Religious Education in the Age of Mass Schooling

Islam and the State: Religious Education in the Age of Mass Schooling Abstract We study the competition between public and religious schools and its consequences for nation building. In the 1970s, a landmark mass schooling effort in Indonesia aimed to secularize education and to curb religious influence in society. The regime built 61,000 public elementary schools, seeking in part to upend a longstanding Islamic school system. Using novel data on Islamic school construction and curriculum, we identify short-run effects on exposed cohorts as well as dynamic, long-run effects on education markets. While primary enrollment shifted towards state schools, religious education increased on net as Islamic secondary schools absorbed the higher demand for continued education. The Islamic sector not only entered new markets to compete with the state but also increased religious curriculum at newly created schools. While exposed cohorts are not more attached to secular principles, they report greater religiosity and transmit these religious values to the next generation. Overall, the ideological competition in education undermined the nation-building impacts of mass schooling. 1 Introduction Providing education is one of the central missions of modern states. Yet, mass public schooling is a recent historical phenomenon. For centuries, religious organizations dominated education markets across Western Europe, the Middle East, and elsewhere. In many countries, the state overturned this dominance through secularization policies with notable examples in the Ferry Laws in France or Kemalist reforms in Turkey. In other countries, religious schools still cater to a large share of the student population. Across these settings, little is known about how competition between state and religious schools shapes identity and nation building (Alesina et al., 2019). Understanding this process is of central importance in contexts where religious schools account for a large share of private education. The literature on school competition has generally overlooked the religious dimension of education as a key margin of school choice. Yet, these choices may be crucial in shaping cultural transmission (Bisin and Verdier, 2000). Other recent work has examined the link between schooling reforms and ideology, but has not explored the competitive response to state expansion in education markets (Bandiera et al., 2019; Cantoni et al., 2017). In this paper, we study how religious school choice and competition between state and religious schools affect cultural change. We explore the consequences of mass public schooling for identity and nation building in the world’s largest Muslim country. We show that Islamic schools counteracted the secularizing impacts of mass schooling. Indonesia provides a rich context for understanding how religious schools endure in modernizing states. Millions of Indonesians were educated in religious institutions historically, and around one-fifth of students attended Islamic schools in 2019 (see Table 1). Yet, in the 1970s, the country underwent a uniquely ambitious expansion of its public schooling system through the celebrated Sekolah Dasar (SD) Presidential Instruction (INPRES). This policy oversaw the construction of more than 61,000 elementary schools—many of them in areas where Islamic institutions long played an important role. Beginning with Duflo (2001), a large literature has studied the impacts of SD INPRES on human capital and development. However, the policy also entailed political objectives in the wake of turmoil in the 1960s that saw the demise of the Communist movement and the growing strength of Islamic institutions. Rooted in the Suharto regime’s hostility towards organized Islam, SD INPRES was designed to foster nation building and to curb religious influence in society (Boland, 1982; Kelabora, 1976). Mass secular schooling would hasten the transition to a single national curriculum and render Islamic schools irrelevant. We study how the religious sector adapted and responded to this effort. Our analysis identifies short-term effects of the policy on exposed cohorts as well as dynamic, longterm effects on education markets with varying INPRES school construction in the 1970s. Several novel data sources allow us to explore, for the first time, how the policy shaped multiple dimensions of schooling content. These data include nationally-representative surveys capturing Islamic education as well as administrative records on the universe of schools with date and location of establishment. The latter record nearly 220,000 secular and 80,000 Islamic schools. This allows us to characterize the evolution of demand- and supply-side responses over the ensuing decades. For some schools, we also observe a breakdown of curriculum hours in 2019, which we use to quantify the religious content of subject matter and to measure long-run differences in ideological differentiation. 1 Perhaps unexpectedly, public school expansion increased overall exposure to Islamic education. Intuitively, SD INPRES decreased attendance in primary Islamic day schools (madrasa) in the short run. However, religious schools absorbed some of the increased demand for secondary schooling that resulted from mass primary schooling. As many as three-quarters of Islamic junior secondary school students originated in public primary schools. Ultimately, this demand effect increased the likelihood that INPRES-exposed cohorts would attend an Islamic school throughout their educational years. In other words, SD INPRES increased not only years of schooling—as identified in prior work—but also exposure to Islamic education, which was arguably contrary to the regime’s objective. These school choices were shaped by the Islamic sector response. Overall, Islamic school construction increased in districts with greater SD INPRES intensity. In the short-run, secondary madrasa entered to capitalize on growing demand for continued schooling among graduates from INPRES elementary schools. We observe this strategic entry within very local education markets, below the district level. Secondary madrasa continued to enter high-INPRES markets differentially over the ensuing decades. While elementary madrasa did not enter immediately to compete with nearby INPRES schools, they began to do so systematically around the mid-1980s. A simple Stackelberg competition framework provides intuition for this more surprising result. Strategic complementarities between the two education systems led the Islamic sector to increase its supply of schools in response to SD INPRES. As a result, the state expansion in education markets failed to crowd out Islamic schools.1 In addition to the quantity response, Islamic schools entering high-INPRES districts after the program provided greater curriculum differentiation at the primary and junior secondary level. We measure differentiation based on classroom hours devoted to Islamic subjects, e.g., Islamic law (fiqh), theology (aqidah), and ethics (akhlaq), as well as Arabic instruction. The increase in Islamic content comes at the expense of subjects emphasized in the standard curriculum, including study of the national language and Pancasila, the secular ideology of the state. Such differentiation may have been welfare-enhancing to the extent that variation in Islamic content addressed heterogeneous preferences for different types of schooling. Overall, though, the quantity and ideological responses of the Islamic sector counteracted the state’s efforts to homogenize and secularize education. These results open a new window into the celebrated SD INPRES program and show how the ideological effects of mass schooling depend on the response of non-state schools. We establish the plausibility of the parallel trends assumption not only for years of schooling as in Duflo (2001) but also for religious schooling rates at all instruction levels, and for establishments of new Islamic schools. Although the state may have targeted SD INPRES towards districts with a greater prevalence of Islamic schools, it did not target on the basis of differential trends in Islamic school construction or religious schooling rates prior to the 1970s. While these patterns lend themselves to a causal interpretation, we 1 Several potential mechanisms underlie these strategic complementarities. First, transitions between secular and religious schools are common. In the Indonesian Family Life Survey (IFLS), 31% of students graduating from Islamic primary (at age 12) attend secular junior secondary between ages 12–15, and 48% of students graduating from Islamic junior secondary at age 15 attend secular senior secondary between ages 15–18. Second, many secondary madrasa are built in the same physical location as primary madrasa to take advantage of lower costs and increased demand. Third, elementary madrasa may have been more attractive than INPRES elementary schools for conservative families that remained reluctant to send their daughters to secular schools, especially after the adoption of a headscarf ban in public schools in 1982 (see Section 5.2). 2 note that the long-run effects capture a complex set of responses that unfold well after the 1970s policy shock. Like with any difference-in-difference, the farther one moves from the intervention period, the less compelling the causal inference. Nevertheless, given that mass schooling programs like SD INPRES occur at critical junctures of development, the dynamics we identify may reflect trajectories that would not have otherwise emerged absent the program. While the state also expanded secondary schools after SD INPRES, it was not able to fully counteract the Islamic sector’s ability to capture new primary school graduates. There are at least two explanations for this differential response. On the state side, a budgetary shock in the early 1980s due to declining oil revenue led to major cutbacks in education spending. The regime may also have prioritized primary school expansion because it believed that indoctrination was best realized at that level. Meanwhile, the Islamic sector was able to leverage its major charitable institution, known as the waqf, to expand educational infrastructure. This protected revenue stream, built on private charity, helps Islamic organizations compete with the state in the education sector across the Muslim world. We show that the Islamic school supply response was, in fact, stronger in districts with a larger waqf base prior to SD INPRES.2 Using a combination of electoral, census, and survey data, we further characterize the legacy of SD INPRES for identity and nation building. Overall, the policy did not shore up support for the Suharto regime, and it did not increase attachment to standard markers of Indonesian identity. Strikingly, SD INPRES did not benefit Suharto’s political party, Golkar, in the 1977 and 1982 elections, nor after 1987 when affected cohorts began to vote. Instead, Islamic parties, the main opposition, gained in districts with greater INPRES intensity, consistent with backlash against the secular state. In the long run, schoolage exposure to SD INPRES did not increase support for Pancasila, use of the national language, or affinity with secular principles. Instead, exposed cohorts report greater attachment to Islam: they are more likely to be literate in Arabic, a core part of the curriculum in Islamic schools, and also exhibit greater piety across a range of Islamic practices. Moreover, the patterns of Islamic school choice we observe among exposed cohorts are passed on to children in the next generation, who are more likely to inherit the religious identity of their parents. Together, these results suggest that the policy fell short of its ideological objectives through a combination of exposure to religious education and increased transmission of Islamic values. On the supply side, the increased provision of religious schooling allowed the Islamic sector to attract large numbers of SD INPRES graduates. On the demand side, the secularization of education prompted some families to prioritize religious school choice and cultural transmission (Bisin et al., 2020). As a result, mass public schooling did not bolster national identity or support for the secular state, and it did not decrease religiosity in the long run. At the same time, the greater piety among affected cohorts was not accompanied by greater support for sharia law. Ultimately, the state failed to curb religious influence in society, but it may have been more effective at stifling Islamism inside the classroom. In this regard, some of our findings stand in contrast to Bazzi et al. (2020) who show that a resource 2 In practice, the Islamic sector is comprised of schools run by large Islamic foundations or organizations (such as Muhammadiyah and Nahdlatul Ulama) as well as independent establishments funded through their own autonomous waqf endowments. We do not distinguish between these two types of schools, but note that the existence of large institutional actors implies some degree of coordination within the otherwise decentralized Islamic school system. 