Citation
Whitacre, Travis (2023). Hidden in Plain Sight: Prevalence and Impact of ADHD Underdiagnosis. Southern Economic Association 2023 Annual Conference. New Orleans, LA.Abstract
While diagnostic errors are prevalent throughout health and mental health fields, little is known about their long-term consequences. Here, I assess the long-term education and labor market effects of undiagnosed ADHD during childhood. Doing so is challenging for two reasons. First, a child's true ADHD status is unobserved; only their diagnosed status is known. Second, even if diagnostic errors are observed, they are likely akin to non-classical misclassification errors and therefore endogenous. To overcome these challenges, I provide a theoretical model, extending on Marquardt (2020) and Chan Jr et. al (2019), of diagnosis to understand the sources of diagnostic error. The theory maps clear causal mechanisms in which genetic characteristics passed on from the parent affects true ADHD status, ADHD status and demographic characteristics affect the probability of diagnosis, and all three of these (ADHD, diagnosis, and demographic characteristics) determine adult labor market and education outcomes.The model is then estimable by extending the partial observability model developed in Nguimkeu et al. (2019) through a bivariate probit model. Using data from ADD Health, I am able to predict true ADHD status and the probability of a false negative diagnosis for a national sample of adolescents using detailed data on genetic, hereditary, demographic, and socioeconomic characteristics. The partial observability model uses the fact that, under the assumption of no false positives, an observed diagnosis can only happen if a child truly has ADHD and the child is diagnosed with ADHD correctly. Thus, using data only on observed diagnosis, the model is able to recover determinants of both true ADHD status and receiving a correct diagnosis. This results in predicted probabilities of ADHD, false negative diagnosis, and a conditional false negative diagnosis for every individual within the sample.
The analysis has striking results. First, I find that ADHD is underdiagnosed, in which there is a 0.78 conditional probability of a false negative, or going undiagnosed given true predicted ADHD status. Second, large diagnostic gaps exist based on sex, race, ethnicity, and parental socioeconomic status. The racial gaps cannot be fully explained by socioeconomic background. Third, I find that both ADHD and a false negative diagnosis are highly detrimental for human capital development, and that large improvements can be made with correct diagnosis. I conduct an empirical monte carlo to test for the robustness of these results for varying rates of false positives.
These results provide evidence of viscous intergenerational cycles in which inequity in mental health care is driven by and reinforces socioeconomic inequities. To further explore these results, I consider which types of school and neighborhood characteristics might drive higher school level false negative rates. With merged common core data, I use bivariate OLS and LASSO methods used in Finkelstein et. al (2016) and find that it is primarily economic neighborhood characteristics as well as racial and ethnic demographics which drive higher school level false negative rates. A targeted policy recommendation from this paper would include improving health insurance access.