Intellectual disability
The population benchmark used by the 2019 review was about 50%; newer estimates vary by age, place and ascertainment.
01 — A cross-section of the field
Researchers reviewed 301 studies and found a remarkably narrow participant pool.
The review covered original research published during 2016 in high-impact autism-specific journals. Across the papers, 100,245 participants were identified as autistic. Only 165 studies reported enough intellectual-ability information for the main meta-analysis.
301
studies examinedInterventions, biology, epidemiology, psychology, social research and diagnostic-scale studies—all sampled at one point in time.
Minimally verbal or non-speaking
Population estimates commonly cited at the time were 25–30%; only 50 studies reported language status.
The bars reproduce the benchmarks and random-effects estimates used in Russell et al. (2019). They describe a 2016 literature cross-section, not every autism study before or since.
02 — Four ways the gap compounds
Underrepresentation was common, underreported and then generalized away.
“Missing” has two meanings here. Some people were excluded or not recruited. In many other papers, the data needed to know whether they were represented were simply not reported.
03 — How a narrow sample becomes a broad claim
No single paper has to be dishonest for the evidence base to drift.
Those requirements may be scientifically convenient but can make participation inaccessible.
If ID and language status are absent from the title or abstract, the boundary becomes hard to see.
Later papers inherit the broad label without inheriting the sample limitation.
The consequence is not merely a less diverse sample. If intellectual disability or language level is related to a treatment response, brain measure or developmental pathway, the estimate for the included group may not transfer to the excluded one.
04 — Inclusion is a research method
The fix is not to pretend every study can include everyone.
Some questions are intentionally about a specific subgroup. The problem is claiming population-wide knowledge from subgroup evidence without making the boundary visible. The review’s recommendations are practical.
Work with participants, families and service providers; allow home visits and flexible scheduling.
Pilot non-speaking response options, sensory accommodations and procedures that do not require unnecessary verbal ability.
Put ID and language characteristics—and their limitations—in abstracts, titles and interpretation.
The sharpest insight is not “half of autism is missing.” Population proportions change across datasets. The durable finding is the size of the mismatch: the people with some of the highest support needs were represented at only 6% and 2% in the studies that reported enough data.