2. Background on coral bleaching

A coral reef is a marine ecosystem in tropical seas characterized by reef-building corals. Covering less than 0.1 per cent of the word's ocean area, coral reefs form some of the Earth's most diverse ecosystems as they provide food and shelter to about 25 per cent of the ocean's fish (Spalding and Grenfell, Reference Spalding and Grenfell1997; NOAA, 2019). During the past few decades, the coral reefs have been threatened by many factors ranging from rising ocean temperatures to human actions, imposing risks on both the marine ecosystem and humans.

In this paper, we focus on coral bleaching, a natural phenomenon in which coral reefs are weakened due mainly to abnormally high sea surface temperatures (SST). When the SST exceeds a threshold, corals release symbiotic algae, which photosynthesize and feed the corals (Brown, Reference Brown1997), resulting in bleached colors. If the SST reverts back to a normal range in a short period of time, corals can usually regain their colors, recover and survive. On the other hand, if the SST anomaly lasts over 4–6 weeks, the corals usually die (Hoegh-Guldberg, Reference Hoegh-Guldberg1999; Wilkinson and Hodgson, Reference Wilkinson and Hodgson1999), and the reefs may or may not recover. If they recover, the recoveries take at least five years (e.g., Wilkinson and Hodgson, Reference Wilkinson and Hodgson1999; Graham et al., Reference Graham, Wilson, Jennings, Polunin, Robinson, Bijoux and Daw2007; Gilmour et al., Reference Gilmour, Smith, Heyward, Baird and Pratchett2013). The recovery process begins with new coral larvae or polyps settling into the old reef structure. Then, these new corals grow, usually at a very slow rate (Barnes and Hughes, Reference Barnes and Hughes1999; Veron and Stafford-Smith, Reference Veron and Stafford-Smith2000).Footnote ⁴ In the worst case scenario, the reefs are permanently damaged, and the corals will be replaced by algae.

Severe coral bleaching events are usually associated with deteriorating fish stocks, with severity varying by species and location.Footnote ⁵ Coral bleaching usually leads to a decline in coral reef species within three years of the bleaching event, and coral bleaching can also cause distress for non-reef species through ecological relationships.

These adverse effects of coral bleaching on marine resources negatively impact humans in a variety of ways. First, coral bleaching might decrease catch and income in the fisheries sector as fish stocks decline. Frequent and severe bleaching events also impose risks on future food security because fish is an important source of animal protein especially in coastal communities. Second, bleached corals and deteriorating reef fish stocks are less attractive for recreational activities such as SCUBA diving and snorkeling. Finally, coral reefs play a crucial role in shoreline preservation, and weakened reefs may result in more land erosion in the future.

This paper focuses on the mass coral bleaching in 1998 and the fisheries sector in Indonesia. During the severe El Niño in 1997–1998, ocean temperatures in the tropical zone around the world rose significantly, leading to the world's first widespread mass coral bleaching.Footnote ⁶ Numerous coral bleaching spots were found in the Indian and the Pacific Oceans spanning from the eastern coast of Africa to as far as Australia. Wilkinson (Reference Wilkinson2000) estimates that 16 per cent of the world's corals were destroyed during this massive bleaching event. In Indonesia, coral bleaching was reported in West Sumatra, the south shore of Central Java, Bali and Lombok area, and Southern Sulawesi. The coral mortality rate in Bali area was estimated to be around 50 per cent (Goreau et al., Reference Goreau, McClanahan, Hayes and Strong2000). This bleaching episode is the first large-scale coral bleaching to be reported for Indonesia.Footnote ⁷ Households were unlikely to anticipate it because even scientists could predict it only a few days in advance (Hoegh-Guldberg, Reference Hoegh-Guldberg1999). For this reason, this bleaching event gives us a unique opportunity to study a climate shock prior to any adaptations.

