NimStim facial expression database – neurotypical bias
A few comments on the 2008 NimStim database (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3474329/pdf/nihms408635.pdf), a large and racially diverse reference database for face- and emotion-recognition tasks
Facial expressions for ‘happy’, ‘sad’, ‘angry’, ‘fearful’, ‘surprised’, ‘disgusted’, ‘neutral’ and ‘calm’ are represented, with open- and closed-mouth variants for all but ‘surprised’, and an additional ‘exuberant’ variant for ‘happy’
The face stimuli themselves are posed photographs of 42* professional actors from New York City, 18 female and 25 male, aged 21–30
Twenty-five of those actors were European-American, 10 African-American, 6 Asian-American and two Latinx-American – but there was a deliberate decision to focus on the US (and specifically New York) to minimise effects of inter-cultural variation
(I was somewhat surprised to see a paper from just 15 years ago uncritically using the racially motivated terms ‘Caucasian’ and ‘sub-Saharan Africa’, even if these were in reference to earlier papers that may have used these terms)
*originally 43, but one actor didn’t meet validation thresholds
[1/2]
NimStim facial expression database – neurotypical bias
The NimStim database was validated by 81 participants, 51 female and 30 male, aged 18–35, with over 70% being European American – a mixture of 34 New Yorkers and 47 undergraduates from a Midwestern US liberal arts college
Validation used a forced-choice method: the participants had to choose one of the eight facial expression terms (‘happy’, ‘sad’, ‘angry’, ‘fearful’, ‘surprised’, ‘disgusted’, ‘neutral’ and ‘calm’) that the actors had been given, or ‘none of the above’, for each face in the database
I’ll skip over the statistical analysis – mainly because it would take me too long to understand it! – but I see a couple of potentially significant problems
First, as noted in the paper, these are posed photographs
People are pretty bad at judging emotions from context-free still images (see e.g. https://doi.org/10.1038/s41467-021-25352-6)
And a professional actor portraying a static emotion is likely to produce a relatively unnatural performance
Second, there’s the double empathy problem (https://kar.kent.ac.uk/62639/): specifically, why should Autistic people be expected to read emotions from facial expressions validated by neurotypical people?
Why do I say they were neurotypical?
I can’t be 100% sure, but it’s the word *healthy*: the validators are described as ‘untrained, healthy adult research participants’
Am I wrong to assume that ‘healthy’ is code for neurotypical here?
[2/2]