Human development and disaster mortality: evidence from India

Natural disasters disrupt economic activities, damage assets and infrastructure, and can push households into poverty traps, with long-term adverse effects on well-being and human capital. Prior evidence shows disaster mortality is higher in developing countries due to limited coping capacity, weaker infrastructure, and socioeconomic vulnerabilities. Focusing solely on per capita income (PCI) is insufficient as PCI does not capture the multidimensional aspects of development. This study argues that a composite measure—HDI, incorporating income, health, and education—better reflects inclusive development relevant to disaster risk reduction. India faces recurrent floods and cyclones with substantial human and economic losses, and large interstate heterogeneity in development and exposure. Research question: Do higher levels of human development reduce human fatalities from floods and cyclones, and over what time horizons? Purpose and contributions: Using panel data for 19 Indian states (1983–2011), the study (i) examines the effect of HDI at time t on flood mortality in t+1, t+2, t+3; (ii) extends to cyclone-related deaths over the same horizons; (iii) assesses total fatalities (floods+cyclones); (iv) incorporates measures of disaster occurrence/severity; and (v) addresses HDI endogeneity via an IV Poisson control function approach using state-wise drought-prone area as an instrument. The paper posits that higher HDI should lower mortality by enabling relocation, improving individual and institutional preparedness, strengthening infrastructure, and enhancing access to education and health services.

The literature links development to disaster impacts through multiple channels. Studies document that higher PCI is generally associated with fewer disaster deaths, but the relationship can be non-linear and mediated by governance and inequality (Kahn, 2005; Kellenberg & Mobarak, 2008; Strömberg, 2007; Fankhauser & McDermott, 2014; Raschky, 2008). Developing countries bear disproportionate human losses due to limited early warning, evacuation systems, and infrastructure (UNDRR 2022; Bradshaw, 2003). Vulnerability correlates with low income, weak health and education, and institutional capacity (Nirupama, 2012; Zhou et al., 2014). HDI emerges as a determinant of resilience: higher HDI is associated with lower casualties (Feng et al., 2014; Prasojo et al., 2021; Baradan et al., 2019). Government effectiveness and investment in quality infrastructure reduce damages and deaths (Tennant & Gilmore, 2020; Taghizadeh-Hesary et al., 2019, 2021; Aschauer, 1990; Aldrich, 2023). Inequality can exacerbate disaster outcomes and impede collective mitigation (Anbarci et al., 2005; Cappelli et al., 2021). Disasters also have persistent effects on human capital, especially for children and the poor (Alderman et al., 2006; Anttila-Hughes & Hsiang, 2013; Baez et al., 2010). The review highlights a gap on reverse causality: while many examine disaster impacts on HDI, fewer assess how HDI improvements reduce direct non-market losses (fatalities). This study addresses that gap in the Indian context.

Data and variables: Panel dataset for 19 Indian states from 1983 to 2011. Disaster fatalities (floods, cyclones) are from Accidental Deaths & Suicides in India (ADSI), Government of India. Flood-affected areas from Central Water Commission (CWC) reports. HDI (inequality-adjusted) is compiled from Mukherjee et al. (2016) for years 1983, 1987, 1993, 1999, 2004, 2009, 2011. To mitigate endogeneity and align timing, HDI at year t is matched with outcomes averaged over subsequent windows: for HDI in 1983, average fatalities are computed for 1984–1986; for 1987, the average over 1988–1992; and similarly for later rounds. Controls include: lagged credit-deposit ratio (financial development; EPW database), lagged forest cover (ecological resilience; Land Use Statistics), per capita social security and calamity expenditure (PCSCE; Reserve Bank of India), rainfall (IMD Pune), state population (census years with linear interpolation), and indicators of flood/cyclone severity and affected areas (IMD Disastrous Weather Events reports). Drought-prone area by state is also compiled (per Parida, 2020). Econometric strategy: Because fatalities are non-negative counts and over-dispersed, the study estimates fixed-effects (FE) Poisson models as the main specification, with FE Negative Binomial as robustness. Models include region fixed effects (to absorb time-invariant regional heterogeneity) and year fixed effects (to capture national time-varying shocks/policies). Standard errors use sandwich (robust) estimators to allow for deviations from Poisson assumptions. Core specifications: - Flood fatalities at t+1: FF_it+1 = exp{β1 HDI_it + β2 CR_it−1 + β3 FC_it−1 + β4 ln(PCSCE)_it−1 + β5 Z_it + α_i + λ_t + μ_it}. - Cyclone fatalities at t+1: CF_it+1 analogous. - Total fatalities (flood+cyclone): TF_it+1 analogous, with inclusion of relevant severity/occurrence controls in Z_it. Endogeneity and IV (Control Function): HDI may be endogenous due to reverse causality (e.g., disasters depress income/health/education) and omitted time-varying factors. Instrument: state-wise drought-prone area (SDPA), expected to be negatively correlated with HDI but affecting fatalities only through HDI. First stage: HDI_it = π1 SDPA_it + controls + region and year fixed effects. Residuals from the first stage are included in the FE Poisson structural equation (IV Poisson via control function) to purge endogeneity. Tests: Significant negative SDPA coefficient in the first stage supports instrument relevance; significance of the residual term confirms endogeneity. Average marginal effects (AME) are computed to express impacts in counts of expected fatalities.

