## wpiea2020027-print-pdf

## Source details

**Canonical URL:** [wpiea2020027-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020027-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020027-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020027-print-pdf.pdf.json)

---

### Key findings and magnitudes
- In Emerging and Developing Countries (EMDCs), one fifth of girls are married before the age of 18 (UNICEF Global Database 2018).
- Reducing child marriage by one standard deviation (16.74 percentage points) can increase per capita growth by 0.66 percentage points per year (simultaneous equations model).
- If child marriage were ended today, long-term annual per capita growth in EMDCs would increase by 1.05 percentage points (linear relationship between child marriage and growth).
- A decrease in child marriage by one standard deviation (16.74 percent) can increase per capita growth by 0.33 percentage points per year in the basic growth regression with only initial income and child marriage (regression 1, Table 2).
- A one standard deviation reduction in 10-year lagged child marriage improves the human capital index by 0.35, which then improves per capita growth by 0.36 percentage points.
- A one standard deviation improvement in the 5-year lagged human capital index (0.51) reduces child marriage by 13.54 percentage points and increases per capita growth by 0.53 percentage points per year—suggesting almost 80 percent of the effect of child marriage on growth is accounted for by lagged education.
- A one standard deviation improvement in the 10-year lagged life expectancy at birth (9.28 years) increases the human capital index by 0.19, which reduces child marriage by 5.20 percentage points and raises per capita growth by 0.20 percentage points per year.
- Under a long-run equilibrium interpretation, a one standard deviation reduction in child marriage (16.74 percentage points) could raise annual per capita growth by 1.02 percentage points.
- A one standard deviation decrease in the log of per capita GDP (0.41) can increase per capita growth by 0.67 percentage points per year (simultaneous model reports a similar magnitude: 0.72 percentage points in long-run interpretation).

### Socioeconomic context and consequences
- Child marriage substantially increases fertility and population growth and reduces educational attainment resulting in lower earnings and productivity.
- Child marriage harms the health and nutrition of mothers and their children, is associated with increased family violence and gender inequality, and reduces women’s decision-making power (Wodon et al., 2017).
- Ending child marriage is a Sustainable Development Goal for 2030 with vital impacts on poverty-reduction, inequality, and economic growth.
- Previous estimates: ending child marriage by 2030 in a group of 106 EMDCs would save, on average, about $300 billion per year between now and 2030 from reduced population growth and about $75 billion per year from improved child health (Wodon et al., 2017).
- Savadogo and Wodon (2017) estimate EMDCs would gain, by 2030, an average of 1 percent of family earnings from improved salaries and productivity. Chaaban and Cunningham (2011) estimate allowing girls in EMDCs to complete education could increase lifetime earnings by up to 1.5 percent per year.

### Modeling approach
- Simultaneous equations model applied to account for interdependence between child marriage, education, and health, and to quantify correlation between child marriage and long-term per capita GDP growth.
- Basic cross-country growth equation (Sala-i-Martin et al. (2004) framework):
  - G_{t,t+T} = θ0 + θ1 ln(y_t) + θ X_t + ε_t, with T=20 (growth from 1997 to 2017); cross-sectional sample of 112 countries (attribute of observations reduced to 76 due to data attrition).
- Child marriage definition: percentage of women aged 20 to 24 years who were first married or in a union before age 18 (UNICEF Global Database); each country has at least one child marriage statistic between 1986 and 2017; averages or single observations used to proxy the 1997 value given high temporal stability.
- Simultaneous model system (T-year model, h chosen as 5 years):
  - Growth equation: G_{t,t+T} depends on He_t (health), Edu_t (education), CM_t (child marriage), ln(y_t), and X1_t.
  - Child marriage equation: CM_t depends on He_{t-h}, Edu_{t-h}, and X2_t.
  - Education equation: Edu_{t-h} depends on He_{t-2h} and CM_{t-2h}.
- Estimation method: three-stage least squares to address endogeneity and correlated disturbances across equations.
- Key proxies and controls:
  - Human capital index based on years of schooling and returns to education (Penn World Table 9).
  - Life expectancy at birth used as proxy for health (World Bank WDI).
  - Control variables include log of per capita GDP, East Asia dummy, share of extractable commodities in exports, fraction of tropical area, trade openness, relative price of investment, share of government consumption, terms of trade, credit to private sector (with some variables averaged over prior periods to ensure exogeneity).

