COVIDcast Signal Changelog
When Delphi makes substantial changes to a signal’s computation, we will typically publish it under a new name rather than revising the existing signal. By contrast, the changes documented here are bug fixes, pipeline repairs, and minor enhancements.
Changes are also announced via the COVIDcast API mailing list, and we strongly recommend that anyone using the API to download data regularly should subscribe to this list. We will only use the list to announce API changes, data corrections, and other information relevant to API users.
For current information on how up-to-date our public health and behavior data is, how much of the country is covered by our public health and behavior data, and the data versioning effects present in our public health and behavior data, please see Delphi’s Indicator Status Overview page.
Table of contents
- COVIDcast Signal Changelog
Sources and Signals
doctor-visits
9 November 2020
We went from a custom geo-mapping file (for aggregating from county->(msa, hrr, state)) to a central geo file based on rigorously sourced US census data.
fb-survey
3 June 2020
Duplicate survey weights had corrupted historical figures for the following signals and dates. The correct data has been restored to the API.
raw_wclicounty: 20200406, 20200408, 20200410, 20200430hrr: 20200406, 20200408-20200410, 20200430msa: 20200408, 20200410, 20200430state: 20200408-20200410, 20200430
smoothed_wclicounty: 20200406, 20200408-20200414, 20200430-20200506hrr: 20200406-20200415, 20200430-20200506msa: 20200408-20200414, 20200430-20200506state: 20200408-20200416, 20200430-20200506
20 November 2020
Due to a bug in our data processing system, estimates of the percentage of people reporting that they have been tested for COVID-19 calculated before October 8th were incorrect.
We incorrectly treated an answer of “no” as a missing response, which affected the smoothed_tested_14d and smoothed_wtested_14d signals from the fb-survey source.
As of Nov. 20th, the error has been corrected and all affected data reissued.
hospital-admissions
20 October 2020
We went from a custom geo mapping file (for aggregating from zip->(county, msa, hrr, state)) to a central geo file based on rigorously sourced US census data.
jhu-csse
3 June 2020
We now include figures on Puerto Rico for all jhu-csse signals at the state level.
1 September 2020
NY Boroughs county FIPS (36005, 36047, 36061, 36081, 36085) are now split in proportion to the population of each county, instead of being reported in aggregate in FIPS 36061.
7 October 2020
The following changes were made to all jhu-csse signals related to geocoding:
- NY Boroughs county FIPS (36005, 36047, 36061, 36081, 36085) are now differentiated by JHU,
- “Unassigned” and “Out of State” counts are assigned to a megaFIPS at the county level (XX000 where XX is the state FIPS code) and will now be incorporated into the state totals
- The split of the Kansas City, Missouri into its four sub-counties is now done with a more precise population proportion,
- The split of the Dukes and Nantucket, Massachusetts FIPS codes into the two sub-counties is now done with a more precise population proportion,
- Some counts are reported in the Utah territories JHU UIDs 84070015-84070020. These are now aggregated properly into the state megaFIPS 49000, which will be aggregated in state-level information.
indicator-combination
3 June 2020
Standard errors are now included in the nmf_day_doc_fbc_fbs_ght signal for all geo levels and dates, representing the estimated uncertainty in this signal. This uncertainty comes because the signal is a combination of other signals based on survey estimates or other estimates with margins of error.
nmf_day_doc_fbc_fbs_ght- all geo levels
- all dates
- all geo levels
12 October 2020
The 10 October 2020 issue of all indicator-combination deaths signals has been removed from the API. These signals are primarily constructed of USAFacts data, whose 10 October 2020 issue was discovered to be corrupt on 11 October and repaired on 12 October. Subsequent issues have adequate coverage of all regions and dates included in the 10 October issue, so this change only affects forecasters who intend to pull training data with an as_of or issues parameter set to 20201010.
quidel
22 October 2020
We went from a custom geo mapping file (for aggregating from zip->(county, msa, hrr, state)) to a central geo file based on rigorously sourced US census data.
safegraph
19 April 2021
The Safegraph social distancing metrics are no longer being updated. Weekly patterns are still available.
3 November 2020
We went from a custom geo mapping file (for aggregating from county->state) to a central geo file based on rigorously sourced US census data.
usa-facts
3 November 2020
We went from a custom geo mapping file (for aggregating from county->(msa, hrr, state)) to a central geo file based on rigorously sourced US census data.
12 October 2020
The 10 October 2020 issue of all usa-facts deaths signals has been removed from the API. The file for deaths provided by USAFacts on 10 October included case data instead. The resulting spurious 100× increase in magnitude of COVIDcast usa-facts deaths signals was noticed on 11 October and repaired on 12 October. Subsequent issues have adequate coverage of all regions and dates included in the 10 October issue, so this change only affects forecasters that intend to pull training data with an as_of or issues parameter set to 20201010.