Storms, early education, and human capital

Pelli, Martino; Tschopp, Jeanne Working Paper Storms, early education, and human capital ADB Economics Working Paper Series, No. 743 Provided in Cooperation with: Asian Development Bank (ADB), Manila Suggested Citation: Pelli, Martino; Tschopp, Jeanne (2024) : Storms, early education, and human capital, ADB Economics Working Paper Series, No. 743, Asian Development Bank (ADB), Manila, https://doi.org/10.22617/WPS240462-2 This Version is available at: https://hdl.handle.net/10419/305430 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ ASIAN DEVELOPMENT BANK ASIAN DEVELOPMENT BANK 6 ADB Avenue, Mandaluyong City 1550 Metro Manila, Philippines www.adb.org STORMS, EARLY EDUCATION, AND HUMAN CAPITAL Martino Pelli and Jeanne Tschopp ADB ECONOMICS WORKING PAPER SERIES NO. 743 October 2024 Storms, Early Education, and Human Capital This paper examines the impact of school-age exposure to storms on education and employment outcomes in India. Using wind exposure histories, the paper shows that exposure to an average storm can cause a 2.4 percentage point increase in educational delays, a 2 percentage point drop in post-secondary education attainment, and a 1.6 percentage point decline in regular salaried employment. The paper also highlights the role of damaged school infrastructure and declining household income in driving these outcomes. About the Asian Development Bank ADB is committed to achieving a prosperous, inclusive, resilient, and sustainable Asia and the Pacific, while sustaining its efforts to eradicate extreme poverty. Established in 1966, it is owned by 68 members —49 from the region. Its main instruments for helping its developing member countries are policy dialogue, loans, equity investments, guarantees, grants, and technical assistance. ASIAN DEVELOPMENT BANK The ADB Economics Working Paper Series presents research in progress to elicit comments and encourage debate on development issues in Asia and the Pacific. The views expressed are those of the authors and do not necessarily reflect the views and policies of ADB or its Board of Governors or the governments they represent. ADB Economics Working Paper Series Storms, Early Education, and Human Capital Martino Pelli and Jeanne Tschopp No. 743 | October 2024 Martino Pelli ([email protected]) is a senior economist at the Economic Research and Development Impact Department, Asian Development Bank. Jeanne Tschopp ([email protected]) is an assistant professor at the Department of Economics, University of Bern. Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) © 2024 Asian Development Bank 6 ADB Avenue, Mandaluyong City, 1550 Metro Manila, Philippines Tel +63 2 8632 4444; Fax +63 2 8636 2444 www.adb.org Some rights reserved. Published in 2024. ISSN 2313-6537 (print), 2313-6545 (PDF) Publication Stock No. WPS240462-2 DOI: http://dx.doi.org/10.22617/WPS240462-2 The views expressed in this publication are those of the authors and do not necessarily reflect the views and policies ofthe Asian Development Bank (ADB) or its Board of Governors or the governments they represent. ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The mention of specific companies or products of manufacturers does not imply that they are endorsed or recommended by ADB in preference to others of a similar nature that are not mentioned. By making any designation of or reference to a particular territory or geographic area inthis document, ADB does not intend to make any judgments as to the legal or other status of any territory or area. This publication is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) https://creativecommons.org/licenses/by/3.0/igo/. By using the content of this publication, you agree to be bound bytheterms of this license. For attribution, translations, adaptations, and permissions, please read the provisions andterms of use at https://www.adb.org/terms-use#openaccess. This CC license does not apply to non-ADB copyright materials in this publication. If the material is attributed toanother source, please contact the copyright owner or publisher of that source for permission to reproduce it. ADB cannot be held liable for any claims that arise as a result of your use of the material. Please contact [email protected] if you have questions or comments with respect to content, or if you wish toobtain copyright permission for your intended use that does not fall within these terms, or for permission to use theADB logo. Corrigenda to ADB publications may be found at http://www.adb.org/publications/corrigenda. ABSTRACT This paper explores how school-age exposure to storms impacts the education and primary activity status of young adults in India. Using a cross-sectional cohort study based on wind exposure histories, we find evidence of a significant deskilling of areas vulnerable to climate change-related risks. Specifically, our results show a 2.4 percentage point increase in the probability of accruing educational delays, a 2 percentage point decline in post-secondary education achievement, and a 1.6 percentage point reduction in obtaining regular salaried jobs. Additionally, our study provides evidence that degraded school infrastructure and declining household income contribute to these findings. Keywords: climate change, storms, education, human capital JEL codes: I25, O12, Q54 We are grateful to the Social Sciences and Humanities Research Council (SSHRC, grant number 039367), and the Swiss National Foundation (SNF, grant number 100018_192553) for their financial support. We thank, without implicating them, Teevrat Garg, Blaise Melly, Ben Sand, Eric Strobl. All remaining errors are ours. 1 Introduction Lowand middle-income nations confront an inordinately elevated risk of catastrophes, as they are both more exposed to climate-related hazards and possess diminished resilience (Dell et al., 2014).1This deficiency in resilience renders children particularly susceptible, a concerning fact given that extreme weather events are anticipated to escalate due to climate change (IPCC, 2023; Emanuel, 2021). Our study delves into the enduring adverse consequences of school-age exposure to tropical storms and cyclones on education and pursuits during early adulthood in India. We scrutinize potential causal pathways spanning the school years, a crucial phase demonstrated to shape lifetime earnings (e.g., Oreopoulos, 2007; Angrist and Krueger, 1991). In this paper, we begin by evaluating the ramifications of storm exposure during school-age on educational outcomes in both the short and long run, using a cross-sectional cohort study based on the 2018 release of the Periodic Labour Force Survey (PLFS). We assess educational outcomes by considering years of schooling and the highest educational level achieved. To capture the cumulative effects of storms throughout school years, we devise a continuous treatment that aggregates wind exposure histories for each district and cohort born between 1985-1995. Contrasting with other environmental impact studies (e.g., Ebenstein et al., 2016; Deuchert and Felfe, 2015), our focus centers on long-term exposure to storms, which enables us to gauge the consequences of climate change as opposed to mere weather variability. Our results indicate that long-run storm exposure during school years causes educational delays, and exerts significant lasting effects on educational attainment and career choices in early adulthood. An average storm exposure during school years yields an increased likelihood of experiencing an educational delay by 2.4 percentage points, which translates to a 7.25% rise in the fraction of delayed individuals. We also observe a notable decline in the number of individuals attaining post-secondary education. An average exposure leads to a decrease in this probability by 2 percentage points, corresponding to a 7.35% reduction in the proportion of individuals with this level of education. Furthermore, we identify detrimental effects of storm exposure on labor market outcomes. An average school-age exposure results in a 1.6 percentage-point decrease 1Developing countries’ reduced resilience stems from factors such as insufficient infrastructure, weak social safety nets, market failures like absent credit and insurance markets, and an absence of effective early warning systems and comprehensive disaster risk management. Refer to Hallegatte et al. (2020) for a comprehensive review of why economically disadvantaged populations are disproportionately impacted by natural hazards and related disasters. in the proportion of individuals employed as regular workers, while concurrently causing a similar increase in the share of individuals occupied with domestic duties. These statistics equate to an 8% reduction in the share of individuals employed as regular workers and a 4.8% rise in the share of individuals involved in domestic duties. Additionally, we observe that an average exposure incurs a 3.9% reduction in hourly wages. Severe tropical storms, once uncommon but now progressively prevalent due to climate change, amplify all these findings. Our results remain robust through a series of checks, encompassing a falsification test and alternative specifications of school-age exposure to storms. We also demonstrate that our estimates are not influenced by early-life exposure to storms or other environmental