Mis-Detection Controlled Estimation: Explaining Variations in the Observed Incidence of Autism

Abstract

We estimate the prevalence of true and falsely-observed autism using a maximum likelihood technique that accounts for the presence of both false positive and false negative realizations of the disorder's underlying incidence. This approach, which we label mis-detection corrected estimation (MCE), is the maximum likelihood equivalent of a two-part model where both the first and second stages are binary outcomes. Using the 2007 and 2011 National Survey of Children's Health, we estimate the impact of factors that may be driving variation in autism diagnoses, including individual-level demographics and state-level policies such as education quality, health care access, and mental health parity. Results show that age, access to hospitals, unemployment, and poverty increase the likelihood that autism is detected, while black and Hispanic children are less likely to be detected. The only policy measure that improves detection is autism-specific mental health parity legislation. We also estimate the rate of false positives across selected variables and find that girls, blacks, Hispanics, and children in poverty are more likely to be mis-detected.