3 windfall for Islamic institutions in the 1960s caused a shift towards Islamism. While waqf endowments empowered Islamists in the 1960s, they also enabled mainstream Islamic actors to counteract the state’s attempt to secularize education a decade later, and ultimately to strengthen religious identity. The distinct shifts in piety and religious politics across the two studies have organizational and cultural roots. In the 1960s, the Islamists, long-repressed by the state, benefitted most from an increase in their resource base. By contrast, in the 1970s, moderate organizations like Muhammadiyah and Nahdlatul Ulama, with their extensive resources and long history of providing Islamic education, were best positioned to capitalize on the growing demand for secondary education that resulted from mass primary schooling. Moreover, unlike the 1960s land reform that targeted agrarian elites, mass secular education challenged religious identity and values for society at large. Many families invested in preserving their religious identity by choosing religious schools and by transmitting religious values to their children. Across the two studies, the resources and legal protections available to Islamic institutions may have proven decisive in their resilience against the secularization of society. Historically, religious institutions in Europe often did not benefit from such protections. The same regimes that enforced sweeping secularization reforms in education were also careful to strip the Church of its physical assets. In contrast to the Muslim world, this may have contributed to the relative demise of religious organizations in education and other markets. Related Literature. There is growing evidence on the crucial role of education policy in nation building.3 Some studies show that education fosters civic engagement (Dee, 2004; Larreguy and Marshall, 2017). Many show that education weakens religiosity (e.g., Glaeser and Sacerdote, 2008; Hungerman, 2014; Mocan and Pogorelova, 2017), while others show that mass schooling led to the decline of church attendance in historical Germany (Becker et al., 2017) and to lower piety in contemporary Turkey (Gulesci and Meyersson, 2016). Our paper is among the first to link mass schooling to greater religiosity, at the expense of secular nation-building objectives. Our key innovation lies in understanding how the state’s historical competitor in education provision—religious organizations—responded to mass schooling. Two recent studies explore the effects of education reforms in France and Turkey. Squicciarini (2020) shows how the Catholic Church, through investments in religious schools, slowed the diffusion of technical knowledge in 19th century France. Sakalli (2019) shows that religious families in Turkey pushed back against state efforts at secularization by removing their children from public schools. In contrast, we investigate competition between Islam and the state after one of the largest school expansion programs ever implemented. Ultimately, the Islamic sector response contributed to the program’s limited impacts on nation building. Prior work on SD INPRES has not explored the Islamic sector response or the program’s nationbuilding consequences. Recent work by Akresh et al. (2018) and Mazumder et al. (2019) identify the long-term and intergenerational effects on similar outcomes as Duflo (2001), while Ashraf et al. (2020) show that the policy had large effects on education for women from ethnic groups with a bride price 3 Alesina et al. (2019) describe the experiences of European states and provide a model formalizing the role of mass schooling. Cantoni et al. (2017) study how a curriculum reform affected political attitudes in China. Bandiera et al. (2019) link the rise of public schooling to immigration in the U.S. Other related work can be found across the social sciences with leading examples in political science (Ansell and Lindvall, 2013; Paglayan, 2017, 2018) and sociology (Meyer et al., 1979). 4 tradition. Martinez-Bravo (2017), Roth and Sumarto (2015), and Rohner and Saia (2019) study the impacts on governance, intergroup tolerance, and conflict, respectively. While these studies also show how education affects political economy outcomes, we explore how religion and politics shape educational outcomes and, in turn, the long-term consequences of mass schooling. Our findings offer insight into the general equilibrium effects of education policy in societies with a strong religious schooling sector. With the benefit of new data, we show how the endogenous response of the Islamic sector works against the homogenizing effects of mass schooling. These insights also advance the literature on religious schooling and its consequences for religious cultural transmission. The education literature in the U.S. has explored the returns to Catholic schooling (Altonji et al., 2005; Neal, 1997). Andrabi et al. (2006) and Berman and Stepanyan (2004) provide descriptive background on Islamic schooling in Pakistan and a range of Muslim countries, respectively. Few studies in this literature distinguish between private and religious schools, which often pursue distinct ideological objectives. In the tradition of Bisin and Verdier (2000, 2001), many argue that parents make school choices so as to influence the horizontal or “oblique” transmission of cultural values. For example, Cohen-Zada (2006) models religious school choice as a function of religious group shares in society. Carvalho and Koyama (2016) describe how historically marginalized communities may underinvest in education as a form of cultural resistance. We show that parents responded to a secular education policy by increasing the scope for religious cultural transmission through their choice of religious schools. 2 Background: Islam and Education in Indonesia Indonesia’s longstanding dual education system reflects the enduring role of religious schools in a country home to more than 225 million Muslims.4 This section provides background on religious schools, education policy and the SD INPRES program, school curricula, and education markets. 2.1 Typology of Islamic Schools Indonesia’s education system is comprised of secular and religious schools. Secular education is provided by public as well as private schools—76% of secular schools are public, but 90% are at the primary level and only 50% at the senior secondary level. Secular schools fall under the regulatory authority of the Ministry of Education and Culture (Kemdikbud or MEC) since the 1970s (see Section 2.2). Parallel to secular schools, there are two types of Islamic schools—madrasa and pesantren—that played a major role in the transmission of human capital and culture for much of Indonesian history. Madrasa are day schools that use pedagogical methods similar to secular schools but offer substantially more religious content in their curriculum (see Section 2.4). There is an exact correspondence between education levels in the madrasa system and the secular system. Elementary madrasa (Madrasah Ibtidaiyah or MI) correspond to public elementary schools (Sekolah Dasar or SD), while junior secondary madrasa (Madrasah Tsanawiyah or MTs) and senior secondary madrasa (Madrasah Aliyah or MA) are the Islamic counterparts to junior (Sekolah Menengah Pertama or SMP) and senior (Sekolah Menengah Atas 4 Nearly ninety percent of Indonesians are Muslim with Christians being the largest minority religion. 5 or SMA) secondary schools, respectively. Outside this nomenclature, other schools known as Madrasa Diniyah, often operating as afternoon schools (akin to Bible study), exclusively teach Islamic subjects. Pesantren are boarding schools devoted to the study of Islam. Similar to Christian seminaries, many pesantren are geared towards producing religious scholars, though they typically offer instruction across multiple levels of education beginning as young as age 6. Compared to madrasa, pesantren tend to have more religious instruction, less regulatory oversight, and a more politically active orientation. The vast majority of Islamic schools are privately run. This includes all pesantren and 92% of madrasa in 2019. The small number of state-run madrasa largely originate out of a central government initiative in the late 1950s to take over Islamic schools run by provincial governments. In 1967, the regime invited all private madrasa to become state-run and gain access to additional funding. This effort fell flat as most Islamic schools opted to remain private. The Islamic sector’s strong desire to maintain independence is at the heart of Indonesia’s dual education system. 2.2 SD INPRES and the Origins of the Dual System The literature provides rich background on the SD INPRES program. Here, we provide additional details on the historical context of state efforts to confront Islamic schools. Origins of the Dual System. At independence in 1945, amidst a wider debate about the place of Islam in Indonesia’s constitution, the state established a secular education system. Religious schools were placed under the purview of the Ministry of Religious Affairs (Kemenag or MORA), which sought to extend its influence over madrasa (Kelabora, 1976). For example, in 1958 a major reform effort failed to limit religious instruction time to 21–28% of study hours. Throughout the 1950s and 1960s, “the strong commitment of the Muslim community to having their own education system . . . made it impossible for the government to replace Islamic schools with non-religious schools” (Zuhdi, 2006, p. 75). In the early years of Suharto’s New Order regime, nation building became a central priority. At first, Islamic leaders were associated with this effort, having helped defeat Communist forces during the political upheaval of the mid-1960s. As testament to this alliance, the government made 2–4 hours of mandatory religious instruction in public schools in 1967. However, the regime stance towards Islamic education rapidly changed as it embraced an overarching policy of suppressing political Islam (Boland, 1982). This manifested in the decision to force four existing Islamic political organizations into the single umbrella United Development Party (Partai Persatuan Pembangunan or PPP) in 1973. Less than a decade later, the government adopted a ban on religious veiling inside public schools (see Section 5.2). Sekolah Dasar (SD) INPRES. It is in this context that the government launched SD INPRES. Equipped with windfall oil revenues from the early 1970s, the Suharto regime prioritized development spending. This included a large allocation for primary school construction to meet the new compulsory requirement introduced in 1973. The Presidential Instruction No. 10/1973 and subsequent yearly decrees specified the funding allocated to each district as a function of the child population not enrolled in school. Each school package funded the construction of a primary school for grades 1–6 with six classrooms. In total, up to 61,000 schools were constructed between 1973–80 under the program, with each district 6 receiving anywhere between 16 and 824 new elementary (SD) schools.5 The school