Chaijaroen (Reference Chaijaroen2019) studies the economic impacts of the mass coral bleaching in 1998 on the Indonesian fisheries sector. This bleaching event is associated with a large reduction in income among the affected fishery households in 2000, two years after the event occurred. As a result of this income shock, the affected households experienced an overall reduction in consumption in 2000; the fall in protein consumption was the largest among all consumption types. In response to the shocks, the affected households migrated more in 2000, but they were not able to increase their labor supply or switch to a new industry until 2007. Given these large income and protein consumption shocks in Chaijaroen (Reference Chaijaroen2019), coral bleaching might have imposed significant risks on child development. This paper explores these risks and elaborates on how government policies can be used to mitigate them.

3. Empirical framework

The goal of this paper is to evaluate the effects of coral bleaching on fertility and child development outcomes. These effects are estimated using a triple differences (DDD) framework on data from a household survey and a scientific paper on coral bleaching (Goreau et al., Reference Goreau, McClanahan, Hayes and Strong2000). In this empirical framework, those affected by the 1998 coral bleaching, i.e., the treatment group, consist of households that worked in fisheries and lived in areas with reported coral bleaching in 1997. This treatment group is compared against all other households in the survey while taking into account differences across sectors and areas of residence. We will elaborate on this empirical framework in this section.

3.1. Data

This paper combines data from the IFLS household survey with data on coral bleaching and rainfall. The IFLS is a panel dataset that surveyed 7,224 Indonesian households in 1993. These households together with their spin-offs were resurveyed in 1997, 2000, 2007 and 2014. The IFLS covers 13 out of the 27 provinces of Indonesia and is representative of around 83 per cent of the Indonesian population. The dataset contains diverse socio-economic information on households, adult household members, children, women, and their communities. This paper utilizes data from the household, children, and women modules of the 1993, 1997, 2000 and 2007 waves of data (Frankenberg and Karoly, Reference Frankenberg and Karoly1995; Frankenberg and Duncan, Reference Frankenberg and Duncan2000; Strauss et al., Reference Strauss, Beegle, Sikoki, Dwiyanto, Herawati and Witoelar2004, Reference Strauss, Witoelar, Sikoki and Wattie2009).

The main source of coral bleaching data is the reported bleaching spots in Goreau et al. (Reference Goreau, McClanahan, Hayes and Strong2000). The paper collected coral bleaching reports from long-time reef observers around the world. This helps ensure that any damages reported were due to coral bleaching, not some other causes. However, it is possible that some coral bleaching might not get reported and potentially cause measurement errors. For this reason, SST anomalies from satellite maps are used as another measure of coral bleaching for robustness checks (see online appendix for details). Both measures of coral bleaching are merged with the IFLS data at a coastal area level, defined as one ocean coastline in one province. The exception to this definition is Bali and West Nusa Tengara – these two provinces are treated as one coastal area because they are small islands in the same area with similar SST anomalies.

Table 1 exhibits summary statistics based on the 1997 wave of data and shows the basic characteristics of our sample. Most households in the sample were led by middle-aged males with an average age of around 40 years old. Almost none of the household heads had a college degree. In the children sample, the summary statistics for anthropometric outcomes suggest that the children's health was below the World Health Organization (WHO) standards as indicated by the negative averages of height-for-age and weight-for-height $z$-scores in 1997. The enrollment rate for elementary school averaged 95 per cent for all children. In the women sample, above 80 per cent were married, and almost none had a college degree. When comparing the treatment group with the rest of the sample, we find that the treatment group differed from the rest on some fronts. These differences and how they affect our identification will be further discussed towards the end of this section.

Table 1.

Summary statistics for key variables by groups before coral bleaching

Notes. $P$-values are from unpaired t-tests for differences in means between the treatment group and the control group. Mother's height is in centimeters. Education and age are in years. Height-for-age and weight-for-height are $z$-scores.

