national integration of minorities Father-Son Sample
and common language which are closely related to educational out­ Variables 1946− 1960 1961− 1975 1976− 1990
comes. For example, in rural areas, ethnic languages had been widely Javanese
used. According to a survey conducted in 1976 showed that only 26 % of Father’s Years of schooling 6.28 6.89 7.45
Indonesian school children used the Indonesian national language at Son’s Years of schooling 8.02 9.66 10.61
Sundanese
home (Mani, 1980). Besides, anecdotal evidences based on challenges Father’s Years of schooling 6.59 7.36 7.69
Chinese Indonesians faced such as sudden changes in language of in­ Son’s Years of schooling 8.03 9.92 10.21
struction in 1965 (Suryadinata, 1988) and being non-native to the Other Southern Sumatrans
Indonesian national language among other ethnicities may affect their Father’s Years of schooling 6.45 7.92 7.84
Son’s Years of schooling 8.19 10.15 10.13
educational outcomes across generations7. On the other hand, different
Batak
assimilation policies between the indigenous population (amalgamation Father’s Years of schooling 7.59 7.83 8.50
model) and minorities (absorption model) might have led to creating Son’s Years of schooling 8.71 11.05 11.67
different intergenerational persistence patterns, especially for those Betawi
born between 1950 and 1970. Father’s Years of schooling 5.03 6.97 7.39
Son’s Years of schooling 7.87 10.10 10.88
For this part of the analysis, we used the broader cohort range as it Minang
has three major benefits, first it makes the analysis less complicated and Father’s Years of schooling 7.04 7.68 8.67
secondly we are able to have more observations and lastly it helps us to Son’s Years of schooling 8.76 10.43 11.98
visualize the result in a precise way. We focus on six major ethnicities Other Ethinicities
Father’s Years of schooling 6.24 6.49 7.27
Son’s Years of schooling 7.65 9.45 10.42

Source: Author’s calculation from IFLS (Notes: We used broader cohorts here
7
In our ethnicity analysis we do not examine the Chinese Indonesians due to and calculated average years of schooling for both fathers’ and sons’ education
lower number of observations. distribution by ethnicity).

12
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Table A2
Mobility estimates on ethnic groups.
Father-Son Sample

Variables 1946− 1960 1961− 1975 1976− 1990

Javanese
ρ
̂ 0.404** (0.043) 0.435**(0.012) 0.520**(0.010)
R2 0.119 0.250 0.271
Observations 610 2946 5866
Sundanese
ρ
̂ 0.469**(0.071) 0.424**(0.023) 0.563**(0.044)
R2 0.136 0.225 0.312
Observations 241 924 1663
Other Southern Sumatrans
ρ
̂ 0.112 (0.379) 0.337**(0.058) 0.574**(0.036)
R2 0.004 0.139 0.264
Observations 47 189 528
Batak
ρ
̂ 0.066(0.122) 0.298**(0.051) 0.365**(0.036)
R2 0.005 0.099 0.151
Observations 52 300 704
Betawi
ρ
̂ 0.354** (0.117) 0.349**(0.036) 0.362** (0.033)
R2 0.138 0.178 0.145
Observations 46 282 664
Minang
ρ
̂ 0.198 (0.175) 0.266** (0.044) 0.323**(0.030)
R2 0.027 0.082 0.126
Observations 77 366 802
Other Ethinicities
ρ
̂ 0.382** (0.051) 0.402** (0.018) 0.468** (0.012)
R2 0.141 0.204 0.244
Observations 395 1962 4578

Robust standard errors in parentheses.

** p < 0.01, * p < 0.05, + p < 0.1.

where we have enough observations and rest of the ethnic groups are Indonesian population which is 40.2 %, followed by 15.4 % Sundanese
grouped as other ethnicities. For each ethnicity we have three cohorts, and rest of the ethnic group remains less than 4%. Since IFLS is not
each with 15 years gap starting from 1946 to 1990. As per native ethnicity representative of Indonesia so our estimates from this part of
Indonesian (Census, 2009) Javanese constitutes the biggest chunk of the analysis can not be generalized. Although estimates can not be

Fig. A1. Upward Mobility Index over Ethnicities.

13
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Fig. A2. Downward Mobility Index over Ethnicities.

Fig. A3. Stationary Distribution Plot over Ethnicities.