factors such as precipitation and temperature during ages 5-15. These findings show that prolonged exposure to extreme environmental shocks during school years may contribute to the progressive deskilling of regions more susceptible to climate-change related risks. This degradation of skills will exacerbate inequalities, underscoring additional costs of climate change that have not been extensively examined thus far. To devise appropriate mitigating policies, we must comprehend the underlying pathways that drive our primary results. In the second part of the paper, we employ supplementary datasets (Consumer Pyramids and District Information System for Education) to probe the short-term mechanisms by which storms may influence long-term educational outcomes, focusing on their impact on household income and school infrastructure damage. Our findings reveal the presence of both demand and supply shocks in the schooling sector. Firstly, using panel local projections, we find that following an average storm, household income progressively declines, reaching levels approximately 8% below pre-disaster incomes 10 months post-shock. Secondly, we show that school closures significantly escalate in the aftermath of an average storm, with the proportion of closed schools surging by 7.4% within two years. However, the impact of an average exposure on the proportion of well-maintained classrooms and reliable electricity availability at schools is modest, albeit statistically significant. Lastly, we observe a reduction in primary school attendance and a decline in academic performance among middle school students, which is consistent with a negative income shock and a reduction in schooling demand. These findings offer indirect evidence that the enduring consequences of storms on education extend beyond the mere physical damages to schools, highlighting the significance of broadening post-disaster policies beyond reconstruction efforts and enhancing social safety nets (see Deryugina, 2017, for the importance of social safety nets in developed countries). Specifically, our results propose that financial transfers 2 ought to be paired with policies advocating sustained education and post-disaster school enrollment, potentially by conditioning cash transfers on school attendance. Furthermore, social policies, such as unemployment insurance, could prove instrumental in fostering resilience and risk management in urban areas, which are observed to be particularly vulnerable following storms. Our paper enhances the body of research examining the economic consequences of environmental disturbances during childhood in developing nations. Thus far, the literature on this subject has concentrated on three primary analytical approaches. Firstly, a significant portion of research investigates the short-term effects of concurrent shocks (e.g., Spencer et al., 2016; Björkman-Nyqvist, 2013; Jensen, 2000). Secondly, another branch of literature explores the deleterious repercussions of environmental shocks experienced in utero or early life (up to 4 years old). These shocks have been linked to a range of short and long-term outcomes, including various aspects of adult life, such as health, education, wealth, and offspring outcomes (e.g., Chang et al., 2022; Hyland and Russ, 2019; Rosales-Rueda, 2018; Akresh et al., 2017; Dinkelman, 2017; Maccini and Yang, 2009). The third analytical approach delves into the long-term consequences of short-term incidents occurring later in life, beyond ages 0-4. This line of research typically focuses on either singular events, as illustrated by Shidiqi et al. (2023), Groppo and Kraehnert (2017), and Deuchert and Felfe (2015), or on short-term occurrences coinciding with crucial moments for individuals, such as high-stakes exam days, as seen in studies by Park (2022) and Ebenstein et al. (2016). Our paper makes two main contributions to this body of research. First, by examining long-term exposure, our findings indicate a potential progressive deskilling in areas more susceptible to climate change-related risks – an additional cost associated with climate change that has not yet been emphasized in the literature. Second, we concentrate on the impact of these risks during the school-age years, highlighting the significance of this period in shaping human capital formation. Recognizing the distinction between infancy and school-age exposure is crucial, as the mechanisms through which adverse shocks affect long-term education likely differ. While in utero disruptions are known to influence human capital via children’s health, disasters during school-age years are more likely to impact long-term educational attainment through changes in household income and schooling infrastructure. Our study is most closely related to Deuchert and Felfe (2015) and Shidiqi et al. (2023). Deuchert and Felfe (2015) investigates the shortand long-term effects of Super Typhoon Mike on educational outcomes in Cebu Island, Philippines. The study reveals a 3 negative and enduring impact on education, alongside a gradual reallocation of funds from education to reconstruction efforts. We build upon these findings by concentrating on long-term exposure through a continuous measure rather than a binary damage indicator, offering valuable insights into the labor market outcomes of affected children in early adulthood and presenting an in-depth analysis of the factors underlying long-term educational delays.2In a similar vein, Shidiqi et al. (2023) examines the aftermath of a potent earthquake in Yogyakarta, Indonesia, in 2006, observing its repercussions on educational delays and attainments. Like cyclones, the earthquake instigated educational setbacks and decreased the likelihood of fulfilling mandatory education. However, contrary to cyclones, earthquakes appear to hinder education predominantly through the destruction of school facilities. Lastly, our study contributes to the literature by conducting an extensive examination of the mechanisms connecting school-age exposure to storms with long-term human capital degradation, specifically focusing on income and schooling infrastructure channels. To our knowledge, these channels have predominantly been explored independently (see Baez et al., 2010, for a review). Our findings on the income channel relate to the body of research that reveals the negative impact of economic recessions on education (e.g., Stuart, 2022). The majority of studies addressing this channel have employed difference-in-difference estimations. In contrast, local projections offer advantages in cases of multiple treatments, making them a suitable alternative to the difference-in-difference approach for addressing dynamic treatment effects. To the best of our knowledge, few papers have utilized this methodology to study environmental shocks.3 Although it is undeniable that disasters lead to infrastructure disruptions and damages (see e.g. Hallegatte et al., 2019), there are limited studies specifically estimating the effects of disasters on educational facilities. Our results indirectly highlight the importance of adequate school infrastructure for long-term education and labor 2In a related study, Shah and Steinberg (2017) employs rainfall as a proxy for wages in rural India, demonstrating that higher rainfalls, which correlate with higher wages, influence human capital accumulation differently depending on the child’s age. This research is part of a broader literature on the impact of economic downturns on education (e.g., Stuart, 2022). While the study also scrutinizes the school-age period, its primary objective differs from ours, as it seeks to evaluate how favorable economic conditions alter the opportunity cost of schooling and, consequently, the incentives for children to attend school. 3Local projections have been first proposed by Jorda (2005). They have been used widely in empirical macroeconomics and more recently in the context of environmental economics (see e.g. Barattieri et al., 2023; Roth Tran and Wilson, 2023; Naguib et al., 2022). As discussed by Dube et al. (2023), they can be used as an alternative methodology to deal with the issue of dynamic treatment effects that arises with the difference-in-difference approach in the case of multiple treatments. 