KEY TAKEAWAYS
  • In mid-April, national job losses peaked at 15% of pre-pandemic employment, recovering somewhat to 13% in mid-May
  • Job losses most severely affected services with high interpersonal proximity, and goods and services whose consumption is nonessential in the sense of being discretionary or deferrable
  • Job losses in manufacturing and construction appear to be a secondary effect of pandemic-induced employment contractions in services
  • As of mid-April, state job losses ranged from 8% to above 20% of pre-pandemic employment
  • Job losses through mid-April show geographic clustering at the state level
  • Simple analyses of sector mix and pandemic intensity explain differences in job losses for some but not all states

Introduction

In this report, we survey employment data from national and state jobs reports from the US Bureau of Labor Statistics for April and May. These contain the first comprehensive data on employment losses during the coronavirus (COVID-19) pandemic to date. The goals of this report are to

Historical Background

Roughly three months after the first wave of coronavirus lockdowns in the United States, we are now beginning to gain clarity about actual job losses to date due to the pandemic. During the third week of March, nearly 3 million people filed new claims for unemployment insurance—a record high since tracking began in 1967 and more than 11 times the average weekly rate for the previous 12 months. That record lasted for just one week. More than 6 million new unemployment claims were filed in each of the following two weeks before gradually tapering to 1.6 million new claims in the last week of May, as shown in Figure 1.

Through March and April, as the pandemic rapidly spread through the United States, weekly unemployment claims were the main, and virtually the only, source of data on job losses. While total new unemployment claims provided a reasonable proxy for lost jobs during the pandemic’s early weeks, this metric is subject to two caveats. First, some workers who lost their jobs were either ineligible for unemployment insurance or were eligible but did not file a claim. Second, and importantly as the pandemic progressed, some workers who filed claims for unemployment insurance in March and April have gone back to work since then. At best, unemployment claims are a serviceable, although flawed, proxy for job losses— but the only data at hand until recently.

Since the last issue of the Quarterly Economics Briefing, the US Bureau of Labor Statistics has published monthly jobs reports containing data on labor market conditions for April and May.The BLS publishes two monthly jobs reports, a national jobs report and a state jobs report, both appearing in the month following the month to which the report applies. Thus, for example, the two April jobs reports contain labor market data pertaining to the one-month period ending as of mid-April and are published in May. We will refer to a jobs report by the latest month for which it contains data, not by the month in which it is published. As of this writing, we now have available both the national and state jobs reports for April and the national jobs report for May (the state jobs report for May is scheduled to appear later in June). These reports provide snapshots of employment levels at mid-month in April and May, as indicated on the timeline in Figure 1, and offer the first comprehensive data about job losses from the coronavirus pandemic as it has developed so far.

National Perspectives: Aggregate Job Losses and Key Sectors

From mid-February to mid-April, US private nonfarm job losses totaled 19.8 million, or 15.4% of pre-pandemic employment in mid-February. By mid-May, however, total US job losses relative to February stood at 16.2 million, a net recovery of 3.6 million jobs from mid-April to mid-May.Throughout this report, we measure job losses as the difference between private non-farm employment, seasonally unadjusted, from the jobs report for the indicated month and from the jobs report for February, the last month before extensive layoffs began in March. A postscript to this report explains why seasonally unadjusted employment data provide a more accurate depiction of employment changes over the relatively brief pandemic period than seasonally adjusted data. The employment rebound from April to May, amounting to 18% of total job losses as of mid-April, was surprising and unexpected when first published on June 5. Barring a resurgence of coronavirus infections as state economies gradually reopen, it appears that the peak for US job losses was reached and passed in April. As of mid-June, reports have appeared of coronavirus resurgence in several states, including Arizona, California, Florida, and Texas.

Table 1 summarizes national employment levels and job losses for various sectors of the economy as of mid-April and mid-May. The first three columns show seasonally unadjusted private non-farm employment levels as of mid-February (the last jobs report before pandemic-related layoffs in March), mid-April, and mid-May. The next two columns show national job losses in each sector in mid-April and mid-May as a percentage of their pre-pandemic February employment levels. The following column orders sectors by their share of total employment in February. The last two columns give the contribution of jobs lost in each sector as a share of total lost jobs as of mid-April and mid-May.

