US Strategy Weekly: A BLS Data JOLT

President Trump’s firing of BLS Commissioner Erika McEntarfer may have been unfair for several reasons and we doubt that July’s report was politically motivated. But we wonder if our shock to the Commissioner’s firing is because she works in the public sector. Making large and consistent errors in forecasts when working in the private sector could easily result in one being fired for cause. But regardless of the fairness of this event, it placed a much-needed spotlight on the messiness of government data. It may result in some necessary changes and hopefully result in more reliable data.

The monthly employment report has always been an “estimate” of jobs built on a survey, and while the data has always been imprecise, the estimates have simply gotten worse in our view. We have known this for years, which is why we focus on broad based trends in the data rather than any individual data point. And if you want to use government data, or any data properly, you must first understand its methodology and its weaknesses.

The BLS reported a paltry 73,000 jobs were added to the workforce in the month of July, but more shockingly, previous months were reduced by 258,000 jobs. This was a record level revision and the BLS attributed it to additional data and a recalculation of the seasonal adjustment. The unemployment rate increased slightly to 4.2%. The report showed that government employees declined by 10,000 but there were 14,100 job losses in the federal government excluding the post office. State employment increased by 5,000. The media reported that the response rate to surveys in the household survey declined from 90% in 2019 to 67%, but that survey impacts the unemployment rate not the 258,000 job revision in payrolls. And the media also attributed the poor response rate to federal job cuts. However, this is nonsensical since numbers are reported by individual state departments of labor, where July’s report shows that jobs increased! We would also note that the BLS website shows that the response rate for the establishment job survey fell from 69% in 2015 to 42.6% in March 2025 and the household survey response rate fell from 88% to 68% in the same period. See page 4. This is one of the main weaknesses in the jobs data and the responsibility resides with the state agencies.

First, it is important to understand that both surveys are samples, and from this sample, an algorithm estimates the monthly numbers. And like all models, it will also provide the statistical error factor. In the household survey this statistical error factor is 600,000 and in the establishment survey it is 136,000. This means that the estimate of 73,000 new jobs in July falls within a “confidence range” of plus 209,000 to negative 63,000. This may shock most people; but it has always been true.

All in all, there are four adjustments made to employment data and as the BLS noted last week, seasonal adjustments were revised in the July report. Three adjustments are made to the surveys each month: seasonal adjustments, additional or revised survey data, and the birth death adjustment. First, there is a strong seasonal pattern of job losses in July (education/summer recess) and January (private sector layoffs) and the seasonal adjustment smooths this trend. Seasonal adjustments are revised regularly, as new data is accumulated. In our opinion, BLS mathematicians should have ignored data from the pandemic period since the mandated shut down of businesses disrupted all the normal seasonal patterns. However, this was not done, and seasonal adjustments have become problematic in the aftermath. See page 6. Second, data trickles in with a lag, so numbers are revised once new data is reported. Third, the birth death adjustment is used to adjust for the number of jobs added by business creation or jobs lost due to business closures. This data is not captured by the states in a timely manner, so a birth death adjustment is derived from historical information. Note that the estimate of births and deaths is applied to the not-seasonally-adjusted number and in July it added 257,000 jobs. (The not-seasonally adjusted establishment number in July was a loss of 1.05 million jobs.) The fourth adjustment to the data is a quarterly and/or annual revision to population estimates which comes from the Census Bureau. There will be a quarterly census adjustment to data in September. In sum, understanding how the data is created can be daunting, but using the data can be perilous for the uninformed.

The negative revisions in the establishment survey to the months of May and June severely damaged the trends in the jobs data. Whereas, the 3-month average of job creation had been 126,670 in April, it fell to 35,330 in July. In our opinion, when the 3-month average of job creation is over 100,000 it represents a good job market and when it exceeds 250,000 it defines a very strong job market. Conversely, a negative 3-month average of job creation has historically been a sign of a recession. This is what makes the sharp decline over the last three months so ominous. Although both the establishment and household surveys show job growth to be positive year-over-year, both trends are now below average and declining. While the 3-month average in the household survey can be quite volatile, it fell into negative territory in July! In short, the changes seen in the July jobs report were significant and have a distinct recessionary tilt. August’s employment report, released in September ahead of the FOMC meeting, will be important in terms of clarifying the weakness now appearing in the job market. See page 7.

Unemployment rates by level of education were particularly interesting in July. They show unemployment rates rising for high school and college graduates but falling for those with less than a high school degree and for those with some college or an associate degree. Note that the labor force for those with a bachelor’s degree is the largest and it nearly equals the sum of high school and associate degree graduates. It is nearly eight times as large as those with no high school degree. Moreover, it is an important category and typically has the lowest unemployment rate of all categories. In July the unemployment rate for those with a bachelor’s degree rose to 3.1% while the rate for those with an associate degree fell to 3.0%. It was an unusual convergence of unemployment rates. See page 9.

The ISM manufacturing index was 48.0 in July, down from 49.0 and most all components, with the exception of prices paid to suppliers, were below 50, an indication of contraction. The brightest spot in July was the increase in production from 50.3 to 51.4. The nonmanufacturing survey inched lower to 50.1 showing a tepid expansion. However, the employment indices were most important. The employment index for nonmanufacturing was below 50 for the fourth time in the last five months. As a result, both surveys show employment to be under 50 and contracting. See page 13. This is another indication that the job market may be weakening.

There were positive economic reports this week such as the second quarter PCE index at 2.1% (page 10) and a 3.0% increase in second quarter GDP (page 11). Second quarter earnings releases continue to beat expectations and consensus forecasts for 2025 and 2026 continue to rise. However, the real driver of equity prices in August may be the expectation that the fed funds rate will be cut at the September 16-17 meeting. History suggests that stocks rise in the months following the first rate cut, particularly if it is the first of a series of rate cuts. In short, despite the weakness seen in revised jobs data, the equity market is expected to be higher in the second half of the year.

Technical indicators have been bullish since May. Our 25-day up/down volume oscillator has not confirmed the new highs recorded in August, which suggests there could be a near-term pullback in the indices, but we would be a buyer on dips.

Gail Dudack

Click to Download