Autistic and
non-autistic
observers
judged autistic
suspects alike —
a forensic test of
double empathy

A test the theory seldom faces

The double empathy problem is, in the words of the study discussed here, an “apparently quite widely accepted” idea. Milton’s argument, set out in 2012, is that breakdowns in understanding between autistic and non-autistic people run both ways: if non-autistic people struggle to read the minds and intentions of autistic people, the opposite is likely to apply when autistic people try to read non-autistic people. In this Directory’s reading, it moved the problem from the autistic person to the gap between people.

On 21 September, Law and Human Behavior published a study whose title announces a result against it: Evidence for Expectancy Violations Theory but Not the Double Empathy Hypothesis. Neil Brewer, Carmen Lucas, Katie Logos, Robyn Young and Alliyza Lim, at Flinders and Adelaide universities, set out to run what they call a test the hypothesis has “seldom been subjected to”. This Directory has leaned on double empathy before. That is the reason to read this carefully, at its own strength.

What the experiments did

The suspects came from an earlier study in which autistic and non-autistic adults took part in a simulated online chatroom about illegal hacking and were then interviewed about it, on video. From those recordings the team picked interviewees who showed relatively high or low levels of one of two behaviours: difficulty interpreting figurative language (Experiment 1) or speech hesitations (Experiment 2). They deliberately chose suspects to minimise group differences in four other behaviours, such as gaze and facial expression, and included men and women in each group.

The observers were recruited online through Prolific. Experiment 1 had 212 autistic and 215 non-autistic observers; Experiment 2 had 125 of each. Autistic observers had to report a formal diagnosis — those who self-identified or were seeking a diagnosis were excluded. Each observer watched interviews with two male and two female suspects from one diagnostic group, and rated them for suspiciousness, credibility and likely guilt, and for how far their behaviour violated expectations.

The double empathy prediction the authors set up is specific. If non-autistic observers judge autistic suspects more harshly because they misread them, autistic observers should show the reverse — judging the autistic suspects more favourably than the non-autistic ones. Statistically, that is an interaction between who is watching and who is being watched.

What they found

That interaction did not appear. In both experiments it was non-significant and very small, and Bayes factors gave moderate evidence for the null. In the authors’ summary, for judgements of suspiciousness, credibility and likely guilt, “it did not matter whether suspects and observers shared the same diagnostic status.” Autistic and non-autistic observers did not differ reliably in how they rated autistic suspects — apart from a weak, gender-linked interaction in Experiment 2 with only anecdotal Bayesian support.

Other findings are less tidy than the title:

• In Experiment 1, autistic suspects were judged more negatively than non-autistic suspects overall, but the effect was weak. It was driven by men: male autistic suspects were judged more harshly than male non-autistic suspects, while female autistic suspects were judged less harshly than female non-autistic suspects.
• In Experiment 2, autistic suspects were not judged more negatively overall. There was a weak interaction: when hesitations were more pronounced, autistic suspects tended to be judged more harshly; when they were less pronounced, the pattern reversed. The Bayesian support was only anecdotal, and the authors urge caution about how reliable the effect is.
• Neither figurative language difficulties nor speech hesitations had a significant effect on their own.
• What was strongly associated with harsher judgements, whatever the suspect’s diagnosis, was the observer’s sense that the suspect’s behaviour violated expectations.

The authors’ public significance statement puts it plainly: anyone whose behaviour in a forensic interview “does not align with ‘normal’ expectations may be judged harshly.” The quotation marks around normal are theirs.

What kind of test this is

The study is careful and its null is real. It should be counted. What it counts against is worth stating precisely.

Double empathy is a claim about mutual understanding between people who are interacting. This study tested third-party observers watching short recorded interviews and rating strangers on credibility and guilt. The authors anticipate the objection that judging suspicion and credibility falls outside what double empathy covers, and argue against it: theory-of-mind measures are associated with detecting suspicious behaviour, and studies cited in support of double empathy used similar interpersonal judgements. They add that testing such explanations will require further research. That is a fair reply about the kind of judgement. It does not remove the difference between reading a stranger from the outside and understanding someone you are talking with.

The authors also list their own limits. The study used only two male and two female suspects from each group, so the gender pattern may reflect those particular people. The suspects were chosen to minimise the other behaviours that distinguish autistic and non-autistic interviewees, which the authors say may have “artificially curtailed” autism-related differences. The interviews were online and brief.

Separately, the hyperscanning review shows how rarely the missing design has been run: of its 15 studies with autistic participants, ten included a comparison group of non-autistic pairs and three included only mixed autistic–non-autistic pairs — and it is mainly the trait-based studies that begin to test matched pairs.

So the honest position is this. One strong prediction drawn from double empathy — that autistic observers will judge autistic strangers more favourably in a credibility task — has now failed a direct test, in two experiments. The broader claim, that understanding breaks down across the gap rather than inside the autistic person, has not been refuted, because the design that could refute it — autistic people interacting with autistic people, measured against mixed pairs — remains rare. Holding both is not hedging. It is what the evidence currently supports.

The finding that should worry police and courts

Set the theoretical argument aside and the study’s most practical result is the one named first in its title. Behaviour that violates what observers expect to see in an interview was strongly associated with harsher judgements of suspiciousness, credibility and guilt — for everyone, autistic or not.

Those expectations are, in this Directory’s reading, the problem, and they are not a neutral instrument. They are an average of how people are expected to present — the norm that nobody quite embodies but everyone is measured against. Separately, male autistic suspects were judged more harshly in Experiment 1. But across both experiments, being autistic was not significantly linked to observers’ sense that expectations had been violated: the paths from suspect group to expectancy violations were non-significant in both experiments. The authors’ own recommendation is that police, judges and jurors should not rely on “extraevidential factors” when assessing credibility. The deeper implication is that demeanour-based credibility assessment is itself an extraevidential factor dressed up as judgement — and that the question worth asking of any forensic interview is not who understands whom, but whose way of presenting the process has quietly decided to believe.

Citations

Brewer, N., Lucas, C. A., Logos, K. et al. (2026) — How behavioral differences between autistic and nonautistic suspects in forensic interviews affect observers’ evaluations: Evidence for expectancy violations theory but not the double empathy hypothesis — Law and Human Behavior

Sarovic, D., Sohrabpour, A., Tosato, T. et al. (2026) — A PRISMA Systematic Review of Hyperscanning in Autism — Review Journal of Autism and Developmental Disorders