Droughts: Learning continuity and recovery

Chatterjee, Somdeep; Katewa, Neeraj Working Paper Droughts: Learning continuity and recovery ADBI Working Paper, No. 1466 Provided in Cooperation with: Asian Development Bank Institute (ADBI), Tokyo Suggested Citation: Chatterjee, Somdeep; Katewa, Neeraj (2024) : Droughts: Learning continuity and recovery, ADBI Working Paper, No. 1466, Asian Development Bank Institute (ADBI), Tokyo, https://doi.org/10.56506/IWFG1305 This Version is available at: https://hdl.handle.net/10419/305428 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-nc-nd/3.0/igo/ ADBI Working Paper 1466 Chatterjee and Katewa ADBI Working Paper Series DROUGHTS: LEARNING CONTINUITY AND RECOVERY Somdeep Chatterjee and Neeraj Katewa No. 1466 July 2024 Asian Development Bank Institute ADBI Working Paper 1466 Chatterjee and Katewa The Working Paper series is a continuation of the formerly named Discussion Paper series; the numbering of the papers continued without interruption or change. ADBI’s working papers reflect initial ideas on a topic and are posted online for discussion. Some working papers may develop into other forms of publication. Suggested citation: Chatterjee, S., and N. Katewa. 2024. Droughts: Learning Continuity and Recovery. ADBI Working Paper 1466. Tokyo: Asian Development Bank Institute. Available: https://doi.org/10.56506/IWFG1305 Please contact the authors for information about this paper. Emails: [email protected], [email protected] Somdeep Chatterjee is an Assistant Professor at the Indian Institute of Management, Calcutta, Kolkata, West Bengal, India. Neeraj Katewa is a Doctoral Student at the Indian Institute of Management Lucknow, Uttar Pradesh, India. The views expressed in this paper are the views of the author and do not necessarily reflect the views or policies of ADBI, ADB, its Board of Directors, or the governments they represent. ADBI does not guarantee the accuracy of the data included in this paper and accepts no responsibility for any consequences of their use. Terminology used may not necessarily be consistent with ADB official terms. Working papers are subject to formal revision and correction before they are finalized and considered published. The authors thank Daniel Suryadarma, Tetsushi Sonobe, Albert Park, Harry Patrinos, M. Niaz Asadullah, and various conference participants at ADBI, Japan, for useful suggestions and inputs. Any remaining errors are our own. Asian Development Bank Institute Kasumigaseki Building, 8th Floor 3-2-5 Kasumigaseki, Chiyoda-ku Tokyo 100-6008, Japan Tel: +81-3-3593-5500 Fax: +81-3-3593-5571 URL: www.adbi.org E-mail: [email protected] © 2024 Asian Development Bank Institute ADBI Working Paper 1466 Chatterjee and Katewa Abstract In this paper, we study the nuanced relationship between drought, academic performance, and the efficacy of preventive measures, focusing on two drought-prone regions from the states of Maharashtra and Karnataka in India. Utilizing an event study framework relying on difference-in-differences regressions, we address the recurrent and non-permanent nature of droughts, elucidating their impact on learning outcomes over time. In Maharashtra, which lacks preventive schemes, successive droughts lead to a significant drop in standardized test scores for math and reading skills, followed by intermittent recovery and further interruptions. The school dropout rate also exhibits a similar pattern. Contrarily, Karnataka’s existing government-led preventive programs demonstrate resilience in academic achievement during drought periods. Keywords: climate variability, human capital investment, educational attainment JEL Classification: I21, O12, Q54, Q58 1 Introduction A major consequence of climate change is prolonged extreme weather events (Scoones, 1992; Angassa and Oba, 2008), including the recurrence of natural disasters such as droughts (Parry et al., 2010; IPCC, 2014; World Bank, 2019). While this has direct implications for global development in general (Stocker et al., 2013), among the various types of natural disasters, drought is perceived as the most severe hazard affecting agricultural productivity (Wilhite and Buchanan-Smith, 2005; Arshad et al., 2017; Arshad et al., 2018). Given that a considerable proportion of the population in lowand middle-income countries depends on agriculture either directly or indirectly (Mahendra Dev, 2012; Mehar et al., 2016), these adverse impacts can precipitate forced migration (Munshi, 2003; Henry et al., 2003; Feng et al., 2010; Dillon et al., 2011; Gray and Mueller, 2012; Mbaye, 2017) and affect household incomes (Paxson, 1992; Miguel et al., 2004; Skoufias et al., 2012; Keshavarz and Karami, 2013), thereby subsequently influencing household consumption expenditures and investments in human capital, such as investments in education (Benson and Clay, 2004; Fitzsimons, 2007; Duryea et al., 2007; Barrios et al., 2008; Bjorkman-Nyqvist, 2013; Agamile et al., 2021). While this literature seems to suggest that there is a negative impact of poor rainfall on education outcomes, more generally, the evidence on the impacts of such extreme weather events on human capital outcomes is mixed (Randell and Gray, 2016; Shah and Steinberg, 2017). For instance, there is some evidence that water scarcity induced by droughts could lead to higher opportunity costs of enrollment and results in a transition away from school participation towards labor market participation (Glory and Nsikak-Abasi, 2013). Other studies have attempted to link extreme weather events to outcomes such as school dropout rates (Khalili, 2020), education expenditures (Amjath-Babu et al., 2016), and the labor supply (Graff Zivin and Neidell, 2014). Moreover, the literature highlights the multifaceted impact of early-life rainfall on socioeconomic, health, and educational outcomes (Maccini and Yang, 2009). However, the estimates of educational attainment among students as a consequence of exposure to these shocks remain limited, with the exception of Shah and Steinberg (2017), who utilize test scores as a proxy for human capital attainment and demonstrate that positive rainfall shocks result in reduced school enrollment and attendance, along with diminished overall test scores for school-aged children.1 Our paper is closely related to this, and we study an analogous outcome variable, i.e., standardized test scores. However, unlike Shah and Steinberg (2017), who concentrate on gauging the average annual 1Interestingly, an alternative viewpoint is presented here by Shah and Steinberg (2017), emphasizing the importance of human capital investments and the alternative value of time. A substitute for education is engagement in agricultural activities or household responsibilities. Consequently, during years with favorable monsoons, households experience an increased opportunity cost for keeping their children enrolled in school, as the children could contribute to the household’s earnings if they were not in school. Shah and Steinberg’s findings from India indicate that children are more likely to drop out of school and perform poorly during good rainfall years, suggesting that higher rainfall negatively affects their academic performance (or conversely, students attend school and perform better during poor rainfall years). 