4 indicating if the individual is a female, a first-born child and Hindu respectively.12 δdand δb are sets of district FE and cohort FE, respectively. Although our data are cross-sectional, we introduce a district-specific linear relationship across cohorts, τd, that accounts for differential trends in district-level education policies and regional disparities in economic growth. Finally, iis the error term.13 Incorporating district fixed effects in all our model specifications provides us with a measure of storm exposure that is conditionally exogenous. This allows us to account for the possibility that certain areas, such as coastal regions, may be more susceptible to hazards. Earlier studies have demonstrated that the occurrence of a cyclone does not provide any information on the probability of observing a similar event in the same location in the future (see e.g., Pielke et al., 2008; Elsner and Bossak, 2001). Therefore, it is impossible to predict the occurrence and exact path of storms, conditional on location. By using district FE, we are left with random realizations of storms, purging any correlation between locational economic decisions and the local distribution of storm exposure.14 Furthermore, our use of winds exclusively to construct exposures can be considered exogenous. Although other storms’ hazards include floods and surges, their impacts are affected by land management and deforestation, which have been shown to factor into people’s settling decisions (Petkov, 2022). We use the PLFS survey weights to weight our observations, and we cluster standard errors at the state level. Clustering at the state level is appropriate because education funding and programs are primarily administered at the state level.15 State-level clustering also takes into account spatial correlations within a state and time correlations in the exposure index resulting from the fact that the same storm affects multiple birth-year cohorts simultaneously.16 12We include birth order as a control variable in our analysis, as certain studies have indicated that birth order may have an impact on parental investments in education (see e.g., Black et al., 2005). We do not include controls for household headship, marital status, rural residency, or household size, as each of these variables could be affected by school-age exposure to storms and may lead to a bad-control issue if included. 13It is worth noting that our outcomes are observed only once for each individual in the cross-sectional PLFS dataset, and thus our specifications are neither dynamic nor staggered difference-in-differences models despite the inclusion of different cohorts. Instead, we adopt a cross-sectional cohort study approach, where we retrospectively evaluate individuals’ exposure histories between ages 5-15. 14Evidence suggests that, even with climate change, any signal will appear in the distribution of storm activity very gradually (see e.g., Emanuel, 2011). Therefore, it is unlikely that economic agents are aware of changes in the local distribution of storm exposure. 15http://countrystudies.us/india/37.htm 16We work with 35 clusters, including 28 states and 7 union territories. Tables D.4 and D.5 in the Online Appendix show that our results remain consistent when using district clustering or district-cohort clustering. 11 Results. In Table 1, Panels A and B present the results for equation (3). Panel A reports the results for educational delay measured as the difference between the reported years of schooling and the expected number of years based on reported educational attainment. Column (1) presents the baseline results, which suggest that exposure to storms leads to a statistically significant delay in completing a given level of education. The estimated delay for a child with unit exposure is 0.43 years on average, which translates to roughly a 5-month delay. Figure E.2 of the Online Appendix shows that while unit values in the exposure index are exceptional in our sample period, they are observed in Odisha due to the 1999 BOB 06 super cyclone, the most severe and destructive tropical cyclone recorded in India from 1990 to 2000. Although extremely severe cyclonic storms are rare, recent events such as storms Phailin and Fani in Odisha in 2013 and 2019 respectively, super cyclone Amphan in West Bengal in 2020, and severe cyclonic storms Tauktae and Yaas in Gujarat and West Bengal and Odisha respectively in 2021, indicate an increasing frequency of such events. Therefore, it is important to provide an interpretation of our estimates for large (unit) values of the exposure index, as it informs on the educational long-term delays that the current generation of school-attending kids may face. If we use the average exposure in the sample to interpret our results, we find that the educational delay is approximately nine school days.17 We examine the robustness of our results by using different sets of fixed effects (FE) in columns (3) and (4). In column (3), we include state-cohort FE, which allows the economic conditions of a state at the time of a cohort’s birth to affect long-term educational delays. Although the estimate is less precise and smaller than the baseline, it remains qualitatively similar. In the last column, we add state-policy FE, which account for the introduction of the new National Policy on Education in 1986 and its amendment in 1992. The policy aimed to provide compulsory education for all children up to the age of 14 and was effectively adopted at the state level. The interaction term covers cohorts born in 1985, those born between 1986 and 1991, and those born after 1991. Results are similar to those in column (2) and remain statistically significant at the 5% level. In Panel B of Table 1, we estimate a linear probability model to examine the impact of school-age exposure to storms on the likelihood of experiencing an educational delay of at least one year. The results are consistent across specifications and indicate that exposure to storms during schooling years increases the likelihood of experiencing an educational delay. Specifically, in the baseline specification (column 1), a unit exposure (which is likely to be driven by severe events) is associated with a 24 percentage point increase in the 17With an average storm exposure of 0.1in our sample, 42 weeks per year, and a 5-day school week, this number is computed as 0.43 ·0.1·42 ·5. 12 probability of accumulating an educational delay (i.e. repeating a year or dropping out). On the other hand, an average exposure increases this probability by 2.4 percentage points. In our sample, the proportion of individuals with an educational delay of at least one year is 0.331 (as shown in Panel B of Table D.3 of the Online Appendix). Based on the baseline estimate in Panel B, we can infer that this proportion would increase by approximately 72.5% (to a share of 0.571) in the case of an extreme cyclonic storm exposure, and by 7.25% (to a share of 0.355) in the case of an average exposure. 3.2 Educational Attainment In Panel C of Table 1, we investigate whether storms not only cause educational delays, but also affect the likelihood of completing a given level of education. We use an ordered logit model with a categorical variable representing reported educational attainment (0=below primary, 1=primary school, 2=middle school, 3=secondary education, 4=above-secondary education), where category 0 includes individuals who received some education but did not complete primary school. The same set of variables as in equation (3) is included as controls. The first column of the table displays the ordered logit estimates, while columns (2)-(6) report the marginal effects of school-age exposure to storms for each category of schooling. All of the estimates are statistically significant and represent the percentage point changes in the probability of completing a certain level of education in the case of unit school-age exposure to storms. In general, positive exposure to storms increases the probability of not completing primary school and completing at most primary and middle school, while decreasing the probability of achieving secondary and post-secondary education. Specifically, our findings show that unit exposure reduces the likelihood of achieving post-secondary education by 20 percentage points. This translates to a 2 percentage point reduction in the event of an average exposure. To give a sense of the scale of our findings, let us consider children who experienced the 1999 BOB 06 super cyclone during their schooling years (i.e., Cbd = 1). Based on the educational attainment proportions in Table D.3 of the Online Appendix, the estimates in Panel C of Table 1 suggest that the percentage of individuals who did not complete primary school (with at most primary education) would increase from 2.7% to 7.3% (or 9.8% to 20.8%), while the percentage of individuals who obtained post-secondary education (with at most secondary education) would decline from 27.2% to 7.2% (or 36.5% to 31%). Therefore, it can be inferred that exposure to the 1999 super cyclone likely resulted in a significant increase in individuals lacking basic education. Even for an 13 average storm exposure, these percentages remain significant. For example, the proportion of individuals with post-secondary education would decrease by 7.35%, while the percentage of individuals with only primary school education would increase by 11%. In Figure E.3 in the Online Appendix, we utilize the estimates obtained from the ordered logit model to visualize the predicted probabilities of achieving a particular level of education across the range of storm exposures from 0 to 1, along with their 95% confidence intervals. The overall findings from this analysis reveal that storms result in a leftward shift in the distribution of educational attainment, which is particularly concerning for developing nations such as India, where the distribution of skills is already heavily skewed to the left. 3.3 Type of Activity We expect that the educational disruption caused by storms during compulsory schooling would affect the type of labor market activities individuals perform in early adulthood, as certain types of jobs require higher levels of education or at least basic reading, writing, and computing skills. To investigate this issue, we estimate a reduced-form specification of school-age exposure to storms on an indicator variable for each type