On June 5, the Bureau of Labor Statistics (BLS) released the Employment Situation Summary or “jobs report” which described the situation of the labor market as it stood on the week of May 12.
The report made headlines. According to the household survey, there had been an increase in the number of Americans employed by 3.8 million from the prior month, equivalent to 1.5% of the US adult population. Though this increase represents only a fraction of the 23 million jobs lost in April, the number was better than expected, and it led the Dow to gain around 800 points to reach its highest point since February. The report was heralded by Trump, who called it “rocket-like” and caused a minor commotion as some in social media speculated (without evidence) about political interference in these numbers.
The expectations were low because of the sustained high numbers of new Unemployment Insurance (UI) claims. The number of new UI claims usually tracks the number of new people who recently lost a job, and hence are useful markers for overall job loss. However, this time, many of the UI claims seen in April and May were delayed claims from people who had lost their job in April, but who hadn’t been able to get their claims processed due to many states’ UI offices being overwhelmed (due to the extraordinary surge in claims, the changes in rules and additional programs brought about by the CARES Act).
For a trader who paid attention to the Understanding Coronavirus in America (UCA) tracker instead of the claims data, June 5 could have had been a very good day. According to our data, there were 3.7 million more Americans working in the weeks surrounding June 12, almost exactly matching the estimate from the BLS.
The Understanding Coronavirus in America is a tracking panel that surveys a representative sample of Americans roughly every two weeks. The survey was launched on March 10, and includes a national bi-weekly questionnaire on employment, health behavior, and other topics. It is nested in the Understanding America Study, a sample of about nine thousand people, which is nationally representative. An important feature of the Understanding America Study compared to many other sources is that it is sampled from physical addresses, and provides internet access to households that don’t have it. By comparison, many other surveys are “opt-in”, and while they calculate weights to match population on some demographic characteristics, they recruit only people who self-select on-line, and hence they miss important swathes of the population. Within the UAS, a subset of more than 7,000 individuals are part of the UCA tracking panel.
Surveys at the scale of UCA are not often used to track employment levels at a monthly frequency. The margin of error of a sample of this size is high relative to the usual month to month changes in the employment numbers. In “normal times,” whether the country adds two, three or four hundred thousand is important to understand the health of the economy. But these are not normal times. The drop in employment in April was 23 million. In May, the “surprise” was that four million jobs were added, while most economists expected a net loss of jobs.
When the uncertainty of the future of the labor market is of this magnitude, a well-designed survey of a nationally representative sample can be useful. The UCA surveys ask respondents about their current employment status. Depending on the labor force status in the prior survey, the respondents are asked whether there have been changes in that status. For instance, if a respondent was on a temporary lay-off in the prior survey, the new survey asks whether she has been recalled, started working on a different job, etc. This allows us to construct an estimate of the number of Americans working.
The data are used to produce a graph with a daily estimate. The microdata, however, is also made publicly available to researchers upon registration and signing of a data user agreement. For this blog post, we use the microdata from the entire two-week surveys closest to the 12th of each month, which is the time period in each month during which the Bureau of Labor Statistics surveys Americans about their employment situation.
Our estimates are close to what the Bureau of Labor Statistics produces. The Employment-to-population ratios published by the Bureau for Labor Statistics for February, March, April and May are 61.1%, 60%, 51.3%, and 52.8% respectively. Our job numbers for March through May are 61.9%, 50.2%, and 51.6%. Note that we pick up the improvement from April to May, with a difference of only 0.1 percentage points.
One thing to note, however, is that if we calculate the job loss between March and April for instance, then we find that (61.9- 50.2)/61.9=18.9% have lost their job. Similarly, if we compare May to March, we calculate a job loss of (61.9-51.6)/61.9=16.6%. These numbers are higher than those from the BLS. Why are our job loss numbers much more dramatic? As has been reported extensively in the press by now, BLS analysis has shown that the Census Bureau interviewers (the Census does the household interviews for the BLS) have misclassified many respondents who “don’t work because of the coronavirus” as having a job, while these should have been classified as unemployed.
Given this issue with misclassification, there is an added uncertainty about the topline for employment in the next report. If the misclassified group gets smaller, then some of the “true” employment gains will not show in their topline employment, while they may show in ours.
With this caveat in mind, our data shows that employment growth in June continues at about the same pace as in May. We predict that an additional 1.6% of the U.S. adult population, or 4.1 million people, gained employment between the week of May 12 and the week of June 12.