Design-Based Inference
An approach to uncertainty in which randomness comes from the assignment mechanism (which units are treated), with potential outcomes treated as fixed — as opposed to sampling-based inference, where randomness comes from drawing a sample from a larger population. The distinction matters for what the standard error means: when the sample is (nearly) the whole population of interest, sampling-based standard errors can be too large, and the relevant uncertainty is design-based (about the counterfactual assignment), sometimes implying smaller or differently-computed standard errors. This reframing underlies randomization inference in experiments, the local-randomization view of RD, and design-based DiD.
Relied on by
Randomized experiments, regression analysis where the population is fully observed, design-based DiD and RD.