2 precipitation and its current or next-year effects, our objective is two-fold. First, we intend to analyze the repercussions of drought events in regions prone to drought over an extended period. We show that droughts lead to learning losses as well, although in the context of our study, this is imprecisely estimated. Second, we take advantage of this approach to investigate recovery times and assess the efficacy of preventive measures. Essentially, we compare and contrast two distinct regions directly affected by droughts in the context of India. However, one of the regions had in place certain measures targeted towards coping with the aftermath of the drought event, whereas the other did not. We use these settings as a natural experiment to analyze the recovery from drought in these regions in the context of the estimated learning losses, which in turn allows us to estimate the efficacy of public policy designed to cope with the droughts. The Indian Meteorological Department (IMD) has established meteorological subdivisions, henceforth referred to as subdivisions (distinct from administrative districts, hereinafter called districts), based on similarities in climate and monsoon rainfall patterns. These subdivisions form the basis of our identification strategy. Unlike controlled experiments or policy implementations, drought transcends administrative boundaries, affecting broader geographical areas with diffused exposure. We acknowledge the non-permanent nature of drought and the fact that impact regions may change with each drought event. Therefore, to address this issue, we focus on subdivisions that are drought-prone and faced persistent drought in the decade starting from the year 2011. These drought events, along with their severity and the regions they affect, have been well documented in government reports, news articles, and other research papers. Additionally, we empirically validate the required relevance assumption for our identification strategy, confirming that these drought-prone subdivisions faced severe drought compared to the rest of their respective states during the study period. Consequently, the chosen setting for our study is the semi-arid regions of the Deccan Plateau, specifically Marathwada in Maharashtra and North Interior Karnataka (NIK), which face the highest frequency of severe droughts in India. Although administratively both subdivisions are in different states, they share meteorological borders and belong to the same arid zone of the Deccan Plateau, experiencing similar climatic conditions. Thus, they encountered comparable drought incidents in terms of intensity and timing during the study period, as corroborated by our dataset and secondary sources. While these neighboring meteorological subdivisions share climatic similarities, they also provide a compelling backdrop for our comparative analysis of the recovery and coping strategies in the aftermath of droughts. This is because, unlike Maharashtra, the state of Karnataka had explicit government programs in place to assist with coping with these events. Essentially, this creates a natural experiment for studying our question. Our identification strategy relies on separately comparing drought-prone areas in the affected states to the non-drought areas before and after the drought events, over a period of time. We perform separate event studies for Maharashtra and Karnataka to find suggestive evidence of learning losses in both 3 states as a result of the drought, but with different rates of recovery, with Karnataka outperforming Maharashtra. This lends support to the hypothesis that public policy directed towards recovery from these extreme weather events is effective in helping a region to recuperate from learning losses. We use data on test scores from the Annual Status of Education Report (ASER) in two states, Maharashtra and Karnataka, comprising 0.45 million observations between 2008 and 2018. We find that, in Maharashtra, a significant decline in math and reading scores occurred in the aftermath of consecutive drought years in 2011–2012. Although there was a gradual recovery during the normal year of 2013, this progress was disrupted by another drought in 2014–2015. Conversely, in Karnataka, where preventive programs were in place, academic scores experienced a relatively modest decline following consecutive drought years. Notably, the recovery of learning losses after successive drought years was more prompt in the drought-prone region of Karnataka compared to Maharashtra. This observation underscores that the implementation of government-led drought preventive programs can enhance the resilience of academic scores, indicative of educational outcomes, against the adverse impacts of consecutive meteorological droughts to a certain extent. Additionally, the findings regarding dropout rates further support this assertion. In Maharashtra, we compare Marathwada with the rest of the state to understand the impact of drought and the recovery period in the absence of preventive measures (we call this our baseline scenario). During the relevant study period, the state did not have any specific scheme or program to prevent the adverse effects of rainfall shortages, unlike the neighboring state of Karnataka, to the best of our knowledge. The Karnataka state government, however, implemented specific preventive initiatives such as the Sujala scheme and Project Bhoochetana to address water scarcity and the drought impact. Work under these projects was completed before the consecutive drought events starting in 2011. Additionally, before the consecutive drought years of 2014–2015, the government had also rolled out the next phase of these initiatives, along with newer initiatives supported by the World Bank. These drought preventive efforts were expected to have resulted in improved access to resources, enhanced agricultural productivity, and increased household income, demonstrating a commitment to sustainable water management. Therefore, the estimated impacts of droughts on learning for the state of Karnataka, where we compare NIK with the rest of the state, would likely be masked by the effectiveness of public sector preventive measures in drought-prone areas. Consequently, based on assumptions about the counterfactual, we fail to reject the hypothesis that the dampened learning losses in Karnataka and the phenomenal recovery may be attributed to these public policy measures. The rest of the paper is structured as follows: Section 2 offers background information on the study’s focus area. In Section 3, we detail the sources of our data. Section 4 examines the empirical framework, encompassing the identification strategy and methodology. In Section 5, we first validate identification assumptions and then present the main findings, along with supporting evidence and robustness test results. 