### Empirical results: growth equation and child marriage interactions
- Basic growth regressions (Table 2):
  - Regression 1 (only initial income and child marriage): child marriage coefficient negative and significant; a 1 SD reduction (16.74 percent) raises per capita growth by 0.33 percentage points per year; R^2 = 0.16.
  - Regression 2 (Sala-i-Martin theoretical equation without child marriage): human capital index positive and significant; East Asia dummy positive and significant; log per capita GDP, fraction of tropical area, and share of government consumption negative and significant; R^2 = 0.45.
  - Regression 3 (theoretical controls plus child marriage): child marriage negative but not significant, suggesting correlations between child marriage and other explanatory variables.
  - Child marriage squared not significant — linear relationship assumed.
- Child marriage regression (Table 3):
  - GDP per capita, East Asia dummy, and lagged human capital are significantly and negatively linked to child marriage.
  - Fraction of tropical area is significantly and positively correlated with child marriage.
  - Results robust to lag changes (5 to 7 years) and addition of terms of trade, credit to private sector, labor force participation.
- Simultaneous equations model (Table 5):
  - Child marriage has a significant and negative direct impact on growth: 1 SD reduction (16.74 percentage points) → per capita growth +0.66 percentage points per year.
  - Indirect/intergenerational channel: reduction in child marriage lagged 10 years raises human capital and later growth (0.35 increase in human capital → 0.36 percentage points growth).
  - Initial wealth impact comparable to child marriage: 1 SD decrease in ln per capita GDP (0.41) → per capita growth +0.67 percentage points per year.
  - Share of government consumption: 1 SD decline (11.67 percentage points) → per capita growth +0.23 percentage points per year.
  - Lagged human capital (0.51 SD) reduces child marriage by 13.54 percentage points → increases per capita growth by 0.53 percentage points per year (education accounts for ~80 percent of child marriage effect on growth).
  - Lagged health: 1 SD improvement in life expectancy (9.28 years) → human capital +0.19 → child marriage -5.20 percentage points → growth +0.20 percentage points per year.
  - The 5-year lagged life expectancy effect on child marriage is significant and positive (possible interpretation: better survival rates lead to more children being married absent other policy changes).
  - Results robust to inclusion of inequality and governance proxies, varying lag lengths (h=7), using child marriage by age 15, adding credit to private sector and labor force participation as regressors, and removing East Asia dummy.