expansion program was accompanied by mass hires of teachers and the removal of primary school fees in 1977. The expansion of the state school system entailed both developmental and political objectives. SD INPRES aimed at secularizing and homogenizing primary education. Civic education was to supplant certain Islamic subjects, while instruction was to take place in the national language, Bahasa Indonesia, rather than the local ethnic languages or Arabic.6 The goal was to build a citizenry steeped in the inclusive Pancasila ideology and invested in the national identity. A World Bank (1989) report notes that “. . . public education was viewed by the Government as a key medium for promoting national unity and national values—first, through instruction in Pancasila, and next through instruction in the national language, Bahasa Indonesia” (p. 14), and that “[i]n so large and dispersed a country . . . policymakers have consistently looked to neighborhood primary schools as vehicles for national integration” (p. 35). Parallel to the school expansion program, a 1972 decree stipulated that all formal education must be administered by the Ministry of Education. In the context of the regime’s growing hostility towards political Islam, this was interpreted as an attack on the status and independence of Islamic schools: “While there was no clear statement concerning the status of the Islamic schools . . . Muslim leaders interpreted that the Presidential Decree was intended, among other things, to weaken the status of the Islamic educational institutions. Since the decree did not specifically clarify the status of Islamic educational institutions, they assumed that the government was trying to eliminate these latter through the application of a so-called pendidikan satu atap (“single roof education”) policy.” (Zuhdi, 2006, p.89) The reform was strongly opposed by Muslim leaders and ultimately abandoned as part of a compromise between the MEC and MORA. In 1975, the government recognized the special status of Islamic education, allowing madrasa to remain under MORA authority. Subsequent reforms in 1984 and 1989 provided further recognition to Islamic school graduates on par with their secular school counterparts. Ultimately, though, the reform left the dual system in place: as of writing, Islamic schools remain under the MORA, which monitors quality and curriculum. 2.3 Curriculum Differences Between Secular and Islamic Schools Islamic schools teach a range of religious subjects that are not covered in secular schools. There are five core subjects: Islamic law (fiqh), Islamic doctrine and ethics (aqidah and akhlaq), Qur’an and traditions of Prophet Muhammad (hadith), Arabic language, and history of the Prophets (qisa al-anbiya). Zuhdi (2006) provides examples of curriculum timetables in elementary madrasa, pesantren, and public schools in the 1950s. Grade 6 students in the latter spent a total of 2 hours per week in religious education, whereas those in Islamic schools spent anywhere from 25 to 40 percent of instruction time on religious subjects. 5 The Presidential Decrees for 1973–74 (INPRES 10/1973 and 6/1974), 1975–76 (6/1975 and 3/1976), 1977–78 (3/1977 and 6/1978) and 1979–80 (12/1979 and 6/1980) authorized grants for 6,000, 10,000, 15,000, and 14,000 new schools, respectively. The total of these appears closer to the figure of 45,874 INPRES schools appearing in 1980 village-level administrative data known as Podes. In Tables A.6 and A.7, we show that the program’s impacts on years of schooling and religious schooling are robust to using this alternative measure of INPRES intensity (as in Martinez-Bravo, 2017). 6 Bahasa Indonesia is based on the minority ethnic language of Malay, which was spoken by only 5 percent of the country when it was chosen as the national language by leaders of the independence movement in 1928. 7 These sharp patterns of curriculum differentiation can be seen in contemporary data. Secular public schools largely adhere to 2 hours of religious instruction per week. Meanwhile, data described in Section 3 show that madrasa devote 26% of instruction hours to religious content on average with more hours at higher grade levels. There is considerable variation across madrasa (standard deviation of 6%) but a roughly equal breakdown in hours across the five subjects above, including Arabic. At the same time, only 5% of instruction is devoted to Pancasila and Civic Education and an additional 5% to the study of Indonesian language and literature. This large gap between hours devoted to Islamic content versus civics and the national language distinguishes student experiences in Islamic versus secular schools. 2.4 Education Markets and Religious School Choice Before introducing our data, we make three remarks on education markets and the scope for school choice. First, Islamic schools comprise a majority of all private schools (more than 60% nationally in 2019). Moreover, in many local markets, private school choice is tantamount to Islamic school choice. While secular private schools are important in some areas, they provide a very different learning experience from their religious private counterparts under the MORA. Second, at the local level, one finds considerable scope for religious versus secular school choice. For primary school, the village—home to 2,500–3,500 people on average—is the relevant education market. Here, we see, in 1990 for example, that 95% of villages with an elementary madrasa also have an elementary secular school. For secondary school, the education market often spans multiple villages, but even at this level, one finds that 43% of villages with an Islamic middle school also have a secular middle school.7 Given these patterns of local competition, it is not surprising that Islamic and secular school students report traveling similar distances to school in survey data from 2015 (Susenas). Third, while Islamic and state schools may differentiate on cost as well, such differentiation appears limited. We can only see this in contemporary data (Susenas 2015), which shows average annual costs, at the primary level, of roughly USD 20 for Islamic schools and USD 21 for state schools. At the middle school level, annual costs average USD 34 for state schools and USD 29 for madrasa. These figures suggest ample scope for competition across secular and religious schools along various margins besides cost. 3 Data We draw upon several new data sources that allow for the first systematic analysis of how SD INPRES affected education markets over the short- and long-run. We combine survey data on Islamic education with administrative data on Islamic school construction to shed new light on both the demand- and the supply-side response to the policy. With data on school curriculum, we characterize different margins of adjustment to mass schooling efforts by the state. Additional data sources help understand how the policy shaped identity and nation building over the long run. Survey Data on Schooling. We measure Islamic school attendance and other measures of education 7 These figures are based on contemporaneous Podes administrative data, described in the following section. 8 status using six rounds of the National Socioeconomic Survey (Susenas), collected between 2012–2018. The Supas 1995 intercensal survey data used by Duflo (2001) did not include information on Islamic education. While Susenas has reported breakdowns of madrasa and secular education since the late 1990s, the 2012 round was the first to include information on birthplace, which is needed to identify childhood exposure to SD INPRES. Additionally, we can link (co-resident) children’s schooling to their parents’ exposure to SD INPRES in the 1970s. One limitation of Susenas is that it only records the type (Islamic vs. secular) of school for the final level of attainment and hence misses potentially informative patterns of switching across Islamic and secular schools throughout one’s educational years. We revisit this issue in Section 5, where we also use the Indonesia Family Life Survey (IFLS) for validation purposes. The IFLS is a rich longitudinal survey spanning 1993 to 2014, and, unlike Susenas, it records the type of schooling for each year of education. However, the IFLS is limited in geographic scope, which often frustrates analyses of policies with district-level variation like SD INPRES. Table 1 reports estimates of Islamic schooling in the IFLS, Susenas, and administrative records. Together, these sources point to a sizable Islamic education sector. School Registries. We use newly compiled administrative data from MORA comprising the universe of madrasa and pesantren (see Appendix C for details). In total, there are 52,398 formal madrasa, 82,871 madrasa diniyah (informal Qur’an study schools), and 25,938 pesantren active in 2019 with establishment dates spanning more than 100 years. Roughly one-third of Islamic school students are enrolled in pesantren and two-thirds in madrasa, according to enrollment records (column 5 of Table 1). Madrasa are further subdivided into three levels of instruction: elementary or MI (25,533 schools), junior secondary or MTs (18,101 schools), and senior secondary or MA (8,764 schools). We rely on an analogous registry of secular schools maintained by the MEC. These data comprise 219,145 schools and include date of establishment, grade level, and private/public status. We address potential concerns about survival bias in these registries using a triennial administrative census of villages (known as Podes) beginning in 1980. Each of these school registries includes details on the location of establishment. Most of our analysis focuses on the district because (i) this is the level at which the SD INPRES policy rule varies, and (ii) analyzing school choice using Susenas is only feasible at this level. However, we also explore Islam–state competition at the village and subdistrict level when examining the supply-side response to SD INPRES. While pesantren may constitute an important part of the Islamic sector response to SD INPRES, they are more difficult to study than madrasa. The Susenas data do not record pesantren attendance. Nor does the MORA registry clarify the level at which a given pesantren organizes its instruction; many, in fact, teach students of all ages under one roof. Moreover, pesantren do not follow the national exams or provide public information on their course offerings. Nevertheless, it is evident that pesantren are sharply differentiated from state schools on curriculum and other dimensions of learning. School Curriculum. We study curriculum using an online registry of schools, called Sistem Informasi Aplikasi Pendidikan (SIAP). This database includes detailed breakdowns of madrasa curriculum with hourby-hour subject timetables each week. While the data cover nearly 20% of madrasa, secular schools do not yet report to SIAP. The timetables provide a unique window into the learning environment at Islamic 9 schools. Our main interest lies in time allocated to (i) Islamic subjects, including Arabic language and literature, (ii) Pancasila/civic education, and (iii) Indonesian language and literature. Downstream Outcomes. We explore political impacts beginning with electoral returns for the state party of the Suharto regime, Golkar, and the Islamic opposition beginning in 1971, the last election prior to SD INPRES. We examine the ideology of legislative candidates in the 2019 election using text from online campaign documents. These include appeals to the faith (e.g., Islam, Muslim, umma, sharia) and references to Pancasila and related Indonesian nation-building concepts.8 We also construct linguistic proxies for religious and national identity. Susenas 2012–2018 reports Arabic literacy. The complete-count 2010 Population Census reports whether the national language, Bahasa Indonesia, is the main language spoken at home. This is distinct from speaking ability: nearly 90% of Indonesians are able to speak the national language, but only 20% use it as the main language inside the home. We view Indonesian use at home as a measure of national affinity, reflecting greater attachment to national as opposed to ethnic or religious identity (see Bazzi et al., 2019, for validation). Finally, we measure Islamic piety and preferences using a nationally-representative survey conducted in 2008 by Pepinsky et al. (2018), who sample 10 individuals from each contemporary district. The survey captures a host of Islamic practices (e.g., fasting, paying zakat) and political preferences (e.g., support for sharia law). It also provides a measure of support for Pancasila. 