Context

COVIDcast displays signals related to COVID-19 activity levels across the United States, derived from a variety of anonymized, aggregated data sources made available by multiple partners.

One of COVIDcast streams displays results for a CMU-run symptom survey, advertised through Facebook.

Content

This dataset is gathered using the delphi-epidata API and contains covidcast_meta and covidcast datasources.

Presently the dataset contains fb-survey data signal which is based on CMU-run symptom surveys, advertised through Facebook. Using this survey data, CMU estimate the percentage of people in a given location, on a given day that have CLI (covid-like illness = fever, along with cough, or shortness of breath, or difficulty breathing), and separately, that have ILI (influenza-like illness = fever, along with cough or sore throat).

Dataset Description

Files are organized in folders based on the spatial resolution of fb-survey data (state, county, hrr, msa).

Each file contains the percentage of people in a given location, on a given day that have CLI or ILI. Data consists of raw and smoothed estimates and is gathered for all time values available at delphi-epidata.

Each file contains the following columns: - geo_value - location code - time_value - time unit (e.g. date) over which underlying events happened - direction - trend classifier (+1 -> increasing, 0 steady or not determined, -1 -> decreasing) - value - value (statistic) derived from the underlying data source - stderr - standard error of the statistic with respect to its sampling distribution, null when not applicable - sample_size - number of "data points" used in computing the statistic, null when not applicable

Additionally, the dataset contains the most recent covidcast_meta where you can find the summary statistics for fb-survey data.

This dataset contains large files which can be used to reproduce the results in McDonald, D.J., Bien, J., Green, A., Hu, A.J., DeFries, N., Hyun, S., Oliveira, N.L., Sharpnack, J., Tang, J., Tibshirani, R., Ventura, V., Wasserman, L., and Tibshirani, R.J. “Can Auxiliary Indicators Improve COVID-19 Forecasting and Hotspot Prediction?,” Proceedings of the National Academy of Sciences, 2021. https://doi.org/10.1101/2021.06.22.21259346 Short-term forecasts of traditional streams from public health reporting (such as cases, hospitalizations, and deaths) are a key input to public health decision-making during a pandemic. Since early 2020, our research group has worked with data partners to collect, curate, and make publicly available numerous real-time COVID-19 indicators, providing multiple views of pandemic activity in the U.S. This paper studies the utility of five such indicators---derived from de-identified medical insurance claims, self-reported symptoms from online surveys, and COVID-related Google search activity---from a forecasting perspective. For each indicator, we ask whether its inclusion in an autoregressive (AR) model leads to improved predictive accuracy relative to the same model excluding it. Such an AR model, without external features, is already competitive with many top COVID-19 forecasting models in use today. Our analysis reveals that (a) inclusion of each of these five indicators improves on the overall predictive accuracy of the AR model; (b) predictive gains are in general most pronounced during times in which COVID cases are trending in "flat" or "down" directions; (c) one indicator, based on Google searches, seems to be particularly helpful during "up" trends. Complete descriptions as well as code are available from https://github.com/cmu-delphi/covidcast-pnas/ and are permanently accessible at https://doi.org/10.5281/zenodo.5639567. This material is based on work supported by gifts from Facebook, Google.org, the McCune Foundation, and Optum.