of activity in Panel A of Table 2. For example, in column (1), the dependent variable is a dummy variable equal to 1 if the main activity of individual iis regular work. We include the same set of controls as in equation (3) for each type of activity. Our estimates suggest that individuals who were exposed to storms during their schooling years are less likely to work as regular salaried workers and more likely to perform domestic duties. However, we find no statistically significant effect on the likelihood of being a casual worker, self-employed, or an unpaid family worker. To illustrate the magnitude of our results, let us consider the labor market impacts associated with the average positive exposure in our sample (i.e., Cbd = 0.1). The estimate in column (1) implies a 1.6 percentage point reduction in the probability of being a regular worker. According to Panel D of Table D.3 of the Online Appendix, 19.6% of individuals in our sample are engaged in regular work. Thus, the estimate in column (1) implies an 8% decrease in the probability of being employed as a regular worker. The estimate in column (5) indicates a 1.6 percentage point increase in the likelihood of performing domestic duties as the primary activity in early adulthood, which corresponds to a 4.8% change when taking the share of individuals involved in domestic duties (i.e., a share of 0.33) as a baseline. These effects are more pronounced for children who experienced more severe exposures. For example, with unit exposures and taking the 14 same baseline shares, the estimates imply changes of approximately 80% and 48% for regular work and domestic duties, respectively. In summary, an average storm increases the probability of experiencing a schooling delay by 2.4 percentage points, while concurrently reducing the likelihood of completing post-secondary education by 2 percentage points. These findings are congruent with a 1.6 percentage-point decrease in the probability of securing a regular salaried position, suggesting that both delays in schooling and a skill reduction are likely contributing factors. In Panel B of Table 2, we investigate whether positive exposure to storms is associated with lower wages and longer hours of work. In column (1), we restrict the sample to workers who receive a salary, which explains the drop in sample size. We find no evidence that, conditional on being employed as a regular worker, school-age exposure to storms has a permanent effect on wages. However, the subsample of workers with a positive salary is a selected one since exposure to storms reduces the probability of being a regular worker (see column 1 of Panel A). To address this issue, we run a Tobit estimation and report the average marginal effect (AME) on wages, evaluated at the means of the covariates, in column (2) of Panel B. The estimate shows a negative effect on wages, which is statistically significant at the 10% level. This suggests that, on average, a super storm causes a 39% decline in hourly wages, or taking the average exposure, a 3.9% wage drop. This result is consistent with the fact that storms increase educational delays and reduce the probability of completing higher education. Column (3) of Panel B shows the results on hours of work, focusing on individuals reporting positive hours of work and a positive salary, as in column (1). In column (4), we report the corresponding AME from a Tobit estimation, once again evaluated at the means of the covariates. We find no evidence that school-age exposure to storms has a permanent effect on hours of work. The disruption of education caused by storms is likely to widen income and social disparities across different districts and age groups in the long run. Our findings suggest that this increase in inequality is primarily driven by changes in qualifications and types of employment, leading to less secure and potentially lower-paying work. Additionally, disparities along the income distribution may further increase as those who experience the largest delays in education often come from vulnerable social groups. 15 3.4 Robustness In this section, we present a series of robustness checks. We focus on the results related to education and refer readers to Online Appendix C for an analysis of the primary activity status of individuals. Early-life Exposure to Storms. It is well-documented in the literature that shocks experienced in early life can have long-lasting negative impacts on health, education, and labor market outcomes (see e.g., Almond et al., 2018, for a recent survey of this literature). To verify that our findings are attributable to shocks during school-age years rather than earlier shocks, we augment our baseline specification by incorporating a measure of storm exposure during the early years of life (0-4 years old), a period considered critical for skill formation in developmental psychology, epidemiology, and economics (see e.g., Duque et al., 2019; Heckman, 2008; Knudsen et al., 2006). The measure is similar to our baseline index (see equation 1), but specifically focuses on early life. Table 3 presents results from this exercise. In column (1), we present our baseline specification. Including storms which took place in early years reduces the sample to birthyear cohorts 1990-1995. Column (2) shows that restricting the sample to individuals born after 1989 does not affect our baseline results. In column (3), we replace school-age (Cbd) with early-life exposure. Estimates are statistically insignificant for all outcomes, except for educational attainment, suggesting that for individuals who actually received some formal schooling, storms in early life have little impact on educational delays. However, this does not necessarily mean that early-life exposure to storms has no impact on education; it may still reduce the probability of receiving a formal education, which we are unable to assess. Finally, in column (4), we include both early-life and school-age exposures simultaneously. The inclusion of the early-life measure does not impact the estimates of interest, suggesting that our results are not driven by storm shocks that occurred in years 0-4, and that school-age years are crucial for long-term human capital formation.18 Falsification Test. To confirm the validity of our identification strategy, we conduct a falsification test by randomly assigning storms to the sample. We shuffle the measure of exposure to storms across the entire sample and substitute this randomized variable for the actual exposure measure in the baseline specification. We anticipate that the results 18We also control for after-school storm exposure by summing yearly exposures over the after-school period up to 2018 in the analysis of primary activity status (see Table D.7 in the Online Appendix). Our baseline estimates remain unchanged even after adding this control. 16 of this exercise will yield mostly statistically insignificant estimates for the variable of interest, while leaving the statistical significance of the estimates for other variables largely unchanged. We repeat this process 1,000 times and record the t-statistics and p-values for each iteration. We present the results of the falsification test in Figure 1, which visually shows the distribution of t-statistics for the coefficient of interest. Panel A focuses on the regression on years of educational delay, while Panel B considers educational delay as a binary variable. Panel C shows the results for educational attainment. The histograms represent the distribution of t-statistics across the 1,000 repetitions of the falsification exercise, with the red vertical line indicating the t-statistic of the baseline estimates (3.01, 2.96, and -3.79, respectively). We observe that the majority of the distribution falls within the -1.96 and 1.96 boundaries, indicating that most of the coefficients obtained through the falsely-attributed storms are statistically insignificant. Removing Extreme Exposures. Table 4 examines the sensitivity of our results to extreme values of exposure. The baseline results are shown in column (1). In column (2), we exclude individuals from Odisha, which has unusually high values of exposure due to the 1999 super cyclone BOB 06. The results obtained from this subsample are similar to the baseline estimates. Finally, in column (3), we exclude all winds with values above the 95th percentile of the wind speed distribution. As anticipated, this leads to smaller effect sizes and less precise estimates. Climate Controls. To account for the potential influence of general climate conditions during childhood on human capital formation and long-term outcomes, we include controls for local climate effects such as precipitation and temperature in columns (4) and (5) of Table 4. To this end, we augment the baseline specification by introducing a district-specific variable measuring the average annual precipitation (in millimeters) between ages 5 and 15. Additionally, we include controls for the average temperature (in ◦C) and the number of days that children in a particular district were exposed to different temperature ranges (0-10, 10-20, 20-30, and above 30◦C) during their school-age years. We obtain the raw temperature and precipitation data from the ERA5-Land archive, accessed through the Google Earth Engine, and then aggregate them at the district level.19 19The ERA5-Land data is generated by researchers at the European Centre for Medium-Term Weather Forecasting (Muñoz Sabater et al., 2019). It is a climate reanalysis dataset that provides hourly weather information with a spatial resolution of 0.1×0.1 degrees, which is approximately 10×10 kilometers, covering the period from 1981 to the present. 