Revising the Payroll Survey Benchmark: What To Expect


Revising the Payroll Survey Benchmark: What To Expect

October 6, 2004 3 min read
Tim Kane
Tim Kane is a Visiting Fellow at The Heritage Foundation’s Center for...

This Friday, the Bureau of Labor Statistics (BLS) will publish its regular monthly Employment Situation report, with statistics on payroll job growth as well as the national unemployment rate. But this particular report will be special, and not just because of the Presidential debate on the economy 12 hours later. Friday's BLS report will include an announcement about the annual "benchmark" revision to the payroll survey data. The benchmarking will not be made official until January 2005, but Friday's report will reveal the likely size of the change. A large upward revision would reflect what the household employment survey has been showing for months: that the job market is stronger than the payroll survey portrays and that more Americans are working today than ever before.


How Benchmarking Works

While the payroll survey is based on a sample of employers, the benchmark actually uses the entire population of employment records for March of every year to confirm the employment level. And from this benchmark, sample-based estimates that were made in previous months are revised.


For example, the benchmark may call for an increase of 120,000 jobs above the payroll survey's count from March 2004. This would be implemented by adding 10,000 jobs to each month's tally, all the way back to April 2003. Estimates for each month following the benchmark would be increased by the entire 120,000.


BLS also recalculates "birth and death" estimates for firms as a part of its benchmark revision.


And finally, BLS reruns seasonal adjustment and revises five years of seasonally adjusted estimates. The technical footnotes from BLS describe the process this way:

The sample-based estimates from the establishment survey are adjusted once a year (on a lagged basis) to universe counts of payroll employment obtained from administrative records of the unemployment insurance program. The difference between the March sample-based employment estimates and the March universe counts is known as a benchmark revision, and serves as a rough proxy for total survey error. The new benchmarks also incorporate changes in the classification of industries. Over the past decade, the benchmark revision for total nonfarm employment has averaged 0.3 percent, ranging from zero to 0.7 percent.

Total nonfarm employment in March was 130,630,000. If the benchmark revision comes in at the average, today's payroll numbers would be boosted by 391,890 jobs. And if the revision matches its high, today's payroll numbers would be boosted by 914,410 jobs.


BLS's fuller explanation of the benchmarking process is available online here:


What Happened Last Time

Historically, BLS released its benchmark report in early June with the May Employment Situation report. But by the fall of 2003, improvements in its methods allowed BLS to move its benchmark report up to January, with a pre-announcement in October.


Last year, many analysts anticipated the payroll jobs figures would be revised sharply upward, but they were actually benchmarked downward by 122,000. That is, there were 122,000 fewer jobs in March 2003 than estimated from the CES sample. However, the October 2003 announcement had predicted a downward benchmark of 145,000. So while the announcement is not exact, it tends to be very close.


Many analysts again predict that BLS will announce a large upward revision on Friday. Other economic data, besides payrolls, describe an economy that is stronger than the payroll survey reflects. One conclusion is that the payroll survey is simply suffering from a sampling error that will be corrected once the sample is replaced by a full population count. But that logic failed in 2003 for one simple reason: whatever problem plagues the payroll survey may not be limited to its sampling methodology.


The Heritage Foundation has documented that the payroll survey suffers from major structural problems. (See, for example, "Diverging Employment Data: A Critical View of the Payroll Survey.") One key finding is that payrolls-whether sample or population-do not count the self-employed, consultants, LLCs, and the like, all of which are mainstays of the new economy. Another structural problem is that payrolls systematically double-count due to job-changing, which makes the payroll survey's job estimates vulnerable to changing turnover rates. Understandably, job turnover declined sharply following 9/11, but prior to then, payrolls may have been inflated by as many as 1 million jobs.


For these reasons, the payroll survey benchmark announcement is unpredictable. And the reasons for its unpredictability should be taken as a grain of salt for overall payroll employment estimates as well. Unlike the payroll numbers, however, the unemployment rate is effectively carved in stone. It will never be revised. And during 2004, the unemployment rate has held steady around 5.5 percent, near its natural rate. The unemployment rate announced on Friday is likely to remain near that low level.

Tim Kane, Ph.D., is Research Fellow in the Center for Data Analysis at The Heritage Foundation.


Tim Kane