Long-term and Intergenerational Effects of Education: Evidence from - - PowerPoint PPT Presentation

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Long-term and Intergenerational Effects of Education: Evidence from School Construction in Indonesia

Richard Akresh§, Daniel Halim¥, Marieke Kleemans§

§University of Illinois, Urbana-Champaign ¥World Bank

August 2019

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Motivation

• Governments in developing countries spend 1 trillion dollars

annually on education, households spend hundreds of billions more (Glewwe and Muralidharan, 2016)

• Macroeconomic growth models stressing the importance of

schooling for economic development

• Challenge in measuring causal effects of schooling
• Direction of causality and effects differ across identification

strategies (e.g. Bills and Klenow, 2000, Foster and Rosenzweig, 1996)

• Major strides forward with randomized experiments but

sometimes difficult to generalize and often focus on short-run effects only (McEwan, 2015)

• Questions remain whether effects persist or fade over time

(Evans and Ngatia, 2018, Blattman et al, 2018)

• Do effects spillover to the next generation? Little evidence on

intergenerational transmission of schooling in developing country setting

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

This paper

• Research Question: What are the long-term and

intergenerational effects of additional schooling as a child?

• School construction program in 1970s Indonesia provides

natural experiment (Duflo, 2001)

• Calculate wage return to education for working age male in

1995

• Exploit variation across districts in the number of schools built

and across birth cohorts in their exposure to the new schools

• Use 2016 nationally representative household survey

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

This paper

• Study effects of education on a range of outcomes, many of

which not previously studied

• Persistence of effects 43 years after the program using 2016

nationally representative cross-sectional data

• Study intergenerational effects of children whose parents were

exposed to the program

• Gender and marriage market dynamics
• For school construction that has large up-front costs and

benefits dispersed over time, can perform a detailed cost-benefit analysis

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Preview of Results

• First generation
• Large effects on education
• Men more likely to be formal workers, work outside agriculture,

and migrate

• Women more likely to migrate and have fewer children
• Households: improved living standards and pay more taxes
• Second generation
• Parents transmit effects to next generation: particularly large

increases in secondary and tertiary education

• Mother’s effect larger than father’s
• Effect on daughters seems larger than sons
• Policy implications: School construction pays for itself in

higher government taxes

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Institutional Context

• Indonesia is the 4th most populous country in the world, 261

million people

• 7th lagest economy in the world in terms of total GDP at PPP
• GPD per capita at PPP is $12,432, and 5.0% GDP growth

rate in 2016

• In 1972, enrollment rates were low in Indonesia at 71% among

primary school-aged children, large disparities across regions

• Average adult in 65 out of 290 districts did not complete

primary education (Census 1971)

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Institutional Context

• Between 1973 and 1979, Indonesia built 61,800 INPRES

primary schools with the goal to reduce regional disparities

• On average, program added over 200 schools per district or

two schools for every 1,000 children of primary school age

• Duflo (2001) finds gains in educational attainment and

short-term wage effects in 1995 for men exposed to program.

• Program also featured in Ashraf et al, 2018, Karachiwalla and

Palloni, 2019, Bharati et al, 2019, Mazumder et al, 2019, and Rohner and Saia, 2019.

Number of school constructed per 1,000 children between 1973 and 1979

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Data

• National Socioeconomic Survey, Susenas 2016
• Nationally representative household survey conducted by

Indonesian Central Bureau of Statistics

• Covers all 34 provinces and all 511 districts in Indonesia
• 109,847 households and 143,790 individuals interviewed for

cohorts we focus on

• District of birth data
• Rich dataset with detailed questions on education, labor force

participation (work, sector, hours worked), migration, expenditures, health, housing, assets, nutrition, taxes, demographics, and education/work of children

• Average age ≈ 50 for first generation; ≈ 15 for second

generation

Summary Statistics 1st Generation Summary Statistics 2nd Generation Susenas 2016

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Summary Statistics 1st Generation

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Identification Strategy

• Following Duflo (2001), use difference-in-differences in which

year and district of birth jointly determine exposure to school construction

• Children young enough in 1974 could benefit from the

program, but older cohorts would not

• More INPRES schools constructed in regions with lower

schooling enrollment at baseline

• Specifically, compare children aged 2-6 and 12-17 in 1974

across high and low intensity program regions

• Identifying assuming: the change in outcomes across birth

cohorts in the regions that built many schools would have been the same as the change across birth cohorts in the regions that did not build many schools

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Difference-in-differences

yijt = α + β Schoolj · Youngit + (XjB

′

t) γt + µj + δt + εijt

• yijt is the outcome of individual i born in district j in year t
• Schoolj is the number of INPRES schools constructed per

