The Occupational Employment and Wage Statistics survey, the only establishment survey that reports wages by occupation across industries, began in 1971. The intellectual context was what Pennsylvania Senator Joseph Clark called the “Manpower Revolution,” which was the idea that structural and frictional unemployment could be significantly reduced through a targeted mix of development assistance, education and training programs, and rationalization of hiring. Tax cuts wouldn’t do the job alone. 

For labor-market planners, understanding the trends in occupational employment and earnings was necessary for targeting funds for training and development assistance. The idea appealed across the political spectrum. Clark was a champion of the Great Society who had sponsored Kennedy-era legislation to assist deindustrializing areas—the 1961 Area Redevelopment and the 1962 Manpower Development and Training Acts—and was a major ally of organized labor. For conservatives, such as Federal Reserve Chairman Arthur Burns, earnings data by occupation and industry helped to substantiate theories of “voluntary” unemployment allegedly caused by too-generous unemployment insurance and social security payments. 

Today the kernel of the Manpower Revolution of the sixties has grown to dominate our understanding of the labor market, whether in the theory of a “skills gap,” a “labor shortage,” or that of labor-market “monopsony.” Yet our primary federal tool for measuring wages by occupation and industry has withered through decades of underuse. During the 1990s, the OEWS survey was fundamentally redesigned to accommodate employers in the era of the Clinton administration’s welfare reform and “workfare” policies. As a result, the survey has lost much of its potential use today for workers and labor-market planners. It’s time to consider overhauling the survey to enable the public to get a better vision of the labor market in action. 

The History of the OEWS Survey Design

The method of the OEWS survey begins with the unemployment insurance system. In coordination with state workforce agencies, BLS selects a sample of establishments from administrative data produced by unemployment insurance programs. The sample is then divided by sector and industry, with BLS and state agencies surveying establishments by mail and by phone. The broader project was originally to emulate the “active labor market policy” of the Swedish Social Democratic Party, in which a central employment service facilitated the re-allocation of labor into expanding industries and out of contracting industries. Primary users of early OES data included the Department of Labor’s Manpower Administration, augmented by the 1962 Manpower Training and Development Act; the Department of Health, Education, and Welfare; and local governments funded by the 1963 Vocational Education Act. As BLS describes the survey’s purpose today, “Uses of the data include evaluating employment and wages by industry, occupation, and geographic area; updating prevailing wages for foreign labor certification; projecting occupational employment for the nation, states, and areas; informing vocational planning; estimating social security receipts, as an input to calculating reimbursement rates for Medicare and Medicaid providers; identifying STEM related employment and wages for the National Science Foundation; calculating occupational injury rates; serving as an input to the Employment Cost Index and to the President’s Pay Agent report,” among other uses.

Since 1975, two federal agencies have been the primary users of BLS’s OES data. The first is the Employment Training Administration (ETA), which that year absorbed the Manpower Administration of the 1950s and 1960s. By the early 1980s, ETA expanded to oversee the Trade Adjustment Assistance program, Registered Apprenticeships, the Job Corps, and a variety of training grants for employers. Both workers and program administrators can use data on wage rates for the occupations targeted by these programs. The second OES user was the Office of Foreign Labor Certification, formerly the Alien Certification Program, which required prevailing wage data for the various visa programs it helped to administer—the H-2B, H-1B, H-1B1 and E-3 work visa programs. This ensured that imported labor would not be used to alter wage patterns in a given industry or region. 

At the request of the Alien Certification Program, in 1996 BLS made fundamental changes to the OES survey to meet a number of demands from the program’s employer constituency. First, the Alien Certification Program requested occupation earnings for 770 categories of worker. Second, these data users asked occupational earnings to be specified by city or Metropolitan Statistical Area. Third, and most consequently, employers asked that they not be contacted by BLS more than once every three years. “This was done to limit the response burden on any one employer,” two BLS staticians later recounted. This “constraint” on data collection, the officials explained, “created additional estimation issues as only one-third of the certainty units could be sampled each year.”  Thus, while the sample size was expanded to accommodate all 770 occupational categories for the Alien Certification Program, nearly doubling to over a million establishments, it was also broken up to be measured at different points in time in order to reduce “respondent burden.”

The effect of the new estimation and survey methods was to render occupational earnings data useless for comparison across time. Every six months, the agency contacts one sixth of its sample. It then releases annual estimates of occupational earnings by MSA and industry based on the entire sample. For any year, five-sixths of the sample are adjusted according to the BLS Producers’-Cost Index (PCI) to make them current and comparable to the most recent data. The PCI is a national index: the effect of this estimation method is to make occupational wages in a given category appear to rise across regions at the same rate, whether or not wages actually behave in this manner. 

Within BLS, the inability to use OEWS data to make comparisons through time has long been recognized as a limitation of the survey. For this reason, in recent years BLS initiated “a long-term research project that resulted in a recommendation for a new sample design and estimation method” to make our national wage data into a time series. In 2019, BLS staff published the model that will be the basis for the new estimation data, which predicts wages by occupation according to similar establishments in the underlying administrative data from the unemployment-insurance system. In 2022, OEWS data will be published using this model, enabling wage data to be compared back to at least 2015. However, as the staff note, “a new sample design has not been implemented,” meaning the current methods of selecting and contacting establishments will not change. The new estimation method enables the creation of time series data “without changing the sample design.” This leaves considerable room for improvement in the empirical basis of data. 

