The Institutional Foundations Of Religious Politics (Figures)

F IGURES F IGURE I Illustration of the Difference-in-Discontinuity Design Contemporary Waqf and Islamism A. Waqf Land ←not expropriable expropriable→ 1500 few 5-9 ha holdings many 5-9 ha holdings 10 pesantren, 2008 | island FE waqf, % of zoned land, 2003 | island FE ←not expropriable B. Islamic Boarding Schools expropriable→ few 5-9 ha holdings many 5-9 ha holdings 1000 5 0 500 0 -5 -500 250 300 350 400 450 population density in 1960 500 550 250 300 ←not expropriable ←not expropriable expropriable→ 10 few 5-9 ha holdings many 5-9 ha holdings Sharia regulations, 1998-2013 | island FE Islamist vote share, 1999-2014 | island FE 500 550 500 550 D. Sharia Regulations C. Islamist Vote Share .1 350 400 450 population density in 1960 .05 0 -.05 expropriable→ few 5-9 ha holdings many 5-9 ha holdings 5 0 -5 250 300 350 400 450 population density in 1960 500 550 250 300 350 400 450 population density in 1960 Notes: This figure illustrates how the difference in outcomes between districts with “few” and “many” marginal 5–9 ha landholdings (below and above the median, respectively) changes at the 400 people/km2 cutoff, above which 5–9 ha holdings are expropriable and below which they are not. We partial out island fixed effects from the outcomes before plotting these figures, in keeping with our within-island research design. We restrict the graph to districts in the 250–550 people/km2 range for presentational purposes, but the above/below median split is based on the full sample. Our estimating equation (1) uses the full, continuous variation in 5–9 ha holdings whereas this figure splits the full sample of districts into the above and below median 5–9 ha holdings. As such, this should be read only as an approximation to the identifying variation in the regression results reported in subsequent figures and tables. The curves are local linear regressions with an Epanechnikov kernel and bandwidth of 50. F IGURE II Timing of Waqf Endowments Difference-in-Discontinuity, β: new waqf-endowed mosque land (ha), annualized A. Waqf Land Endowments in Newly Created Mosques 1.25 1 .75 .5 .25 0 1920 1930 1940 1950 1960 1970 decade of establishment 1980 1990 2000 1990 2000 Difference-in-Discontinuity, β: number of students at new pesantren, annualized B. Size of Newly Created Pesantren 600 400 200 0 1920 1930 1940 1950 1960 1970 decade of establishment 1980 Notes: This figure shows difference-in-discontinuity estimates based on equation (1) fully interacted with decade dummies. The unit of observation is a district–year from 1920 to 2009, and the dependent variable is (A) the amount of waqf land allocated to newly established mosques, and (B) total student enrollment in newly established pesantren. Each coefficient (circle) can be read as the annual effect of an additional 100 marginal expropriable landholdings (MEH) on the given outcome in the given decade. Both outcomes are constructed from Ministry of Religion administrative records. Standard errors are clustered by district, and the vertical bars represent 90% confidence intervals. In (A), we can reject the null that the coefficient for the 1960s is the same as the coefficient for the 1950s with a p-value equal to 0.010. The corresponding p-value for (B) is 0.093. TABLES TABLE I C ONTEMPORARY Waqf L AND AND Waqf -E NDOWED I NSTITUTIONS Waqf Land hectares % total % zoned (1) (2) (3) Waqf -Endowed Institutions mosques pesantren madrasas (4) (5) (6) expropriation intensity 0.207** (0.097) 0.251** (0.107) 3.286** (1.483) 2.956*** (0.985) 0.532** (0.264) 1.024** (0.410) Number of Villages Number of Districts Dependent Variable Mean R2 55,200 191 0.848 0.033 55,200 191 0.518 0.005 55,200 191 6.127 0.044 48,978 189 3.921 0.234 48,978 189 0.503 0.165 48,978 189 0.787 0.195 Notes: This table reports estimates of equation (1). Waqf land is observed in the 2003 Podes village survey and measured in hectares (inverse-hyperbolic sine transformed) in column (1), as % of total land in column (2), and as % of zoned land in column (3). Waqf -endowed institutions are observed at the village level in 2008 and include: the number of mosques in column (4), the number of Islamic boarding schools (pesantren) in column (5), and the number of Islamic day schools (madrasa) in column (6). Expropriation intensity denotes the interaction of an indicator equal to one for districts above 400 people/km2 in 1960 (Above400) with the number of marginal expropriable holdings (M EH, in 100s) in the size category subject to redistribution above this cutoff according to the Basic Agrarian Law, namely holdings between 5 and 9 hectares. The coefficients on each of the own terms are included in the regression but not reported here. The specification includes island fixed effects and a cubic polynomial in 1960 population density interacted separately with the two land reform exposure variables (Above400 and M EH) and their interaction. The sample size drops in columns (4)–(6) as the data could not be linked to the baseline 2003 data for certain villages as a result of changes in administrative codes and boundaries. Including a combined measure of pesantren and madrasa for the complete sample in 2003 yields a coefficient of 0.689 (0.270)**. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE II I SLAMIST PARTY S UPPORT IN L EGISLATIVE E LECTIONS Village 1999 Top 3 Finish (1) District Level 1999 1999-2014 Vote Share Vote Share (2) (3) A. Islamist Parties expropriation intensity 0.158* (0.083) 0.049** (0.024) 0.044** (0.020) Number of Villages Number of Districts Dependent Variable Mean R2 55,200 191 0.516 0.051 – 189 0.136 0.205 – 191 0.154 0.222 B. Moderate Islamic Parties expropriation intensity -0.129 (0.083) -0.024 (0.037) -0.008 (0.023) Number of Villages Number of Districts Dependent Variable Mean R2 55,200 191 0.468 0.204 – 189 0.170 0.387 – 191 0.152 0.339 C. Secular Parties expropriation intensity -0.009 (0.028) -0.025 (0.044) -0.036 (0.031) Number of Villages Number of Districts Dependent Variable Mean R2 55,200 191 0.932 0.270 – 189 0.695 0.414 – 191 0.695 0.327 Notes: This table reports estimates of equation (1). The dependent variable is: an indicator of whether a given political family finished in the top 3 in the 1999 national legislative elections, as observed in the 2003 Podes (column 1), district-level vote shares in the 1999 elections (column 2), and district-level vote shares in each of the 1999, 2004, 2009 and 2014 elections stacked into a single pooled regression with year fixed effects (column 3). The top panel reports effects on electoral support for Islamist parties, which include the United Development Party (PPP), the Prosperous Justice Party (PKS), and the Crescent Star Party (PBB). In the middle panel, electoral support is defined with respect to moderate Islamic parties: the National Mandate Party (PAN) and the National Awakening Party (PKB). Unlike Islamist parties, which rejected Pancasila in 1999 and had advocated for including Islamic law in the Indonesian Constitution, these two parties have pluralistic ideologies that embrace Pancasila, the secular-nationalist doctrine of the Indonesian state. The bottom panel looks at effects on support for all other, secular parties, including the Indonesian Party of Struggle (PDI-P) and the Golkar Party. See Appendix Tables A.5 and A.6 for party-specific outcomes. Regressions in column (1) are at the village level while regressions in columns (2) and (3) are at the district and district×election year level, respectively. See the notes to Table I for additional details on the specification and Appendix Table A.1 for summary statistics. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE III I SLAMISM AND L OCAL G OVERNANCE expropriation intensity Number of Districts R2 Local Relig. Gov. Officials (1) Sharia Regulations (2) Sharia Microfinance (3) 0.614** (0.266) 0.600** (0.289) 0.346 (0.237) 1.153* (0.603) 0.825** (0.397) 0.045 (0.424) 0.837** (0.349) 1.119*** (0.357) 191 0.289 191 0.136 189 0.435 80 0.436 80 0.688 80 0.451 114 0.268 114 0.304 Islamic Court Cases re: Waqf Inherit. Marital (4) (5) (6) Islamist Vigilante Activity Incidents Casualties (7) (8) Notes: This table reports estimates of equation (1) for several outcomes, all of which are standardized: the number of district-level government employees dedicated to managing religious affairs in the local Ministry of Religion office (column 1), the number of sharia regulations adopted in the district between 1998–2013 (column 2), the share of villages in the district with at least one sharia-based microfinance institution (column 3), the number of sharia court cases related to waqf, inheritance, and marriage/divorce (columns 4–6, respectively), and the number of incidents (column 7) and casualties (column 8) inflicted by the Islamic Defenders Front (FPI), an Islamist vigilante group. All regressions are run at the district level. The sample size is smaller in columns (4)–(8) due to missing data on court cases and Islamist vigilante activity. Expropriation intensity is not significantly associated with attrition in either dataset (see text for details). See the notes to Table I for additional details on the specification and Appendix Table A.1 for summary statistics. