Ronald Kessler Seminar, March 4, 2026

Ronald Kessler, a prominent psychiatric epidemiologist, discusses the challenges and potential solutions for using online panels in epidemiologic research. He begins by highlighting the historical context of survey sampling issues, referencing the 1948 Dewey-Truman election prediction failure. He contrasts high-quality probability surveys like the National Election Study with modern low-response-rate surveys such as the Household Pulse Survey during COVID-19, which produced inaccurate small-area estimates. Kessler emphasizes the inadequacy of current government surveys for mental health, noting that most rely on superficial questions and lack validation. He mentions two high-quality national surveys (NCS-R and NESARC) but notes their data are outdated. He then introduces probability-based panels as a middle ground, but cautions that their cumulative response rates are often low (around 10%). He proposes hybrid surveys that combine a large online panel with a smaller, high-quality probability sample, using the latter as a calibration anchor. He illustrates how auxiliary data from the American Community Survey can help adjust for nonresponse bias. Throughout, he stresses the importance of accurate prevalence estimates for mental disorders and warns against relying on nonprobability surveys that can produce wildly inflated estimates (e.g., an 800% increase in anxiety during COVID).