### Robustness checks and selected econometric statistics
- Basic growth equation (robustness variants):
  - Log of the per capita GDP coefficient reported values include -1.26***, -1.54***, -1.28***, -1.24***.
  - Human Capital Index coefficients reported as 0.76***, 0.87***, 0.79***.
  - Female school enrollment, secondary: 0.01** (0.00).
  - Fraction of tropical area: -0.61***, -0.39**, -0.51***, -0.61***.
  - Trade openness in IV regression: -0.01** (noted as puzzling and possibly due to small sample size of 54).
  - Observations and fit statistics: Observations 76, 54, 76, 72 across variants; R2 reported as 0.45, 0.48, 0.46, 0.44; P-value 0.00 in reported variants.
- Regressions of child marriage on explanatory variables:
  - Log of the per capita GDP coefficients reported as -13.01*, -10.58, -13.04*, -12.30, -13.00*.
  - Human Capital Index (5-year lag): -11.27***, -11.11***, -12.52***, -11.32***.
  - Fraction of tropical area positive and significant across variants (e.g., 10.81***, 10.22***, 10.95***, 11.32***, 10.78***).
  - Observations typically 77 (some variants 73); R2 reported as 0.56 across variants; P-value 0.00.
- Simultaneous equations model (selected coefficients):
  - Growth (t,t+T): Child marriage (t) effects reported as -0.04**, -0.06**, -0.11**, -0.03**, -0.06*** across models.
  - Log of the per capita GDP (t) in growth equation: -1.63***, -1.77***, -1.72***, -1.56***, -1.85***.
  - In the child marriage equation: Log of the per capita GDP (t) coefficients reported as -7.89**, -7.44**, -1.32, -8.35**, -8.37**; Human Capital Index (t-h) reported as -26.80***, -27.65***, -13.30***, -26.73***, -26.95***.
  - Human Capital equation: Life expectancy at birth (t-2h) 0.02***, 0.02***, 0.01*, 0.02***, 0.02***; Child marriage (t-2h) -0.02*** across variants.
  - Observations typically 74 (some variants 71); P-value 0.00 in reported variants.
- Additional robustness notes:
  - Defining child marriage as marriage before age 15 (instead of age 18) produces small differences: the effect of initial wealth on marriage before age 15 is not significant, and the marginal effect of the human capital index is smaller than in the baseline.
  - Removing the East Asia dummy in the growth equation marginally increases the long-run effect of health on growth.
  - Models controlling for institutions (rule of law, government effectiveness) were estimated; variables are scores from the Worldwide Governance Indicators database normalized so that 0 and 1 correspond to the worse and the best governance situations.

### Policy implications and priorities
- Ending child marriage yields substantial long-term per capita growth benefits—directly and via improvements in education and health.
- Education policies are crucial: lagged education produces the largest channel through which child marriage reductions raise growth (almost 80 percent of the effect).
- Health improvements matter mainly through their positive effect on education, and thus on child marriage and growth.
- Multifaceted strategies required to end child marriage:
  - Empowerment of girls (information, skills, support networks).
  - Community education and outreach to parents and communities on the costs of child marriage.
  - Improved educational quality and access for girls.
  - Economic incentives to parents and social assistance to poor and marginalized children.
  - Legal and policy frameworks and enforcement; adoption of laws alone has not ended the practice (World Bank, 2017).
- Complementary measures: remove policies that expel pregnant girls from school; stepped-up outreach and education to governments and affected communities; regional and multilateral initiatives are starting to drive policy change.

### Data availability (Appendix 1)
- Appendix 1 lists country-year availability of child marriage data (UNICEF Global Databases); selected country examples:
  - Afghanistan: 2013, 2015
  - Bangladesh: 1994, 1997, 2000, 2004, 2007, 2011, 2013, 2014
  - India: 1993, 1999, 2006, 2015, 2016
  - Nigeria: 1990, 2003, 2008, 2013, 2017
  - Ethiopia: 2000, 2005, 2011, 2016
  - Pakistan: 1991, 2006, 2012
  - Peru: 1986, 1992, 1996, 2000, 2004, 2007, 2009, 2010, 2011, 2012, 2014
- Full Appendix 1 provides a comprehensive country-by-year inventory (page reference: 19).

### Limitations and avenues for future research
- The model quantifies correlations and interdependencies but cannot establish strict causality; micro-analyses and randomized experiments could provide causal insights.
- Better frequency and granularity of child marriage and education data (e.g., annual child marriage, enrollment rates) would improve identification and interpretation.
- Panel analyses at subnational levels (state-level child marriage data) could be informative where higher-frequency data exist.

### Conclusions
- Child marriage is prevalent across many EMDCs and has serious socioeconomic consequences across fertility, health, education, and earnings.
- A simultaneous equations approach shows that reducing child marriage could significantly improve economic growth directly and indirectly, with education representing the primary channel.
- Quantitatively, a one standard deviation reduction in child marriage (16.74 percentage points) raises long-term annual per capita real GDP growth by 0.66 percentage points (simultaneous model) and could raise it by about 1.05 percentage points if child marriage were ended today (long-term linear interpretation).