14
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

generalized but this dimension of our analysis open us new area of there is increase in the value of stationary distribution in every cohort.
research in context of Indonesia. Therefore, with the availability of right Hence, we document higher stationary distribution for university edu­
dataset this particular dimension can be reinvestigated in the future. cation in Minang in comparison to other ethnicities. This pattern is
In data we observed, historically some of the ethnicities had higher different from rest of the ethnic groups, thus, over time stationary dis­
asset holdings such as Other South Sumatrans but in the most recent tribution kept on changing for every education stage and this is non-
cohort we observe Minang has higher wealth in terms of average asset convergence.
holdings. Thus, in Table A1 we observe highest change for Minang in
average years of schooling for sons’ which is 3.22 years. Whereas, in Appendix B
particular two population wise biggest ethnicities which are Javanese
and Sundanese for them the change in the education distribution of fa­
thers is also not remarkable. Although, it can be argued when time
started fathers in these two particular group were already located at
higher end of education distribution. Moreover, we document highly
Table B1
educated ethnic groups for both fathers and sons education distribution
Mobility estimates over regions using wave 1.
is Minang, however, from 1961− 1975 we observe fathers in Other South
Father-Son Sample
Sumatrans have highest average years of schooling.
In Table A2 we report the persistence estimates for six major eth­ Regions 1946− 1960 1961− 1975 1976− 1990
nicities of Indonesia. In context of observing the trend in persistence, the West Java
0.536** (0.028) 0.548** (0.024) 0.642** (0.057)
results are very different to our estimates what we observe in nationwide
ρ
̂
0.014
sample. For example, in some of the ethnic groups such as Javanese and R 2 0.263 0.390 0.214
other ethnicities there is increasing and significant trend in persistence Observations 956 967 1432
estimates but for some of the ethnic groups like Betawi and Sundanese Central Java
we observe no such trend. We observe higher persistence in bigger ρ
̂ 0.429** (0.035) 0.524** (0.031) 0.162* (0.066)
ethnic groups like Javanese, Sundanese and Other Sumatrans but in R2 0.178 0.258 0.005
small groups as we observe higher variation in the distribution of edu­ Observations 742 704 1124
cation in Table A1, thus, our persistence estimates compliment those North Sumatra
0.310** (0.036) 0.427** (0.042) 0.442** (0.094)
facts because small ethnicities are least persistent. In particular to our ρ
̂
R2 0.150 0.170 0.015
last cohort we found Other Sumatrans to be the most persistent ethnic
group with 0.574 correlation coefficient and Minang was least persistent Observations 493 507 913
Jakarta
ethnicity with 0.323 correlation coefficient. 0.429** (0.030) 0.480** (0.027) 0.122 (0.067)
ρ
̂
Similar to decomposition exercise for nationwide sample here our
R2 0.225 0.276 0.004
motivation remains same due to unclear trend in persistence of different
Observations 638 802 863
ethnic groups we obtained Term B and further by using equation (4) and South Sulawesi
(5) we construct upward and downward mobility index. In Fig. A1 we ρ
̂ 0.372** (0.081) 0.353** (0.046) 0.430**(0.032)
plot upward mobility index for the concerned ethnic groups. Here we R2 0.155 0.166 0.110
observe starting with first cohort there are substantial differences in Observations 377 360 572
upward probabilities but during 1961− 1975 we observe close conver­ Lampung
gence from all the ethnic groups. This implies that probability of son ρ
̂ 0.282** (0.059) 0.415** (0.068) 0.060 (0.118)
achieving more education than his father increased in all ethnic groups. R2 0.073 0.154 0.001
It is important to note this is the same time when government of Observations 300 249 465
Indonesia initiated massive school expansion program and started South Sumatra
0.400** (0.047) 0.428** (0.065) 0.489** (0.098)
providing free education up till higher secondary. It is also argued by
ρ
̂
R2 0.149 0.133 0.027
Hertz and Jayasundera (2007) that school expansion policy in mid-70′ s
Observations 432 342 585
implemented by the Government of Indonesia significantly enhanced
West Sumatra
the education attainment and major contribution is driven by sons of less ρ
̂ 0.363** (0.057) 0.325** (0.059) 0.056 (0.109)
educated parents. Interestingly, as we move along the last cohort again 0.073 0.096 0.001
R2
we notice the differences in upward probabilities index of all the eth­
Observations 274 316 521
nicities in the analysis. In Fig. A2 we document downward mobility West Nusa Tenggara
index and there we observe the similar change in 1961− 1975 cohort ρ
̂ 0.509** (0.048) 0.545** (0.047) 0.189* (0.081)
which means decline in probability that son will attain less education R2 0.186 0.227 0.009
than his father. This declining is more apparent in less-populated groups Observations 439 393 691
like Batak, Betawi and Minang because their initial index values are very Yogyakarta
high. From this perspective, we have concluded that education is an ρ
̂ 0.299** (0.051) 0.418** (0.031) 0.442** (0.034)

