Message-passing tests: what they actually measure (and don't)

A message-passing test goes like this: a nonspeaking person is shown information their communication partner does not have — a picture, an object, a short phrase. The partner is then asked to support the person while they spell or type what they saw. If the answer is correct, the reasoning goes, the message must have come from the nonspeaker. If it is wrong, the message must have come from the partner.

It sounds airtight. It is not.

What the test actually measures

Message-passing tests measure performance under conditions of high anxiety, unfamiliar environments, unfamiliar partners, novel motor demands, and the explicit knowledge that the person is being tested to determine whether they are real. For someone with whole-body apraxia and chronic sensory dysregulation, those are exactly the conditions under which motor planning collapses. We have decades of evidence that apraxic performance is brittle in unfamiliar contexts — that fact is not in dispute in any other corner of the motor-planning literature.

So when a message-passing test fails, we know one of two things happened. Either the person could not author the message, or the person could not execute the motor sequence to spell it under those specific conditions. The test cannot tell the difference. It was not designed to.

The IQ-test parallel

This is the same category of error that produced a century of bad inferences from IQ tests applied to populations they were not designed for. An IQ test administered in a second language, under time pressure, by a stranger, in a strange room, to a person with a motor impairment, does not measure intelligence. It measures the person's ability to perform on that test.

Message-passing has the same shape. It is a single high-stakes performance task, scored binary, with no error analysis, no sensitivity to motor variability, and no comparison condition against the person's own baseline. No serious psychometrician would accept that design for any other claim.

What a fair test would look like

• Repeated trials over weeks, not a single sitting.
• The person's regular partner, in their regular environment, with their regular regulation supports in place.
• Trial-by-trial analysis of motor execution (not just final answer correctness).
• Within-subject controls — comparing the same person's performance on known versus unknown content, not comparing them to a typing-fluent norm.
• Pre-registered protocols co-designed with nonspeaking adults who actually use the methods.

None of that is exotic. It is just basic methodology. The field has not done it yet, and the absence of that work is being cited as evidence that the people in question cannot communicate. That is not how science works.