COVID Act Now (CAN)
- Source name:
covid-act-now - Earliest issue available: 2021-03-25
- Number of data revisions since 19 May 2020: 0
- Date of last change: Never
- Available for: county, hrr, msa, state, hhs, nation (see geography coding docs)
- Time type: day (see date format docs)
- License: CC BY-NC
The COVID Act Now (CAN) data source provides COVID-19 testing statistics, such as positivity rates and total tests performed. The county-level positivity rates and test totals are pulled directly from CAN. While CAN provides this data potentially from multiple sources, we only use data sourced from the CDC’s COVID-19 Integrated County View.
| Signal | Description |
|---|---|
pcr_specimen_positivity_rate |
Proportion of PCR specimens tested that have a positive result |
pcr_specimen_total_tests |
Total number of PCR specimens tested |
Table of contents
Estimation
The quantities received from CAN / CDC are the county-level positivity rate and total tests, which are based on the counts of PCR specimens tested. In particular, they are also already smoothed with a 7-day-average.
For a fixed location \(i\) and time \(t\), let \(Y_{it}\) denote the number of PCR specimens tested that have a positive result. Let \(N_{it}\) denote the total number of PCR specimens tested. Let \(p_{it}\) be the PCR-specimen positivity rate as a binomial proportion.
At the county-level, \(p_{it}\) and \(N_{it}\) are taken directly from CAN / CDC without modification. For the sake of aggregating to other geographical levels, we estimate each county’s \(Y_{it}\) as
\[\hat{Y}_{it} = N_{it} p_{it}\]Let \(G\) be a set of counties with associated population weights \(w_i\) for each county \(i \in G\). At other geographical levels, the total number of tests \(N_{Gt}\) and our estimate of the positivity rate \(\hat{p}_{Gt}\) is:
\[\begin{aligned} N_{Gt} &= \sum_{i \in G} w_i N_{it} \\ \hat{p}_{Gt} &= \frac{ \sum_{i \in G} w_i \hat{Y}_{it} }{ N_{Gt} } \end{aligned}\]Note that in order to more closely mirror the CAN data source, no pseudo-observations (such as we would typically apply for a Jeffreys prior) are used in the estimation of these proportions. Instead of using the Jeffreys prior to more accurately estimate the standard error when the value is zero, we suppress all entries with zero sample size (total tests performed) and leave those with 0 positive results as-is.
Standard Error
The estimated standard errors are estimated by:
\[\begin{aligned} \hat{se}(p_{it}) &= \sqrt{\frac{p_{it} (1 - p_{it})}{N_{it}}} \\ \hat{se}(\hat{p}_{Gt}) &= \sqrt{\frac{\hat{p}_{Gt} (1 - \hat{p}_{Gt})}{N_{Gt}}} \end{aligned}\]Smoothing
No additional smoothing is done to avoid double-smoothing, since the data pulled from CAN / CDC is already smoothed with a 7-day-average.
Limitations
Estimates for geographical levels beyond counties may be inaccurate due to how aggregations are done on smoothed values instead of the raw values. Ideally we would aggregate raw values then smooth, but the raw values are not accessible in this case.
The positivity rate here should not be interpreted as the population positivity rate as the testing performed are typically not randomly sampled, especially for early data with lower testing volumes.
A few counties, most notably in California, are also not covered by this data source.
Entries with zero total tests performed are also suppressed, even if it was actually the case that no tests were performed for the day.
Lag and Backfill
The lag for these signals varies depending on the reporting patterns of individual counties. Most counties have their latest data report with a lag of 2 days, while others can take 9 days or more in the case of California counties.
These signals are also backfilled as backlogged test results could get assigned to older 7-day timeframes. Most recent test positivity rates do not change substantially with backfill (having a median delta of close to 0). However, most recent total tests performed is expected to increase in later data revisions (having a median increase of 7%). Values more than 5 days in the past are expected to remain fairly static (with total tests performed having a median increase of 1% of less), as most major revisions have already occurred.
Source and Licensing
County-level testing data is scraped by CAN from the CDC’s COVID-19 Integrated County View, and made available through CAN’s API.