With the rise in COVID-19 cases, some states are rolling back openings of businesses. When combined with PPP programs running out next month, this could lead to another spike in unemployment claims. Since early April, AGS has released a weekly dataset on unemployment estimates down to the block group, free of charge.
Over the last few months, we have learned a great deal about estimating the number of unemployment claims. While we do not pretend to have “on the ground” information to support these estimates, our initial tests on the data to date suggest that our methodology is a reasonable one – we are focusing on the distribution of employment by occupation and using a series of estimates of vulnerability curves to simulate what is being reported at a national and state level. Our first methodology relied heavily on our own Estimates and Projections from the 19B release, along with the weekly release of data from the Bureau of Labor Statistics.
In late April, just a few weeks in, we changed the methodology to implement more controls, given the drastic differences at the state level in terms of new unemployment claims as a percentage of the state labor force. We added a second layer of vulnerability to account for losses by industry by developing a “sector vulnerability” score for each industry, then creating an overall score for each block group. At this point, we were seeing significant impacts on food service, hospitality, but also all workers who live in areas whose economies are highly dependent upon tourism.
In May, we once again adjusted the methodology. In order to maintain relative comparability with monthly state unemployment estimates produced by the BLS, we have adjusted model to attempt to account for both exits from the labor force and re-entrants.
When we saw states reopening in June, we knew that the methodology needed to change again, as the unemployment rate declined from 14.7% in mid-April to 13.3% in mid-May. What is particularly difficult from a methodology viewpoint is that the state level monthly estimates lag nearly a full month behind the state level UI claims and total covered unemployed, and that those weekly numbers are subject to considerable inter-state variation. For example, Georgia has a labor force of about 5 million people, of which nearly 2.5 million had requested UI benefits. Despite that, the monthly unemployment figure for Georgia for mid-April did not show the massive increase in unemployment that would have been expected. As a result, we made adjustments using a simple relationship at the state level by using a multiplier that relates the change in the “insured unemployed” state figure over time to the total number of new claims.
We remain committed to producing these estimates for as long as the demand continues. We hope that these maps and datasets can be found useful by researchers analyzing the economic impacts of COVID-19, or as a supplement to our semi-annual demographic releases in this rapidly evolving crisis. These datasets may be freely downloaded for research use, but may not be resold without written permission from AGS. You can find the data here.
Published works utilizing this dataset should identify the source as: Applied Geographic Solutions, Inc., Thousand Oaks, California and reference the weekly release as indicated by the last date field in the file.