4. Fertility

The literature on the effects of shocks on fertility offers a mixed set of findings – some shocks lead to an increase in fertility, while others cause fertility to fall. This section explores how coral bleaching, which leads to income and protein consumption shocks, affects fertility in the short and medium terms. We use the first four waves of the IFLS – 1993, 1997, 2000 and 2007 – and a DDD framework to estimate the effects. Timing wise, the 1993 and 1997 waves were surveyed before coral bleaching happened in 1998, so they serve as pre-treatment waves. The 2000 wave data were collected roughly two years after the bleaching event, so this wave of data allows for the estimation of short-run effects. The 2007 wave then reveals information on medium-run effects. Our sample includes all women, regardless of their exposures to the fisheries sector and the coral bleaching event, provided that they are of child-bearing age. For each woman $i$ living in a household $h$ and province $p$, let $Y_{it}$ be the woman's number of births given in period $t$. Then, the main estimating equation can be written as

(1)\begin{align} Y_{it} & = \beta_0 + \sum_{w = 2000, 2007} (\beta^w_1 CB_p Fish_h Post^{w}_t + \beta_2^w Fish_h Post^{w}_t \nonumber\\ & \quad + \beta_3^w CB_p Post^{w}_t + \beta_{4}^w Post^{w}_t)+\beta_5 CB_p Fish_h + \beta_6 CB_p \nonumber\\ & \quad + \beta_7 Fish_h + X'_{iht} \delta +\gamma_h + \eta_p + \lambda_t + \epsilon_{it}, \end{align}

where $CB_p$ is a dummy indicator for living in the coral bleaching area in 1997, $Fish_h$ is a dummy indicator for working in the fisheries sector in 1997, and $Post^{w}_t$ are dummy indicators for waves 2000 and 2007. $X_{iht}$ is a vector of household and individual control covariates which include the woman's age, education, and marital status as well as the household head's sex, age, and education. $\gamma _h$, $\eta _p$, and $\lambda _t$ are household, province, and wave fixed effects, respectively. Our main coefficients of interest are $\beta ^{2000}_1$ and $\beta ^{2007}_1$, which illustrate how coral bleaching changes $Y_{it}$ in the treatment group relative to all other households while taking into account the differences across the sectors (fisheries vs non-fisheries) and the areas (bleaching vs non-bleaching).

Pregnancy history data in the ever-married women modules are used in the main estimation in this section. One caveat of this set of data is that not all of the women who had ever been married in the sampled households were surveyed. In the first wave of the IFLS, only the household head, his/her spouse, and a random 25 per cent of the remaining adult household members were selected for adult individual-level interviews. This sampling scheme was expanded to cover more individuals in the subsequent waves but still did not cover every ever-married child-bearing-age woman. For this reason, a similar analysis is performed using birth information from household rosters to ensure that results are robust. These alternative results are very similar to the main results and are available upon request.

We focus on live births that occurred within 19 months of the earliest survey date in each wave. This 19-month window is derived from the timing of the shock in 1998 in relation to the earliest survey date in 2000. Specifically, only babies that were conceived during or after coral bleaching (February 1998) should be considered as part of the treatment group, so the birth window for 2000 is restricted to 9 months after February 1998 or November 1998 onward. The 2000 IFLS was first surveyed in June 2000, so this time restriction yields a window of 19 months before the first survey date. In addition, only women in the child-bearing age range of 15–49 years old are included in our analysis.

Table 2 presents regression results on live birthsFootnote ¹⁰ based on (1). $\hat {\beta }_1^{2000}$, i.e., the coefficient on $Bleach*Fish*Post^{2000}$, in all columns suggest that women in the treatment group had around 0.13–0.17 more children than other women in 2000.Footnote ¹¹ This is a significant increase considering the average births of 0.11 per woman in the treatment group in 1997. $\hat {\beta }_1^{2007}$, on the other hand, are not statistically significant at the usual significance levels.

Table 2.