4 The conclusion summarizes our findings and outlines future research avenues. 2 Background India, characterized by its extensive geographical span and diverse climates, is particularly vulnerable to the impacts of climate change (MOEFCC, 2018). Among these, drought emerges as a persistent threat with widespread repercussions (Dai, 2013). It is widely acknowledged that drought has a detrimental impact on the economy, with the Global Assessment Report (GAR) revealing that severe droughts can lead to an estimated 2–5% reduction in India’s gross domestic product (GDP). The phenomenon of drought is intricate and relies on interactions with various hydrological parameters such as evaporation, precipitation, infiltration, runoff, and surface water and groundwater storage (Sirda¸s and Sen, 2003; Esfahanian et al., 2017; Kim et al., 2017). Based on these parameters, droughts are commonly classified into meteorological, agricultural, hydrological, and socio-economic categories (Heim, 2002). Meteorological drought is triggered by high temperatures and low precipitation, and it results in water scarcity. This, in turn, leads to agricultural drought, causing stress on crops and impacting people’s livelihoods, the ecological environment, and socio-economic conditions (Uttaruk and Laosuwan, 2017; Du et al., 2018). 2.1 Drought-prone Zones in India Drought is a non-permanent and recurring phenomenon caused by a prolonged deficiency in rainfall compared to long-term average conditions in a given area (Patel and Yadav, 2015; Sreekesh et al., 2019). This is especially prevalent in semi-arid ecosystems (Hind and Marwan, 2010; Mohammad et al., 2015; Hussein, 2018; Mohammad et al., 2018; Sandeep et al., 2021). Within India’s climatic landscape, the southern semi-arid zone, spanning the states of Maharashtra, Karnataka, Telangana, and Andhra Pradesh, faces climatic vulnerability due to its location on the leeward side of the Western Ghats. The Western Ghats, a formidable mountain range running parallel to the western coast, creates a rain shadow in the eastern region, known as the Deccan Plateau (Mohamed et al., 2019). Because of where it is located, the Deccan Plateau often does not get enough water, making it dry on the leeward side of the Western Ghats, which block rain clouds coming from the Arabian Sea. Consequently, water scarcity persists as a significant challenge for the states in the Deccan Plateau2. 2Refer to Figure 2 of the case study titled “Drought Characteristics over the Deccan Plateau Region of India,” by Srinivasan et al., in the GAR Special Report on Drought 2021 by the United Nations Office for Disaster Risk Reduction. The published documents of the case study can be accessed here. In this case study, Figure 2 presents the areas showing composite rainfall deviations from the climatological normal (1981–2010) from June–September of very dry years (2002, 2004, and 2015). It highlights that the most affected areas in the two states, Maharashtra and Karnataka, fall within their respective most-droughtprone meteorological subdivisions, i.e., Marathwada in Maharashtra and North Interior Karnataka (NIK) in Karnataka. 5 Low and irregular rainfall, along with extreme temperatures and intense solar radiation, makes these semiarid regions in southern peninsular India highly vulnerable (INECC, 2010). Water scarcity is prevalent, with low groundwater tables, minimal rainfall, and high water runoff. The primary water source throughout the year in these regions is small and medium amounts of stored water. Consequently, the Global Assessment Report (GAR) on Drought 2021, released by the UN Office for Disaster Risk Reduction, identifies the Deccan region as experiencing the highest frequency (more than 6%) of severe droughts in all of India. The severe repercussions of drought-related crop failures are vividly highlighted by a concerning statistic in the most recent report from the National Crime Records Bureau (NCRB) in the year 2021—49.6% of farmer suicides in India are concentrated in the state of Maharashtra, followed by Karnataka (Accidental Deaths & Suicides in India (ADSI), 2022). Notably, both states are situated in the geographical region beneath the Western Ghats on the Deccan Plateau. This region, identified as experiencing the highest frequency (more than 6%) of severe droughts in all of India, according to the Global Assessment Report (GAR), further emphasizes the correlation between geographical location, drought situations, and their distressing consequences. Maharashtra is the third largest state in India based on geographical area and the highest contributor to India’s GDP at 14% (Deshpande, 2023). Agriculture plays an important role in the economy of Maharashtra. The total population of the state is 112 million according to the 2011 census, of which 57.8% are dependent on agriculture. About 84% of the total area under agriculture in the state is directly dependent on monsoon rainfall (Economic Survey of Maharashtra 2012-13, 2013). At the same time, a substantial portion of the state lies in the semi-arid regions in the rain shadow of the Western Ghats (Todmal, 2019). Thus, drought is one of the major natural disasters that pose a great threat to the state economy and agricultural development. To study and forecast weather, the state has been divided into four meteorological subdivisions3: Konkan (west coast), Madhya Maharashtra, Marathwada, and Vidarbha. Meteorological subdivisions are created by the Indian Meteorological Department based on similar climate and monsoon rainfall patterns (Kelkar and Sreejith, 2020). Historically, Maharashtra has been plagued by numerous drought events, such as in the years 1972, 2000, and 2004, and most recently, Maharastra faced a distressing series of consecutive droughts from 2011–2012 (Maharashtra Ahead, 2013), 2014–2015 (Kulkarni et al., 2016), and 2018–2019 (Singh et al., 2022). When a drought event occurs for two or more consecutive years, then it is considered to be a persistent drought event (Amrit et al., 2018). Marathwada, a region in arid portions of Maharashtra, has particularly been the epicenter of such persistent droughts. Nearly 73.83% of the population of the Marathwada region depends on agriculture as their primary source of income (Kelkar, 2013). Nonetheless, the irrigation provision in this area is quite deficient, 3Refer to the Indian Meteorological Department’s website for more detail: https://mausam.imd.gov.in/. 