4 Empirical Strategy This section elaborates our approach to identifying the individual- and school-level responses to SD INPRES. We defer identification checks to the results sections that follow. 4.1 Religious School Attendance and Downstream Outcomes First, we identify effects of the SD INPRES school expansion program on religious schooling using the standard difference-in-differences specification from Duflo (2001): yijt = α + β(IN P RESj × youngit ) + (X0j Ωt )0 Θ + µj + δt + εijt , (1) where i, j, t denote individual, district of birth, and year of birth; IN P RESj measures elementary public schools constructed per 1,000 children from 1973 to 1978; youngit = 1 for individuals aged 2–6 in 1974; µj and δt are district and cohort fixed effects, respectively; and X0j Ωt captures cohort effects interacted with the district’s children population, school enrollment, and exposure to a large governmental water and sanitation program, all in 1971.9 Like Duflo (2001), we compare individuals aged 2–6 (exposed cohorts) 8 The following are examples of nation-building appeals in candidate platforms: “[ensuring the] life of the democratic and just nation according to Pancasila and the 1945 constitution,” and “defending and maintaining Pancasila ideology and the existence of the unity of the Republic of Indonesia”. See Appendix C for further details. 9 Our core sample comprises 275 districts based on boundaries at the time of SD INPRES in the 1970s. In specifications with controls for the water and sanitation program, Duflo (2001) reports 283 districts based on boundaries as of 1995, by which time 8 districts from the 1970s had split in two. 10 with those aged 12–17 (comparison cohorts) in 1974. This specification identifies short-term effects for directly exposed cohorts. In a second specification, we compare cohorts aged 6 or less (exposed) with cohorts aged 12 or more in 1974 (comparison). This captures longer-term effects, inclusive of the market response to SD INPRES. In both specifications, we exclude partially exposed cohorts, aged 7–11 in 1974, as in Duflo (2001). We also trace out the response over time by estimating cohort-specific β. Our interest lies in how SD INPRES shaped Islamic school choice. At the primary level, the expansion of the public sector should have pushed students away from elementary madrasa, the closest substitute in the religious sector. That is, we expect a negative effect of INPRES intensity on elementary Islamic school attendance for exposed cohorts. At the same time, the increase in primary completion rates identified in prior work could have caused greater demand for secondary schooling. With the state focused on expanding primary education, secondary Islamic schools would have been well-positioned to capitalize on this demand shock. For this reason, we expect the policy might have increased secondary Islamic school attendance. Whether the substitution effect at the primary level outweighs the demand effect at the secondary level is an empirical question. The answer tells us about the overall net effect of state school expansion on exposure to Islamic education. We also estimate equation (1) for the broader set of downstream outcomes described in the previous section. These reduced form estimates capture causal effects of SD INPRES on ideology and identity among exposed cohorts. We defer interpretation of the reduced form in this case to Section 7. 4.2 Supply-Side Responses To identify supply-side responses to the expansion of the public school system, we estimate: yijt = α + β(IN P RESj × P ost1972t ) + (X0j Ωt )0 Θ + µj + δt + εijt , (2) where yijt is a variable defined for type of school i, district j, and year of establishment t; P ost1972it is an indicator for panel years after 1972; and X0j Ωt includes year dummies interacted with the same districtlevel baseline covariates as those in equation (1), namely the district’s children population, school enrollment, and exposure to a large governmental water and sanitation program in 1971. We also estimate event-study analogues of equation (2) that replace P ost1972 with semi-decade dummies. We first estimate equation (2) on a balanced district–year panel, using Islamic school entry as the dependent variable. Here, yijt denotes the number of new schools of type i—elementary, junior secondary, and senior secondary madrasa as well as pesantren and madrasa diniyah—created per district–year and per 1,000 children in 1971. This specification identifies the change in the number of Islamic school establishments in districts with greater INPRES intensity relative to other districts after the program began. In a separate analysis, we explore Islamic school entry profiles in response to SD INPRES at the village level. We use a multinomial logit specification where the outcomes capture combinations of madrasa entry at the primary and secondary level. Compared to our main district-level analysis, this specification measures competition within more localized education markets. This village-level analysis is more descriptive in nature as the within-district variation in SD INPRES may reflect endogenous 11 targeting on the part of district governments. We also estimate competitive responses to SD INPRES in terms of curriculum differentiation. In this case, we estimate equation (2) on an unbalanced district-year panel (see Section 6.2). We are interested in the ideological content of the curriculum as reflected in study hours across subjects (e.g., Islam versus Pancasila, Indonesian versus Arabic). Under the assumption that curriculum remains stable within a given school over time, this specification identifies changes in the ideological leaning of schools established in districts with greater INPRES intensity after the program began. 5 Effects on Religious Schooling This section presents our first set of results pertaining to religious school choice. 5.1 Religious Schooling by Level Table 2 reports the effects of SD INPRES on binary indicators of madrasa attendance. The outcomes in panel (a) equal one if the respondent’s highest level of education is elementary Islamic (columns 1– 2), junior secondary Islamic (columns 3–4), or senior secondary Islamic (columns 5–6). One concern with such measures is that the share of respondents completing Islamic education could be increasing simply because overall education levels are increasing after SD INPRES. Thus, in panel (b), we look at a different measure equal to one if the respondent completed Islamic elementary, junior secondary, or senior secondary, conditional on completing exactly the relevant years of education for each level (6 years for elementary, 9 for junior secondary, and 12 for senior secondary). These measures capture the share of Islamic education at each instruction level. While such conditioning may be endogenous, these alternative measures alleviate concerns that our results in panel (a) are driven by the general increase in education levels. Across panels, and in all results that follow, we cluster standard errors at the historic 1970s district level of SD INPRES policy variation. At the elementary level, the policy pulled students away from madrasa and pushed them towards the state system. Among cohorts aged 2–6 in 1974, INPRES intensity reduces the likelihood of Islamic primary attendance by approximately 10% (column 1). This substitution effect becomes smaller in magnitude for the long-run cohort comparison (column 2). Similar patterns arise for the conditional measure in panel (b). The weaker substitution effect in column 2 may be due to an increase in Islamic school construction over the medium to long run, a mechanism explored in Section 6.10 At the secondary level, Islamic schools absorbed some of the increased demand for post-primary education. This effect is apparent both in the short term (columns 3 and 5) and the long term (columns 4 and 6). However, the longer-term effect is more than twice as large in magnitude, which again points to a potential supply-side response by the Islamic sector. The auxiliary IFLS data provides a striking 10 Examples of this can be found in the Indonesian-language literature, e.g., Darmaningtyas (2004) notes: “. . . the tension between government and the clerics that had built schools in the form of pesantren persisted during the entire New Order Era. As a result, many SD INPRES [primary schools] in Madura [a region of East Java] have few students, because communities prefer schools built by religious leaders.” 12 summary statistic highlighting the importance of the demand channel: 78% of those that attended Islamic secondary schools did so after completing secular primary schools. The estimates in columns 3–6 suggest that SD INPRES may have catalyzed this type of schooling trajectory and ultimately increased exposure to Islamic education. These results are borne out with less parametric structure in Figure 1. The graphs show the fraction of Susenas respondents in each cohort reporting elementary (panel a), junior secondary (panel b), or senior secondary (panel c) Islamic school as their highest level of education, separately for high-INPRES and low-INPRES districts. Appendix Figure A.1 reports the corresponding graphs with Islamic education defined conditional on completing the relevant years of schooling. These figures show the same key patterns as Table 2. High-INPRES districts experience a short-run substitution away from elementary Islamic schools and a long-run increase in the completion of secondary Islamic schooling. Both patterns begin to materialize for those born after 1968, the first cohort fully exposed to SD INPRES. 5.2 More (Islamic) Schooling SD INPRES increased not only total years of education but also net exposure to Islamic education. Column 1 of Table 3 (panel a) shows that each primary school constructed per 1,000 children increased years of schooling by around 0.14 years. The corresponding male-specific estimate of 0.17 years in Appendix Table A.3 lies between the range of estimated effects for men in Duflo (2001)—0.12 to 0.19—based on the intercensal survey (Supas) from 1995. The effect size roughly doubles when expanding the sample to include cohorts younger and older than the narrow-exposure window in the baseline (column 2). Ultimately, the increase in secondary Islamic schooling exceeds the substitution effect towards secular schools at the primary level. In the short-run, each additional INPRES school increased the likelihood of Islamic schooling by 5% (column 3), and this grows by a factor of 4 over the long run (column 4). The same holds for the likelihood of any Islamic schooling conditional on completing the relevant years of education as in panel (b) of Table 2. While close to zero in the short run (column 5), the effect of SD INPRES is positive and significant in the long run (column 6). Together, the estimates in columns 3–6 of Table 3 (panel a) are consistent with those in panel (d) of Figure 1: high-INPRES districts experience a diverging trend in the share of students completing any Islamic education. Panel (b) of Table 3 sheds further light on the counterfactual schooling outcomes, showing that SD INPRES increased Islamic education by inducing certain types of families to pursue additional schooling for their children. Some families would have sent their children to Islamic school in the absence of the policy. Others may have enrolled their children precisely because of the changes in schooling options brought by INPRES. We explore this latter, complier population by instrumenting for years of schooling using the difference-in-difference term in equation (1).11 This identifies the local average treatment effect of INPRES on Islamic schooling among compliers, namely children who received additional schooling as a result of the policy. The positive IV estimates in columns 3–6 of panel (b) show that children induced by SD INPRES to attain greater schooling were also more likely to receive an Islamic education. This effect is driven by the increase in secondary Islamic education, since INPRES exposure increases junior 11 This is the same instrumentation strategy that Duflo (2001) uses to study the Mincerian returns to years of schooling. 