The FRED Blog has discussed why employment in retail and postal services peak around the Christmas holidays. Today we tap into a 2018 research piece from the Federal Reserve Bank of Chicago to discuss why construction employment may follow different patterns across states.

Our FRED graph above shows monthly employment data reported by the US Bureau of Labor Statistics. The solid lines are the numbers (in thousands) of employed workers in construction between May 2016 and June 2026 in two states: Kentucky in purple and Minnesota in blue. The dashed lines are the same employment figures adjusted for the seasonal impact of factors, such as weather, that affect overall economic activity in that industry. We picked those two states to make our point because they have markedly different weather during the winter and summer months.

Note there are far more construction employees in Minnesota than in Kentucky because the Northern state is more populous than the Southern / Midwestern state. Given this size difference, comparing annual employment peaks and throughs between states isn’t easy or straightforward.

To better tell the story behind the numbers, we created a second FRED graph that plots the size of the seasonal changes in employment as a fraction of the seasonally adjusted employment figures between May 2016 and June 2026. This graph shows similar overall patterns for both states: increases through August and declines until February, but a much wider seasonal range in Minnesota than in Kentucky. (That is, much higher highs and much lower lows for Minnesota.) This difference could be related to the ability to work through more of the winter months in the South relative to the North.

If you want to dive further into these patterns, check out the state-level employment data by industry in FRED.

How these graphs were created: Search FRED for “All Employees: Construction in Minnesota” and find the seasonally adjusted series (MNCONS). Click “Edit Graph” and navigate to the “Add Line” tab. Search for “All Employees: Construction in Minnesota” and find the non-seasonally adjusted series (MNCONSN) and click “Add Data Series.” Repeat with the seasonally adjusted (KYCONS) and not seasonally adjusted (KYCONSN) series for “All Employees: Construction in Kentucky” to complete the graph.

Suggested by Alison Booth and Diego Mendez-Carbajo.

FRED has not only current data series but also historic data series, some of which are very old. As we’ve mentioned on this blog, the oldest series in FRED is population data for the United Kingdom, dating back to 1086! The next 41 oldest series are also from the United Kingdom. Now that the United States is celebrating a quarter millennium of existence, let’s look at the oldest US series in FRED.

Our FRED graph above shows data on federal public expenditures, specifically for public works, that go as far back as 1791. More categories for public expenditures are available, with a couple also dating back to 1791.

The graph is dominated by the huge increase in military expenditures that came with WWI. Note that the data are calculated in current prices—that is, prices at the time, which have not been adjusted for inflation. And there’s no price series that is old enough to help reconcile this issue. Indeed, the data were computed retrospectively in 1920, which means that the data collection practices used were obviously not up to modern standards. In fact, it seems some liberties were taken with the definitions of the data. For example, the little bump in 1904 corresponds to the purchase of the Panama Canal and land around it: $40 million to France and $10 million to Panama.

How this graph was created: From the FRED home page, go to the “Browse Data By” list on the right side and choose “Category.” From the Categories page, go to the “Browse Data” list on the left side and choose “All Series.” From the All Series page, change “Sort by Popularity” on the right side of the listing to “Sort by Obs Start.” From this oldest series list, choose the first US series.

Suggested by Christian Zimmermann.

The recent AI boom has renewed debate about how official statistics capture changes in productivity growth and how any lags or limitations might impact monetary policy. For example, in the 1990s, some believed that low measures of productivity weren’t capturing the true benefits from new technology.

Back in 1996

Federal Reserve Chair Alan Greenspan argued in 1996 that the productivity gains associated with the information and communications technology boom were not yet visible in the official data. That judgment helped support his case for delaying preemptive interest-rate increases.

Shortly afterward, the 1999 comprehensive revision of the National Income and Product Accounts began treating software expenditures as capital investment. Together with other statistical changes, this revision raised estimates of the productivity growth that had occurred during the 1990s, bringing the official data closer in line with the acceleration in productivity that Greenspan believed had been under way.

The data, as seen in 1996, 2000, and 2026

Our ALFRED graph above compares three vintages of labor productivity growth data: The blue bars reflect early data available in September 1996. The green bars reflect the revised data available in February 2000, which incorporate the 1999 NIPA revision. The orange bars reflect the most-current data available at the time of this writing, as of June 2026.

• As of September 1996, the data indicated that labor productivity had grown by an average of just 0.89% between 1989 and 1995.
• By February 2000, average labor productivity growth for that same time period had been raised to 1.40%.
• As of  June 2026, after more revisions, it stands at 1.51%.

This comparison shows how weak measures of productivity growth appeared in real time and how subsequent revisions substantially altered the historical picture.

How this graph was created: Search ALFRED for “Nonfarm Business Sector: Labor Productivity (Output per Hour) for All Workers” and select the series with ID OPHNFB. Open “Edit Graph” and add the series three times. Set the as-of dates to September 10, 1996; February 8, 2000; and June 4, 2026. For each series, change the units to “Percent Change from Year Ago,” set the frequency to “Annual,” and use “Average” as the aggregation method. Adjust the observation period to begin in 1989 and end in 1995.

Suggested by Hannah Rubinton.