Uses of Labor Market Statistics

The labor market reforms of the 1990s worked largely to strengthen employers’ hands in hiring and managing workers. Work requirements for social security and unemployment insurance (UI) programs, for example, increase the number of job applicants and give employers greater discretion in hiring decisions. Another example of this move to separate hiring decisions from the influence of workers was the creation of “America’s Job Bank” in 1995, a national listing service the ETA created to rationalize its numerous employment programs and those of the states into a single “one-stop shop” for employers. A database of employer-provided openings was one device to police unemployment-insurance “fraud,” given that unemployed workers were required to demonstrate job applications as a condition of their unemployment assistance. Growing out of this effort to ensure work requirements for UI recipients was the Job Openings and Labor Turnover Survey (JOLTS), which Congress piloted in 1998 and established in 2001. 

There are widely divergent uses for occupational statistics. Designing a survey and estimation method depends on these intended uses. Under the Davis-Bacon Act, for example, Congress mandates that public projects employ construction labor at prevailing wages. This information is produced by the Branch of Construction Wage Determination in the Wage and Hour Division of the Department of Labor—rather than the OES survey of the BLS. The construction unions which help to enforce the Davis-Bacon Act are some of the primary users of this data. This division in data production reflects the underlying division of the American labor market between collective and individual wage bargaining. 

Who are today’s occupational employment statistics produced for? The design of the OEWS until 2022 has rendered comparisons of wages through time meaningful only if one can reasonably assume that wages for given types of jobs increase at the same rate in different regions. This assumption rarely holds, given the size of the national economy, the different growth rates across industries and regions, and the different institutions of wage determination across both. But current data was still useful, and mandatory for employers hiring guestworkers. The survey reflected this purpose, since employers of immigrant labor required annually accurate data, but had little need for chronological analysis or desire to regularly supply information for a federal survey. 

In 2007, the George W. Bush administration discontinued America’s Job Bank and replaced its listings with those of DirectEmployers Association, a syndicate of 200 corporations who contract with commercial listing services, such as Monster, Indeed, Google, etc. For large employers, patterns in occupational earnings are apparently best kept private. In 2008, the Bush administration’s DOL began allowing employers to submit private wage surveys for certain types of guest worker visas—a controversial decision fought by migrant-worker organizations throughout the Obama administration. This difference in how prevailing wages are calculated gives employers an incentive to reclassify certain work to recruit foreign labor at lower wages. For certain skilled occupations cut out of the general guestworker laws, the very bureau that requested the current form of the OEWS on behalf of employers no longer requires them to use it. 

Employers’ demands to have specific earnings data for work visas came into direct conflict with producing empirical data on changing wage patterns or using regional occupational earnings trends in national employment and training planning. Those same demands eventually dispensed with public data requirements altogether. Thus, the OEWS survey’s ostensible purposes no longer exist. 

This is unfortunate for those who imagine structural solutions to the problem of labor allocation beyond the general macroeconomic policies of raising demand. Now more than ever the US public requires accurate information about wages across occupations and regions. Workers, as the great majority of the public, have a special interest in this information. Union federations hoping to establish regional or national wage rates, for example, can use such data to deploy organizing resources and influence local union wage policies. Employers seeking to hire labor should have better access to data about their competitors in hiring. Community colleges and training programs should be able to plan in the interests of the public, rather than merely as appendages of the largest, most influential, and powerful employers and donors in their region or industry. If organizations are to understand their private planning in a national context, there must be public data with which to compare. 

The new estimation method BLS will use in the 2022 OEWS data will offer the beginning of a solution. For the first time in decades, federal wage data by occupation and region will be usable as a time series. Yet giving such data a firmer empirical grounding requires redesigning the OEWS sample, either reducing the number of occupations or dramatically expanding staff. Redesigning the sample may be a particularly ornery task, given the joint state-federal nature of the unemployment-insurance system on which the underlying data is based: methods must be altered not only for BLS and its six regional offices, but for the state workforce agencies with which they collaborate to contact establishments. 

The BLS has not published any indications of whether or when such a redesign of the OEWS sample will occur. Since 2008, the Bureau of Labor Statistics budget has provided funds for around 500 full-time employees for its Labor Force Statistics program, which includes OEWS. In the same period, the population has grown some 10 percent. In its budget request for the 2022 fiscal year, the Department of Labor is admirably requesting a modest expansion to the Labor Force Statistics program. But the amount of full-time equivalent payroll for statisticians and field is considerably less than the increase in population. The public would benefit from much more.

To give a sense of our nation’s current data priorities, consider that the Research and Development Budget for the Facebook corporation was $18.4 billion in 2020, according to filings required by the SEC. For six months of 2021 alone, Facebook spent $11.2 billion on research and development. For the 2021 fiscal year, by comparison, the congressional appropriation for the entire Bureau of Labor Statistics was $655 million, or 1/28 of the Facebook research budget. 

Determining which occupational categories should be measured and for what purposes, of course, must depend on the representatives of data users. One possible use, for example, is for an industry board to determine wages, such as has been recently proposed for both app-service employment and fast food franchises in state legislatures in California and New York. Without comparison to comparable occupational categories, what basis is there for wage administration?  If workers are to be one such voluntary participant in the labor market, then workers’ representatives must have some say over how the OES survey should be conducted in the future. Ultimately the standard for improvement depends on the needs for data users.