Beginning in January 2021, the U.S. government prioritized ensuring continuity of learning for all students during the COVID-19 pandemic (1). To estimate the extent of COVID-19–associated school disruptions, CDC and the Johns Hopkins University Applied Physics Laboratory used a Hidden Markov Model (HMM) (2) statistical approach to estimate the most likely actual learning modality based on patterns observed in past data, accounting for conflicting or missing information and systematic Internet searches (3) for COVID-19–related school closures. This information was used to assess how many U.S. schools were open, and in which learning modalities, during August 1–September 17, 2021. Learning modalities included 1) full in-person learning, 2) a hybrid of in-person and remote learning, and 3) full remote learning.
Multiple data sources were combined to estimate the learning modality for public and public charter school districts in the U.S. using HMM; sources included Burbio, MCH Strategic Data, American Enterprise Institute–Return to Learn, and state dashboards. Weekly learning modalities (full in-person, hybrid, and full remote) during August 1, 2020–July 31, 2021 were used to select the optimal weights for each reported modality in order to infer the most likely actual learning modality. The trained HMM was applied weekly during August 1–September 17, 2021. In addition to using HMM, since February 2020, CDC has also tracked district and individual public and private school closures attributed to COVID-19 and estimated the number of students and teachers affected by these closures. School closure data were obtained via daily systematic Internet searches, as described previously (3), which identified publicly announced COVID-19–related closures lasting ≥1 day. School or district closure was defined as a transition from being open to being closed for in-person instruction. Fully in-person and hybrid (i.e., latter includes both in-person and remote) learning modalities were classified as open; fully remote learning modalities (if stated as offered during closure) were classified as closed. Closure dates and reasons were recorded and linked to publicly available education data. HMM was fitted using the Pomegranate module (version 0.14.3) for Python (version 3.7.6). COVID-SC data were imported into SAS (version 9.4; SAS Institute) for Analysis. These activities were reviewed by CDC and were conducted consistent with applicable federal law and CDC policy.
Suggested citation for this article: Parks SE, Zviedrite N, Budzyn SE, et al. COVID-19–Related School Closures and Learning Modality Changes — U.S. August 1–September 17, 2021. MMWR Morb Mortal Wkly Rep. ePub: 24 September 2021.