Can You Trust the U.S. Unemployment Rate?

The latest employment report from the U.S. Bureau of Labor Statistics (BLS) was undeniably a shocker. The addition of 2.5 million jobs and decline in the unemployment rate in May foiled expectations of a continued deterioration in labor market conditions due to the COVID-19 pandemic. For many, surprise gave way to charges of bungling or, more darkly, assertions of political manipulation.

The BLS itself first flagged the existence of a reporting issue back in March, noting a flaw in the calculation of the unemployment rate. Dubbed a "misclassification error," it was responsible for producing an unemployment rate for May about 3 percentage points lower than it should have been, meaning the official rate should be 16.4 percent instead of the 13.3 percent reported.

Going back to March, the unemployment rate should have been 5.4 percent instead of the official 4.4 percent rate; in April, it should have been 19.7 percent instead of the reported 14.7 percent.

The June 5 report triggered questions of how the misclassification error happened, how the unemployment rate is reconciled with the number of people claiming unemployment insurance benefits and who is ultimately considered unemployed in the official count.

While a debate about the Trump administration's intentions with labor market data rages across blog posts and the Twittersphere, the simple takeaways appear to be that:

• The BLS—with its notes and explanations attached to each report—is practicing transparency.
• Collecting employment data during a pandemic has given rise to new challenges.
• The unemployment rate is troublingly high—but declined by 3.3 percentage points from April to May.

Erica Groshen, an economist at the Cornell University School of Industrial and Labor Relations in Ithaca, N.Y., and commissioner of the BLS from 2013 to 2017 under President Barack Obama, was also surprised by the report's findings but does not question the integrity of the agency or the numbers.

Groshen discussed with SHRM Online how the monthly unemployment rate is calculated and how the misclassifications occurred and said it would be "basically impossible" to tamper with the reporting or the data without detection.

SHRM Online: How is the monthly unemployment rate calculated?

Groshen: The unemployment rate is arrived at by means of a household survey of 60,000 U.S. households each month, conducted by hundreds of interviewers from the U.S. Census Bureau.

The process is hierarchical and mutually exclusive. First, the interviewers determine if the person [and others in the household] is employed or not. The basic question asked is "During the week that contains the 12th of the month, did you work for pay or profit?" If the person worked for at least one hour during that time, they are considered employed. Probing follow-up questions are meant to clarify the person's situation. If the person responds that they did not work for pay during that week, there are a series of questions to establish whether they are either unemployed or outside the labor force. The unemployed fall into basically two types: those actively looking for work or people on temporary layoff—a category which is usually small, but that's not the case now. If someone is on temporary layoff, they have a strong expectation that they will be called back to work within the next six months.

If you are either employed or unemployed, you are counted in the labor force. If you're neither of those, you are out of the labor force. The unemployment rate then is the sum of the number of unemployed people divided by the number of the people in the labor force.

SHRM Online: Can you explain how the official unemployment rate differs from what some call the "real" unemployment rate?

Groshen: First, it must be understood that receiving unemployment insurance benefits is not a factor in determining the unemployment rate. The Census Bureau interviewers do not ask whether people have applied for benefits or are receiving benefits. There is no relationship between the Department of Labor data on unemployment insurance benefits and the monthly BLS survey for the construction of the unemployment rate. [Editor's Note: The latest weekly unemployment insurance benefits data shows that about 30 million people are collecting benefits, although about 10 million of those are self-employed. The BLS shows 21 million people counted as unemployed.]

The unemployment rate has the virtue of being tried-and-true, having been applied consistently for 80 years. But the BLS also produces a set of six measures of labor market underutilization that range from narrow to broad, referred to as U-1 through U-6. The official unemployment rate is the U-3 measure. What is often called the "real" unemployment rate, or U-6, is the broadest and includes:

• Those who are unemployed by the U-3 definition.
• Those who want a full-time job but can only find part-time work or who've had their hours cut due to economic reasons.
• Those marginally attached to the labor force, who may be out of work for personal reasons but say they want a job.
• Discouraged workers who say they want to work but are not actively looking for a job because they believe they can't find one.

The U-6 measure will always be larger than the unemployment rate. Particularly in this recession, you see a much wider gap between the U-6 and the U-3 than seen normally.

SHRM Online: What is the so-called misclassification error?

Groshen: OK, imagine the interviewer asks you "Are you employed?" You were working when suddenly your employer was forced to shut down and placed you on temporary layoff. Being laid off due to a pandemic-related shutdown is a new situation for everyone, including the interviewers. I think it's reasonable that many people, including self-employed people, didn't consider themselves laid off, even though they weren't working. There is a category for people like that, intended for people who because of their own personal circumstances are not at work—they're sick, they're on jury duty, they're on vacation, things like that. Employees not at work because of COVID-19-related shutdowns ended up being counted as employed but not at work, when they should have been counted as unemployed on temporary layoff.

Neither the questions nor the interviewers have changed. Instead, when circumstances changed to something no one has ever encountered before, people responded to the survey in a way that led them to be misclassified.

SHRM Online: How did it happen?

Groshen: Here's my take. In advance of the March survey, remembering a similar misclassification issue in October 2013 during the federal government shutdown, the BLS issued special COVID-19 guidance to the Census Bureau interviewers to ensure that temporary layoffs would be properly counted as such. When the March survey responses arrived, BLS staff checked the data and saw a problem. Their best estimate is that about 1.4 million workers on furlough answered questions as if they were employed but out on some kind of leave. About 39 percent of workers on temporary layoff were likely misclassified in March.

In preparation for the April data collection, the BLS reiterated its previous COVID-19 guidance. Unfortunately, the intervention failed to resolve the problem. Although a smaller share of temporary layoffs appears to have been misclassified than in March [30 percent], the scale of layoffs was much larger in April, so 7.5 million temporary layoffs were likely misclassified.

For May, the BLS redoubled its efforts, providing additional supervisor-led training to interviewers. Yet, the problem persisted. The share of temporary layoffs likely misclassified declined further to 25 percent, but that still resulted in about 4.9 million likely misclassifications.

The impact of the misclassifications is consequential, yet the BLS has announced that it will not use its misclassification estimates to alter the official unemployment rate for March, April or May. One reason is that changing respondents' answers violates its practices. Changing answers carries the real risk of setting an unwelcome precedent. In addition, the public could rightfully be less certain that published statistics reflect the underlying data collected.

Going forward, I expect misclassifications to subside.

SHRM Online: Is it possible to manipulate the monthly employment data for political reasons?

Groshen: No. This is a much more automated process than most people realize. And the culture at BLS is one of complete allegiance to the integrity of the numbers. That's reinforced by the agency's directives, principles and practices. The only presidential appointee in the entire agency is the commissioner, with a four-year fixed term, which means he or she does not serve at the pleasure of the president. He or she cannot be fired by the president. The commissioner does not see any statistics until they are final, nor does he or she have access to the underlying microdata. Frankly, even if I had had access, I wouldn't have known how to manipulate it.

The survey data is collected by hundreds of interviewers independently, who enter their responses to the Census Bureau for collation. It is then delivered to BLS staff about a week before the data is released. The commissioner does receive the topline numbers on the Monday evening of the week the report is released but has no way to change them.

Also, the public use file with all the survey responses is made available to researchers. The microdata would have to be changed to match the tables in the report or it would eventually be discovered. It would have to be a conspiracy of enormous proportion including Census and BLS staff. It's not conceivable on a practical basis and notably there have been no red flags raised, like whistleblowers, staff resignations or calls to the media.

By Roy Maurer

June 16, 2020