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. D EMAND FOR AND TABLE IV S UPPLY OF R ELIGIOUS P OLITICS A. D EMAND : S URVEY R ESPONSES Data Source: Outcome: expropriation intensity Number of Individuals Number of Districts Dependent Variable Mean R2 IFLS Survey candidate [...] very impt. in determining vote religion religiosity (1) (2) Muslim president very impt. (3) Pepinsky et al. Survey Religiosity Sharia Law president support support very impt. objective subjective (4) (5) (6) 0.092*** (0.026) 0.083*** (0.029) 0.117*** (0.044) 0.088** (0.043) 0.056* (0.033) 0.062 (0.052) 43,965 157 0.394 0.084 43,965 157 0.406 0.076 1,825 129 0.659 0.052 1,822 129 0.769 0.035 1,840 129 0.429 0.087 1,709 129 0.600 0.082 B. S UPPLY: L EGISLATIVE C ANDIDATE E NTRY Outcome: expropriation intensity Number of Districts Dependent Variable Mean R2 campaign on Islam & Sharia (7) Hajj experience (8) religious scholar (9) occupation: teacher student (10) (11) teacher/student Islam campaign (12) 9.624*** (3.318) 43.406*** (15.805) 0.809** (0.322) 14.319*** (5.117) 5.692 (4.241) 1.411*** (0.461) 191 7.1 0.160 191 45.3 0.267 191 0.3 0.324 191 16.9 0.309 191 23.6 0.239 191 0.7 0.183 Notes: This table reports estimates of equation (1) for demand-side survey responses (panel a) and supply-side legislator responses (panel b). In Panel A, columns (1)–(2) look at two outcomes from the Indonesia Family Life Survey (IFLS): an indicator for whether the religion (column 1) and the religiosity (column 2) of a candidate is a very important factor in individual voting decisions. Panel A, columns (3)–(6) look at outcomes from the Pepinsky, Liddle and Mujani (2018) survey data: whether respondents say the president being Muslim (column 3) and being religious (column 4) is very important; and whether they support an index of specific Sharia-inspired legal regulations (column 5) and generically support Sharia law (column 6). These two surveys do not cover all districts in Indonesia, hence the smaller district sample size. These individual-level regressions control for gender, age and age squared, and an indicator for urban; the IFLS regression additionally controls for survey wave fixed effects. None of these added controls materially affect the results. The regressions in column (1)–(4) are also restricted to Muslims (see Table V). In Panel B, we look at data on legislators’ profiles in the 2019 election: the number of candidates for district parliament that mention Islam- or Sharia-related concepts in their campaign platforms (column 7), indicate their status as Hajj pilgrims in their formal name listed on the ballot (column 8), hail from a religious scholar background (column 9), report “teacher” as their primary occupation (column 10), report “student” as their primary occupation (column 11), and the number of students and teachers with Islamic campaign platforms as measured in column 7 (column 12). See the notes to Table I for additional details on the specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE V I SLAMIC P IETY AND P RACTICE IFLS Survey Source: Outcome: Am I Muslim? Am I Very Religious? (1) expropriation intensity Number of Individuals Number of Districts Dependent Variable Mean R2 Pepinsky et al. Survey Am I Muslim Pray 5x/day Fast Ramadan Read Quran Friday Mosque Pray Sunna Prayer Group Pay Zakat (2) Trust Co-Muslims More (3) (4) (5) (6) (7) (8) (9) (10) (11) Practices Index cols. 5-11 (12) 0.033 (0.050) -0.003 (0.006) -0.004 (0.014) -0.002 (0.032) 0.024 (0.043) 0.063 (0.043) -0.046 (0.060) -0.065 (0.048) -0.036 (0.046) 0.006 (0.055) 0.030 (0.056) -0.014 (0.039) 45,296 158 0.899 0.244 43,965 157 0.102 0.037 40,727 157 0.119 0.094 2,047 137 0.903 0.329 1,847 129 0.688 0.132 1,848 129 0.822 0.075 1,843 129 0.276 0.035 1,842 129 0.236 0.076 1,829 129 0.180 0.056 1,841 128 0.259 0.048 1,841 129 0.831 0.048 1,848 129 0.446 0.087 Notes: This table reports estimates of equation (1) for the following outcomes: from the IFLS, a dummy for being a Muslim (column 1), self-reported religiosity (column 2), and relative trust towards co-Muslims and non-Muslims (column 3). From the Pepinsky, Liddle and Mujani (2018) data, we look at individuals who self-report as Muslim (column 4), individuals who report praying 5 times a day (column 5), fasting during Ramadan (column 6), reading the Qur’an (column 7), always attending Friday prayer (column 8), attending non-mandatory Sunnah prayers (column 9), being part of a prayer group (column 10), and paying zakat collection group (col. 11). We also pool all practices in columns (5)–(11) in an index, and regress this index on our main specification in column (12). All regressions except those in columns (1) and (4) are restricted to Muslims. All regressions control for gender, age and age squared, and an indicator for urban; the IFLS regression additionally controls for survey wave fixed effects. None of these added controls materially affect the results. See the notes to Table I for additional details on the specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. E FFECTS OF Waqf -E NDOWED TABLE VI L AND ON R ELIGIOUS I NSTITUTIONS Waqf -Endowed Institutions mosques pesantren madrasas (1) (2) (3) AND P OLITICS Religious Politics Islamist Party Min. Religion Top 3, 1999 Vote Shr. 99-14 Employees (4) (5) (6) Sharia Regulations (7) A. O RDINARY L EAST S QUARES waqf land Number of Villages/Districts Dependent Variable Mean R2 0.614*** (0.067) 0.129*** (0.021) 0.177*** (0.028) 0.037*** (0.007) 0.034*** (0.008) 0.102* (0.060) 0.527** (0.229) 48,710 3.930 0.231 48,710 0.505 0.166 48,710 0.791 0.188 55,200 0.516 0.052 191 0.154 0.306 191 6.474 0.362 191 1.681 0.139 B. I NSTRUMENTAL VARIABLES waqf land [weak-instrument-robust p-value] Number of Villages/Districts Dependent Variable Mean First Stage Effective F Statistic Underidentification Test, p-value 14.612** (6.967) [0.003] 2.632** (1.083) [0.045] 5.062** (2.199) [0.013] 0.766** (0.369) [0.057] 0.148* (0.087) [0.017] 0.654* (0.373) [0.095] 2.633 (2.065) [0.156] 48,710 3.930 4.5 0.040 48,710 0.505 4.5 0.040 48,710 0.791 4.5 0.040 55,200 0.516 4.6 0.038 191 0.154 7.3 0.016 191 6.474 5.8 0.010 191 1.681 5.8 0.010 Notes: This table reports estimates of equation (2). Panel A reports OLS estimates, and Panel B reports IV estimates, where we use the difference-in-discontinuity term from equation (1), Above400 × M EH, as an instrument for waqf land (inverse-hyperbolic-sine transformed as in Column (1) of Table I). All other controls in equation (1) are retained in the OLS to ensure comparability with the IV. Columns (1)–(3) examine institutional outcomes from Table I: the number of mosques (column 1), Islamic boarding schools (pesantren, column 2), and Islamic non-boarding schools (madrasa, column 3) observed at the village level in 2008. Columns (4)–(5) look at outcomes from Table II: an indicator for whether Islamist parties finished in the top 3 in the 1999 national legislative elections (column 4), and the district-level vote share received by Islamist parties across the 1999, 2004, 2009 and 2014 elections (column 5). Columns (6)–(7) examine outcomes from Table III: the log number of local government employees dedicated to managing religious affairs (column 6), and the number of sharia regulations adopted in the district between 1998–2013 (column 7). The null of the underidentification test is that the equation is underidentified. The weak-instrument robust p-value is based on the Anderson-Rubin test. See the notes to Table I for additional details on the specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. TABLE VII Waqf L AND , A GRICULTURAL D EVELOPMENT, AND M ISALLOCATION log agric. GDP/capita (1) log agric. productivity (2) capital/labor ratio (3) capital/land ratio (4) light intensity (5) A. R EDUCED F ORM expropriation intensity Number of Villages Number of Districts Dependent Variable Mean R2 -0.577*** (0.215) -0.113* (0.059) -0.729** (0.298) -0.050** (0.025) 0.112 (0.103) 47,598 191 13.012 0.060 47,598 191 -0.473 0.158 47,598 191 2.223 0.044 47,597 191 0.111 0.021 41,459 187 1.648 0.658 B. O RDINARY L EAST S QUARES waqf land -0.011 (0.021) 0.016 (0.013) 0.008 (0.050) -0.006** (0.003) -0.011 (0.011) Number of Villages Number of Districts Dependent Variable Mean R2 47,249 191 13.0 0.053 47,249 191 -0.470 0.159 47,249 191 2.239 0.044 47,248 191 0.112 0.020 41,151 187 1.657 0.657 C. I NSTRUMENTAL VARIABLES waqf land [weak-IV-robust p-value] Number of Villages Number of Districts Dependent Variable Mean First Stage Effective F Stat Underidentification Test, p-value -2.526** (1.013) [0.008] -0.495 (0.335) [0.058] -3.219** (1.444) [0.015] -0.221* (0.122) [0.049] 0.563 (0.495) [0.272] 47,249 191 13.0 6.4 0.012 47,249 191 -0.470 6.4 0.012 47,249 191 2.239 6.4 0.012 47,248 191 0.112 6.4 0.012 41,151 187 1.657 4.5 0.038 Notes: This table reports estimates from equation (1) in Panel A, and estimates from equation (2) in Panels B and C using OLS and IV, respectively, as in Table VI. The village-level dependent variable, measured in 2003, is: the log of agricultural GDP per capita (column 1), the revenue-weighted log of output per hectare by crop (column 2), the total number of agricultural machines per capita (column 3) and per hectare planted (column 4), and nighttime light intensity (inverse-hyperbolic-sine transformed). The sample in columns (1)–(4) is restricted to villages with positive production of at least one crop. The sample drops in column (5) due to problems merging luminosity shapefiles with the administrative data in 2003. The null of the underidentification test is that the equation is underidentified. The weak-instrument robust p-value is based on the Anderson-Rubin test. See