6 of 2011, the subdivisions prone to drought (Marathwada in Maharashtra and NIK in Karnataka) received significantly lower rainfall compared to the rest of the regions of their respective states. For instance, in Maharashtra, in the year 2010, which was considered a normal year, this difference amounted to 97.734 mm. By 2011, this difference had increased to 168.6887 mm. In percentage terms, drought-prone regions in Maharashtra experienced a drastic 31.56% decrease in rainfall in the year 2011 compared to the previous year, whereas the rest of Maharashtra only saw a modest 2.52% decrease. Similarly, in Karnataka, droughtprone areas experienced a notable 29.41% drop in rainfall in 2011 from the previous year, while the rest of the state actually saw an increase of 7.28% during the same period. This substantial drop in rainfall persisted in the following year, marking two consecutive drought years for Marathwada and NIK. While other regions in both states also faced a shortfall in rainfall in 2012, it is crucial to note that even before this dip, the rainfall in these drought-prone subdivisions was consistently lower than that in the rest of their respective states. The examination of the raw means of rainfall data, coupled with evidence from the background section, instills confidence that Marathwada and NIK were indeed severely impacted by drought in Maharashtra and Karnataka starting from 2011. However, we further rigorously test this in an event study setting and the results are presented in Figures A1 and A2, with a detailed discussion provided later in subsection 5.1. 4.2 Methodology Using the ordinary least squares (OLS) method to evaluate the impact of droughts is problematic. There can be unobservable factors that can lead to omitted variable bias. For instance, comparing children in the drought-prone division with those in other regions of the state might be confounded by pre-existing disparities between the two areas, possibly skewing outcomes. Similarly, comparing post-drought-period outcomes with pre-drought-period outcomes within the drought-prone division might yield biased results due to various factors, including shifts in the macroeconomic landscape over time. To address these issues, we propose an event study framework using the difference-in-differences methodology. As a result, we leverage spatio-temporal variations arising from persistent drought occurrences to identify their causal impact on learning outcomes and the recovery period for learning losses. Our analysis focuses on a series of drought events starting in 2011 in drought-prone subdivisions of Maharashtra and Karnataka. Given that our primary dataset starts in 2008, a year not marked by severe drought conditions, we have designated 2008 as the reference period (base year) for our analysis within this framework. To estimate the causal impact of drought on learning outcomes and the recovery period for learning losses, we employ the following regression equation for each child ifrom district din state sduring the time 13 period t: T estScoreidst =αd+β1·(Droughtds ∗T imet) + β2·(T imet) + βi·(Xi) + µidst.(1) In Equation 1, T estScore represents the standardized reading or math test scores for child iin time t. αdrepresents district fixed effects; it controls for time-invariant district specific characteristics. T imetis a dummy variable with eight versions, each designed to facilitate the construction of an event study framework. Across these versions, it assumes a value of 0 for the base year 2008 and subsequently takes the value 1 for each year from 2009 to 2018 (excluding 2015 and 2017, as our primary dataset lacks data for these years). Droughtds is a dummy variable indicating whether the individual belongs to the drought-prone subdivision of state s. In the context of Maharashtra, this subdivision corresponds to Marathwada, while in Karnataka, it is NIK. As we know that drought is non-permanent in nature and is not restricted by administrative boundaries, to avoid contamination in the control group, we excluded the districts in state sthat faced severe drought during the period of study but are not a part of the drought-prone meteorological subdivision in state s. Demographic controls represented by Xencompass the child’s age, the child’s gender, the family size, and the mother’s education. Standard errors are adjusted for district-level clustering. The interpretation of the coefficient β1yields the difference in mean test scores for time t compared to the base year, differentiating between children from the drought-prone subdivisions and those from other regions of the state s. The identifying assumption implies that in the counterfactual scenario, the estimated coefficient would be statistically insignificant. This is demonstrated in the results section. 5 Results In this section, we employ an event study difference-in-differences framework, as described in the methodology section (subsection 4.2), to present the outcomes of our analysis regarding drought incidents in the droughtprone areas of Maharashtra and Karnataka, starting from the year 2011. 5.1 Validating the Treatment Status To confirm that our identification of drought-prone regions is reasonable, we use rainfall data from our primary dataset as the dependent variable in the main empirical equation (Equation 1), discussed in the methodology section (subsection 4.2). This is executed for both states, encompassing all years starting from 2009 with respect to the base year 2008 in the event study setup. Figures A1 and A2 plot the estimated coefficients for the difference in the average annual rainfall for each time period compared to the base year, 14 distinguishing between the Marathwada region and the rest of the regions in Maharashtra, as well as the NIK region and the rest of the regions in Karnataka, respectively. These figures indicate that prior to 2011, the difference in the average annual rainfall between the droughtprone subdivisions and the rest of the regions in their respective states was not statistically significant when compared to the base year. However, in 2011, a statistically significant decline in the average annual rainfall is observed in the drought-prone subdivisions compared to the rest of the regions, as indicated by the negative coefficients. However, in the year 2012, a marginal positive coefficient is observed for the Marathwada region in comparison to the rest of Maharashtra when compared to the base year. However, this observed increase is not statistically significant and does not indicate an actual rise in rainfall. Instead, Marathwada experienced a decline in rainfall for the second consecutive year, and the positive coefficient is influenced by other regions in Maharashtra also facing shortages in rainfall during that year (refer to Figure A2 of the online appendix). A similar pattern is noted for Karnataka in 2012, where a non-significant positive coefficient is observed, and the raw means of the rainfall data align with the actual scenario (Figure A3 of the online appendix). 