important factor for these minorities to be successful. Moreover, human R2 0.082 0.288 0.265
capital upgrading of these minorities contributes to Indonesian econ­ Observations 374 351 380
Bali
omy. Now it has become even important to find whether there exists
ρ
̂ 0.328** (0.057) 0.356** (0.027) 0.162 (0.100)
some long run convergence for different stages of education attainment
R2 0.108 0.167 0.006
in these ethnicities or not.
Observations 257 332 410
Therefore, in order to investigate the long run convergence, we plot
East Java
stationary distribution in Fig. A3 for six major ethnicities. We observe ρ
̂ 0.555** (0.031) 0.510** (0.032) 0.246** (0.065)
increase in long run probability towards attainment of university edu­ R2 0.224 0.249 0.014
cation in all the ethnic groups. But differential patterns negate the ex­ Observations 986 851 1126
istence of convergence about various stages of education attainment or
even across ethnic groups. For example, in ethnic group Minang we Robust standard errors in parentheses.
observe over time decline in stationary distribution for no-schooling,
primary, middle, and higher secondary but for university education

15
S.H. Raza and U. Aytun International Journal of Educational Development 81 (2021) 102348

Table B2
Direction of mobility estimates by using all five wave and first wave (first wave is provincial representative).
First to Second Cohort Second to Third Cohort

West Java
Direction of coefficient in all five waves ↑ ↑
Direction of coefficient in the first wave ↑ ↑
Central Java
Direction of coefficient in all five waves ↓ ↑
Direction of coefficient in the first wave ↑ ↓
North Sumatra
Direction of coefficient in all five waves ↑ ↑
Direction of coefficient in the first wave ↑ ↑
Jakarta
Direction of coefficient in all five waves ↑ ↑
Direction of coefficient in the first wave ↑ Inconclusive as the estimate of 3rd cohort is insignificant
South Sulawesi
Direction of coefficient in all five waves ↓ ↑
Direction of coefficient in the first wave ↓ ↑
Lampung
Direction of coefficient in all five waves ↓ ↑
Direction of coefficient in the first wave ↑ Inconclusive as the estimate of 3rd cohort is insignificant
South Sumatra
Direction of coefficient in all five waves ↑ ↑
Direction of coefficient in the first wave ↑ ↑
West Sumatra
Direction of coefficient in all five waves ↓ ↓
Direction of coefficient in the first wave ↓ Inconclusive as the estimate of 3rd cohort is insignificant
West Nusa Tenggara
Direction of coefficient in all five waves ↑ ↑
Direction of coefficient in the first wave ↑ Inconclusive as the estimate of 3rd cohort is insignificant
Yogyakarta
Direction of coefficient in all five waves ↑ ↑
Direction of coefficient in the first wave ↑ ↑
Bali
Direction of coefficient in all five waves ↓ ↑
Direction of coefficient in the first wave ↑ Inconclusive as the estimate of 3rd cohort is insignificant
East Java
Direction of coefficient in all five waves ↓ ↓
Direction of coefficient in the first wave ↓ ↓

