Labor Force Statistics During the COVID-19 Pandemic: Challenges in Using the Current Population Survey to Determine the Unemployment Rate
By Sanjukta Chaudhuri
After a record 128 months of steady economic expansion (from June 2009 through February 2020), the United States has entered a period of labor market upheaval due to the COVID-19 pandemic. Efforts to slow the spread of the virus have had an impact unprecedented in size and speed on the labor market. As a result, conventional methodologies that rely heavily on survey data to estimate labor market measures may not adequately capture the rapid employment changes we've seen during this crisis. For Minnesota's Labor Market Information (LMI) office and the Bureau of Labor Statistics (BLS), the situation has presented unique challenges in measuring unemployment.
Nationally as well as in Minnesota, various orders shutting down or limiting business operations to ensure social distancing have resulted in layoffs, reduced hours, a surge in Unemployment Insurance (UI) applications and reduced labor force participation. In April, the U.S. unemployment rate surged to 14.5%, up 10.3 percentage points from March, and in Minnesota our official unemployment rate went from 2.9% (seasonally adjusted and revised) for March to 8.7% (seasonally adjusted and revised) for April. However, the official unemployment rate may not capture the immediate extent of the downturn.
This is because the Current Population Survey (CPS), which is the primary data source used to generate monthly labor force measures, like the unemployment rate, at the national and state levels is not designed to capture rapid changes but instead trends over time. The CPS is a monthly household survey sponsored jointly by the U.S. Census Bureau and the U.S. BLS that has been running since the 1930s. The CPS is one of the oldest and most trusted surveys in the U.S. It generates a wealth of information on the labor force including participants and non-participants, employment, unemployment, industry and occupation of employment, educational attainment and demographics including race and ethnicity. The CPS data are used to develop alternative measures of unemployment including the number of discouraged workers, duration of unemployment, job seeking activities, number of part-time workers resulting from economic reasons, reasons for not working and reasons for not seeking employment.
The monthly fieldwork for the CPS is conducted during the "survey week", which includes the 19th of the month. The "reference" week is the previous week, which includes the 12th of the month. Survey administrators use in-person and telephone interviews to ask respondents about their labor force status during the previous week, i.e. the reference week. Households from all fifty states and D.C. are selected through scientific survey methods. A total of 60,000 households are surveyed every month. National results from the monthly CPS are published two weeks after the survey week in the employment situation release. In May 2020, 870 households were surveyed in Minnesota.
The following are six reasons why the unemployment rate, generated from the CPS, may not have done a good job at capturing the full extent of the labor market downturn during the first months of the current crisis.
1. Survey technique issues: Normally, the CPS is administered through in-person and telephone interviews. Due to social distancing measures, only telephone interviews were conducted. This resulted in a lower response rate than normal. For the CPS overall, in March the response rate was 73 percent, ten percentage points below normal. In April, the response dropped to 70 percent, while in May, the response rate was even lower at 67 percent, about 15 percentage points lower than pre-pandemic months.
Compared to the overall CPS, Minnesota's response rates were higher, although not as high as the pre-pandemic average and much lower than normal between March and May. Between 2008 and 2019 the average monthly household response rate for Minnesota's CPS was 91.6%. However, it dropped in March, April and May to 79.6%, 81.4% and 79.8% respectively: all three are the lowest on record for Minnesota since 2008.
Despite lower response rates, BLS was able to create estimates that were statistically valid and reliable. However, in small and medium size states with small samples, like Minnesota (our sample is 900 households monthly) a suppressed response rate can make survey results less reliable statistically, but not invalid or insignificant.
2. Temporary lay-off vs. employed but absent from work: With temporary closures beginning March 17 and a Stay at Home order in effect from March 28 through May 17, many workers were laid off on a temporary basis. These workers should ideally be categorized as unemployed. However, in many cases they were instead reported as being employed but absent from work. At the national level, this created an error in measuring the number of unemployed workers, which should have been much higher.
In Minnesota the number of observations in the category "employed, but not at work for other reasons" was higher than average during the pandemic vs. the pre-pandemic months. Table 1 shows over-the-year comparisons for March, April, and May (years 2016 through 2020). In May, for example, the number of observations in this category was 57,513 which was 49,465 more than the average of May in the previous four years. The same conclusions can be drawn for March and April.
The possibility of misclassification means that unemployment might be higher than estimated in Minnesota. Table 2 shows the adjusted numbers of employed and unemployed participants that subtracts the employed but absent from work from the number of employed and adds them to the number of unemployed. In all three months, the unemployment rate would be much higher than published estimates show. In May, for example, the number of unemployed would increase by 49,465, increasing the unemployment rate from 10.8% to 12.4%. In April, the unemployment rate would increase from 8.9% to 10.9%.
Table 1. CPS Employed, but Not at Work for Other Reasons, Minnesota
|Difference Between 2020 and 2016-2019 Average
|Source: Current Population Survey and Bureau of Labor Statistics
Table 2. CPS Data Adjusted for Employed but Not at Work for Other Reasons, Minnesota
|Adjustment for misclassification
|Change in Employed
|Change in Unemployed
|Source: Current Population Survey and Bureau of Labor Statistics
3. Unemployed vs. out of labor force: In order to be counted as unemployed, a worker must be willing and able to work and actively seeking employment or must be on temporary layoff waiting to be recalled by their employer. There are at least two reasons why unemployed workers in this category might have been misclassified as not in the labor force rather than unemployed. First, an unemployed worker may want to actively seek work but is involuntarily not seeking a job due to business establishments being closed. Second, the Stay at Home order restricted mobility and required social distancing, making job search impractical in many cases.
4. Increase in labor underutilization: With stay-at-home orders reducing demand in many industries, a typical reaction of business establishments is to cut hours for workers. These cuts in hours have increased the number of people working part-time for economic reasons, a measure of labor underutilization, also known as involuntarily part time. Although not officially unemployed, these underutilized workers nonetheless share many of the symptoms of unemployment. The alternative measures of unemployment that take labor underutilization into account should be even more closely examined during this period.
5. Empirical relationships: The empirical relationships between explanatory variables, including labor force (CPS), Unemployment Insurance claims, and payroll jobs (Current Employment Statistics) are critical for seasonally adjusted monthly labor force statistics. Estimates are seasonally adjusted to account for regular and predictable seasonal changes in the labor market over the course of a year. The estimation of seasonally adjusted level changes in labor force, employed, and unemployed rely heavily on a stable relationship over time between the covariates. In other words, the methodology is not designed to capture sudden wild swings in the labor market like we're seeing now.
One such wild swing is the sudden surge in UI claims. The increase in claims was so large and occurred so quickly that the past empirical relationships on which the econometric models relied may not be suitable in the current situation.
6. CARES Act: The CARES Act provides for special UI compensation programs during the pandemic, namely the Pandemic Unemployment Assistance (PUA) and Pandemic Emergency Unemployment Compensation (PEUC). These programs will likely create changes for many months in the historic relationship between payroll, UI claims, and unemployment. The pandemic programs relax the conventional qualifications and time limits for drawing unemployment compensation, with the result of temporarily changing the very definition of unemployment. Specifically, both programs provide flexibility to the requirement that individuals actively seek work while obtaining UI benefits.
For the foreseeable future, Minnesota's LMI office will continue to use a combination of monthly labor force statistics, UI claims, and alternative measures of unemployment to draw conclusions about the true impact of COVID-19 on the labor market.