17 Results are shown in columns (4) and (5) of Table 4. Column (4) reports the baseline specification estimated on the subset of the sample for which climate variables are available. In column (5), we include controls for precipitation and temperature, as described earlier. Across all panels, the coefficient of interest remains precisely estimated and qualitatively similar to the baseline results, albeit with a slightly smaller size effect. Migration. One potential source of bias arises from measurement error in storm exposure attributed to migration. In fact, we determine storm exposure based on individuals’ residential locations at the time of the survey, as we lack information about their residence during their school years. If individuals have relocated to a different district by 2018, our measure may contain errors, particularly if either the district they left or the one they moved to had experienced storms during their compulsory schooling. To address this concern, we conducted an analysis using supplementary data sets, demonstrating that permanent out-of-district migration is relatively rare and that the occurrence of a storm tends to reduce the probability of migration (see Online Appendix A and Footnote 7). As an additional test, focusing on the PLFS sample, we propose excluding married females from the analysis (column 6 of Table 4). In India, marriage is a primary driver of migration, especially for women who typically move to live with their husband’s family. However, this test is admittedly approximate, as it is unclear whether all married women have migrated out of their district, and it is also uncertain whether unmarried women have not migrated. Similarly, we observe that educated individuals had to move more frequently, often for educational purposes. Thus, we also suggest excluding highly educated individuals from the sample (column 7 of Table 4), noting that estimated delays will only pertain to individuals with at most secondary education. As observed in columns (6) and (7) of Table 4, the key estimates remain qualitatively similar to the baseline results, although they are somewhat less precisely estimated in the former column, possibly due to the exclusion of approximately 40% of the sample. Education Controls. To account for the fact that individuals within educational categories may share observable characteristics or have similar abilities that make them more or less likely to experience educational delay, we propose two alternative approaches.20 20Since school-age exposure to storms impacts both educational delay and educational attainment, using educational categories fixed effects would cause a bad-control problem 18 The first approach involves using the predicted probability of completing a reported level of education, conditional on observable individual characteristics, as a proxy for educational attainment. Specifically, we estimate a linear probability model on a set of individual characteristics, such as the individual’s gender, year of birth, whether they are a first-born child, and their religion (Hindu or non-Hindu) as predictors, along with interactions of these variables. To avoid a bad control issue in the final regression, we focus on a subsample of states with zero exposure to storms between 1990 and 2010, and restrict ourselves to individual characteristics that are unlikely to be affected by storms. We use these estimates to predict the probability of completing each level of education for each individual in the full (baseline) sample and use the corresponding probability as a proxy for educational attainment in the final regression. The result of this exercise is presented in column (2) of Table 5 and is highly comparable to the baseline estimates presented in column (1). Second, we propose including fixed effects capturing parental education, which has been shown to be a significant predictor of children’s educational achievements (Björklund and Salvanes, 2011; Guryan et al., 2008). Parental education is less likely to be affected by children’s exposure to storms, although there is a possibility that parents enrolled in university may have young children attending primary school. This may be particularly true for relatively young parents. However, it is implausible that parental education overlaps with children’s compulsory schooling at low levels of education. One potential drawback of this approach is that parental education is only observable if both the individual and their parents live in the same household. Additionally, since married women often move in with their husbands’ families, the sample may include relatively more males than in the baseline. We begin by replicating the baseline approach on the subset of the sample with available data on parental education (column 3 of the table). Despite the smaller sample size (representing only 45% of the initial sample), the coefficients are very similar to the baseline estimates. In column (4), we include fixed effects for parental education, and the estimates remain nearly identical across specifications. Alternative Measures of School-age Exposure to Storms. In Table 6, we explore alternative specifications of Cbd. The first alternative (column 2) measures exposure by summing over the squares of yearly exposures (i.e., Pt=b+15 t=b+5 x2 dt), which assigns more weight to stronger exposures. In contrast, the baseline measure simply sums over exposures, treating a district-cohort exposed to multiple small storms the same as a district-cohort exposed to one violent and potentially destructive storm. By summing 19 across squares, we can differentiate between storm intensities when aggregating over the years. However, multiple exposures over time in a given district are rare (75.5% of individuals with positive exposure experienced only one storm between ages 5 and 15), so we do not expect this alternative specification to substantially alter our results. Second, we experiment with a different functional form to capture the relationship between the force exerted by winds on structures and wind speed exposure. Emanuel (2011) notes that there are physical reasons to believe that damages to building infrastructure are related to wind speed exposure in a cubic manner. To account for this, we use a cubic specification in columns (3) and (5) of Table 6, where we replace the square in Equation 2 with a cube. Third, we modify the wind speed threshold that defines a storm. In the paper, we use a benchmark threshold of 50 knots, based on Emanuel (2011), which is likely to cause damages. However, in columns (4) and (5) of Table 6, we raise the threshold to 64 knots, equivalent to a category 1 cyclone on the Saffir-Simpson scale. In column (6) of Table 6, we eliminate the threshold altogether and include all winds that occur during a cyclone event. Finally, in column (7) of Table 6, we compute the maximum wind speed hitting each district using the HURRECON wind field model (see Online Appendix B.2 for more details), following Boose et al. (2004) instead of Deppermann (1947). Overall, the estimates in Table 6 largely resemble the baseline results, except for column (6), where measuring exposure directly with winds introduces a downward bias. This is likely due to the assignment of non-zero exposure values for districts with only mild wind speeds, which hardly cause any damage. 4 Income and Infrastructure Channels Our study indicates that for school-age children, exposure to storms leads to educational delays and has negative long-term consequences for both academic achievements and career prospects. This is concerning, as a decline in human capital formation contributes to a deskilling of the population, potentially impeding economic growth. To develop effective policy recommendations addressing the consequences of storms, we next explore the ways in which education may be impacted. Storms can affect human capital through two primary channels: shifts in schooling demand and changes in schooling supply.21 21See Baez et al. (2010) for a comprehensive review of the literature on the channels through which disasters damage human capital. 