How many Americans currently hold jobs? And how does that compare with previous months, or even previous years?
For nearly a century, the Bureau of Labor Statistics, part of the US Department of Labor, has calculated the number of Americans who currently hold a job. Their main survey, also known as PAYEMS ("Payroll employees, seasonally adjusted"), tries to capture everyone who has a private-sector or government job, other than farm workers, the self employed, and people in the military.
PAYEMS numbers are usually released on the first Friday of each month, and reflect information from the previous month. The numbers are collected by retrieving employment data from a sample of more than 120,000 employers across the United States. If employment goes up, that's seen as a good sign, although it also points to a tighter labor market, which could lead to inflation. If employment goes down, that's seen as a bad sign for the economy, although it does ease inflation pressure.
Because PAYEMS is a sample, and because employers don't always have the most accurate information at hand, it's customary for each month's jobs report not only to include the latest data, but to revise previous months' data. So the July 2025 data, which was released on Friday, August 1st, included revisions to the May and June 2025 data.
Those revisions dramatically changed the view of the US economy, because it seemed that the BLS had overestimated the number of jobs by 250,000. And that's just the revision from June's numbers; the May numbers were also revised downward, but "only" by 125,000. It would thus seem that employment in the US is slowing down. Given the number of recent layoffs, and the persistent reports of trouble people are having finding a job, the data makes sense.
That these numbers really spooked the Trump administration, which has been claiming that the economy is booming. In response to this jobs report, President Trump claimed the numbers were politically motivated, and fired Erika McEntarfer, the head statistician at the Bureau of Labor Statistics (https://www.washingtonpost.com/business/2025/08/05/trump-jobs-data-firing-labor-statistics/). This has led to widespread fears that US government data, which has long been seen as non-partisan and reliable, might now be subject to partisan whims and interpretations.
This week, we'll look at the latest jobs report, and try to understand how big of a revision we actually saw in its historical context.
Data and five questions
You might expect that we could get this week's data from FRED (https://fred.stlouisfed.org/), the St. Louis Federal Reserve's online economic data portal. However, FRED only keeps the most recent revision of data. We need to instead use ALFRED (https://alfred.stlouisfed.org/), which keeps archived data sets. It indicates the "vintage" of data, so that we can keep track of which figures were released when, and compare them with other revisions.
You can get the data by going to ALFRED's page for PAYEMS (https://alfred.stlouisfed.org/series?seid=PAYEMS).
Click on the "download" button, and ask for "all vintages." We want the data measured in thousands of people, and for both observations and vintages to start in January, 1965. I retrieved the data in Excel format. We'll use this Excel file for this week's analysis.
This weeks' learning goals include working with Excel files, using window functions, dates and times, and plotting with Plotly.
Paid subscribers, including members of my LernerPython+data plan, can download the data from a link at the bottom of this page.
I'll be back tomorrow with complete solutions and explanations.
Meanwhile, here are my five questions and tasks:
- Create a Pandas data frame using the data in the Excel spreadsheet. The index should contain
datetimedata from theobservation_datecolumn. The columns should also be havedatetimevalues, based on the final six digits of each column's name. What does each row represent? What does each column represent? - What were the July PAYEMS numbers that we got on Friday, August 1st for July? What were the revised numbers for May and June? By how much did the June numbers change since last month's report? By how much did the May numbers change since the June report? What sorts of adjustments did we see in May's data, first in July and then in August?