5.2 Main Findings As discussed in the background section (Section 2), the Maharashtra government lacked preventive schemes to address drought. Despite the implementation of the Jalyukt Shivar scheme after consecutive drought years in 2014–2015, its effectiveness was questioned. This circumstance offers an opportunity to investigate the impact of drought in the absence of preventive measures. To accomplish this, we employ the difference-in-differences regression for Maharashtra in an event study framework outlined in Equation 1, as detailed in the methodology section (subsection 4.2), to examine the causal impact of drought on learning outcomes. Equation 1 is separately applied to reading and math scores for the various years considered in our study, with 2008 as the base year. The results for child math scores are presented in Figure 1, and those for child reading scores are reported in Figure 2 based on the estimation of our regression equation. Panels 1 and 2 of Table A1 of the appendix contain the full regression table corresponding to these figures. We observe a statistically significant drop in math scores following consecutive drought years in 2011– 2012. In 2012, the decline in math scores is approximately 0.22 σpoints compared to the base year, as compared to the base year between the average child in Marathwada and the rest of Maharashtra after accounting for time-invariant district-specific effects. This learning loss in math scores gradually begins to recover during the normal year of 2013, only to be followed by another consecutive drought affecting the exposed group in 2014–2015. A similar impact is discernible in reading scores, with a notable drop in the 15 year 2012. Subsequently, recovery is observed until the next consecutive drought in 2014–2015 impacts the exposed group. The impact of drought on the academic performance of exposed groups, specifically in mathematics and reading scores, exhibits a similarity in magnitude. However, given that mathematics scores are lower than reading scores in Maharashtra, and particularly in Marathwada, the negative impact is more pronounced in mathematics. Furthermore, the recovery from the impact of drought appears to be swifter in reading scores than in mathematics scores. This disparity indicates that children may already possess weaker mathematical skills compared to reading skills, amplifying the detrimental effects of drought on mathematics performance. Moreover, we observe that despite drought occurring in both 2011 and 2012, the effects on academic performance begin to manifest from 2012 onwards. This suggests a temporal lag of one year between the onset of drought and its discernible influence on learning outcomes. Given this hypothesis, for the second consecutive drought (which occurred in 2014–2015), we would anticipate observing its substantial impact on test scores in the year 2015. However, since ASER data for 2015 and 2017 are not publicly available, we were unable to comprehensively assess the effects of the second consecutive drought. Despite incomplete data, we still observe a slight decrease in math scores, hindering recovery from prior drought years, alongside a significant decline in reading scores. 5.2.1 Effectiveness of Drought Preventive Strategies by the Public Sector16 As discussed earlier, Karnataka, being a neighboring state, shares a similar climate and climatic challenges with Maharashtra, especially in the neighboring meteorological subdivisions (Marathwada and NIK). However, unlike Maharashtra, Karnataka had existing drought preventive programs before a series of consecutive droughts starting in 2011. This presents an opportunity to assess the effectiveness of Karnataka’s preventive strategies. With both states facing comparable climatic conditions, this analysis of the neighboring meteorological division (NIK) allows us to evaluate the impact of drought on learning outcomes with preventive measures in place. To achieve this objective, we employ the difference-in-differences regression for Karnataka in an event study framework outlined in Equation 1, as detailed in the methodology section (subsection 4.2), to assess the influence of drought on learning outcomes with preventive measures implemented. Equation 1 is independently applied to reading and math scores for the different years under examination in our study for Karnataka, with 2008 serving as the reference year. The outcomes for child math scores are depicted in 16We also examine the role of the private sector in drought-prone areas to address water scarcity; unlike government efforts, the private sector often targets localized areas. Given the constraints of data and the time lag in educational outcomes, we analyze expenditure patterns. However, the main focus of our paper on educational outcomes leads us to place this analysis of private sector drought prevention effectiveness in the online appendix. 16 Figure 4, and the results for child reading scores are illustrated in Figure 5, derived from the estimation of our regression equation. The full regression table corresponding to these figures is provided in panels 1 and 2 of Table A2 of the appendix. In contrast to Maharashtra, the academic achievement of an average child in the exposed group within Karnataka, i.e., the NIK meteorological subdivision, appears resilient to the impact of successive meteorological droughts. While there is a slight downturn in the reading scores in 2011 following consecutive drought years in 2011–2012, this dip is neither economically nor statistically significant. Moreover, the recovery from this dip is swift when compared to the baseline scenario. However, there could have been a potential issue with interpreting our findings if Karnataka inherently excelled in education compared to Maharashtra, beyond drought mitigation efforts. However, our summary statistics indicate that Maharashtra slightly outperforms Karnataka in education, with the reverse not holding true. Therefore, Maharashtra cannot be characterized as a poor performer in terms of learning outcomes compared to Karnataka. In fact, Marathwada, within Maharashtra, also outperforms the NIK subdivision of Karnataka. Additionally, Maharashtra and Marathwada receive slightly more rainfall than Karnataka and NIK, respectively (refer to Figures A2 and A3 in the online appendix). Furthermore, as the latter half of the decade unfolds, discernible positive trends emerge, indicating that the initiatives implemented in drought-prone areas of the state are yielding results. The academic performance of children from the drought-prone division (NIK) exhibits an upward trajectory compared to the baseline year of 2008. 