Abstract

This study makes a novel contribution to the climate–economics literature by systematically demonstrating how rainfall data choice and spatial processing decisions can alter econometric estimates of drought impacts on household food insecurity. While many applied studies assume that rainfall datasets are interchangeable or differ only by scale, the analysis here shows that empirical results may depend as much on dataset resolution, observational inputs, and interpolation methods as on the underlying relationship being studied. Using nationally representative household survey data from Mali and Chad, rainfall shocks are measured with the Standardized Precipitation Index and assigned to households using three common spatial interpolation techniques: bilinear interpolation, nearest-neighbor assignment, and zonal aggregation. The study compares across multiple rainfall products representing satellite-based, gauge-based, blended, and reanalysis sources, thereby providing the first comprehensive assessment of how these technical decisions influence estimates of climate impacts on household welfare. The results reveal two central findings. First, in models without household and community controls, estimated coefficients display substantial heterogeneity, ranging from strongly negative values to weak or even positive associations that contradict theoretical expectations. This volatility is especially pronounced for coarse-resolution and satellite-only products, while higher-resolution blended sources show relatively greater stability. Importantly, the heterogeneity is not limited to coefficient size but also extends to the relative ranking of datasets: across interpolation methods, the order of estimated effects shifts and sometimes reverses, demonstrating that rainfall products are not affine transformations of one another but instead embody differential measurement error. Second, when household- and community-level controls are added, the heterogeneity largely disappears, and coefficients converge toward consistently negative values consistent with theory that higher rainfall reduces food insecurity by improving agricultural outcomes and household resources. The inclusion of controls dampens within-method variation and cross-dataset instability, underscoring the importance of local socioeconomic context for empirical precision. Nonetheless, differences in magnitude remain, reflecting dataset resolution and construction features. High-resolution, gauge-calibrated products consistently yield stronger and more stable coefficients, while coarse or satellite-only sources continue to produce weaker and noisier associations. Interpolation methods also matter: zonal aggregation amplifies coefficients by smoothing local anomalies, bilinear interpolation moderates effects by weighting across nearby grid cells, and nearest-neighbor assignment preserves localized variation but remains more volatile. A particularly important contribution of this study is to highlight the instability of ordinality, or dataset rankings, across both interpolation methods and model specifications. Without controls, rankings differ markedly depending on the interpolation technique, and with controls, rankings shift again, sometimes reversing entirely. This finding has broad implications for applied research: conclusions about which datasets perform “best” or yield the strongest associations with welfare outcomes cannot be taken for granted and may depend heavily on processing decisions. For applied economics, the implications are profound. Because researchers often treat rainfall data choice and interpolation as technical details, there is a risk that empirical findings could vary substantially simply due to unreported or arbitrary methodological choices. In practice, this means that two otherwise similar studies might report divergent conclusions not because the underlying relationship differs, but because of the rainfall dataset and processing methods used. Such sensitivity also creates scope for selective reporting, whether intentional or inadvertent, when researchers choose rainfall sources post hoc. At the same time, the study points toward constructive solutions. Greater transparency in documenting rainfall data selection, spatial processing, and model specification, coupled with systematic robustness checks across multiple datasets, can greatly strengthen the credibility and reproducibility of empirical findings. By carefully acknowledging how measurement error and spatial assignment shape results, researchers can provide more reliable evidence to inform policy design in climate-vulnerable contexts. In summary, this study advances the literature by showing that rainfall shocks are indeed linked to household food insecurity in Mali and Chad in theoretically consistent ways, but that the stability and strength of these estimates critically depend on dataset resolution, observational inputs, and interpolation methods. The broader lesson is that climate–economics research must treat data choice and processing not as secondary technicalities but as central determinants of inference, making transparency and robustness indispensable for credible policy-relevant evidence.