This paper revisits the instrumental variable (IV) approach in Lipscomb et al. (2013, LMB) to study the impacts of electrification. We first make corrections to the construction of the dataset, including the modelled IV. We find that results differ in repeated runs of the simulations, with mostly weak instruments. After making corrections to the data, the main outcome, the Human Development Index, and most originally identified mechanisms turn insignificant, even after allowing for revisions along robustness checks proposed in LMB. We second develop a framework that accounts for weak IVs and discourages specification screening. Applying it to LMB, we find that most theoretically justified specifications yield insignificant results. The proposed framework is transferable to other IV applications to reduce potential bias stemming from researcher’s or replicator’s discretion.