5.3 Effect on School Dropout Rates The primary focus of this study is to assess the effects of drought on learning outcomes. One significant pathway through which drought may influence learning outcomes is by compelling children to withdraw from school. In the context of climate-related disasters such as droughts, households facing water scarcity encounter increased opportunity costs associated with maintaining their children’s enrollment in school (Glory and Nsikak-Abasi, 2013). To investigate this phenomenon and to provide additional support for our primary findings about learning outcomes, we employ the event study framework delineated in Equation 1 (see subsection 4.2) to examine the causal impact of drought on school dropout rates. We apply Equation 1 separately to dropout rates in Maharashtra and Karnataka, presenting the corresponding results in Figures 3 and 6 for Maharashtra and Karnataka, respectively. The full regression tables corresponding to these figures are provided in panel 3 of Appendix Tables A1 and A2. 17 Our analysis reveals a statistically significant increase in dropout rates during consecutive drought years in 2011–2012 in the drought-prone regions of Maharashtra, namely the Marathwada meteorological subdivision, followed by a recovery in subsequent years. In contrast, the school participation in the exposed group within Karnataka, namely the NIK meteorological subdivision, appears resilient to the impact of successive meteorological droughts. Moreover, positive trends emerge in the latter years, indicating the effectiveness of initiatives implemented in drought-prone areas of the state. These findings regarding the impact of drought on school participation complement our primary results on learning outcomes, providing additional support and confidence in our conclusions. 5.4 Results from an Alternative Identification Strategy In our main event study framework (Equation 1), we employed difference-in-difference regression equations to assess the impact of drought in comparison to regions less prone to drought within a state over time. Specifically, we conducted separate analyses for Maharashtra and Karnataka to examine the effects of drought with and without public sector preventive measures in place. By running two distinct event studies for each state, we avoided assuming that Maharashtra and Karnataka were identical in their counterfactual conditions. In this subsection, we propose an alternative identification strategy utilizing triple difference regression equations within the same event study framework. This approach allows for a more comprehensive comparison between states and facilitates the comparison of the drought-prone regions with their corresponding non-drought-prone regions within a single equation. Additionally, it allows us to test the robustness of our estimates. The proposed alternative identification strategy for each child ifrom district din state sduring the time period tis as follows: Yidst =αd+β1·(Droughtd∗States∗T imet) + β2·(Droughtd∗States) + β3·(States∗T imet) +β4·(Droughtd∗T imet) + β5·(States) + β6·(T imet) + βi·(Xi) + µidst, (2) where Yidst denotes the outcome variable for child iat time t.T imetis a dummy variable, varying across eight versions to create an event study framework. It assumes a value of 0 for the base year 2008 and a value of 1 for each subsequent year from 2009 to 2018, excluding 2015 and 2017 due to data unavailability in our primary dataset. Droughtdis a dummy variable indicating whether a child belongs to the drought-prone subdivisions or not, taking the value 1 for individuals from either Marathwada or NIK. Statesis a dummy variable that is assigned the value 1 for Maharashtra and 0 for Karnataka. Control variables consistent with Equation 1 are included, denoted by Xi.αdrepresents district fixed effects, and standard errors are clustered 18 at the district level. The results of the above-mentioned alternative identification strategy, depicted in Figures 7, 8, and 9 for the math score, reading score, and dropout rates, respectively, provide a combined picture of the earlier findings presented separately for both states. The triple difference point estimates illustrated in these figures reveal that, over time, prior to consecutive drought years, there existed no significant disparity between the two states concerning learning outcomes and school attendance. However, following the consecutive drought years of 2011–2012, the trends indicate that the drought-prone regions of Maharashtra face learning losses and higher school dropout rates, while the drought-prone regions of Karnataka display resilience and progress due to public sector interventions in drought-prone areas. 5.5 Results from an Alternative Reference Year In our primary empirical framework, we established 2008 as the baseline year for our event study, as our primary dataset starts from that year. However, in this particular subsection, for the robustness of our estimates, we shift the baseline year to 2010, the year immediately preceding a series of consecutive drought events. Like the year 2008, the year 2010 was also not marked by severe drought conditions. Consequently, Equation 1, as elaborated in the methodology section (subsection 4.2), remains unchanged except for the modification of the T imetdummy variable. In this analysis, the T imetvariable serves as a dummy variable with eight versions, each designed to construct an event study framework. Across these versions, it is assigned a value of 0 for the year 2010, which serves as the baseline year for all versions. Subsequently, for each version, the dummy variable takes on a value of 1 for one year at a time, namely 2008, 2009, 2011, 2012, 2013, 2014, 2016, and 2018. The outcomes of this analysis utilizing the alternative base year for the states of Maharashtra and Karnataka are depicted in Figures 10–15, where Figures 10, 11, and 12 pertain to math scores, reading scores, and dropout rates for Maharashtra, and Figures 13, 14, and 15 correspond to the same outcome variables for Karnataka. Notably, these findings for both states align closely with the primary results obtained with 2008 as the baseline year. Thus, it is evident that our