13 and senior secondary Islamic education but decreases elementary Islamic education in the reduced form (see Table 2). In other words, INPRES increased exposure to Islamic education by enabling children who otherwise would not have reached junior secondary school to reach this higher level of education. The Gender Dimension. Overall, SD INPRES increased exposure to Islamic education among both men and women. The effects may even be slightly larger for women, at least over the long run (see Appendix Table A.2).12 Importantly, the significant effects on Islamic schooling for both genders increase the scope for intergenerational transmission within the family—a hypothesis we explore in Section 7.4. Religious school choice among women is of particular interest. In 1982, the Suharto regime imposed a ban on the Islamic veil (hijab) in public schools. A government decree standardized the use of school uniforms in the country, which in effect amounted to a crackdown on veiling (Jo, 2020; Shofia, 2020). Women wishing to wear a headscarf would have faced a choice between transferring to an Islamic school or dropping out of school. The substitution effect in columns 1–2 of Table 2 could have been offset by this headscarf ban, specifically for female students. We explore this hypothesis in Appendix Table A.4, where we interact equation (1) with exposure to the headscarf ban.13 Specifically, INPRES-exposed women who would have been too young to complete their primary education before the ban may have transferred to an Islamic school after 1982. The top row in Appendix Table A.4 shows women exposed to the ban were indeed more likely to complete an Islamic elementary education relative to other cohorts exposed to SD INPRES. This holds whether we look at the unconditional likelihood of completing an Islamic primary education (columns 1–2) or the conditional measure (columns 3–4). This result supports the view that Islamic schools contribute to address heterogeneous preferences—potentially cutting across genders—for different types of schooling. Attending vs. Completing Islamic Education. One potential concern with these results is the low share of individuals with Islamic schooling reported in the Susenas data. Indeed, Table 1 shows that exposure to Islamic schooling is considerably higher in other sources. In the IFLS, Islamic education rates range from 11% in primary to 23% in junior secondary (20% across all levels, and 25% among enrolled cohorts). Administrative enrollment records for 2019 similarly show attendance rates ranging from 13% in primary to 23% in junior secondary (21% overall). There are two reasons why the Susenas data may lead us to underestimate the effects of SD INPRES on Islamic school exposure. First, Susenas indicates whether the final year of education took place in a madrasa. If some of those attending secular secondary schools attended elementary madrasa, the Susenas measures would be understated.14 Second, Susenas does not allow respondents to indicate pesantren 12 These estimates are consistent with different parental preferences over the religious content of schooling for boys and girls. This resonates with Meyersson (2014) who finds that Islamist mayors increased educational achievement among women in Turkey. In our context, the smaller short-run effect on years of schooling for women (see Appendix Table A.3) suggests parents may have been initially more reluctant to send their daughters to the newly created public schools. Madrasa would then have provided a more acceptable alternative for girls’ education in these families. 13 The regression reported in Appendix Table A.4 interacts INPRES intensity and the indicator for treated cohorts (aged 2-6 in 1974) with a gender dummy and a dummy for individuals aged less than 12 in 1982. All the relevant two-way and three-way interactions are included in the regression but their output suppressed. 14 The IFLS suggests that these switcher populations could be sizable: among those continuing after Islamic primary school, 31% attend secular junior secondary schools, and among those continuing after Islamic junior secondary school, 48% attend 14 attendance. The large supply response among pesantren that we identify below suggests that this population could be important. As a validation exercise in Appendix Table A.5, we estimate the effects of SD INPRES on years of Islamic education among Muslim respondents in the IFLS. Unlike Susenas, the IFLS reports the type of education completed at every instruction level and also reports pesantren attendance. Overall, SD INPRES decreased the likelihood to have completed an Islamic elementary education (columns 1–2) as well as total years of Islamic elementary education (columns 7–8). On the other hand, the policy increased attendance rates in Islamic secondary schools at the junior and the senior level (columns 3–6) and increased years of secondary Islamic education overall (columns 9–10). Reassuringly, these patterns mirror those obtained using the Susenas data. 5.3 Identification Checks Our core results on school choice are robust to key concerns about causal inference. First, SD INPRES was not systematically allocated towards districts with different preexisting trends in Islamic schooling. Figure 2 demonstrates the absence of pre-trends in primary and secondary Islamic school attainment, respectively. These graphs estimate cohort-specific β in equation (1), coloring the exposed and control cohorts in gray and light gray, respectively, with the partially exposed cohorts in white. Figures 2 (a) and (c) show the short-run time-path, and (b) and (d) correspond to the long-run specification. Figure 3 presents analogous patterns for total years of schooling and any Islamic education. Meanwhile, Figure 4, discussed below, presents complementary evidence against pre-trends in Islamic school construction. This is not to imply that the regime was entirely blind to regional variation in the size of the Islamic education sector. In fact, Appendix Table A.1 shows that the government allocated proportionally more INPRES schools to districts with a greater prevalence of Islamic schools as of 1972.15 What the government did not do at the time was strategically target districts where the Islamic education sector was rapidly expanding. The lack of pre-trends in Figures 2–4 is consistent with this interpretation. 5.4 Why the Supply-Side Matters In sum, SD INPRES caused an increase in Islamic education that was fueled by secondary Islamic schools’ absorption of new primary school graduates. Duflo (2004) conjectures that “the program affected mostly primary school completion, whereas omitted factors would have affected other levels of schooling.” We show in the following section that a crucial omitted factor lies in the supply response of the Islamic education sector. Here, we present motivating evidence. Appendix Table A.8 regresses outcomes from Tables 2 and 3 on school construction from 1973 to 1978. Odd-numbered columns replicate the baseline specification. Even-numbered columns include analogous interactions of the young cohort indicator with state secondary, Islamic primary, and Islamic secular senior secondary schools. Appendix Figure A.4 further illustrates this phenomenon using repeated cohorts from Susenas, e.g., 12% of respondents born in 1998 attended a madrasa in 2012, but only 7% of the same cohort reported having completed Islamic schooling in 2018. 15 This level difference is evident at the individual level in Figure 1 for cohorts born before 1968. 15 secondary schools constructed over the same period. These endogenous supply measures exhibit informative correlations with schooling outcomes. The likelihood of completing any secondary schooling strongly correlates with the entry of state and Islamic secondary schools, but not with new primary schools (column 4). Any effect of SD INPRES on post-primary completion rates may have come from additional, correlated responses by the state and the Islamic sector to the initial policy shock. Moreover, Islamic secondary school completion is shaped not only by INPRES primary school entry but also by entry of Islamic primary and secondary schools (column 6). It is these entry decisions that underlie the LATE in Table 3: greater years of schooling came with greater exposure to Islamic education. 6 Supply-Side Responses To better understand the effects of SD INPRES on school choice, we now explore how the program shaped supply-side dynamics in education markets.16 The Islamic sector responded to the primary school construction boom in two ways. First, Islamic society, equipped with charitable endowments (waqf ), expanded its educational presence in districts with greater SD INPRES intensity—starting with junior secondary schools, and ultimately at all levels of schooling. Second, Islamic schools entering highINPRES districts after the policy provide a greater volume of Islamic content, and a smaller volume of civic education and instruction in the national language. Together, these results show how a landmark mass schooling effort transformed education markets over the ensuing decades. 6.1 More Islamic Schools This section examines the Islamic sector’s response along the extensive margin: construction of new schools. The outcomes of interest include the number of new elementary, junior secondary, and senior secondary madrasa establishments. We also look at new madrasa diniyah and pesantren, which are Islamic afternoon schools and boarding schools, respectively. The numbers of schools of each type created per district-year are divided by the 1971 children population, analogous to the SD INPRES intensity measure. In Table 4, panel (a) shows greater entry of Islamic schools in high-INPRES districts: elementary (column 1), junior secondary (column 2), and senior secondary madrasa (column 3), as well as diniyah (column 4) and pesantren (column 5). One additional INPRES school per 1,000 children is associated with 4 more Islamic elementary and junior secondary schools per year. Tracing out these effects over time, we uncover a dynamic Islamic sector response to the state’s primary school expansion. Figure 4 estimates the effects of SD INPRES by semi-decade using an eventstudy approach. High-INPRES districts experience a steadily diverging trend in entry of new secondary madrasa and pesantren (Figure 4, panels b–d). A similar pattern holds for elementary madrasa (panel a). Figure 4(f) shows that Islamic school entry outpaced additional state school entry from the 1980s onward in these high-INPRES districts. This was fueled by private madrasa (see Appendix Figure A.2). 16 In what follows, while often referring to the “supply response”, we acknowledge that the long-run expansion of Islamic education in high-INPRES regions may well be due to changes in demand among the originally-exposed cohorts as they raise children of their own, as seen in Section 7.4. 