Effects of coral bleaching on fertility

Notes. Province-clustered standard errors are in parentheses. *, **, and *** denote statistical significance at the 0.1, 0.05, and 0.01 levels, respectively. The sample is women who were 15–49 years old in the 1993, 1997, 2000 and 2007 waves of data. The dependent variable is the number of children born within 19 months of the earliest interview date. All models include the woman's age, education, and marital status. Columns 2 and 4 also include household head's sex, education, and age. Wave and province fixed effects are included in all models. Columns 3 and 4 also contain household fixed effects.

The results presented in this section are consistent with one of the scenarios in Gary Becker's seminal fertility model. Becker and Lewis (Reference Becker and Lewis1973) propose a framework linking income to the quantity and quality of children. In their model, a decrease in income lowers the demand for children through the direct income effect. The decrease in income would also result in a fall in child quality and hence would decrease the shadow price of the quantity of children. The lowered shadow price could then lead to an increase in the quantity of children through the substitution effect. The empirical results in the next section suggest that the quality of children declined after a fall in income, and the results presented in this section support the case where the substitution effect is greater than the income effect. That is, the affected households might have substituted the quantity over the quality aspects when they make their reproductive decisions.

The positive effect of coral bleaching on births is consistent with the empirical literature on certain shocks. In the United States, Dehejia and Lleras-Muney (Reference Dehejia and Lleras-Muney2004) study fertility and unemployment rates and find that the substitution effect is strong for white mothers with low-education. Consequently, more births are observed from this group of mothers as the unemployment rate increases. Some shocks are also associated with a temporary increase in fertility. For example, Hurricane Mitch was associated with higher odds of births in Nicaragua (Davis, Reference Davis2017). Blackouts can also cause an increase in fertility (e.g., Burlando, Reference Burlando2014a, Reference Burlando2014). In both cases, the increase in fertility is due to a timing shift of births rather than an increase in the total number of births.

The increase in fertility in this paper, however, is not in line with some findings in the literature on Indonesia. Kim and Prskawetz (Reference Kim and Prskawetz2010) study different kinds of shocks and find that most of the shocks do not affect fertility. Only unemployment is associated with an increase in fertility. Sellers and Gray (Reference Sellers and Gray2019) find that delays in monsoon onset led to a small increase in fertility intention and a fall in births. Both papers conjecture that households attempted to use the quantity of children to smooth consumption in times of shocks.

The observed increase in fertility in this paper is likely a temporary shift of fertility timing rather than a permanent increase in the quantity of children as a means for consumption smoothing. We find evidence for an increase in contraceptive use and a decrease in fertility intention in 2000. In addition, the total fertility in 2007 does not differ between the treatment and control groups.Footnote ¹² The affected households were likely to view coral bleaching as a temporary shock during which the opportunity cost of children has gone down, so they decided to bring forward their birth timing.

The difference in the findings on fertility may also stem from inherent differences between coral bleaching and other shocks. First, rainfall shocks and droughts affect women more than coral bleaching does. Both men and women work in agriculture in Indonesia, while fisheries is a male-dominant industry. With deteriorated marine resources, households in fisheries did not increase their labor supply in 2000 but did so in 2007 (Chaijaroen, Reference Chaijaroen2019). This implies a lower opportunity cost of children among the fishery households relative to those in agriculture, especially in 2000. Second, the adverse effects of climate shocks in fisheries might be easier to mitigate than those in agriculture. A severe drought may slash a wide variety of the food supply – from rice, a main food staple, to fruits, vegetables, and plant-based animal feed. A decline in availability of protein from fish, on the other hand, might be compensated by other sources of protein or even other food groups. Agriculture is also a much larger industry than fisheries, so shocks in the agriculture sector are more likely to spread and spill to other industries.

Nonetheless, evidence on child outcomes, to be discussed in the next section, suggests that the affected households were not able to fully overcome the adverse effects of coral bleaching. The effects of coral bleaching, either on marine resources or on humans, are longer-term and harder to be reversed than those of most shocks in the literature. For example, rainfall varies every growing season, and blackouts, even the longest ones, last only months. Humans can rebuild after natural disasters, but humans cannot directly mend the deteriorated marine resources. Therefore, our finding on fertility raises the questions of whether the affected households might have misconceived the effects of coral bleaching as short-term and easily recoverable, and whether policy makers should have stepped in to help with fertility planning.