• Higher HDI reduces disaster fatalities. FE Poisson estimates show: • Flood fatalities decline by about 2.2%, 3.7–4.9% over t+1 to t+3 per one-unit increase in HDI (coefficients: −0.022, −0.037, −0.049). AME: expected flood deaths fall by approximately 85 (t+1), 141 (t+2), and 173 (t+3). • Cyclone fatalities decline with HDI: coefficients are negative and significant especially by t+3 (about −5%). AME: expected cyclone deaths fall by about 12 (t+1), 6 (t+2), and 69 (t+3). • Total fatalities (floods+cyclones) fall with HDI. AME: roughly 109 (t+1), 183 (t+2), and 247 (t+3) fewer expected deaths per one-unit increase in HDI. - Robustness: FE Negative Binomial estimates yield similar negative and significant effects of HDI on flood, cyclone, and total fatalities; AME magnitudes are comparable. - IV (Control Function) results: Using drought-prone area as an instrument, HDI remains negative and significant in flood, cyclone, and combined fatalities models, corroborating a causal interpretation. First-stage estimates show SDPA strongly and negatively associated with HDI; the control-function residual term is significant, indicating endogeneity. - Other determinants: • Higher per capita social and calamity expenditure tends to reduce fatalities (often significant, especially at longer horizons and in cyclones). • Greater credit availability (lagged credit ratio) is generally associated with fewer fatalities, particularly in Negative Binomial results. • Higher population is associated with more cyclone fatalities. • Greater disaster severity/occurrence indicators significantly raise fatalities. • Forest cover effects are mixed; some specifications suggest protective effects.

The findings directly address the research question by demonstrating that improvements in human development—capturing income, health, and education—are associated with and causally reduce mortality from floods and cyclones. This supports the view that focusing solely on raising PCI is insufficient; multidimensional progress enhances preparedness, enables mobility away from high-risk zones, strengthens demand for and use of resilient infrastructure, and improves individual capacity to respond to warnings. The significance of government social/calamity expenditure and financial development indicates that institutional responsiveness and access to credit complement human development in reducing fatalities. Results remain robust across models and under an IV strategy, suggesting that endogeneity does not overturn the conclusion. The ecological and institutional context also matters: population pressure and hazard severity increase deaths, while investments in resilience (infrastructure, early warning, DRR spending) mitigate them. For India, heterogeneity across states (e.g., low-HDI states like Bihar, Uttar Pradesh, Odisha) aligns with higher mortality burdens, underscoring equity-focused, state-specific policy design. The study contributes to disaster economics by highlighting the reverse channel—how HDI reduces direct non-market losses (deaths)—and quantifies medium-term effects over multiple years.

The study shows that Indian states with higher inequality-adjusted HDI experience significantly fewer fatalities from floods and cyclones, with effects strengthening over subsequent years. Using FE Poisson, FE Negative Binomial, and IV Poisson (control function) models, the results are consistent and robust. Policy implications: prioritize investments in human capital—education, healthcare, and living standards—as core components of disaster risk reduction. Strengthen ex-ante budget allocations for DRR; enhance early warning and forecasting; build flood- and cyclone-resilient infrastructure; expand social protection and institutional capacity; increase forest cover and improve environmental management; and deepen financial inclusion to facilitate household and community resilience investments. Comprehensive, long-term disaster management strategies that integrate human development with infrastructure and institutional improvements can substantially reduce mortality in hazard-prone regions.

• Temporal and measurement constraints: HDI is observed only in discrete rounds (1983, 1987, 1993, 1999, 2004, 2009, 2011), requiring averaging of fatalities over subsequent windows and potentially smoothing short-term dynamics. - Scope of hazards: The analysis focuses on floods and cyclones; results may not generalize to other hazards (e.g., earthquakes, heatwaves). - Data aggregation: State-level panel data may mask within-state heterogeneity in exposure, vulnerability, and institutional capacity. - Endogeneity mitigation relies on an instrument (state-wise drought-prone area) with standard exclusion restrictions; while first-stage relevance is strong, untestable exclusion assumptions remain. - Population is interpolated between census years, which may introduce measurement error. - Official disaster and expenditure data may contain reporting or classification inconsistencies across states and years.