16 Several robustness checks point to a causal interpretation of the Islamic sector response. First, note the lack of pre-trends in Islamic school construction in Figure 4, mirroring the patterns in Islamic school completion rates in Figure 2. Second, Appendix Figure A.3 suggests that the private Islamic sector’s response is distinct from other private sector responses.17 Some private secular schools enter in response to SD INPRES, but such entry is most concentrated at the primary level and follows a different (and more muted) post-1970s trajectory than the Islamic sector. Third, the patterns are unlikely to be an artifact of survivor bias in the 2019 registry of Islamic schools. Appendix Table A.10 shows that the increase in Islamic school entry after the 1970s can be seen in historical administrative data (from Podes 1980, 1983, 1990, 1993) that is not subject to the attrition biases inherent to contemporary administrative registries.18 Local Competition. Table 5 explores the strategic Islamic sector response within local education markets. Islamic organizations may respond locally to SD INPRES entry in their village or in neighboring ones within a subdistrict. Using a multinomial logit formulation, we consider four distinct competition profiles: no entry, elementary madrasa entry, junior secondary madrasa entry, and both elementary and junior secondary madrasa entry.19 We report marginal effects with no entry being the reference category. Table 5 suggests distinct short- and medium-run supply responses by the Islamic sector. In the short run, new junior secondary madrasa capitalized on demand for continued education among SD INPRES graduates. Villages with SD INPRES entry between 1973 and 1978 are 50% more likely to have built only an Islamic junior secondary school by 1983 (column 2, panel a). SD INPRES entry in nearby villages within the same subdistrict is associated with greater junior secondary madrasa entry in one’s own village: moving from the 25th to the 75th percentile of subdistrict saturation (excluding one’s own village) shifts such entry by nearly 30%. This is consistent with secondary schools serving students from more than one village. At the same time, there is no differential elementary madrasa entry (columns 1 and 3, panel a). In other words, the Islamic sector focused its short-run efforts on absorbing some of the growing demand for post-primary education. By contrast, in the remaining years of the Suharto era, the Islamic sector not only built more junior secondary schools but also began to compete locally at the primary level. This medium-run response can be seen in panel (b) of Table 5, which looks at Islamic school entry from 1984 to 1998. SD INPRES construction in the 1970s is associated with an increase in the likelihood of elementary madrasa construction in the 1980s and 1990s within the same village (column 2). More junior secondary madrasa enter in villages with SD INPRES schools (column 3), and also do so in tandem with elementary madrasa (column 4). These findings corroborate the event-study path in Figure 4. While suggestive of a causal competitive response, the results in Table 5 should be interpreted more descriptively than our district-level results. The plausibly exogenous policy variation lies at the district level where school construction funds were allocated based on the school-age population and enrollment rate. Within district, these funds may be allocated endogenously across villages and subdistricts. One concern might be that district officials with particular secularizing ambition targeted SD INPRES schools 17 There are 41,969 private non-Islamic schools under MEC authority in 2019 (see Appendix C). The first round of Podes was in 1976, but this data does not distinguish Islamic schools. 19 Similar insights obtain when allowing for all 8 possible combinations across the three grade levels, including senior secondary. 18 17 towards villages with Islamic schools (see Appendix Table A.1 for evidence of such targeting at the national-to-district level). We address this concern in Appendix Table A.9, where the same patterns of local competition hold when conditioning on Islamic schools being present in the village before 1973. Interpretation. Overall, the Islamic sector responded to SD INPRES by building more schools. Ap- pendix B offers a theoretical foundation for this result. Under basic assumptions about the shape of demand for schooling, religious and secular education can act as strategic complements (Bulow et al., 1985). We provide a simple model that generates this prediction. In this setup, the state and the Islamic sector compete in a Stackelberg game where the former is leader and the latter is follower. An outward shift in demand for schooling causes the state to increase its supply of schools. This increases the supply of religious schools since the Islamic sector’s best response is upward sloping. In practice, these strategic complementarities may arise from a variety of mechanisms seen in our setting, including, among others, transitions in and out of the Islamic schooling system, co-location of primary and secondary madrasa, and comparative advantage of madrasa in providing female education in conservative communities. Financing New Islamic Schools. These supply results beg the question of how the Islamic sector financed such an educational expansion? For decades, private Islamic organizations had financed schools through the use of inalienable waqf land endowments. In 1963, 88% of all private elementary madrasa were run by Islamic organizations (Lee, 1995). One of the largest, Muhammadiyah, controlled over 3,000 hectares of waqf property supporting more than 4,300 madrasa by 2004 (Jahar, 2005). We show here that waqf endowments helped fuel the Islamic sector response to SD INPRES. Panel (b) of Table 4 interacts the relevant terms in equation (2) with a proxy for waqf land at the district level in 1972: land endowed in waqf to support mosques. While a small subset of all waqf land, this measure is the best available proxy in the time period and is likely proportional to total waqf land in a given locality, which includes waqf land that directly supported religious schools.20 Since waqf land correlates with the Muslim share in the local population, we also interact the relevant terms in equation (2) with the share of Muslim individuals among cohorts born by 1972 (observed in the 1976 census). Districts with more waqf endowments experience a larger Islamic school supply response to SD INPRES (columns 1–3). One additional km2 of waqf land is associated with 0.5 more elementary madrasa over the ensuing 25 years. Reassuringly, districts with a larger Muslim share see a larger Islamic school supply response to SD INPRES. This is distinct from the heterogeneous effect of initial waqf endowments and perhaps consistent with a backlash against the secularization effort embodied in the policy. Overall, the capital held in waqf enabled a strong, positive infrastructure response by the Islamic sector. Why, though, did Islamic organizations not expand their schooling network prior to INPRES if indeed they had the resources to do so? Part of the answer lies in the politics of competition between Islam and the state. Absent efforts to secularize local education, Islamic leaders might have had weaker incentives to push into new markets or contest existing ones. This is the essence of the simple model described in Appendix B. It also resonates with the policy context in which the state was not only expanding access 20 See Bazzi et al. (2020) for a detailed discussion of this measure, which comes from administrative data collected by MORA. 18 to secular schools but also pushing to secularize Islamic schools (see Section 2.2). As we show next, the Islamic sector responded to this push not only along an extensive margin but also an ideological one. 6.2 Ideological Differentiation Table 6 shows that Islamic schools created in high-INPRES districts after 1972 provide greater religious content and study of Arabic at the expense of civic education and the national language. Here, we estimate an unbalanced district-level panel where each observation is a mean outcome across all schools entering a given grade level in a given year. Pooling across levels, we find that SD INPRES is associated with an increase in the share of weekly instruction time devoted to Islamic subjects (panel a, column 1). At both the primary and junior secondary levels, each additional INPRES school is associated with a 5% increase in Islamic content among newly created Islamic schools (panel a, columns 2 and 3). At the junior secondary level, part of this increase in Islamic content is achieved through a reduction in classroom time devoted to Pancasila and civic education (panel b, column 3). Panels (c) and (d) show similar patterns of substitution for the share of instruction hours dedicated to Arabic and Bahasa Indonesia, respectively. Each additional INPRES school is associated with a 6% increase in Arabic instruction at the primary level (panel c, column 2) and a 5% decrease in Bahasa instruction at the junior secondary level (panel d, column 3).21 In Appendix Table A.11, we show that the increase in Islamic content and Arabic instruction, as well as the corresponding decrease in civic education and Bahasa instruction, hold when measuring total instruction hours. This is important insofar as Islamic schools might have increased total classroom time to accommodate other material besides religious subjects. Together, Tables 6 and A.11 suggest that instruction hours dedicated to Islamic content and Arabic crowd out civic education and study of the national language—two important inputs to the homogenizing function of mass public schooling. Note that our difference-in-difference-based interpretation hinges on the stability of school curricula. That is, we assume that the curriculum observed in 2019 is highly correlated with that observed in a school’s initial year of operation. It is not possible to validate this assumption, but there are reasons to think that a school’s curriculum is closely attached to its ideology, which likely has persistent features tied to the identity of founders. Moreover, given the legacy of conservative schools’ opposition to state oversight, we suspect that the madrasa included in the SIAP registry are those with less Islamic content and hence more likely to be compliant with government-recommended curriculum. This could work against our findings, presuming that such selective reporting is differential in high-INPRES districts.22 Quality. Do these patterns of curriculum differentiation affect the quality of learning in Islamic schools created in the wake of SD INPRES? The higher volume of instruction dedicated to Islamic content and 21 Despite these shifts at the primary and junior secondary level, we find different patterns at the senior secondary level where SD INPRES is associated with a reduction in Islamic content and an increase in Pancasila and Arabic instruction (panels a–c, column 4). This goes against some of the findings elsewhere but may be an artifact of the small number of senior secondary schools in SIAP. It also hints at a possible secularization of senior secondary Islamic schools aimed at capturing junior secondary graduates intent on going on to university where proficiency in traditional non-Islamic subjects is essential. 22 We find some evidence against differential reporting. For example, madrasa created after 1972 in high-INPRES districts are no more or less likely to report to SIAP, using the baseline supply-side regression specification in equation (2). Nor do we observe differential reporting to SIAP by more conservative madrasa—as inferred from Islamist-oriented school names. 19 Arabic learning comes at the expense of studying standard subjects required to pass national exams. Appendix Table A.12 shows that students in Islamic schools devoting more classroom time to religious subjects exhibit weaker performance on standardized math and science tests. Our identification strategy does not allow us to disentangle this particular mechanism (or differentiation along a quality margin) from selection on ability. However, in Appendix Table A.13, we show that Islamic junior secondary schools created after 1972 in high-INPRES districts exhibit lower contemporary test scores than those created prior to the program (though this is imprecise, see column 2). There is also a larger test score gap between Islamic and non-Islamic schools in high-INPRES districts among schools created after 1972 (column 4). These results are consistent with both different sorting on ability as well as a change in instructional quality across Islamic and non-Islamic schools after SD INPRES. 