5. Child outcomes

This section discusses the effects of coral bleaching on child development with a focus on anthropometric and schooling outcomes. The results in this section generally suggest that child development has been hampered after the coral bleaching in 1998, at least in the short run when the affected households experienced income and protein consumption shocks. By 2007, the income shock has recovered, but some evidence still indicates that the affected children were more likely to fail a grade in school despite a higher enrollment rate than the other children.

5.1. Anthropometric outcomes

The anthropometric outcomes considered in this paper are standardized height-for-age and weight-for-height, and dummy indicators for severe stunting and wasting during the first five years of life. Severe stunting is defined as the height-for-age that is at least three standard deviations below the median. Similarly, severe wasting is when the weight-for-height is at least three standard deviations below the median. The standardization of weight and height is based on the WHO scales which are available for ages up to five years old and for height up to 120 centimeters.

Data from the children modules in the 1997 and 2000 waves of the IFLS are used to estimate the effects of coral bleaching on child anthropometric outcomes. The estimation is based on a triple differences framework in which the sample includes children who were 0–5 years old in the 1997 and 2000 waves of data. The treatment group consists of children who were born to the households affected by coral bleaching between 1995–2000.

It is worth noting that we cannot follow the common identification strategy in the early life shock literature where the exact age at the time of shock is interacted with the shock exposure because the exact timing of the shocks from coral bleaching is unknown. According to the marine science literature, the effects of coral bleaching on fish stocks take time to manifest; however, the exact length of time varies by reef site and depends on reef conditions, local fish stock profile, and bleaching severity. In the case of the 1998 bleaching event in Indonesia, there were no data or studies on fish stock to the best of our knowledge. It is only known that income and protein consumption shocks were realized by 2000 (Chaijaroen, Reference Chaijaroen2019).

Let $Y_{it}$ be the outcome of interest for child $i$ living in household $h$, province $p$, and period $t$, then the main estimating equation takes the form

(2)\begin{align} Y_{it} & = \beta_0 + \beta_1 CB_p Fish_h Post_t + \beta_2 CB_p Fish_h + \beta_3 CB_p Post_t \nonumber\\ & \quad + \beta_4 Fish_h Post_t +\beta_5 CB_p + \beta_6 Fish_h + \beta_7 Post_t + X'_{iht} \delta + \gamma_h + \epsilon_{it}, \end{align}

where $CB_p$ is a dummy indicator for living in coral bleaching areas in 1997, $Fish_h$ is a dummy indicator for the household's working in the fisheries sector in 1997, and $Post_t$ is a dummy indicator for the 2000 wave. $X_{iht}$ is a vector of household and individual control covariates which include the child's sex, age and race; the household head's sex, age and education; and the mother's education and height. Finally, $\gamma _h$ denotes household fixed effects.

Table 3 illustrates the results on anthropometric outcomes. To control for household-specific time-invariant factors such as upbringing and genetic factors, household fixed effects are included in columns 5–8. The results in columns 1 and 5 suggest that the 1998 coral bleaching does not statistically affect weight-for-height. Column 2 indicates that the bleaching event is associated with about half a standard deviation decrease in height-for-age. However, when the household fixed effects are included in column 6, the effect is not statistically significant. Both models for wasting (columns 3 and 7) suggest that coral bleaching might be associated with an increase in the likelihood of severe wasting, but the coefficients are not statistically significant. Finally, columns 4 and 8 point toward an increase in the likelihood of severe stunting after coral bleaching. Specifically, the results in column 8 indicate that children exposed to coral bleaching were 29.6 percentage points more likely to be severely stunted than other children after controlling for household time-invariant unobservables.

Table 3.

