Figure 4 separates the total hours reductions documented in Figures 1 - 3 into three channels: shutdowns, layoffs, and cuts in hours. We define firms as having fully shut down in a given week if the Homebase data records zero employees clocking in at that firm during that week. We identify a worker as having been laid off in a given week if that employee works zero hours at a firm which is still operating. We define hours cuts as the reduction in hours, relative to that initial baseline, among workers still employed at still operating firms. The figure distinguishes which fraction of the percent change in hours each week since early February is attributable to these three forms of hours reductions. The total number of hours worked in the last week of March are less than half what they were in late January. Most of that reduction is due to firms fully shutting down or asking retained employees to work fewer hours. A smaller percentage is due to firms laying off a portion of their workforce. This suggests that the principal driver of unemployment claims is total firm shut downs. It also suggests that even still employed workers are suffering a cutback in their hours.
One important caveat to this decomposition is what we refer to as a firm shut-down is a shut-down of Homebase measured employment. If firms employ workers that do not schedule their time using Homebase and some of these workers remain employed, some of the hours losses that we attribute to shut-downs may instead be properly attributed to layoffs. Another caveat is that the "hours cut" category includes all workers with positive hours during a given week. Firms that shut down in the middle of a week, as well as workers who are laid off mid-week, will be counted in this category in the first week, and will not appear in the correct category until the first week in which they have zero hours.
Methodology
Our analyses are based on data on hours worked at the establishment-worker-day level generously made available by Homebase. These data extend from January 1, 2020 through March 28, 2020. We aggregate the Homebase data to the firm-MSA-industry-day level. We restrict the sample to firms whose employees worked at least 80 hours between January 19 and February 1 and to states for which we observe at least 50 such firms. We refer to this two-week window as the “base period. ” All analyses weight firms by their total hours during the base period.
In our analyses of weekly outcomes (e.g., Figures 1 and 4), we normalize each firm’s hours by dividing by the average hours worked per week over the base period at the firm. In our analyses of daily outcomes (e.g. Figures 2 and 3), we normalize by dividing by the average value of the outcome at the given firm on the same day of the week during our base period. For example, if total hours for a firm on Friday, March 13 was 100 and total hours for the same firm on Friday, January 24 and Friday, January 31 was 300, (150 on each day), the outcome variable total hours’ value would be .66. (This is 100 divided by (300/2), the average Friday hours in the base period.)
We use the data compiled by The New York Times on the timing of stay-at-home and shelter-in-place orders in different states.
[1] It’s important to emphasize that reducing activity and interactions between people is the goal of stay-at-home and shelter-in-place orders, so the earlier reduction in employment, particularly among businesses involving significant closer-contact such as Beauty and Personal Care and Leisure and entertainment, can be interpreted as the laws working as intended. Other work, for example from Villas-Boas, et al (2020) suggests that, by reducing activity and mobility, stay-at-home and shelter-in-place orders may reduce hospitalizations and death due to COVID-19.
Check back to Rustandy's Coronavirus Social Impact Research page for the latest results. Read the press announcement.
Authors
Alexander W. Bartik, Assistant Professor Economics, University of Illinois at Urbana-Champaign, and Research Affiliate, UChicago’s Poverty Lab; Marianne Bertrand, Chris P. Dialynas Distinguished Service Professor of Economics, University of Chicago Booth School of Business, and Faculty Director, Chicago Booth's Rustandy Center for Social Sector Innovation and UChicago’s Poverty Lab; Feng Lin, Research Professional, Chicago Booth; Jesse Rothstein, Professor of Public Policy and Economics, University of California, Berkeley, and Director, Institute for Research on Labor and Employment (IRLE) and California Policy Lab; and Matt Unrath, PhD Candidate, Goldman School of Public Policy, UC Berkeley, and Research Fellow, California Policy Lab
Acknowledgements
We thank Homebase and Ray Sandza in particular for generously allowing access to their data and sharing their time to answer questions and help us understand the data. We also thank Jingwei Maggie Li, Salma Nassar, and Greg Saldutte at Booth's Rustandy Center for Social Sector Innovation and Manal Saleh at the Poverty Lab for excellent assistance on this project and Michael Stepner for comments.