7 Mass Schooling and Nation Building Like most mass schooling efforts, Indonesia’s entailed significant political and ideological objectives. This section shows that such ambition may have come up short, frustrated in part by the dynamic supply-side response of the Islamic education sector and by families’ efforts to transmit religious values to their children in an increasingly secular political environment. We show here that SD INPRES failed to increase support for the Suharto regime and set in motion a significant shift in religious identity and preferences that ultimately worked against the state’s secular nation-building agenda. In what follows, we maintain our focus on the reduced form. This allows for compelling causal inference but requires careful interpretation. In particular, we do not disentangle the direct effect of SD INPRES exposure from that of the increased Islamic school presence. For example, the policy could have affected piety through increased exposure to religion in state schools (with their mandatory 2 hours). While this is possible, the nexus of results below suggests that Islamic school choice likely played a significant role in shaping the legacy of SD INPRES. Without such a strong role, it is difficult to explain why INPRES exposure increased religiosity without simultaneously increasing attachment to the national identity, or why exposed cohorts were more likely to send their own children to Islamic schools. 7.1 Support for the New Order Regime In the short run, a major development initiative like SD INPRES could have bolstered electoral support for Suharto and the New Order. We explore legislative election results during this period (in 1971, 1977, 1982, 1987, and 1992) and after Indonesia’s democratic transition (in 1999, 2004, and 2009).23 Only three parties were allowed to compete under the New Order after 1971: Suharto’s Golkar party, the Muslim umbrella United Development Party (PPP),24 and the secular nationalist Indonesian Democratic Party 23 24 The final election of the Suharto era was in 1997, but we could not obtain district-level records from this round. The Suharto regime forced all Islamic political parties to combine under the PPP in 1973 while also mandating that “Islam” not be allowed in the party name. In the 1971 election, we capture the Islamic vote share by combining all four Islamic parties that were later subsumed under PPP: Nahdatul Ulama (NU), the Muslim Party of Indonesia (Parmusi), the Islamic Association Party of Indonesia (PSII) and the Islamic Education Movement (Perti). NU was the second-highest ranked party in that election (after Golkar) with 18% of the vote. 20 (PDI). Golkar obtained 70% of the vote on average across all New Order elections, while the PPP was the main opposition with 21% of the vote. After 1999, both Golkar and the PPP garner much smaller vote shares due to the proliferation of parties on both the secular and religious sides of the political spectrum. Surprisingly, SD INPRES did not increase electoral support for the regime in high-INPRES districts. The 1971 round was the only New Order election conducted before school construction ensued. Elections held in 1977 and 1982 would have been indirectly affected by the policy (e.g., through the increased presence and influence of teachers), while the exposed cohorts aged less than 6 in 1974 would have first voted in 1987. In panel (a) of Figure 5, Suharto’s Golkar party experiences a marked decline in electoral support from 1971 to 1977 in high-INPRES districts: each additional INPRES school per 1,000 children is associated with a 2–4 percentage point (p.p.) decline in the Golkar vote share (relative to the mean of 65% in 1971). This effect appears as early as 1977 and persists until 1992. The Islamic opposition captured some of the declining support for Golkar. We see this for the PPP vote share in absolute terms (panel b) and relative to Golkar (panel c). One explanation could be that the PPP captured general opposition sentiment. Indeed, the effect of INPRES intensity on support for the PPP becomes noisier after 1999, when the PPP was no longer the main vehicle for opposition aspirations. However, another explanation is that the Islamic sector pushed back against secularization, which was most salient in districts with greater INPRES school construction. The decline in Golkar support as early as 1977 is consistent with this pushback. If instead these electoral shifts had been slower to materialize, it would have been difficult to rule out an alternative explanation, namely that INPRES created a more educated citizenry that was simply more opposed to the regime’s authoritarian ambition. The more plausible explanation for Figure 5 is that the Islamic sector mobilized not only by building more religious schools but also by coordinating political opposition through its own expanding school network. Overall, Figure 5 shows that SD INPRES did not boost support for the Suharto regime during the New Order era. Even under an autocratic regime with tightly controlled elections, mass schooling failed to indoctrinate voters and instead fostered support for the main opposition party, the Islam-based PPP. Table 8, discussed below, provides long-run, cohort-based evidence on legislative candidate entry that aligns with these early electoral divisions. After 1999, the nature of the Indonesian state changes dramatically. Many more political parties compete and capture opposition sentiment. Moreover, ideological attachment to the Indonesian nation no longer conflates with support for the New Order regime. Understanding the long-term impacts of SD INPRES on identity and nation building thus requires looking across a range of indicators. 7.2 Effects on National and Religious Identity As the Islamic schooling system expanded, it counteracted the regime’s intent to build a secular Indonesian identity. Table 7 provides initial evidence of a shift in religious and political ideology. Panel (a) explores dimensions of secular identity across Muslim and non-Muslim citizens, while panel (b) examines religious piety and practice among Muslims. Of course, a high level of Islamic piety is not incompatible with a high level of attachment to the secular state. Yet, as we show, INPRES exposure is associated with greater piety but not with greater support for secular principles, which, together, point to a role for the 21 Islamic sector response in explaining the long-run effects on nation building. We first examine a standard marker of an individual’s attachment to the national identity in multilingual countries: the use of the national language at home. With the complete-count 2010 Population Census, we observe nearly 32 million individuals in the original cohorts aged 2–6 and 12–17 in 1974. We find null effects of SD INPRES across the full population (column 1). However, this null masks a religious divide: 16% of Muslims use Indonesian as the main language at home compared to 28% of non-Muslims.25 Among Muslims, INPRES-exposed cohorts report slightly less home use of the national language (column 2), while affected non-Muslim cohorts exhibit little response (column 3). These weak effects are striking given that INPRES schools aimed to promote a single Indonesian identity built around a common language. To be sure, SD INPRES did increase Indonesian proficiency, including among Muslims (see Appendix Table A.14, columns 1–3). However, it did not increase attachment to Bahasa Indonesia inside the home. For those exposed to Islamic education as a result of SD INPRES, immersion in the national language may have been crowded out by the study of Arabic, the language of the Qur’an. Table 6 showed that schools created in high-INPRES districts after 1972 devote more classroom time to Arabic and less to Indonesian language and literature. Table 7 shows that SD INPRES increased Arabic knowledge among exposed cohorts (column 4). Columns 5 and 6 show that the positive effects are driven by those with any Islamic education (two-thirds of whom report Arabic literacy, compared to one-third with secular education).26 Importantly, like the distinction between national language ability versus home use, here too we can clarify the identity content of Arabic literacy: Appendix Table A.14 (columns 4–9) shows that SD INPRES increased literacy in the Latin alphabet (on which Indonesian is based) but did not increase literacy in other languages besides Arabic. Moreover, conditional on years of schooling (fixed effects), Arabic literacy is 20–30 p.p. higher for those with Islamic education (see Appendix Table A.15). These shifts in national and religious identity are accompanied by broader changes in piety. In panel (b) of Table 7, we look at a range of Islamic practices recorded by Pepinsky et al. (2018): praying 5 times a day (column 1), fasting during Ramadan (column 2), reading the Qur’an (column 3), attending Friday prayer (column 4), performing Sunna prayers (column 5), joining prayer groups known as pengajian (column 6), and paying zakat (column 7). While several of these are standard indicators of piety, respondents vary widely in their practice. For example, 83% report paying zakat and 81% fast during Ramadan, while only 23% always attend Friday prayer and 18% perform non-obligatory Sunna prayer. Column 8 pools all practices into a single index. Overall, we find positive effects of INPRES exposure on most measures.27 The effects are somewhat larger for religious practices with a social dimension (e.g., attending Friday prayer), perhaps reflecting the fact that individuals socialized in madrasa more regularly practice their faith in a group setting within their community. 25 Importantly, using this same Population Census data, we find a precise zero effect of SD INPRES on the likelihood of being Muslim: -0.0003(0.0011) relative to a mean of 0.878. 26 We switch between sample splitting on religion and on religious schooling across outcomes in panel (a) because Susenas does not record religion, and the 2010 Population Census does not report Islamic schooling but only the level completed. 27 These effects differ somewhat from those in Table 8 of Rohner and Saia (2019). Using the IFLS and a more expansive definition of INPRES exposure that includes the partially-exposed cohorts, they show that INPRES increased religiosity (on a selfreported 5-point scale) but had null effects on prayer. Our findings across a richer set of outcomes measuring piety and practice are consistent with their finding on religiosity but not with respect to prayer. 