Article summary
Dropping subjects after random assignment because they fail a manipulation check is common across the social sciences, usually intended to restrict estimates to those thought to understand the prompt. This practice can, however, introduce serious bias and obscures what the experiment actually reveals unless handled differently.