Accounting for speed-accuracy trade-offs in developmental prosopagnosia

Despite severe everyday problems recognising faces, some individuals with developmental prosopagnosia (DP) achieve typical accuracy scores on objective face recognition tests. This has led to calls to also examine response time (RT) since RT is often longer in DP relative to control participants. Here, 24 DPs and 110 age-matched controls completed four tasks: The Cambridge Face Memory Test (CFMT); the Cambridge Bicycle Memory Test; an Old New Faces delayed recognition task, and a Famous Faces Test. We used accuracy and the Balanced Integration Score (BIS), a measure that adjusts accuracy for RT (calculated as Z accuracy minus Z RT), to classify our sample at group and individual level. Subjective face recognition ability was assessed using the PI20 questionnaire and interviews. Old New Faces data showed a larger between-group effect size and stronger evidence for group differences using BIS compared with accuracy (d = 1.24, BF10 = 5,090,000 and d = 0.60, BF10 = 28.8, respectively). On a derived global (averaged) memory measure, we similarly observed a larger between-group effect size and stronger evidence for group differences using BIS (Cohen’s d = 1.63, BF10 = 280,000,000) vs accuracy alone (Cohen’s d = 1.39, BF10 = 2,830,000). On this global face memory measure, 16/24 DPs showed a major impairment of at least 1.7 standard deviations below age-matched control means using BIS compared with only 5/24 DPs using accuracy. Logistic regression indicated that a model incorporating the BIS measures was the most sensitive for classifying DP, showing the highest area under the curve (AUC = .904) and correctly classifying 68.2% of the DP group. For comparison, the ‘gold-standard’ CFMT accuracy score alone correctly classified only 16.7% of the DP group. BIS is thus a sensitive novel measure for accounting for speed-accuracy trade-offs that can otherwise mask impairment measured only by accuracy in DP.