Theory, external validity, and experimental inference: Some conjectures

It is often argued that experiments are strong on causal identification (internal validity) but weak on generalizability (external validity). One widely accepted way to limit threats to external validity is to incorporate as much variation in the background conditions and in the covariates as possible through replication. Another strategy is to make the theoretical foundations of the experiment more explicit. The latter requires that we develop trajectories of experiments that are consistent with a theoretical argument. In other words, new experiments should not simply consist of changing the context of old ones, but do so in ways that explicitly test various aspects of a theory in a coherent way.