20 middle school. The findings demonstrate that storms have a substantial impact on attendance for primary school children in levels C3 to C5, corresponding to ages 8 to 11. The estimated reductions in attendance are sizable, with a drop of approximately 15% for a super storm and 0.75% for an average storm. However, no statistically significant impacts on attendance are observed for any other levels. Overall, our findings align with previous research indicating that weather shocks and disasters triggered by natural hazards lead to a drop in school enrollments in developing countries (see, for instance, Jensen, 2000). Furthermore, our findings lend support to the idea that a disaster-induced negative income shock prompts parents to withdraw their children from school. This could be due to financial constraints, as families may no longer have the means to afford schooling, or because their children are required to work to supplement the household income. This hypothesis that disasters increase the incidence of working children is well-supported by the literature (see, for instance, Baez et al., 2010; De Janvry et al., 2006, for a review). The imprecise estimates for levels C1 and C2 suggest that children at these levels may remain in school because they are either too young to work or unable to contribute to reconstruction efforts due to physical limitations. Additionally, the lack of effects on levels C6-C8 may be attributed to the dedication of poorer parents whose children have advanced to this level to ensure that their children complete middle school. Furthermore, since wealthier children are more commonly found in middle school, they may be less affected by the income shock, as our income shock story suggests. Despite the lack of impact on attendance at the middle school level, older students may still be compelled to work or work more frequently after school and on weekends. This may reduce the amount of time available for studying, increase fatigue, reduce concentration in school, and ultimately lead to a decline in academic performance. We explore this possibility in Panel B of Table 8, which examines examination results in the final year of primary (C5) and middle (C8) school.29 Columns (1) and (2) focus on the log average number of students who appeared for the exam, while the subsequent two columns examine those who passed the exam. Columns (6) and (7) show the effect on the log average number of students who scored above 60 Our results indicate that there is no effect on exam appearance or performance for students at level C5. However, in accordance with our income shock narrative, we discover that middle school students experience a decline in academic performance. Fewer students appear for the exam, and even fewer pass it. Additionally, the number of students who pass with a grade above 60% also decreases following a storm. 29Examination results for other years of schooling are not available. 27 The estimates are consistent across columns, and unlike the results for school attendance, they reveal substantial effects of about 15% for an average storm. This implies that even a moderate storm can have significant impacts on education. These findings are particularly important given that a decline in grades may ultimately result in educational delays, although we cannot formally investigate this possibility with our available data. Taken together, our results suggest that the primary reason for the estimated long-term educational delays is the significant decrease in income, resulting in reduced attendance for primary school children and decreased academic performance for middle school children. While the decline in attendance for an average storm is relatively small, the deterioration of academic performance is substantial. Specifically, the average storm results in a 15% reduction in the number of children who appear for the exam, pass the exam, and receive a good grade. These results align with previous research that has demonstrated the adverse contemporaneous effects of disasters triggered by natural hazards and extreme weather shocks on education in developing countries (see e.g., Deuchert and Felfe, 2015; Spencer et al., 2016). 5 Conclusion In this study, we examine the impact of storm exposure during school years on long-term educational attainment and primary activity status among young adults in India. Our findings indicate that individuals who experienced a storm during these critical years are more likely to experience educational delays and less likely to complete higher education. Furthermore, we observe a decrease in the likelihood of securing regular salaried employment and an increase in the probability of engaging in domestic duties as a primary activity. Our results also provide indirect evidence that the enduring effects of school-age storm exposure can be attributed to both the degradation of educational infrastructure and a decline in household demand for schooling due to reduced income. These findings align with the notion that adverse income shocks can lead to an increase in working children, manifested as decreased school attendance for primary students and diminished academic performance for middle schoolers. Overall, our study underscores the importance of robust social safety nets and the need to extend post-disaster policies beyond mere reconstruction efforts. Such policies should integrate financial transfers with educational initiatives, such as cash transfers conditional on school attendance and enhanced school support. Local projections of 28 household income reveal that social policies, including unemployment insurance, could be instrumental in building resilience and managing risk in urban areas, which are particularly vulnerable following storms. While our results offer valuable insights, two caveats warrant consideration, suggesting that our findings should be interpreted as lower bounds of the true effects. First, our estimations do not account for the poorest individuals who are likely not enrolled in school due to their families’ low-income status. Including this segment, which tends to be disproportionately affected by disasters, would likely yield larger estimates. Second, our data only accounts for individuals who survived the storm and its aftermath until 2018. Although storm-related fatalities have been relatively contained in recent years, it is important to recognize that our results may be subject to survivorship bias. However, we do not anticipate this issue to significantly affect our findings. 29 Tables and Figures Table 1: Educational Delay and Educational Attainment Educational delay (1) (2) (3) Panel A: # of years School-age exposure 0.43∗∗∗ 0.22∗0.28∗∗ (0.14) (0.13) (0.10) Panel B: yes=1, no=0 School-age exposure 0.24∗∗∗ 0.20∗∗ 0.18∗∗ (0.080) (0.084) (0.072) Controls Yes Yes Yes State-cohort FE No Yes No State-policy FE No No Yes Observations 70,003 70,003 70,003 Panel A: Mean dep. var. 0.52 0.52 0.52 Panel B: Mean dep. var. 0.33 0.33 0.33 Educational attainment Logit Below Primary Middle Secondary Above-secondary estimates primary school school education education (1) (2) (3) (4) (5) (6) Panel C: Educ. attainment School-age exposure -1.18∗∗∗ 0.046∗∗∗ 0.11∗∗∗ 0.096∗∗∗ -0.055∗∗∗ -0.20∗∗∗ (0.31) (0.011) (0.029) (0.027) (0.015) (0.052) Controls Yes Yes Yes Yes Yes Yes Observations 70,003 70,003 70,003 70,003 70,003 70,003 Mean dep. var. 0.027 0.098 0.239 0.365 0.272 Notes: Panel A and B show results on educational delay. In Panel A, educational delay is calculated as the difference between the reported years of schooling and the minimum number of years required in the schooling system to attain the reported educational level. In Panel B, educational delay is measured using a dummy variable that takes a value of one if the delay is at least one year. In Panel C, educational attainment is a categorical variable indicating the reported level of education (0=no formal schooling, 1=primary school, 2=middle school, 3=secondary education, 4=above-secondary education), with category 0 including individuals who received some education but did not complete primary school. Column (1) shows the results from an ordered logit estimation where the dependent variable is a categorical variable indicating the reported educational attainment. Columns (2) to (6) report the marginal effects of childhood exposure to storms for each category of schooling. Controls include district FE, cohort FE, district trends, and individual controls, including dummy variables indicating if the individual is female, first-born, and Hindu.Policy FE used in the interaction terms include three FE corresponding to cohorts born in 1985, those born between 1986 and 1991, and those born after 1991. ∗p < 0.10, ∗∗ p < 0.05,∗∗∗ p < 0.01. Standard errors are clustered at the state level. Source: Authors’ estimates. 30 Table 2: Type of Activity, Wages and Hours Worked Regular work Casual labor Self-employed Unpaid family work Domestic duties (1) (2) (3) (4) (5) Panel A: Type of activity School-age exposure -0.16∗∗ -0.0093 -0.046 0.016 0.16∗∗∗ (0.076) (0.075) (0.059) (0.067) (0.056) Controls Yes Yes Yes Yes Yes Observations 70003 70003 70003 70003 70003 Mean dep. var 0.196 0.093 0.132 0.079 0.329 Log hourly wages Hours of work Log hourly wages AME Tobit Hours of work AME Tobit (1) (2) (3) (4) Panel B: Wage & hours worked School-age exposure 0.018 -0.393∗3.57 -4.33 (0.18) (0.230) (3.70) (4.684) Controls Yes Yes Yes Yes Observations 29,089 70,003 29,089 70,003 Mean dep. var 3.71 53.60 Notes: In Panel A, the dependent variable is a dummy variable that takes a value of 1 if the main activity of the individual is to perform regular work (column 1), casual labor (column 2), self-employment (column 3), work as an unpaid family worker (column 4), or perform domestic duties (column 5), as described in the Section 2. In Panel B, the dependent variable is the individual’s logarithm of (real) hourly wage in rupees (columns 1 and 2) and hours worked (columns 3 and 4). Columns (1) and (3) estimate the effect of childhood exposure to storms using a subsample of individuals who report wages and hours worked. This subsample mainly consists of individuals engaged in regular work and casual labor, and the number of observations may differ slightly from that presented in the summary statistics due to singleton observations. In columns (2) and (4), we report the average marginal effects (AME) on wages and hours worked, respectively, evaluated at the means of the covariates, using a Tobit estimation. Controls include district FE, cohort FE, district trends, and individual controls, including dummy variables indicating if the individual is female, first-born, and Hindu.