References Hertz, T., Jayasundera, T., 2007. School Construction and Intergenerational Mobility in
Indonesia (No. 2007-18).
Hertz, T., Jayasundera, T., Piraino, P., Selcuk, S., Smith, N., Verashchagina, A., 2007. The
Akresh, R., Halim, D., Kleemans, M., 2018. Long-term and Intergenerational Effects of
inheritance of educational inequality: international comparisons and fifty-year
Education: Evidence From School Construction in Indonesia (No. w25265). National
trends. BE J. Econ. Anal. Policy 7 (2).
Bureau of Economic Research.
Hill, C., 2008. International business: competing in the global market place. Strateg. Dir.
Al-Samarrai, S., 2013. Local Governance and Education Performance: A Survey of the
24 (9).
Quality of Local Education Governance in 50 Indonesian Districts, Human
Huang, J., 2013. Intergenerational transmission of educational attainment: the role of
Development. World Bank, Jakarta.
household assets. Econ. Educ. Rev. 33, 112–123.
Aspinall, E., Klinken, G.V., 2011. The State and Illegality in Indonesia, p. 328. Brill.
Kanbur, R., Wagstaff, A., 2014. How Useful Is Inequality of Opportunity As a Policy
Aydemir, A.B., Yazici, H., 2019. Intergenerational education mobility and the level of
Construct?.
development. Eur. Econ. Rev. 116, 160–185.
Kwenda, P., Ntuli, M., Gwatidzo, T., 2015. Temporal developments in intergenerational
Azam, M., Bhatt, V., 2015. Like father, like son? Intergenerational educational mobility
transmission of education: case for black South Africans. Res. Soc. Stratif. Mobil. 42,
in India. Demography 52 (6), 1929–1959.
96–113.
Balisacan, A.M., Pernia, E.M., Asra, A., 2003. Revisiting growth and poverty reduction in
Mani, A., 1980. Determinants of Educational Aspirations Among Indonesian Youth. Ph.D.
Indonesia: what do subnational data show? Bull. Indones. Econ. Stud. 39 (3),
dissertation submitted to the University of Wisconsin, Madison.
329–351.
Neidhöfer, G., Serrano, J., Gasparini, L., 2018. Educational inequality and
Bennington, L., Habir, A.D., 2003. Human resource management in Indonesia. Hum.
intergenerational mobility in Latin America: a new database. J. Dev. Econ. 134,
Resour. Manage. Rev. 13 (3), 373–392.
329–349.
Björklund, A., Salvanes, K.G., 2011. Education and family background: mechanisms and
Organisation for Economic Co-operation and Development (OECD) & Asian
policies. In: Handbook of the Economics of Education, vol. 3. Elsevier, pp. 201–247.
Development Bank, 2015. Education in Indonesia. Rising to the Challenge.
Black, S.E., Devereux, P.J., 2010. Recent Developments in Intergenerational Mobility
Prasiwi, Y., Susandari, S., 2015. Studi Deskriptif Mengenai Adaptasi Sosiokultural Dan
(No. w15889). National Bureau of Economic Research.
Psikologikal Pada Mahasiswa Etnik Minang Dan Batak Di Bandung.
Black, S.E., Devereux, P.J., Salvanes, K.G., 2011. Older and wiser? Birth order and IQ of
Ramdani, R., Borualogo, I.S., Hamdan, S.R., 2015. Kebanggaan atas Identitas Etnik pada
young men. CESifo Econ. Stud. 57 (1), 103–120.
Mahasiswa Perantau Kelompok Etnik Minang dan Batak di Bandung (Ethnic Identity
Checchi, D., García-Peñalosa, C., 2008. Labour market institutions and income
of Migrated Students from Minang and Batak Ethnic Group in Bandung). In:
inequality. Econ. Policy 23 (56), 602–649.
Prosiding Penelitian Sivitas Akademika Universitas Islam Bandung, 2, pp. 455–459.
Checchi, D., Fiorio, C.V., Leonardi, M., 2013. Intergenerational persistence of
Ranasinghe, R., 2015. The transmission of education across generations: evidence from
educational attainment in Italy. Econ. Lett. 118 (1), 229–232.
Australia. B. E. Anal. Policy 15 (4), 1893–1917.
Chetty, R., Hendren, N., Kline, P., Saez, E., 2014. Where is the land of opportunity? The
Roemer, J.E., 1998. Theories of Distributive Justice. Harvard University Press.
geography of intergenerational mobility in the United States. Q. J. Econ. 129 (4),
Rosser, A., 2018. Beyond Access: Making Indonesia’s Education System Work.
1553–1623.
Suryadinata, L., 1988. Government policy and National integration in Indonesia.
Daude, C., 2011. Ascendance by descendants? On intergenerational education mobility
Southeast Asian J. Soc. Sci. 16 (2).
in Latin America. In: OECD Development Centre Working Papers, 297, p. 1.
World Bank, 2005. World Development Report, 2006. World Bank & Oxford University
EDUCATION SYSTEM PROFILES, 2019. Education in Indonesia. https://wenr.wes.
Press, Washington & New York.
org/2019/03/education-in-indonesia-2.

16