Religious Education in the Age of Mass Schooling Appendix B

B A Stackelberg Model of Competition in Education Markets This section describes a simple model to rationalize the increase in the supply of Islamic schools in response to SD INPRES. The model leverages insights from Bulow et al. (1985) and Marini and Rodano (2013) on strategic complementarities in Cournot and Stackelberg duopolies. Suppose two players j = s, i compete in a Stackelberg game. The state (s) is the Stackelberg leader and the Islamic sector (i) is the follower. Both players maximize the number of students enrolled in their respective schools. The payoff of player j is: π(qj , Q) = (1 + Q)−b qj where P (Q) = (1 + Q)−b , b > 1 is the inverse demand for schooling and Q = qs + qi is the total supply of schools across both sectors. We solve recursively for a Stackelberg (subgame perfect) equilibrium. The Islamic sector solves: ri (qs ) = arg max(1 + qs + qi )−b qi , (B.1) qi taking the state’s choice of qs as given. The FOC with respect to qi yields: ri (qs ) = 1 + qs = qi , b−1 (B.2) which implies that i and s are strategic complements (ri (qs ) is upward sloping). Given the Islamic sector’s best response, the state solves: qs = arg max(1 + qs + ri (qs ))−b qs (B.3) qs The equilibrium number of state schools is then: qs∗ = 1 , b−1 (B.4) which implies that qi∗ = b/(b − 1)2 Islamic schools are produced in equilibrium. Now, suppose that an outward shift in demand for education leads the state to increase its provision 0 of schools. Specifically, s and i now face inverse demand P (Q) = (1 + Q)−b , b > b0 > 1. In this case, the state produces 1 b0 −1 > 1 b−1 schools and the Islamic sector responds by supplying b0 (b0 −1)2 > b (b−1)2 schools. This result provides a microfoundation for the Islamic sector’s positive supply response discussed in Section 6.1 1 Of course, the model makes several simplifying assumptions, including an assumption of zero marginal costs. This assumption ensures a closed-form solution but may not be innocuous. In particular, the results in Section 6.1 suggest that the Islamic sector indeed may have faced different costs of constructing new madrasa across districts owing to differences in the availability of Islamic charitable assets (waqf ). Proof of B.2: The FOC of B.1 yields −b(1 + qs + qi )−b−1 qi + (1 + qs + qi )−b = (1 + qs + qi )−b−1 (−bqi + 1 + qi + qs ) = 0 ⇒ qi = Proof of B.4: The FOC of B.3 yields     1 + qs −b−1 1 + qs −b 1 + qs + qs + 1 + qs + =0 b−1 b−1       1 + qs −b−1 b 1 + qs 1 + qs + −b qs + 1 + qs + =0 b−1 b−1 b−1  −b b b−1 −b2 qs + b − 1 + bqs − qs + 1 + qs = 0 −b2 qs + b + bqs = 0 qs = i’s equilibrium strategy is then obtained by plugging B.4 into B.2. 1 b−1 1 + qs b−1 C Data Sources and Construction We describe here the key variables and data sources used in the paper. Education: Survey and Administrative Data Surveys. We measure years and type of schooling using the annual National Socioeconomic Survey (Susenas) from 2012, 2013, 2014, 2016, 2017, and 2018. These enumerate schooling measures for all household members and also record the birth district for each, which we merge with the district-level INPRES intensity measure collected by Duflo (2001). We additionally use Islamic school attendance data from the Indonesia Family Life Survey (IFLS) in 1993, 1997, 2000, 2007, and 2014. The IFLS is too limited geographically for our econometric analysis, but we use it for descriptive purposes in Table 1 and elsewhere in the text. Susenas reports the type of education (Islamic or secular) for the final level of schooling certification (primary, junior secondary, and senior secondary) as well as the final year of schooling attended if falling between certification levels. Our measure of Islamic schooling is based on the union of these two, but results are nearly identical when restricting to final level certified or final level attended. For example, some individuals report completing secular primary school and attending two years of Islamic junior secondary but not completing the full three years at that level. Our approach identifies this individual as having secular primary school and, separately, Islamic junior secondary school. Registries. We use data from numerous administrative sources provided by the Government of In- donesia. Table 1 used data on total non-pesantren enrollment in 2019 from the Ministry of Education (MEC) and Ministry of Religious Affairs (MORA) as reported at the following website: http: //apkapm.data.kemdikbud.go.id (accessed March 22, 2020). Pesantren enrollment in 2019 is computed from school-level records that we scraped from the MORA portal: https://ditpdpontren. kemenag.go.id/pbsb/ (accessed November 15, 2018). These records also indicate the district and year of establishment for each pesantren (see Bazzi et al., 2020, for additional details). Data on madrasa come from MORA registries provided to us by MORA officials in August 2019 and January/February 2020.1 These include village, district, and year of establishment for all formal madrasa (primary, junior secondary, and senior secondary) as well as informal madrasa diniyah. The latter are entirely privately-run. The former are majority private with a small fraction (around 8%) that are publiclyrun by MORA. Overall, 6% of madrasa and 22% of pesantren, respectively, have missing establishment years. This missing-ness is uncorrelated with SD INPRES intensity. Data on non-Islamic schools come from a MEC registry known by its Indonesian acronym Dapodik.2 These data include village, district, and year of establishment for all formal schools not administered by MORA. These include 166,257 publicly-run schools and 52,888 privately-run schools. Among the latter, 10,919 schools have Islamic names, indicating that they are likely religious schools operating under the 1 We are grateful to the following individuals for graciously sharing these data: Dodi Irawan, Aziz Saleh, Dr. Abdullah Faqih, and Doni Wibowo. 