the notes to Table I for additional details on the specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. A LTERNATIVE E XPLANATIONS TABLE VIII (I): L AND I NEQUALITY AND D EMOGRAPHIC C HANGE ∆ Landholdings Distribution 1963–1980 1963–1985 1963–1990 no. >5 ha λ no. >5 ha λ no. >5 ha λ (1) (2) (3) (4) (5) (6) expropriation intensity Number of Districts R2 ∆ Demographics, 1961-71 population sex ratio (7) (8) -0.075 (0.179) -0.206 (0.192) -0.011 (0.154) -0.222 (0.170) 0.068 (0.214) -0.144 (0.190) -0.090 (0.102) -0.323 (0.237) 191 0.086 191 0.404 191 0.131 191 0.395 191 0.076 191 0.359 168 0.169 168 0.157 Notes: This table reports estimates of equation (1) for district-level measures of changes in the land distribution since 1963: changes in the number of 5+ hectare holdings and land dispersion (λ) between 1963 and 1980 (columns 1 and 2), 1963 and 1985 (columns 3 and 4); and 1963–1990 (column 5 and 6). Recall that the estimated Pareto parameters, λ, are decreasing in in dispersion/inequality. We also consider district-level population growth between 1961 and 1971 (column 7) growth in the male-to-female sex ratio between 1961 and 1971 (column 8). The sample size is smaller in columns (7)–(8) due to uncovered districts in the 1971 Population Census. All dependent variables are normalized to have mean zero and standard deviation one. See the notes to Table I for additional details on the specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. TABLE IX A LTERNATIVE E XPLANATIONS (II): S CHOOLING AND P UBLIC G OODS expropriation intensity Number of Villages Number of Districts Dep. Var. Mean R2 INPRES Schools (1) Mean Yrs. Schooling (2) -0.018 (0.260) -0.030 (0.259) -0.027 (0.017) -0.003 (0.008) 0.007 (0.008) 0.005 (0.015) -0.012 (0.029) 191 191 2.233 0.289 46,147 188 4.676 0.126 46,628 190 0.372 0.109 46,628 190 0.123 0.159 46,628 190 0.092 0.114 41,437 187 0.392 0.088 41,437 187 0.497 0.264 Population Share with . . . School Primary Junior Sec. Senior Sec. (3) (4) (5) Public Goods Index Health Infrastructure (6) (7) Notes: This table reports estimates of equation (1) for the following outcomes: the number of public primary schools per 1,000 students constructed at the district level by the Suharto government from 1973–78 as part of the INPRES program (column 1); the mean years of schooling in 2000 (column 2); the share of the village population in 2000 with primary schooling (column 3), with junior secondary schooling (column 4), with senior secondary schooling (column 5); village-level indices, ranging from zero to one, capturing the presence of health public goods including doctors, midwives, and health clinics (column 6) and infrastructure public goods including 4-wheel road access, safe water, sewage, garbage collection, and kerosene supply (column 7). These indices are based on the mean across all rounds of Podes from 1999 to 2014. The sample size is smaller in columns (5) and (6) due to changing boundaries and merging difficulties. See the notes to Table I for additional details on the specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. A A.1 Further Empirical Results Descriptive Statistics TABLE A.1 Summary Statistics Mean SD N Source 342.38 .309 392 .846 22.7 1.2 .40 .15 .136 .154 .170 .152 777.9 1.7 .058 .61 208 10,840 .30 .31 7.1 45.3 .26 16.9 23.6 0.73 754.34 .463 577 .335 29.5 2.5 .24 .14 .098 .086 .142 .106 480.3 2.6 0.073 1.30 624 10,005 .91 1.03 10.8 47.0 .64 16.7 23.7 1.56 191 191 191 191 191 191 191 191 189 746 189 746 191 191 189 80 80 80 114 114 191 191 191 191 191 191 1961 Population Census 1963 Agricultural Census 1963 Agricultural Census 1963 Agricultural Census Min. Relig. Affairs (Kemenag) Min. Relig. Affairs (Kemenag) Electoral Commission Electoral Commission Electoral Commission Electoral Commission Electoral Commission Electoral Commission Min. Relig. Affairs (Kemenag) Buehler (2016) 2017 Podes Relig. Courts Info. Syst. (SIPP) Relig. Courts Info. Syst. (SIPP) Relig. Courts Info. Syst. (SIPP) Violence Monitoring Syst. (SNPK) Violence Monitoring Syst. (SNPK) Electoral Commission Site Electoral Commission Site Electoral Commission Site Electoral Commission Site Electoral Commission Site Electoral Commission Site 3.44 .52 6.13 3.92 .50 .787 .516 669.2 10.0 46.6 39.6 2.97 20.8 4.67 1.31 1.83 .50 620.6 21.8 56.2 55,200 55,200 55,200 48,978 48,978 48,978 55,200 49,311 55,200 55,200 2003 Podes 2003 Podes 2003 Podes 2008 Podes 2008 Podes 2008 Podes 2003 Podes Google Maps/GIS 2003 Podes 2003 Podes .394 .406 .659 .769 .439 .600 .489 .491 .474 .421 .261 .490 43,965 43,965 1,825 1,822 1,840 1,709 IFLS IFLS Pepinsky et al. (2018) Pepinsky et al. (2018) Pepinsky et al. (2018) Pepinsky et al. (2018) District-level data Historic Population density Above 400 ppl/km2 threshold Expropriable holdings (5-9 hectares) Baseline land dispersion Mosques by 1920 Pesantren by 1920 Islamic vote 1955–57 Communist vote 1955–57 Islamist vote share, 1999 Islamist vote share, 1999–2014 Islamic moderate vote share, 1999 Islamic moderate vote share, 1999-2014 Local employees Min. Relig. Affairs Sharia regulations Islamic microfinance Islamic court waqf cases Islamic court inheritance cases Islamic court marital cases Islamist Vigilante (FPI) incidents Islamist Vigilante (FPI) casualties Candidates campaigning on Islam Candidates with Hajj experience Candidates who are religious scholars Candidates who are teachers Candidates who are students Candidates who are students/teachers campaigning on Islam Village-level data Waqf land in village (hectares) Waqf in village (% total land) Waqf in village (% zoned land) Mosques in village Pesantren in village Madrasas in village Islamist parties in top 3, 1999 Distance to nearest shrine (km) Dist. to subdistrict office (km) Dist. to district office (km) Individual-level survey data Candidate religion very important Candidate religiosity very important Muslim president very important Religiosity president very important Support Sharia law (objective) Support Sharia law (subjective) Notes: This table reports means, standard deviations (SD), and sample sizes for dependent and independent variables used in the paper and appendix. For a full elaboration of sources and variable construction, see Appendix B. F IGURE A.1 Distribution of Establishment Dates for Mosques and Pesantren A. Mosques B. Pesantren Notes: This figure reports the distribution of establishment dates for the universe of (a) mosques and (b) pesantren. Data from the Ministry of Religious Affairs. A.2 Identification Checks F IGURE A.2 Continuity of Marginal Expropriable Holdings (M EH) Around the 400 Cutoff number of marginal expropriable landholdings 600 400 200 0 250 300 350 400 450 population density in 1960 500 550 Notes: This figure illustrates the continuity of marginal expropriable landholdings (5-9 ha) across the 400 people/km2 cutoff. We restrict attention to districts in the 250–550 people/km2 range for presentational purposes. The curves are local linear regressions with an Epanechnikov kernel and bandwidth of 50. F IGURE A.3 Population Density: McCrary (2008) Test estimated density .006 .004 .002 0 0 200 400 600 800 population density, 1960 1000 1200 Notes: This figure reports the McCrary (2008) test for manipulation of the running variable, population density in 1960. The graph reveals no evidence of such manipulation. The figure excludes three districts with population density above 1300 people/km2 for presentational purposes. TABLE A.2 Balance on Time-Invariant and Pre-Reform Covariates N Mean Expropriation intensity Mosques by 1920 191 Pesantren by 1920 191 Islamist vote, 1955/57 160 Masyumi vote, 1955/57 160 NU vote, 1955/57 160 Darul Islam events 191 Ethnic Arab population, 1930 Pop. Census 180 Ethnic European population, 1930 Pop. Census 180 Ethnic Chinese population, 1930 Pop. Census 180 Communist vote, 1955/57 160 Mean rainfall shocks, 1955-59 191 Mean rainfall shocks, 1960-65 191 Baseline land dispersion, 1963 191 Number of males, 1961 Pop. Census 191 Number of females, 1961 Pop. Census 191 Number of farms, 1963 Ag. Census 189 Dryland area (ha), 1963 Ag. Census 190 Wetland area (Ha), 1963 Ag. Census 189 Total agricultural area (ha), 1963 Ag. Census 190 22.7 (29.47) 1.15 (2.49) .4 (.24) .22 (.2) .13 (.15) .88 (2.86) .67 (1.21) .83 (2.22) 6.27 (13.61) .15 (.14) .05 (.05) -.08 (.05) .85 (.34) 215.64 (150.54) 221.37 (157.82) 61.16 (41.14) 42.9 (40.58) 22.27 (17.28) 64.78 (47.49) .29 (.45) 287.76 (1794.06) .1 (.3) 669.15 (620.57) 10.01 (21.78) 46.53 (56.07) .44 (.5) 5.879 (7.511) -1.454 (1.513) .027 (.059) .015 (.014) .012 (.052) .576 (.537) .515 (.396) .122 (.338) 4.592 (3.797) -.072 (.05) .007 (.015) .010 (.011) .07 (.054) 25.554 (54.555) 23.4 (56.158) 2.76 (12.61) -10.041 (8.877) 5.901 (3.979) -4.18 (9.765) -.035 (.042) 22.145 (149.08) .036** (.014) 69.197 (60.192) 1.725 (1.122) 9.012* (5.331) -.018 (.016) Village located on hill 55200 Altitude 55200 Village located on beach 55200 Distance to nearest shrine (km) 48956 Distance to subdistrict office 55200 Distance to district office 55200 Java-Bali-NTB 55200 Notes: This table reports balance checks on baseline covariates (either time invariant or measured prior to the land reform) regressed on our baseline specification. Across columns, we report the number of observations (N ), mean and standard deviation in parentheses, and the difference-in-discontinuity coefficient from equation ( 1). The male and female population and land totals are divided by 1,000 for presentational