The complexities of combining federal autism data reveal new challenges in state reporting

Building a comprehensive autism report across US states highlights significant issues in aligning federal data sources, emphasising the importance of precise measurement and categorisation methods.

The real challenge in building a public state-by-state autism report was not combining the numbers, but knowing which numbers should not be combined. The project drew on three federal datasets: the Education Department’s IDEA Section 618 counts for children ages 5 to 21 receiving special education services, Census population estimates for the same age band and the CDC’s Autism and Developmental Disabilities Monitoring Network, which tracks autism in defined communities rather than whole states. According to the Education Department and the CDC, those sources are designed for different purposes, even though they can appear to fit neatly together at first glance.

For the identification rate, the calculation itself was simple: autism counts from IDEA divided by the Census population ages 5 to 21, then multiplied by 1,000. But the important data rules were more exacting. Iowa and New Mexico do not report special education by autism category, so their rows had to be marked not applicable rather than treated as zero. Using zero would have created a false ranking. The age band also had to match exactly; a broader child population would have distorted every result in the table.

The CDC’s ADDM figures were the join the project deliberately refused. ADDM is a surveillance system built around specific communities, often counties or metropolitan areas, not states. The agency says it is intended to measure prevalence, identification practices and demographic patterns in those communities, and its latest national estimate is about 1 in 31 children aged 8, or 3.2%. That makes it useful context, but not a clean fourth column beside a state-wide identification rate. The project therefore kept ADDM in its own field, labelled clearly as a site figure and left out of default ranking. The broader lesson is that two datasets sharing a state code are not necessarily joinable: the population, geography and measurement method all have to align first.