22 Together, the results in Table 7 suggest that SD INPRES may have increased Islamic identity at the expense of a secular national identity. For those attending Islamic schools, this could have occurred through learning Arabic as well as Islamic thought and practice, particularly in the impressionable years of primary and secondary schooling. Meanwhile, for those attending state schools, this could have occurred through greater exposure to Islamic-educated peers in one’s community or engagement with the Islamic sector outside formal schooling (e.g., through parental inputs or attendance of madrasa diniyah or mosque-based youth groups). We explore these mechanisms in Section 7.4, but fully disentangling them is beyond the scope of our identification strategy and data. Nevertheless, their combined effects help explain how the Islamic sector response could have been large enough to bolster religious identity over the long run and, in turn, to shape ideology and political preferences as we show next. 7.3 Effects on Political Attitudes and Ideology Table 8 explores additional downstream effects of SD INPRES on political ideology among citizens (panel a) and politicians (panel b). First, we consider a direct measure of citizen support for Pancasila, the secular national ideology advanced through states schools. The Pepinsky et al. (2018) survey asks respondents whether Pancasila is the best ideology for the nation or whether it should be replaced with another more suitable ideology. Column 1 in panel (a) shows that SD INPRES had a fairly precise null effect on support for Pancasila, which stands at 84% across the population. However, this masks a large divide between Muslims (83%) and non-Muslims (93%). This gulf widens for those exposed to SD INPRES: affected non-Muslim cohorts exhibit greater support for Pancasila than non-affected cohorts (column 2) whereas affected Muslim cohorts exhibit a small but precise null response (column 3). These results are suggestive of differential effects across groups with varying exposure to the Islamic education sector. Reassuringly, Appendix Table A.16 shows that individuals in this survey exhibit similar Islamic schooling outcomes as those in the baseline Susenas sample from Tables 2 and 3. While INPRES exposure did not increase support for Pancasila among the majority Muslim population, it also did not spur support for conservative Islamist ideology as an alternative foundation of the state. We demonstrate this using two measures of support for Islamic law, again drawing on the Pepinsky et al. (2018) survey. The first, subjective measure in column 4 is an indicator for individuals reporting strong or very strong support for sharia as the foundation of the state. The second, objective measure in column 5 takes the mean across indicators of support for different dimensions of sharia: corporal punishment for crime, prohibition of interest, mandatory hijab, supporting polygamy, punish adultery with stoning, and punish apostasy with death. Across both outcomes, we find null effects of SD INPRES on affected cohorts of Muslim citizens. Appendix Tables A.17 and A.18 provide further evidence using the six sub-components of the sharia index as well as other measures of support for Islamist ideology. The bottom panel (b) of Table 8 provides analogous evidence on long-run ideology among politicians. We estimate the effects of INPRES exposure on candidate entry in the 2019 legislative elections, restricting to the original cohorts (2–6 versus 12–17 in 1974, respectively).28 INPRES-exposed cohorts are 28 Legislative candidates are required to have at least a primary education. Hence, the results here apply to a population for whom the secondary school response is more important than any first-order effects on primary education. 23 significantly less likely to run on a Golkar ticket and more likely to run on a PPP ticket (columns 1 and 2).29 In other words, the short-run effects on support for Golkar and the PPP seen in Figure 5 persisted over the long run among affected cohorts of political candidates. This is despite both parties being considerably less popular than in the New Order era when the PPP was the main opposition. Furthermore, INPRES-exposed candidates, across all parties, are less likely to campaign on Pancasila or related nationbuilding themes (column 3). However, they are no more likely to campaign on Islamic themes (column 4), and indeed religious appeals are not confounded with nation-building ones (column 5). 7.4 Intergenerational Transmission of Religious Values In this final section, we highlight the role of intergenerational cultural transmission in shaping the legacy of SD INPRES for religious schooling and values. The cohorts originally exposed to SD INPRES were more likely to have attended an Islamic school and to subsequently report greater engagement with Islam. Two generations after INPRES schools were built, attendance in Islamic schools remained very high: in 2019, 21% of pupils were enrolled in a madrasa or a pesantren (Table 1). This suggests that the shifts in religious identity and preferences for Islamic schooling set in motion by SD INPRES were likely passed on to future generations. Two types of religious cultural transmission could have taken place among the cohorts exposed to mass public schooling. On the one hand, parents wishing to maintain a strong religious identity inside their household (potentially as a result of attending an Islamic school themselves) could have sent their own children to an Islamic school. This would have influenced children’s religiosity by affecting the type of peers with whom a child would interact—a process that the literature since Bisin and Verdier (2000, 2001) has described as horizontal or “oblique” cultural transmission. On the other hand, parents could also have invested in greater religious socialization at home for fear that children would fail to maintain strong religious values in a fast-secularizing society. Such vertical transmission could either complement or substitute for religious school choice. Horizontal Transmission. We explore the “oblique” transmission hypothesis in panel (a) of Table 9, which reports intergenerational effects on Islamic school choice using the complete enumeration of household members’ schooling in Susenas. This analysis focuses on the original cohorts in Tables 2 and 3 but is restricted to those with co-resident children older than 18 (i.e., those who have already completed schooling).30 To allow for maximal sample coverage, we estimate the effects of each parent’s exposure separately rather than restricting to the particular subsample of kids with both parents in the original cohorts. Both parent’s exposures to SD INPRES in the 1970s are associated with an increase in children’s likelihood of completing secondary Islamic education (columns 3–6). However, these children are also less likely to complete primary Islamic education (columns 1–2), much like their parents in column 1 of Table 2. Parents may be choosing to replicate their own educational trajectories for their children a few decades later. Appendix Table A.19 provides direct evidence of intergenerational persistence in Islamic 29 30 No other party affiliations admit significant effects. This sample is comprised largely of ethnic groups with strong (post-marital) intergenerational co-residence norms. Whether our findings generalize to adult children living separately from their parents cannot be answered with available data. 24 schooling: the likelihood that a child completes Islamic schooling is 20 p.p. higher when either parent has an Islamic education background (columns 4 and 8). These results hint at a distinction between preferences and constraints. The parents of exposed cohorts who sent their children to public primary schools in the 1970s were often constrained to choose Islamic secondary schools for continued education. As those children became parents themselves, they did not face the same constrained school choice set as both Islamic primary and public secondary schools had expanded over time (see Section 6). Yet, these parents still chose Islamic schools for their children. In other words, the constraints faced by parents in the 1970s may have led to a shift in the demand for Islamic schooling of their grandchildren several decades later. Vertical Transmission. Parents directly exposed to the secularization of education could also have invested in greater religious socialization within their household. In panel (b) of Table 9, we study two mechanisms for vertical religious transmission. First, we show that SD INPRES increased assortative mating (homogamy) among religiously educated households, which could have facilitated vertical religious transmission within the household.31 Column 1 of Table 9 (panel b) shows that exposed male cohorts are more likely to marry women with Islamic schooling. This could be due to matching within Islamic schools, matching post-schooling, or arranged marriages by parents who sent their children to an Islamic school. It could also be an indirect consequence of the slightly larger effect of SD INPRES on madrasa education for girls (see Section 5.2). The effects are null for women’s marital choice, perhaps because women face greater constraints in selecting partners (column 2). Overall, this increased mixing of Islamic-educated individuals in the marriage market, in turn, has implications for the transmission of religious values within the household. In columns 3–6 of panel (b), we explore the transmission of religious values operating outside the Islamic school system. As our proxy for engagement with Islam, we use the Arabic literacy status of parents and children measured in Susenas. We showed in Section 7.2 that SD INPRES increased Arabic literacy among affected cohorts. In columns 3–4, our dependent variable is a dummy for all 3 members of a nuclear household (the father, the mother, and the child) being literate in Arabic. Both a father’s and a mother’s exposure to SD INPRES increase the likelihood that the entire household is literate in Arabic, reflecting both assortative mating between the parents and a higher likelihood of religious cultural transmission involving children. Finally, in columns 5–6, we look at an indicator for the child’s Arabic literacy in the subsample of (parental) respondents who are literate in Arabic and whose child has received no Islamic schooling. While this sample split is endogenous to INPRES exposure, this test provides suggestive evidence of vertical religious transmission outside the Islamic school classroom. Indeed, among the subset of parents literate in Arabic, we find that children educated outside the Islamic school system are more likely to be literate in Arabic when the parents were directly exposed to SD INPRES. This could be due to direct instruction inside the home, extracurricular education at the local mosque or madrasa diniyah, or both. Overall, parents exposed to mass public schooling ensure that their children maintain a strong religious 31 A large literature studies the role of homogamy in religious cultural transmission (see Bisin et al., 2020, for a review). 25 identity both through their choice of school, and by investing in greater vertical transmission of religious values within the family. 8 Conclusion One of the most ambitious educational policies ever implemented, SD INPRES pursued developmental as well as ideological objectives. A large literature documents the policy’s substantial and long-lasting effects on human capital. In this paper, we provide the first comprehensive investigation of its effects on education markets as well as its nation-building legacy. Our appraisal of the policy’s long-term ideological impacts takes into account the competitive response of Islamic schools. Before the 1970s, the Indonesian state had sought to suppress the country’s long-standing Islamic education sector to facilitate the emergence of a secular national identity. SD INPRES was also designed with this goal in mind, in the context of a young political regime focused on promoting national unity and its own legitimacy. Our findings point to some surprising consequences of mass schooling. Despite its enormous investments, the Suharto regime reaped little electoral gain from SD INPRES. Nor did the mass public school expansion strengthen national identity or support for the secular state over the long run. This runs counter to the nation-building effects of mass schooling seen elsewhere historically. Part of the difference here lies in the Islamic sector response, which counteracted state investments in secular education by capturing new primary graduates from state schools, by building more religious schools, and by expanding Islamic content inside the classroom. In the long run, this increased piety may have crowded out Indonesian identity without representing a genuine threat to the legitimacy of the Indonesian state. Indeed, increased religiosity was not accompanied by increased support for Islamist ideology. The different patterns of piety and politics between this study and Bazzi et al. (2020) highlight the vast diversity within Indonesian Islam that persists to this day. While schools borne out of the 1960s shock advocate a more fundamentalist ideology emphasizing the importance of sharia law, those borne out of market competition with SD INPRES schools in the 1970s provide a curriculum more accommodative of the secular state, despite their large volume of religious content. Successive Indonesian governments have successfully capitalized on these divisions. Moderate establishments are co-opted and incorporated in the mainstream education system while those promoting more radical ideologies are marginalized. Yet, more than fifty years after SD INPRES attempted to eliminate it, the dual system remains. 26