Unconscious mental content in implicit evaluation: Evidence from misprediction

The tide is turning against a major idea in social psychology: that implicit evaluations reflect mental content that lies beyond conscious awareness. This view is being reconsidered in light of mounting evidence that people can predict their own implicit evaluations with high accuracy. However, there are reasons to question whether such predictive accuracy reflects introspective access. First, prior studies have relied almost exclusively on familiar targets (e.g., racial groups), allowing predictions to be informed by background knowledge (e.g., knowing that a group is stigmatized) rather than introspection. Second, implicit and explicit evaluations have been highly correlated in prior work, enabling accurate predictions simply by assuming that implicit evaluations mirror explicit ones. Here we report eight experiments (five preregistered; N = 6,794) designed to minimize these non-introspective routes to predictive accuracy. We introduced participants to novel targets and shifted implicit and explicit evaluations of these targets in opposite directions, rendering explicit evaluations an unreliable cue. Under these conditions, predictive accuracy ranged from low to non-existent: participants frequently anticipated shifts in their implicit evaluations in the opposite direction of the actual chance. These results generalized across two learning paradigms (impression formation and attribute conditioning), two implicit evaluation measures (IAT and EPT), and between-participant and within-participant designs. We consider multiple interpretations of these findings, including the possibility that implicit evaluations reflect mental content that is largely inaccessible to conscious awareness.