Validity of Inferred Differences at Low Effect Sizes

Establish whether Value Faces’ system-inferred differences in demonstrated values are valid when the standardized cross-context effect size is not statistically significant, specifically in the regime where $|d| < 0.22$.

Background

The evaluation validates system inferences against participants’ perceptions over the range of effect sizes represented in the study, but it excludes the narrow region in which inferred differences are not statistically significant. In that regime, a discrepancy between system inference and user perception could reflect genuine variation below conscious awareness or overfitting to conversational noise. Because user perception may not provide reliable ground truth there, the validity of the system’s inferences remains unresolved.