Table 1 shows clearly that job losses from the coronavirus pandemic have differed dramatically across different sectors of the economy. All sectors sustained job losses, but some experienced job losses disproportionately greater than their share in total employment. Most strongly impacted, by order of their share of total job losses in mid-April, were Accommodation and Food Services, Education and Health Services, Retail Trade, Professional and Business Services, Manufacturing, and Other Services.Arts, Entertainment, and Recreation suffered the greatest percentage job loss of all sectors in Table 1, but accounts for only a small share of total employment. With the exception of Manufacturing, all of the entries in this list are service sectors. Most striking is the Accommodations and Food Services sector, which accounted for 11% of national employment in February but more than 33% of jobs lost as of mid-April. Not all service sectors were strongly impacted, however. Wholesale Trade, Transportation and Warehousing, Information, and Real Estate services suffered relatively mild job losses; Finance and Insurance was virtually unaffected at the national level.

Job Losses in Service Sectors—Proximity and Essentiality. Comparing job losses from February to mid-April across different service sectors, there appear to be two coincident drivers: physical proximity and essentiality. Physical proximity refers to the degree of interpersonal contact associated with a service or activity, clearly a deterrent to economic activity during an outbreak of infectious disease.Please see our companion report on the degree of physical proximity across different economic sectors. Essentiality refers to the degree to which a service cannot be postponed; services that are more discretionary or capable of postponement are less essential in this terminology.Our use of the term “essentiality” relates to observed behavior and is distinct from the concept of essential services as formally defined by regulatory guidelines, executive orders, or legislative enactments. For example, certain medical services designated as essential according to state regulatory guidelines nonetheless experienced a loss of demand as patients postponed or cancelled treatment. Overall, service sectors that suffered the greatest pandemic-related job losses are those involving a relatively high degree of interpersonal proximity (Food Services; Retail Trade); those that are nonessential or capable of postponement (Accommodation, especially associated with travel; Arts, Entertainment, and Recreation); or those that have both attributes together (certain Education and Health Services, certain Professional and Business Services).

Job Losses in Manufacturing and Construction—Secondary Effects. The combination of physical proximity and essentiality explains job losses reasonably well in most service sectors. But what about manufacturing and construction, both of which contributed materially to job losses through mid-April? For these sectors, the story appears to be more about discretionary postponement of major purchases and projects than about physical proximity.

We interpret employment losses in manufacturing industries through mid-April as an instance of what economists think of as a secondary, or equilibrium, effect of the primary shock to service employment that initiated the coronavirus recession. From a causal perspective, employment losses from reduced economic activity in service sectors directly impacted by the pandemic—the primary shock—thereby reduce aggregate income and consequently demand for other goods and services—the secondary effect. Driven by the loss of income, secondary reductions in demand most strongly impact those goods and services whose consumption is discretionary, deferrable, or expensive.

The distinction between primary and secondary employment impacts—the former referring to direct demand (and supply) shocks to coronavirus-affected services and the latter referring to consequent income effects impacting the demand for goods and services more broadly—is important to understanding the channels by which the coronavirus pandemic has caused employment losses, but not to their timing. When layoffs began in earnest in mid-March, employers and consumers anticipated events and reacted quickly. As a result, primary and secondary employment losses happened more or less simultaneously and both are evident in the data for mid-April.

In Manufacturing, job losses from February to April were dominated by big losses in Transportation Equipment, mainly passenger vehicles and light trucks. Other Manufacturing subsectors, especially nondurables, were less strongly impacted.

This makes intuitive sense. The purchase of a new motor vehicle is one of the biggest expenditures most households will ever undertake, behind home ownership. For households out of work or at risk of being so, it is an easy choice to hold off buying a new car, or making other big-ticket expenditures, until circumstances improve. We see employment losses in Manufacturing, especially Transportation Equipment, as a secondary effect of the coronavirus recession, logically consequent to its primary impact cutting employment in service sectors with a high degree of physical proximity.