1,000 children between 1973 and 1979

• Youngit is a dummy if individual i is aged 2-6 in 1974
• Xj is a vector of district controls
• B

′

t is a vector of birth year dummies

• µj are district fixed effects
• δt are birth year fixed effects
• Standard errors clustered at the district level

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Difference-in-differences Second Generation

yijtca = α + β Schoolj · Youngit + (XjB

′

t) γt + µj + δt + θa + εijtca

• yijtca is the outcome of child c who is age a, born to parent i

who was born in district j in year t

• Schoolj is the number of INPRES schools constructed per

1,000 children in the father’s or mother’s birth district between 1973 and 1979

• Youngit is a dummy if the father or mother belongs to the

young cohort (ages 2-6 in 1974)

• θa are child c’s age fixed effects
• Standard errors clustered at the father’s or mother’s birth

district

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Parallel trends on men’s years of education

Parallel Trends All Variables

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Strategies to Address Large Number of Outcomes

• Creation of indexes for families of outcomes
• Index of standardised outcomes relative to old cohorts in

low-intensity areas amongst a family of outcomes (Kling, Liebman and Katz, 2007)

• Following Banerjee et al. (2015) and Ajayi and Ross (2017)
• Multiple hypothesis testing
• False discovery rate (FDR) to allow inference when many tests

are being conducted (Benjamini and Hochberg, 1995)

• FDR allows researcher to tolerate a certain number of tests to

be incorrectly discovered

• FDR adjusted q-value of 0.05 implies that 5% of significant

tests will result in false positives, compared with unadjusted p-value of 0.05 implying that 5% of all tests will result in false positives

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Effects on Indexes of Long-run Outcomes

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Effect of School Construction on Education

• Duflo (2001) difference-in-differences in Indonesia
• Handa (2002) difference-in-differences in Mozambique
• Alderman, Kim, Orazem (2003) RCT in Pakistan
• Burde and Linden (2013) in Afghanistan
• Kazianga, Levy, Linden, Sloan (2013) RD in Burkina Faso
• Khanna (2018) in India

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Long-run Effects on Education

Outcomes: Education Mean/SD Effect on: Men Women Men Women Years of schooling 8.022 7.105 0.268*** 0.234*** (4.230) (4.215) (0.047) [0.000] (0.042) [0.000] Completed Primary 0.813 0.727 0.026*** 0.041*** (0.390) (0.446) (0.006) [0.000] (0.006) [0.000] Completed Lower Secondary 0.385 0.312 0.023*** 0.008 (0.487) (0.463) (0.006) [0.000] (0.007) [0.422] Completed Upper Secondary 0.338 0.261 0.026*** 0.005 (0.473) (0.439) (0.006) [0.000] (0.006) [0.422] Completed Tertiary 0.095 0.077

• 0.001
• 0.003

(0.293) (0.267) (0.003) [0.741] (0.003) [0.422]

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Long-run Effects on Education

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Education, Labor Market Outcomes, and Migration

• Large literature on the causal relationship between education

and labor market outcomes (Heckman, Humprhies, and Veramendi, 2018, Glaeser and Lu, 2018)

• Long-term effects of secondary schools access (Duflo, Dupas,

Kremer, 2017)

• Large literature on education and migration has focused on

selection into migration

• Empirical evidence for Indonesia (Hicks, Kleemans, Li and

Miguel, 2018) and for developing countries in general (Young, 2013) shows positive selection from rural to urban areas and negative selection from urban to rural

• Little known about the causal relationship between education

and migration: does an exogenous shift in education lead to more or less migration?

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Long-run Effects on Labor Market Outcomes and Migration

Outcomes: Labor Market Mean/SD Effect on: and Migration Men Women Men Women

Work 0.948 0.638 0.006** 0.003 (0.223) (0.481) (0.003) [0.080] (0.005) [0.953] Work hours 40.981 36.227 0.258 0.157 (17.115) (18.792) (0.158) [0.101] (0.208) [0.953] Formal worker 0.327 0.236 0.011***

• 0.005

(0.469) (0.425) (0.004) [0.032] (0.005) [0.953] Non-agriculture sector 0.560 0.547 0.012*** 0.002 (0.496) (0.498) (0.005) [0.032] (0.005) [0.953] Service sector 0.364 0.459 0.010***

• 0.000

(0.481) (0.498) (0.004) [0.032] (0.006) [0.953] Migrant 0.273 0.245 0.007** 0.008** (0.445) (0.430) (0.003) [0.085] (0.003) [0.166] Local migration 0.108 0.106 0.005* 0.005** (0.310) (0.307) (0.003) [0.101] (0.003) [0.229]

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Long-run Effects on Living Standards