conclusions remain consistent irrespective of the choice of the baseline year. 6 Discussion In our study, we examine the effects of drought and its recovery with and without preventative measures, using an event study approach. Since droughts are temporary and their impact areas may vary with each occurrence, we focus on meteorological subdivisions prone to drought. Specifically, we concentrate on the 19 semi-arid regions of the Deccan Plateau in two neighboring states, Maharashtra and Karnataka, which experience frequent severe droughts in India. We provide empirical evidence supporting the relevance assumption for our strategy that the drought-prone subdivisions experienced severe drought compared to other areas within their respective states during the study period. We compare the occurrence of drought in these drought-prone subdivisions between the two states. In Maharashtra, we observe learning losses in its drought-prone subdivisions following drought events, while in Karnataka, such losses are not evident. We hypothesize that this discrepancy is due to the presence of drought-preventative measures in one state and their absence in the other. However, one concern may arise regarding the interpretation of our findings if Karnataka simply provides better education. Our analysis, employing a triple difference setup to compare both states and their drought-prone regions with a baseline year, disproves this concern. In fact, our summary statistics indicate that, on average, Maharashtra performed slightly better in educational outcomes compared to Karnataka. Additionally, drought-prone regions within Maharashtra also demonstrated superior academic performance compared to those in Karnataka. Another concern could relate to the drought intensity and the timeline of drought occurrence in the drought-prone regions of both states. However, despite being administratively distinct, the meteorological subdivisions in both states share borders and belong to the same arid zone, experiencing similar climatic conditions and encountering comparable drought incidents. Moreover, we also demonstrate using our dataset and through secondary sources that both regions experienced similar drought incidents in terms of intensity and timing during the study period. Thus, our main findings regarding test scores and school participation support the hypothesis that preventive measures in one state contribute to education resilience in its drought-prone regions. These preventive measures were primarily targeting agriculture, a crucial pathway through which meteorological drought affects human capital investment in agriculture-dependent economies. Therefore, our study suggests that with such preventive measures in place, the drought-prone regions in agriculture-dependent economies may still face meteorological droughts, but the severity of their impact may be mitigated. Furthermore, we acknowledge the role of corporate initiatives in combating water crises in droughtprone regions. While private sector efforts complement those of the public sector, they often operate on a smaller scale and with limited geographic coverage. Recognizing the potential limitations of district-level observation and the time required for test scores to manifest, we shift our focus to expenditure patterns. Since the main focus of our paper is on educational outcomes, we place this analysis of private sector drought prevention effectiveness and the associated results in the online appendix, highlighting increased educationrelated expenses in regions with a higher density of private sector drought prevention projects. However, we caution against drawing causal conclusions due to limitations in data granularity and the potential influence 20 of other private sector projects on our comparison group. Moving forward, primary research focusing on collecting data at lower administrative levels could provide more targeted insights into the effectiveness of private sector drought preventive measures. Our study highlights the importance of collective efforts involving all stakeholders in addressing water scarcity. While our findings demonstrate the positive impacts of government measures on education resilience, they also indicate the need for further research to assess the causal impact of private sector initiatives. Such efforts would contribute to a comprehensive understanding of the collective actions needed to address water scarcity effectively. 21 References Accidental Deaths & Suicides in India (ADSI). (2022). NCRB. https://ncrb.gov.in/en/accidental-deathssuicides-india-adsi Agamile, P., Dimova, R., & Golan, J. (2021). Crop choice, drought and gender: New insights from smallholders’ response to weather shocks in rural Uganda. Journal of Agricultural Economics, 72(3), 829–856. https://doi.org/10.1111/14779552.12427 Amjath-Babu, T. S., Krupnik, T. J., Aravindakshan, S., Arshad, M., & Kaechele, H. (2016). Climate change and indicators of probable shifts in the consumption portfolios of dryland farmers in Sub-Saharan Africa: Implications for policy. Ecological Indicators, 67, 830–838. Amrit, K., Pandey, R. P., & Mishra, S. K. (2018). Assessment of meteorological drought characteristics over Central India. Sustainable Water Resources Management, 4(4), 999–1010. https://doi.org/10.1007/s40899-017-02055 Angassa, A., & Oba, G. (2008). Herder perceptions on impacts of range enclosures, crop farming, fire ban and bush encroachment on the rangelands of Borana, Southern Ethiopia. Human Ecology, 36, 201–215. Arshad, M., K¨achele, H., Krupnik, T. J., Amjath-Babu, T. S., Aravindakshan, S., Abbas, A., ... & M¨uller, K. (2017). Climate variability, farmland value, and farmers’ perceptions of climate change: implications for adaptation in rural Pakistan. International Journal of Sustainable Development & World Ecology, 24(6), 532-544. Arshad, M., Amjath-Babu, T. S., Aravindakshan, S., Krupnik, T. J., Toussaint, V., K¨achele, H., & M¨uller, K. (2018). Climatic variability and thermal stress in Pakistan’s rice and wheat systems: A stochastic frontier and quantile regression analysis of economic efficiency. Ecological Indicators, 89, 496–506. Barrios, S., Ouattara, B., & Strobl, E. (2008). The impact of climatic change on agricultural production: Is it different for Africa? Food Policy, 33(4), 287–298. https://doi.org/10.1016/j.foodpol.2008.01.003 Benson, C., & Clay, E. J. (2004). Understanding the economic and financial impacts of natural disasters (No. 4). World Bank Publications. Bhandari, D. C., Meghwal, P. R., & Lodha, S. (2014). Horticulture based production systems in Indian arid regions. In Sustainable Development and Biodiversity (pp. 19–49). Springer International Publishing. Bj¨orkman-Nyqvist, M. (2013). Income shocks and gender gaps