purposes. * p<0.1, ** p<0.05, *** p<0.01. A.3 Waqf Establishments across Time Periods Table A.3 reports results based on aggregating decadal estimates in Figure II. The data is at the district-by-year level. Columns 1 and 3 look at the entire 1920–2009 period. Columns 2 and 4 examine solely the two decades around the land reform (1950s and 1960s). The first row reports the main effect of expropriation intensity, i.e., the estimate associated with expropriation intensity in the pre-land reform period. The second row reports the interaction between expropriation intensity and an indicator for the post-1960 period. TABLE A.3 Waqf Land and Waqf -Endowed Institutions, 1920–2009 Waqf Land, New Mosques 1920-2009 1950-69 (1) (2) expropriation intensity (pre) expropriation intensity×post-1960 Number of Districts Number of Observations Dependent Variable Mean Students, New Pesantren 1920-2009 1950-69 (3) (4) 0.075 (0.056) 0.490*** (0.174) 0.039 (0.088) 0.189*** (0.073) 18.75 (13.81) 197.31** (78.0) 13.38 (25.98) 61.94* (36.71) 191 17,190 0.551 191 3,820 0.435 191 17,190 158.9 191 3,820 64.3 Notes: This table reports estimates of equation (1) fully interacted with a post-1960 dummy. The dependent variables are measures of waqf endowed land given to new mosques and of the number of students enrolled in new pesantren between 1920–2009, as in Figure II. These measures vary at the district-year level. See the notes to Table I for additional details on the specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. A.4 Alternative Inference Procedures TABLE A.4 Alternative Approaches to Inference Waqf land, ha (1) expropriation intensity clustering by 1960 district, baseline clustering by 1960 district, wild bootstrap p-value clustering by 1960 district + Young (2016) effective d.o.f.-adj. Conley (1999) spatial HAC, 100 km bandwidth Conley (1999) spatial HAC, 300 km bandwidth Number of Observations Waqf -Endowed Institutions mosques pesantren madrasas (2) (3) (4) Religious Politics Islamist Party Min. Religion Top 3, 1999 Vote Shr. 99-14 Employees (5) (6) (7) Sharia Regulations (8) Vigilante Casualties (9) 0.207 2.956 0.532 1.024 0.158 0.047 0.614 0.600 1.119 (0.097)** [0.046]** (0.115)* (0.102)** (0.118)** (0.985)*** [0.004]*** (1.174)** (1.006)*** (0.219)*** (0.264)** [0.055]* – (0.330) (0.089)*** (0.410)** [0.036]** (0.489)* (0.254)*** – (0.083)* [0.098]* (0.098) (0.104) (0.079)* (0.022)** [0.062]* (0.026)* (0.020)** (0.014)*** (0.266)** [0.022]** (0.295)* (0.216)*** (0.189)*** (0.289)** [0.057]* (0.320)* (0.358)* (0.343)* (0.357)*** [0.032]** (0.408)** (0.303)*** (0.207)*** 55,200 48,978 48,978 48,978 55,200 191 191 191 114 Notes: This table re-estimates core results from Tables I–III using alternative approaches to inference besides the baseline clustering by 1960, the level at which the land reform policy varies. After that baseline, we use the wild cluster bootstrap, reporting the p-value. The Young (2016) adjustment accounts for the effective degrees of freedom implied by the residual variation. The Conley (1999) spatial HAC allows for correlated unobservables across districts or villages within 100 and 300 km of the given district or village centroid. The missing standard error (“–”) is due to a matrix computational failure. * p<0.1, ** p<0.05, *** p<0.01. A.5 Party-by-Party Electoral Outcomes Tables A.5 and A.6 report results for individual political parties in the 1999 and the 2014 legislative elections, respectively. The 3 Islamist parties running in these elections are the United Development Party (Partai Persatuan Pembangunan) or PPP, the Prosperous Justice Party (Partai Keadilan Sejahtera) or PKS, and the Crescent Star Party (Partai Bulan Bintang) or PBB. These parties advocate for an Islamic state based on sharia law and reject Pancasila, the doctrine promoting a secular and inclusive vision for Indonesia. The two moderate Islamic parties competing in these elections are the National Mandate Party (Partai Amanat Nasional) or PAN and the National Awakening Party (Partai Kebangkitan Bangsa) or PKB. Both parties endorsed Pancasila prior to the 1999 election. Secular parties include the Indonesian Democratic Party of Struggle (Partai Demokrasi Indonesia Perjuangan, the party of President Joko Widodo) or PDI-P and the Golkar Party (Partai Golongan Karya, the party of former President Suharto), as well as (in the 2014 election) the Great Indonesia Movement Party (Partai Gerakan Indonesia Raya) or Gerindra, the Democrat Party (PD), the Nasdem Party (Partai Nasdem), the People’s Conscience Party (Partai Hati Nurani Rakyat) or Hanura, and the Indonesian Justice and Unity Party (Partai Keadilan dan Persatuan Indonesia) or PKIP, a split from the Golkar party. TABLE A.5 Party-by-Party Voting Results, Top 7 (out of 48) Parties in the 1999 Election Islamist PKS (2) PPP (1) PBB (3) Moderate Islamic PAN PKB (4) (5) Secular Golkar PDIP (6) (7) Panel A: Village-Level Top 3 Finish (col. 1 of Table II) expropriation intensity 0.159* (0.082) -0.000 (0.002) -0.000 (0.004) -0.005 (0.018) -0.114 (0.089) -0.037 (0.062) 0.002 (0.049) Number of Villages Dependent Variable Mean R2 55,200 0.503 0.047 55,200 0.004 0.001 55,200 0.021 0.020 55,200 0.145 0.056 55,200 0.351 0.309 55,200 0.829 0.127 55,200 0.807 0.210 Panel B: District-Level Vote Share (col. 2 of Table II) expropriation intensity Number of Districts Dependent Variable Mean R2 0.042** (0.021) 0.004 (0.003) 0.003 (0.003) 0.013 (0.008) -0.038 (0.043) -0.001 (0.020) -0.030 (0.033) 189 0.107 0.178 189 0.010 0.392 189 0.019 0.154 189 0.060 0.540 189 0.110 0.373 189 0.276 0.749 189 0.301 0.427 Notes: This table reports estimates of equation (1) using specifications analogous to those in columns 1 and 2 of Table II, further disaggregating the three groups of parties reported in that table. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE A.6 Party-by-Party Voting Results, All 12 Parties in the 2014 Election expropriation intensity Number of Districts Dependent Variable Mean R2 Moderate Islamic PAN PKB (4) (5) Golkar (6) PDIP (7) Gerindra (8) Secular Demokrat (9) NasDem (10) Hanura (11) PKPI (12) 0.007* (0.004) 0.006 (0.008) -0.026 (0.019) 0.014 (0.033) -0.032* (0.018) 0.016 (0.011) -0.013 (0.009) -0.005 (0.012) 0.007 (0.006) -0.001 (0.003) 175 0.028 0.347 175 0.084 0.330 175 0.087 0.411 175 0.151 0.244 175 0.153 0.418 175 0.108 0.151 175 0.097 0.233 175 0.071 0.233 175 0.062 0.399 175 0.025 0.482 PPP (1) Islamist PKS (2) PBB (3) 0.014 (0.012) 0.014* (0.008) 175 0.071 0.265 175 0.064 0.234 Notes: This table reports estimates of equation (1) using a specification analogous to that in column 2 of Table II but for the 2014 Legislative Election and further disaggregating the three groups of parties reported in that table. We are missing data for 16 districts that did not post complete voting outcomes to the Indonesian National Electoral Commission website from which we obtained the data (via Nicholas Kuipers). * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. A.6 Individual-Level Voting Outcomes TABLE A.7 Voting at the Individual Level in the 2004 Legislative Election (1) piety index Number of Individuals Number of Districts Dependent Variable Mean (3) 0.172*** (0.047) 0.075** (0.030) 0.151*** (0.039) 0.079** (0.031) 1,398 128 0.122 1,398 128 0.122 1,398 128 0.181 0.132*** (0.037) expropriation intensity 1,398 128 0.122 Voted for . . . Party Moderate Islamic (4) (5) (6) Islamist (2) (7) -0.286*** (0.057) -0.145** (0.067) 0.151*** (0.045) -0.142** (0.064) 1,398 128 0.181 1,398 128 0.181 1,398 128 0.670 Secular (8) (9) 0.045 (0.070) -0.292*** (0.056) 0.039 (0.066) 1,398 128 0.670 1,398 128 0.670 Notes: This table reports estimates of equation (1) for voting outcomes in the 2004 legislative election as reported in the Pepinsky, Liddle and Mujani (2018) survey data. The dependent variable is an indicator equal to one if the individual voted for an Islamist party (columns 1–3), a moderate Islamic party (columns 4–6), and a secular party (columns 7–9). See the notes to Table II for these party classifications. The piety index is based on the summary index used in column 12 of Table V, capturing a range of obligatory and non-obligatory Islamic practices. The specifications are otherwise identical to those using the Pepinsky, Liddle and Mujani (2018) data in Tables IV and V. