Signal detection theory – what helps preventing misdiagnoses and false positives in general can’t be bad for diagnosing Amusia either, one would think. Our very own Molly Henry and her former supervisor Devin McAuley now demonstrate in a just-accepted paper
Failure to apply signal detection theory to the Montreal Battery of Evaluation of Amusia may misdiagnose amusia
in Music Perception that this is indeed the case:
They show that analyses based on confidence ratings and ROC-curves outperforms simple percentage correct in diagnosing Amusia.
Here is the abstract, and watch out for the full paper to appear soon:
This article considers a signal detection theory (SDT) approach to evaluation of performance on the Montreal Battery of Evaluation of Amusia (MBEA). One hundred fifty-five individuals completed the original binary-response version of the MBEA (n = 62) or a confidence rating version (MBEA‑C; n = 93). Confidence ratings afforded construction of empirical receiver operator characteristic (ROC) curves and derivation of bias-free performance measures against which we compared the standard performance metric, proportion correct (PC), and an alternative signal detection metric, d’. Across the board, PC was tainted by response bias, and underestimated performance as indexed by Az, a nonparametric ROC-based performance measure. Signal detection analyses further revealed that some individuals performing worse than the standard PC-based cutoff for amusia diagnosis showed large response biases. Given that PC is contaminated by response bias, this suggests the possibility that categorizing individuals as having amusia or not using a PC-based cutoff may inadvertently misclassify some individuals with normal perceptual sensitivity as amusic simply because they performed with large response biases. In line with this possibility, a comparison of amusia classification using d’- and PC-based cutoffs showed potential misclassification of 33% of the examined cases.