Small area variations in four measures of poverty among Indian households: Econometric analysis of National Family Health Survey 2019-2021

India has witnessed significant poverty reduction in recent decades, yet this progress has been uneven across the country. The 2011 census indicated that approximately 22% of the Indian population lived on less than USD 1.90 per day, highlighting the persistent challenge of poverty. This study investigates the spatial distribution of poverty within India to better understand the heterogeneity of poverty and inform the design of effective poverty eradication policies. The research uses four distinct measures of household poverty: (a) bottom 10th wealth percentile; (b) bottom 20th wealth percentile; (c) below the poverty line (BPL) status, as determined by possession of a BPL card providing subsidized grain; and (d) multidimensional poverty (MDP), encompassing deprivations in health, education, and living standards. The study leverages the rich dataset of the fifth round of the National Family Health Survey (NFHS-5), conducted from 2019 to 2021, to analyze these variations at multiple geographic levels.

Existing research has documented substantial income and wealth inequality within India, examining disparities between districts and states. Studies have explored various contributing factors, including inter-caste inequality, regional agricultural variations, and differences in infrastructure investments. However, the fine-grained spatial variation of poverty within districts and between clusters remains understudied. This gap in understanding motivates the current study to delve into smaller geographic units and their contribution to the overall picture of poverty in India. Understanding the spatial distribution of poverty at the cluster level is crucial for effective policy targeting, as previous studies have shown the critical role of clusters in shaping various social and health outcomes in India.

The analysis utilizes data from NFHS-5, collected through a two-stage cluster sampling strategy. The primary sampling units (PSUs) were clusters of adjacent households. The first stage involved selecting rural and urban clusters, while the second stage involved selecting households from within these clusters. The dataset includes information on 2,795,894 household members nested within 30,170 clusters, 707 districts, and 36 states/union territories. Four-level variance component models were employed to decompose the geographic variation in poverty across clusters, districts, and states. These models considered the binary nature of the four poverty measures. The proportion of variation attributable to each geographic level was calculated by dividing the variance at that level by the total geographic variation. Precision-weighted estimates were generated for clusters and districts for each outcome. The standard deviations of these cluster-level values were then calculated by district to assess within-district and between-cluster variations in poverty.

The study revealed several key findings. First, moderate to strong positive correlations were observed between the four measures of poverty at the district level. Second, the largest share of geographic variation was attributed to states for bottom 10th and 20th wealth percentiles and BPL households, while clusters accounted for the largest share of variation for MDP. Districts consistently showed the least variation across all four measures. Third, a wide range of within-district between-cluster standard deviations was observed for all four poverty measures, highlighting substantial small area variation. Maps illustrating the geographic distribution of each measure at the district and cluster levels revealed clusters of poverty in northern, central, and parts of eastern India. Fourth, significant positive correlations were found between district-level poverty prevalence and cluster-level standard deviations for the bottom 10th and 20th wealth percentiles and MDP. A slight negative correlation existed for BPL.

The findings underscore the significance of considering smaller geographic units, particularly clusters, when designing and implementing anti-poverty policies. The substantial within-district variation highlights the need to move beyond a one-size-fits-all approach and tailor interventions to specific local contexts. The study's results also illuminate the multidimensional nature of poverty, as different poverty measures exhibit varying geographic patterns. This emphasizes the need to consider multiple dimensions of poverty to gain a comprehensive understanding of its distribution and design effective, targeted interventions.

This study expands upon existing research by demonstrating substantial small area variation in poverty within districts and between clusters in India. Policymakers should consider this small-scale heterogeneity when designing and implementing anti-poverty programs to ensure equitable advancement and avoid a one-size-fits-all approach. Future research should investigate the contextual factors contributing to these variations and explore how policies can be tailored to address specific local needs.

Two key limitations of this study are the self-reported nature of some household wealth questions in the NFHS (though NFHS data is generally considered high quality), and the lack of adjustment for sociodemographic factors like caste and household head education in the precision-weighted estimates, which could introduce bias. These factors should be considered in future research.