We interpret the decline in Construction employment similarly. More than most other sectors, job losses in Construction by state from February to April were strongly correlated with overall state job losses during the same period. This is consistent with the idea that pending construction projects were suspended or postponed in response to local economic conditions—states with higher employment losses overall also tended to cut Construction employment more severely. As with Manufacturing, we see job losses in Construction as a secondary, demand-related effect of the coronavirus recession.

The Jobs Recovery from April to May. It is too early to say whether the jobs recovery from mid-April to mid-May marks the beginning of a rapid, V-shaped end of the coronavirus recession, or simply the first, relatively easy reversals of wholesale layoffs during March and April. In almost every sector represented in Table 1, sector shares of total job losses were changed little from April to May. This indicates that nearly all sectors recovered by mid-May had the same proportion, roughly 18%, of their peak job losses as of mid-April.

An apparent exception is Construction, where May employment is almost back to its February level. However, our use of seasonally unadjusted employment data is misleading in this instance. In an ordinary year, Construction employment increases significantly from February to May – during 2019, this increase was 6.8% at the national level.For 2019, 7,062,000 Construction employment in February; 7,540,000 in May (not seasonally adjusted). While up from its low in April, national Construction employment nonetheless fell by 2.3% from February to May in 2020. Relative to 2019 performance for the same period, this corresponds to a loss of expected Construction employment of 9% (2.3% + 6.8%) from February to May in 2020.

State Comparisons—Geography is Important

This section compares job losses from the coronavirus recession at the state level. Because state-level detail is available only in the April jobs report (the state jobs report for May is due out in late June), our discussion focuses on job losses during the initial stage of the coronavirus recession, from mid-February to mid-April. The national jobs report for May indicates that the countrywide low point for job losses to date was registered in the data for mid-April; consequently, we expect to see improvements in mid-April job losses for most states when their employment data for May becomes available.

All states suffered severe job losses during this period, but individual experience varied substantially. State job losses as a proportion of pre-pandemic February employment ranged from highs above 20% to lows around 8%.

Figure 2 groups states according to their severity of job losses in mid-April. States indicated as High and Low are those whose job losses as of mid-April were 25% above or below the national average.Both national and state jobs reports for April show a 14.9% national employment loss from mid-February to mid-April. Revisions in the May national jobs report to the data for April raised the national employment loss from mid-February to mid-April to 15.4%. Our state categorizations in Figure 2 are based on April state-level data and therefore reflect the 14.9% national job loss from February to April. National data presented in Table 1 incorporate the May revisions and therefore show a 15.4% national job loss over the same period. The apparent inconsistency is the result of different vintages of available data at national and state levels of detail. Mid-High and Mid-Low states are those with job losses either higher or lower than the national average, but by less than 25%.

State variations in employment losses through mid-April show clear geographic clustering. States with the highest employment losses (more than 18.6%) are clustered in the Northeast (New York, New Jersey, Rhode Island, Vermont) together with a few other states (Michigan, Hawaii, Nevada) for which the underlying drivers of job losses are quite clear. Michigan’s economy, though more diversified than in the past, has very high exposure to automotive manufacturing; and in addition, the state had coronavirus hot spots in Detroit and surrounding counties. Both Nevada and Hawaii rely heavily on out-of-state tourism, which was devastated in both states at the start of the pandemic.

Similarly, states with the lowest employment losses (less than 11.2%) form a contiguous cluster along the axis from North Dakota to Oklahoma, with Virginia as the sole exception. The western states in this cluster are located interior to national boundaries (except for North Dakota), do not have major trade hubs, and have relatively small urban population centers.

States that experienced employment losses closer to the national average also exhibit geographic clustering, though less dramatically. Mid-High employment loss states (from 14.9% to 18.6%) follow a rough diagonal from Maine to Kentucky, with the addition of Wisconsin, Louisiana, California, and Washington. The last three states have significant exposure to trade via major shipping ports; in addition, Louisiana and Washington were early hot spots for coronavirus.

Mid-Low employment loss states (from 11.2% to 14.9%) include Western, Southwestern, and Midwestern states in a rough ring around the Low employment loss states; Southeastern states from Mississippi eastward to North Carolina and Florida; as well as Alaska, West Virginia, Maryland, and Washington, DC. In addition to being most numerous, states in the Mid-Low employment loss group are also the most diverse in terms of their economic profiles.