Effects of coral bleaching on anthropometric outcomes in 2000

Notes. Province clustered standard errors are in parentheses. *, **, and *** denote statistical significance at the 0.1, 0.05, and 0.01 levels, respectively. The sample is children who were 0–5 years old in the 1997 and 2000 waves. The dependent variables are standardized weight-for-height (WHZ) and height-for-age (HAZ), and dummy indicators for severe malnutrition based on the WHZ and HAZ ($Z < -3$, wasting and stunting, respectively). All models include the child's sex, age, race; the household head's sex, age, and education; as well as the mother's education and height as control covariates. Age and province fixed effects are included in all models. Models 5-8 also contain household fixed effects.

5.2. Schooling outcomes

Early-life shocks and their effects on health outcomes in the first few years of life can translate into poor schooling and other outcomes later in life (e.g., Kim and Prskawetz, Reference Kim and Prskawetz2010). In this section, we investigate how an exposure to coral bleaching in early childhood affects three schooling outcomes: 1) current enrollment status ($Enroll$), 2) whether a child has ever enrolled in school ($Ever$ $Enrolled$), and 3) whether a child has ever failed a grade in school ($Fail$). These effects are estimated using a triple differences framework on the 1997 and 2007 waves of the IFLS. The models in this section compare elementary school age children (7–12 years old) in 2007 with those of the same ages in 1997. The estimating equation is similar to (2) except that $Post_t$ is now a dummy indicator for wave 2007. $X_{iht}$ in this section includes the child's age dummy indicators and sex as well as the household head's sex, age, and education.

Table 4 contains regression results for the aforementioned schooling outcomes. Columns 1–4 show the results from models without household fixed effects while columns 5–7 illustrate the results with the household fixed effects. The treatment coefficients in the enrollment models are all positive, implying that an improvement in income after coral bleaching might increase school enrollment. For example, column 1 suggests that coral bleaching increases the chance that a child is currently enrolled in school by 23.4 percentage points.

Table 4.

Effects of coral bleaching on schooling outcomes in 2007

Notes. Province clustered standard errors are in parentheses. *, **, and *** denote statistical significance at the 0.1, 0.05, and 0.01 levels, respectively. The sample is children who were 7–12 years old in the 1997 and 2007 waves. The dependent variables are dummy variables equal to 1 if a child is currently enrolled in school, if a child has ever enrolled in school, and if a child has ever failed a grade in school. All models include the child's sex, and the household head's sex, age, and education. Age (in years) and province fixed effects are included in all models. Models 5–7 also contain household fixed effects.

On the performance side, the treatment coefficients for $Fail$ are all positive. This, together with the results on enrollment, implies that children who had been exposed to coral bleaching were more likely to fail a grade in school even when they had an equal or even better chance to be in school. The lower performance might be due to two plausible reasons. First, an early childhood exposure to coral bleaching could have led to poor nutrition and health outcomes, imposing long-term risks on later life outcomes. Second, parents might have had less time to care for their children in 2007 as they started increasing their work hours.

6. Comparison with rainfall shocks and policy implications

In this section, coral bleaching is compared against one of the most common economic shocks in the literature, rainfall shocks. Compared to rainfall shocks, coral bleaching is more long-term and novel, so the known policy implications in this literature might not apply to coral bleaching. Specifically, rainfall shocks usually last just a few months whereas the effects of coral bleaching on fish stocks span several years or even become permanent. Humans have experienced rainfall variations for as long as we have existed as a species, and we have learned to adapt in various ways to these shocks. Coral bleaching, on the other hand, is relatively new. The first known coral bleaching events were recorded in the 1980s, and the more severe ones only started in the late 1990s. With climate change, more long-term and novel shocks like coral bleaching are expected to occur. We should therefore investigate how these shocks differ from those studied in the current literature to shed light on relevant policies that would mitigate similar future climate shocks.