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. A.7 A.7.1 Further Robustness Checks Alternative Specifications We elaborate here on the robustness checks on our core Tables I, II, and III as described in Section VI.A.. In the tables described below, we report estimates from our baseline specification (1) in the top row, and we describe alternative specification choices in the first column. Each subsequent column reports the difference-in-discontinuity term estimated from a different regression. In Tables A.8, A.9, and A.10, we check the robustness of our main results to different versions of equation (1) where we include: baseline agricultural controls from the 1963 Agricultural Census (number of males, females, and farms, total irrigated land area, and total dry land area); baseline political controls (the vote share received by Islamist parties and by the Communist Party in the 1950s elections, and violent events associated with the Darul Islam rebellion); baseline Islamic organizations (the number of mosques and pesantren in the district by 1920); and province fixed effects. In Table A.8, where regressions are estimated at the village level, we also show robustness to including village-level geographic controls (altitude, beach location, distance to the nearest subdistrict capital and district capital). All controls are interacted with the Above400 dummy and the number of M EH. We also show robustness of our core results to excluding Sulawesi and Sumatra, motivated by concerns for the quality of the 1963 Agricultural Census data. The subsequent two rows of Tables A.8, A.9, and A.10 test the robustness of our main results to the assumption that landholdings follow a Pareto distribution. We do this by computing lower and upper bounds on the number of M EH in each district, and by estimating the difference-indiscontinuity at these bounds. For the upper bound, note that holdings between 5-9 ha are bounded above by the number of holdings above 5 ha, which is observed in the Census tabulations. We therefore estimate a specification where we use all holdings above 5 hectares in lieu of estimated holdings between 5–9 hectares. Our core results are robust (and sometimes stronger) when we use this actual measure of large landholdings. A plausible lower bound can be obtained from the following simple procedure. We first calculate the total amount of land held in holdings of size 5 ha and above as the difference between total agricultural land in the district and land held in holdings under 5 ha. Next, we compute the largest possible number of holdings with size 9 ha and above, which is simply equal to the total land area calculated in the first step divided by 9. The difference between the observed number of holdings above 5 ha and the maximum number of holdings above 9 ha (set to zero if negative) provides a lower bound on the number of holdings between 59 ha. In Tables A.8, A.9, and A.10, we show that our core results are mostly robust to using this lower bound. The difference-in-discontinuity coefficient remains positive throughout but is sometimes imprecisely estimated. This is not surprising since marginal expropriable holdings increase the amount of identifying variation we can exploit in each district. Finally, Tables A.8, A.9, and A.10 report a simple placebo check where we look for a discontinuous jump in outcomes at 500 people/km2 , which was not a relevant cutoff in the land reform. Here we interact a dummy for districts above 500 people/km2 (instead of 400 in our main specification) with the number of holdings between 5–9 hectares. As expected, we do not find any evidence that this interaction is associated with the contemporary prevalence of waqf lands, waqf -endowed institutions, voting for Islamist parties, and Islamist influence on local governance. TABLE A.8 Waqf Land and Waqf -Endowed Institutions: Alternative Specifications hectares (1) Waqf Land % total % zoned (2) (3) Waqf -Endowed Institutions mosques pesantren madrasas (4) (5) (6) Baseline specification (Table 1) 0.207** (0.097) 0.251** (0.107) 3.286** (1.483) 2.956*** (0.985) 0.532** (0.264) 1.024** (0.410) Controls: Agriculture 0.177* (0.102) 0.274** (0.114) 3.032** (1.442) 3.385*** (0.768) 0.420* (0.214) 1.077** (0.499) Controls: Geography 0.196** (0.098) 0.250** (0.106) 3.329** (1.480) 3.234*** (0.928) 0.539** (0.259) 1.041** (0.412) Controls: 1955-57 Vote Shares + Darul Islam 0.154* (0.082) 0.168* (0.088) 2.958** (1.475) 2.561*** (0.797) 0.261 (0.228) 0.687** (0.288) Controls: Mosques & Pesantren in 1920 0.124 (0.108) 0.279*** (0.099) 4.718** (1.921) 2.548*** (0.870) 0.422* (0.235) 1.117** (0.474) Province Fixed Effects 0.204** (0.083) 0.265** (0.105) 2.537** (1.228) 1.490 (1.053) 0.366 (0.252) 0.712** (0.333) No Sulawesi/Sumatra 0.199** (0.097) 0.206* (0.107) 2.014** (0.965) 3.224*** (1.141) 0.575** (0.287) 1.122** (0.465) Lower bound on MEH 0.092 (0.089) 0.078 (0.098) 1.695 (1.133) 2.735** (1.073) 0.164 (0.217) 0.841** (0.424) Upper bound on MEH (all 5+ holdings) 0.076** (0.034) 0.096*** (0.036) 1.002** (0.449) 0.952*** (0.332) 0.146* (0.087) 0.416*** (0.159) Placebo: expropriation at 500 ppl/km2 0.027 (0.063) -0.043 (0.065) 1.547 (1.834) 0.604 (1.677) 0.023 (0.239) 0.305 (0.285) Number of Villages Number of Districts Dependent Variable Mean R2 55200 191 0.848 0.033 55200 191 0.518 0.005 55200 191 6.127 0.044 48978 189 3.921 0.234 48978 189 0.503 0.165 48978 189 0.787 0.195 Notes: This table reports estimates from variants of equation (1) using the same outcomes as in Table I. Geographic controls include altitude, coastal location, distance to the nearest subdistrict capital and the district capital. Agricultural controls include number of males, number of females, number of farms, total irrigated land area, and total dry land area measured in the 1963 Agricultural Census. Vote shares refer to Islamic parties and the Communist Party in the 1955/57 legislative elections. Controls are interacted with the Above400 dummy and the number of marginal expropriable holdings (M EH). The R2 reported in the bottom panel corresponds to the baseline specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE A.9 Islamist Party Support: Alternative Specifications Village 1999 Top 3 Finish (1) District Level 1999 1999-2014 Vote Share Vote Share (2) (3) Islamist Parties Baseline specification (Table 2) 0.158* (0.083) 0.049** (0.024) 0.044** (0.020) Controls: Agriculture 0.197** (0.094) 0.056* (0.030) 0.053** (0.024) Controls: 1955-57 Vote Shares + Darul Islam 0.112* (0.061) 0.023 (0.017) 0.027** (0.012) Controls: Mosques & Pesantren by 1920 0.128 (0.094) 0.039 (0.030) 0.044* (0.025) Province Fixed Effects 0.090 (0.092) 0.034 (0.028) 0.031 (0.023) No Sulawesi/Sumatra 0.211** (0.094) 0.057** (0.027) 0.049** (0.022) Lower bound on MEH 0.177** (0.077) 0.041* (0.024) 0.034* (0.018) Upper bound on MEH (all 5+ holdings) 0.050* (0.029) 0.015* (0.009) 0.014* (0.007) Placebo: expropriation at 500 ppl/km2 -0.015 (0.083) 0.001 (0.027) -0.007 (0.020) Number of Villages Number of Districts Dependent Variable Mean R2 55200 191 0.516 0.051 – 189 0.136 0.205 – 191 0.154 0.222 Notes: This table reports estimates from variants of equation (1) using the same outcomes as in Panel A of Table II. The estimate in column (3) pools across all four quinquennial elections. Agricultural controls include number of males, females, and farms, total irrigated land area, and total dry land area. Vote shares refer to Islamic parties and the Communist Party in the 1955/57 legislative elections. Controls are interacted with the Above400 dummy and the number of marginal expropriable holdings (M EH). The R2 reported in the bottom panel corresponds to the baseline specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE A.10 Islamism and Local Governance: Alternative Specifications Local Relig. Gov. Officials (1) Sharia Regulations (2) Sharia Banking (3) Baseline specification (Table 3) 0.614** (0.266) 0.600** (0.289) 0.346 (0.237) 1.153* (0.603) 0.825** (0.397) 0.045 (0.424) 0.837** (0.349) 1.119*** (0.357) Controls: Agriculture 0.495* (0.269) 0.649 (0.406) 0.803** (0.316) 1.152** (0.493) 0.660* (0.362) 0.030 (0.295) 0.364 (0.527) 0.878 (0.602) Controls: 1955-57 Vote Shares + Darul Islam 0.504* (0.256) 0.388 (0.278) 0.399* (0.239) 0.924** (0.441) 0.909 (0.580) -0.136 (0.425) 0.794* (0.418) 1.063*** (0.371) Controls: Mosques & Pesantren by 1920 0.534* (0.306) 0.311 (0.294) 0.402* (0.206) 1.131** (0.489) 0.743* (0.395) -0.099 (0.657) 1.042*** (0.350) 1.272*** (0.403) Province Fixed Effects 0.098 (0.243) 0.414** (0.180) 0.085 (0.243) 0.907 (0.755) 0.237 (0.409) -0.763 (0.590) 0.881** (0.358) 1.178*** (0.369) No Sulawesi/Sumatra 0.541* (0.281) 0.631* (0.321) 0.332 (0.261) 1.669*** (0.530) 0.788 (0.475) -0.033 (0.497) 0.972*** (0.354) 1.302*** (0.341) Lower bound on expropriable holdings 0.347 (0.285) 0.361 (0.288) 0.470* (0.257) 0.505 (0.580) 0.495 (0.323) -0.145 (0.445) 0.561 (0.467) 0.773* (0.435) Upper bound on expropriable holdings 0.239*** (0.091) 0.216** (0.106) 0.141 (0.089) 0.314 (0.194) 0.290* (0.148) 0.010 (0.139) 0.260** (0.130) 0.341** (0.148) Placebo: expropriation at 500 ppl/km2 0.122 (0.318) -0.101 (0.472) -0.197 (0.254) 0.054 (0.523) -0.017 (0.314) 0.732 (0.519) -0.860* (0.512) -1.308 (1.057) 191 0.289 191 0.136 189 0.435 80 0.436 80 0.688 80 0.451 114 0.268 114 0.304 Number of Districts R2 Islamic Court Cases re: Waqf Inherit. Divorce (4) (5) (6) Islamist Vigilante Activity Incidents Casualties (7) (8) Notes: This table reports estimates from variants of equation (1) using the same standardized outcomes as in Table III. Agricultural controls include number of males, females, and farms, total irrigated land area, and total dry land area. Vote shares refer to Islamic parties and the Communist Party in the 1955/57 legislative elections. Controls are interacted with the Above400 dummy and the number of marginal expropriable holdings (M EH). The R2 reported in the bottom panel corresponds to the baseline specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. A.7.2 RD Robustness Checks In Tables A.11, A.12, and A.13, we show that our main results are mostly robust to alternative parametrizations of the RD specification. For comparison, the top row of each of these tables reports estimates from the baseline specification in Tables I, II, and III, respectively. We then report the following checks. First, we vary the degree of the polynomial in the running variable (1960 population density) in equation (1). Our main results are robust to alternative polynomials (linear, quadratic, and quartic) besides the cubic specification used in the baseline estimates. Second, we