Outcomes: Living Standards Mean/SD Treatment Effect on: Men Women Men Women Total Expenditures (Rp10k) 391.649 375.616 0.021*** 0.032*** (352.495) (343.823) (0.007) [0.010] (0.007) [0.000] Food Expenditures (Rp10k) 194.443 184.222 0.014** 0.028*** (120.447) (121.110) (0.007) [0.036] (0.007) [0.000] Non-food (Rp10k) 197.206 191.393 0.027*** 0.039*** (271.884) (261.111) (0.008) [0.004] (0.008) [0.000] Non-food/Total 44.592 45.144 0.287*** 0.237*** (13.376) (13.751) (0.110) [0.024] (0.102) [0.021] Education Expenditures(Rp10k) 13.971 12.202 0.160** 0.193** (33.167) (30.346) (0.064) [0.024] (0.076) [0.011] Taxes (Rp10k) 4.749 4.552 0.078*** 0.123*** (11.433) (10.743) (0.017) [0.000] (0.019) [0.000] Tax Components Health and Nutrition Housing and Assets Welfare Programs

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Long-run Effects on Marriage

Outcome Mean/SD Effect on: Men Women Men Women Age of first marriage 25.219 (5.022) 20.888 (4.788) 0.058 (0.053) [0.867] 0.050 (0.059) [0.745] Children 0-14 0.910 (1.059) 0.559 (0.868)

• 0.012

(0.017) [0.867]

• 0.035**

(0.016) [0.559] Spouse’s characteristics Years of schooling 7.635 7.426 0.180*** 0.116*** (4.081) (4.192) (0.046) [0.000] (0.043) [0.028] Completed Primary 0.797 0.773 0.038*** 0.025*** (0.402) (0.419) (0.006) [0.000] (0.005) [0.000] Literate 0.939 0.944 0.027*** 0.016*** (0.239) (0.230) (0.006) [0.000] (0.004) [0.001] Literature Spouse’s Labor Market Characteristics

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Assortative Matching

Outcome Mean/SD Treatment Effect on: Men Women Men Women Spouse’s years of schooling 7.635 7.426 0.180*** 0.116*** (4.081) (4.192) (0.046) [0.000] (0.043) [0.028] Spouse’s years of schooling percentile 40.62 (28.49) 32.44 (29.29) 0.625** (0.303) [0.115] 0.448 (0.298) [0.133] Education gap 0.432 (3.707) 0.112 (3.600) 0.093* (0.049) [0.115]

• 0.078**

(0.038) [0.086] Education percentile gap 36.96 (31.18) 34.39 (30.08) 0.461 (0.402) [0.252]

• 0.738**

(0.322) [0.065]

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Education and Intergenerational Outcomes

• Large literature on intergenerational transmission of human

capital showing considerable persistence in economic

• pportunity as a source of increased inequality, perhaps

especially in low-income settings (review by Black and Devereux, 2011)

• In addition to primary focus on estimation of correlation and

elasticities, increased emphasis on causal relationship

• Using changes in education policies, compulsory schooling laws

(Chevalier (2004), Oreopoulos et al (2008), Currie and Moretti (2003)), and environmental shocks (Black et al., forthcoming)

• Very limited evidence from developing countries
• Large degree of intergenerational persistence in economic
• utcomes: will school construction improve intergenerational

mobility?

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Intergenerational Effects and Assortative Matching

• Following Behrman and Rosenzweig (2002) consider a linear

reduced-form equation of schooling of child i in family j: Sc

ij = δ1Sf j + δ2Sm j + γ1X f j + γ2X m j

+ εc

ij

(i = individual, j = family, c = child, f = father, m = mother)

• Emphasizes interrelationship of parent characterisctics

affecting child schooling: parent characteristics will be correlated with each other due to nonrandom matching in the marriage market

• Assortative matching relationships:

Sf = f (Sm, X m) and X f = g(Sm, X m)

• Empirical association between parent and child schooling

reflects not only direct relationship but also manifest itself through assortative matching

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Intergenerational Effects on Education

Outcomes: Child Education Mean/SD Effect of Exposure: Fathers Mothers Fathers on Children Mothers on Children Years of schooling 7.967 8.854 0.097*** 0.169*** (4.340) (4.278) (0.032) [0.014] (0.045) [0.001] Completed Primary 0.637 0.728 0.000 0.001 (0.481) (0.445) (0.002) [0.928] (0.003) [0.796] Completed Lower Secondary 0.413 0.504 0.006* 0.015*** (0.492) (0.500) (0.003) [0.171] (0.005) [0.006] Completed Upper Secondary 0.217 0.300 0.009** 0.014*** (0.412) (0.458) (0.004) [0.061] (0.005) [0.013] Completed Tertiary 0.041 0.064 0.004* 0.008** (0.198) (0.245) (0.002) [0.171] (0.003) [0.044] Age-for-grade 0.835 0.789 0.011*** 0.018*** (0.371) (0.408) (0.004) [0.030] (0.005) [0.002] Second Generation Estimating Equation Intergenerational Effects on Wellbeing

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Intergenerational Effects on Education

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Intergenerational Effects on Education by Gender

Extended cohorts Including non-coresidents

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Mediation Analysis

What observable characteristics change the coefficient on the intergenerational transmission of education?

