Beyond Isolated Effects: Rethinking Causal Questions in the Presence of Linked Attributes

When researchers study how people evaluate individuals, groups, or scenarios, they often attempt to isolate the effect of one cue ``net of'' another by randomizing attributes in conjoint or vignette designs. We show that such isolated effects are neither conceptually coherent nor empirically identifiable when the attributes are cognitively linked. Because linked attributes jointly define how a profile is evaluated, altering one cue necessarily changes the interaction with other, linked attributes, especially its level of atypicality—the extent to which the profile violates stereotype-based expectations. Focusing on the party-policy debate in polarization research using two pre-registered conjoint experiments, we demonstrate the impossibility of isolating net effects of linked attributes. We urge a pivot toward theorizing and estimating conditional or joint effects that take attribute linkage seriously. This shift aligns causal questions with cognitive reality and yields more interpretable evidence on how multiple cues jointly drive evaluation.