vary the bandwidth around the population density cutoff of 400 people/km2 , with bandwidths ranging from 100 to 300. The difference-in-discontinuity estimate remains positive and significant in most of these specifications. At the lower end, we see the limits of the identifying variation afforded by the policy as we are left with too few districts (46 out of 191) to conduct statistically well-powered difference-in-discontinuity regressions with the associated interactions. TABLE A.11 Waqf Land and Waqf -Endowed Institutions: RD Specification Checks hectares (1) Waqf Land % total % zoned (2) (3) Waqf -Endowed Institutions mosques pesantren madrasas (4) (5) (6) Baseline specification (Table 1) 0.207** (0.097) 0.251** (0.107) 3.286** (1.483) 2.956*** (0.985) 0.532** (0.264) 1.024** (0.410) Linear in density 0.155*** (0.052) 0.143*** (0.050) 0.822 (1.008) 0.865 (0.732) 0.223 (0.145) 0.546** (0.214) Quadratic in density 0.109 (0.071) 0.195** (0.075) 0.230 (1.314) 1.733** (0.826) 0.305 (0.199) 0.740** (0.366) Quartic in density 0.128 (0.078) 0.155 (0.099) 1.060 (1.205) 2.213** (0.922) 0.340 (0.286) 0.935** (0.376) Local linear, Bandwidth: 100 0.224* (0.113) 0.152 (0.159) 2.310 (1.635) 1.262 (1.394) 0.788 (0.590) 0.997** (0.429) Villages Districts 12,304 46 12,304 46 12,304 46 12,089 46 12,089 46 12,089 46 Local linear, Bandwidth: 200 0.223** (0.090) 0.182* (0.104) 1.969** (0.970) 3.228*** (1.092) 0.742** (0.343) 1.206*** (0.428) Villages Districts 18,917 71 18,917 71 18,917 71 18,406 71 18,406 71 18,406 71 Local linear, Bandwidth: 300 0.197** (0.080) 0.277*** (0.091) 0.379 (1.195) 2.308*** (0.774) 0.522* (0.297) 0.939*** (0.345) Villages Districts 24,563 97 24,563 97 24,563 97 23,739 96 23,739 96 23,739 96 Number of Villages Number of Districts Dependent Variable Mean R2 55200 191 0.848 0.033 55200 191 0.518 0.005 55200 191 6.127 0.044 48978 189 3.921 0.234 48978 189 0.503 0.165 48978 189 0.787 0.195 Notes: This table reports estimates from variants of equation (1) using the same outcomes as in Table I. Each row reports results from a different specification. In the second to fourth row, we vary the degree of the polynomial in the running variable (1960 population density) to alternative polynomials (linear, quadratic, and quartic) besides the cubic specification used in the baseline estimates. In subsequent rows, we vary the bandwidth around the 400 people/km2 cutoff, with bandwidths ranging from 100 to 300. The bottom panel reports regression statistics from the baseline specification. See Section A.7.2 for a detailed description of each specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE A.12 Islamist Party Support: RD Specification Checks Village 1999 Top 3 Finish (1) District Level 1999 1999-2014 Vote Share Avg. Vote Share (2) (3) Islamist Parties Baseline specification (Table 2) 0.158* (0.083) 0.049** (0.024) 0.044** (0.020) Linear in density 0.114** (0.046) 0.030** (0.014) 0.035*** (0.011) Quadratic in density 0.099 (0.065) 0.029 (0.018) 0.036** (0.014) Quartic in density 0.179* (0.092) 0.051** (0.024) 0.045** (0.019) Local linear, Bandwidth: 100 0.088 (0.143) 0.056 (0.049) 0.054 (0.037) Observations Districts 12,304 46 46 46 179 46 Local linear, Bandwidth: 200 0.143* (0.086) 0.062** (0.027) 0.054** (0.021) Observations Districts 18,917 71 71 71 277 71 Local linear, Bandwidth: 300 0.116 (0.073) 0.039* (0.023) 0.043** (0.018) Observations Districts 24,563 97 97 97 380 97 Number of Villages Number of Districts Dependent Variable Mean R2 55200 191 0.516 0.051 – 189 0.136 0.205 – 191 0.154 0.222 Notes: This table reports estimates from variants of equation (1). Islamist parties are defined as in Table II. Each row reports results from a different specification. In the second to fourth row, we vary the degree of the polynomial in the running variable (1960 population density) to alternative polynomials (linear, quadratic, and quartic) besides the cubic specification used in the baseline estimates. In subsequent rows, we vary the bandwidth around the 400 people/km2 cutoff, with bandwidths ranging from 100 to 300. The bottom panel reports regression statistics from the baseline specification. See Section A.7.2 for a detailed description of each specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE A.13 Islamism and Local Governance: RD Specification Checks Local Relig. Gov. Officials (1) Sharia Regulations (2) Sharia Banking (3) Baseline specification (Table 3) 0.614** (0.266) 0.600** (0.289) 0.346 (0.237) 1.153* (0.603) 0.825** (0.397) 0.045 (0.424) 0.837** (0.349) 1.119*** (0.357) Linear in density 0.283** (0.134) 0.269 (0.190) -0.217 (0.136) 0.516* (0.296) 0.218 (0.305) 0.015 (0.264) 0.472** (0.191) 0.480*** (0.175) Quadratic in density 0.315* (0.179) 0.445* (0.228) 0.066 (0.172) 0.761* (0.432) 0.548 (0.501) -0.015 (0.343) 0.513* (0.269) 0.718** (0.291) Quartic in density 0.165 (0.264) 0.439 (0.318) -0.087 (0.264) 1.047* (0.612) 0.665 (0.501) -0.044 (0.380) 0.481 (0.679) 0.599 (0.546) Local linear, Bandwidth: 100 0.478 (0.487) 0.915** (0.421) 0.228 (0.365) 2.785*** (0.657) 0.531 (0.565) -0.013 (0.416) 0.815 (0.889) 1.288* (0.725) 46 46 46 22 22 22 33 33 0.555* (0.290) 0.891** (0.342) 0.355 (0.251) 1.193 (0.712) 0.684 (0.479) -0.355 (0.439) 0.679 (0.547) 0.935* (0.484) 71 71 71 37 37 37 47 47 0.235 (0.252) 0.510* (0.279) -0.019 (0.210) 0.973 (0.589) 0.664 (0.468) -0.278 (0.342) 0.519 (0.346) 0.577* (0.325) 97 97 96 47 47 47 62 62 191 0.289 191 0.136 189 0.435 80 0.436 80 0.688 80 0.451 114 0.268 114 0.304 Districts Local linear, Bandwidth: 200 Districts Local linear, Bandwidth: 300 Districts Number of Districts R2 Islamic Court Cases re: Waqf Inherit. Divorce (4) (5) (6) Islamist Vigilante Activity Incidents Casualties (7) (8) Notes: This table reports estimates fron variants of equation (1) using the same standardized outcomes as in Table III. In the second to fourth row, we vary the degree of the polynomial in the running variable (1960 population density) to alternative polynomials (linear, quadratic, and quartic) besides the cubic specification used in the baseline estimates. In subsequent rows, we vary the bandwidth around the 400 people/km2 cutoff, with bandwidths ranging from 100 to 300. The bottom panel reports regression statistics from the baseline specification. See Section A.7.2 for a detailed description of each specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. A.7.3 Other Land Reform Policy Cutoffs In Tables A.14, A.15, and A.16, we probe the role of other cutoffs used in the 1960 land reform. While the core of our analysis focuses on the 400 people/km2 cutoff, nominally the reform used two other cutoffs at 50 and 250 people/km2 , as discussed in Sections II.A. and IV.B.. We proceed in two ways. First, we estimate a version of equation (1) where we “pool” all three cutoffs and each district is matched to the nearest cutoff. The specification takes the following form: = α + γ0 Abovej + γ1 M EHj + β(Abovej × M EHj ) + yij 3 X cdj + islandj (A.1) d=1 + 3 X cdj g(densityj )[δ0d + δ1d Abovej + δ2d M EHj + δ3d (Abovej × M EHj )] + εij d=1 where c = 1, 2, 3 is a set of indicators equal to 1 if threshold d is the nearest threshold (d = 1, 2, 3 denote the cutoffs at 50, 250, and 400, respectively), and Abovej is equal to 1 if district j is above the population density threshold of the nearest cutoff. In this case, marginal expropriable holdings (M EH) are defined for each cutoff as follows: holdings between 5–9 hectares at the 400 cutoff, holdings between 7.5–12 hectares at the 250 cutoff, and holdings between 10–20 hectares at the 50 cutoff. The difference-in-discontinuity estimate remains positive in all tables, and is statistically significant across most columns of Table A.14 and A.16. However, the fact that considerably less expropriation actually occurred at these lower cutoffs implies that the results delivered in this specification are much more imprecise. Second, we look for discontinuities in outcomes at each cutoff. The specification in this case is: yij = α + γ050 Above50j + γ150 M EH50j + β 50 (Above50j × M EH50j ) + γ0250 Above250j γ0400 Above400j + + 3 X + + γ1250 M EH250j γ1400 M EH400j +β 250 (Above250j × M EH250j ) +β 400 (Above400j × M EH400j ) (A.2) g(Ddj ) [δ0d + δ1d Above{d}j + δ2d M EH{d}j + δ3d (Above{d}j × M EH{d}j )] d=1 + islandj + εij , where Ddj represent the distance to cutoff d for district j and d = 1, 2, 3 denote the cutoffs at 50, 250, and 400, respectively. Here, to improve precision we use non-overlapping intervals of MEH at each cutoff, namely 5–9 ha at 400, 9–12 ha at 250, and 12–20 ha at 50 people/km2 . The bottom panel of Tables A.14, A.15, and A.16 reports the corresponding estimates of β 50 , β 250 , and β 400 . These coefficients are jointly significant when looking at land under waqf (Table A.14 column 1). Again, the results are less precise than when focusing solely on the 400 cutoff where expropriation intensity was most binding. As discussed in Section VI.B., these patterns are in line with the historical record of policy implementation, which suggests that by the time the land reform was halted in the mid1960s it had mostly not reached regions of the country where the other cutoffs were most pervasive. TABLE A.14 Waqf Land and Waqf -Endowed Institutions: Other BAL Cutoffs Waqf Land hectares % total % zoned (1) (2) (3) Waqf -Endowed Institutions mosques pesantren madrasas (4) (5) (6) Baseline specification (Table 1) 0.207** (0.097) 0.251** (0.107) 3.286** (1.483) 