What Are the Margins of Error in the BLS Monthly Jobs Reports?

Following the incredulous May jobs report, I wondered: What are the margins of error assumed by the BLS? Here are the answers.

The above chart is from BLS Current Employment Statistics (CES).

90 Percent Confidence Ranges

Those numbers are for the first release. The numbers vary slightly by revision.

I asked the BLS for the 2023 confidence numbers and received a prompt answer including an Excel spreadsheet.

Thanks!

BLS 2023 CES Monthly Jobs Report 90% Confidence Levels

Those are seasonally-adjusted confidence levels. I do not have unadjusted confidence levels but they should not be wildly different. Because ….

If the unadjusted numbers are not accurate, the seasonally-adjusted numbers cannot be accurate either.

3-Month and 1-Year 90 Percent Confidence Ranges for 2023

• The 3-month 90 percent confidence level is +- 207,300.
• The 1-year 90 percent confidence level is +- 377,500.

I am most interested in the 3-month and 12-month numbers because QCEW numbers are quarterly.

QCEW (Quarterly Census of Employment and Wages) is a count of Unemployment Insurance (UI) administrative records submitted by 11.9 million establishments. That’s about 99 percent of the data.

Nonfarm Payrolls are are from the Establishment Survey (CES). The sample survey was 666,000 individual worksites in 2023. That’s about 5.6 percent of the data.

QCEW will be far more accurate than the BLS monthly convolutions but QCEW lags CES by about 5 months.

How Much Does the BLS Overstate Monthly Jobs?

Note: The numbers in all of my charts are unadjusted. They need to be because QCEW numbers are not seasonally adjusted.

I discussed above QCEW chart on June 6 in How Much Does the BLS Overstate Monthly Jobs? Here’s the Answer

In retrospect, by “overstate” I should say “differ” although in this case I am confident overstate is correct.

CES vs QCEW Full Year 2023

• CES: 155,211,000 to 158,269,000 (+3.06 million)
• QCEW: 152,525,000 to 154,848,000 (+2.32 million)

I do not have confidence levels on QCEW but they should be much more accurate than CES.

For the full year, the difference between QCEW and CES is 735,000.

I struggle to believe the one-year 90 percent confidence range of +-377,500 as stated by the BLS.

Quarter-Over-Quarter Change in Nonfarm Payrolls vs QCEW

What we really need to see is the difference between the reports. Thus …

Quarter-Over-Quarter Change in Nonfarm Payrolls CES Minus QCEW

Quarter-Over-Quarter Notes

• For 2023 Q3, the CES report shows a gain of 3,000. QCEW says the loss is 400,000. The discrepancy is 403,000.
• For 2023 Q4, the CES report shows a gain of 1,424,000. QCEW shows a gain of 820,000. The discrepancy is 604,000.

For the last three years, the quarter-over-quarter discrepancy between the CES change and the QCEW change ranges from -514,000 to +604,000.

The BLS states that the 90 percent 3-month CES confidence level for 2023 is +-307,300.

Once again, I struggle to believe the stated CES confidence levels since QCEW rates to be more accurate.

Let’s look at the year-over-year numbers again. For ease in discussion, I repeat the chart.

Year-Over-Year Chance in Nonfarm Payrolls vs QCEW

Unlike quarter-over-quarter changes, that pattern does not seem random.

There appears to be a clear bias at economic turns for CES to undercount jobs coming out of recessions and overstate them headed into recessions.

Otherwise, discrepancies roughly offset.

Directional Bias Questions

• Has something fundamentally changed since Covid?
• Have mass retirements by baby boomers impacted the numbers?
• What about the CES Birth-Death model at turns?

The birth-death adjustment is the BLS estimate of business jobs created by new businesses or lost when businesses shut down.

Mass layoffs have nothing to do with birth-death adjustments, nor do bankruptcies unless companies shut their doors.

Birth-Death Adjustment 2024

The BLS birth-death adjustment added 231,000 jobs to the unadjusted total in May. It added 603,000 to the unadjusted numbers for the year.

Please do not subtract these unadjusted numbers from the reported seasonally adjusted numbers! The Birth-death model did not add 231,00 jobs to the May report as widely reported.

Were there 603,000 new jobs this year created by new businesses? 231,000 in May?

Birth-Death Math for May 2024

• The unadjusted total payroll for May including the B/D adjustment is 158,918,000 (158.9 million)
• Of that unadjusted total, 231,000 (231 thousand) were from the birth death adjustment
• The unadjusted total before the addition was 158,687,000

The adjustment added 231,000 / 158,687,0000 or 0.15 percent to the unadjusted number.

If the birth-death numbers are randomly wrong, then 0.15 percent does not matter much. But what if the errors are not random?

For the sake of argument, what if the error is half that, a seemingly insignificant 0.075 percent per month, but in the same direction for 6 months?

The cumulative error for six months would then be 0.45 percentage points. Is that enough to matter?

Y-O-Y Percent Percentage Change in Nonfarm Payrolls vs QCEW

In the last six months, the discrepancy between CES and QCEW has gone from -0.10 percent to 0.45 percent.

But 0.45 percent is 735,000 jobs. That’s 122,500 jobs a month.

Curiously, cumulative January-May birth-death adjustment is 603,000. That’s 120,600 a month.

But those birth-death numbers are for 2024. Let’s revisit with 2023 birth-death numbers.

Birth-Death Adjustment 2023

Is it possible that we are headed into recession with far fewer net birth-death jobs than expected?

A historical check, however, shows a recent tendency of the BLS to understate the number of jobs in the birth-death revisions.

Preliminary vs Benchmark Birth-Death Annual Totals

The only negative revisions were in recession years. And those revisions don’t suggest the major discrepancies we now see.

Also, the BLS has revised its methodology in the above timeframe.

For January through May 2024 we have +603,000 preliminary.

For 2023 we have 1,266 preliminary.

2023 Preliminary Birth-Death Numbers

• Q1: 3,000
• Q2: 635,000
• Q3: 264,000
• Q4: 364,000

The birth-death numbers may be a source of the CES-QCEW discrepancy, perhaps a likely source. But unless the benchmark revision is massive, it’s probably not the sole explanation for the discrepancy.

Finally, the 2023 Q4 QCEW number is preliminary. Expect lots of revisions. I believe negative.

Payrolls Rise 272,000 Employment Drop 408,000

The data is confusing and conflicting in many ways.

For discussion, please see Another Bizarro Jobs Report – Payrolls Rise 272,000 Employment Drop 408,000

Nonfarm Payrolls vs Employment Gains Since May 2023

• Nonfarm Payrolls: 2,756,000
• Employment Level: +376,000
• Full-Time Employment: -1,163,000

In the last year, jobs are up 2.8 million while full-time employment is down 1.2 million

Similarly, there is a huge discrepancy between Gross Domestic Product (GDP) vs Gross Domestic Income (GDI)

Real GDP vs Real GDI

GDP and GDI are two measures of the same thing. Income received should match product produced.

However, income is historically low vs GDP.

For discussion of the GDP-GDI discrepancy, please see More Soft Economic Data, Q1 GDP Revised Lower, Q4 GDI Significantly Lower

GDI lends credence to the idea that the household survey (employment) is correct, not the CES establishment survey (nonfarm payrolls) with its assumed birth-death adjustments.

Importantly, QCEW fits in the picture as well.

QCEW, the Household Survey employment numbers, and GDI are in sync. The outlier is the nonfarm payroll report.

On that basis, coupled with weakening trends in hard data, I repeat my unpopular opinion: A Second-Quarter Recession This Year Looks Increasingly Likely

I am passing this post to the BLS to see if I can get some comments.