From simulation to reality: Empirical validation of the superiority of da in indexing sensitivity in recognition-memory experiments.

The selection of an accuracy (sensitivity) measure is a pivotal decision faced by researchers who use binary judgments.An abundance of measures has been developed to index sensitivity independent of participants' bias.However, recent Monte Carlo simulations ran in our lab found that all commonly used measures confound sensitivity with bias, with the exception of our proposed version of the signal-detection measure d-sub-a (denoted d a ), based on confidence ratings rather than binary responses.Nevertheless, empirical confirmation in real-world experiments is needed to validate any measure.To this end, we ran two large-scale recognition studies that can serve as a proof of concept to all binary tasks.Sensitivity was not manipulated at encoding, while we manipulated bias at test.When comparing these iso-sensitive conditions, valid measures should yield erroneous significant results at approximately 5% (or α).Using resampling, we repeatedly ran the experiment to derive rates of Type-I errors.Only d a , but not P r , A′, or d′, demonstrated the desired 5% rate of Type-I errors across different experimental designs, sample sizes, and number of trials.We also used our data to assess statistical power.We compared d a 's ability to detect genuine sensitivity differences and found detection rates similar to all other measures tested.Combined with our simulation results, these empirical findings provide compelling evidence that d a is a valid and powerful sensitivity measure.Our findings offer a validated measure that can reduce the hazards of the measurement crisis that plague recognition-memory research.Our large data sets, available online, provide opportunities for future analyses.