To estimate the effects of rainfall shocks, data from the IFLS are merged with rainfall data from the University of Delaware Air Temperature & Precipitation database. We then follow Maccini and Yang (Reference Maccini and Yang2009) and construct the rainfall variable as the total rainfall in each child's birth year.Footnote ¹⁴ When estimating the effects of the rainfall shocks, we allow the effects to depend on whether a household is agriculture-based. Let $Y_{it}$ be the outcome of interest for child $i$ living in household $h$ and year $t$, then the estimating equation can be written as

(3)\begin{equation} Y_{it} = \beta_0 + \beta_1 Rain_{it} + \beta_2 Agri_{ht} + \beta_3 Rain_{it}*Agri_{ht} + X'_{iht}\delta + \gamma_h +\epsilon_{it}, \end{equation}

where $X_{iht}$ is a vector of control covariates, and $\gamma _h$ is household fixed effects.

Table 5 contains key coefficients from (3) for anthropometric outcomes and illustrates the lesser effects of rainfall relative to coral bleaching. In particular, a transitory rainfall shock at birth only affects weight, a measure of short-term growth, but not other anthropometric measures. Specifically, the only treatment coefficient that is statistically significant at conventional significance levels is that on the interaction term in the WHZ model without household fixed effects (column 1), suggesting that rainfall has a positive effect on weight only among agriculture-based households. The average rainfall in this sample is 184 centimeters and the standard deviation is 48.97 centimeters, so a one standard deviation increase in rainfall is associated with an increase in weight by 0.17 standard deviations.

Table 5.

Effects of birth year rainfall on anthropometric outcomes in 2000

Notes. Clustered standard errors are in parentheses. * and *** denote statistical significance at the 0.1 and 0.01 levels, respectively. The sample is children who were 0–5 years old in the 2000 wave of data. The dependent variables are standardized weight-for-height and height-for-age as well as dummy indicators for severe malnutrition based on the WHZ and HAZ ($Z < -3$). $Rainfall$ is birth year wet season rainfall in centimeters at a province level. $Agri$ is a dummy variable equal to 1 if the household engaged in agriculture in 2000. All models include the child's sex and race, the household head's sex, age, and education, as well as the mother's education and height as control covariates. Age (in years) and province fixed effects are included in all models.

This effect of rainfall on weight is relatively small compared to the effects of coral bleaching on height, a measure of long-term growth, and severe malnutrition. For example, coral bleaching increases the likelihood of severe malnutrition (severe wasting and severe stunting) by 4–30 percentage points. Coral bleaching is also associated with a 0.53 standard deviation decrease in standardized height, while the effects of birth year rainfall on standardized height and severe malnutrition are not statistically significant.

Relative to rainfall in birth year, an early-life exposure to coral bleaching also has stronger effects on schooling outcomes. Table 6 contains key coefficients from (3) for schooling outcomes. Similar to most of the results in table 5, the coefficients in this table are not statistically significant.

Table 6.

Effects of birth year rainfall on schooling outcomes in 2007

Notes. Province clustered standard errors are in parentheses. *** denotes statistical significance at the 0.01 level. The sample is children who were 7–12 years old in 2007. The dependent variables are dummy indicators equal to 1 if a child is currently enrolled in school, if a child has ever enrolled in school, and if a child has ever failed a grade in school. $Rainfall$ is birth year wet season rainfall in centimeters at a provincial level. $Agri$ is a dummy variable equal to 1 if the household engaged in agriculture in 2000. All models include the child's sex, and the household head's sex, age, and education. Age (in years) and province fixed effects are included in all models.

The results on rainfall shocks and anthropometric outcomes are consistent with Reference Cornwell and InderCornwell and Inder's (Reference Cornwell and Inder2015) nutritional effect finding in which an increase in rainfall boosts food availability and improves child health in rural Indonesia. Both the anthropometric and schooling results in this paper, however, are not consistent with the findings in Maccini and Yang (Reference Maccini and Yang2009), due probably to structural differences across time. Maccini and Yang (Reference Maccini and Yang2009) focuses on adults aged 25 and above in 2007. Their sample was exposed to the rainfall shocks at least 18 years before the sample in this paper. In a developing country setting, many factors could have changed over this time span. For example, irrigation and technological advancement might have lessened any adverse effects that low rainfall might have on agricultural income. In addition, healthcare might have significantly improved over the years, so small shocks in nutrition can be smoothed out.