*Source: IMF working paper (content unit spanning I. INTRODUCTION, II. MODELING CHILD MARRIAGE AND GROWTH, and III. CONCLUSIONS).*

### Appendix 1: Availability of Child Marriage Data ................................................................ 19

### Appendix 1: Availability of Child Marriage Data

### Appendix overview
- Appendix 1 title: Availability of Child Marriage Data ................................................................ 19

### Content pointers
- Indicates coverage or discussion of available data sources and availability issues for child marriage (exact contents not provided in source excerpt).
- Page reference: 19

---

### Appendix 2: Robustness Checks for Estimation of the Basic Growth Equation

### Appendix overview
- Appendix 2 title: Robustness Checks for Estimation of the Basic Growth Equation .................... 21

### Content pointers
- Signals robustness analyses related to the basic growth equation estimation (exact methods and results not provided in source excerpt).
- Page reference: 21

---

### Appendix 3: Robustness Checks for Regressions of Child marriage on Explantory Variables from Basic Growth Equations

### Appendix overview
- Appendix 3 title: Robustness Checks for Regressions of Child marriage on Explantory Variables 
  from Basic Growth Equations .............................................................................................. 22

### Content pointers
- Indicates robustness checks for regressions linking child marriage to explanatory variables drawn from basic growth equations (exact methods and results not provided in source excerpt).
- Page reference: 22

---

### Appendix 4: Robustness Checks for Estimation of the Simultaneous Equations Model

### Appendix overview
- Appendix 4 title: Robustness Checks for Estimation of the Simultaneous Equations Model ........ 23

### Content pointers
- Indicates robustness analyses for a simultaneous equations model estimation (exact methods and results not provided in source excerpt).
- Page reference: 23

*Source: wpiea2020027-print-pdf - Appendix 1: Availability of Child Marriage Data ................................................................ 19*

### References .............................................................................................................

### I.   INTRODUCTION

### Key findings and magnitudes
- In Emerging and Developing Countries (EMDCs), one fifth of girls are married before the age of 18 (UNICEF Global Database 2018).
- Reducing child marriage by one standard deviation (16.74 percentage points) can increase per capita growth by 0.66 percentage points per year (simultaneous equations model).
- If child marriage were ended today, long-term annual per capita growth in EMDCs would increase by 1.05 percentage points (linear relationship between child marriage and growth).
- A decrease in child marriage by one standard deviation (16.74 percent) can increase per capita growth by 0.33 percentage points per year in the basic growth regression with only initial income and child marriage (regression 1, Table 2).
- A one standard deviation reduction in 10-year lagged child marriage improves the human capital index by 0.35, which then improves per capita growth by 0.36 percentage points.
- A one standard deviation improvement in the 5-year lagged human capital index (0.51) reduces child marriage by 13.54 percentage points and increases per capita growth by 0.53 percentage points per year—suggesting almost 80 percent of the effect of child marriage on growth is accounted for by lagged education.
- A one standard deviation improvement in the 10-year lagged life expectancy at birth (9.28 years) increases the human capital index by 0.19, which reduces child marriage by 5.20 percentage points and raises per capita growth by 0.20 percentage points per year.
- Under a long-run equilibrium interpretation, a one standard deviation reduction in child marriage (16.74 percentage points) could raise annual per capita growth by 1.02 percentage points.
- A one standard deviation decrease in the log of per capita GDP (0.41) can increase per capita growth by 0.67 percentage points per year (simultaneous model reports a similar magnitude: 0.72 percentage points in long-run interpretation).