2 We are grateful to Wisnu Harto Adiwijoyo for graciously sharing these data. MEC instead of MORA. These schools are subject to different regulations on curriculum and also have access to other sources of state funding than the Islamic schools under MORA oversight. We distinguish secular from Islamic-named private schools in the MEC data by identifying the latter as having any of the following terms appearing in the school name: Islam, Darussalam, Darul, Muhammada, Salam, Sunna, Kuran, Jihad, Umma, Madrasa Halal, or Imam. We use this distinction to examine private secular schools in Appendix Figure A.3. We measure curriculum content at the school–grade level using data from the Sistem Informasi Aplikasi Pendidikan (SIAP) registry of schools. We scraped data from this registry’s online portal over several months in Fall 2019: http://siap-sekolah.com/. As of April 2020, SIAP only included detailed curriculum timetables for madrasa. We link these madrasa to the MORA registry using school IDs reported in both sources. The SIAP report detailed course timetables for every hour of every schoolday in a typical week for the 2018–2019 academic year. There are over 3,000 distinct course titles with many being (spelling) variations on the same topic. We coded up each course as being Islamic or non-Islamic and also identified courses associated with civic education and Pancasila, which are known by their Indonesian acronym of PPKN. These course codings are available upon request. SIAP includes data for around one-fifth of all madrasa, but as noted in footnote 22, this selective reporting likely works against our core findings with respect to INPRES intensity. We measure test scores using data collected by the MEC on the national exam scores in 2014 for science and math. We scraped these data in March 2015 from the MEC portal: http://referensi. data.kemdikbud.go.id. We link these data to the Dapodik and MORA registries using school IDs available across datasets. Electoral Outcomes: Vote Shares and Legislative Candidates Vote Shares. First, we draw upon district-level vote shares by party from the national legislative elections in 1971, 1977, 1982, 1987, 1992, 1999, 2004, and 2009. These data were graciously shared with us by individuals that worked with Dwight King. In 1971, one observes the following Islamic parties: NU, PSII, Perti, and the Muslim Party of Indonesia (Partai Muslimin Indonesia or Parmusi). From 1977 to 1992, the only Islamic party was the United Development Party (Partai Persatuan Pembangunan or PPP), which was forged out of a forced merger of the four Islamic parties contesting the 1971 election. We study the vote shares for the PPP and the Suharto regime party, Golkar. Legislative Candidates. We use data on the universe of legislative candidates in the 2019 election. Thanks to Nicholas Kuipers for scraping and sharing these data from the Indonesian Electoral Commission: http://www.kpu.go.id/. These include candidates for national, provincial, and district legislatures. We use information on candidate age, district, and party ticket. We also categorize their campaign motivation and platform statements as appealing to Islamic themes as reflected in the following words: umma, dawah, Muslim, Islam, sharia, and jihad. We separately classify appeals to nation building as reflected in the following words: Pancasila, Indonesia, NKRI, bangsa (nation), bhinneka (diversity), and satuan (unitary). The latter three terms are staples in the nation-building corpus of Indonesian leaders and literature. NKRI is an acronym for the Indonesian homeland in a popular nationalistic slogan. Linguistic Proxies for Identity We proxy for national identity using an indicator of whether an individual speaks the national language, Bahasa Indonesia, as his/her main language at home (instead of his/her native ethnic language). This is distinct from Indonesian speaking ability, which we also observe. These data—along with religion, age, and district of birth—are recorded in the complete-count 2010 Population Census, which we obtained from the Harvard Library. We view Arabic language proficiency as one indicator of Islamic identity. The Susenas data described above record literacy in Latin, Arabic, and other alphabets. Religiosity and Religious Political Preferences We use rich individual-level survey data from Pepinsky, Liddle and Mujani (2018), which is based on a 2008 survey conducted by the authors in which 10 individuals were sampled from each contemporary district. These data include individual age, religion, years and type of education, a host of questions on Islamic piety, practice, and political preferences. Seven Islamic practices are explored in Table 7. The survey also record dimensions of support for Islamic law (sharia) and religious politics more generally. We also use a measure of stated support for Pancasila.