Last week, 2.2 million workers applied for unemployment benefits. This is the twelfth week in a row that initial unemployment claims are have been more than twice the worst week of the Great Recession.

Of the 2.2 million who applied for unemployment benefits last week, 1.5 million applied for regular state unemployment insurance (UI), and 0.7 million applied for Pandemic Unemployment Assistance (PUA). PUA is the federal program for workers who are out of work because of the virus but who are not eligible for regular UI (e.g., the self-employed). At this point, only 42 states and Puerto Rico are reporting PUA claims. This means PUA claims are still being undercounted.

How is it that we are seeing large numbers of initial unemployment claims now, when the jobs report from last Friday shows we added jobs in May? One key thing is the fact that the unemployment benefits numbers don’t account for changes in hiring. If there are a large number of layoffs, there can still be job growth if there is also a lot of hiring (or rehiring). Further, some unemployment claims since April may be from people who actually lost their job in March or April but didn’t apply right away (perhaps because they couldn’t get through the system).

Many commentators are still reporting the cumulative number of initial regular state UI claims over the last 12 weeks as a measure of how many people are out of work because of the virus. I believe we should abandon that approach because it ignores PUA—and is thus an understatement on that front—but overstates things in other ways (for example, some who were laid off and applied for UI in March or April may now be going back to work). Instead, we can calculate the total number of workers who are either on unemployment benefits, or have applied and are waiting to see if they will get benefits, in the following way:

A total of 18.9 million workers had made it through at least the first round of regular state UI processing as of May 30 (these are known as “continued claims” or “insured unemployment”), and 3.2 million had filed initial UI claims on top of that, but had not yet made it through the first round of processing. And, 9.7 million workers had made it through at least the first round of PUA processing by May 23, and 2.8 million had filed initial PUA claims on top of that but had not yet made it through the first round of processing. Even further, by May 23, another 773,000 workers had made it through at least the first round of processing in one of the other unemployment benefits programs, or had filed initial claims in other programs on top of that. The largest of the other programs are Pandemic Emergency Unemployment Compensation—which is available to workers who have exhausted their regular state benefits—and Short-Time Compensation, which is an alternative to layoffs where employers reduce work hours rather than laying off workers, and workers get partial benefits. Altogether, that’s 35.4 million workers who are either on unemployment benefits or who have applied very recently and are waiting to see if they will get benefits. That is more than one in five U.S. workers.

It’s important to note that of the 35.4 million workers “on” unemployment benefits, more than a third (35.4%) are on PUA. This certainly underscores how enormous the gaps are in our regular state unemployment insurance programs.

One question I’ve gotten since the release of the May jobs data last Friday is, “How do May’s UI claims square with May’s jobs numbers?” Using the May jobs data, I calculate that in mid-May there were 32.5 million workers who were either officially unemployed or were out of work as a result of the virus but were being counted as either employed or out of the labor force. I arrive at that 32.5 million by summing the officially unemployed (21.0 million), the 4.9 million workers that the Bureau of Labor Statistics (BLS) notes are temporarily unemployed workers who are being misclassified as “employed not at work” (see question 13 here for more explanation), and the 6.6 million who are out of work as a result of the virus but are being counted as having dropped out of the labor force because they aren’t actively seeking work. (I calculate the 6.6 million by subtracting the actual level of the labor force in May, 158.2 million, from what the level of the labor force would likely have been if the virus hadn’t happened, 164.8 million, which I calculate by multiplying the May working-age population by the February, pre-virus labor force participation rate.)

On May 16, there were 30.0 million workers claiming unemployment benefits in all programs, not including Short-Time Compensation. Comparing that to the 32.5 million workers who were either officially unemployed or were otherwise out of work as a result of the virus in mid-May suggests a very high share of people out of work as a result of the virus are claiming unemployment benefits. However, this direct comparison is almost surely overstating that share. For example, some part-time workers are receiving benefits, but they aren’t counted in the 32.5 million, because that figure only includes people who aren’t working. Also, as I mention below, there may be some double-counting of benefits claims. But nevertheless, it does appear that a large share of people who are out of work as a result of the virus are claiming unemployment benefits and that is very good news.

All in all, today’s UI data highlight the deep recession we are now in. It’s important to remember that this recession is going to exacerbate existing racial inequalities by causing greater job loss in Black and Hispanic households than in white households.

Policymakers need to do much, much more to fight this recession and set our economy up for a strong recovery, which will not happen without intervention. For example, without massive federal aid to state and local governments, 5.3 million workers in the public and private sector will likely lose their jobs by the end of 2021. Further, the across-the-board $600 increase in weekly unemployment benefits, which was one of the most effective parts of the CARES Act on both humanitarian and economic grounds, should be extended well past its expiration at the end of July. We also need increased support for safety net programs, including SNAP (or food stamps). And importantly, we can’t turn off federal relief too early. The expiration of relief provisions should be tied to actual economic conditions. Assigning arbitrary end dates to provisions to sustain the economy makes no sense when there is so much uncertainty about how the recovery will unfold. Automatic triggers (where provisions phase out as the unemployment rate falls or the employment-to-population ratio rises) would alleviate the very real threat that we turn off federal aid when the economy is still too weak, hamstringing the recovery.

Note on the unemployment benefits data: Claims for UI and PUA should be completely non-overlapping because that is how the Department of Labor has directed state agencies to report them. However, some states may be misreporting claims so there may be some double-counting. Also, I focus on the not-seasonally-adjusted numbers for regular state UI claims because the way DOL does seasonal adjustments of unemployment insurance claims data is distortionary at a time like this. Claims from all other programs, including PUA, are only available on an unadjusted basis.