in education: Evidence from Uganda. Journal of Development Economics, 105, 237–253. https://doi.org/10.1016/j.jdeveco.2013.07.013 Chatterjee, S., Hastawala, S., & Kamal, J. (2023). (En-)‘lightening’children: Assessing the impacts of access to electricity on learning achievement levels. Review of Development Economics. DAC (2009). Manual of Drought Management, Department of Agriculture & Cooperation, Government of India Dai, A. (2013). Increasing drought under global warming in observations and models. Nature Climate Change, 3(1), 52–58. https://doi.org/10.1038/nclimate1633 Department of Agriculture and Farmers Welfare, Government of India. (2016, December). Manual for Drought Management. https://agriwelfare.gov.in/Documents/Updated%20Drought%20Manual 0.pdf 22 Figure 2: Results for reading scores from baseline scenario Note: The figure is based on Equation 1, comparing Marathwada and the rest of Maharashtra in the event study framework with 2008 as the reference period. All points represent coefficients from different regressions of reading test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. Figure 3: Results for dropout rate from baseline scenario Note: The figure is based on Equation 1, comparing Marathwada and the rest of Maharashtra in the event study framework with 2008 as the reference period. All points represent coefficients from different regressions of the dropout rate and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 29 Figure 4: Results for math scores with public sector drought preventive measures Note: The figure is based on Equation 2, comparing NIK and the rest of Karnataka in the event study framework with 2008 as the reference period. All points represent coefficients from different regressions of math test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 30 Figure 5: Results for reading scores with public sector drought preventive measures Note: The figure is based on Equation 2, comparing NIK and the rest of Karnataka in the event study framework with 2008 as the reference period. All points represent coefficients from different regressions of reading test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. Figure 6: Results for dropout rate with public sector drought preventive measures Note: The figure is based on Equation 2, comparing NIK and the rest of Karnataka in the event study framework with 2008 as the reference period. All points represent coefficients from different regressions of the dropout rate and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 31 Figure 7: Results for math scores using alternative identification strategy Note: The figure is based on Equation 2, comparing drought-prone regions against those not prone to drought in two states, namely Maharashtra and Karnataka, over time. This analysis is conducted within an event study framework, with the year 2008 as the reference period. All points represent coefficients from different regressions of math test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 32 Figure 8: Results for reading scores using alternative identification strategy Note: The figure is based on Equation 2, comparing drought-prone regions against those not prone to drought in two states, namely Maharashtra and Karnataka, over time. This analysis is conducted within an event study framework, with the year 2008 as the reference period. All points represent coefficients from different regressions of reading test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. Figure 9: Results for dropout rates using alternative identification strategy Note: The figure is based on Equation 2, comparing drought-prone regions against those not prone to drought in two states, namely Maharashtra and Karnataka, over time. This analysis is conducted within an event study framework, with the year 2008 as the reference period. All points represent coefficients from different regressions of dropout rates and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 33 Figure 10: Results using alternative reference year for math scores from baseline scenario Note: The figure is based on Equation 1, comparing Marathwada and the rest of Maharashtra in the event study framework with 2010 as the reference period. All points represent coefficients from different regressions of math test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. Figure 11: Results using alternative reference year for reading scores from baseline scenario Note: The figure is based on Equation 1, comparing Marathwada and the rest of Maharashtra in the event study framework with 2010 as the reference period. All points represent coefficients from different regressions of reading test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 34 Figure 12: Results using alternative reference year for dropout rate from baseline scenario Note: The figure is based on Equation 1, comparing Marathwada and the rest of Maharashtra in the event study framework with 2010 as the reference period. All points represent coefficients from different regressions of the dropout rate and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 35 Figure 13: Results using alternative reference year for math scores with public sector drought preventive measures Note: The figure is based on Equation 1, comparing NIK and the rest of Karnataka in the event study framework with 2010 as the reference period. All points represent coefficients from different regressions of math test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 36 Figure 14: Results using alternative reference year for reading scores with public sector drought preventive measures Note: The figure is based on Equation 1, comparing NIK and the rest of Karnataka in the event study framework with 2010 as the reference period. All points represent coefficients from different regressions of reading test scores and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. Figure 15: Results using alternative reference year for dropout rate with public sector drought preventive measures Note: The figure is based on Equation 1, comparing NIK and the rest of Karnataka in the event study framework with 2010 as the reference period. All points represent coefficients from different regressions of the dropout rate and include demographic controls and district fixed effects. Vertical lines represent 90% confidence intervals. 37 Appendix Figure A1: Results for rainfall using primary empirical framework Note: The figure is based on Equation 1, comparing rainfall (in millimeters) in Marathwada and the rest of Maharashtra in the event study framework with 2008 as the reference period. All points represent coefficients from different regressions of average annual rainfall. Vertical lines represent 90% confidence intervals. 38