Mediator: None Work/ Migration Living Standards Tax Housing & Assets Health Inv. Health Marriage All Panel A: Father Schools constructed * Young cohort 0.097*** 0.080** 0.082*** 0.084*** 0.082*** 0.086*** 0.096*** 0.056* 0.055** (0.032) (0.031) (0.031) (0.030) (0.028) (0.031) (0.032) (0.028) (0.026) Mediator 0.386*** 0.652*** 0.539*** 0.772*** 0.165*** 0.045*** 0.616*** (0.015) (0.017) (0.017) (0.019) (0.013) (0.011) (0.014) Panel B: Mother Schools constructed * Young cohort 0.169*** 0.166*** 0.130*** 0.138*** 0.133*** 0.156*** 0.168*** 0.137*** 0.111*** (0.045) (0.045) (0.044) (0.041) (0.039) (0.043) (0.045) (0.043) (0.039) Mediator 0.353*** 0.852*** 0.721*** 1.037*** 0.189*** 0.062*** 0.602*** (0.014) (0.019) (0.022) (0.019) (0.014) (0.015) (0.023)

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Mediation analysis

What spouse’s characteristics change the coefficient on the intergenerational transmission of education?

Spouse’s Mediator: None Years of Schooling Completed Primary Literate Formal worker Non-Agri sector Migrant No health complaint All Panel A: Father Schools constructed * Young cohort 0.097*** 0.069** 0.044 0.049* 0.068* 0.070** 0.090*** 0.092*** 0.020 (0.032) (0.030) (0.028) (0.027) (0.036) (0.034) (0.032) (0.032) (0.032) Mediator 0.146*** 1.381*** 1.798*** 0.627*** 0.940*** 0.330***

• 0.037*

(0.004) (0.042) (0.087) (0.031) (0.030) (0.027) (0.020) Panel B: Mother Schools constructed * Young cohort 0.169*** 0.126*** 0.115*** 0.125*** 0.130*** 0.126*** 0.152*** 0.155*** 0.091** (0.045) (0.045) (0.043) (0.044) (0.047) (0.044) (0.046) (0.046) (0.043) Mediator 0.176*** 1.567*** 1.906*** 0.677*** 1.007*** 0.452***

• 0.092***

(0.004) (0.040) (0.079) (0.035) (0.034) (0.032) (0.027)

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Cost Benefit Analysis

• Cost large and upfront: building schools, training teachers,

recurring teacher salary

• Benefits accrued over time
• Improved living standards: Compare the program’s costs with

the overall welfare benefits for the affected population: Does the economy benefit from improved living standards?

• Increased tax revenue: Can increases in government taxes

alone offset the government’s cost of the program?

• Large net benefits when include improved living standards.

Internal rates of return 13-21% and benefits outweigh costs within 17-30 years after schools built

• Across a range of different parameter estimates, school

construction leads to increased government tax revenues that

• ffset school construction costs within 25-40 years

Figure

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Cost Benefit Analysis

Alternative scenarios (1) (3) (5) (8) (10) Costs Schools construction 0.78 0.78 0.78 0.78 0.78 Teachers training 0.12 0.12 0.12 0.12 0.12 Teachers' salaries 1.65 2.95 2.95 6.10 12.19 Benefits Paid by cohorts born in 1968-1980 1968-1980 1968-1980 1968-2000 1968-2000 Tax receipts 9.00 7.32 9.16 18.14 20.74 Net Benefit (Benefits - Costs) 6.56 3.58 5.42 11.25 7.76 Breakeven year 1998 2007 2009 2018 2031 Living standards 61.69 53.18 65.46 128.34 146.49 Net Benefit (Benefits - Costs) 59.24 49.44 61.72 121.45 133.50 Breakeven year 1990 1992 1994 1999 2003 Internal rate of return Tax receipts 10.48 7.68 8.10 8.05 6.37 Living Standards 20.68 17.69 16.84 15.26 13.15

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Cost Benefit Analysis

Taxes Living Standards Taxes Living Standards

• 10

40 90 140 Net Benefits (Billions USD) 1980 2000 2020 2040 2060 2080 Year Scenario 5 Scenario 10

