Normative data for accuracy and response times at the computerized Benton Facial Recognition Test (BFRT-c)

Abstract

We report normative data from a large (N=307) sample of young adult participants tested with a computerized version of the long form of the classical Benton Facial Recognition Test (BFRT; Benton & Van Allen, 1968). The BFRT-c requires to match a target face photograph to one or three of 6 face photographs simultaneously presented. Percent accuracy at the BFRT-c (81%-83%) is below ceiling yet well above chance level, with little interindividual variance in this typical population sample, as expected from a sensitive clinical test. While split-half reliability on accuracy rates is relatively low due to the large variability in difficulty across items, correct response times (RTs) measured in this version – completed in 3 minutes on average - provide a reliable and critical complementary measure of performance at individual unfamiliar face matching. In line with previous observations from other measures, females outperform male participants at the BFRT-c, especially for female faces. In general, performance is also lower following lighting changes than head rotations, in line with previous studies emphasizing limited ability at matching pictures of unfamiliar faces with important variations in lighting direction. Overall, this normative data set supports the validity of the BFRT-c as a key component of a battery of tests identifying clinical impairments at individual face recognition such as observed in acquired prosopagnosia. However, beyond global indexes of performance based on accuracy rates only, this analysis strongly recommends full consideration of the time taken to match individual faces as well as the variability in performance across items.

Understanding human individuation of unfamiliar faces with oddball fast periodic visual stimulation and electroencephalography

“…Subsequently, researchers developed behavioral tests with schematic faces or photographs of unfamiliar faces in order to evaluate and characterize FI ability in neurotypical populations (e.g., Laughery, Alexander, & Lane, 1971; Yin, 1969; see Ellis, 1975 for a review of early studies). Many laboratories developed a variety of computerized assessments of this function in the following decades, and a number of these digital tests are now widely available, together with normative data (e.g., the Cambridge Face Memory Test [CFMT]: Bowles et al., 2009; Duchaine & Nakayama, 2006; see also Russell, Duchaine, & Nakayama, 2009 for a longer version of the test, the CFMT+; the Cambridge Face Perception Test [CFPT]: Duchaine, Germine, & Nakayama, 2007; the Glasgow Face Matching Test [GFMT]: Burton, White, & McNeill, 2010; the Caledonian Face Test: Logan, Wilkinson, Wilson, Gordon, & Loffler, 2016; the Kent Face Matching Test [KFMT]: Fysch & Bindemann, 2018; and the computerized version of the Benton Face Recognition Test [BFRT]: Rossion & Michel, 2018, adapted from the original test of Benton & Van Allen, 1968). However, whether these behavioral tests possess the virtues listed above for adequately measuring FI, especially in terms of high validity, is questionable.…”

“…As for the BFRT, it has long been known that brain‐damaged patients with substantial impairment for language comprehension show a notably high frequency of defect at this test (Hamsher, Levin, & Benton, 1979). As a consequence of this uncertain validity, it is unclear whether the large inter‐individual variability observed with these tests (e.g., in the CFMT: Bowles et al., 2009; Palermo et al., 2017; CFMT and GFMT: McCaffery, Robertson, Young, & Burton, 2018; BFRT: Rossion & Michel, 2018) or with other explicit behavioral measures of FI (e.g., Bindemann, Avetisyan, & Rakow, 2012; Estudillo & Bindemann, 2014; Stacchi, Huguenin‐Elie, Caldara, & Ramon, 2020; Wilhelm et al, 2010) in the normal population reflects genuine variability at the FI function or variability of these general factors. As performance is only modestly correlated across different behavioral tests in the normal adult population (e.g., McCaffery et al., 2018; Stacchi et al., 2020; Wilhelm et al, 2010), the weight of general factors involved in these various tasks could be predominant.…”

“…Additionally, the absence of time constraints in these tests could also favor unnatural, analytical processing of the face images or other deviant task strategies (e.g., focusing exclusively on the mouth or hairline). Sensitivity is also limited in behavioral tests that do not consider response time, which certainly carry relevant information about FI ability (Rossion & Michel, 2018; Wilhelm et al., 2010; but see Meyer, Schmitz, Wilhelm, & Hildebrandt, 2019). However, accuracy and response time are difficult to combine or interpret together (Bruyer & Brisbaert, 2011; Liesefled & Janczyk, 2019).…”

Guess who? Facial identity discrimination training improves face memory in typically developing children.

“…Because accuracy can obscure more fine-grained differences in performance in face perception tasks (e.g. Rossion & Michel, 2018), we also considered reaction times in the matching tasks in a separate ANCOVA (controlling for age). For each participant, average response latency was calculated for correct trials, excluding those that differed from the participant's mean response latency by more than three standard deviations, or were less than 150 ms in duration.…”

Twenty years of investigation with the case of prosopagnosia PS to understand human face identity recognition. Part I: Function

“…Even when there is no explicit encoding procedure and requirement to hold an individual face in memory in the absence of the stimulus, PS is impaired at individuating faces, for instance at the Benton Face Recognition Test (BFRT, Benton & van Allen, 1968), requiring matching pictures of unfamiliar faces across head rotation and lighting direction changes. However, she performs well above chance level at these tests, and her deficit appears only clearly when considering her extremely prolonged response times (RTs), for instance at the electronic version of the BFRT (BFRT-c; Rossion & Michel, 2018; see also Busigny & Rossion, 2010;and Fysh & Ramon, 2022 for a recent discussion of this issue). Relative to typical participants, PS's impairment in accuracy rates and RTs at matching/discriminating pictures of unfamiliar faces for their identity has been documented in tens of experiments, as shown for example in section 4 below.…”