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01. Standard errors are clustered at the state level. Source: Authors’ estimates. 31 Table 3: Controlling for Early-life Exposure (Education) Baseline Sub-sample Early-life School & early-life (1) (2) (3) (4) Panel A: Educ. delay: # of years School-age exposure 0.43∗∗∗ 0.35∗∗ 0.35∗∗ (0.14) (0.15) (0.14) Early-life exposure -0.16 -0.16 (0.14) (0.14) Panel B: Educ. delay: yes=1, no=0 School-age exposure 0.24∗∗∗ 0.38∗∗∗ 0.38∗∗∗ (0.080) (0.088) (0.087) Early-life exposure -0.086 -0.088 (0.080) (0.080) Panel C: Educ. attainment School-age exposure -1.18∗∗∗ -1.22∗∗∗ -1.22∗∗∗ (0.31) (0.44) (0.43) Early-life exposure 0.82∗∗∗ 0.83∗∗∗ (0.13) (0.15) Controls Yes Yes Yes Yes Observations 70,003 41,892 41,892 41,892 Panel A: Mean dep. var. 0.52 Panel B: Mean dep. var. 0.33 Notes: The table presents results when controlling for early-life exposure to storms. In Panel A, educational delay is calculated as the difference between reported years of schooling and the minimum number of years required in the schooling system to attain the reported educational level. In Panel B, educational delay is measured using a dummy variable that takes a value of 1 if the delay is at least one year. In Panel C, educational attainment is a categorical variable indicating the reported level of education (0=no formal schooling, 1=primary school, 2=middle school, 3=secondary education, 4=above-secondary education), with category 0 including individuals who received some education but did not complete primary school. Column (1) shows baseline estimates, while column (2) presents results for the baseline specification estimated on a subsample of individuals born after 1989. In columns (3) and (4), the focus is on the same subsample. Column (3) replaces the school-age exposure measure with the early-life exposure index, and column (4) includes both measures simultaneously. Controls include district FE, cohort FE, district trends, and individual controls, including dummy variables indicating if the individual is female, first-born, and Hindu. ∗ p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01. Standard errors are clustered at the state level. Source: Authors’ estimates. 32 Table 4: Additional Robustness (Education) Excl. Excl. Sub-sample Climate Excl. Excl. Baseline Orissa extremes (climate) controls females high educ. (1) (2) (3) (4) (5) (6) (7) Panel A: Educ. delay: # of years School-age exposure 0.43∗∗∗ 0.53∗∗ 0.29∗∗ 0.44∗∗∗ 0.22∗∗ 0.40∗0.52∗∗ (0.14) (0.22) (0.15) (0.16) (0.097) (0.20) (0.23) Panel B: Educ. delay: yes=1, no=0 School-age exposure 0.24∗∗∗ 0.23∗0.15∗0.24∗∗∗ 0.18∗∗ 0.24∗∗ 0.28∗∗ (0.080) (0.14) (0.077) (0.084) (0.071) (0.10) (0.11) Panel C: Educ. attainment School-age exposure -1.18∗∗∗ -0.74∗∗∗ -0.65#-1.30∗∗∗ -0.81∗∗ -0.86#-1.20∗∗∗ (0.31) (0.27) (0.41) (0.24) (0.39) (0.53) (0.41) Controls Yes Yes Yes Yes Yes Yes Yes Climate controls No No No No Yes No No Observations 70,003 67,770 70,003 66,702 66,702 43,989 50,981 Panel A: Mean dep. var. 0.52 Panel B: Mean dep. var. 0.33 Notes: The table presents results after removing extreme exposures and controlling for climate variables. In Panel A, educational delay is calculated as the difference between reported years of schooling and the minimum number of years required in the schooling system to attain the reported educational level. In Panel B, educational delay is measured using a dummy variable that takes a value of 1 if the delay is at least one year. In Panel C, educational attainment is a categorical variable indicating the reported level of education (0=no formal schooling, 1=primary school, 2=middle school, 3=secondary education, 4=abovesecondary education), with category 0 including individuals who received some education but did not complete primary school. Column (1) shows baseline estimates, while column (2) presents results for the baseline specification estimated on a subsample of individuals located outside Orissa. In column (3), we recompute the exposure index by removing all winds with values above the 95th percentile of the wind distribution. Column (4) replicates the baseline specification on a subsample for which climate variables are available. In column (5), we include climate controls, such as a district-specific measure capturing the average yearly precipitation (in millimeters) experienced between ages 5-15. Additionally, we include the average temperature (in ◦C) and the number of exposure days within temperature bins (0-10, 10-20, 20-30, and above 30◦C) to which children of a given district were exposed during school age. Column (6) excludes married females and column (7) removes individuals with abovesecondary education from the sample. Controls include district FE, cohort FE, district trends, and individual controls, including dummy variables indicating if the individual is female, first-born, and Hindu.#p < 0.012,∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01. Standard errors are clustered at the state level. Source: Authors’ estimates. 33 Table 5: Educational Controls (Education) Predicted Parental Baseline educ. attainment Sub-sample education (1) (2) (3) (4) Panel A: Educ. delay: # of years School-age exposure 0.43∗∗∗ 0.43∗∗∗ 0.37∗0.37∗∗ (0.14) (0.15) (0.19) (0.18) Panel B: Educ. delay: yes=1, no=0 School-age exposure 0.24∗∗∗ 0.24∗∗∗ 0.24∗∗ 0.24∗∗ (0.080) (0.080) (0.11) (0.11) Panel C: Educ. attainment School-age exposure -1.18∗∗∗ -1.17∗∗∗ -0.79 -0.91∗ (0.31) (0.31) (0.59) (0.47) Controls Yes Yes Yes Yes Predicted educ. attainment No Yes No No Parental education No No No Yes Observations 70,003 70,003 31,243 31,243 Panel A: Mean dep. var. 0.52 Panel B: Mean dep. var. 0.33 Notes: The table presents results on educational delay with the addition of educational controls. In Panel A, educational delay is calculated as the difference between reported years of schooling and the minimum number of years required in the schooling system to attain the reported educational level. In Panel B, educational delay is measured using a dummy variable that takes a value of 1 if the delay is at least one year. In Panel C, educational attainment is a categorical variable indicating the reported level of education (0=no formal schooling, 1=primary school, 2=middle school, 3=secondary education, 4=above-secondary education), with category 0 including individuals who received some education but did not complete primary school. Column (1) shows baseline estimates, while column (2) controls for the individual’s predicted probability of completing the reported level of education. Column (3) replicates the baseline specification on a subsample for which parental education is available. In column (4), the same sample as in column (3) is used, and parental education is additionally controlled for. Controls include district FE, cohort FE, district trends, and individual controls, including dummy variables indicating if the individual is female, first-born, and Hindu.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01. Standard errors are clustered at the state level. Source: Authors’ estimates. 34 Table 6: Alternative Measures of Storm Exposure (Education) Baseline Sum of squares 50, cubic 64, square 64, cubic All winds HURRECON (1) (2) (3) (4) (5) (6) (7) Panel A: Educ. delay: # of years School-age exposure 0.43∗∗∗ 0.43∗∗ 0.48∗∗ 0.48∗∗ 0.42∗∗ 0.077∗∗∗ 0.39∗∗ (0.14) (0.19) (0.21) (0.18) (0.19) (0.024) (0.16) Panel B: Educ. delay: yes=1, no=0 School-age exposure 0.24∗∗∗ 0.30∗∗∗ 0.30∗∗∗ 0.29∗∗∗ 0.30∗∗∗ 0.044∗∗∗ 0.27∗∗∗ (0.080) (0.077) (0.084) (0.078) (0.076) (0.014) (0.064) Panel C: Educ. attainment School-age exposure -1.18∗∗∗ -1.69∗∗∗ -1.65∗∗∗ -1.59∗∗∗ -1.68∗∗∗ 0.013 -1.24∗∗∗ (0.31) (0.27) (0.14) (0.11) (0.26) (0.085) (0.39) Controls Yes Yes Yes Yes Yes Yes Yes Observations 70,003 70,003 70,003 70,003 70,003 70,003 70,003 Panel A: Mean dep. var. 0.52 0.52 0.52 0.52 0.52 0.52 0.52 Panel B: Mean dep. var. 0.33 0.33 0.33 0.33 0.33 0.33 0.33 Notes: The table presents results on educational delay with the use of alternative specifications of the school-age exposure to storms. In Panel A, educational delay is calculated as the difference between reported years of schooling and the minimum number of years required to attain the reported level of education. In Panel B, educational delay is measured using a dummy variable that takes a value of 1 if the delay is at least one year. In Panel C, educational attainment is a categorical variable indicating the reported level of education (0=no formal schooling, 1=primary school, 2=middle school, 3=secondary education, 4=above-secondary education), with category 0 including individuals who received some education but did not complete primary school. Column (1) shows baseline estimates, while columns (2)-(7) present results based on alternative specifications of storm exposure. Specifically, in column (2), storm exposure is calculated using the sum of the squares of yearly exposures. In column (3), storm exposure is calculated using a threshold of 50 knots and a cube. In column (4), exposure is calculated using a threshold of 64 knots and a square, and in column (5), exposure is calculated using a threshold of 64 knots and a cube. Column (6) computes exposure using all winds, and finally, in column (7), exposure is computed using the HURRECON model, a threshold of 50 knots, and a square. Controls include district FE, cohort FE, district trends, and individual controls, including dummy variables indicating if the individual is female, first-born, and Hindu.∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01. Standard errors are clustered at the state level. Source: Authors’ estimates. . 35 Table 7: Damages to School Facilities and School Destruction Log avg. # of classrooms Share of schools in good conditions with electricity without electricity with unreliable electricity (1) (2) (3) (4) Panel A: Damages to school facilities Storm exposure -0.101∗∗ -0.056∗∗∗ 0.045∗∗∗ 0.011∗∗ (0.047) (0.0082) (0.011) (0.0045) Controls Yes Yes Yes Yes Observations 153,789 153,789 153,789 153,789 Mean dep. var. 4.36 0.688 0.285 0.026 Exit share of schools Share of buildings under construction (1) (2) (3) (4) (5) (6) Panel B: School destruction Storm exposure -0.0068 -0.0046 0.0084 -0.019∗∗∗ -0.025∗∗∗ -0.029∗∗∗ (0.0089) (0.0091) (0.011) (0.0023) (0.0036) (0.0048) Storm exposure(t−1) 0.016∗∗∗ 0.029∗∗∗ -0.053∗∗∗ -0.056∗∗∗ (0.0036) (0.0054) (0.0084) (0.0093) Storm exposure(t−2) 0.067∗∗∗ -0.023∗∗∗ (0.019) (0.0056) Controls Yes Yes Yes Yes Yes Yes Observations 109,831 92,918 76,109 109,831 92,918 76,109 Mean dep. var. 0.026 0.008 Notes: Panel A presents results on damages to school facilities. In column (1), the dependent variable is the log of the average number of classrooms in good conditions, with the average taken across schools at the pincode-year level. In columns (2) and (3), the dependent variable is the share of schools with and without electricity, respectively, in a postal code-year. In the last column, the dependent variable is the share of schools with unreliable electricity. Panel B shows results on school destruction. The dependent variable is the share of existing schools and the share of school buildings under construction in columns (1)-(3) and (4)-(6), respectively, with shares expressed relative to the number of schools in 2010 and taken within a postal code-year. In both panels, storm exposure is computed from wind exposures at the postal code level using a quadratic damage function and a 50 knots threshold. In Panel B, the number of observations differs slightly from that presented in the summary statistics due to singleton observations that are dropped in the estimations. In column (1) of Panel A, the mean dependent variable at the bottom of the table is presented without logs. Controls include postal code FE and district-year FE. ∗p < 0.10,∗∗ p < 0.05,∗∗∗ p < 0.01. Standard errors are clustered at the state level. Source: Authors’ estimates. 36 Hsiang, S. M. (2010). Temperatures and Cyclones Strongly Associated with Economic Production in the Caribbean and Central America. Proceedings of the National Academy of Sciences, 107(35):15367–15372. Hsu, S. and Zhongde, Y. (1998). A Note on the Radius of Maximum Wind for Hurricanes. Journal of Coastal Research, 14(2):667–668. Hyland, M. and Russ, J. (2019). Water as Destiny – The Long-term Impacts of Drought in Sub-Saharan Africa. World Development, 115:30–45. IPCC (2023). Climate Change 2023: Synthesis Report. Contribution of Working Groups I, II and III to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change [core writing team, h. lee and j. romero (eds.)]. IPCC, Geneva, Switzerland, 184 pp. Jensen, R. (2000). Agricultural Volatility and Investments in Children. American Economic Review, 90(2):399–404. Jorda, O. (2005). Estimation and Inference of Impulse Responses by Local Projections. American Economic Review, 95:161–182. Keerthiratne, S. and Tol, R. S. J. (2018). Impact of Natural Disasters on Income Inequality in Sri Lanka. World Development, 105:217–230. Knapp, K. R., Kruk, M. C., Levinson, D. H., Diamond, H. J., and Neumann, C. J. (2010). The International Best Track Archive for Climate Stewardship (IBTrACS): Unifying Tropical Cyclone Best Track Data. Bulletin of the American Meteorological Society, 91:363–376. Knudsen, E. I., Heckman, J. J., Cameron, J. L., and Shonkoff, J. P. (2006). Economic, Neurobiological, and Behavioral Perspectives on Building America’s Future Workforce. Proceedings of the National Academy of Sciences, 103(27):10155–10162. Maccini, S. and Yang, D. (2009). Under the Weather: Health, Schooling, and Economic Consequences of Early-Life Rainfall. American Economic Review, 99(3):1006–1026. Mueller, V., Gray, C., and Kosec, K. (2014). Heat Stress Increases Long-term Human Migration in Rural Pakistan. Nature Climate Change, 4:182–185. Muñoz Sabater, J. et al. (2019). ERA5-Land hourly data from 1981 to present. Copernicus Climate Change Service (C3S) Climate Data Store (CDS), 10. Munshi, K. and Rosenzweig, M. (2016). Networks and Misallocation: Insurance, Migration, and the Rural-Urban Wage Gap. The American Economic Review, 106(1):46–98. Naguib, C., Poirier, D., Pelli, M., and Tschopp, J. (2022). The Impact of Cyclones on Local Economic Growth: Evidence from Local Projections. Economic Letters, 220:110871. Neria, Y., Nandi, A., and Galea, S. (2008). Post-Traumatic Stress Disorder Following Disasters: A Systematic Review. Psychological medicine, 38(4):467. 43 Oreopoulos, P. (2007). Do Dropouts Drop out too Soon? Wealth, Health and Happiness from Compulsory Schooling. Journal of Public Economics, 91(11):2213–2229. Parag, M. and Yang, D. (2020). Taken by Storm: Hurricanes, Migrant Networks, and US Immigration. American Economic Journal: Applied Economics, 12(2):250–77. Park, R. J. (2022). Hot Temperature and High-Stakes Performance. Journal of Human Resources, 57(2):400–434. Pelli, M. and Tschopp, J. (2017). Comparative Advantage, Capital Destruction, and Hurricanes. Journal of International Economics, 108(C):315–337. Pelli, M., Tschopp, J., Bezmaternykh, N., and Eklou, K. M. (2023). In the Eye of the Storm: Firms and Capital Destruction in India. Journal of Urban Economics, 134:103529. Petkov, I. (2022). Weather Shocks, Population, and Housing Prices: the Role of Expectation Revisions. Economics of Disasters and Climate Change, 6(3):495–540. Pielke, R., Landsea, C., Mayfield, M., Laver, J., and Pasch, R. (2008). Hurricanes and Global Warming. American Meteorological Society, pages 1571–1575. Rosales-Rueda, M. (2018). The Impact of Early Life Shocks on Human Capital Formation: Evidence from El Niño Floods in Ecuador. Journal of Health Economics, 62:13–44. Roth Tran, B. and Wilson, D. (2023). The Local Economic Impact of Natural Disasters. Working Paper 2020-34, Federal Reserve Bank of San Francisco. Shah, M. and Steinberg, B. M. (2017). Drought of Opportunities: Contemporaneous and Long-Term Impacts of Rainfall Shocks on Human Capital. Journal of Political Economy, 125(2):527–561. Shakya, S., Basnet, S., and Paudel, J. (2022). Natural Disasters and Labor Migration: Evidence from Nepal’s Earthquake. World Development, 151:105748. Sheldon, T. L. and Zhan, C. (2022). The Impact of Hurricanes and Floods on Domestic Migration. Journal of Environmental Economics and Management, 115:102726. Shidiqi, K.-A., Di Paolo, A., and Álvaro Choi (2023). Earthquake exposure and schooling: Impacts and mechanisms. Economics of Education Review, 94:102397. Simpson, R. and Riehl, H. (1981). The Hurricane and Its Impact. Louisiana State University Press. Spencer, N., Polachek, S., and Strobl, E. (2016). How Do Hurricanes Impact Scholastic Achievement? A Caribbean Perspective. Natural Hazards, 84:1437–1462. Stuart, B. A. (2022). The Long-Run Effects of Recessions on Education and Income. American Economic Journal: Applied Economics, 14(1):42–74. Topalova, P. (2010). Factor Immobility and Regional Impacts of Trade Liberalization: Evidence on Poverty from India. American Economic Journal: Applied Economics, 2(4):1–41. 44 ASIAN DEVELOPMENT BANK ASIAN DEVELOPMENT BANK 6 ADB Avenue, Mandaluyong City 1550 Metro Manila, Philippines www.adb.org STORMS, EARLY EDUCATION, AND HUMAN CAPITAL Martino Pelli and Jeanne Tschopp ADB ECONOMICS WORKING PAPER SERIES NO. 743 October 2024 Storms, Early Education, and Human Capital This paper examines the impact of school-age exposure to storms on education and employment outcomes in India. Using wind exposure histories, the paper shows that exposure to an average storm can cause a 2.4 percentage point increase in educational delays, a 2 percentage point drop in post-secondary education attainment, and a 1.6 percentage point decline in regular salaried employment. The paper also highlights the role of damaged school infrastructure and declining household income in driving these outcomes. About the Asian Development Bank ADB is committed to achieving a prosperous, inclusive, resilient, and sustainable Asia and the Pacific, while sustaining its efforts to eradicate extreme poverty. Established in 1966, it is owned by 69 members —49 from the region. Its main instruments for helping its developing member countries are policy dialogue, loans, equity investments, guarantees, grants, and technical assistance.