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Robustness checks

• Including non-coresident children for second generation

analysis

Bounding exercises

• Possible general equilibrium effects

General Equilibrium

• Placebo regressions: compare old cohort (ages 12-17 in 1974)

versus even older cohort (ages 18-24 in 1974)

Placebo

• Extended cohort definitions: added younger cohorts

(post-1972), middle cohorts (1963-1967), and older cohorts (pre-1957)

Extended cohort definitions

• Expenditures: estimated using inverse hyperbolic sine, logs,

and nominal amounts; estimated using per capita and total household expenditures

Alternative expenditure transformations

• Alternative control variables: estimated regressions on indexes

without the main control variables

Alternative control variables

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Conclusion (1)

• (Very) long-term impacts of schooling on wide range of
• utcomes: education, employment, migration, expenditures,

health, taxes

• Important gender differences
• Years of schooling increases by ¼ of a year for men and

women

• For women, concentrated in primary school
• For men also significant effects for higher education levels
• Large employment effects
• Adult men more likely to be working and in the formal sector
• Men and women more likely to have migrated
• Households in which either head or spouse received more

education have: higher expenditures, more assets, improved housing, pay higher taxes, improved nutrition, increased health expenditures, but little observable improvement in health

Introduction Empirical Strategy 1G Results 2G Results CBA Robustness Conclusion

Conclusion (2)

• Parents transmit these effects to next generation: increases in

children’s years of schooling, particularly large increases in secondary and tertiary education

• Parent’s exposure leads to 0.10-0.17 extra years of child

education

• Mother’s effect seems larger than father’s
• Effect on daughters seems larger than sons
• Intergenerational transmission of human capital appear to be

driven by changes in marriage partner’s characteristics, with spouses having more education and improved labor market

• utcomes
• Policy implications: School construction pays for itself in

higher government taxes

Extra Slides

Appendix

Extra Slides

Difference-in-differences

yijtca = α + β Schoolj · Youngit + (XjB

′

t) γt + µj + δt + θa + εijtca

• yijtca is the outcome of child c who is age a, born to parent i

who was born in district j in year t

• Schoolj is the number of INPRES schools constructed per

1,000 children in the father’s or mother’s birth district between 1973 and 1979

• Youngit is a dummy if the father or mother belongs to the

young cohort (ages 2-6 in 1974)

• θa are child c’s age fixed effects
• Standard errors clustered at the father’s or mother’s birth

district

Back

Extra Slides

Parallel trends men

Back

Extra Slides

Parallel trends women

Back

Extra Slides

Summary Statistics 1st Generation

Back

Extra Slides

Summary Statistics 2nd Generation

Back

Extra Slides

Long-run Effects of School Construction

Extra Slides

Effect of School Construction on Education

• Duflo (2001) difference-in-differences in Indonesia
• Handa (2002) difference-in-differences in Mozambique
• Alderman, Kim, Orazem (2003) RCT in Pakistan
• Burde and Linden (2013) in Afghanistan
• Kazianga, Levy, Linden, Sloan (2013) RD in Burkina Faso
• Khanna (2018) in India

Extra Slides

Long-run Effects on Taxes

Outcomes: Taxes Mean/SD Effect on: Men Women Men Women Total (Rp10k) 4.749 4.552 0.078*** 0.123*** (11.433) (10.743) (0.017) [0.000] (0.019) [0.000] Land & building (Rp10k) 0.465 0.506 0.041* 0.075*** (2.742) (2.446) (0.022) [0.120] (0.021) [0.000] Vehicle (Rp10k) 3.610 3.398 0.154*** 0.267*** (8.076) (7.821) (0.047) [0.003] (0.052) [0.000] Local (Rp10k) 0.469 0.468 0.048 0.082** (2.259) (2.074) (0.033) [0.148] (0.039) [0.036] Back

Extra Slides

Education and Health

• Correlation between more education and better health
• Literature finds mixed results when measuring causal

relationship

• Some studies find changes in labor market outcomes, but no

effects on mortality or self-reported health (Malamud, Mitrut, Pop-Eleches, 2018)

• Health and mortality in US (Lleras-Muney, 2005, Mazumder,

2008), UK (Oreopoulous, 2006; Clark and Royer, 2013), Sweden (Meghir et al., forthcoming)