Is There a Simple Explanation for State Variation in Employment Losses?

The short answer to this question is that simple explanations work reasonably well for some states, but not very well for all states. In this section, we consider a few such candidates.

Economic Sector Mix. The logic of explaining a state’s employment experience via sector mix is straightforward. Since employment in certain sectors—for example, high-proximity or nonessential services, discretionary manufactures, and construction—were strongly impacted at the national level, then states with more pre-pandemic employment in these sectors ought to be most strongly affected, and vice-versa. In fact, this explanation works well for Nevada and Hawaii, with very high exposure to out-of-state tourism; and for Michigan, with high exposure to automotive manufacturing; but not so well in other states. It is possible to identify negative employment impacts as of mid-April in states with high tourism exposure (including Vermont, Florida, Louisiana, Montana, Colorado, and Wyoming) and in states with high automotive exposure (including Ohio, Kentucky, and Indiana), but these are secondary drivers of the affected states’ employment losses overall.

Our analysis suggests that in almost every state, employment loss through mid-April was driven by variation in the magnitude of employment loss across all sectors, rather than sector mix. That is, states with higher employment losses tended to have higher employment across all economic sectors, not higher concentrations in the sectors hardest hit by the pandemic.

Pandemic Severity. Here too the logic is simple. Coronavirus infection rates varied significantly across states; therefore, should not employment losses vary directly with the pandemic’s intensity at the state level? A quick look at the map in Figure 2 suggests that this explanation has merit. States with known coronavirus hot spots, for example, New York, New Jersey, Michigan, Louisiana, and Washington, all experienced employment losses above the national average. However, other states with relatively high infection rates per capita experienced employment losses below the national average, for example Colorado, Georgia, and Indiana. And some states with low infection rates per capita, including Maine, New Hampshire, Ohio, Kentucky, and Wisconsin, had higher-than-average employment losses. Once again, the simple explanation is a bit too simple.

A closer examination of the data reveals a more subtle story. Recall that our state employment data is for mid-April, roughly six weeks into the coronavirus pandemic. At that time, diagnosed cases were accelerating rapidly across the country, but regulatory responses in most states were still at an early stage. Most states issued stay-at-home orders in early April, but significant differences in states’ regulatory responses to the pandemic had not emerged at that time.

States did differ, however, in their rates of diagnosed coronavirus cases per capita (see Figure 3). As of mid-April, per 100,000 population:

We find that average state employment losses were positively related across these groups, but almost uncorrelated within each group. In simple terms, this means that big differences in case rates—on the order of 100 cases or more per 100,000 population—do predict increased layoffs and employment losses across states, but small differences are not impactful. Interestingly, we also find the widest variation in employment losses among 32 states whose mid-April case rates were below 100. These ranged from 9% or less (Oklahoma and South Dakota) to around 18% (Maine and New Hampshire).

We think that a fuller explanation of interstate employment losses from the coronavirus recession will take into account the pandemic’s severity by state, together with the degree to which the state is either open or closed—that is, the degree to which the state is economically self-contained or dependent on flows of goods and services going beyond its own borders. For states with high coronavirus case rates, the imperative to lock down economic activity dominates. But for states with low coronavirus case rates, variations in employment losses may have more to do with the pandemic’s effects on out-of-state trading partners —disrupting supply chains or reducing external demand for the home state’s goods and services.

For the moment, this remains an intriguing hypothesis deserving of more in-depth research. We will return to it in a forthcoming issue of the Quarterly Economics Briefing.

Postscript: A Note on Seasonally Unadjusted Data

We rely on seasonally unadjusted employment data because these are actual employment levels, and hence their differences represent actual job losses. In contrast, seasonally adjusted employment numbers are hypothetical numbers obtained by normalizing actual (unadjusted) employment levels for seasonal fluctuations across different sectors.

For example, seasonally adjusted construction employment is higher than actual construction employment during off-peak winter months and lower than actual construction employment during peak summer months. Seasonal adjustment makes sense when comparing employment data across different seasons or years but is misleading in the current context where we want to count actual job losses for approximately 12 weeks from mid-February to mid-May. We have noted an exception for construction, where seasonal employment increases significantly from February to May in a typical year.