2.956*** (0.985) 0.532** (0.264) 1.024** (0.410) Pooling all cutoffs 0.173** (0.075) 0.150*** (0.052) 1.104 (1.144) 1.316** (0.602) 0.141 (0.154) 0.266 (0.164) Expropriation at 50 ppl/km2 0.050 (0.049) 0.004 (0.711) 0.194** (0.088) -0.023 (0.028) 0.201 (0.732) 0.325** (0.128) 0.968 (0.799) -5.583 (8.488) 1.317 (0.976) 0.003 (0.244) 0.050 (7.205) 1.948** (0.872) 0.027 (0.042) -0.020 (1.087) 0.354 (0.260) -0.065 (0.044) -4.054*** (1.298) 0.690** (0.301) 55200 191 0.848 0.033 55200 191 0.518 0.005 55200 191 6.127 0.044 48978 189 3.921 0.234 48978 189 0.503 0.165 48978 189 0.787 0.195 Expropriation at 250 ppl/km2 Expropriation at 400 ppl/km2 Number of Villages Number of Districts Dependent Variable Mean R2 Notes: This table examines the same outcomes as in Table I. We first report estimates from equation (A.1) where each district is matched to the nearest cutoff. In the bottom panel, we report estimates of β 50 , β 250 , and β 400 from equation (A.2). See Section A.7.3 for a detailed description of each specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE A.15 Islamist Party Support in Legislative Elections: Other BAL Cutoffs Village 1999 Top 3 Finish (1) District Level 1999 1999-2014 Vote Share Avg. Vote Share (2) (3) Islamist Parties Baseline specification (Table 2) 0.158* (0.083) 0.049** (0.024) 0.044** (0.020) Pooling all cutoffs 0.011 (0.059) 0.025 (0.022) 0.024 (0.020) Expropriation at 50 ppl/km2 -0.033 (0.035) 0.197 (0.576) 0.168* (0.097) -0.002 (0.009) 0.025 (0.112) 0.055* (0.028) -0.005 (0.006) -0.088 (0.077) 0.054** (0.025) 55200 191 0.516 0.051 – 189 0.136 0.205 – 191 0.154 0.222 Expropriation at 250 ppl/km2 Expropriation at 400 ppl/km2 Number of Villages Number of Districts Dependent Variable Mean R2 Notes: This table examines the same outcomes as in Table II. We first report estimates from equation (A.1) where each district is matched to the nearest cutoff. In the bottom panel, we report estimates of β 50 , β 250 , and β 400 from equation (A.2). See Section A.7.3 for a detailed description of each specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors clustered by 1960 district. TABLE A.16 Islamism and Local Governance: Other BAL Cutoffs Local Relig. Gov. Officials (1) Sharia Regulations (2) Sharia Banking (3) Baseline specification (Table 3) 0.614** (0.266) 0.600** (0.289) 0.346 (0.237) 1.153* (0.603) 0.825** (0.397) 0.045 (0.424) 0.837** (0.349) 1.119*** (0.357) Pooling all cutoffs 0.215 (0.182) 0.073 (0.238) 0.131 (0.180) 0.851 (0.680) 1.066* (0.554) 0.522 (0.445) 1.025*** (0.315) 1.487*** (0.213) Expropriation at 50 ppl/km2 -0.073 (0.100) -0.606 (1.216) 0.283 (0.288) 0.077 (0.126) -2.745* (1.598) 0.630* (0.339) 0.036 (0.104) -2.384 (1.544) 0.704* (0.370) -0.365 (1.021) 5.404* (3.097) 2.112* (1.214) 0.405 (0.980) 5.408* (3.129) -1.329 (1.377) -1.298* (0.705) 9.320*** (2.907) -0.903 (0.622) -0.331 (0.216) 1.833 (2.164) 1.194** (0.599) -0.093 (0.131) 2.115 (1.579) 1.581*** (0.414) 191 0.289 191 0.136 189 0.435 80 0.436 80 0.688 80 0.451 114 0.268 114 0.304 Expropriation at 250 ppl/km2 Expropriation at 400 ppl/km2 Number of Districts R2 Islamic Court Cases re: Waqf Inherit. Divorce (4) (5) (6) Islamist Vigilante Activity Incidents Casualties (7) (8) Notes: This table examines the same standardized outcomes as in Table III. All regressions are run at the district level. We first report estimates from equation (A.1) where each district is matched to the nearest cutoff. In the bottom panel, we report estimates of β 50 , β 250 , and β 400 from equation (A.2). See Section A.7.3 for a detailed description of each specification. * p<0.1, ** p<0.05, *** p<0.01. Robust standard errors. B Data Sources and Construction We describe here the key variables and data sources used in the paper. Historic Population Density We measure historic district-level population density using tabulations from the 1961 Population Census available in island-level hard-bound report, Sensus Penduduk 1961 in the Central Bureau of Statistics (Badan Pusat Statistik or BPS) library in Jakarta. The specific table that we use is titled “Sensus Penduduk Republik Indonesia 1961. Angka Sementara Penduduk Indonesia Menurut Jenis Kelamin. Per Daerah Tingkat II.” We use ArcGIS to construct the area of historical 1961 districts by amalgamating later districts back to their 1961 boundaries. Using the resulting population density, we classify districts into the four categories discussed in the paper. Historic Landholdings We measure the historic landholdings distribution and number of large holders using the 1963 Agricultural Census conducted for the purposes of assessing landholdings in order to implement the stipulated reform. We digitized district-level tabulations available in a report by BPS, Sensus Pertanian 1963, with the table “Number of farms by size of area” (“Banjaknja usaha pertanian rakjat menurut golongan luas tanah”). While the raw data from the Agricultural Census are no longer available, these tabulations provide sufficient granularity to estimate (with noise) the number of large landholders in each district that would be affected by the land reform. For each district, we observe the number of holders with 0.1–0.49 hectares (ha), 0.5–0.99 ha, 1–1.49 ha, 1.5–1.99 ha, 2–2.99 ha, 3–3.99 ha, 4–4.99 ha, and greater than or equal to 5 ha. As detailed in Section IV.A., assuming that landholdings L follow a Pareto distribution with probability density function λLλ L−λ−1 , we can estimate the number of landholders in different bins above 5 ha. Given the Pareto formulation, the distribution parameter λ holds over all truncated segments of the distribution. As such, we can use the bins below 5 ha to recover the shape of the distribution above 5 ha where we do not know the number of landholders in each affected size bin subject to redistribution based on the four density cutoffs. We proceed in two steps. First, we estimate the Pareto shape parameter, λ, for each district using b to back out a maximum likelihood procedure for landholdings L ∈ [0.1, 5) ha. Second, we use λ the number of landholders with 5–9 ha, 9–12 ha, 12–20 ha, and > 20 ha, following the stipulated cutoffs. Concretely, we multiply the number N5 of landholders in the ≥ 5 ha bin by the share of the district’s total landholding distribution in the given range based on the Pareto cumulative
λb distribution function (e.g., for 5–9 ha, this is given by [1 − 59 ] × N5 ). Although we are not able to estimate these marginal landholdings separately by wetland and dryland, we are able to control for the total number of farms, total wetland area (ha), and total dryland area (ha) using district-level tabulations elsewhere in the Sensus Pertanian 1963 report under the table titled “Farm area, average size of Farm and Paddy area” (“Luas tanah Pertanian Rakjat dan luas panenan padi”). We measure the post-land reform distribution of landholdings using the 1980 and 1990 Population Censuses as well as the 1985 Intercensal Survey (Supas). These are the first three Census/Intercensus rounds that include measures of total landholdings owned by each household. We use the samples available on IPUMS International and estimate the Pareto landholdings dispersion parameter λ, for all landholdings above 0.1 ha. These estimates are at the district-level, at which the pop- ulation summary statistics are representative, and hence directly comparable with the tabulations from 1963. Contemporary Landholdings, Including Waqf We measure contemporary landholdings using the 2003 Agricultural Census. We use this universal census data to estimate Pareto shape parameters, λ, for every village and also to construct a measure capturing the share of all households with greater than 0.1 ha over which λ is estimated. See Bazzi (2017) for details on the data and estimation procedure, which differs from that used for the coarser, binned 1963 Agricultural Census data. We use the 2003 administrative village census (Potensi Desa or Podes) to measure the total land area under waqf status overall, as a share of total land, and as a share of zoned land. Contemporary Islamic Institutions We use the Podes 2003 and 2008 data to construct village-level measures of Islamic institutions. The 2003 data include the total number of Islamic schools (pesantren and madrasa), and the 2008 data disaggregate the two. Both rounds include the number of mosques. We also draw upon administrative data from the Ministry of Religion to measure (i) the amount of waqf land allocated to new mosques and (ii) the number of students enrolled in pesantren, both by year and district of establishment. We scrape these administrative data from the web: (i) is from http://simas.kemenag.go.id/ and (ii) is from https://ditpdpontren.kemenag. go.id/pbsb/. The number of pesantren established by 1920 is winsorized at the 99th percentile due to data concerns over one district (Jember), where 71% of pesantren are implausibly reported as having been established in 1900. Electoral Outcomes We draw upon several sources to measure historic and contemporary electoral outcomes. First, we draw upon district-level vote shares by party from the national legislative elections in 1955, 1999, 2004, 2009, and 2014. These data were graciously shared with us by individuals that worked with Dwight King. Several districts are missing data for the 1955 elections. We therefore supplement the 1955 legislative election data with data from the 1957 district legislative elections that were held in select districts. There