Extra Slides

Long-run Effects on Health Investments

Outcomes: Health Investment Mean/SD Effect on: Men Women Men Women Total health expenditure (Rp10k) 7.517 7.961 0.071* 0.055 (34.130) (35.245) (0.038) [0.114] (0.041) [0.185] Preventive measures (Rp10k) 0.744 0.671 0.242*** 0.193*** (3.225) (3.135) (0.068) [0.002] (0.071) [0.013] Family planning (Rp10k) 0.286 0.219 0.321*** 0.226*** (0.872) (0.856) (0.061) [0.000] (0.071) [0.008] Private hospital (Rp10k) 2.101 2.200 0.048** 0.075*** (20.718) (22.266) (0.023) [0.114] (0.024) [0.008] Health insurance (Rp10k) 3.821 3.635 0.083 0.142*** (16.425) (14.047) (0.055) [0.134] (0.048) [0.009]

Extra Slides

Long-run Effects on Nutrition

Outcomes: Food Intake Mean/SD Effect on: Men Women Men Women Calorie (IHS) 260.915 249.699 0.005 0.018*** (106.001) (109.833) (0.004) [0.301] (0.005) [0.001] Protein (IHS) 7.116 6.831 0.006 0.018*** (3.254) (3.330) (0.005) [0.301] (0.005) [0.001] Fat (IHS) 6.074 5.810 0.011** 0.023*** (3.110) (3.150) (0.004) [0.061] (0.006) [0.000] Carbohydrate (IHS) 40.869 39.040 0.005 0.017*** (17.728) (18.245) (0.004) [0.301] (0.005) [0.001]

Extra Slides

Long-run Effects on Health Outcomes

Outcomes: Health Outcomes Mean/SD Effect on: Men Women Men Women No health complaint 0.690 0.646 0.004 0.003 (0.463) (0.478) (0.004) [0.352] (0.004) [0.771] Non-disrupted days 28.851 28.801 0.042 0.027 (4.012) (4.064) (0.028) [0.266] (0.033) [0.771] No severe health complaint 0.951 0.949 0.005***

• 0.001

(0.216) (0.221) (0.002) [0.025] (0.002) [0.771] Back

Extra Slides

Education and Demographics

• Reduction in fertility (Osili and Long, 2008)
• Improved quality of marriage partner, small effect on fertility

(McCrary and Royer, 2011)

• Reduced teen pregnancy and teen marriage (Duflo, Dupas,

Kremer, 2015)

• Lower teen fertility, but no impact on completed fertility,

increased spousal education (Geruso and Royer, 2018)

Back

Extra Slides

Long-run Effects on Marriage (2)

Outcome Mean/SD Effect on: Men Women Men Women Spouse’s characteristics Work 0.614 (0.487) 0.918 (0.275) 0.003 (0.005) [0.867]

• 0.001

(0.004) [0.745] Formal worker 0.239 (0.427) 0.293 (0.455) 0.003 (0.005) [0.867] 0.005 (0.005) [0.745] Non-agriculture sector 0.555 (0.497) 0.517 (0.500) 0.006 (0.005) [0.867] 0.005 (0.007) [0.745] Migrant 0.260 (0.439) 0.277 (0.448) 0.007* (0.004) [0.387] 0.008** (0.003) [0.164] No health complaint 0.694 (0.461) 0.646 (0.478)

• 0.001

(0.004) [0.867] 0.006 (0.005) [0.745] Back

Extra Slides

Assortative Matching

Outcome Mean/SD Treatment Effect on: Men Women Men Women Spouse’s years of schooling 7.635 7.426 0.180*** 0.116*** (4.081) (4.192) (0.046) [0.000] (0.043) [0.028] Spouse’s years of schooling percentile 40.62 (28.49) 32.44 (29.29) 0.625** (0.303) [0.115] 0.448 (0.298) [0.133] Education gap 0.432 (3.707) 0.112 (3.600) 0.093* (0.049) [0.115]

• 0.078**

(0.038) [0.086] Education percentile gap 36.96 (31.18) 34.39 (30.08) 0.461 (0.402) [0.252]

• 0.738**

(0.322) [0.065] Back

Extra Slides

Long-run Effects on Housing and Assets

Outcomes: Housing and Mean/SD Effect on: Assets Men Women Men Women Urban 0.425 0.438

• 0.001

0.002 (0.494) (0.496) (0.004) [0.822] (0.004) [0.576] Rent equivalent (Rp10k) 42.991 43.085 0.012 0.028*** (56.342) (56.573) (0.008) [0.293] (0.008) [0.001] Floor area (sq-m) 79.894 81.355 1.229** 1.480*** (58.651) (59.726) (0.566) [0.119] (0.510) [0.011] Utilities (Rp10k) 15.714 15.729 0.051** 0.085*** (20.983) (21.796) (0.022) [0.102] (0.024) [0.002] Asset index (PCA)