### Socioeconomic context and consequences
- Child marriage substantially increases fertility and population growth and reduces educational attainment resulting in lower earnings and productivity.
- Child marriage harms the health and nutrition of mothers and their children, is associated with increased family violence and gender inequality, and reduces women’s decision-making power (Wodon et al., 2017).
- Ending child marriage is a Sustainable Development Goal for 2030 with vital impacts on poverty-reduction, inequality, and economic growth.
- Previous estimates: ending child marriage by 2030 in a group of 106 EMDCs would save, on average, about $300 billion per year between now and 2030 from reduced population growth and about $75 billion per year from improved child health (Wodon et al., 2017).
- Savadogo and Wodon (2017) estimate EMDCs would gain, by 2030, an average of 1 percent of family earnings from improved salaries and productivity. Chaaban and Cunningham (2011) estimate allowing girls in EMDCs to complete education could increase lifetime earnings by up to 1.5 percent per year.

---

### Modeling approach
- A simultaneous equations model is applied to account for interdependence between child marriage, education, and health, and to quantify the correlation between child marriage and long-term per capita GDP growth.
- A basic cross-country growth equation follows Sala-i-Martin et al. (2004):
  - G_{t,t+T} = θ0 + θ1 ln(y_t) + θ X_t + ε_t, with T=20 (growth from 1997 to 2017) and a cross-sectional sample of 112 countries (attribute of observations reduced to 76 due to data attrition).
- Child marriage is defined as the percentage of women aged 20 to 24 years who were first married or in a union before age 18 (UNICEF Global Database). Each country has at least one child marriage statistic between 1986 and 2017; averages or single observations are used to proxy the 1997 value given high temporal stability.
- Simultaneous model system (T-year model, h chosen as 5 years):
  - Growth equation: G_{t,t+T} depends on He_t (health), Edu_t (education), CM_t (child marriage), ln(y_t), and X1_t.
  - Child marriage equation: CM_t depends on He_{t-h}, Edu_{t-h}, and X2_t.
  - Education equation: Edu_{t-h} depends on He_{t-2h} and CM_{t-2h}.
- Estimation method: three-stage least squares to address endogeneity and correlated disturbances across equations.
- Key proxies:
  - Human capital index based on years of schooling and returns to education (Penn World Table 9).
  - Life expectancy at birth used as proxy for health (World Bank WDI).
  - Control variables include log of per capita GDP, East Asia dummy, share of extractable commodities in exports, fraction of tropical area, trade openness, relative price of investment, share of government consumption, terms of trade, credit to private sector (with some variables averaged over prior periods to ensure exogeneity).

---

### Empirical results: growth equation and child marriage interactions
- Basic growth regressions (Table 2):
  - Regression 1 (only initial income and child marriage): child marriage coefficient negative and significant; a 1 SD reduction (16.74 percent) raises per capita growth by 0.33 percentage points per year; R^2 = 0.16.
  - Regression 2 (Sala-i-Martin theoretical equation without child marriage): human capital index positive and significant; East Asia dummy positive and significant; log per capita GDP, fraction of tropical area, and share of government consumption negative and significant; R^2 = 0.45.
  - Regression 3 (theoretical controls plus child marriage): child marriage negative but not significant, suggesting correlations between child marriage and other explanatory variables.
  - Child marriage squared not significant — linear relationship assumed.
- Child marriage regression (Table 3):
  - GDP per capita, East Asia dummy, and lagged human capital are significantly and negatively linked to child marriage.
  - Fraction of tropical area is significantly and positively correlated with child marriage.
  - Results robust to lag changes (5 to 7 years) and addition of terms of trade, credit to private sector, labor force participation.
- Simultaneous equations model (Table 5):
  - Child marriage has a significant and negative direct impact on growth: 1 SD reduction (16.74 percentage points) → per capita growth +0.66 percentage points per year.
  - Indirect/intergenerational channel: reduction in child marriage lagged 10 years raises human capital and later growth (0.35 increase in human capital → 0.36 percentage points growth).
  - Initial wealth impact comparable to child marriage: 1 SD decrease in ln per capita GDP (0.41) → per capita growth +0.67 percentage points per year.
  - Share of government consumption: 1 SD decline (11.67 percentage points) → per capita growth +0.23 percentage points per year.
  - Lagged human capital (0.51 SD) reduces child marriage by 13.54 percentage points → increases per capita growth by 0.53 percentage points per year (education accounts for ~80 percent of child marriage effect on growth).
  - Lagged health: 1 SD improvement in life expectancy (9.28 years) → human capital +0.19 → child marriage -5.20 percentage points → growth +0.20 percentage points per year.
  - The 5-year lagged life expectancy effect on child marriage is significant and positive (possible interpretation: better survival rates lead to more children being married absent other policy changes).
  - Results robust to inclusion of inequality and governance proxies, varying lag lengths (h=7), using child marriage by age 15, adding credit to private sector and labor force participation as regressors, and removing East Asia dummy.

