A crowdsourcing approach to identifying community behavior during the COVID Pandemic allowed us to learn useful information about actual practices in a real-time basis and the subsequent analysis confirmed that the data gathered had a real-time relationship with COVID infection outcomes which validates the usefulness of the data and the collection methods. This presentation will focus upon the methods employed in the data collection and analysis phases of this project.

Computational forecasts of COVID-19 infections may benefit from temporal signals associated with human behavior. Non-pharmaceutical interventions (NPI) have been shown to reduce the spread of an infectious agent, but accurate information about how the general public interprets and acts upon guidelines developed by public health officials is rarely collected. For 36 weeks from September, 2020 to April 2021, we asked two crowds twenty-one questions about their perceptions of their community’s adherence to NPI and public health guidelines and collected over 10,000 responses.

We produced weekly highlights of the survey responses about NPI adherence in the U.S. The responses are separated by state and compared to state level statistics regarding the level of COVID-19 infections from the Johns Hopkins COVID database for the same time period. At the end of the 8 months, we produced a summary report covering the entire period.

After the data collection phase was completed, further analysis was performed on the data. Crowd- sourced NPI signals were mapped to a mean perception of adherence—or MEPA—and included in computational forecasts. Several MEPA signals linearly correlated with one through four week ahead incident cases of COVID-19 at the US national level. Including questions related to masking, testing, and limiting large gatherings increased out of sample predictive performance for 1-3 week ahead probabilistic forecasts of incident cases of COVID-19 when compared to model that was trained on only past incident cases. In addition, we found that MEPA signals could be clustered which suggests a more focused survey may have sufficed and provided similar performance.

We concluded that crowdsourced perceptions of non-pharmaceutical adherence may be an important signal to improve forecasts of the trajectory of an infectious agent and increase public health situational awareness.

Find the Q&A here: Q&A on 'Data Driven ERM'