• 0.035
• 0.069

0.030* 0.040** (1.868) (1.882) (0.017) [0.223] (0.015) [0.020] Back

Extra Slides

Long-run Effects on Welfare Program Participation

Outcomes: Mean/SD Effect on: Not a Recipient of Welfare Program Men Women Men Women Cash Transfer 0.959 0.961 0.002 0.001 (0.197) (0.194) (0.002) [0.742] (0.002) [0.914] Rice for Poor 0.608 0.594

• 0.002

0.009* (0.488) (0.491) (0.004) [0.850] (0.005) [0.200] Poor Student's Assistance 0.844 0.873 0.001

• 0.000

(0.363) (0.333) (0.004) [0.850] (0.004) [0.914] Social Protection Card 0.814 0.820 0.001 0.000 (0.389) (0.384) (0.004) [0.850] (0.004) [0.914] Back

Extra Slides

Interngenerational Effects on Wellbeing Second Generation

Outcomes: Child welfare Mean/SD Effect of Exposure: Fathers Mothers Fathers on Children Mothers on Children Non-work days 5.317 4.820 0.044** 0.031 (2.670) (2.865) (0.021) [0.136] (0.019) [0.463] Non-work hours 156.679 153.047 0.299* 0.215 (19.704) (21.597) (0.157) [0.173] (0.151) [0.463] No health complaint 0.797 0.823

• 0.008***

0.004 (0.402) (0.382) (0.003) [0.042] (0.003) [0.463] Non-disrupted days 29.492 29.550

• 0.026*

0.007 (2.086) (2.067) (0.016) [0.198] (0.015) [0.893] No severe health complaint 0.978 0.980

• 0.000
• 0.000

(0.147) (0.140) (0.001) [0.751] (0.001) [0.893] Back

Extra Slides

Intergenerational Effects on Education by Gender

Dependent variable: Years of Schooling Parents born between: 1957-1962 and 1968-1972 1950- 1980 1957-1962 and 1968-1972 1950- 1980 1957-1962 and 1968-1972 1950- 1980 2nd generation children: Sons and Daughters Sons Only Daughters Only Father exposed 0.001 (0.038) 0.044** (0.021)

• 0.038

(0.049) 0.042 (0.026) 0.036 (0.051) 0.046** (0.023) Mother exposed 0.160*** (0.059) 0.118*** (0.035) 0.139** (0.069) 0.094** (0.040) 0.188*** (0.072) 0.140*** (0.038) Father = Mother (p-value) 0.046 0.050 0.076 0.267 0.134 0.026 Mean 8.674 7.827 8.575 7.787 8.796 7.875 Observations 44,105 246,466 24,366 133,896 19,739 112,570 Back

Extra Slides

Cost-Benefit Scenario

1994 2014 10 20 30 40 50 Net Benefits (Billions USD) 1970 1980 1990 2000 2010 2020 2030 2040 2050 Year Taxes Living Standards Back

Extra Slides

Placebo Effect of School Construction on Indexes

• .025

.025 .05 .075 .1 First Generation Second Generation Education Work/ Migration Living standards Tax Housing/ Assets Nutrition Health investment Health Marriage Market Not on Welfare program Education Child wellbeing Female Male

Back

Extra Slides

Effects on Indexes with Extended Cohort Sample

• .025

.025 .05 .075 .1 First Generation Second Generation Education Work/ Migration Living standards Tax Housing/ Assets Nutrition Health investment Health Marriage Market Not on Welfare program Education Child wellbeing Female Male

Back

Extra Slides

Effects on Indexes with Alternative Control Variables

• .025

.025 .05 .075 .1 First Generation Second Generation Education Work/ Migration Living standards Tax Housing/ Assets Nutrition Health investment Health Marriage Market Not on Welfare program Education Child wellbeing Female Male

Back

Extra Slides

Effects on Expenditures using Alternatives Transformations

Total expenditure IHS Total Log Nominal IHS Per-capita (1) (2) (3) (4) Panel A: Father Schools constructed * Young cohort 0.021*** (0.007) 0.021*** (0.007) 9.882*** (3.628) 0.016** (0.007) Observations 68,687 68,687 68,687 68,687 Mean 391.649 391.649 391.649 391.649 Panel B: Mother Schools constructed * Young cohort 0.032*** (0.007) 0.032*** (0.007) 11.022*** (2.583) 0.018*** (0.007) Observations 66,249 66,249 66,249 66,249 Mean 375.616 375.616 375.616 375.616 Back

Extra Slides

Including non-coresident children

• Trade-off between the selection of which individuals remain in

a household, and are therefore in the survey, and what age they need to be to finish higher levels of school

Alternative upper-bound ages

• Susenas only includes information on individuals currently

residing in a given household but not on family members living elsewhere. Three bounding analyses:

• 1. Extreme bounds in which we assume all non-co-resident

children have parents who are or are not exposed

Table

• 2. Use auxiliary data (IFLS) to obtain fraction of children at each

age born to old and young cohort parents among all children no longer living with their parents. Randomly assign at each age non-co-resident children in the Susenas data to either old

• r young cohort parents and exclude the rest from the
• regression. Simulate 1,000 times

Figure

• 3. Repeat analyses in IFLS

Table Back

Extra Slides

Including non-coresident children

• Trade-off between the selection of which individuals remain in

a household, and are therefore in the survey, and what age they need to be to finish higher levels of school

Alternative upper-bound ages

• Susenas only includes information on individuals currently

residing in a given household but not on family members living elsewhere. Three bounding analyses:

• 1. Extreme bounds in which we assume all non-co-resident

children have parents who are or are not exposed

Table

• 2. Use auxiliary data (IFLS) to obtain fraction of children at each

age born to old and young cohort parents among all children no longer living with their parents. Randomly assign at each age non-co-resident children in the Susenas data to either old

• r young cohort parents and exclude the rest from the
• regression. Simulate 1,000 times

Figure

• 3. Repeat analyses in IFLS

Table Back

Extra Slides

Simulated Exposure Assignment

Susenas Susenas with Extreme Assumptions IFLS (1) (2) (3) (4) (5) (6) Assume Not Exposed Assume Exposed All Stayers Movers Panel A: Father Schools constructed * Young cohort 0.097*** (0.032) 0.021 (0.016) 0.000 (0.014) 0.103 (0.104) 0.030 (0.109)

• 0.020

(0.251) Observations 120,838 644,675 644,675 6,186 4,048 2,138 Panel B: Mother Schools constructed * Young cohort 0.169*** (0.045) 0.052*** (0.017) 0.030* (0.017) 0.300** (0.147) 0.539*** (0.128) 0.126 (0.239) Observations 105,523 644,675 644,675 7,227 3,756 3,471 Back

Extra Slides

Simulated Exposure Assignment

0.011 0.047 0.018 0.065 10 20 30 40 .02 .04 .06 .08 .1 Estimated Treatment Effect on Child's Education Father Mother

Note: estimates from 1000 random sample drawings; 5th and 95th percentiles indicated.

Back

Extra Slides

Alternative Upper-Bound Ages

Back

Extra Slides

Possible General Equilibrium Effects

• If school construction increases the young cohort’s education

in high intensity regions, this could affect individuals who are not exposed to school construction

• Could bias our results leading to an over- or under-estimate of

the true effect

• If young cohorts are substitutes for the older cohorts in the

labor market, then this could drive down wages for the older cohorts who are competing with them → overestimate true effect

• Adjusting our estimates by the magnitudes found by Duflo

(2004) does not significantly alter our results

• If young cohorts are complements for the old cohorts, then the
• lder cohorts benefit by having more educated younger cohorts

in their location → underestimate true effect (Khanna, 2018)

Back

Extra Slides

Advantage of using Susenas 2016 data

• 2 critical data issues about Susenas 2016 relevant for our

analysis

• Includes information on individual’s district of birth
• Most household surveys only have information on individual’s

current district of residence

• Sample is sufficiently large to precisely estimate observed

relationships

• Indonesian Family Life Survey (IFLS) has data on district of

birth, but not large enough to detect effects of school construction (Appendix Table B.1 and Appendix Table B.2)

• 2 key differences between Susenas and IFLS (geographic

coverage and number of observations)

• Map of Indonesia with districts in gray indicating IFLS

coverage

Back

Extra Slides

Marriage market ’puzzle’

Increased living standards for women in the absence of own labor market improvements

Mediator: None Years of Schooling Completed Primary Literate Work Formal worker Non- agriculture sector Migrant All Panel A: Father 0.021*** 0.009 0.007 0.010 0.020*** 0.019** 0.017** 0.018** 0.008 Schools constructed * Young cohort (0.007) (0.008) (0.007) (0.007) (0.008) (0.009) (0.008) (0.008) (0.009) Mediator 0.062*** 0.337*** 0.374*** 0.002 0.371*** 0.480*** 0.273*** (0.001) (0.009) (0.014) (0.008) (0.014) (0.010) (0.011) Panel B: Mother 0.032*** 0.016** 0.015** 0.017** 0.023*** 0.018** 0.018** 0.021*** 0.012 Schools constructed * Young cohort (0.007) (0.008) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) (0.007) Mediator 0.063*** 0.351*** 0.380*** -0.057*** 0.319*** 0.412*** 0.293*** (0.001) (0.009) (0.015) (0.013) (0.012) (0.009) (0.012)