---

### Policy implications and priorities
- Ending child marriage yields substantial long-term per capita growth benefits—directly and via improvements in education and health.
- Education policies are crucial: lagged education produces the largest channel through which child marriage reductions raise growth (almost 80 percent of the effect).
- Health improvements matter mainly through their positive effect on education, and thus on child marriage and growth.
- Multifaceted strategies are required to end child marriage: empowerment of girls (information, skills, support networks), community education, improved educational quality and access for girls, economic incentives to parents, and legal/policy frameworks. No single strategy is sufficient; comprehensive programs and investments are lacking.
- Adoption of laws alone has not ended the practice (World Bank, 2017); regional and multilateral initiatives (e.g., World Bank Sahel Women Empowerment and Demographic regional project; African Union, UNFPA, UNICEF initiatives) are beginning to drive policy change.

---

### Limitations and avenues for future research
- The model quantifies correlations and interdependencies but cannot establish strict causality; micro-analyses and randomized experiments could provide causal insights.
- Better frequency and granularity of child marriage and education data (e.g., annual child marriage, enrollment rates) would improve identification and interpretation.
- Panel analyses at subnational levels (state-level child marriage data) could be informative where higher-frequency data exist.

### CONCLUSIONS
- Child marriage is prevalent across many EMDCs and has serious socioeconomic consequences across fertility, health, education, and earnings.
- A simultaneous equations approach shows that reducing child marriage could significantly improve economic growth directly and indirectly, with education representing the primary channel.
- Quantitatively, a one standard deviation reduction in child marriage (16.74 percentage points) raises long-term annual per capita real GDP growth by 0.66 percentage points (simultaneous model) and could raise it by about 1.05 percentage points if child marriage were ended today (long-term linear interpretation).

_Italic: Source: IMF working paper (content unit spanning I. INTRODUCTION, II. MODELING CHILD MARRIAGE AND GROWTH, and III. CONCLUSIONS)._

### 0.66 percentage points. This means that if child marriage were ended today, the long-term

### wpiea2020027-print-pdf - 0.66 percentage points. This means that if child marriage were ended today, the long-term

### Key findings on growth and child marriage
- The estimate "0.66 percentage points." (text fragment present in the source).
- If child marriage were ended today, the long-term annual per capita growth in EMDCs would increase by 1.05 percentage points.
- Almost 80 percent of the effect of child marriage on growth is accounted for by past education.
- An improvement in the human capital index (reflecting years of schooling and returns to education) by one standard deviation (0.51) could raise per capita growth by 0.53 percentage points per year.
- Results support a strong relationship between child marriage and economic growth, acknowledging there is always some margin of error around cross-country econometric estimates.