are still some historic districts with no voting data from the 1950s, and for these 20 districts, we impute the vote shares for neighboring districts so as to retain the largest possible sample of districts when including this control in robustness checks. We digitize the latter from raw electoral reports obtained from files shared with us by Donald Hindley. Second, we use the 2003 Podes, which records the top 1, 2, and 3 ranked parties at the village-level in the first post-Suharto legislative election held in 1999. Unfortunately, the vote shares themselves are not reported. We categorize parties based on conventions put forward in the political science literature on Indonesia, including numerous works by Dwight King and R. William Liddle as well as a seminal article by Baswedan (2004) aligning parties in the post-Suharto era along a spectrum of Islamist leanings. In 1955 and 1957, we define Islamist parties as Masyumi, the Indonesian Islamic Union Party (Partai Serikat Islam Indonesia or PSII), the Islamic Educators Association (Perhimpunan Tarbiyah Islamiyah or Perti), and Nahdlatul Ulama (NU). While the first post-Sukarno election in 1971 saw several Islamic parties (NU, PSII, Perti, and the Muslim Party of Indonesia; Masyumi was banned in 1961), thereafter the Suharto regime allowed only a single Islamic party in the United Development Party (Partai Persatuan Pembangunan or PPP). From 1999 onward, we follow Baswedan (2004) in classifying Islamic parties. We consider as moderate Islamic parties the National Mandate Party (Partai Amanat Nasional or PAN) and the National Awakening Party (Partai Kebangkitan Bangsa or PKB), both of which initially adopted the national ideology of Pancasila prior to the 1999 election when parties were allowed to choose whether or not to embrace this for the first time in the post-Suharto era. The PKB is the successor to a large part of the former NU political wing, which disbanded from the PPP in 1984. The Islamist parties include the PPP, the Prosperous Justice Party (Partai Keadilan Sejahtera or PKS), and the Crescent Star Party (Partai Bulan Bintang or PBB). All three parties rejected Pancasila, including the PPP which was forced to accept Pancasila during the Suharto era. In sum, the PPP, PBB, and PKS can be seen as traditional Islamist parties whereas the PKB and PAN are Islamic albeit inclusive and non-Islamist in their orientation. While the particular leanings of these parties change over time and until today, this rough breakdown lines up with most historical and contemporary accounts by political observers. There are numerous non-Islamic parties, nearly all of which are secular (except a few tiny Christian parties). We lump all of these parties into the residual, secular category. For a full elaboration, see Appendix A.5. In 1955/57, we also observe the Communist vote share. Religious Political Preferences, Piety, and Practice We use the Indonesia Family Life Survey (IFLS) rounds 4–5 in 2007 and 2014/15, respectively, to measure individual-level religious political preferences. These include the following questions: (i) In an election, having a candidate with the same religion as yours makes it [...] to vote for him/her. (... is a 1 to 5 scale ranging from very likely to very unlikely; and (ii) In an election, if the candidates have the same religion as yours, how important is the religiosity of a candidate in influencing your decision to vote for him/her? A more religious candidate make [...] to vote for him/her. (... is the same scale as (iv)). The IFLS also includes questions about tolerance towards other religious faiths living in one’s village, neighborhood, house, and family as well as building a house worship nearby. We take the mean of these five questions which range on a 1 to 4 scale from very happy with to very opposed. We also consider two other variables capturing religiosity (“Am I a very a religious person?”) and interreligious trust (“Do I trust members of other religious faiths less than those of my own?”). 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 district. The data include numerous questions we use in Tables IV, V, and A.7. This includes religious political preferences (e.g., how important is the religion and religiosity of the President of Indonesia, support specific sharia regulations) as well as a host of questions about Islamic practice (e.g., fasting, paying zakat). Sharia Regulations We use data from Buehler (2016), Appendix 1, pp. 215–220 on the number of Sharia regulations adopted by district between 1998 and 2013. We use the total number of regulations, inclusive of legislative and executive branch regulations. Islamic Microfinance We use the Podes 2017 data to measure the prevalence of Islamic microfinance institutions known as Baitul Maal wat Tamwil or BMT. These data are available at the village level. We aggregate to the district level in Table III and related robustness tables, in keeping with the unit of analysis for other outcomes in that table. Legislator Profiles In Table IV, we consider three measures capturing the religious appeal 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/. The first measure captures whether the candidate’s online campaign statement appeals to Islamic or Islamist themes. These include umma, dawah, Muslim, Islam, sharia, and jihad. The second measure captures whether the candidate’s official name listed on the ballot includes an honorific title (Haji or simply H) signaling their prior Hajj pilgrimage to Mecca. The third measure captures whether the candidate’s official name listed on the ballot includes an honorific title signaling their status as a relgiious scholar (Kyai or Kyai Haji or simply KH). Islamic Courts We scrape data from web portals for every district-level Islamic court that reports such information online through the Religious Courts Information System (SIPP). A complete list of these portals is available upon request. The data, which vary in years of coverage, include date of filing and type of case. Islamist Vigilante Activity We capture the number of incidents and casualties due to violent activities by the Islamic Defenders Front (FPI) using data from the National Violence Monitoring System (or SNPK by its Indonesian acronym). The event-based data include a textual description of the underlying media report, and we search for terms related to the FPI. We include counts over the entire period the data are available, beginning in 1998 through 2014. The SNPK do not cover all regions of the country and hence the more limited sample size for this analysis. Local Religious Bureaucracy We digitized tabulations of the number of different types of bureaucrats working in the district-level Ministry of Religious Affairs in 2018. These come from “Dalam Angka” reports available in pdf on the Ministry of Religious Affairs website. Historic Demographics We use the Sensus Penduduk 1961 report noted above to control for the total number of men and women in each district as of 1960 before the land reform. We use the 1971 Population Census to construct age-specific male-to-female sex ratios. The data come from IPUMS International, and we use the population weights to go from the sample constructed by IPUMS to the historic district-level total male and female population. We also construct district population growth between 1961 and 1971 using this data. We also use the 1930 Population Census to measure the number of ethnic Arab and Chinese in each historic district. These data come from digitizing 1930 Population Census tabulations from Dutch archives. Islamic Holy Sites We use Google maps to identify the latitude/longitude of holy sites for the nine “saints” of Indonesian Islam in Java known as Wali Songo (the “nine saints” in the Javanese language). These nine men are credited with being the vanguard for the advance of Islamic across Java historically. We then compute the great circle distance of each village to the nearest holy site. Islamist Insurgency in the 1950s We use the book by van Dijk (1981) on the Darul Islam rebellion to hand-code measures of the Islamist rebellion that took place across Indonesia in the 1950s. We construct indicators for whether any violence took place in a given district. Contemporary Demographics We use the universal coverage 2000 Population Census to capture mean years of schooling and the share with different levels of education. Contemporary Agricultural Production, Development, and Public Goods We use the triennial Podes to construct several village-level proxies for development and public goods. Using the 2003 round, we construct the price-weighted agricultural output and total agricultural productivity (with price weights coming the FAO, see Bazzi et al. (2016) for details). We also measure the total number of capital machines for farming in each village in 2003. We construct an index of locally provided health and infrastructure public goods using all six Podes rounds from 1999 to 2014. The infrastructure index is based on that used in Martinez-Bravo (2017). Following Henderson et al. (2012), we capture a summary measure development based on the share of the village with any nightlights as observed from NOAA satellites in 2000. Geographic Controls We use the Podes 2003 data to construct the following geographic controls: indicators for whether the village is located on a hill or on a beach, the altitude in meters, and the distance to subdistrict and district capitals in kilometers.