What is also striking about the comparison of coral bleaching with rainfall shocks is the increase in fertility following the exposure to coral bleaching. The rainfall shocks in the Indonesian literature, on the other hand, are associated with a decrease in fertility (Sellers and Gray, Reference Sellers and Gray2019). Both coral bleaching and rainfall shocks hamper child development, but the effects are greater in the case of coral bleaching. This finding then raises the question of whether the effects of coral bleaching were underestimated, and stronger policy interventions should have been implemented.

Given that the increase in fertility is a timing shift of births, rather than an increase in total fertility, policies should mostly focus on providing support for mothers and children. Policy on fertility, if any, should emphasize the demand side as data on contraceptive use do not show any supply-side problems. Specifically, the policy should educate parents about the possible long-term effects of coral bleaching so they could correctly assess the costs associated with having children.

Right around the time of the coral bleaching in 1998, several policies were rolled out to fight against poverty and economic hardship following the 1997 Asian Financial Crisis, yet these policies were not greatly effective in eliminating the adverse effects of coral bleaching on child development. According to Daly and Fane (Reference Daly and Fane2002), the largest anti-poverty program during that time was the subsidized rice program OPK (Operasi Pasar Khusus), aiming to provide nutrition to the poor. Scholarship and school grants as well as healthcare cards and a supplementary nutrition program were also implemented during the same time. However, the results in this paper still indicate some malnutrition for those at the lower end of the distribution as well as an increased likelihood of failing a grade in school, suggesting that the existing policies were not particularly effective in mitigating the effects of coral bleaching.

There are at least three main problems with the policies during the 1997 crisis. First, the policies could not reach some of the poorest due to several glitches in policy coverage and targeting (Daly and Fane, Reference Daly and Fane2002). Second, policies aiming to mitigate malnutrition following coral bleaching should have emphasized protein supplement, but the main food subsidy program at that time focused only on rice. Third, increasing enrollment through scholarships and school grants alone might not suffice in improving education outcomes. Future policies should therefore address these issues.

Since the Asian Financial Crisis, many other effective programs have been implemented in Indonesia; some of them could potentially help mitigate the effects of coral bleaching. Given that coral bleaching affected children through both the reductions in income and nutrition intakes, a conditional cash transfer (CCT) program with conditions on healthcare and school participation can be effective. For instance, Program Keluarga Harapan (PKH), a household-level CCT program, was rolled out in 2007 and contained conditions on iron supplement, postnatal care, child immunization and health checkups, and school attendance. In the same year, the government also started a village-level CCT program, Generasi, which provided annual block grants to support health and education. These programs help improve preventive healthcare utilization, increase protein consumption and nutrition, as well as decrease wasting and stunting (Kusuma et al., Reference Kusuma, McConnell, Berman and Cohen2007a, Reference Kusuma, Thabrany, Hidayat, McConnell, Berman and Cohen2007; Aizawa, Reference Aizawa2020; Cahyadi et al., Reference Cahyadi, Hanna, Olken, Prima, Satriawan and Syamsulhakim2020). If these programs had been rolled out right after the coral bleaching event, some adverse effects of the shocks might have been mitigated.

In summary, the comparison of the 1998 coral bleaching to birth-year rainfall shocks suggests that the coral bleaching event had larger negative effects on child development. This, together with the increase in fertility after the event, calls for better policy interventions. Based on a literature survey of policies in Indonesia and the findings in this paper, policies addressing coral bleaching should focus on protein intakes, preventive healthcare for mothers and children, and student performance in school. For example, a CCT with conditions on those aspects could work well. This policy implication is directly applicable to climate shocks that result in nutrition and income losses, such as ocean acidification and shocks that permanently change agricultural production patterns. The income aspects of the results in this paper also apply to other long-term income shocks such as a permanent job loss.