### Policy recommendations and implications
- Prioritize education policies when developing comprehensive strategies to end child marriage, given the strong interdependent relationship between education and child marriage.
- Improve education quality and increase the number of years of schooling; complementary policies required include:
  - Social assistance to poor and marginalized children.
  - Addressing high rates of school dropout for girls due to pregnancy, including removing policies that expel pregnant girls from school.
- Strengthen healthcare policies because better health outcomes indirectly support higher levels of and returns to education, which help end child marriage.
- Stepped-up outreach and education to governments and affected communities (especially parents) on the costs of child marriage will be critical.

### Data availability (Appendix 1)
- Appendix 1 lists country-year availability of child marriage data for many countries (source: UNICEF Global Databases). Examples from the list include:
  - Afghanistan: 2013, 2015
  - Bangladesh: 1994, 1997, 2000, 2004, 2007, 2011, 2013, 2014
  - India: 1993, 1999, 2006, 2015, 2016
  - Nigeria: 1990, 2003, 2008, 2013, 2017
  - Ethiopia: 2000, 2005, 2011, 2016
  - Pakistan: 1991, 2006, 2012
  - Peru: 1986, 1992, 1996, 2000, 2004, 2007, 2009, 2010, 2011, 2012, 2014
  - (Full Appendix 1 provides a comprehensive country-by-year inventory.)

### Robustness checks and econometric results (selected)
- Basic growth equation (robustness checks):
  - Log of the per capita GDP coefficient ranges reported, e.g., -1.26***, -1.54***, -1.28***, -1.24*** (standard errors in parentheses).
  - Human Capital Index coefficients reported as 0.76***, 0.87***, 0.79*** (with standard errors).
  - Female school enrollment, secondary: 0.01** (0.00).
  - Fraction of tropical area: -0.61***, -0.39**, -0.51***, -0.61***.
  - Trade openness in IV regression: -0.01** (puzzling sign noted and attributed to possible selection bias due to small sample size of 54).
  - Observations and fit statistics: Observations 76, 54, 76, 72 across variants; R2 reported as 0.45, 0.48, 0.46, 0.44; P-value 0.00 in all reported variants.
- Regressions of child marriage on explanatory variables:
  - Log of the per capita GDP coefficients reported (e.g., -13.01*, -10.58, -13.04*, -12.30, -13.00*).
  - Human Capital Index (5-year lag): -11.27***, -11.11***, -12.52***, -11.32***.
  - Fraction of tropical area positive and significant across variants (e.g., 10.81***, 10.22***, 10.95***, 11.32***, 10.78***).
  - Observations typically 77 (with some variants 73); R2 reported as 0.56 across variants; P-value 0.00.
- Simultaneous equations model (selected coefficients and robustness):
  - Growth (t,t+T): Child marriage (t) effects reported as -0.04**, -0.06**, -0.11**, -0.03**, -0.06*** across models.
  - Log of the per capita GDP (t) in growth equation: -1.63***, -1.77***, -1.72***, -1.56***, -1.85***.
  - In the child marriage equation: Log of the per capita GDP (t) coefficients reported as -7.89**, -7.44**, -1.32, -8.35**, -8.37**; Human Capital Index (t-h) reported as -26.80***, -27.65***, -13.30***, -26.73***, -26.95***.
  - Human Capital equation: Life expectancy at birth (t-2h) 0.02***, 0.02***, 0.01*, 0.02***, 0.02***; Child marriage (t-2h) -0.02*** across variants.
  - Observations typically 74 (some variants 71); P-value 0.00 in reported variants.
- Additional robustness notes:
  - Defining child marriage as marriage before age 15 (instead of age 18) produces small differences: the effect of initial wealth on marriage before age 15 is not significant, and the marginal effect of the human capital index is smaller than in the baseline.
  - Removing the East Asia dummy in the growth equation marginally increases the long-run effect of health on growth.
  - Models controlling for institutions (rule of law, government effectiveness) were estimated; variables are scores from the Worldwide Governance Indicators database normalized so that 0 and 1 correspond to the worse and the best governance situations.

*Source: wpiea2020027-print-pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020027-print-pdf.pdf_